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>
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
38 int restore_backup(struct supertype
*st
,
39 struct mdinfo
*content
,
49 int disk_count
= next_spare
+ working_disks
;
50 char *backup_file
= *backup_filep
;
52 dprintf("Called restore_backup()\n");
53 fdlist
= xmalloc(sizeof(int) * disk_count
);
55 enable_fds(next_spare
);
56 for (i
= 0; i
< next_spare
; i
++)
58 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
64 fd
= dev_open(buf
, O_RDWR
);
66 if (dev
->disk
.raid_disk
>= 0)
67 fdlist
[dev
->disk
.raid_disk
] = fd
;
69 fdlist
[next_spare
++] = fd
;
73 backup_file
= locate_backup(content
->sys_name
);
74 *backup_filep
= backup_file
;
77 if (st
->ss
->external
&& st
->ss
->recover_backup
)
78 err
= st
->ss
->recover_backup(st
, content
);
80 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
81 backup_file
, verbose
> 0);
83 while (next_spare
> 0) {
85 if (fdlist
[next_spare
] >= 0)
86 close(fdlist
[next_spare
]);
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
92 pr_err("Possibly you need to specify a --backup-file\n");
96 dprintf("restore_backup() returns status OK.\n");
100 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
108 * This means that we need to *find* all other devices.
115 struct supertype
*st
= NULL
;
116 char *subarray
= NULL
;
118 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
119 pr_err("cannot get array info for %s\n", devname
);
123 if (info
.array
.level
!= -1) {
124 pr_err("can only add devices to linear arrays\n");
128 st
= super_by_fd(fd
, &subarray
);
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info
.array
.major_version
);
136 pr_err("Cannot grow linear sub-arrays yet\n");
142 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
144 pr_err("cannot open %s\n", newdev
);
149 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
150 pr_err("%s is not a block device!\n", newdev
);
155 /* now check out all the devices and make sure we can read the
157 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
158 mdu_disk_info_t disk
;
161 st
->ss
->free_super(st
);
164 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
165 pr_err("cannot get device detail for device %d\n",
171 dv
= map_dev(disk
.major
, disk
.minor
, 1);
173 pr_err("cannot find device file for device %d\n",
179 fd2
= dev_open(dv
, O_RDWR
);
181 pr_err("cannot open device file %s\n", dv
);
187 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
188 pr_err("cannot find super block on %s\n", dv
);
196 /* Ok, looks good. Lets update the superblock and write it out to
200 info
.disk
.number
= d
;
201 info
.disk
.major
= major(stb
.st_rdev
);
202 info
.disk
.minor
= minor(stb
.st_rdev
);
203 info
.disk
.raid_disk
= d
;
204 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
205 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
208 if (st
->ss
->store_super(st
, nfd
)) {
209 pr_err("Cannot store new superblock on %s\n",
216 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
217 pr_err("Cannot add new disk to this array\n");
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
224 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
225 pr_err("cannot get array info for %s\n", devname
);
230 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
231 mdu_disk_info_t disk
;
235 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
236 pr_err("cannot get device detail for device %d\n",
240 dv
= map_dev(disk
.major
, disk
.minor
, 1);
242 pr_err("cannot find device file for device %d\n",
246 fd2
= dev_open(dv
, O_RDWR
);
248 pr_err("cannot open device file %s\n", dv
);
251 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
252 pr_err("cannot find super block on %s\n", dv
);
256 info
.array
.raid_disks
= nd
+1;
257 info
.array
.nr_disks
= nd
+1;
258 info
.array
.active_disks
= nd
+1;
259 info
.array
.working_disks
= nd
+1;
261 st
->ss
->update_super(st
, &info
, "linear-grow-update", 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 int vers
= md_get_version(fd
);
292 unsigned long long bitmapsize
, array_size
;
295 major
= BITMAP_MAJOR_HOSTENDIAN
;
296 pr_err("Warning - bitmaps created on this kernel are not portable\n"
297 " between different architectures. Consider upgrading the Linux kernel.\n");
300 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
302 pr_err("Memory allocation failure.\n");
304 pr_err("bitmaps not supported by this kernel.\n");
307 if (bmf
.pathname
[0]) {
308 if (strcmp(s
->bitmap_file
,"none")==0) {
309 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
310 pr_err("failed to remove bitmap %s\n",
316 pr_err("%s already has a bitmap (%s)\n",
317 devname
, bmf
.pathname
);
320 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
321 pr_err("cannot get array status for %s\n", devname
);
324 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
325 if (strcmp(s
->bitmap_file
, "none")==0) {
326 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
327 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
328 pr_err("failed to remove internal bitmap.\n");
333 pr_err("Internal bitmap already present on %s\n",
338 if (strcmp(s
->bitmap_file
, "none") == 0) {
339 pr_err("no bitmap found on %s\n", devname
);
342 if (array
.level
<= 0) {
343 pr_err("Bitmaps not meaningful with level %s\n",
344 map_num(pers
, array
.level
)?:"of this array");
347 bitmapsize
= array
.size
;
349 if (get_dev_size(fd
, NULL
, &array_size
) &&
350 array_size
> (0x7fffffffULL
<<9)) {
351 /* Array is big enough that we cannot trust array.size
352 * try other approaches
354 bitmapsize
= get_component_size(fd
);
356 if (bitmapsize
== 0) {
357 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
361 if (array
.level
== 10) {
362 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
363 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
366 st
= super_by_fd(fd
, &subarray
);
368 pr_err("Cannot understand version %d.%d\n",
369 array
.major_version
, array
.minor_version
);
373 pr_err("Cannot add bitmaps to sub-arrays yet\n");
378 if (strcmp(s
->bitmap_file
, "internal") == 0) {
381 int offset_setable
= 0;
383 if (st
->ss
->add_internal_bitmap
== NULL
) {
384 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
387 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
390 for (d
=0; d
< st
->max_devs
; d
++) {
391 mdu_disk_info_t disk
;
394 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
396 if (disk
.major
== 0 &&
399 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
401 dv
= map_dev(disk
.major
, disk
.minor
, 1);
403 int fd2
= dev_open(dv
, O_RDWR
);
406 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
407 if (st
->ss
->add_internal_bitmap(
409 &s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
410 bitmapsize
, offset_setable
,
413 st
->ss
->write_bitmap(st
, fd2
);
415 pr_err("failed to create internal bitmap - chunksize problem.\n");
423 if (offset_setable
) {
424 st
->ss
->getinfo_super(st
, mdi
, NULL
);
425 sysfs_init(mdi
, fd
, NULL
);
426 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
429 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
430 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
434 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
435 pr_err("failed to set internal bitmap.\n");
442 int max_devs
= st
->max_devs
;
444 /* try to load a superblock */
445 for (d
= 0; d
< max_devs
; d
++) {
446 mdu_disk_info_t disk
;
450 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
452 if ((disk
.major
==0 && disk
.minor
==0) ||
453 (disk
.state
& (1<<MD_DISK_REMOVED
)))
455 dv
= map_dev(disk
.major
, disk
.minor
, 1);
458 fd2
= dev_open(dv
, O_RDONLY
);
460 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
462 st
->ss
->uuid_from_super(st
, uuid
);
469 pr_err("cannot find UUID for array!\n");
472 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
473 c
->delay
, s
->write_behind
, bitmapsize
, major
)) {
476 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
478 pr_err("weird: %s cannot be opened\n",
482 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
485 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
486 pr_err("Cannot set bitmap file for %s: %s\n",
487 devname
, strerror(err
));
496 * When reshaping an array we might need to backup some data.
497 * This is written to all spares with a 'super_block' describing it.
498 * The superblock goes 4K from the end of the used space on the
500 * It if written after the backup is complete.
501 * It has the following structure.
504 static struct mdp_backup_super
{
505 char magic
[16]; /* md_backup_data-1 or -2 */
508 /* start/sizes in 512byte sectors */
509 __u64 devstart
; /* address on backup device/file of data */
512 __u32 sb_csum
; /* csum of preceeding bytes. */
514 __u64 devstart2
; /* offset in to data of second section */
517 __u32 sb_csum2
; /* csum of preceeding bytes. */
519 } __attribute__((aligned(512))) bsb
, bsb2
;
521 static __u32
bsb_csum(char *buf
, int len
)
525 for (i
= 0; i
< len
; i
++)
526 csum
= (csum
<<3) + buf
[0];
527 return __cpu_to_le32(csum
);
530 static int check_idle(struct supertype
*st
)
532 /* Check that all member arrays for this container, or the
533 * container of this array, are idle
535 char *container
= (st
->container_devnm
[0]
536 ? st
->container_devnm
: st
->devnm
);
537 struct mdstat_ent
*ent
, *e
;
540 ent
= mdstat_read(0, 0);
541 for (e
= ent
; e
; e
= e
->next
) {
542 if (!is_container_member(e
, container
))
544 if (e
->percent
>= 0) {
553 static int freeze_container(struct supertype
*st
)
555 char *container
= (st
->container_devnm
[0]
556 ? st
->container_devnm
: st
->devnm
);
561 if (block_monitor(container
, 1)) {
562 pr_err("failed to freeze container\n");
569 static void unfreeze_container(struct supertype
*st
)
571 char *container
= (st
->container_devnm
[0]
572 ? st
->container_devnm
: st
->devnm
);
574 unblock_monitor(container
, 1);
577 static int freeze(struct supertype
*st
)
579 /* Try to freeze resync/rebuild on this array/container.
580 * Return -1 if the array is busy,
581 * return -2 container cannot be frozen,
582 * return 0 if this kernel doesn't support 'frozen'
583 * return 1 if it worked.
585 if (st
->ss
->external
)
586 return freeze_container(st
);
588 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
594 /* Need to clear any 'read-auto' status */
595 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
596 strncmp(buf
, "read-auto", 9) == 0)
597 sysfs_set_str(sra
, NULL
, "array_state", "clean");
599 err
= sysfs_freeze_array(sra
);
605 static void unfreeze(struct supertype
*st
)
607 if (st
->ss
->external
)
608 return unfreeze_container(st
);
610 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
614 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0
615 && strcmp(buf
, "frozen\n") == 0) {
616 printf("unfreeze\n");
617 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
623 static void wait_reshape(struct mdinfo
*sra
)
625 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
631 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
632 strncmp(action
, "reshape", 7) == 0)
633 sysfs_wait(fd
, NULL
);
637 static int reshape_super(struct supertype
*st
, unsigned long long size
,
638 int level
, int layout
, int chunksize
, int raid_disks
,
639 int delta_disks
, char *backup_file
, char *dev
,
640 int direction
, int verbose
)
642 /* nothing extra to check in the native case */
643 if (!st
->ss
->external
)
645 if (!st
->ss
->reshape_super
||
646 !st
->ss
->manage_reshape
) {
647 pr_err("%s metadata does not support reshape\n",
652 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
653 raid_disks
, delta_disks
, backup_file
, dev
,
657 static void sync_metadata(struct supertype
*st
)
659 if (st
->ss
->external
) {
660 if (st
->update_tail
) {
661 flush_metadata_updates(st
);
662 st
->update_tail
= &st
->updates
;
664 st
->ss
->sync_metadata(st
);
668 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
670 /* when dealing with external metadata subarrays we need to be
671 * prepared to handle EAGAIN. The kernel may need to wait for
672 * mdmon to mark the array active so the kernel can handle
673 * allocations/writeback when preparing the reshape action
674 * (md_allow_write()). We temporarily disable safe_mode_delay
675 * to close a race with the array_state going clean before the
676 * next write to raid_disks / stripe_cache_size
681 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
683 (strcmp(name
, "raid_disks") != 0 &&
684 strcmp(name
, "stripe_cache_size") != 0))
685 return sysfs_set_num(sra
, NULL
, name
, n
);
687 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
690 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
691 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
692 if (rc
< 0 && errno
== EAGAIN
) {
693 ping_monitor(container
);
694 /* if we get EAGAIN here then the monitor is not active
697 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
699 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
703 int start_reshape(struct mdinfo
*sra
, int already_running
,
704 int before_data_disks
, int data_disks
)
707 unsigned long long sync_max_to_set
;
709 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
710 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
711 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
712 sra
->reshape_progress
);
713 if (before_data_disks
<= data_disks
)
714 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
716 sync_max_to_set
= (sra
->component_size
* data_disks
717 - sra
->reshape_progress
) / data_disks
;
718 if (!already_running
)
719 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
720 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
721 if (!already_running
&& err
== 0) {
724 err
= sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
727 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
732 void abort_reshape(struct mdinfo
*sra
)
734 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
735 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
736 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
737 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
738 sysfs_set_num(sra
, NULL
, "sync_min", 0);
739 // It isn't safe to reset sync_max as we aren't monitoring.
740 // Array really should be stopped at this point.
743 int remove_disks_for_takeover(struct supertype
*st
,
748 struct mdinfo
*remaining
;
751 if (sra
->array
.level
== 10)
752 nr_of_copies
= layout
& 0xff;
753 else if (sra
->array
.level
== 1)
754 nr_of_copies
= sra
->array
.raid_disks
;
758 remaining
= sra
->devs
;
760 /* for each 'copy', select one device and remove from the list. */
761 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
762 struct mdinfo
**diskp
;
765 /* Find a working device to keep */
766 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
767 struct mdinfo
*disk
= *diskp
;
769 if (disk
->disk
.raid_disk
< slot
)
771 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
773 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
775 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
777 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
780 /* We have found a good disk to use! */
782 disk
->next
= sra
->devs
;
791 if (slot
< sra
->array
.raid_disks
) {
792 /* didn't find all slots */
798 sra
->devs
= remaining
;
802 /* Remove all 'remaining' devices from the array */
804 struct mdinfo
*sd
= remaining
;
805 remaining
= sd
->next
;
807 sysfs_set_str(sra
, sd
, "state", "faulty");
808 sysfs_set_str(sra
, sd
, "slot", "none");
809 /* for external metadata disks should be removed in mdmon */
810 if (!st
->ss
->external
)
811 sysfs_set_str(sra
, sd
, "state", "remove");
812 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
813 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
814 sd
->next
= sra
->devs
;
820 void reshape_free_fdlist(int *fdlist
,
821 unsigned long long *offsets
,
826 for (i
= 0; i
< size
; i
++)
834 int reshape_prepare_fdlist(char *devname
,
838 unsigned long blocks
,
841 unsigned long long *offsets
)
847 for (d
= 0; d
<= nrdisks
; d
++)
850 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
851 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
853 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
854 char *dn
= map_dev(sd
->disk
.major
,
856 fdlist
[sd
->disk
.raid_disk
]
857 = dev_open(dn
, O_RDONLY
);
858 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
859 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
860 pr_err("%s: cannot open component %s\n",
861 devname
, dn
? dn
: "-unknown-");
865 } else if (backup_file
== NULL
) {
867 char *dn
= map_dev(sd
->disk
.major
,
869 fdlist
[d
] = dev_open(dn
, O_RDWR
);
870 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
872 pr_err("%s: cannot open component %s\n",
873 devname
, dn
? dn
: "-unknown-");
884 int reshape_open_backup_file(char *backup_file
,
889 unsigned long long *offsets
,
893 /* Return 1 on success, 0 on any form of failure */
894 /* need to check backup file is large enough */
900 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
904 pr_err("%s: cannot create backup file %s: %s\n",
905 devname
, backup_file
, strerror(errno
));
908 /* Guard against backup file being on array device.
909 * If array is partitioned or if LVM etc is in the
910 * way this will not notice, but it is better than
913 fstat(*fdlist
, &stb
);
916 if (stb
.st_rdev
== dev
) {
917 pr_err("backup file must NOT be on the array being reshaped.\n");
923 for (i
=0; i
< blocks
+ 8 ; i
++) {
924 if (write(*fdlist
, buf
, 512) != 512) {
925 pr_err("%s: cannot create backup file %s: %s\n",
926 devname
, backup_file
, strerror(errno
));
930 if (fsync(*fdlist
) != 0) {
931 pr_err("%s: cannot create backup file %s: %s\n",
932 devname
, backup_file
, strerror(errno
));
936 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
937 char *bu
= make_backup(sys_name
);
938 if (symlink(backup_file
, bu
))
939 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
947 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
948 unsigned int ndata
, unsigned int odata
)
950 unsigned long a
, b
, blocks
;
951 /* So how much do we need to backup.
952 * We need an amount of data which is both a whole number of
953 * old stripes and a whole number of new stripes.
954 * So LCM for (chunksize*datadisks).
956 a
= (ochunk
/512) * odata
;
957 b
= (nchunk
/512) * ndata
;
960 /* LCM == product / GCD */
961 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
966 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
968 /* Based on the current array state in info->array and
969 * the changes in info->new_* etc, determine:
970 * - whether the change is possible
971 * - Intermediate level/raid_disks/layout
972 * - whether a restriping reshape is needed
973 * - number of sectors in minimum change unit. This
974 * will cover a whole number of stripes in 'before' and
977 * Return message if the change should be rejected
978 * NULL if the change can be achieved
980 * This can be called as part of starting a reshape, or
981 * when assembling an array that is undergoing reshape.
983 int near
, far
, offset
, copies
;
985 int old_chunk
, new_chunk
;
986 /* delta_parity records change in number of devices
987 * caused by level change
989 int delta_parity
= 0;
991 memset(re
, 0, sizeof(*re
));
993 /* If a new level not explicitly given, we assume no-change */
994 if (info
->new_level
== UnSet
)
995 info
->new_level
= info
->array
.level
;
998 switch (info
->new_level
) {
1004 /* chunk size is meaningful, must divide component_size
1007 if (info
->component_size
% (info
->new_chunk
/512)) {
1008 unsigned long long shrink
= info
->component_size
;
1009 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1010 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1011 info
->new_chunk
/1024, info
->component_size
/2);
1012 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1014 pr_err("will shrink the array so the given chunk size would work.\n");
1019 return "chunk size not meaningful for this level";
1022 info
->new_chunk
= info
->array
.chunk_size
;
1024 switch (info
->array
.level
) {
1026 return "No reshape is possibly for this RAID level";
1028 if (info
->delta_disks
!= UnSet
)
1029 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1031 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1033 /* RAID1 can convert to RAID1 with different disks, or
1034 * raid5 with 2 disks, or
1037 if (info
->new_level
> 1 &&
1038 (info
->component_size
& 7))
1039 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1040 if (info
->new_level
== 0) {
1041 if (info
->delta_disks
!= UnSet
&&
1042 info
->delta_disks
!= 0)
1043 return "Cannot change number of disks with RAID1->RAID0 conversion";
1045 re
->before
.data_disks
= 1;
1046 re
->after
.data_disks
= 1;
1049 if (info
->new_level
== 1) {
1050 if (info
->delta_disks
== UnSet
)
1051 /* Don't know what to do */
1052 return "no change requested for Growing RAID1";
1056 if (info
->array
.raid_disks
== 2 &&
1057 info
->new_level
== 5) {
1060 re
->before
.data_disks
= 1;
1061 if (info
->delta_disks
!= UnSet
&&
1062 info
->delta_disks
!= 0)
1063 re
->after
.data_disks
= 1 + info
->delta_disks
;
1065 re
->after
.data_disks
= 1;
1066 if (re
->after
.data_disks
< 1)
1067 return "Number of disks too small for RAID5";
1069 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1070 info
->array
.chunk_size
= 65536;
1073 /* Could do some multi-stage conversions, but leave that to
1076 return "Impossibly level change request for RAID1";
1079 /* RAID10 can be converted from near mode to
1080 * RAID0 by removing some devices.
1081 * It can also be reshaped if the kernel supports
1084 switch (info
->new_level
) {
1086 if ((info
->array
.layout
& ~0xff) != 0x100)
1087 return "Cannot Grow RAID10 with far/offset layout";
1088 /* number of devices must be multiple of number of copies */
1089 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1090 return "RAID10 layout too complex for Grow operation";
1092 new_disks
= (info
->array
.raid_disks
1093 / (info
->array
.layout
& 0xff));
1094 if (info
->delta_disks
== UnSet
)
1095 info
->delta_disks
= (new_disks
1096 - info
->array
.raid_disks
);
1098 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1099 return "New number of raid-devices impossible for RAID10";
1100 if (info
->new_chunk
&&
1101 info
->new_chunk
!= info
->array
.chunk_size
)
1102 return "Cannot change chunk-size with RAID10 Grow";
1106 re
->before
.data_disks
= new_disks
;
1107 re
->after
.data_disks
= re
->before
.data_disks
;
1111 near
= info
->array
.layout
& 0xff;
1112 far
= (info
->array
.layout
>> 8) & 0xff;
1113 offset
= info
->array
.layout
& 0x10000;
1114 if (far
> 1 && !offset
)
1115 return "Cannot reshape RAID10 in far-mode";
1116 copies
= near
* far
;
1118 old_chunk
= info
->array
.chunk_size
* far
;
1120 if (info
->new_layout
== UnSet
)
1121 info
->new_layout
= info
->array
.layout
;
1123 near
= info
->new_layout
& 0xff;
1124 far
= (info
->new_layout
>> 8) & 0xff;
1125 offset
= info
->new_layout
& 0x10000;
1126 if (far
> 1 && !offset
)
1127 return "Cannot reshape RAID10 to far-mode";
1128 if (near
* far
!= copies
)
1129 return "Cannot change number of copies when reshaping RAID10";
1131 if (info
->delta_disks
== UnSet
)
1132 info
->delta_disks
= 0;
1133 new_disks
= (info
->array
.raid_disks
+
1136 new_chunk
= info
->new_chunk
* far
;
1139 re
->before
.layout
= info
->array
.layout
;
1140 re
->before
.data_disks
= info
->array
.raid_disks
;
1141 re
->after
.layout
= info
->new_layout
;
1142 re
->after
.data_disks
= new_disks
;
1143 /* For RAID10 we don't do backup but do allow reshape,
1144 * so set backup_blocks to INVALID_SECTORS rather than
1146 * And there is no need to synchronise stripes on both
1147 * 'old' and 'new'. So the important
1148 * number is the minimum data_offset difference
1149 * which is the larger of (offset copies * chunk).
1151 re
->backup_blocks
= INVALID_SECTORS
;
1152 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1153 if (new_disks
< re
->before
.data_disks
&&
1154 info
->space_after
< re
->min_offset_change
)
1155 /* Reduce component size by one chunk */
1156 re
->new_size
= (info
->component_size
-
1157 re
->min_offset_change
);
1159 re
->new_size
= info
->component_size
;
1160 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1164 return "RAID10 can only be changed to RAID0";
1167 /* RAID0 can be converted to RAID10, or to RAID456 */
1168 if (info
->new_level
== 10) {
1169 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1170 /* Assume near=2 layout */
1171 info
->new_layout
= 0x102;
1172 info
->delta_disks
= info
->array
.raid_disks
;
1174 if (info
->new_layout
== UnSet
) {
1175 int copies
= 1 + (info
->delta_disks
1176 / info
->array
.raid_disks
);
1177 if (info
->array
.raid_disks
* (copies
-1)
1178 != info
->delta_disks
)
1179 return "Impossible number of devices for RAID0->RAID10";
1180 info
->new_layout
= 0x100 + copies
;
1182 if (info
->delta_disks
== UnSet
) {
1183 int copies
= info
->new_layout
& 0xff;
1184 if (info
->new_layout
!= 0x100 + copies
)
1185 return "New layout impossible for RAID0->RAID10";;
1186 info
->delta_disks
= (copies
- 1) *
1187 info
->array
.raid_disks
;
1189 if (info
->new_chunk
&&
1190 info
->new_chunk
!= info
->array
.chunk_size
)
1191 return "Cannot change chunk-size with RAID0->RAID10";
1194 re
->before
.data_disks
= (info
->array
.raid_disks
+
1196 re
->after
.data_disks
= re
->before
.data_disks
;
1197 re
->before
.layout
= info
->new_layout
;
1201 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1202 * a raid4 style layout of the final level.
1204 switch (info
->new_level
) {
1209 re
->before
.layout
= 0;
1214 re
->before
.layout
= ALGORITHM_PARITY_N
;
1215 if (info
->new_layout
== UnSet
)
1216 info
->new_layout
= map_name(r5layout
, "default");
1221 re
->before
.layout
= ALGORITHM_PARITY_N
;
1222 if (info
->new_layout
== UnSet
)
1223 info
->new_layout
= map_name(r6layout
, "default");
1226 return "Impossible level change requested";
1228 re
->before
.data_disks
= info
->array
.raid_disks
;
1229 /* determining 'after' layout happens outside this 'switch' */
1233 info
->array
.layout
= ALGORITHM_PARITY_N
;
1235 switch (info
->new_level
) {
1239 re
->level
= info
->array
.level
;
1240 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1241 re
->before
.layout
= info
->array
.layout
;
1245 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1246 re
->before
.layout
= info
->array
.layout
;
1251 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1252 switch (info
->array
.layout
) {
1253 case ALGORITHM_LEFT_ASYMMETRIC
:
1254 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1256 case ALGORITHM_RIGHT_ASYMMETRIC
:
1257 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1259 case ALGORITHM_LEFT_SYMMETRIC
:
1260 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1262 case ALGORITHM_RIGHT_SYMMETRIC
:
1263 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1265 case ALGORITHM_PARITY_0
:
1266 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1268 case ALGORITHM_PARITY_N
:
1269 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1272 return "Cannot convert an array with this layout";
1276 if (info
->array
.raid_disks
!= 2)
1277 return "Can only convert a 2-device array to RAID1";
1278 if (info
->delta_disks
!= UnSet
&&
1279 info
->delta_disks
!= 0)
1280 return "Cannot set raid_disk when converting RAID5->RAID1";
1282 info
->new_chunk
= 0;
1285 return "Impossible level change requested";
1289 switch (info
->new_level
) {
1295 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1296 re
->before
.layout
= info
->array
.layout
;
1299 return "Impossible level change requested";
1304 /* If we reached here then it looks like a re-stripe is
1305 * happening. We have determined the intermediate level
1306 * and initial raid_disks/layout and stored these in 're'.
1308 * We need to deduce the final layout that can be atomically
1309 * converted to the end state.
1311 switch (info
->new_level
) {
1313 /* We can only get to RAID0 from RAID4 or RAID5
1314 * with appropriate layout and one extra device
1316 if (re
->level
!= 4 && re
->level
!= 5)
1317 return "Cannot covert to RAID0 from this level";
1319 switch (re
->level
) {
1321 re
->before
.layout
= 0;
1322 re
->after
.layout
= 0;
1325 re
->after
.layout
= ALGORITHM_PARITY_N
;
1331 /* We can only get to RAID4 from RAID5 */
1332 if (re
->level
!= 4 && re
->level
!= 5)
1333 return "Cannot convert to RAID4 from this level";
1335 switch (re
->level
) {
1337 re
->after
.layout
= 0;
1340 re
->after
.layout
= ALGORITHM_PARITY_N
;
1346 /* We get to RAID5 from RAID5 or RAID6 */
1347 if (re
->level
!= 5 && re
->level
!= 6)
1348 return "Cannot convert to RAID5 from this level";
1350 switch (re
->level
) {
1352 if (info
->new_layout
== UnSet
)
1353 re
->after
.layout
= re
->before
.layout
;
1355 re
->after
.layout
= info
->new_layout
;
1358 if (info
->new_layout
== UnSet
)
1359 info
->new_layout
= re
->before
.layout
;
1361 /* after.layout needs to be raid6 version of new_layout */
1362 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1363 re
->after
.layout
= ALGORITHM_PARITY_N
;
1366 char *ls
= map_num(r5layout
, info
->new_layout
);
1369 /* Current RAID6 layout has a RAID5
1372 strcat(strcpy(layout
, ls
), "-6");
1373 l
= map_name(r6layout
, layout
);
1375 return "Cannot find RAID6 layout to convert to";
1377 /* Current RAID6 has no equivalent.
1378 * If it is already a '-6' layout we
1379 * can leave it unchanged, else we must
1382 ls
= map_num(r6layout
, info
->new_layout
);
1384 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1385 return "Please specify new layout";
1386 l
= info
->new_layout
;
1388 re
->after
.layout
= l
;
1394 /* We must already be at level 6 */
1396 return "Impossible level change";
1397 if (info
->new_layout
== UnSet
)
1398 re
->after
.layout
= info
->array
.layout
;
1400 re
->after
.layout
= info
->new_layout
;
1403 return "Impossible level change requested";
1405 if (info
->delta_disks
== UnSet
)
1406 info
->delta_disks
= delta_parity
;
1408 re
->after
.data_disks
= (re
->before
.data_disks
1411 switch (re
->level
) {
1412 case 6: re
->parity
= 2;
1415 case 5: re
->parity
= 1;
1417 default: re
->parity
= 0;
1420 /* So we have a restripe operation, we need to calculate the number
1421 * of blocks per reshape operation.
1423 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1424 if (info
->new_chunk
== 0)
1425 info
->new_chunk
= info
->array
.chunk_size
;
1426 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1427 re
->after
.layout
== re
->before
.layout
&&
1428 info
->new_chunk
== info
->array
.chunk_size
) {
1429 /* Nothing to change, can change level immediately. */
1430 re
->level
= info
->new_level
;
1431 re
->backup_blocks
= 0;
1434 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1435 /* chunk and layout changes make no difference */
1436 re
->level
= info
->new_level
;
1437 re
->backup_blocks
= 0;
1441 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1442 get_linux_version() < 2006032)
1443 return "in-place reshape is not safe before 2.6.32 - sorry.";
1445 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1446 get_linux_version() < 2006030)
1447 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1449 re
->backup_blocks
= compute_backup_blocks(
1450 info
->new_chunk
, info
->array
.chunk_size
,
1451 re
->after
.data_disks
,
1452 re
->before
.data_disks
);
1453 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1455 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1459 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1462 struct mdinfo
*info
;
1466 if ((st
== NULL
) || (sra
== NULL
))
1469 if (text_version
== NULL
)
1470 text_version
= sra
->text_version
;
1471 subarray
= strchr(text_version
+1, '/')+1;
1472 info
= st
->ss
->container_content(st
, subarray
);
1474 unsigned long long current_size
= 0;
1475 unsigned long long new_size
=
1476 info
->custom_array_size
/2;
1478 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1479 new_size
> current_size
) {
1480 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1482 dprintf("Error: Cannot set array size");
1485 dprintf("Array size changed");
1487 dprintf_cont(" from %llu to %llu.\n",
1488 current_size
, new_size
);
1492 dprintf("Error: set_array_size(): info pointer in NULL\n");
1497 static int reshape_array(char *container
, int fd
, char *devname
,
1498 struct supertype
*st
, struct mdinfo
*info
,
1499 int force
, struct mddev_dev
*devlist
,
1500 unsigned long long data_offset
,
1501 char *backup_file
, int verbose
, int forked
,
1502 int restart
, int freeze_reshape
);
1503 static int reshape_container(char *container
, char *devname
,
1505 struct supertype
*st
,
1506 struct mdinfo
*info
,
1508 char *backup_file
, int verbose
,
1509 int forked
, int restart
, int freeze_reshape
);
1511 int Grow_reshape(char *devname
, int fd
,
1512 struct mddev_dev
*devlist
,
1513 unsigned long long data_offset
,
1514 struct context
*c
, struct shape
*s
)
1516 /* Make some changes in the shape of an array.
1517 * The kernel must support the change.
1519 * There are three different changes. Each can trigger
1520 * a resync or recovery so we freeze that until we have
1521 * requested everything (if kernel supports freezing - 2.6.30).
1523 * - change size (i.e. component_size)
1525 * - change layout/chunksize/ndisks
1527 * The last can require a reshape. It is different on different
1528 * levels so we need to check the level before actioning it.
1529 * Some times the level change needs to be requested after the
1530 * reshape (e.g. raid6->raid5, raid5->raid0)
1533 struct mdu_array_info_s array
;
1535 struct supertype
*st
;
1536 char *subarray
= NULL
;
1540 char *container
= NULL
;
1543 struct mddev_dev
*dv
;
1549 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1550 pr_err("%s is not an active md array - aborting\n",
1554 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10
1555 && (array
.level
< 4 || array
.level
> 6)) {
1556 pr_err("--grow --data-offset not yet supported\n");
1561 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1562 pr_err("cannot change component size at the same time as other changes.\n"
1563 " Change size first, then check data is intact before making other changes.\n");
1567 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&& array
.level
> 1 &&
1568 get_linux_version() < 2006032 &&
1569 !check_env("MDADM_FORCE_FEWER")) {
1570 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1571 " Please use a newer kernel\n");
1575 st
= super_by_fd(fd
, &subarray
);
1577 pr_err("Unable to determine metadata format for %s\n", devname
);
1580 if (s
->raiddisks
> st
->max_devs
) {
1581 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1585 /* in the external case we need to check that the requested reshape is
1586 * supported, and perform an initial check that the container holds the
1587 * pre-requisite spare devices (mdmon owns final validation)
1589 if (st
->ss
->external
) {
1593 container
= st
->container_devnm
;
1594 cfd
= open_dev_excl(st
->container_devnm
);
1596 container
= st
->devnm
;
1598 cfd
= open_dev_excl(st
->devnm
);
1602 pr_err("Unable to open container for %s\n",
1608 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1611 pr_err("Cannot read superblock for %s\n",
1617 /* check if operation is supported for metadata handler */
1618 if (st
->ss
->container_content
) {
1619 struct mdinfo
*cc
= NULL
;
1620 struct mdinfo
*content
= NULL
;
1622 cc
= st
->ss
->container_content(st
, subarray
);
1623 for (content
= cc
; content
; content
= content
->next
) {
1624 int allow_reshape
= 1;
1626 /* check if reshape is allowed based on metadata
1627 * indications stored in content.array.status
1629 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
1631 if (content
->array
.state
1632 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))
1634 if (!allow_reshape
) {
1635 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1636 devname
, container
);
1644 if (mdmon_running(container
))
1645 st
->update_tail
= &st
->updates
;
1649 for (dv
= devlist
; dv
; dv
= dv
->next
)
1651 if (s
->raiddisks
> array
.raid_disks
&&
1652 array
.spare_disks
+added_disks
< (s
->raiddisks
- array
.raid_disks
) &&
1654 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1655 " Use --force to over-ride this check.\n",
1656 s
->raiddisks
- array
.raid_disks
,
1657 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1658 array
.spare_disks
+ added_disks
);
1662 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
1663 | GET_STATE
| GET_VERSION
);
1665 if (st
->ss
->external
&& subarray
== NULL
) {
1666 array
.level
= LEVEL_CONTAINER
;
1667 sra
->array
.level
= LEVEL_CONTAINER
;
1670 pr_err("failed to read sysfs parameters for %s\n",
1674 frozen
= freeze(st
);
1676 /* freeze() already spewed the reason */
1679 } else if (frozen
< 0) {
1680 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1685 /* ========= set size =============== */
1686 if (s
->size
> 0 && (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1687 unsigned long long orig_size
= get_component_size(fd
)/2;
1688 unsigned long long min_csize
;
1690 int raid0_takeover
= 0;
1693 orig_size
= (unsigned) array
.size
;
1695 if (orig_size
== 0) {
1696 pr_err("Cannot set device size in this type of array.\n");
1701 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1702 devname
, APPLY_METADATA_CHANGES
, c
->verbose
> 0)) {
1707 if (st
->ss
->external
) {
1708 /* metadata can have size limitation
1709 * update size value according to metadata information
1711 struct mdinfo
*sizeinfo
=
1712 st
->ss
->container_content(st
, subarray
);
1714 unsigned long long new_size
=
1715 sizeinfo
->custom_array_size
/2;
1716 int data_disks
= get_data_disks(
1717 sizeinfo
->array
.level
,
1718 sizeinfo
->array
.layout
,
1719 sizeinfo
->array
.raid_disks
);
1720 new_size
/= data_disks
;
1721 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size
, new_size
,
1722 new_size
* data_disks
);
1724 sysfs_free(sizeinfo
);
1728 /* Update the size of each member device in case
1729 * they have been resized. This will never reduce
1730 * below the current used-size. The "size" attribute
1731 * understands '0' to mean 'max'.
1735 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
1736 if (sysfs_set_num(sra
, mdi
, "size",
1737 s
->size
== MAX_SIZE
? 0 : s
->size
) < 0) {
1738 /* Probably kernel refusing to let us
1739 * reduce the size - not an error.
1743 if (array
.not_persistent
== 0 &&
1744 array
.major_version
== 0 &&
1745 get_linux_version() < 3001000) {
1746 /* Dangerous to allow size to exceed 2TB */
1747 unsigned long long csize
;
1748 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
1749 if (csize
>= 2ULL*1024*1024*1024)
1750 csize
= 2ULL*1024*1024*1024;
1751 if ((min_csize
== 0 || (min_csize
1758 pr_err("Cannot set size on array members.\n");
1759 goto size_change_error
;
1761 if (min_csize
&& s
->size
> min_csize
) {
1762 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1764 goto size_change_error
;
1766 if (min_csize
&& s
->size
== MAX_SIZE
) {
1767 /* Don't let the kernel choose a size - it will get
1770 pr_err("Limited v0.90 array to 2TB per device\n");
1771 s
->size
= min_csize
;
1773 if (st
->ss
->external
) {
1774 if (sra
->array
.level
== 0) {
1775 rv
= sysfs_set_str(sra
, NULL
, "level",
1779 /* get array parametes after takeover
1780 * to chane one parameter at time only
1782 rv
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
1785 /* make sure mdmon is
1786 * aware of the new level */
1787 if (!mdmon_running(st
->container_devnm
))
1788 start_mdmon(st
->container_devnm
);
1789 ping_monitor(container
);
1790 if (mdmon_running(st
->container_devnm
) &&
1791 st
->update_tail
== NULL
)
1792 st
->update_tail
= &st
->updates
;
1795 if (s
->size
== MAX_SIZE
)
1797 array
.size
= s
->size
;
1798 if (s
->size
& ~INT32_MAX
) {
1799 /* got truncated to 32bit, write to
1800 * component_size instead
1803 rv
= sysfs_set_num(sra
, NULL
,
1804 "component_size", s
->size
);
1808 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1810 /* manage array size when it is managed externally
1812 if ((rv
== 0) && st
->ss
->external
)
1813 rv
= set_array_size(st
, sra
, sra
->text_version
);
1816 if (raid0_takeover
) {
1817 /* do not recync non-existing parity,
1818 * we will drop it anyway
1820 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
1821 /* go back to raid0, drop parity disk
1823 sysfs_set_str(sra
, NULL
, "level", "raid0");
1824 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1831 /* restore metadata */
1832 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1833 UnSet
, NULL
, devname
,
1834 ROLLBACK_METADATA_CHANGES
,
1837 pr_err("Cannot set device size for %s: %s\n",
1838 devname
, strerror(err
));
1840 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1841 cont_err("Bitmap must be removed before size can be changed\n");
1845 if (s
->assume_clean
) {
1846 /* This will fail on kernels older than 3.0 unless
1847 * a backport has been arranged.
1850 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
1851 pr_err("--assume-clean not supported with --grow on this kernel\n");
1853 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1854 s
->size
= get_component_size(fd
)/2;
1856 s
->size
= array
.size
;
1857 if (c
->verbose
>= 0) {
1858 if (s
->size
== orig_size
)
1859 pr_err("component size of %s unchanged at %lluK\n",
1862 pr_err("component size of %s has been set to %lluK\n",
1866 } else if (array
.level
!= LEVEL_CONTAINER
) {
1867 s
->size
= get_component_size(fd
)/2;
1869 s
->size
= array
.size
;
1872 /* See if there is anything else to do */
1873 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
1874 (s
->layout_str
== NULL
) &&
1875 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
1876 data_offset
== INVALID_SECTORS
&&
1877 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
1878 /* Nothing more to do */
1879 if (!changed
&& c
->verbose
>= 0)
1880 pr_err("%s: no change requested\n",
1885 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1886 * current implementation assumes that following conditions must be met:
1889 * - near_copies == 2
1891 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
1892 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1893 (s
->level
== 0 && array
.level
== 1 && sra
)) {
1895 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1897 dprintf("Array cannot be reshaped\n");
1903 /* Make sure mdmon has seen the device removal
1904 * and updated metadata before we continue with
1908 ping_monitor(container
);
1911 memset(&info
, 0, sizeof(info
));
1913 sysfs_init(&info
, fd
, NULL
);
1914 strcpy(info
.text_version
, sra
->text_version
);
1915 info
.component_size
= s
->size
*2;
1916 info
.new_level
= s
->level
;
1917 info
.new_chunk
= s
->chunk
* 1024;
1918 if (info
.array
.level
== LEVEL_CONTAINER
) {
1919 info
.delta_disks
= UnSet
;
1920 info
.array
.raid_disks
= s
->raiddisks
;
1921 } else if (s
->raiddisks
)
1922 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
1924 info
.delta_disks
= UnSet
;
1925 if (s
->layout_str
== NULL
) {
1926 info
.new_layout
= UnSet
;
1927 if (info
.array
.level
== 6 &&
1928 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1929 info
.array
.layout
>= 16) {
1930 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
1931 cont_err("during the reshape, please specify --layout=preserve\n");
1932 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1936 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
1937 strcmp(s
->layout_str
, "normalize") == 0) {
1938 /* If we have a -6 RAID6 layout, remove the '-6'. */
1939 info
.new_layout
= UnSet
;
1940 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1942 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1943 h
= strrchr(l
, '-');
1944 if (h
&& strcmp(h
, "-6") == 0) {
1946 info
.new_layout
= map_name(r6layout
, l
);
1949 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
1953 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
1954 /* This means that a non-standard RAID6 layout
1957 * - When reshape a RAID6 (e.g. adding a device)
1958 * which is in a non-standard layout, it is OK
1959 * to preserve that layout.
1960 * - When converting a RAID5 to RAID6, leave it in
1961 * the XXX-6 layout, don't re-layout.
1963 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
1964 info
.new_layout
= info
.array
.layout
;
1965 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
1967 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
1969 info
.new_layout
= map_name(r6layout
, l
);
1971 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
1976 int l
= info
.new_level
;
1978 l
= info
.array
.level
;
1981 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
1984 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
1987 info
.new_layout
= parse_layout_10(s
->layout_str
);
1990 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
1993 pr_err("layout not meaningful with this level\n");
1997 if (info
.new_layout
== UnSet
) {
1998 pr_err("layout %s not understood for this level\n",
2005 if (array
.level
== LEVEL_FAULTY
) {
2006 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2007 pr_err("cannot change level of Faulty device\n");
2011 pr_err("cannot set chunksize of Faulty device\n");
2014 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2015 pr_err("cannot set raid_disks of Faulty device\n");
2018 if (s
->layout_str
) {
2019 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2020 dprintf("Cannot get array information.\n");
2023 array
.layout
= info
.new_layout
;
2024 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2025 pr_err("failed to set new layout\n");
2027 } else if (c
->verbose
>= 0)
2028 printf("layout for %s set to %d\n",
2029 devname
, array
.layout
);
2031 } else if (array
.level
== LEVEL_CONTAINER
) {
2032 /* This change is to be applied to every array in the
2033 * container. This is only needed when the metadata imposes
2034 * restraints of the various arrays in the container.
2035 * Currently we only know that IMSM requires all arrays
2036 * to have the same number of devices so changing the
2037 * number of devices (On-Line Capacity Expansion) must be
2038 * performed at the level of the container
2040 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2041 c
->force
, c
->backup_file
, c
->verbose
, 0, 0, 0);
2044 /* get spare devices from external metadata
2046 if (st
->ss
->external
) {
2047 struct mdinfo
*info2
;
2049 info2
= st
->ss
->container_content(st
, subarray
);
2051 info
.array
.spare_disks
=
2052 info2
->array
.spare_disks
;
2057 /* Impose these changes on a single array. First
2058 * check that the metadata is OK with the change. */
2060 if (reshape_super(st
, 0, info
.new_level
,
2061 info
.new_layout
, info
.new_chunk
,
2062 info
.array
.raid_disks
, info
.delta_disks
,
2063 c
->backup_file
, devname
, APPLY_METADATA_CHANGES
,
2069 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2070 devlist
, data_offset
, c
->backup_file
, c
->verbose
,
2081 /* verify_reshape_position()
2082 * Function checks if reshape position in metadata is not farther
2083 * than position in md.
2085 * 0 : not valid sysfs entry
2086 * it can be caused by not started reshape, it should be started
2087 * by reshape array or raid0 array is before takeover
2088 * -1 : error, reshape position is obviously wrong
2089 * 1 : success, reshape progress correct or updated
2091 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2097 /* read sync_max, failure can mean raid0 array */
2098 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2102 unsigned long long position
= strtoull(buf
, &ep
, 0);
2104 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2105 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2106 position
*= get_data_disks(level
,
2108 info
->array
.raid_disks
);
2109 if (info
->reshape_progress
< position
) {
2110 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2111 info
->reshape_progress
, position
);
2112 info
->reshape_progress
= position
;
2114 } else if (info
->reshape_progress
> position
) {
2115 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2116 position
, info
->reshape_progress
);
2119 dprintf("Reshape position in md and metadata are the same;");
2123 } else if (rv
== 0) {
2124 /* for valid sysfs entry, 0-length content
2125 * should be indicated as error
2133 static unsigned long long choose_offset(unsigned long long lo
,
2134 unsigned long long hi
,
2135 unsigned long long min
,
2136 unsigned long long max
)
2138 /* Choose a new offset between hi and lo.
2139 * It must be between min and max, but
2140 * we would prefer something near the middle of hi/lo, and also
2141 * prefer to be aligned to a big power of 2.
2143 * So we start with the middle, then for each bit,
2144 * starting at '1' and increasing, if it is set, we either
2145 * add it or subtract it if possible, preferring the option
2146 * which is furthest from the boundary.
2148 * We stop once we get a 1MB alignment. As units are in sectors,
2149 * 1MB = 2*1024 sectors.
2151 unsigned long long choice
= (lo
+ hi
) / 2;
2152 unsigned long long bit
= 1;
2154 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2155 unsigned long long bigger
, smaller
;
2156 if (! (bit
& choice
))
2158 bigger
= choice
+ bit
;
2159 smaller
= choice
- bit
;
2160 if (bigger
> max
&& smaller
< min
)
2164 else if (smaller
< min
)
2166 else if (hi
- bigger
> smaller
- lo
)
2174 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2175 char *devname
, int delta_disks
,
2176 unsigned long long data_offset
,
2177 unsigned long long min
,
2183 unsigned long long before
, after
;
2185 /* Need to find min space before and after so same is used
2188 before
= UINT64_MAX
;
2190 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2194 struct supertype
*st2
;
2195 struct mdinfo info2
;
2197 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2199 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2200 dfd
= dev_open(dn
, O_RDONLY
);
2202 pr_err("%s: cannot open component %s\n",
2203 devname
, dn
? dn
: "-unknown-");
2206 st2
= dup_super(st
);
2207 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2211 pr_err("%s: cannot get superblock from %s\n",
2215 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2216 st2
->ss
->free_super(st2
);
2218 if (info2
.space_before
== 0 &&
2219 info2
.space_after
== 0) {
2220 /* Metadata doesn't support data_offset changes */
2222 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2226 if (before
> info2
.space_before
)
2227 before
= info2
.space_before
;
2228 if (after
> info2
.space_after
)
2229 after
= info2
.space_after
;
2231 if (data_offset
!= INVALID_SECTORS
) {
2233 if (info2
.data_offset
== data_offset
) {
2234 pr_err("%s: already has that data_offset\n",
2238 if (data_offset
< info2
.data_offset
)
2242 } else if ((data_offset
<= info2
.data_offset
&& dir
== 1) ||
2243 (data_offset
>= info2
.data_offset
&& dir
== -1)) {
2244 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2250 if (before
== UINT64_MAX
)
2251 /* impossible really, there must be no devices */
2254 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2255 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2256 unsigned long long new_data_offset
;
2258 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2260 if (delta_disks
< 0) {
2261 /* Don't need any space as array is shrinking
2262 * just move data_offset up by min
2264 if (data_offset
== INVALID_SECTORS
)
2265 new_data_offset
= sd
->data_offset
+ min
;
2267 if (data_offset
< sd
->data_offset
+ min
) {
2268 pr_err("--data-offset too small for %s\n",
2272 new_data_offset
= data_offset
;
2274 } else if (delta_disks
> 0) {
2275 /* need space before */
2279 pr_err("Insufficient head-space for reshape on %s\n",
2283 if (data_offset
== INVALID_SECTORS
)
2284 new_data_offset
= sd
->data_offset
- min
;
2286 if (data_offset
> sd
->data_offset
- min
) {
2287 pr_err("--data-offset too large for %s\n",
2291 new_data_offset
= data_offset
;
2295 /* can move up or down. If 'data_offset'
2296 * was set we would have already decided,
2297 * so just choose direction with most space.
2304 sysfs_set_str(sra
, NULL
, "reshape_direction",
2305 dir
== 1 ? "backwards" : "forwards");
2307 /* Increase data offset */
2311 pr_err("Insufficient tail-space for reshape on %s\n",
2315 if (data_offset
!= INVALID_SECTORS
&&
2316 data_offset
< sd
->data_offset
+ min
) {
2317 pr_err("--data-offset too small on %s\n",
2321 if (data_offset
!= INVALID_SECTORS
)
2322 new_data_offset
= data_offset
;
2324 new_data_offset
= choose_offset(sd
->data_offset
,
2325 sd
->data_offset
+ after
,
2326 sd
->data_offset
+ min
,
2327 sd
->data_offset
+ after
);
2329 /* Decrease data offset */
2333 pr_err("insufficient head-room on %s\n",
2337 if (data_offset
!= INVALID_SECTORS
&&
2338 data_offset
< sd
->data_offset
- min
) {
2339 pr_err("--data-offset too small on %s\n",
2343 if (data_offset
!= INVALID_SECTORS
)
2344 new_data_offset
= data_offset
;
2346 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2348 sd
->data_offset
- before
,
2349 sd
->data_offset
- min
);
2352 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2353 if (err
< 0 && errno
== E2BIG
) {
2354 /* try again after increasing data size to max */
2355 err
= sysfs_set_num(sra
, sd
, "size", 0);
2356 if (err
< 0 && errno
== EINVAL
&&
2357 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2358 /* some kernels have a bug where you cannot
2359 * use '0' on spare devices. */
2360 sysfs_set_num(sra
, sd
, "size",
2361 (sra
->component_size
+ after
)/2);
2363 err
= sysfs_set_num(sra
, sd
, "new_offset",
2367 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2368 pr_err("data-offset is too big for %s\n",
2372 if (sd
== sra
->devs
&&
2373 (errno
== ENOENT
|| errno
== E2BIG
))
2374 /* Early kernel, no 'new_offset' file,
2375 * or kernel doesn't like us.
2376 * For RAID5/6 this is not fatal
2379 pr_err("Cannot set new_offset for %s\n",
2388 /* Just use a backup file */
2392 static int raid10_reshape(char *container
, int fd
, char *devname
,
2393 struct supertype
*st
, struct mdinfo
*info
,
2394 struct reshape
*reshape
,
2395 unsigned long long data_offset
,
2396 int force
, int verbose
)
2398 /* Changing raid_disks, layout, chunksize or possibly
2399 * just data_offset for a RAID10.
2400 * We must always change data_offset. We change by at least
2401 * ->min_offset_change which is the largest of the old and new
2403 * If raid_disks is increasing, then data_offset must decrease
2404 * by at least this copy size.
2405 * If raid_disks is unchanged, data_offset must increase or
2406 * decrease by at least min_offset_change but preferably by much more.
2407 * We choose half of the available space.
2408 * If raid_disks is decreasing, data_offset must increase by
2409 * at least min_offset_change. To allow of this, component_size
2410 * must be decreased by the same amount.
2412 * So we calculate the required minimum and direction, possibly
2413 * reduce the component_size, then iterate through the devices
2414 * and set the new_data_offset.
2415 * If that all works, we set chunk_size, layout, raid_disks, and start
2419 unsigned long long min
;
2422 sra
= sysfs_read(fd
, NULL
,
2423 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2426 pr_err("%s: Cannot get array details from sysfs\n",
2430 min
= reshape
->min_offset_change
;
2432 if (info
->delta_disks
)
2433 sysfs_set_str(sra
, NULL
, "reshape_direction",
2434 info
->delta_disks
< 0 ? "backwards" : "forwards");
2435 if (info
->delta_disks
< 0 &&
2436 info
->space_after
< min
) {
2437 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2438 (sra
->component_size
-
2441 pr_err("cannot reduce component size\n");
2445 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
, data_offset
,
2448 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2449 cont_err("supported on this kernel\n");
2455 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2457 if (!err
&& sysfs_set_num(sra
, NULL
, "layout", reshape
->after
.layout
) < 0)
2459 if (!err
&& sysfs_set_num(sra
, NULL
, "raid_disks",
2460 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2462 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2465 pr_err("Cannot set array shape for %s\n",
2468 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2469 cont_err(" Bitmap must be removed before shape can be changed\n");
2479 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2481 struct mdinfo
*sra
, *sd
;
2482 /* Initialisation to silence compiler warning */
2483 unsigned long long min_space_before
= 0, min_space_after
= 0;
2486 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2489 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2492 struct supertype
*st2
;
2493 struct mdinfo info2
;
2495 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2497 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2498 dfd
= dev_open(dn
, O_RDONLY
);
2501 st2
= dup_super(st
);
2502 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2508 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2509 st2
->ss
->free_super(st2
);
2512 min_space_before
> info2
.space_before
)
2513 min_space_before
= info2
.space_before
;
2515 min_space_after
> info2
.space_after
)
2516 min_space_after
= info2
.space_after
;
2519 if (sd
== NULL
&& !first
) {
2520 info
->space_after
= min_space_after
;
2521 info
->space_before
= min_space_before
;
2526 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2527 struct mdinfo
*info
,
2528 int disks
, unsigned long long blocks
)
2530 /* Check that the internal stripe cache is
2531 * large enough, or it won't work.
2532 * It must hold at least 4 stripes of the larger
2535 unsigned long cache
;
2536 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2537 cache
*= 4; /* 4 stripes minimum */
2538 cache
/= 512; /* convert to sectors */
2539 /* make sure there is room for 'blocks' with a bit to spare */
2540 if (cache
< 16 + blocks
/ disks
)
2541 cache
= 16 + blocks
/ disks
;
2542 cache
/= (4096/512); /* Covert from sectors to pages */
2544 if (sra
->cache_size
< cache
)
2545 subarray_set_num(container
, sra
, "stripe_cache_size",
2549 static int impose_reshape(struct mdinfo
*sra
,
2550 struct mdinfo
*info
,
2551 struct supertype
*st
,
2554 char *devname
, char *container
,
2555 struct reshape
*reshape
)
2557 struct mdu_array_info_s array
;
2559 sra
->new_chunk
= info
->new_chunk
;
2562 /* for external metadata checkpoint saved by mdmon can be lost
2563 * or missed /due to e.g. crash/. Check if md is not during
2564 * restart farther than metadata points to.
2565 * If so, this means metadata information is obsolete.
2567 if (st
->ss
->external
)
2568 verify_reshape_position(info
, reshape
->level
);
2569 sra
->reshape_progress
= info
->reshape_progress
;
2571 sra
->reshape_progress
= 0;
2572 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2573 /* start from the end of the new array */
2574 sra
->reshape_progress
= (sra
->component_size
2575 * reshape
->after
.data_disks
);
2578 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2579 if (info
->array
.chunk_size
== info
->new_chunk
&&
2580 reshape
->before
.layout
== reshape
->after
.layout
&&
2581 st
->ss
->external
== 0) {
2582 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2583 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2585 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2588 pr_err("Cannot set device shape for %s: %s\n",
2589 devname
, strerror(errno
));
2592 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2593 cont_err("Bitmap must be removed before shape can be changed\n");
2597 } else if (!restart
) {
2598 /* set them all just in case some old 'new_*' value
2599 * persists from some earlier problem.
2602 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2604 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2605 reshape
->after
.layout
) < 0)
2607 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2608 reshape
->after
.data_disks
+
2609 reshape
->parity
) < 0)
2612 pr_err("Cannot set device shape for %s\n",
2616 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2617 cont_err("Bitmap must be removed before shape can be changed\n");
2626 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2629 struct mdu_array_info_s array
;
2631 sysfs_init(&info
, fd
, NULL
);
2633 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2635 (array
.level
>= 4 && array
.level
<= 6)) {
2636 /* To convert to RAID0 we need to fail and
2637 * remove any non-data devices. */
2640 int data_disks
= array
.raid_disks
- 1;
2641 if (array
.level
== 6)
2643 if (array
.level
== 5 &&
2644 array
.layout
!= ALGORITHM_PARITY_N
)
2646 if (array
.level
== 6 &&
2647 array
.layout
!= ALGORITHM_PARITY_N_6
)
2649 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2650 /* First remove any spares so no recovery starts */
2651 for (d
= 0, found
= 0;
2652 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2654 mdu_disk_info_t disk
;
2656 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2658 if (disk
.major
== 0 && disk
.minor
== 0)
2661 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2662 && disk
.raid_disk
< data_disks
)
2665 ioctl(fd
, HOT_REMOVE_DISK
,
2666 makedev(disk
.major
, disk
.minor
));
2668 /* Now fail anything left */
2669 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2670 for (d
= 0, found
= 0;
2671 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2674 mdu_disk_info_t disk
;
2676 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2678 if (disk
.major
== 0 && disk
.minor
== 0)
2681 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2682 && disk
.raid_disk
< data_disks
)
2685 ioctl(fd
, SET_DISK_FAULTY
,
2686 makedev(disk
.major
, disk
.minor
));
2688 while (ioctl(fd
, HOT_REMOVE_DISK
,
2689 makedev(disk
.major
, disk
.minor
)) < 0
2696 c
= map_num(pers
, level
);
2698 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2701 pr_err("%s: could not set level to %s\n",
2704 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2705 cont_err("Bitmap must be removed before level can be changed\n");
2709 pr_err("level of %s changed to %s\n",
2716 static void catch_term(int sig
)
2721 static int continue_via_systemd(char *devnm
)
2723 int skipped
, i
, pid
, status
;
2725 /* In a systemd/udev world, it is best to get systemd to
2726 * run "mdadm --grow --continue" rather than running in the
2731 /* FIXME yuk. CLOSE_EXEC?? */
2733 for (i
= 3; skipped
< 20; i
++)
2739 /* Don't want to see error messages from
2740 * systemctl. If the service doesn't exist,
2741 * we fork ourselves.
2744 open("/dev/null", O_WRONLY
);
2745 snprintf(pathbuf
, sizeof(pathbuf
), "mdadm-grow-continue@%s.service",
2747 status
= execl("/usr/bin/systemctl", "systemctl",
2750 status
= execl("/bin/systemctl", "systemctl", "start",
2753 case -1: /* Just do it ourselves. */
2755 default: /* parent - good */
2756 pid
= wait(&status
);
2757 if (pid
>= 0 && status
== 0)
2763 static int reshape_array(char *container
, int fd
, char *devname
,
2764 struct supertype
*st
, struct mdinfo
*info
,
2765 int force
, struct mddev_dev
*devlist
,
2766 unsigned long long data_offset
,
2767 char *backup_file
, int verbose
, int forked
,
2768 int restart
, int freeze_reshape
)
2770 struct reshape reshape
;
2773 int orig_level
= UnSet
;
2777 struct mdu_array_info_s array
;
2780 struct mddev_dev
*dv
;
2784 unsigned long long *offsets
= NULL
;
2788 unsigned long blocks
;
2789 unsigned long long array_size
;
2791 struct mdinfo
*sra
= NULL
;
2794 /* when reshaping a RAID0, the component_size might be zero.
2795 * So try to fix that up.
2797 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2798 dprintf("Cannot get array information.\n");
2801 if (array
.level
== 0 && info
->component_size
== 0) {
2802 get_dev_size(fd
, NULL
, &array_size
);
2803 info
->component_size
= array_size
/ array
.raid_disks
;
2806 if (array
.level
== 10)
2807 /* Need space_after info */
2808 get_space_after(fd
, st
, info
);
2810 if (info
->reshape_active
) {
2811 int new_level
= info
->new_level
;
2812 info
->new_level
= UnSet
;
2813 if (info
->delta_disks
> 0)
2814 info
->array
.raid_disks
-= info
->delta_disks
;
2815 msg
= analyse_change(devname
, info
, &reshape
);
2816 info
->new_level
= new_level
;
2817 if (info
->delta_disks
> 0)
2818 info
->array
.raid_disks
+= info
->delta_disks
;
2820 /* Make sure the array isn't read-only */
2821 ioctl(fd
, RESTART_ARRAY_RW
, 0);
2823 msg
= analyse_change(devname
, info
, &reshape
);
2825 /* if msg == "", error has already been printed */
2827 pr_err("%s\n", msg
);
2831 (reshape
.level
!= info
->array
.level
||
2832 reshape
.before
.layout
!= info
->array
.layout
||
2833 reshape
.before
.data_disks
+ reshape
.parity
2834 != info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
2835 pr_err("reshape info is not in native format - cannot continue.\n");
2839 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
2840 !((sysfs_get_str(info
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0) &&
2841 (strncmp(buf
, "reshape", 7) == 0))) {
2842 /* When reshape is restarted from '0', very begin of array
2843 * it is possible that for external metadata reshape and array
2844 * configuration doesn't happen.
2845 * Check if md has the same opinion, and reshape is restarted
2846 * from 0. If so, this is regular reshape start after reshape
2847 * switch in metadata to next array only.
2849 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
2850 (info
->reshape_progress
== 0))
2854 /* reshape already started. just skip to monitoring the reshape */
2855 if (reshape
.backup_blocks
== 0)
2857 if (restart
& RESHAPE_NO_BACKUP
)
2860 /* Need 'sra' down at 'started:' */
2861 sra
= sysfs_read(fd
, NULL
,
2862 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
2865 pr_err("%s: Cannot get array details from sysfs\n",
2871 backup_file
= locate_backup(sra
->sys_name
);
2875 /* The container is frozen but the array may not be.
2876 * So freeze the array so spares don't get put to the wrong use
2877 * FIXME there should probably be a cleaner separation between
2878 * freeze_array and freeze_container.
2880 sysfs_freeze_array(info
);
2881 /* Check we have enough spares to not be degraded */
2883 for (dv
= devlist
; dv
; dv
=dv
->next
)
2885 spares_needed
= max(reshape
.before
.data_disks
,
2886 reshape
.after
.data_disks
)
2887 + reshape
.parity
- array
.raid_disks
;
2890 info
->new_level
> 1 && info
->array
.level
> 1 &&
2891 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
2892 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2893 " Use --force to over-ride this check.\n",
2895 spares_needed
== 1 ? "" : "s",
2896 info
->array
.spare_disks
+ added_disks
);
2899 /* Check we have enough spares to not fail */
2900 spares_needed
= max(reshape
.before
.data_disks
,
2901 reshape
.after
.data_disks
)
2903 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
2904 spares_needed
> info
->array
.spare_disks
+added_disks
) {
2905 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2907 spares_needed
== 1 ? "" : "s",
2908 info
->array
.spare_disks
+ added_disks
);
2912 if (reshape
.level
!= array
.level
) {
2913 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
2916 info
->new_layout
= UnSet
; /* after level change,
2917 * layout is meaningless */
2918 orig_level
= array
.level
;
2919 sysfs_freeze_array(info
);
2921 if (reshape
.level
> 0 && st
->ss
->external
) {
2922 /* make sure mdmon is aware of the new level */
2923 if (mdmon_running(container
))
2924 flush_mdmon(container
);
2926 if (!mdmon_running(container
))
2927 start_mdmon(container
);
2928 ping_monitor(container
);
2929 if (mdmon_running(container
) &&
2930 st
->update_tail
== NULL
)
2931 st
->update_tail
= &st
->updates
;
2934 /* ->reshape_super might have chosen some spares from the
2935 * container that it wants to be part of the new array.
2936 * We can collect them with ->container_content and give
2937 * them to the kernel.
2939 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
2940 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2941 struct mdinfo
*info2
=
2942 st
->ss
->container_content(st
, subarray
);
2946 sysfs_init(info2
, fd
, st
->devnm
);
2947 /* When increasing number of devices, we need to set
2948 * new raid_disks before adding these, or they might
2951 if (reshape
.backup_blocks
&&
2952 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
2953 subarray_set_num(container
, info2
, "raid_disks",
2954 reshape
.after
.data_disks
+
2956 for (d
= info2
->devs
; d
; d
= d
->next
) {
2957 if (d
->disk
.state
== 0 &&
2958 d
->disk
.raid_disk
>= 0) {
2959 /* This is a spare that wants to
2960 * be part of the array.
2962 add_disk(fd
, st
, info2
, d
);
2968 /* We might have been given some devices to add to the
2969 * array. Now that the array has been changed to the right
2970 * level and frozen, we can safely add them.
2973 Manage_subdevs(devname
, fd
, devlist
, verbose
,
2976 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
2977 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
2978 if (reshape
.backup_blocks
== 0) {
2979 /* No restriping needed, but we might need to impose
2980 * some more changes: layout, raid_disks, chunk_size
2982 /* read current array info */
2983 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2984 dprintf("Cannot get array information.\n");
2987 /* compare current array info with new values and if
2988 * it is different update them to new */
2989 if (info
->new_layout
!= UnSet
&&
2990 info
->new_layout
!= array
.layout
) {
2991 array
.layout
= info
->new_layout
;
2992 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2993 pr_err("failed to set new layout\n");
2995 } else if (verbose
>= 0)
2996 printf("layout for %s set to %d\n",
2997 devname
, array
.layout
);
2999 if (info
->delta_disks
!= UnSet
&&
3000 info
->delta_disks
!= 0 &&
3001 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
3002 array
.raid_disks
+= info
->delta_disks
;
3003 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
3004 pr_err("failed to set raid disks\n");
3006 } else if (verbose
>= 0) {
3007 printf("raid_disks for %s set to %d\n",
3008 devname
, array
.raid_disks
);
3011 if (info
->new_chunk
!= 0 &&
3012 info
->new_chunk
!= array
.chunk_size
) {
3013 if (sysfs_set_num(info
, NULL
,
3014 "chunk_size", info
->new_chunk
) != 0) {
3015 pr_err("failed to set chunk size\n");
3017 } else if (verbose
>= 0)
3018 printf("chunk size for %s set to %d\n",
3019 devname
, array
.chunk_size
);
3026 * There are three possibilities.
3027 * 1/ The array will shrink.
3028 * We need to ensure the reshape will pause before reaching
3029 * the 'critical section'. We also need to fork and wait for
3030 * that to happen. When it does we
3031 * suspend/backup/complete/unfreeze
3033 * 2/ The array will not change size.
3034 * This requires that we keep a backup of a sliding window
3035 * so that we can restore data after a crash. So we need
3036 * to fork and monitor progress.
3037 * In future we will allow the data_offset to change, so
3038 * a sliding backup becomes unnecessary.
3040 * 3/ The array will grow. This is relatively easy.
3041 * However the kernel's restripe routines will cheerfully
3042 * overwrite some early data before it is safe. So we
3043 * need to make a backup of the early parts of the array
3044 * and be ready to restore it if rebuild aborts very early.
3045 * For externally managed metadata, we still need a forked
3046 * child to monitor the reshape and suspend IO over the region
3047 * that is being reshaped.
3049 * We backup data by writing it to one spare, or to a
3050 * file which was given on command line.
3052 * In each case, we first make sure that storage is available
3053 * for the required backup.
3055 * - request the shape change.
3056 * - fork to handle backup etc.
3058 /* Check that we can hold all the data */
3059 get_dev_size(fd
, NULL
, &array_size
);
3060 if (reshape
.new_size
< (array_size
/512)) {
3061 pr_err("this change will reduce the size of the array.\n"
3062 " use --grow --array-size first to truncate array.\n"
3063 " e.g. mdadm --grow %s --array-size %llu\n",
3064 devname
, reshape
.new_size
/2);
3068 if (array
.level
== 10) {
3069 /* Reshaping RAID10 does not require any data backup by
3070 * user-space. Instead it requires that the data_offset
3071 * is changed to avoid the need for backup.
3072 * So this is handled very separately
3075 /* Nothing to do. */
3077 return raid10_reshape(container
, fd
, devname
, st
, info
,
3078 &reshape
, data_offset
,
3081 sra
= sysfs_read(fd
, NULL
,
3082 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3085 pr_err("%s: Cannot get array details from sysfs\n",
3091 switch(set_new_data_offset(sra
, st
, devname
,
3092 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3094 reshape
.min_offset_change
, 1)) {
3098 /* Updated data_offset, so it's easy now */
3099 update_cache_size(container
, sra
, info
,
3100 min(reshape
.before
.data_disks
,
3101 reshape
.after
.data_disks
),
3102 reshape
.backup_blocks
);
3104 /* Right, everything seems fine. Let's kick things off.
3108 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3109 devname
, container
, &reshape
) < 0)
3111 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3113 if (errno
!= EINVAL
) {
3114 pr_err("Failed to initiate reshape!\n");
3117 /* revert data_offset and try the old way */
3118 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3119 sysfs_set_num(sra
, sd
, "new_offset",
3121 sysfs_set_str(sra
, NULL
, "reshape_direction",
3126 if (info
->new_level
== reshape
.level
)
3128 /* need to adjust level when reshape completes */
3130 case -1: /* ignore error, but don't wait */
3132 default: /* parent */
3140 fd
= open_dev(sra
->sys_name
);
3142 impose_level(fd
, info
->new_level
, devname
, verbose
);
3144 case 1: /* Couldn't set data_offset, try the old way */
3145 if (data_offset
!= INVALID_SECTORS
) {
3146 pr_err("Cannot update data_offset on this array\n");
3153 /* Decide how many blocks (sectors) for a reshape
3154 * unit. The number we have so far is just a minimum
3156 blocks
= reshape
.backup_blocks
;
3157 if (reshape
.before
.data_disks
==
3158 reshape
.after
.data_disks
) {
3159 /* Make 'blocks' bigger for better throughput, but
3160 * not so big that we reject it below.
3161 * Try for 16 megabytes
3163 while (blocks
* 32 < sra
->component_size
&&
3167 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3169 if (blocks
>= sra
->component_size
/2) {
3170 pr_err("%s: Something wrong - reshape aborted\n",
3175 /* Now we need to open all these devices so we can read/write.
3177 nrdisks
= max(reshape
.before
.data_disks
,
3178 reshape
.after
.data_disks
) + reshape
.parity
3179 + sra
->array
.spare_disks
;
3180 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3181 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3183 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3184 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
3185 nrdisks
, blocks
, backup_file
,
3190 if ((st
->ss
->manage_reshape
== NULL
) ||
3191 (st
->ss
->recover_backup
== NULL
)) {
3192 if (backup_file
== NULL
) {
3193 if (reshape
.after
.data_disks
<=
3194 reshape
.before
.data_disks
) {
3195 pr_err("%s: Cannot grow - need backup-file\n",
3197 pr_err(" Please provide one with \"--backup=...\"\n");
3199 } else if (sra
->array
.spare_disks
== 0) {
3200 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3204 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3206 fdlist
+d
, offsets
+d
,
3215 update_cache_size(container
, sra
, info
,
3216 min(reshape
.before
.data_disks
, reshape
.after
.data_disks
),
3219 /* Right, everything seems fine. Let's kick things off.
3220 * If only changing raid_disks, use ioctl, else use
3225 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3226 devname
, container
, &reshape
) < 0)
3229 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3230 reshape
.after
.data_disks
);
3232 pr_err("Cannot %s reshape for %s\n",
3233 restart
? "continue" : "start",
3238 sysfs_set_str(sra
, NULL
, "array_state", "active");
3239 if (freeze_reshape
) {
3243 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3244 sra
->reshape_progress
);
3248 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3249 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3256 /* Now we just need to kick off the reshape and watch, while
3257 * handling backups of the data...
3258 * This is all done by a forked background process.
3260 switch(forked
? 0 : fork()) {
3262 pr_err("Cannot run child to monitor reshape: %s\n",
3276 /* If another array on the same devices is busy, the
3277 * reshape will wait for them. This would mean that
3278 * the first section that we suspend will stay suspended
3279 * for a long time. So check on that possibility
3280 * by looking for "DELAYED" in /proc/mdstat, and if found,
3284 struct mdstat_ent
*mds
, *m
;
3286 mds
= mdstat_read(1, 0);
3287 for (m
= mds
; m
; m
= m
->next
)
3288 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3290 m
->percent
== RESYNC_DELAYED
)
3293 /* Haven't started the reshape thread
3300 if (delayed
== 1 && get_linux_version() < 3007000) {
3301 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3302 " You might experience problems until other reshapes complete.\n");
3306 mdstat_wait(30 - (delayed
-1) * 25);
3310 if (check_env("MDADM_GROW_VERIFY"))
3311 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3314 mlockall(MCL_FUTURE
);
3316 signal(SIGTERM
, catch_term
);
3318 if (st
->ss
->external
) {
3319 /* metadata handler takes it from here */
3320 done
= st
->ss
->manage_reshape(
3321 fd
, sra
, &reshape
, st
, blocks
,
3323 d
- odisks
, fdlist
+odisks
,
3326 done
= child_monitor(
3327 fd
, sra
, &reshape
, st
, blocks
,
3329 d
- odisks
, fdlist
+odisks
,
3335 if (backup_file
&& done
) {
3337 bul
= make_backup(sra
->sys_name
);
3340 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3348 unlink(backup_file
);
3355 if (!st
->ss
->external
&&
3356 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
3357 && info
->custom_array_size
) &&
3358 info
->new_level
== reshape
.level
&&
3360 /* no need to wait for the reshape to finish as
3361 * there is nothing more to do.
3368 if (st
->ss
->external
) {
3369 /* Re-load the metadata as much could have changed */
3370 int cfd
= open_dev(st
->container_devnm
);
3372 flush_mdmon(container
);
3373 st
->ss
->free_super(st
);
3374 st
->ss
->load_container(st
, cfd
, container
);
3379 /* set new array size if required customer_array_size is used
3382 if (reshape
.before
.data_disks
!=
3383 reshape
.after
.data_disks
&&
3384 info
->custom_array_size
)
3385 set_array_size(st
, info
, info
->text_version
);
3387 if (info
->new_level
!= reshape
.level
) {
3389 fd
= open(devname
, O_RDONLY
);
3390 impose_level(fd
, info
->new_level
, devname
, verbose
);
3392 if (info
->new_level
== 0)
3393 st
->update_tail
= NULL
;
3405 if (orig_level
!= UnSet
&& sra
) {
3406 c
= map_num(pers
, orig_level
);
3407 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3408 pr_err("aborting level change\n");
3416 /* mdfd handle is passed to be closed in child process (after fork).
3418 int reshape_container(char *container
, char *devname
,
3420 struct supertype
*st
,
3421 struct mdinfo
*info
,
3423 char *backup_file
, int verbose
,
3424 int forked
, int restart
, int freeze_reshape
)
3426 struct mdinfo
*cc
= NULL
;
3428 char last_devnm
[32] = "";
3430 /* component_size is not meaningful for a container,
3431 * so pass '0' meaning 'no change'
3434 reshape_super(st
, 0, info
->new_level
,
3435 info
->new_layout
, info
->new_chunk
,
3436 info
->array
.raid_disks
, info
->delta_disks
,
3437 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3445 /* ping monitor to be sure that update is on disk
3447 ping_monitor(container
);
3449 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3450 if (continue_via_systemd(container
))
3453 switch (forked
? 0 : fork()) {
3454 case -1: /* error */
3455 perror("Cannot fork to complete reshape\n");
3458 default: /* parent */
3459 if (!freeze_reshape
)
3460 printf("%s: multi-array reshape continues in background\n", Name
);
3467 /* close unused handle in child process
3473 /* For each member array with reshape_active,
3474 * we need to perform the reshape.
3475 * We pick the first array that needs reshaping and
3476 * reshape it. reshape_array() will re-read the metadata
3477 * so the next time through a different array should be
3478 * ready for reshape.
3479 * It is possible that the 'different' array will not
3480 * be assembled yet. In that case we simple exit.
3481 * When it is assembled, the mdadm which assembles it
3482 * will take over the reshape.
3484 struct mdinfo
*content
;
3486 struct mdstat_ent
*mdstat
;
3492 cc
= st
->ss
->container_content(st
, NULL
);
3494 for (content
= cc
; content
; content
= content
->next
) {
3496 if (!content
->reshape_active
)
3499 subarray
= strchr(content
->text_version
+1, '/')+1;
3500 mdstat
= mdstat_by_subdev(subarray
, container
);
3503 if (mdstat
->active
== 0) {
3504 pr_err("Skipping inactive array %s.\n",
3506 free_mdstat(mdstat
);
3515 devid
= devnm2devid(mdstat
->devnm
);
3516 adev
= map_dev(major(devid
), minor(devid
), 0);
3518 adev
= content
->text_version
;
3520 fd
= open_dev(mdstat
->devnm
);
3522 pr_err("Device %s cannot be opened for reshape.\n", adev
);
3526 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3527 /* Do not allow for multiple reshape_array() calls for
3529 * It can happen when reshape_array() returns without
3530 * error, when reshape is not finished (wrong reshape
3531 * starting/continuation conditions). Mdmon doesn't
3532 * switch to next array in container and reentry
3533 * conditions for the same array occur.
3534 * This is possibly interim until the behaviour of
3535 * reshape_array is resolved().
3537 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3541 strcpy(last_devnm
, mdstat
->devnm
);
3543 sysfs_init(content
, fd
, mdstat
->devnm
);
3545 if (mdmon_running(container
))
3546 flush_mdmon(container
);
3548 rv
= reshape_array(container
, fd
, adev
, st
,
3549 content
, force
, NULL
, INVALID_SECTORS
,
3550 backup_file
, verbose
, 1, restart
,
3554 if (freeze_reshape
) {
3563 if (mdmon_running(container
))
3564 flush_mdmon(container
);
3573 * We run a child process in the background which performs the following
3575 * - wait for resync to reach a certain point
3576 * - suspend io to the following section
3577 * - backup that section
3578 * - allow resync to proceed further
3580 * - discard the backup.
3582 * When are combined in slightly different ways in the three cases.
3584 * - suspend/backup/allow/wait/resume/discard
3586 * - allow/wait/suspend/backup/allow/wait/resume/discard
3588 * - wait/resume/discard/suspend/backup/allow
3590 * suspend/backup/allow always come together
3591 * wait/resume/discard do too.
3592 * For the same-size case we have two backups to improve flow.
3596 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3597 unsigned long long backup_point
,
3598 unsigned long long wait_point
,
3599 unsigned long long *suspend_point
,
3600 unsigned long long *reshape_completed
, int *frozen
)
3602 /* This function is called repeatedly by the reshape manager.
3603 * It determines how much progress can safely be made and allows
3605 * - 'info' identifies the array and particularly records in
3606 * ->reshape_progress the metadata's knowledge of progress
3607 * This is a sector offset from the start of the array
3608 * of the next array block to be relocated. This number
3609 * may increase from 0 or decrease from array_size, depending
3610 * on the type of reshape that is happening.
3611 * Note that in contrast, 'sync_completed' is a block count of the
3612 * reshape so far. It gives the distance between the start point
3613 * (head or tail of device) and the next place that data will be
3614 * written. It always increases.
3615 * - 'reshape' is the structure created by analyse_change
3616 * - 'backup_point' shows how much the metadata manager has backed-up
3617 * data. For reshapes with increasing progress, it is the next address
3618 * to be backed up, previous addresses have been backed-up. For
3619 * decreasing progress, it is the earliest address that has been
3620 * backed up - later address are also backed up.
3621 * So addresses between reshape_progress and backup_point are
3622 * backed up providing those are in the 'correct' order.
3623 * - 'wait_point' is an array address. When reshape_completed
3624 * passes this point, progress_reshape should return. It might
3625 * return earlier if it determines that ->reshape_progress needs
3626 * to be updated or further backup is needed.
3627 * - suspend_point is maintained by progress_reshape and the caller
3628 * should not touch it except to initialise to zero.
3629 * It is an array address and it only increases in 2.6.37 and earlier.
3630 * This makes it difficult to handle reducing reshapes with
3631 * external metadata.
3632 * However: it is similar to backup_point in that it records the
3633 * other end of a suspended region from reshape_progress.
3634 * it is moved to extend the region that is safe to backup and/or
3636 * - reshape_completed is read from sysfs and returned. The caller
3637 * should copy this into ->reshape_progress when it has reason to
3638 * believe that the metadata knows this, and any backup outside this
3642 * 1 if more data from backup_point - but only as far as suspend_point,
3643 * should be backed up
3644 * 0 if things are progressing smoothly
3645 * -1 if the reshape is finished because it is all done,
3646 * -2 if the reshape is finished due to an error.
3649 int advancing
= (reshape
->after
.data_disks
3650 >= reshape
->before
.data_disks
);
3651 unsigned long long need_backup
; /* All data between start of array and
3652 * here will at some point need to
3655 unsigned long long read_offset
, write_offset
;
3656 unsigned long long write_range
;
3657 unsigned long long max_progress
, target
, completed
;
3658 unsigned long long array_size
= (info
->component_size
3659 * reshape
->before
.data_disks
);
3663 /* First, we unsuspend any region that is now known to be safe.
3664 * If suspend_point is on the 'wrong' side of reshape_progress, then
3665 * we don't have or need suspension at the moment. This is true for
3666 * native metadata when we don't need to back-up.
3669 if (info
->reshape_progress
<= *suspend_point
)
3670 sysfs_set_num(info
, NULL
, "suspend_lo",
3671 info
->reshape_progress
);
3673 /* Note: this won't work in 2.6.37 and before.
3674 * Something somewhere should make sure we don't need it!
3676 if (info
->reshape_progress
>= *suspend_point
)
3677 sysfs_set_num(info
, NULL
, "suspend_hi",
3678 info
->reshape_progress
);
3681 /* Now work out how far it is safe to progress.
3682 * If the read_offset for ->reshape_progress is less than
3683 * 'blocks' beyond the write_offset, we can only progress as far
3685 * Otherwise we can progress until the write_offset for the new location
3686 * reaches (within 'blocks' of) the read_offset at the current location.
3687 * However that region must be suspended unless we are using native
3689 * If we need to suspend more, we limit it to 128M per device, which is
3690 * rather arbitrary and should be some time-based calculation.
3692 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3693 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3694 write_range
= info
->new_chunk
/512;
3695 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3696 need_backup
= array_size
;
3698 need_backup
= reshape
->backup_blocks
;
3700 if (read_offset
< write_offset
+ write_range
)
3701 max_progress
= backup_point
;
3705 reshape
->after
.data_disks
;
3707 if (read_offset
> write_offset
- write_range
)
3708 /* Can only progress as far as has been backed up,
3709 * which must be suspended */
3710 max_progress
= backup_point
;
3711 else if (info
->reshape_progress
<= need_backup
)
3712 max_progress
= backup_point
;
3714 if (info
->array
.major_version
>= 0)
3715 /* Can progress until backup is needed */
3716 max_progress
= need_backup
;
3718 /* Can progress until metadata update is required */
3721 reshape
->after
.data_disks
;
3722 /* but data must be suspended */
3723 if (max_progress
< *suspend_point
)
3724 max_progress
= *suspend_point
;
3729 /* We know it is safe to progress to 'max_progress' providing
3730 * it is suspended or we are using native metadata.
3731 * Consider extending suspend_point 128M per device if it
3732 * is less than 64M per device beyond reshape_progress.
3733 * But always do a multiple of 'blocks'
3734 * FIXME this is too big - it takes to long to complete
3737 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3738 reshape
->after
.data_disks
);
3739 target
/= reshape
->backup_blocks
;
3742 target
*= reshape
->backup_blocks
;
3744 /* For externally managed metadata we always need to suspend IO to
3745 * the area being reshaped so we regularly push suspend_point forward.
3746 * For native metadata we only need the suspend if we are going to do
3750 if ((need_backup
> info
->reshape_progress
3751 || info
->array
.major_version
< 0) &&
3752 *suspend_point
< info
->reshape_progress
+ target
) {
3753 if (need_backup
< *suspend_point
+ 2 * target
)
3754 *suspend_point
= need_backup
;
3755 else if (*suspend_point
+ 2 * target
< array_size
)
3756 *suspend_point
+= 2 * target
;
3758 *suspend_point
= array_size
;
3759 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
3760 if (max_progress
> *suspend_point
)
3761 max_progress
= *suspend_point
;
3764 if (info
->array
.major_version
>= 0) {
3765 /* Only need to suspend when about to backup */
3766 if (info
->reshape_progress
< need_backup
* 2 &&
3767 *suspend_point
> 0) {
3769 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
3770 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
3773 /* Need to suspend continually */
3774 if (info
->reshape_progress
< *suspend_point
)
3775 *suspend_point
= info
->reshape_progress
;
3776 if (*suspend_point
+ target
< info
->reshape_progress
)
3777 /* No need to move suspend region yet */;
3779 if (*suspend_point
>= 2 * target
)
3780 *suspend_point
-= 2 * target
;
3783 sysfs_set_num(info
, NULL
, "suspend_lo",
3786 if (max_progress
< *suspend_point
)
3787 max_progress
= *suspend_point
;
3791 /* now set sync_max to allow that progress. sync_max, like
3792 * sync_completed is a count of sectors written per device, so
3793 * we find the difference between max_progress and the start point,
3794 * and divide that by after.data_disks to get a sync_max
3796 * At the same time we convert wait_point to a similar number
3797 * for comparing against sync_completed.
3799 /* scale down max_progress to per_disk */
3800 max_progress
/= reshape
->after
.data_disks
;
3801 /* Round to chunk size as some kernels give an erroneously high number */
3802 max_progress
/= info
->new_chunk
/512;
3803 max_progress
*= info
->new_chunk
/512;
3804 /* And round to old chunk size as the kernel wants that */
3805 max_progress
/= info
->array
.chunk_size
/512;
3806 max_progress
*= info
->array
.chunk_size
/512;
3807 /* Limit progress to the whole device */
3808 if (max_progress
> info
->component_size
)
3809 max_progress
= info
->component_size
;
3810 wait_point
/= reshape
->after
.data_disks
;
3812 /* switch from 'device offset' to 'processed block count' */
3813 max_progress
= info
->component_size
- max_progress
;
3814 wait_point
= info
->component_size
- wait_point
;
3818 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
3820 /* Now wait. If we have already reached the point that we were
3821 * asked to wait to, don't wait at all, else wait for any change.
3822 * We need to select on 'sync_completed' as that is the place that
3823 * notifications happen, but we are really interested in
3824 * 'reshape_position'
3826 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
3828 goto check_progress
;
3830 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3831 goto check_progress
;
3833 while (completed
< max_progress
&& completed
< wait_point
) {
3834 /* Check that sync_action is still 'reshape' to avoid
3835 * waiting forever on a dead array
3838 if (sysfs_get_str(info
, NULL
, "sync_action",
3840 strncmp(action
, "reshape", 7) != 0)
3842 /* Some kernels reset 'sync_completed' to zero
3843 * before setting 'sync_action' to 'idle'.
3844 * So we need these extra tests.
3846 if (completed
== 0 && advancing
3847 && strncmp(action
, "idle", 4) == 0
3848 && info
->reshape_progress
> 0)
3850 if (completed
== 0 && !advancing
3851 && strncmp(action
, "idle", 4) == 0
3852 && info
->reshape_progress
< (info
->component_size
3853 * reshape
->after
.data_disks
))
3855 sysfs_wait(fd
, NULL
);
3856 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3857 goto check_progress
;
3859 /* Some kernels reset 'sync_completed' to zero,
3860 * we need to have real point we are in md.
3861 * So in that case, read 'reshape_position' from sysfs.
3863 if (completed
== 0) {
3864 unsigned long long reshapep
;
3866 if (sysfs_get_str(info
, NULL
, "sync_action",
3868 strncmp(action
, "idle", 4) == 0 &&
3869 sysfs_get_ll(info
, NULL
,
3870 "reshape_position", &reshapep
) == 0)
3871 *reshape_completed
= reshapep
;
3873 /* some kernels can give an incorrectly high
3874 * 'completed' number, so round down */
3875 completed
/= (info
->new_chunk
/512);
3876 completed
*= (info
->new_chunk
/512);
3877 /* Convert 'completed' back in to a 'progress' number */
3878 completed
*= reshape
->after
.data_disks
;
3880 completed
= (info
->component_size
3881 * reshape
->after
.data_disks
3883 *reshape_completed
= completed
;
3888 /* We return the need_backup flag. Caller will decide
3889 * how much - a multiple of ->backup_blocks up to *suspend_point
3892 return need_backup
> info
->reshape_progress
;
3894 return need_backup
>= info
->reshape_progress
;
3897 /* if we couldn't read a number from sync_completed, then
3898 * either the reshape did complete, or it aborted.
3899 * We can tell which by checking for 'none' in reshape_position.
3900 * If it did abort, then it might immediately restart if it
3901 * it was just a device failure that leaves us degraded but
3904 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
3905 || strncmp(buf
, "none", 4) != 0) {
3906 /* The abort might only be temporary. Wait up to 10
3907 * seconds for fd to contain a valid number again.
3911 unsigned long long new_sync_max
;
3912 while (fd
>= 0 && rv
< 0 && wait
> 0) {
3913 if (sysfs_wait(fd
, &wait
) != 1)
3915 switch (sysfs_fd_get_ll(fd
, &completed
)) {
3917 /* all good again */
3919 /* If "sync_max" is no longer max_progress
3920 * we need to freeze things
3922 sysfs_get_ll(info
, NULL
, "sync_max", &new_sync_max
);
3923 *frozen
= (new_sync_max
!= max_progress
);
3925 case -2: /* read error - abort */
3932 return rv
; /* abort */
3934 /* Maybe racing with array shutdown - check state */
3937 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
3938 || strncmp(buf
, "inactive", 8) == 0
3939 || strncmp(buf
, "clear",5) == 0)
3940 return -2; /* abort */
3941 return -1; /* complete */
3945 /* FIXME return status is never checked */
3946 static int grow_backup(struct mdinfo
*sra
,
3947 unsigned long long offset
, /* per device */
3948 unsigned long stripes
, /* per device, in old chunks */
3949 int *sources
, unsigned long long *offsets
,
3950 int disks
, int chunk
, int level
, int layout
,
3951 int dests
, int *destfd
, unsigned long long *destoffsets
,
3952 int part
, int *degraded
,
3955 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3956 * to storage 'destfd' (offset 'destoffsets'), after first
3957 * suspending IO. Then allow resync to continue
3958 * over the suspended section.
3959 * Use part 'part' of the backup-super-block.
3964 unsigned long long ll
;
3966 //printf("offset %llu\n", offset);
3972 /* Check that array hasn't become degraded, else we might backup the wrong data */
3973 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
3974 return -1; /* FIXME this error is ignored */
3975 new_degraded
= (int)ll
;
3976 if (new_degraded
!= *degraded
) {
3977 /* check each device to ensure it is still working */
3979 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3980 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
3982 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
3984 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
3985 strstr(sbuf
, "faulty") ||
3986 strstr(sbuf
, "in_sync") == NULL
) {
3987 /* this device is dead */
3988 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
3989 if (sd
->disk
.raid_disk
>= 0 &&
3990 sources
[sd
->disk
.raid_disk
] >= 0) {
3991 close(sources
[sd
->disk
.raid_disk
]);
3992 sources
[sd
->disk
.raid_disk
] = -1;
3997 *degraded
= new_degraded
;
4000 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4001 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4003 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4004 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4007 bsb
.magic
[15] = '2';
4008 for (i
= 0; i
< dests
; i
++)
4010 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
4012 lseek64(destfd
[i
], destoffsets
[i
], 0);
4014 rv
= save_stripes(sources
, offsets
,
4015 disks
, chunk
, level
, layout
,
4017 offset
*512*odata
, stripes
* chunk
* odata
,
4022 bsb
.mtime
= __cpu_to_le64(time(0));
4023 for (i
= 0; i
< dests
; i
++) {
4024 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4026 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4027 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4028 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4029 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4032 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
4033 != destoffsets
[i
] - 4096)
4035 if (write(destfd
[i
], &bsb
, 512) != 512)
4037 if (destoffsets
[i
] > 4096) {
4038 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4039 destoffsets
[i
]+stripes
*chunk
*odata
)
4041 if (write(destfd
[i
], &bsb
, 512) != 512)
4051 /* in 2.6.30, the value reported by sync_completed can be
4052 * less that it should be by one stripe.
4053 * This only happens when reshape hits sync_max and pauses.
4054 * So allow wait_backup to either extent sync_max further
4055 * than strictly necessary, or return before the
4056 * sync has got quite as far as we would really like.
4057 * This is what 'blocks2' is for.
4058 * The various caller give appropriate values so that
4061 /* FIXME return value is often ignored */
4062 static int forget_backup(int dests
, int *destfd
,
4063 unsigned long long *destoffsets
,
4067 * Erase backup 'part' (which is 0 or 1)
4073 bsb
.arraystart2
= __cpu_to_le64(0);
4074 bsb
.length2
= __cpu_to_le64(0);
4076 bsb
.arraystart
= __cpu_to_le64(0);
4077 bsb
.length
= __cpu_to_le64(0);
4079 bsb
.mtime
= __cpu_to_le64(time(0));
4081 for (i
= 0; i
< dests
; i
++) {
4082 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4083 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4084 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4085 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4086 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4087 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4088 destoffsets
[i
]-4096)
4091 write(destfd
[i
], &bsb
, 512) != 512)
4098 static void fail(char *msg
)
4101 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4102 rv
|= (write(2, "\n", 1) != 1);
4106 static char *abuf
, *bbuf
;
4107 static unsigned long long abuflen
;
4108 static void validate(int afd
, int bfd
, unsigned long long offset
)
4110 /* check that the data in the backup against the array.
4111 * This is only used for regression testing and should not
4112 * be used while the array is active
4116 lseek64(bfd
, offset
- 4096, 0);
4117 if (read(bfd
, &bsb2
, 512) != 512)
4118 fail("cannot read bsb");
4119 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4120 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4121 fail("first csum bad");
4122 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4123 fail("magic is bad");
4124 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4125 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4126 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4127 fail("second csum bad");
4129 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4130 fail("devstart is wrong");
4133 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4135 if (abuflen
< len
) {
4139 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4140 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4142 /* just stop validating on mem-alloc failure */
4147 lseek64(bfd
, offset
, 0);
4148 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4149 //printf("len %llu\n", len);
4150 fail("read first backup failed");
4152 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4153 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4154 fail("read first from array failed");
4155 if (memcmp(bbuf
, abuf
, len
) != 0) {
4158 printf("offset=%llu len=%llu\n",
4159 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4160 for (i
=0; i
<len
; i
++)
4161 if (bbuf
[i
] != abuf
[i
]) {
4162 printf("first diff byte %d\n", i
);
4166 fail("data1 compare failed");
4170 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4172 if (abuflen
< len
) {
4176 abuf
= xmalloc(abuflen
);
4177 bbuf
= xmalloc(abuflen
);
4180 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4181 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4182 fail("read second backup failed");
4183 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4184 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4185 fail("read second from array failed");
4186 if (memcmp(bbuf
, abuf
, len
) != 0)
4187 fail("data2 compare failed");
4191 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4192 struct supertype
*st
, unsigned long blocks
,
4193 int *fds
, unsigned long long *offsets
,
4194 int dests
, int *destfd
, unsigned long long *destoffsets
)
4196 /* Monitor a reshape where backup is being performed using
4197 * 'native' mechanism - either to a backup file, or
4198 * to some space in a spare.
4202 unsigned long long speed
;
4203 unsigned long long suspend_point
, array_size
;
4204 unsigned long long backup_point
, wait_point
;
4205 unsigned long long reshape_completed
;
4207 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
4208 int part
= 0; /* The next part of the backup area to fill. It may already
4209 * be full, so we need to check */
4210 int level
= reshape
->level
;
4211 int layout
= reshape
->before
.layout
;
4212 int data
= reshape
->before
.data_disks
;
4213 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4214 int chunk
= sra
->array
.chunk_size
;
4216 unsigned long stripes
;
4220 /* set up the backup-super-block. This requires the
4221 * uuid from the array.
4223 /* Find a superblock */
4224 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4228 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4230 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4231 devfd
= dev_open(dn
, O_RDONLY
);
4234 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4240 pr_err("Cannot find a superblock\n");
4244 memset(&bsb
, 0, 512);
4245 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4246 st
->ss
->uuid_from_super(st
, uuid
);
4247 memcpy(bsb
.set_uuid
, uuid
, 16);
4248 bsb
.mtime
= __cpu_to_le64(time(0));
4249 bsb
.devstart2
= blocks
;
4251 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4252 reshape
->before
.data_disks
;
4254 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4255 /* Don't start the 'reshape' */
4257 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4258 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4259 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4263 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4264 backup_point
= sra
->reshape_progress
;
4267 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4268 backup_point
= reshape
->backup_blocks
;
4269 suspend_point
= array_size
;
4275 /* Want to return as soon the oldest backup slot can
4276 * be released as that allows us to start backing up
4277 * some more, providing suspend_point has been
4278 * advanced, which it should have.
4281 wait_point
= array_size
;
4282 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4283 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4284 __le64_to_cpu(bsb
.length
));
4285 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4286 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4287 __le64_to_cpu(bsb
.length2
));
4290 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4291 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4292 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4293 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4296 reshape_completed
= sra
->reshape_progress
;
4297 rv
= progress_reshape(sra
, reshape
,
4298 backup_point
, wait_point
,
4299 &suspend_point
, &reshape_completed
,
4301 /* external metadata would need to ping_monitor here */
4302 sra
->reshape_progress
= reshape_completed
;
4304 /* Clear any backup region that is before 'here' */
4306 if (__le64_to_cpu(bsb
.length
) > 0 &&
4307 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4308 __le64_to_cpu(bsb
.length
)))
4309 forget_backup(dests
, destfd
,
4311 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4312 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4313 __le64_to_cpu(bsb
.length2
)))
4314 forget_backup(dests
, destfd
,
4317 if (__le64_to_cpu(bsb
.length
) > 0 &&
4318 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4319 forget_backup(dests
, destfd
,
4321 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4322 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4323 forget_backup(dests
, destfd
,
4333 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4334 /* No longer need to monitor this reshape */
4335 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4341 unsigned long long offset
;
4342 unsigned long actual_stripes
;
4343 /* Need to backup some data.
4344 * If 'part' is not used and the desired
4345 * backup size is suspended, do a backup,
4346 * then consider the next part.
4348 /* Check that 'part' is unused */
4349 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4351 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4354 offset
= backup_point
/ data
;
4355 actual_stripes
= stripes
;
4357 if (offset
+ actual_stripes
* (chunk
/512) >
4358 sra
->component_size
)
4359 actual_stripes
= ((sra
->component_size
- offset
)
4361 if (offset
+ actual_stripes
* (chunk
/512) >
4365 if (offset
< actual_stripes
* (chunk
/512))
4366 actual_stripes
= offset
/ (chunk
/512);
4367 offset
-= actual_stripes
* (chunk
/512);
4368 if (offset
< suspend_point
/data
)
4371 if (actual_stripes
== 0)
4373 grow_backup(sra
, offset
, actual_stripes
,
4375 disks
, chunk
, level
, layout
,
4376 dests
, destfd
, destoffsets
,
4377 part
, °raded
, buf
);
4378 validate(afd
, destfd
[0], destoffsets
[0]);
4379 /* record where 'part' is up to */
4382 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4384 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4388 /* FIXME maybe call progress_reshape one more time instead */
4389 /* remove any remaining suspension */
4390 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4391 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4392 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4393 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4395 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4396 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4402 * If any spare contains md_back_data-1 which is recent wrt mtime,
4403 * write that data into the array and update the super blocks with
4404 * the new reshape_progress
4406 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
4407 char *backup_file
, int verbose
)
4411 unsigned long long *offsets
;
4412 unsigned long long nstripe
, ostripe
;
4415 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4416 if (info
->array
.level
== 6) odata
--; /* number of data disks */
4417 ndata
= info
->array
.raid_disks
- 1;
4418 if (info
->new_level
== 6) ndata
--;
4420 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4422 if (info
->delta_disks
<= 0)
4423 /* Didn't grow, so the backup file must have
4427 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4428 struct mdinfo dinfo
;
4431 char *devname
, namebuf
[20];
4432 unsigned long long lo
, hi
;
4434 /* This was a spare and may have some saved data on it.
4435 * Load the superblock, find and load the
4436 * backup_super_block.
4437 * If either fail, go on to next device.
4438 * If the backup contains no new info, just return
4439 * else restore data and update all superblocks
4441 if (i
== old_disks
-1) {
4442 fd
= open(backup_file
, O_RDONLY
);
4444 pr_err("backup file %s inaccessible: %s\n",
4445 backup_file
, strerror(errno
));
4448 devname
= backup_file
;
4453 if (st
->ss
->load_super(st
, fd
, NULL
))
4456 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4457 st
->ss
->free_super(st
);
4460 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4462 pr_err("Cannot seek on device %d\n", i
);
4463 continue; /* Cannot seek */
4465 sprintf(namebuf
, "device-%d", i
);
4468 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4470 pr_err("Cannot read from %s\n", devname
);
4471 continue; /* Cannot read */
4473 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4474 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4476 pr_err("No backup metadata on %s\n", devname
);
4479 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4481 pr_err("Bad backup-metadata checksum on %s\n", devname
);
4482 continue; /* bad checksum */
4484 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4485 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4487 pr_err("Bad backup-metadata checksum2 on %s\n", devname
);
4488 continue; /* Bad second checksum */
4490 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4492 pr_err("Wrong uuid on backup-metadata on %s\n", devname
);
4493 continue; /* Wrong uuid */
4496 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4497 * sometimes they aren't... So allow considerable flexability in matching, and allow
4498 * this test to be overridden by an environment variable.
4500 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
4501 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
4502 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4503 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4504 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4505 (unsigned long)info
->array
.utime
);
4507 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4508 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4509 continue; /* time stamp is too bad */
4513 if (bsb
.magic
[15] == '1') {
4514 if (bsb
.length
== 0)
4516 if (info
->delta_disks
>= 0) {
4517 /* reshape_progress is increasing */
4518 if (__le64_to_cpu(bsb
.arraystart
)
4519 + __le64_to_cpu(bsb
.length
)
4520 < info
->reshape_progress
) {
4523 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4524 continue; /* No new data here */
4527 /* reshape_progress is decreasing */
4528 if (__le64_to_cpu(bsb
.arraystart
) >=
4529 info
->reshape_progress
)
4530 goto nonew
; /* No new data here */
4533 if (bsb
.length
== 0 && bsb
.length2
== 0)
4535 if (info
->delta_disks
>= 0) {
4536 /* reshape_progress is increasing */
4537 if ((__le64_to_cpu(bsb
.arraystart
)
4538 + __le64_to_cpu(bsb
.length
)
4539 < info
->reshape_progress
)
4541 (__le64_to_cpu(bsb
.arraystart2
)
4542 + __le64_to_cpu(bsb
.length2
)
4543 < info
->reshape_progress
))
4544 goto nonew
; /* No new data here */
4546 /* reshape_progress is decreasing */
4547 if (__le64_to_cpu(bsb
.arraystart
) >=
4548 info
->reshape_progress
&&
4549 __le64_to_cpu(bsb
.arraystart2
) >=
4550 info
->reshape_progress
)
4551 goto nonew
; /* No new data here */
4554 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4557 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4559 continue; /* Cannot seek */
4561 /* There should be a duplicate backup superblock 4k before here */
4562 if (lseek64(fd
, -4096, 1) < 0 ||
4563 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4564 goto second_fail
; /* Cannot find leading superblock */
4565 if (bsb
.magic
[15] == '1')
4566 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4568 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4569 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4570 goto second_fail
; /* Cannot find leading superblock */
4572 /* Now need the data offsets for all devices. */
4573 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4574 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4577 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4578 /* FIXME should be this be an error */
4580 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4581 st
->ss
->free_super(st
);
4582 offsets
[j
] = dinfo
.data_offset
* 512;
4584 printf("%s: restoring critical section\n", Name
);
4586 if (restore_stripes(fdlist
, offsets
,
4587 info
->array
.raid_disks
,
4591 fd
, __le64_to_cpu(bsb
.devstart
)*512,
4592 __le64_to_cpu(bsb
.arraystart
)*512,
4593 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4594 /* didn't succeed, so giveup */
4596 pr_err("Error restoring backup from %s\n",
4602 if (bsb
.magic
[15] == '2' &&
4603 restore_stripes(fdlist
, offsets
,
4604 info
->array
.raid_disks
,
4608 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
4609 __le64_to_cpu(bsb
.devstart2
)*512,
4610 __le64_to_cpu(bsb
.arraystart2
)*512,
4611 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4612 /* didn't succeed, so giveup */
4614 pr_err("Error restoring second backup from %s\n",
4622 /* Ok, so the data is restored. Let's update those superblocks. */
4626 lo
= __le64_to_cpu(bsb
.arraystart
);
4627 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4629 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4630 unsigned long long lo1
, hi1
;
4631 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4632 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4636 } else if (lo
< lo1
)
4642 (info
->reshape_progress
< lo
||
4643 info
->reshape_progress
> hi
))
4644 /* backup does not affect reshape_progress*/ ;
4645 else if (info
->delta_disks
>= 0) {
4646 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4647 __le64_to_cpu(bsb
.length
);
4648 if (bsb
.magic
[15] == '2') {
4649 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
4650 __le64_to_cpu(bsb
.length2
);
4651 if (p2
> info
->reshape_progress
)
4652 info
->reshape_progress
= p2
;
4655 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4656 if (bsb
.magic
[15] == '2') {
4657 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
4658 if (p2
< info
->reshape_progress
)
4659 info
->reshape_progress
= p2
;
4662 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4665 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4667 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4668 dinfo
.reshape_progress
= info
->reshape_progress
;
4669 st
->ss
->update_super(st
, &dinfo
,
4670 "_reshape_progress",
4672 st
->ss
->store_super(st
, fdlist
[j
]);
4673 st
->ss
->free_super(st
);
4677 /* Didn't find any backup data, try to see if any
4680 if (info
->delta_disks
< 0) {
4681 /* When shrinking, the critical section is at the end.
4682 * So see if we are before the critical section.
4684 unsigned long long first_block
;
4685 nstripe
= ostripe
= 0;
4687 while (ostripe
>= nstripe
) {
4688 ostripe
+= info
->array
.chunk_size
/ 512;
4689 first_block
= ostripe
* odata
;
4690 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4691 (info
->new_chunk
/512);
4694 if (info
->reshape_progress
>= first_block
)
4697 if (info
->delta_disks
> 0) {
4698 /* See if we are beyond the critical section. */
4699 unsigned long long last_block
;
4700 nstripe
= ostripe
= 0;
4702 while (nstripe
>= ostripe
) {
4703 nstripe
+= info
->new_chunk
/ 512;
4704 last_block
= nstripe
* ndata
;
4705 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4706 (info
->array
.chunk_size
/512);
4709 if (info
->reshape_progress
>= last_block
)
4712 /* needed to recover critical section! */
4714 pr_err("Failed to find backup of critical section\n");
4718 int Grow_continue_command(char *devname
, int fd
,
4719 char *backup_file
, int verbose
)
4722 struct supertype
*st
= NULL
;
4723 struct mdinfo
*content
= NULL
;
4724 struct mdinfo array
;
4725 char *subarray
= NULL
;
4726 struct mdinfo
*cc
= NULL
;
4727 struct mdstat_ent
*mdstat
= NULL
;
4731 dprintf("Grow continue from command line called for %s\n",
4734 st
= super_by_fd(fd
, &subarray
);
4735 if (!st
|| !st
->ss
) {
4736 pr_err("Unable to determine metadata format for %s\n",
4740 dprintf("Grow continue is run for ");
4741 if (st
->ss
->external
== 0) {
4743 dprintf_cont("native array (%s)\n", devname
);
4744 if (ioctl(fd
, GET_ARRAY_INFO
, &array
.array
) < 0) {
4745 pr_err("%s is not an active md array - aborting\n", devname
);
4747 goto Grow_continue_command_exit
;
4750 /* Need to load a superblock.
4751 * FIXME we should really get what we need from
4754 for (d
= 0; d
< MAX_DISKS
; d
++) {
4755 mdu_disk_info_t disk
;
4759 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
4761 if (disk
.major
== 0 && disk
.minor
== 0)
4763 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
4765 dv
= map_dev(disk
.major
, disk
.minor
, 1);
4768 fd2
= dev_open(dv
, O_RDONLY
);
4771 err
= st
->ss
->load_super(st
, fd2
, NULL
);
4773 /* invalidate fd2 to avoid possible double close() */
4779 if (d
== MAX_DISKS
) {
4780 pr_err("Unable to load metadata for %s\n",
4783 goto Grow_continue_command_exit
;
4785 st
->ss
->getinfo_super(st
, content
, NULL
);
4790 dprintf_cont("subarray (%s)\n", subarray
);
4791 container
= st
->container_devnm
;
4792 cfd
= open_dev_excl(st
->container_devnm
);
4794 container
= st
->devnm
;
4796 cfd
= open_dev_excl(st
->devnm
);
4797 dprintf_cont("container (%s)\n", container
);
4801 pr_err("Unable to open container for %s\n", devname
);
4803 goto Grow_continue_command_exit
;
4806 /* find in container array under reshape
4808 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
4810 pr_err("Cannot read superblock for %s\n",
4813 goto Grow_continue_command_exit
;
4816 cc
= st
->ss
->container_content(st
, subarray
);
4817 for (content
= cc
; content
; content
= content
->next
) {
4819 int allow_reshape
= 1;
4821 if (content
->reshape_active
== 0)
4823 /* The decision about array or container wide
4824 * reshape is taken in Grow_continue based
4825 * content->reshape_active state, therefore we
4826 * need to check_reshape based on
4827 * reshape_active and subarray name
4829 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
4831 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
4832 (content
->array
.state
4833 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
4836 if (!allow_reshape
) {
4837 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4838 devname
, container
);
4840 goto Grow_continue_command_exit
;
4843 array
= strchr(content
->text_version
+1, '/')+1;
4844 mdstat
= mdstat_by_subdev(array
, container
);
4847 if (mdstat
->active
== 0) {
4848 pr_err("Skipping inactive array %s.\n",
4850 free_mdstat(mdstat
);
4857 pr_err("Unable to determine reshaped array for %s\n", devname
);
4859 goto Grow_continue_command_exit
;
4861 fd2
= open_dev(mdstat
->devnm
);
4863 pr_err("cannot open (%s)\n", mdstat
->devnm
);
4865 goto Grow_continue_command_exit
;
4868 sysfs_init(content
, fd2
, mdstat
->devnm
);
4870 /* start mdmon in case it is not running
4872 if (!mdmon_running(container
))
4873 start_mdmon(container
);
4874 ping_monitor(container
);
4876 if (mdmon_running(container
))
4877 st
->update_tail
= &st
->updates
;
4879 pr_err("No mdmon found. Grow cannot continue.\n");
4881 goto Grow_continue_command_exit
;
4885 /* verify that array under reshape is started from
4888 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
4890 goto Grow_continue_command_exit
;
4895 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
4897 Grow_continue_command_exit
:
4902 st
->ss
->free_super(st
);
4903 free_mdstat(mdstat
);
4910 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
4911 char *backup_file
, int forked
, int freeze_reshape
)
4915 if (!info
->reshape_active
)
4918 if (st
->ss
->external
) {
4919 int cfd
= open_dev(st
->container_devnm
);
4924 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
4926 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
4927 st
, info
, 0, backup_file
,
4929 1 | info
->reshape_active
,
4932 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
4933 NULL
, INVALID_SECTORS
,
4934 backup_file
, 0, forked
,
4935 1 | info
->reshape_active
,
4941 char *make_backup(char *name
)
4943 char *base
= "backup_file-";
4947 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
4948 fname
= xmalloc(len
);
4949 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
4953 char *locate_backup(char *name
)
4955 char *fl
= make_backup(name
);
4958 if (stat(fl
, &stb
) == 0 &&
4959 S_ISREG(stb
.st_mode
))