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 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
621 static void wait_reshape(struct mdinfo
*sra
)
623 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
629 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
630 strncmp(action
, "reshape", 7) == 0)
631 sysfs_wait(fd
, NULL
);
635 static int reshape_super(struct supertype
*st
, unsigned long long size
,
636 int level
, int layout
, int chunksize
, int raid_disks
,
637 int delta_disks
, char *backup_file
, char *dev
,
638 int direction
, int verbose
)
640 /* nothing extra to check in the native case */
641 if (!st
->ss
->external
)
643 if (!st
->ss
->reshape_super
||
644 !st
->ss
->manage_reshape
) {
645 pr_err("%s metadata does not support reshape\n",
650 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
651 raid_disks
, delta_disks
, backup_file
, dev
,
655 static void sync_metadata(struct supertype
*st
)
657 if (st
->ss
->external
) {
658 if (st
->update_tail
) {
659 flush_metadata_updates(st
);
660 st
->update_tail
= &st
->updates
;
662 st
->ss
->sync_metadata(st
);
666 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
668 /* when dealing with external metadata subarrays we need to be
669 * prepared to handle EAGAIN. The kernel may need to wait for
670 * mdmon to mark the array active so the kernel can handle
671 * allocations/writeback when preparing the reshape action
672 * (md_allow_write()). We temporarily disable safe_mode_delay
673 * to close a race with the array_state going clean before the
674 * next write to raid_disks / stripe_cache_size
679 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
681 (strcmp(name
, "raid_disks") != 0 &&
682 strcmp(name
, "stripe_cache_size") != 0))
683 return sysfs_set_num(sra
, NULL
, name
, n
);
685 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
688 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
689 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
690 if (rc
< 0 && errno
== EAGAIN
) {
691 ping_monitor(container
);
692 /* if we get EAGAIN here then the monitor is not active
695 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
697 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
701 int start_reshape(struct mdinfo
*sra
, int already_running
,
702 int before_data_disks
, int data_disks
)
705 unsigned long long sync_max_to_set
;
707 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
708 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
709 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
710 sra
->reshape_progress
);
711 if (before_data_disks
<= data_disks
)
712 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
714 sync_max_to_set
= (sra
->component_size
* data_disks
715 - sra
->reshape_progress
) / data_disks
;
716 if (!already_running
)
717 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
718 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
719 if (!already_running
&& err
== 0) {
722 err
= sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
725 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
730 void abort_reshape(struct mdinfo
*sra
)
732 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
733 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
734 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
735 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
736 sysfs_set_num(sra
, NULL
, "sync_min", 0);
737 // It isn't safe to reset sync_max as we aren't monitoring.
738 // Array really should be stopped at this point.
741 int remove_disks_for_takeover(struct supertype
*st
,
746 struct mdinfo
*remaining
;
749 if (sra
->array
.level
== 10)
750 nr_of_copies
= layout
& 0xff;
751 else if (sra
->array
.level
== 1)
752 nr_of_copies
= sra
->array
.raid_disks
;
756 remaining
= sra
->devs
;
758 /* for each 'copy', select one device and remove from the list. */
759 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
760 struct mdinfo
**diskp
;
763 /* Find a working device to keep */
764 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
765 struct mdinfo
*disk
= *diskp
;
767 if (disk
->disk
.raid_disk
< slot
)
769 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
771 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
773 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
775 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
778 /* We have found a good disk to use! */
780 disk
->next
= sra
->devs
;
789 if (slot
< sra
->array
.raid_disks
) {
790 /* didn't find all slots */
796 sra
->devs
= remaining
;
800 /* Remove all 'remaining' devices from the array */
802 struct mdinfo
*sd
= remaining
;
803 remaining
= sd
->next
;
805 sysfs_set_str(sra
, sd
, "state", "faulty");
806 sysfs_set_str(sra
, sd
, "slot", "none");
807 /* for external metadata disks should be removed in mdmon */
808 if (!st
->ss
->external
)
809 sysfs_set_str(sra
, sd
, "state", "remove");
810 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
811 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
812 sd
->next
= sra
->devs
;
818 void reshape_free_fdlist(int *fdlist
,
819 unsigned long long *offsets
,
824 for (i
= 0; i
< size
; i
++)
832 int reshape_prepare_fdlist(char *devname
,
836 unsigned long blocks
,
839 unsigned long long *offsets
)
845 for (d
= 0; d
<= nrdisks
; d
++)
848 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
849 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
851 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
852 sd
->disk
.raid_disk
< raid_disks
) {
853 char *dn
= map_dev(sd
->disk
.major
,
855 fdlist
[sd
->disk
.raid_disk
]
856 = dev_open(dn
, O_RDONLY
);
857 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
858 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
859 pr_err("%s: cannot open component %s\n",
860 devname
, dn
? dn
: "-unknown-");
864 } else if (backup_file
== NULL
) {
866 char *dn
= map_dev(sd
->disk
.major
,
868 fdlist
[d
] = dev_open(dn
, O_RDWR
);
869 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
871 pr_err("%s: cannot open component %s\n",
872 devname
, dn
? dn
: "-unknown-");
883 int reshape_open_backup_file(char *backup_file
,
888 unsigned long long *offsets
,
892 /* Return 1 on success, 0 on any form of failure */
893 /* need to check backup file is large enough */
899 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
903 pr_err("%s: cannot create backup file %s: %s\n",
904 devname
, backup_file
, strerror(errno
));
907 /* Guard against backup file being on array device.
908 * If array is partitioned or if LVM etc is in the
909 * way this will not notice, but it is better than
912 fstat(*fdlist
, &stb
);
915 if (stb
.st_rdev
== dev
) {
916 pr_err("backup file must NOT be on the array being reshaped.\n");
922 for (i
=0; i
< blocks
+ 8 ; i
++) {
923 if (write(*fdlist
, buf
, 512) != 512) {
924 pr_err("%s: cannot create backup file %s: %s\n",
925 devname
, backup_file
, strerror(errno
));
929 if (fsync(*fdlist
) != 0) {
930 pr_err("%s: cannot create backup file %s: %s\n",
931 devname
, backup_file
, strerror(errno
));
935 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
936 char *bu
= make_backup(sys_name
);
937 if (symlink(backup_file
, bu
))
938 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
946 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
947 unsigned int ndata
, unsigned int odata
)
949 unsigned long a
, b
, blocks
;
950 /* So how much do we need to backup.
951 * We need an amount of data which is both a whole number of
952 * old stripes and a whole number of new stripes.
953 * So LCM for (chunksize*datadisks).
955 a
= (ochunk
/512) * odata
;
956 b
= (nchunk
/512) * ndata
;
959 /* LCM == product / GCD */
960 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
965 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
967 /* Based on the current array state in info->array and
968 * the changes in info->new_* etc, determine:
969 * - whether the change is possible
970 * - Intermediate level/raid_disks/layout
971 * - whether a restriping reshape is needed
972 * - number of sectors in minimum change unit. This
973 * will cover a whole number of stripes in 'before' and
976 * Return message if the change should be rejected
977 * NULL if the change can be achieved
979 * This can be called as part of starting a reshape, or
980 * when assembling an array that is undergoing reshape.
982 int near
, far
, offset
, copies
;
984 int old_chunk
, new_chunk
;
985 /* delta_parity records change in number of devices
986 * caused by level change
988 int delta_parity
= 0;
990 memset(re
, 0, sizeof(*re
));
992 /* If a new level not explicitly given, we assume no-change */
993 if (info
->new_level
== UnSet
)
994 info
->new_level
= info
->array
.level
;
997 switch (info
->new_level
) {
1003 /* chunk size is meaningful, must divide component_size
1006 if (info
->component_size
% (info
->new_chunk
/512)) {
1007 unsigned long long shrink
= info
->component_size
;
1008 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1009 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1010 info
->new_chunk
/1024, info
->component_size
/2);
1011 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1013 pr_err("will shrink the array so the given chunk size would work.\n");
1018 return "chunk size not meaningful for this level";
1021 info
->new_chunk
= info
->array
.chunk_size
;
1023 switch (info
->array
.level
) {
1025 return "No reshape is possibly for this RAID level";
1027 if (info
->delta_disks
!= UnSet
)
1028 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1030 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1032 /* RAID1 can convert to RAID1 with different disks, or
1033 * raid5 with 2 disks, or
1036 if (info
->new_level
> 1 &&
1037 (info
->component_size
& 7))
1038 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1039 if (info
->new_level
== 0) {
1040 if (info
->delta_disks
!= UnSet
&&
1041 info
->delta_disks
!= 0)
1042 return "Cannot change number of disks with RAID1->RAID0 conversion";
1044 re
->before
.data_disks
= 1;
1045 re
->after
.data_disks
= 1;
1048 if (info
->new_level
== 1) {
1049 if (info
->delta_disks
== UnSet
)
1050 /* Don't know what to do */
1051 return "no change requested for Growing RAID1";
1055 if (info
->array
.raid_disks
== 2 &&
1056 info
->new_level
== 5) {
1059 re
->before
.data_disks
= 1;
1060 if (info
->delta_disks
!= UnSet
&&
1061 info
->delta_disks
!= 0)
1062 re
->after
.data_disks
= 1 + info
->delta_disks
;
1064 re
->after
.data_disks
= 1;
1065 if (re
->after
.data_disks
< 1)
1066 return "Number of disks too small for RAID5";
1068 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1069 info
->array
.chunk_size
= 65536;
1072 /* Could do some multi-stage conversions, but leave that to
1075 return "Impossibly level change request for RAID1";
1078 /* RAID10 can be converted from near mode to
1079 * RAID0 by removing some devices.
1080 * It can also be reshaped if the kernel supports
1083 switch (info
->new_level
) {
1085 if ((info
->array
.layout
& ~0xff) != 0x100)
1086 return "Cannot Grow RAID10 with far/offset layout";
1087 /* number of devices must be multiple of number of copies */
1088 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1089 return "RAID10 layout too complex for Grow operation";
1091 new_disks
= (info
->array
.raid_disks
1092 / (info
->array
.layout
& 0xff));
1093 if (info
->delta_disks
== UnSet
)
1094 info
->delta_disks
= (new_disks
1095 - info
->array
.raid_disks
);
1097 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1098 return "New number of raid-devices impossible for RAID10";
1099 if (info
->new_chunk
&&
1100 info
->new_chunk
!= info
->array
.chunk_size
)
1101 return "Cannot change chunk-size with RAID10 Grow";
1105 re
->before
.data_disks
= new_disks
;
1106 re
->after
.data_disks
= re
->before
.data_disks
;
1110 near
= info
->array
.layout
& 0xff;
1111 far
= (info
->array
.layout
>> 8) & 0xff;
1112 offset
= info
->array
.layout
& 0x10000;
1113 if (far
> 1 && !offset
)
1114 return "Cannot reshape RAID10 in far-mode";
1115 copies
= near
* far
;
1117 old_chunk
= info
->array
.chunk_size
* far
;
1119 if (info
->new_layout
== UnSet
)
1120 info
->new_layout
= info
->array
.layout
;
1122 near
= info
->new_layout
& 0xff;
1123 far
= (info
->new_layout
>> 8) & 0xff;
1124 offset
= info
->new_layout
& 0x10000;
1125 if (far
> 1 && !offset
)
1126 return "Cannot reshape RAID10 to far-mode";
1127 if (near
* far
!= copies
)
1128 return "Cannot change number of copies when reshaping RAID10";
1130 if (info
->delta_disks
== UnSet
)
1131 info
->delta_disks
= 0;
1132 new_disks
= (info
->array
.raid_disks
+
1135 new_chunk
= info
->new_chunk
* far
;
1138 re
->before
.layout
= info
->array
.layout
;
1139 re
->before
.data_disks
= info
->array
.raid_disks
;
1140 re
->after
.layout
= info
->new_layout
;
1141 re
->after
.data_disks
= new_disks
;
1142 /* For RAID10 we don't do backup but do allow reshape,
1143 * so set backup_blocks to INVALID_SECTORS rather than
1145 * And there is no need to synchronise stripes on both
1146 * 'old' and 'new'. So the important
1147 * number is the minimum data_offset difference
1148 * which is the larger of (offset copies * chunk).
1150 re
->backup_blocks
= INVALID_SECTORS
;
1151 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1152 if (new_disks
< re
->before
.data_disks
&&
1153 info
->space_after
< re
->min_offset_change
)
1154 /* Reduce component size by one chunk */
1155 re
->new_size
= (info
->component_size
-
1156 re
->min_offset_change
);
1158 re
->new_size
= info
->component_size
;
1159 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1163 return "RAID10 can only be changed to RAID0";
1166 /* RAID0 can be converted to RAID10, or to RAID456 */
1167 if (info
->new_level
== 10) {
1168 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1169 /* Assume near=2 layout */
1170 info
->new_layout
= 0x102;
1171 info
->delta_disks
= info
->array
.raid_disks
;
1173 if (info
->new_layout
== UnSet
) {
1174 int copies
= 1 + (info
->delta_disks
1175 / info
->array
.raid_disks
);
1176 if (info
->array
.raid_disks
* (copies
-1)
1177 != info
->delta_disks
)
1178 return "Impossible number of devices for RAID0->RAID10";
1179 info
->new_layout
= 0x100 + copies
;
1181 if (info
->delta_disks
== UnSet
) {
1182 int copies
= info
->new_layout
& 0xff;
1183 if (info
->new_layout
!= 0x100 + copies
)
1184 return "New layout impossible for RAID0->RAID10";;
1185 info
->delta_disks
= (copies
- 1) *
1186 info
->array
.raid_disks
;
1188 if (info
->new_chunk
&&
1189 info
->new_chunk
!= info
->array
.chunk_size
)
1190 return "Cannot change chunk-size with RAID0->RAID10";
1193 re
->before
.data_disks
= (info
->array
.raid_disks
+
1195 re
->after
.data_disks
= re
->before
.data_disks
;
1196 re
->before
.layout
= info
->new_layout
;
1200 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1201 * a raid4 style layout of the final level.
1203 switch (info
->new_level
) {
1208 re
->before
.layout
= 0;
1213 re
->before
.layout
= ALGORITHM_PARITY_N
;
1214 if (info
->new_layout
== UnSet
)
1215 info
->new_layout
= map_name(r5layout
, "default");
1220 re
->before
.layout
= ALGORITHM_PARITY_N
;
1221 if (info
->new_layout
== UnSet
)
1222 info
->new_layout
= map_name(r6layout
, "default");
1225 return "Impossible level change requested";
1227 re
->before
.data_disks
= info
->array
.raid_disks
;
1228 /* determining 'after' layout happens outside this 'switch' */
1232 info
->array
.layout
= ALGORITHM_PARITY_N
;
1234 switch (info
->new_level
) {
1238 re
->level
= info
->array
.level
;
1239 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1240 re
->before
.layout
= info
->array
.layout
;
1244 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1245 re
->before
.layout
= info
->array
.layout
;
1250 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1251 switch (info
->array
.layout
) {
1252 case ALGORITHM_LEFT_ASYMMETRIC
:
1253 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1255 case ALGORITHM_RIGHT_ASYMMETRIC
:
1256 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1258 case ALGORITHM_LEFT_SYMMETRIC
:
1259 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1261 case ALGORITHM_RIGHT_SYMMETRIC
:
1262 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1264 case ALGORITHM_PARITY_0
:
1265 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1267 case ALGORITHM_PARITY_N
:
1268 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1271 return "Cannot convert an array with this layout";
1275 if (info
->array
.raid_disks
!= 2)
1276 return "Can only convert a 2-device array to RAID1";
1277 if (info
->delta_disks
!= UnSet
&&
1278 info
->delta_disks
!= 0)
1279 return "Cannot set raid_disk when converting RAID5->RAID1";
1281 info
->new_chunk
= 0;
1284 return "Impossible level change requested";
1288 switch (info
->new_level
) {
1294 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1295 re
->before
.layout
= info
->array
.layout
;
1298 return "Impossible level change requested";
1303 /* If we reached here then it looks like a re-stripe is
1304 * happening. We have determined the intermediate level
1305 * and initial raid_disks/layout and stored these in 're'.
1307 * We need to deduce the final layout that can be atomically
1308 * converted to the end state.
1310 switch (info
->new_level
) {
1312 /* We can only get to RAID0 from RAID4 or RAID5
1313 * with appropriate layout and one extra device
1315 if (re
->level
!= 4 && re
->level
!= 5)
1316 return "Cannot covert to RAID0 from this level";
1318 switch (re
->level
) {
1320 re
->before
.layout
= 0;
1321 re
->after
.layout
= 0;
1324 re
->after
.layout
= ALGORITHM_PARITY_N
;
1330 /* We can only get to RAID4 from RAID5 */
1331 if (re
->level
!= 4 && re
->level
!= 5)
1332 return "Cannot convert to RAID4 from this level";
1334 switch (re
->level
) {
1336 re
->after
.layout
= 0;
1339 re
->after
.layout
= ALGORITHM_PARITY_N
;
1345 /* We get to RAID5 from RAID5 or RAID6 */
1346 if (re
->level
!= 5 && re
->level
!= 6)
1347 return "Cannot convert to RAID5 from this level";
1349 switch (re
->level
) {
1351 if (info
->new_layout
== UnSet
)
1352 re
->after
.layout
= re
->before
.layout
;
1354 re
->after
.layout
= info
->new_layout
;
1357 if (info
->new_layout
== UnSet
)
1358 info
->new_layout
= re
->before
.layout
;
1360 /* after.layout needs to be raid6 version of new_layout */
1361 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1362 re
->after
.layout
= ALGORITHM_PARITY_N
;
1365 char *ls
= map_num(r5layout
, info
->new_layout
);
1368 /* Current RAID6 layout has a RAID5
1371 strcat(strcpy(layout
, ls
), "-6");
1372 l
= map_name(r6layout
, layout
);
1374 return "Cannot find RAID6 layout to convert to";
1376 /* Current RAID6 has no equivalent.
1377 * If it is already a '-6' layout we
1378 * can leave it unchanged, else we must
1381 ls
= map_num(r6layout
, info
->new_layout
);
1383 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1384 return "Please specify new layout";
1385 l
= info
->new_layout
;
1387 re
->after
.layout
= l
;
1393 /* We must already be at level 6 */
1395 return "Impossible level change";
1396 if (info
->new_layout
== UnSet
)
1397 re
->after
.layout
= info
->array
.layout
;
1399 re
->after
.layout
= info
->new_layout
;
1402 return "Impossible level change requested";
1404 if (info
->delta_disks
== UnSet
)
1405 info
->delta_disks
= delta_parity
;
1407 re
->after
.data_disks
= (re
->before
.data_disks
1410 switch (re
->level
) {
1411 case 6: re
->parity
= 2;
1414 case 5: re
->parity
= 1;
1416 default: re
->parity
= 0;
1419 /* So we have a restripe operation, we need to calculate the number
1420 * of blocks per reshape operation.
1422 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1423 if (info
->new_chunk
== 0)
1424 info
->new_chunk
= info
->array
.chunk_size
;
1425 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1426 re
->after
.layout
== re
->before
.layout
&&
1427 info
->new_chunk
== info
->array
.chunk_size
) {
1428 /* Nothing to change, can change level immediately. */
1429 re
->level
= info
->new_level
;
1430 re
->backup_blocks
= 0;
1433 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1434 /* chunk and layout changes make no difference */
1435 re
->level
= info
->new_level
;
1436 re
->backup_blocks
= 0;
1440 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1441 get_linux_version() < 2006032)
1442 return "in-place reshape is not safe before 2.6.32 - sorry.";
1444 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1445 get_linux_version() < 2006030)
1446 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1448 re
->backup_blocks
= compute_backup_blocks(
1449 info
->new_chunk
, info
->array
.chunk_size
,
1450 re
->after
.data_disks
,
1451 re
->before
.data_disks
);
1452 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1454 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1458 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1461 struct mdinfo
*info
;
1465 if ((st
== NULL
) || (sra
== NULL
))
1468 if (text_version
== NULL
)
1469 text_version
= sra
->text_version
;
1470 subarray
= strchr(text_version
+1, '/')+1;
1471 info
= st
->ss
->container_content(st
, subarray
);
1473 unsigned long long current_size
= 0;
1474 unsigned long long new_size
=
1475 info
->custom_array_size
/2;
1477 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1478 new_size
> current_size
) {
1479 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1481 dprintf("Error: Cannot set array size");
1484 dprintf("Array size changed");
1486 dprintf_cont(" from %llu to %llu.\n",
1487 current_size
, new_size
);
1491 dprintf("Error: set_array_size(): info pointer in NULL\n");
1496 static int reshape_array(char *container
, int fd
, char *devname
,
1497 struct supertype
*st
, struct mdinfo
*info
,
1498 int force
, struct mddev_dev
*devlist
,
1499 unsigned long long data_offset
,
1500 char *backup_file
, int verbose
, int forked
,
1501 int restart
, int freeze_reshape
);
1502 static int reshape_container(char *container
, char *devname
,
1504 struct supertype
*st
,
1505 struct mdinfo
*info
,
1507 char *backup_file
, int verbose
,
1508 int forked
, int restart
, int freeze_reshape
);
1510 int Grow_reshape(char *devname
, int fd
,
1511 struct mddev_dev
*devlist
,
1512 unsigned long long data_offset
,
1513 struct context
*c
, struct shape
*s
)
1515 /* Make some changes in the shape of an array.
1516 * The kernel must support the change.
1518 * There are three different changes. Each can trigger
1519 * a resync or recovery so we freeze that until we have
1520 * requested everything (if kernel supports freezing - 2.6.30).
1522 * - change size (i.e. component_size)
1524 * - change layout/chunksize/ndisks
1526 * The last can require a reshape. It is different on different
1527 * levels so we need to check the level before actioning it.
1528 * Some times the level change needs to be requested after the
1529 * reshape (e.g. raid6->raid5, raid5->raid0)
1532 struct mdu_array_info_s array
;
1534 struct supertype
*st
;
1535 char *subarray
= NULL
;
1539 char *container
= NULL
;
1542 struct mddev_dev
*dv
;
1548 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1549 pr_err("%s is not an active md array - aborting\n",
1553 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10
1554 && (array
.level
< 4 || array
.level
> 6)) {
1555 pr_err("--grow --data-offset not yet supported\n");
1560 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1561 pr_err("cannot change component size at the same time as other changes.\n"
1562 " Change size first, then check data is intact before making other changes.\n");
1566 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&& array
.level
> 1 &&
1567 get_linux_version() < 2006032 &&
1568 !check_env("MDADM_FORCE_FEWER")) {
1569 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1570 " Please use a newer kernel\n");
1574 st
= super_by_fd(fd
, &subarray
);
1576 pr_err("Unable to determine metadata format for %s\n", devname
);
1579 if (s
->raiddisks
> st
->max_devs
) {
1580 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1584 /* in the external case we need to check that the requested reshape is
1585 * supported, and perform an initial check that the container holds the
1586 * pre-requisite spare devices (mdmon owns final validation)
1588 if (st
->ss
->external
) {
1592 container
= st
->container_devnm
;
1593 cfd
= open_dev_excl(st
->container_devnm
);
1595 container
= st
->devnm
;
1597 cfd
= open_dev_excl(st
->devnm
);
1601 pr_err("Unable to open container for %s\n",
1607 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1610 pr_err("Cannot read superblock for %s\n",
1616 /* check if operation is supported for metadata handler */
1617 if (st
->ss
->container_content
) {
1618 struct mdinfo
*cc
= NULL
;
1619 struct mdinfo
*content
= NULL
;
1621 cc
= st
->ss
->container_content(st
, subarray
);
1622 for (content
= cc
; content
; content
= content
->next
) {
1623 int allow_reshape
= 1;
1625 /* check if reshape is allowed based on metadata
1626 * indications stored in content.array.status
1628 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
1630 if (content
->array
.state
1631 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))
1633 if (!allow_reshape
) {
1634 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1635 devname
, container
);
1643 if (mdmon_running(container
))
1644 st
->update_tail
= &st
->updates
;
1648 for (dv
= devlist
; dv
; dv
= dv
->next
)
1650 if (s
->raiddisks
> array
.raid_disks
&&
1651 array
.spare_disks
+added_disks
< (s
->raiddisks
- array
.raid_disks
) &&
1653 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1654 " Use --force to over-ride this check.\n",
1655 s
->raiddisks
- array
.raid_disks
,
1656 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1657 array
.spare_disks
+ added_disks
);
1661 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
1662 | GET_STATE
| GET_VERSION
);
1664 if (st
->ss
->external
&& subarray
== NULL
) {
1665 array
.level
= LEVEL_CONTAINER
;
1666 sra
->array
.level
= LEVEL_CONTAINER
;
1669 pr_err("failed to read sysfs parameters for %s\n",
1673 frozen
= freeze(st
);
1675 /* freeze() already spewed the reason */
1678 } else if (frozen
< 0) {
1679 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1684 /* ========= set size =============== */
1685 if (s
->size
> 0 && (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1686 unsigned long long orig_size
= get_component_size(fd
)/2;
1687 unsigned long long min_csize
;
1689 int raid0_takeover
= 0;
1692 orig_size
= (unsigned) array
.size
;
1694 if (orig_size
== 0) {
1695 pr_err("Cannot set device size in this type of array.\n");
1700 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1701 devname
, APPLY_METADATA_CHANGES
, c
->verbose
> 0)) {
1706 if (st
->ss
->external
) {
1707 /* metadata can have size limitation
1708 * update size value according to metadata information
1710 struct mdinfo
*sizeinfo
=
1711 st
->ss
->container_content(st
, subarray
);
1713 unsigned long long new_size
=
1714 sizeinfo
->custom_array_size
/2;
1715 int data_disks
= get_data_disks(
1716 sizeinfo
->array
.level
,
1717 sizeinfo
->array
.layout
,
1718 sizeinfo
->array
.raid_disks
);
1719 new_size
/= data_disks
;
1720 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size
, new_size
,
1721 new_size
* data_disks
);
1723 sysfs_free(sizeinfo
);
1727 /* Update the size of each member device in case
1728 * they have been resized. This will never reduce
1729 * below the current used-size. The "size" attribute
1730 * understands '0' to mean 'max'.
1734 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
1735 if (sysfs_set_num(sra
, mdi
, "size",
1736 s
->size
== MAX_SIZE
? 0 : s
->size
) < 0) {
1737 /* Probably kernel refusing to let us
1738 * reduce the size - not an error.
1742 if (array
.not_persistent
== 0 &&
1743 array
.major_version
== 0 &&
1744 get_linux_version() < 3001000) {
1745 /* Dangerous to allow size to exceed 2TB */
1746 unsigned long long csize
;
1747 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
1748 if (csize
>= 2ULL*1024*1024*1024)
1749 csize
= 2ULL*1024*1024*1024;
1750 if ((min_csize
== 0 || (min_csize
1757 pr_err("Cannot set size on array members.\n");
1758 goto size_change_error
;
1760 if (min_csize
&& s
->size
> min_csize
) {
1761 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1763 goto size_change_error
;
1765 if (min_csize
&& s
->size
== MAX_SIZE
) {
1766 /* Don't let the kernel choose a size - it will get
1769 pr_err("Limited v0.90 array to 2TB per device\n");
1770 s
->size
= min_csize
;
1772 if (st
->ss
->external
) {
1773 if (sra
->array
.level
== 0) {
1774 rv
= sysfs_set_str(sra
, NULL
, "level",
1778 /* get array parameters after takeover
1779 * to change one parameter at time only
1781 rv
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
1784 /* make sure mdmon is
1785 * aware of the new level */
1786 if (!mdmon_running(st
->container_devnm
))
1787 start_mdmon(st
->container_devnm
);
1788 ping_monitor(container
);
1789 if (mdmon_running(st
->container_devnm
) &&
1790 st
->update_tail
== NULL
)
1791 st
->update_tail
= &st
->updates
;
1794 if (s
->size
== MAX_SIZE
)
1796 array
.size
= s
->size
;
1797 if (s
->size
& ~INT32_MAX
) {
1798 /* got truncated to 32bit, write to
1799 * component_size instead
1802 rv
= sysfs_set_num(sra
, NULL
,
1803 "component_size", s
->size
);
1807 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1809 /* manage array size when it is managed externally
1811 if ((rv
== 0) && st
->ss
->external
)
1812 rv
= set_array_size(st
, sra
, sra
->text_version
);
1815 if (raid0_takeover
) {
1816 /* do not recync non-existing parity,
1817 * we will drop it anyway
1819 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
1820 /* go back to raid0, drop parity disk
1822 sysfs_set_str(sra
, NULL
, "level", "raid0");
1823 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1830 /* restore metadata */
1831 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1832 UnSet
, NULL
, devname
,
1833 ROLLBACK_METADATA_CHANGES
,
1836 pr_err("Cannot set device size for %s: %s\n",
1837 devname
, strerror(err
));
1839 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1840 cont_err("Bitmap must be removed before size can be changed\n");
1844 if (s
->assume_clean
) {
1845 /* This will fail on kernels older than 3.0 unless
1846 * a backport has been arranged.
1849 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
1850 pr_err("--assume-clean not supported with --grow on this kernel\n");
1852 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1853 s
->size
= get_component_size(fd
)/2;
1855 s
->size
= array
.size
;
1856 if (c
->verbose
>= 0) {
1857 if (s
->size
== orig_size
)
1858 pr_err("component size of %s unchanged at %lluK\n",
1861 pr_err("component size of %s has been set to %lluK\n",
1865 } else if (array
.level
!= LEVEL_CONTAINER
) {
1866 s
->size
= get_component_size(fd
)/2;
1868 s
->size
= array
.size
;
1871 /* See if there is anything else to do */
1872 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
1873 (s
->layout_str
== NULL
) &&
1874 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
1875 data_offset
== INVALID_SECTORS
&&
1876 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
1877 /* Nothing more to do */
1878 if (!changed
&& c
->verbose
>= 0)
1879 pr_err("%s: no change requested\n",
1884 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1885 * current implementation assumes that following conditions must be met:
1888 * - near_copies == 2
1890 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
1891 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1892 (s
->level
== 0 && array
.level
== 1 && sra
)) {
1894 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1896 dprintf("Array cannot be reshaped\n");
1902 /* Make sure mdmon has seen the device removal
1903 * and updated metadata before we continue with
1907 ping_monitor(container
);
1910 memset(&info
, 0, sizeof(info
));
1912 sysfs_init(&info
, fd
, NULL
);
1913 strcpy(info
.text_version
, sra
->text_version
);
1914 info
.component_size
= s
->size
*2;
1915 info
.new_level
= s
->level
;
1916 info
.new_chunk
= s
->chunk
* 1024;
1917 if (info
.array
.level
== LEVEL_CONTAINER
) {
1918 info
.delta_disks
= UnSet
;
1919 info
.array
.raid_disks
= s
->raiddisks
;
1920 } else if (s
->raiddisks
)
1921 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
1923 info
.delta_disks
= UnSet
;
1924 if (s
->layout_str
== NULL
) {
1925 info
.new_layout
= UnSet
;
1926 if (info
.array
.level
== 6 &&
1927 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1928 info
.array
.layout
>= 16) {
1929 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
1930 cont_err("during the reshape, please specify --layout=preserve\n");
1931 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1935 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
1936 strcmp(s
->layout_str
, "normalize") == 0) {
1937 /* If we have a -6 RAID6 layout, remove the '-6'. */
1938 info
.new_layout
= UnSet
;
1939 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1941 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1942 h
= strrchr(l
, '-');
1943 if (h
&& strcmp(h
, "-6") == 0) {
1945 info
.new_layout
= map_name(r6layout
, l
);
1948 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
1952 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
1953 /* This means that a non-standard RAID6 layout
1956 * - When reshape a RAID6 (e.g. adding a device)
1957 * which is in a non-standard layout, it is OK
1958 * to preserve that layout.
1959 * - When converting a RAID5 to RAID6, leave it in
1960 * the XXX-6 layout, don't re-layout.
1962 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
1963 info
.new_layout
= info
.array
.layout
;
1964 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
1966 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
1968 info
.new_layout
= map_name(r6layout
, l
);
1970 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
1975 int l
= info
.new_level
;
1977 l
= info
.array
.level
;
1980 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
1983 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
1986 info
.new_layout
= parse_layout_10(s
->layout_str
);
1989 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
1992 pr_err("layout not meaningful with this level\n");
1996 if (info
.new_layout
== UnSet
) {
1997 pr_err("layout %s not understood for this level\n",
2004 if (array
.level
== LEVEL_FAULTY
) {
2005 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2006 pr_err("cannot change level of Faulty device\n");
2010 pr_err("cannot set chunksize of Faulty device\n");
2013 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2014 pr_err("cannot set raid_disks of Faulty device\n");
2017 if (s
->layout_str
) {
2018 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2019 dprintf("Cannot get array information.\n");
2022 array
.layout
= info
.new_layout
;
2023 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2024 pr_err("failed to set new layout\n");
2026 } else if (c
->verbose
>= 0)
2027 printf("layout for %s set to %d\n",
2028 devname
, array
.layout
);
2030 } else if (array
.level
== LEVEL_CONTAINER
) {
2031 /* This change is to be applied to every array in the
2032 * container. This is only needed when the metadata imposes
2033 * restraints of the various arrays in the container.
2034 * Currently we only know that IMSM requires all arrays
2035 * to have the same number of devices so changing the
2036 * number of devices (On-Line Capacity Expansion) must be
2037 * performed at the level of the container
2039 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2040 c
->force
, c
->backup_file
, c
->verbose
, 0, 0, 0);
2043 /* get spare devices from external metadata
2045 if (st
->ss
->external
) {
2046 struct mdinfo
*info2
;
2048 info2
= st
->ss
->container_content(st
, subarray
);
2050 info
.array
.spare_disks
=
2051 info2
->array
.spare_disks
;
2056 /* Impose these changes on a single array. First
2057 * check that the metadata is OK with the change. */
2059 if (reshape_super(st
, 0, info
.new_level
,
2060 info
.new_layout
, info
.new_chunk
,
2061 info
.array
.raid_disks
, info
.delta_disks
,
2062 c
->backup_file
, devname
, APPLY_METADATA_CHANGES
,
2068 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2069 devlist
, data_offset
, c
->backup_file
, c
->verbose
,
2080 /* verify_reshape_position()
2081 * Function checks if reshape position in metadata is not farther
2082 * than position in md.
2084 * 0 : not valid sysfs entry
2085 * it can be caused by not started reshape, it should be started
2086 * by reshape array or raid0 array is before takeover
2087 * -1 : error, reshape position is obviously wrong
2088 * 1 : success, reshape progress correct or updated
2090 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2096 /* read sync_max, failure can mean raid0 array */
2097 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2101 unsigned long long position
= strtoull(buf
, &ep
, 0);
2103 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2104 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2105 position
*= get_data_disks(level
,
2107 info
->array
.raid_disks
);
2108 if (info
->reshape_progress
< position
) {
2109 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2110 info
->reshape_progress
, position
);
2111 info
->reshape_progress
= position
;
2113 } else if (info
->reshape_progress
> position
) {
2114 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2115 position
, info
->reshape_progress
);
2118 dprintf("Reshape position in md and metadata are the same;");
2122 } else if (rv
== 0) {
2123 /* for valid sysfs entry, 0-length content
2124 * should be indicated as error
2132 static unsigned long long choose_offset(unsigned long long lo
,
2133 unsigned long long hi
,
2134 unsigned long long min
,
2135 unsigned long long max
)
2137 /* Choose a new offset between hi and lo.
2138 * It must be between min and max, but
2139 * we would prefer something near the middle of hi/lo, and also
2140 * prefer to be aligned to a big power of 2.
2142 * So we start with the middle, then for each bit,
2143 * starting at '1' and increasing, if it is set, we either
2144 * add it or subtract it if possible, preferring the option
2145 * which is furthest from the boundary.
2147 * We stop once we get a 1MB alignment. As units are in sectors,
2148 * 1MB = 2*1024 sectors.
2150 unsigned long long choice
= (lo
+ hi
) / 2;
2151 unsigned long long bit
= 1;
2153 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2154 unsigned long long bigger
, smaller
;
2155 if (! (bit
& choice
))
2157 bigger
= choice
+ bit
;
2158 smaller
= choice
- bit
;
2159 if (bigger
> max
&& smaller
< min
)
2163 else if (smaller
< min
)
2165 else if (hi
- bigger
> smaller
- lo
)
2173 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2174 char *devname
, int delta_disks
,
2175 unsigned long long data_offset
,
2176 unsigned long long min
,
2182 unsigned long long before
, after
;
2184 /* Need to find min space before and after so same is used
2187 before
= UINT64_MAX
;
2189 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2193 struct supertype
*st2
;
2194 struct mdinfo info2
;
2196 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2198 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2199 dfd
= dev_open(dn
, O_RDONLY
);
2201 pr_err("%s: cannot open component %s\n",
2202 devname
, dn
? dn
: "-unknown-");
2205 st2
= dup_super(st
);
2206 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2210 pr_err("%s: cannot get superblock from %s\n",
2214 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2215 st2
->ss
->free_super(st2
);
2217 if (info2
.space_before
== 0 &&
2218 info2
.space_after
== 0) {
2219 /* Metadata doesn't support data_offset changes */
2221 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2225 if (before
> info2
.space_before
)
2226 before
= info2
.space_before
;
2227 if (after
> info2
.space_after
)
2228 after
= info2
.space_after
;
2230 if (data_offset
!= INVALID_SECTORS
) {
2232 if (info2
.data_offset
== data_offset
) {
2233 pr_err("%s: already has that data_offset\n",
2237 if (data_offset
< info2
.data_offset
)
2241 } else if ((data_offset
<= info2
.data_offset
&& dir
== 1) ||
2242 (data_offset
>= info2
.data_offset
&& dir
== -1)) {
2243 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2249 if (before
== UINT64_MAX
)
2250 /* impossible really, there must be no devices */
2253 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2254 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2255 unsigned long long new_data_offset
;
2257 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2259 if (delta_disks
< 0) {
2260 /* Don't need any space as array is shrinking
2261 * just move data_offset up by min
2263 if (data_offset
== INVALID_SECTORS
)
2264 new_data_offset
= sd
->data_offset
+ min
;
2266 if (data_offset
< sd
->data_offset
+ min
) {
2267 pr_err("--data-offset too small for %s\n",
2271 new_data_offset
= data_offset
;
2273 } else if (delta_disks
> 0) {
2274 /* need space before */
2278 pr_err("Insufficient head-space for reshape on %s\n",
2282 if (data_offset
== INVALID_SECTORS
)
2283 new_data_offset
= sd
->data_offset
- min
;
2285 if (data_offset
> sd
->data_offset
- min
) {
2286 pr_err("--data-offset too large for %s\n",
2290 new_data_offset
= data_offset
;
2294 /* can move up or down. If 'data_offset'
2295 * was set we would have already decided,
2296 * so just choose direction with most space.
2303 sysfs_set_str(sra
, NULL
, "reshape_direction",
2304 dir
== 1 ? "backwards" : "forwards");
2306 /* Increase data offset */
2310 pr_err("Insufficient tail-space for reshape on %s\n",
2314 if (data_offset
!= INVALID_SECTORS
&&
2315 data_offset
< sd
->data_offset
+ min
) {
2316 pr_err("--data-offset too small on %s\n",
2320 if (data_offset
!= INVALID_SECTORS
)
2321 new_data_offset
= data_offset
;
2323 new_data_offset
= choose_offset(sd
->data_offset
,
2324 sd
->data_offset
+ after
,
2325 sd
->data_offset
+ min
,
2326 sd
->data_offset
+ after
);
2328 /* Decrease data offset */
2332 pr_err("insufficient head-room on %s\n",
2336 if (data_offset
!= INVALID_SECTORS
&&
2337 data_offset
< sd
->data_offset
- min
) {
2338 pr_err("--data-offset too small on %s\n",
2342 if (data_offset
!= INVALID_SECTORS
)
2343 new_data_offset
= data_offset
;
2345 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2347 sd
->data_offset
- before
,
2348 sd
->data_offset
- min
);
2351 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2352 if (err
< 0 && errno
== E2BIG
) {
2353 /* try again after increasing data size to max */
2354 err
= sysfs_set_num(sra
, sd
, "size", 0);
2355 if (err
< 0 && errno
== EINVAL
&&
2356 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2357 /* some kernels have a bug where you cannot
2358 * use '0' on spare devices. */
2359 sysfs_set_num(sra
, sd
, "size",
2360 (sra
->component_size
+ after
)/2);
2362 err
= sysfs_set_num(sra
, sd
, "new_offset",
2366 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2367 pr_err("data-offset is too big for %s\n",
2371 if (sd
== sra
->devs
&&
2372 (errno
== ENOENT
|| errno
== E2BIG
))
2373 /* Early kernel, no 'new_offset' file,
2374 * or kernel doesn't like us.
2375 * For RAID5/6 this is not fatal
2378 pr_err("Cannot set new_offset for %s\n",
2387 /* Just use a backup file */
2391 static int raid10_reshape(char *container
, int fd
, char *devname
,
2392 struct supertype
*st
, struct mdinfo
*info
,
2393 struct reshape
*reshape
,
2394 unsigned long long data_offset
,
2395 int force
, int verbose
)
2397 /* Changing raid_disks, layout, chunksize or possibly
2398 * just data_offset for a RAID10.
2399 * We must always change data_offset. We change by at least
2400 * ->min_offset_change which is the largest of the old and new
2402 * If raid_disks is increasing, then data_offset must decrease
2403 * by at least this copy size.
2404 * If raid_disks is unchanged, data_offset must increase or
2405 * decrease by at least min_offset_change but preferably by much more.
2406 * We choose half of the available space.
2407 * If raid_disks is decreasing, data_offset must increase by
2408 * at least min_offset_change. To allow of this, component_size
2409 * must be decreased by the same amount.
2411 * So we calculate the required minimum and direction, possibly
2412 * reduce the component_size, then iterate through the devices
2413 * and set the new_data_offset.
2414 * If that all works, we set chunk_size, layout, raid_disks, and start
2418 unsigned long long min
;
2421 sra
= sysfs_read(fd
, NULL
,
2422 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2425 pr_err("%s: Cannot get array details from sysfs\n",
2429 min
= reshape
->min_offset_change
;
2431 if (info
->delta_disks
)
2432 sysfs_set_str(sra
, NULL
, "reshape_direction",
2433 info
->delta_disks
< 0 ? "backwards" : "forwards");
2434 if (info
->delta_disks
< 0 &&
2435 info
->space_after
< min
) {
2436 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2437 (sra
->component_size
-
2440 pr_err("cannot reduce component size\n");
2444 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
, data_offset
,
2447 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2448 cont_err("supported on this kernel\n");
2454 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2456 if (!err
&& sysfs_set_num(sra
, NULL
, "layout", reshape
->after
.layout
) < 0)
2458 if (!err
&& sysfs_set_num(sra
, NULL
, "raid_disks",
2459 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2461 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2464 pr_err("Cannot set array shape for %s\n",
2467 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2468 cont_err(" Bitmap must be removed before shape can be changed\n");
2478 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2480 struct mdinfo
*sra
, *sd
;
2481 /* Initialisation to silence compiler warning */
2482 unsigned long long min_space_before
= 0, min_space_after
= 0;
2485 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2488 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2491 struct supertype
*st2
;
2492 struct mdinfo info2
;
2494 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2496 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2497 dfd
= dev_open(dn
, O_RDONLY
);
2500 st2
= dup_super(st
);
2501 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2507 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2508 st2
->ss
->free_super(st2
);
2511 min_space_before
> info2
.space_before
)
2512 min_space_before
= info2
.space_before
;
2514 min_space_after
> info2
.space_after
)
2515 min_space_after
= info2
.space_after
;
2518 if (sd
== NULL
&& !first
) {
2519 info
->space_after
= min_space_after
;
2520 info
->space_before
= min_space_before
;
2525 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2526 struct mdinfo
*info
,
2527 int disks
, unsigned long long blocks
)
2529 /* Check that the internal stripe cache is
2530 * large enough, or it won't work.
2531 * It must hold at least 4 stripes of the larger
2534 unsigned long cache
;
2535 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2536 cache
*= 4; /* 4 stripes minimum */
2537 cache
/= 512; /* convert to sectors */
2538 /* make sure there is room for 'blocks' with a bit to spare */
2539 if (cache
< 16 + blocks
/ disks
)
2540 cache
= 16 + blocks
/ disks
;
2541 cache
/= (4096/512); /* Convert from sectors to pages */
2543 if (sra
->cache_size
< cache
)
2544 subarray_set_num(container
, sra
, "stripe_cache_size",
2548 static int impose_reshape(struct mdinfo
*sra
,
2549 struct mdinfo
*info
,
2550 struct supertype
*st
,
2553 char *devname
, char *container
,
2554 struct reshape
*reshape
)
2556 struct mdu_array_info_s array
;
2558 sra
->new_chunk
= info
->new_chunk
;
2561 /* for external metadata checkpoint saved by mdmon can be lost
2562 * or missed /due to e.g. crash/. Check if md is not during
2563 * restart farther than metadata points to.
2564 * If so, this means metadata information is obsolete.
2566 if (st
->ss
->external
)
2567 verify_reshape_position(info
, reshape
->level
);
2568 sra
->reshape_progress
= info
->reshape_progress
;
2570 sra
->reshape_progress
= 0;
2571 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2572 /* start from the end of the new array */
2573 sra
->reshape_progress
= (sra
->component_size
2574 * reshape
->after
.data_disks
);
2577 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2578 if (info
->array
.chunk_size
== info
->new_chunk
&&
2579 reshape
->before
.layout
== reshape
->after
.layout
&&
2580 st
->ss
->external
== 0) {
2581 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2582 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2584 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2587 pr_err("Cannot set device shape for %s: %s\n",
2588 devname
, strerror(errno
));
2591 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2592 cont_err("Bitmap must be removed before shape can be changed\n");
2596 } else if (!restart
) {
2597 /* set them all just in case some old 'new_*' value
2598 * persists from some earlier problem.
2601 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2603 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2604 reshape
->after
.layout
) < 0)
2606 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2607 reshape
->after
.data_disks
+
2608 reshape
->parity
) < 0)
2611 pr_err("Cannot set device shape for %s\n",
2615 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2616 cont_err("Bitmap must be removed before shape can be changed\n");
2625 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2628 struct mdu_array_info_s array
;
2630 sysfs_init(&info
, fd
, NULL
);
2632 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2634 (array
.level
>= 4 && array
.level
<= 6)) {
2635 /* To convert to RAID0 we need to fail and
2636 * remove any non-data devices. */
2639 int data_disks
= array
.raid_disks
- 1;
2640 if (array
.level
== 6)
2642 if (array
.level
== 5 &&
2643 array
.layout
!= ALGORITHM_PARITY_N
)
2645 if (array
.level
== 6 &&
2646 array
.layout
!= ALGORITHM_PARITY_N_6
)
2648 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2649 /* First remove any spares so no recovery starts */
2650 for (d
= 0, found
= 0;
2651 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2653 mdu_disk_info_t disk
;
2655 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2657 if (disk
.major
== 0 && disk
.minor
== 0)
2660 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2661 && disk
.raid_disk
< data_disks
)
2664 ioctl(fd
, HOT_REMOVE_DISK
,
2665 makedev(disk
.major
, disk
.minor
));
2667 /* Now fail anything left */
2668 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2669 for (d
= 0, found
= 0;
2670 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2673 mdu_disk_info_t disk
;
2675 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2677 if (disk
.major
== 0 && disk
.minor
== 0)
2680 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2681 && disk
.raid_disk
< data_disks
)
2684 ioctl(fd
, SET_DISK_FAULTY
,
2685 makedev(disk
.major
, disk
.minor
));
2687 while (ioctl(fd
, HOT_REMOVE_DISK
,
2688 makedev(disk
.major
, disk
.minor
)) < 0
2695 c
= map_num(pers
, level
);
2697 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2700 pr_err("%s: could not set level to %s\n",
2703 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2704 cont_err("Bitmap must be removed before level can be changed\n");
2708 pr_err("level of %s changed to %s\n",
2715 static void catch_term(int sig
)
2720 static int continue_via_systemd(char *devnm
)
2722 int skipped
, i
, pid
, status
;
2724 /* In a systemd/udev world, it is best to get systemd to
2725 * run "mdadm --grow --continue" rather than running in the
2730 /* FIXME yuk. CLOSE_EXEC?? */
2732 for (i
= 3; skipped
< 20; i
++)
2738 /* Don't want to see error messages from
2739 * systemctl. If the service doesn't exist,
2740 * we fork ourselves.
2743 open("/dev/null", O_WRONLY
);
2744 snprintf(pathbuf
, sizeof(pathbuf
), "mdadm-grow-continue@%s.service",
2746 status
= execl("/usr/bin/systemctl", "systemctl",
2749 status
= execl("/bin/systemctl", "systemctl", "start",
2752 case -1: /* Just do it ourselves. */
2754 default: /* parent - good */
2755 pid
= wait(&status
);
2756 if (pid
>= 0 && status
== 0)
2762 static int reshape_array(char *container
, int fd
, char *devname
,
2763 struct supertype
*st
, struct mdinfo
*info
,
2764 int force
, struct mddev_dev
*devlist
,
2765 unsigned long long data_offset
,
2766 char *backup_file
, int verbose
, int forked
,
2767 int restart
, int freeze_reshape
)
2769 struct reshape reshape
;
2772 int orig_level
= UnSet
;
2776 struct mdu_array_info_s array
;
2779 struct mddev_dev
*dv
;
2783 unsigned long long *offsets
= NULL
;
2787 unsigned long blocks
;
2788 unsigned long long array_size
;
2790 struct mdinfo
*sra
= NULL
;
2793 /* when reshaping a RAID0, the component_size might be zero.
2794 * So try to fix that up.
2796 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2797 dprintf("Cannot get array information.\n");
2800 if (array
.level
== 0 && info
->component_size
== 0) {
2801 get_dev_size(fd
, NULL
, &array_size
);
2802 info
->component_size
= array_size
/ array
.raid_disks
;
2805 if (array
.level
== 10)
2806 /* Need space_after info */
2807 get_space_after(fd
, st
, info
);
2809 if (info
->reshape_active
) {
2810 int new_level
= info
->new_level
;
2811 info
->new_level
= UnSet
;
2812 if (info
->delta_disks
> 0)
2813 info
->array
.raid_disks
-= info
->delta_disks
;
2814 msg
= analyse_change(devname
, info
, &reshape
);
2815 info
->new_level
= new_level
;
2816 if (info
->delta_disks
> 0)
2817 info
->array
.raid_disks
+= info
->delta_disks
;
2819 /* Make sure the array isn't read-only */
2820 ioctl(fd
, RESTART_ARRAY_RW
, 0);
2822 msg
= analyse_change(devname
, info
, &reshape
);
2824 /* if msg == "", error has already been printed */
2826 pr_err("%s\n", msg
);
2830 (reshape
.level
!= info
->array
.level
||
2831 reshape
.before
.layout
!= info
->array
.layout
||
2832 reshape
.before
.data_disks
+ reshape
.parity
2833 != info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
2834 pr_err("reshape info is not in native format - cannot continue.\n");
2838 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
2839 !((sysfs_get_str(info
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0) &&
2840 (strncmp(buf
, "reshape", 7) == 0))) {
2841 /* When reshape is restarted from '0', very begin of array
2842 * it is possible that for external metadata reshape and array
2843 * configuration doesn't happen.
2844 * Check if md has the same opinion, and reshape is restarted
2845 * from 0. If so, this is regular reshape start after reshape
2846 * switch in metadata to next array only.
2848 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
2849 (info
->reshape_progress
== 0))
2853 /* reshape already started. just skip to monitoring the reshape */
2854 if (reshape
.backup_blocks
== 0)
2856 if (restart
& RESHAPE_NO_BACKUP
)
2859 /* Need 'sra' down at 'started:' */
2860 sra
= sysfs_read(fd
, NULL
,
2861 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
2864 pr_err("%s: Cannot get array details from sysfs\n",
2870 backup_file
= locate_backup(sra
->sys_name
);
2874 /* The container is frozen but the array may not be.
2875 * So freeze the array so spares don't get put to the wrong use
2876 * FIXME there should probably be a cleaner separation between
2877 * freeze_array and freeze_container.
2879 sysfs_freeze_array(info
);
2880 /* Check we have enough spares to not be degraded */
2882 for (dv
= devlist
; dv
; dv
=dv
->next
)
2884 spares_needed
= max(reshape
.before
.data_disks
,
2885 reshape
.after
.data_disks
)
2886 + reshape
.parity
- array
.raid_disks
;
2889 info
->new_level
> 1 && info
->array
.level
> 1 &&
2890 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
2891 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2892 " Use --force to over-ride this check.\n",
2894 spares_needed
== 1 ? "" : "s",
2895 info
->array
.spare_disks
+ added_disks
);
2898 /* Check we have enough spares to not fail */
2899 spares_needed
= max(reshape
.before
.data_disks
,
2900 reshape
.after
.data_disks
)
2902 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
2903 spares_needed
> info
->array
.spare_disks
+added_disks
) {
2904 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2906 spares_needed
== 1 ? "" : "s",
2907 info
->array
.spare_disks
+ added_disks
);
2911 if (reshape
.level
!= array
.level
) {
2912 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
2915 info
->new_layout
= UnSet
; /* after level change,
2916 * layout is meaningless */
2917 orig_level
= array
.level
;
2918 sysfs_freeze_array(info
);
2920 if (reshape
.level
> 0 && st
->ss
->external
) {
2921 /* make sure mdmon is aware of the new level */
2922 if (mdmon_running(container
))
2923 flush_mdmon(container
);
2925 if (!mdmon_running(container
))
2926 start_mdmon(container
);
2927 ping_monitor(container
);
2928 if (mdmon_running(container
) &&
2929 st
->update_tail
== NULL
)
2930 st
->update_tail
= &st
->updates
;
2933 /* ->reshape_super might have chosen some spares from the
2934 * container that it wants to be part of the new array.
2935 * We can collect them with ->container_content and give
2936 * them to the kernel.
2938 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
2939 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2940 struct mdinfo
*info2
=
2941 st
->ss
->container_content(st
, subarray
);
2945 sysfs_init(info2
, fd
, st
->devnm
);
2946 /* When increasing number of devices, we need to set
2947 * new raid_disks before adding these, or they might
2950 if (reshape
.backup_blocks
&&
2951 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
2952 subarray_set_num(container
, info2
, "raid_disks",
2953 reshape
.after
.data_disks
+
2955 for (d
= info2
->devs
; d
; d
= d
->next
) {
2956 if (d
->disk
.state
== 0 &&
2957 d
->disk
.raid_disk
>= 0) {
2958 /* This is a spare that wants to
2959 * be part of the array.
2961 add_disk(fd
, st
, info2
, d
);
2967 /* We might have been given some devices to add to the
2968 * array. Now that the array has been changed to the right
2969 * level and frozen, we can safely add them.
2972 Manage_subdevs(devname
, fd
, devlist
, verbose
,
2975 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
2976 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
2977 if (reshape
.backup_blocks
== 0) {
2978 /* No restriping needed, but we might need to impose
2979 * some more changes: layout, raid_disks, chunk_size
2981 /* read current array info */
2982 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2983 dprintf("Cannot get array information.\n");
2986 /* compare current array info with new values and if
2987 * it is different update them to new */
2988 if (info
->new_layout
!= UnSet
&&
2989 info
->new_layout
!= array
.layout
) {
2990 array
.layout
= info
->new_layout
;
2991 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2992 pr_err("failed to set new layout\n");
2994 } else if (verbose
>= 0)
2995 printf("layout for %s set to %d\n",
2996 devname
, array
.layout
);
2998 if (info
->delta_disks
!= UnSet
&&
2999 info
->delta_disks
!= 0 &&
3000 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
3001 array
.raid_disks
+= info
->delta_disks
;
3002 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
3003 pr_err("failed to set raid disks\n");
3005 } else if (verbose
>= 0) {
3006 printf("raid_disks for %s set to %d\n",
3007 devname
, array
.raid_disks
);
3010 if (info
->new_chunk
!= 0 &&
3011 info
->new_chunk
!= array
.chunk_size
) {
3012 if (sysfs_set_num(info
, NULL
,
3013 "chunk_size", info
->new_chunk
) != 0) {
3014 pr_err("failed to set chunk size\n");
3016 } else if (verbose
>= 0)
3017 printf("chunk size for %s set to %d\n",
3018 devname
, array
.chunk_size
);
3025 * There are three possibilities.
3026 * 1/ The array will shrink.
3027 * We need to ensure the reshape will pause before reaching
3028 * the 'critical section'. We also need to fork and wait for
3029 * that to happen. When it does we
3030 * suspend/backup/complete/unfreeze
3032 * 2/ The array will not change size.
3033 * This requires that we keep a backup of a sliding window
3034 * so that we can restore data after a crash. So we need
3035 * to fork and monitor progress.
3036 * In future we will allow the data_offset to change, so
3037 * a sliding backup becomes unnecessary.
3039 * 3/ The array will grow. This is relatively easy.
3040 * However the kernel's restripe routines will cheerfully
3041 * overwrite some early data before it is safe. So we
3042 * need to make a backup of the early parts of the array
3043 * and be ready to restore it if rebuild aborts very early.
3044 * For externally managed metadata, we still need a forked
3045 * child to monitor the reshape and suspend IO over the region
3046 * that is being reshaped.
3048 * We backup data by writing it to one spare, or to a
3049 * file which was given on command line.
3051 * In each case, we first make sure that storage is available
3052 * for the required backup.
3054 * - request the shape change.
3055 * - fork to handle backup etc.
3057 /* Check that we can hold all the data */
3058 get_dev_size(fd
, NULL
, &array_size
);
3059 if (reshape
.new_size
< (array_size
/512)) {
3060 pr_err("this change will reduce the size of the array.\n"
3061 " use --grow --array-size first to truncate array.\n"
3062 " e.g. mdadm --grow %s --array-size %llu\n",
3063 devname
, reshape
.new_size
/2);
3067 if (array
.level
== 10) {
3068 /* Reshaping RAID10 does not require any data backup by
3069 * user-space. Instead it requires that the data_offset
3070 * is changed to avoid the need for backup.
3071 * So this is handled very separately
3074 /* Nothing to do. */
3076 return raid10_reshape(container
, fd
, devname
, st
, info
,
3077 &reshape
, data_offset
,
3080 sra
= sysfs_read(fd
, NULL
,
3081 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3084 pr_err("%s: Cannot get array details from sysfs\n",
3090 switch(set_new_data_offset(sra
, st
, devname
,
3091 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3093 reshape
.min_offset_change
, 1)) {
3097 /* Updated data_offset, so it's easy now */
3098 update_cache_size(container
, sra
, info
,
3099 min(reshape
.before
.data_disks
,
3100 reshape
.after
.data_disks
),
3101 reshape
.backup_blocks
);
3103 /* Right, everything seems fine. Let's kick things off.
3107 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3108 devname
, container
, &reshape
) < 0)
3110 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3112 if (errno
!= EINVAL
) {
3113 pr_err("Failed to initiate reshape!\n");
3116 /* revert data_offset and try the old way */
3117 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3118 sysfs_set_num(sra
, sd
, "new_offset",
3120 sysfs_set_str(sra
, NULL
, "reshape_direction",
3125 if (info
->new_level
== reshape
.level
)
3127 /* need to adjust level when reshape completes */
3129 case -1: /* ignore error, but don't wait */
3131 default: /* parent */
3139 fd
= open_dev(sra
->sys_name
);
3141 impose_level(fd
, info
->new_level
, devname
, verbose
);
3143 case 1: /* Couldn't set data_offset, try the old way */
3144 if (data_offset
!= INVALID_SECTORS
) {
3145 pr_err("Cannot update data_offset on this array\n");
3152 /* Decide how many blocks (sectors) for a reshape
3153 * unit. The number we have so far is just a minimum
3155 blocks
= reshape
.backup_blocks
;
3156 if (reshape
.before
.data_disks
==
3157 reshape
.after
.data_disks
) {
3158 /* Make 'blocks' bigger for better throughput, but
3159 * not so big that we reject it below.
3160 * Try for 16 megabytes
3162 while (blocks
* 32 < sra
->component_size
&&
3166 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3168 if (blocks
>= sra
->component_size
/2) {
3169 pr_err("%s: Something wrong - reshape aborted\n",
3174 /* Now we need to open all these devices so we can read/write.
3176 nrdisks
= max(reshape
.before
.data_disks
,
3177 reshape
.after
.data_disks
) + reshape
.parity
3178 + sra
->array
.spare_disks
;
3179 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3180 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3182 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3183 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
3184 nrdisks
, blocks
, backup_file
,
3189 if ((st
->ss
->manage_reshape
== NULL
) ||
3190 (st
->ss
->recover_backup
== NULL
)) {
3191 if (backup_file
== NULL
) {
3192 if (reshape
.after
.data_disks
<=
3193 reshape
.before
.data_disks
) {
3194 pr_err("%s: Cannot grow - need backup-file\n",
3196 pr_err(" Please provide one with \"--backup=...\"\n");
3198 } else if (d
== odisks
) {
3199 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3203 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3205 fdlist
+d
, offsets
+d
,
3214 update_cache_size(container
, sra
, info
,
3215 min(reshape
.before
.data_disks
, reshape
.after
.data_disks
),
3218 /* Right, everything seems fine. Let's kick things off.
3219 * If only changing raid_disks, use ioctl, else use
3224 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3225 devname
, container
, &reshape
) < 0)
3228 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3229 reshape
.after
.data_disks
);
3231 pr_err("Cannot %s reshape for %s\n",
3232 restart
? "continue" : "start",
3237 sysfs_set_str(sra
, NULL
, "array_state", "active");
3238 if (freeze_reshape
) {
3242 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3243 sra
->reshape_progress
);
3247 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3248 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3255 /* Now we just need to kick off the reshape and watch, while
3256 * handling backups of the data...
3257 * This is all done by a forked background process.
3259 switch(forked
? 0 : fork()) {
3261 pr_err("Cannot run child to monitor reshape: %s\n",
3275 /* If another array on the same devices is busy, the
3276 * reshape will wait for them. This would mean that
3277 * the first section that we suspend will stay suspended
3278 * for a long time. So check on that possibility
3279 * by looking for "DELAYED" in /proc/mdstat, and if found,
3283 struct mdstat_ent
*mds
, *m
;
3285 mds
= mdstat_read(1, 0);
3286 for (m
= mds
; m
; m
= m
->next
)
3287 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3289 m
->percent
== RESYNC_DELAYED
)
3292 /* Haven't started the reshape thread
3299 if (delayed
== 1 && get_linux_version() < 3007000) {
3300 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3301 " You might experience problems until other reshapes complete.\n");
3305 mdstat_wait(30 - (delayed
-1) * 25);
3309 if (check_env("MDADM_GROW_VERIFY"))
3310 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3313 mlockall(MCL_FUTURE
);
3315 signal(SIGTERM
, catch_term
);
3317 if (st
->ss
->external
) {
3318 /* metadata handler takes it from here */
3319 done
= st
->ss
->manage_reshape(
3320 fd
, sra
, &reshape
, st
, blocks
,
3322 d
- odisks
, fdlist
+odisks
,
3325 done
= child_monitor(
3326 fd
, sra
, &reshape
, st
, blocks
,
3328 d
- odisks
, fdlist
+odisks
,
3334 if (backup_file
&& done
) {
3336 bul
= make_backup(sra
->sys_name
);
3339 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3347 unlink(backup_file
);
3354 if (!st
->ss
->external
&&
3355 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
3356 && info
->custom_array_size
) &&
3357 info
->new_level
== reshape
.level
&&
3359 /* no need to wait for the reshape to finish as
3360 * there is nothing more to do.
3367 if (st
->ss
->external
) {
3368 /* Re-load the metadata as much could have changed */
3369 int cfd
= open_dev(st
->container_devnm
);
3371 flush_mdmon(container
);
3372 st
->ss
->free_super(st
);
3373 st
->ss
->load_container(st
, cfd
, container
);
3378 /* set new array size if required customer_array_size is used
3381 if (reshape
.before
.data_disks
!=
3382 reshape
.after
.data_disks
&&
3383 info
->custom_array_size
)
3384 set_array_size(st
, info
, info
->text_version
);
3386 if (info
->new_level
!= reshape
.level
) {
3388 fd
= open(devname
, O_RDONLY
);
3389 impose_level(fd
, info
->new_level
, devname
, verbose
);
3391 if (info
->new_level
== 0)
3392 st
->update_tail
= NULL
;
3404 if (orig_level
!= UnSet
&& sra
) {
3405 c
= map_num(pers
, orig_level
);
3406 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3407 pr_err("aborting level change\n");
3415 /* mdfd handle is passed to be closed in child process (after fork).
3417 int reshape_container(char *container
, char *devname
,
3419 struct supertype
*st
,
3420 struct mdinfo
*info
,
3422 char *backup_file
, int verbose
,
3423 int forked
, int restart
, int freeze_reshape
)
3425 struct mdinfo
*cc
= NULL
;
3427 char last_devnm
[32] = "";
3429 /* component_size is not meaningful for a container,
3430 * so pass '0' meaning 'no change'
3433 reshape_super(st
, 0, info
->new_level
,
3434 info
->new_layout
, info
->new_chunk
,
3435 info
->array
.raid_disks
, info
->delta_disks
,
3436 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3444 /* ping monitor to be sure that update is on disk
3446 ping_monitor(container
);
3448 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3449 if (continue_via_systemd(container
))
3452 switch (forked
? 0 : fork()) {
3453 case -1: /* error */
3454 perror("Cannot fork to complete reshape\n");
3457 default: /* parent */
3458 if (!freeze_reshape
)
3459 printf("%s: multi-array reshape continues in background\n", Name
);
3466 /* close unused handle in child process
3472 /* For each member array with reshape_active,
3473 * we need to perform the reshape.
3474 * We pick the first array that needs reshaping and
3475 * reshape it. reshape_array() will re-read the metadata
3476 * so the next time through a different array should be
3477 * ready for reshape.
3478 * It is possible that the 'different' array will not
3479 * be assembled yet. In that case we simple exit.
3480 * When it is assembled, the mdadm which assembles it
3481 * will take over the reshape.
3483 struct mdinfo
*content
;
3485 struct mdstat_ent
*mdstat
;
3491 cc
= st
->ss
->container_content(st
, NULL
);
3493 for (content
= cc
; content
; content
= content
->next
) {
3495 if (!content
->reshape_active
)
3498 subarray
= strchr(content
->text_version
+1, '/')+1;
3499 mdstat
= mdstat_by_subdev(subarray
, container
);
3502 if (mdstat
->active
== 0) {
3503 pr_err("Skipping inactive array %s.\n",
3505 free_mdstat(mdstat
);
3514 devid
= devnm2devid(mdstat
->devnm
);
3515 adev
= map_dev(major(devid
), minor(devid
), 0);
3517 adev
= content
->text_version
;
3519 fd
= open_dev(mdstat
->devnm
);
3521 pr_err("Device %s cannot be opened for reshape.\n", adev
);
3525 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3526 /* Do not allow for multiple reshape_array() calls for
3528 * It can happen when reshape_array() returns without
3529 * error, when reshape is not finished (wrong reshape
3530 * starting/continuation conditions). Mdmon doesn't
3531 * switch to next array in container and reentry
3532 * conditions for the same array occur.
3533 * This is possibly interim until the behaviour of
3534 * reshape_array is resolved().
3536 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3540 strcpy(last_devnm
, mdstat
->devnm
);
3542 sysfs_init(content
, fd
, mdstat
->devnm
);
3544 if (mdmon_running(container
))
3545 flush_mdmon(container
);
3547 rv
= reshape_array(container
, fd
, adev
, st
,
3548 content
, force
, NULL
, INVALID_SECTORS
,
3549 backup_file
, verbose
, 1, restart
,
3553 if (freeze_reshape
) {
3562 if (mdmon_running(container
))
3563 flush_mdmon(container
);
3572 * We run a child process in the background which performs the following
3574 * - wait for resync to reach a certain point
3575 * - suspend io to the following section
3576 * - backup that section
3577 * - allow resync to proceed further
3579 * - discard the backup.
3581 * When are combined in slightly different ways in the three cases.
3583 * - suspend/backup/allow/wait/resume/discard
3585 * - allow/wait/suspend/backup/allow/wait/resume/discard
3587 * - wait/resume/discard/suspend/backup/allow
3589 * suspend/backup/allow always come together
3590 * wait/resume/discard do too.
3591 * For the same-size case we have two backups to improve flow.
3595 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3596 unsigned long long backup_point
,
3597 unsigned long long wait_point
,
3598 unsigned long long *suspend_point
,
3599 unsigned long long *reshape_completed
, int *frozen
)
3601 /* This function is called repeatedly by the reshape manager.
3602 * It determines how much progress can safely be made and allows
3604 * - 'info' identifies the array and particularly records in
3605 * ->reshape_progress the metadata's knowledge of progress
3606 * This is a sector offset from the start of the array
3607 * of the next array block to be relocated. This number
3608 * may increase from 0 or decrease from array_size, depending
3609 * on the type of reshape that is happening.
3610 * Note that in contrast, 'sync_completed' is a block count of the
3611 * reshape so far. It gives the distance between the start point
3612 * (head or tail of device) and the next place that data will be
3613 * written. It always increases.
3614 * - 'reshape' is the structure created by analyse_change
3615 * - 'backup_point' shows how much the metadata manager has backed-up
3616 * data. For reshapes with increasing progress, it is the next address
3617 * to be backed up, previous addresses have been backed-up. For
3618 * decreasing progress, it is the earliest address that has been
3619 * backed up - later address are also backed up.
3620 * So addresses between reshape_progress and backup_point are
3621 * backed up providing those are in the 'correct' order.
3622 * - 'wait_point' is an array address. When reshape_completed
3623 * passes this point, progress_reshape should return. It might
3624 * return earlier if it determines that ->reshape_progress needs
3625 * to be updated or further backup is needed.
3626 * - suspend_point is maintained by progress_reshape and the caller
3627 * should not touch it except to initialise to zero.
3628 * It is an array address and it only increases in 2.6.37 and earlier.
3629 * This makes it difficult to handle reducing reshapes with
3630 * external metadata.
3631 * However: it is similar to backup_point in that it records the
3632 * other end of a suspended region from reshape_progress.
3633 * it is moved to extend the region that is safe to backup and/or
3635 * - reshape_completed is read from sysfs and returned. The caller
3636 * should copy this into ->reshape_progress when it has reason to
3637 * believe that the metadata knows this, and any backup outside this
3641 * 1 if more data from backup_point - but only as far as suspend_point,
3642 * should be backed up
3643 * 0 if things are progressing smoothly
3644 * -1 if the reshape is finished because it is all done,
3645 * -2 if the reshape is finished due to an error.
3648 int advancing
= (reshape
->after
.data_disks
3649 >= reshape
->before
.data_disks
);
3650 unsigned long long need_backup
; /* All data between start of array and
3651 * here will at some point need to
3654 unsigned long long read_offset
, write_offset
;
3655 unsigned long long write_range
;
3656 unsigned long long max_progress
, target
, completed
;
3657 unsigned long long array_size
= (info
->component_size
3658 * reshape
->before
.data_disks
);
3662 /* First, we unsuspend any region that is now known to be safe.
3663 * If suspend_point is on the 'wrong' side of reshape_progress, then
3664 * we don't have or need suspension at the moment. This is true for
3665 * native metadata when we don't need to back-up.
3668 if (info
->reshape_progress
<= *suspend_point
)
3669 sysfs_set_num(info
, NULL
, "suspend_lo",
3670 info
->reshape_progress
);
3672 /* Note: this won't work in 2.6.37 and before.
3673 * Something somewhere should make sure we don't need it!
3675 if (info
->reshape_progress
>= *suspend_point
)
3676 sysfs_set_num(info
, NULL
, "suspend_hi",
3677 info
->reshape_progress
);
3680 /* Now work out how far it is safe to progress.
3681 * If the read_offset for ->reshape_progress is less than
3682 * 'blocks' beyond the write_offset, we can only progress as far
3684 * Otherwise we can progress until the write_offset for the new location
3685 * reaches (within 'blocks' of) the read_offset at the current location.
3686 * However that region must be suspended unless we are using native
3688 * If we need to suspend more, we limit it to 128M per device, which is
3689 * rather arbitrary and should be some time-based calculation.
3691 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3692 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3693 write_range
= info
->new_chunk
/512;
3694 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3695 need_backup
= array_size
;
3697 need_backup
= reshape
->backup_blocks
;
3699 if (read_offset
< write_offset
+ write_range
)
3700 max_progress
= backup_point
;
3704 reshape
->after
.data_disks
;
3706 if (read_offset
> write_offset
- write_range
)
3707 /* Can only progress as far as has been backed up,
3708 * which must be suspended */
3709 max_progress
= backup_point
;
3710 else if (info
->reshape_progress
<= need_backup
)
3711 max_progress
= backup_point
;
3713 if (info
->array
.major_version
>= 0)
3714 /* Can progress until backup is needed */
3715 max_progress
= need_backup
;
3717 /* Can progress until metadata update is required */
3720 reshape
->after
.data_disks
;
3721 /* but data must be suspended */
3722 if (max_progress
< *suspend_point
)
3723 max_progress
= *suspend_point
;
3728 /* We know it is safe to progress to 'max_progress' providing
3729 * it is suspended or we are using native metadata.
3730 * Consider extending suspend_point 128M per device if it
3731 * is less than 64M per device beyond reshape_progress.
3732 * But always do a multiple of 'blocks'
3733 * FIXME this is too big - it takes to long to complete
3736 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3737 reshape
->after
.data_disks
);
3738 target
/= reshape
->backup_blocks
;
3741 target
*= reshape
->backup_blocks
;
3743 /* For externally managed metadata we always need to suspend IO to
3744 * the area being reshaped so we regularly push suspend_point forward.
3745 * For native metadata we only need the suspend if we are going to do
3749 if ((need_backup
> info
->reshape_progress
3750 || info
->array
.major_version
< 0) &&
3751 *suspend_point
< info
->reshape_progress
+ target
) {
3752 if (need_backup
< *suspend_point
+ 2 * target
)
3753 *suspend_point
= need_backup
;
3754 else if (*suspend_point
+ 2 * target
< array_size
)
3755 *suspend_point
+= 2 * target
;
3757 *suspend_point
= array_size
;
3758 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
3759 if (max_progress
> *suspend_point
)
3760 max_progress
= *suspend_point
;
3763 if (info
->array
.major_version
>= 0) {
3764 /* Only need to suspend when about to backup */
3765 if (info
->reshape_progress
< need_backup
* 2 &&
3766 *suspend_point
> 0) {
3768 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
3769 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
3772 /* Need to suspend continually */
3773 if (info
->reshape_progress
< *suspend_point
)
3774 *suspend_point
= info
->reshape_progress
;
3775 if (*suspend_point
+ target
< info
->reshape_progress
)
3776 /* No need to move suspend region yet */;
3778 if (*suspend_point
>= 2 * target
)
3779 *suspend_point
-= 2 * target
;
3782 sysfs_set_num(info
, NULL
, "suspend_lo",
3785 if (max_progress
< *suspend_point
)
3786 max_progress
= *suspend_point
;
3790 /* now set sync_max to allow that progress. sync_max, like
3791 * sync_completed is a count of sectors written per device, so
3792 * we find the difference between max_progress and the start point,
3793 * and divide that by after.data_disks to get a sync_max
3795 * At the same time we convert wait_point to a similar number
3796 * for comparing against sync_completed.
3798 /* scale down max_progress to per_disk */
3799 max_progress
/= reshape
->after
.data_disks
;
3800 /* Round to chunk size as some kernels give an erroneously high number */
3801 max_progress
/= info
->new_chunk
/512;
3802 max_progress
*= info
->new_chunk
/512;
3803 /* And round to old chunk size as the kernel wants that */
3804 max_progress
/= info
->array
.chunk_size
/512;
3805 max_progress
*= info
->array
.chunk_size
/512;
3806 /* Limit progress to the whole device */
3807 if (max_progress
> info
->component_size
)
3808 max_progress
= info
->component_size
;
3809 wait_point
/= reshape
->after
.data_disks
;
3811 /* switch from 'device offset' to 'processed block count' */
3812 max_progress
= info
->component_size
- max_progress
;
3813 wait_point
= info
->component_size
- wait_point
;
3817 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
3819 /* Now wait. If we have already reached the point that we were
3820 * asked to wait to, don't wait at all, else wait for any change.
3821 * We need to select on 'sync_completed' as that is the place that
3822 * notifications happen, but we are really interested in
3823 * 'reshape_position'
3825 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
3827 goto check_progress
;
3829 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3830 goto check_progress
;
3832 while (completed
< max_progress
&& completed
< wait_point
) {
3833 /* Check that sync_action is still 'reshape' to avoid
3834 * waiting forever on a dead array
3837 if (sysfs_get_str(info
, NULL
, "sync_action",
3839 strncmp(action
, "reshape", 7) != 0)
3841 /* Some kernels reset 'sync_completed' to zero
3842 * before setting 'sync_action' to 'idle'.
3843 * So we need these extra tests.
3845 if (completed
== 0 && advancing
3846 && strncmp(action
, "idle", 4) == 0
3847 && info
->reshape_progress
> 0)
3849 if (completed
== 0 && !advancing
3850 && strncmp(action
, "idle", 4) == 0
3851 && info
->reshape_progress
< (info
->component_size
3852 * reshape
->after
.data_disks
))
3854 sysfs_wait(fd
, NULL
);
3855 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3856 goto check_progress
;
3858 /* Some kernels reset 'sync_completed' to zero,
3859 * we need to have real point we are in md.
3860 * So in that case, read 'reshape_position' from sysfs.
3862 if (completed
== 0) {
3863 unsigned long long reshapep
;
3865 if (sysfs_get_str(info
, NULL
, "sync_action",
3867 strncmp(action
, "idle", 4) == 0 &&
3868 sysfs_get_ll(info
, NULL
,
3869 "reshape_position", &reshapep
) == 0)
3870 *reshape_completed
= reshapep
;
3872 /* some kernels can give an incorrectly high
3873 * 'completed' number, so round down */
3874 completed
/= (info
->new_chunk
/512);
3875 completed
*= (info
->new_chunk
/512);
3876 /* Convert 'completed' back in to a 'progress' number */
3877 completed
*= reshape
->after
.data_disks
;
3879 completed
= (info
->component_size
3880 * reshape
->after
.data_disks
3882 *reshape_completed
= completed
;
3887 /* We return the need_backup flag. Caller will decide
3888 * how much - a multiple of ->backup_blocks up to *suspend_point
3891 return need_backup
> info
->reshape_progress
;
3893 return need_backup
>= info
->reshape_progress
;
3896 /* if we couldn't read a number from sync_completed, then
3897 * either the reshape did complete, or it aborted.
3898 * We can tell which by checking for 'none' in reshape_position.
3899 * If it did abort, then it might immediately restart if it
3900 * it was just a device failure that leaves us degraded but
3903 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
3904 || strncmp(buf
, "none", 4) != 0) {
3905 /* The abort might only be temporary. Wait up to 10
3906 * seconds for fd to contain a valid number again.
3910 unsigned long long new_sync_max
;
3911 while (fd
>= 0 && rv
< 0 && wait
> 0) {
3912 if (sysfs_wait(fd
, &wait
) != 1)
3914 switch (sysfs_fd_get_ll(fd
, &completed
)) {
3916 /* all good again */
3918 /* If "sync_max" is no longer max_progress
3919 * we need to freeze things
3921 sysfs_get_ll(info
, NULL
, "sync_max", &new_sync_max
);
3922 *frozen
= (new_sync_max
!= max_progress
);
3924 case -2: /* read error - abort */
3931 return rv
; /* abort */
3933 /* Maybe racing with array shutdown - check state */
3936 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
3937 || strncmp(buf
, "inactive", 8) == 0
3938 || strncmp(buf
, "clear",5) == 0)
3939 return -2; /* abort */
3940 return -1; /* complete */
3944 /* FIXME return status is never checked */
3945 static int grow_backup(struct mdinfo
*sra
,
3946 unsigned long long offset
, /* per device */
3947 unsigned long stripes
, /* per device, in old chunks */
3948 int *sources
, unsigned long long *offsets
,
3949 int disks
, int chunk
, int level
, int layout
,
3950 int dests
, int *destfd
, unsigned long long *destoffsets
,
3951 int part
, int *degraded
,
3954 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3955 * to storage 'destfd' (offset 'destoffsets'), after first
3956 * suspending IO. Then allow resync to continue
3957 * over the suspended section.
3958 * Use part 'part' of the backup-super-block.
3963 unsigned long long ll
;
3965 //printf("offset %llu\n", offset);
3971 /* Check that array hasn't become degraded, else we might backup the wrong data */
3972 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
3973 return -1; /* FIXME this error is ignored */
3974 new_degraded
= (int)ll
;
3975 if (new_degraded
!= *degraded
) {
3976 /* check each device to ensure it is still working */
3978 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3979 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
3981 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
3983 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
3984 strstr(sbuf
, "faulty") ||
3985 strstr(sbuf
, "in_sync") == NULL
) {
3986 /* this device is dead */
3987 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
3988 if (sd
->disk
.raid_disk
>= 0 &&
3989 sources
[sd
->disk
.raid_disk
] >= 0) {
3990 close(sources
[sd
->disk
.raid_disk
]);
3991 sources
[sd
->disk
.raid_disk
] = -1;
3996 *degraded
= new_degraded
;
3999 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4000 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4002 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4003 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4006 bsb
.magic
[15] = '2';
4007 for (i
= 0; i
< dests
; i
++)
4009 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
4011 lseek64(destfd
[i
], destoffsets
[i
], 0);
4013 rv
= save_stripes(sources
, offsets
,
4014 disks
, chunk
, level
, layout
,
4016 offset
*512*odata
, stripes
* chunk
* odata
,
4021 bsb
.mtime
= __cpu_to_le64(time(0));
4022 for (i
= 0; i
< dests
; i
++) {
4023 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4025 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4026 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4027 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4028 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4031 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
4032 != destoffsets
[i
] - 4096)
4034 if (write(destfd
[i
], &bsb
, 512) != 512)
4036 if (destoffsets
[i
] > 4096) {
4037 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4038 destoffsets
[i
]+stripes
*chunk
*odata
)
4040 if (write(destfd
[i
], &bsb
, 512) != 512)
4050 /* in 2.6.30, the value reported by sync_completed can be
4051 * less that it should be by one stripe.
4052 * This only happens when reshape hits sync_max and pauses.
4053 * So allow wait_backup to either extent sync_max further
4054 * than strictly necessary, or return before the
4055 * sync has got quite as far as we would really like.
4056 * This is what 'blocks2' is for.
4057 * The various caller give appropriate values so that
4060 /* FIXME return value is often ignored */
4061 static int forget_backup(int dests
, int *destfd
,
4062 unsigned long long *destoffsets
,
4066 * Erase backup 'part' (which is 0 or 1)
4072 bsb
.arraystart2
= __cpu_to_le64(0);
4073 bsb
.length2
= __cpu_to_le64(0);
4075 bsb
.arraystart
= __cpu_to_le64(0);
4076 bsb
.length
= __cpu_to_le64(0);
4078 bsb
.mtime
= __cpu_to_le64(time(0));
4080 for (i
= 0; i
< dests
; i
++) {
4081 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4082 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4083 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4084 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4085 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4086 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4087 destoffsets
[i
]-4096)
4090 write(destfd
[i
], &bsb
, 512) != 512)
4097 static void fail(char *msg
)
4100 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4101 rv
|= (write(2, "\n", 1) != 1);
4105 static char *abuf
, *bbuf
;
4106 static unsigned long long abuflen
;
4107 static void validate(int afd
, int bfd
, unsigned long long offset
)
4109 /* check that the data in the backup against the array.
4110 * This is only used for regression testing and should not
4111 * be used while the array is active
4115 lseek64(bfd
, offset
- 4096, 0);
4116 if (read(bfd
, &bsb2
, 512) != 512)
4117 fail("cannot read bsb");
4118 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4119 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4120 fail("first csum bad");
4121 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4122 fail("magic is bad");
4123 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4124 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4125 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4126 fail("second csum bad");
4128 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4129 fail("devstart is wrong");
4132 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4134 if (abuflen
< len
) {
4138 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4139 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4141 /* just stop validating on mem-alloc failure */
4146 lseek64(bfd
, offset
, 0);
4147 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4148 //printf("len %llu\n", len);
4149 fail("read first backup failed");
4151 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4152 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4153 fail("read first from array failed");
4154 if (memcmp(bbuf
, abuf
, len
) != 0) {
4157 printf("offset=%llu len=%llu\n",
4158 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4159 for (i
=0; i
<len
; i
++)
4160 if (bbuf
[i
] != abuf
[i
]) {
4161 printf("first diff byte %d\n", i
);
4165 fail("data1 compare failed");
4169 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4171 if (abuflen
< len
) {
4175 abuf
= xmalloc(abuflen
);
4176 bbuf
= xmalloc(abuflen
);
4179 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4180 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4181 fail("read second backup failed");
4182 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4183 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4184 fail("read second from array failed");
4185 if (memcmp(bbuf
, abuf
, len
) != 0)
4186 fail("data2 compare failed");
4190 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4191 struct supertype
*st
, unsigned long blocks
,
4192 int *fds
, unsigned long long *offsets
,
4193 int dests
, int *destfd
, unsigned long long *destoffsets
)
4195 /* Monitor a reshape where backup is being performed using
4196 * 'native' mechanism - either to a backup file, or
4197 * to some space in a spare.
4201 unsigned long long speed
;
4202 unsigned long long suspend_point
, array_size
;
4203 unsigned long long backup_point
, wait_point
;
4204 unsigned long long reshape_completed
;
4206 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
4207 int part
= 0; /* The next part of the backup area to fill. It may already
4208 * be full, so we need to check */
4209 int level
= reshape
->level
;
4210 int layout
= reshape
->before
.layout
;
4211 int data
= reshape
->before
.data_disks
;
4212 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4213 int chunk
= sra
->array
.chunk_size
;
4215 unsigned long stripes
;
4219 /* set up the backup-super-block. This requires the
4220 * uuid from the array.
4222 /* Find a superblock */
4223 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4227 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4229 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4230 devfd
= dev_open(dn
, O_RDONLY
);
4233 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4239 pr_err("Cannot find a superblock\n");
4243 memset(&bsb
, 0, 512);
4244 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4245 st
->ss
->uuid_from_super(st
, uuid
);
4246 memcpy(bsb
.set_uuid
, uuid
, 16);
4247 bsb
.mtime
= __cpu_to_le64(time(0));
4248 bsb
.devstart2
= blocks
;
4250 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4251 reshape
->before
.data_disks
;
4253 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4254 /* Don't start the 'reshape' */
4256 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4257 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4258 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4262 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4263 backup_point
= sra
->reshape_progress
;
4266 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4267 backup_point
= reshape
->backup_blocks
;
4268 suspend_point
= array_size
;
4274 /* Want to return as soon the oldest backup slot can
4275 * be released as that allows us to start backing up
4276 * some more, providing suspend_point has been
4277 * advanced, which it should have.
4280 wait_point
= array_size
;
4281 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4282 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4283 __le64_to_cpu(bsb
.length
));
4284 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4285 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4286 __le64_to_cpu(bsb
.length2
));
4289 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4290 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4291 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4292 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4295 reshape_completed
= sra
->reshape_progress
;
4296 rv
= progress_reshape(sra
, reshape
,
4297 backup_point
, wait_point
,
4298 &suspend_point
, &reshape_completed
,
4300 /* external metadata would need to ping_monitor here */
4301 sra
->reshape_progress
= reshape_completed
;
4303 /* Clear any backup region that is before 'here' */
4305 if (__le64_to_cpu(bsb
.length
) > 0 &&
4306 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4307 __le64_to_cpu(bsb
.length
)))
4308 forget_backup(dests
, destfd
,
4310 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4311 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4312 __le64_to_cpu(bsb
.length2
)))
4313 forget_backup(dests
, destfd
,
4316 if (__le64_to_cpu(bsb
.length
) > 0 &&
4317 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4318 forget_backup(dests
, destfd
,
4320 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4321 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4322 forget_backup(dests
, destfd
,
4332 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4333 /* No longer need to monitor this reshape */
4334 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4340 unsigned long long offset
;
4341 unsigned long actual_stripes
;
4342 /* Need to backup some data.
4343 * If 'part' is not used and the desired
4344 * backup size is suspended, do a backup,
4345 * then consider the next part.
4347 /* Check that 'part' is unused */
4348 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4350 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4353 offset
= backup_point
/ data
;
4354 actual_stripes
= stripes
;
4356 if (offset
+ actual_stripes
* (chunk
/512) >
4357 sra
->component_size
)
4358 actual_stripes
= ((sra
->component_size
- offset
)
4360 if (offset
+ actual_stripes
* (chunk
/512) >
4364 if (offset
< actual_stripes
* (chunk
/512))
4365 actual_stripes
= offset
/ (chunk
/512);
4366 offset
-= actual_stripes
* (chunk
/512);
4367 if (offset
< suspend_point
/data
)
4370 if (actual_stripes
== 0)
4372 grow_backup(sra
, offset
, actual_stripes
,
4374 disks
, chunk
, level
, layout
,
4375 dests
, destfd
, destoffsets
,
4376 part
, °raded
, buf
);
4377 validate(afd
, destfd
[0], destoffsets
[0]);
4378 /* record where 'part' is up to */
4381 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4383 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4387 /* FIXME maybe call progress_reshape one more time instead */
4388 /* remove any remaining suspension */
4389 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4390 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4391 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4392 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4394 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4395 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4401 * If any spare contains md_back_data-1 which is recent wrt mtime,
4402 * write that data into the array and update the super blocks with
4403 * the new reshape_progress
4405 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
4406 char *backup_file
, int verbose
)
4410 unsigned long long *offsets
;
4411 unsigned long long nstripe
, ostripe
;
4414 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4415 if (info
->array
.level
== 6) odata
--; /* number of data disks */
4416 ndata
= info
->array
.raid_disks
- 1;
4417 if (info
->new_level
== 6) ndata
--;
4419 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4421 if (info
->delta_disks
<= 0)
4422 /* Didn't grow, so the backup file must have
4426 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4427 struct mdinfo dinfo
;
4430 char *devname
, namebuf
[20];
4431 unsigned long long lo
, hi
;
4433 /* This was a spare and may have some saved data on it.
4434 * Load the superblock, find and load the
4435 * backup_super_block.
4436 * If either fail, go on to next device.
4437 * If the backup contains no new info, just return
4438 * else restore data and update all superblocks
4440 if (i
== old_disks
-1) {
4441 fd
= open(backup_file
, O_RDONLY
);
4443 pr_err("backup file %s inaccessible: %s\n",
4444 backup_file
, strerror(errno
));
4447 devname
= backup_file
;
4452 if (st
->ss
->load_super(st
, fd
, NULL
))
4455 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4456 st
->ss
->free_super(st
);
4459 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4461 pr_err("Cannot seek on device %d\n", i
);
4462 continue; /* Cannot seek */
4464 sprintf(namebuf
, "device-%d", i
);
4467 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4469 pr_err("Cannot read from %s\n", devname
);
4470 continue; /* Cannot read */
4472 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4473 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4475 pr_err("No backup metadata on %s\n", devname
);
4478 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4480 pr_err("Bad backup-metadata checksum on %s\n", devname
);
4481 continue; /* bad checksum */
4483 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4484 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4486 pr_err("Bad backup-metadata checksum2 on %s\n", devname
);
4487 continue; /* Bad second checksum */
4489 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4491 pr_err("Wrong uuid on backup-metadata on %s\n", devname
);
4492 continue; /* Wrong uuid */
4495 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4496 * sometimes they aren't... So allow considerable flexability in matching, and allow
4497 * this test to be overridden by an environment variable.
4499 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
4500 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
4501 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4502 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4503 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4504 (unsigned long)info
->array
.utime
);
4506 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4507 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4508 continue; /* time stamp is too bad */
4512 if (bsb
.magic
[15] == '1') {
4513 if (bsb
.length
== 0)
4515 if (info
->delta_disks
>= 0) {
4516 /* reshape_progress is increasing */
4517 if (__le64_to_cpu(bsb
.arraystart
)
4518 + __le64_to_cpu(bsb
.length
)
4519 < info
->reshape_progress
) {
4522 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4523 continue; /* No new data here */
4526 /* reshape_progress is decreasing */
4527 if (__le64_to_cpu(bsb
.arraystart
) >=
4528 info
->reshape_progress
)
4529 goto nonew
; /* No new data here */
4532 if (bsb
.length
== 0 && bsb
.length2
== 0)
4534 if (info
->delta_disks
>= 0) {
4535 /* reshape_progress is increasing */
4536 if ((__le64_to_cpu(bsb
.arraystart
)
4537 + __le64_to_cpu(bsb
.length
)
4538 < info
->reshape_progress
)
4540 (__le64_to_cpu(bsb
.arraystart2
)
4541 + __le64_to_cpu(bsb
.length2
)
4542 < info
->reshape_progress
))
4543 goto nonew
; /* No new data here */
4545 /* reshape_progress is decreasing */
4546 if (__le64_to_cpu(bsb
.arraystart
) >=
4547 info
->reshape_progress
&&
4548 __le64_to_cpu(bsb
.arraystart2
) >=
4549 info
->reshape_progress
)
4550 goto nonew
; /* No new data here */
4553 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4556 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4558 continue; /* Cannot seek */
4560 /* There should be a duplicate backup superblock 4k before here */
4561 if (lseek64(fd
, -4096, 1) < 0 ||
4562 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4563 goto second_fail
; /* Cannot find leading superblock */
4564 if (bsb
.magic
[15] == '1')
4565 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4567 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4568 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4569 goto second_fail
; /* Cannot find leading superblock */
4571 /* Now need the data offsets for all devices. */
4572 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4573 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4576 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4577 /* FIXME should be this be an error */
4579 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4580 st
->ss
->free_super(st
);
4581 offsets
[j
] = dinfo
.data_offset
* 512;
4583 printf("%s: restoring critical section\n", Name
);
4585 if (restore_stripes(fdlist
, offsets
,
4586 info
->array
.raid_disks
,
4590 fd
, __le64_to_cpu(bsb
.devstart
)*512,
4591 __le64_to_cpu(bsb
.arraystart
)*512,
4592 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4593 /* didn't succeed, so giveup */
4595 pr_err("Error restoring backup from %s\n",
4601 if (bsb
.magic
[15] == '2' &&
4602 restore_stripes(fdlist
, offsets
,
4603 info
->array
.raid_disks
,
4607 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
4608 __le64_to_cpu(bsb
.devstart2
)*512,
4609 __le64_to_cpu(bsb
.arraystart2
)*512,
4610 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4611 /* didn't succeed, so giveup */
4613 pr_err("Error restoring second backup from %s\n",
4621 /* Ok, so the data is restored. Let's update those superblocks. */
4625 lo
= __le64_to_cpu(bsb
.arraystart
);
4626 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4628 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4629 unsigned long long lo1
, hi1
;
4630 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4631 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4635 } else if (lo
< lo1
)
4641 (info
->reshape_progress
< lo
||
4642 info
->reshape_progress
> hi
))
4643 /* backup does not affect reshape_progress*/ ;
4644 else if (info
->delta_disks
>= 0) {
4645 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4646 __le64_to_cpu(bsb
.length
);
4647 if (bsb
.magic
[15] == '2') {
4648 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
4649 __le64_to_cpu(bsb
.length2
);
4650 if (p2
> info
->reshape_progress
)
4651 info
->reshape_progress
= p2
;
4654 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4655 if (bsb
.magic
[15] == '2') {
4656 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
4657 if (p2
< info
->reshape_progress
)
4658 info
->reshape_progress
= p2
;
4661 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4664 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4666 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4667 dinfo
.reshape_progress
= info
->reshape_progress
;
4668 st
->ss
->update_super(st
, &dinfo
,
4669 "_reshape_progress",
4671 st
->ss
->store_super(st
, fdlist
[j
]);
4672 st
->ss
->free_super(st
);
4676 /* Didn't find any backup data, try to see if any
4679 if (info
->delta_disks
< 0) {
4680 /* When shrinking, the critical section is at the end.
4681 * So see if we are before the critical section.
4683 unsigned long long first_block
;
4684 nstripe
= ostripe
= 0;
4686 while (ostripe
>= nstripe
) {
4687 ostripe
+= info
->array
.chunk_size
/ 512;
4688 first_block
= ostripe
* odata
;
4689 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4690 (info
->new_chunk
/512);
4693 if (info
->reshape_progress
>= first_block
)
4696 if (info
->delta_disks
> 0) {
4697 /* See if we are beyond the critical section. */
4698 unsigned long long last_block
;
4699 nstripe
= ostripe
= 0;
4701 while (nstripe
>= ostripe
) {
4702 nstripe
+= info
->new_chunk
/ 512;
4703 last_block
= nstripe
* ndata
;
4704 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4705 (info
->array
.chunk_size
/512);
4708 if (info
->reshape_progress
>= last_block
)
4711 /* needed to recover critical section! */
4713 pr_err("Failed to find backup of critical section\n");
4717 int Grow_continue_command(char *devname
, int fd
,
4718 char *backup_file
, int verbose
)
4721 struct supertype
*st
= NULL
;
4722 struct mdinfo
*content
= NULL
;
4723 struct mdinfo array
;
4724 char *subarray
= NULL
;
4725 struct mdinfo
*cc
= NULL
;
4726 struct mdstat_ent
*mdstat
= NULL
;
4730 dprintf("Grow continue from command line called for %s\n",
4733 st
= super_by_fd(fd
, &subarray
);
4734 if (!st
|| !st
->ss
) {
4735 pr_err("Unable to determine metadata format for %s\n",
4739 dprintf("Grow continue is run for ");
4740 if (st
->ss
->external
== 0) {
4742 dprintf_cont("native array (%s)\n", devname
);
4743 if (ioctl(fd
, GET_ARRAY_INFO
, &array
.array
) < 0) {
4744 pr_err("%s is not an active md array - aborting\n", devname
);
4746 goto Grow_continue_command_exit
;
4749 /* Need to load a superblock.
4750 * FIXME we should really get what we need from
4753 for (d
= 0; d
< MAX_DISKS
; d
++) {
4754 mdu_disk_info_t disk
;
4758 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
4760 if (disk
.major
== 0 && disk
.minor
== 0)
4762 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
4764 dv
= map_dev(disk
.major
, disk
.minor
, 1);
4767 fd2
= dev_open(dv
, O_RDONLY
);
4770 err
= st
->ss
->load_super(st
, fd2
, NULL
);
4772 /* invalidate fd2 to avoid possible double close() */
4778 if (d
== MAX_DISKS
) {
4779 pr_err("Unable to load metadata for %s\n",
4782 goto Grow_continue_command_exit
;
4784 st
->ss
->getinfo_super(st
, content
, NULL
);
4789 dprintf_cont("subarray (%s)\n", subarray
);
4790 container
= st
->container_devnm
;
4791 cfd
= open_dev_excl(st
->container_devnm
);
4793 container
= st
->devnm
;
4795 cfd
= open_dev_excl(st
->devnm
);
4796 dprintf_cont("container (%s)\n", container
);
4800 pr_err("Unable to open container for %s\n", devname
);
4802 goto Grow_continue_command_exit
;
4805 /* find in container array under reshape
4807 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
4809 pr_err("Cannot read superblock for %s\n",
4812 goto Grow_continue_command_exit
;
4815 cc
= st
->ss
->container_content(st
, subarray
);
4816 for (content
= cc
; content
; content
= content
->next
) {
4818 int allow_reshape
= 1;
4820 if (content
->reshape_active
== 0)
4822 /* The decision about array or container wide
4823 * reshape is taken in Grow_continue based
4824 * content->reshape_active state, therefore we
4825 * need to check_reshape based on
4826 * reshape_active and subarray name
4828 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
4830 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
4831 (content
->array
.state
4832 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
4835 if (!allow_reshape
) {
4836 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4837 devname
, container
);
4839 goto Grow_continue_command_exit
;
4842 array
= strchr(content
->text_version
+1, '/')+1;
4843 mdstat
= mdstat_by_subdev(array
, container
);
4846 if (mdstat
->active
== 0) {
4847 pr_err("Skipping inactive array %s.\n",
4849 free_mdstat(mdstat
);
4856 pr_err("Unable to determine reshaped array for %s\n", devname
);
4858 goto Grow_continue_command_exit
;
4860 fd2
= open_dev(mdstat
->devnm
);
4862 pr_err("cannot open (%s)\n", mdstat
->devnm
);
4864 goto Grow_continue_command_exit
;
4867 sysfs_init(content
, fd2
, mdstat
->devnm
);
4869 /* start mdmon in case it is not running
4871 if (!mdmon_running(container
))
4872 start_mdmon(container
);
4873 ping_monitor(container
);
4875 if (mdmon_running(container
))
4876 st
->update_tail
= &st
->updates
;
4878 pr_err("No mdmon found. Grow cannot continue.\n");
4880 goto Grow_continue_command_exit
;
4884 /* verify that array under reshape is started from
4887 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
4889 goto Grow_continue_command_exit
;
4894 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
4896 Grow_continue_command_exit
:
4901 st
->ss
->free_super(st
);
4902 free_mdstat(mdstat
);
4909 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
4910 char *backup_file
, int forked
, int freeze_reshape
)
4914 if (!info
->reshape_active
)
4917 if (st
->ss
->external
) {
4918 int cfd
= open_dev(st
->container_devnm
);
4923 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
4925 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
4926 st
, info
, 0, backup_file
,
4928 1 | info
->reshape_active
,
4931 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
4932 NULL
, INVALID_SECTORS
,
4933 backup_file
, 0, forked
,
4934 1 | info
->reshape_active
,
4940 char *make_backup(char *name
)
4942 char *base
= "backup_file-";
4946 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
4947 fname
= xmalloc(len
);
4948 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
4952 char *locate_backup(char *name
)
4954 char *fl
= make_backup(name
);
4957 if (stat(fl
, &stb
) == 0 &&
4958 S_ISREG(stb
.st_mode
))