2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
38 int restore_backup(struct supertype
*st
,
39 struct mdinfo
*content
,
49 int disk_count
= next_spare
+ working_disks
;
50 char *backup_file
= *backup_filep
;
52 dprintf("Called restore_backup()\n");
53 fdlist
= xmalloc(sizeof(int) * disk_count
);
55 enable_fds(next_spare
);
56 for (i
= 0; i
< next_spare
; i
++)
58 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
64 fd
= dev_open(buf
, O_RDWR
);
66 if (dev
->disk
.raid_disk
>= 0)
67 fdlist
[dev
->disk
.raid_disk
] = fd
;
69 fdlist
[next_spare
++] = fd
;
73 backup_file
= locate_backup(content
->sys_name
);
74 *backup_filep
= backup_file
;
77 if (st
->ss
->external
&& st
->ss
->recover_backup
)
78 err
= st
->ss
->recover_backup(st
, content
);
80 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
81 backup_file
, verbose
> 0);
83 while (next_spare
> 0) {
85 if (fdlist
[next_spare
] >= 0)
86 close(fdlist
[next_spare
]);
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
92 pr_err("Possibly you need to specify a --backup-file\n");
96 dprintf("restore_backup() returns status OK.\n");
100 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
108 * This means that we need to *find* all other devices.
115 struct supertype
*st
= NULL
;
116 char *subarray
= NULL
;
118 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
119 pr_err("cannot get array info for %s\n", devname
);
123 if (info
.array
.level
!= -1) {
124 pr_err("can only add devices to linear arrays\n");
128 st
= super_by_fd(fd
, &subarray
);
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info
.array
.major_version
);
136 pr_err("Cannot grow linear sub-arrays yet\n");
142 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
144 pr_err("cannot open %s\n", newdev
);
149 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
150 pr_err("%s is not a block device!\n", newdev
);
155 /* now check out all the devices and make sure we can read the
157 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
158 mdu_disk_info_t disk
;
161 st
->ss
->free_super(st
);
164 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
165 pr_err("cannot get device detail for device %d\n",
171 dv
= map_dev(disk
.major
, disk
.minor
, 1);
173 pr_err("cannot find device file for device %d\n",
179 fd2
= dev_open(dv
, O_RDWR
);
181 pr_err("cannot open device file %s\n", dv
);
187 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
188 pr_err("cannot find super block on %s\n", dv
);
196 /* Ok, looks good. Lets update the superblock and write it out to
200 info
.disk
.number
= d
;
201 info
.disk
.major
= major(stb
.st_rdev
);
202 info
.disk
.minor
= minor(stb
.st_rdev
);
203 info
.disk
.raid_disk
= d
;
204 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
205 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
208 if (st
->ss
->store_super(st
, nfd
)) {
209 pr_err("Cannot store new superblock on %s\n",
216 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
217 pr_err("Cannot add new disk to this array\n");
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
224 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
225 pr_err("cannot get array info for %s\n", devname
);
230 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
231 mdu_disk_info_t disk
;
235 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
236 pr_err("cannot get device detail for device %d\n",
240 dv
= map_dev(disk
.major
, disk
.minor
, 1);
242 pr_err("cannot find device file for device %d\n",
246 fd2
= dev_open(dv
, O_RDWR
);
248 pr_err("cannot open device file %s\n", dv
);
251 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
252 pr_err("cannot find super block on %s\n", dv
);
256 info
.array
.raid_disks
= nd
+1;
257 info
.array
.nr_disks
= nd
+1;
258 info
.array
.active_disks
= nd
+1;
259 info
.array
.working_disks
= nd
+1;
261 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
264 if (st
->ss
->store_super(st
, fd2
)) {
265 pr_err("Cannot store new superblock on %s\n", dv
);
275 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
286 mdu_bitmap_file_t bmf
;
287 mdu_array_info_t array
;
288 struct supertype
*st
;
289 char *subarray
= NULL
;
290 int major
= BITMAP_MAJOR_HI
;
291 int vers
= md_get_version(fd
);
292 unsigned long long bitmapsize
, array_size
;
295 major
= BITMAP_MAJOR_HOSTENDIAN
;
296 pr_err("Warning - bitmaps created on this kernel are not portable\n"
297 " between different architectures. Consider upgrading the Linux kernel.\n");
300 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
302 pr_err("Memory allocation failure.\n");
304 pr_err("bitmaps not supported by this kernel.\n");
307 if (bmf
.pathname
[0]) {
308 if (strcmp(s
->bitmap_file
,"none")==0) {
309 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
310 pr_err("failed to remove bitmap %s\n",
316 pr_err("%s already has a bitmap (%s)\n",
317 devname
, bmf
.pathname
);
320 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
321 pr_err("cannot get array status for %s\n", devname
);
324 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
325 if (strcmp(s
->bitmap_file
, "none")==0) {
326 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
327 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
328 pr_err("failed to remove internal bitmap.\n");
333 pr_err("Internal bitmap already present on %s\n",
338 if (strcmp(s
->bitmap_file
, "none") == 0) {
339 pr_err("no bitmap found on %s\n", devname
);
342 if (array
.level
<= 0) {
343 pr_err("Bitmaps not meaningful with level %s\n",
344 map_num(pers
, array
.level
)?:"of this array");
347 bitmapsize
= array
.size
;
349 if (get_dev_size(fd
, NULL
, &array_size
) &&
350 array_size
> (0x7fffffffULL
<<9)) {
351 /* Array is big enough that we cannot trust array.size
352 * try other approaches
354 bitmapsize
= get_component_size(fd
);
356 if (bitmapsize
== 0) {
357 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
361 if (array
.level
== 10) {
362 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
363 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
366 st
= super_by_fd(fd
, &subarray
);
368 pr_err("Cannot understand version %d.%d\n",
369 array
.major_version
, array
.minor_version
);
373 pr_err("Cannot add bitmaps to sub-arrays yet\n");
378 if (strcmp(s
->bitmap_file
, "internal") == 0) {
381 int offset_setable
= 0;
383 if (st
->ss
->add_internal_bitmap
== NULL
) {
384 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
387 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
390 for (d
=0; d
< st
->max_devs
; d
++) {
391 mdu_disk_info_t disk
;
394 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
396 if (disk
.major
== 0 &&
399 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
401 dv
= map_dev(disk
.major
, disk
.minor
, 1);
403 int fd2
= dev_open(dv
, O_RDWR
);
406 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
407 if (st
->ss
->add_internal_bitmap(
409 &s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
410 bitmapsize
, offset_setable
,
413 st
->ss
->write_bitmap(st
, fd2
);
415 pr_err("failed to create internal bitmap - chunksize problem.\n");
423 if (offset_setable
) {
424 st
->ss
->getinfo_super(st
, mdi
, NULL
);
425 sysfs_init(mdi
, fd
, NULL
);
426 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
429 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
430 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
434 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
435 pr_err("failed to set internal bitmap.\n");
442 int max_devs
= st
->max_devs
;
444 /* try to load a superblock */
445 for (d
= 0; d
< max_devs
; d
++) {
446 mdu_disk_info_t disk
;
450 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
452 if ((disk
.major
==0 && disk
.minor
==0) ||
453 (disk
.state
& (1<<MD_DISK_REMOVED
)))
455 dv
= map_dev(disk
.major
, disk
.minor
, 1);
458 fd2
= dev_open(dv
, O_RDONLY
);
460 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
462 st
->ss
->uuid_from_super(st
, uuid
);
469 pr_err("cannot find UUID for array!\n");
472 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
473 c
->delay
, s
->write_behind
, bitmapsize
, major
)) {
476 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
478 pr_err("weird: %s cannot be opened\n",
482 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
485 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
486 pr_err("Cannot set bitmap file for %s: %s\n",
487 devname
, strerror(err
));
496 * When reshaping an array we might need to backup some data.
497 * This is written to all spares with a 'super_block' describing it.
498 * The superblock goes 4K from the end of the used space on the
500 * It if written after the backup is complete.
501 * It has the following structure.
504 static struct mdp_backup_super
{
505 char magic
[16]; /* md_backup_data-1 or -2 */
508 /* start/sizes in 512byte sectors */
509 __u64 devstart
; /* address on backup device/file of data */
512 __u32 sb_csum
; /* csum of preceeding bytes. */
514 __u64 devstart2
; /* offset in to data of second section */
517 __u32 sb_csum2
; /* csum of preceeding bytes. */
519 } __attribute__((aligned(512))) bsb
, bsb2
;
521 static __u32
bsb_csum(char *buf
, int len
)
525 for (i
= 0; i
< len
; i
++)
526 csum
= (csum
<<3) + buf
[0];
527 return __cpu_to_le32(csum
);
530 static int check_idle(struct supertype
*st
)
532 /* Check that all member arrays for this container, or the
533 * container of this array, are idle
535 char *container
= (st
->container_devnm
[0]
536 ? st
->container_devnm
: st
->devnm
);
537 struct mdstat_ent
*ent
, *e
;
540 ent
= mdstat_read(0, 0);
541 for (e
= ent
; e
; e
= e
->next
) {
542 if (!is_container_member(e
, container
))
544 if (e
->percent
>= 0) {
553 static int freeze_container(struct supertype
*st
)
555 char *container
= (st
->container_devnm
[0]
556 ? st
->container_devnm
: st
->devnm
);
561 if (block_monitor(container
, 1)) {
562 pr_err("failed to freeze container\n");
569 static void unfreeze_container(struct supertype
*st
)
571 char *container
= (st
->container_devnm
[0]
572 ? st
->container_devnm
: st
->devnm
);
574 unblock_monitor(container
, 1);
577 static int freeze(struct supertype
*st
)
579 /* Try to freeze resync/rebuild on this array/container.
580 * Return -1 if the array is busy,
581 * return -2 container cannot be frozen,
582 * return 0 if this kernel doesn't support 'frozen'
583 * return 1 if it worked.
585 if (st
->ss
->external
)
586 return freeze_container(st
);
588 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
594 /* Need to clear any 'read-auto' status */
595 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
596 strncmp(buf
, "read-auto", 9) == 0)
597 sysfs_set_str(sra
, NULL
, "array_state", "clean");
599 err
= sysfs_freeze_array(sra
);
605 static void unfreeze(struct supertype
*st
)
607 if (st
->ss
->external
)
608 return unfreeze_container(st
);
610 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
614 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0
615 && strcmp(buf
, "frozen\n") == 0) {
616 printf("unfreeze\n");
617 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
623 static void wait_reshape(struct mdinfo
*sra
)
625 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
631 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
632 strncmp(action
, "reshape", 7) == 0)
633 sysfs_wait(fd
, NULL
);
637 static int reshape_super(struct supertype
*st
, unsigned long long size
,
638 int level
, int layout
, int chunksize
, int raid_disks
,
639 int delta_disks
, char *backup_file
, char *dev
,
640 int direction
, int verbose
)
642 /* nothing extra to check in the native case */
643 if (!st
->ss
->external
)
645 if (!st
->ss
->reshape_super
||
646 !st
->ss
->manage_reshape
) {
647 pr_err("%s metadata does not support reshape\n",
652 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
653 raid_disks
, delta_disks
, backup_file
, dev
,
657 static void sync_metadata(struct supertype
*st
)
659 if (st
->ss
->external
) {
660 if (st
->update_tail
) {
661 flush_metadata_updates(st
);
662 st
->update_tail
= &st
->updates
;
664 st
->ss
->sync_metadata(st
);
668 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
670 /* when dealing with external metadata subarrays we need to be
671 * prepared to handle EAGAIN. The kernel may need to wait for
672 * mdmon to mark the array active so the kernel can handle
673 * allocations/writeback when preparing the reshape action
674 * (md_allow_write()). We temporarily disable safe_mode_delay
675 * to close a race with the array_state going clean before the
676 * next write to raid_disks / stripe_cache_size
681 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
683 (strcmp(name
, "raid_disks") != 0 &&
684 strcmp(name
, "stripe_cache_size") != 0))
685 return sysfs_set_num(sra
, NULL
, name
, n
);
687 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
690 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
691 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
692 if (rc
< 0 && errno
== EAGAIN
) {
693 ping_monitor(container
);
694 /* if we get EAGAIN here then the monitor is not active
697 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
699 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
703 int start_reshape(struct mdinfo
*sra
, int already_running
,
704 int before_data_disks
, int data_disks
)
707 unsigned long long sync_max_to_set
;
709 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
710 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
711 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
712 sra
->reshape_progress
);
713 if (before_data_disks
<= data_disks
)
714 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
716 sync_max_to_set
= (sra
->component_size
* data_disks
717 - sra
->reshape_progress
) / data_disks
;
718 if (!already_running
)
719 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
720 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
721 if (!already_running
&& err
== 0) {
724 err
= sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
727 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
732 void abort_reshape(struct mdinfo
*sra
)
734 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
735 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
736 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
737 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
738 sysfs_set_num(sra
, NULL
, "sync_min", 0);
739 // It isn't safe to reset sync_max as we aren't monitoring.
740 // Array really should be stopped at this point.
743 int remove_disks_for_takeover(struct supertype
*st
,
748 struct mdinfo
*remaining
;
751 if (sra
->array
.level
== 10)
752 nr_of_copies
= layout
& 0xff;
753 else if (sra
->array
.level
== 1)
754 nr_of_copies
= sra
->array
.raid_disks
;
758 remaining
= sra
->devs
;
760 /* for each 'copy', select one device and remove from the list. */
761 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
762 struct mdinfo
**diskp
;
765 /* Find a working device to keep */
766 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
767 struct mdinfo
*disk
= *diskp
;
769 if (disk
->disk
.raid_disk
< slot
)
771 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
773 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
775 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
777 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
780 /* We have found a good disk to use! */
782 disk
->next
= sra
->devs
;
791 if (slot
< sra
->array
.raid_disks
) {
792 /* didn't find all slots */
798 sra
->devs
= remaining
;
802 /* Remove all 'remaining' devices from the array */
804 struct mdinfo
*sd
= remaining
;
805 remaining
= sd
->next
;
807 sysfs_set_str(sra
, sd
, "state", "faulty");
808 sysfs_set_str(sra
, sd
, "slot", "none");
809 /* for external metadata disks should be removed in mdmon */
810 if (!st
->ss
->external
)
811 sysfs_set_str(sra
, sd
, "state", "remove");
812 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
813 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
814 sd
->next
= sra
->devs
;
820 void reshape_free_fdlist(int *fdlist
,
821 unsigned long long *offsets
,
826 for (i
= 0; i
< size
; i
++)
834 int reshape_prepare_fdlist(char *devname
,
838 unsigned long blocks
,
841 unsigned long long *offsets
)
847 for (d
= 0; d
<= nrdisks
; d
++)
850 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
851 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
853 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
854 sd
->disk
.raid_disk
< raid_disks
) {
855 char *dn
= map_dev(sd
->disk
.major
,
857 fdlist
[sd
->disk
.raid_disk
]
858 = dev_open(dn
, O_RDONLY
);
859 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
860 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
861 pr_err("%s: cannot open component %s\n",
862 devname
, dn
? dn
: "-unknown-");
866 } else if (backup_file
== NULL
) {
868 char *dn
= map_dev(sd
->disk
.major
,
870 fdlist
[d
] = dev_open(dn
, O_RDWR
);
871 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
873 pr_err("%s: cannot open component %s\n",
874 devname
, dn
? dn
: "-unknown-");
885 int reshape_open_backup_file(char *backup_file
,
890 unsigned long long *offsets
,
894 /* Return 1 on success, 0 on any form of failure */
895 /* need to check backup file is large enough */
901 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
905 pr_err("%s: cannot create backup file %s: %s\n",
906 devname
, backup_file
, strerror(errno
));
909 /* Guard against backup file being on array device.
910 * If array is partitioned or if LVM etc is in the
911 * way this will not notice, but it is better than
914 fstat(*fdlist
, &stb
);
917 if (stb
.st_rdev
== dev
) {
918 pr_err("backup file must NOT be on the array being reshaped.\n");
924 for (i
=0; i
< blocks
+ 8 ; i
++) {
925 if (write(*fdlist
, buf
, 512) != 512) {
926 pr_err("%s: cannot create backup file %s: %s\n",
927 devname
, backup_file
, strerror(errno
));
931 if (fsync(*fdlist
) != 0) {
932 pr_err("%s: cannot create backup file %s: %s\n",
933 devname
, backup_file
, strerror(errno
));
937 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
938 char *bu
= make_backup(sys_name
);
939 if (symlink(backup_file
, bu
))
940 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
948 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
949 unsigned int ndata
, unsigned int odata
)
951 unsigned long a
, b
, blocks
;
952 /* So how much do we need to backup.
953 * We need an amount of data which is both a whole number of
954 * old stripes and a whole number of new stripes.
955 * So LCM for (chunksize*datadisks).
957 a
= (ochunk
/512) * odata
;
958 b
= (nchunk
/512) * ndata
;
961 /* LCM == product / GCD */
962 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
967 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
969 /* Based on the current array state in info->array and
970 * the changes in info->new_* etc, determine:
971 * - whether the change is possible
972 * - Intermediate level/raid_disks/layout
973 * - whether a restriping reshape is needed
974 * - number of sectors in minimum change unit. This
975 * will cover a whole number of stripes in 'before' and
978 * Return message if the change should be rejected
979 * NULL if the change can be achieved
981 * This can be called as part of starting a reshape, or
982 * when assembling an array that is undergoing reshape.
984 int near
, far
, offset
, copies
;
986 int old_chunk
, new_chunk
;
987 /* delta_parity records change in number of devices
988 * caused by level change
990 int delta_parity
= 0;
992 memset(re
, 0, sizeof(*re
));
994 /* If a new level not explicitly given, we assume no-change */
995 if (info
->new_level
== UnSet
)
996 info
->new_level
= info
->array
.level
;
999 switch (info
->new_level
) {
1005 /* chunk size is meaningful, must divide component_size
1008 if (info
->component_size
% (info
->new_chunk
/512)) {
1009 unsigned long long shrink
= info
->component_size
;
1010 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1011 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1012 info
->new_chunk
/1024, info
->component_size
/2);
1013 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1015 pr_err("will shrink the array so the given chunk size would work.\n");
1020 return "chunk size not meaningful for this level";
1023 info
->new_chunk
= info
->array
.chunk_size
;
1025 switch (info
->array
.level
) {
1027 return "No reshape is possibly for this RAID level";
1029 if (info
->delta_disks
!= UnSet
)
1030 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1032 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1034 /* RAID1 can convert to RAID1 with different disks, or
1035 * raid5 with 2 disks, or
1038 if (info
->new_level
> 1 &&
1039 (info
->component_size
& 7))
1040 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1041 if (info
->new_level
== 0) {
1042 if (info
->delta_disks
!= UnSet
&&
1043 info
->delta_disks
!= 0)
1044 return "Cannot change number of disks with RAID1->RAID0 conversion";
1046 re
->before
.data_disks
= 1;
1047 re
->after
.data_disks
= 1;
1050 if (info
->new_level
== 1) {
1051 if (info
->delta_disks
== UnSet
)
1052 /* Don't know what to do */
1053 return "no change requested for Growing RAID1";
1057 if (info
->array
.raid_disks
== 2 &&
1058 info
->new_level
== 5) {
1061 re
->before
.data_disks
= 1;
1062 if (info
->delta_disks
!= UnSet
&&
1063 info
->delta_disks
!= 0)
1064 re
->after
.data_disks
= 1 + info
->delta_disks
;
1066 re
->after
.data_disks
= 1;
1067 if (re
->after
.data_disks
< 1)
1068 return "Number of disks too small for RAID5";
1070 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1071 info
->array
.chunk_size
= 65536;
1074 /* Could do some multi-stage conversions, but leave that to
1077 return "Impossibly level change request for RAID1";
1080 /* RAID10 can be converted from near mode to
1081 * RAID0 by removing some devices.
1082 * It can also be reshaped if the kernel supports
1085 switch (info
->new_level
) {
1087 if ((info
->array
.layout
& ~0xff) != 0x100)
1088 return "Cannot Grow RAID10 with far/offset layout";
1089 /* number of devices must be multiple of number of copies */
1090 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1091 return "RAID10 layout too complex for Grow operation";
1093 new_disks
= (info
->array
.raid_disks
1094 / (info
->array
.layout
& 0xff));
1095 if (info
->delta_disks
== UnSet
)
1096 info
->delta_disks
= (new_disks
1097 - info
->array
.raid_disks
);
1099 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1100 return "New number of raid-devices impossible for RAID10";
1101 if (info
->new_chunk
&&
1102 info
->new_chunk
!= info
->array
.chunk_size
)
1103 return "Cannot change chunk-size with RAID10 Grow";
1107 re
->before
.data_disks
= new_disks
;
1108 re
->after
.data_disks
= re
->before
.data_disks
;
1112 near
= info
->array
.layout
& 0xff;
1113 far
= (info
->array
.layout
>> 8) & 0xff;
1114 offset
= info
->array
.layout
& 0x10000;
1115 if (far
> 1 && !offset
)
1116 return "Cannot reshape RAID10 in far-mode";
1117 copies
= near
* far
;
1119 old_chunk
= info
->array
.chunk_size
* far
;
1121 if (info
->new_layout
== UnSet
)
1122 info
->new_layout
= info
->array
.layout
;
1124 near
= info
->new_layout
& 0xff;
1125 far
= (info
->new_layout
>> 8) & 0xff;
1126 offset
= info
->new_layout
& 0x10000;
1127 if (far
> 1 && !offset
)
1128 return "Cannot reshape RAID10 to far-mode";
1129 if (near
* far
!= copies
)
1130 return "Cannot change number of copies when reshaping RAID10";
1132 if (info
->delta_disks
== UnSet
)
1133 info
->delta_disks
= 0;
1134 new_disks
= (info
->array
.raid_disks
+
1137 new_chunk
= info
->new_chunk
* far
;
1140 re
->before
.layout
= info
->array
.layout
;
1141 re
->before
.data_disks
= info
->array
.raid_disks
;
1142 re
->after
.layout
= info
->new_layout
;
1143 re
->after
.data_disks
= new_disks
;
1144 /* For RAID10 we don't do backup but do allow reshape,
1145 * so set backup_blocks to INVALID_SECTORS rather than
1147 * And there is no need to synchronise stripes on both
1148 * 'old' and 'new'. So the important
1149 * number is the minimum data_offset difference
1150 * which is the larger of (offset copies * chunk).
1152 re
->backup_blocks
= INVALID_SECTORS
;
1153 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1154 if (new_disks
< re
->before
.data_disks
&&
1155 info
->space_after
< re
->min_offset_change
)
1156 /* Reduce component size by one chunk */
1157 re
->new_size
= (info
->component_size
-
1158 re
->min_offset_change
);
1160 re
->new_size
= info
->component_size
;
1161 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1165 return "RAID10 can only be changed to RAID0";
1168 /* RAID0 can be converted to RAID10, or to RAID456 */
1169 if (info
->new_level
== 10) {
1170 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1171 /* Assume near=2 layout */
1172 info
->new_layout
= 0x102;
1173 info
->delta_disks
= info
->array
.raid_disks
;
1175 if (info
->new_layout
== UnSet
) {
1176 int copies
= 1 + (info
->delta_disks
1177 / info
->array
.raid_disks
);
1178 if (info
->array
.raid_disks
* (copies
-1)
1179 != info
->delta_disks
)
1180 return "Impossible number of devices for RAID0->RAID10";
1181 info
->new_layout
= 0x100 + copies
;
1183 if (info
->delta_disks
== UnSet
) {
1184 int copies
= info
->new_layout
& 0xff;
1185 if (info
->new_layout
!= 0x100 + copies
)
1186 return "New layout impossible for RAID0->RAID10";;
1187 info
->delta_disks
= (copies
- 1) *
1188 info
->array
.raid_disks
;
1190 if (info
->new_chunk
&&
1191 info
->new_chunk
!= info
->array
.chunk_size
)
1192 return "Cannot change chunk-size with RAID0->RAID10";
1195 re
->before
.data_disks
= (info
->array
.raid_disks
+
1197 re
->after
.data_disks
= re
->before
.data_disks
;
1198 re
->before
.layout
= info
->new_layout
;
1202 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1203 * a raid4 style layout of the final level.
1205 switch (info
->new_level
) {
1210 re
->before
.layout
= 0;
1215 re
->before
.layout
= ALGORITHM_PARITY_N
;
1216 if (info
->new_layout
== UnSet
)
1217 info
->new_layout
= map_name(r5layout
, "default");
1222 re
->before
.layout
= ALGORITHM_PARITY_N
;
1223 if (info
->new_layout
== UnSet
)
1224 info
->new_layout
= map_name(r6layout
, "default");
1227 return "Impossible level change requested";
1229 re
->before
.data_disks
= info
->array
.raid_disks
;
1230 /* determining 'after' layout happens outside this 'switch' */
1234 info
->array
.layout
= ALGORITHM_PARITY_N
;
1236 switch (info
->new_level
) {
1240 re
->level
= info
->array
.level
;
1241 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1242 re
->before
.layout
= info
->array
.layout
;
1246 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1247 re
->before
.layout
= info
->array
.layout
;
1252 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1253 switch (info
->array
.layout
) {
1254 case ALGORITHM_LEFT_ASYMMETRIC
:
1255 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1257 case ALGORITHM_RIGHT_ASYMMETRIC
:
1258 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1260 case ALGORITHM_LEFT_SYMMETRIC
:
1261 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1263 case ALGORITHM_RIGHT_SYMMETRIC
:
1264 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1266 case ALGORITHM_PARITY_0
:
1267 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1269 case ALGORITHM_PARITY_N
:
1270 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1273 return "Cannot convert an array with this layout";
1277 if (info
->array
.raid_disks
!= 2)
1278 return "Can only convert a 2-device array to RAID1";
1279 if (info
->delta_disks
!= UnSet
&&
1280 info
->delta_disks
!= 0)
1281 return "Cannot set raid_disk when converting RAID5->RAID1";
1283 info
->new_chunk
= 0;
1286 return "Impossible level change requested";
1290 switch (info
->new_level
) {
1296 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1297 re
->before
.layout
= info
->array
.layout
;
1300 return "Impossible level change requested";
1305 /* If we reached here then it looks like a re-stripe is
1306 * happening. We have determined the intermediate level
1307 * and initial raid_disks/layout and stored these in 're'.
1309 * We need to deduce the final layout that can be atomically
1310 * converted to the end state.
1312 switch (info
->new_level
) {
1314 /* We can only get to RAID0 from RAID4 or RAID5
1315 * with appropriate layout and one extra device
1317 if (re
->level
!= 4 && re
->level
!= 5)
1318 return "Cannot covert to RAID0 from this level";
1320 switch (re
->level
) {
1322 re
->before
.layout
= 0;
1323 re
->after
.layout
= 0;
1326 re
->after
.layout
= ALGORITHM_PARITY_N
;
1332 /* We can only get to RAID4 from RAID5 */
1333 if (re
->level
!= 4 && re
->level
!= 5)
1334 return "Cannot convert to RAID4 from this level";
1336 switch (re
->level
) {
1338 re
->after
.layout
= 0;
1341 re
->after
.layout
= ALGORITHM_PARITY_N
;
1347 /* We get to RAID5 from RAID5 or RAID6 */
1348 if (re
->level
!= 5 && re
->level
!= 6)
1349 return "Cannot convert to RAID5 from this level";
1351 switch (re
->level
) {
1353 if (info
->new_layout
== UnSet
)
1354 re
->after
.layout
= re
->before
.layout
;
1356 re
->after
.layout
= info
->new_layout
;
1359 if (info
->new_layout
== UnSet
)
1360 info
->new_layout
= re
->before
.layout
;
1362 /* after.layout needs to be raid6 version of new_layout */
1363 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1364 re
->after
.layout
= ALGORITHM_PARITY_N
;
1367 char *ls
= map_num(r5layout
, info
->new_layout
);
1370 /* Current RAID6 layout has a RAID5
1373 strcat(strcpy(layout
, ls
), "-6");
1374 l
= map_name(r6layout
, layout
);
1376 return "Cannot find RAID6 layout to convert to";
1378 /* Current RAID6 has no equivalent.
1379 * If it is already a '-6' layout we
1380 * can leave it unchanged, else we must
1383 ls
= map_num(r6layout
, info
->new_layout
);
1385 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1386 return "Please specify new layout";
1387 l
= info
->new_layout
;
1389 re
->after
.layout
= l
;
1395 /* We must already be at level 6 */
1397 return "Impossible level change";
1398 if (info
->new_layout
== UnSet
)
1399 re
->after
.layout
= info
->array
.layout
;
1401 re
->after
.layout
= info
->new_layout
;
1404 return "Impossible level change requested";
1406 if (info
->delta_disks
== UnSet
)
1407 info
->delta_disks
= delta_parity
;
1409 re
->after
.data_disks
= (re
->before
.data_disks
1412 switch (re
->level
) {
1413 case 6: re
->parity
= 2;
1416 case 5: re
->parity
= 1;
1418 default: re
->parity
= 0;
1421 /* So we have a restripe operation, we need to calculate the number
1422 * of blocks per reshape operation.
1424 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1425 if (info
->new_chunk
== 0)
1426 info
->new_chunk
= info
->array
.chunk_size
;
1427 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1428 re
->after
.layout
== re
->before
.layout
&&
1429 info
->new_chunk
== info
->array
.chunk_size
) {
1430 /* Nothing to change, can change level immediately. */
1431 re
->level
= info
->new_level
;
1432 re
->backup_blocks
= 0;
1435 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1436 /* chunk and layout changes make no difference */
1437 re
->level
= info
->new_level
;
1438 re
->backup_blocks
= 0;
1442 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1443 get_linux_version() < 2006032)
1444 return "in-place reshape is not safe before 2.6.32 - sorry.";
1446 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1447 get_linux_version() < 2006030)
1448 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1450 re
->backup_blocks
= compute_backup_blocks(
1451 info
->new_chunk
, info
->array
.chunk_size
,
1452 re
->after
.data_disks
,
1453 re
->before
.data_disks
);
1454 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1456 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1460 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1463 struct mdinfo
*info
;
1467 if ((st
== NULL
) || (sra
== NULL
))
1470 if (text_version
== NULL
)
1471 text_version
= sra
->text_version
;
1472 subarray
= strchr(text_version
+1, '/')+1;
1473 info
= st
->ss
->container_content(st
, subarray
);
1475 unsigned long long current_size
= 0;
1476 unsigned long long new_size
=
1477 info
->custom_array_size
/2;
1479 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1480 new_size
> current_size
) {
1481 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1483 dprintf("Error: Cannot set array size");
1486 dprintf("Array size changed");
1488 dprintf_cont(" from %llu to %llu.\n",
1489 current_size
, new_size
);
1493 dprintf("Error: set_array_size(): info pointer in NULL\n");
1498 static int reshape_array(char *container
, int fd
, char *devname
,
1499 struct supertype
*st
, struct mdinfo
*info
,
1500 int force
, struct mddev_dev
*devlist
,
1501 unsigned long long data_offset
,
1502 char *backup_file
, int verbose
, int forked
,
1503 int restart
, int freeze_reshape
);
1504 static int reshape_container(char *container
, char *devname
,
1506 struct supertype
*st
,
1507 struct mdinfo
*info
,
1509 char *backup_file
, int verbose
,
1510 int forked
, int restart
, int freeze_reshape
);
1512 int Grow_reshape(char *devname
, int fd
,
1513 struct mddev_dev
*devlist
,
1514 unsigned long long data_offset
,
1515 struct context
*c
, struct shape
*s
)
1517 /* Make some changes in the shape of an array.
1518 * The kernel must support the change.
1520 * There are three different changes. Each can trigger
1521 * a resync or recovery so we freeze that until we have
1522 * requested everything (if kernel supports freezing - 2.6.30).
1524 * - change size (i.e. component_size)
1526 * - change layout/chunksize/ndisks
1528 * The last can require a reshape. It is different on different
1529 * levels so we need to check the level before actioning it.
1530 * Some times the level change needs to be requested after the
1531 * reshape (e.g. raid6->raid5, raid5->raid0)
1534 struct mdu_array_info_s array
;
1536 struct supertype
*st
;
1537 char *subarray
= NULL
;
1541 char *container
= NULL
;
1544 struct mddev_dev
*dv
;
1550 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1551 pr_err("%s is not an active md array - aborting\n",
1555 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10
1556 && (array
.level
< 4 || array
.level
> 6)) {
1557 pr_err("--grow --data-offset not yet supported\n");
1562 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1563 pr_err("cannot change component size at the same time as other changes.\n"
1564 " Change size first, then check data is intact before making other changes.\n");
1568 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&& array
.level
> 1 &&
1569 get_linux_version() < 2006032 &&
1570 !check_env("MDADM_FORCE_FEWER")) {
1571 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1572 " Please use a newer kernel\n");
1576 st
= super_by_fd(fd
, &subarray
);
1578 pr_err("Unable to determine metadata format for %s\n", devname
);
1581 if (s
->raiddisks
> st
->max_devs
) {
1582 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1586 /* in the external case we need to check that the requested reshape is
1587 * supported, and perform an initial check that the container holds the
1588 * pre-requisite spare devices (mdmon owns final validation)
1590 if (st
->ss
->external
) {
1594 container
= st
->container_devnm
;
1595 cfd
= open_dev_excl(st
->container_devnm
);
1597 container
= st
->devnm
;
1599 cfd
= open_dev_excl(st
->devnm
);
1603 pr_err("Unable to open container for %s\n",
1609 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1612 pr_err("Cannot read superblock for %s\n",
1618 /* check if operation is supported for metadata handler */
1619 if (st
->ss
->container_content
) {
1620 struct mdinfo
*cc
= NULL
;
1621 struct mdinfo
*content
= NULL
;
1623 cc
= st
->ss
->container_content(st
, subarray
);
1624 for (content
= cc
; content
; content
= content
->next
) {
1625 int allow_reshape
= 1;
1627 /* check if reshape is allowed based on metadata
1628 * indications stored in content.array.status
1630 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
1632 if (content
->array
.state
1633 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))
1635 if (!allow_reshape
) {
1636 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1637 devname
, container
);
1645 if (mdmon_running(container
))
1646 st
->update_tail
= &st
->updates
;
1650 for (dv
= devlist
; dv
; dv
= dv
->next
)
1652 if (s
->raiddisks
> array
.raid_disks
&&
1653 array
.spare_disks
+added_disks
< (s
->raiddisks
- array
.raid_disks
) &&
1655 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1656 " Use --force to over-ride this check.\n",
1657 s
->raiddisks
- array
.raid_disks
,
1658 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1659 array
.spare_disks
+ added_disks
);
1663 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
1664 | GET_STATE
| GET_VERSION
);
1666 if (st
->ss
->external
&& subarray
== NULL
) {
1667 array
.level
= LEVEL_CONTAINER
;
1668 sra
->array
.level
= LEVEL_CONTAINER
;
1671 pr_err("failed to read sysfs parameters for %s\n",
1675 frozen
= freeze(st
);
1677 /* freeze() already spewed the reason */
1680 } else if (frozen
< 0) {
1681 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1686 /* ========= set size =============== */
1687 if (s
->size
> 0 && (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1688 unsigned long long orig_size
= get_component_size(fd
)/2;
1689 unsigned long long min_csize
;
1691 int raid0_takeover
= 0;
1694 orig_size
= (unsigned) array
.size
;
1696 if (orig_size
== 0) {
1697 pr_err("Cannot set device size in this type of array.\n");
1702 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1703 devname
, APPLY_METADATA_CHANGES
, c
->verbose
> 0)) {
1708 if (st
->ss
->external
) {
1709 /* metadata can have size limitation
1710 * update size value according to metadata information
1712 struct mdinfo
*sizeinfo
=
1713 st
->ss
->container_content(st
, subarray
);
1715 unsigned long long new_size
=
1716 sizeinfo
->custom_array_size
/2;
1717 int data_disks
= get_data_disks(
1718 sizeinfo
->array
.level
,
1719 sizeinfo
->array
.layout
,
1720 sizeinfo
->array
.raid_disks
);
1721 new_size
/= data_disks
;
1722 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size
, new_size
,
1723 new_size
* data_disks
);
1725 sysfs_free(sizeinfo
);
1729 /* Update the size of each member device in case
1730 * they have been resized. This will never reduce
1731 * below the current used-size. The "size" attribute
1732 * understands '0' to mean 'max'.
1736 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
1737 if (sysfs_set_num(sra
, mdi
, "size",
1738 s
->size
== MAX_SIZE
? 0 : s
->size
) < 0) {
1739 /* Probably kernel refusing to let us
1740 * reduce the size - not an error.
1744 if (array
.not_persistent
== 0 &&
1745 array
.major_version
== 0 &&
1746 get_linux_version() < 3001000) {
1747 /* Dangerous to allow size to exceed 2TB */
1748 unsigned long long csize
;
1749 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
1750 if (csize
>= 2ULL*1024*1024*1024)
1751 csize
= 2ULL*1024*1024*1024;
1752 if ((min_csize
== 0 || (min_csize
1759 pr_err("Cannot set size on array members.\n");
1760 goto size_change_error
;
1762 if (min_csize
&& s
->size
> min_csize
) {
1763 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1765 goto size_change_error
;
1767 if (min_csize
&& s
->size
== MAX_SIZE
) {
1768 /* Don't let the kernel choose a size - it will get
1771 pr_err("Limited v0.90 array to 2TB per device\n");
1772 s
->size
= min_csize
;
1774 if (st
->ss
->external
) {
1775 if (sra
->array
.level
== 0) {
1776 rv
= sysfs_set_str(sra
, NULL
, "level",
1780 /* get array parameters after takeover
1781 * to change one parameter at time only
1783 rv
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
1786 /* make sure mdmon is
1787 * aware of the new level */
1788 if (!mdmon_running(st
->container_devnm
))
1789 start_mdmon(st
->container_devnm
);
1790 ping_monitor(container
);
1791 if (mdmon_running(st
->container_devnm
) &&
1792 st
->update_tail
== NULL
)
1793 st
->update_tail
= &st
->updates
;
1796 if (s
->size
== MAX_SIZE
)
1798 array
.size
= s
->size
;
1799 if (s
->size
& ~INT32_MAX
) {
1800 /* got truncated to 32bit, write to
1801 * component_size instead
1804 rv
= sysfs_set_num(sra
, NULL
,
1805 "component_size", s
->size
);
1809 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1811 /* manage array size when it is managed externally
1813 if ((rv
== 0) && st
->ss
->external
)
1814 rv
= set_array_size(st
, sra
, sra
->text_version
);
1817 if (raid0_takeover
) {
1818 /* do not recync non-existing parity,
1819 * we will drop it anyway
1821 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
1822 /* go back to raid0, drop parity disk
1824 sysfs_set_str(sra
, NULL
, "level", "raid0");
1825 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1832 /* restore metadata */
1833 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1834 UnSet
, NULL
, devname
,
1835 ROLLBACK_METADATA_CHANGES
,
1838 pr_err("Cannot set device size for %s: %s\n",
1839 devname
, strerror(err
));
1841 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1842 cont_err("Bitmap must be removed before size can be changed\n");
1846 if (s
->assume_clean
) {
1847 /* This will fail on kernels older than 3.0 unless
1848 * a backport has been arranged.
1851 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
1852 pr_err("--assume-clean not supported with --grow on this kernel\n");
1854 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1855 s
->size
= get_component_size(fd
)/2;
1857 s
->size
= array
.size
;
1858 if (c
->verbose
>= 0) {
1859 if (s
->size
== orig_size
)
1860 pr_err("component size of %s unchanged at %lluK\n",
1863 pr_err("component size of %s has been set to %lluK\n",
1867 } else if (array
.level
!= LEVEL_CONTAINER
) {
1868 s
->size
= get_component_size(fd
)/2;
1870 s
->size
= array
.size
;
1873 /* See if there is anything else to do */
1874 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
1875 (s
->layout_str
== NULL
) &&
1876 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
1877 data_offset
== INVALID_SECTORS
&&
1878 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
1879 /* Nothing more to do */
1880 if (!changed
&& c
->verbose
>= 0)
1881 pr_err("%s: no change requested\n",
1886 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1887 * current implementation assumes that following conditions must be met:
1890 * - near_copies == 2
1892 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
1893 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1894 (s
->level
== 0 && array
.level
== 1 && sra
)) {
1896 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1898 dprintf("Array cannot be reshaped\n");
1904 /* Make sure mdmon has seen the device removal
1905 * and updated metadata before we continue with
1909 ping_monitor(container
);
1912 memset(&info
, 0, sizeof(info
));
1914 sysfs_init(&info
, fd
, NULL
);
1915 strcpy(info
.text_version
, sra
->text_version
);
1916 info
.component_size
= s
->size
*2;
1917 info
.new_level
= s
->level
;
1918 info
.new_chunk
= s
->chunk
* 1024;
1919 if (info
.array
.level
== LEVEL_CONTAINER
) {
1920 info
.delta_disks
= UnSet
;
1921 info
.array
.raid_disks
= s
->raiddisks
;
1922 } else if (s
->raiddisks
)
1923 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
1925 info
.delta_disks
= UnSet
;
1926 if (s
->layout_str
== NULL
) {
1927 info
.new_layout
= UnSet
;
1928 if (info
.array
.level
== 6 &&
1929 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1930 info
.array
.layout
>= 16) {
1931 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
1932 cont_err("during the reshape, please specify --layout=preserve\n");
1933 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1937 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
1938 strcmp(s
->layout_str
, "normalize") == 0) {
1939 /* If we have a -6 RAID6 layout, remove the '-6'. */
1940 info
.new_layout
= UnSet
;
1941 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1943 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1944 h
= strrchr(l
, '-');
1945 if (h
&& strcmp(h
, "-6") == 0) {
1947 info
.new_layout
= map_name(r6layout
, l
);
1950 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
1954 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
1955 /* This means that a non-standard RAID6 layout
1958 * - When reshape a RAID6 (e.g. adding a device)
1959 * which is in a non-standard layout, it is OK
1960 * to preserve that layout.
1961 * - When converting a RAID5 to RAID6, leave it in
1962 * the XXX-6 layout, don't re-layout.
1964 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
1965 info
.new_layout
= info
.array
.layout
;
1966 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
1968 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
1970 info
.new_layout
= map_name(r6layout
, l
);
1972 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
1977 int l
= info
.new_level
;
1979 l
= info
.array
.level
;
1982 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
1985 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
1988 info
.new_layout
= parse_layout_10(s
->layout_str
);
1991 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
1994 pr_err("layout not meaningful with this level\n");
1998 if (info
.new_layout
== UnSet
) {
1999 pr_err("layout %s not understood for this level\n",
2006 if (array
.level
== LEVEL_FAULTY
) {
2007 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2008 pr_err("cannot change level of Faulty device\n");
2012 pr_err("cannot set chunksize of Faulty device\n");
2015 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2016 pr_err("cannot set raid_disks of Faulty device\n");
2019 if (s
->layout_str
) {
2020 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2021 dprintf("Cannot get array information.\n");
2024 array
.layout
= info
.new_layout
;
2025 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2026 pr_err("failed to set new layout\n");
2028 } else if (c
->verbose
>= 0)
2029 printf("layout for %s set to %d\n",
2030 devname
, array
.layout
);
2032 } else if (array
.level
== LEVEL_CONTAINER
) {
2033 /* This change is to be applied to every array in the
2034 * container. This is only needed when the metadata imposes
2035 * restraints of the various arrays in the container.
2036 * Currently we only know that IMSM requires all arrays
2037 * to have the same number of devices so changing the
2038 * number of devices (On-Line Capacity Expansion) must be
2039 * performed at the level of the container
2041 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2042 c
->force
, c
->backup_file
, c
->verbose
, 0, 0, 0);
2045 /* get spare devices from external metadata
2047 if (st
->ss
->external
) {
2048 struct mdinfo
*info2
;
2050 info2
= st
->ss
->container_content(st
, subarray
);
2052 info
.array
.spare_disks
=
2053 info2
->array
.spare_disks
;
2058 /* Impose these changes on a single array. First
2059 * check that the metadata is OK with the change. */
2061 if (reshape_super(st
, 0, info
.new_level
,
2062 info
.new_layout
, info
.new_chunk
,
2063 info
.array
.raid_disks
, info
.delta_disks
,
2064 c
->backup_file
, devname
, APPLY_METADATA_CHANGES
,
2070 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2071 devlist
, data_offset
, c
->backup_file
, c
->verbose
,
2082 /* verify_reshape_position()
2083 * Function checks if reshape position in metadata is not farther
2084 * than position in md.
2086 * 0 : not valid sysfs entry
2087 * it can be caused by not started reshape, it should be started
2088 * by reshape array or raid0 array is before takeover
2089 * -1 : error, reshape position is obviously wrong
2090 * 1 : success, reshape progress correct or updated
2092 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2098 /* read sync_max, failure can mean raid0 array */
2099 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2103 unsigned long long position
= strtoull(buf
, &ep
, 0);
2105 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2106 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2107 position
*= get_data_disks(level
,
2109 info
->array
.raid_disks
);
2110 if (info
->reshape_progress
< position
) {
2111 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2112 info
->reshape_progress
, position
);
2113 info
->reshape_progress
= position
;
2115 } else if (info
->reshape_progress
> position
) {
2116 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2117 position
, info
->reshape_progress
);
2120 dprintf("Reshape position in md and metadata are the same;");
2124 } else if (rv
== 0) {
2125 /* for valid sysfs entry, 0-length content
2126 * should be indicated as error
2134 static unsigned long long choose_offset(unsigned long long lo
,
2135 unsigned long long hi
,
2136 unsigned long long min
,
2137 unsigned long long max
)
2139 /* Choose a new offset between hi and lo.
2140 * It must be between min and max, but
2141 * we would prefer something near the middle of hi/lo, and also
2142 * prefer to be aligned to a big power of 2.
2144 * So we start with the middle, then for each bit,
2145 * starting at '1' and increasing, if it is set, we either
2146 * add it or subtract it if possible, preferring the option
2147 * which is furthest from the boundary.
2149 * We stop once we get a 1MB alignment. As units are in sectors,
2150 * 1MB = 2*1024 sectors.
2152 unsigned long long choice
= (lo
+ hi
) / 2;
2153 unsigned long long bit
= 1;
2155 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2156 unsigned long long bigger
, smaller
;
2157 if (! (bit
& choice
))
2159 bigger
= choice
+ bit
;
2160 smaller
= choice
- bit
;
2161 if (bigger
> max
&& smaller
< min
)
2165 else if (smaller
< min
)
2167 else if (hi
- bigger
> smaller
- lo
)
2175 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2176 char *devname
, int delta_disks
,
2177 unsigned long long data_offset
,
2178 unsigned long long min
,
2184 unsigned long long before
, after
;
2186 /* Need to find min space before and after so same is used
2189 before
= UINT64_MAX
;
2191 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2195 struct supertype
*st2
;
2196 struct mdinfo info2
;
2198 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2200 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2201 dfd
= dev_open(dn
, O_RDONLY
);
2203 pr_err("%s: cannot open component %s\n",
2204 devname
, dn
? dn
: "-unknown-");
2207 st2
= dup_super(st
);
2208 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2212 pr_err("%s: cannot get superblock from %s\n",
2216 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2217 st2
->ss
->free_super(st2
);
2219 if (info2
.space_before
== 0 &&
2220 info2
.space_after
== 0) {
2221 /* Metadata doesn't support data_offset changes */
2223 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2227 if (before
> info2
.space_before
)
2228 before
= info2
.space_before
;
2229 if (after
> info2
.space_after
)
2230 after
= info2
.space_after
;
2232 if (data_offset
!= INVALID_SECTORS
) {
2234 if (info2
.data_offset
== data_offset
) {
2235 pr_err("%s: already has that data_offset\n",
2239 if (data_offset
< info2
.data_offset
)
2243 } else if ((data_offset
<= info2
.data_offset
&& dir
== 1) ||
2244 (data_offset
>= info2
.data_offset
&& dir
== -1)) {
2245 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2251 if (before
== UINT64_MAX
)
2252 /* impossible really, there must be no devices */
2255 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2256 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2257 unsigned long long new_data_offset
;
2259 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2261 if (delta_disks
< 0) {
2262 /* Don't need any space as array is shrinking
2263 * just move data_offset up by min
2265 if (data_offset
== INVALID_SECTORS
)
2266 new_data_offset
= sd
->data_offset
+ min
;
2268 if (data_offset
< sd
->data_offset
+ min
) {
2269 pr_err("--data-offset too small for %s\n",
2273 new_data_offset
= data_offset
;
2275 } else if (delta_disks
> 0) {
2276 /* need space before */
2280 pr_err("Insufficient head-space for reshape on %s\n",
2284 if (data_offset
== INVALID_SECTORS
)
2285 new_data_offset
= sd
->data_offset
- min
;
2287 if (data_offset
> sd
->data_offset
- min
) {
2288 pr_err("--data-offset too large for %s\n",
2292 new_data_offset
= data_offset
;
2296 /* can move up or down. If 'data_offset'
2297 * was set we would have already decided,
2298 * so just choose direction with most space.
2305 sysfs_set_str(sra
, NULL
, "reshape_direction",
2306 dir
== 1 ? "backwards" : "forwards");
2308 /* Increase data offset */
2312 pr_err("Insufficient tail-space for reshape on %s\n",
2316 if (data_offset
!= INVALID_SECTORS
&&
2317 data_offset
< sd
->data_offset
+ min
) {
2318 pr_err("--data-offset too small on %s\n",
2322 if (data_offset
!= INVALID_SECTORS
)
2323 new_data_offset
= data_offset
;
2325 new_data_offset
= choose_offset(sd
->data_offset
,
2326 sd
->data_offset
+ after
,
2327 sd
->data_offset
+ min
,
2328 sd
->data_offset
+ after
);
2330 /* Decrease data offset */
2334 pr_err("insufficient head-room on %s\n",
2338 if (data_offset
!= INVALID_SECTORS
&&
2339 data_offset
< sd
->data_offset
- min
) {
2340 pr_err("--data-offset too small on %s\n",
2344 if (data_offset
!= INVALID_SECTORS
)
2345 new_data_offset
= data_offset
;
2347 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2349 sd
->data_offset
- before
,
2350 sd
->data_offset
- min
);
2353 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2354 if (err
< 0 && errno
== E2BIG
) {
2355 /* try again after increasing data size to max */
2356 err
= sysfs_set_num(sra
, sd
, "size", 0);
2357 if (err
< 0 && errno
== EINVAL
&&
2358 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2359 /* some kernels have a bug where you cannot
2360 * use '0' on spare devices. */
2361 sysfs_set_num(sra
, sd
, "size",
2362 (sra
->component_size
+ after
)/2);
2364 err
= sysfs_set_num(sra
, sd
, "new_offset",
2368 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2369 pr_err("data-offset is too big for %s\n",
2373 if (sd
== sra
->devs
&&
2374 (errno
== ENOENT
|| errno
== E2BIG
))
2375 /* Early kernel, no 'new_offset' file,
2376 * or kernel doesn't like us.
2377 * For RAID5/6 this is not fatal
2380 pr_err("Cannot set new_offset for %s\n",
2389 /* Just use a backup file */
2393 static int raid10_reshape(char *container
, int fd
, char *devname
,
2394 struct supertype
*st
, struct mdinfo
*info
,
2395 struct reshape
*reshape
,
2396 unsigned long long data_offset
,
2397 int force
, int verbose
)
2399 /* Changing raid_disks, layout, chunksize or possibly
2400 * just data_offset for a RAID10.
2401 * We must always change data_offset. We change by at least
2402 * ->min_offset_change which is the largest of the old and new
2404 * If raid_disks is increasing, then data_offset must decrease
2405 * by at least this copy size.
2406 * If raid_disks is unchanged, data_offset must increase or
2407 * decrease by at least min_offset_change but preferably by much more.
2408 * We choose half of the available space.
2409 * If raid_disks is decreasing, data_offset must increase by
2410 * at least min_offset_change. To allow of this, component_size
2411 * must be decreased by the same amount.
2413 * So we calculate the required minimum and direction, possibly
2414 * reduce the component_size, then iterate through the devices
2415 * and set the new_data_offset.
2416 * If that all works, we set chunk_size, layout, raid_disks, and start
2420 unsigned long long min
;
2423 sra
= sysfs_read(fd
, NULL
,
2424 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2427 pr_err("%s: Cannot get array details from sysfs\n",
2431 min
= reshape
->min_offset_change
;
2433 if (info
->delta_disks
)
2434 sysfs_set_str(sra
, NULL
, "reshape_direction",
2435 info
->delta_disks
< 0 ? "backwards" : "forwards");
2436 if (info
->delta_disks
< 0 &&
2437 info
->space_after
< min
) {
2438 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2439 (sra
->component_size
-
2442 pr_err("cannot reduce component size\n");
2446 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
, data_offset
,
2449 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2450 cont_err("supported on this kernel\n");
2456 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2458 if (!err
&& sysfs_set_num(sra
, NULL
, "layout", reshape
->after
.layout
) < 0)
2460 if (!err
&& sysfs_set_num(sra
, NULL
, "raid_disks",
2461 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2463 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2466 pr_err("Cannot set array shape for %s\n",
2469 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2470 cont_err(" Bitmap must be removed before shape can be changed\n");
2480 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2482 struct mdinfo
*sra
, *sd
;
2483 /* Initialisation to silence compiler warning */
2484 unsigned long long min_space_before
= 0, min_space_after
= 0;
2487 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2490 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2493 struct supertype
*st2
;
2494 struct mdinfo info2
;
2496 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2498 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2499 dfd
= dev_open(dn
, O_RDONLY
);
2502 st2
= dup_super(st
);
2503 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2509 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2510 st2
->ss
->free_super(st2
);
2513 min_space_before
> info2
.space_before
)
2514 min_space_before
= info2
.space_before
;
2516 min_space_after
> info2
.space_after
)
2517 min_space_after
= info2
.space_after
;
2520 if (sd
== NULL
&& !first
) {
2521 info
->space_after
= min_space_after
;
2522 info
->space_before
= min_space_before
;
2527 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2528 struct mdinfo
*info
,
2529 int disks
, unsigned long long blocks
)
2531 /* Check that the internal stripe cache is
2532 * large enough, or it won't work.
2533 * It must hold at least 4 stripes of the larger
2536 unsigned long cache
;
2537 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2538 cache
*= 4; /* 4 stripes minimum */
2539 cache
/= 512; /* convert to sectors */
2540 /* make sure there is room for 'blocks' with a bit to spare */
2541 if (cache
< 16 + blocks
/ disks
)
2542 cache
= 16 + blocks
/ disks
;
2543 cache
/= (4096/512); /* Convert from sectors to pages */
2545 if (sra
->cache_size
< cache
)
2546 subarray_set_num(container
, sra
, "stripe_cache_size",
2550 static int impose_reshape(struct mdinfo
*sra
,
2551 struct mdinfo
*info
,
2552 struct supertype
*st
,
2555 char *devname
, char *container
,
2556 struct reshape
*reshape
)
2558 struct mdu_array_info_s array
;
2560 sra
->new_chunk
= info
->new_chunk
;
2563 /* for external metadata checkpoint saved by mdmon can be lost
2564 * or missed /due to e.g. crash/. Check if md is not during
2565 * restart farther than metadata points to.
2566 * If so, this means metadata information is obsolete.
2568 if (st
->ss
->external
)
2569 verify_reshape_position(info
, reshape
->level
);
2570 sra
->reshape_progress
= info
->reshape_progress
;
2572 sra
->reshape_progress
= 0;
2573 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2574 /* start from the end of the new array */
2575 sra
->reshape_progress
= (sra
->component_size
2576 * reshape
->after
.data_disks
);
2579 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2580 if (info
->array
.chunk_size
== info
->new_chunk
&&
2581 reshape
->before
.layout
== reshape
->after
.layout
&&
2582 st
->ss
->external
== 0) {
2583 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2584 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2586 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2589 pr_err("Cannot set device shape for %s: %s\n",
2590 devname
, strerror(errno
));
2593 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2594 cont_err("Bitmap must be removed before shape can be changed\n");
2598 } else if (!restart
) {
2599 /* set them all just in case some old 'new_*' value
2600 * persists from some earlier problem.
2603 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2605 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2606 reshape
->after
.layout
) < 0)
2608 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2609 reshape
->after
.data_disks
+
2610 reshape
->parity
) < 0)
2613 pr_err("Cannot set device shape for %s\n",
2617 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2618 cont_err("Bitmap must be removed before shape can be changed\n");
2627 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2630 struct mdu_array_info_s array
;
2632 sysfs_init(&info
, fd
, NULL
);
2634 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2636 (array
.level
>= 4 && array
.level
<= 6)) {
2637 /* To convert to RAID0 we need to fail and
2638 * remove any non-data devices. */
2641 int data_disks
= array
.raid_disks
- 1;
2642 if (array
.level
== 6)
2644 if (array
.level
== 5 &&
2645 array
.layout
!= ALGORITHM_PARITY_N
)
2647 if (array
.level
== 6 &&
2648 array
.layout
!= ALGORITHM_PARITY_N_6
)
2650 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2651 /* First remove any spares so no recovery starts */
2652 for (d
= 0, found
= 0;
2653 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2655 mdu_disk_info_t disk
;
2657 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2659 if (disk
.major
== 0 && disk
.minor
== 0)
2662 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2663 && disk
.raid_disk
< data_disks
)
2666 ioctl(fd
, HOT_REMOVE_DISK
,
2667 makedev(disk
.major
, disk
.minor
));
2669 /* Now fail anything left */
2670 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2671 for (d
= 0, found
= 0;
2672 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2675 mdu_disk_info_t disk
;
2677 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2679 if (disk
.major
== 0 && disk
.minor
== 0)
2682 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2683 && disk
.raid_disk
< data_disks
)
2686 ioctl(fd
, SET_DISK_FAULTY
,
2687 makedev(disk
.major
, disk
.minor
));
2689 while (ioctl(fd
, HOT_REMOVE_DISK
,
2690 makedev(disk
.major
, disk
.minor
)) < 0
2697 c
= map_num(pers
, level
);
2699 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2702 pr_err("%s: could not set level to %s\n",
2705 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2706 cont_err("Bitmap must be removed before level can be changed\n");
2710 pr_err("level of %s changed to %s\n",
2717 static void catch_term(int sig
)
2722 static int continue_via_systemd(char *devnm
)
2724 int skipped
, i
, pid
, status
;
2726 /* In a systemd/udev world, it is best to get systemd to
2727 * run "mdadm --grow --continue" rather than running in the
2732 /* FIXME yuk. CLOSE_EXEC?? */
2734 for (i
= 3; skipped
< 20; i
++)
2740 /* Don't want to see error messages from
2741 * systemctl. If the service doesn't exist,
2742 * we fork ourselves.
2745 open("/dev/null", O_WRONLY
);
2746 snprintf(pathbuf
, sizeof(pathbuf
), "mdadm-grow-continue@%s.service",
2748 status
= execl("/usr/bin/systemctl", "systemctl",
2751 status
= execl("/bin/systemctl", "systemctl", "start",
2754 case -1: /* Just do it ourselves. */
2756 default: /* parent - good */
2757 pid
= wait(&status
);
2758 if (pid
>= 0 && status
== 0)
2764 static int reshape_array(char *container
, int fd
, char *devname
,
2765 struct supertype
*st
, struct mdinfo
*info
,
2766 int force
, struct mddev_dev
*devlist
,
2767 unsigned long long data_offset
,
2768 char *backup_file
, int verbose
, int forked
,
2769 int restart
, int freeze_reshape
)
2771 struct reshape reshape
;
2774 int orig_level
= UnSet
;
2778 struct mdu_array_info_s array
;
2781 struct mddev_dev
*dv
;
2785 unsigned long long *offsets
= NULL
;
2789 unsigned long blocks
;
2790 unsigned long long array_size
;
2792 struct mdinfo
*sra
= NULL
;
2795 /* when reshaping a RAID0, the component_size might be zero.
2796 * So try to fix that up.
2798 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2799 dprintf("Cannot get array information.\n");
2802 if (array
.level
== 0 && info
->component_size
== 0) {
2803 get_dev_size(fd
, NULL
, &array_size
);
2804 info
->component_size
= array_size
/ array
.raid_disks
;
2807 if (array
.level
== 10)
2808 /* Need space_after info */
2809 get_space_after(fd
, st
, info
);
2811 if (info
->reshape_active
) {
2812 int new_level
= info
->new_level
;
2813 info
->new_level
= UnSet
;
2814 if (info
->delta_disks
> 0)
2815 info
->array
.raid_disks
-= info
->delta_disks
;
2816 msg
= analyse_change(devname
, info
, &reshape
);
2817 info
->new_level
= new_level
;
2818 if (info
->delta_disks
> 0)
2819 info
->array
.raid_disks
+= info
->delta_disks
;
2821 /* Make sure the array isn't read-only */
2822 ioctl(fd
, RESTART_ARRAY_RW
, 0);
2824 msg
= analyse_change(devname
, info
, &reshape
);
2826 /* if msg == "", error has already been printed */
2828 pr_err("%s\n", msg
);
2832 (reshape
.level
!= info
->array
.level
||
2833 reshape
.before
.layout
!= info
->array
.layout
||
2834 reshape
.before
.data_disks
+ reshape
.parity
2835 != info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
2836 pr_err("reshape info is not in native format - cannot continue.\n");
2840 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
2841 !((sysfs_get_str(info
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0) &&
2842 (strncmp(buf
, "reshape", 7) == 0))) {
2843 /* When reshape is restarted from '0', very begin of array
2844 * it is possible that for external metadata reshape and array
2845 * configuration doesn't happen.
2846 * Check if md has the same opinion, and reshape is restarted
2847 * from 0. If so, this is regular reshape start after reshape
2848 * switch in metadata to next array only.
2850 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
2851 (info
->reshape_progress
== 0))
2855 /* reshape already started. just skip to monitoring the reshape */
2856 if (reshape
.backup_blocks
== 0)
2858 if (restart
& RESHAPE_NO_BACKUP
)
2861 /* Need 'sra' down at 'started:' */
2862 sra
= sysfs_read(fd
, NULL
,
2863 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
2866 pr_err("%s: Cannot get array details from sysfs\n",
2872 backup_file
= locate_backup(sra
->sys_name
);
2876 /* The container is frozen but the array may not be.
2877 * So freeze the array so spares don't get put to the wrong use
2878 * FIXME there should probably be a cleaner separation between
2879 * freeze_array and freeze_container.
2881 sysfs_freeze_array(info
);
2882 /* Check we have enough spares to not be degraded */
2884 for (dv
= devlist
; dv
; dv
=dv
->next
)
2886 spares_needed
= max(reshape
.before
.data_disks
,
2887 reshape
.after
.data_disks
)
2888 + reshape
.parity
- array
.raid_disks
;
2891 info
->new_level
> 1 && info
->array
.level
> 1 &&
2892 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
2893 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2894 " Use --force to over-ride this check.\n",
2896 spares_needed
== 1 ? "" : "s",
2897 info
->array
.spare_disks
+ added_disks
);
2900 /* Check we have enough spares to not fail */
2901 spares_needed
= max(reshape
.before
.data_disks
,
2902 reshape
.after
.data_disks
)
2904 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
2905 spares_needed
> info
->array
.spare_disks
+added_disks
) {
2906 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2908 spares_needed
== 1 ? "" : "s",
2909 info
->array
.spare_disks
+ added_disks
);
2913 if (reshape
.level
!= array
.level
) {
2914 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
2917 info
->new_layout
= UnSet
; /* after level change,
2918 * layout is meaningless */
2919 orig_level
= array
.level
;
2920 sysfs_freeze_array(info
);
2922 if (reshape
.level
> 0 && st
->ss
->external
) {
2923 /* make sure mdmon is aware of the new level */
2924 if (mdmon_running(container
))
2925 flush_mdmon(container
);
2927 if (!mdmon_running(container
))
2928 start_mdmon(container
);
2929 ping_monitor(container
);
2930 if (mdmon_running(container
) &&
2931 st
->update_tail
== NULL
)
2932 st
->update_tail
= &st
->updates
;
2935 /* ->reshape_super might have chosen some spares from the
2936 * container that it wants to be part of the new array.
2937 * We can collect them with ->container_content and give
2938 * them to the kernel.
2940 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
2941 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2942 struct mdinfo
*info2
=
2943 st
->ss
->container_content(st
, subarray
);
2947 sysfs_init(info2
, fd
, st
->devnm
);
2948 /* When increasing number of devices, we need to set
2949 * new raid_disks before adding these, or they might
2952 if (reshape
.backup_blocks
&&
2953 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
2954 subarray_set_num(container
, info2
, "raid_disks",
2955 reshape
.after
.data_disks
+
2957 for (d
= info2
->devs
; d
; d
= d
->next
) {
2958 if (d
->disk
.state
== 0 &&
2959 d
->disk
.raid_disk
>= 0) {
2960 /* This is a spare that wants to
2961 * be part of the array.
2963 add_disk(fd
, st
, info2
, d
);
2969 /* We might have been given some devices to add to the
2970 * array. Now that the array has been changed to the right
2971 * level and frozen, we can safely add them.
2974 Manage_subdevs(devname
, fd
, devlist
, verbose
,
2977 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
2978 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
2979 if (reshape
.backup_blocks
== 0) {
2980 /* No restriping needed, but we might need to impose
2981 * some more changes: layout, raid_disks, chunk_size
2983 /* read current array info */
2984 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2985 dprintf("Cannot get array information.\n");
2988 /* compare current array info with new values and if
2989 * it is different update them to new */
2990 if (info
->new_layout
!= UnSet
&&
2991 info
->new_layout
!= array
.layout
) {
2992 array
.layout
= info
->new_layout
;
2993 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2994 pr_err("failed to set new layout\n");
2996 } else if (verbose
>= 0)
2997 printf("layout for %s set to %d\n",
2998 devname
, array
.layout
);
3000 if (info
->delta_disks
!= UnSet
&&
3001 info
->delta_disks
!= 0 &&
3002 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
3003 array
.raid_disks
+= info
->delta_disks
;
3004 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
3005 pr_err("failed to set raid disks\n");
3007 } else if (verbose
>= 0) {
3008 printf("raid_disks for %s set to %d\n",
3009 devname
, array
.raid_disks
);
3012 if (info
->new_chunk
!= 0 &&
3013 info
->new_chunk
!= array
.chunk_size
) {
3014 if (sysfs_set_num(info
, NULL
,
3015 "chunk_size", info
->new_chunk
) != 0) {
3016 pr_err("failed to set chunk size\n");
3018 } else if (verbose
>= 0)
3019 printf("chunk size for %s set to %d\n",
3020 devname
, array
.chunk_size
);
3027 * There are three possibilities.
3028 * 1/ The array will shrink.
3029 * We need to ensure the reshape will pause before reaching
3030 * the 'critical section'. We also need to fork and wait for
3031 * that to happen. When it does we
3032 * suspend/backup/complete/unfreeze
3034 * 2/ The array will not change size.
3035 * This requires that we keep a backup of a sliding window
3036 * so that we can restore data after a crash. So we need
3037 * to fork and monitor progress.
3038 * In future we will allow the data_offset to change, so
3039 * a sliding backup becomes unnecessary.
3041 * 3/ The array will grow. This is relatively easy.
3042 * However the kernel's restripe routines will cheerfully
3043 * overwrite some early data before it is safe. So we
3044 * need to make a backup of the early parts of the array
3045 * and be ready to restore it if rebuild aborts very early.
3046 * For externally managed metadata, we still need a forked
3047 * child to monitor the reshape and suspend IO over the region
3048 * that is being reshaped.
3050 * We backup data by writing it to one spare, or to a
3051 * file which was given on command line.
3053 * In each case, we first make sure that storage is available
3054 * for the required backup.
3056 * - request the shape change.
3057 * - fork to handle backup etc.
3059 /* Check that we can hold all the data */
3060 get_dev_size(fd
, NULL
, &array_size
);
3061 if (reshape
.new_size
< (array_size
/512)) {
3062 pr_err("this change will reduce the size of the array.\n"
3063 " use --grow --array-size first to truncate array.\n"
3064 " e.g. mdadm --grow %s --array-size %llu\n",
3065 devname
, reshape
.new_size
/2);
3069 if (array
.level
== 10) {
3070 /* Reshaping RAID10 does not require any data backup by
3071 * user-space. Instead it requires that the data_offset
3072 * is changed to avoid the need for backup.
3073 * So this is handled very separately
3076 /* Nothing to do. */
3078 return raid10_reshape(container
, fd
, devname
, st
, info
,
3079 &reshape
, data_offset
,
3082 sra
= sysfs_read(fd
, NULL
,
3083 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3086 pr_err("%s: Cannot get array details from sysfs\n",
3092 switch(set_new_data_offset(sra
, st
, devname
,
3093 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3095 reshape
.min_offset_change
, 1)) {
3099 /* Updated data_offset, so it's easy now */
3100 update_cache_size(container
, sra
, info
,
3101 min(reshape
.before
.data_disks
,
3102 reshape
.after
.data_disks
),
3103 reshape
.backup_blocks
);
3105 /* Right, everything seems fine. Let's kick things off.
3109 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3110 devname
, container
, &reshape
) < 0)
3112 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3114 if (errno
!= EINVAL
) {
3115 pr_err("Failed to initiate reshape!\n");
3118 /* revert data_offset and try the old way */
3119 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3120 sysfs_set_num(sra
, sd
, "new_offset",
3122 sysfs_set_str(sra
, NULL
, "reshape_direction",
3127 if (info
->new_level
== reshape
.level
)
3129 /* need to adjust level when reshape completes */
3131 case -1: /* ignore error, but don't wait */
3133 default: /* parent */
3141 fd
= open_dev(sra
->sys_name
);
3143 impose_level(fd
, info
->new_level
, devname
, verbose
);
3145 case 1: /* Couldn't set data_offset, try the old way */
3146 if (data_offset
!= INVALID_SECTORS
) {
3147 pr_err("Cannot update data_offset on this array\n");
3154 /* Decide how many blocks (sectors) for a reshape
3155 * unit. The number we have so far is just a minimum
3157 blocks
= reshape
.backup_blocks
;
3158 if (reshape
.before
.data_disks
==
3159 reshape
.after
.data_disks
) {
3160 /* Make 'blocks' bigger for better throughput, but
3161 * not so big that we reject it below.
3162 * Try for 16 megabytes
3164 while (blocks
* 32 < sra
->component_size
&&
3168 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3170 if (blocks
>= sra
->component_size
/2) {
3171 pr_err("%s: Something wrong - reshape aborted\n",
3176 /* Now we need to open all these devices so we can read/write.
3178 nrdisks
= max(reshape
.before
.data_disks
,
3179 reshape
.after
.data_disks
) + reshape
.parity
3180 + sra
->array
.spare_disks
;
3181 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3182 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3184 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3185 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
3186 nrdisks
, blocks
, backup_file
,
3191 if ((st
->ss
->manage_reshape
== NULL
) ||
3192 (st
->ss
->recover_backup
== NULL
)) {
3193 if (backup_file
== NULL
) {
3194 if (reshape
.after
.data_disks
<=
3195 reshape
.before
.data_disks
) {
3196 pr_err("%s: Cannot grow - need backup-file\n",
3198 pr_err(" Please provide one with \"--backup=...\"\n");
3200 } else if (d
== odisks
) {
3201 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3205 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3207 fdlist
+d
, offsets
+d
,
3216 update_cache_size(container
, sra
, info
,
3217 min(reshape
.before
.data_disks
, reshape
.after
.data_disks
),
3220 /* Right, everything seems fine. Let's kick things off.
3221 * If only changing raid_disks, use ioctl, else use
3226 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3227 devname
, container
, &reshape
) < 0)
3230 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3231 reshape
.after
.data_disks
);
3233 pr_err("Cannot %s reshape for %s\n",
3234 restart
? "continue" : "start",
3239 sysfs_set_str(sra
, NULL
, "array_state", "active");
3240 if (freeze_reshape
) {
3244 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3245 sra
->reshape_progress
);
3249 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3250 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3257 /* Now we just need to kick off the reshape and watch, while
3258 * handling backups of the data...
3259 * This is all done by a forked background process.
3261 switch(forked
? 0 : fork()) {
3263 pr_err("Cannot run child to monitor reshape: %s\n",
3277 /* If another array on the same devices is busy, the
3278 * reshape will wait for them. This would mean that
3279 * the first section that we suspend will stay suspended
3280 * for a long time. So check on that possibility
3281 * by looking for "DELAYED" in /proc/mdstat, and if found,
3285 struct mdstat_ent
*mds
, *m
;
3287 mds
= mdstat_read(1, 0);
3288 for (m
= mds
; m
; m
= m
->next
)
3289 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3291 m
->percent
== RESYNC_DELAYED
)
3294 /* Haven't started the reshape thread
3301 if (delayed
== 1 && get_linux_version() < 3007000) {
3302 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3303 " You might experience problems until other reshapes complete.\n");
3307 mdstat_wait(30 - (delayed
-1) * 25);
3311 if (check_env("MDADM_GROW_VERIFY"))
3312 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3315 mlockall(MCL_FUTURE
);
3317 signal(SIGTERM
, catch_term
);
3319 if (st
->ss
->external
) {
3320 /* metadata handler takes it from here */
3321 done
= st
->ss
->manage_reshape(
3322 fd
, sra
, &reshape
, st
, blocks
,
3324 d
- odisks
, fdlist
+odisks
,
3327 done
= child_monitor(
3328 fd
, sra
, &reshape
, st
, blocks
,
3330 d
- odisks
, fdlist
+odisks
,
3336 if (backup_file
&& done
) {
3338 bul
= make_backup(sra
->sys_name
);
3341 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3349 unlink(backup_file
);
3356 if (!st
->ss
->external
&&
3357 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
3358 && info
->custom_array_size
) &&
3359 info
->new_level
== reshape
.level
&&
3361 /* no need to wait for the reshape to finish as
3362 * there is nothing more to do.
3369 if (st
->ss
->external
) {
3370 /* Re-load the metadata as much could have changed */
3371 int cfd
= open_dev(st
->container_devnm
);
3373 flush_mdmon(container
);
3374 st
->ss
->free_super(st
);
3375 st
->ss
->load_container(st
, cfd
, container
);
3380 /* set new array size if required customer_array_size is used
3383 if (reshape
.before
.data_disks
!=
3384 reshape
.after
.data_disks
&&
3385 info
->custom_array_size
)
3386 set_array_size(st
, info
, info
->text_version
);
3388 if (info
->new_level
!= reshape
.level
) {
3390 fd
= open(devname
, O_RDONLY
);
3391 impose_level(fd
, info
->new_level
, devname
, verbose
);
3393 if (info
->new_level
== 0)
3394 st
->update_tail
= NULL
;
3406 if (orig_level
!= UnSet
&& sra
) {
3407 c
= map_num(pers
, orig_level
);
3408 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3409 pr_err("aborting level change\n");
3417 /* mdfd handle is passed to be closed in child process (after fork).
3419 int reshape_container(char *container
, char *devname
,
3421 struct supertype
*st
,
3422 struct mdinfo
*info
,
3424 char *backup_file
, int verbose
,
3425 int forked
, int restart
, int freeze_reshape
)
3427 struct mdinfo
*cc
= NULL
;
3429 char last_devnm
[32] = "";
3431 /* component_size is not meaningful for a container,
3432 * so pass '0' meaning 'no change'
3435 reshape_super(st
, 0, info
->new_level
,
3436 info
->new_layout
, info
->new_chunk
,
3437 info
->array
.raid_disks
, info
->delta_disks
,
3438 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3446 /* ping monitor to be sure that update is on disk
3448 ping_monitor(container
);
3450 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3451 if (continue_via_systemd(container
))
3454 switch (forked
? 0 : fork()) {
3455 case -1: /* error */
3456 perror("Cannot fork to complete reshape\n");
3459 default: /* parent */
3460 if (!freeze_reshape
)
3461 printf("%s: multi-array reshape continues in background\n", Name
);
3468 /* close unused handle in child process
3474 /* For each member array with reshape_active,
3475 * we need to perform the reshape.
3476 * We pick the first array that needs reshaping and
3477 * reshape it. reshape_array() will re-read the metadata
3478 * so the next time through a different array should be
3479 * ready for reshape.
3480 * It is possible that the 'different' array will not
3481 * be assembled yet. In that case we simple exit.
3482 * When it is assembled, the mdadm which assembles it
3483 * will take over the reshape.
3485 struct mdinfo
*content
;
3487 struct mdstat_ent
*mdstat
;
3493 cc
= st
->ss
->container_content(st
, NULL
);
3495 for (content
= cc
; content
; content
= content
->next
) {
3497 if (!content
->reshape_active
)
3500 subarray
= strchr(content
->text_version
+1, '/')+1;
3501 mdstat
= mdstat_by_subdev(subarray
, container
);
3504 if (mdstat
->active
== 0) {
3505 pr_err("Skipping inactive array %s.\n",
3507 free_mdstat(mdstat
);
3516 devid
= devnm2devid(mdstat
->devnm
);
3517 adev
= map_dev(major(devid
), minor(devid
), 0);
3519 adev
= content
->text_version
;
3521 fd
= open_dev(mdstat
->devnm
);
3523 pr_err("Device %s cannot be opened for reshape.\n", adev
);
3527 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3528 /* Do not allow for multiple reshape_array() calls for
3530 * It can happen when reshape_array() returns without
3531 * error, when reshape is not finished (wrong reshape
3532 * starting/continuation conditions). Mdmon doesn't
3533 * switch to next array in container and reentry
3534 * conditions for the same array occur.
3535 * This is possibly interim until the behaviour of
3536 * reshape_array is resolved().
3538 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3542 strcpy(last_devnm
, mdstat
->devnm
);
3544 sysfs_init(content
, fd
, mdstat
->devnm
);
3546 if (mdmon_running(container
))
3547 flush_mdmon(container
);
3549 rv
= reshape_array(container
, fd
, adev
, st
,
3550 content
, force
, NULL
, INVALID_SECTORS
,
3551 backup_file
, verbose
, 1, restart
,
3555 if (freeze_reshape
) {
3564 if (mdmon_running(container
))
3565 flush_mdmon(container
);
3574 * We run a child process in the background which performs the following
3576 * - wait for resync to reach a certain point
3577 * - suspend io to the following section
3578 * - backup that section
3579 * - allow resync to proceed further
3581 * - discard the backup.
3583 * When are combined in slightly different ways in the three cases.
3585 * - suspend/backup/allow/wait/resume/discard
3587 * - allow/wait/suspend/backup/allow/wait/resume/discard
3589 * - wait/resume/discard/suspend/backup/allow
3591 * suspend/backup/allow always come together
3592 * wait/resume/discard do too.
3593 * For the same-size case we have two backups to improve flow.
3597 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3598 unsigned long long backup_point
,
3599 unsigned long long wait_point
,
3600 unsigned long long *suspend_point
,
3601 unsigned long long *reshape_completed
, int *frozen
)
3603 /* This function is called repeatedly by the reshape manager.
3604 * It determines how much progress can safely be made and allows
3606 * - 'info' identifies the array and particularly records in
3607 * ->reshape_progress the metadata's knowledge of progress
3608 * This is a sector offset from the start of the array
3609 * of the next array block to be relocated. This number
3610 * may increase from 0 or decrease from array_size, depending
3611 * on the type of reshape that is happening.
3612 * Note that in contrast, 'sync_completed' is a block count of the
3613 * reshape so far. It gives the distance between the start point
3614 * (head or tail of device) and the next place that data will be
3615 * written. It always increases.
3616 * - 'reshape' is the structure created by analyse_change
3617 * - 'backup_point' shows how much the metadata manager has backed-up
3618 * data. For reshapes with increasing progress, it is the next address
3619 * to be backed up, previous addresses have been backed-up. For
3620 * decreasing progress, it is the earliest address that has been
3621 * backed up - later address are also backed up.
3622 * So addresses between reshape_progress and backup_point are
3623 * backed up providing those are in the 'correct' order.
3624 * - 'wait_point' is an array address. When reshape_completed
3625 * passes this point, progress_reshape should return. It might
3626 * return earlier if it determines that ->reshape_progress needs
3627 * to be updated or further backup is needed.
3628 * - suspend_point is maintained by progress_reshape and the caller
3629 * should not touch it except to initialise to zero.
3630 * It is an array address and it only increases in 2.6.37 and earlier.
3631 * This makes it difficult to handle reducing reshapes with
3632 * external metadata.
3633 * However: it is similar to backup_point in that it records the
3634 * other end of a suspended region from reshape_progress.
3635 * it is moved to extend the region that is safe to backup and/or
3637 * - reshape_completed is read from sysfs and returned. The caller
3638 * should copy this into ->reshape_progress when it has reason to
3639 * believe that the metadata knows this, and any backup outside this
3643 * 1 if more data from backup_point - but only as far as suspend_point,
3644 * should be backed up
3645 * 0 if things are progressing smoothly
3646 * -1 if the reshape is finished because it is all done,
3647 * -2 if the reshape is finished due to an error.
3650 int advancing
= (reshape
->after
.data_disks
3651 >= reshape
->before
.data_disks
);
3652 unsigned long long need_backup
; /* All data between start of array and
3653 * here will at some point need to
3656 unsigned long long read_offset
, write_offset
;
3657 unsigned long long write_range
;
3658 unsigned long long max_progress
, target
, completed
;
3659 unsigned long long array_size
= (info
->component_size
3660 * reshape
->before
.data_disks
);
3664 /* First, we unsuspend any region that is now known to be safe.
3665 * If suspend_point is on the 'wrong' side of reshape_progress, then
3666 * we don't have or need suspension at the moment. This is true for
3667 * native metadata when we don't need to back-up.
3670 if (info
->reshape_progress
<= *suspend_point
)
3671 sysfs_set_num(info
, NULL
, "suspend_lo",
3672 info
->reshape_progress
);
3674 /* Note: this won't work in 2.6.37 and before.
3675 * Something somewhere should make sure we don't need it!
3677 if (info
->reshape_progress
>= *suspend_point
)
3678 sysfs_set_num(info
, NULL
, "suspend_hi",
3679 info
->reshape_progress
);
3682 /* Now work out how far it is safe to progress.
3683 * If the read_offset for ->reshape_progress is less than
3684 * 'blocks' beyond the write_offset, we can only progress as far
3686 * Otherwise we can progress until the write_offset for the new location
3687 * reaches (within 'blocks' of) the read_offset at the current location.
3688 * However that region must be suspended unless we are using native
3690 * If we need to suspend more, we limit it to 128M per device, which is
3691 * rather arbitrary and should be some time-based calculation.
3693 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3694 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3695 write_range
= info
->new_chunk
/512;
3696 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3697 need_backup
= array_size
;
3699 need_backup
= reshape
->backup_blocks
;
3701 if (read_offset
< write_offset
+ write_range
)
3702 max_progress
= backup_point
;
3706 reshape
->after
.data_disks
;
3708 if (read_offset
> write_offset
- write_range
)
3709 /* Can only progress as far as has been backed up,
3710 * which must be suspended */
3711 max_progress
= backup_point
;
3712 else if (info
->reshape_progress
<= need_backup
)
3713 max_progress
= backup_point
;
3715 if (info
->array
.major_version
>= 0)
3716 /* Can progress until backup is needed */
3717 max_progress
= need_backup
;
3719 /* Can progress until metadata update is required */
3722 reshape
->after
.data_disks
;
3723 /* but data must be suspended */
3724 if (max_progress
< *suspend_point
)
3725 max_progress
= *suspend_point
;
3730 /* We know it is safe to progress to 'max_progress' providing
3731 * it is suspended or we are using native metadata.
3732 * Consider extending suspend_point 128M per device if it
3733 * is less than 64M per device beyond reshape_progress.
3734 * But always do a multiple of 'blocks'
3735 * FIXME this is too big - it takes to long to complete
3738 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3739 reshape
->after
.data_disks
);
3740 target
/= reshape
->backup_blocks
;
3743 target
*= reshape
->backup_blocks
;
3745 /* For externally managed metadata we always need to suspend IO to
3746 * the area being reshaped so we regularly push suspend_point forward.
3747 * For native metadata we only need the suspend if we are going to do
3751 if ((need_backup
> info
->reshape_progress
3752 || info
->array
.major_version
< 0) &&
3753 *suspend_point
< info
->reshape_progress
+ target
) {
3754 if (need_backup
< *suspend_point
+ 2 * target
)
3755 *suspend_point
= need_backup
;
3756 else if (*suspend_point
+ 2 * target
< array_size
)
3757 *suspend_point
+= 2 * target
;
3759 *suspend_point
= array_size
;
3760 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
3761 if (max_progress
> *suspend_point
)
3762 max_progress
= *suspend_point
;
3765 if (info
->array
.major_version
>= 0) {
3766 /* Only need to suspend when about to backup */
3767 if (info
->reshape_progress
< need_backup
* 2 &&
3768 *suspend_point
> 0) {
3770 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
3771 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
3774 /* Need to suspend continually */
3775 if (info
->reshape_progress
< *suspend_point
)
3776 *suspend_point
= info
->reshape_progress
;
3777 if (*suspend_point
+ target
< info
->reshape_progress
)
3778 /* No need to move suspend region yet */;
3780 if (*suspend_point
>= 2 * target
)
3781 *suspend_point
-= 2 * target
;
3784 sysfs_set_num(info
, NULL
, "suspend_lo",
3787 if (max_progress
< *suspend_point
)
3788 max_progress
= *suspend_point
;
3792 /* now set sync_max to allow that progress. sync_max, like
3793 * sync_completed is a count of sectors written per device, so
3794 * we find the difference between max_progress and the start point,
3795 * and divide that by after.data_disks to get a sync_max
3797 * At the same time we convert wait_point to a similar number
3798 * for comparing against sync_completed.
3800 /* scale down max_progress to per_disk */
3801 max_progress
/= reshape
->after
.data_disks
;
3802 /* Round to chunk size as some kernels give an erroneously high number */
3803 max_progress
/= info
->new_chunk
/512;
3804 max_progress
*= info
->new_chunk
/512;
3805 /* And round to old chunk size as the kernel wants that */
3806 max_progress
/= info
->array
.chunk_size
/512;
3807 max_progress
*= info
->array
.chunk_size
/512;
3808 /* Limit progress to the whole device */
3809 if (max_progress
> info
->component_size
)
3810 max_progress
= info
->component_size
;
3811 wait_point
/= reshape
->after
.data_disks
;
3813 /* switch from 'device offset' to 'processed block count' */
3814 max_progress
= info
->component_size
- max_progress
;
3815 wait_point
= info
->component_size
- wait_point
;
3819 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
3821 /* Now wait. If we have already reached the point that we were
3822 * asked to wait to, don't wait at all, else wait for any change.
3823 * We need to select on 'sync_completed' as that is the place that
3824 * notifications happen, but we are really interested in
3825 * 'reshape_position'
3827 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
3829 goto check_progress
;
3831 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3832 goto check_progress
;
3834 while (completed
< max_progress
&& completed
< wait_point
) {
3835 /* Check that sync_action is still 'reshape' to avoid
3836 * waiting forever on a dead array
3839 if (sysfs_get_str(info
, NULL
, "sync_action",
3841 strncmp(action
, "reshape", 7) != 0)
3843 /* Some kernels reset 'sync_completed' to zero
3844 * before setting 'sync_action' to 'idle'.
3845 * So we need these extra tests.
3847 if (completed
== 0 && advancing
3848 && strncmp(action
, "idle", 4) == 0
3849 && info
->reshape_progress
> 0)
3851 if (completed
== 0 && !advancing
3852 && strncmp(action
, "idle", 4) == 0
3853 && info
->reshape_progress
< (info
->component_size
3854 * reshape
->after
.data_disks
))
3856 sysfs_wait(fd
, NULL
);
3857 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3858 goto check_progress
;
3860 /* Some kernels reset 'sync_completed' to zero,
3861 * we need to have real point we are in md.
3862 * So in that case, read 'reshape_position' from sysfs.
3864 if (completed
== 0) {
3865 unsigned long long reshapep
;
3867 if (sysfs_get_str(info
, NULL
, "sync_action",
3869 strncmp(action
, "idle", 4) == 0 &&
3870 sysfs_get_ll(info
, NULL
,
3871 "reshape_position", &reshapep
) == 0)
3872 *reshape_completed
= reshapep
;
3874 /* some kernels can give an incorrectly high
3875 * 'completed' number, so round down */
3876 completed
/= (info
->new_chunk
/512);
3877 completed
*= (info
->new_chunk
/512);
3878 /* Convert 'completed' back in to a 'progress' number */
3879 completed
*= reshape
->after
.data_disks
;
3881 completed
= (info
->component_size
3882 * reshape
->after
.data_disks
3884 *reshape_completed
= completed
;
3889 /* We return the need_backup flag. Caller will decide
3890 * how much - a multiple of ->backup_blocks up to *suspend_point
3893 return need_backup
> info
->reshape_progress
;
3895 return need_backup
>= info
->reshape_progress
;
3898 /* if we couldn't read a number from sync_completed, then
3899 * either the reshape did complete, or it aborted.
3900 * We can tell which by checking for 'none' in reshape_position.
3901 * If it did abort, then it might immediately restart if it
3902 * it was just a device failure that leaves us degraded but
3905 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
3906 || strncmp(buf
, "none", 4) != 0) {
3907 /* The abort might only be temporary. Wait up to 10
3908 * seconds for fd to contain a valid number again.
3912 unsigned long long new_sync_max
;
3913 while (fd
>= 0 && rv
< 0 && wait
> 0) {
3914 if (sysfs_wait(fd
, &wait
) != 1)
3916 switch (sysfs_fd_get_ll(fd
, &completed
)) {
3918 /* all good again */
3920 /* If "sync_max" is no longer max_progress
3921 * we need to freeze things
3923 sysfs_get_ll(info
, NULL
, "sync_max", &new_sync_max
);
3924 *frozen
= (new_sync_max
!= max_progress
);
3926 case -2: /* read error - abort */
3933 return rv
; /* abort */
3935 /* Maybe racing with array shutdown - check state */
3938 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
3939 || strncmp(buf
, "inactive", 8) == 0
3940 || strncmp(buf
, "clear",5) == 0)
3941 return -2; /* abort */
3942 return -1; /* complete */
3946 /* FIXME return status is never checked */
3947 static int grow_backup(struct mdinfo
*sra
,
3948 unsigned long long offset
, /* per device */
3949 unsigned long stripes
, /* per device, in old chunks */
3950 int *sources
, unsigned long long *offsets
,
3951 int disks
, int chunk
, int level
, int layout
,
3952 int dests
, int *destfd
, unsigned long long *destoffsets
,
3953 int part
, int *degraded
,
3956 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3957 * to storage 'destfd' (offset 'destoffsets'), after first
3958 * suspending IO. Then allow resync to continue
3959 * over the suspended section.
3960 * Use part 'part' of the backup-super-block.
3965 unsigned long long ll
;
3967 //printf("offset %llu\n", offset);
3973 /* Check that array hasn't become degraded, else we might backup the wrong data */
3974 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
3975 return -1; /* FIXME this error is ignored */
3976 new_degraded
= (int)ll
;
3977 if (new_degraded
!= *degraded
) {
3978 /* check each device to ensure it is still working */
3980 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3981 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
3983 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
3985 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
3986 strstr(sbuf
, "faulty") ||
3987 strstr(sbuf
, "in_sync") == NULL
) {
3988 /* this device is dead */
3989 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
3990 if (sd
->disk
.raid_disk
>= 0 &&
3991 sources
[sd
->disk
.raid_disk
] >= 0) {
3992 close(sources
[sd
->disk
.raid_disk
]);
3993 sources
[sd
->disk
.raid_disk
] = -1;
3998 *degraded
= new_degraded
;
4001 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4002 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4004 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4005 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4008 bsb
.magic
[15] = '2';
4009 for (i
= 0; i
< dests
; i
++)
4011 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
4013 lseek64(destfd
[i
], destoffsets
[i
], 0);
4015 rv
= save_stripes(sources
, offsets
,
4016 disks
, chunk
, level
, layout
,
4018 offset
*512*odata
, stripes
* chunk
* odata
,
4023 bsb
.mtime
= __cpu_to_le64(time(0));
4024 for (i
= 0; i
< dests
; i
++) {
4025 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4027 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4028 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4029 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4030 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4033 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
4034 != destoffsets
[i
] - 4096)
4036 if (write(destfd
[i
], &bsb
, 512) != 512)
4038 if (destoffsets
[i
] > 4096) {
4039 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4040 destoffsets
[i
]+stripes
*chunk
*odata
)
4042 if (write(destfd
[i
], &bsb
, 512) != 512)
4052 /* in 2.6.30, the value reported by sync_completed can be
4053 * less that it should be by one stripe.
4054 * This only happens when reshape hits sync_max and pauses.
4055 * So allow wait_backup to either extent sync_max further
4056 * than strictly necessary, or return before the
4057 * sync has got quite as far as we would really like.
4058 * This is what 'blocks2' is for.
4059 * The various caller give appropriate values so that
4062 /* FIXME return value is often ignored */
4063 static int forget_backup(int dests
, int *destfd
,
4064 unsigned long long *destoffsets
,
4068 * Erase backup 'part' (which is 0 or 1)
4074 bsb
.arraystart2
= __cpu_to_le64(0);
4075 bsb
.length2
= __cpu_to_le64(0);
4077 bsb
.arraystart
= __cpu_to_le64(0);
4078 bsb
.length
= __cpu_to_le64(0);
4080 bsb
.mtime
= __cpu_to_le64(time(0));
4082 for (i
= 0; i
< dests
; i
++) {
4083 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4084 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4085 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4086 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4087 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4088 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4089 destoffsets
[i
]-4096)
4092 write(destfd
[i
], &bsb
, 512) != 512)
4099 static void fail(char *msg
)
4102 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4103 rv
|= (write(2, "\n", 1) != 1);
4107 static char *abuf
, *bbuf
;
4108 static unsigned long long abuflen
;
4109 static void validate(int afd
, int bfd
, unsigned long long offset
)
4111 /* check that the data in the backup against the array.
4112 * This is only used for regression testing and should not
4113 * be used while the array is active
4117 lseek64(bfd
, offset
- 4096, 0);
4118 if (read(bfd
, &bsb2
, 512) != 512)
4119 fail("cannot read bsb");
4120 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4121 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4122 fail("first csum bad");
4123 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4124 fail("magic is bad");
4125 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4126 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4127 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4128 fail("second csum bad");
4130 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4131 fail("devstart is wrong");
4134 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4136 if (abuflen
< len
) {
4140 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4141 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4143 /* just stop validating on mem-alloc failure */
4148 lseek64(bfd
, offset
, 0);
4149 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4150 //printf("len %llu\n", len);
4151 fail("read first backup failed");
4153 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4154 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4155 fail("read first from array failed");
4156 if (memcmp(bbuf
, abuf
, len
) != 0) {
4159 printf("offset=%llu len=%llu\n",
4160 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4161 for (i
=0; i
<len
; i
++)
4162 if (bbuf
[i
] != abuf
[i
]) {
4163 printf("first diff byte %d\n", i
);
4167 fail("data1 compare failed");
4171 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4173 if (abuflen
< len
) {
4177 abuf
= xmalloc(abuflen
);
4178 bbuf
= xmalloc(abuflen
);
4181 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4182 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4183 fail("read second backup failed");
4184 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4185 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4186 fail("read second from array failed");
4187 if (memcmp(bbuf
, abuf
, len
) != 0)
4188 fail("data2 compare failed");
4192 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4193 struct supertype
*st
, unsigned long blocks
,
4194 int *fds
, unsigned long long *offsets
,
4195 int dests
, int *destfd
, unsigned long long *destoffsets
)
4197 /* Monitor a reshape where backup is being performed using
4198 * 'native' mechanism - either to a backup file, or
4199 * to some space in a spare.
4203 unsigned long long speed
;
4204 unsigned long long suspend_point
, array_size
;
4205 unsigned long long backup_point
, wait_point
;
4206 unsigned long long reshape_completed
;
4208 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
4209 int part
= 0; /* The next part of the backup area to fill. It may already
4210 * be full, so we need to check */
4211 int level
= reshape
->level
;
4212 int layout
= reshape
->before
.layout
;
4213 int data
= reshape
->before
.data_disks
;
4214 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4215 int chunk
= sra
->array
.chunk_size
;
4217 unsigned long stripes
;
4221 /* set up the backup-super-block. This requires the
4222 * uuid from the array.
4224 /* Find a superblock */
4225 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4229 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4231 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4232 devfd
= dev_open(dn
, O_RDONLY
);
4235 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4241 pr_err("Cannot find a superblock\n");
4245 memset(&bsb
, 0, 512);
4246 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4247 st
->ss
->uuid_from_super(st
, uuid
);
4248 memcpy(bsb
.set_uuid
, uuid
, 16);
4249 bsb
.mtime
= __cpu_to_le64(time(0));
4250 bsb
.devstart2
= blocks
;
4252 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4253 reshape
->before
.data_disks
;
4255 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4256 /* Don't start the 'reshape' */
4258 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4259 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4260 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4264 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4265 backup_point
= sra
->reshape_progress
;
4268 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4269 backup_point
= reshape
->backup_blocks
;
4270 suspend_point
= array_size
;
4276 /* Want to return as soon the oldest backup slot can
4277 * be released as that allows us to start backing up
4278 * some more, providing suspend_point has been
4279 * advanced, which it should have.
4282 wait_point
= array_size
;
4283 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4284 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4285 __le64_to_cpu(bsb
.length
));
4286 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4287 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4288 __le64_to_cpu(bsb
.length2
));
4291 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4292 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4293 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4294 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4297 reshape_completed
= sra
->reshape_progress
;
4298 rv
= progress_reshape(sra
, reshape
,
4299 backup_point
, wait_point
,
4300 &suspend_point
, &reshape_completed
,
4302 /* external metadata would need to ping_monitor here */
4303 sra
->reshape_progress
= reshape_completed
;
4305 /* Clear any backup region that is before 'here' */
4307 if (__le64_to_cpu(bsb
.length
) > 0 &&
4308 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4309 __le64_to_cpu(bsb
.length
)))
4310 forget_backup(dests
, destfd
,
4312 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4313 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4314 __le64_to_cpu(bsb
.length2
)))
4315 forget_backup(dests
, destfd
,
4318 if (__le64_to_cpu(bsb
.length
) > 0 &&
4319 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4320 forget_backup(dests
, destfd
,
4322 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4323 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4324 forget_backup(dests
, destfd
,
4334 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4335 /* No longer need to monitor this reshape */
4336 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4342 unsigned long long offset
;
4343 unsigned long actual_stripes
;
4344 /* Need to backup some data.
4345 * If 'part' is not used and the desired
4346 * backup size is suspended, do a backup,
4347 * then consider the next part.
4349 /* Check that 'part' is unused */
4350 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4352 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4355 offset
= backup_point
/ data
;
4356 actual_stripes
= stripes
;
4358 if (offset
+ actual_stripes
* (chunk
/512) >
4359 sra
->component_size
)
4360 actual_stripes
= ((sra
->component_size
- offset
)
4362 if (offset
+ actual_stripes
* (chunk
/512) >
4366 if (offset
< actual_stripes
* (chunk
/512))
4367 actual_stripes
= offset
/ (chunk
/512);
4368 offset
-= actual_stripes
* (chunk
/512);
4369 if (offset
< suspend_point
/data
)
4372 if (actual_stripes
== 0)
4374 grow_backup(sra
, offset
, actual_stripes
,
4376 disks
, chunk
, level
, layout
,
4377 dests
, destfd
, destoffsets
,
4378 part
, °raded
, buf
);
4379 validate(afd
, destfd
[0], destoffsets
[0]);
4380 /* record where 'part' is up to */
4383 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4385 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4389 /* FIXME maybe call progress_reshape one more time instead */
4390 /* remove any remaining suspension */
4391 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4392 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4393 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4394 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4396 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4397 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4403 * If any spare contains md_back_data-1 which is recent wrt mtime,
4404 * write that data into the array and update the super blocks with
4405 * the new reshape_progress
4407 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
4408 char *backup_file
, int verbose
)
4412 unsigned long long *offsets
;
4413 unsigned long long nstripe
, ostripe
;
4416 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4417 if (info
->array
.level
== 6) odata
--; /* number of data disks */
4418 ndata
= info
->array
.raid_disks
- 1;
4419 if (info
->new_level
== 6) ndata
--;
4421 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4423 if (info
->delta_disks
<= 0)
4424 /* Didn't grow, so the backup file must have
4428 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4429 struct mdinfo dinfo
;
4432 char *devname
, namebuf
[20];
4433 unsigned long long lo
, hi
;
4435 /* This was a spare and may have some saved data on it.
4436 * Load the superblock, find and load the
4437 * backup_super_block.
4438 * If either fail, go on to next device.
4439 * If the backup contains no new info, just return
4440 * else restore data and update all superblocks
4442 if (i
== old_disks
-1) {
4443 fd
= open(backup_file
, O_RDONLY
);
4445 pr_err("backup file %s inaccessible: %s\n",
4446 backup_file
, strerror(errno
));
4449 devname
= backup_file
;
4454 if (st
->ss
->load_super(st
, fd
, NULL
))
4457 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4458 st
->ss
->free_super(st
);
4461 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4463 pr_err("Cannot seek on device %d\n", i
);
4464 continue; /* Cannot seek */
4466 sprintf(namebuf
, "device-%d", i
);
4469 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4471 pr_err("Cannot read from %s\n", devname
);
4472 continue; /* Cannot read */
4474 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4475 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4477 pr_err("No backup metadata on %s\n", devname
);
4480 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4482 pr_err("Bad backup-metadata checksum on %s\n", devname
);
4483 continue; /* bad checksum */
4485 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4486 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4488 pr_err("Bad backup-metadata checksum2 on %s\n", devname
);
4489 continue; /* Bad second checksum */
4491 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4493 pr_err("Wrong uuid on backup-metadata on %s\n", devname
);
4494 continue; /* Wrong uuid */
4497 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4498 * sometimes they aren't... So allow considerable flexability in matching, and allow
4499 * this test to be overridden by an environment variable.
4501 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
4502 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
4503 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4504 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4505 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4506 (unsigned long)info
->array
.utime
);
4508 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4509 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4510 continue; /* time stamp is too bad */
4514 if (bsb
.magic
[15] == '1') {
4515 if (bsb
.length
== 0)
4517 if (info
->delta_disks
>= 0) {
4518 /* reshape_progress is increasing */
4519 if (__le64_to_cpu(bsb
.arraystart
)
4520 + __le64_to_cpu(bsb
.length
)
4521 < info
->reshape_progress
) {
4524 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4525 continue; /* No new data here */
4528 /* reshape_progress is decreasing */
4529 if (__le64_to_cpu(bsb
.arraystart
) >=
4530 info
->reshape_progress
)
4531 goto nonew
; /* No new data here */
4534 if (bsb
.length
== 0 && bsb
.length2
== 0)
4536 if (info
->delta_disks
>= 0) {
4537 /* reshape_progress is increasing */
4538 if ((__le64_to_cpu(bsb
.arraystart
)
4539 + __le64_to_cpu(bsb
.length
)
4540 < info
->reshape_progress
)
4542 (__le64_to_cpu(bsb
.arraystart2
)
4543 + __le64_to_cpu(bsb
.length2
)
4544 < info
->reshape_progress
))
4545 goto nonew
; /* No new data here */
4547 /* reshape_progress is decreasing */
4548 if (__le64_to_cpu(bsb
.arraystart
) >=
4549 info
->reshape_progress
&&
4550 __le64_to_cpu(bsb
.arraystart2
) >=
4551 info
->reshape_progress
)
4552 goto nonew
; /* No new data here */
4555 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4558 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4560 continue; /* Cannot seek */
4562 /* There should be a duplicate backup superblock 4k before here */
4563 if (lseek64(fd
, -4096, 1) < 0 ||
4564 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4565 goto second_fail
; /* Cannot find leading superblock */
4566 if (bsb
.magic
[15] == '1')
4567 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4569 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4570 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4571 goto second_fail
; /* Cannot find leading superblock */
4573 /* Now need the data offsets for all devices. */
4574 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4575 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4578 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4579 /* FIXME should be this be an error */
4581 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4582 st
->ss
->free_super(st
);
4583 offsets
[j
] = dinfo
.data_offset
* 512;
4585 printf("%s: restoring critical section\n", Name
);
4587 if (restore_stripes(fdlist
, offsets
,
4588 info
->array
.raid_disks
,
4592 fd
, __le64_to_cpu(bsb
.devstart
)*512,
4593 __le64_to_cpu(bsb
.arraystart
)*512,
4594 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4595 /* didn't succeed, so giveup */
4597 pr_err("Error restoring backup from %s\n",
4603 if (bsb
.magic
[15] == '2' &&
4604 restore_stripes(fdlist
, offsets
,
4605 info
->array
.raid_disks
,
4609 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
4610 __le64_to_cpu(bsb
.devstart2
)*512,
4611 __le64_to_cpu(bsb
.arraystart2
)*512,
4612 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4613 /* didn't succeed, so giveup */
4615 pr_err("Error restoring second backup from %s\n",
4623 /* Ok, so the data is restored. Let's update those superblocks. */
4627 lo
= __le64_to_cpu(bsb
.arraystart
);
4628 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4630 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4631 unsigned long long lo1
, hi1
;
4632 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4633 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4637 } else if (lo
< lo1
)
4643 (info
->reshape_progress
< lo
||
4644 info
->reshape_progress
> hi
))
4645 /* backup does not affect reshape_progress*/ ;
4646 else if (info
->delta_disks
>= 0) {
4647 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4648 __le64_to_cpu(bsb
.length
);
4649 if (bsb
.magic
[15] == '2') {
4650 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
4651 __le64_to_cpu(bsb
.length2
);
4652 if (p2
> info
->reshape_progress
)
4653 info
->reshape_progress
= p2
;
4656 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4657 if (bsb
.magic
[15] == '2') {
4658 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
4659 if (p2
< info
->reshape_progress
)
4660 info
->reshape_progress
= p2
;
4663 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4666 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4668 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4669 dinfo
.reshape_progress
= info
->reshape_progress
;
4670 st
->ss
->update_super(st
, &dinfo
,
4671 "_reshape_progress",
4673 st
->ss
->store_super(st
, fdlist
[j
]);
4674 st
->ss
->free_super(st
);
4678 /* Didn't find any backup data, try to see if any
4681 if (info
->delta_disks
< 0) {
4682 /* When shrinking, the critical section is at the end.
4683 * So see if we are before the critical section.
4685 unsigned long long first_block
;
4686 nstripe
= ostripe
= 0;
4688 while (ostripe
>= nstripe
) {
4689 ostripe
+= info
->array
.chunk_size
/ 512;
4690 first_block
= ostripe
* odata
;
4691 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4692 (info
->new_chunk
/512);
4695 if (info
->reshape_progress
>= first_block
)
4698 if (info
->delta_disks
> 0) {
4699 /* See if we are beyond the critical section. */
4700 unsigned long long last_block
;
4701 nstripe
= ostripe
= 0;
4703 while (nstripe
>= ostripe
) {
4704 nstripe
+= info
->new_chunk
/ 512;
4705 last_block
= nstripe
* ndata
;
4706 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4707 (info
->array
.chunk_size
/512);
4710 if (info
->reshape_progress
>= last_block
)
4713 /* needed to recover critical section! */
4715 pr_err("Failed to find backup of critical section\n");
4719 int Grow_continue_command(char *devname
, int fd
,
4720 char *backup_file
, int verbose
)
4723 struct supertype
*st
= NULL
;
4724 struct mdinfo
*content
= NULL
;
4725 struct mdinfo array
;
4726 char *subarray
= NULL
;
4727 struct mdinfo
*cc
= NULL
;
4728 struct mdstat_ent
*mdstat
= NULL
;
4732 dprintf("Grow continue from command line called for %s\n",
4735 st
= super_by_fd(fd
, &subarray
);
4736 if (!st
|| !st
->ss
) {
4737 pr_err("Unable to determine metadata format for %s\n",
4741 dprintf("Grow continue is run for ");
4742 if (st
->ss
->external
== 0) {
4744 dprintf_cont("native array (%s)\n", devname
);
4745 if (ioctl(fd
, GET_ARRAY_INFO
, &array
.array
) < 0) {
4746 pr_err("%s is not an active md array - aborting\n", devname
);
4748 goto Grow_continue_command_exit
;
4751 /* Need to load a superblock.
4752 * FIXME we should really get what we need from
4755 for (d
= 0; d
< MAX_DISKS
; d
++) {
4756 mdu_disk_info_t disk
;
4760 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
4762 if (disk
.major
== 0 && disk
.minor
== 0)
4764 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
4766 dv
= map_dev(disk
.major
, disk
.minor
, 1);
4769 fd2
= dev_open(dv
, O_RDONLY
);
4772 err
= st
->ss
->load_super(st
, fd2
, NULL
);
4774 /* invalidate fd2 to avoid possible double close() */
4780 if (d
== MAX_DISKS
) {
4781 pr_err("Unable to load metadata for %s\n",
4784 goto Grow_continue_command_exit
;
4786 st
->ss
->getinfo_super(st
, content
, NULL
);
4791 dprintf_cont("subarray (%s)\n", subarray
);
4792 container
= st
->container_devnm
;
4793 cfd
= open_dev_excl(st
->container_devnm
);
4795 container
= st
->devnm
;
4797 cfd
= open_dev_excl(st
->devnm
);
4798 dprintf_cont("container (%s)\n", container
);
4802 pr_err("Unable to open container for %s\n", devname
);
4804 goto Grow_continue_command_exit
;
4807 /* find in container array under reshape
4809 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
4811 pr_err("Cannot read superblock for %s\n",
4814 goto Grow_continue_command_exit
;
4817 cc
= st
->ss
->container_content(st
, subarray
);
4818 for (content
= cc
; content
; content
= content
->next
) {
4820 int allow_reshape
= 1;
4822 if (content
->reshape_active
== 0)
4824 /* The decision about array or container wide
4825 * reshape is taken in Grow_continue based
4826 * content->reshape_active state, therefore we
4827 * need to check_reshape based on
4828 * reshape_active and subarray name
4830 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
4832 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
4833 (content
->array
.state
4834 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
4837 if (!allow_reshape
) {
4838 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4839 devname
, container
);
4841 goto Grow_continue_command_exit
;
4844 array
= strchr(content
->text_version
+1, '/')+1;
4845 mdstat
= mdstat_by_subdev(array
, container
);
4848 if (mdstat
->active
== 0) {
4849 pr_err("Skipping inactive array %s.\n",
4851 free_mdstat(mdstat
);
4858 pr_err("Unable to determine reshaped array for %s\n", devname
);
4860 goto Grow_continue_command_exit
;
4862 fd2
= open_dev(mdstat
->devnm
);
4864 pr_err("cannot open (%s)\n", mdstat
->devnm
);
4866 goto Grow_continue_command_exit
;
4869 sysfs_init(content
, fd2
, mdstat
->devnm
);
4871 /* start mdmon in case it is not running
4873 if (!mdmon_running(container
))
4874 start_mdmon(container
);
4875 ping_monitor(container
);
4877 if (mdmon_running(container
))
4878 st
->update_tail
= &st
->updates
;
4880 pr_err("No mdmon found. Grow cannot continue.\n");
4882 goto Grow_continue_command_exit
;
4886 /* verify that array under reshape is started from
4889 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
4891 goto Grow_continue_command_exit
;
4896 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
4898 Grow_continue_command_exit
:
4903 st
->ss
->free_super(st
);
4904 free_mdstat(mdstat
);
4911 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
4912 char *backup_file
, int forked
, int freeze_reshape
)
4916 if (!info
->reshape_active
)
4919 if (st
->ss
->external
) {
4920 int cfd
= open_dev(st
->container_devnm
);
4925 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
4927 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
4928 st
, info
, 0, backup_file
,
4930 1 | info
->reshape_active
,
4933 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
4934 NULL
, INVALID_SECTORS
,
4935 backup_file
, 0, forked
,
4936 1 | info
->reshape_active
,
4942 char *make_backup(char *name
)
4944 char *base
= "backup_file-";
4948 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
4949 fname
= xmalloc(len
);
4950 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
4954 char *locate_backup(char *name
)
4956 char *fl
= make_backup(name
);
4959 if (stat(fl
, &stb
) == 0 &&
4960 S_ISREG(stb
.st_mode
))