2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 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>
28 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
29 #error no endian defined
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
38 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
40 /* Add a device to an active array.
41 * Currently, just extend a linear array.
42 * This requires writing a new superblock on the
43 * new device, calling the kernel to add the device,
44 * and if that succeeds, update the superblock on
46 * This means that we need to *find* all other devices.
53 struct supertype
*st
= NULL
;
54 char *subarray
= NULL
;
56 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
57 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
61 if (info
.array
.level
!= -1) {
62 fprintf(stderr
, Name
": can only add devices to linear arrays\n");
66 st
= super_by_fd(fd
, &subarray
);
68 fprintf(stderr
, Name
": cannot handle arrays with superblock version %d\n", info
.array
.major_version
);
73 fprintf(stderr
, Name
": Cannot grow linear sub-arrays yet\n");
78 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
80 fprintf(stderr
, Name
": cannot open %s\n", newdev
);
85 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
86 fprintf(stderr
, Name
": %s is not a block device!\n", newdev
);
91 /* now check out all the devices and make sure we can read the superblock */
92 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
96 st
->ss
->free_super(st
);
99 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
100 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
106 dv
= map_dev(disk
.major
, disk
.minor
, 1);
108 fprintf(stderr
, Name
": cannot find device file for device %d\n",
114 fd2
= dev_open(dv
, O_RDWR
);
116 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
122 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
123 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
131 /* Ok, looks good. Lets update the superblock and write it out to
135 info
.disk
.number
= d
;
136 info
.disk
.major
= major(stb
.st_rdev
);
137 info
.disk
.minor
= minor(stb
.st_rdev
);
138 info
.disk
.raid_disk
= d
;
139 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
140 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
143 if (st
->ss
->store_super(st
, nfd
)) {
144 fprintf(stderr
, Name
": Cannot store new superblock on %s\n",
151 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
152 fprintf(stderr
, Name
": Cannot add new disk to this array\n");
155 /* Well, that seems to have worked.
156 * Now go through and update all superblocks
159 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
160 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
165 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
166 mdu_disk_info_t disk
;
170 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
171 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
175 dv
= map_dev(disk
.major
, disk
.minor
, 1);
177 fprintf(stderr
, Name
": cannot find device file for device %d\n",
181 fd2
= dev_open(dv
, O_RDWR
);
183 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
186 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
187 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
191 info
.array
.raid_disks
= nd
+1;
192 info
.array
.nr_disks
= nd
+1;
193 info
.array
.active_disks
= nd
+1;
194 info
.array
.working_disks
= nd
+1;
196 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
199 if (st
->ss
->store_super(st
, fd2
)) {
200 fprintf(stderr
, Name
": Cannot store new superblock on %s\n", dv
);
210 int Grow_addbitmap(char *devname
, int fd
, char *file
, int chunk
, int delay
, int write_behind
, int force
)
213 * First check that array doesn't have a bitmap
214 * Then create the bitmap
217 * For internal bitmaps, we need to check the version,
218 * find all the active devices, and write the bitmap block
221 mdu_bitmap_file_t bmf
;
222 mdu_array_info_t array
;
223 struct supertype
*st
;
224 char *subarray
= NULL
;
225 int major
= BITMAP_MAJOR_HI
;
226 int vers
= md_get_version(fd
);
227 unsigned long long bitmapsize
, array_size
;
230 major
= BITMAP_MAJOR_HOSTENDIAN
;
231 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel"
232 " are not portable\n"
233 " between different architectures. Consider upgrading"
234 " the Linux kernel.\n");
237 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
239 fprintf(stderr
, Name
": Memory allocation failure.\n");
241 fprintf(stderr
, Name
": bitmaps not supported by this kernel.\n");
244 if (bmf
.pathname
[0]) {
245 if (strcmp(file
,"none")==0) {
246 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
247 fprintf(stderr
, Name
": failed to remove bitmap %s\n",
253 fprintf(stderr
, Name
": %s already has a bitmap (%s)\n",
254 devname
, bmf
.pathname
);
257 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
258 fprintf(stderr
, Name
": cannot get array status for %s\n", devname
);
261 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
262 if (strcmp(file
, "none")==0) {
263 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
264 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
265 fprintf(stderr
, Name
": failed to remove internal bitmap.\n");
270 fprintf(stderr
, Name
": Internal bitmap already present on %s\n",
275 if (strcmp(file
, "none") == 0) {
276 fprintf(stderr
, Name
": no bitmap found on %s\n", devname
);
279 if (array
.level
<= 0) {
280 fprintf(stderr
, Name
": Bitmaps not meaningful with level %s\n",
281 map_num(pers
, array
.level
)?:"of this array");
284 bitmapsize
= array
.size
;
286 if (get_dev_size(fd
, NULL
, &array_size
) &&
287 array_size
> (0x7fffffffULL
<<9)) {
288 /* Array is big enough that we cannot trust array.size
289 * try other approaches
291 bitmapsize
= get_component_size(fd
);
293 if (bitmapsize
== 0) {
294 fprintf(stderr
, Name
": Cannot reliably determine size of array to create bitmap - sorry.\n");
298 if (array
.level
== 10) {
299 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
300 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
303 st
= super_by_fd(fd
, &subarray
);
305 fprintf(stderr
, Name
": Cannot understand version %d.%d\n",
306 array
.major_version
, array
.minor_version
);
310 fprintf(stderr
, Name
": Cannot add bitmaps to sub-arrays yet\n");
315 if (strcmp(file
, "internal") == 0) {
317 if (st
->ss
->add_internal_bitmap
== NULL
) {
318 fprintf(stderr
, Name
": Internal bitmaps not supported "
319 "with %s metadata\n", st
->ss
->name
);
322 for (d
=0; d
< st
->max_devs
; d
++) {
323 mdu_disk_info_t disk
;
326 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
328 if (disk
.major
== 0 &&
331 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
333 dv
= map_dev(disk
.major
, disk
.minor
, 1);
335 int fd2
= dev_open(dv
, O_RDWR
);
338 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
339 if (st
->ss
->add_internal_bitmap(
341 &chunk
, delay
, write_behind
,
342 bitmapsize
, 0, major
)
344 st
->ss
->write_bitmap(st
, fd2
);
346 fprintf(stderr
, Name
": failed to create internal bitmap - chunksize problem.\n");
354 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
355 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
358 ": Cannot add bitmap while array is"
359 " resyncing or reshaping etc.\n");
360 fprintf(stderr
, Name
": failed to set internal bitmap.\n");
367 int max_devs
= st
->max_devs
;
369 /* try to load a superblock */
370 for (d
=0; d
<max_devs
; d
++) {
371 mdu_disk_info_t disk
;
375 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
377 if ((disk
.major
==0 && disk
.minor
==0) ||
378 (disk
.state
& (1<<MD_DISK_REMOVED
)))
380 dv
= map_dev(disk
.major
, disk
.minor
, 1);
382 fd2
= dev_open(dv
, O_RDONLY
);
384 st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
386 st
->ss
->uuid_from_super(st
, uuid
);
392 fprintf(stderr
, Name
": cannot find UUID for array!\n");
395 if (CreateBitmap(file
, force
, (char*)uuid
, chunk
,
396 delay
, write_behind
, bitmapsize
, major
)) {
399 bitmap_fd
= open(file
, O_RDWR
);
401 fprintf(stderr
, Name
": weird: %s cannot be opened\n",
405 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
409 ": Cannot add bitmap while array is"
410 " resyncing or reshaping etc.\n");
411 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
412 devname
, strerror(err
));
422 * When reshaping an array we might need to backup some data.
423 * This is written to all spares with a 'super_block' describing it.
424 * The superblock goes 4K from the end of the used space on the
426 * It if written after the backup is complete.
427 * It has the following structure.
430 static struct mdp_backup_super
{
431 char magic
[16]; /* md_backup_data-1 or -2 */
434 /* start/sizes in 512byte sectors */
435 __u64 devstart
; /* address on backup device/file of data */
438 __u32 sb_csum
; /* csum of preceeding bytes. */
440 __u64 devstart2
; /* offset in to data of second section */
443 __u32 sb_csum2
; /* csum of preceeding bytes. */
445 } __attribute__((aligned(512))) bsb
, bsb2
;
447 static __u32
bsb_csum(char *buf
, int len
)
451 for (i
=0; i
<len
; i
++)
452 csum
= (csum
<<3) + buf
[0];
453 return __cpu_to_le32(csum
);
456 static int check_idle(struct supertype
*st
)
458 /* Check that all member arrays for this container, or the
459 * container of this array, are idle
461 int container_dev
= (st
->container_dev
!= NoMdDev
462 ? st
->container_dev
: st
->devnum
);
464 struct mdstat_ent
*ent
, *e
;
467 fmt_devname(container
, container_dev
);
468 ent
= mdstat_read(0, 0);
469 for (e
= ent
; e
; e
= e
->next
) {
470 if (!is_container_member(e
, container
))
472 if (e
->percent
>= 0) {
481 static int freeze_container(struct supertype
*st
)
483 int container_dev
= (st
->container_dev
!= NoMdDev
484 ? st
->container_dev
: st
->devnum
);
490 fmt_devname(container
, container_dev
);
492 if (block_monitor(container
, 1)) {
493 fprintf(stderr
, Name
": failed to freeze container\n");
500 static void unfreeze_container(struct supertype
*st
)
502 int container_dev
= (st
->container_dev
!= NoMdDev
503 ? st
->container_dev
: st
->devnum
);
506 fmt_devname(container
, container_dev
);
508 unblock_monitor(container
, 1);
511 static int freeze(struct supertype
*st
)
513 /* Try to freeze resync/rebuild on this array/container.
514 * Return -1 if the array is busy,
515 * return -2 container cannot be frozen,
516 * return 0 if this kernel doesn't support 'frozen'
517 * return 1 if it worked.
519 if (st
->ss
->external
)
520 return freeze_container(st
);
522 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
527 err
= sysfs_freeze_array(sra
);
533 static void unfreeze(struct supertype
*st
)
535 if (st
->ss
->external
)
536 return unfreeze_container(st
);
538 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
541 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
543 fprintf(stderr
, Name
": failed to unfreeze array\n");
548 static void wait_reshape(struct mdinfo
*sra
)
550 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
556 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
557 strncmp(action
, "reshape", 7) == 0) {
561 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
566 static int reshape_super(struct supertype
*st
, long long size
, int level
,
567 int layout
, int chunksize
, int raid_disks
,
568 int delta_disks
, char *backup_file
, char *dev
,
571 /* nothing extra to check in the native case */
572 if (!st
->ss
->external
)
574 if (!st
->ss
->reshape_super
||
575 !st
->ss
->manage_reshape
) {
576 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
581 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
582 raid_disks
, delta_disks
, backup_file
, dev
,
586 static void sync_metadata(struct supertype
*st
)
588 if (st
->ss
->external
) {
589 if (st
->update_tail
) {
590 flush_metadata_updates(st
);
591 st
->update_tail
= &st
->updates
;
593 st
->ss
->sync_metadata(st
);
597 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
599 /* when dealing with external metadata subarrays we need to be
600 * prepared to handle EAGAIN. The kernel may need to wait for
601 * mdmon to mark the array active so the kernel can handle
602 * allocations/writeback when preparing the reshape action
603 * (md_allow_write()). We temporarily disable safe_mode_delay
604 * to close a race with the array_state going clean before the
605 * next write to raid_disks / stripe_cache_size
610 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
612 (strcmp(name
, "raid_disks") != 0 &&
613 strcmp(name
, "stripe_cache_size") != 0))
614 return sysfs_set_num(sra
, NULL
, name
, n
);
616 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
619 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
620 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
621 if (rc
< 0 && errno
== EAGAIN
) {
622 ping_monitor(container
);
623 /* if we get EAGAIN here then the monitor is not active
626 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
628 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
632 int start_reshape(struct mdinfo
*sra
, int already_running
)
635 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
636 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
637 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
638 if (!already_running
)
639 sysfs_set_num(sra
, NULL
, "sync_min", 0);
640 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", 0);
641 if (!already_running
)
642 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
647 void abort_reshape(struct mdinfo
*sra
)
649 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
650 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
651 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
652 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
653 sysfs_set_num(sra
, NULL
, "sync_min", 0);
654 sysfs_set_str(sra
, NULL
, "sync_max", "max");
657 int remove_disks_for_takeover(struct supertype
*st
,
662 struct mdinfo
*remaining
;
665 if (sra
->array
.level
== 10)
666 nr_of_copies
= layout
& 0xff;
667 else if (sra
->array
.level
== 1)
668 nr_of_copies
= sra
->array
.raid_disks
;
672 remaining
= sra
->devs
;
674 /* for each 'copy', select one device and remove from the list. */
675 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
676 struct mdinfo
**diskp
;
679 /* Find a working device to keep */
680 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
681 struct mdinfo
*disk
= *diskp
;
683 if (disk
->disk
.raid_disk
< slot
)
685 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
687 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
689 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
691 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
694 /* We have found a good disk to use! */
696 disk
->next
= sra
->devs
;
705 if (slot
< sra
->array
.raid_disks
) {
706 /* didn't find all slots */
712 sra
->devs
= remaining
;
716 /* Remove all 'remaining' devices from the array */
718 struct mdinfo
*sd
= remaining
;
719 remaining
= sd
->next
;
721 sysfs_set_str(sra
, sd
, "state", "faulty");
722 sysfs_set_str(sra
, sd
, "slot", "none");
723 /* for external metadata disks should be removed in mdmon */
724 if (!st
->ss
->external
)
725 sysfs_set_str(sra
, sd
, "state", "remove");
726 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
727 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
728 sd
->next
= sra
->devs
;
734 void reshape_free_fdlist(int *fdlist
,
735 unsigned long long *offsets
,
740 for (i
= 0; i
< size
; i
++)
748 int reshape_prepare_fdlist(char *devname
,
752 unsigned long blocks
,
755 unsigned long long *offsets
)
760 for (d
= 0; d
<= nrdisks
; d
++)
763 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
764 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
766 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
767 char *dn
= map_dev(sd
->disk
.major
,
769 fdlist
[sd
->disk
.raid_disk
]
770 = dev_open(dn
, O_RDONLY
);
771 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
772 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
774 Name
": %s: cannot open component %s\n",
775 devname
, dn
? dn
: "-unknown-");
779 } else if (backup_file
== NULL
) {
781 char *dn
= map_dev(sd
->disk
.major
,
783 fdlist
[d
] = dev_open(dn
, O_RDWR
);
784 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
786 fprintf(stderr
, Name
": %s: cannot open component %s\n",
787 devname
, dn
? dn
: "-unknown-");
798 int reshape_open_backup_file(char *backup_file
,
803 unsigned long long *offsets
,
806 /* Return 1 on success, 0 on any form of failure */
807 /* need to check backup file is large enough */
813 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
817 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
818 devname
, backup_file
, strerror(errno
));
821 /* Guard against backup file being on array device.
822 * If array is partitioned or if LVM etc is in the
823 * way this will not notice, but it is better than
826 fstat(*fdlist
, &stb
);
829 if (stb
.st_rdev
== dev
) {
830 fprintf(stderr
, Name
": backup file must NOT be"
831 " on the array being reshaped.\n");
837 for (i
=0; i
< blocks
+ 8 ; i
++) {
838 if (write(*fdlist
, buf
, 512) != 512) {
839 fprintf(stderr
, Name
": %s: cannot create"
840 " backup file %s: %s\n",
841 devname
, backup_file
, strerror(errno
));
845 if (fsync(*fdlist
) != 0) {
846 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
847 devname
, backup_file
, strerror(errno
));
854 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
855 unsigned int ndata
, unsigned int odata
)
857 unsigned long a
, b
, blocks
;
858 /* So how much do we need to backup.
859 * We need an amount of data which is both a whole number of
860 * old stripes and a whole number of new stripes.
861 * So LCM for (chunksize*datadisks).
863 a
= (ochunk
/512) * odata
;
864 b
= (nchunk
/512) * ndata
;
872 /* LCM == product / GCD */
873 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
878 char *analyse_change(struct mdinfo
*info
, struct reshape
*re
)
880 /* Based on the current array state in info->array and
881 * the changes in info->new_* etc, determine:
882 * - whether the change is possible
883 * - Intermediate level/raid_disks/layout
884 * - whether a restriping reshape is needed
885 * - number of sectors in minimum change unit. This
886 * will cover a whole number of stripes in 'before' and
889 * Return message if the change should be rejected
890 * NULL if the change can be achieved
892 * This can be called as part of starting a reshape, or
893 * when assembling an array that is undergoing reshape.
896 /* delta_parity records change in number of devices
897 * caused by level change
899 int delta_parity
= 0;
901 /* If a new level not explicitly given, we assume no-change */
902 if (info
->new_level
== UnSet
)
903 info
->new_level
= info
->array
.level
;
906 switch (info
->new_level
) {
912 /* chunk size is meaningful, must divide component_size
915 if (info
->component_size
% (info
->new_chunk
/512))
916 return "New chunk size does not"
917 " divide component size";
920 return "chunk size not meaningful for this level";
923 info
->new_chunk
= info
->array
.chunk_size
;
925 switch (info
->array
.level
) {
927 /* RAID1 can convert to RAID1 with different disks, or
928 * raid5 with 2 disks, or
931 if (info
->new_level
== 0) {
932 if (info
->delta_disks
!= UnSet
&&
933 info
->delta_disks
!= 0)
934 return "Cannot change number of disks "
935 "with RAID1->RAID0 conversion";
937 re
->before
.data_disks
= 1;
938 re
->after
.data_disks
= 1;
939 re
->before
.layout
= 0;
940 re
->backup_blocks
= 0;
944 if (info
->new_level
== 1) {
945 if (info
->delta_disks
== UnSet
)
946 /* Don't know what to do */
947 return "no change requested for Growing RAID1";
949 re
->backup_blocks
= 0;
953 if (info
->array
.raid_disks
== 2 &&
954 info
->new_level
== 5) {
955 if (info
->delta_disks
!= UnSet
&&
956 info
->delta_disks
!= 0)
957 return "Cannot change number of disks "
958 "with RAID1->RAID5 conversion";
960 re
->before
.data_disks
= 1;
961 re
->after
.data_disks
= 1;
962 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
963 info
->array
.chunk_size
= 65536;
966 /* Could do some multi-stage conversions, but leave that to
969 return "Impossibly level change request for RAID1";
972 /* RAID10 can only be converted from near mode to
973 * RAID0 by removing some devices
975 if ((info
->array
.layout
& ~0xff) != 0x100)
976 return "Cannot Grow RAID10 with far/offset layout";
977 /* number of devices must be multiple of number of copies */
978 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
979 return "RAID10 layout too complex for Grow operation";
981 if (info
->new_level
!= 0)
982 return "RAID10 can only be changed to RAID0";
983 new_disks
= (info
->array
.raid_disks
984 / (info
->array
.layout
& 0xff));
985 if (info
->delta_disks
== UnSet
)
986 info
->delta_disks
= (new_disks
987 - info
->array
.raid_disks
);
989 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
990 return "New number of raid-devices impossible for RAID10";
991 if (info
->new_chunk
&&
992 info
->new_chunk
!= info
->array
.chunk_size
)
993 return "Cannot change chunk-size with RAID10 Grow";
998 re
->before
.data_disks
= new_disks
;
999 re
->after
.data_disks
= re
->before
.data_disks
;
1000 re
->before
.layout
= 0;
1001 re
->backup_blocks
= 0;
1005 /* RAID0 can be converted to RAID10, or to RAID456 */
1006 if (info
->new_level
== 10) {
1007 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1008 /* Assume near=2 layout */
1009 info
->new_layout
= 0x102;
1010 info
->delta_disks
= info
->array
.raid_disks
;
1012 if (info
->new_layout
== UnSet
) {
1013 int copies
= 1 + (info
->delta_disks
1014 / info
->array
.raid_disks
);
1015 if (info
->array
.raid_disks
* (copies
-1)
1016 != info
->delta_disks
)
1017 return "Impossible number of devices"
1018 " for RAID0->RAID10";
1019 info
->new_layout
= 0x100 + copies
;
1021 if (info
->delta_disks
== UnSet
) {
1022 int copies
= info
->new_layout
& 0xff;
1023 if (info
->new_layout
!= 0x100 + copies
)
1024 return "New layout impossible"
1025 " for RAID0->RAID10";;
1026 info
->delta_disks
= (copies
- 1) *
1027 info
->array
.raid_disks
;
1029 if (info
->new_chunk
&&
1030 info
->new_chunk
!= info
->array
.chunk_size
)
1031 return "Cannot change chunk-size with RAID0->RAID10";
1035 re
->before
.data_disks
= (info
->array
.raid_disks
+
1037 re
->after
.data_disks
= re
->before
.data_disks
;
1038 re
->before
.layout
= info
->new_layout
;
1039 re
->backup_blocks
= 0;
1043 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1044 * a raid4 style layout of the final level.
1046 switch (info
->new_level
) {
1051 re
->before
.layout
= 0;
1056 re
->before
.layout
= ALGORITHM_PARITY_N
;
1061 re
->before
.layout
= ALGORITHM_PARITY_N
;
1064 return "Impossible level change requested";
1066 re
->before
.data_disks
= info
->array
.raid_disks
;
1067 /* determining 'after' layout happens outside this 'switch' */
1071 info
->array
.layout
= ALGORITHM_PARITY_N
;
1073 switch (info
->new_level
) {
1077 re
->level
= info
->array
.level
;
1078 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1079 re
->before
.layout
= info
->array
.layout
;
1083 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1084 re
->before
.layout
= info
->array
.layout
;
1089 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1090 switch (info
->array
.layout
) {
1091 case ALGORITHM_LEFT_ASYMMETRIC
:
1092 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1094 case ALGORITHM_RIGHT_ASYMMETRIC
:
1095 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1097 case ALGORITHM_LEFT_SYMMETRIC
:
1098 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1100 case ALGORITHM_RIGHT_SYMMETRIC
:
1101 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1103 case ALGORITHM_PARITY_0
:
1104 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1106 case ALGORITHM_PARITY_N
:
1107 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1110 return "Cannot convert an array with this layout";
1114 if (info
->array
.raid_disks
!= 2)
1115 return "Can only convert a 2-device array to RAID1";
1116 if (info
->delta_disks
!= UnSet
&&
1117 info
->delta_disks
!= 0)
1118 return "Cannot set raid_disk when "
1119 "converting RAID5->RAID1";
1123 return "Impossible level change requested";
1127 switch (info
->new_level
) {
1133 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1134 re
->before
.layout
= info
->array
.layout
;
1137 return "Impossible level change requested";
1142 /* If we reached here then it looks like a re-stripe is
1143 * happening. We have determined the intermediate level
1144 * and initial raid_disks/layout and stored these in 're'.
1146 * We need to deduce the final layout that can be atomically
1147 * converted to the end state.
1149 switch (info
->new_level
) {
1151 /* We can only get to RAID0 from RAID4 or RAID5
1152 * with appropriate layout and one extra device
1154 if (re
->level
!= 4 && re
->level
!= 5)
1155 return "Cannot covert to RAID0 from this level";
1157 switch (re
->level
) {
1159 re
->after
.layout
= 0 ; break;
1161 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1166 /* We can only get to RAID4 from RAID5 */
1167 if (re
->level
!= 4 && re
->level
!= 5)
1168 return "Cannot convert to RAID4 from this level";
1170 switch (re
->level
) {
1172 re
->after
.layout
= 0 ; break;
1174 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1179 /* We get to RAID5 for RAID5 or RAID6 */
1180 if (re
->level
!= 5 && re
->level
!= 6)
1181 return "Cannot convert to RAID5 from this level";
1183 switch (re
->level
) {
1185 if (info
->new_layout
== UnSet
)
1186 re
->after
.layout
= re
->before
.layout
;
1188 re
->after
.layout
= info
->new_layout
;
1191 if (info
->new_layout
== UnSet
)
1192 info
->new_layout
= re
->before
.layout
;
1194 /* after.layout needs to be raid6 version of new_layout */
1195 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1196 re
->after
.layout
= ALGORITHM_PARITY_N
;
1199 char *ls
= map_num(r5layout
, info
->new_layout
);
1201 strcat(strcpy(layout
, ls
), "-6");
1202 l
= map_name(r6layout
, layout
);
1204 return "Cannot find RAID6 layout"
1206 re
->after
.layout
= l
;
1212 /* We must already be at level 6 */
1214 return "Impossible level change";
1215 if (info
->new_layout
== UnSet
)
1216 re
->after
.layout
= info
->array
.layout
;
1218 re
->after
.layout
= info
->new_layout
;
1221 return "Impossible level change requested";
1223 if (info
->delta_disks
== UnSet
)
1224 info
->delta_disks
= delta_parity
;
1226 re
->after
.data_disks
= (re
->before
.data_disks
1229 switch (re
->level
) {
1230 case 6: re
->parity
= 2; break;
1232 case 5: re
->parity
= 1; break;
1233 default: re
->parity
= 0; break;
1235 /* So we have a restripe operation, we need to calculate the number
1236 * of blocks per reshape operation.
1238 if (info
->new_chunk
== 0)
1239 info
->new_chunk
= info
->array
.chunk_size
;
1240 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1241 re
->after
.layout
== re
->before
.layout
&&
1242 info
->new_chunk
== info
->array
.chunk_size
) {
1243 /* Nothing to change */
1244 re
->backup_blocks
= 0;
1247 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1248 /* chunk and layout changes make no difference */
1249 re
->backup_blocks
= 0;
1253 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1254 get_linux_version() < 2006032)
1255 return "in-place reshape is not safe before 2.6.32 - sorry.";
1257 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1258 get_linux_version() < 2006030)
1259 return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
1261 re
->backup_blocks
= compute_backup_blocks(
1262 info
->new_chunk
, info
->array
.chunk_size
,
1263 re
->after
.data_disks
,
1264 re
->before
.data_disks
);
1266 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1270 static int reshape_array(char *container
, int fd
, char *devname
,
1271 struct supertype
*st
, struct mdinfo
*info
,
1272 int force
, char *backup_file
, int quiet
, int forked
,
1274 static int reshape_container(char *container
, int cfd
, char *devname
,
1275 struct supertype
*st
,
1276 struct mdinfo
*info
,
1281 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1283 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1286 /* Make some changes in the shape of an array.
1287 * The kernel must support the change.
1289 * There are three different changes. Each can trigger
1290 * a resync or recovery so we freeze that until we have
1291 * requested everything (if kernel supports freezing - 2.6.30).
1293 * - change size (i.e. component_size)
1295 * - change layout/chunksize/ndisks
1297 * The last can require a reshape. It is different on different
1298 * levels so we need to check the level before actioning it.
1299 * Some times the level change needs to be requested after the
1300 * reshape (e.g. raid6->raid5, raid5->raid0)
1303 struct mdu_array_info_s array
;
1305 struct supertype
*st
;
1306 char *subarray
= NULL
;
1310 char *container
= NULL
;
1311 char container_buf
[20];
1317 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1318 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1324 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1325 fprintf(stderr
, Name
": cannot change component size at the same time "
1326 "as other changes.\n"
1327 " Change size first, then check data is intact before "
1328 "making other changes.\n");
1332 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1333 get_linux_version() < 2006032 &&
1334 !check_env("MDADM_FORCE_FEWER")) {
1335 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1336 " Please use a newer kernel\n");
1340 st
= super_by_fd(fd
, &subarray
);
1342 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1345 if (raid_disks
> st
->max_devs
) {
1346 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1347 " beyond %d\n", st
->max_devs
);
1351 /* in the external case we need to check that the requested reshape is
1352 * supported, and perform an initial check that the container holds the
1353 * pre-requisite spare devices (mdmon owns final validation)
1355 if (st
->ss
->external
) {
1360 container_dev
= st
->container_dev
;
1361 cfd
= open_dev_excl(st
->container_dev
);
1363 container_dev
= st
->devnum
;
1365 cfd
= open_dev_excl(st
->devnum
);
1369 fprintf(stderr
, Name
": Unable to open container for %s\n",
1375 fmt_devname(container_buf
, container_dev
);
1376 container
= container_buf
;
1378 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1381 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1387 if (mdmon_running(container_dev
))
1388 st
->update_tail
= &st
->updates
;
1391 if (raid_disks
> array
.raid_disks
&&
1392 array
.spare_disks
< (raid_disks
- array
.raid_disks
) &&
1395 Name
": Need %d spare%s to avoid degraded array,"
1396 " and only have %d.\n"
1397 " Use --force to over-ride this check.\n",
1398 raid_disks
- array
.raid_disks
,
1399 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1404 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
1405 | GET_STATE
| GET_VERSION
);
1407 if (st
->ss
->external
&& subarray
== NULL
) {
1408 array
.level
= LEVEL_CONTAINER
;
1409 sra
->array
.level
= LEVEL_CONTAINER
;
1412 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1416 frozen
= freeze(st
);
1418 /* freeze() already spewed the reason */
1420 } else if (frozen
< 0) {
1421 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1422 " be reshaped\n", devname
);
1426 /* ========= set size =============== */
1427 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1428 long long orig_size
= array
.size
;
1430 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1437 if (array
.size
!= size
) {
1438 /* got truncated to 32bit, write to
1439 * component_size instead
1442 rv
= sysfs_set_num(sra
, NULL
,
1443 "component_size", size
);
1447 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1451 /* restore metadata */
1452 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1453 UnSet
, NULL
, devname
, !quiet
) == 0)
1455 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1456 devname
, strerror(err
));
1458 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1459 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1463 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1464 size
= get_component_size(fd
)/2;
1468 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
1471 } else if (array
.level
!= LEVEL_CONTAINER
) {
1472 size
= get_component_size(fd
)/2;
1477 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1478 * current implementation assumes that following conditions must be met:
1481 * - near_copies == 2
1483 if ((level
== 0 && array
.level
== 10 && sra
&&
1484 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1485 (level
== 0 && array
.level
== 1 && sra
)) {
1487 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1489 dprintf(Name
": Array cannot be reshaped\n");
1495 /* FIXME this is added with no justification - why is it here */
1496 ping_monitor(container
);
1500 sysfs_init(&info
, fd
, NoMdDev
);
1501 strcpy(info
.text_version
, sra
->text_version
);
1502 info
.component_size
= size
*2;
1503 info
.new_level
= level
;
1504 info
.new_chunk
= chunksize
* 1024;
1506 info
.delta_disks
= raid_disks
- info
.array
.raid_disks
;
1508 info
.delta_disks
= UnSet
;
1509 if (layout_str
== NULL
) {
1510 info
.new_layout
= UnSet
;
1511 if (info
.array
.level
== 6 &&
1512 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1513 info
.array
.layout
>= 16) {
1514 fprintf(stderr
, Name
1515 ": %s has a non-standard layout. If you"
1516 " wish to preserve this\n"
1517 " during the reshape, please specify"
1518 " --layout=preserve\n"
1519 " If you want to change it, specify a"
1520 " layout or use --layout=normalise\n",
1525 } else if (strcmp(layout_str
, "normalise") == 0 ||
1526 strcmp(layout_str
, "normalize") == 0) {
1527 /* If we have a -6 RAID6 layout, remove the '-6'. */
1528 info
.new_layout
= UnSet
;
1529 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1531 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1532 h
= strrchr(l
, '-');
1533 if (h
&& strcmp(h
, "-6") == 0) {
1535 info
.new_layout
= map_name(r6layout
, l
);
1538 } else if (strcmp(layout_str
, "preserve") == 0) {
1539 info
.new_layout
= UnSet
;
1541 int l
= info
.new_level
;
1543 l
= info
.array
.level
;
1546 info
.new_layout
= map_name(r5layout
, layout_str
);
1549 info
.new_layout
= map_name(r6layout
, layout_str
);
1552 info
.new_layout
= parse_layout_10(layout_str
);
1555 info
.new_layout
= parse_layout_faulty(layout_str
);
1558 fprintf(stderr
, Name
": layout not meaningful"
1559 " with this level\n");
1563 if (info
.new_layout
== UnSet
) {
1564 fprintf(stderr
, Name
": layout %s not understood"
1565 " for this level\n",
1572 if (array
.level
== LEVEL_CONTAINER
) {
1573 /* This change is to be applied to every array in the
1574 * container. This is only needed when the metadata imposes
1575 * restraints of the various arrays in the container.
1576 * Currently we only know that IMSM requires all arrays
1577 * to have the same number of devices so changing the
1578 * number of devices (On-Line Capacity Expansion) must be
1579 * performed at the level of the container
1581 rv
= reshape_container(container
, fd
, devname
, st
, &info
,
1582 force
, backup_file
, quiet
);
1585 /* get spare devices from external metadata
1587 if (st
->ss
->external
) {
1588 struct mdinfo
*info2
;
1590 info2
= st
->ss
->container_content(st
, subarray
);
1592 info
.array
.spare_disks
=
1593 info2
->array
.spare_disks
;
1598 /* Impose these changes on a single array. First
1599 * check that the metadata is OK with the change. */
1601 if (reshape_super(st
, info
.component_size
, info
.new_level
,
1602 info
.new_layout
, info
.new_chunk
,
1603 info
.array
.raid_disks
, info
.delta_disks
,
1604 backup_file
, devname
, quiet
)) {
1609 rv
= reshape_array(container
, fd
, devname
, st
, &info
, force
,
1610 backup_file
, quiet
, 0, 0);
1619 static int reshape_array(char *container
, int fd
, char *devname
,
1620 struct supertype
*st
, struct mdinfo
*info
,
1622 char *backup_file
, int quiet
, int forked
,
1625 struct reshape reshape
;
1628 int orig_level
= UnSet
;
1631 struct mdu_array_info_s array
;
1635 unsigned long long *offsets
;
1639 unsigned long blocks
;
1640 unsigned long cache
;
1641 unsigned long long array_size
;
1643 struct mdinfo
*sra
= NULL
;
1645 msg
= analyse_change(info
, &reshape
);
1647 fprintf(stderr
, Name
": %s\n", msg
);
1650 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1651 dprintf("Canot get array information.\n");
1656 /* reshape already started. just skip to monitoring the reshape */
1657 if (reshape
.backup_blocks
== 0)
1661 /* The container is frozen but the array may not be.
1662 * So freeze the array so spares don't get put to the wrong use
1663 * FIXME there should probably be a cleaner separation between
1664 * freeze_array and freeze_container.
1666 sysfs_freeze_array(info
);
1667 spares_needed
= max(reshape
.before
.data_disks
,
1668 reshape
.after
.data_disks
)
1669 + reshape
.parity
- array
.raid_disks
;
1672 info
->new_level
> 1 &&
1673 spares_needed
> info
->array
.spare_disks
) {
1675 Name
": Need %d spare%s to avoid degraded array,"
1676 " and only have %d.\n"
1677 " Use --force to over-ride this check.\n",
1679 spares_needed
== 1 ? "" : "s",
1680 info
->array
.spare_disks
);
1684 if (reshape
.level
!= info
->array
.level
) {
1685 char *c
= map_num(pers
, reshape
.level
);
1690 err
= sysfs_set_str(info
, NULL
, "level", c
);
1693 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1696 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1697 fprintf(stderr
, " Bitmap must be removed"
1698 " before level can be changed\n");
1702 fprintf(stderr
, Name
": level of %s changed to %s\n",
1704 orig_level
= info
->array
.level
;
1705 sysfs_freeze_array(info
);
1707 if (reshape
.level
> 0 && st
->ss
->external
) {
1708 /* make sure mdmon is aware of the new level */
1709 if (!mdmon_running(st
->container_dev
))
1710 start_mdmon(st
->container_dev
);
1711 ping_monitor(container
);
1714 /* ->reshape_super might have chosen some spares from the
1715 * container that it wants to be part of the new array.
1716 * We can collect them with ->container_content and give
1717 * them to the kernel.
1719 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
1720 char *subarray
= strchr(info
->text_version
+1, '/')+1;
1721 struct mdinfo
*info2
=
1722 st
->ss
->container_content(st
, subarray
);
1726 sysfs_init(info2
, fd
, st
->devnum
);
1727 for (d
= info2
->devs
; d
; d
= d
->next
) {
1728 if (d
->disk
.state
== 0 &&
1729 d
->disk
.raid_disk
>= 0) {
1730 /* This is a spare that wants to
1731 * be part of the array.
1733 add_disk(fd
, st
, info2
, d
);
1740 if (reshape
.backup_blocks
== 0) {
1741 /* No restriping needed, but we might need to impose
1742 * some more changes: layout, raid_disks, chunk_size
1744 /* read current array info */
1745 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1746 dprintf("Canot get array information.\n");
1749 /* compare current array info with new values and if
1750 * it is different update them to new */
1751 if (info
->new_layout
!= UnSet
&&
1752 info
->new_layout
!= array
.layout
) {
1753 array
.layout
= info
->new_layout
;
1754 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1755 fprintf(stderr
, Name
": failed to set new layout\n");
1758 printf("layout for %s set to %d\n",
1759 devname
, array
.layout
);
1761 if (info
->delta_disks
!= UnSet
&&
1762 info
->delta_disks
!= 0 &&
1763 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
1764 array
.raid_disks
+= info
->delta_disks
;
1765 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1766 fprintf(stderr
, Name
": failed to set raid disks\n");
1768 } else if (!quiet
) {
1769 printf("raid_disks for %s set to %d\n",
1770 devname
, array
.raid_disks
);
1773 if (info
->new_chunk
!= 0 &&
1774 info
->new_chunk
!= array
.chunk_size
) {
1775 if (sysfs_set_num(info
, NULL
,
1776 "chunk_size", info
->new_chunk
) != 0) {
1777 fprintf(stderr
, Name
": failed to set chunk size\n");
1780 printf("chunk size for %s set to %d\n",
1781 devname
, array
.chunk_size
);
1788 * There are three possibilities.
1789 * 1/ The array will shrink.
1790 * We need to ensure the reshape will pause before reaching
1791 * the 'critical section'. We also need to fork and wait for
1792 * that to happen. When it does we
1793 * suspend/backup/complete/unfreeze
1795 * 2/ The array will not change size.
1796 * This requires that we keep a backup of a sliding window
1797 * so that we can restore data after a crash. So we need
1798 * to fork and monitor progress.
1799 * In future we will allow the data_offset to change, so
1800 * a sliding backup becomes unnecessary.
1802 * 3/ The array will grow. This is relatively easy.
1803 * However the kernel's restripe routines will cheerfully
1804 * overwrite some early data before it is safe. So we
1805 * need to make a backup of the early parts of the array
1806 * and be ready to restore it if rebuild aborts very early.
1807 * For externally managed metadata, we still need a forked
1808 * child to monitor the reshape and suspend IO over the region
1809 * that is being reshaped.
1811 * We backup data by writing it to one spare, or to a
1812 * file which was given on command line.
1814 * In each case, we first make sure that storage is available
1815 * for the required backup.
1817 * - request the shape change.
1818 * - fork to handle backup etc.
1821 /* Check that we can hold all the data */
1822 get_dev_size(fd
, NULL
, &array_size
);
1823 if (reshape
.new_size
< (array_size
/512)) {
1825 Name
": this change will reduce the size of the array.\n"
1826 " use --grow --array-size first to truncate array.\n"
1827 " e.g. mdadm --grow %s --array-size %llu\n",
1828 devname
, reshape
.new_size
/2);
1832 sra
= sysfs_read(fd
, 0,
1833 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
1836 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
1841 /* Decide how many blocks (sectors) for a reshape
1842 * unit. The number we have so far is just a minimum
1844 blocks
= reshape
.backup_blocks
;
1845 if (reshape
.before
.data_disks
==
1846 reshape
.after
.data_disks
) {
1847 /* Make 'blocks' bigger for better throughput, but
1848 * not so big that we reject it below.
1849 * Try for 16 megabytes
1851 while (blocks
* 32 < sra
->component_size
&&
1855 fprintf(stderr
, Name
": Need to backup %luK of critical "
1856 "section..\n", blocks
/2);
1858 if (blocks
>= sra
->component_size
/2) {
1859 fprintf(stderr
, Name
": %s: Something wrong"
1860 " - reshape aborted\n",
1865 /* Now we need to open all these devices so we can read/write.
1867 nrdisks
= max(reshape
.before
.data_disks
,
1868 reshape
.after
.data_disks
) + reshape
.parity
1869 + sra
->array
.spare_disks
;
1870 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1871 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1872 if (!fdlist
|| !offsets
) {
1873 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1877 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
1878 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
1879 nrdisks
, blocks
, backup_file
,
1884 if (backup_file
== NULL
) {
1885 if (reshape
.after
.data_disks
<= reshape
.before
.data_disks
) {
1887 Name
": %s: Cannot grow - need backup-file\n",
1890 } else if (sra
->array
.spare_disks
== 0) {
1891 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1892 "backup-file to backup critical section\n",
1897 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
1899 fdlist
+d
, offsets
+d
, restart
)) {
1905 /* lastly, check that the internal stripe cache is
1906 * large enough, or it won't work.
1907 * It must hold at least 4 stripes of the larger
1910 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
1911 cache
*= 4; /* 4 stripes minimum */
1912 cache
/= 512; /* convert to sectors */
1913 disks
= min(reshape
.before
.data_disks
, reshape
.after
.data_disks
);
1914 /* make sure there is room for 'blocks' with a bit to spare */
1915 if (cache
< 16 + blocks
/ disks
)
1916 cache
= 16 + blocks
/ disks
;
1917 cache
/= (4096/512); /* Covert from sectors to pages */
1919 if (sra
->cache_size
< cache
)
1920 subarray_set_num(container
, sra
, "stripe_cache_size",
1923 /* Right, everything seems fine. Let's kick things off.
1924 * If only changing raid_disks, use ioctl, else use
1929 sra
->new_chunk
= info
->new_chunk
;
1931 if (info
->reshape_active
)
1932 sra
->reshape_progress
= info
->reshape_progress
;
1934 sra
->reshape_progress
= 0;
1935 if (reshape
.after
.data_disks
< reshape
.before
.data_disks
)
1936 /* start from the end of the new array */
1937 sra
->reshape_progress
= (sra
->component_size
1938 * reshape
.after
.data_disks
);
1941 if (info
->array
.chunk_size
== info
->new_chunk
&&
1942 reshape
.before
.layout
== reshape
.after
.layout
&&
1943 st
->ss
->external
== 0) {
1944 /* use SET_ARRAY_INFO but only if reshape hasn't started */
1945 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1946 array
.raid_disks
= reshape
.after
.data_disks
+ reshape
.parity
;
1947 if (!info
->reshape_active
&&
1948 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1952 Name
": Cannot set device shape for %s: %s\n",
1953 devname
, strerror(errno
));
1956 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1958 " Bitmap must be removed before"
1959 " shape can be changed\n");
1963 } else if (info
->reshape_active
&& !st
->ss
->external
) {
1964 /* We don't need to set anything here for internal
1965 * metadata, and for kernels before 2.6.38 we can
1969 /* set them all just in case some old 'new_*' value
1970 * persists from some earlier problem.
1971 * We even set them when restarting in the middle. They will
1972 * already be set in that case so this will be a no-op,
1973 * but it is hard to tell the difference.
1976 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
1978 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
1979 reshape
.after
.layout
) < 0)
1981 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
1982 reshape
.after
.data_disks
+
1983 reshape
.parity
) < 0)
1986 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1990 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1992 " Bitmap must be removed before"
1993 " shape can be changed\n");
1998 err
= start_reshape(sra
, (info
->reshape_active
&& !st
->ss
->external
));
2000 fprintf(stderr
, Name
": Cannot start reshape for %s\n",
2005 sysfs_set_str(sra
, NULL
, "array_state", "active");
2007 /* Now we just need to kick off the reshape and watch, while
2008 * handling backups of the data...
2009 * This is all done by a forked background process.
2011 switch(forked
? 0 : fork()) {
2013 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
2024 if (check_env("MDADM_GROW_VERIFY"))
2025 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
2028 mlockall(MCL_FUTURE
);
2030 if (st
->ss
->external
) {
2031 /* metadata handler takes it from here */
2032 done
= st
->ss
->manage_reshape(
2033 fd
, sra
, &reshape
, st
, blocks
,
2035 d
- odisks
, fdlist
+odisks
,
2038 done
= child_monitor(
2039 fd
, sra
, &reshape
, st
, blocks
,
2041 d
- odisks
, fdlist
+odisks
,
2044 if (backup_file
&& done
)
2045 unlink(backup_file
);
2051 if (!st
->ss
->external
&&
2052 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
2053 && info
->custom_array_size
) &&
2054 info
->new_level
== reshape
.level
&&
2056 /* no need to wait for the reshape to finish as
2057 * there is nothing more to do.
2063 if (st
->ss
->external
) {
2064 /* Re-load the metadata as much could have changed */
2065 int cfd
= open_dev(st
->container_dev
);
2067 ping_monitor(container
);
2068 st
->ss
->free_super(st
);
2069 st
->ss
->load_container(st
, cfd
, container
);
2074 /* set new array size if required customer_array_size is used
2077 if (reshape
.before
.data_disks
!=
2078 reshape
.after
.data_disks
&&
2079 info
->custom_array_size
) {
2080 struct mdinfo
*info2
;
2081 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2083 info2
= st
->ss
->container_content(st
, subarray
);
2085 unsigned long long current_size
= 0;
2086 unsigned long long new_size
=
2087 info2
->custom_array_size
/2;
2089 if (sysfs_get_ll(sra
,
2092 ¤t_size
) == 0 &&
2093 new_size
> current_size
) {
2094 if (sysfs_set_num(sra
, NULL
,
2095 "array_size", new_size
)
2097 dprintf("Error: Cannot"
2100 dprintf("Array size "
2102 dprintf(" from %llu to %llu.\n",
2103 current_size
, new_size
);
2109 if (info
->new_level
!= reshape
.level
) {
2111 c
= map_num(pers
, info
->new_level
);
2113 err
= sysfs_set_str(sra
, NULL
, "level", c
);
2115 fprintf(stderr
, Name\
2116 ": %s: could not set level "
2117 "to %s\n", devname
, c
);
2127 if (orig_level
!= UnSet
&& sra
) {
2128 c
= map_num(pers
, orig_level
);
2129 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
2130 fprintf(stderr
, Name
": aborting level change\n");
2137 int reshape_container(char *container
, int cfd
, char *devname
,
2138 struct supertype
*st
,
2139 struct mdinfo
*info
,
2144 struct mdinfo
*cc
= NULL
;
2146 /* component_size is not meaningful for a container,
2147 * so pass '-1' meaning 'no change'
2149 if (reshape_super(st
, -1, info
->new_level
,
2150 info
->new_layout
, info
->new_chunk
,
2151 info
->array
.raid_disks
, info
->delta_disks
,
2152 backup_file
, devname
, quiet
)) {
2159 /* ping monitor to be sure that update is on disk
2161 ping_monitor(container
);
2164 case -1: /* error */
2165 perror("Cannot fork to complete reshape\n");
2168 default: /* parent */
2169 printf(Name
": multi-array reshape continues in background\n");
2176 /* For each member array with reshape_active,
2177 * we need to perform the reshape.
2178 * We pick the first array that needs reshaping and
2179 * reshape it. reshape_array() will re-read the metadata
2180 * so the next time through a different array should be
2181 * ready for reshape.
2183 struct mdinfo
*content
;
2186 struct mdstat_ent
*mdstat
;
2191 cc
= st
->ss
->container_content(st
, NULL
);
2193 for (content
= cc
; content
; content
= content
->next
) {
2195 if (!content
->reshape_active
)
2198 subarray
= strchr(content
->text_version
+1, '/')+1;
2199 mdstat
= mdstat_by_subdev(subarray
,
2200 devname2devnum(container
));
2208 fd
= open_dev(mdstat
->devnum
);
2211 adev
= map_dev(dev2major(mdstat
->devnum
),
2212 dev2minor(mdstat
->devnum
),
2215 adev
= content
->text_version
;
2217 sysfs_init(content
, fd
, mdstat
->devnum
);
2219 rv
= reshape_array(container
, fd
, adev
, st
,
2221 backup_file
, quiet
, 1, 0);
2232 * We run a child process in the background which performs the following
2234 * - wait for resync to reach a certain point
2235 * - suspend io to the following section
2236 * - backup that section
2237 * - allow resync to proceed further
2239 * - discard the backup.
2241 * When are combined in slightly different ways in the three cases.
2243 * - suspend/backup/allow/wait/resume/discard
2245 * - allow/wait/suspend/backup/allow/wait/resume/discard
2247 * - wait/resume/discard/suspend/backup/allow
2249 * suspend/backup/allow always come together
2250 * wait/resume/discard do too.
2251 * For the same-size case we have two backups to improve flow.
2255 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
2256 unsigned long long backup_point
,
2257 unsigned long long wait_point
,
2258 unsigned long long *suspend_point
,
2259 unsigned long long *reshape_completed
)
2261 /* This function is called repeatedly by the reshape manager.
2262 * It determines how much progress can safely be made and allows
2264 * - 'info' identifies the array and particularly records in
2265 * ->reshape_progress the metadata's knowledge of progress
2266 * This is a sector offset from the start of the array
2267 * of the next array block to be relocated. This number
2268 * may increase from 0 or decrease from array_size, depending
2269 * on the type of reshape that is happening.
2270 * Note that in contrast, 'sync_completed' is a block count of the
2271 * reshape so far. It gives the distance between the start point
2272 * (head or tail of device) and the next place that data will be
2273 * written. It always increases.
2274 * - 'reshape' is the structure created by analyse_change
2275 * - 'backup_point' shows how much the metadata manager has backed-up
2276 * data. For reshapes with increasing progress, it is the next address
2277 * to be backed up, previous addresses have been backed-up. For
2278 * decreasing progress, it is the earliest address that has been
2279 * backed up - later address are also backed up.
2280 * So addresses between reshape_progress and backup_point are
2281 * backed up providing those are in the 'correct' order.
2282 * - 'wait_point' is an array address. When reshape_completed
2283 * passes this point, progress_reshape should return. It might
2284 * return earlier if it determines that ->reshape_progress needs
2285 * to be updated or further backup is needed.
2286 * - suspend_point is maintained by progress_reshape and the caller
2287 * should not touch it except to initialise to zero.
2288 * It is an array address and it only increases in 2.6.37 and earlier.
2289 * This makes it difficult to handle reducing reshapes with
2290 * external metadata.
2291 * However: it is similar to backup_point in that it records the
2292 * other end of a suspended region from reshape_progress.
2293 * it is moved to extend the region that is safe to backup and/or
2295 * - reshape_completed is read from sysfs and returned. The caller
2296 * should copy this into ->reshape_progress when it has reason to
2297 * believe that the metadata knows this, and any backup outside this
2301 * 1 if more data from backup_point - but only as far as suspend_point,
2302 * should be backed up
2303 * 0 if things are progressing smoothly
2304 * -1 if the reshape is finished, either because it is all done,
2305 * or due to an error.
2308 int advancing
= (reshape
->after
.data_disks
2309 >= reshape
->before
.data_disks
);
2310 unsigned long long need_backup
; /* All data between start of array and
2311 * here will at some point need to
2314 unsigned long long read_offset
, write_offset
;
2315 unsigned long long write_range
;
2316 unsigned long long max_progress
, target
, completed
;
2317 unsigned long long array_size
= (info
->component_size
2318 * reshape
->before
.data_disks
);
2322 /* First, we unsuspend any region that is now known to be safe.
2323 * If suspend_point is on the 'wrong' side of reshape_progress, then
2324 * we don't have or need suspension at the moment. This is true for
2325 * native metadata when we don't need to back-up.
2328 if (info
->reshape_progress
<= *suspend_point
)
2329 sysfs_set_num(info
, NULL
, "suspend_lo",
2330 info
->reshape_progress
);
2332 /* Note: this won't work in 2.6.37 and before.
2333 * Something somewhere should make sure we don't need it!
2335 if (info
->reshape_progress
>= *suspend_point
)
2336 sysfs_set_num(info
, NULL
, "suspend_hi",
2337 info
->reshape_progress
);
2340 /* Now work out how far it is safe to progress.
2341 * If the read_offset for ->reshape_progress is less than
2342 * 'blocks' beyond the write_offset, we can only progress as far
2344 * Otherwise we can progress until the write_offset for the new location
2345 * reaches (within 'blocks' of) the read_offset at the current location.
2346 * However that region must be suspended unless we are using native
2348 * If we need to suspend more, we limit it to 128M per device, which is
2349 * rather arbitrary and should be some time-based calculation.
2351 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
2352 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
2353 write_range
= info
->new_chunk
/512;
2354 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2355 need_backup
= array_size
;
2357 need_backup
= reshape
->backup_blocks
;
2359 if (read_offset
< write_offset
+ write_range
)
2360 max_progress
= backup_point
;
2364 reshape
->after
.data_disks
;
2366 if (read_offset
> write_offset
- write_range
)
2367 /* Can only progress as far as has been backed up,
2368 * which must be suspended */
2369 max_progress
= backup_point
;
2370 else if (info
->reshape_progress
<= need_backup
)
2371 max_progress
= backup_point
;
2373 if (info
->array
.major_version
>= 0)
2374 /* Can progress until backup is needed */
2375 max_progress
= need_backup
;
2377 /* Can progress until metadata update is required */
2380 reshape
->after
.data_disks
;
2381 /* but data must be suspended */
2382 if (max_progress
< *suspend_point
)
2383 max_progress
= *suspend_point
;
2388 /* We know it is safe to progress to 'max_progress' providing
2389 * it is suspended or we are using native metadata.
2390 * Consider extending suspend_point 128M per device if it
2391 * is less than 64M per device beyond reshape_progress.
2392 * But always do a multiple of 'blocks'
2393 * FIXME this is too big - it takes to long to complete
2396 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
2397 reshape
->after
.data_disks
);
2398 target
/= reshape
->backup_blocks
;
2401 target
*= reshape
->backup_blocks
;
2403 /* For externally managed metadata we always need to suspend IO to
2404 * the area being reshaped so we regularly push suspend_point forward.
2405 * For native metadata we only need the suspend if we are going to do
2409 if ((need_backup
> info
->reshape_progress
2410 || info
->array
.major_version
< 0) &&
2411 *suspend_point
< info
->reshape_progress
+ target
) {
2412 if (need_backup
< *suspend_point
+ 2 * target
)
2413 *suspend_point
= need_backup
;
2414 else if (*suspend_point
+ 2 * target
< array_size
)
2415 *suspend_point
+= 2 * target
;
2417 *suspend_point
= array_size
;
2418 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
2419 if (max_progress
> *suspend_point
)
2420 max_progress
= *suspend_point
;
2423 if (info
->array
.major_version
>= 0) {
2424 /* Only need to suspend when about to backup */
2425 if (info
->reshape_progress
< need_backup
* 2 &&
2426 *suspend_point
> 0) {
2428 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
2429 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
2432 /* Need to suspend continually */
2433 if (info
->reshape_progress
< *suspend_point
)
2434 *suspend_point
= info
->reshape_progress
;
2435 if (*suspend_point
+ target
< info
->reshape_progress
)
2436 /* No need to move suspend region yet */;
2438 if (*suspend_point
>= 2 * target
)
2439 *suspend_point
-= 2 * target
;
2442 sysfs_set_num(info
, NULL
, "suspend_lo",
2445 if (max_progress
< *suspend_point
)
2446 max_progress
= *suspend_point
;
2450 /* now set sync_max to allow that progress. sync_max, like
2451 * sync_completed is a count of sectors written per device, so
2452 * we find the difference between max_progress and the start point,
2453 * and divide that by after.data_disks to get a sync_max
2455 * At the same time we convert wait_point to a similar number
2456 * for comparing against sync_completed.
2458 /* scale down max_progress to per_disk */
2459 max_progress
/= reshape
->after
.data_disks
;
2460 /* Round to chunk size as some kernels give an erroneously high number */
2461 max_progress
/= info
->new_chunk
/512;
2462 max_progress
*= info
->new_chunk
/512;
2463 /* And round to old chunk size as the kernel wants that */
2464 max_progress
/= info
->array
.chunk_size
/512;
2465 max_progress
*= info
->array
.chunk_size
/512;
2466 /* Limit progress to the whole device */
2467 if (max_progress
> info
->component_size
)
2468 max_progress
= info
->component_size
;
2469 wait_point
/= reshape
->after
.data_disks
;
2471 /* switch from 'device offset' to 'processed block count' */
2472 max_progress
= info
->component_size
- max_progress
;
2473 wait_point
= info
->component_size
- wait_point
;
2476 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
2478 /* Now wait. If we have already reached the point that we were
2479 * asked to wait to, don't wait at all, else wait for any change.
2480 * We need to select on 'sync_completed' as that is the place that
2481 * notifications happen, but we are really interested in
2482 * 'reshape_position'
2484 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
2486 goto check_progress
;
2488 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2490 goto check_progress
;
2492 while (completed
< max_progress
&& completed
< wait_point
) {
2493 /* Check that sync_action is still 'reshape' to avoid
2494 * waiting forever on a dead array
2498 if (sysfs_get_str(info
, NULL
, "sync_action",
2500 strncmp(action
, "reshape", 7) != 0)
2502 /* Some kernels reset 'sync_completed' to zero
2503 * before setting 'sync_action' to 'idle'.
2504 * So we need these extra tests.
2506 if (completed
== 0 && advancing
2507 && info
->reshape_progress
> 0)
2509 if (completed
== 0 && !advancing
2510 && info
->reshape_progress
< (info
->component_size
2511 * reshape
->after
.data_disks
))
2515 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2516 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2518 goto check_progress
;
2521 /* Some kernels reset 'sync_completed' to zero,
2522 * we need to have real point we are in md
2525 completed
= max_progress
;
2527 /* some kernels can give an incorrectly high 'completed' number */
2528 completed
/= (info
->new_chunk
/512);
2529 completed
*= (info
->new_chunk
/512);
2530 /* Convert 'completed' back in to a 'progress' number */
2531 completed
*= reshape
->after
.data_disks
;
2533 completed
= info
->component_size
* reshape
->after
.data_disks
2536 *reshape_completed
= completed
;
2540 /* We return the need_backup flag. Caller will decide
2541 * how much - a multiple of ->backup_blocks up to *suspend_point
2544 return need_backup
> info
->reshape_progress
;
2546 return need_backup
>= info
->reshape_progress
;
2549 /* if we couldn't read a number from sync_completed, then
2550 * either the reshape did complete, or it aborted.
2551 * We can tell which by checking for 'none' in reshape_position.
2554 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
2555 || strncmp(buf
, "none", 4) != 0)
2556 return -2; /* abort */
2558 /* Maybe racing with array shutdown - check state */
2559 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
2560 || strncmp(buf
, "inactive", 8) == 0
2561 || strncmp(buf
, "clear",5) == 0)
2562 return -2; /* abort */
2563 return -1; /* complete */
2568 /* FIXME return status is never checked */
2569 static int grow_backup(struct mdinfo
*sra
,
2570 unsigned long long offset
, /* per device */
2571 unsigned long stripes
, /* per device, in old chunks */
2572 int *sources
, unsigned long long *offsets
,
2573 int disks
, int chunk
, int level
, int layout
,
2574 int dests
, int *destfd
, unsigned long long *destoffsets
,
2575 int part
, int *degraded
,
2578 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2579 * to storage 'destfd' (offset 'destoffsets'), after first
2580 * suspending IO. Then allow resync to continue
2581 * over the suspended section.
2582 * Use part 'part' of the backup-super-block.
2587 unsigned long long ll
;
2589 //printf("offset %llu\n", offset);
2595 /* Check that array hasn't become degraded, else we might backup the wrong data */
2596 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
2597 return -1; /* FIXME this error is ignored */
2598 new_degraded
= (int)ll
;
2599 if (new_degraded
!= *degraded
) {
2600 /* check each device to ensure it is still working */
2602 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2603 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2605 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2607 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2608 strstr(sbuf
, "faulty") ||
2609 strstr(sbuf
, "in_sync") == NULL
) {
2610 /* this device is dead */
2611 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2612 if (sd
->disk
.raid_disk
>= 0 &&
2613 sources
[sd
->disk
.raid_disk
] >= 0) {
2614 close(sources
[sd
->disk
.raid_disk
]);
2615 sources
[sd
->disk
.raid_disk
] = -1;
2620 *degraded
= new_degraded
;
2623 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2624 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2626 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2627 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2630 bsb
.magic
[15] = '2';
2631 for (i
= 0; i
< dests
; i
++)
2633 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
2635 lseek64(destfd
[i
], destoffsets
[i
], 0);
2637 rv
= save_stripes(sources
, offsets
,
2638 disks
, chunk
, level
, layout
,
2640 offset
*512*odata
, stripes
* chunk
* odata
,
2645 bsb
.mtime
= __cpu_to_le64(time(0));
2646 for (i
= 0; i
< dests
; i
++) {
2647 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2649 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2650 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2651 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2652 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2655 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
2656 != destoffsets
[i
] - 4096)
2658 if (write(destfd
[i
], &bsb
, 512) != 512)
2660 if (destoffsets
[i
] > 4096) {
2661 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
2662 destoffsets
[i
]+stripes
*chunk
*odata
)
2664 if (write(destfd
[i
], &bsb
, 512) != 512)
2674 /* in 2.6.30, the value reported by sync_completed can be
2675 * less that it should be by one stripe.
2676 * This only happens when reshape hits sync_max and pauses.
2677 * So allow wait_backup to either extent sync_max further
2678 * than strictly necessary, or return before the
2679 * sync has got quite as far as we would really like.
2680 * This is what 'blocks2' is for.
2681 * The various caller give appropriate values so that
2684 /* FIXME return value is often ignored */
2685 static int forget_backup(
2686 int dests
, int *destfd
, unsigned long long *destoffsets
,
2690 * Erase backup 'part' (which is 0 or 1)
2696 bsb
.arraystart2
= __cpu_to_le64(0);
2697 bsb
.length2
= __cpu_to_le64(0);
2699 bsb
.arraystart
= __cpu_to_le64(0);
2700 bsb
.length
= __cpu_to_le64(0);
2702 bsb
.mtime
= __cpu_to_le64(time(0));
2704 for (i
= 0; i
< dests
; i
++) {
2705 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2706 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2707 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2708 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2709 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2710 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
2711 destoffsets
[i
]-4096)
2714 write(destfd
[i
], &bsb
, 512) != 512)
2721 static void fail(char *msg
)
2724 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
2725 rv
|= (write(2, "\n", 1) != 1);
2729 static char *abuf
, *bbuf
;
2730 static unsigned long long abuflen
;
2731 static void validate(int afd
, int bfd
, unsigned long long offset
)
2733 /* check that the data in the backup against the array.
2734 * This is only used for regression testing and should not
2735 * be used while the array is active
2739 lseek64(bfd
, offset
- 4096, 0);
2740 if (read(bfd
, &bsb2
, 512) != 512)
2741 fail("cannot read bsb");
2742 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
2743 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
2744 fail("first csum bad");
2745 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
2746 fail("magic is bad");
2747 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
2748 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
2749 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
2750 fail("second csum bad");
2752 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
2753 fail("devstart is wrong");
2756 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
2758 if (abuflen
< len
) {
2762 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
2763 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
2765 /* just stop validating on mem-alloc failure */
2770 lseek64(bfd
, offset
, 0);
2771 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
2772 //printf("len %llu\n", len);
2773 fail("read first backup failed");
2775 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
2776 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2777 fail("read first from array failed");
2778 if (memcmp(bbuf
, abuf
, len
) != 0) {
2781 printf("offset=%llu len=%llu\n",
2782 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
2783 for (i
=0; i
<len
; i
++)
2784 if (bbuf
[i
] != abuf
[i
]) {
2785 printf("first diff byte %d\n", i
);
2789 fail("data1 compare failed");
2793 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
2795 if (abuflen
< len
) {
2799 abuf
= malloc(abuflen
);
2800 bbuf
= malloc(abuflen
);
2803 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2804 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2805 fail("read second backup failed");
2806 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2807 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2808 fail("read second from array failed");
2809 if (memcmp(bbuf
, abuf
, len
) != 0)
2810 fail("data2 compare failed");
2814 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
2815 struct supertype
*st
, unsigned long blocks
,
2816 int *fds
, unsigned long long *offsets
,
2817 int dests
, int *destfd
, unsigned long long *destoffsets
)
2819 /* Monitor a reshape where backup is being performed using
2820 * 'native' mechanism - either to a backup file, or
2821 * to some space in a spare.
2825 unsigned long long speed
;
2826 unsigned long long suspend_point
, array_size
;
2827 unsigned long long backup_point
, wait_point
;
2828 unsigned long long reshape_completed
;
2830 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
2831 int part
= 0; /* The next part of the backup area to fill. It may already
2832 * be full, so we need to check */
2833 int level
= reshape
->level
;
2834 int layout
= reshape
->before
.layout
;
2835 int data
= reshape
->before
.data_disks
;
2836 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
2837 int chunk
= sra
->array
.chunk_size
;
2839 unsigned long stripes
;
2841 /* set up the backup-super-block. This requires the
2842 * uuid from the array.
2844 /* Find a superblock */
2845 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2849 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2851 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
2852 devfd
= dev_open(dn
, O_RDONLY
);
2855 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
2861 fprintf(stderr
, Name
": Cannot find a superblock\n");
2865 memset(&bsb
, 0, 512);
2866 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2867 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
2868 bsb
.mtime
= __cpu_to_le64(time(0));
2869 bsb
.devstart2
= blocks
;
2871 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
2872 reshape
->before
.data_disks
;
2874 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2875 /* Don't start the 'reshape' */
2877 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
2878 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2879 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2883 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
2884 backup_point
= sra
->reshape_progress
;
2887 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
2888 backup_point
= reshape
->backup_blocks
;
2889 suspend_point
= array_size
;
2895 /* Want to return as soon the oldest backup slot can
2896 * be released as that allows us to start backing up
2897 * some more, providing suspend_point has been
2898 * advanced, which it should have.
2901 wait_point
= array_size
;
2902 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
2903 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
2904 __le64_to_cpu(bsb
.length
));
2905 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
2906 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
2907 __le64_to_cpu(bsb
.length2
));
2910 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
2911 wait_point
= __le64_to_cpu(bsb
.arraystart
);
2912 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
2913 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
2916 rv
= progress_reshape(sra
, reshape
,
2917 backup_point
, wait_point
,
2918 &suspend_point
, &reshape_completed
);
2919 /* external metadata would need to ping_monitor here */
2920 sra
->reshape_progress
= reshape_completed
;
2922 /* Clear any backup region that is before 'here' */
2924 if (reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
2925 __le64_to_cpu(bsb
.length
)))
2926 forget_backup(dests
, destfd
,
2928 if (reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
2929 __le64_to_cpu(bsb
.length2
)))
2930 forget_backup(dests
, destfd
,
2933 if (reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
2934 forget_backup(dests
, destfd
,
2936 if (reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
2937 forget_backup(dests
, destfd
,
2948 unsigned long long offset
;
2949 unsigned long actual_stripes
;
2950 /* Need to backup some data.
2951 * If 'part' is not used and the desired
2952 * backup size is suspended, do a backup,
2953 * then consider the next part.
2955 /* Check that 'part' is unused */
2956 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
2958 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
2961 offset
= backup_point
/ data
;
2962 actual_stripes
= stripes
;
2964 if (offset
+ actual_stripes
* (chunk
/512) >
2965 sra
->component_size
)
2966 actual_stripes
= ((sra
->component_size
- offset
)
2968 if (offset
+ actual_stripes
* (chunk
/512) >
2972 if (offset
< actual_stripes
* (chunk
/512))
2973 actual_stripes
= offset
/ (chunk
/512);
2974 offset
-= actual_stripes
* (chunk
/512);
2975 if (offset
< suspend_point
/data
)
2978 if (actual_stripes
== 0)
2980 grow_backup(sra
, offset
, actual_stripes
,
2982 disks
, chunk
, level
, layout
,
2983 dests
, destfd
, destoffsets
,
2984 part
, °raded
, buf
);
2985 validate(afd
, destfd
[0], destoffsets
[0]);
2986 /* record where 'part' is up to */
2989 backup_point
+= actual_stripes
* (chunk
/512) * data
;
2991 backup_point
-= actual_stripes
* (chunk
/512) * data
;
2995 /* FIXME maybe call progress_reshape one more time instead */
2996 abort_reshape(sra
); /* remove any remaining suspension */
2997 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2998 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
3004 * If any spare contains md_back_data-1 which is recent wrt mtime,
3005 * write that data into the array and update the super blocks with
3006 * the new reshape_progress
3008 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
3009 char *backup_file
, int verbose
)
3013 unsigned long long *offsets
;
3014 unsigned long long nstripe
, ostripe
;
3017 if (info
->new_level
!= info
->array
.level
)
3018 return 1; /* Cannot handle level changes (they are instantaneous) */
3020 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
3021 if (info
->array
.level
== 6) odata
--; /* number of data disks */
3022 ndata
= info
->array
.raid_disks
- 1;
3023 if (info
->new_level
== 6) ndata
--;
3025 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
3027 if (info
->delta_disks
<= 0)
3028 /* Didn't grow, so the backup file must have
3032 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
3033 struct mdinfo dinfo
;
3036 char *devname
, namebuf
[20];
3037 unsigned long long lo
, hi
;
3039 /* This was a spare and may have some saved data on it.
3040 * Load the superblock, find and load the
3041 * backup_super_block.
3042 * If either fail, go on to next device.
3043 * If the backup contains no new info, just return
3044 * else restore data and update all superblocks
3046 if (i
== old_disks
-1) {
3047 fd
= open(backup_file
, O_RDONLY
);
3049 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
3050 backup_file
, strerror(errno
));
3053 devname
= backup_file
;
3058 if (st
->ss
->load_super(st
, fd
, NULL
))
3061 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3062 st
->ss
->free_super(st
);
3065 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
3067 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
3068 continue; /* Cannot seek */
3070 sprintf(namebuf
, "device-%d", i
);
3073 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
3075 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
3076 continue; /* Cannot read */
3078 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
3079 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
3081 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
3084 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
3086 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
3087 continue; /* bad checksum */
3089 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
3090 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
3092 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
3093 continue; /* Bad second checksum */
3095 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
3097 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
3098 continue; /* Wrong uuid */
3101 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3102 * sometimes they aren't... So allow considerable flexability in matching, and allow
3103 * this test to be overridden by an environment variable.
3105 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
3106 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
3107 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3108 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
3109 "for array with timestamp %lu\n",
3110 (unsigned long)__le64_to_cpu(bsb
.mtime
),
3111 (unsigned long)info
->array
.utime
);
3114 fprintf(stderr
, Name
": too-old timestamp on "
3115 "backup-metadata on %s\n", devname
);
3116 continue; /* time stamp is too bad */
3120 if (bsb
.magic
[15] == '1') {
3121 if (bsb
.length
== 0)
3123 if (info
->delta_disks
>= 0) {
3124 /* reshape_progress is increasing */
3125 if (__le64_to_cpu(bsb
.arraystart
)
3126 + __le64_to_cpu(bsb
.length
)
3127 < info
->reshape_progress
) {
3130 fprintf(stderr
, Name
3131 ": backup-metadata found on %s but is not needed\n", devname
);
3132 continue; /* No new data here */
3135 /* reshape_progress is decreasing */
3136 if (__le64_to_cpu(bsb
.arraystart
) >=
3137 info
->reshape_progress
)
3138 goto nonew
; /* No new data here */
3141 if (bsb
.length
== 0 && bsb
.length2
== 0)
3143 if (info
->delta_disks
>= 0) {
3144 /* reshape_progress is increasing */
3145 if ((__le64_to_cpu(bsb
.arraystart
)
3146 + __le64_to_cpu(bsb
.length
)
3147 < info
->reshape_progress
)
3149 (__le64_to_cpu(bsb
.arraystart2
)
3150 + __le64_to_cpu(bsb
.length2
)
3151 < info
->reshape_progress
))
3152 goto nonew
; /* No new data here */
3154 /* reshape_progress is decreasing */
3155 if (__le64_to_cpu(bsb
.arraystart
) >=
3156 info
->reshape_progress
&&
3157 __le64_to_cpu(bsb
.arraystart2
) >=
3158 info
->reshape_progress
)
3159 goto nonew
; /* No new data here */
3162 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
3165 fprintf(stderr
, Name
3166 ": Failed to verify secondary backup-metadata block on %s\n",
3168 continue; /* Cannot seek */
3170 /* There should be a duplicate backup superblock 4k before here */
3171 if (lseek64(fd
, -4096, 1) < 0 ||
3172 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
3173 goto second_fail
; /* Cannot find leading superblock */
3174 if (bsb
.magic
[15] == '1')
3175 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
3177 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
3178 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
3179 goto second_fail
; /* Cannot find leading superblock */
3181 /* Now need the data offsets for all devices. */
3182 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
3183 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
3186 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3187 /* FIXME should be this be an error */
3189 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3190 st
->ss
->free_super(st
);
3191 offsets
[j
] = dinfo
.data_offset
* 512;
3193 printf(Name
": restoring critical section\n");
3195 if (restore_stripes(fdlist
, offsets
,
3196 info
->array
.raid_disks
,
3200 fd
, __le64_to_cpu(bsb
.devstart
)*512,
3201 __le64_to_cpu(bsb
.arraystart
)*512,
3202 __le64_to_cpu(bsb
.length
)*512)) {
3203 /* didn't succeed, so giveup */
3205 fprintf(stderr
, Name
": Error restoring backup from %s\n",
3210 if (bsb
.magic
[15] == '2' &&
3211 restore_stripes(fdlist
, offsets
,
3212 info
->array
.raid_disks
,
3216 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
3217 __le64_to_cpu(bsb
.devstart2
)*512,
3218 __le64_to_cpu(bsb
.arraystart2
)*512,
3219 __le64_to_cpu(bsb
.length2
)*512)) {
3220 /* didn't succeed, so giveup */
3222 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
3228 /* Ok, so the data is restored. Let's update those superblocks. */
3232 lo
= __le64_to_cpu(bsb
.arraystart
);
3233 hi
= lo
+ __le64_to_cpu(bsb
.length
);
3235 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
3236 unsigned long long lo1
, hi1
;
3237 lo1
= __le64_to_cpu(bsb
.arraystart2
);
3238 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
3242 } else if (lo
< lo1
)
3248 (info
->reshape_progress
< lo
||
3249 info
->reshape_progress
> hi
))
3250 /* backup does not affect reshape_progress*/ ;
3251 else if (info
->delta_disks
>= 0) {
3252 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
3253 __le64_to_cpu(bsb
.length
);
3254 if (bsb
.magic
[15] == '2') {
3255 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
3256 __le64_to_cpu(bsb
.length2
);
3257 if (p2
> info
->reshape_progress
)
3258 info
->reshape_progress
= p2
;
3261 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
3262 if (bsb
.magic
[15] == '2') {
3263 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
3264 if (p2
< info
->reshape_progress
)
3265 info
->reshape_progress
= p2
;
3268 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
3269 if (fdlist
[j
] < 0) continue;
3270 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3272 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3273 dinfo
.reshape_progress
= info
->reshape_progress
;
3274 st
->ss
->update_super(st
, &dinfo
,
3275 "_reshape_progress",
3277 st
->ss
->store_super(st
, fdlist
[j
]);
3278 st
->ss
->free_super(st
);
3282 /* Didn't find any backup data, try to see if any
3285 if (info
->delta_disks
< 0) {
3286 /* When shrinking, the critical section is at the end.
3287 * So see if we are before the critical section.
3289 unsigned long long first_block
;
3290 nstripe
= ostripe
= 0;
3292 while (ostripe
>= nstripe
) {
3293 ostripe
+= info
->array
.chunk_size
/ 512;
3294 first_block
= ostripe
* odata
;
3295 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
3296 (info
->new_chunk
/512);
3299 if (info
->reshape_progress
>= first_block
)
3302 if (info
->delta_disks
> 0) {
3303 /* See if we are beyond the critical section. */
3304 unsigned long long last_block
;
3305 nstripe
= ostripe
= 0;
3307 while (nstripe
>= ostripe
) {
3308 nstripe
+= info
->new_chunk
/ 512;
3309 last_block
= nstripe
* ndata
;
3310 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
3311 (info
->array
.chunk_size
/512);
3314 if (info
->reshape_progress
>= last_block
)
3317 /* needed to recover critical section! */
3319 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
3323 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
3326 int err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
3329 return reshape_array(NULL
, mdfd
, "array", st
, info
, 1, backup_file
, 0, 0, 1);