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 char *backup_file
, char *dev
, int verbose
)
570 /* nothing extra to check in the native case */
571 if (!st
->ss
->external
)
573 if (!st
->ss
->reshape_super
||
574 !st
->ss
->manage_reshape
) {
575 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
580 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
581 raid_disks
, backup_file
, dev
, verbose
);
584 static void sync_metadata(struct supertype
*st
)
586 if (st
->ss
->external
) {
587 if (st
->update_tail
) {
588 flush_metadata_updates(st
);
589 st
->update_tail
= &st
->updates
;
591 st
->ss
->sync_metadata(st
);
595 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
597 /* when dealing with external metadata subarrays we need to be
598 * prepared to handle EAGAIN. The kernel may need to wait for
599 * mdmon to mark the array active so the kernel can handle
600 * allocations/writeback when preparing the reshape action
601 * (md_allow_write()). We temporarily disable safe_mode_delay
602 * to close a race with the array_state going clean before the
603 * next write to raid_disks / stripe_cache_size
608 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
610 (strcmp(name
, "raid_disks") != 0 &&
611 strcmp(name
, "stripe_cache_size") != 0))
612 return sysfs_set_num(sra
, NULL
, name
, n
);
614 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
617 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
618 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
619 if (rc
< 0 && errno
== EAGAIN
) {
620 ping_monitor(container
);
621 /* if we get EAGAIN here then the monitor is not active
624 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
626 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
630 int start_reshape(struct mdinfo
*sra
)
633 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
634 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
635 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
636 /* Setting sync_min can fail if the recovery is already 'running',
637 * which can happen when restarting an array which is reshaping.
638 * So don't worry about errors here */
639 sysfs_set_num(sra
, NULL
, "sync_min", 0);
640 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", 0);
641 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
646 void abort_reshape(struct mdinfo
*sra
)
648 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
649 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
650 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
651 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
652 sysfs_set_num(sra
, NULL
, "sync_min", 0);
653 sysfs_set_str(sra
, NULL
, "sync_max", "max");
656 int remove_disks_for_takeover(struct supertype
*st
,
661 struct mdinfo
*remaining
;
664 if (sra
->array
.level
== 10)
665 nr_of_copies
= layout
& 0xff;
666 else if (sra
->array
.level
== 1)
667 nr_of_copies
= sra
->array
.raid_disks
;
671 remaining
= sra
->devs
;
673 /* for each 'copy', select one device and remove from the list. */
674 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
675 struct mdinfo
**diskp
;
678 /* Find a working device to keep */
679 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
680 struct mdinfo
*disk
= *diskp
;
682 if (disk
->disk
.raid_disk
< slot
)
684 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
686 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
688 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
690 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
693 /* We have found a good disk to use! */
695 disk
->next
= sra
->devs
;
704 if (slot
< sra
->array
.raid_disks
) {
705 /* didn't find all slots */
711 sra
->devs
= remaining
;
715 /* Remove all 'remaining' devices from the array */
717 struct mdinfo
*sd
= remaining
;
718 remaining
= sd
->next
;
720 sysfs_set_str(sra
, sd
, "state", "faulty");
721 sysfs_set_str(sra
, sd
, "slot", "none");
722 /* for external metadata disks should be removed in mdmon */
723 if (!st
->ss
->external
)
724 sysfs_set_str(sra
, sd
, "state", "remove");
725 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
726 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
727 sd
->next
= sra
->devs
;
733 void reshape_free_fdlist(int *fdlist
,
734 unsigned long long *offsets
,
739 for (i
= 0; i
< size
; i
++)
747 int reshape_prepare_fdlist(char *devname
,
751 unsigned long blocks
,
754 unsigned long long *offsets
)
759 for (d
= 0; d
<= nrdisks
; d
++)
762 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
763 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
765 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
766 char *dn
= map_dev(sd
->disk
.major
,
768 fdlist
[sd
->disk
.raid_disk
]
769 = dev_open(dn
, O_RDONLY
);
770 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
771 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
773 Name
": %s: cannot open component %s\n",
774 devname
, dn
? dn
: "-unknown-");
778 } else if (backup_file
== NULL
) {
780 char *dn
= map_dev(sd
->disk
.major
,
782 fdlist
[d
] = dev_open(dn
, O_RDWR
);
783 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
785 fprintf(stderr
, Name
": %s: cannot open component %s\n",
786 devname
, dn
? dn
: "-unknown-");
797 int reshape_open_backup_file(char *backup_file
,
802 unsigned long long *offsets
,
805 /* Return 1 on success, 0 on any form of failure */
806 /* need to check backup file is large enough */
812 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
816 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
817 devname
, backup_file
, strerror(errno
));
820 /* Guard against backup file being on array device.
821 * If array is partitioned or if LVM etc is in the
822 * way this will not notice, but it is better than
825 fstat(*fdlist
, &stb
);
828 if (stb
.st_rdev
== dev
) {
829 fprintf(stderr
, Name
": backup file must NOT be"
830 " on the array being reshaped.\n");
836 for (i
=0; i
< blocks
+ 1 ; i
++) {
837 if (write(*fdlist
, buf
, 512) != 512) {
838 fprintf(stderr
, Name
": %s: cannot create"
839 " backup file %s: %s\n",
840 devname
, backup_file
, strerror(errno
));
844 if (fsync(*fdlist
) != 0) {
845 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
846 devname
, backup_file
, strerror(errno
));
853 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
854 unsigned int ndata
, unsigned int odata
)
856 unsigned long a
, b
, blocks
;
857 /* So how much do we need to backup.
858 * We need an amount of data which is both a whole number of
859 * old stripes and a whole number of new stripes.
860 * So LCM for (chunksize*datadisks).
862 a
= (ochunk
/512) * odata
;
863 b
= (nchunk
/512) * ndata
;
871 /* LCM == product / GCD */
872 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
877 char *analyse_change(struct mdinfo
*info
, struct reshape
*re
)
879 /* Based on the current array state in info->array and
880 * the changes in info->new_* etc, determine:
881 * - whether the change is possible
882 * - Intermediate level/raid_disks/layout
883 * - whether a restriping reshape is needed
884 * - number of sectors in minimum change unit. This
885 * will cover a whole number of stripes in 'before' and
888 * Return message if the change should be rejected
889 * NULL if the change can be achieved
891 * This can be called as part of starting a reshape, or
892 * when assembling an array that is undergoing reshape.
896 /* If a new level not explicitly given, we assume no-change */
897 if (info
->new_level
== UnSet
)
898 info
->new_level
= info
->array
.level
;
901 switch (info
->new_level
) {
907 /* chunk size is meaningful, must divide component_size
910 if (info
->component_size
% (info
->new_chunk
/512))
911 return "New chunk size does not"
912 " divide component size";
915 return "chunk size not meaningful for this level";
918 info
->new_chunk
= info
->array
.chunk_size
;
920 switch (info
->array
.level
) {
922 /* RAID1 can convert to RAID1 with different disks, or
923 * raid5 with 2 disks, or
926 if (info
->new_level
== 0) {
928 re
->before
.data_disks
= 1;
929 re
->after
.data_disks
= 1;
930 re
->before
.layout
= 0;
931 re
->backup_blocks
= 0;
935 if (info
->new_level
== 1) {
936 if (info
->delta_disks
== UnSet
)
937 /* Don't know what to do */
938 return "no change requested for Growing RAID1";
940 re
->backup_blocks
= 0;
944 if (info
->array
.raid_disks
== 2 &&
945 info
->new_level
== 5) {
947 re
->before
.data_disks
= 1;
948 re
->after
.data_disks
= 1;
949 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
950 info
->array
.chunk_size
= 65536;
953 /* Could do some multi-stage conversions, but leave that to
956 return "Impossibly level change request for RAID1";
959 /* RAID10 can only be converted from near mode to
960 * RAID0 by removing some devices
962 if ((info
->array
.layout
& ~0xff) != 0x100)
963 return "Cannot Grow RAID10 with far/offset layout";
964 /* number of devices must be multiple of number of copies */
965 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
966 return "RAID10 layout too complex for Grow operation";
968 if (info
->new_level
!= 0)
969 return "RAID10 can only be changed to RAID0";
970 new_disks
= (info
->array
.raid_disks
971 / (info
->array
.layout
& 0xff));
972 if (info
->delta_disks
== UnSet
) {
973 info
->delta_disks
= (new_disks
974 - info
->array
.raid_disks
);
976 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
977 return "New number of raid-devices impossible for RAID10";
978 if (info
->new_chunk
&&
979 info
->new_chunk
!= info
->array
.chunk_size
)
980 return "Cannot change chunk-size with RAID10 Grow";
985 re
->before
.data_disks
= new_disks
;
986 re
->after
.data_disks
= re
->before
.data_disks
;
987 re
->before
.layout
= 0;
988 re
->backup_blocks
= 0;
992 /* RAID0 can be converted to RAID10, or to RAID456 */
993 if (info
->new_level
== 10) {
994 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
995 /* Assume near=2 layout */
996 info
->new_layout
= 0x102;
997 info
->delta_disks
= info
->array
.raid_disks
;
999 if (info
->new_layout
== UnSet
) {
1000 int copies
= 1 + (info
->delta_disks
1001 / info
->array
.raid_disks
);
1002 if (info
->array
.raid_disks
* (copies
-1)
1003 != info
->delta_disks
)
1004 return "Impossible number of devices"
1005 " for RAID0->RAID10";
1006 info
->new_layout
= 0x100 + copies
;
1008 if (info
->delta_disks
== UnSet
) {
1009 int copies
= info
->new_layout
& 0xff;
1010 if (info
->new_layout
!= 0x100 + copies
)
1011 return "New layout impossible"
1012 " for RAID0->RAID10";;
1013 info
->delta_disks
= (copies
- 1) *
1014 info
->array
.raid_disks
;
1016 if (info
->new_chunk
&&
1017 info
->new_chunk
!= info
->array
.chunk_size
)
1018 return "Cannot change chunk-size with RAID0->RAID10";
1022 re
->before
.data_disks
= (info
->array
.raid_disks
+
1024 re
->after
.data_disks
= re
->before
.data_disks
;
1025 re
->before
.layout
= info
->new_layout
;
1026 re
->backup_blocks
= 0;
1030 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1031 * a raid4 style layout of the final level.
1033 switch (info
->new_level
) {
1037 re
->before
.layout
= 0;
1041 re
->before
.layout
= ALGORITHM_PARITY_N
;
1045 re
->before
.layout
= ALGORITHM_PARITY_N
;
1048 return "Impossible level change requested";
1050 re
->before
.data_disks
= info
->array
.raid_disks
;
1051 /* determining 'after' layout happens outside this 'switch' */
1055 info
->array
.layout
= ALGORITHM_PARITY_N
;
1057 switch (info
->new_level
) {
1059 re
->level
= info
->array
.level
;
1060 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1061 re
->before
.layout
= info
->array
.layout
;
1065 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1066 re
->before
.layout
= info
->array
.layout
;
1070 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1071 switch (info
->array
.layout
) {
1072 case ALGORITHM_LEFT_ASYMMETRIC
:
1073 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1075 case ALGORITHM_RIGHT_ASYMMETRIC
:
1076 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1078 case ALGORITHM_LEFT_SYMMETRIC
:
1079 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1081 case ALGORITHM_RIGHT_SYMMETRIC
:
1082 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1084 case ALGORITHM_PARITY_0
:
1085 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1087 case ALGORITHM_PARITY_N
:
1088 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1091 return "Cannot convert an array with this layout";
1095 if (info
->array
.raid_disks
!= 2)
1096 return "Can only convert a 2-device array to RAID1";
1100 return "Impossible level change requested";
1104 switch (info
->new_level
) {
1109 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1110 re
->before
.layout
= info
->array
.layout
;
1113 return "Impossible level change requested";
1118 /* If we reached here then it looks like a re-stripe is
1119 * happening. We have determined the intermediate level
1120 * and initial raid_disks/layout and stored these in 're'.
1122 * We need to deduce the final layout that can be atomically
1123 * converted to the end state.
1125 switch (info
->new_level
) {
1127 /* We can only get to RAID0 from RAID4 or RAID5
1128 * with appropriate layout and one extra device
1130 if (re
->level
!= 4 && re
->level
!= 5)
1131 return "Cannot covert to RAID0 from this level";
1132 if (info
->delta_disks
== UnSet
)
1133 re
->after
.data_disks
= re
->before
.data_disks
;
1135 re
->after
.data_disks
=
1136 info
->array
.raid_disks
+ info
->delta_disks
;
1137 switch (re
->level
) {
1139 re
->after
.layout
= 0 ; break;
1141 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1146 /* We can only get to RAID4 from RAID5 */
1147 if (re
->level
!= 4 && re
->level
!= 5)
1148 return "Cannot convert to RAID4 from this level";
1149 if (info
->delta_disks
== UnSet
)
1150 re
->after
.data_disks
= re
->before
.data_disks
;
1152 re
->after
.data_disks
=
1153 re
->before
.data_disks
+ info
->delta_disks
;
1154 switch (re
->level
) {
1156 re
->after
.layout
= 0 ; break;
1158 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1163 /* We get to RAID5 for RAID5 or RAID6 */
1164 if (re
->level
!= 5 && re
->level
!= 6)
1165 return "Cannot convert to RAID5 from this level";
1166 if (info
->delta_disks
== UnSet
)
1167 re
->after
.data_disks
= re
->before
.data_disks
;
1168 else if (re
->level
== 5)
1169 re
->after
.data_disks
=
1170 re
->before
.data_disks
+ info
->delta_disks
;
1172 re
->after
.data_disks
=
1173 info
->array
.raid_disks
+ info
->delta_disks
- 1;
1174 switch (re
->level
) {
1176 if (info
->new_layout
== UnSet
)
1177 re
->after
.layout
= re
->before
.layout
;
1179 re
->after
.layout
= info
->new_layout
;
1182 if (info
->new_layout
== UnSet
)
1183 info
->new_layout
= re
->before
.layout
;
1185 /* after.layout needs to be raid6 version of new_layout */
1186 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1187 re
->after
.layout
= ALGORITHM_PARITY_N
;
1190 char *ls
= map_num(r5layout
, info
->new_layout
);
1192 strcat(strcpy(layout
, ls
), "-6");
1193 l
= map_name(r6layout
, layout
);
1195 return "Cannot find RAID6 layout"
1197 re
->after
.layout
= l
;
1203 /* We must already be at level 6 */
1205 return "Impossible level change";
1206 if (info
->delta_disks
== UnSet
)
1207 re
->after
.data_disks
= re
->before
.data_disks
;
1209 re
->after
.data_disks
= (info
->array
.raid_disks
+
1210 info
->delta_disks
) - 2;
1211 if (info
->new_layout
== UnSet
)
1212 re
->after
.layout
= info
->array
.layout
;
1214 re
->after
.layout
= info
->new_layout
;
1217 return "Impossible level change requested";
1219 switch (re
->level
) {
1220 case 6: re
->parity
= 2; break;
1222 case 5: re
->parity
= 1; break;
1223 default: re
->parity
= 0; break;
1225 /* So we have a restripe operation, we need to calculate the number
1226 * of blocks per reshape operation.
1228 if (info
->new_chunk
== 0)
1229 info
->new_chunk
= info
->array
.chunk_size
;
1230 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1231 re
->after
.layout
== re
->before
.layout
&&
1232 info
->new_chunk
== info
->array
.chunk_size
) {
1233 /* Nothing to change */
1234 re
->backup_blocks
= 0;
1237 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1238 /* chunk and layout changes make no difference */
1239 re
->backup_blocks
= 0;
1243 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1244 get_linux_version() < 2006032)
1245 return "in-place reshape is not safe before 2.6.32 - sorry.";
1247 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1248 get_linux_version() < 2006030)
1249 return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
1251 re
->backup_blocks
= compute_backup_blocks(
1252 info
->new_chunk
, info
->array
.chunk_size
,
1253 re
->after
.data_disks
,
1254 re
->before
.data_disks
);
1256 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1260 static int reshape_array(char *container
, int fd
, char *devname
,
1261 struct supertype
*st
, struct mdinfo
*info
,
1262 int force
, char *backup_file
, int quiet
, int forked
,
1264 static int reshape_container(char *container
, int cfd
, char *devname
,
1265 struct supertype
*st
,
1266 struct mdinfo
*info
,
1271 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1273 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1276 /* Make some changes in the shape of an array.
1277 * The kernel must support the change.
1279 * There are three different changes. Each can trigger
1280 * a resync or recovery so we freeze that until we have
1281 * requested everything (if kernel supports freezing - 2.6.30).
1283 * - change size (i.e. component_size)
1285 * - change layout/chunksize/ndisks
1287 * The last can require a reshape. It is different on different
1288 * levels so we need to check the level before actioning it.
1289 * Some times the level change needs to be requested after the
1290 * reshape (e.g. raid6->raid5, raid5->raid0)
1293 struct mdu_array_info_s array
;
1295 struct supertype
*st
;
1296 char *subarray
= NULL
;
1300 char *container
= NULL
;
1301 char container_buf
[20];
1307 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1308 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1314 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1315 fprintf(stderr
, Name
": cannot change component size at the same time "
1316 "as other changes.\n"
1317 " Change size first, then check data is intact before "
1318 "making other changes.\n");
1322 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1323 get_linux_version() < 2006032 &&
1324 !check_env("MDADM_FORCE_FEWER")) {
1325 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1326 " Please use a newer kernel\n");
1330 st
= super_by_fd(fd
, &subarray
);
1332 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1335 if (raid_disks
> st
->max_devs
) {
1336 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1337 " beyond %d\n", st
->max_devs
);
1341 /* in the external case we need to check that the requested reshape is
1342 * supported, and perform an initial check that the container holds the
1343 * pre-requisite spare devices (mdmon owns final validation)
1345 if (st
->ss
->external
) {
1350 container_dev
= st
->container_dev
;
1351 cfd
= open_dev_excl(st
->container_dev
);
1353 container_dev
= st
->devnum
;
1355 cfd
= open_dev_excl(st
->devnum
);
1359 fprintf(stderr
, Name
": Unable to open container for %s\n",
1365 fmt_devname(container_buf
, container_dev
);
1366 container
= container_buf
;
1368 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1371 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1377 if (mdmon_running(container_dev
))
1378 st
->update_tail
= &st
->updates
;
1381 if (raid_disks
> array
.raid_disks
&&
1382 array
.spare_disks
< (raid_disks
- array
.raid_disks
) &&
1385 Name
": Need %d spare%s to avoid degraded array,"
1386 " and only have %d.\n"
1387 " Use --force to over-ride this check.\n",
1388 raid_disks
- array
.raid_disks
,
1389 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1394 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
1395 | GET_STATE
| GET_VERSION
);
1397 if (st
->ss
->external
&& subarray
== NULL
) {
1398 array
.level
= LEVEL_CONTAINER
;
1399 sra
->array
.level
= LEVEL_CONTAINER
;
1402 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1406 frozen
= freeze(st
);
1408 /* freeze() already spewed the reason */
1410 } else if (frozen
< 0) {
1411 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1412 " be reshaped\n", devname
);
1416 /* ========= set size =============== */
1417 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1418 long long orig_size
= array
.size
;
1420 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1426 if (array
.size
!= size
) {
1427 /* got truncated to 32bit, write to
1428 * component_size instead
1431 rv
= sysfs_set_num(sra
, NULL
,
1432 "component_size", size
);
1436 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1440 /* restore metadata */
1441 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1442 NULL
, devname
, !quiet
) == 0)
1444 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1445 devname
, strerror(err
));
1447 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1448 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1452 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1453 size
= get_component_size(fd
)/2;
1457 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
1460 } else if (array
.level
!= LEVEL_CONTAINER
) {
1461 size
= get_component_size(fd
)/2;
1466 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1467 * current implementation assumes that following conditions must be met:
1470 * - near_copies == 2
1472 if ((level
== 0 && array
.level
== 10 && sra
&&
1473 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1474 (level
== 0 && array
.level
== 1 && sra
)) {
1476 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1478 dprintf(Name
": Array cannot be reshaped\n");
1484 /* FIXME this is added with no justification - why is it here */
1485 ping_monitor(container
);
1489 sysfs_init(&info
, fd
, NoMdDev
);
1490 strcpy(info
.text_version
, sra
->text_version
);
1491 info
.component_size
= size
*2;
1492 info
.new_level
= level
;
1493 info
.new_chunk
= chunksize
* 1024;
1495 info
.delta_disks
= raid_disks
- info
.array
.raid_disks
;
1497 info
.delta_disks
= UnSet
;
1498 if (layout_str
== NULL
) {
1499 info
.new_layout
= UnSet
;
1500 if (info
.array
.level
== 6 &&
1501 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1502 info
.array
.layout
>= 16) {
1503 fprintf(stderr
, Name
1504 ": %s has a non-standard layout. If you"
1505 " wish to preserve this\n"
1506 " during the reshape, please specify"
1507 " --layout=preserve\n"
1508 " If you want to change it, specify a"
1509 " layout or use --layout=normalise\n",
1514 } else if (strcmp(layout_str
, "normalise") == 0 ||
1515 strcmp(layout_str
, "normalize") == 0) {
1516 /* If we have a -6 RAID6 layout, remove the '-6'. */
1517 info
.new_layout
= UnSet
;
1518 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1520 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1521 h
= strrchr(l
, '-');
1522 if (h
&& strcmp(h
, "-6") == 0) {
1524 info
.new_layout
= map_name(r6layout
, l
);
1527 } else if (strcmp(layout_str
, "preserve") == 0) {
1528 info
.new_layout
= UnSet
;
1530 int l
= info
.new_level
;
1532 l
= info
.array
.level
;
1535 info
.new_layout
= map_name(r5layout
, layout_str
);
1538 info
.new_layout
= map_name(r6layout
, layout_str
);
1541 info
.new_layout
= parse_layout_10(layout_str
);
1544 info
.new_layout
= parse_layout_faulty(layout_str
);
1547 fprintf(stderr
, Name
": layout not meaningful"
1548 " with this level\n");
1552 if (info
.new_layout
== UnSet
) {
1553 fprintf(stderr
, Name
": layout %s not understood"
1554 " for this level\n",
1561 if (array
.level
== LEVEL_CONTAINER
) {
1562 /* This change is to be applied to every array in the
1563 * container. This is only needed when the metadata imposes
1564 * restraints of the various arrays in the container.
1565 * Currently we only know that IMSM requires all arrays
1566 * to have the same number of devices so changing the
1567 * number of devices (On-Line Capacity Expansion) must be
1568 * performed at the level of the container
1570 rv
= reshape_container(container
, fd
, devname
, st
, &info
,
1571 force
, backup_file
, quiet
);
1574 /* Impose these changes on a single array. First
1575 * check that the metadata is OK with the change. */
1577 if (reshape_super(st
, info
.component_size
, info
.new_level
,
1578 info
.new_layout
, info
.new_chunk
,
1579 info
.array
.raid_disks
+ info
.delta_disks
,
1580 backup_file
, devname
, quiet
)) {
1585 rv
= reshape_array(container
, fd
, devname
, st
, &info
, force
,
1586 backup_file
, quiet
, 0, 0);
1595 static int reshape_array(char *container
, int fd
, char *devname
,
1596 struct supertype
*st
, struct mdinfo
*info
,
1598 char *backup_file
, int quiet
, int forked
,
1601 struct reshape reshape
;
1604 int orig_level
= UnSet
;
1607 struct mdu_array_info_s array
;
1611 unsigned long long *offsets
;
1615 unsigned long blocks
;
1616 unsigned long cache
;
1617 unsigned long long array_size
;
1619 struct mdinfo
*sra
= NULL
;
1621 msg
= analyse_change(info
, &reshape
);
1623 fprintf(stderr
, Name
": %s\n", msg
);
1626 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1627 dprintf("Canot get array information.\n");
1632 /* reshape already started. just skip to monitoring the reshape */
1633 if (reshape
.backup_blocks
== 0)
1637 /* The container is frozen but the array may not be.
1638 * So freeze the array so spares don't get put to the wrong use
1639 * FIXME there should probably be a cleaner separation between
1640 * freeze_array and freeze_container.
1642 sysfs_freeze_array(info
);
1643 spares_needed
= max(reshape
.before
.data_disks
,
1644 reshape
.after
.data_disks
)
1645 + reshape
.parity
- array
.raid_disks
;
1648 info
->new_level
> 1 &&
1649 spares_needed
> info
->array
.spare_disks
) {
1651 Name
": Need %d spare%s to avoid degraded array,"
1652 " and only have %d.\n"
1653 " Use --force to over-ride this check.\n",
1655 spares_needed
== 1 ? "" : "s",
1656 info
->array
.spare_disks
);
1660 if (reshape
.level
!= info
->array
.level
) {
1661 char *c
= map_num(pers
, reshape
.level
);
1666 err
= sysfs_set_str(info
, NULL
, "level", c
);
1669 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1672 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1673 fprintf(stderr
, " Bitmap must be removed"
1674 " before level can be changed\n");
1678 fprintf(stderr
, Name
": level of %s changed to %s\n",
1680 orig_level
= info
->array
.level
;
1682 if (reshape
.level
> 0 && st
->ss
->external
) {
1683 /* make sure mdmon is aware of the new level */
1684 if (!mdmon_running(st
->container_dev
))
1685 start_mdmon(st
->container_dev
);
1686 ping_monitor(container
);
1689 /* ->reshape_super might have chosen some spares from the
1690 * container that it wants to be part of the new array.
1691 * We can collect them with ->container_content and give
1692 * them to the kernel.
1694 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
1695 char *subarray
= strchr(info
->text_version
+1, '/')+1;
1696 struct mdinfo
*info2
=
1697 st
->ss
->container_content(st
, subarray
);
1701 sysfs_init(info2
, fd
, st
->devnum
);
1702 for (d
= info2
->devs
; d
; d
= d
->next
) {
1703 if (d
->disk
.state
== 0 &&
1704 d
->disk
.raid_disk
>= 0) {
1705 /* This is a spare that wants to
1706 * be part of the array.
1708 add_disk(fd
, st
, info2
, d
);
1715 if (reshape
.backup_blocks
== 0) {
1716 /* No restriping needed, but we might need to impose
1717 * some more changes: layout, raid_disks, chunk_size
1719 /* read current array info */
1720 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1721 dprintf("Canot get array information.\n");
1724 /* compare current array info with new values and if
1725 * it is different update them to new */
1726 if (info
->new_layout
!= UnSet
&&
1727 info
->new_layout
!= array
.layout
) {
1728 array
.layout
= info
->new_layout
;
1729 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1730 fprintf(stderr
, Name
": failed to set new layout\n");
1733 printf("layout for %s set to %d\n",
1734 devname
, array
.layout
);
1736 if (info
->delta_disks
!= UnSet
&&
1737 info
->delta_disks
!= 0 &&
1738 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
1739 array
.raid_disks
+= info
->delta_disks
;
1740 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1741 fprintf(stderr
, Name
": failed to set raid disks\n");
1743 } else if (!quiet
) {
1744 printf("raid_disks for %s set to %d\n",
1745 devname
, array
.raid_disks
);
1748 if (info
->new_chunk
!= 0 &&
1749 info
->new_chunk
!= array
.chunk_size
) {
1750 if (sysfs_set_num(info
, NULL
,
1751 "chunk_size", info
->new_chunk
) != 0) {
1752 fprintf(stderr
, Name
": failed to set chunk size\n");
1755 printf("chunk size for %s set to %d\n",
1756 devname
, array
.chunk_size
);
1763 * There are three possibilities.
1764 * 1/ The array will shrink.
1765 * We need to ensure the reshape will pause before reaching
1766 * the 'critical section'. We also need to fork and wait for
1767 * that to happen. When it does we
1768 * suspend/backup/complete/unfreeze
1770 * 2/ The array will not change size.
1771 * This requires that we keep a backup of a sliding window
1772 * so that we can restore data after a crash. So we need
1773 * to fork and monitor progress.
1774 * In future we will allow the data_offset to change, so
1775 * a sliding backup becomes unnecessary.
1777 * 3/ The array will grow. This is relatively easy.
1778 * However the kernel's restripe routines will cheerfully
1779 * overwrite some early data before it is safe. So we
1780 * need to make a backup of the early parts of the array
1781 * and be ready to restore it if rebuild aborts very early.
1782 * For externally managed metadata, we still need a forked
1783 * child to monitor the reshape and suspend IO over the region
1784 * that is being reshaped.
1786 * We backup data by writing it to one spare, or to a
1787 * file which was given on command line.
1789 * In each case, we first make sure that storage is available
1790 * for the required backup.
1792 * - request the shape change.
1793 * - fork to handle backup etc.
1796 /* Check that we can hold all the data */
1797 get_dev_size(fd
, NULL
, &array_size
);
1798 if (reshape
.new_size
< (array_size
/512)) {
1800 Name
": this change will reduce the size of the array.\n"
1801 " use --grow --array-size first to truncate array.\n"
1802 " e.g. mdadm --grow %s --array-size %llu\n",
1803 devname
, reshape
.new_size
/2);
1807 sra
= sysfs_read(fd
, 0,
1808 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
1811 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
1816 /* Decide how many blocks (sectors) for a reshape
1817 * unit. The number we have so far is just a minimum
1819 blocks
= reshape
.backup_blocks
;
1820 if (reshape
.before
.data_disks
==
1821 reshape
.after
.data_disks
) {
1822 /* Make 'blocks' bigger for better throughput, but
1823 * not so big that we reject it below.
1824 * Try for 16 megabytes
1826 while (blocks
* 32 < sra
->component_size
&&
1830 fprintf(stderr
, Name
": Need to backup %luK of critical "
1831 "section..\n", blocks
/2);
1833 if (blocks
>= sra
->component_size
/2) {
1834 fprintf(stderr
, Name
": %s: Something wrong"
1835 " - reshape aborted\n",
1840 /* Now we need to open all these devices so we can read/write.
1842 nrdisks
= array
.raid_disks
+ sra
->array
.spare_disks
;
1843 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1844 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1845 if (!fdlist
|| !offsets
) {
1846 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1850 d
= reshape_prepare_fdlist(devname
, sra
, array
.raid_disks
,
1851 nrdisks
, blocks
, backup_file
,
1856 if (backup_file
== NULL
) {
1857 if (reshape
.after
.data_disks
<= reshape
.before
.data_disks
) {
1859 Name
": %s: Cannot grow - need backup-file\n",
1862 } else if (sra
->array
.spare_disks
== 0) {
1863 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1864 "backup-file to backup critical section\n",
1869 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
1871 fdlist
+d
, offsets
+d
, restart
)) {
1877 /* lastly, check that the internal stripe cache is
1878 * large enough, or it won't work.
1879 * It must hold at least 4 stripes of the larger
1882 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
1883 cache
*= 4; /* 4 stripes minimum */
1884 cache
/= 512; /* convert to sectors */
1885 disks
= min(reshape
.before
.data_disks
, reshape
.after
.data_disks
);
1886 /* make sure there is room for 'blocks' with a bit to spare */
1887 if (cache
< 16 + blocks
/ disks
)
1888 cache
= 16 + blocks
/ disks
;
1889 cache
/= (4096/512); /* Covert from sectors to pages */
1891 if (sra
->cache_size
< cache
)
1892 subarray_set_num(container
, sra
, "stripe_cache_size",
1895 /* Right, everything seems fine. Let's kick things off.
1896 * If only changing raid_disks, use ioctl, else use
1901 sra
->new_chunk
= info
->new_chunk
;
1903 if (info
->reshape_active
)
1904 sra
->reshape_progress
= info
->reshape_progress
;
1906 sra
->reshape_progress
= 0;
1907 if (reshape
.after
.data_disks
< reshape
.before
.data_disks
)
1908 /* start from the end of the new array */
1909 sra
->reshape_progress
= (sra
->component_size
1910 * reshape
.after
.data_disks
);
1913 if (info
->array
.chunk_size
== info
->new_chunk
&&
1914 reshape
.before
.layout
== reshape
.after
.layout
&&
1915 st
->ss
->external
== 0) {
1916 /* use SET_ARRAY_INFO but only if reshape hasn't started */
1917 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1918 array
.raid_disks
= reshape
.after
.data_disks
+ reshape
.parity
;
1919 if (!info
->reshape_active
&&
1920 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1924 Name
": Cannot set device shape for %s: %s\n",
1925 devname
, strerror(errno
));
1928 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1930 " Bitmap must be removed before"
1931 " shape can be changed\n");
1935 } else if (info
->reshape_active
&& !st
->ss
->external
) {
1936 /* We don't need to set anything here for internal
1937 * metadata, and for kernels before 2.6.38 we can
1941 /* set them all just in case some old 'new_*' value
1942 * persists from some earlier problem.
1943 * We even set them when restarting in the middle. They will
1944 * already be set in that case so this will be a no-op,
1945 * but it is hard to tell the difference.
1948 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
1950 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
1951 reshape
.after
.layout
) < 0)
1953 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
1954 reshape
.after
.data_disks
+
1955 reshape
.parity
) < 0)
1958 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1962 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1964 " Bitmap must be removed before"
1965 " shape can be changed\n");
1970 err
= start_reshape(sra
);
1972 fprintf(stderr
, Name
": Cannot start reshape for %s\n",
1977 sysfs_set_str(sra
, NULL
, "array_state", "active");
1979 /* Now we just need to kick off the reshape and watch, while
1980 * handling backups of the data...
1981 * This is all done by a forked background process.
1983 switch(forked
? 0 : fork()) {
1985 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
1996 if (check_env("MDADM_GROW_VERIFY"))
1997 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
2000 mlockall(MCL_FUTURE
);
2002 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
2004 if (st
->ss
->external
) {
2005 /* metadata handler takes it from here */
2006 done
= st
->ss
->manage_reshape(
2007 fd
, sra
, &reshape
, st
, blocks
,
2009 d
- odisks
, fdlist
+odisks
,
2012 done
= child_monitor(
2013 fd
, sra
, &reshape
, st
, blocks
,
2015 d
- odisks
, fdlist
+odisks
,
2018 if (backup_file
&& done
)
2019 unlink(backup_file
);
2025 if (!st
->ss
->external
&&
2026 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
2027 && info
->custom_array_size
) &&
2028 info
->new_level
== reshape
.level
&&
2030 /* no need to wait for the reshape to finish as
2031 * there is nothing more to do.
2037 if (st
->ss
->external
) {
2038 /* Re-load the metadata as much could have changed */
2039 int cfd
= open_dev(st
->container_dev
);
2041 ping_monitor(container
);
2042 st
->ss
->free_super(st
);
2043 st
->ss
->load_container(st
, cfd
, container
);
2048 /* set new array size if required customer_array_size is used
2051 if (reshape
.before
.data_disks
!=
2052 reshape
.after
.data_disks
&&
2053 info
->custom_array_size
) {
2054 struct mdinfo
*info2
;
2055 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2057 info2
= st
->ss
->container_content(st
, subarray
);
2059 unsigned long long current_size
= 0;
2060 unsigned long long new_size
=
2061 info2
->custom_array_size
/2;
2063 if (sysfs_get_ll(sra
,
2066 ¤t_size
) == 0 &&
2067 new_size
> current_size
) {
2068 if (sysfs_set_num(sra
, NULL
,
2069 "array_size", new_size
)
2071 dprintf("Error: Cannot"
2074 dprintf("Array size "
2076 dprintf(" from %llu to %llu.\n",
2077 current_size
, new_size
);
2083 if (info
->new_level
!= reshape
.level
) {
2085 c
= map_num(pers
, info
->new_level
);
2087 err
= sysfs_set_str(sra
, NULL
, "level", c
);
2089 fprintf(stderr
, Name\
2090 ": %s: could not set level "
2091 "to %s\n", devname
, c
);
2100 if (orig_level
!= UnSet
&& sra
) {
2101 c
= map_num(pers
, orig_level
);
2102 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
2103 fprintf(stderr
, Name
": aborting level change\n");
2110 int reshape_container(char *container
, int cfd
, char *devname
,
2111 struct supertype
*st
,
2112 struct mdinfo
*info
,
2117 struct mdinfo
*cc
= NULL
;
2119 /* component_size is not meaningful for a container,
2120 * so pass '-1' meaning 'no change'
2122 if (reshape_super(st
, -1, info
->new_level
,
2123 info
->new_layout
, info
->new_chunk
,
2124 info
->array
.raid_disks
+ info
->delta_disks
,
2125 backup_file
, devname
, quiet
))
2130 /* ping monitor to be sure that update is on disk
2132 ping_monitor(container
);
2135 case -1: /* error */
2136 perror("Cannot fork to complete reshape\n");
2138 default: /* parent */
2139 printf(Name
": multi-array reshape continues in background\n");
2146 /* For each member array with reshape_active,
2147 * we need to perform the reshape.
2148 * We pick the first array that needs reshaping and
2149 * reshape it. reshape_array() will re-read the metadata
2150 * so the next time through a different array should be
2151 * ready for reshape.
2153 struct mdinfo
*content
;
2156 struct mdstat_ent
*mdstat
;
2161 cc
= st
->ss
->container_content(st
, NULL
);
2163 for (content
= cc
; content
; content
= content
->next
) {
2165 if (!content
->reshape_active
)
2168 subarray
= strchr(content
->text_version
+1, '/')+1;
2169 mdstat
= mdstat_by_subdev(subarray
,
2170 devname2devnum(container
));
2178 fd
= open_dev(mdstat
->devnum
);
2181 adev
= map_dev(dev2major(mdstat
->devnum
),
2182 dev2minor(mdstat
->devnum
),
2185 adev
= content
->text_version
;
2187 sysfs_init(content
, fd
, mdstat
->devnum
);
2189 rv
= reshape_array(container
, fd
, adev
, st
,
2191 backup_file
, quiet
, 1, 0);
2202 * We run a child process in the background which performs the following
2204 * - wait for resync to reach a certain point
2205 * - suspend io to the following section
2206 * - backup that section
2207 * - allow resync to proceed further
2209 * - discard the backup.
2211 * When are combined in slightly different ways in the three cases.
2213 * - suspend/backup/allow/wait/resume/discard
2215 * - allow/wait/suspend/backup/allow/wait/resume/discard
2217 * - wait/resume/discard/suspend/backup/allow
2219 * suspend/backup/allow always come together
2220 * wait/resume/discard do too.
2221 * For the same-size case we have two backups to improve flow.
2225 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
2226 unsigned long long backup_point
,
2227 unsigned long long wait_point
,
2228 unsigned long long *suspend_point
,
2229 unsigned long long *reshape_completed
)
2231 /* This function is called repeatedly by the reshape manager.
2232 * It determines how much progress can safely be made and allows
2234 * - 'info' identifies the array and particularly records in
2235 * ->reshape_progress the metadata's knowledge of progress
2236 * This is a sector offset from the start of the array
2237 * of the next array block to be relocated. This number
2238 * may increase from 0 or decrease from array_size, depending
2239 * on the type of reshape that is happening.
2240 * Note that in contrast, 'sync_completed' is a block count of the
2241 * reshape so far. It gives the distance between the start point
2242 * (head or tail of device) and the next place that data will be
2243 * written. It always increases.
2244 * - 'reshape' is the structure created by analyse_change
2245 * - 'backup_point' shows how much the metadata manager has backed-up
2246 * data. For reshapes with increasing progress, it is the next address
2247 * to be backed up, previous addresses have been backed-up. For
2248 * decreasing progress, it is the earliest address that has been
2249 * backed up - later address are also backed up.
2250 * So addresses between reshape_progress and backup_point are
2251 * backed up providing those are in the 'correct' order.
2252 * - 'wait_point' is an array address. When reshape_completed
2253 * passes this point, progress_reshape should return. It might
2254 * return earlier if it determines that ->reshape_progress needs
2255 * to be updated or further backup is needed.
2256 * - suspend_point is maintained by progress_reshape and the caller
2257 * should not touch it except to initialise to zero.
2258 * It is an array address and it only increases in 2.6.37 and earlier.
2259 * This makes it difficult to handle reducing reshapes with
2260 * external metadata.
2261 * However: it is similar to backup_point in that it records the
2262 * other end of a suspended region from reshape_progress.
2263 * it is moved to extend the region that is safe to backup and/or
2265 * - reshape_completed is read from sysfs and returned. The caller
2266 * should copy this into ->reshape_progress when it has reason to
2267 * believe that the metadata knows this, and any backup outside this
2271 * 1 if more data from backup_point - but only as far as suspend_point,
2272 * should be backed up
2273 * 0 if things are progressing smoothly
2274 * -1 if the reshape is finished, either because it is all done,
2275 * or due to an error.
2278 int advancing
= (reshape
->after
.data_disks
2279 >= reshape
->before
.data_disks
);
2280 unsigned long long need_backup
; /* All data between start of array and
2281 * here will at some point need to
2284 unsigned long long read_offset
, write_offset
;
2285 unsigned long long write_range
;
2286 unsigned long long max_progress
, target
, completed
;
2287 unsigned long long array_size
= (info
->component_size
2288 * reshape
->before
.data_disks
);
2292 /* First, we unsuspend any region that is now known to be safe.
2293 * If suspend_point is on the 'wrong' side of reshape_progress, then
2294 * we don't have or need suspension at the moment. This is true for
2295 * native metadata when we don't need to back-up.
2298 if (info
->reshape_progress
<= *suspend_point
)
2299 sysfs_set_num(info
, NULL
, "suspend_lo",
2300 info
->reshape_progress
);
2302 /* Note: this won't work in 2.6.37 and before.
2303 * Something somewhere should make sure we don't need it!
2305 if (info
->reshape_progress
>= *suspend_point
)
2306 sysfs_set_num(info
, NULL
, "suspend_hi",
2307 info
->reshape_progress
);
2310 /* Now work out how far it is safe to progress.
2311 * If the read_offset for ->reshape_progress is less than
2312 * 'blocks' beyond the write_offset, we can only progress as far
2314 * Otherwise we can progress until the write_offset for the new location
2315 * reaches (within 'blocks' of) the read_offset at the current location.
2316 * However that region must be suspended unless we are using native
2318 * If we need to suspend more, we limit it to 128M per device, which is
2319 * rather arbitrary and should be some time-based calculation.
2321 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
2322 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
2323 write_range
= info
->new_chunk
/512;
2324 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2325 need_backup
= array_size
;
2327 need_backup
= reshape
->backup_blocks
;
2329 if (read_offset
< write_offset
+ write_range
)
2330 max_progress
= backup_point
;
2334 reshape
->after
.data_disks
;
2336 if (read_offset
> write_offset
- write_range
)
2337 /* Can only progress as far as has been backed up,
2338 * which must be suspended */
2339 max_progress
= backup_point
;
2340 else if (info
->reshape_progress
<= need_backup
)
2341 max_progress
= backup_point
;
2343 if (info
->array
.major_version
>= 0)
2344 /* Can progress until backup is needed */
2345 max_progress
= need_backup
;
2347 /* Can progress until metadata update is required */
2350 reshape
->after
.data_disks
;
2351 /* but data must be suspended */
2352 if (max_progress
< *suspend_point
)
2353 max_progress
= *suspend_point
;
2358 /* We know it is safe to progress to 'max_progress' providing
2359 * it is suspended or we are using native metadata.
2360 * Consider extending suspend_point 128M per device if it
2361 * is less than 64M per device beyond reshape_progress.
2362 * But always do a multiple of 'blocks'
2363 * FIXME this is too big - it takes to long to complete
2366 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
2367 reshape
->after
.data_disks
);
2368 target
/= reshape
->backup_blocks
;
2371 target
*= reshape
->backup_blocks
;
2373 /* For externally managed metadata we always need to suspend IO to
2374 * the area being reshaped so we regularly push suspend_point forward.
2375 * For native metadata we only need the suspend if we are going to do
2379 if ((need_backup
> info
->reshape_progress
2380 || info
->array
.major_version
< 0) &&
2381 *suspend_point
< info
->reshape_progress
+ target
) {
2382 if (need_backup
< *suspend_point
+ 2 * target
)
2383 *suspend_point
= need_backup
;
2384 else if (*suspend_point
+ 2 * target
< array_size
)
2385 *suspend_point
+= 2 * target
;
2387 *suspend_point
= array_size
;
2388 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
2389 if (max_progress
> *suspend_point
)
2390 max_progress
= *suspend_point
;
2393 if (info
->array
.major_version
>= 0) {
2394 /* Only need to suspend when about to backup */
2395 if (info
->reshape_progress
< need_backup
* 2 &&
2396 *suspend_point
> 0) {
2398 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
2399 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
2402 /* Need to suspend continually */
2403 if (info
->reshape_progress
< *suspend_point
)
2404 *suspend_point
= info
->reshape_progress
;
2405 if (*suspend_point
+ target
< info
->reshape_progress
)
2406 /* No need to move suspend region yet */;
2408 if (*suspend_point
>= 2 * target
)
2409 *suspend_point
-= 2 * target
;
2412 sysfs_set_num(info
, NULL
, "suspend_lo",
2415 if (max_progress
< *suspend_point
)
2416 max_progress
= *suspend_point
;
2420 /* now set sync_max to allow that progress. sync_max, like
2421 * sync_completed is a count of sectors written per device, so
2422 * we find the difference between max_progress and the start point,
2423 * and divide that by after.data_disks to get a sync_max
2425 * At the same time we convert wait_point to a similar number
2426 * for comparing against sync_completed.
2428 /* scale down max_progress to per_disk */
2429 max_progress
/= reshape
->after
.data_disks
;
2430 /* Round to chunk size as some kernels give an erroneously high number */
2431 max_progress
/= info
->new_chunk
/512;
2432 max_progress
*= info
->new_chunk
/512;
2433 /* Limit progress to the whole device */
2434 if (max_progress
> info
->component_size
)
2435 max_progress
= info
->component_size
;
2436 wait_point
/= reshape
->after
.data_disks
;
2438 /* switch from 'device offset' to 'processed block count' */
2439 max_progress
= info
->component_size
- max_progress
;
2440 wait_point
= info
->component_size
- wait_point
;
2443 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
2445 /* Now wait. If we have already reached the point that we were
2446 * asked to wait to, don't wait at all, else wait for any change.
2447 * We need to select on 'sync_completed' as that is the place that
2448 * notifications happen, but we are really interested in
2449 * 'reshape_position'
2451 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
2453 goto check_progress
;
2455 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2457 goto check_progress
;
2459 while (completed
< max_progress
&& completed
< wait_point
) {
2460 /* Check that sync_action is still 'reshape' to avoid
2461 * waiting forever on a dead array
2465 if (sysfs_get_str(info
, NULL
, "sync_action",
2467 strncmp(action
, "reshape", 7) != 0)
2469 /* Some kernels reset 'sync_completed' to zero
2470 * before setting 'sync_action' to 'idle'.
2471 * So we need these extra tests.
2473 if (completed
== 0 && advancing
2474 && info
->reshape_progress
> 0)
2476 if (completed
== 0 && !advancing
2477 && info
->reshape_progress
< (info
->component_size
2478 * reshape
->after
.data_disks
))
2482 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2483 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2485 goto check_progress
;
2488 /* Some kernels reset 'sync_completed' to zero,
2489 * we need to have real point we are in md
2492 completed
= max_progress
;
2494 /* some kernels can give an incorrectly high 'completed' number */
2495 completed
/= (info
->new_chunk
/512);
2496 completed
*= (info
->new_chunk
/512);
2497 /* Convert 'completed' back in to a 'progress' number */
2498 completed
*= reshape
->after
.data_disks
;
2500 completed
= info
->component_size
* reshape
->after
.data_disks
2503 *reshape_completed
= completed
;
2507 /* We return the need_backup flag. Caller will decide
2508 * how much - a multiple of ->backup_blocks up to *suspend_point
2511 return need_backup
> info
->reshape_progress
;
2513 return need_backup
>= info
->reshape_progress
;
2516 /* if we couldn't read a number from sync_completed, then
2517 * either the reshape did complete, or it aborted.
2518 * We can tell which by checking for 'none' in reshape_position.
2521 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
2522 || strncmp(buf
, "none", 4) != 0)
2523 return -2; /* abort */
2525 /* Maybe racing with array shutdown - check state */
2526 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
2527 || strncmp(buf
, "inactive", 8) == 0
2528 || strncmp(buf
, "clear",5) == 0)
2529 return -2; /* abort */
2530 return -1; /* complete */
2535 /* FIXME return status is never checked */
2536 static int grow_backup(struct mdinfo
*sra
,
2537 unsigned long long offset
, /* per device */
2538 unsigned long stripes
, /* per device, in old chunks */
2539 int *sources
, unsigned long long *offsets
,
2540 int disks
, int chunk
, int level
, int layout
,
2541 int dests
, int *destfd
, unsigned long long *destoffsets
,
2542 int part
, int *degraded
,
2545 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2546 * to storage 'destfd' (offset 'destoffsets'), after first
2547 * suspending IO. Then allow resync to continue
2548 * over the suspended section.
2549 * Use part 'part' of the backup-super-block.
2554 unsigned long long ll
;
2556 //printf("offset %llu\n", offset);
2562 /* Check that array hasn't become degraded, else we might backup the wrong data */
2563 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
2564 return -1; /* FIXME this error is ignored */
2565 new_degraded
= (int)ll
;
2566 if (new_degraded
!= *degraded
) {
2567 /* check each device to ensure it is still working */
2569 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2570 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2572 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2574 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2575 strstr(sbuf
, "faulty") ||
2576 strstr(sbuf
, "in_sync") == NULL
) {
2577 /* this device is dead */
2578 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2579 if (sd
->disk
.raid_disk
>= 0 &&
2580 sources
[sd
->disk
.raid_disk
] >= 0) {
2581 close(sources
[sd
->disk
.raid_disk
]);
2582 sources
[sd
->disk
.raid_disk
] = -1;
2587 *degraded
= new_degraded
;
2590 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2591 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2593 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2594 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2597 bsb
.magic
[15] = '2';
2598 for (i
= 0; i
< dests
; i
++)
2600 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
2602 lseek64(destfd
[i
], destoffsets
[i
], 0);
2604 rv
= save_stripes(sources
, offsets
,
2605 disks
, chunk
, level
, layout
,
2607 offset
*512*odata
, stripes
* chunk
* odata
,
2612 bsb
.mtime
= __cpu_to_le64(time(0));
2613 for (i
= 0; i
< dests
; i
++) {
2614 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2616 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2617 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2618 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2619 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2622 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
2623 != destoffsets
[i
] - 4096)
2625 if (write(destfd
[i
], &bsb
, 512) != 512)
2627 if (destoffsets
[i
] > 4096) {
2628 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
2629 destoffsets
[i
]+stripes
*chunk
*odata
)
2631 if (write(destfd
[i
], &bsb
, 512) != 512)
2641 /* in 2.6.30, the value reported by sync_completed can be
2642 * less that it should be by one stripe.
2643 * This only happens when reshape hits sync_max and pauses.
2644 * So allow wait_backup to either extent sync_max further
2645 * than strictly necessary, or return before the
2646 * sync has got quite as far as we would really like.
2647 * This is what 'blocks2' is for.
2648 * The various caller give appropriate values so that
2651 /* FIXME return value is often ignored */
2652 static int forget_backup(
2653 int dests
, int *destfd
, unsigned long long *destoffsets
,
2657 * Erase backup 'part' (which is 0 or 1)
2663 bsb
.arraystart2
= __cpu_to_le64(0);
2664 bsb
.length2
= __cpu_to_le64(0);
2666 bsb
.arraystart
= __cpu_to_le64(0);
2667 bsb
.length
= __cpu_to_le64(0);
2669 bsb
.mtime
= __cpu_to_le64(time(0));
2671 for (i
= 0; i
< dests
; i
++) {
2672 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2673 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2674 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2675 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2676 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2677 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
2678 destoffsets
[i
]-4096)
2681 write(destfd
[i
], &bsb
, 512) != 512)
2688 static void fail(char *msg
)
2691 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
2692 rv
|= (write(2, "\n", 1) != 1);
2696 static char *abuf
, *bbuf
;
2697 static unsigned long long abuflen
;
2698 static void validate(int afd
, int bfd
, unsigned long long offset
)
2700 /* check that the data in the backup against the array.
2701 * This is only used for regression testing and should not
2702 * be used while the array is active
2706 lseek64(bfd
, offset
- 4096, 0);
2707 if (read(bfd
, &bsb2
, 512) != 512)
2708 fail("cannot read bsb");
2709 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
2710 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
2711 fail("first csum bad");
2712 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
2713 fail("magic is bad");
2714 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
2715 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
2716 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
2717 fail("second csum bad");
2719 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
2720 fail("devstart is wrong");
2723 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
2725 if (abuflen
< len
) {
2729 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
2730 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
2732 /* just stop validating on mem-alloc failure */
2737 lseek64(bfd
, offset
, 0);
2738 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
2739 //printf("len %llu\n", len);
2740 fail("read first backup failed");
2742 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
2743 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2744 fail("read first from array failed");
2745 if (memcmp(bbuf
, abuf
, len
) != 0) {
2748 printf("offset=%llu len=%llu\n",
2749 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
2750 for (i
=0; i
<len
; i
++)
2751 if (bbuf
[i
] != abuf
[i
]) {
2752 printf("first diff byte %d\n", i
);
2756 fail("data1 compare failed");
2760 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
2762 if (abuflen
< len
) {
2766 abuf
= malloc(abuflen
);
2767 bbuf
= malloc(abuflen
);
2770 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2771 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2772 fail("read second backup failed");
2773 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2774 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2775 fail("read second from array failed");
2776 if (memcmp(bbuf
, abuf
, len
) != 0)
2777 fail("data2 compare failed");
2781 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
2782 struct supertype
*st
, unsigned long blocks
,
2783 int *fds
, unsigned long long *offsets
,
2784 int dests
, int *destfd
, unsigned long long *destoffsets
)
2786 /* Monitor a reshape where backup is being performed using
2787 * 'native' mechanism - either to a backup file, or
2788 * to some space in a spare.
2792 unsigned long long speed
;
2793 unsigned long long suspend_point
, array_size
;
2794 unsigned long long backup_point
, wait_point
;
2795 unsigned long long reshape_completed
;
2797 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
2798 int part
= 0; /* The next part of the backup area to fill. It may already
2799 * be full, so we need to check */
2800 int level
= reshape
->level
;
2801 int layout
= reshape
->before
.layout
;
2802 int data
= reshape
->before
.data_disks
;
2803 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
2804 int chunk
= sra
->array
.chunk_size
;
2806 unsigned long stripes
;
2808 /* set up the backup-super-block. This requires the
2809 * uuid from the array.
2811 /* Find a superblock */
2812 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2816 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2818 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
2819 devfd
= dev_open(dn
, O_RDONLY
);
2822 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
2828 fprintf(stderr
, Name
": Cannot find a superblock\n");
2832 memset(&bsb
, 0, 512);
2833 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2834 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
2835 bsb
.mtime
= __cpu_to_le64(time(0));
2836 bsb
.devstart2
= blocks
;
2838 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
2839 reshape
->before
.data_disks
;
2841 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2842 /* Don't start the 'reshape' */
2844 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
2845 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2846 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2850 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
2851 backup_point
= sra
->reshape_progress
;
2854 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
2855 backup_point
= reshape
->backup_blocks
;
2856 suspend_point
= array_size
;
2862 /* Want to return as soon the oldest backup slot can
2863 * be released as that allows us to start backing up
2864 * some more, providing suspend_point has been
2865 * advanced, which it should have.
2868 wait_point
= array_size
;
2869 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
2870 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
2871 __le64_to_cpu(bsb
.length
));
2872 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
2873 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
2874 __le64_to_cpu(bsb
.length2
));
2877 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
2878 wait_point
= __le64_to_cpu(bsb
.arraystart
);
2879 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
2880 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
2883 rv
= progress_reshape(sra
, reshape
,
2884 backup_point
, wait_point
,
2885 &suspend_point
, &reshape_completed
);
2886 /* external metadata would need to ping_monitor here */
2887 sra
->reshape_progress
= reshape_completed
;
2889 /* Clear any backup region that is before 'here' */
2891 if (reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
2892 __le64_to_cpu(bsb
.length
)))
2893 forget_backup(dests
, destfd
,
2895 if (reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
2896 __le64_to_cpu(bsb
.length2
)))
2897 forget_backup(dests
, destfd
,
2900 if (reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
2901 forget_backup(dests
, destfd
,
2903 if (reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
2904 forget_backup(dests
, destfd
,
2915 unsigned long long offset
;
2916 unsigned long actual_stripes
;
2917 /* Need to backup some data.
2918 * If 'part' is not used and the desired
2919 * backup size is suspended, do a backup,
2920 * then consider the next part.
2922 /* Check that 'part' is unused */
2923 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
2925 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
2928 offset
= backup_point
/ data
;
2929 actual_stripes
= stripes
;
2931 if (offset
+ actual_stripes
* (chunk
/512) >
2932 sra
->component_size
)
2933 actual_stripes
= ((sra
->component_size
- offset
)
2935 if (offset
+ actual_stripes
* (chunk
/512) >
2939 if (offset
< actual_stripes
* (chunk
/512))
2940 actual_stripes
= offset
/ (chunk
/512);
2941 offset
-= actual_stripes
* (chunk
/512);
2942 if (offset
< suspend_point
/data
)
2945 grow_backup(sra
, offset
, actual_stripes
,
2947 disks
, chunk
, level
, layout
,
2948 dests
, destfd
, destoffsets
,
2949 part
, °raded
, buf
);
2950 validate(afd
, destfd
[0], destoffsets
[0]);
2951 /* record where 'part' is up to */
2954 backup_point
+= actual_stripes
* (chunk
/512) * data
;
2956 backup_point
-= actual_stripes
* (chunk
/512) * data
;
2960 /* FIXME maybe call progress_reshape one more time instead */
2961 abort_reshape(sra
); /* remove any remaining suspension */
2962 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2963 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
2969 * If any spare contains md_back_data-1 which is recent wrt mtime,
2970 * write that data into the array and update the super blocks with
2971 * the new reshape_progress
2973 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
2974 char *backup_file
, int verbose
)
2978 unsigned long long *offsets
;
2979 unsigned long long nstripe
, ostripe
;
2982 if (info
->new_level
!= info
->array
.level
)
2983 return 1; /* Cannot handle level changes (they are instantaneous) */
2985 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
2986 if (info
->array
.level
== 6) odata
--; /* number of data disks */
2987 ndata
= info
->array
.raid_disks
- 1;
2988 if (info
->new_level
== 6) ndata
--;
2990 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
2992 if (info
->delta_disks
<= 0)
2993 /* Didn't grow, so the backup file must have
2997 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
2998 struct mdinfo dinfo
;
3001 char *devname
, namebuf
[20];
3002 unsigned long long lo
, hi
;
3004 /* This was a spare and may have some saved data on it.
3005 * Load the superblock, find and load the
3006 * backup_super_block.
3007 * If either fail, go on to next device.
3008 * If the backup contains no new info, just return
3009 * else restore data and update all superblocks
3011 if (i
== old_disks
-1) {
3012 fd
= open(backup_file
, O_RDONLY
);
3014 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
3015 backup_file
, strerror(errno
));
3018 devname
= backup_file
;
3023 if (st
->ss
->load_super(st
, fd
, NULL
))
3026 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3027 st
->ss
->free_super(st
);
3030 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
3032 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
3033 continue; /* Cannot seek */
3035 sprintf(namebuf
, "device-%d", i
);
3038 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
3040 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
3041 continue; /* Cannot read */
3043 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
3044 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
3046 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
3049 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
3051 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
3052 continue; /* bad checksum */
3054 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
3055 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
3057 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
3058 continue; /* Bad second checksum */
3060 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
3062 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
3063 continue; /* Wrong uuid */
3066 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3067 * sometimes they aren't... So allow considerable flexability in matching, and allow
3068 * this test to be overridden by an environment variable.
3070 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
3071 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
3072 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3073 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
3074 "for array with timestamp %lu\n",
3075 (unsigned long)__le64_to_cpu(bsb
.mtime
),
3076 (unsigned long)info
->array
.utime
);
3079 fprintf(stderr
, Name
": too-old timestamp on "
3080 "backup-metadata on %s\n", devname
);
3081 continue; /* time stamp is too bad */
3085 if (bsb
.magic
[15] == '1') {
3086 if (bsb
.length
== 0)
3088 if (info
->delta_disks
>= 0) {
3089 /* reshape_progress is increasing */
3090 if (__le64_to_cpu(bsb
.arraystart
)
3091 + __le64_to_cpu(bsb
.length
)
3092 < info
->reshape_progress
) {
3095 fprintf(stderr
, Name
3096 ": backup-metadata found on %s but is not needed\n", devname
);
3097 continue; /* No new data here */
3100 /* reshape_progress is decreasing */
3101 if (__le64_to_cpu(bsb
.arraystart
) >=
3102 info
->reshape_progress
)
3103 goto nonew
; /* No new data here */
3106 if (bsb
.length
== 0 && bsb
.length2
== 0)
3108 if (info
->delta_disks
>= 0) {
3109 /* reshape_progress is increasing */
3110 if ((__le64_to_cpu(bsb
.arraystart
)
3111 + __le64_to_cpu(bsb
.length
)
3112 < info
->reshape_progress
)
3114 (__le64_to_cpu(bsb
.arraystart2
)
3115 + __le64_to_cpu(bsb
.length2
)
3116 < info
->reshape_progress
))
3117 goto nonew
; /* No new data here */
3119 /* reshape_progress is decreasing */
3120 if (__le64_to_cpu(bsb
.arraystart
) >=
3121 info
->reshape_progress
&&
3122 __le64_to_cpu(bsb
.arraystart2
) >=
3123 info
->reshape_progress
)
3124 goto nonew
; /* No new data here */
3127 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
3130 fprintf(stderr
, Name
3131 ": Failed to verify secondary backup-metadata block on %s\n",
3133 continue; /* Cannot seek */
3135 /* There should be a duplicate backup superblock 4k before here */
3136 if (lseek64(fd
, -4096, 1) < 0 ||
3137 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
3138 goto second_fail
; /* Cannot find leading superblock */
3139 if (bsb
.magic
[15] == '1')
3140 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
3142 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
3143 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
3144 goto second_fail
; /* Cannot find leading superblock */
3146 /* Now need the data offsets for all devices. */
3147 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
3148 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
3151 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3152 /* FIXME should be this be an error */
3154 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3155 st
->ss
->free_super(st
);
3156 offsets
[j
] = dinfo
.data_offset
* 512;
3158 printf(Name
": restoring critical section\n");
3160 if (restore_stripes(fdlist
, offsets
,
3161 info
->array
.raid_disks
,
3165 fd
, __le64_to_cpu(bsb
.devstart
)*512,
3166 __le64_to_cpu(bsb
.arraystart
)*512,
3167 __le64_to_cpu(bsb
.length
)*512)) {
3168 /* didn't succeed, so giveup */
3170 fprintf(stderr
, Name
": Error restoring backup from %s\n",
3175 if (bsb
.magic
[15] == '2' &&
3176 restore_stripes(fdlist
, offsets
,
3177 info
->array
.raid_disks
,
3181 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
3182 __le64_to_cpu(bsb
.devstart2
)*512,
3183 __le64_to_cpu(bsb
.arraystart2
)*512,
3184 __le64_to_cpu(bsb
.length2
)*512)) {
3185 /* didn't succeed, so giveup */
3187 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
3193 /* Ok, so the data is restored. Let's update those superblocks. */
3197 lo
= __le64_to_cpu(bsb
.arraystart
);
3198 hi
= lo
+ __le64_to_cpu(bsb
.length
);
3200 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
3201 unsigned long long lo1
, hi1
;
3202 lo1
= __le64_to_cpu(bsb
.arraystart2
);
3203 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
3207 } else if (lo
< lo1
)
3213 (info
->reshape_progress
< lo
||
3214 info
->reshape_progress
> hi
))
3215 /* backup does not affect reshape_progress*/ ;
3216 else if (info
->delta_disks
>= 0) {
3217 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
3218 __le64_to_cpu(bsb
.length
);
3219 if (bsb
.magic
[15] == '2') {
3220 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
3221 __le64_to_cpu(bsb
.length2
);
3222 if (p2
> info
->reshape_progress
)
3223 info
->reshape_progress
= p2
;
3226 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
3227 if (bsb
.magic
[15] == '2') {
3228 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
3229 if (p2
< info
->reshape_progress
)
3230 info
->reshape_progress
= p2
;
3233 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
3234 if (fdlist
[j
] < 0) continue;
3235 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3237 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3238 dinfo
.reshape_progress
= info
->reshape_progress
;
3239 st
->ss
->update_super(st
, &dinfo
,
3240 "_reshape_progress",
3242 st
->ss
->store_super(st
, fdlist
[j
]);
3243 st
->ss
->free_super(st
);
3247 /* Didn't find any backup data, try to see if any
3250 if (info
->delta_disks
< 0) {
3251 /* When shrinking, the critical section is at the end.
3252 * So see if we are before the critical section.
3254 unsigned long long first_block
;
3255 nstripe
= ostripe
= 0;
3257 while (ostripe
>= nstripe
) {
3258 ostripe
+= info
->array
.chunk_size
/ 512;
3259 first_block
= ostripe
* odata
;
3260 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
3261 (info
->new_chunk
/512);
3264 if (info
->reshape_progress
>= first_block
)
3267 if (info
->delta_disks
> 0) {
3268 /* See if we are beyond the critical section. */
3269 unsigned long long last_block
;
3270 nstripe
= ostripe
= 0;
3272 while (nstripe
>= ostripe
) {
3273 nstripe
+= info
->new_chunk
/ 512;
3274 last_block
= nstripe
* ndata
;
3275 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
3276 (info
->array
.chunk_size
/512);
3279 if (info
->reshape_progress
>= last_block
)
3282 /* needed to recover critical section! */
3284 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
3288 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
3291 int err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
3294 return reshape_array(NULL
, mdfd
, "array", st
, info
, 1, backup_file
, 0, 0, 1);