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
, int already_running
)
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 if (!already_running
)
637 sysfs_set_num(sra
, NULL
, "sync_min", 0);
638 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", 0);
639 if (!already_running
)
640 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
645 void abort_reshape(struct mdinfo
*sra
)
647 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
648 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
649 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
650 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
651 sysfs_set_num(sra
, NULL
, "sync_min", 0);
652 sysfs_set_str(sra
, NULL
, "sync_max", "max");
655 int remove_disks_for_takeover(struct supertype
*st
,
660 struct mdinfo
*remaining
;
663 if (sra
->array
.level
== 10)
664 nr_of_copies
= layout
& 0xff;
665 else if (sra
->array
.level
== 1)
666 nr_of_copies
= sra
->array
.raid_disks
;
670 remaining
= sra
->devs
;
672 /* for each 'copy', select one device and remove from the list. */
673 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
674 struct mdinfo
**diskp
;
677 /* Find a working device to keep */
678 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
679 struct mdinfo
*disk
= *diskp
;
681 if (disk
->disk
.raid_disk
< slot
)
683 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
685 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
687 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
689 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
692 /* We have found a good disk to use! */
694 disk
->next
= sra
->devs
;
703 if (slot
< sra
->array
.raid_disks
) {
704 /* didn't find all slots */
710 sra
->devs
= remaining
;
714 /* Remove all 'remaining' devices from the array */
716 struct mdinfo
*sd
= remaining
;
717 remaining
= sd
->next
;
719 sysfs_set_str(sra
, sd
, "state", "faulty");
720 sysfs_set_str(sra
, sd
, "slot", "none");
721 /* for external metadata disks should be removed in mdmon */
722 if (!st
->ss
->external
)
723 sysfs_set_str(sra
, sd
, "state", "remove");
724 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
725 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
726 sd
->next
= sra
->devs
;
732 void reshape_free_fdlist(int *fdlist
,
733 unsigned long long *offsets
,
738 for (i
= 0; i
< size
; i
++)
746 int reshape_prepare_fdlist(char *devname
,
750 unsigned long blocks
,
753 unsigned long long *offsets
)
758 for (d
= 0; d
<= nrdisks
; d
++)
761 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
762 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
764 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
765 char *dn
= map_dev(sd
->disk
.major
,
767 fdlist
[sd
->disk
.raid_disk
]
768 = dev_open(dn
, O_RDONLY
);
769 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
770 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
772 Name
": %s: cannot open component %s\n",
773 devname
, dn
? dn
: "-unknown-");
777 } else if (backup_file
== NULL
) {
779 char *dn
= map_dev(sd
->disk
.major
,
781 fdlist
[d
] = dev_open(dn
, O_RDWR
);
782 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
784 fprintf(stderr
, Name
": %s: cannot open component %s\n",
785 devname
, dn
? dn
: "-unknown-");
796 int reshape_open_backup_file(char *backup_file
,
801 unsigned long long *offsets
,
804 /* Return 1 on success, 0 on any form of failure */
805 /* need to check backup file is large enough */
811 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
815 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
816 devname
, backup_file
, strerror(errno
));
819 /* Guard against backup file being on array device.
820 * If array is partitioned or if LVM etc is in the
821 * way this will not notice, but it is better than
824 fstat(*fdlist
, &stb
);
827 if (stb
.st_rdev
== dev
) {
828 fprintf(stderr
, Name
": backup file must NOT be"
829 " on the array being reshaped.\n");
835 for (i
=0; i
< blocks
+ 8 ; i
++) {
836 if (write(*fdlist
, buf
, 512) != 512) {
837 fprintf(stderr
, Name
": %s: cannot create"
838 " backup file %s: %s\n",
839 devname
, backup_file
, strerror(errno
));
843 if (fsync(*fdlist
) != 0) {
844 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
845 devname
, backup_file
, strerror(errno
));
852 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
853 unsigned int ndata
, unsigned int odata
)
855 unsigned long a
, b
, blocks
;
856 /* So how much do we need to backup.
857 * We need an amount of data which is both a whole number of
858 * old stripes and a whole number of new stripes.
859 * So LCM for (chunksize*datadisks).
861 a
= (ochunk
/512) * odata
;
862 b
= (nchunk
/512) * ndata
;
870 /* LCM == product / GCD */
871 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
876 char *analyse_change(struct mdinfo
*info
, struct reshape
*re
)
878 /* Based on the current array state in info->array and
879 * the changes in info->new_* etc, determine:
880 * - whether the change is possible
881 * - Intermediate level/raid_disks/layout
882 * - whether a restriping reshape is needed
883 * - number of sectors in minimum change unit. This
884 * will cover a whole number of stripes in 'before' and
887 * Return message if the change should be rejected
888 * NULL if the change can be achieved
890 * This can be called as part of starting a reshape, or
891 * when assembling an array that is undergoing reshape.
895 /* If a new level not explicitly given, we assume no-change */
896 if (info
->new_level
== UnSet
)
897 info
->new_level
= info
->array
.level
;
900 switch (info
->new_level
) {
906 /* chunk size is meaningful, must divide component_size
909 if (info
->component_size
% (info
->new_chunk
/512))
910 return "New chunk size does not"
911 " divide component size";
914 return "chunk size not meaningful for this level";
917 info
->new_chunk
= info
->array
.chunk_size
;
919 switch (info
->array
.level
) {
921 /* RAID1 can convert to RAID1 with different disks, or
922 * raid5 with 2 disks, or
925 if (info
->new_level
== 0) {
927 re
->before
.data_disks
= 1;
928 re
->after
.data_disks
= 1;
929 re
->before
.layout
= 0;
930 re
->backup_blocks
= 0;
934 if (info
->new_level
== 1) {
935 if (info
->delta_disks
== UnSet
)
936 /* Don't know what to do */
937 return "no change requested for Growing RAID1";
939 re
->backup_blocks
= 0;
943 if (info
->array
.raid_disks
== 2 &&
944 info
->new_level
== 5) {
946 re
->before
.data_disks
= 1;
947 re
->after
.data_disks
= 1;
948 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
949 info
->array
.chunk_size
= 65536;
952 /* Could do some multi-stage conversions, but leave that to
955 return "Impossibly level change request for RAID1";
958 /* RAID10 can only be converted from near mode to
959 * RAID0 by removing some devices
961 if ((info
->array
.layout
& ~0xff) != 0x100)
962 return "Cannot Grow RAID10 with far/offset layout";
963 /* number of devices must be multiple of number of copies */
964 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
965 return "RAID10 layout too complex for Grow operation";
967 if (info
->new_level
!= 0)
968 return "RAID10 can only be changed to RAID0";
969 new_disks
= (info
->array
.raid_disks
970 / (info
->array
.layout
& 0xff));
971 if (info
->delta_disks
== UnSet
) {
972 info
->delta_disks
= (new_disks
973 - info
->array
.raid_disks
);
975 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
976 return "New number of raid-devices impossible for RAID10";
977 if (info
->new_chunk
&&
978 info
->new_chunk
!= info
->array
.chunk_size
)
979 return "Cannot change chunk-size with RAID10 Grow";
984 re
->before
.data_disks
= new_disks
;
985 re
->after
.data_disks
= re
->before
.data_disks
;
986 re
->before
.layout
= 0;
987 re
->backup_blocks
= 0;
991 /* RAID0 can be converted to RAID10, or to RAID456 */
992 if (info
->new_level
== 10) {
993 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
994 /* Assume near=2 layout */
995 info
->new_layout
= 0x102;
996 info
->delta_disks
= info
->array
.raid_disks
;
998 if (info
->new_layout
== UnSet
) {
999 int copies
= 1 + (info
->delta_disks
1000 / info
->array
.raid_disks
);
1001 if (info
->array
.raid_disks
* (copies
-1)
1002 != info
->delta_disks
)
1003 return "Impossible number of devices"
1004 " for RAID0->RAID10";
1005 info
->new_layout
= 0x100 + copies
;
1007 if (info
->delta_disks
== UnSet
) {
1008 int copies
= info
->new_layout
& 0xff;
1009 if (info
->new_layout
!= 0x100 + copies
)
1010 return "New layout impossible"
1011 " for RAID0->RAID10";;
1012 info
->delta_disks
= (copies
- 1) *
1013 info
->array
.raid_disks
;
1015 if (info
->new_chunk
&&
1016 info
->new_chunk
!= info
->array
.chunk_size
)
1017 return "Cannot change chunk-size with RAID0->RAID10";
1021 re
->before
.data_disks
= (info
->array
.raid_disks
+
1023 re
->after
.data_disks
= re
->before
.data_disks
;
1024 re
->before
.layout
= info
->new_layout
;
1025 re
->backup_blocks
= 0;
1029 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1030 * a raid4 style layout of the final level.
1032 switch (info
->new_level
) {
1036 re
->before
.layout
= 0;
1040 re
->before
.layout
= ALGORITHM_PARITY_N
;
1044 re
->before
.layout
= ALGORITHM_PARITY_N
;
1047 return "Impossible level change requested";
1049 re
->before
.data_disks
= info
->array
.raid_disks
;
1050 /* determining 'after' layout happens outside this 'switch' */
1054 info
->array
.layout
= ALGORITHM_PARITY_N
;
1056 switch (info
->new_level
) {
1058 re
->level
= info
->array
.level
;
1059 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1060 re
->before
.layout
= info
->array
.layout
;
1064 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1065 re
->before
.layout
= info
->array
.layout
;
1069 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1070 switch (info
->array
.layout
) {
1071 case ALGORITHM_LEFT_ASYMMETRIC
:
1072 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1074 case ALGORITHM_RIGHT_ASYMMETRIC
:
1075 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1077 case ALGORITHM_LEFT_SYMMETRIC
:
1078 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1080 case ALGORITHM_RIGHT_SYMMETRIC
:
1081 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1083 case ALGORITHM_PARITY_0
:
1084 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1086 case ALGORITHM_PARITY_N
:
1087 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1090 return "Cannot convert an array with this layout";
1094 if (info
->array
.raid_disks
!= 2)
1095 return "Can only convert a 2-device array to RAID1";
1099 return "Impossible level change requested";
1103 switch (info
->new_level
) {
1108 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1109 re
->before
.layout
= info
->array
.layout
;
1112 return "Impossible level change requested";
1117 /* If we reached here then it looks like a re-stripe is
1118 * happening. We have determined the intermediate level
1119 * and initial raid_disks/layout and stored these in 're'.
1121 * We need to deduce the final layout that can be atomically
1122 * converted to the end state.
1124 switch (info
->new_level
) {
1126 /* We can only get to RAID0 from RAID4 or RAID5
1127 * with appropriate layout and one extra device
1129 if (re
->level
!= 4 && re
->level
!= 5)
1130 return "Cannot covert to RAID0 from this level";
1131 if (info
->delta_disks
== UnSet
)
1132 re
->after
.data_disks
= re
->before
.data_disks
;
1134 re
->after
.data_disks
=
1135 info
->array
.raid_disks
+ info
->delta_disks
;
1136 switch (re
->level
) {
1138 re
->after
.layout
= 0 ; break;
1140 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1145 /* We can only get to RAID4 from RAID5 */
1146 if (re
->level
!= 4 && re
->level
!= 5)
1147 return "Cannot convert to RAID4 from this level";
1148 if (info
->delta_disks
== UnSet
)
1149 re
->after
.data_disks
= re
->before
.data_disks
;
1151 re
->after
.data_disks
=
1152 re
->before
.data_disks
+ info
->delta_disks
;
1153 switch (re
->level
) {
1155 re
->after
.layout
= 0 ; break;
1157 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1162 /* We get to RAID5 for RAID5 or RAID6 */
1163 if (re
->level
!= 5 && re
->level
!= 6)
1164 return "Cannot convert to RAID5 from this level";
1165 if (info
->delta_disks
== UnSet
)
1166 re
->after
.data_disks
= re
->before
.data_disks
;
1167 else if (re
->level
== 5)
1168 re
->after
.data_disks
=
1169 re
->before
.data_disks
+ info
->delta_disks
;
1171 re
->after
.data_disks
=
1172 info
->array
.raid_disks
+ info
->delta_disks
- 1;
1173 switch (re
->level
) {
1175 if (info
->new_layout
== UnSet
)
1176 re
->after
.layout
= re
->before
.layout
;
1178 re
->after
.layout
= info
->new_layout
;
1181 if (info
->new_layout
== UnSet
)
1182 info
->new_layout
= re
->before
.layout
;
1184 /* after.layout needs to be raid6 version of new_layout */
1185 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1186 re
->after
.layout
= ALGORITHM_PARITY_N
;
1189 char *ls
= map_num(r5layout
, info
->new_layout
);
1191 strcat(strcpy(layout
, ls
), "-6");
1192 l
= map_name(r6layout
, layout
);
1194 return "Cannot find RAID6 layout"
1196 re
->after
.layout
= l
;
1202 /* We must already be at level 6 */
1204 return "Impossible level change";
1205 if (info
->delta_disks
== UnSet
)
1206 re
->after
.data_disks
= re
->before
.data_disks
;
1208 re
->after
.data_disks
= (info
->array
.raid_disks
+
1209 info
->delta_disks
) - 2;
1210 if (info
->new_layout
== UnSet
)
1211 re
->after
.layout
= info
->array
.layout
;
1213 re
->after
.layout
= info
->new_layout
;
1216 return "Impossible level change requested";
1218 switch (re
->level
) {
1219 case 6: re
->parity
= 2; break;
1221 case 5: re
->parity
= 1; break;
1222 default: re
->parity
= 0; break;
1224 /* So we have a restripe operation, we need to calculate the number
1225 * of blocks per reshape operation.
1227 if (info
->new_chunk
== 0)
1228 info
->new_chunk
= info
->array
.chunk_size
;
1229 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1230 re
->after
.layout
== re
->before
.layout
&&
1231 info
->new_chunk
== info
->array
.chunk_size
) {
1232 /* Nothing to change */
1233 re
->backup_blocks
= 0;
1236 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1237 /* chunk and layout changes make no difference */
1238 re
->backup_blocks
= 0;
1242 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1243 get_linux_version() < 2006032)
1244 return "in-place reshape is not safe before 2.6.32 - sorry.";
1246 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1247 get_linux_version() < 2006030)
1248 return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
1250 re
->backup_blocks
= compute_backup_blocks(
1251 info
->new_chunk
, info
->array
.chunk_size
,
1252 re
->after
.data_disks
,
1253 re
->before
.data_disks
);
1255 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1259 static int reshape_array(char *container
, int fd
, char *devname
,
1260 struct supertype
*st
, struct mdinfo
*info
,
1261 int force
, char *backup_file
, int quiet
, int forked
,
1263 static int reshape_container(char *container
, int cfd
, char *devname
,
1264 struct supertype
*st
,
1265 struct mdinfo
*info
,
1270 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1272 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1275 /* Make some changes in the shape of an array.
1276 * The kernel must support the change.
1278 * There are three different changes. Each can trigger
1279 * a resync or recovery so we freeze that until we have
1280 * requested everything (if kernel supports freezing - 2.6.30).
1282 * - change size (i.e. component_size)
1284 * - change layout/chunksize/ndisks
1286 * The last can require a reshape. It is different on different
1287 * levels so we need to check the level before actioning it.
1288 * Some times the level change needs to be requested after the
1289 * reshape (e.g. raid6->raid5, raid5->raid0)
1292 struct mdu_array_info_s array
;
1294 struct supertype
*st
;
1295 char *subarray
= NULL
;
1299 char *container
= NULL
;
1300 char container_buf
[20];
1306 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1307 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1313 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1314 fprintf(stderr
, Name
": cannot change component size at the same time "
1315 "as other changes.\n"
1316 " Change size first, then check data is intact before "
1317 "making other changes.\n");
1321 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1322 get_linux_version() < 2006032 &&
1323 !check_env("MDADM_FORCE_FEWER")) {
1324 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1325 " Please use a newer kernel\n");
1329 st
= super_by_fd(fd
, &subarray
);
1331 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1334 if (raid_disks
> st
->max_devs
) {
1335 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1336 " beyond %d\n", st
->max_devs
);
1340 /* in the external case we need to check that the requested reshape is
1341 * supported, and perform an initial check that the container holds the
1342 * pre-requisite spare devices (mdmon owns final validation)
1344 if (st
->ss
->external
) {
1349 container_dev
= st
->container_dev
;
1350 cfd
= open_dev_excl(st
->container_dev
);
1352 container_dev
= st
->devnum
;
1354 cfd
= open_dev_excl(st
->devnum
);
1358 fprintf(stderr
, Name
": Unable to open container for %s\n",
1364 fmt_devname(container_buf
, container_dev
);
1365 container
= container_buf
;
1367 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1370 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1376 if (mdmon_running(container_dev
))
1377 st
->update_tail
= &st
->updates
;
1380 if (raid_disks
> array
.raid_disks
&&
1381 array
.spare_disks
< (raid_disks
- array
.raid_disks
) &&
1384 Name
": Need %d spare%s to avoid degraded array,"
1385 " and only have %d.\n"
1386 " Use --force to over-ride this check.\n",
1387 raid_disks
- array
.raid_disks
,
1388 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1393 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
1394 | GET_STATE
| GET_VERSION
);
1396 if (st
->ss
->external
&& subarray
== NULL
) {
1397 array
.level
= LEVEL_CONTAINER
;
1398 sra
->array
.level
= LEVEL_CONTAINER
;
1401 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1405 frozen
= freeze(st
);
1407 /* freeze() already spewed the reason */
1409 } else if (frozen
< 0) {
1410 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1411 " be reshaped\n", devname
);
1415 /* ========= set size =============== */
1416 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1417 long long orig_size
= array
.size
;
1419 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1425 if (array
.size
!= size
) {
1426 /* got truncated to 32bit, write to
1427 * component_size instead
1430 rv
= sysfs_set_num(sra
, NULL
,
1431 "component_size", size
);
1435 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1439 /* restore metadata */
1440 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1441 NULL
, devname
, !quiet
) == 0)
1443 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1444 devname
, strerror(err
));
1446 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1447 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1451 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1452 size
= get_component_size(fd
)/2;
1456 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
1459 } else if (array
.level
!= LEVEL_CONTAINER
) {
1460 size
= get_component_size(fd
)/2;
1465 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1466 * current implementation assumes that following conditions must be met:
1469 * - near_copies == 2
1471 if ((level
== 0 && array
.level
== 10 && sra
&&
1472 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1473 (level
== 0 && array
.level
== 1 && sra
)) {
1475 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1477 dprintf(Name
": Array cannot be reshaped\n");
1483 /* FIXME this is added with no justification - why is it here */
1484 ping_monitor(container
);
1488 sysfs_init(&info
, fd
, NoMdDev
);
1489 strcpy(info
.text_version
, sra
->text_version
);
1490 info
.component_size
= size
*2;
1491 info
.new_level
= level
;
1492 info
.new_chunk
= chunksize
* 1024;
1494 info
.delta_disks
= raid_disks
- info
.array
.raid_disks
;
1496 info
.delta_disks
= UnSet
;
1497 if (layout_str
== NULL
) {
1498 info
.new_layout
= UnSet
;
1499 if (info
.array
.level
== 6 &&
1500 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1501 info
.array
.layout
>= 16) {
1502 fprintf(stderr
, Name
1503 ": %s has a non-standard layout. If you"
1504 " wish to preserve this\n"
1505 " during the reshape, please specify"
1506 " --layout=preserve\n"
1507 " If you want to change it, specify a"
1508 " layout or use --layout=normalise\n",
1513 } else if (strcmp(layout_str
, "normalise") == 0 ||
1514 strcmp(layout_str
, "normalize") == 0) {
1515 /* If we have a -6 RAID6 layout, remove the '-6'. */
1516 info
.new_layout
= UnSet
;
1517 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1519 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1520 h
= strrchr(l
, '-');
1521 if (h
&& strcmp(h
, "-6") == 0) {
1523 info
.new_layout
= map_name(r6layout
, l
);
1526 } else if (strcmp(layout_str
, "preserve") == 0) {
1527 info
.new_layout
= UnSet
;
1529 int l
= info
.new_level
;
1531 l
= info
.array
.level
;
1534 info
.new_layout
= map_name(r5layout
, layout_str
);
1537 info
.new_layout
= map_name(r6layout
, layout_str
);
1540 info
.new_layout
= parse_layout_10(layout_str
);
1543 info
.new_layout
= parse_layout_faulty(layout_str
);
1546 fprintf(stderr
, Name
": layout not meaningful"
1547 " with this level\n");
1551 if (info
.new_layout
== UnSet
) {
1552 fprintf(stderr
, Name
": layout %s not understood"
1553 " for this level\n",
1560 if (array
.level
== LEVEL_CONTAINER
) {
1561 /* This change is to be applied to every array in the
1562 * container. This is only needed when the metadata imposes
1563 * restraints of the various arrays in the container.
1564 * Currently we only know that IMSM requires all arrays
1565 * to have the same number of devices so changing the
1566 * number of devices (On-Line Capacity Expansion) must be
1567 * performed at the level of the container
1569 rv
= reshape_container(container
, fd
, devname
, st
, &info
,
1570 force
, backup_file
, quiet
);
1573 /* Impose these changes on a single array. First
1574 * check that the metadata is OK with the change. */
1576 if (reshape_super(st
, info
.component_size
, info
.new_level
,
1577 info
.new_layout
, info
.new_chunk
,
1578 info
.array
.raid_disks
+ info
.delta_disks
,
1579 backup_file
, devname
, quiet
)) {
1584 rv
= reshape_array(container
, fd
, devname
, st
, &info
, force
,
1585 backup_file
, quiet
, 0, 0);
1594 static int reshape_array(char *container
, int fd
, char *devname
,
1595 struct supertype
*st
, struct mdinfo
*info
,
1597 char *backup_file
, int quiet
, int forked
,
1600 struct reshape reshape
;
1603 int orig_level
= UnSet
;
1606 struct mdu_array_info_s array
;
1610 unsigned long long *offsets
;
1614 unsigned long blocks
;
1615 unsigned long cache
;
1616 unsigned long long array_size
;
1618 struct mdinfo
*sra
= NULL
;
1620 msg
= analyse_change(info
, &reshape
);
1622 fprintf(stderr
, Name
": %s\n", msg
);
1625 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1626 dprintf("Canot get array information.\n");
1631 /* reshape already started. just skip to monitoring the reshape */
1632 if (reshape
.backup_blocks
== 0)
1636 /* The container is frozen but the array may not be.
1637 * So freeze the array so spares don't get put to the wrong use
1638 * FIXME there should probably be a cleaner separation between
1639 * freeze_array and freeze_container.
1641 sysfs_freeze_array(info
);
1642 spares_needed
= max(reshape
.before
.data_disks
,
1643 reshape
.after
.data_disks
)
1644 + reshape
.parity
- array
.raid_disks
;
1647 info
->new_level
> 1 &&
1648 spares_needed
> info
->array
.spare_disks
) {
1650 Name
": Need %d spare%s to avoid degraded array,"
1651 " and only have %d.\n"
1652 " Use --force to over-ride this check.\n",
1654 spares_needed
== 1 ? "" : "s",
1655 info
->array
.spare_disks
);
1659 if (reshape
.level
!= info
->array
.level
) {
1660 char *c
= map_num(pers
, reshape
.level
);
1665 err
= sysfs_set_str(info
, NULL
, "level", c
);
1668 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1671 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1672 fprintf(stderr
, " Bitmap must be removed"
1673 " before level can be changed\n");
1677 fprintf(stderr
, Name
": level of %s changed to %s\n",
1679 orig_level
= info
->array
.level
;
1680 sysfs_freeze_array(info
);
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
= max(reshape
.before
.data_disks
,
1843 reshape
.after
.data_disks
) + reshape
.parity
1844 + sra
->array
.spare_disks
;
1845 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1846 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1847 if (!fdlist
|| !offsets
) {
1848 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1852 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
1853 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
1854 nrdisks
, blocks
, backup_file
,
1859 if (backup_file
== NULL
) {
1860 if (reshape
.after
.data_disks
<= reshape
.before
.data_disks
) {
1862 Name
": %s: Cannot grow - need backup-file\n",
1865 } else if (sra
->array
.spare_disks
== 0) {
1866 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1867 "backup-file to backup critical section\n",
1872 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
1874 fdlist
+d
, offsets
+d
, restart
)) {
1880 /* lastly, check that the internal stripe cache is
1881 * large enough, or it won't work.
1882 * It must hold at least 4 stripes of the larger
1885 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
1886 cache
*= 4; /* 4 stripes minimum */
1887 cache
/= 512; /* convert to sectors */
1888 disks
= min(reshape
.before
.data_disks
, reshape
.after
.data_disks
);
1889 /* make sure there is room for 'blocks' with a bit to spare */
1890 if (cache
< 16 + blocks
/ disks
)
1891 cache
= 16 + blocks
/ disks
;
1892 cache
/= (4096/512); /* Covert from sectors to pages */
1894 if (sra
->cache_size
< cache
)
1895 subarray_set_num(container
, sra
, "stripe_cache_size",
1898 /* Right, everything seems fine. Let's kick things off.
1899 * If only changing raid_disks, use ioctl, else use
1904 sra
->new_chunk
= info
->new_chunk
;
1906 if (info
->reshape_active
)
1907 sra
->reshape_progress
= info
->reshape_progress
;
1909 sra
->reshape_progress
= 0;
1910 if (reshape
.after
.data_disks
< reshape
.before
.data_disks
)
1911 /* start from the end of the new array */
1912 sra
->reshape_progress
= (sra
->component_size
1913 * reshape
.after
.data_disks
);
1916 if (info
->array
.chunk_size
== info
->new_chunk
&&
1917 reshape
.before
.layout
== reshape
.after
.layout
&&
1918 st
->ss
->external
== 0) {
1919 /* use SET_ARRAY_INFO but only if reshape hasn't started */
1920 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1921 array
.raid_disks
= reshape
.after
.data_disks
+ reshape
.parity
;
1922 if (!info
->reshape_active
&&
1923 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1927 Name
": Cannot set device shape for %s: %s\n",
1928 devname
, strerror(errno
));
1931 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1933 " Bitmap must be removed before"
1934 " shape can be changed\n");
1938 } else if (info
->reshape_active
&& !st
->ss
->external
) {
1939 /* We don't need to set anything here for internal
1940 * metadata, and for kernels before 2.6.38 we can
1944 /* set them all just in case some old 'new_*' value
1945 * persists from some earlier problem.
1946 * We even set them when restarting in the middle. They will
1947 * already be set in that case so this will be a no-op,
1948 * but it is hard to tell the difference.
1951 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
1953 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
1954 reshape
.after
.layout
) < 0)
1956 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
1957 reshape
.after
.data_disks
+
1958 reshape
.parity
) < 0)
1961 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1965 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1967 " Bitmap must be removed before"
1968 " shape can be changed\n");
1973 err
= start_reshape(sra
, (info
->reshape_active
&& !st
->ss
->external
));
1975 fprintf(stderr
, Name
": Cannot start reshape for %s\n",
1980 sysfs_set_str(sra
, NULL
, "array_state", "active");
1982 /* Now we just need to kick off the reshape and watch, while
1983 * handling backups of the data...
1984 * This is all done by a forked background process.
1986 switch(forked
? 0 : fork()) {
1988 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
1999 if (check_env("MDADM_GROW_VERIFY"))
2000 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
2003 mlockall(MCL_FUTURE
);
2005 if (st
->ss
->external
) {
2006 /* metadata handler takes it from here */
2007 done
= st
->ss
->manage_reshape(
2008 fd
, sra
, &reshape
, st
, blocks
,
2010 d
- odisks
, fdlist
+odisks
,
2013 done
= child_monitor(
2014 fd
, sra
, &reshape
, st
, blocks
,
2016 d
- odisks
, fdlist
+odisks
,
2019 if (backup_file
&& done
)
2020 unlink(backup_file
);
2026 if (!st
->ss
->external
&&
2027 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
2028 && info
->custom_array_size
) &&
2029 info
->new_level
== reshape
.level
&&
2031 /* no need to wait for the reshape to finish as
2032 * there is nothing more to do.
2038 if (st
->ss
->external
) {
2039 /* Re-load the metadata as much could have changed */
2040 int cfd
= open_dev(st
->container_dev
);
2042 ping_monitor(container
);
2043 st
->ss
->free_super(st
);
2044 st
->ss
->load_container(st
, cfd
, container
);
2049 /* set new array size if required customer_array_size is used
2052 if (reshape
.before
.data_disks
!=
2053 reshape
.after
.data_disks
&&
2054 info
->custom_array_size
) {
2055 struct mdinfo
*info2
;
2056 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2058 info2
= st
->ss
->container_content(st
, subarray
);
2060 unsigned long long current_size
= 0;
2061 unsigned long long new_size
=
2062 info2
->custom_array_size
/2;
2064 if (sysfs_get_ll(sra
,
2067 ¤t_size
) == 0 &&
2068 new_size
> current_size
) {
2069 if (sysfs_set_num(sra
, NULL
,
2070 "array_size", new_size
)
2072 dprintf("Error: Cannot"
2075 dprintf("Array size "
2077 dprintf(" from %llu to %llu.\n",
2078 current_size
, new_size
);
2084 if (info
->new_level
!= reshape
.level
) {
2086 c
= map_num(pers
, info
->new_level
);
2088 err
= sysfs_set_str(sra
, NULL
, "level", c
);
2090 fprintf(stderr
, Name\
2091 ": %s: could not set level "
2092 "to %s\n", devname
, c
);
2102 if (orig_level
!= UnSet
&& sra
) {
2103 c
= map_num(pers
, orig_level
);
2104 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
2105 fprintf(stderr
, Name
": aborting level change\n");
2112 int reshape_container(char *container
, int cfd
, char *devname
,
2113 struct supertype
*st
,
2114 struct mdinfo
*info
,
2119 struct mdinfo
*cc
= NULL
;
2121 /* component_size is not meaningful for a container,
2122 * so pass '-1' meaning 'no change'
2124 if (reshape_super(st
, -1, info
->new_level
,
2125 info
->new_layout
, info
->new_chunk
,
2126 info
->array
.raid_disks
+ info
->delta_disks
,
2127 backup_file
, devname
, quiet
)) {
2134 /* ping monitor to be sure that update is on disk
2136 ping_monitor(container
);
2139 case -1: /* error */
2140 perror("Cannot fork to complete reshape\n");
2143 default: /* parent */
2144 printf(Name
": multi-array reshape continues in background\n");
2151 /* For each member array with reshape_active,
2152 * we need to perform the reshape.
2153 * We pick the first array that needs reshaping and
2154 * reshape it. reshape_array() will re-read the metadata
2155 * so the next time through a different array should be
2156 * ready for reshape.
2158 struct mdinfo
*content
;
2161 struct mdstat_ent
*mdstat
;
2166 cc
= st
->ss
->container_content(st
, NULL
);
2168 for (content
= cc
; content
; content
= content
->next
) {
2170 if (!content
->reshape_active
)
2173 subarray
= strchr(content
->text_version
+1, '/')+1;
2174 mdstat
= mdstat_by_subdev(subarray
,
2175 devname2devnum(container
));
2183 fd
= open_dev(mdstat
->devnum
);
2186 adev
= map_dev(dev2major(mdstat
->devnum
),
2187 dev2minor(mdstat
->devnum
),
2190 adev
= content
->text_version
;
2192 sysfs_init(content
, fd
, mdstat
->devnum
);
2194 rv
= reshape_array(container
, fd
, adev
, st
,
2196 backup_file
, quiet
, 1, 0);
2207 * We run a child process in the background which performs the following
2209 * - wait for resync to reach a certain point
2210 * - suspend io to the following section
2211 * - backup that section
2212 * - allow resync to proceed further
2214 * - discard the backup.
2216 * When are combined in slightly different ways in the three cases.
2218 * - suspend/backup/allow/wait/resume/discard
2220 * - allow/wait/suspend/backup/allow/wait/resume/discard
2222 * - wait/resume/discard/suspend/backup/allow
2224 * suspend/backup/allow always come together
2225 * wait/resume/discard do too.
2226 * For the same-size case we have two backups to improve flow.
2230 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
2231 unsigned long long backup_point
,
2232 unsigned long long wait_point
,
2233 unsigned long long *suspend_point
,
2234 unsigned long long *reshape_completed
)
2236 /* This function is called repeatedly by the reshape manager.
2237 * It determines how much progress can safely be made and allows
2239 * - 'info' identifies the array and particularly records in
2240 * ->reshape_progress the metadata's knowledge of progress
2241 * This is a sector offset from the start of the array
2242 * of the next array block to be relocated. This number
2243 * may increase from 0 or decrease from array_size, depending
2244 * on the type of reshape that is happening.
2245 * Note that in contrast, 'sync_completed' is a block count of the
2246 * reshape so far. It gives the distance between the start point
2247 * (head or tail of device) and the next place that data will be
2248 * written. It always increases.
2249 * - 'reshape' is the structure created by analyse_change
2250 * - 'backup_point' shows how much the metadata manager has backed-up
2251 * data. For reshapes with increasing progress, it is the next address
2252 * to be backed up, previous addresses have been backed-up. For
2253 * decreasing progress, it is the earliest address that has been
2254 * backed up - later address are also backed up.
2255 * So addresses between reshape_progress and backup_point are
2256 * backed up providing those are in the 'correct' order.
2257 * - 'wait_point' is an array address. When reshape_completed
2258 * passes this point, progress_reshape should return. It might
2259 * return earlier if it determines that ->reshape_progress needs
2260 * to be updated or further backup is needed.
2261 * - suspend_point is maintained by progress_reshape and the caller
2262 * should not touch it except to initialise to zero.
2263 * It is an array address and it only increases in 2.6.37 and earlier.
2264 * This makes it difficult to handle reducing reshapes with
2265 * external metadata.
2266 * However: it is similar to backup_point in that it records the
2267 * other end of a suspended region from reshape_progress.
2268 * it is moved to extend the region that is safe to backup and/or
2270 * - reshape_completed is read from sysfs and returned. The caller
2271 * should copy this into ->reshape_progress when it has reason to
2272 * believe that the metadata knows this, and any backup outside this
2276 * 1 if more data from backup_point - but only as far as suspend_point,
2277 * should be backed up
2278 * 0 if things are progressing smoothly
2279 * -1 if the reshape is finished, either because it is all done,
2280 * or due to an error.
2283 int advancing
= (reshape
->after
.data_disks
2284 >= reshape
->before
.data_disks
);
2285 unsigned long long need_backup
; /* All data between start of array and
2286 * here will at some point need to
2289 unsigned long long read_offset
, write_offset
;
2290 unsigned long long write_range
;
2291 unsigned long long max_progress
, target
, completed
;
2292 unsigned long long array_size
= (info
->component_size
2293 * reshape
->before
.data_disks
);
2297 /* First, we unsuspend any region that is now known to be safe.
2298 * If suspend_point is on the 'wrong' side of reshape_progress, then
2299 * we don't have or need suspension at the moment. This is true for
2300 * native metadata when we don't need to back-up.
2303 if (info
->reshape_progress
<= *suspend_point
)
2304 sysfs_set_num(info
, NULL
, "suspend_lo",
2305 info
->reshape_progress
);
2307 /* Note: this won't work in 2.6.37 and before.
2308 * Something somewhere should make sure we don't need it!
2310 if (info
->reshape_progress
>= *suspend_point
)
2311 sysfs_set_num(info
, NULL
, "suspend_hi",
2312 info
->reshape_progress
);
2315 /* Now work out how far it is safe to progress.
2316 * If the read_offset for ->reshape_progress is less than
2317 * 'blocks' beyond the write_offset, we can only progress as far
2319 * Otherwise we can progress until the write_offset for the new location
2320 * reaches (within 'blocks' of) the read_offset at the current location.
2321 * However that region must be suspended unless we are using native
2323 * If we need to suspend more, we limit it to 128M per device, which is
2324 * rather arbitrary and should be some time-based calculation.
2326 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
2327 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
2328 write_range
= info
->new_chunk
/512;
2329 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2330 need_backup
= array_size
;
2332 need_backup
= reshape
->backup_blocks
;
2334 if (read_offset
< write_offset
+ write_range
)
2335 max_progress
= backup_point
;
2339 reshape
->after
.data_disks
;
2341 if (read_offset
> write_offset
- write_range
)
2342 /* Can only progress as far as has been backed up,
2343 * which must be suspended */
2344 max_progress
= backup_point
;
2345 else if (info
->reshape_progress
<= need_backup
)
2346 max_progress
= backup_point
;
2348 if (info
->array
.major_version
>= 0)
2349 /* Can progress until backup is needed */
2350 max_progress
= need_backup
;
2352 /* Can progress until metadata update is required */
2355 reshape
->after
.data_disks
;
2356 /* but data must be suspended */
2357 if (max_progress
< *suspend_point
)
2358 max_progress
= *suspend_point
;
2363 /* We know it is safe to progress to 'max_progress' providing
2364 * it is suspended or we are using native metadata.
2365 * Consider extending suspend_point 128M per device if it
2366 * is less than 64M per device beyond reshape_progress.
2367 * But always do a multiple of 'blocks'
2368 * FIXME this is too big - it takes to long to complete
2371 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
2372 reshape
->after
.data_disks
);
2373 target
/= reshape
->backup_blocks
;
2376 target
*= reshape
->backup_blocks
;
2378 /* For externally managed metadata we always need to suspend IO to
2379 * the area being reshaped so we regularly push suspend_point forward.
2380 * For native metadata we only need the suspend if we are going to do
2384 if ((need_backup
> info
->reshape_progress
2385 || info
->array
.major_version
< 0) &&
2386 *suspend_point
< info
->reshape_progress
+ target
) {
2387 if (need_backup
< *suspend_point
+ 2 * target
)
2388 *suspend_point
= need_backup
;
2389 else if (*suspend_point
+ 2 * target
< array_size
)
2390 *suspend_point
+= 2 * target
;
2392 *suspend_point
= array_size
;
2393 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
2394 if (max_progress
> *suspend_point
)
2395 max_progress
= *suspend_point
;
2398 if (info
->array
.major_version
>= 0) {
2399 /* Only need to suspend when about to backup */
2400 if (info
->reshape_progress
< need_backup
* 2 &&
2401 *suspend_point
> 0) {
2403 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
2404 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
2407 /* Need to suspend continually */
2408 if (info
->reshape_progress
< *suspend_point
)
2409 *suspend_point
= info
->reshape_progress
;
2410 if (*suspend_point
+ target
< info
->reshape_progress
)
2411 /* No need to move suspend region yet */;
2413 if (*suspend_point
>= 2 * target
)
2414 *suspend_point
-= 2 * target
;
2417 sysfs_set_num(info
, NULL
, "suspend_lo",
2420 if (max_progress
< *suspend_point
)
2421 max_progress
= *suspend_point
;
2425 /* now set sync_max to allow that progress. sync_max, like
2426 * sync_completed is a count of sectors written per device, so
2427 * we find the difference between max_progress and the start point,
2428 * and divide that by after.data_disks to get a sync_max
2430 * At the same time we convert wait_point to a similar number
2431 * for comparing against sync_completed.
2433 /* scale down max_progress to per_disk */
2434 max_progress
/= reshape
->after
.data_disks
;
2435 /* Round to chunk size as some kernels give an erroneously high number */
2436 max_progress
/= info
->new_chunk
/512;
2437 max_progress
*= info
->new_chunk
/512;
2438 /* And round to old chunk size as the kernel wants that */
2439 max_progress
/= info
->array
.chunk_size
/512;
2440 max_progress
*= info
->array
.chunk_size
/512;
2441 /* Limit progress to the whole device */
2442 if (max_progress
> info
->component_size
)
2443 max_progress
= info
->component_size
;
2444 wait_point
/= reshape
->after
.data_disks
;
2446 /* switch from 'device offset' to 'processed block count' */
2447 max_progress
= info
->component_size
- max_progress
;
2448 wait_point
= info
->component_size
- wait_point
;
2451 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
2453 /* Now wait. If we have already reached the point that we were
2454 * asked to wait to, don't wait at all, else wait for any change.
2455 * We need to select on 'sync_completed' as that is the place that
2456 * notifications happen, but we are really interested in
2457 * 'reshape_position'
2459 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
2461 goto check_progress
;
2463 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2465 goto check_progress
;
2467 while (completed
< max_progress
&& completed
< wait_point
) {
2468 /* Check that sync_action is still 'reshape' to avoid
2469 * waiting forever on a dead array
2473 if (sysfs_get_str(info
, NULL
, "sync_action",
2475 strncmp(action
, "reshape", 7) != 0)
2477 /* Some kernels reset 'sync_completed' to zero
2478 * before setting 'sync_action' to 'idle'.
2479 * So we need these extra tests.
2481 if (completed
== 0 && advancing
2482 && info
->reshape_progress
> 0)
2484 if (completed
== 0 && !advancing
2485 && info
->reshape_progress
< (info
->component_size
2486 * reshape
->after
.data_disks
))
2490 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2491 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2493 goto check_progress
;
2496 /* Some kernels reset 'sync_completed' to zero,
2497 * we need to have real point we are in md
2500 completed
= max_progress
;
2502 /* some kernels can give an incorrectly high 'completed' number */
2503 completed
/= (info
->new_chunk
/512);
2504 completed
*= (info
->new_chunk
/512);
2505 /* Convert 'completed' back in to a 'progress' number */
2506 completed
*= reshape
->after
.data_disks
;
2508 completed
= info
->component_size
* reshape
->after
.data_disks
2511 *reshape_completed
= completed
;
2515 /* We return the need_backup flag. Caller will decide
2516 * how much - a multiple of ->backup_blocks up to *suspend_point
2519 return need_backup
> info
->reshape_progress
;
2521 return need_backup
>= info
->reshape_progress
;
2524 /* if we couldn't read a number from sync_completed, then
2525 * either the reshape did complete, or it aborted.
2526 * We can tell which by checking for 'none' in reshape_position.
2529 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
2530 || strncmp(buf
, "none", 4) != 0)
2531 return -2; /* abort */
2533 /* Maybe racing with array shutdown - check state */
2534 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
2535 || strncmp(buf
, "inactive", 8) == 0
2536 || strncmp(buf
, "clear",5) == 0)
2537 return -2; /* abort */
2538 return -1; /* complete */
2543 /* FIXME return status is never checked */
2544 static int grow_backup(struct mdinfo
*sra
,
2545 unsigned long long offset
, /* per device */
2546 unsigned long stripes
, /* per device, in old chunks */
2547 int *sources
, unsigned long long *offsets
,
2548 int disks
, int chunk
, int level
, int layout
,
2549 int dests
, int *destfd
, unsigned long long *destoffsets
,
2550 int part
, int *degraded
,
2553 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2554 * to storage 'destfd' (offset 'destoffsets'), after first
2555 * suspending IO. Then allow resync to continue
2556 * over the suspended section.
2557 * Use part 'part' of the backup-super-block.
2562 unsigned long long ll
;
2564 //printf("offset %llu\n", offset);
2570 /* Check that array hasn't become degraded, else we might backup the wrong data */
2571 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
2572 return -1; /* FIXME this error is ignored */
2573 new_degraded
= (int)ll
;
2574 if (new_degraded
!= *degraded
) {
2575 /* check each device to ensure it is still working */
2577 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2578 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2580 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2582 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2583 strstr(sbuf
, "faulty") ||
2584 strstr(sbuf
, "in_sync") == NULL
) {
2585 /* this device is dead */
2586 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2587 if (sd
->disk
.raid_disk
>= 0 &&
2588 sources
[sd
->disk
.raid_disk
] >= 0) {
2589 close(sources
[sd
->disk
.raid_disk
]);
2590 sources
[sd
->disk
.raid_disk
] = -1;
2595 *degraded
= new_degraded
;
2598 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2599 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2601 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2602 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2605 bsb
.magic
[15] = '2';
2606 for (i
= 0; i
< dests
; i
++)
2608 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
2610 lseek64(destfd
[i
], destoffsets
[i
], 0);
2612 rv
= save_stripes(sources
, offsets
,
2613 disks
, chunk
, level
, layout
,
2615 offset
*512*odata
, stripes
* chunk
* odata
,
2620 bsb
.mtime
= __cpu_to_le64(time(0));
2621 for (i
= 0; i
< dests
; i
++) {
2622 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2624 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2625 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2626 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2627 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2630 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
2631 != destoffsets
[i
] - 4096)
2633 if (write(destfd
[i
], &bsb
, 512) != 512)
2635 if (destoffsets
[i
] > 4096) {
2636 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
2637 destoffsets
[i
]+stripes
*chunk
*odata
)
2639 if (write(destfd
[i
], &bsb
, 512) != 512)
2649 /* in 2.6.30, the value reported by sync_completed can be
2650 * less that it should be by one stripe.
2651 * This only happens when reshape hits sync_max and pauses.
2652 * So allow wait_backup to either extent sync_max further
2653 * than strictly necessary, or return before the
2654 * sync has got quite as far as we would really like.
2655 * This is what 'blocks2' is for.
2656 * The various caller give appropriate values so that
2659 /* FIXME return value is often ignored */
2660 static int forget_backup(
2661 int dests
, int *destfd
, unsigned long long *destoffsets
,
2665 * Erase backup 'part' (which is 0 or 1)
2671 bsb
.arraystart2
= __cpu_to_le64(0);
2672 bsb
.length2
= __cpu_to_le64(0);
2674 bsb
.arraystart
= __cpu_to_le64(0);
2675 bsb
.length
= __cpu_to_le64(0);
2677 bsb
.mtime
= __cpu_to_le64(time(0));
2679 for (i
= 0; i
< dests
; i
++) {
2680 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2681 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2682 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2683 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2684 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2685 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
2686 destoffsets
[i
]-4096)
2689 write(destfd
[i
], &bsb
, 512) != 512)
2696 static void fail(char *msg
)
2699 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
2700 rv
|= (write(2, "\n", 1) != 1);
2704 static char *abuf
, *bbuf
;
2705 static unsigned long long abuflen
;
2706 static void validate(int afd
, int bfd
, unsigned long long offset
)
2708 /* check that the data in the backup against the array.
2709 * This is only used for regression testing and should not
2710 * be used while the array is active
2714 lseek64(bfd
, offset
- 4096, 0);
2715 if (read(bfd
, &bsb2
, 512) != 512)
2716 fail("cannot read bsb");
2717 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
2718 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
2719 fail("first csum bad");
2720 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
2721 fail("magic is bad");
2722 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
2723 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
2724 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
2725 fail("second csum bad");
2727 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
2728 fail("devstart is wrong");
2731 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
2733 if (abuflen
< len
) {
2737 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
2738 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
2740 /* just stop validating on mem-alloc failure */
2745 lseek64(bfd
, offset
, 0);
2746 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
2747 //printf("len %llu\n", len);
2748 fail("read first backup failed");
2750 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
2751 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2752 fail("read first from array failed");
2753 if (memcmp(bbuf
, abuf
, len
) != 0) {
2756 printf("offset=%llu len=%llu\n",
2757 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
2758 for (i
=0; i
<len
; i
++)
2759 if (bbuf
[i
] != abuf
[i
]) {
2760 printf("first diff byte %d\n", i
);
2764 fail("data1 compare failed");
2768 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
2770 if (abuflen
< len
) {
2774 abuf
= malloc(abuflen
);
2775 bbuf
= malloc(abuflen
);
2778 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2779 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2780 fail("read second backup failed");
2781 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2782 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2783 fail("read second from array failed");
2784 if (memcmp(bbuf
, abuf
, len
) != 0)
2785 fail("data2 compare failed");
2789 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
2790 struct supertype
*st
, unsigned long blocks
,
2791 int *fds
, unsigned long long *offsets
,
2792 int dests
, int *destfd
, unsigned long long *destoffsets
)
2794 /* Monitor a reshape where backup is being performed using
2795 * 'native' mechanism - either to a backup file, or
2796 * to some space in a spare.
2800 unsigned long long speed
;
2801 unsigned long long suspend_point
, array_size
;
2802 unsigned long long backup_point
, wait_point
;
2803 unsigned long long reshape_completed
;
2805 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
2806 int part
= 0; /* The next part of the backup area to fill. It may already
2807 * be full, so we need to check */
2808 int level
= reshape
->level
;
2809 int layout
= reshape
->before
.layout
;
2810 int data
= reshape
->before
.data_disks
;
2811 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
2812 int chunk
= sra
->array
.chunk_size
;
2814 unsigned long stripes
;
2816 /* set up the backup-super-block. This requires the
2817 * uuid from the array.
2819 /* Find a superblock */
2820 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2824 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2826 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
2827 devfd
= dev_open(dn
, O_RDONLY
);
2830 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
2836 fprintf(stderr
, Name
": Cannot find a superblock\n");
2840 memset(&bsb
, 0, 512);
2841 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2842 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
2843 bsb
.mtime
= __cpu_to_le64(time(0));
2844 bsb
.devstart2
= blocks
;
2846 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
2847 reshape
->before
.data_disks
;
2849 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2850 /* Don't start the 'reshape' */
2852 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
2853 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2854 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2858 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
2859 backup_point
= sra
->reshape_progress
;
2862 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
2863 backup_point
= reshape
->backup_blocks
;
2864 suspend_point
= array_size
;
2870 /* Want to return as soon the oldest backup slot can
2871 * be released as that allows us to start backing up
2872 * some more, providing suspend_point has been
2873 * advanced, which it should have.
2876 wait_point
= array_size
;
2877 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
2878 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
2879 __le64_to_cpu(bsb
.length
));
2880 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
2881 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
2882 __le64_to_cpu(bsb
.length2
));
2885 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
2886 wait_point
= __le64_to_cpu(bsb
.arraystart
);
2887 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
2888 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
2891 rv
= progress_reshape(sra
, reshape
,
2892 backup_point
, wait_point
,
2893 &suspend_point
, &reshape_completed
);
2894 /* external metadata would need to ping_monitor here */
2895 sra
->reshape_progress
= reshape_completed
;
2897 /* Clear any backup region that is before 'here' */
2899 if (reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
2900 __le64_to_cpu(bsb
.length
)))
2901 forget_backup(dests
, destfd
,
2903 if (reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
2904 __le64_to_cpu(bsb
.length2
)))
2905 forget_backup(dests
, destfd
,
2908 if (reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
2909 forget_backup(dests
, destfd
,
2911 if (reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
2912 forget_backup(dests
, destfd
,
2923 unsigned long long offset
;
2924 unsigned long actual_stripes
;
2925 /* Need to backup some data.
2926 * If 'part' is not used and the desired
2927 * backup size is suspended, do a backup,
2928 * then consider the next part.
2930 /* Check that 'part' is unused */
2931 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
2933 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
2936 offset
= backup_point
/ data
;
2937 actual_stripes
= stripes
;
2939 if (offset
+ actual_stripes
* (chunk
/512) >
2940 sra
->component_size
)
2941 actual_stripes
= ((sra
->component_size
- offset
)
2943 if (offset
+ actual_stripes
* (chunk
/512) >
2947 if (offset
< actual_stripes
* (chunk
/512))
2948 actual_stripes
= offset
/ (chunk
/512);
2949 offset
-= actual_stripes
* (chunk
/512);
2950 if (offset
< suspend_point
/data
)
2953 if (actual_stripes
== 0)
2955 grow_backup(sra
, offset
, actual_stripes
,
2957 disks
, chunk
, level
, layout
,
2958 dests
, destfd
, destoffsets
,
2959 part
, °raded
, buf
);
2960 validate(afd
, destfd
[0], destoffsets
[0]);
2961 /* record where 'part' is up to */
2964 backup_point
+= actual_stripes
* (chunk
/512) * data
;
2966 backup_point
-= actual_stripes
* (chunk
/512) * data
;
2970 /* FIXME maybe call progress_reshape one more time instead */
2971 abort_reshape(sra
); /* remove any remaining suspension */
2972 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2973 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
2979 * If any spare contains md_back_data-1 which is recent wrt mtime,
2980 * write that data into the array and update the super blocks with
2981 * the new reshape_progress
2983 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
2984 char *backup_file
, int verbose
)
2988 unsigned long long *offsets
;
2989 unsigned long long nstripe
, ostripe
;
2992 if (info
->new_level
!= info
->array
.level
)
2993 return 1; /* Cannot handle level changes (they are instantaneous) */
2995 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
2996 if (info
->array
.level
== 6) odata
--; /* number of data disks */
2997 ndata
= info
->array
.raid_disks
- 1;
2998 if (info
->new_level
== 6) ndata
--;
3000 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
3002 if (info
->delta_disks
<= 0)
3003 /* Didn't grow, so the backup file must have
3007 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
3008 struct mdinfo dinfo
;
3011 char *devname
, namebuf
[20];
3012 unsigned long long lo
, hi
;
3014 /* This was a spare and may have some saved data on it.
3015 * Load the superblock, find and load the
3016 * backup_super_block.
3017 * If either fail, go on to next device.
3018 * If the backup contains no new info, just return
3019 * else restore data and update all superblocks
3021 if (i
== old_disks
-1) {
3022 fd
= open(backup_file
, O_RDONLY
);
3024 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
3025 backup_file
, strerror(errno
));
3028 devname
= backup_file
;
3033 if (st
->ss
->load_super(st
, fd
, NULL
))
3036 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3037 st
->ss
->free_super(st
);
3040 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
3042 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
3043 continue; /* Cannot seek */
3045 sprintf(namebuf
, "device-%d", i
);
3048 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
3050 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
3051 continue; /* Cannot read */
3053 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
3054 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
3056 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
3059 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
3061 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
3062 continue; /* bad checksum */
3064 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
3065 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
3067 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
3068 continue; /* Bad second checksum */
3070 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
3072 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
3073 continue; /* Wrong uuid */
3076 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3077 * sometimes they aren't... So allow considerable flexability in matching, and allow
3078 * this test to be overridden by an environment variable.
3080 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
3081 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
3082 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3083 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
3084 "for array with timestamp %lu\n",
3085 (unsigned long)__le64_to_cpu(bsb
.mtime
),
3086 (unsigned long)info
->array
.utime
);
3089 fprintf(stderr
, Name
": too-old timestamp on "
3090 "backup-metadata on %s\n", devname
);
3091 continue; /* time stamp is too bad */
3095 if (bsb
.magic
[15] == '1') {
3096 if (bsb
.length
== 0)
3098 if (info
->delta_disks
>= 0) {
3099 /* reshape_progress is increasing */
3100 if (__le64_to_cpu(bsb
.arraystart
)
3101 + __le64_to_cpu(bsb
.length
)
3102 < info
->reshape_progress
) {
3105 fprintf(stderr
, Name
3106 ": backup-metadata found on %s but is not needed\n", devname
);
3107 continue; /* No new data here */
3110 /* reshape_progress is decreasing */
3111 if (__le64_to_cpu(bsb
.arraystart
) >=
3112 info
->reshape_progress
)
3113 goto nonew
; /* No new data here */
3116 if (bsb
.length
== 0 && bsb
.length2
== 0)
3118 if (info
->delta_disks
>= 0) {
3119 /* reshape_progress is increasing */
3120 if ((__le64_to_cpu(bsb
.arraystart
)
3121 + __le64_to_cpu(bsb
.length
)
3122 < info
->reshape_progress
)
3124 (__le64_to_cpu(bsb
.arraystart2
)
3125 + __le64_to_cpu(bsb
.length2
)
3126 < info
->reshape_progress
))
3127 goto nonew
; /* No new data here */
3129 /* reshape_progress is decreasing */
3130 if (__le64_to_cpu(bsb
.arraystart
) >=
3131 info
->reshape_progress
&&
3132 __le64_to_cpu(bsb
.arraystart2
) >=
3133 info
->reshape_progress
)
3134 goto nonew
; /* No new data here */
3137 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
3140 fprintf(stderr
, Name
3141 ": Failed to verify secondary backup-metadata block on %s\n",
3143 continue; /* Cannot seek */
3145 /* There should be a duplicate backup superblock 4k before here */
3146 if (lseek64(fd
, -4096, 1) < 0 ||
3147 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
3148 goto second_fail
; /* Cannot find leading superblock */
3149 if (bsb
.magic
[15] == '1')
3150 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
3152 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
3153 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
3154 goto second_fail
; /* Cannot find leading superblock */
3156 /* Now need the data offsets for all devices. */
3157 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
3158 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
3161 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3162 /* FIXME should be this be an error */
3164 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3165 st
->ss
->free_super(st
);
3166 offsets
[j
] = dinfo
.data_offset
* 512;
3168 printf(Name
": restoring critical section\n");
3170 if (restore_stripes(fdlist
, offsets
,
3171 info
->array
.raid_disks
,
3175 fd
, __le64_to_cpu(bsb
.devstart
)*512,
3176 __le64_to_cpu(bsb
.arraystart
)*512,
3177 __le64_to_cpu(bsb
.length
)*512)) {
3178 /* didn't succeed, so giveup */
3180 fprintf(stderr
, Name
": Error restoring backup from %s\n",
3185 if (bsb
.magic
[15] == '2' &&
3186 restore_stripes(fdlist
, offsets
,
3187 info
->array
.raid_disks
,
3191 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
3192 __le64_to_cpu(bsb
.devstart2
)*512,
3193 __le64_to_cpu(bsb
.arraystart2
)*512,
3194 __le64_to_cpu(bsb
.length2
)*512)) {
3195 /* didn't succeed, so giveup */
3197 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
3203 /* Ok, so the data is restored. Let's update those superblocks. */
3207 lo
= __le64_to_cpu(bsb
.arraystart
);
3208 hi
= lo
+ __le64_to_cpu(bsb
.length
);
3210 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
3211 unsigned long long lo1
, hi1
;
3212 lo1
= __le64_to_cpu(bsb
.arraystart2
);
3213 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
3217 } else if (lo
< lo1
)
3223 (info
->reshape_progress
< lo
||
3224 info
->reshape_progress
> hi
))
3225 /* backup does not affect reshape_progress*/ ;
3226 else if (info
->delta_disks
>= 0) {
3227 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
3228 __le64_to_cpu(bsb
.length
);
3229 if (bsb
.magic
[15] == '2') {
3230 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
3231 __le64_to_cpu(bsb
.length2
);
3232 if (p2
> info
->reshape_progress
)
3233 info
->reshape_progress
= p2
;
3236 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
3237 if (bsb
.magic
[15] == '2') {
3238 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
3239 if (p2
< info
->reshape_progress
)
3240 info
->reshape_progress
= p2
;
3243 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
3244 if (fdlist
[j
] < 0) continue;
3245 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3247 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3248 dinfo
.reshape_progress
= info
->reshape_progress
;
3249 st
->ss
->update_super(st
, &dinfo
,
3250 "_reshape_progress",
3252 st
->ss
->store_super(st
, fdlist
[j
]);
3253 st
->ss
->free_super(st
);
3257 /* Didn't find any backup data, try to see if any
3260 if (info
->delta_disks
< 0) {
3261 /* When shrinking, the critical section is at the end.
3262 * So see if we are before the critical section.
3264 unsigned long long first_block
;
3265 nstripe
= ostripe
= 0;
3267 while (ostripe
>= nstripe
) {
3268 ostripe
+= info
->array
.chunk_size
/ 512;
3269 first_block
= ostripe
* odata
;
3270 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
3271 (info
->new_chunk
/512);
3274 if (info
->reshape_progress
>= first_block
)
3277 if (info
->delta_disks
> 0) {
3278 /* See if we are beyond the critical section. */
3279 unsigned long long last_block
;
3280 nstripe
= ostripe
= 0;
3282 while (nstripe
>= ostripe
) {
3283 nstripe
+= info
->new_chunk
/ 512;
3284 last_block
= nstripe
* ndata
;
3285 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
3286 (info
->array
.chunk_size
/512);
3289 if (info
->reshape_progress
>= last_block
)
3292 /* needed to recover critical section! */
3294 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
3298 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
3301 int err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
3304 return reshape_array(NULL
, mdfd
, "array", st
, info
, 1, backup_file
, 0, 0, 1);