2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
28 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
29 #error no endian defined
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
38 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
40 /* Add a device to an active array.
41 * Currently, just extend a linear array.
42 * This requires writing a new superblock on the
43 * new device, calling the kernel to add the device,
44 * and if that succeeds, update the superblock on
46 * This means that we need to *find* all other devices.
53 struct supertype
*st
= NULL
;
54 char *subarray
= NULL
;
56 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
57 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
61 if (info
.array
.level
!= -1) {
62 fprintf(stderr
, Name
": can only add devices to linear arrays\n");
66 st
= super_by_fd(fd
, &subarray
);
68 fprintf(stderr
, Name
": cannot handle arrays with superblock version %d\n", info
.array
.major_version
);
73 fprintf(stderr
, Name
": Cannot grow linear sub-arrays yet\n");
78 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
80 fprintf(stderr
, Name
": cannot open %s\n", newdev
);
85 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
86 fprintf(stderr
, Name
": %s is not a block device!\n", newdev
);
91 /* now check out all the devices and make sure we can read the superblock */
92 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
96 st
->ss
->free_super(st
);
99 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
100 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
106 dv
= map_dev(disk
.major
, disk
.minor
, 1);
108 fprintf(stderr
, Name
": cannot find device file for device %d\n",
114 fd2
= dev_open(dv
, O_RDWR
);
116 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
122 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
123 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
131 /* Ok, looks good. Lets update the superblock and write it out to
135 info
.disk
.number
= d
;
136 info
.disk
.major
= major(stb
.st_rdev
);
137 info
.disk
.minor
= minor(stb
.st_rdev
);
138 info
.disk
.raid_disk
= d
;
139 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
140 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
143 if (st
->ss
->store_super(st
, nfd
)) {
144 fprintf(stderr
, Name
": Cannot store new superblock on %s\n",
151 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
152 fprintf(stderr
, Name
": Cannot add new disk to this array\n");
155 /* Well, that seems to have worked.
156 * Now go through and update all superblocks
159 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
160 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
165 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
166 mdu_disk_info_t disk
;
170 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
171 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
175 dv
= map_dev(disk
.major
, disk
.minor
, 1);
177 fprintf(stderr
, Name
": cannot find device file for device %d\n",
181 fd2
= dev_open(dv
, O_RDWR
);
183 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
186 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
187 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
191 info
.array
.raid_disks
= nd
+1;
192 info
.array
.nr_disks
= nd
+1;
193 info
.array
.active_disks
= nd
+1;
194 info
.array
.working_disks
= nd
+1;
196 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
199 if (st
->ss
->store_super(st
, fd2
)) {
200 fprintf(stderr
, Name
": Cannot store new superblock on %s\n", dv
);
210 int Grow_addbitmap(char *devname
, int fd
, char *file
, int chunk
, int delay
, int write_behind
, int force
)
213 * First check that array doesn't have a bitmap
214 * Then create the bitmap
217 * For internal bitmaps, we need to check the version,
218 * find all the active devices, and write the bitmap block
221 mdu_bitmap_file_t bmf
;
222 mdu_array_info_t array
;
223 struct supertype
*st
;
224 char *subarray
= NULL
;
225 int major
= BITMAP_MAJOR_HI
;
226 int vers
= md_get_version(fd
);
227 unsigned long long bitmapsize
, array_size
;
230 major
= BITMAP_MAJOR_HOSTENDIAN
;
231 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel"
232 " are not portable\n"
233 " between different architectures. Consider upgrading"
234 " the Linux kernel.\n");
237 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
239 fprintf(stderr
, Name
": Memory allocation failure.\n");
241 fprintf(stderr
, Name
": bitmaps not supported by this kernel.\n");
244 if (bmf
.pathname
[0]) {
245 if (strcmp(file
,"none")==0) {
246 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
247 fprintf(stderr
, Name
": failed to remove bitmap %s\n",
253 fprintf(stderr
, Name
": %s already has a bitmap (%s)\n",
254 devname
, bmf
.pathname
);
257 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
258 fprintf(stderr
, Name
": cannot get array status for %s\n", devname
);
261 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
262 if (strcmp(file
, "none")==0) {
263 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
264 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
265 fprintf(stderr
, Name
": failed to remove internal bitmap.\n");
270 fprintf(stderr
, Name
": Internal bitmap already present on %s\n",
275 if (strcmp(file
, "none") == 0) {
276 fprintf(stderr
, Name
": no bitmap found on %s\n", devname
);
279 if (array
.level
<= 0) {
280 fprintf(stderr
, Name
": Bitmaps not meaningful with level %s\n",
281 map_num(pers
, array
.level
)?:"of this array");
284 bitmapsize
= array
.size
;
286 if (get_dev_size(fd
, NULL
, &array_size
) &&
287 array_size
> (0x7fffffffULL
<<9)) {
288 /* Array is big enough that we cannot trust array.size
289 * try other approaches
291 bitmapsize
= get_component_size(fd
);
293 if (bitmapsize
== 0) {
294 fprintf(stderr
, Name
": Cannot reliably determine size of array to create bitmap - sorry.\n");
298 if (array
.level
== 10) {
299 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
300 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
303 st
= super_by_fd(fd
, &subarray
);
305 fprintf(stderr
, Name
": Cannot understand version %d.%d\n",
306 array
.major_version
, array
.minor_version
);
310 fprintf(stderr
, Name
": Cannot add bitmaps to sub-arrays yet\n");
315 if (strcmp(file
, "internal") == 0) {
317 if (st
->ss
->add_internal_bitmap
== NULL
) {
318 fprintf(stderr
, Name
": Internal bitmaps not supported "
319 "with %s metadata\n", st
->ss
->name
);
322 for (d
=0; d
< st
->max_devs
; d
++) {
323 mdu_disk_info_t disk
;
326 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
328 if (disk
.major
== 0 &&
331 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
333 dv
= map_dev(disk
.major
, disk
.minor
, 1);
335 int fd2
= dev_open(dv
, O_RDWR
);
338 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
339 if (st
->ss
->add_internal_bitmap(
341 &chunk
, delay
, write_behind
,
342 bitmapsize
, 0, major
)
344 st
->ss
->write_bitmap(st
, fd2
);
346 fprintf(stderr
, Name
": failed to create internal bitmap - chunksize problem.\n");
354 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
355 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
358 ": Cannot add bitmap while array is"
359 " resyncing or reshaping etc.\n");
360 fprintf(stderr
, Name
": failed to set internal bitmap.\n");
367 int max_devs
= st
->max_devs
;
369 /* try to load a superblock */
370 for (d
=0; d
<max_devs
; d
++) {
371 mdu_disk_info_t disk
;
375 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
377 if ((disk
.major
==0 && disk
.minor
==0) ||
378 (disk
.state
& (1<<MD_DISK_REMOVED
)))
380 dv
= map_dev(disk
.major
, disk
.minor
, 1);
382 fd2
= dev_open(dv
, O_RDONLY
);
384 st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
386 st
->ss
->uuid_from_super(st
, uuid
);
392 fprintf(stderr
, Name
": cannot find UUID for array!\n");
395 if (CreateBitmap(file
, force
, (char*)uuid
, chunk
,
396 delay
, write_behind
, bitmapsize
, major
)) {
399 bitmap_fd
= open(file
, O_RDWR
);
401 fprintf(stderr
, Name
": weird: %s cannot be opened\n",
405 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
409 ": Cannot add bitmap while array is"
410 " resyncing or reshaping etc.\n");
411 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
412 devname
, strerror(err
));
422 * When reshaping an array we might need to backup some data.
423 * This is written to all spares with a 'super_block' describing it.
424 * The superblock goes 4K from the end of the used space on the
426 * It if written after the backup is complete.
427 * It has the following structure.
430 static struct mdp_backup_super
{
431 char magic
[16]; /* md_backup_data-1 or -2 */
434 /* start/sizes in 512byte sectors */
435 __u64 devstart
; /* address on backup device/file of data */
438 __u32 sb_csum
; /* csum of preceeding bytes. */
440 __u64 devstart2
; /* offset in to data of second section */
443 __u32 sb_csum2
; /* csum of preceeding bytes. */
445 } __attribute__((aligned(512))) bsb
, bsb2
;
447 static __u32
bsb_csum(char *buf
, int len
)
451 for (i
=0; i
<len
; i
++)
452 csum
= (csum
<<3) + buf
[0];
453 return __cpu_to_le32(csum
);
456 static int check_idle(struct supertype
*st
)
458 /* Check that all member arrays for this container, or the
459 * container of this array, are idle
461 int container_dev
= (st
->container_dev
!= NoMdDev
462 ? st
->container_dev
: st
->devnum
);
464 struct mdstat_ent
*ent
, *e
;
467 fmt_devname(container
, container_dev
);
468 ent
= mdstat_read(0, 0);
469 for (e
= ent
; e
; e
= e
->next
) {
470 if (!is_container_member(e
, container
))
472 if (e
->percent
>= 0) {
481 static int freeze_container(struct supertype
*st
)
483 int container_dev
= (st
->container_dev
!= NoMdDev
484 ? st
->container_dev
: st
->devnum
);
490 fmt_devname(container
, container_dev
);
492 if (block_monitor(container
, 1)) {
493 fprintf(stderr
, Name
": failed to freeze container\n");
500 static void unfreeze_container(struct supertype
*st
)
502 int container_dev
= (st
->container_dev
!= NoMdDev
503 ? st
->container_dev
: st
->devnum
);
506 fmt_devname(container
, container_dev
);
508 unblock_monitor(container
, 1);
511 static int freeze(struct supertype
*st
)
513 /* Try to freeze resync/rebuild on this array/container.
514 * Return -1 if the array is busy,
515 * return -2 container cannot be frozen,
516 * return 0 if this kernel doesn't support 'frozen'
517 * return 1 if it worked.
519 if (st
->ss
->external
)
520 return freeze_container(st
);
522 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
527 err
= sysfs_freeze_array(sra
);
533 static void unfreeze(struct supertype
*st
)
535 if (st
->ss
->external
)
536 return unfreeze_container(st
);
538 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
541 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
543 fprintf(stderr
, Name
": failed to unfreeze array\n");
548 static void wait_reshape(struct mdinfo
*sra
)
550 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
556 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
557 strncmp(action
, "reshape", 7) == 0) {
561 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
566 static int reshape_super(struct supertype
*st
, long long size
, int level
,
567 int layout
, int chunksize
, int raid_disks
,
568 int delta_disks
, char *backup_file
, char *dev
,
571 /* nothing extra to check in the native case */
572 if (!st
->ss
->external
)
574 if (!st
->ss
->reshape_super
||
575 !st
->ss
->manage_reshape
) {
576 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
581 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
582 raid_disks
, delta_disks
, backup_file
, dev
,
586 static void sync_metadata(struct supertype
*st
)
588 if (st
->ss
->external
) {
589 if (st
->update_tail
) {
590 flush_metadata_updates(st
);
591 st
->update_tail
= &st
->updates
;
593 st
->ss
->sync_metadata(st
);
597 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
599 /* when dealing with external metadata subarrays we need to be
600 * prepared to handle EAGAIN. The kernel may need to wait for
601 * mdmon to mark the array active so the kernel can handle
602 * allocations/writeback when preparing the reshape action
603 * (md_allow_write()). We temporarily disable safe_mode_delay
604 * to close a race with the array_state going clean before the
605 * next write to raid_disks / stripe_cache_size
610 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
612 (strcmp(name
, "raid_disks") != 0 &&
613 strcmp(name
, "stripe_cache_size") != 0))
614 return sysfs_set_num(sra
, NULL
, name
, n
);
616 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
619 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
620 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
621 if (rc
< 0 && errno
== EAGAIN
) {
622 ping_monitor(container
);
623 /* if we get EAGAIN here then the monitor is not active
626 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
628 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
632 int start_reshape(struct mdinfo
*sra
, int already_running
)
635 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
636 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
637 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
638 if (!already_running
)
639 sysfs_set_num(sra
, NULL
, "sync_min", 0);
640 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", 0);
641 if (!already_running
)
642 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
647 void abort_reshape(struct mdinfo
*sra
)
649 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
650 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
651 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
652 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
653 sysfs_set_num(sra
, NULL
, "sync_min", 0);
654 sysfs_set_str(sra
, NULL
, "sync_max", "max");
657 int remove_disks_for_takeover(struct supertype
*st
,
662 struct mdinfo
*remaining
;
665 if (sra
->array
.level
== 10)
666 nr_of_copies
= layout
& 0xff;
667 else if (sra
->array
.level
== 1)
668 nr_of_copies
= sra
->array
.raid_disks
;
672 remaining
= sra
->devs
;
674 /* for each 'copy', select one device and remove from the list. */
675 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
676 struct mdinfo
**diskp
;
679 /* Find a working device to keep */
680 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
681 struct mdinfo
*disk
= *diskp
;
683 if (disk
->disk
.raid_disk
< slot
)
685 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
687 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
689 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
691 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
694 /* We have found a good disk to use! */
696 disk
->next
= sra
->devs
;
705 if (slot
< sra
->array
.raid_disks
) {
706 /* didn't find all slots */
712 sra
->devs
= remaining
;
716 /* Remove all 'remaining' devices from the array */
718 struct mdinfo
*sd
= remaining
;
719 remaining
= sd
->next
;
721 sysfs_set_str(sra
, sd
, "state", "faulty");
722 sysfs_set_str(sra
, sd
, "slot", "none");
723 /* for external metadata disks should be removed in mdmon */
724 if (!st
->ss
->external
)
725 sysfs_set_str(sra
, sd
, "state", "remove");
726 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
727 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
728 sd
->next
= sra
->devs
;
734 void reshape_free_fdlist(int *fdlist
,
735 unsigned long long *offsets
,
740 for (i
= 0; i
< size
; i
++)
748 int reshape_prepare_fdlist(char *devname
,
752 unsigned long blocks
,
755 unsigned long long *offsets
)
760 for (d
= 0; d
<= nrdisks
; d
++)
763 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
764 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
766 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
767 char *dn
= map_dev(sd
->disk
.major
,
769 fdlist
[sd
->disk
.raid_disk
]
770 = dev_open(dn
, O_RDONLY
);
771 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
772 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
774 Name
": %s: cannot open component %s\n",
775 devname
, dn
? dn
: "-unknown-");
779 } else if (backup_file
== NULL
) {
781 char *dn
= map_dev(sd
->disk
.major
,
783 fdlist
[d
] = dev_open(dn
, O_RDWR
);
784 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
786 fprintf(stderr
, Name
": %s: cannot open component %s\n",
787 devname
, dn
? dn
: "-unknown-");
798 int reshape_open_backup_file(char *backup_file
,
803 unsigned long long *offsets
,
806 /* Return 1 on success, 0 on any form of failure */
807 /* need to check backup file is large enough */
813 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
817 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
818 devname
, backup_file
, strerror(errno
));
821 /* Guard against backup file being on array device.
822 * If array is partitioned or if LVM etc is in the
823 * way this will not notice, but it is better than
826 fstat(*fdlist
, &stb
);
829 if (stb
.st_rdev
== dev
) {
830 fprintf(stderr
, Name
": backup file must NOT be"
831 " on the array being reshaped.\n");
837 for (i
=0; i
< blocks
+ 8 ; i
++) {
838 if (write(*fdlist
, buf
, 512) != 512) {
839 fprintf(stderr
, Name
": %s: cannot create"
840 " backup file %s: %s\n",
841 devname
, backup_file
, strerror(errno
));
845 if (fsync(*fdlist
) != 0) {
846 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
847 devname
, backup_file
, strerror(errno
));
854 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
855 unsigned int ndata
, unsigned int odata
)
857 unsigned long a
, b
, blocks
;
858 /* So how much do we need to backup.
859 * We need an amount of data which is both a whole number of
860 * old stripes and a whole number of new stripes.
861 * So LCM for (chunksize*datadisks).
863 a
= (ochunk
/512) * odata
;
864 b
= (nchunk
/512) * ndata
;
872 /* LCM == product / GCD */
873 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
878 char *analyse_change(struct mdinfo
*info
, struct reshape
*re
)
880 /* Based on the current array state in info->array and
881 * the changes in info->new_* etc, determine:
882 * - whether the change is possible
883 * - Intermediate level/raid_disks/layout
884 * - whether a restriping reshape is needed
885 * - number of sectors in minimum change unit. This
886 * will cover a whole number of stripes in 'before' and
889 * Return message if the change should be rejected
890 * NULL if the change can be achieved
892 * This can be called as part of starting a reshape, or
893 * when assembling an array that is undergoing reshape.
896 /* delta_parity records change in number of devices
897 * caused by level change
899 int delta_parity
= 0;
901 /* If a new level not explicitly given, we assume no-change */
902 if (info
->new_level
== UnSet
)
903 info
->new_level
= info
->array
.level
;
906 switch (info
->new_level
) {
912 /* chunk size is meaningful, must divide component_size
915 if (info
->component_size
% (info
->new_chunk
/512))
916 return "New chunk size does not"
917 " divide component size";
920 return "chunk size not meaningful for this level";
923 info
->new_chunk
= info
->array
.chunk_size
;
925 switch (info
->array
.level
) {
927 /* RAID1 can convert to RAID1 with different disks, or
928 * raid5 with 2 disks, or
931 if (info
->new_level
== 0) {
932 if (info
->delta_disks
!= UnSet
&&
933 info
->delta_disks
!= 0)
934 return "Cannot change number of disks "
935 "with RAID1->RAID0 conversion";
937 re
->before
.data_disks
= 1;
938 re
->after
.data_disks
= 1;
939 re
->before
.layout
= 0;
940 re
->backup_blocks
= 0;
944 if (info
->new_level
== 1) {
945 if (info
->delta_disks
== UnSet
)
946 /* Don't know what to do */
947 return "no change requested for Growing RAID1";
949 re
->backup_blocks
= 0;
953 if (info
->array
.raid_disks
== 2 &&
954 info
->new_level
== 5) {
957 re
->before
.data_disks
= 1;
958 if (info
->delta_disks
!= UnSet
&&
959 info
->delta_disks
!= 0)
960 re
->after
.data_disks
= 1 + info
->delta_disks
;
962 re
->after
.data_disks
= 1;
963 if (re
->after
.data_disks
< 1)
964 return "Number of disks too small for RAID5";
966 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
967 info
->array
.chunk_size
= 65536;
970 /* Could do some multi-stage conversions, but leave that to
973 return "Impossibly level change request for RAID1";
976 /* RAID10 can only be converted from near mode to
977 * RAID0 by removing some devices
979 if ((info
->array
.layout
& ~0xff) != 0x100)
980 return "Cannot Grow RAID10 with far/offset layout";
981 /* number of devices must be multiple of number of copies */
982 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
983 return "RAID10 layout too complex for Grow operation";
985 if (info
->new_level
!= 0)
986 return "RAID10 can only be changed to RAID0";
987 new_disks
= (info
->array
.raid_disks
988 / (info
->array
.layout
& 0xff));
989 if (info
->delta_disks
== UnSet
)
990 info
->delta_disks
= (new_disks
991 - info
->array
.raid_disks
);
993 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
994 return "New number of raid-devices impossible for RAID10";
995 if (info
->new_chunk
&&
996 info
->new_chunk
!= info
->array
.chunk_size
)
997 return "Cannot change chunk-size with RAID10 Grow";
1002 re
->before
.data_disks
= new_disks
;
1003 re
->after
.data_disks
= re
->before
.data_disks
;
1004 re
->before
.layout
= 0;
1005 re
->backup_blocks
= 0;
1009 /* RAID0 can be converted to RAID10, or to RAID456 */
1010 if (info
->new_level
== 10) {
1011 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1012 /* Assume near=2 layout */
1013 info
->new_layout
= 0x102;
1014 info
->delta_disks
= info
->array
.raid_disks
;
1016 if (info
->new_layout
== UnSet
) {
1017 int copies
= 1 + (info
->delta_disks
1018 / info
->array
.raid_disks
);
1019 if (info
->array
.raid_disks
* (copies
-1)
1020 != info
->delta_disks
)
1021 return "Impossible number of devices"
1022 " for RAID0->RAID10";
1023 info
->new_layout
= 0x100 + copies
;
1025 if (info
->delta_disks
== UnSet
) {
1026 int copies
= info
->new_layout
& 0xff;
1027 if (info
->new_layout
!= 0x100 + copies
)
1028 return "New layout impossible"
1029 " for RAID0->RAID10";;
1030 info
->delta_disks
= (copies
- 1) *
1031 info
->array
.raid_disks
;
1033 if (info
->new_chunk
&&
1034 info
->new_chunk
!= info
->array
.chunk_size
)
1035 return "Cannot change chunk-size with RAID0->RAID10";
1039 re
->before
.data_disks
= (info
->array
.raid_disks
+
1041 re
->after
.data_disks
= re
->before
.data_disks
;
1042 re
->before
.layout
= info
->new_layout
;
1043 re
->backup_blocks
= 0;
1047 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1048 * a raid4 style layout of the final level.
1050 switch (info
->new_level
) {
1055 re
->before
.layout
= 0;
1060 re
->before
.layout
= ALGORITHM_PARITY_N
;
1065 re
->before
.layout
= ALGORITHM_PARITY_N
;
1068 return "Impossible level change requested";
1070 re
->before
.data_disks
= info
->array
.raid_disks
;
1071 /* determining 'after' layout happens outside this 'switch' */
1075 info
->array
.layout
= ALGORITHM_PARITY_N
;
1077 switch (info
->new_level
) {
1081 re
->level
= info
->array
.level
;
1082 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1083 re
->before
.layout
= info
->array
.layout
;
1087 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1088 re
->before
.layout
= info
->array
.layout
;
1093 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1094 switch (info
->array
.layout
) {
1095 case ALGORITHM_LEFT_ASYMMETRIC
:
1096 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1098 case ALGORITHM_RIGHT_ASYMMETRIC
:
1099 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1101 case ALGORITHM_LEFT_SYMMETRIC
:
1102 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1104 case ALGORITHM_RIGHT_SYMMETRIC
:
1105 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1107 case ALGORITHM_PARITY_0
:
1108 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1110 case ALGORITHM_PARITY_N
:
1111 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1114 return "Cannot convert an array with this layout";
1118 if (info
->array
.raid_disks
!= 2)
1119 return "Can only convert a 2-device array to RAID1";
1120 if (info
->delta_disks
!= UnSet
&&
1121 info
->delta_disks
!= 0)
1122 return "Cannot set raid_disk when "
1123 "converting RAID5->RAID1";
1127 return "Impossible level change requested";
1131 switch (info
->new_level
) {
1137 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1138 re
->before
.layout
= info
->array
.layout
;
1141 return "Impossible level change requested";
1146 /* If we reached here then it looks like a re-stripe is
1147 * happening. We have determined the intermediate level
1148 * and initial raid_disks/layout and stored these in 're'.
1150 * We need to deduce the final layout that can be atomically
1151 * converted to the end state.
1153 switch (info
->new_level
) {
1155 /* We can only get to RAID0 from RAID4 or RAID5
1156 * with appropriate layout and one extra device
1158 if (re
->level
!= 4 && re
->level
!= 5)
1159 return "Cannot covert to RAID0 from this level";
1161 switch (re
->level
) {
1163 re
->after
.layout
= 0 ; break;
1165 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1170 /* We can only get to RAID4 from RAID5 */
1171 if (re
->level
!= 4 && re
->level
!= 5)
1172 return "Cannot convert to RAID4 from this level";
1174 switch (re
->level
) {
1176 re
->after
.layout
= 0 ; break;
1178 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1183 /* We get to RAID5 for RAID5 or RAID6 */
1184 if (re
->level
!= 5 && re
->level
!= 6)
1185 return "Cannot convert to RAID5 from this level";
1187 switch (re
->level
) {
1189 if (info
->new_layout
== UnSet
)
1190 re
->after
.layout
= re
->before
.layout
;
1192 re
->after
.layout
= info
->new_layout
;
1195 if (info
->new_layout
== UnSet
)
1196 info
->new_layout
= re
->before
.layout
;
1198 /* after.layout needs to be raid6 version of new_layout */
1199 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1200 re
->after
.layout
= ALGORITHM_PARITY_N
;
1203 char *ls
= map_num(r5layout
, info
->new_layout
);
1205 strcat(strcpy(layout
, ls
), "-6");
1206 l
= map_name(r6layout
, layout
);
1208 return "Cannot find RAID6 layout"
1210 re
->after
.layout
= l
;
1216 /* We must already be at level 6 */
1218 return "Impossible level change";
1219 if (info
->new_layout
== UnSet
)
1220 re
->after
.layout
= info
->array
.layout
;
1222 re
->after
.layout
= info
->new_layout
;
1225 return "Impossible level change requested";
1227 if (info
->delta_disks
== UnSet
)
1228 info
->delta_disks
= delta_parity
;
1230 re
->after
.data_disks
= (re
->before
.data_disks
1233 switch (re
->level
) {
1234 case 6: re
->parity
= 2; break;
1236 case 5: re
->parity
= 1; break;
1237 default: re
->parity
= 0; break;
1239 /* So we have a restripe operation, we need to calculate the number
1240 * of blocks per reshape operation.
1242 if (info
->new_chunk
== 0)
1243 info
->new_chunk
= info
->array
.chunk_size
;
1244 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1245 re
->after
.layout
== re
->before
.layout
&&
1246 info
->new_chunk
== info
->array
.chunk_size
) {
1247 /* Nothing to change */
1248 re
->backup_blocks
= 0;
1251 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1252 /* chunk and layout changes make no difference */
1253 re
->backup_blocks
= 0;
1257 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1258 get_linux_version() < 2006032)
1259 return "in-place reshape is not safe before 2.6.32 - sorry.";
1261 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1262 get_linux_version() < 2006030)
1263 return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
1265 re
->backup_blocks
= compute_backup_blocks(
1266 info
->new_chunk
, info
->array
.chunk_size
,
1267 re
->after
.data_disks
,
1268 re
->before
.data_disks
);
1270 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1274 static int reshape_array(char *container
, int fd
, char *devname
,
1275 struct supertype
*st
, struct mdinfo
*info
,
1276 int force
, struct mddev_dev
*devlist
,
1277 char *backup_file
, int quiet
, int forked
,
1279 static int reshape_container(char *container
, char *devname
,
1280 struct supertype
*st
,
1281 struct mdinfo
*info
,
1284 int quiet
, int restart
);
1286 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1288 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1289 struct mddev_dev
*devlist
,
1292 /* Make some changes in the shape of an array.
1293 * The kernel must support the change.
1295 * There are three different changes. Each can trigger
1296 * a resync or recovery so we freeze that until we have
1297 * requested everything (if kernel supports freezing - 2.6.30).
1299 * - change size (i.e. component_size)
1301 * - change layout/chunksize/ndisks
1303 * The last can require a reshape. It is different on different
1304 * levels so we need to check the level before actioning it.
1305 * Some times the level change needs to be requested after the
1306 * reshape (e.g. raid6->raid5, raid5->raid0)
1309 struct mdu_array_info_s array
;
1311 struct supertype
*st
;
1312 char *subarray
= NULL
;
1316 char *container
= NULL
;
1317 char container_buf
[20];
1320 struct mddev_dev
*dv
;
1326 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1327 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1333 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1334 fprintf(stderr
, Name
": cannot change component size at the same time "
1335 "as other changes.\n"
1336 " Change size first, then check data is intact before "
1337 "making other changes.\n");
1341 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1342 get_linux_version() < 2006032 &&
1343 !check_env("MDADM_FORCE_FEWER")) {
1344 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1345 " Please use a newer kernel\n");
1349 st
= super_by_fd(fd
, &subarray
);
1351 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1354 if (raid_disks
> st
->max_devs
) {
1355 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1356 " beyond %d\n", st
->max_devs
);
1360 /* in the external case we need to check that the requested reshape is
1361 * supported, and perform an initial check that the container holds the
1362 * pre-requisite spare devices (mdmon owns final validation)
1364 if (st
->ss
->external
) {
1369 container_dev
= st
->container_dev
;
1370 cfd
= open_dev_excl(st
->container_dev
);
1372 container_dev
= st
->devnum
;
1374 cfd
= open_dev_excl(st
->devnum
);
1378 fprintf(stderr
, Name
": Unable to open container for %s\n",
1384 fmt_devname(container_buf
, container_dev
);
1385 container
= container_buf
;
1387 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1390 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1396 if (mdmon_running(container_dev
))
1397 st
->update_tail
= &st
->updates
;
1401 for (dv
= devlist
; dv
; dv
= dv
->next
)
1403 if (raid_disks
> array
.raid_disks
&&
1404 array
.spare_disks
+added_disks
< (raid_disks
- array
.raid_disks
) &&
1407 Name
": Need %d spare%s to avoid degraded array,"
1408 " and only have %d.\n"
1409 " Use --force to over-ride this check.\n",
1410 raid_disks
- array
.raid_disks
,
1411 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1412 array
.spare_disks
+ added_disks
);
1416 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
1417 | GET_STATE
| GET_VERSION
);
1419 if (st
->ss
->external
&& subarray
== NULL
) {
1420 array
.level
= LEVEL_CONTAINER
;
1421 sra
->array
.level
= LEVEL_CONTAINER
;
1424 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1428 frozen
= freeze(st
);
1430 /* freeze() already spewed the reason */
1432 } else if (frozen
< 0) {
1433 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1434 " be reshaped\n", devname
);
1438 /* ========= set size =============== */
1439 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1440 long long orig_size
= array
.size
;
1443 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1450 /* Update the size of each member device in case
1451 * they have been resized. This will never reduce
1452 * below the current used-size. The "size" attribute
1453 * understand '0' to mean 'max'.
1455 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
)
1456 sysfs_set_num(sra
, mdi
, "size", size
);
1459 if (array
.size
!= size
) {
1460 /* got truncated to 32bit, write to
1461 * component_size instead
1464 rv
= sysfs_set_num(sra
, NULL
,
1465 "component_size", size
);
1469 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1473 /* restore metadata */
1474 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1475 UnSet
, NULL
, devname
, !quiet
) == 0)
1477 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1478 devname
, strerror(err
));
1480 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1481 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1485 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1486 size
= get_component_size(fd
)/2;
1490 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
1493 } else if (array
.level
!= LEVEL_CONTAINER
) {
1494 size
= get_component_size(fd
)/2;
1499 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1500 * current implementation assumes that following conditions must be met:
1503 * - near_copies == 2
1505 if ((level
== 0 && array
.level
== 10 && sra
&&
1506 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1507 (level
== 0 && array
.level
== 1 && sra
)) {
1509 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1511 dprintf(Name
": Array cannot be reshaped\n");
1517 /* FIXME this is added with no justification - why is it here */
1518 ping_monitor(container
);
1521 memset(&info
, 0, sizeof(info
));
1523 sysfs_init(&info
, fd
, NoMdDev
);
1524 strcpy(info
.text_version
, sra
->text_version
);
1525 info
.component_size
= size
*2;
1526 info
.new_level
= level
;
1527 info
.new_chunk
= chunksize
* 1024;
1528 if (info
.array
.level
== LEVEL_CONTAINER
) {
1529 info
.delta_disks
= UnSet
;
1530 info
.array
.raid_disks
= raid_disks
;
1531 } else if (raid_disks
)
1532 info
.delta_disks
= raid_disks
- info
.array
.raid_disks
;
1534 info
.delta_disks
= UnSet
;
1535 if (layout_str
== NULL
) {
1536 info
.new_layout
= UnSet
;
1537 if (info
.array
.level
== 6 &&
1538 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1539 info
.array
.layout
>= 16) {
1540 fprintf(stderr
, Name
1541 ": %s has a non-standard layout. If you"
1542 " wish to preserve this\n"
1543 " during the reshape, please specify"
1544 " --layout=preserve\n"
1545 " If you want to change it, specify a"
1546 " layout or use --layout=normalise\n",
1551 } else if (strcmp(layout_str
, "normalise") == 0 ||
1552 strcmp(layout_str
, "normalize") == 0) {
1553 /* If we have a -6 RAID6 layout, remove the '-6'. */
1554 info
.new_layout
= UnSet
;
1555 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1557 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1558 h
= strrchr(l
, '-');
1559 if (h
&& strcmp(h
, "-6") == 0) {
1561 info
.new_layout
= map_name(r6layout
, l
);
1564 } else if (strcmp(layout_str
, "preserve") == 0) {
1565 info
.new_layout
= UnSet
;
1567 int l
= info
.new_level
;
1569 l
= info
.array
.level
;
1572 info
.new_layout
= map_name(r5layout
, layout_str
);
1575 info
.new_layout
= map_name(r6layout
, layout_str
);
1578 info
.new_layout
= parse_layout_10(layout_str
);
1581 info
.new_layout
= parse_layout_faulty(layout_str
);
1584 fprintf(stderr
, Name
": layout not meaningful"
1585 " with this level\n");
1589 if (info
.new_layout
== UnSet
) {
1590 fprintf(stderr
, Name
": layout %s not understood"
1591 " for this level\n",
1598 if (array
.level
== LEVEL_CONTAINER
) {
1599 /* This change is to be applied to every array in the
1600 * container. This is only needed when the metadata imposes
1601 * restraints of the various arrays in the container.
1602 * Currently we only know that IMSM requires all arrays
1603 * to have the same number of devices so changing the
1604 * number of devices (On-Line Capacity Expansion) must be
1605 * performed at the level of the container
1607 rv
= reshape_container(container
, devname
, st
, &info
,
1608 force
, backup_file
, quiet
, 0);
1611 /* get spare devices from external metadata
1613 if (st
->ss
->external
) {
1614 struct mdinfo
*info2
;
1616 info2
= st
->ss
->container_content(st
, subarray
);
1618 info
.array
.spare_disks
=
1619 info2
->array
.spare_disks
;
1624 /* Impose these changes on a single array. First
1625 * check that the metadata is OK with the change. */
1627 if (reshape_super(st
, info
.component_size
, info
.new_level
,
1628 info
.new_layout
, info
.new_chunk
,
1629 info
.array
.raid_disks
, info
.delta_disks
,
1630 backup_file
, devname
, quiet
)) {
1635 rv
= reshape_array(container
, fd
, devname
, st
, &info
, force
,
1636 devlist
, backup_file
, quiet
, 0, 0);
1645 static int reshape_array(char *container
, int fd
, char *devname
,
1646 struct supertype
*st
, struct mdinfo
*info
,
1647 int force
, struct mddev_dev
*devlist
,
1648 char *backup_file
, int quiet
, int forked
,
1651 struct reshape reshape
;
1654 int orig_level
= UnSet
;
1657 struct mdu_array_info_s array
;
1660 struct mddev_dev
*dv
;
1664 unsigned long long *offsets
;
1668 unsigned long blocks
;
1669 unsigned long cache
;
1670 unsigned long long array_size
;
1672 struct mdinfo
*sra
= NULL
;
1674 /* when reshaping a RAID0, the component_size might be zero.
1675 * So try to fix that up.
1677 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1678 dprintf("Cannot get array information.\n");
1681 if (array
.level
== 0 && info
->component_size
== 0) {
1682 get_dev_size(fd
, NULL
, &array_size
);
1683 info
->component_size
= array_size
/ array
.raid_disks
;
1686 if (info
->reshape_active
) {
1687 int new_level
= info
->new_level
;
1688 info
->new_level
= UnSet
;
1689 info
->array
.raid_disks
-= info
->delta_disks
;
1690 msg
= analyse_change(info
, &reshape
);
1691 info
->new_level
= new_level
;
1692 info
->array
.raid_disks
+= info
->delta_disks
;
1694 /* Make sure the array isn't read-only */
1695 ioctl(fd
, RESTART_ARRAY_RW
, 0);
1697 msg
= analyse_change(info
, &reshape
);
1699 fprintf(stderr
, Name
": %s\n", msg
);
1703 (reshape
.level
!= info
->array
.level
||
1704 reshape
.before
.layout
!= info
->array
.layout
||
1705 reshape
.before
.data_disks
+ reshape
.parity
1706 != info
->array
.raid_disks
- info
->delta_disks
)) {
1707 fprintf(stderr
, Name
": reshape info is not in native format -"
1708 " cannot continue.\n");
1713 /* reshape already started. just skip to monitoring the reshape */
1714 if (reshape
.backup_blocks
== 0)
1718 /* The container is frozen but the array may not be.
1719 * So freeze the array so spares don't get put to the wrong use
1720 * FIXME there should probably be a cleaner separation between
1721 * freeze_array and freeze_container.
1723 sysfs_freeze_array(info
);
1724 /* Check we have enough spares to not be degraded */
1726 for (dv
= devlist
; dv
; dv
=dv
->next
)
1728 spares_needed
= max(reshape
.before
.data_disks
,
1729 reshape
.after
.data_disks
)
1730 + reshape
.parity
- array
.raid_disks
;
1733 info
->new_level
> 1 && info
->array
.level
> 1 &&
1734 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
1736 Name
": Need %d spare%s to avoid degraded array,"
1737 " and only have %d.\n"
1738 " Use --force to over-ride this check.\n",
1740 spares_needed
== 1 ? "" : "s",
1741 info
->array
.spare_disks
+ added_disks
);
1744 /* Check we have enough spares to not fail */
1745 spares_needed
= max(reshape
.before
.data_disks
,
1746 reshape
.after
.data_disks
)
1748 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
1749 spares_needed
> info
->array
.spare_disks
+added_disks
) {
1751 Name
": Need %d spare%s to create working array,"
1752 " and only have %d.\n",
1754 spares_needed
== 1 ? "" : "s",
1755 info
->array
.spare_disks
+ added_disks
);
1759 if (reshape
.level
!= array
.level
) {
1760 char *c
= map_num(pers
, reshape
.level
);
1765 err
= sysfs_set_str(info
, NULL
, "level", c
);
1768 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1771 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1772 fprintf(stderr
, " Bitmap must be removed"
1773 " before level can be changed\n");
1777 fprintf(stderr
, Name
": level of %s changed to %s\n",
1779 orig_level
= array
.level
;
1780 sysfs_freeze_array(info
);
1782 if (reshape
.level
> 0 && st
->ss
->external
) {
1783 /* make sure mdmon is aware of the new level */
1784 if (!mdmon_running(st
->container_dev
))
1785 start_mdmon(st
->container_dev
);
1786 ping_monitor(container
);
1789 /* ->reshape_super might have chosen some spares from the
1790 * container that it wants to be part of the new array.
1791 * We can collect them with ->container_content and give
1792 * them to the kernel.
1794 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
1795 char *subarray
= strchr(info
->text_version
+1, '/')+1;
1796 struct mdinfo
*info2
=
1797 st
->ss
->container_content(st
, subarray
);
1801 sysfs_init(info2
, fd
, st
->devnum
);
1802 /* When increasing number of devices, we need to set
1803 * new raid_disks before adding these, or they might
1806 if (reshape
.backup_blocks
&&
1807 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
1808 subarray_set_num(container
, info2
, "raid_disks",
1809 reshape
.after
.data_disks
+
1811 for (d
= info2
->devs
; d
; d
= d
->next
) {
1812 if (d
->disk
.state
== 0 &&
1813 d
->disk
.raid_disk
>= 0) {
1814 /* This is a spare that wants to
1815 * be part of the array.
1817 add_disk(fd
, st
, info2
, d
);
1823 /* We might have been given some devices to add to the
1824 * array. Now that the array has been changed to the right
1825 * level and frozen, we can safely add them.
1828 Manage_subdevs(devname
, fd
, devlist
, !quiet
,
1831 if (reshape
.backup_blocks
== 0) {
1832 /* No restriping needed, but we might need to impose
1833 * some more changes: layout, raid_disks, chunk_size
1835 /* read current array info */
1836 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1837 dprintf("Cannot get array information.\n");
1840 /* compare current array info with new values and if
1841 * it is different update them to new */
1842 if (info
->new_layout
!= UnSet
&&
1843 info
->new_layout
!= array
.layout
) {
1844 array
.layout
= info
->new_layout
;
1845 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1846 fprintf(stderr
, Name
": failed to set new layout\n");
1849 printf("layout for %s set to %d\n",
1850 devname
, array
.layout
);
1852 if (info
->delta_disks
!= UnSet
&&
1853 info
->delta_disks
!= 0 &&
1854 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
1855 array
.raid_disks
+= info
->delta_disks
;
1856 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1857 fprintf(stderr
, Name
": failed to set raid disks\n");
1859 } else if (!quiet
) {
1860 printf("raid_disks for %s set to %d\n",
1861 devname
, array
.raid_disks
);
1864 if (info
->new_chunk
!= 0 &&
1865 info
->new_chunk
!= array
.chunk_size
) {
1866 if (sysfs_set_num(info
, NULL
,
1867 "chunk_size", info
->new_chunk
) != 0) {
1868 fprintf(stderr
, Name
": failed to set chunk size\n");
1871 printf("chunk size for %s set to %d\n",
1872 devname
, array
.chunk_size
);
1879 * There are three possibilities.
1880 * 1/ The array will shrink.
1881 * We need to ensure the reshape will pause before reaching
1882 * the 'critical section'. We also need to fork and wait for
1883 * that to happen. When it does we
1884 * suspend/backup/complete/unfreeze
1886 * 2/ The array will not change size.
1887 * This requires that we keep a backup of a sliding window
1888 * so that we can restore data after a crash. So we need
1889 * to fork and monitor progress.
1890 * In future we will allow the data_offset to change, so
1891 * a sliding backup becomes unnecessary.
1893 * 3/ The array will grow. This is relatively easy.
1894 * However the kernel's restripe routines will cheerfully
1895 * overwrite some early data before it is safe. So we
1896 * need to make a backup of the early parts of the array
1897 * and be ready to restore it if rebuild aborts very early.
1898 * For externally managed metadata, we still need a forked
1899 * child to monitor the reshape and suspend IO over the region
1900 * that is being reshaped.
1902 * We backup data by writing it to one spare, or to a
1903 * file which was given on command line.
1905 * In each case, we first make sure that storage is available
1906 * for the required backup.
1908 * - request the shape change.
1909 * - fork to handle backup etc.
1912 /* Check that we can hold all the data */
1913 get_dev_size(fd
, NULL
, &array_size
);
1914 if (reshape
.new_size
< (array_size
/512)) {
1916 Name
": this change will reduce the size of the array.\n"
1917 " use --grow --array-size first to truncate array.\n"
1918 " e.g. mdadm --grow %s --array-size %llu\n",
1919 devname
, reshape
.new_size
/2);
1923 sra
= sysfs_read(fd
, 0,
1924 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
1927 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
1932 /* Decide how many blocks (sectors) for a reshape
1933 * unit. The number we have so far is just a minimum
1935 blocks
= reshape
.backup_blocks
;
1936 if (reshape
.before
.data_disks
==
1937 reshape
.after
.data_disks
) {
1938 /* Make 'blocks' bigger for better throughput, but
1939 * not so big that we reject it below.
1940 * Try for 16 megabytes
1942 while (blocks
* 32 < sra
->component_size
&&
1946 fprintf(stderr
, Name
": Need to backup %luK of critical "
1947 "section..\n", blocks
/2);
1949 if (blocks
>= sra
->component_size
/2) {
1950 fprintf(stderr
, Name
": %s: Something wrong"
1951 " - reshape aborted\n",
1956 /* Now we need to open all these devices so we can read/write.
1958 nrdisks
= max(reshape
.before
.data_disks
,
1959 reshape
.after
.data_disks
) + reshape
.parity
1960 + sra
->array
.spare_disks
;
1961 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1962 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1963 if (!fdlist
|| !offsets
) {
1964 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1968 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
1969 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
1970 nrdisks
, blocks
, backup_file
,
1975 if (backup_file
== NULL
) {
1976 if (reshape
.after
.data_disks
<= reshape
.before
.data_disks
) {
1978 Name
": %s: Cannot grow - need backup-file\n",
1981 } else if (sra
->array
.spare_disks
== 0) {
1982 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1983 "backup-file to backup critical section\n",
1988 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
1990 fdlist
+d
, offsets
+d
, restart
)) {
1996 /* lastly, check that the internal stripe cache is
1997 * large enough, or it won't work.
1998 * It must hold at least 4 stripes of the larger
2001 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2002 cache
*= 4; /* 4 stripes minimum */
2003 cache
/= 512; /* convert to sectors */
2004 disks
= min(reshape
.before
.data_disks
, reshape
.after
.data_disks
);
2005 /* make sure there is room for 'blocks' with a bit to spare */
2006 if (cache
< 16 + blocks
/ disks
)
2007 cache
= 16 + blocks
/ disks
;
2008 cache
/= (4096/512); /* Covert from sectors to pages */
2010 if (sra
->cache_size
< cache
)
2011 subarray_set_num(container
, sra
, "stripe_cache_size",
2014 /* Right, everything seems fine. Let's kick things off.
2015 * If only changing raid_disks, use ioctl, else use
2020 sra
->new_chunk
= info
->new_chunk
;
2023 sra
->reshape_progress
= info
->reshape_progress
;
2025 sra
->reshape_progress
= 0;
2026 if (reshape
.after
.data_disks
< reshape
.before
.data_disks
)
2027 /* start from the end of the new array */
2028 sra
->reshape_progress
= (sra
->component_size
2029 * reshape
.after
.data_disks
);
2032 if (info
->array
.chunk_size
== info
->new_chunk
&&
2033 reshape
.before
.layout
== reshape
.after
.layout
&&
2034 st
->ss
->external
== 0) {
2035 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2036 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2037 array
.raid_disks
= reshape
.after
.data_disks
+ reshape
.parity
;
2039 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2043 Name
": Cannot set device shape for %s: %s\n",
2044 devname
, strerror(errno
));
2047 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2049 " Bitmap must be removed before"
2050 " shape can be changed\n");
2054 } else if (!restart
) {
2055 /* set them all just in case some old 'new_*' value
2056 * persists from some earlier problem.
2059 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2061 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2062 reshape
.after
.layout
) < 0)
2064 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2065 reshape
.after
.data_disks
+
2066 reshape
.parity
) < 0)
2069 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
2073 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2075 " Bitmap must be removed before"
2076 " shape can be changed\n");
2081 err
= start_reshape(sra
, restart
);
2084 Name
": Cannot %s reshape for %s\n",
2085 restart
? "continue" : "start",
2090 sysfs_set_str(sra
, NULL
, "array_state", "active");
2092 /* Now we just need to kick off the reshape and watch, while
2093 * handling backups of the data...
2094 * This is all done by a forked background process.
2096 switch(forked
? 0 : fork()) {
2098 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
2109 if (check_env("MDADM_GROW_VERIFY"))
2110 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
2113 mlockall(MCL_FUTURE
);
2115 if (st
->ss
->external
) {
2116 /* metadata handler takes it from here */
2117 done
= st
->ss
->manage_reshape(
2118 fd
, sra
, &reshape
, st
, blocks
,
2120 d
- odisks
, fdlist
+odisks
,
2123 done
= child_monitor(
2124 fd
, sra
, &reshape
, st
, blocks
,
2126 d
- odisks
, fdlist
+odisks
,
2129 if (backup_file
&& done
)
2130 unlink(backup_file
);
2136 if (!st
->ss
->external
&&
2137 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
2138 && info
->custom_array_size
) &&
2139 info
->new_level
== reshape
.level
&&
2141 /* no need to wait for the reshape to finish as
2142 * there is nothing more to do.
2148 if (st
->ss
->external
) {
2149 /* Re-load the metadata as much could have changed */
2150 int cfd
= open_dev(st
->container_dev
);
2152 ping_monitor(container
);
2153 st
->ss
->free_super(st
);
2154 st
->ss
->load_container(st
, cfd
, container
);
2159 /* set new array size if required customer_array_size is used
2162 if (reshape
.before
.data_disks
!=
2163 reshape
.after
.data_disks
&&
2164 info
->custom_array_size
) {
2165 struct mdinfo
*info2
;
2166 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2168 info2
= st
->ss
->container_content(st
, subarray
);
2170 unsigned long long current_size
= 0;
2171 unsigned long long new_size
=
2172 info2
->custom_array_size
/2;
2174 if (sysfs_get_ll(sra
,
2177 ¤t_size
) == 0 &&
2178 new_size
> current_size
) {
2179 if (sysfs_set_num(sra
, NULL
,
2180 "array_size", new_size
)
2182 dprintf("Error: Cannot"
2185 dprintf("Array size "
2187 dprintf(" from %llu to %llu.\n",
2188 current_size
, new_size
);
2194 if (info
->new_level
!= reshape
.level
) {
2196 c
= map_num(pers
, info
->new_level
);
2198 err
= sysfs_set_str(sra
, NULL
, "level", c
);
2200 fprintf(stderr
, Name\
2201 ": %s: could not set level "
2202 "to %s\n", devname
, c
);
2212 if (orig_level
!= UnSet
&& sra
) {
2213 c
= map_num(pers
, orig_level
);
2214 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
2215 fprintf(stderr
, Name
": aborting level change\n");
2222 int reshape_container(char *container
, char *devname
,
2223 struct supertype
*st
,
2224 struct mdinfo
*info
,
2227 int quiet
, int restart
)
2229 struct mdinfo
*cc
= NULL
;
2232 /* component_size is not meaningful for a container,
2233 * so pass '-1' meaning 'no change'
2236 reshape_super(st
, -1, info
->new_level
,
2237 info
->new_layout
, info
->new_chunk
,
2238 info
->array
.raid_disks
, info
->delta_disks
,
2239 backup_file
, devname
, quiet
)) {
2246 /* ping monitor to be sure that update is on disk
2248 ping_monitor(container
);
2251 case -1: /* error */
2252 perror("Cannot fork to complete reshape\n");
2255 default: /* parent */
2256 printf(Name
": multi-array reshape continues in background\n");
2263 /* For each member array with reshape_active,
2264 * we need to perform the reshape.
2265 * We pick the first array that needs reshaping and
2266 * reshape it. reshape_array() will re-read the metadata
2267 * so the next time through a different array should be
2268 * ready for reshape.
2269 * It is possible that the 'different' array will not
2270 * be assembled yet. In that case we simple exit.
2271 * When it is assembled, the mdadm which assembles it
2272 * will take over the reshape.
2274 struct mdinfo
*content
;
2276 struct mdstat_ent
*mdstat
;
2281 cc
= st
->ss
->container_content(st
, NULL
);
2283 for (content
= cc
; content
; content
= content
->next
) {
2285 if (!content
->reshape_active
)
2288 subarray
= strchr(content
->text_version
+1, '/')+1;
2289 mdstat
= mdstat_by_subdev(subarray
,
2290 devname2devnum(container
));
2298 fd
= open_dev(mdstat
->devnum
);
2301 adev
= map_dev(dev2major(mdstat
->devnum
),
2302 dev2minor(mdstat
->devnum
),
2305 adev
= content
->text_version
;
2307 sysfs_init(content
, fd
, mdstat
->devnum
);
2309 rv
= reshape_array(container
, fd
, adev
, st
,
2310 content
, force
, NULL
,
2311 backup_file
, quiet
, 1, restart
);
2324 * We run a child process in the background which performs the following
2326 * - wait for resync to reach a certain point
2327 * - suspend io to the following section
2328 * - backup that section
2329 * - allow resync to proceed further
2331 * - discard the backup.
2333 * When are combined in slightly different ways in the three cases.
2335 * - suspend/backup/allow/wait/resume/discard
2337 * - allow/wait/suspend/backup/allow/wait/resume/discard
2339 * - wait/resume/discard/suspend/backup/allow
2341 * suspend/backup/allow always come together
2342 * wait/resume/discard do too.
2343 * For the same-size case we have two backups to improve flow.
2347 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
2348 unsigned long long backup_point
,
2349 unsigned long long wait_point
,
2350 unsigned long long *suspend_point
,
2351 unsigned long long *reshape_completed
)
2353 /* This function is called repeatedly by the reshape manager.
2354 * It determines how much progress can safely be made and allows
2356 * - 'info' identifies the array and particularly records in
2357 * ->reshape_progress the metadata's knowledge of progress
2358 * This is a sector offset from the start of the array
2359 * of the next array block to be relocated. This number
2360 * may increase from 0 or decrease from array_size, depending
2361 * on the type of reshape that is happening.
2362 * Note that in contrast, 'sync_completed' is a block count of the
2363 * reshape so far. It gives the distance between the start point
2364 * (head or tail of device) and the next place that data will be
2365 * written. It always increases.
2366 * - 'reshape' is the structure created by analyse_change
2367 * - 'backup_point' shows how much the metadata manager has backed-up
2368 * data. For reshapes with increasing progress, it is the next address
2369 * to be backed up, previous addresses have been backed-up. For
2370 * decreasing progress, it is the earliest address that has been
2371 * backed up - later address are also backed up.
2372 * So addresses between reshape_progress and backup_point are
2373 * backed up providing those are in the 'correct' order.
2374 * - 'wait_point' is an array address. When reshape_completed
2375 * passes this point, progress_reshape should return. It might
2376 * return earlier if it determines that ->reshape_progress needs
2377 * to be updated or further backup is needed.
2378 * - suspend_point is maintained by progress_reshape and the caller
2379 * should not touch it except to initialise to zero.
2380 * It is an array address and it only increases in 2.6.37 and earlier.
2381 * This makes it difficult to handle reducing reshapes with
2382 * external metadata.
2383 * However: it is similar to backup_point in that it records the
2384 * other end of a suspended region from reshape_progress.
2385 * it is moved to extend the region that is safe to backup and/or
2387 * - reshape_completed is read from sysfs and returned. The caller
2388 * should copy this into ->reshape_progress when it has reason to
2389 * believe that the metadata knows this, and any backup outside this
2393 * 1 if more data from backup_point - but only as far as suspend_point,
2394 * should be backed up
2395 * 0 if things are progressing smoothly
2396 * -1 if the reshape is finished because it is all done,
2397 * -2 if the reshape is finished due to an error.
2400 int advancing
= (reshape
->after
.data_disks
2401 >= reshape
->before
.data_disks
);
2402 unsigned long long need_backup
; /* All data between start of array and
2403 * here will at some point need to
2406 unsigned long long read_offset
, write_offset
;
2407 unsigned long long write_range
;
2408 unsigned long long max_progress
, target
, completed
;
2409 unsigned long long array_size
= (info
->component_size
2410 * reshape
->before
.data_disks
);
2414 /* First, we unsuspend any region that is now known to be safe.
2415 * If suspend_point is on the 'wrong' side of reshape_progress, then
2416 * we don't have or need suspension at the moment. This is true for
2417 * native metadata when we don't need to back-up.
2420 if (info
->reshape_progress
<= *suspend_point
)
2421 sysfs_set_num(info
, NULL
, "suspend_lo",
2422 info
->reshape_progress
);
2424 /* Note: this won't work in 2.6.37 and before.
2425 * Something somewhere should make sure we don't need it!
2427 if (info
->reshape_progress
>= *suspend_point
)
2428 sysfs_set_num(info
, NULL
, "suspend_hi",
2429 info
->reshape_progress
);
2432 /* Now work out how far it is safe to progress.
2433 * If the read_offset for ->reshape_progress is less than
2434 * 'blocks' beyond the write_offset, we can only progress as far
2436 * Otherwise we can progress until the write_offset for the new location
2437 * reaches (within 'blocks' of) the read_offset at the current location.
2438 * However that region must be suspended unless we are using native
2440 * If we need to suspend more, we limit it to 128M per device, which is
2441 * rather arbitrary and should be some time-based calculation.
2443 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
2444 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
2445 write_range
= info
->new_chunk
/512;
2446 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2447 need_backup
= array_size
;
2449 need_backup
= reshape
->backup_blocks
;
2451 if (read_offset
< write_offset
+ write_range
)
2452 max_progress
= backup_point
;
2456 reshape
->after
.data_disks
;
2458 if (read_offset
> write_offset
- write_range
)
2459 /* Can only progress as far as has been backed up,
2460 * which must be suspended */
2461 max_progress
= backup_point
;
2462 else if (info
->reshape_progress
<= need_backup
)
2463 max_progress
= backup_point
;
2465 if (info
->array
.major_version
>= 0)
2466 /* Can progress until backup is needed */
2467 max_progress
= need_backup
;
2469 /* Can progress until metadata update is required */
2472 reshape
->after
.data_disks
;
2473 /* but data must be suspended */
2474 if (max_progress
< *suspend_point
)
2475 max_progress
= *suspend_point
;
2480 /* We know it is safe to progress to 'max_progress' providing
2481 * it is suspended or we are using native metadata.
2482 * Consider extending suspend_point 128M per device if it
2483 * is less than 64M per device beyond reshape_progress.
2484 * But always do a multiple of 'blocks'
2485 * FIXME this is too big - it takes to long to complete
2488 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
2489 reshape
->after
.data_disks
);
2490 target
/= reshape
->backup_blocks
;
2493 target
*= reshape
->backup_blocks
;
2495 /* For externally managed metadata we always need to suspend IO to
2496 * the area being reshaped so we regularly push suspend_point forward.
2497 * For native metadata we only need the suspend if we are going to do
2501 if ((need_backup
> info
->reshape_progress
2502 || info
->array
.major_version
< 0) &&
2503 *suspend_point
< info
->reshape_progress
+ target
) {
2504 if (need_backup
< *suspend_point
+ 2 * target
)
2505 *suspend_point
= need_backup
;
2506 else if (*suspend_point
+ 2 * target
< array_size
)
2507 *suspend_point
+= 2 * target
;
2509 *suspend_point
= array_size
;
2510 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
2511 if (max_progress
> *suspend_point
)
2512 max_progress
= *suspend_point
;
2515 if (info
->array
.major_version
>= 0) {
2516 /* Only need to suspend when about to backup */
2517 if (info
->reshape_progress
< need_backup
* 2 &&
2518 *suspend_point
> 0) {
2520 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
2521 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
2524 /* Need to suspend continually */
2525 if (info
->reshape_progress
< *suspend_point
)
2526 *suspend_point
= info
->reshape_progress
;
2527 if (*suspend_point
+ target
< info
->reshape_progress
)
2528 /* No need to move suspend region yet */;
2530 if (*suspend_point
>= 2 * target
)
2531 *suspend_point
-= 2 * target
;
2534 sysfs_set_num(info
, NULL
, "suspend_lo",
2537 if (max_progress
< *suspend_point
)
2538 max_progress
= *suspend_point
;
2542 /* now set sync_max to allow that progress. sync_max, like
2543 * sync_completed is a count of sectors written per device, so
2544 * we find the difference between max_progress and the start point,
2545 * and divide that by after.data_disks to get a sync_max
2547 * At the same time we convert wait_point to a similar number
2548 * for comparing against sync_completed.
2550 /* scale down max_progress to per_disk */
2551 max_progress
/= reshape
->after
.data_disks
;
2552 /* Round to chunk size as some kernels give an erroneously high number */
2553 max_progress
/= info
->new_chunk
/512;
2554 max_progress
*= info
->new_chunk
/512;
2555 /* And round to old chunk size as the kernel wants that */
2556 max_progress
/= info
->array
.chunk_size
/512;
2557 max_progress
*= info
->array
.chunk_size
/512;
2558 /* Limit progress to the whole device */
2559 if (max_progress
> info
->component_size
)
2560 max_progress
= info
->component_size
;
2561 wait_point
/= reshape
->after
.data_disks
;
2563 /* switch from 'device offset' to 'processed block count' */
2564 max_progress
= info
->component_size
- max_progress
;
2565 wait_point
= info
->component_size
- wait_point
;
2568 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
2570 /* Now wait. If we have already reached the point that we were
2571 * asked to wait to, don't wait at all, else wait for any change.
2572 * We need to select on 'sync_completed' as that is the place that
2573 * notifications happen, but we are really interested in
2574 * 'reshape_position'
2576 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
2578 goto check_progress
;
2580 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2582 goto check_progress
;
2584 while (completed
< max_progress
&& completed
< wait_point
) {
2585 /* Check that sync_action is still 'reshape' to avoid
2586 * waiting forever on a dead array
2590 if (sysfs_get_str(info
, NULL
, "sync_action",
2592 strncmp(action
, "reshape", 7) != 0)
2594 /* Some kernels reset 'sync_completed' to zero
2595 * before setting 'sync_action' to 'idle'.
2596 * So we need these extra tests.
2598 if (completed
== 0 && advancing
2599 && info
->reshape_progress
> 0)
2601 if (completed
== 0 && !advancing
2602 && info
->reshape_progress
< (info
->component_size
2603 * reshape
->after
.data_disks
))
2607 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2608 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2610 goto check_progress
;
2613 /* Some kernels reset 'sync_completed' to zero,
2614 * we need to have real point we are in md
2617 completed
= max_progress
;
2619 /* some kernels can give an incorrectly high 'completed' number */
2620 completed
/= (info
->new_chunk
/512);
2621 completed
*= (info
->new_chunk
/512);
2622 /* Convert 'completed' back in to a 'progress' number */
2623 completed
*= reshape
->after
.data_disks
;
2625 completed
= info
->component_size
* reshape
->after
.data_disks
2628 *reshape_completed
= completed
;
2632 /* We return the need_backup flag. Caller will decide
2633 * how much - a multiple of ->backup_blocks up to *suspend_point
2636 return need_backup
> info
->reshape_progress
;
2638 return need_backup
>= info
->reshape_progress
;
2641 /* if we couldn't read a number from sync_completed, then
2642 * either the reshape did complete, or it aborted.
2643 * We can tell which by checking for 'none' in reshape_position.
2646 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
2647 || strncmp(buf
, "none", 4) != 0)
2648 return -2; /* abort */
2650 /* Maybe racing with array shutdown - check state */
2651 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
2652 || strncmp(buf
, "inactive", 8) == 0
2653 || strncmp(buf
, "clear",5) == 0)
2654 return -2; /* abort */
2655 return -1; /* complete */
2660 /* FIXME return status is never checked */
2661 static int grow_backup(struct mdinfo
*sra
,
2662 unsigned long long offset
, /* per device */
2663 unsigned long stripes
, /* per device, in old chunks */
2664 int *sources
, unsigned long long *offsets
,
2665 int disks
, int chunk
, int level
, int layout
,
2666 int dests
, int *destfd
, unsigned long long *destoffsets
,
2667 int part
, int *degraded
,
2670 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2671 * to storage 'destfd' (offset 'destoffsets'), after first
2672 * suspending IO. Then allow resync to continue
2673 * over the suspended section.
2674 * Use part 'part' of the backup-super-block.
2679 unsigned long long ll
;
2681 //printf("offset %llu\n", offset);
2687 /* Check that array hasn't become degraded, else we might backup the wrong data */
2688 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
2689 return -1; /* FIXME this error is ignored */
2690 new_degraded
= (int)ll
;
2691 if (new_degraded
!= *degraded
) {
2692 /* check each device to ensure it is still working */
2694 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2695 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2697 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2699 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2700 strstr(sbuf
, "faulty") ||
2701 strstr(sbuf
, "in_sync") == NULL
) {
2702 /* this device is dead */
2703 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2704 if (sd
->disk
.raid_disk
>= 0 &&
2705 sources
[sd
->disk
.raid_disk
] >= 0) {
2706 close(sources
[sd
->disk
.raid_disk
]);
2707 sources
[sd
->disk
.raid_disk
] = -1;
2712 *degraded
= new_degraded
;
2715 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2716 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2718 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2719 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2722 bsb
.magic
[15] = '2';
2723 for (i
= 0; i
< dests
; i
++)
2725 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
2727 lseek64(destfd
[i
], destoffsets
[i
], 0);
2729 rv
= save_stripes(sources
, offsets
,
2730 disks
, chunk
, level
, layout
,
2732 offset
*512*odata
, stripes
* chunk
* odata
,
2737 bsb
.mtime
= __cpu_to_le64(time(0));
2738 for (i
= 0; i
< dests
; i
++) {
2739 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2741 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2742 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2743 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2744 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2747 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
2748 != destoffsets
[i
] - 4096)
2750 if (write(destfd
[i
], &bsb
, 512) != 512)
2752 if (destoffsets
[i
] > 4096) {
2753 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
2754 destoffsets
[i
]+stripes
*chunk
*odata
)
2756 if (write(destfd
[i
], &bsb
, 512) != 512)
2766 /* in 2.6.30, the value reported by sync_completed can be
2767 * less that it should be by one stripe.
2768 * This only happens when reshape hits sync_max and pauses.
2769 * So allow wait_backup to either extent sync_max further
2770 * than strictly necessary, or return before the
2771 * sync has got quite as far as we would really like.
2772 * This is what 'blocks2' is for.
2773 * The various caller give appropriate values so that
2776 /* FIXME return value is often ignored */
2777 static int forget_backup(
2778 int dests
, int *destfd
, unsigned long long *destoffsets
,
2782 * Erase backup 'part' (which is 0 or 1)
2788 bsb
.arraystart2
= __cpu_to_le64(0);
2789 bsb
.length2
= __cpu_to_le64(0);
2791 bsb
.arraystart
= __cpu_to_le64(0);
2792 bsb
.length
= __cpu_to_le64(0);
2794 bsb
.mtime
= __cpu_to_le64(time(0));
2796 for (i
= 0; i
< dests
; i
++) {
2797 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2798 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2799 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2800 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2801 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2802 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
2803 destoffsets
[i
]-4096)
2806 write(destfd
[i
], &bsb
, 512) != 512)
2813 static void fail(char *msg
)
2816 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
2817 rv
|= (write(2, "\n", 1) != 1);
2821 static char *abuf
, *bbuf
;
2822 static unsigned long long abuflen
;
2823 static void validate(int afd
, int bfd
, unsigned long long offset
)
2825 /* check that the data in the backup against the array.
2826 * This is only used for regression testing and should not
2827 * be used while the array is active
2831 lseek64(bfd
, offset
- 4096, 0);
2832 if (read(bfd
, &bsb2
, 512) != 512)
2833 fail("cannot read bsb");
2834 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
2835 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
2836 fail("first csum bad");
2837 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
2838 fail("magic is bad");
2839 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
2840 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
2841 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
2842 fail("second csum bad");
2844 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
2845 fail("devstart is wrong");
2848 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
2850 if (abuflen
< len
) {
2854 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
2855 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
2857 /* just stop validating on mem-alloc failure */
2862 lseek64(bfd
, offset
, 0);
2863 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
2864 //printf("len %llu\n", len);
2865 fail("read first backup failed");
2867 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
2868 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2869 fail("read first from array failed");
2870 if (memcmp(bbuf
, abuf
, len
) != 0) {
2873 printf("offset=%llu len=%llu\n",
2874 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
2875 for (i
=0; i
<len
; i
++)
2876 if (bbuf
[i
] != abuf
[i
]) {
2877 printf("first diff byte %d\n", i
);
2881 fail("data1 compare failed");
2885 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
2887 if (abuflen
< len
) {
2891 abuf
= malloc(abuflen
);
2892 bbuf
= malloc(abuflen
);
2895 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2896 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2897 fail("read second backup failed");
2898 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2899 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2900 fail("read second from array failed");
2901 if (memcmp(bbuf
, abuf
, len
) != 0)
2902 fail("data2 compare failed");
2906 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
2907 struct supertype
*st
, unsigned long blocks
,
2908 int *fds
, unsigned long long *offsets
,
2909 int dests
, int *destfd
, unsigned long long *destoffsets
)
2911 /* Monitor a reshape where backup is being performed using
2912 * 'native' mechanism - either to a backup file, or
2913 * to some space in a spare.
2917 unsigned long long speed
;
2918 unsigned long long suspend_point
, array_size
;
2919 unsigned long long backup_point
, wait_point
;
2920 unsigned long long reshape_completed
;
2922 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
2923 int part
= 0; /* The next part of the backup area to fill. It may already
2924 * be full, so we need to check */
2925 int level
= reshape
->level
;
2926 int layout
= reshape
->before
.layout
;
2927 int data
= reshape
->before
.data_disks
;
2928 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
2929 int chunk
= sra
->array
.chunk_size
;
2931 unsigned long stripes
;
2933 /* set up the backup-super-block. This requires the
2934 * uuid from the array.
2936 /* Find a superblock */
2937 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2941 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2943 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
2944 devfd
= dev_open(dn
, O_RDONLY
);
2947 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
2953 fprintf(stderr
, Name
": Cannot find a superblock\n");
2957 memset(&bsb
, 0, 512);
2958 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2959 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
2960 bsb
.mtime
= __cpu_to_le64(time(0));
2961 bsb
.devstart2
= blocks
;
2963 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
2964 reshape
->before
.data_disks
;
2966 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2967 /* Don't start the 'reshape' */
2969 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
2970 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2971 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2975 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
2976 backup_point
= sra
->reshape_progress
;
2979 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
2980 backup_point
= reshape
->backup_blocks
;
2981 suspend_point
= array_size
;
2987 /* Want to return as soon the oldest backup slot can
2988 * be released as that allows us to start backing up
2989 * some more, providing suspend_point has been
2990 * advanced, which it should have.
2993 wait_point
= array_size
;
2994 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
2995 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
2996 __le64_to_cpu(bsb
.length
));
2997 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
2998 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
2999 __le64_to_cpu(bsb
.length2
));
3002 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
3003 wait_point
= __le64_to_cpu(bsb
.arraystart
);
3004 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
3005 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
3008 rv
= progress_reshape(sra
, reshape
,
3009 backup_point
, wait_point
,
3010 &suspend_point
, &reshape_completed
);
3011 /* external metadata would need to ping_monitor here */
3012 sra
->reshape_progress
= reshape_completed
;
3014 /* Clear any backup region that is before 'here' */
3016 if (__le64_to_cpu(bsb
.length
) > 0 &&
3017 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
3018 __le64_to_cpu(bsb
.length
)))
3019 forget_backup(dests
, destfd
,
3021 if (__le64_to_cpu(bsb
.length2
) > 0 &&
3022 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
3023 __le64_to_cpu(bsb
.length2
)))
3024 forget_backup(dests
, destfd
,
3027 if (__le64_to_cpu(bsb
.length
) > 0 &&
3028 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
3029 forget_backup(dests
, destfd
,
3031 if (__le64_to_cpu(bsb
.length2
) > 0 &&
3032 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
3033 forget_backup(dests
, destfd
,
3042 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
3043 /* No longer need to monitor this reshape */
3049 unsigned long long offset
;
3050 unsigned long actual_stripes
;
3051 /* Need to backup some data.
3052 * If 'part' is not used and the desired
3053 * backup size is suspended, do a backup,
3054 * then consider the next part.
3056 /* Check that 'part' is unused */
3057 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
3059 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
3062 offset
= backup_point
/ data
;
3063 actual_stripes
= stripes
;
3065 if (offset
+ actual_stripes
* (chunk
/512) >
3066 sra
->component_size
)
3067 actual_stripes
= ((sra
->component_size
- offset
)
3069 if (offset
+ actual_stripes
* (chunk
/512) >
3073 if (offset
< actual_stripes
* (chunk
/512))
3074 actual_stripes
= offset
/ (chunk
/512);
3075 offset
-= actual_stripes
* (chunk
/512);
3076 if (offset
< suspend_point
/data
)
3079 if (actual_stripes
== 0)
3081 grow_backup(sra
, offset
, actual_stripes
,
3083 disks
, chunk
, level
, layout
,
3084 dests
, destfd
, destoffsets
,
3085 part
, °raded
, buf
);
3086 validate(afd
, destfd
[0], destoffsets
[0]);
3087 /* record where 'part' is up to */
3090 backup_point
+= actual_stripes
* (chunk
/512) * data
;
3092 backup_point
-= actual_stripes
* (chunk
/512) * data
;
3096 /* FIXME maybe call progress_reshape one more time instead */
3097 abort_reshape(sra
); /* remove any remaining suspension */
3098 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3099 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
3105 * If any spare contains md_back_data-1 which is recent wrt mtime,
3106 * write that data into the array and update the super blocks with
3107 * the new reshape_progress
3109 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
3110 char *backup_file
, int verbose
)
3114 unsigned long long *offsets
;
3115 unsigned long long nstripe
, ostripe
;
3118 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
3119 if (info
->array
.level
== 6) odata
--; /* number of data disks */
3120 ndata
= info
->array
.raid_disks
- 1;
3121 if (info
->new_level
== 6) ndata
--;
3123 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
3125 if (info
->delta_disks
<= 0)
3126 /* Didn't grow, so the backup file must have
3130 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
3131 struct mdinfo dinfo
;
3134 char *devname
, namebuf
[20];
3135 unsigned long long lo
, hi
;
3137 /* This was a spare and may have some saved data on it.
3138 * Load the superblock, find and load the
3139 * backup_super_block.
3140 * If either fail, go on to next device.
3141 * If the backup contains no new info, just return
3142 * else restore data and update all superblocks
3144 if (i
== old_disks
-1) {
3145 fd
= open(backup_file
, O_RDONLY
);
3147 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
3148 backup_file
, strerror(errno
));
3151 devname
= backup_file
;
3156 if (st
->ss
->load_super(st
, fd
, NULL
))
3159 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3160 st
->ss
->free_super(st
);
3163 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
3165 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
3166 continue; /* Cannot seek */
3168 sprintf(namebuf
, "device-%d", i
);
3171 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
3173 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
3174 continue; /* Cannot read */
3176 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
3177 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
3179 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
3182 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
3184 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
3185 continue; /* bad checksum */
3187 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
3188 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
3190 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
3191 continue; /* Bad second checksum */
3193 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
3195 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
3196 continue; /* Wrong uuid */
3199 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3200 * sometimes they aren't... So allow considerable flexability in matching, and allow
3201 * this test to be overridden by an environment variable.
3203 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
3204 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
3205 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3206 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
3207 "for array with timestamp %lu\n",
3208 (unsigned long)__le64_to_cpu(bsb
.mtime
),
3209 (unsigned long)info
->array
.utime
);
3212 fprintf(stderr
, Name
": too-old timestamp on "
3213 "backup-metadata on %s\n", devname
);
3214 continue; /* time stamp is too bad */
3218 if (bsb
.magic
[15] == '1') {
3219 if (bsb
.length
== 0)
3221 if (info
->delta_disks
>= 0) {
3222 /* reshape_progress is increasing */
3223 if (__le64_to_cpu(bsb
.arraystart
)
3224 + __le64_to_cpu(bsb
.length
)
3225 < info
->reshape_progress
) {
3228 fprintf(stderr
, Name
3229 ": backup-metadata found on %s but is not needed\n", devname
);
3230 continue; /* No new data here */
3233 /* reshape_progress is decreasing */
3234 if (__le64_to_cpu(bsb
.arraystart
) >=
3235 info
->reshape_progress
)
3236 goto nonew
; /* No new data here */
3239 if (bsb
.length
== 0 && bsb
.length2
== 0)
3241 if (info
->delta_disks
>= 0) {
3242 /* reshape_progress is increasing */
3243 if ((__le64_to_cpu(bsb
.arraystart
)
3244 + __le64_to_cpu(bsb
.length
)
3245 < info
->reshape_progress
)
3247 (__le64_to_cpu(bsb
.arraystart2
)
3248 + __le64_to_cpu(bsb
.length2
)
3249 < info
->reshape_progress
))
3250 goto nonew
; /* No new data here */
3252 /* reshape_progress is decreasing */
3253 if (__le64_to_cpu(bsb
.arraystart
) >=
3254 info
->reshape_progress
&&
3255 __le64_to_cpu(bsb
.arraystart2
) >=
3256 info
->reshape_progress
)
3257 goto nonew
; /* No new data here */
3260 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
3263 fprintf(stderr
, Name
3264 ": Failed to verify secondary backup-metadata block on %s\n",
3266 continue; /* Cannot seek */
3268 /* There should be a duplicate backup superblock 4k before here */
3269 if (lseek64(fd
, -4096, 1) < 0 ||
3270 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
3271 goto second_fail
; /* Cannot find leading superblock */
3272 if (bsb
.magic
[15] == '1')
3273 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
3275 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
3276 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
3277 goto second_fail
; /* Cannot find leading superblock */
3279 /* Now need the data offsets for all devices. */
3280 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
3281 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
3284 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3285 /* FIXME should be this be an error */
3287 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3288 st
->ss
->free_super(st
);
3289 offsets
[j
] = dinfo
.data_offset
* 512;
3291 printf(Name
": restoring critical section\n");
3293 if (restore_stripes(fdlist
, offsets
,
3294 info
->array
.raid_disks
,
3298 fd
, __le64_to_cpu(bsb
.devstart
)*512,
3299 __le64_to_cpu(bsb
.arraystart
)*512,
3300 __le64_to_cpu(bsb
.length
)*512)) {
3301 /* didn't succeed, so giveup */
3303 fprintf(stderr
, Name
": Error restoring backup from %s\n",
3308 if (bsb
.magic
[15] == '2' &&
3309 restore_stripes(fdlist
, offsets
,
3310 info
->array
.raid_disks
,
3314 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
3315 __le64_to_cpu(bsb
.devstart2
)*512,
3316 __le64_to_cpu(bsb
.arraystart2
)*512,
3317 __le64_to_cpu(bsb
.length2
)*512)) {
3318 /* didn't succeed, so giveup */
3320 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
3326 /* Ok, so the data is restored. Let's update those superblocks. */
3330 lo
= __le64_to_cpu(bsb
.arraystart
);
3331 hi
= lo
+ __le64_to_cpu(bsb
.length
);
3333 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
3334 unsigned long long lo1
, hi1
;
3335 lo1
= __le64_to_cpu(bsb
.arraystart2
);
3336 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
3340 } else if (lo
< lo1
)
3346 (info
->reshape_progress
< lo
||
3347 info
->reshape_progress
> hi
))
3348 /* backup does not affect reshape_progress*/ ;
3349 else if (info
->delta_disks
>= 0) {
3350 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
3351 __le64_to_cpu(bsb
.length
);
3352 if (bsb
.magic
[15] == '2') {
3353 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
3354 __le64_to_cpu(bsb
.length2
);
3355 if (p2
> info
->reshape_progress
)
3356 info
->reshape_progress
= p2
;
3359 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
3360 if (bsb
.magic
[15] == '2') {
3361 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
3362 if (p2
< info
->reshape_progress
)
3363 info
->reshape_progress
= p2
;
3366 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
3367 if (fdlist
[j
] < 0) continue;
3368 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3370 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3371 dinfo
.reshape_progress
= info
->reshape_progress
;
3372 st
->ss
->update_super(st
, &dinfo
,
3373 "_reshape_progress",
3375 st
->ss
->store_super(st
, fdlist
[j
]);
3376 st
->ss
->free_super(st
);
3380 /* Didn't find any backup data, try to see if any
3383 if (info
->delta_disks
< 0) {
3384 /* When shrinking, the critical section is at the end.
3385 * So see if we are before the critical section.
3387 unsigned long long first_block
;
3388 nstripe
= ostripe
= 0;
3390 while (ostripe
>= nstripe
) {
3391 ostripe
+= info
->array
.chunk_size
/ 512;
3392 first_block
= ostripe
* odata
;
3393 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
3394 (info
->new_chunk
/512);
3397 if (info
->reshape_progress
>= first_block
)
3400 if (info
->delta_disks
> 0) {
3401 /* See if we are beyond the critical section. */
3402 unsigned long long last_block
;
3403 nstripe
= ostripe
= 0;
3405 while (nstripe
>= ostripe
) {
3406 nstripe
+= info
->new_chunk
/ 512;
3407 last_block
= nstripe
* ndata
;
3408 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
3409 (info
->array
.chunk_size
/512);
3412 if (info
->reshape_progress
>= last_block
)
3415 /* needed to recover critical section! */
3417 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
3421 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
3425 char *container
= NULL
;
3428 err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
3431 if (st
->ss
->external
) {
3432 fmt_devname(buf
, st
->container_dev
);
3436 if (!mdmon_running(st
->container_dev
))
3437 start_mdmon(st
->container_dev
);
3438 ping_monitor_by_id(st
->container_dev
);
3441 if (info
->reshape_active
== 2) {
3442 int cfd
= open_dev(st
->container_dev
);
3445 st
->ss
->load_container(st
, cfd
, container
);
3447 return reshape_container(container
, NULL
,
3448 st
, info
, 0, backup_file
,
3452 return reshape_array(container
, mdfd
, "array", st
, info
, 1,
3453 NULL
, backup_file
, 0, 0, 1);