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 child_grow(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
457 int *fds
, unsigned long long *offsets
,
458 int disks
, int chunk
, int level
, int layout
, int data
,
459 int dests
, int *destfd
, unsigned long long *destoffsets
);
460 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
461 int *fds
, unsigned long long *offsets
,
462 int disks
, int chunk
, int level
, int layout
, int data
,
463 int dests
, int *destfd
, unsigned long long *destoffsets
);
464 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
465 int *fds
, unsigned long long *offsets
,
466 unsigned long long start
,
467 int disks
, int chunk
, int level
, int layout
, int data
,
468 int dests
, int *destfd
, unsigned long long *destoffsets
);
470 static int check_idle(struct supertype
*st
)
472 /* Check that all member arrays for this container, or the
473 * container of this array, are idle
475 int container_dev
= (st
->container_dev
!= NoMdDev
476 ? st
->container_dev
: st
->devnum
);
478 struct mdstat_ent
*ent
, *e
;
481 fmt_devname(container
, container_dev
);
482 ent
= mdstat_read(0, 0);
483 for (e
= ent
; e
; e
= e
->next
) {
484 if (!is_container_member(e
, container
))
486 if (e
->percent
>= 0) {
495 static int freeze_container(struct supertype
*st
)
497 int container_dev
= (st
->container_dev
!= NoMdDev
498 ? st
->container_dev
: st
->devnum
);
504 fmt_devname(container
, container_dev
);
506 if (block_monitor(container
, 1)) {
507 fprintf(stderr
, Name
": failed to freeze container\n");
514 static void unfreeze_container(struct supertype
*st
)
516 int container_dev
= (st
->container_dev
!= NoMdDev
517 ? st
->container_dev
: st
->devnum
);
520 fmt_devname(container
, container_dev
);
522 unblock_monitor(container
, 1);
525 static int freeze(struct supertype
*st
)
527 /* Try to freeze resync/rebuild on this array/container.
528 * Return -1 if the array is busy,
529 * return -2 container cannot be frozen,
530 * return 0 if this kernel doesn't support 'frozen'
531 * return 1 if it worked.
533 if (st
->ss
->external
)
534 return freeze_container(st
);
536 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
541 err
= sysfs_freeze_array(sra
);
547 static void unfreeze(struct supertype
*st
, int frozen
)
549 /* If 'frozen' is 1, unfreeze the array */
553 if (st
->ss
->external
)
554 return unfreeze_container(st
);
556 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
559 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
561 fprintf(stderr
, Name
": failed to unfreeze array\n");
566 static void wait_reshape(struct mdinfo
*sra
)
568 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
574 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
575 strncmp(action
, "reshape", 7) == 0) {
579 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
584 static int reshape_super(struct supertype
*st
, long long size
, int level
,
585 int layout
, int chunksize
, int raid_disks
,
586 char *backup_file
, char *dev
, int verbose
)
588 /* nothing extra to check in the native case */
589 if (!st
->ss
->external
)
591 if (!st
->ss
->reshape_super
||
592 !st
->ss
->manage_reshape
) {
593 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
598 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
599 raid_disks
, backup_file
, dev
, verbose
);
602 static void sync_metadata(struct supertype
*st
)
604 if (st
->ss
->external
) {
605 if (st
->update_tail
) {
606 flush_metadata_updates(st
);
607 st
->update_tail
= &st
->updates
;
609 st
->ss
->sync_metadata(st
);
613 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
615 /* when dealing with external metadata subarrays we need to be
616 * prepared to handle EAGAIN. The kernel may need to wait for
617 * mdmon to mark the array active so the kernel can handle
618 * allocations/writeback when preparing the reshape action
619 * (md_allow_write()). We temporarily disable safe_mode_delay
620 * to close a race with the array_state going clean before the
621 * next write to raid_disks / stripe_cache_size
626 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
627 if (strcmp(name
, "raid_disks") != 0 &&
628 strcmp(name
, "stripe_cache_size") != 0)
629 return sysfs_set_num(sra
, NULL
, name
, n
);
631 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
634 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
635 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
636 if (rc
< 0 && errno
== EAGAIN
) {
637 ping_monitor(container
);
638 /* if we get EAGAIN here then the monitor is not active
641 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
643 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
647 static int reshape_container_raid_disks(char *container
, int raid_disks
)
649 /* for each subarray switch to a raid level that can
650 * support the reshape, and set raid disks
652 struct mdstat_ent
*ent
, *e
;
653 int changed
= 0, rv
= 0, err
= 0;
655 ent
= mdstat_read(1, 0);
657 fprintf(stderr
, Name
": unable to read /proc/mdstat\n");
662 for (e
= ent
; e
; e
= e
->next
) {
665 int level
, takeover_delta
= 0;
667 if (!is_container_member(e
, container
))
670 level
= map_name(pers
, e
->level
);
672 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
675 /* metadata records 'orig_level' */
676 rv
= sysfs_set_num(sub
, NULL
, "level", 4);
681 /* we want spares to be used for capacity
682 * expansion, not rebuild
698 sub
= sysfs_read(-1, e
->devnum
, GET_CHUNK
|GET_CACHE
);
701 cache
= (sub
->array
.chunk_size
/ 4096) * 4;
702 if (cache
> sub
->cache_size
)
703 rv
= subarray_set_num(container
, sub
,
704 "stripe_cache_size", cache
);
712 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
716 rv
= subarray_set_num(container
, sub
, "raid_disks",
717 raid_disks
+ takeover_delta
);
731 ": failed to initiate container reshape%s%s\n",
732 err
? ": " : "", err
? strerror(err
) : "");
739 static void revert_container_raid_disks(struct supertype
*st
, int fd
, char *container
)
741 /* we failed to prepare all subarrays in the container for
742 * reshape, so cancel the changes and restore the nominal raid
745 struct mdstat_ent
*ent
, *e
;
747 ent
= mdstat_read(0, 0);
750 ": failed to read /proc/mdstat while aborting reshape\n");
754 if (st
->ss
->load_container(st
, fd
, NULL
)) {
756 ": failed read metadata while aborting reshape\n");
761 for (e
= ent
; e
; e
= e
->next
) {
762 int level_fixed
= 0, disks_fixed
= 0;
763 struct mdinfo
*sub
, *prev
;
766 if (!is_container_member(e
, container
))
769 subarray
= to_subarray(e
, container
);
770 prev
= st
->ss
->container_content(st
, subarray
);
772 /* changing level might change raid_disks so we do it
773 * first and then check if raid_disks still needs fixing
775 if (map_name(pers
, e
->level
) != prev
->array
.level
) {
776 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
778 !sysfs_set_num(sub
, NULL
, "level", prev
->array
.level
))
784 sub
= sysfs_read(-1, e
->devnum
, GET_DISKS
);
785 if (sub
&& sub
->array
.raid_disks
!= prev
->array
.raid_disks
) {
786 if (!subarray_set_num(container
, sub
, "raid_disks",
787 prev
->array
.raid_disks
))
793 if (!disks_fixed
|| !level_fixed
)
795 ": failed to restore %s to a %d-disk %s array\n",
796 e
->dev
, prev
->array
.raid_disks
,
797 map_num(pers
, prev
->array
.level
));
800 st
->ss
->free_super(st
);
804 int remove_disks_on_raid10_to_raid0_takeover(struct supertype
*st
,
809 struct mdinfo
*remaining
;
812 nr_of_copies
= layout
& 0xff;
814 remaining
= sra
->devs
;
816 /* for each 'copy', select one device and remove from the list. */
817 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
818 struct mdinfo
**diskp
;
821 /* Find a working device to keep */
822 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
823 struct mdinfo
*disk
= *diskp
;
825 if (disk
->disk
.raid_disk
< slot
)
827 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
829 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
831 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
833 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
836 /* We have found a good disk to use! */
838 disk
->next
= sra
->devs
;
847 if (slot
< sra
->array
.raid_disks
) {
848 /* didn't find all slots */
854 sra
->devs
= remaining
;
858 /* Remove all 'remaining' devices from the array */
860 struct mdinfo
*sd
= remaining
;
861 remaining
= sd
->next
;
863 sysfs_set_str(sra
, sd
, "state", "faulty");
864 sysfs_set_str(sra
, sd
, "slot", "none");
865 sysfs_set_str(sra
, sd
, "state", "remove");
866 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
867 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
868 sd
->next
= sra
->devs
;
874 void reshape_free_fdlist(int *fdlist
,
875 unsigned long long *offsets
,
880 for (i
= 0; i
< size
; i
++)
888 int reshape_prepare_fdlist(char *devname
,
892 unsigned long blocks
,
895 unsigned long long *offsets
)
900 for (d
= 0; d
<= nrdisks
; d
++)
903 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
904 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
906 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
907 char *dn
= map_dev(sd
->disk
.major
,
909 fdlist
[sd
->disk
.raid_disk
]
910 = dev_open(dn
, O_RDONLY
);
911 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
912 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
914 Name
": %s: cannot open component %s\n",
915 devname
, dn
? dn
: "-unknown-");
919 } else if (backup_file
== NULL
) {
921 char *dn
= map_dev(sd
->disk
.major
,
923 fdlist
[d
] = dev_open(dn
, O_RDWR
);
924 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
926 fprintf(stderr
, Name
": %s: cannot open component %s\n",
927 devname
, dn
? dn
: "-unknown-");
938 int reshape_open_backup_file(char *backup_file
,
943 unsigned long long *offsets
)
945 /* Return 1 on success, 0 on any form of failure */
946 /* need to check backup file is large enough */
952 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|O_EXCL
,
956 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
957 devname
, backup_file
, strerror(errno
));
960 /* Guard against backup file being on array device.
961 * If array is partitioned or if LVM etc is in the
962 * way this will not notice, but it is better than
965 fstat(*fdlist
, &stb
);
968 if (stb
.st_rdev
== dev
) {
969 fprintf(stderr
, Name
": backup file must NOT be"
970 " on the array being reshaped.\n");
976 for (i
=0; i
< blocks
+ 1 ; i
++) {
977 if (write(*fdlist
, buf
, 512) != 512) {
978 fprintf(stderr
, Name
": %s: cannot create"
979 " backup file %s: %s\n",
980 devname
, backup_file
, strerror(errno
));
984 if (fsync(*fdlist
) != 0) {
985 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
986 devname
, backup_file
, strerror(errno
));
993 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
994 unsigned int ndata
, unsigned int odata
)
996 unsigned long a
, b
, blocks
;
997 /* So how much do we need to backup.
998 * We need an amount of data which is both a whole number of
999 * old stripes and a whole number of new stripes.
1000 * So LCM for (chunksize*datadisks).
1002 a
= (ochunk
/512) * odata
;
1003 b
= (nchunk
/512) * ndata
;
1011 /* LCM == product / GCD */
1012 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
1018 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1020 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1023 /* Make some changes in the shape of an array.
1024 * The kernel must support the change.
1026 * There are three different changes. Each can trigger
1027 * a resync or recovery so we freeze that until we have
1028 * requested everything (if kernel supports freezing - 2.6.30).
1030 * - change size (i.e. component_size)
1032 * - change layout/chunksize/ndisks
1034 * The last can require a reshape. It is different on different
1035 * levels so we need to check the level before actioning it.
1036 * Some times the level change needs to be requested after the
1037 * reshape (e.g. raid6->raid5, raid5->raid0)
1040 struct mdu_array_info_s array
, orig
;
1043 struct supertype
*st
;
1044 char *subarray
= NULL
;
1047 int nlayout
, olayout
;
1049 unsigned int ndata
, odata
;
1050 int orig_level
= UnSet
;
1051 char alt_layout
[40];
1053 unsigned long long *offsets
;
1058 unsigned long blocks
, stripes
;
1059 unsigned long cache
;
1060 unsigned long long array_size
;
1062 char *container
= NULL
;
1069 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1070 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1076 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1077 fprintf(stderr
, Name
": cannot change component size at the same time "
1078 "as other changes.\n"
1079 " Change size first, then check data is intact before "
1080 "making other changes.\n");
1084 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1085 get_linux_version() < 2006032 &&
1086 !check_env("MDADM_FORCE_FEWER")) {
1087 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1088 " Please use a newer kernel\n");
1092 st
= super_by_fd(fd
, &subarray
);
1094 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1097 if (raid_disks
> st
->max_devs
) {
1098 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1099 " beyond %d\n", st
->max_devs
);
1103 /* in the external case we need to check that the requested reshape is
1104 * supported, and perform an initial check that the container holds the
1105 * pre-requisite spare devices (mdmon owns final validation)
1107 if (st
->ss
->external
) {
1112 container_dev
= st
->container_dev
;
1113 cfd
= open_dev_excl(st
->container_dev
);
1114 } else if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 ||
1117 Name
": %s is a container, only 'raid-devices' can be changed\n",
1121 container_dev
= st
->devnum
;
1123 cfd
= open_dev_excl(st
->devnum
);
1127 fprintf(stderr
, Name
": Unable to open container for %s\n",
1133 container
= devnum2devname(st
->devnum
);
1135 fprintf(stderr
, Name
": Could not determine container name\n");
1141 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1143 rv
= st
->ss
->load_super(st
, cfd
, NULL
);
1145 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1151 if (mdmon_running(container_dev
))
1152 st
->update_tail
= &st
->updates
;
1155 if (raid_disks
> array
.raid_disks
&&
1156 array
.spare_disks
< (raid_disks
- array
.raid_disks
) &&
1159 Name
": Need %d spare%s to avoid degraded array,"
1160 " and only have %d.\n"
1161 " Use --force to over-ride this check.\n",
1162 raid_disks
- array
.raid_disks
,
1163 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1168 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
| GET_STATE
);
1170 if (st
->ss
->external
&& subarray
== NULL
) {
1171 array
.level
= LEVEL_CONTAINER
;
1172 sra
->array
.level
= LEVEL_CONTAINER
;
1175 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1179 frozen
= freeze(st
);
1181 /* freeze() already spewed the reason */
1183 } else if (frozen
< 0) {
1184 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1185 " be reshaped\n", devname
);
1189 /* ========= set size =============== */
1190 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1191 long long orig_size
= array
.size
;
1193 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1199 if (array
.size
!= size
) {
1200 /* got truncated to 32bit, write to
1201 * component_size instead
1204 rv
= sysfs_set_num(sra
, NULL
,
1205 "component_size", size
);
1209 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1213 /* restore metadata */
1214 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1215 NULL
, devname
, !quiet
) == 0)
1217 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1218 devname
, strerror(err
));
1220 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1221 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1225 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1226 size
= get_component_size(fd
)/2;
1230 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
1233 } else if (array
.level
!= LEVEL_CONTAINER
) {
1234 size
= get_component_size(fd
)/2;
1239 /* ========= check for Raid10 -> Raid0 conversion ===============
1240 * current implemenation assumes that following conditions must be met:
1242 * - near_copies == 2
1244 if (level
== 0 && array
.level
== 10 &&
1245 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) {
1247 err
= remove_disks_on_raid10_to_raid0_takeover(st
, sra
, array
.layout
);
1249 dprintf(Name
": Array cannot be reshaped\n");
1258 /* ======= set level =========== */
1259 if (level
!= UnSet
&& level
!= array
.level
) {
1260 /* Trying to change the level.
1261 * We might need to change layout first and schedule a
1262 * level change for later.
1263 * Level changes that can happen immediately are:
1264 * 0->4,5,6 1->5 4->5,6 5->1,6
1265 * Level changes that need a layout change first are:
1266 * 6->5,4,0 : need a -6 layout, or parity-last
1267 * 5->4,0 : need parity-last
1269 if ((array
.level
== 6 || array
.level
== 5) &&
1270 (level
== 5 || level
== 4 || level
== 0)) {
1271 /* Don't change level yet, but choose intermediate
1275 if (layout_str
== NULL
)
1276 switch (array
.layout
) {
1277 case ALGORITHM_LEFT_ASYMMETRIC
:
1278 case ALGORITHM_LEFT_ASYMMETRIC_6
:
1279 case ALGORITHM_ROTATING_N_RESTART
:
1280 layout_str
= "left-asymmetric-6";
1282 case ALGORITHM_LEFT_SYMMETRIC
:
1283 case ALGORITHM_LEFT_SYMMETRIC_6
:
1284 case ALGORITHM_ROTATING_N_CONTINUE
:
1285 layout_str
= "left-symmetric-6";
1287 case ALGORITHM_RIGHT_ASYMMETRIC
:
1288 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
1289 case ALGORITHM_ROTATING_ZERO_RESTART
:
1290 layout_str
= "right-asymmetric-6";
1292 case ALGORITHM_RIGHT_SYMMETRIC
:
1293 case ALGORITHM_RIGHT_SYMMETRIC_6
:
1294 layout_str
= "right-symmetric-6";
1296 case ALGORITHM_PARITY_0
:
1297 case ALGORITHM_PARITY_0_6
:
1298 layout_str
= "parity-first-6";
1300 case ALGORITHM_PARITY_N
:
1301 layout_str
= "parity-last";
1304 fprintf(stderr
, Name
": %s: cannot"
1305 "convert layout to RAID5 equivalent\n",
1311 int l
= map_name(r5layout
, layout_str
);
1313 fprintf(stderr
, Name
": %s: layout '%s' not recognised\n",
1314 devname
, layout_str
);
1318 if (l
!= ALGORITHM_PARITY_N
) {
1319 /* need the -6 version */
1320 char *ls
= map_num(r5layout
, l
);
1321 strcat(strcpy(alt_layout
, ls
),
1323 layout_str
= alt_layout
;
1327 /* The final raid6->raid5 conversion
1328 * will reduce the number of disks,
1329 * so now we need to aim higher
1333 layout_str
= "parity-last";
1335 /* Level change is a simple takeover. In the external
1336 * case we don't check with the metadata handler until
1337 * we establish what the final layout will be. If the
1338 * level change is disallowed we will revert to
1339 * orig_level without disturbing the metadata, otherwise
1340 * we will send an update.
1342 c
= map_num(pers
, level
);
1344 rv
= 1;/* not possible */
1348 /* Need to check there are enough spares */
1349 int spares_needed
= 0;
1350 switch (array
.level
* 16 + level
) {
1352 spares_needed
= 1; break;
1354 spares_needed
= 2; break;
1356 spares_needed
= 1; break;
1358 spares_needed
= 2; break;
1360 spares_needed
= 1; break;
1362 if (raid_disks
> array
.raid_disks
)
1363 spares_needed
+= raid_disks
-array
.raid_disks
;
1364 if (spares_needed
> array
.spare_disks
) {
1366 Name
": Need %d spare%s to avoid"
1367 " degraded array, and only have %d.\n"
1368 " Use --force to over-ride"
1371 spares_needed
== 1 ? "" : "s",
1377 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1380 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1383 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1384 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1389 orig_level
= orig
.level
;
1390 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1391 if (layout_str
== NULL
&&
1392 orig
.level
== 5 && level
== 6 &&
1393 array
.layout
!= orig
.layout
)
1394 layout_str
= map_num(r5layout
, orig
.layout
);
1396 fprintf(stderr
, Name
" level of %s changed to %s\n",
1402 /* ========= set shape (chunk_size / layout / ndisks) ============== */
1403 /* Check if layout change is a no-op */
1404 switch (array
.level
) {
1406 if (layout_str
&& array
.layout
== map_name(r5layout
, layout_str
))
1410 if (layout_str
== NULL
&&
1411 ((chunksize
&& chunksize
* 1024 != array
.chunk_size
) ||
1412 (raid_disks
&& raid_disks
!= array
.raid_disks
)) &&
1413 array
.layout
>= 16) {
1414 fprintf(stderr
, Name
1415 ": %s has a non-standard layout. If you wish to preserve this\n"
1416 " during the reshape, please specify --layout=preserve\n"
1417 " If you want to change it, specify a layout or use --layout=normalise\n",
1423 (strcmp(layout_str
, "normalise") == 0 ||
1424 strcmp(layout_str
, "normalize") == 0)) {
1426 strcpy(alt_layout
, map_num(r6layout
, array
.layout
));
1427 hyphen
= strrchr(alt_layout
, '-');
1428 if (hyphen
&& strcmp(hyphen
, "-6") == 0) {
1430 layout_str
= alt_layout
;
1434 if (layout_str
&& array
.layout
== map_name(r6layout
, layout_str
))
1436 if (layout_str
&& strcmp(layout_str
, "preserve") == 0)
1440 if (layout_str
== NULL
1441 && (chunksize
== 0 || chunksize
*1024 == array
.chunk_size
)
1442 && (raid_disks
== 0 || raid_disks
== array
.raid_disks
)) {
1443 if (reshape_super(st
, -1, level
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1449 if (level
!= UnSet
&& level
!= array
.level
) {
1450 /* Looks like this level change doesn't need
1451 * a reshape after all.
1453 c
= map_num(pers
, level
);
1455 rv
= sysfs_set_str(sra
, NULL
, "level", c
);
1458 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1461 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1462 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1466 } else if (!changed
&& !quiet
)
1467 fprintf(stderr
, Name
": %s: no change requested\n",
1470 if (st
->ss
->external
&& !mdmon_running(st
->container_dev
) &&
1472 start_mdmon(st
->container_dev
);
1473 ping_monitor(container
);
1478 c
= map_num(pers
, array
.level
);
1479 if (c
== NULL
) c
= "-unknown-";
1480 switch (array
.level
) {
1481 default: /* raid0, linear, multipath cannot be reconfigured */
1482 fprintf(stderr
, Name
": %s array %s cannot be reshaped.\n",
1484 /* TODO raid0 raiddisks can be reshaped via raid4 */
1487 case LEVEL_CONTAINER
: {
1490 /* double check that we are not changing anything but raid_disks */
1491 if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 || level
!= UnSet
) {
1493 Name
": %s is a container, only 'raid-devices' can be changed\n",
1499 if (reshape_super(st
, -1, UnSet
, UnSet
, 0, raid_disks
,
1500 backup_file
, devname
, !quiet
)) {
1505 count
= reshape_container_raid_disks(container
, raid_disks
);
1507 revert_container_raid_disks(st
, fd
, container
);
1510 } else if (count
== 0) {
1512 fprintf(stderr
, Name
1513 ": no active subarrays to reshape\n");
1518 if (!mdmon_running(st
->devnum
)) {
1519 start_mdmon(st
->devnum
);
1520 ping_monitor(container
);
1521 st
->update_tail
= &st
->updates
;
1524 /* give mdmon a chance to allocate spares */
1525 ping_manager(container
);
1527 /* manage_reshape takes care of releasing the array(s) */
1528 st
->ss
->manage_reshape(st
, backup_file
);
1532 case LEVEL_FAULTY
: /* only 'layout' change is permitted */
1534 if (chunksize
|| raid_disks
) {
1535 fprintf(stderr
, Name
": %s: Cannot change chunksize or disks of a 'faulty' array\n",
1540 if (layout_str
== NULL
)
1541 break; /* nothing to do.... */
1543 array
.layout
= parse_layout_faulty(layout_str
);
1544 if (array
.layout
< 0) {
1545 fprintf(stderr
, Name
": %s: layout %s not understood for 'faulty' array\n",
1546 devname
, layout_str
);
1550 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1551 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
1552 devname
, strerror(errno
));
1555 printf("layout for %s set to %d\n", devname
, array
.layout
);
1558 case 1: /* only raid_disks can each be changed. */
1560 if (chunksize
|| layout_str
!= NULL
) {
1561 fprintf(stderr
, Name
": %s: Cannot change chunk size or layout for a RAID1 array.\n",
1566 if (raid_disks
> 0) {
1567 if (reshape_super(st
, -1, UnSet
, UnSet
, 0, raid_disks
,
1568 NULL
, devname
, !quiet
)) {
1573 array
.raid_disks
= raid_disks
;
1574 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1575 fprintf(stderr
, Name
": Cannot set raid-devices for %s: %s\n",
1576 devname
, strerror(errno
));
1587 * layout/chunksize/raid_disks can be changed
1588 * though the kernel may not support it all.
1591 fprintf(stderr
, Name
": Cannot reshape subarrays yet\n");
1596 * There are three possibilities.
1597 * 1/ The array will shrink.
1598 * We need to ensure the reshape will pause before reaching
1599 * the 'critical section'. We also need to fork and wait for
1600 * that to happen. When it does we
1601 * suspend/backup/complete/unfreeze
1603 * 2/ The array will not change size.
1604 * This requires that we keep a backup of a sliding window
1605 * so that we can restore data after a crash. So we need
1606 * to fork and monitor progress.
1608 * 3/ The array will grow. This is relatively easy.
1609 * However the kernel's restripe routines will cheerfully
1610 * overwrite some early data before it is safe. So we
1611 * need to make a backup of the early parts of the array
1612 * and be ready to restore it if rebuild aborts very early.
1614 * We backup data by writing it to one spare, or to a
1615 * file which was given on command line.
1617 * [FOLLOWING IS OLD AND PARTLY WRONG]
1618 * So: we enumerate the devices in the array and
1619 * make sure we can open all of them.
1620 * Then we freeze the early part of the array and
1621 * backup to the various spares.
1622 * Then we request changes and start the reshape.
1623 * Monitor progress until it has passed the danger zone.
1624 * and finally invalidate the copied data and unfreeze the
1625 * start of the array.
1627 * In each case, we first make sure that storage is available
1628 * for the required backup.
1630 * - request the shape change.
1631 * - for to handle backup etc.
1633 nchunk
= ochunk
= array
.chunk_size
;
1634 nlayout
= olayout
= array
.layout
;
1635 ndisks
= odisks
= array
.raid_disks
;
1638 nchunk
= chunksize
* 1024;
1639 if (size
% chunksize
) {
1640 fprintf(stderr
, Name
": component size %lluK is not"
1641 " a multiple of chunksize %dK\n",
1646 if (layout_str
!= NULL
)
1647 switch(array
.level
) {
1648 case 4: /* ignore layout */
1651 nlayout
= map_name(r5layout
, layout_str
);
1652 if (nlayout
== UnSet
) {
1653 fprintf(stderr
, Name
": layout %s not understood for raid5.\n",
1661 nlayout
= map_name(r6layout
, layout_str
);
1662 if (nlayout
== UnSet
) {
1663 fprintf(stderr
, Name
": layout %s not understood for raid6.\n",
1670 if (raid_disks
) ndisks
= raid_disks
;
1674 if (array
.level
== 6) {
1675 odata
--; /* number of data disks */
1679 if (odata
== ndata
&&
1680 get_linux_version() < 2006032) {
1681 fprintf(stderr
, Name
": in-place reshape is not safe before 2.6.32, sorry.\n");
1685 /* Check that we can hold all the data */
1686 get_dev_size(fd
, NULL
, &array_size
);
1687 if (ndata
* (unsigned long long)size
< (array_size
/1024)) {
1688 fprintf(stderr
, Name
": this change will reduce the size of the array.\n"
1689 " use --grow --array-size first to truncate array.\n"
1690 " e.g. mdadm --grow %s --array-size %llu\n",
1691 devname
, ndata
* size
);
1696 blocks
= compute_backup_blocks(nchunk
, ochunk
, ndata
, odata
);
1699 sra
= sysfs_read(fd
, 0,
1700 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
1704 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
1710 if (ndata
== odata
) {
1711 /* Make 'blocks' bigger for better throughput, but
1712 * not so big that we reject it below.
1713 * Try for 16 megabytes
1715 while (blocks
* 32 < sra
->component_size
&&
1719 fprintf(stderr
, Name
": Need to backup %luK of critical "
1720 "section..\n", blocks
/2);
1722 if (blocks
>= sra
->component_size
/2) {
1723 fprintf(stderr
, Name
": %s: Something wrong - reshape aborted\n",
1728 nrdisks
= array
.raid_disks
+ sra
->array
.spare_disks
;
1729 /* Now we need to open all these devices so we can read/write.
1731 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1732 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1733 if (!fdlist
|| !offsets
) {
1734 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1739 d
= reshape_prepare_fdlist(devname
, sra
, array
.raid_disks
,
1740 nrdisks
, blocks
, backup_file
,
1746 if (backup_file
== NULL
) {
1747 if (st
->ss
->external
&& !st
->ss
->manage_reshape
) {
1748 fprintf(stderr
, Name
": %s Grow operation not supported by %s metadata\n",
1749 devname
, st
->ss
->name
);
1753 if (ndata
<= odata
) {
1754 fprintf(stderr
, Name
": %s: Cannot grow - need backup-file\n",
1758 } else if (sra
->array
.spare_disks
== 0) {
1759 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1760 "backup-file to backup critical section\n",
1765 if (d
== array
.raid_disks
) {
1766 fprintf(stderr
, Name
": %s: No spare device for backup\n",
1772 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
1774 fdlist
+d
, offsets
+d
)) {
1781 /* check that the operation is supported by the metadata */
1782 if (reshape_super(st
, -1, level
, nlayout
, nchunk
, ndisks
,
1783 backup_file
, devname
, !quiet
)) {
1788 /* lastly, check that the internal stripe cache is
1789 * large enough, or it won't work.
1792 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
1793 cache
= cache
* 4 / 4096;
1794 if (cache
< blocks
/ 8 / odisks
+ 16)
1795 /* Make it big enough to hold 'blocks' */
1796 cache
= blocks
/ 8 / odisks
+ 16;
1797 if (sra
->cache_size
< cache
)
1798 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
1800 /* Right, everything seems fine. Let's kick things off.
1801 * If only changing raid_disks, use ioctl, else use
1805 if (ochunk
== nchunk
&& olayout
== nlayout
) {
1806 array
.raid_disks
= ndisks
;
1807 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1810 fprintf(stderr
, Name
": Cannot set device shape for %s: %s\n",
1811 devname
, strerror(errno
));
1812 if (ndisks
< odisks
&&
1813 get_linux_version() < 2006030)
1814 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1816 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1817 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1822 /* set them all just in case some old 'new_*' value
1823 * persists from some earlier problem
1825 int err
= err
; /* only used if rv==1, and always set if
1826 * rv==1, so initialisation not needed,
1827 * despite gcc warning
1829 if (sysfs_set_num(sra
, NULL
, "chunk_size", nchunk
) < 0)
1830 rv
= 1, err
= errno
;
1831 if (!rv
&& sysfs_set_num(sra
, NULL
, "layout", nlayout
) < 0)
1832 rv
= 1, err
= errno
;
1833 if (!rv
&& sysfs_set_num(sra
, NULL
, "raid_disks", ndisks
) < 0)
1834 rv
= 1, err
= errno
;
1836 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1838 if (get_linux_version() < 2006030)
1839 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1841 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1842 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1847 if (ndisks
== 2 && odisks
== 2) {
1848 /* No reshape is needed in this trivial case */
1853 if (st
->ss
->external
) {
1854 /* metadata handler takes it from here */
1855 ping_manager(container
);
1856 st
->ss
->manage_reshape(st
, backup_file
);
1861 /* set up the backup-super-block. This requires the
1862 * uuid from the array.
1864 /* Find a superblock */
1865 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1869 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1871 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1872 devfd
= dev_open(dn
, O_RDONLY
);
1875 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
1881 fprintf(stderr
, Name
": %s: Cannot find a superblock\n",
1887 memset(&bsb
, 0, 512);
1888 memcpy(bsb
.magic
, "md_backup_data-1", 16);
1889 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
1890 bsb
.mtime
= __cpu_to_le64(time(0));
1891 bsb
.devstart2
= blocks
;
1892 stripes
= blocks
/ (ochunk
/512) / odata
;
1893 /* Now we just need to kick off the reshape and watch, while
1894 * handling backups of the data...
1895 * This is all done by a forked background process.
1900 if (check_env("MDADM_GROW_VERIFY"))
1901 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
1904 mlockall(MCL_FUTURE
);
1907 done
= child_grow(fd
, sra
, stripes
,
1909 odisks
, ochunk
, array
.level
, olayout
, odata
,
1910 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1911 else if (odata
> ndata
)
1912 done
= child_shrink(fd
, sra
, stripes
,
1914 odisks
, ochunk
, array
.level
, olayout
, odata
,
1915 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1917 done
= child_same_size(fd
, sra
, stripes
,
1920 odisks
, ochunk
, array
.level
, olayout
, odata
,
1921 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1922 if (backup_file
&& done
)
1923 unlink(backup_file
);
1924 if (level
!= UnSet
&& level
!= array
.level
) {
1925 /* We need to wait for the reshape to finish
1926 * (which will have happened unless odata < ndata)
1927 * and then set the level
1930 c
= map_num(pers
, level
);
1932 exit(0);/* not possible */
1936 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1938 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1943 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
1948 /* The child will take care of unfreezing the array */
1957 if (rv
&& orig_level
!= UnSet
&& sra
) {
1958 c
= map_num(pers
, orig_level
);
1959 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
1960 fprintf(stderr
, Name
": aborting level change\n");
1962 unfreeze(st
, frozen
);
1967 * We run a child process in the background which performs the following
1969 * - wait for resync to reach a certain point
1970 * - suspend io to the following section
1971 * - backup that section
1972 * - allow resync to proceed further
1974 * - discard the backup.
1976 * When are combined in slightly different ways in the three cases.
1978 * - suspend/backup/allow/wait/resume/discard
1980 * - allow/wait/suspend/backup/allow/wait/resume/discard
1982 * - wait/resume/discard/suspend/backup/allow
1984 * suspend/backup/allow always come together
1985 * wait/resume/discard do too.
1986 * For the same-size case we have two backups to improve flow.
1990 /* FIXME return status is never checked */
1991 static int grow_backup(struct mdinfo
*sra
,
1992 unsigned long long offset
, /* per device */
1993 unsigned long stripes
, /* per device */
1994 int *sources
, unsigned long long *offsets
,
1995 int disks
, int chunk
, int level
, int layout
,
1996 int dests
, int *destfd
, unsigned long long *destoffsets
,
1997 int part
, int *degraded
,
2000 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2001 * to storage 'destfd' (offset 'destoffsets'), after first
2002 * suspending IO. Then allow resync to continue
2003 * over the suspended section.
2004 * Use part 'part' of the backup-super-block.
2009 unsigned long long ll
;
2011 //printf("offset %llu\n", offset);
2016 sysfs_set_num(sra
, NULL
, "suspend_hi", (offset
+ stripes
* (chunk
/512)) * odata
);
2017 /* Check that array hasn't become degraded, else we might backup the wrong data */
2018 sysfs_get_ll(sra
, NULL
, "degraded", &ll
);
2019 new_degraded
= (int)ll
;
2020 if (new_degraded
!= *degraded
) {
2021 /* check each device to ensure it is still working */
2023 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2024 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2026 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2028 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2029 strstr(sbuf
, "faulty") ||
2030 strstr(sbuf
, "in_sync") == NULL
) {
2031 /* this device is dead */
2032 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2033 if (sd
->disk
.raid_disk
>= 0 &&
2034 sources
[sd
->disk
.raid_disk
] >= 0) {
2035 close(sources
[sd
->disk
.raid_disk
]);
2036 sources
[sd
->disk
.raid_disk
] = -1;
2041 *degraded
= new_degraded
;
2044 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2045 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2047 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2048 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2051 bsb
.magic
[15] = '2';
2052 for (i
= 0; i
< dests
; i
++)
2054 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
2056 lseek64(destfd
[i
], destoffsets
[i
], 0);
2058 rv
= save_stripes(sources
, offsets
,
2059 disks
, chunk
, level
, layout
,
2061 offset
*512*odata
, stripes
* chunk
* odata
,
2066 bsb
.mtime
= __cpu_to_le64(time(0));
2067 for (i
= 0; i
< dests
; i
++) {
2068 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2070 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2071 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2072 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2073 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2076 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
2077 != destoffsets
[i
] - 4096)
2079 if (write(destfd
[i
], &bsb
, 512) != 512)
2081 if (destoffsets
[i
] > 4096) {
2082 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
2083 destoffsets
[i
]+stripes
*chunk
*odata
)
2085 if (write(destfd
[i
], &bsb
, 512) != 512)
2095 /* in 2.6.30, the value reported by sync_completed can be
2096 * less that it should be by one stripe.
2097 * This only happens when reshape hits sync_max and pauses.
2098 * So allow wait_backup to either extent sync_max further
2099 * than strictly necessary, or return before the
2100 * sync has got quite as far as we would really like.
2101 * This is what 'blocks2' is for.
2102 * The various caller give appropriate values so that
2105 /* FIXME return value is often ignored */
2106 static int wait_backup(struct mdinfo
*sra
,
2107 unsigned long long offset
, /* per device */
2108 unsigned long long blocks
, /* per device */
2109 unsigned long long blocks2
, /* per device - hack */
2110 int dests
, int *destfd
, unsigned long long *destoffsets
,
2113 /* Wait for resync to pass the section that was backed up
2114 * then erase the backup and allow IO
2116 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
2117 unsigned long long completed
;
2123 sysfs_set_num(sra
, NULL
, "sync_max", offset
+ blocks
+ blocks2
);
2125 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
2127 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2131 while (completed
< offset
+ blocks
) {
2136 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2137 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2141 if (sysfs_get_str(sra
, NULL
, "sync_action",
2143 strncmp(action
, "reshape", 7) != 0)
2149 bsb
.arraystart2
= __cpu_to_le64(0);
2150 bsb
.length2
= __cpu_to_le64(0);
2152 bsb
.arraystart
= __cpu_to_le64(0);
2153 bsb
.length
= __cpu_to_le64(0);
2155 bsb
.mtime
= __cpu_to_le64(time(0));
2157 for (i
= 0; i
< dests
; i
++) {
2158 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2159 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2160 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2161 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2162 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2163 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
2164 destoffsets
[i
]-4096)
2167 write(destfd
[i
], &bsb
, 512) != 512)
2174 static void fail(char *msg
)
2177 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
2178 rv
|= (write(2, "\n", 1) != 1);
2182 static char *abuf
, *bbuf
;
2183 static unsigned long long abuflen
;
2184 static void validate(int afd
, int bfd
, unsigned long long offset
)
2186 /* check that the data in the backup against the array.
2187 * This is only used for regression testing and should not
2188 * be used while the array is active
2192 lseek64(bfd
, offset
- 4096, 0);
2193 if (read(bfd
, &bsb2
, 512) != 512)
2194 fail("cannot read bsb");
2195 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
2196 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
2197 fail("first csum bad");
2198 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
2199 fail("magic is bad");
2200 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
2201 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
2202 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
2203 fail("second csum bad");
2205 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
2206 fail("devstart is wrong");
2209 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
2211 if (abuflen
< len
) {
2215 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
2216 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
2218 /* just stop validating on mem-alloc failure */
2223 lseek64(bfd
, offset
, 0);
2224 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
2225 //printf("len %llu\n", len);
2226 fail("read first backup failed");
2228 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
2229 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2230 fail("read first from array failed");
2231 if (memcmp(bbuf
, abuf
, len
) != 0) {
2234 printf("offset=%llu len=%llu\n",
2235 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
2236 for (i
=0; i
<len
; i
++)
2237 if (bbuf
[i
] != abuf
[i
]) {
2238 printf("first diff byte %d\n", i
);
2242 fail("data1 compare failed");
2246 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
2248 if (abuflen
< len
) {
2252 abuf
= malloc(abuflen
);
2253 bbuf
= malloc(abuflen
);
2256 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2257 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2258 fail("read second backup failed");
2259 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2260 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2261 fail("read second from array failed");
2262 if (memcmp(bbuf
, abuf
, len
) != 0)
2263 fail("data2 compare failed");
2267 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2268 int *fds
, unsigned long long *offsets
,
2269 int disks
, int chunk
, int level
, int layout
, int data
,
2270 int dests
, int *destfd
, unsigned long long *destoffsets
)
2275 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2276 /* Don't start the 'reshape' */
2278 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2279 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2280 grow_backup(sra
, 0, stripes
,
2281 fds
, offsets
, disks
, chunk
, level
, layout
,
2282 dests
, destfd
, destoffsets
,
2284 validate(afd
, destfd
[0], destoffsets
[0]);
2285 wait_backup(sra
, 0, stripes
* (chunk
/ 512), stripes
* (chunk
/ 512),
2286 dests
, destfd
, destoffsets
,
2288 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2290 /* FIXME this should probably be numeric */
2291 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2295 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2296 int *fds
, unsigned long long *offsets
,
2297 int disks
, int chunk
, int level
, int layout
, int data
,
2298 int dests
, int *destfd
, unsigned long long *destoffsets
)
2301 unsigned long long start
;
2305 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2307 start
= sra
->component_size
- stripes
* (chunk
/512);
2308 sysfs_set_num(sra
, NULL
, "sync_max", start
);
2309 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
2310 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2311 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2312 rv
= wait_backup(sra
, 0, start
- stripes
* (chunk
/512), stripes
* (chunk
/512),
2313 dests
, destfd
, destoffsets
, 0);
2316 grow_backup(sra
, 0, stripes
,
2318 disks
, chunk
, level
, layout
,
2319 dests
, destfd
, destoffsets
,
2321 validate(afd
, destfd
[0], destoffsets
[0]);
2322 wait_backup(sra
, start
, stripes
*(chunk
/512), 0,
2323 dests
, destfd
, destoffsets
, 0);
2324 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2326 /* FIXME this should probably be numeric */
2327 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2331 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2332 int *fds
, unsigned long long *offsets
,
2333 unsigned long long start
,
2334 int disks
, int chunk
, int level
, int layout
, int data
,
2335 int dests
, int *destfd
, unsigned long long *destoffsets
)
2337 unsigned long long size
;
2338 unsigned long tailstripes
= stripes
;
2341 unsigned long long speed
;
2345 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2348 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2349 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2351 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2352 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2354 grow_backup(sra
, start
, stripes
,
2356 disks
, chunk
, level
, layout
,
2357 dests
, destfd
, destoffsets
,
2359 grow_backup(sra
, (start
+ stripes
) * (chunk
/512), stripes
,
2361 disks
, chunk
, level
, layout
,
2362 dests
, destfd
, destoffsets
,
2364 validate(afd
, destfd
[0], destoffsets
[0]);
2366 start
+= stripes
* 2; /* where to read next */
2367 size
= sra
->component_size
/ (chunk
/512);
2368 while (start
< size
) {
2369 if (wait_backup(sra
, (start
-stripes
*2)*(chunk
/512),
2370 stripes
*(chunk
/512), 0,
2371 dests
, destfd
, destoffsets
,
2374 sysfs_set_num(sra
, NULL
, "suspend_lo", start
*(chunk
/512) * data
);
2375 if (start
+ stripes
> size
)
2376 tailstripes
= (size
- start
);
2378 grow_backup(sra
, start
*(chunk
/512), tailstripes
,
2380 disks
, chunk
, level
, layout
,
2381 dests
, destfd
, destoffsets
,
2382 part
, °raded
, buf
);
2385 validate(afd
, destfd
[0], destoffsets
[0]);
2387 if (wait_backup(sra
, (start
-stripes
*2) * (chunk
/512), stripes
* (chunk
/512), 0,
2388 dests
, destfd
, destoffsets
,
2391 sysfs_set_num(sra
, NULL
, "suspend_lo", ((start
-stripes
)*(chunk
/512)) * data
);
2392 wait_backup(sra
, (start
-stripes
) * (chunk
/512), tailstripes
* (chunk
/512), 0,
2393 dests
, destfd
, destoffsets
,
2395 sysfs_set_num(sra
, NULL
, "suspend_lo", (size
*(chunk
/512)) * data
);
2396 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
2402 * If any spare contains md_back_data-1 which is recent wrt mtime,
2403 * write that data into the array and update the super blocks with
2404 * the new reshape_progress
2406 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
2407 char *backup_file
, int verbose
)
2411 unsigned long long *offsets
;
2412 unsigned long long nstripe
, ostripe
;
2415 if (info
->new_level
!= info
->array
.level
)
2416 return 1; /* Cannot handle level changes (they are instantaneous) */
2418 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
2419 if (info
->array
.level
== 6) odata
--; /* number of data disks */
2420 ndata
= info
->array
.raid_disks
- 1;
2421 if (info
->new_level
== 6) ndata
--;
2423 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
2425 if (info
->delta_disks
<= 0)
2426 /* Didn't grow, so the backup file must have
2430 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
2431 struct mdinfo dinfo
;
2434 char *devname
, namebuf
[20];
2436 /* This was a spare and may have some saved data on it.
2437 * Load the superblock, find and load the
2438 * backup_super_block.
2439 * If either fail, go on to next device.
2440 * If the backup contains no new info, just return
2441 * else restore data and update all superblocks
2443 if (i
== old_disks
-1) {
2444 fd
= open(backup_file
, O_RDONLY
);
2446 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
2447 backup_file
, strerror(errno
));
2450 devname
= backup_file
;
2455 if (st
->ss
->load_super(st
, fd
, NULL
))
2458 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2459 st
->ss
->free_super(st
);
2462 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
2464 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
2465 continue; /* Cannot seek */
2467 sprintf(namebuf
, "device-%d", i
);
2470 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
2472 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
2473 continue; /* Cannot read */
2475 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
2476 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
2478 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
2481 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
2483 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
2484 continue; /* bad checksum */
2486 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
2487 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
2489 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
2490 continue; /* Bad second checksum */
2492 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
2494 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
2495 continue; /* Wrong uuid */
2498 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2499 * sometimes they aren't... So allow considerable flexability in matching, and allow
2500 * this test to be overridden by an environment variable.
2502 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
2503 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
2504 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2505 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
2506 "for array with timestamp %lu\n",
2507 (unsigned long)__le64_to_cpu(bsb
.mtime
),
2508 (unsigned long)info
->array
.utime
);
2511 fprintf(stderr
, Name
": too-old timestamp on "
2512 "backup-metadata on %s\n", devname
);
2513 continue; /* time stamp is too bad */
2517 if (bsb
.magic
[15] == '1') {
2518 if (info
->delta_disks
>= 0) {
2519 /* reshape_progress is increasing */
2520 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2521 info
->reshape_progress
) {
2524 fprintf(stderr
, Name
": backup-metadata found on %s but is not needed\n", devname
);
2525 continue; /* No new data here */
2528 /* reshape_progress is decreasing */
2529 if (__le64_to_cpu(bsb
.arraystart
) >=
2530 info
->reshape_progress
)
2531 goto nonew
; /* No new data here */
2534 if (info
->delta_disks
>= 0) {
2535 /* reshape_progress is increasing */
2536 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2537 info
->reshape_progress
&&
2538 __le64_to_cpu(bsb
.arraystart2
) + __le64_to_cpu(bsb
.length2
) <
2539 info
->reshape_progress
)
2540 goto nonew
; /* No new data here */
2542 /* reshape_progress is decreasing */
2543 if (__le64_to_cpu(bsb
.arraystart
) >=
2544 info
->reshape_progress
&&
2545 __le64_to_cpu(bsb
.arraystart2
) >=
2546 info
->reshape_progress
)
2547 goto nonew
; /* No new data here */
2550 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
2553 fprintf(stderr
, Name
": Failed to verify secondary backup-metadata block on %s\n",
2555 continue; /* Cannot seek */
2557 /* There should be a duplicate backup superblock 4k before here */
2558 if (lseek64(fd
, -4096, 1) < 0 ||
2559 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
2560 goto second_fail
; /* Cannot find leading superblock */
2561 if (bsb
.magic
[15] == '1')
2562 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
2564 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
2565 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
2566 goto second_fail
; /* Cannot find leading superblock */
2568 /* Now need the data offsets for all devices. */
2569 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
2570 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
2573 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2574 /* FIXME should be this be an error */
2576 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2577 st
->ss
->free_super(st
);
2578 offsets
[j
] = dinfo
.data_offset
* 512;
2580 printf(Name
": restoring critical section\n");
2582 if (restore_stripes(fdlist
, offsets
,
2583 info
->array
.raid_disks
,
2587 fd
, __le64_to_cpu(bsb
.devstart
)*512,
2588 __le64_to_cpu(bsb
.arraystart
)*512,
2589 __le64_to_cpu(bsb
.length
)*512)) {
2590 /* didn't succeed, so giveup */
2592 fprintf(stderr
, Name
": Error restoring backup from %s\n",
2597 if (bsb
.magic
[15] == '2' &&
2598 restore_stripes(fdlist
, offsets
,
2599 info
->array
.raid_disks
,
2603 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
2604 __le64_to_cpu(bsb
.devstart2
)*512,
2605 __le64_to_cpu(bsb
.arraystart2
)*512,
2606 __le64_to_cpu(bsb
.length2
)*512)) {
2607 /* didn't succeed, so giveup */
2609 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
2615 /* Ok, so the data is restored. Let's update those superblocks. */
2617 if (info
->delta_disks
>= 0) {
2618 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
2619 __le64_to_cpu(bsb
.length
);
2620 if (bsb
.magic
[15] == '2') {
2621 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
2622 __le64_to_cpu(bsb
.length2
);
2623 if (p2
> info
->reshape_progress
)
2624 info
->reshape_progress
= p2
;
2627 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
2628 if (bsb
.magic
[15] == '2') {
2629 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
2630 if (p2
< info
->reshape_progress
)
2631 info
->reshape_progress
= p2
;
2634 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
2635 if (fdlist
[j
] < 0) continue;
2636 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2638 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2639 dinfo
.reshape_progress
= info
->reshape_progress
;
2640 st
->ss
->update_super(st
, &dinfo
,
2641 "_reshape_progress",
2643 st
->ss
->store_super(st
, fdlist
[j
]);
2644 st
->ss
->free_super(st
);
2648 /* Didn't find any backup data, try to see if any
2651 if (info
->delta_disks
< 0) {
2652 /* When shrinking, the critical section is at the end.
2653 * So see if we are before the critical section.
2655 unsigned long long first_block
;
2656 nstripe
= ostripe
= 0;
2658 while (ostripe
>= nstripe
) {
2659 ostripe
+= info
->array
.chunk_size
/ 512;
2660 first_block
= ostripe
* odata
;
2661 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
2662 (info
->new_chunk
/512);
2665 if (info
->reshape_progress
>= first_block
)
2668 if (info
->delta_disks
> 0) {
2669 /* See if we are beyond the critical section. */
2670 unsigned long long last_block
;
2671 nstripe
= ostripe
= 0;
2673 while (nstripe
>= ostripe
) {
2674 nstripe
+= info
->new_chunk
/ 512;
2675 last_block
= nstripe
* ndata
;
2676 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
2677 (info
->array
.chunk_size
/512);
2680 if (info
->reshape_progress
>= last_block
)
2683 /* needed to recover critical section! */
2685 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
2689 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
2692 /* Array is assembled and ready to be started, but
2693 * monitoring is probably required.
2696 * - set upper bound for resync
2697 * - initialise the 'suspend' boundaries
2698 * - switch to read-write
2699 * - fork and continue monitoring
2703 unsigned long long backup_offsets
[1];
2704 int odisks
, ndisks
, ochunk
, nchunk
,odata
,ndata
;
2705 unsigned long a
,b
,blocks
,stripes
;
2708 unsigned long long *offsets
;
2710 struct mdinfo
*sra
, *sd
;
2712 unsigned long cache
;
2715 err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
2719 /* make sure reshape doesn't progress until we are ready */
2720 sysfs_set_str(info
, NULL
, "sync_max", "0");
2721 sysfs_set_str(info
, NULL
, "array_state", "active"); /* FIXME or clean */
2723 sra
= sysfs_read(-1, devname2devnum(info
->sys_name
),
2724 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
2729 /* ndisks is not growing, so raid_disks is old and +delta is new */
2730 odisks
= info
->array
.raid_disks
;
2731 ndisks
= odisks
+ info
->delta_disks
;
2734 if (info
->array
.level
== 6) {
2738 ochunk
= info
->array
.chunk_size
;
2739 nchunk
= info
->new_chunk
;
2741 a
= (ochunk
/512) * odata
;
2742 b
= (nchunk
/512) * ndata
;
2750 /* LCM == product / GCD */
2751 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
2754 while (blocks
* 32 < sra
->component_size
&&
2757 stripes
= blocks
/ (info
->array
.chunk_size
/512) / odata
;
2759 /* check that the internal stripe cache is
2760 * large enough, or it won't work.
2762 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
2763 cache
= cache
* 4 / 4096;
2764 if (cache
< blocks
/ 8 / odisks
+ 16)
2765 /* Make it big enough to hold 'blocks' */
2766 cache
= blocks
/ 8 / odisks
+ 16;
2767 if (sra
->cache_size
< cache
)
2768 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
2771 memset(&bsb
, 0, 512);
2772 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2773 memcpy(&bsb
.set_uuid
, info
->uuid
, 16);
2774 bsb
.mtime
= __cpu_to_le64(time(0));
2775 bsb
.devstart2
= blocks
;
2777 backup_fd
= open(backup_file
, O_RDWR
|O_CREAT
, S_IRUSR
| S_IWUSR
);
2778 if (backup_fd
< 0) {
2779 fprintf(stderr
, Name
": Cannot open backup file %s\n",
2780 backup_file
?: "- no backup-file given");
2783 backup_list
[0] = backup_fd
;
2784 backup_offsets
[0] = 8 * 512;
2785 fds
= malloc(odisks
* sizeof(fds
[0]));
2786 offsets
= malloc(odisks
* sizeof(offsets
[0]));
2787 for (d
=0; d
<odisks
; d
++)
2790 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2791 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2793 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2794 char *dn
= map_dev(sd
->disk
.major
,
2796 fds
[sd
->disk
.raid_disk
]
2797 = dev_open(dn
, O_RDONLY
);
2798 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
2799 if (fds
[sd
->disk
.raid_disk
] < 0) {
2800 fprintf(stderr
, Name
": %s: cannot open component %s\n",
2801 info
->sys_name
, dn
?dn
:"-unknown-");
2812 mlockall(MCL_FUTURE
);
2813 if (info
->delta_disks
< 0)
2814 done
= child_shrink(-1, info
, stripes
,
2816 info
->array
.raid_disks
,
2817 info
->array
.chunk_size
,
2818 info
->array
.level
, info
->array
.layout
,
2820 1, backup_list
, backup_offsets
);
2821 else if (info
->delta_disks
== 0) {
2822 /* The 'start' is a per-device stripe number.
2823 * reshape_progress is a per-array sector number.
2824 * So divide by ndata * chunk_size
2826 unsigned long long start
= info
->reshape_progress
/ ndata
;
2827 start
/= (info
->array
.chunk_size
/512);
2828 done
= child_same_size(-1, info
, stripes
,
2831 info
->array
.raid_disks
,
2832 info
->array
.chunk_size
,
2833 info
->array
.level
, info
->array
.layout
,
2835 1, backup_list
, backup_offsets
);
2837 if (backup_file
&& done
)
2838 unlink(backup_file
);
2839 /* FIXME should I intuit a level change */
2842 fprintf(stderr
, Name
": Cannot run child to continue monitoring reshape: %s\n",