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
;
232 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
233 " between different architectured. Consider upgrading 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) {
356 fprintf(stderr
, Name
": failed to set internal bitmap.\n");
363 int max_devs
= st
->max_devs
;
365 /* try to load a superblock */
366 for (d
=0; d
<max_devs
; d
++) {
367 mdu_disk_info_t disk
;
371 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
373 if ((disk
.major
==0 && disk
.minor
==0) ||
374 (disk
.state
& (1<<MD_DISK_REMOVED
)))
376 dv
= map_dev(disk
.major
, disk
.minor
, 1);
378 fd2
= dev_open(dv
, O_RDONLY
);
380 st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
382 st
->ss
->uuid_from_super(st
, uuid
);
388 fprintf(stderr
, Name
": cannot find UUID for array!\n");
391 if (CreateBitmap(file
, force
, (char*)uuid
, chunk
,
392 delay
, write_behind
, bitmapsize
, major
)) {
395 bitmap_fd
= open(file
, O_RDWR
);
397 fprintf(stderr
, Name
": weird: %s cannot be opened\n",
401 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
402 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
403 devname
, strerror(errno
));
413 * When reshaping an array we might need to backup some data.
414 * This is written to all spares with a 'super_block' describing it.
415 * The superblock goes 4K from the end of the used space on the
417 * It if written after the backup is complete.
418 * It has the following structure.
421 static struct mdp_backup_super
{
422 char magic
[16]; /* md_backup_data-1 or -2 */
425 /* start/sizes in 512byte sectors */
426 __u64 devstart
; /* address on backup device/file of data */
429 __u32 sb_csum
; /* csum of preceeding bytes. */
431 __u64 devstart2
; /* offset in to data of second section */
434 __u32 sb_csum2
; /* csum of preceeding bytes. */
436 } __attribute__((aligned(512))) bsb
, bsb2
;
438 static __u32
bsb_csum(char *buf
, int len
)
442 for (i
=0; i
<len
; i
++)
443 csum
= (csum
<<3) + buf
[0];
444 return __cpu_to_le32(csum
);
447 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
448 int *fds
, unsigned long long *offsets
,
449 int disks
, int chunk
, int level
, int layout
, int data
,
450 int dests
, int *destfd
, unsigned long long *destoffsets
);
451 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
452 int *fds
, unsigned long long *offsets
,
453 int disks
, int chunk
, int level
, int layout
, int data
,
454 int dests
, int *destfd
, unsigned long long *destoffsets
);
455 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
456 int *fds
, unsigned long long *offsets
,
457 unsigned long long start
,
458 int disks
, int chunk
, int level
, int layout
, int data
,
459 int dests
, int *destfd
, unsigned long long *destoffsets
);
461 static int freeze_container(struct supertype
*st
)
463 int container_dev
= (st
->container_dev
!= NoMdDev
464 ? st
->container_dev
: st
->devnum
);
465 char *container
= devnum2devname(container_dev
);
469 ": could not determine container name, freeze aborted\n");
473 if (block_monitor(container
, 1)) {
474 fprintf(stderr
, Name
": failed to freeze container\n");
481 static void unfreeze_container(struct supertype
*st
)
483 int container_dev
= (st
->container_dev
!= NoMdDev
484 ? st
->container_dev
: st
->devnum
);
485 char *container
= devnum2devname(container_dev
);
489 ": could not determine container name, unfreeze aborted\n");
493 unblock_monitor(container
, 1);
496 static int freeze(struct supertype
*st
)
498 /* Try to freeze resync/rebuild on this array/container.
499 * Return -1 if the array is busy,
500 * return -2 container cannot be frozen,
501 * return 0 if this kernel doesn't support 'frozen'
502 * return 1 if it worked.
504 if (st
->ss
->external
)
505 return freeze_container(st
);
507 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
512 err
= sysfs_freeze_array(sra
);
518 static void unfreeze(struct supertype
*st
, int frozen
)
520 /* If 'frozen' is 1, unfreeze the array */
524 if (st
->ss
->external
)
525 return unfreeze_container(st
);
527 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
530 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
532 fprintf(stderr
, Name
": failed to unfreeze array\n");
537 static void wait_reshape(struct mdinfo
*sra
)
539 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
546 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
548 if (sysfs_fd_get_str(fd
, action
, 20) < 0) {
552 } while (strncmp(action
, "reshape", 7) == 0);
555 static int reshape_super(struct supertype
*st
, long long size
, int level
,
556 int layout
, int chunksize
, int raid_disks
,
557 char *backup_file
, char *dev
, int verbose
)
559 /* nothing extra to check in the native case */
560 if (!st
->ss
->external
)
562 if (!st
->ss
->reshape_super
||
563 !st
->ss
->manage_reshape
) {
564 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
569 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
570 raid_disks
, backup_file
, dev
, verbose
);
573 static void sync_metadata(struct supertype
*st
)
575 if (st
->ss
->external
) {
577 flush_metadata_updates(st
);
579 st
->ss
->sync_metadata(st
);
583 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
585 /* when dealing with external metadata subarrays we need to be
586 * prepared to handle EAGAIN. The kernel may need to wait for
587 * mdmon to mark the array active so the kernel can handle
588 * allocations/writeback when preparing the reshape action
589 * (md_allow_write()). We temporarily disable safe_mode_delay
590 * to close a race with the array_state going clean before the
591 * next write to raid_disks / stripe_cache_size
596 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
597 if (strcmp(name
, "raid_disks") != 0 &&
598 strcmp(name
, "stripe_cache_size") != 0)
599 return sysfs_set_num(sra
, NULL
, name
, n
);
601 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
604 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
605 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
606 if (rc
< 0 && errno
== EAGAIN
) {
607 ping_monitor(container
);
608 /* if we get EAGAIN here then the monitor is not active
611 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
613 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
617 static int reshape_container_raid_disks(char *container
, int raid_disks
)
619 /* for each subarray switch to a raid level that can
620 * support the reshape, and set raid disks
622 struct mdstat_ent
*ent
, *e
;
623 int changed
= 0, rv
= 0, err
= 0;
625 ent
= mdstat_read(1, 0);
627 fprintf(stderr
, Name
": unable to read /proc/mdstat\n");
632 for (e
= ent
; e
; e
= e
->next
) {
635 int level
, takeover_delta
= 0;
637 if (!is_container_member(e
, container
))
640 level
= map_name(pers
, e
->level
);
642 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
645 /* metadata records 'orig_level' */
646 rv
= sysfs_set_num(sub
, NULL
, "level", 4);
651 /* we want spares to be used for capacity
652 * expansion, not rebuild
668 sub
= sysfs_read(-1, e
->devnum
, GET_CHUNK
|GET_CACHE
);
671 cache
= (sub
->array
.chunk_size
/ 4096) * 4;
672 if (cache
> sub
->cache_size
)
673 rv
= subarray_set_num(container
, sub
,
674 "stripe_cache_size", cache
);
682 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
686 rv
= subarray_set_num(container
, sub
, "raid_disks",
687 raid_disks
+ takeover_delta
);
701 ": failed to initiate container reshape%s%s\n",
702 err
? ": " : "", err
? strerror(err
) : "");
709 static void revert_container_raid_disks(struct supertype
*st
, int fd
, char *container
)
711 /* we failed to prepare all subarrays in the container for
712 * reshape, so cancel the changes and restore the nominal raid
715 struct mdstat_ent
*ent
, *e
;
717 ent
= mdstat_read(0, 0);
720 ": failed to read /proc/mdstat while aborting reshape\n");
724 if (st
->ss
->load_container(st
, fd
, NULL
)) {
726 ": failed read metadata while aborting reshape\n");
731 for (e
= ent
; e
; e
= e
->next
) {
732 int level_fixed
= 0, disks_fixed
= 0;
733 struct mdinfo
*sub
, *prev
;
736 if (!is_container_member(e
, container
))
739 subarray
= to_subarray(e
, container
);
740 prev
= st
->ss
->container_content(st
, subarray
);
742 /* changing level might change raid_disks so we do it
743 * first and then check if raid_disks still needs fixing
745 if (map_name(pers
, e
->level
) != prev
->array
.level
) {
746 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
748 !sysfs_set_num(sub
, NULL
, "level", prev
->array
.level
))
754 sub
= sysfs_read(-1, e
->devnum
, GET_DISKS
);
755 if (sub
&& sub
->array
.raid_disks
!= prev
->array
.raid_disks
) {
756 if (!subarray_set_num(container
, sub
, "raid_disks",
757 prev
->array
.raid_disks
))
763 if (!disks_fixed
|| !level_fixed
)
765 ": failed to restore %s to a %d-disk %s array\n",
766 e
->dev
, prev
->array
.raid_disks
,
767 map_num(pers
, prev
->array
.level
));
770 st
->ss
->free_super(st
);
774 int remove_disks_on_raid10_to_raid0_takeover(struct supertype
*st
,
779 struct mdinfo
*remaining
;
782 nr_of_copies
= layout
& 0xff;
784 remaining
= sra
->devs
;
786 /* for each 'copy', select one device and remove from the list. */
787 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
788 struct mdinfo
**diskp
;
791 /* Find a working device to keep */
792 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
793 struct mdinfo
*disk
= *diskp
;
795 if (disk
->disk
.raid_disk
< slot
)
797 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
799 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
801 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
803 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
806 /* We have found a good disk to use! */
808 disk
->next
= sra
->devs
;
817 if (slot
< sra
->array
.raid_disks
) {
818 /* didn't find all slots */
824 sra
->devs
= remaining
;
828 /* Remove all 'remaining' devices from the array */
830 struct mdinfo
*sd
= remaining
;
831 remaining
= sd
->next
;
833 sysfs_set_str(sra
, sd
, "state", "faulty");
834 sysfs_set_str(sra
, sd
, "slot", "none");
835 sysfs_set_str(sra
, sd
, "state", "remove");
836 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
837 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
838 sd
->next
= sra
->devs
;
844 void reshape_free_fdlist(int *fdlist
,
845 unsigned long long *offsets
,
850 for (i
= 0; i
< size
; i
++)
858 int reshape_prepare_fdlist(char *devname
,
862 unsigned long blocks
,
865 unsigned long long *offsets
)
870 for (d
= 0; d
<= nrdisks
; d
++)
873 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
874 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
876 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
877 char *dn
= map_dev(sd
->disk
.major
,
879 fdlist
[sd
->disk
.raid_disk
]
880 = dev_open(dn
, O_RDONLY
);
881 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
882 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
884 Name
": %s: cannot open component %s\n",
885 devname
, dn
? dn
: "-unknown-");
889 } else if (backup_file
== NULL
) {
891 char *dn
= map_dev(sd
->disk
.major
,
893 fdlist
[d
] = dev_open(dn
, O_RDWR
);
894 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
896 fprintf(stderr
, Name
": %s: cannot open component %s\n",
897 devname
, dn
? dn
: "-unknown-");
908 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
909 unsigned int ndata
, unsigned int odata
)
911 unsigned long a
, b
, blocks
;
912 /* So how much do we need to backup.
913 * We need an amount of data which is both a whole number of
914 * old stripes and a whole number of new stripes.
915 * So LCM for (chunksize*datadisks).
917 a
= (ochunk
/512) * odata
;
918 b
= (nchunk
/512) * ndata
;
926 /* LCM == product / GCD */
927 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
933 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
935 int level
, char *layout_str
, int chunksize
, int raid_disks
)
937 /* Make some changes in the shape of an array.
938 * The kernel must support the change.
940 * There are three different changes. Each can trigger
941 * a resync or recovery so we freeze that until we have
942 * requested everything (if kernel supports freezing - 2.6.30).
944 * - change size (i.e. component_size)
946 * - change layout/chunksize/ndisks
948 * The last can require a reshape. It is different on different
949 * levels so we need to check the level before actioning it.
950 * Some times the level change needs to be requested after the
951 * reshape (e.g. raid6->raid5, raid5->raid0)
954 struct mdu_array_info_s array
, orig
;
957 struct supertype
*st
;
958 char *subarray
= NULL
;
961 int nlayout
, olayout
;
963 unsigned int ndata
, odata
;
964 int orig_level
= UnSet
;
967 unsigned long long *offsets
;
972 unsigned long blocks
, stripes
;
974 unsigned long long array_size
;
976 char *container
= NULL
;
983 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
984 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
990 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
991 fprintf(stderr
, Name
": cannot change component size at the same time "
992 "as other changes.\n"
993 " Change size first, then check data is intact before "
994 "making other changes.\n");
998 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
999 get_linux_version() < 2006032 &&
1000 !check_env("MDADM_FORCE_FEWER")) {
1001 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1002 " Please use a newer kernel\n");
1006 st
= super_by_fd(fd
, &subarray
);
1008 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1012 /* in the external case we need to check that the requested reshape is
1013 * supported, and perform an initial check that the container holds the
1014 * pre-requisite spare devices (mdmon owns final validation)
1016 if (st
->ss
->external
) {
1021 container_dev
= st
->container_dev
;
1022 cfd
= open_dev_excl(st
->container_dev
);
1023 } else if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 ||
1026 Name
": %s is a container, only 'raid-devices' can be changed\n",
1030 container_dev
= st
->devnum
;
1032 cfd
= open_dev_excl(st
->devnum
);
1036 fprintf(stderr
, Name
": Unable to open container for %s\n",
1042 container
= devnum2devname(st
->devnum
);
1044 fprintf(stderr
, Name
": Could not determine container name\n");
1050 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1052 rv
= st
->ss
->load_super(st
, cfd
, NULL
);
1054 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1060 if (mdmon_running(container_dev
))
1061 st
->update_tail
= &st
->updates
;
1064 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
| GET_STATE
);
1066 if (st
->ss
->external
&& subarray
== NULL
) {
1067 array
.level
= LEVEL_CONTAINER
;
1068 sra
->array
.level
= LEVEL_CONTAINER
;
1071 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1075 frozen
= freeze(st
);
1077 /* freeze() already spewed the reason */
1079 } else if (frozen
< 0) {
1080 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1081 " be reshaped\n", devname
);
1085 /* ========= set size =============== */
1086 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1087 long long orig_size
= array
.size
;
1089 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1095 if (array
.size
!= size
) {
1096 /* got truncated to 32bit, write to
1097 * component_size instead
1100 rv
= sysfs_set_num(sra
, NULL
,
1101 "component_size", size
);
1105 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1109 /* restore metadata */
1110 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1111 NULL
, devname
, !quiet
) == 0)
1113 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1114 devname
, strerror(err
));
1116 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1117 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1121 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1122 size
= get_component_size(fd
)/2;
1126 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
1129 } else if (array
.level
!= LEVEL_CONTAINER
) {
1130 size
= get_component_size(fd
)/2;
1135 /* ========= check for Raid10 -> Raid0 conversion ===============
1136 * current implemenation assumes that following conditions must be met:
1138 * - near_copies == 2
1140 if (level
== 0 && array
.level
== 10 &&
1141 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) {
1143 err
= remove_disks_on_raid10_to_raid0_takeover(st
, sra
, array
.layout
);
1145 dprintf(Name
": Array cannot be reshaped\n");
1154 /* ======= set level =========== */
1155 if (level
!= UnSet
&& level
!= array
.level
) {
1156 /* Trying to change the level.
1157 * We might need to change layout first and schedule a
1158 * level change for later.
1159 * Level changes that can happen immediately are:
1160 * 0->4,5,6 1->5 4->5,6 5->1,6
1161 * Level changes that need a layout change first are:
1162 * 6->5,4,0 : need a -6 layout, or parity-last
1163 * 5->4,0 : need parity-last
1165 if ((array
.level
== 6 || array
.level
== 5) &&
1166 (level
== 5 || level
== 4 || level
== 0)) {
1167 /* Don't change level yet, but choose intermediate
1171 if (layout_str
== NULL
)
1172 switch (array
.layout
) {
1173 case ALGORITHM_LEFT_ASYMMETRIC
:
1174 case ALGORITHM_LEFT_ASYMMETRIC_6
:
1175 case ALGORITHM_ROTATING_N_RESTART
:
1176 layout_str
= "left-asymmetric-6";
1178 case ALGORITHM_LEFT_SYMMETRIC
:
1179 case ALGORITHM_LEFT_SYMMETRIC_6
:
1180 case ALGORITHM_ROTATING_N_CONTINUE
:
1181 layout_str
= "left-symmetric-6";
1183 case ALGORITHM_RIGHT_ASYMMETRIC
:
1184 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
1185 case ALGORITHM_ROTATING_ZERO_RESTART
:
1186 layout_str
= "right-asymmetric-6";
1188 case ALGORITHM_RIGHT_SYMMETRIC
:
1189 case ALGORITHM_RIGHT_SYMMETRIC_6
:
1190 layout_str
= "right-symmetric-6";
1192 case ALGORITHM_PARITY_0
:
1193 case ALGORITHM_PARITY_0_6
:
1194 layout_str
= "parity-first-6";
1196 case ALGORITHM_PARITY_N
:
1197 layout_str
= "parity-last";
1200 fprintf(stderr
, Name
": %s: cannot"
1201 "convert layout to RAID5 equivalent\n",
1207 int l
= map_name(r5layout
, layout_str
);
1209 fprintf(stderr
, Name
": %s: layout '%s' not recognised\n",
1210 devname
, layout_str
);
1214 if (l
!= ALGORITHM_PARITY_N
) {
1215 /* need the -6 version */
1216 char *ls
= map_num(r5layout
, l
);
1217 strcat(strcpy(alt_layout
, ls
),
1219 layout_str
= alt_layout
;
1223 /* The final raid6->raid5 conversion
1224 * will reduce the number of disks,
1225 * so now we need to aim higher
1229 layout_str
= "parity-last";
1231 /* Level change is a simple takeover. In the external
1232 * case we don't check with the metadata handler until
1233 * we establish what the final layout will be. If the
1234 * level change is disallowed we will revert to
1235 * orig_level without disturbing the metadata, otherwise
1236 * we will send an update.
1238 c
= map_num(pers
, level
);
1240 rv
= 1;/* not possible */
1243 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1246 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1249 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1250 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1255 orig_level
= orig
.level
;
1256 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1257 if (layout_str
== NULL
&&
1258 orig
.level
== 5 && level
== 6 &&
1259 array
.layout
!= orig
.layout
)
1260 layout_str
= map_num(r5layout
, orig
.layout
);
1262 fprintf(stderr
, Name
" level of %s changed to %s\n",
1268 /* ========= set shape (chunk_size / layout / ndisks) ============== */
1269 /* Check if layout change is a no-op */
1270 switch (array
.level
) {
1272 if (layout_str
&& array
.layout
== map_name(r5layout
, layout_str
))
1276 if (layout_str
== NULL
&&
1277 ((chunksize
&& chunksize
* 1024 != array
.chunk_size
) ||
1278 (raid_disks
&& raid_disks
!= array
.raid_disks
)) &&
1279 array
.layout
>= 16) {
1280 fprintf(stderr
, Name
1281 ": %s has a non-standard layout. If you wish to preserve this\n"
1282 " during the reshape, please specify --layout=preserve\n"
1283 " If you want to change it, specify a layout or use --layout=normalise\n",
1289 (strcmp(layout_str
, "normalise") == 0 ||
1290 strcmp(layout_str
, "normalize") == 0)) {
1292 strcpy(alt_layout
, map_num(r6layout
, array
.layout
));
1293 hyphen
= strrchr(alt_layout
, '-');
1294 if (hyphen
&& strcmp(hyphen
, "-6") == 0) {
1296 layout_str
= alt_layout
;
1300 if (layout_str
&& array
.layout
== map_name(r6layout
, layout_str
))
1302 if (layout_str
&& strcmp(layout_str
, "preserve") == 0)
1306 if (layout_str
== NULL
1307 && (chunksize
== 0 || chunksize
*1024 == array
.chunk_size
)
1308 && (raid_disks
== 0 || raid_disks
== array
.raid_disks
)) {
1309 if (reshape_super(st
, -1, level
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1315 if (level
!= UnSet
&& level
!= array
.level
) {
1316 /* Looks like this level change doesn't need
1317 * a reshape after all.
1319 c
= map_num(pers
, level
);
1321 rv
= sysfs_set_str(sra
, NULL
, "level", c
);
1324 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1327 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1328 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1332 } else if (!changed
&& !quiet
)
1333 fprintf(stderr
, Name
": %s: no change requested\n",
1336 if (st
->ss
->external
&& !mdmon_running(st
->container_dev
) &&
1338 start_mdmon(st
->container_dev
);
1339 ping_monitor(container
);
1344 c
= map_num(pers
, array
.level
);
1345 if (c
== NULL
) c
= "-unknown-";
1346 switch (array
.level
) {
1347 default: /* raid0, linear, multipath cannot be reconfigured */
1348 fprintf(stderr
, Name
": %s array %s cannot be reshaped.\n",
1350 /* TODO raid0 raiddisks can be reshaped via raid4 */
1353 case LEVEL_CONTAINER
: {
1356 /* double check that we are not changing anything but raid_disks */
1357 if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 || level
!= UnSet
) {
1359 Name
": %s is a container, only 'raid-devices' can be changed\n",
1365 st
->update_tail
= &st
->updates
;
1366 if (reshape_super(st
, -1, UnSet
, UnSet
, 0, raid_disks
,
1367 backup_file
, devname
, !quiet
)) {
1372 count
= reshape_container_raid_disks(container
, raid_disks
);
1374 revert_container_raid_disks(st
, fd
, container
);
1377 } else if (count
== 0) {
1379 fprintf(stderr
, Name
1380 ": no active subarrays to reshape\n");
1384 if (!mdmon_running(st
->devnum
)) {
1385 start_mdmon(st
->devnum
);
1386 ping_monitor(container
);
1390 /* give mdmon a chance to allocate spares */
1391 ping_manager(container
);
1393 /* manage_reshape takes care of releasing the array(s) */
1394 st
->ss
->manage_reshape(st
, backup_file
);
1398 case LEVEL_FAULTY
: /* only 'layout' change is permitted */
1400 if (chunksize
|| raid_disks
) {
1401 fprintf(stderr
, Name
": %s: Cannot change chunksize or disks of a 'faulty' array\n",
1406 if (layout_str
== NULL
)
1407 break; /* nothing to do.... */
1409 array
.layout
= parse_layout_faulty(layout_str
);
1410 if (array
.layout
< 0) {
1411 fprintf(stderr
, Name
": %s: layout %s not understood for 'faulty' array\n",
1412 devname
, layout_str
);
1416 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1417 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
1418 devname
, strerror(errno
));
1421 printf("layout for %s set to %d\n", devname
, array
.layout
);
1424 case 1: /* only raid_disks can each be changed. */
1426 if (chunksize
|| layout_str
!= NULL
) {
1427 fprintf(stderr
, Name
": %s: Cannot change chunk size or layout for a RAID1 array.\n",
1432 if (raid_disks
> 0) {
1433 if (reshape_super(st
, -1, UnSet
, UnSet
, 0, raid_disks
,
1434 NULL
, devname
, !quiet
)) {
1439 array
.raid_disks
= raid_disks
;
1440 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1441 fprintf(stderr
, Name
": Cannot set raid-devices for %s: %s\n",
1442 devname
, strerror(errno
));
1453 * layout/chunksize/raid_disks can be changed
1454 * though the kernel may not support it all.
1457 fprintf(stderr
, Name
": Cannot reshape subarrays yet\n");
1462 * There are three possibilities.
1463 * 1/ The array will shrink.
1464 * We need to ensure the reshape will pause before reaching
1465 * the 'critical section'. We also need to fork and wait for
1466 * that to happen. When it does we
1467 * suspend/backup/complete/unfreeze
1469 * 2/ The array will not change size.
1470 * This requires that we keep a backup of a sliding window
1471 * so that we can restore data after a crash. So we need
1472 * to fork and monitor progress.
1474 * 3/ The array will grow. This is relatively easy.
1475 * However the kernel's restripe routines will cheerfully
1476 * overwrite some early data before it is safe. So we
1477 * need to make a backup of the early parts of the array
1478 * and be ready to restore it if rebuild aborts very early.
1480 * We backup data by writing it to one spare, or to a
1481 * file which was given on command line.
1483 * [FOLLOWING IS OLD AND PARTLY WRONG]
1484 * So: we enumerate the devices in the array and
1485 * make sure we can open all of them.
1486 * Then we freeze the early part of the array and
1487 * backup to the various spares.
1488 * Then we request changes and start the reshape.
1489 * Monitor progress until it has passed the danger zone.
1490 * and finally invalidate the copied data and unfreeze the
1491 * start of the array.
1493 * In each case, we first make sure that storage is available
1494 * for the required backup.
1496 * - request the shape change.
1497 * - for to handle backup etc.
1499 nchunk
= ochunk
= array
.chunk_size
;
1500 nlayout
= olayout
= array
.layout
;
1501 ndisks
= odisks
= array
.raid_disks
;
1504 nchunk
= chunksize
* 1024;
1505 if (size
% chunksize
) {
1506 fprintf(stderr
, Name
": component size %lluK is not"
1507 " a multiple of chunksize %dK\n",
1512 if (layout_str
!= NULL
)
1513 switch(array
.level
) {
1514 case 4: /* ignore layout */
1517 nlayout
= map_name(r5layout
, layout_str
);
1518 if (nlayout
== UnSet
) {
1519 fprintf(stderr
, Name
": layout %s not understood for raid5.\n",
1527 nlayout
= map_name(r6layout
, layout_str
);
1528 if (nlayout
== UnSet
) {
1529 fprintf(stderr
, Name
": layout %s not understood for raid6.\n",
1536 if (raid_disks
) ndisks
= raid_disks
;
1540 if (array
.level
== 6) {
1541 odata
--; /* number of data disks */
1545 if (odata
== ndata
&&
1546 get_linux_version() < 2006032) {
1547 fprintf(stderr
, Name
": in-place reshape is not safe before 2.6.32, sorry.\n");
1551 /* Check that we can hold all the data */
1552 get_dev_size(fd
, NULL
, &array_size
);
1553 if (ndata
* (unsigned long long)size
< (array_size
/1024)) {
1554 fprintf(stderr
, Name
": this change will reduce the size of the array.\n"
1555 " use --grow --array-size first to truncate array.\n"
1556 " e.g. mdadm --grow %s --array-size %llu\n",
1557 devname
, ndata
* size
);
1562 blocks
= compute_backup_blocks(nchunk
, ochunk
, ndata
, odata
);
1565 sra
= sysfs_read(fd
, 0,
1566 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
1570 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
1576 if (ndata
== odata
) {
1577 /* Make 'blocks' bigger for better throughput, but
1578 * not so big that we reject it below.
1579 * Try for 16 megabytes
1581 while (blocks
* 32 < sra
->component_size
&&
1585 fprintf(stderr
, Name
": Need to backup %luK of critical "
1586 "section..\n", blocks
/2);
1588 if (blocks
>= sra
->component_size
/2) {
1589 fprintf(stderr
, Name
": %s: Something wrong - reshape aborted\n",
1594 nrdisks
= array
.raid_disks
+ sra
->array
.spare_disks
;
1595 /* Now we need to open all these devices so we can read/write.
1597 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1598 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1599 if (!fdlist
|| !offsets
) {
1600 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1605 d
= reshape_prepare_fdlist(devname
, sra
, array
.raid_disks
,
1606 nrdisks
, blocks
, backup_file
,
1612 if (backup_file
== NULL
) {
1613 if (st
->ss
->external
&& !st
->ss
->manage_reshape
) {
1614 fprintf(stderr
, Name
": %s Grow operation not supported by %s metadata\n",
1615 devname
, st
->ss
->name
);
1619 if (ndata
<= odata
) {
1620 fprintf(stderr
, Name
": %s: Cannot grow - need backup-file\n",
1624 } else if (sra
->array
.spare_disks
== 0) {
1625 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1626 "backup-file to backup critical section\n",
1631 if (d
== array
.raid_disks
) {
1632 fprintf(stderr
, Name
": %s: No spare device for backup\n",
1638 /* need to check backup file is large enough */
1640 fdlist
[d
] = open(backup_file
, O_RDWR
|O_CREAT
|O_EXCL
,
1642 offsets
[d
] = 8 * 512;
1643 if (fdlist
[d
] < 0) {
1644 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1645 devname
, backup_file
, strerror(errno
));
1649 memset(buf
, 0, 512);
1650 for (i
=0; i
< (signed)blocks
+ 1 ; i
++) {
1651 if (write(fdlist
[d
], buf
, 512) != 512) {
1652 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1653 devname
, backup_file
, strerror(errno
));
1658 if (fsync(fdlist
[d
]) != 0) {
1659 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1660 devname
, backup_file
, strerror(errno
));
1667 /* check that the operation is supported by the metadata */
1668 if (reshape_super(st
, -1, level
, nlayout
, nchunk
, ndisks
,
1669 backup_file
, devname
, !quiet
)) {
1674 /* lastly, check that the internal stripe cache is
1675 * large enough, or it won't work.
1678 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
1679 cache
= cache
* 4 / 4096;
1680 if (cache
< blocks
/ 8 / odisks
+ 16)
1681 /* Make it big enough to hold 'blocks' */
1682 cache
= blocks
/ 8 / odisks
+ 16;
1683 if (sra
->cache_size
< cache
)
1684 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
1686 /* Right, everything seems fine. Let's kick things off.
1687 * If only changing raid_disks, use ioctl, else use
1691 if (ochunk
== nchunk
&& olayout
== nlayout
) {
1692 array
.raid_disks
= ndisks
;
1693 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1696 fprintf(stderr
, Name
": Cannot set device shape for %s: %s\n",
1697 devname
, strerror(errno
));
1698 if (ndisks
< odisks
&&
1699 get_linux_version() < 2006030)
1700 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1702 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1703 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1708 /* set them all just in case some old 'new_*' value
1709 * persists from some earlier problem
1711 int err
= err
; /* only used if rv==1, and always set if
1712 * rv==1, so initialisation not needed,
1713 * despite gcc warning
1715 if (sysfs_set_num(sra
, NULL
, "chunk_size", nchunk
) < 0)
1716 rv
= 1, err
= errno
;
1717 if (!rv
&& sysfs_set_num(sra
, NULL
, "layout", nlayout
) < 0)
1718 rv
= 1, err
= errno
;
1719 if (!rv
&& sysfs_set_num(sra
, NULL
, "raid_disks", ndisks
) < 0)
1720 rv
= 1, err
= errno
;
1722 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1724 if (get_linux_version() < 2006030)
1725 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1727 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1728 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1733 if (ndisks
== 2 && odisks
== 2) {
1734 /* No reshape is needed in this trivial case */
1739 if (st
->ss
->external
) {
1740 /* metadata handler takes it from here */
1741 ping_manager(container
);
1742 st
->ss
->manage_reshape(st
, backup_file
);
1747 /* set up the backup-super-block. This requires the
1748 * uuid from the array.
1750 /* Find a superblock */
1751 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1755 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1757 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1758 devfd
= dev_open(dn
, O_RDONLY
);
1761 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
1767 fprintf(stderr
, Name
": %s: Cannot find a superblock\n",
1773 memset(&bsb
, 0, 512);
1774 memcpy(bsb
.magic
, "md_backup_data-1", 16);
1775 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
1776 bsb
.mtime
= __cpu_to_le64(time(0));
1777 bsb
.devstart2
= blocks
;
1778 stripes
= blocks
/ (ochunk
/512) / odata
;
1779 /* Now we just need to kick off the reshape and watch, while
1780 * handling backups of the data...
1781 * This is all done by a forked background process.
1786 if (check_env("MDADM_GROW_VERIFY"))
1787 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
1790 mlockall(MCL_FUTURE
);
1793 done
= child_grow(fd
, sra
, stripes
,
1795 odisks
, ochunk
, array
.level
, olayout
, odata
,
1796 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1797 else if (odata
> ndata
)
1798 done
= child_shrink(fd
, sra
, stripes
,
1800 odisks
, ochunk
, array
.level
, olayout
, odata
,
1801 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1803 done
= child_same_size(fd
, sra
, stripes
,
1806 odisks
, ochunk
, array
.level
, olayout
, odata
,
1807 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1808 if (backup_file
&& done
)
1809 unlink(backup_file
);
1810 if (level
!= UnSet
&& level
!= array
.level
) {
1811 /* We need to wait for the reshape to finish
1812 * (which will have happened unless odata < ndata)
1813 * and then set the level
1816 c
= map_num(pers
, level
);
1818 exit(0);/* not possible */
1822 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1824 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1829 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
1834 /* The child will take care of unfreezing the array */
1843 if (rv
&& orig_level
!= UnSet
&& sra
) {
1844 c
= map_num(pers
, orig_level
);
1845 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
1846 fprintf(stderr
, Name
": aborting level change\n");
1848 unfreeze(st
, frozen
);
1853 * We run a child process in the background which performs the following
1855 * - wait for resync to reach a certain point
1856 * - suspend io to the following section
1857 * - backup that section
1858 * - allow resync to proceed further
1860 * - discard the backup.
1862 * When are combined in slightly different ways in the three cases.
1864 * - suspend/backup/allow/wait/resume/discard
1866 * - allow/wait/suspend/backup/allow/wait/resume/discard
1868 * - wait/resume/discard/suspend/backup/allow
1870 * suspend/backup/allow always come together
1871 * wait/resume/discard do too.
1872 * For the same-size case we have two backups to improve flow.
1876 /* FIXME return status is never checked */
1877 static int grow_backup(struct mdinfo
*sra
,
1878 unsigned long long offset
, /* per device */
1879 unsigned long stripes
, /* per device */
1880 int *sources
, unsigned long long *offsets
,
1881 int disks
, int chunk
, int level
, int layout
,
1882 int dests
, int *destfd
, unsigned long long *destoffsets
,
1883 int part
, int *degraded
,
1886 /* Backup 'blocks' sectors at 'offset' on each device of the array,
1887 * to storage 'destfd' (offset 'destoffsets'), after first
1888 * suspending IO. Then allow resync to continue
1889 * over the suspended section.
1890 * Use part 'part' of the backup-super-block.
1895 unsigned long long ll
;
1897 //printf("offset %llu\n", offset);
1902 sysfs_set_num(sra
, NULL
, "suspend_hi", (offset
+ stripes
* (chunk
/512)) * odata
);
1903 /* Check that array hasn't become degraded, else we might backup the wrong data */
1904 sysfs_get_ll(sra
, NULL
, "degraded", &ll
);
1905 new_degraded
= (int)ll
;
1906 if (new_degraded
!= *degraded
) {
1907 /* check each device to ensure it is still working */
1909 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1910 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1912 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
1914 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
1915 strstr(sbuf
, "faulty") ||
1916 strstr(sbuf
, "in_sync") == NULL
) {
1917 /* this device is dead */
1918 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
1919 if (sd
->disk
.raid_disk
>= 0 &&
1920 sources
[sd
->disk
.raid_disk
] >= 0) {
1921 close(sources
[sd
->disk
.raid_disk
]);
1922 sources
[sd
->disk
.raid_disk
] = -1;
1927 *degraded
= new_degraded
;
1930 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
1931 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
1933 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
1934 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
1937 bsb
.magic
[15] = '2';
1938 for (i
= 0; i
< dests
; i
++)
1940 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
1942 lseek64(destfd
[i
], destoffsets
[i
], 0);
1944 rv
= save_stripes(sources
, offsets
,
1945 disks
, chunk
, level
, layout
,
1947 offset
*512*odata
, stripes
* chunk
* odata
,
1952 bsb
.mtime
= __cpu_to_le64(time(0));
1953 for (i
= 0; i
< dests
; i
++) {
1954 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
1956 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
1957 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
1958 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
1959 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
1962 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
1963 != destoffsets
[i
] - 4096)
1965 if (write(destfd
[i
], &bsb
, 512) != 512)
1967 if (destoffsets
[i
] > 4096) {
1968 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
1969 destoffsets
[i
]+stripes
*chunk
*odata
)
1971 if (write(destfd
[i
], &bsb
, 512) != 512)
1981 /* in 2.6.30, the value reported by sync_completed can be
1982 * less that it should be by one stripe.
1983 * This only happens when reshape hits sync_max and pauses.
1984 * So allow wait_backup to either extent sync_max further
1985 * than strictly necessary, or return before the
1986 * sync has got quite as far as we would really like.
1987 * This is what 'blocks2' is for.
1988 * The various caller give appropriate values so that
1991 /* FIXME return value is often ignored */
1992 static int wait_backup(struct mdinfo
*sra
,
1993 unsigned long long offset
, /* per device */
1994 unsigned long long blocks
, /* per device */
1995 unsigned long long blocks2
, /* per device - hack */
1996 int dests
, int *destfd
, unsigned long long *destoffsets
,
1999 /* Wait for resync to pass the section that was backed up
2000 * then erase the backup and allow IO
2002 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
2003 unsigned long long completed
;
2009 sysfs_set_num(sra
, NULL
, "sync_max", offset
+ blocks
+ blocks2
);
2011 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
2017 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2018 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2022 if (sysfs_get_str(sra
, NULL
, "sync_action",
2024 strncmp(action
, "reshape", 7) != 0)
2026 } while (completed
< offset
+ blocks
);
2030 bsb
.arraystart2
= __cpu_to_le64(0);
2031 bsb
.length2
= __cpu_to_le64(0);
2033 bsb
.arraystart
= __cpu_to_le64(0);
2034 bsb
.length
= __cpu_to_le64(0);
2036 bsb
.mtime
= __cpu_to_le64(time(0));
2038 for (i
= 0; i
< dests
; i
++) {
2039 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2040 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2041 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2042 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2043 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2044 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
2045 destoffsets
[i
]-4096)
2048 write(destfd
[i
], &bsb
, 512) != 512)
2055 static void fail(char *msg
)
2058 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
2059 rv
|= (write(2, "\n", 1) != 1);
2063 static char *abuf
, *bbuf
;
2064 static unsigned long long abuflen
;
2065 static void validate(int afd
, int bfd
, unsigned long long offset
)
2067 /* check that the data in the backup against the array.
2068 * This is only used for regression testing and should not
2069 * be used while the array is active
2073 lseek64(bfd
, offset
- 4096, 0);
2074 if (read(bfd
, &bsb2
, 512) != 512)
2075 fail("cannot read bsb");
2076 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
2077 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
2078 fail("first csum bad");
2079 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
2080 fail("magic is bad");
2081 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
2082 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
2083 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
2084 fail("second csum bad");
2086 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
2087 fail("devstart is wrong");
2090 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
2092 if (abuflen
< len
) {
2096 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
2097 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
2099 /* just stop validating on mem-alloc failure */
2104 lseek64(bfd
, offset
, 0);
2105 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
2106 //printf("len %llu\n", len);
2107 fail("read first backup failed");
2109 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
2110 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2111 fail("read first from array failed");
2112 if (memcmp(bbuf
, abuf
, len
) != 0) {
2115 printf("offset=%llu len=%llu\n",
2116 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
2117 for (i
=0; i
<len
; i
++)
2118 if (bbuf
[i
] != abuf
[i
]) {
2119 printf("first diff byte %d\n", i
);
2123 fail("data1 compare failed");
2127 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
2129 if (abuflen
< len
) {
2133 abuf
= malloc(abuflen
);
2134 bbuf
= malloc(abuflen
);
2137 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2138 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2139 fail("read second backup failed");
2140 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2141 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2142 fail("read second from array failed");
2143 if (memcmp(bbuf
, abuf
, len
) != 0)
2144 fail("data2 compare failed");
2148 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2149 int *fds
, unsigned long long *offsets
,
2150 int disks
, int chunk
, int level
, int layout
, int data
,
2151 int dests
, int *destfd
, unsigned long long *destoffsets
)
2156 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2157 /* Don't start the 'reshape' */
2159 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2160 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2161 grow_backup(sra
, 0, stripes
,
2162 fds
, offsets
, disks
, chunk
, level
, layout
,
2163 dests
, destfd
, destoffsets
,
2165 validate(afd
, destfd
[0], destoffsets
[0]);
2166 wait_backup(sra
, 0, stripes
* (chunk
/ 512), stripes
* (chunk
/ 512),
2167 dests
, destfd
, destoffsets
,
2169 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2171 /* FIXME this should probably be numeric */
2172 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2176 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2177 int *fds
, unsigned long long *offsets
,
2178 int disks
, int chunk
, int level
, int layout
, int data
,
2179 int dests
, int *destfd
, unsigned long long *destoffsets
)
2182 unsigned long long start
;
2186 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2188 start
= sra
->component_size
- stripes
* (chunk
/512);
2189 sysfs_set_num(sra
, NULL
, "sync_max", start
);
2190 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
2191 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2192 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2193 rv
= wait_backup(sra
, 0, start
- stripes
* (chunk
/512), stripes
* (chunk
/512),
2194 dests
, destfd
, destoffsets
, 0);
2197 grow_backup(sra
, 0, stripes
,
2199 disks
, chunk
, level
, layout
,
2200 dests
, destfd
, destoffsets
,
2202 validate(afd
, destfd
[0], destoffsets
[0]);
2203 wait_backup(sra
, start
, stripes
*(chunk
/512), 0,
2204 dests
, destfd
, destoffsets
, 0);
2205 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2207 /* FIXME this should probably be numeric */
2208 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2212 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2213 int *fds
, unsigned long long *offsets
,
2214 unsigned long long start
,
2215 int disks
, int chunk
, int level
, int layout
, int data
,
2216 int dests
, int *destfd
, unsigned long long *destoffsets
)
2218 unsigned long long size
;
2219 unsigned long tailstripes
= stripes
;
2222 unsigned long long speed
;
2226 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2229 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2230 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2232 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2233 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2235 grow_backup(sra
, start
, stripes
,
2237 disks
, chunk
, level
, layout
,
2238 dests
, destfd
, destoffsets
,
2240 grow_backup(sra
, (start
+ stripes
) * (chunk
/512), stripes
,
2242 disks
, chunk
, level
, layout
,
2243 dests
, destfd
, destoffsets
,
2245 validate(afd
, destfd
[0], destoffsets
[0]);
2247 start
+= stripes
* 2; /* where to read next */
2248 size
= sra
->component_size
/ (chunk
/512);
2249 while (start
< size
) {
2250 if (wait_backup(sra
, (start
-stripes
*2)*(chunk
/512),
2251 stripes
*(chunk
/512), 0,
2252 dests
, destfd
, destoffsets
,
2255 sysfs_set_num(sra
, NULL
, "suspend_lo", start
*(chunk
/512) * data
);
2256 if (start
+ stripes
> size
)
2257 tailstripes
= (size
- start
);
2259 grow_backup(sra
, start
*(chunk
/512), tailstripes
,
2261 disks
, chunk
, level
, layout
,
2262 dests
, destfd
, destoffsets
,
2263 part
, °raded
, buf
);
2266 validate(afd
, destfd
[0], destoffsets
[0]);
2268 if (wait_backup(sra
, (start
-stripes
*2) * (chunk
/512), stripes
* (chunk
/512), 0,
2269 dests
, destfd
, destoffsets
,
2272 sysfs_set_num(sra
, NULL
, "suspend_lo", ((start
-stripes
)*(chunk
/512)) * data
);
2273 wait_backup(sra
, (start
-stripes
) * (chunk
/512), tailstripes
* (chunk
/512), 0,
2274 dests
, destfd
, destoffsets
,
2276 sysfs_set_num(sra
, NULL
, "suspend_lo", (size
*(chunk
/512)) * data
);
2277 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
2283 * If any spare contains md_back_data-1 which is recent wrt mtime,
2284 * write that data into the array and update the super blocks with
2285 * the new reshape_progress
2287 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
2288 char *backup_file
, int verbose
)
2292 unsigned long long *offsets
;
2293 unsigned long long nstripe
, ostripe
;
2296 if (info
->new_level
!= info
->array
.level
)
2297 return 1; /* Cannot handle level changes (they are instantaneous) */
2299 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
2300 if (info
->array
.level
== 6) odata
--; /* number of data disks */
2301 ndata
= info
->array
.raid_disks
- 1;
2302 if (info
->new_level
== 6) ndata
--;
2304 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
2306 if (info
->delta_disks
<= 0)
2307 /* Didn't grow, so the backup file must have
2311 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
2312 struct mdinfo dinfo
;
2315 char *devname
, namebuf
[20];
2317 /* This was a spare and may have some saved data on it.
2318 * Load the superblock, find and load the
2319 * backup_super_block.
2320 * If either fail, go on to next device.
2321 * If the backup contains no new info, just return
2322 * else restore data and update all superblocks
2324 if (i
== old_disks
-1) {
2325 fd
= open(backup_file
, O_RDONLY
);
2327 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
2328 backup_file
, strerror(errno
));
2331 devname
= backup_file
;
2336 if (st
->ss
->load_super(st
, fd
, NULL
))
2339 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2340 st
->ss
->free_super(st
);
2343 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
2345 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
2346 continue; /* Cannot seek */
2348 sprintf(namebuf
, "device-%d", i
);
2351 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
2353 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
2354 continue; /* Cannot read */
2356 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
2357 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
2359 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
2362 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
2364 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
2365 continue; /* bad checksum */
2367 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
2368 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
2370 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
2371 continue; /* Bad second checksum */
2373 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
2375 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
2376 continue; /* Wrong uuid */
2379 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2380 * sometimes they aren't... So allow considerable flexability in matching, and allow
2381 * this test to be overridden by an environment variable.
2383 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
2384 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
2385 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2386 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
2387 "for array with timestamp %lu\n",
2388 (unsigned long)__le64_to_cpu(bsb
.mtime
),
2389 (unsigned long)info
->array
.utime
);
2392 fprintf(stderr
, Name
": too-old timestamp on "
2393 "backup-metadata on %s\n", devname
);
2394 continue; /* time stamp is too bad */
2398 if (bsb
.magic
[15] == '1') {
2399 if (info
->delta_disks
>= 0) {
2400 /* reshape_progress is increasing */
2401 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2402 info
->reshape_progress
) {
2405 fprintf(stderr
, Name
": backup-metadata found on %s but is not needed\n", devname
);
2406 continue; /* No new data here */
2409 /* reshape_progress is decreasing */
2410 if (__le64_to_cpu(bsb
.arraystart
) >=
2411 info
->reshape_progress
)
2412 goto nonew
; /* No new data here */
2415 if (info
->delta_disks
>= 0) {
2416 /* reshape_progress is increasing */
2417 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2418 info
->reshape_progress
&&
2419 __le64_to_cpu(bsb
.arraystart2
) + __le64_to_cpu(bsb
.length2
) <
2420 info
->reshape_progress
)
2421 goto nonew
; /* No new data here */
2423 /* reshape_progress is decreasing */
2424 if (__le64_to_cpu(bsb
.arraystart
) >=
2425 info
->reshape_progress
&&
2426 __le64_to_cpu(bsb
.arraystart2
) >=
2427 info
->reshape_progress
)
2428 goto nonew
; /* No new data here */
2431 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
2434 fprintf(stderr
, Name
": Failed to verify secondary backup-metadata block on %s\n",
2436 continue; /* Cannot seek */
2438 /* There should be a duplicate backup superblock 4k before here */
2439 if (lseek64(fd
, -4096, 1) < 0 ||
2440 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
2441 goto second_fail
; /* Cannot find leading superblock */
2442 if (bsb
.magic
[15] == '1')
2443 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
2445 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
2446 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
2447 goto second_fail
; /* Cannot find leading superblock */
2449 /* Now need the data offsets for all devices. */
2450 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
2451 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
2454 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2455 /* FIXME should be this be an error */
2457 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2458 st
->ss
->free_super(st
);
2459 offsets
[j
] = dinfo
.data_offset
* 512;
2461 printf(Name
": restoring critical section\n");
2463 if (restore_stripes(fdlist
, offsets
,
2464 info
->array
.raid_disks
,
2468 fd
, __le64_to_cpu(bsb
.devstart
)*512,
2469 __le64_to_cpu(bsb
.arraystart
)*512,
2470 __le64_to_cpu(bsb
.length
)*512)) {
2471 /* didn't succeed, so giveup */
2473 fprintf(stderr
, Name
": Error restoring backup from %s\n",
2478 if (bsb
.magic
[15] == '2' &&
2479 restore_stripes(fdlist
, offsets
,
2480 info
->array
.raid_disks
,
2484 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
2485 __le64_to_cpu(bsb
.devstart2
)*512,
2486 __le64_to_cpu(bsb
.arraystart2
)*512,
2487 __le64_to_cpu(bsb
.length2
)*512)) {
2488 /* didn't succeed, so giveup */
2490 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
2496 /* Ok, so the data is restored. Let's update those superblocks. */
2498 if (info
->delta_disks
>= 0) {
2499 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
2500 __le64_to_cpu(bsb
.length
);
2501 if (bsb
.magic
[15] == '2') {
2502 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
2503 __le64_to_cpu(bsb
.length2
);
2504 if (p2
> info
->reshape_progress
)
2505 info
->reshape_progress
= p2
;
2508 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
2509 if (bsb
.magic
[15] == '2') {
2510 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
2511 if (p2
< info
->reshape_progress
)
2512 info
->reshape_progress
= p2
;
2515 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
2516 if (fdlist
[j
] < 0) continue;
2517 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2519 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2520 dinfo
.reshape_progress
= info
->reshape_progress
;
2521 st
->ss
->update_super(st
, &dinfo
,
2522 "_reshape_progress",
2524 st
->ss
->store_super(st
, fdlist
[j
]);
2525 st
->ss
->free_super(st
);
2529 /* Didn't find any backup data, try to see if any
2532 if (info
->delta_disks
< 0) {
2533 /* When shrinking, the critical section is at the end.
2534 * So see if we are before the critical section.
2536 unsigned long long first_block
;
2537 nstripe
= ostripe
= 0;
2539 while (ostripe
>= nstripe
) {
2540 ostripe
+= info
->array
.chunk_size
/ 512;
2541 first_block
= ostripe
* odata
;
2542 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
2543 (info
->new_chunk
/512);
2546 if (info
->reshape_progress
>= first_block
)
2549 if (info
->delta_disks
> 0) {
2550 /* See if we are beyond the critical section. */
2551 unsigned long long last_block
;
2552 nstripe
= ostripe
= 0;
2554 while (nstripe
>= ostripe
) {
2555 nstripe
+= info
->new_chunk
/ 512;
2556 last_block
= nstripe
* ndata
;
2557 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
2558 (info
->array
.chunk_size
/512);
2561 if (info
->reshape_progress
>= last_block
)
2564 /* needed to recover critical section! */
2566 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
2570 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
2573 /* Array is assembled and ready to be started, but
2574 * monitoring is probably required.
2577 * - set upper bound for resync
2578 * - initialise the 'suspend' boundaries
2579 * - switch to read-write
2580 * - fork and continue monitoring
2584 unsigned long long backup_offsets
[1];
2585 int odisks
, ndisks
, ochunk
, nchunk
,odata
,ndata
;
2586 unsigned long a
,b
,blocks
,stripes
;
2589 unsigned long long *offsets
;
2591 struct mdinfo
*sra
, *sd
;
2593 unsigned long cache
;
2596 err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
2600 /* make sure reshape doesn't progress until we are ready */
2601 sysfs_set_str(info
, NULL
, "sync_max", "0");
2602 sysfs_set_str(info
, NULL
, "array_state", "active"); /* FIXME or clean */
2604 sra
= sysfs_read(-1, devname2devnum(info
->sys_name
),
2605 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
2610 /* ndisks is not growing, so raid_disks is old and +delta is new */
2611 odisks
= info
->array
.raid_disks
;
2612 ndisks
= odisks
+ info
->delta_disks
;
2615 if (info
->array
.level
== 6) {
2619 ochunk
= info
->array
.chunk_size
;
2620 nchunk
= info
->new_chunk
;
2622 a
= (ochunk
/512) * odata
;
2623 b
= (nchunk
/512) * ndata
;
2631 /* LCM == product / GCD */
2632 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
2635 while (blocks
* 32 < sra
->component_size
&&
2638 stripes
= blocks
/ (info
->array
.chunk_size
/512) / odata
;
2640 /* check that the internal stripe cache is
2641 * large enough, or it won't work.
2643 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
2644 cache
= cache
* 4 / 4096;
2645 if (cache
< blocks
/ 8 / odisks
+ 16)
2646 /* Make it big enough to hold 'blocks' */
2647 cache
= blocks
/ 8 / odisks
+ 16;
2648 if (sra
->cache_size
< cache
)
2649 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
2652 memset(&bsb
, 0, 512);
2653 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2654 memcpy(&bsb
.set_uuid
, info
->uuid
, 16);
2655 bsb
.mtime
= __cpu_to_le64(time(0));
2656 bsb
.devstart2
= blocks
;
2658 backup_fd
= open(backup_file
, O_RDWR
|O_CREAT
, S_IRUSR
| S_IWUSR
);
2659 backup_list
[0] = backup_fd
;
2660 backup_offsets
[0] = 8 * 512;
2661 fds
= malloc(odisks
* sizeof(fds
[0]));
2662 offsets
= malloc(odisks
* sizeof(offsets
[0]));
2663 for (d
=0; d
<odisks
; d
++)
2666 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2667 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2669 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2670 char *dn
= map_dev(sd
->disk
.major
,
2672 fds
[sd
->disk
.raid_disk
]
2673 = dev_open(dn
, O_RDONLY
);
2674 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
2675 if (fds
[sd
->disk
.raid_disk
] < 0) {
2676 fprintf(stderr
, Name
": %s: cannot open component %s\n",
2677 info
->sys_name
, dn
?dn
:"-unknown-");
2688 mlockall(MCL_FUTURE
);
2689 if (info
->delta_disks
< 0)
2690 done
= child_shrink(-1, info
, stripes
,
2692 info
->array
.raid_disks
,
2693 info
->array
.chunk_size
,
2694 info
->array
.level
, info
->array
.layout
,
2696 1, backup_list
, backup_offsets
);
2697 else if (info
->delta_disks
== 0) {
2698 /* The 'start' is a per-device stripe number.
2699 * reshape_progress is a per-array sector number.
2700 * So divide by ndata * chunk_size
2702 unsigned long long start
= info
->reshape_progress
/ ndata
;
2703 start
/= (info
->array
.chunk_size
/512);
2704 done
= child_same_size(-1, info
, stripes
,
2707 info
->array
.raid_disks
,
2708 info
->array
.chunk_size
,
2709 info
->array
.level
, info
->array
.layout
,
2711 1, backup_list
, backup_offsets
);
2713 if (backup_file
&& done
)
2714 unlink(backup_file
);
2715 /* FIXME should I intuit a level change */
2718 fprintf(stderr
, Name
": Cannot run child to continue monitoring reshape: %s\n",