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
) {
606 flush_metadata_updates(st
);
608 st
->ss
->sync_metadata(st
);
612 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
614 /* when dealing with external metadata subarrays we need to be
615 * prepared to handle EAGAIN. The kernel may need to wait for
616 * mdmon to mark the array active so the kernel can handle
617 * allocations/writeback when preparing the reshape action
618 * (md_allow_write()). We temporarily disable safe_mode_delay
619 * to close a race with the array_state going clean before the
620 * next write to raid_disks / stripe_cache_size
625 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
626 if (strcmp(name
, "raid_disks") != 0 &&
627 strcmp(name
, "stripe_cache_size") != 0)
628 return sysfs_set_num(sra
, NULL
, name
, n
);
630 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
633 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
634 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
635 if (rc
< 0 && errno
== EAGAIN
) {
636 ping_monitor(container
);
637 /* if we get EAGAIN here then the monitor is not active
640 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
642 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
646 static int reshape_container_raid_disks(char *container
, int raid_disks
)
648 /* for each subarray switch to a raid level that can
649 * support the reshape, and set raid disks
651 struct mdstat_ent
*ent
, *e
;
652 int changed
= 0, rv
= 0, err
= 0;
654 ent
= mdstat_read(1, 0);
656 fprintf(stderr
, Name
": unable to read /proc/mdstat\n");
661 for (e
= ent
; e
; e
= e
->next
) {
664 int level
, takeover_delta
= 0;
666 if (!is_container_member(e
, container
))
669 level
= map_name(pers
, e
->level
);
671 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
674 /* metadata records 'orig_level' */
675 rv
= sysfs_set_num(sub
, NULL
, "level", 4);
680 /* we want spares to be used for capacity
681 * expansion, not rebuild
697 sub
= sysfs_read(-1, e
->devnum
, GET_CHUNK
|GET_CACHE
);
700 cache
= (sub
->array
.chunk_size
/ 4096) * 4;
701 if (cache
> sub
->cache_size
)
702 rv
= subarray_set_num(container
, sub
,
703 "stripe_cache_size", cache
);
711 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
715 rv
= subarray_set_num(container
, sub
, "raid_disks",
716 raid_disks
+ takeover_delta
);
730 ": failed to initiate container reshape%s%s\n",
731 err
? ": " : "", err
? strerror(err
) : "");
738 static void revert_container_raid_disks(struct supertype
*st
, int fd
, char *container
)
740 /* we failed to prepare all subarrays in the container for
741 * reshape, so cancel the changes and restore the nominal raid
744 struct mdstat_ent
*ent
, *e
;
746 ent
= mdstat_read(0, 0);
749 ": failed to read /proc/mdstat while aborting reshape\n");
753 if (st
->ss
->load_container(st
, fd
, NULL
)) {
755 ": failed read metadata while aborting reshape\n");
760 for (e
= ent
; e
; e
= e
->next
) {
761 int level_fixed
= 0, disks_fixed
= 0;
762 struct mdinfo
*sub
, *prev
;
765 if (!is_container_member(e
, container
))
768 subarray
= to_subarray(e
, container
);
769 prev
= st
->ss
->container_content(st
, subarray
);
771 /* changing level might change raid_disks so we do it
772 * first and then check if raid_disks still needs fixing
774 if (map_name(pers
, e
->level
) != prev
->array
.level
) {
775 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
777 !sysfs_set_num(sub
, NULL
, "level", prev
->array
.level
))
783 sub
= sysfs_read(-1, e
->devnum
, GET_DISKS
);
784 if (sub
&& sub
->array
.raid_disks
!= prev
->array
.raid_disks
) {
785 if (!subarray_set_num(container
, sub
, "raid_disks",
786 prev
->array
.raid_disks
))
792 if (!disks_fixed
|| !level_fixed
)
794 ": failed to restore %s to a %d-disk %s array\n",
795 e
->dev
, prev
->array
.raid_disks
,
796 map_num(pers
, prev
->array
.level
));
799 st
->ss
->free_super(st
);
803 int remove_disks_on_raid10_to_raid0_takeover(struct supertype
*st
,
808 struct mdinfo
*remaining
;
811 nr_of_copies
= layout
& 0xff;
813 remaining
= sra
->devs
;
815 /* for each 'copy', select one device and remove from the list. */
816 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
817 struct mdinfo
**diskp
;
820 /* Find a working device to keep */
821 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
822 struct mdinfo
*disk
= *diskp
;
824 if (disk
->disk
.raid_disk
< slot
)
826 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
828 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
830 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
832 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
835 /* We have found a good disk to use! */
837 disk
->next
= sra
->devs
;
846 if (slot
< sra
->array
.raid_disks
) {
847 /* didn't find all slots */
853 sra
->devs
= remaining
;
857 /* Remove all 'remaining' devices from the array */
859 struct mdinfo
*sd
= remaining
;
860 remaining
= sd
->next
;
862 sysfs_set_str(sra
, sd
, "state", "faulty");
863 sysfs_set_str(sra
, sd
, "slot", "none");
864 sysfs_set_str(sra
, sd
, "state", "remove");
865 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
866 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
867 sd
->next
= sra
->devs
;
873 void reshape_free_fdlist(int *fdlist
,
874 unsigned long long *offsets
,
879 for (i
= 0; i
< size
; i
++)
887 int reshape_prepare_fdlist(char *devname
,
891 unsigned long blocks
,
894 unsigned long long *offsets
)
899 for (d
= 0; d
<= nrdisks
; d
++)
902 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
903 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
905 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
906 char *dn
= map_dev(sd
->disk
.major
,
908 fdlist
[sd
->disk
.raid_disk
]
909 = dev_open(dn
, O_RDONLY
);
910 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
911 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
913 Name
": %s: cannot open component %s\n",
914 devname
, dn
? dn
: "-unknown-");
918 } else if (backup_file
== NULL
) {
920 char *dn
= map_dev(sd
->disk
.major
,
922 fdlist
[d
] = dev_open(dn
, O_RDWR
);
923 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
925 fprintf(stderr
, Name
": %s: cannot open component %s\n",
926 devname
, dn
? dn
: "-unknown-");
937 int reshape_open_backup_file(char *backup_file
,
942 unsigned long long *offsets
)
944 /* Return 1 on success, 0 on any form of failure */
945 /* need to check backup file is large enough */
951 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|O_EXCL
,
955 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
956 devname
, backup_file
, strerror(errno
));
959 /* Guard against backup file being on array device.
960 * If array is partitioned or if LVM etc is in the
961 * way this will not notice, but it is better than
964 fstat(*fdlist
, &stb
);
967 if (stb
.st_rdev
== dev
) {
968 fprintf(stderr
, Name
": backup file must NOT be"
969 " on the array being reshaped.\n");
975 for (i
=0; i
< blocks
+ 1 ; i
++) {
976 if (write(*fdlist
, buf
, 512) != 512) {
977 fprintf(stderr
, Name
": %s: cannot create"
978 " backup file %s: %s\n",
979 devname
, backup_file
, strerror(errno
));
983 if (fsync(*fdlist
) != 0) {
984 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
985 devname
, backup_file
, strerror(errno
));
992 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
993 unsigned int ndata
, unsigned int odata
)
995 unsigned long a
, b
, blocks
;
996 /* So how much do we need to backup.
997 * We need an amount of data which is both a whole number of
998 * old stripes and a whole number of new stripes.
999 * So LCM for (chunksize*datadisks).
1001 a
= (ochunk
/512) * odata
;
1002 b
= (nchunk
/512) * ndata
;
1010 /* LCM == product / GCD */
1011 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
1017 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1019 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1022 /* Make some changes in the shape of an array.
1023 * The kernel must support the change.
1025 * There are three different changes. Each can trigger
1026 * a resync or recovery so we freeze that until we have
1027 * requested everything (if kernel supports freezing - 2.6.30).
1029 * - change size (i.e. component_size)
1031 * - change layout/chunksize/ndisks
1033 * The last can require a reshape. It is different on different
1034 * levels so we need to check the level before actioning it.
1035 * Some times the level change needs to be requested after the
1036 * reshape (e.g. raid6->raid5, raid5->raid0)
1039 struct mdu_array_info_s array
, orig
;
1042 struct supertype
*st
;
1043 char *subarray
= NULL
;
1046 int nlayout
, olayout
;
1048 unsigned int ndata
, odata
;
1049 int orig_level
= UnSet
;
1050 char alt_layout
[40];
1052 unsigned long long *offsets
;
1057 unsigned long blocks
, stripes
;
1058 unsigned long cache
;
1059 unsigned long long array_size
;
1061 char *container
= NULL
;
1068 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1069 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1075 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1076 fprintf(stderr
, Name
": cannot change component size at the same time "
1077 "as other changes.\n"
1078 " Change size first, then check data is intact before "
1079 "making other changes.\n");
1083 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1084 get_linux_version() < 2006032 &&
1085 !check_env("MDADM_FORCE_FEWER")) {
1086 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1087 " Please use a newer kernel\n");
1091 st
= super_by_fd(fd
, &subarray
);
1093 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1096 if (raid_disks
> st
->max_devs
) {
1097 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1098 " beyond %d\n", st
->max_devs
);
1102 /* in the external case we need to check that the requested reshape is
1103 * supported, and perform an initial check that the container holds the
1104 * pre-requisite spare devices (mdmon owns final validation)
1106 if (st
->ss
->external
) {
1111 container_dev
= st
->container_dev
;
1112 cfd
= open_dev_excl(st
->container_dev
);
1113 } else if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 ||
1116 Name
": %s is a container, only 'raid-devices' can be changed\n",
1120 container_dev
= st
->devnum
;
1122 cfd
= open_dev_excl(st
->devnum
);
1126 fprintf(stderr
, Name
": Unable to open container for %s\n",
1132 container
= devnum2devname(st
->devnum
);
1134 fprintf(stderr
, Name
": Could not determine container name\n");
1140 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1142 rv
= st
->ss
->load_super(st
, cfd
, NULL
);
1144 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1150 if (mdmon_running(container_dev
))
1151 st
->update_tail
= &st
->updates
;
1154 if (raid_disks
> array
.raid_disks
&&
1155 array
.spare_disks
< (raid_disks
- array
.raid_disks
) &&
1158 Name
": Need %d spare%s to avoid degraded array,"
1159 " and only have %d.\n"
1160 " Use --force to over-ride this check.\n",
1161 raid_disks
- array
.raid_disks
,
1162 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1167 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
| GET_STATE
);
1169 if (st
->ss
->external
&& subarray
== NULL
) {
1170 array
.level
= LEVEL_CONTAINER
;
1171 sra
->array
.level
= LEVEL_CONTAINER
;
1174 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1178 frozen
= freeze(st
);
1180 /* freeze() already spewed the reason */
1182 } else if (frozen
< 0) {
1183 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1184 " be reshaped\n", devname
);
1188 /* ========= set size =============== */
1189 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1190 long long orig_size
= array
.size
;
1192 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1198 if (array
.size
!= size
) {
1199 /* got truncated to 32bit, write to
1200 * component_size instead
1203 rv
= sysfs_set_num(sra
, NULL
,
1204 "component_size", size
);
1208 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1212 /* restore metadata */
1213 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1214 NULL
, devname
, !quiet
) == 0)
1216 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1217 devname
, strerror(err
));
1219 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1220 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1224 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1225 size
= get_component_size(fd
)/2;
1229 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
1232 } else if (array
.level
!= LEVEL_CONTAINER
) {
1233 size
= get_component_size(fd
)/2;
1238 /* ========= check for Raid10 -> Raid0 conversion ===============
1239 * current implemenation assumes that following conditions must be met:
1241 * - near_copies == 2
1243 if (level
== 0 && array
.level
== 10 &&
1244 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) {
1246 err
= remove_disks_on_raid10_to_raid0_takeover(st
, sra
, array
.layout
);
1248 dprintf(Name
": Array cannot be reshaped\n");
1257 /* ======= set level =========== */
1258 if (level
!= UnSet
&& level
!= array
.level
) {
1259 /* Trying to change the level.
1260 * We might need to change layout first and schedule a
1261 * level change for later.
1262 * Level changes that can happen immediately are:
1263 * 0->4,5,6 1->5 4->5,6 5->1,6
1264 * Level changes that need a layout change first are:
1265 * 6->5,4,0 : need a -6 layout, or parity-last
1266 * 5->4,0 : need parity-last
1268 if ((array
.level
== 6 || array
.level
== 5) &&
1269 (level
== 5 || level
== 4 || level
== 0)) {
1270 /* Don't change level yet, but choose intermediate
1274 if (layout_str
== NULL
)
1275 switch (array
.layout
) {
1276 case ALGORITHM_LEFT_ASYMMETRIC
:
1277 case ALGORITHM_LEFT_ASYMMETRIC_6
:
1278 case ALGORITHM_ROTATING_N_RESTART
:
1279 layout_str
= "left-asymmetric-6";
1281 case ALGORITHM_LEFT_SYMMETRIC
:
1282 case ALGORITHM_LEFT_SYMMETRIC_6
:
1283 case ALGORITHM_ROTATING_N_CONTINUE
:
1284 layout_str
= "left-symmetric-6";
1286 case ALGORITHM_RIGHT_ASYMMETRIC
:
1287 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
1288 case ALGORITHM_ROTATING_ZERO_RESTART
:
1289 layout_str
= "right-asymmetric-6";
1291 case ALGORITHM_RIGHT_SYMMETRIC
:
1292 case ALGORITHM_RIGHT_SYMMETRIC_6
:
1293 layout_str
= "right-symmetric-6";
1295 case ALGORITHM_PARITY_0
:
1296 case ALGORITHM_PARITY_0_6
:
1297 layout_str
= "parity-first-6";
1299 case ALGORITHM_PARITY_N
:
1300 layout_str
= "parity-last";
1303 fprintf(stderr
, Name
": %s: cannot"
1304 "convert layout to RAID5 equivalent\n",
1310 int l
= map_name(r5layout
, layout_str
);
1312 fprintf(stderr
, Name
": %s: layout '%s' not recognised\n",
1313 devname
, layout_str
);
1317 if (l
!= ALGORITHM_PARITY_N
) {
1318 /* need the -6 version */
1319 char *ls
= map_num(r5layout
, l
);
1320 strcat(strcpy(alt_layout
, ls
),
1322 layout_str
= alt_layout
;
1326 /* The final raid6->raid5 conversion
1327 * will reduce the number of disks,
1328 * so now we need to aim higher
1332 layout_str
= "parity-last";
1334 /* Level change is a simple takeover. In the external
1335 * case we don't check with the metadata handler until
1336 * we establish what the final layout will be. If the
1337 * level change is disallowed we will revert to
1338 * orig_level without disturbing the metadata, otherwise
1339 * we will send an update.
1341 c
= map_num(pers
, level
);
1343 rv
= 1;/* not possible */
1347 /* Need to check there are enough spares */
1348 int spares_needed
= 0;
1349 switch (array
.level
* 16 + level
) {
1351 spares_needed
= 1; break;
1353 spares_needed
= 2; break;
1355 spares_needed
= 1; break;
1357 spares_needed
= 2; break;
1359 spares_needed
= 1; break;
1361 if (raid_disks
> array
.raid_disks
)
1362 spares_needed
+= raid_disks
-array
.raid_disks
;
1363 if (spares_needed
> array
.spare_disks
) {
1365 Name
": Need %d spare%s to avoid"
1366 " degraded array, and only have %d.\n"
1367 " Use --force to over-ride"
1370 spares_needed
== 1 ? "" : "s",
1376 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1379 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1382 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1383 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1388 orig_level
= orig
.level
;
1389 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1390 if (layout_str
== NULL
&&
1391 orig
.level
== 5 && level
== 6 &&
1392 array
.layout
!= orig
.layout
)
1393 layout_str
= map_num(r5layout
, orig
.layout
);
1395 fprintf(stderr
, Name
" level of %s changed to %s\n",
1401 /* ========= set shape (chunk_size / layout / ndisks) ============== */
1402 /* Check if layout change is a no-op */
1403 switch (array
.level
) {
1405 if (layout_str
&& array
.layout
== map_name(r5layout
, layout_str
))
1409 if (layout_str
== NULL
&&
1410 ((chunksize
&& chunksize
* 1024 != array
.chunk_size
) ||
1411 (raid_disks
&& raid_disks
!= array
.raid_disks
)) &&
1412 array
.layout
>= 16) {
1413 fprintf(stderr
, Name
1414 ": %s has a non-standard layout. If you wish to preserve this\n"
1415 " during the reshape, please specify --layout=preserve\n"
1416 " If you want to change it, specify a layout or use --layout=normalise\n",
1422 (strcmp(layout_str
, "normalise") == 0 ||
1423 strcmp(layout_str
, "normalize") == 0)) {
1425 strcpy(alt_layout
, map_num(r6layout
, array
.layout
));
1426 hyphen
= strrchr(alt_layout
, '-');
1427 if (hyphen
&& strcmp(hyphen
, "-6") == 0) {
1429 layout_str
= alt_layout
;
1433 if (layout_str
&& array
.layout
== map_name(r6layout
, layout_str
))
1435 if (layout_str
&& strcmp(layout_str
, "preserve") == 0)
1439 if (layout_str
== NULL
1440 && (chunksize
== 0 || chunksize
*1024 == array
.chunk_size
)
1441 && (raid_disks
== 0 || raid_disks
== array
.raid_disks
)) {
1442 if (reshape_super(st
, -1, level
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1448 if (level
!= UnSet
&& level
!= array
.level
) {
1449 /* Looks like this level change doesn't need
1450 * a reshape after all.
1452 c
= map_num(pers
, level
);
1454 rv
= sysfs_set_str(sra
, NULL
, "level", c
);
1457 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1460 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1461 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1465 } else if (!changed
&& !quiet
)
1466 fprintf(stderr
, Name
": %s: no change requested\n",
1469 if (st
->ss
->external
&& !mdmon_running(st
->container_dev
) &&
1471 start_mdmon(st
->container_dev
);
1472 ping_monitor(container
);
1477 c
= map_num(pers
, array
.level
);
1478 if (c
== NULL
) c
= "-unknown-";
1479 switch (array
.level
) {
1480 default: /* raid0, linear, multipath cannot be reconfigured */
1481 fprintf(stderr
, Name
": %s array %s cannot be reshaped.\n",
1483 /* TODO raid0 raiddisks can be reshaped via raid4 */
1486 case LEVEL_CONTAINER
: {
1489 /* double check that we are not changing anything but raid_disks */
1490 if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 || level
!= UnSet
) {
1492 Name
": %s is a container, only 'raid-devices' can be changed\n",
1498 st
->update_tail
= &st
->updates
;
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");
1517 if (!mdmon_running(st
->devnum
)) {
1518 start_mdmon(st
->devnum
);
1519 ping_monitor(container
);
1523 /* give mdmon a chance to allocate spares */
1524 ping_manager(container
);
1526 /* manage_reshape takes care of releasing the array(s) */
1527 st
->ss
->manage_reshape(st
, backup_file
);
1531 case LEVEL_FAULTY
: /* only 'layout' change is permitted */
1533 if (chunksize
|| raid_disks
) {
1534 fprintf(stderr
, Name
": %s: Cannot change chunksize or disks of a 'faulty' array\n",
1539 if (layout_str
== NULL
)
1540 break; /* nothing to do.... */
1542 array
.layout
= parse_layout_faulty(layout_str
);
1543 if (array
.layout
< 0) {
1544 fprintf(stderr
, Name
": %s: layout %s not understood for 'faulty' array\n",
1545 devname
, layout_str
);
1549 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1550 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
1551 devname
, strerror(errno
));
1554 printf("layout for %s set to %d\n", devname
, array
.layout
);
1557 case 1: /* only raid_disks can each be changed. */
1559 if (chunksize
|| layout_str
!= NULL
) {
1560 fprintf(stderr
, Name
": %s: Cannot change chunk size or layout for a RAID1 array.\n",
1565 if (raid_disks
> 0) {
1566 if (reshape_super(st
, -1, UnSet
, UnSet
, 0, raid_disks
,
1567 NULL
, devname
, !quiet
)) {
1572 array
.raid_disks
= raid_disks
;
1573 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1574 fprintf(stderr
, Name
": Cannot set raid-devices for %s: %s\n",
1575 devname
, strerror(errno
));
1586 * layout/chunksize/raid_disks can be changed
1587 * though the kernel may not support it all.
1590 fprintf(stderr
, Name
": Cannot reshape subarrays yet\n");
1595 * There are three possibilities.
1596 * 1/ The array will shrink.
1597 * We need to ensure the reshape will pause before reaching
1598 * the 'critical section'. We also need to fork and wait for
1599 * that to happen. When it does we
1600 * suspend/backup/complete/unfreeze
1602 * 2/ The array will not change size.
1603 * This requires that we keep a backup of a sliding window
1604 * so that we can restore data after a crash. So we need
1605 * to fork and monitor progress.
1607 * 3/ The array will grow. This is relatively easy.
1608 * However the kernel's restripe routines will cheerfully
1609 * overwrite some early data before it is safe. So we
1610 * need to make a backup of the early parts of the array
1611 * and be ready to restore it if rebuild aborts very early.
1613 * We backup data by writing it to one spare, or to a
1614 * file which was given on command line.
1616 * [FOLLOWING IS OLD AND PARTLY WRONG]
1617 * So: we enumerate the devices in the array and
1618 * make sure we can open all of them.
1619 * Then we freeze the early part of the array and
1620 * backup to the various spares.
1621 * Then we request changes and start the reshape.
1622 * Monitor progress until it has passed the danger zone.
1623 * and finally invalidate the copied data and unfreeze the
1624 * start of the array.
1626 * In each case, we first make sure that storage is available
1627 * for the required backup.
1629 * - request the shape change.
1630 * - for to handle backup etc.
1632 nchunk
= ochunk
= array
.chunk_size
;
1633 nlayout
= olayout
= array
.layout
;
1634 ndisks
= odisks
= array
.raid_disks
;
1637 nchunk
= chunksize
* 1024;
1638 if (size
% chunksize
) {
1639 fprintf(stderr
, Name
": component size %lluK is not"
1640 " a multiple of chunksize %dK\n",
1645 if (layout_str
!= NULL
)
1646 switch(array
.level
) {
1647 case 4: /* ignore layout */
1650 nlayout
= map_name(r5layout
, layout_str
);
1651 if (nlayout
== UnSet
) {
1652 fprintf(stderr
, Name
": layout %s not understood for raid5.\n",
1660 nlayout
= map_name(r6layout
, layout_str
);
1661 if (nlayout
== UnSet
) {
1662 fprintf(stderr
, Name
": layout %s not understood for raid6.\n",
1669 if (raid_disks
) ndisks
= raid_disks
;
1673 if (array
.level
== 6) {
1674 odata
--; /* number of data disks */
1678 if (odata
== ndata
&&
1679 get_linux_version() < 2006032) {
1680 fprintf(stderr
, Name
": in-place reshape is not safe before 2.6.32, sorry.\n");
1684 /* Check that we can hold all the data */
1685 get_dev_size(fd
, NULL
, &array_size
);
1686 if (ndata
* (unsigned long long)size
< (array_size
/1024)) {
1687 fprintf(stderr
, Name
": this change will reduce the size of the array.\n"
1688 " use --grow --array-size first to truncate array.\n"
1689 " e.g. mdadm --grow %s --array-size %llu\n",
1690 devname
, ndata
* size
);
1695 blocks
= compute_backup_blocks(nchunk
, ochunk
, ndata
, odata
);
1698 sra
= sysfs_read(fd
, 0,
1699 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
1703 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
1709 if (ndata
== odata
) {
1710 /* Make 'blocks' bigger for better throughput, but
1711 * not so big that we reject it below.
1712 * Try for 16 megabytes
1714 while (blocks
* 32 < sra
->component_size
&&
1718 fprintf(stderr
, Name
": Need to backup %luK of critical "
1719 "section..\n", blocks
/2);
1721 if (blocks
>= sra
->component_size
/2) {
1722 fprintf(stderr
, Name
": %s: Something wrong - reshape aborted\n",
1727 nrdisks
= array
.raid_disks
+ sra
->array
.spare_disks
;
1728 /* Now we need to open all these devices so we can read/write.
1730 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1731 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1732 if (!fdlist
|| !offsets
) {
1733 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1738 d
= reshape_prepare_fdlist(devname
, sra
, array
.raid_disks
,
1739 nrdisks
, blocks
, backup_file
,
1745 if (backup_file
== NULL
) {
1746 if (st
->ss
->external
&& !st
->ss
->manage_reshape
) {
1747 fprintf(stderr
, Name
": %s Grow operation not supported by %s metadata\n",
1748 devname
, st
->ss
->name
);
1752 if (ndata
<= odata
) {
1753 fprintf(stderr
, Name
": %s: Cannot grow - need backup-file\n",
1757 } else if (sra
->array
.spare_disks
== 0) {
1758 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1759 "backup-file to backup critical section\n",
1764 if (d
== array
.raid_disks
) {
1765 fprintf(stderr
, Name
": %s: No spare device for backup\n",
1771 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
1773 fdlist
+d
, offsets
+d
)) {
1780 /* check that the operation is supported by the metadata */
1781 if (reshape_super(st
, -1, level
, nlayout
, nchunk
, ndisks
,
1782 backup_file
, devname
, !quiet
)) {
1787 /* lastly, check that the internal stripe cache is
1788 * large enough, or it won't work.
1791 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
1792 cache
= cache
* 4 / 4096;
1793 if (cache
< blocks
/ 8 / odisks
+ 16)
1794 /* Make it big enough to hold 'blocks' */
1795 cache
= blocks
/ 8 / odisks
+ 16;
1796 if (sra
->cache_size
< cache
)
1797 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
1799 /* Right, everything seems fine. Let's kick things off.
1800 * If only changing raid_disks, use ioctl, else use
1804 if (ochunk
== nchunk
&& olayout
== nlayout
) {
1805 array
.raid_disks
= ndisks
;
1806 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1809 fprintf(stderr
, Name
": Cannot set device shape for %s: %s\n",
1810 devname
, strerror(errno
));
1811 if (ndisks
< odisks
&&
1812 get_linux_version() < 2006030)
1813 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1815 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1816 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1821 /* set them all just in case some old 'new_*' value
1822 * persists from some earlier problem
1824 int err
= err
; /* only used if rv==1, and always set if
1825 * rv==1, so initialisation not needed,
1826 * despite gcc warning
1828 if (sysfs_set_num(sra
, NULL
, "chunk_size", nchunk
) < 0)
1829 rv
= 1, err
= errno
;
1830 if (!rv
&& sysfs_set_num(sra
, NULL
, "layout", nlayout
) < 0)
1831 rv
= 1, err
= errno
;
1832 if (!rv
&& sysfs_set_num(sra
, NULL
, "raid_disks", ndisks
) < 0)
1833 rv
= 1, err
= errno
;
1835 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1837 if (get_linux_version() < 2006030)
1838 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1840 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1841 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1846 if (ndisks
== 2 && odisks
== 2) {
1847 /* No reshape is needed in this trivial case */
1852 if (st
->ss
->external
) {
1853 /* metadata handler takes it from here */
1854 ping_manager(container
);
1855 st
->ss
->manage_reshape(st
, backup_file
);
1860 /* set up the backup-super-block. This requires the
1861 * uuid from the array.
1863 /* Find a superblock */
1864 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1868 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1870 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1871 devfd
= dev_open(dn
, O_RDONLY
);
1874 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
1880 fprintf(stderr
, Name
": %s: Cannot find a superblock\n",
1886 memset(&bsb
, 0, 512);
1887 memcpy(bsb
.magic
, "md_backup_data-1", 16);
1888 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
1889 bsb
.mtime
= __cpu_to_le64(time(0));
1890 bsb
.devstart2
= blocks
;
1891 stripes
= blocks
/ (ochunk
/512) / odata
;
1892 /* Now we just need to kick off the reshape and watch, while
1893 * handling backups of the data...
1894 * This is all done by a forked background process.
1899 if (check_env("MDADM_GROW_VERIFY"))
1900 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
1903 mlockall(MCL_FUTURE
);
1906 done
= child_grow(fd
, sra
, stripes
,
1908 odisks
, ochunk
, array
.level
, olayout
, odata
,
1909 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1910 else if (odata
> ndata
)
1911 done
= child_shrink(fd
, sra
, stripes
,
1913 odisks
, ochunk
, array
.level
, olayout
, odata
,
1914 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1916 done
= child_same_size(fd
, sra
, stripes
,
1919 odisks
, ochunk
, array
.level
, olayout
, odata
,
1920 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1921 if (backup_file
&& done
)
1922 unlink(backup_file
);
1923 if (level
!= UnSet
&& level
!= array
.level
) {
1924 /* We need to wait for the reshape to finish
1925 * (which will have happened unless odata < ndata)
1926 * and then set the level
1929 c
= map_num(pers
, level
);
1931 exit(0);/* not possible */
1935 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1937 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1942 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
1947 /* The child will take care of unfreezing the array */
1956 if (rv
&& orig_level
!= UnSet
&& sra
) {
1957 c
= map_num(pers
, orig_level
);
1958 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
1959 fprintf(stderr
, Name
": aborting level change\n");
1961 unfreeze(st
, frozen
);
1966 * We run a child process in the background which performs the following
1968 * - wait for resync to reach a certain point
1969 * - suspend io to the following section
1970 * - backup that section
1971 * - allow resync to proceed further
1973 * - discard the backup.
1975 * When are combined in slightly different ways in the three cases.
1977 * - suspend/backup/allow/wait/resume/discard
1979 * - allow/wait/suspend/backup/allow/wait/resume/discard
1981 * - wait/resume/discard/suspend/backup/allow
1983 * suspend/backup/allow always come together
1984 * wait/resume/discard do too.
1985 * For the same-size case we have two backups to improve flow.
1989 /* FIXME return status is never checked */
1990 static int grow_backup(struct mdinfo
*sra
,
1991 unsigned long long offset
, /* per device */
1992 unsigned long stripes
, /* per device */
1993 int *sources
, unsigned long long *offsets
,
1994 int disks
, int chunk
, int level
, int layout
,
1995 int dests
, int *destfd
, unsigned long long *destoffsets
,
1996 int part
, int *degraded
,
1999 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2000 * to storage 'destfd' (offset 'destoffsets'), after first
2001 * suspending IO. Then allow resync to continue
2002 * over the suspended section.
2003 * Use part 'part' of the backup-super-block.
2008 unsigned long long ll
;
2010 //printf("offset %llu\n", offset);
2015 sysfs_set_num(sra
, NULL
, "suspend_hi", (offset
+ stripes
* (chunk
/512)) * odata
);
2016 /* Check that array hasn't become degraded, else we might backup the wrong data */
2017 sysfs_get_ll(sra
, NULL
, "degraded", &ll
);
2018 new_degraded
= (int)ll
;
2019 if (new_degraded
!= *degraded
) {
2020 /* check each device to ensure it is still working */
2022 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2023 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2025 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2027 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2028 strstr(sbuf
, "faulty") ||
2029 strstr(sbuf
, "in_sync") == NULL
) {
2030 /* this device is dead */
2031 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2032 if (sd
->disk
.raid_disk
>= 0 &&
2033 sources
[sd
->disk
.raid_disk
] >= 0) {
2034 close(sources
[sd
->disk
.raid_disk
]);
2035 sources
[sd
->disk
.raid_disk
] = -1;
2040 *degraded
= new_degraded
;
2043 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2044 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2046 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2047 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2050 bsb
.magic
[15] = '2';
2051 for (i
= 0; i
< dests
; i
++)
2053 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
2055 lseek64(destfd
[i
], destoffsets
[i
], 0);
2057 rv
= save_stripes(sources
, offsets
,
2058 disks
, chunk
, level
, layout
,
2060 offset
*512*odata
, stripes
* chunk
* odata
,
2065 bsb
.mtime
= __cpu_to_le64(time(0));
2066 for (i
= 0; i
< dests
; i
++) {
2067 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2069 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2070 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2071 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2072 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2075 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
2076 != destoffsets
[i
] - 4096)
2078 if (write(destfd
[i
], &bsb
, 512) != 512)
2080 if (destoffsets
[i
] > 4096) {
2081 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
2082 destoffsets
[i
]+stripes
*chunk
*odata
)
2084 if (write(destfd
[i
], &bsb
, 512) != 512)
2094 /* in 2.6.30, the value reported by sync_completed can be
2095 * less that it should be by one stripe.
2096 * This only happens when reshape hits sync_max and pauses.
2097 * So allow wait_backup to either extent sync_max further
2098 * than strictly necessary, or return before the
2099 * sync has got quite as far as we would really like.
2100 * This is what 'blocks2' is for.
2101 * The various caller give appropriate values so that
2104 /* FIXME return value is often ignored */
2105 static int wait_backup(struct mdinfo
*sra
,
2106 unsigned long long offset
, /* per device */
2107 unsigned long long blocks
, /* per device */
2108 unsigned long long blocks2
, /* per device - hack */
2109 int dests
, int *destfd
, unsigned long long *destoffsets
,
2112 /* Wait for resync to pass the section that was backed up
2113 * then erase the backup and allow IO
2115 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
2116 unsigned long long completed
;
2122 sysfs_set_num(sra
, NULL
, "sync_max", offset
+ blocks
+ blocks2
);
2124 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
2126 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2130 while (completed
< offset
+ blocks
) {
2135 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2136 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
2140 if (sysfs_get_str(sra
, NULL
, "sync_action",
2142 strncmp(action
, "reshape", 7) != 0)
2148 bsb
.arraystart2
= __cpu_to_le64(0);
2149 bsb
.length2
= __cpu_to_le64(0);
2151 bsb
.arraystart
= __cpu_to_le64(0);
2152 bsb
.length
= __cpu_to_le64(0);
2154 bsb
.mtime
= __cpu_to_le64(time(0));
2156 for (i
= 0; i
< dests
; i
++) {
2157 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2158 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2159 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2160 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2161 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2162 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
2163 destoffsets
[i
]-4096)
2166 write(destfd
[i
], &bsb
, 512) != 512)
2173 static void fail(char *msg
)
2176 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
2177 rv
|= (write(2, "\n", 1) != 1);
2181 static char *abuf
, *bbuf
;
2182 static unsigned long long abuflen
;
2183 static void validate(int afd
, int bfd
, unsigned long long offset
)
2185 /* check that the data in the backup against the array.
2186 * This is only used for regression testing and should not
2187 * be used while the array is active
2191 lseek64(bfd
, offset
- 4096, 0);
2192 if (read(bfd
, &bsb2
, 512) != 512)
2193 fail("cannot read bsb");
2194 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
2195 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
2196 fail("first csum bad");
2197 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
2198 fail("magic is bad");
2199 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
2200 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
2201 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
2202 fail("second csum bad");
2204 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
2205 fail("devstart is wrong");
2208 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
2210 if (abuflen
< len
) {
2214 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
2215 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
2217 /* just stop validating on mem-alloc failure */
2222 lseek64(bfd
, offset
, 0);
2223 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
2224 //printf("len %llu\n", len);
2225 fail("read first backup failed");
2227 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
2228 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2229 fail("read first from array failed");
2230 if (memcmp(bbuf
, abuf
, len
) != 0) {
2233 printf("offset=%llu len=%llu\n",
2234 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
2235 for (i
=0; i
<len
; i
++)
2236 if (bbuf
[i
] != abuf
[i
]) {
2237 printf("first diff byte %d\n", i
);
2241 fail("data1 compare failed");
2245 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
2247 if (abuflen
< len
) {
2251 abuf
= malloc(abuflen
);
2252 bbuf
= malloc(abuflen
);
2255 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2256 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2257 fail("read second backup failed");
2258 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2259 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2260 fail("read second from array failed");
2261 if (memcmp(bbuf
, abuf
, len
) != 0)
2262 fail("data2 compare failed");
2266 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2267 int *fds
, unsigned long long *offsets
,
2268 int disks
, int chunk
, int level
, int layout
, int data
,
2269 int dests
, int *destfd
, unsigned long long *destoffsets
)
2274 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2275 /* Don't start the 'reshape' */
2277 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2278 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2279 grow_backup(sra
, 0, stripes
,
2280 fds
, offsets
, disks
, chunk
, level
, layout
,
2281 dests
, destfd
, destoffsets
,
2283 validate(afd
, destfd
[0], destoffsets
[0]);
2284 wait_backup(sra
, 0, stripes
* (chunk
/ 512), stripes
* (chunk
/ 512),
2285 dests
, destfd
, destoffsets
,
2287 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2289 /* FIXME this should probably be numeric */
2290 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2294 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2295 int *fds
, unsigned long long *offsets
,
2296 int disks
, int chunk
, int level
, int layout
, int data
,
2297 int dests
, int *destfd
, unsigned long long *destoffsets
)
2300 unsigned long long start
;
2304 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2306 start
= sra
->component_size
- stripes
* (chunk
/512);
2307 sysfs_set_num(sra
, NULL
, "sync_max", start
);
2308 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
2309 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2310 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2311 rv
= wait_backup(sra
, 0, start
- stripes
* (chunk
/512), stripes
* (chunk
/512),
2312 dests
, destfd
, destoffsets
, 0);
2315 grow_backup(sra
, 0, stripes
,
2317 disks
, chunk
, level
, layout
,
2318 dests
, destfd
, destoffsets
,
2320 validate(afd
, destfd
[0], destoffsets
[0]);
2321 wait_backup(sra
, start
, stripes
*(chunk
/512), 0,
2322 dests
, destfd
, destoffsets
, 0);
2323 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2325 /* FIXME this should probably be numeric */
2326 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2330 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2331 int *fds
, unsigned long long *offsets
,
2332 unsigned long long start
,
2333 int disks
, int chunk
, int level
, int layout
, int data
,
2334 int dests
, int *destfd
, unsigned long long *destoffsets
)
2336 unsigned long long size
;
2337 unsigned long tailstripes
= stripes
;
2340 unsigned long long speed
;
2344 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2347 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2348 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2350 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2351 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2353 grow_backup(sra
, start
, stripes
,
2355 disks
, chunk
, level
, layout
,
2356 dests
, destfd
, destoffsets
,
2358 grow_backup(sra
, (start
+ stripes
) * (chunk
/512), stripes
,
2360 disks
, chunk
, level
, layout
,
2361 dests
, destfd
, destoffsets
,
2363 validate(afd
, destfd
[0], destoffsets
[0]);
2365 start
+= stripes
* 2; /* where to read next */
2366 size
= sra
->component_size
/ (chunk
/512);
2367 while (start
< size
) {
2368 if (wait_backup(sra
, (start
-stripes
*2)*(chunk
/512),
2369 stripes
*(chunk
/512), 0,
2370 dests
, destfd
, destoffsets
,
2373 sysfs_set_num(sra
, NULL
, "suspend_lo", start
*(chunk
/512) * data
);
2374 if (start
+ stripes
> size
)
2375 tailstripes
= (size
- start
);
2377 grow_backup(sra
, start
*(chunk
/512), tailstripes
,
2379 disks
, chunk
, level
, layout
,
2380 dests
, destfd
, destoffsets
,
2381 part
, °raded
, buf
);
2384 validate(afd
, destfd
[0], destoffsets
[0]);
2386 if (wait_backup(sra
, (start
-stripes
*2) * (chunk
/512), stripes
* (chunk
/512), 0,
2387 dests
, destfd
, destoffsets
,
2390 sysfs_set_num(sra
, NULL
, "suspend_lo", ((start
-stripes
)*(chunk
/512)) * data
);
2391 wait_backup(sra
, (start
-stripes
) * (chunk
/512), tailstripes
* (chunk
/512), 0,
2392 dests
, destfd
, destoffsets
,
2394 sysfs_set_num(sra
, NULL
, "suspend_lo", (size
*(chunk
/512)) * data
);
2395 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
2401 * If any spare contains md_back_data-1 which is recent wrt mtime,
2402 * write that data into the array and update the super blocks with
2403 * the new reshape_progress
2405 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
2406 char *backup_file
, int verbose
)
2410 unsigned long long *offsets
;
2411 unsigned long long nstripe
, ostripe
;
2414 if (info
->new_level
!= info
->array
.level
)
2415 return 1; /* Cannot handle level changes (they are instantaneous) */
2417 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
2418 if (info
->array
.level
== 6) odata
--; /* number of data disks */
2419 ndata
= info
->array
.raid_disks
- 1;
2420 if (info
->new_level
== 6) ndata
--;
2422 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
2424 if (info
->delta_disks
<= 0)
2425 /* Didn't grow, so the backup file must have
2429 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
2430 struct mdinfo dinfo
;
2433 char *devname
, namebuf
[20];
2435 /* This was a spare and may have some saved data on it.
2436 * Load the superblock, find and load the
2437 * backup_super_block.
2438 * If either fail, go on to next device.
2439 * If the backup contains no new info, just return
2440 * else restore data and update all superblocks
2442 if (i
== old_disks
-1) {
2443 fd
= open(backup_file
, O_RDONLY
);
2445 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
2446 backup_file
, strerror(errno
));
2449 devname
= backup_file
;
2454 if (st
->ss
->load_super(st
, fd
, NULL
))
2457 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2458 st
->ss
->free_super(st
);
2461 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
2463 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
2464 continue; /* Cannot seek */
2466 sprintf(namebuf
, "device-%d", i
);
2469 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
2471 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
2472 continue; /* Cannot read */
2474 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
2475 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
2477 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
2480 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
2482 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
2483 continue; /* bad checksum */
2485 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
2486 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
2488 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
2489 continue; /* Bad second checksum */
2491 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
2493 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
2494 continue; /* Wrong uuid */
2497 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2498 * sometimes they aren't... So allow considerable flexability in matching, and allow
2499 * this test to be overridden by an environment variable.
2501 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
2502 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
2503 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2504 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
2505 "for array with timestamp %lu\n",
2506 (unsigned long)__le64_to_cpu(bsb
.mtime
),
2507 (unsigned long)info
->array
.utime
);
2510 fprintf(stderr
, Name
": too-old timestamp on "
2511 "backup-metadata on %s\n", devname
);
2512 continue; /* time stamp is too bad */
2516 if (bsb
.magic
[15] == '1') {
2517 if (info
->delta_disks
>= 0) {
2518 /* reshape_progress is increasing */
2519 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2520 info
->reshape_progress
) {
2523 fprintf(stderr
, Name
": backup-metadata found on %s but is not needed\n", devname
);
2524 continue; /* No new data here */
2527 /* reshape_progress is decreasing */
2528 if (__le64_to_cpu(bsb
.arraystart
) >=
2529 info
->reshape_progress
)
2530 goto nonew
; /* No new data here */
2533 if (info
->delta_disks
>= 0) {
2534 /* reshape_progress is increasing */
2535 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2536 info
->reshape_progress
&&
2537 __le64_to_cpu(bsb
.arraystart2
) + __le64_to_cpu(bsb
.length2
) <
2538 info
->reshape_progress
)
2539 goto nonew
; /* No new data here */
2541 /* reshape_progress is decreasing */
2542 if (__le64_to_cpu(bsb
.arraystart
) >=
2543 info
->reshape_progress
&&
2544 __le64_to_cpu(bsb
.arraystart2
) >=
2545 info
->reshape_progress
)
2546 goto nonew
; /* No new data here */
2549 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
2552 fprintf(stderr
, Name
": Failed to verify secondary backup-metadata block on %s\n",
2554 continue; /* Cannot seek */
2556 /* There should be a duplicate backup superblock 4k before here */
2557 if (lseek64(fd
, -4096, 1) < 0 ||
2558 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
2559 goto second_fail
; /* Cannot find leading superblock */
2560 if (bsb
.magic
[15] == '1')
2561 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
2563 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
2564 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
2565 goto second_fail
; /* Cannot find leading superblock */
2567 /* Now need the data offsets for all devices. */
2568 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
2569 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
2572 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2573 /* FIXME should be this be an error */
2575 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2576 st
->ss
->free_super(st
);
2577 offsets
[j
] = dinfo
.data_offset
* 512;
2579 printf(Name
": restoring critical section\n");
2581 if (restore_stripes(fdlist
, offsets
,
2582 info
->array
.raid_disks
,
2586 fd
, __le64_to_cpu(bsb
.devstart
)*512,
2587 __le64_to_cpu(bsb
.arraystart
)*512,
2588 __le64_to_cpu(bsb
.length
)*512)) {
2589 /* didn't succeed, so giveup */
2591 fprintf(stderr
, Name
": Error restoring backup from %s\n",
2596 if (bsb
.magic
[15] == '2' &&
2597 restore_stripes(fdlist
, offsets
,
2598 info
->array
.raid_disks
,
2602 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
2603 __le64_to_cpu(bsb
.devstart2
)*512,
2604 __le64_to_cpu(bsb
.arraystart2
)*512,
2605 __le64_to_cpu(bsb
.length2
)*512)) {
2606 /* didn't succeed, so giveup */
2608 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
2614 /* Ok, so the data is restored. Let's update those superblocks. */
2616 if (info
->delta_disks
>= 0) {
2617 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
2618 __le64_to_cpu(bsb
.length
);
2619 if (bsb
.magic
[15] == '2') {
2620 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
2621 __le64_to_cpu(bsb
.length2
);
2622 if (p2
> info
->reshape_progress
)
2623 info
->reshape_progress
= p2
;
2626 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
2627 if (bsb
.magic
[15] == '2') {
2628 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
2629 if (p2
< info
->reshape_progress
)
2630 info
->reshape_progress
= p2
;
2633 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
2634 if (fdlist
[j
] < 0) continue;
2635 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2637 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2638 dinfo
.reshape_progress
= info
->reshape_progress
;
2639 st
->ss
->update_super(st
, &dinfo
,
2640 "_reshape_progress",
2642 st
->ss
->store_super(st
, fdlist
[j
]);
2643 st
->ss
->free_super(st
);
2647 /* Didn't find any backup data, try to see if any
2650 if (info
->delta_disks
< 0) {
2651 /* When shrinking, the critical section is at the end.
2652 * So see if we are before the critical section.
2654 unsigned long long first_block
;
2655 nstripe
= ostripe
= 0;
2657 while (ostripe
>= nstripe
) {
2658 ostripe
+= info
->array
.chunk_size
/ 512;
2659 first_block
= ostripe
* odata
;
2660 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
2661 (info
->new_chunk
/512);
2664 if (info
->reshape_progress
>= first_block
)
2667 if (info
->delta_disks
> 0) {
2668 /* See if we are beyond the critical section. */
2669 unsigned long long last_block
;
2670 nstripe
= ostripe
= 0;
2672 while (nstripe
>= ostripe
) {
2673 nstripe
+= info
->new_chunk
/ 512;
2674 last_block
= nstripe
* ndata
;
2675 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
2676 (info
->array
.chunk_size
/512);
2679 if (info
->reshape_progress
>= last_block
)
2682 /* needed to recover critical section! */
2684 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
2688 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
2691 /* Array is assembled and ready to be started, but
2692 * monitoring is probably required.
2695 * - set upper bound for resync
2696 * - initialise the 'suspend' boundaries
2697 * - switch to read-write
2698 * - fork and continue monitoring
2702 unsigned long long backup_offsets
[1];
2703 int odisks
, ndisks
, ochunk
, nchunk
,odata
,ndata
;
2704 unsigned long a
,b
,blocks
,stripes
;
2707 unsigned long long *offsets
;
2709 struct mdinfo
*sra
, *sd
;
2711 unsigned long cache
;
2714 err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
2718 /* make sure reshape doesn't progress until we are ready */
2719 sysfs_set_str(info
, NULL
, "sync_max", "0");
2720 sysfs_set_str(info
, NULL
, "array_state", "active"); /* FIXME or clean */
2722 sra
= sysfs_read(-1, devname2devnum(info
->sys_name
),
2723 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
2728 /* ndisks is not growing, so raid_disks is old and +delta is new */
2729 odisks
= info
->array
.raid_disks
;
2730 ndisks
= odisks
+ info
->delta_disks
;
2733 if (info
->array
.level
== 6) {
2737 ochunk
= info
->array
.chunk_size
;
2738 nchunk
= info
->new_chunk
;
2740 a
= (ochunk
/512) * odata
;
2741 b
= (nchunk
/512) * ndata
;
2749 /* LCM == product / GCD */
2750 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
2753 while (blocks
* 32 < sra
->component_size
&&
2756 stripes
= blocks
/ (info
->array
.chunk_size
/512) / odata
;
2758 /* check that the internal stripe cache is
2759 * large enough, or it won't work.
2761 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
2762 cache
= cache
* 4 / 4096;
2763 if (cache
< blocks
/ 8 / odisks
+ 16)
2764 /* Make it big enough to hold 'blocks' */
2765 cache
= blocks
/ 8 / odisks
+ 16;
2766 if (sra
->cache_size
< cache
)
2767 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
2770 memset(&bsb
, 0, 512);
2771 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2772 memcpy(&bsb
.set_uuid
, info
->uuid
, 16);
2773 bsb
.mtime
= __cpu_to_le64(time(0));
2774 bsb
.devstart2
= blocks
;
2776 backup_fd
= open(backup_file
, O_RDWR
|O_CREAT
, S_IRUSR
| S_IWUSR
);
2777 if (backup_fd
< 0) {
2778 fprintf(stderr
, Name
": Cannot open backup file %s\n",
2779 backup_file
?: "- no backup-file given");
2782 backup_list
[0] = backup_fd
;
2783 backup_offsets
[0] = 8 * 512;
2784 fds
= malloc(odisks
* sizeof(fds
[0]));
2785 offsets
= malloc(odisks
* sizeof(offsets
[0]));
2786 for (d
=0; d
<odisks
; d
++)
2789 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2790 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2792 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2793 char *dn
= map_dev(sd
->disk
.major
,
2795 fds
[sd
->disk
.raid_disk
]
2796 = dev_open(dn
, O_RDONLY
);
2797 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
2798 if (fds
[sd
->disk
.raid_disk
] < 0) {
2799 fprintf(stderr
, Name
": %s: cannot open component %s\n",
2800 info
->sys_name
, dn
?dn
:"-unknown-");
2811 mlockall(MCL_FUTURE
);
2812 if (info
->delta_disks
< 0)
2813 done
= child_shrink(-1, info
, stripes
,
2815 info
->array
.raid_disks
,
2816 info
->array
.chunk_size
,
2817 info
->array
.level
, info
->array
.layout
,
2819 1, backup_list
, backup_offsets
);
2820 else if (info
->delta_disks
== 0) {
2821 /* The 'start' is a per-device stripe number.
2822 * reshape_progress is a per-array sector number.
2823 * So divide by ndata * chunk_size
2825 unsigned long long start
= info
->reshape_progress
/ ndata
;
2826 start
/= (info
->array
.chunk_size
/512);
2827 done
= child_same_size(-1, info
, stripes
,
2830 info
->array
.raid_disks
,
2831 info
->array
.chunk_size
,
2832 info
->array
.level
, info
->array
.layout
,
2834 1, backup_list
, backup_offsets
);
2836 if (backup_file
&& done
)
2837 unlink(backup_file
);
2838 /* FIXME should I intuit a level change */
2841 fprintf(stderr
, Name
": Cannot run child to continue monitoring reshape: %s\n",