2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 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>
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
38 int restore_backup(struct supertype
*st
,
39 struct mdinfo
*content
,
49 int disk_count
= next_spare
+ working_disks
;
50 char *backup_file
= *backup_filep
;
52 dprintf("Called restore_backup()\n");
53 fdlist
= xmalloc(sizeof(int) * disk_count
);
55 enable_fds(next_spare
);
56 for (i
= 0; i
< next_spare
; i
++)
58 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
64 fd
= dev_open(buf
, O_RDWR
);
66 if (dev
->disk
.raid_disk
>= 0)
67 fdlist
[dev
->disk
.raid_disk
] = fd
;
69 fdlist
[next_spare
++] = fd
;
73 backup_file
= locate_backup(content
->sys_name
);
74 *backup_filep
= backup_file
;
77 if (st
->ss
->external
&& st
->ss
->recover_backup
)
78 err
= st
->ss
->recover_backup(st
, content
);
80 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
81 backup_file
, verbose
> 0);
83 while (next_spare
> 0) {
85 if (fdlist
[next_spare
] >= 0)
86 close(fdlist
[next_spare
]);
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
92 pr_err("Possibly you need to specify a --backup-file\n");
96 dprintf("restore_backup() returns status OK.\n");
100 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
108 * This means that we need to *find* all other devices.
115 struct supertype
*st
= NULL
;
116 char *subarray
= NULL
;
118 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
119 pr_err("cannot get array info for %s\n", devname
);
123 if (info
.array
.level
!= -1) {
124 pr_err("can only add devices to linear arrays\n");
128 st
= super_by_fd(fd
, &subarray
);
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info
.array
.major_version
);
136 pr_err("Cannot grow linear sub-arrays yet\n");
142 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
144 pr_err("cannot open %s\n", newdev
);
149 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
150 pr_err("%s is not a block device!\n", newdev
);
155 /* now check out all the devices and make sure we can read the
157 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
158 mdu_disk_info_t disk
;
161 st
->ss
->free_super(st
);
164 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
165 pr_err("cannot get device detail for device %d\n",
171 dv
= map_dev(disk
.major
, disk
.minor
, 1);
173 pr_err("cannot find device file for device %d\n",
179 fd2
= dev_open(dv
, O_RDWR
);
181 pr_err("cannot open device file %s\n", dv
);
187 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
188 pr_err("cannot find super block on %s\n", dv
);
196 /* Ok, looks good. Lets update the superblock and write it out to
200 info
.disk
.number
= d
;
201 info
.disk
.major
= major(stb
.st_rdev
);
202 info
.disk
.minor
= minor(stb
.st_rdev
);
203 info
.disk
.raid_disk
= d
;
204 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
205 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
208 if (st
->ss
->store_super(st
, nfd
)) {
209 pr_err("Cannot store new superblock on %s\n",
216 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
217 pr_err("Cannot add new disk to this array\n");
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
224 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
225 pr_err("cannot get array info for %s\n", devname
);
230 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
231 mdu_disk_info_t disk
;
235 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
236 pr_err("cannot get device detail for device %d\n",
240 dv
= map_dev(disk
.major
, disk
.minor
, 1);
242 pr_err("cannot find device file for device %d\n",
246 fd2
= dev_open(dv
, O_RDWR
);
248 pr_err("cannot open device file %s\n", dv
);
251 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
252 pr_err("cannot find super block on %s\n", dv
);
256 info
.array
.raid_disks
= nd
+1;
257 info
.array
.nr_disks
= nd
+1;
258 info
.array
.active_disks
= nd
+1;
259 info
.array
.working_disks
= nd
+1;
261 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
264 if (st
->ss
->store_super(st
, fd2
)) {
265 pr_err("Cannot store new superblock on %s\n", dv
);
275 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
286 mdu_bitmap_file_t bmf
;
287 mdu_array_info_t array
;
288 struct supertype
*st
;
289 char *subarray
= NULL
;
290 int major
= BITMAP_MAJOR_HI
;
291 int vers
= md_get_version(fd
);
292 unsigned long long bitmapsize
, array_size
;
295 major
= BITMAP_MAJOR_HOSTENDIAN
;
296 pr_err("Warning - bitmaps created on this kernel are not portable\n"
297 " between different architectures. Consider upgrading the Linux kernel.\n");
301 * We only ever get called if s->bitmap_file is != NULL, so this check
302 * is just here to quiet down static code checkers.
307 if (strcmp(s
->bitmap_file
, "clustered") == 0)
308 major
= BITMAP_MAJOR_CLUSTERED
;
310 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
312 pr_err("Memory allocation failure.\n");
314 pr_err("bitmaps not supported by this kernel.\n");
317 if (bmf
.pathname
[0]) {
318 if (strcmp(s
->bitmap_file
,"none")==0) {
319 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
320 pr_err("failed to remove bitmap %s\n",
326 pr_err("%s already has a bitmap (%s)\n",
327 devname
, bmf
.pathname
);
330 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
331 pr_err("cannot get array status for %s\n", devname
);
334 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
335 if (strcmp(s
->bitmap_file
, "none")==0) {
336 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
337 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
338 if (array
.state
& (1<<MD_SB_CLUSTERED
))
339 pr_err("failed to remove clustered bitmap.\n");
341 pr_err("failed to remove internal bitmap.\n");
346 pr_err("bitmap already present on %s\n", devname
);
350 if (strcmp(s
->bitmap_file
, "none") == 0) {
351 pr_err("no bitmap found on %s\n", devname
);
354 if (array
.level
<= 0) {
355 pr_err("Bitmaps not meaningful with level %s\n",
356 map_num(pers
, array
.level
)?:"of this array");
359 bitmapsize
= array
.size
;
361 if (get_dev_size(fd
, NULL
, &array_size
) &&
362 array_size
> (0x7fffffffULL
<<9)) {
363 /* Array is big enough that we cannot trust array.size
364 * try other approaches
366 bitmapsize
= get_component_size(fd
);
368 if (bitmapsize
== 0) {
369 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
373 if (array
.level
== 10) {
374 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
375 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
378 st
= super_by_fd(fd
, &subarray
);
380 pr_err("Cannot understand version %d.%d\n",
381 array
.major_version
, array
.minor_version
);
385 pr_err("Cannot add bitmaps to sub-arrays yet\n");
390 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
391 strcmp(s
->bitmap_file
, "clustered") == 0) {
394 int offset_setable
= 0;
396 if (st
->ss
->add_internal_bitmap
== NULL
) {
397 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
400 st
->nodes
= c
->nodes
;
401 st
->cluster_name
= c
->homecluster
;
402 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
405 for (d
=0; d
< st
->max_devs
; d
++) {
406 mdu_disk_info_t disk
;
409 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
411 if (disk
.major
== 0 &&
414 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
416 dv
= map_dev(disk
.major
, disk
.minor
, 1);
418 int fd2
= dev_open(dv
, O_RDWR
);
421 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
422 if (st
->ss
->add_internal_bitmap(
424 &s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
425 bitmapsize
, offset_setable
,
428 st
->ss
->write_bitmap(st
, fd2
, NoUpdate
);
430 pr_err("failed to create internal bitmap - chunksize problem.\n");
438 if (offset_setable
) {
439 st
->ss
->getinfo_super(st
, mdi
, NULL
);
440 sysfs_init(mdi
, fd
, NULL
);
441 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
444 if (strcmp(s
->bitmap_file
, "clustered") == 0)
445 array
.state
|= (1<<MD_SB_CLUSTERED
);
446 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
447 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
451 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
452 pr_err("failed to set internal bitmap.\n");
459 int max_devs
= st
->max_devs
;
461 /* try to load a superblock */
462 for (d
= 0; d
< max_devs
; d
++) {
463 mdu_disk_info_t disk
;
467 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
469 if ((disk
.major
==0 && disk
.minor
==0) ||
470 (disk
.state
& (1<<MD_DISK_REMOVED
)))
472 dv
= map_dev(disk
.major
, disk
.minor
, 1);
475 fd2
= dev_open(dv
, O_RDONLY
);
477 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
479 st
->ss
->uuid_from_super(st
, uuid
);
486 pr_err("cannot find UUID for array!\n");
489 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
490 c
->delay
, s
->write_behind
, bitmapsize
, major
)) {
493 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
495 pr_err("weird: %s cannot be opened\n",
499 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
502 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
503 pr_err("Cannot set bitmap file for %s: %s\n",
504 devname
, strerror(err
));
513 * When reshaping an array we might need to backup some data.
514 * This is written to all spares with a 'super_block' describing it.
515 * The superblock goes 4K from the end of the used space on the
517 * It if written after the backup is complete.
518 * It has the following structure.
521 static struct mdp_backup_super
{
522 char magic
[16]; /* md_backup_data-1 or -2 */
525 /* start/sizes in 512byte sectors */
526 __u64 devstart
; /* address on backup device/file of data */
529 __u32 sb_csum
; /* csum of preceeding bytes. */
531 __u64 devstart2
; /* offset in to data of second section */
534 __u32 sb_csum2
; /* csum of preceeding bytes. */
536 } __attribute__((aligned(512))) bsb
, bsb2
;
538 static __u32
bsb_csum(char *buf
, int len
)
542 for (i
= 0; i
< len
; i
++)
543 csum
= (csum
<<3) + buf
[0];
544 return __cpu_to_le32(csum
);
547 static int check_idle(struct supertype
*st
)
549 /* Check that all member arrays for this container, or the
550 * container of this array, are idle
552 char *container
= (st
->container_devnm
[0]
553 ? st
->container_devnm
: st
->devnm
);
554 struct mdstat_ent
*ent
, *e
;
557 ent
= mdstat_read(0, 0);
558 for (e
= ent
; e
; e
= e
->next
) {
559 if (!is_container_member(e
, container
))
561 if (e
->percent
>= 0) {
570 static int freeze_container(struct supertype
*st
)
572 char *container
= (st
->container_devnm
[0]
573 ? st
->container_devnm
: st
->devnm
);
578 if (block_monitor(container
, 1)) {
579 pr_err("failed to freeze container\n");
586 static void unfreeze_container(struct supertype
*st
)
588 char *container
= (st
->container_devnm
[0]
589 ? st
->container_devnm
: st
->devnm
);
591 unblock_monitor(container
, 1);
594 static int freeze(struct supertype
*st
)
596 /* Try to freeze resync/rebuild on this array/container.
597 * Return -1 if the array is busy,
598 * return -2 container cannot be frozen,
599 * return 0 if this kernel doesn't support 'frozen'
600 * return 1 if it worked.
602 if (st
->ss
->external
)
603 return freeze_container(st
);
605 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
611 /* Need to clear any 'read-auto' status */
612 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
613 strncmp(buf
, "read-auto", 9) == 0)
614 sysfs_set_str(sra
, NULL
, "array_state", "clean");
616 err
= sysfs_freeze_array(sra
);
622 static void unfreeze(struct supertype
*st
)
624 if (st
->ss
->external
)
625 return unfreeze_container(st
);
627 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
631 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0
632 && strcmp(buf
, "frozen\n") == 0)
633 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
638 static void wait_reshape(struct mdinfo
*sra
)
640 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
646 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
647 strncmp(action
, "reshape", 7) == 0)
648 sysfs_wait(fd
, NULL
);
652 static int reshape_super(struct supertype
*st
, unsigned long long size
,
653 int level
, int layout
, int chunksize
, int raid_disks
,
654 int delta_disks
, char *backup_file
, char *dev
,
655 int direction
, int verbose
)
657 /* nothing extra to check in the native case */
658 if (!st
->ss
->external
)
660 if (!st
->ss
->reshape_super
||
661 !st
->ss
->manage_reshape
) {
662 pr_err("%s metadata does not support reshape\n",
667 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
668 raid_disks
, delta_disks
, backup_file
, dev
,
672 static void sync_metadata(struct supertype
*st
)
674 if (st
->ss
->external
) {
675 if (st
->update_tail
) {
676 flush_metadata_updates(st
);
677 st
->update_tail
= &st
->updates
;
679 st
->ss
->sync_metadata(st
);
683 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
685 /* when dealing with external metadata subarrays we need to be
686 * prepared to handle EAGAIN. The kernel may need to wait for
687 * mdmon to mark the array active so the kernel can handle
688 * allocations/writeback when preparing the reshape action
689 * (md_allow_write()). We temporarily disable safe_mode_delay
690 * to close a race with the array_state going clean before the
691 * next write to raid_disks / stripe_cache_size
696 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
698 (strcmp(name
, "raid_disks") != 0 &&
699 strcmp(name
, "stripe_cache_size") != 0))
700 return sysfs_set_num(sra
, NULL
, name
, n
);
702 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
705 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
706 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
707 if (rc
< 0 && errno
== EAGAIN
) {
708 ping_monitor(container
);
709 /* if we get EAGAIN here then the monitor is not active
712 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
714 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
718 int start_reshape(struct mdinfo
*sra
, int already_running
,
719 int before_data_disks
, int data_disks
)
722 unsigned long long sync_max_to_set
;
724 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
725 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
726 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
727 sra
->reshape_progress
);
728 if (before_data_disks
<= data_disks
)
729 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
731 sync_max_to_set
= (sra
->component_size
* data_disks
732 - sra
->reshape_progress
) / data_disks
;
733 if (!already_running
)
734 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
735 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
736 if (!already_running
&& err
== 0) {
739 err
= sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
742 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
747 void abort_reshape(struct mdinfo
*sra
)
749 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
751 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
752 * suspend_hi to decrease as well as increase.")
753 * you could only increase suspend_{lo,hi} unless the region they
754 * covered was empty. So to reset to 0, you need to push suspend_lo
755 * up past suspend_hi first. So to maximize the chance of mdadm
756 * working on all kernels, we want to keep doing that.
758 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
759 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
760 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
761 sysfs_set_num(sra
, NULL
, "sync_min", 0);
762 // It isn't safe to reset sync_max as we aren't monitoring.
763 // Array really should be stopped at this point.
766 int remove_disks_for_takeover(struct supertype
*st
,
771 struct mdinfo
*remaining
;
774 if (sra
->array
.level
== 10)
775 nr_of_copies
= layout
& 0xff;
776 else if (sra
->array
.level
== 1)
777 nr_of_copies
= sra
->array
.raid_disks
;
781 remaining
= sra
->devs
;
783 /* for each 'copy', select one device and remove from the list. */
784 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
785 struct mdinfo
**diskp
;
788 /* Find a working device to keep */
789 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
790 struct mdinfo
*disk
= *diskp
;
792 if (disk
->disk
.raid_disk
< slot
)
794 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
796 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
798 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
800 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
803 /* We have found a good disk to use! */
805 disk
->next
= sra
->devs
;
814 if (slot
< sra
->array
.raid_disks
) {
815 /* didn't find all slots */
821 sra
->devs
= remaining
;
825 /* Remove all 'remaining' devices from the array */
827 struct mdinfo
*sd
= remaining
;
828 remaining
= sd
->next
;
830 sysfs_set_str(sra
, sd
, "state", "faulty");
831 sysfs_set_str(sra
, sd
, "slot", "none");
832 /* for external metadata disks should be removed in mdmon */
833 if (!st
->ss
->external
)
834 sysfs_set_str(sra
, sd
, "state", "remove");
835 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
836 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
837 sd
->next
= sra
->devs
;
843 void reshape_free_fdlist(int *fdlist
,
844 unsigned long long *offsets
,
849 for (i
= 0; i
< size
; i
++)
857 int reshape_prepare_fdlist(char *devname
,
861 unsigned long blocks
,
864 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 sd
->disk
.raid_disk
< raid_disks
) {
878 char *dn
= map_dev(sd
->disk
.major
,
880 fdlist
[sd
->disk
.raid_disk
]
881 = dev_open(dn
, O_RDONLY
);
882 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
883 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
884 pr_err("%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 pr_err("%s: cannot open component %s\n",
897 devname
, dn
? dn
: "-unknown-");
908 int reshape_open_backup_file(char *backup_file
,
913 unsigned long long *offsets
,
917 /* Return 1 on success, 0 on any form of failure */
918 /* need to check backup file is large enough */
924 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
928 pr_err("%s: cannot create backup file %s: %s\n",
929 devname
, backup_file
, strerror(errno
));
932 /* Guard against backup file being on array device.
933 * If array is partitioned or if LVM etc is in the
934 * way this will not notice, but it is better than
937 fstat(*fdlist
, &stb
);
940 if (stb
.st_rdev
== dev
) {
941 pr_err("backup file must NOT be on the array being reshaped.\n");
947 for (i
=0; i
< blocks
+ 8 ; i
++) {
948 if (write(*fdlist
, buf
, 512) != 512) {
949 pr_err("%s: cannot create backup file %s: %s\n",
950 devname
, backup_file
, strerror(errno
));
954 if (fsync(*fdlist
) != 0) {
955 pr_err("%s: cannot create backup file %s: %s\n",
956 devname
, backup_file
, strerror(errno
));
960 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
961 char *bu
= make_backup(sys_name
);
962 if (symlink(backup_file
, bu
))
963 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
971 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
972 unsigned int ndata
, unsigned int odata
)
974 unsigned long a
, b
, blocks
;
975 /* So how much do we need to backup.
976 * We need an amount of data which is both a whole number of
977 * old stripes and a whole number of new stripes.
978 * So LCM for (chunksize*datadisks).
980 a
= (ochunk
/512) * odata
;
981 b
= (nchunk
/512) * ndata
;
984 /* LCM == product / GCD */
985 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
990 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
992 /* Based on the current array state in info->array and
993 * the changes in info->new_* etc, determine:
994 * - whether the change is possible
995 * - Intermediate level/raid_disks/layout
996 * - whether a restriping reshape is needed
997 * - number of sectors in minimum change unit. This
998 * will cover a whole number of stripes in 'before' and
1001 * Return message if the change should be rejected
1002 * NULL if the change can be achieved
1004 * This can be called as part of starting a reshape, or
1005 * when assembling an array that is undergoing reshape.
1007 int near
, far
, offset
, copies
;
1009 int old_chunk
, new_chunk
;
1010 /* delta_parity records change in number of devices
1011 * caused by level change
1013 int delta_parity
= 0;
1015 memset(re
, 0, sizeof(*re
));
1017 /* If a new level not explicitly given, we assume no-change */
1018 if (info
->new_level
== UnSet
)
1019 info
->new_level
= info
->array
.level
;
1021 if (info
->new_chunk
)
1022 switch (info
->new_level
) {
1028 /* chunk size is meaningful, must divide component_size
1031 if (info
->component_size
% (info
->new_chunk
/512)) {
1032 unsigned long long shrink
= info
->component_size
;
1033 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1034 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1035 info
->new_chunk
/1024, info
->component_size
/2);
1036 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1038 pr_err("will shrink the array so the given chunk size would work.\n");
1043 return "chunk size not meaningful for this level";
1046 info
->new_chunk
= info
->array
.chunk_size
;
1048 switch (info
->array
.level
) {
1050 return "No reshape is possibly for this RAID level";
1052 if (info
->delta_disks
!= UnSet
)
1053 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1055 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1057 /* RAID1 can convert to RAID1 with different disks, or
1058 * raid5 with 2 disks, or
1061 if (info
->new_level
> 1 &&
1062 (info
->component_size
& 7))
1063 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1064 if (info
->new_level
== 0) {
1065 if (info
->delta_disks
!= UnSet
&&
1066 info
->delta_disks
!= 0)
1067 return "Cannot change number of disks with RAID1->RAID0 conversion";
1069 re
->before
.data_disks
= 1;
1070 re
->after
.data_disks
= 1;
1073 if (info
->new_level
== 1) {
1074 if (info
->delta_disks
== UnSet
)
1075 /* Don't know what to do */
1076 return "no change requested for Growing RAID1";
1080 if (info
->array
.raid_disks
!= 2 &&
1081 info
->new_level
== 5)
1082 return "Can only convert a 2-device array to RAID5";
1083 if (info
->array
.raid_disks
== 2 &&
1084 info
->new_level
== 5) {
1087 re
->before
.data_disks
= 1;
1088 if (info
->delta_disks
!= UnSet
&&
1089 info
->delta_disks
!= 0)
1090 re
->after
.data_disks
= 1 + info
->delta_disks
;
1092 re
->after
.data_disks
= 1;
1093 if (re
->after
.data_disks
< 1)
1094 return "Number of disks too small for RAID5";
1096 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1097 info
->array
.chunk_size
= 65536;
1100 /* Could do some multi-stage conversions, but leave that to
1103 return "Impossibly level change request for RAID1";
1106 /* RAID10 can be converted from near mode to
1107 * RAID0 by removing some devices.
1108 * It can also be reshaped if the kernel supports
1111 switch (info
->new_level
) {
1113 if ((info
->array
.layout
& ~0xff) != 0x100)
1114 return "Cannot Grow RAID10 with far/offset layout";
1115 /* number of devices must be multiple of number of copies */
1116 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1117 return "RAID10 layout too complex for Grow operation";
1119 new_disks
= (info
->array
.raid_disks
1120 / (info
->array
.layout
& 0xff));
1121 if (info
->delta_disks
== UnSet
)
1122 info
->delta_disks
= (new_disks
1123 - info
->array
.raid_disks
);
1125 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1126 return "New number of raid-devices impossible for RAID10";
1127 if (info
->new_chunk
&&
1128 info
->new_chunk
!= info
->array
.chunk_size
)
1129 return "Cannot change chunk-size with RAID10 Grow";
1133 re
->before
.data_disks
= new_disks
;
1134 re
->after
.data_disks
= re
->before
.data_disks
;
1138 near
= info
->array
.layout
& 0xff;
1139 far
= (info
->array
.layout
>> 8) & 0xff;
1140 offset
= info
->array
.layout
& 0x10000;
1141 if (far
> 1 && !offset
)
1142 return "Cannot reshape RAID10 in far-mode";
1143 copies
= near
* far
;
1145 old_chunk
= info
->array
.chunk_size
* far
;
1147 if (info
->new_layout
== UnSet
)
1148 info
->new_layout
= info
->array
.layout
;
1150 near
= info
->new_layout
& 0xff;
1151 far
= (info
->new_layout
>> 8) & 0xff;
1152 offset
= info
->new_layout
& 0x10000;
1153 if (far
> 1 && !offset
)
1154 return "Cannot reshape RAID10 to far-mode";
1155 if (near
* far
!= copies
)
1156 return "Cannot change number of copies when reshaping RAID10";
1158 if (info
->delta_disks
== UnSet
)
1159 info
->delta_disks
= 0;
1160 new_disks
= (info
->array
.raid_disks
+
1163 new_chunk
= info
->new_chunk
* far
;
1166 re
->before
.layout
= info
->array
.layout
;
1167 re
->before
.data_disks
= info
->array
.raid_disks
;
1168 re
->after
.layout
= info
->new_layout
;
1169 re
->after
.data_disks
= new_disks
;
1170 /* For RAID10 we don't do backup but do allow reshape,
1171 * so set backup_blocks to INVALID_SECTORS rather than
1173 * And there is no need to synchronise stripes on both
1174 * 'old' and 'new'. So the important
1175 * number is the minimum data_offset difference
1176 * which is the larger of (offset copies * chunk).
1178 re
->backup_blocks
= INVALID_SECTORS
;
1179 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1180 if (new_disks
< re
->before
.data_disks
&&
1181 info
->space_after
< re
->min_offset_change
)
1182 /* Reduce component size by one chunk */
1183 re
->new_size
= (info
->component_size
-
1184 re
->min_offset_change
);
1186 re
->new_size
= info
->component_size
;
1187 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1191 return "RAID10 can only be changed to RAID0";
1194 /* RAID0 can be converted to RAID10, or to RAID456 */
1195 if (info
->new_level
== 10) {
1196 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1197 /* Assume near=2 layout */
1198 info
->new_layout
= 0x102;
1199 info
->delta_disks
= info
->array
.raid_disks
;
1201 if (info
->new_layout
== UnSet
) {
1202 int copies
= 1 + (info
->delta_disks
1203 / info
->array
.raid_disks
);
1204 if (info
->array
.raid_disks
* (copies
-1)
1205 != info
->delta_disks
)
1206 return "Impossible number of devices for RAID0->RAID10";
1207 info
->new_layout
= 0x100 + copies
;
1209 if (info
->delta_disks
== UnSet
) {
1210 int copies
= info
->new_layout
& 0xff;
1211 if (info
->new_layout
!= 0x100 + copies
)
1212 return "New layout impossible for RAID0->RAID10";;
1213 info
->delta_disks
= (copies
- 1) *
1214 info
->array
.raid_disks
;
1216 if (info
->new_chunk
&&
1217 info
->new_chunk
!= info
->array
.chunk_size
)
1218 return "Cannot change chunk-size with RAID0->RAID10";
1221 re
->before
.data_disks
= (info
->array
.raid_disks
+
1223 re
->after
.data_disks
= re
->before
.data_disks
;
1224 re
->before
.layout
= info
->new_layout
;
1228 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1229 * a raid4 style layout of the final level.
1231 switch (info
->new_level
) {
1236 re
->before
.layout
= 0;
1241 re
->before
.layout
= ALGORITHM_PARITY_N
;
1242 if (info
->new_layout
== UnSet
)
1243 info
->new_layout
= map_name(r5layout
, "default");
1248 re
->before
.layout
= ALGORITHM_PARITY_N
;
1249 if (info
->new_layout
== UnSet
)
1250 info
->new_layout
= map_name(r6layout
, "default");
1253 return "Impossible level change requested";
1255 re
->before
.data_disks
= info
->array
.raid_disks
;
1256 /* determining 'after' layout happens outside this 'switch' */
1260 info
->array
.layout
= ALGORITHM_PARITY_N
;
1262 switch (info
->new_level
) {
1266 re
->level
= info
->array
.level
;
1267 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1268 re
->before
.layout
= info
->array
.layout
;
1272 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1273 re
->before
.layout
= info
->array
.layout
;
1278 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1279 switch (info
->array
.layout
) {
1280 case ALGORITHM_LEFT_ASYMMETRIC
:
1281 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1283 case ALGORITHM_RIGHT_ASYMMETRIC
:
1284 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1286 case ALGORITHM_LEFT_SYMMETRIC
:
1287 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1289 case ALGORITHM_RIGHT_SYMMETRIC
:
1290 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1292 case ALGORITHM_PARITY_0
:
1293 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1295 case ALGORITHM_PARITY_N
:
1296 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1299 return "Cannot convert an array with this layout";
1303 if (info
->array
.raid_disks
!= 2)
1304 return "Can only convert a 2-device array to RAID1";
1305 if (info
->delta_disks
!= UnSet
&&
1306 info
->delta_disks
!= 0)
1307 return "Cannot set raid_disk when converting RAID5->RAID1";
1309 info
->new_chunk
= 0;
1312 return "Impossible level change requested";
1316 switch (info
->new_level
) {
1322 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1323 re
->before
.layout
= info
->array
.layout
;
1326 return "Impossible level change requested";
1331 /* If we reached here then it looks like a re-stripe is
1332 * happening. We have determined the intermediate level
1333 * and initial raid_disks/layout and stored these in 're'.
1335 * We need to deduce the final layout that can be atomically
1336 * converted to the end state.
1338 switch (info
->new_level
) {
1340 /* We can only get to RAID0 from RAID4 or RAID5
1341 * with appropriate layout and one extra device
1343 if (re
->level
!= 4 && re
->level
!= 5)
1344 return "Cannot covert to RAID0 from this level";
1346 switch (re
->level
) {
1348 re
->before
.layout
= 0;
1349 re
->after
.layout
= 0;
1352 re
->after
.layout
= ALGORITHM_PARITY_N
;
1358 /* We can only get to RAID4 from RAID5 */
1359 if (re
->level
!= 4 && re
->level
!= 5)
1360 return "Cannot convert to RAID4 from this level";
1362 switch (re
->level
) {
1364 re
->after
.layout
= 0;
1367 re
->after
.layout
= ALGORITHM_PARITY_N
;
1373 /* We get to RAID5 from RAID5 or RAID6 */
1374 if (re
->level
!= 5 && re
->level
!= 6)
1375 return "Cannot convert to RAID5 from this level";
1377 switch (re
->level
) {
1379 if (info
->new_layout
== UnSet
)
1380 re
->after
.layout
= re
->before
.layout
;
1382 re
->after
.layout
= info
->new_layout
;
1385 if (info
->new_layout
== UnSet
)
1386 info
->new_layout
= re
->before
.layout
;
1388 /* after.layout needs to be raid6 version of new_layout */
1389 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1390 re
->after
.layout
= ALGORITHM_PARITY_N
;
1393 char *ls
= map_num(r5layout
, info
->new_layout
);
1396 /* Current RAID6 layout has a RAID5
1399 strcat(strcpy(layout
, ls
), "-6");
1400 l
= map_name(r6layout
, layout
);
1402 return "Cannot find RAID6 layout to convert to";
1404 /* Current RAID6 has no equivalent.
1405 * If it is already a '-6' layout we
1406 * can leave it unchanged, else we must
1409 ls
= map_num(r6layout
, info
->new_layout
);
1411 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1412 return "Please specify new layout";
1413 l
= info
->new_layout
;
1415 re
->after
.layout
= l
;
1421 /* We must already be at level 6 */
1423 return "Impossible level change";
1424 if (info
->new_layout
== UnSet
)
1425 re
->after
.layout
= info
->array
.layout
;
1427 re
->after
.layout
= info
->new_layout
;
1430 return "Impossible level change requested";
1432 if (info
->delta_disks
== UnSet
)
1433 info
->delta_disks
= delta_parity
;
1435 re
->after
.data_disks
= (re
->before
.data_disks
1438 switch (re
->level
) {
1439 case 6: re
->parity
= 2;
1442 case 5: re
->parity
= 1;
1444 default: re
->parity
= 0;
1447 /* So we have a restripe operation, we need to calculate the number
1448 * of blocks per reshape operation.
1450 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1451 if (info
->new_chunk
== 0)
1452 info
->new_chunk
= info
->array
.chunk_size
;
1453 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1454 re
->after
.layout
== re
->before
.layout
&&
1455 info
->new_chunk
== info
->array
.chunk_size
) {
1456 /* Nothing to change, can change level immediately. */
1457 re
->level
= info
->new_level
;
1458 re
->backup_blocks
= 0;
1461 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1462 /* chunk and layout changes make no difference */
1463 re
->level
= info
->new_level
;
1464 re
->backup_blocks
= 0;
1468 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1469 get_linux_version() < 2006032)
1470 return "in-place reshape is not safe before 2.6.32 - sorry.";
1472 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1473 get_linux_version() < 2006030)
1474 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1476 re
->backup_blocks
= compute_backup_blocks(
1477 info
->new_chunk
, info
->array
.chunk_size
,
1478 re
->after
.data_disks
,
1479 re
->before
.data_disks
);
1480 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1482 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1486 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1489 struct mdinfo
*info
;
1493 if ((st
== NULL
) || (sra
== NULL
))
1496 if (text_version
== NULL
)
1497 text_version
= sra
->text_version
;
1498 subarray
= strchr(text_version
+1, '/')+1;
1499 info
= st
->ss
->container_content(st
, subarray
);
1501 unsigned long long current_size
= 0;
1502 unsigned long long new_size
=
1503 info
->custom_array_size
/2;
1505 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1506 new_size
> current_size
) {
1507 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1509 dprintf("Error: Cannot set array size");
1512 dprintf("Array size changed");
1514 dprintf_cont(" from %llu to %llu.\n",
1515 current_size
, new_size
);
1519 dprintf("Error: set_array_size(): info pointer in NULL\n");
1524 static int reshape_array(char *container
, int fd
, char *devname
,
1525 struct supertype
*st
, struct mdinfo
*info
,
1526 int force
, struct mddev_dev
*devlist
,
1527 unsigned long long data_offset
,
1528 char *backup_file
, int verbose
, int forked
,
1529 int restart
, int freeze_reshape
);
1530 static int reshape_container(char *container
, char *devname
,
1532 struct supertype
*st
,
1533 struct mdinfo
*info
,
1535 char *backup_file
, int verbose
,
1536 int forked
, int restart
, int freeze_reshape
);
1538 int Grow_reshape(char *devname
, int fd
,
1539 struct mddev_dev
*devlist
,
1540 unsigned long long data_offset
,
1541 struct context
*c
, struct shape
*s
)
1543 /* Make some changes in the shape of an array.
1544 * The kernel must support the change.
1546 * There are three different changes. Each can trigger
1547 * a resync or recovery so we freeze that until we have
1548 * requested everything (if kernel supports freezing - 2.6.30).
1550 * - change size (i.e. component_size)
1552 * - change layout/chunksize/ndisks
1554 * The last can require a reshape. It is different on different
1555 * levels so we need to check the level before actioning it.
1556 * Some times the level change needs to be requested after the
1557 * reshape (e.g. raid6->raid5, raid5->raid0)
1560 struct mdu_array_info_s array
;
1562 struct supertype
*st
;
1563 char *subarray
= NULL
;
1567 char *container
= NULL
;
1570 struct mddev_dev
*dv
;
1576 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1577 pr_err("%s is not an active md array - aborting\n",
1581 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10
1582 && (array
.level
< 4 || array
.level
> 6)) {
1583 pr_err("--grow --data-offset not yet supported\n");
1588 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1589 pr_err("cannot change component size at the same time as other changes.\n"
1590 " Change size first, then check data is intact before making other changes.\n");
1594 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&& array
.level
> 1 &&
1595 get_linux_version() < 2006032 &&
1596 !check_env("MDADM_FORCE_FEWER")) {
1597 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1598 " Please use a newer kernel\n");
1602 st
= super_by_fd(fd
, &subarray
);
1604 pr_err("Unable to determine metadata format for %s\n", devname
);
1607 if (s
->raiddisks
> st
->max_devs
) {
1608 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1611 if (s
->level
== 0 &&
1612 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) &&
1613 !(array
.state
& (1<<MD_SB_CLUSTERED
))) {
1614 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
1615 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
1616 pr_err("failed to remove internal bitmap.\n");
1621 /* in the external case we need to check that the requested reshape is
1622 * supported, and perform an initial check that the container holds the
1623 * pre-requisite spare devices (mdmon owns final validation)
1625 if (st
->ss
->external
) {
1629 container
= st
->container_devnm
;
1630 cfd
= open_dev_excl(st
->container_devnm
);
1632 container
= st
->devnm
;
1634 cfd
= open_dev_excl(st
->devnm
);
1638 pr_err("Unable to open container for %s\n",
1644 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1647 pr_err("Cannot read superblock for %s\n",
1653 /* check if operation is supported for metadata handler */
1654 if (st
->ss
->container_content
) {
1655 struct mdinfo
*cc
= NULL
;
1656 struct mdinfo
*content
= NULL
;
1658 cc
= st
->ss
->container_content(st
, subarray
);
1659 for (content
= cc
; content
; content
= content
->next
) {
1660 int allow_reshape
= 1;
1662 /* check if reshape is allowed based on metadata
1663 * indications stored in content.array.status
1665 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
1667 if (content
->array
.state
1668 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))
1670 if (!allow_reshape
) {
1671 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1672 devname
, container
);
1680 if (mdmon_running(container
))
1681 st
->update_tail
= &st
->updates
;
1685 for (dv
= devlist
; dv
; dv
= dv
->next
)
1687 if (s
->raiddisks
> array
.raid_disks
&&
1688 array
.spare_disks
+added_disks
< (s
->raiddisks
- array
.raid_disks
) &&
1690 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1691 " Use --force to over-ride this check.\n",
1692 s
->raiddisks
- array
.raid_disks
,
1693 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1694 array
.spare_disks
+ added_disks
);
1698 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
1699 | GET_STATE
| GET_VERSION
);
1701 if (st
->ss
->external
&& subarray
== NULL
) {
1702 array
.level
= LEVEL_CONTAINER
;
1703 sra
->array
.level
= LEVEL_CONTAINER
;
1706 pr_err("failed to read sysfs parameters for %s\n",
1710 frozen
= freeze(st
);
1712 /* freeze() already spewed the reason */
1715 } else if (frozen
< 0) {
1716 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1721 /* ========= set size =============== */
1722 if (s
->size
> 0 && (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1723 unsigned long long orig_size
= get_component_size(fd
)/2;
1724 unsigned long long min_csize
;
1726 int raid0_takeover
= 0;
1729 orig_size
= (unsigned) array
.size
;
1731 if (orig_size
== 0) {
1732 pr_err("Cannot set device size in this type of array.\n");
1737 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1738 devname
, APPLY_METADATA_CHANGES
, c
->verbose
> 0)) {
1743 if (st
->ss
->external
) {
1744 /* metadata can have size limitation
1745 * update size value according to metadata information
1747 struct mdinfo
*sizeinfo
=
1748 st
->ss
->container_content(st
, subarray
);
1750 unsigned long long new_size
=
1751 sizeinfo
->custom_array_size
/2;
1752 int data_disks
= get_data_disks(
1753 sizeinfo
->array
.level
,
1754 sizeinfo
->array
.layout
,
1755 sizeinfo
->array
.raid_disks
);
1756 new_size
/= data_disks
;
1757 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size
, new_size
,
1758 new_size
* data_disks
);
1760 sysfs_free(sizeinfo
);
1764 /* Update the size of each member device in case
1765 * they have been resized. This will never reduce
1766 * below the current used-size. The "size" attribute
1767 * understands '0' to mean 'max'.
1771 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
1772 if (sysfs_set_num(sra
, mdi
, "size",
1773 s
->size
== MAX_SIZE
? 0 : s
->size
) < 0) {
1774 /* Probably kernel refusing to let us
1775 * reduce the size - not an error.
1779 if (array
.not_persistent
== 0 &&
1780 array
.major_version
== 0 &&
1781 get_linux_version() < 3001000) {
1782 /* Dangerous to allow size to exceed 2TB */
1783 unsigned long long csize
;
1784 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
1785 if (csize
>= 2ULL*1024*1024*1024)
1786 csize
= 2ULL*1024*1024*1024;
1787 if ((min_csize
== 0 || (min_csize
1794 pr_err("Cannot set size on array members.\n");
1795 goto size_change_error
;
1797 if (min_csize
&& s
->size
> min_csize
) {
1798 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1800 goto size_change_error
;
1802 if (min_csize
&& s
->size
== MAX_SIZE
) {
1803 /* Don't let the kernel choose a size - it will get
1806 pr_err("Limited v0.90 array to 2TB per device\n");
1807 s
->size
= min_csize
;
1809 if (st
->ss
->external
) {
1810 if (sra
->array
.level
== 0) {
1811 rv
= sysfs_set_str(sra
, NULL
, "level",
1815 /* get array parameters after takeover
1816 * to change one parameter at time only
1818 rv
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
1821 /* make sure mdmon is
1822 * aware of the new level */
1823 if (!mdmon_running(st
->container_devnm
))
1824 start_mdmon(st
->container_devnm
);
1825 ping_monitor(container
);
1826 if (mdmon_running(st
->container_devnm
) &&
1827 st
->update_tail
== NULL
)
1828 st
->update_tail
= &st
->updates
;
1831 if (s
->size
== MAX_SIZE
)
1833 array
.size
= s
->size
;
1834 if (s
->size
& ~INT32_MAX
) {
1835 /* got truncated to 32bit, write to
1836 * component_size instead
1839 rv
= sysfs_set_num(sra
, NULL
,
1840 "component_size", s
->size
);
1844 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1846 /* manage array size when it is managed externally
1848 if ((rv
== 0) && st
->ss
->external
)
1849 rv
= set_array_size(st
, sra
, sra
->text_version
);
1852 if (raid0_takeover
) {
1853 /* do not recync non-existing parity,
1854 * we will drop it anyway
1856 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
1857 /* go back to raid0, drop parity disk
1859 sysfs_set_str(sra
, NULL
, "level", "raid0");
1860 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1867 /* restore metadata */
1868 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1869 UnSet
, NULL
, devname
,
1870 ROLLBACK_METADATA_CHANGES
,
1873 pr_err("Cannot set device size for %s: %s\n",
1874 devname
, strerror(err
));
1876 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1877 cont_err("Bitmap must be removed before size can be changed\n");
1881 if (s
->assume_clean
) {
1882 /* This will fail on kernels older than 3.0 unless
1883 * a backport has been arranged.
1886 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
1887 pr_err("--assume-clean not supported with --grow on this kernel\n");
1889 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1890 s
->size
= get_component_size(fd
)/2;
1892 s
->size
= array
.size
;
1893 if (c
->verbose
>= 0) {
1894 if (s
->size
== orig_size
)
1895 pr_err("component size of %s unchanged at %lluK\n",
1898 pr_err("component size of %s has been set to %lluK\n",
1902 } else if (array
.level
!= LEVEL_CONTAINER
) {
1903 s
->size
= get_component_size(fd
)/2;
1905 s
->size
= array
.size
;
1908 /* See if there is anything else to do */
1909 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
1910 (s
->layout_str
== NULL
) &&
1911 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
1912 data_offset
== INVALID_SECTORS
&&
1913 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
1914 /* Nothing more to do */
1915 if (!changed
&& c
->verbose
>= 0)
1916 pr_err("%s: no change requested\n",
1921 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1922 * current implementation assumes that following conditions must be met:
1925 * - near_copies == 2
1927 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
1928 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1929 (s
->level
== 0 && array
.level
== 1 && sra
)) {
1931 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1933 dprintf("Array cannot be reshaped\n");
1939 /* Make sure mdmon has seen the device removal
1940 * and updated metadata before we continue with
1944 ping_monitor(container
);
1947 memset(&info
, 0, sizeof(info
));
1949 sysfs_init(&info
, fd
, NULL
);
1950 strcpy(info
.text_version
, sra
->text_version
);
1951 info
.component_size
= s
->size
*2;
1952 info
.new_level
= s
->level
;
1953 info
.new_chunk
= s
->chunk
* 1024;
1954 if (info
.array
.level
== LEVEL_CONTAINER
) {
1955 info
.delta_disks
= UnSet
;
1956 info
.array
.raid_disks
= s
->raiddisks
;
1957 } else if (s
->raiddisks
)
1958 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
1960 info
.delta_disks
= UnSet
;
1961 if (s
->layout_str
== NULL
) {
1962 info
.new_layout
= UnSet
;
1963 if (info
.array
.level
== 6 &&
1964 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1965 info
.array
.layout
>= 16) {
1966 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
1967 cont_err("during the reshape, please specify --layout=preserve\n");
1968 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1972 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
1973 strcmp(s
->layout_str
, "normalize") == 0) {
1974 /* If we have a -6 RAID6 layout, remove the '-6'. */
1975 info
.new_layout
= UnSet
;
1976 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1978 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1979 h
= strrchr(l
, '-');
1980 if (h
&& strcmp(h
, "-6") == 0) {
1982 info
.new_layout
= map_name(r6layout
, l
);
1985 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
1989 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
1990 /* This means that a non-standard RAID6 layout
1993 * - When reshape a RAID6 (e.g. adding a device)
1994 * which is in a non-standard layout, it is OK
1995 * to preserve that layout.
1996 * - When converting a RAID5 to RAID6, leave it in
1997 * the XXX-6 layout, don't re-layout.
1999 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2000 info
.new_layout
= info
.array
.layout
;
2001 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2003 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
2005 info
.new_layout
= map_name(r6layout
, l
);
2007 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2012 int l
= info
.new_level
;
2014 l
= info
.array
.level
;
2017 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2020 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2023 info
.new_layout
= parse_layout_10(s
->layout_str
);
2026 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2029 pr_err("layout not meaningful with this level\n");
2033 if (info
.new_layout
== UnSet
) {
2034 pr_err("layout %s not understood for this level\n",
2041 if (array
.level
== LEVEL_FAULTY
) {
2042 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2043 pr_err("cannot change level of Faulty device\n");
2047 pr_err("cannot set chunksize of Faulty device\n");
2050 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2051 pr_err("cannot set raid_disks of Faulty device\n");
2054 if (s
->layout_str
) {
2055 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2056 dprintf("Cannot get array information.\n");
2059 array
.layout
= info
.new_layout
;
2060 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2061 pr_err("failed to set new layout\n");
2063 } else if (c
->verbose
>= 0)
2064 printf("layout for %s set to %d\n",
2065 devname
, array
.layout
);
2067 } else if (array
.level
== LEVEL_CONTAINER
) {
2068 /* This change is to be applied to every array in the
2069 * container. This is only needed when the metadata imposes
2070 * restraints of the various arrays in the container.
2071 * Currently we only know that IMSM requires all arrays
2072 * to have the same number of devices so changing the
2073 * number of devices (On-Line Capacity Expansion) must be
2074 * performed at the level of the container
2080 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2081 c
->force
, c
->backup_file
, c
->verbose
, 0, 0, 0);
2084 /* get spare devices from external metadata
2086 if (st
->ss
->external
) {
2087 struct mdinfo
*info2
;
2089 info2
= st
->ss
->container_content(st
, subarray
);
2091 info
.array
.spare_disks
=
2092 info2
->array
.spare_disks
;
2097 /* Impose these changes on a single array. First
2098 * check that the metadata is OK with the change. */
2100 if (reshape_super(st
, 0, info
.new_level
,
2101 info
.new_layout
, info
.new_chunk
,
2102 info
.array
.raid_disks
, info
.delta_disks
,
2103 c
->backup_file
, devname
, APPLY_METADATA_CHANGES
,
2109 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2110 devlist
, data_offset
, c
->backup_file
, c
->verbose
,
2121 /* verify_reshape_position()
2122 * Function checks if reshape position in metadata is not farther
2123 * than position in md.
2125 * 0 : not valid sysfs entry
2126 * it can be caused by not started reshape, it should be started
2127 * by reshape array or raid0 array is before takeover
2128 * -1 : error, reshape position is obviously wrong
2129 * 1 : success, reshape progress correct or updated
2131 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2137 /* read sync_max, failure can mean raid0 array */
2138 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2142 unsigned long long position
= strtoull(buf
, &ep
, 0);
2144 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2145 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2146 position
*= get_data_disks(level
,
2148 info
->array
.raid_disks
);
2149 if (info
->reshape_progress
< position
) {
2150 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2151 info
->reshape_progress
, position
);
2152 info
->reshape_progress
= position
;
2154 } else if (info
->reshape_progress
> position
) {
2155 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2156 position
, info
->reshape_progress
);
2159 dprintf("Reshape position in md and metadata are the same;");
2163 } else if (rv
== 0) {
2164 /* for valid sysfs entry, 0-length content
2165 * should be indicated as error
2173 static unsigned long long choose_offset(unsigned long long lo
,
2174 unsigned long long hi
,
2175 unsigned long long min
,
2176 unsigned long long max
)
2178 /* Choose a new offset between hi and lo.
2179 * It must be between min and max, but
2180 * we would prefer something near the middle of hi/lo, and also
2181 * prefer to be aligned to a big power of 2.
2183 * So we start with the middle, then for each bit,
2184 * starting at '1' and increasing, if it is set, we either
2185 * add it or subtract it if possible, preferring the option
2186 * which is furthest from the boundary.
2188 * We stop once we get a 1MB alignment. As units are in sectors,
2189 * 1MB = 2*1024 sectors.
2191 unsigned long long choice
= (lo
+ hi
) / 2;
2192 unsigned long long bit
= 1;
2194 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2195 unsigned long long bigger
, smaller
;
2196 if (! (bit
& choice
))
2198 bigger
= choice
+ bit
;
2199 smaller
= choice
- bit
;
2200 if (bigger
> max
&& smaller
< min
)
2204 else if (smaller
< min
)
2206 else if (hi
- bigger
> smaller
- lo
)
2214 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2215 char *devname
, int delta_disks
,
2216 unsigned long long data_offset
,
2217 unsigned long long min
,
2223 unsigned long long before
, after
;
2225 /* Need to find min space before and after so same is used
2228 before
= UINT64_MAX
;
2230 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2234 struct supertype
*st2
;
2235 struct mdinfo info2
;
2237 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2239 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2240 dfd
= dev_open(dn
, O_RDONLY
);
2242 pr_err("%s: cannot open component %s\n",
2243 devname
, dn
? dn
: "-unknown-");
2246 st2
= dup_super(st
);
2247 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2251 pr_err("%s: cannot get superblock from %s\n",
2255 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2256 st2
->ss
->free_super(st2
);
2258 if (info2
.space_before
== 0 &&
2259 info2
.space_after
== 0) {
2260 /* Metadata doesn't support data_offset changes */
2262 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2266 if (before
> info2
.space_before
)
2267 before
= info2
.space_before
;
2268 if (after
> info2
.space_after
)
2269 after
= info2
.space_after
;
2271 if (data_offset
!= INVALID_SECTORS
) {
2273 if (info2
.data_offset
== data_offset
) {
2274 pr_err("%s: already has that data_offset\n",
2278 if (data_offset
< info2
.data_offset
)
2282 } else if ((data_offset
<= info2
.data_offset
&& dir
== 1) ||
2283 (data_offset
>= info2
.data_offset
&& dir
== -1)) {
2284 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2290 if (before
== UINT64_MAX
)
2291 /* impossible really, there must be no devices */
2294 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2295 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2296 unsigned long long new_data_offset
;
2298 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2300 if (delta_disks
< 0) {
2301 /* Don't need any space as array is shrinking
2302 * just move data_offset up by min
2304 if (data_offset
== INVALID_SECTORS
)
2305 new_data_offset
= sd
->data_offset
+ min
;
2307 if (data_offset
< sd
->data_offset
+ min
) {
2308 pr_err("--data-offset too small for %s\n",
2312 new_data_offset
= data_offset
;
2314 } else if (delta_disks
> 0) {
2315 /* need space before */
2319 pr_err("Insufficient head-space for reshape on %s\n",
2323 if (data_offset
== INVALID_SECTORS
)
2324 new_data_offset
= sd
->data_offset
- min
;
2326 if (data_offset
> sd
->data_offset
- min
) {
2327 pr_err("--data-offset too large for %s\n",
2331 new_data_offset
= data_offset
;
2335 /* can move up or down. If 'data_offset'
2336 * was set we would have already decided,
2337 * so just choose direction with most space.
2344 sysfs_set_str(sra
, NULL
, "reshape_direction",
2345 dir
== 1 ? "backwards" : "forwards");
2347 /* Increase data offset */
2351 pr_err("Insufficient tail-space for reshape on %s\n",
2355 if (data_offset
!= INVALID_SECTORS
&&
2356 data_offset
< sd
->data_offset
+ min
) {
2357 pr_err("--data-offset too small on %s\n",
2361 if (data_offset
!= INVALID_SECTORS
)
2362 new_data_offset
= data_offset
;
2364 new_data_offset
= choose_offset(sd
->data_offset
,
2365 sd
->data_offset
+ after
,
2366 sd
->data_offset
+ min
,
2367 sd
->data_offset
+ after
);
2369 /* Decrease data offset */
2373 pr_err("insufficient head-room on %s\n",
2377 if (data_offset
!= INVALID_SECTORS
&&
2378 data_offset
< sd
->data_offset
- min
) {
2379 pr_err("--data-offset too small on %s\n",
2383 if (data_offset
!= INVALID_SECTORS
)
2384 new_data_offset
= data_offset
;
2386 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2388 sd
->data_offset
- before
,
2389 sd
->data_offset
- min
);
2392 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2393 if (err
< 0 && errno
== E2BIG
) {
2394 /* try again after increasing data size to max */
2395 err
= sysfs_set_num(sra
, sd
, "size", 0);
2396 if (err
< 0 && errno
== EINVAL
&&
2397 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2398 /* some kernels have a bug where you cannot
2399 * use '0' on spare devices. */
2400 sysfs_set_num(sra
, sd
, "size",
2401 (sra
->component_size
+ after
)/2);
2403 err
= sysfs_set_num(sra
, sd
, "new_offset",
2407 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2408 pr_err("data-offset is too big for %s\n",
2412 if (sd
== sra
->devs
&&
2413 (errno
== ENOENT
|| errno
== E2BIG
))
2414 /* Early kernel, no 'new_offset' file,
2415 * or kernel doesn't like us.
2416 * For RAID5/6 this is not fatal
2419 pr_err("Cannot set new_offset for %s\n",
2428 /* Just use a backup file */
2432 static int raid10_reshape(char *container
, int fd
, char *devname
,
2433 struct supertype
*st
, struct mdinfo
*info
,
2434 struct reshape
*reshape
,
2435 unsigned long long data_offset
,
2436 int force
, int verbose
)
2438 /* Changing raid_disks, layout, chunksize or possibly
2439 * just data_offset for a RAID10.
2440 * We must always change data_offset. We change by at least
2441 * ->min_offset_change which is the largest of the old and new
2443 * If raid_disks is increasing, then data_offset must decrease
2444 * by at least this copy size.
2445 * If raid_disks is unchanged, data_offset must increase or
2446 * decrease by at least min_offset_change but preferably by much more.
2447 * We choose half of the available space.
2448 * If raid_disks is decreasing, data_offset must increase by
2449 * at least min_offset_change. To allow of this, component_size
2450 * must be decreased by the same amount.
2452 * So we calculate the required minimum and direction, possibly
2453 * reduce the component_size, then iterate through the devices
2454 * and set the new_data_offset.
2455 * If that all works, we set chunk_size, layout, raid_disks, and start
2459 unsigned long long min
;
2462 sra
= sysfs_read(fd
, NULL
,
2463 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2466 pr_err("%s: Cannot get array details from sysfs\n",
2470 min
= reshape
->min_offset_change
;
2472 if (info
->delta_disks
)
2473 sysfs_set_str(sra
, NULL
, "reshape_direction",
2474 info
->delta_disks
< 0 ? "backwards" : "forwards");
2475 if (info
->delta_disks
< 0 &&
2476 info
->space_after
< min
) {
2477 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2478 (sra
->component_size
-
2481 pr_err("cannot reduce component size\n");
2485 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
, data_offset
,
2488 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2489 cont_err("supported on this kernel\n");
2495 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2497 if (!err
&& sysfs_set_num(sra
, NULL
, "layout", reshape
->after
.layout
) < 0)
2499 if (!err
&& sysfs_set_num(sra
, NULL
, "raid_disks",
2500 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2502 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2505 pr_err("Cannot set array shape for %s\n",
2508 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2509 cont_err(" Bitmap must be removed before shape can be changed\n");
2519 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2521 struct mdinfo
*sra
, *sd
;
2522 /* Initialisation to silence compiler warning */
2523 unsigned long long min_space_before
= 0, min_space_after
= 0;
2526 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2529 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2532 struct supertype
*st2
;
2533 struct mdinfo info2
;
2535 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2537 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2538 dfd
= dev_open(dn
, O_RDONLY
);
2541 st2
= dup_super(st
);
2542 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2548 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2549 st2
->ss
->free_super(st2
);
2552 min_space_before
> info2
.space_before
)
2553 min_space_before
= info2
.space_before
;
2555 min_space_after
> info2
.space_after
)
2556 min_space_after
= info2
.space_after
;
2559 if (sd
== NULL
&& !first
) {
2560 info
->space_after
= min_space_after
;
2561 info
->space_before
= min_space_before
;
2566 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2567 struct mdinfo
*info
,
2568 int disks
, unsigned long long blocks
)
2570 /* Check that the internal stripe cache is
2571 * large enough, or it won't work.
2572 * It must hold at least 4 stripes of the larger
2575 unsigned long cache
;
2576 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2577 cache
*= 4; /* 4 stripes minimum */
2578 cache
/= 512; /* convert to sectors */
2579 /* make sure there is room for 'blocks' with a bit to spare */
2580 if (cache
< 16 + blocks
/ disks
)
2581 cache
= 16 + blocks
/ disks
;
2582 cache
/= (4096/512); /* Convert from sectors to pages */
2584 if (sra
->cache_size
< cache
)
2585 subarray_set_num(container
, sra
, "stripe_cache_size",
2589 static int impose_reshape(struct mdinfo
*sra
,
2590 struct mdinfo
*info
,
2591 struct supertype
*st
,
2594 char *devname
, char *container
,
2595 struct reshape
*reshape
)
2597 struct mdu_array_info_s array
;
2599 sra
->new_chunk
= info
->new_chunk
;
2602 /* for external metadata checkpoint saved by mdmon can be lost
2603 * or missed /due to e.g. crash/. Check if md is not during
2604 * restart farther than metadata points to.
2605 * If so, this means metadata information is obsolete.
2607 if (st
->ss
->external
)
2608 verify_reshape_position(info
, reshape
->level
);
2609 sra
->reshape_progress
= info
->reshape_progress
;
2611 sra
->reshape_progress
= 0;
2612 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2613 /* start from the end of the new array */
2614 sra
->reshape_progress
= (sra
->component_size
2615 * reshape
->after
.data_disks
);
2618 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2619 if (info
->array
.chunk_size
== info
->new_chunk
&&
2620 reshape
->before
.layout
== reshape
->after
.layout
&&
2621 st
->ss
->external
== 0) {
2622 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2623 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2625 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2628 pr_err("Cannot set device shape for %s: %s\n",
2629 devname
, strerror(errno
));
2632 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2633 cont_err("Bitmap must be removed before shape can be changed\n");
2637 } else if (!restart
) {
2638 /* set them all just in case some old 'new_*' value
2639 * persists from some earlier problem.
2642 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2644 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2645 reshape
->after
.layout
) < 0)
2647 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2648 reshape
->after
.data_disks
+
2649 reshape
->parity
) < 0)
2652 pr_err("Cannot set device shape for %s\n",
2656 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2657 cont_err("Bitmap must be removed before shape can be changed\n");
2666 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2669 struct mdu_array_info_s array
;
2671 sysfs_init(&info
, fd
, NULL
);
2673 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2675 (array
.level
>= 4 && array
.level
<= 6)) {
2676 /* To convert to RAID0 we need to fail and
2677 * remove any non-data devices. */
2680 int data_disks
= array
.raid_disks
- 1;
2681 if (array
.level
== 6)
2683 if (array
.level
== 5 &&
2684 array
.layout
!= ALGORITHM_PARITY_N
)
2686 if (array
.level
== 6 &&
2687 array
.layout
!= ALGORITHM_PARITY_N_6
)
2689 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2690 /* First remove any spares so no recovery starts */
2691 for (d
= 0, found
= 0;
2692 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2694 mdu_disk_info_t disk
;
2696 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2698 if (disk
.major
== 0 && disk
.minor
== 0)
2701 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2702 && disk
.raid_disk
< data_disks
)
2705 ioctl(fd
, HOT_REMOVE_DISK
,
2706 makedev(disk
.major
, disk
.minor
));
2708 /* Now fail anything left */
2709 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2710 for (d
= 0, found
= 0;
2711 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2714 mdu_disk_info_t disk
;
2716 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2718 if (disk
.major
== 0 && disk
.minor
== 0)
2721 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2722 && disk
.raid_disk
< data_disks
)
2725 ioctl(fd
, SET_DISK_FAULTY
,
2726 makedev(disk
.major
, disk
.minor
));
2728 while (ioctl(fd
, HOT_REMOVE_DISK
,
2729 makedev(disk
.major
, disk
.minor
)) < 0
2736 c
= map_num(pers
, level
);
2738 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2741 pr_err("%s: could not set level to %s\n",
2744 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2745 cont_err("Bitmap must be removed before level can be changed\n");
2749 pr_err("level of %s changed to %s\n",
2756 static void catch_term(int sig
)
2761 static int continue_via_systemd(char *devnm
)
2763 int skipped
, i
, pid
, status
;
2765 /* In a systemd/udev world, it is best to get systemd to
2766 * run "mdadm --grow --continue" rather than running in the
2771 /* FIXME yuk. CLOSE_EXEC?? */
2773 for (i
= 3; skipped
< 20; i
++)
2779 /* Don't want to see error messages from
2780 * systemctl. If the service doesn't exist,
2781 * we fork ourselves.
2784 open("/dev/null", O_WRONLY
);
2785 snprintf(pathbuf
, sizeof(pathbuf
), "mdadm-grow-continue@%s.service",
2787 status
= execl("/usr/bin/systemctl", "systemctl",
2790 status
= execl("/bin/systemctl", "systemctl", "start",
2793 case -1: /* Just do it ourselves. */
2795 default: /* parent - good */
2796 pid
= wait(&status
);
2797 if (pid
>= 0 && status
== 0)
2803 static int reshape_array(char *container
, int fd
, char *devname
,
2804 struct supertype
*st
, struct mdinfo
*info
,
2805 int force
, struct mddev_dev
*devlist
,
2806 unsigned long long data_offset
,
2807 char *backup_file
, int verbose
, int forked
,
2808 int restart
, int freeze_reshape
)
2810 struct reshape reshape
;
2813 int orig_level
= UnSet
;
2817 struct mdu_array_info_s array
;
2820 struct mddev_dev
*dv
;
2824 unsigned long long *offsets
= NULL
;
2828 unsigned long blocks
;
2829 unsigned long long array_size
;
2831 struct mdinfo
*sra
= NULL
;
2834 /* when reshaping a RAID0, the component_size might be zero.
2835 * So try to fix that up.
2837 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2838 dprintf("Cannot get array information.\n");
2841 if (array
.level
== 0 && info
->component_size
== 0) {
2842 get_dev_size(fd
, NULL
, &array_size
);
2843 info
->component_size
= array_size
/ array
.raid_disks
;
2846 if (array
.level
== 10)
2847 /* Need space_after info */
2848 get_space_after(fd
, st
, info
);
2850 if (info
->reshape_active
) {
2851 int new_level
= info
->new_level
;
2852 info
->new_level
= UnSet
;
2853 if (info
->delta_disks
> 0)
2854 info
->array
.raid_disks
-= info
->delta_disks
;
2855 msg
= analyse_change(devname
, info
, &reshape
);
2856 info
->new_level
= new_level
;
2857 if (info
->delta_disks
> 0)
2858 info
->array
.raid_disks
+= info
->delta_disks
;
2860 /* Make sure the array isn't read-only */
2861 ioctl(fd
, RESTART_ARRAY_RW
, 0);
2863 msg
= analyse_change(devname
, info
, &reshape
);
2865 /* if msg == "", error has already been printed */
2867 pr_err("%s\n", msg
);
2871 (reshape
.level
!= info
->array
.level
||
2872 reshape
.before
.layout
!= info
->array
.layout
||
2873 reshape
.before
.data_disks
+ reshape
.parity
2874 != info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
2875 pr_err("reshape info is not in native format - cannot continue.\n");
2879 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
2880 !((sysfs_get_str(info
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0) &&
2881 (strncmp(buf
, "reshape", 7) == 0))) {
2882 /* When reshape is restarted from '0', very begin of array
2883 * it is possible that for external metadata reshape and array
2884 * configuration doesn't happen.
2885 * Check if md has the same opinion, and reshape is restarted
2886 * from 0. If so, this is regular reshape start after reshape
2887 * switch in metadata to next array only.
2889 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
2890 (info
->reshape_progress
== 0))
2894 /* reshape already started. just skip to monitoring the reshape */
2895 if (reshape
.backup_blocks
== 0)
2897 if (restart
& RESHAPE_NO_BACKUP
)
2900 /* Need 'sra' down at 'started:' */
2901 sra
= sysfs_read(fd
, NULL
,
2902 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
2905 pr_err("%s: Cannot get array details from sysfs\n",
2911 backup_file
= locate_backup(sra
->sys_name
);
2915 /* The container is frozen but the array may not be.
2916 * So freeze the array so spares don't get put to the wrong use
2917 * FIXME there should probably be a cleaner separation between
2918 * freeze_array and freeze_container.
2920 sysfs_freeze_array(info
);
2921 /* Check we have enough spares to not be degraded */
2923 for (dv
= devlist
; dv
; dv
=dv
->next
)
2925 spares_needed
= max(reshape
.before
.data_disks
,
2926 reshape
.after
.data_disks
)
2927 + reshape
.parity
- array
.raid_disks
;
2930 info
->new_level
> 1 && info
->array
.level
> 1 &&
2931 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
2932 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2933 " Use --force to over-ride this check.\n",
2935 spares_needed
== 1 ? "" : "s",
2936 info
->array
.spare_disks
+ added_disks
);
2939 /* Check we have enough spares to not fail */
2940 spares_needed
= max(reshape
.before
.data_disks
,
2941 reshape
.after
.data_disks
)
2943 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
2944 spares_needed
> info
->array
.spare_disks
+added_disks
) {
2945 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2947 spares_needed
== 1 ? "" : "s",
2948 info
->array
.spare_disks
+ added_disks
);
2952 if (reshape
.level
!= array
.level
) {
2953 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
2956 info
->new_layout
= UnSet
; /* after level change,
2957 * layout is meaningless */
2958 orig_level
= array
.level
;
2959 sysfs_freeze_array(info
);
2961 if (reshape
.level
> 0 && st
->ss
->external
) {
2962 /* make sure mdmon is aware of the new level */
2963 if (mdmon_running(container
))
2964 flush_mdmon(container
);
2966 if (!mdmon_running(container
))
2967 start_mdmon(container
);
2968 ping_monitor(container
);
2969 if (mdmon_running(container
) &&
2970 st
->update_tail
== NULL
)
2971 st
->update_tail
= &st
->updates
;
2974 /* ->reshape_super might have chosen some spares from the
2975 * container that it wants to be part of the new array.
2976 * We can collect them with ->container_content and give
2977 * them to the kernel.
2979 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
2980 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2981 struct mdinfo
*info2
=
2982 st
->ss
->container_content(st
, subarray
);
2986 sysfs_init(info2
, fd
, st
->devnm
);
2987 /* When increasing number of devices, we need to set
2988 * new raid_disks before adding these, or they might
2991 if (reshape
.backup_blocks
&&
2992 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
2993 subarray_set_num(container
, info2
, "raid_disks",
2994 reshape
.after
.data_disks
+
2996 for (d
= info2
->devs
; d
; d
= d
->next
) {
2997 if (d
->disk
.state
== 0 &&
2998 d
->disk
.raid_disk
>= 0) {
2999 /* This is a spare that wants to
3000 * be part of the array.
3002 add_disk(fd
, st
, info2
, d
);
3008 /* We might have been given some devices to add to the
3009 * array. Now that the array has been changed to the right
3010 * level and frozen, we can safely add them.
3013 if (Manage_subdevs(devname
, fd
, devlist
, verbose
,
3018 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3019 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3020 if (reshape
.backup_blocks
== 0) {
3021 /* No restriping needed, but we might need to impose
3022 * some more changes: layout, raid_disks, chunk_size
3024 /* read current array info */
3025 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
3026 dprintf("Cannot get array information.\n");
3029 /* compare current array info with new values and if
3030 * it is different update them to new */
3031 if (info
->new_layout
!= UnSet
&&
3032 info
->new_layout
!= array
.layout
) {
3033 array
.layout
= info
->new_layout
;
3034 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
3035 pr_err("failed to set new layout\n");
3037 } else if (verbose
>= 0)
3038 printf("layout for %s set to %d\n",
3039 devname
, array
.layout
);
3041 if (info
->delta_disks
!= UnSet
&&
3042 info
->delta_disks
!= 0 &&
3043 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
3044 array
.raid_disks
+= info
->delta_disks
;
3045 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
3046 pr_err("failed to set raid disks\n");
3048 } else if (verbose
>= 0) {
3049 printf("raid_disks for %s set to %d\n",
3050 devname
, array
.raid_disks
);
3053 if (info
->new_chunk
!= 0 &&
3054 info
->new_chunk
!= array
.chunk_size
) {
3055 if (sysfs_set_num(info
, NULL
,
3056 "chunk_size", info
->new_chunk
) != 0) {
3057 pr_err("failed to set chunk size\n");
3059 } else if (verbose
>= 0)
3060 printf("chunk size for %s set to %d\n",
3061 devname
, array
.chunk_size
);
3068 * There are three possibilities.
3069 * 1/ The array will shrink.
3070 * We need to ensure the reshape will pause before reaching
3071 * the 'critical section'. We also need to fork and wait for
3072 * that to happen. When it does we
3073 * suspend/backup/complete/unfreeze
3075 * 2/ The array will not change size.
3076 * This requires that we keep a backup of a sliding window
3077 * so that we can restore data after a crash. So we need
3078 * to fork and monitor progress.
3079 * In future we will allow the data_offset to change, so
3080 * a sliding backup becomes unnecessary.
3082 * 3/ The array will grow. This is relatively easy.
3083 * However the kernel's restripe routines will cheerfully
3084 * overwrite some early data before it is safe. So we
3085 * need to make a backup of the early parts of the array
3086 * and be ready to restore it if rebuild aborts very early.
3087 * For externally managed metadata, we still need a forked
3088 * child to monitor the reshape and suspend IO over the region
3089 * that is being reshaped.
3091 * We backup data by writing it to one spare, or to a
3092 * file which was given on command line.
3094 * In each case, we first make sure that storage is available
3095 * for the required backup.
3097 * - request the shape change.
3098 * - fork to handle backup etc.
3100 /* Check that we can hold all the data */
3101 get_dev_size(fd
, NULL
, &array_size
);
3102 if (reshape
.new_size
< (array_size
/512)) {
3103 pr_err("this change will reduce the size of the array.\n"
3104 " use --grow --array-size first to truncate array.\n"
3105 " e.g. mdadm --grow %s --array-size %llu\n",
3106 devname
, reshape
.new_size
/2);
3110 if (array
.level
== 10) {
3111 /* Reshaping RAID10 does not require any data backup by
3112 * user-space. Instead it requires that the data_offset
3113 * is changed to avoid the need for backup.
3114 * So this is handled very separately
3117 /* Nothing to do. */
3119 return raid10_reshape(container
, fd
, devname
, st
, info
,
3120 &reshape
, data_offset
,
3123 sra
= sysfs_read(fd
, NULL
,
3124 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3127 pr_err("%s: Cannot get array details from sysfs\n",
3133 switch(set_new_data_offset(sra
, st
, devname
,
3134 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3136 reshape
.min_offset_change
, 1)) {
3140 /* Updated data_offset, so it's easy now */
3141 update_cache_size(container
, sra
, info
,
3142 min(reshape
.before
.data_disks
,
3143 reshape
.after
.data_disks
),
3144 reshape
.backup_blocks
);
3146 /* Right, everything seems fine. Let's kick things off.
3150 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3151 devname
, container
, &reshape
) < 0)
3153 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3155 if (errno
!= EINVAL
) {
3156 pr_err("Failed to initiate reshape!\n");
3159 /* revert data_offset and try the old way */
3160 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3161 sysfs_set_num(sra
, sd
, "new_offset",
3163 sysfs_set_str(sra
, NULL
, "reshape_direction",
3168 if (info
->new_level
== reshape
.level
)
3170 /* need to adjust level when reshape completes */
3172 case -1: /* ignore error, but don't wait */
3174 default: /* parent */
3182 fd
= open_dev(sra
->sys_name
);
3184 impose_level(fd
, info
->new_level
, devname
, verbose
);
3186 case 1: /* Couldn't set data_offset, try the old way */
3187 if (data_offset
!= INVALID_SECTORS
) {
3188 pr_err("Cannot update data_offset on this array\n");
3195 /* Decide how many blocks (sectors) for a reshape
3196 * unit. The number we have so far is just a minimum
3198 blocks
= reshape
.backup_blocks
;
3199 if (reshape
.before
.data_disks
==
3200 reshape
.after
.data_disks
) {
3201 /* Make 'blocks' bigger for better throughput, but
3202 * not so big that we reject it below.
3203 * Try for 16 megabytes
3205 while (blocks
* 32 < sra
->component_size
&&
3209 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3211 if (blocks
>= sra
->component_size
/2) {
3212 pr_err("%s: Something wrong - reshape aborted\n",
3217 /* Now we need to open all these devices so we can read/write.
3219 nrdisks
= max(reshape
.before
.data_disks
,
3220 reshape
.after
.data_disks
) + reshape
.parity
3221 + sra
->array
.spare_disks
;
3222 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3223 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3225 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3226 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
3227 nrdisks
, blocks
, backup_file
,
3232 if ((st
->ss
->manage_reshape
== NULL
) ||
3233 (st
->ss
->recover_backup
== NULL
)) {
3234 if (backup_file
== NULL
) {
3235 if (reshape
.after
.data_disks
<=
3236 reshape
.before
.data_disks
) {
3237 pr_err("%s: Cannot grow - need backup-file\n",
3239 pr_err(" Please provide one with \"--backup=...\"\n");
3241 } else if (d
== odisks
) {
3242 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3246 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3248 fdlist
+d
, offsets
+d
,
3257 update_cache_size(container
, sra
, info
,
3258 min(reshape
.before
.data_disks
, reshape
.after
.data_disks
),
3261 /* Right, everything seems fine. Let's kick things off.
3262 * If only changing raid_disks, use ioctl, else use
3267 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3268 devname
, container
, &reshape
) < 0)
3271 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3272 reshape
.after
.data_disks
);
3274 pr_err("Cannot %s reshape for %s\n",
3275 restart
? "continue" : "start",
3280 sysfs_set_str(sra
, NULL
, "array_state", "active");
3281 if (freeze_reshape
) {
3285 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3286 sra
->reshape_progress
);
3290 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3291 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3298 /* Now we just need to kick off the reshape and watch, while
3299 * handling backups of the data...
3300 * This is all done by a forked background process.
3302 switch(forked
? 0 : fork()) {
3304 pr_err("Cannot run child to monitor reshape: %s\n",
3318 /* If another array on the same devices is busy, the
3319 * reshape will wait for them. This would mean that
3320 * the first section that we suspend will stay suspended
3321 * for a long time. So check on that possibility
3322 * by looking for "DELAYED" in /proc/mdstat, and if found,
3326 struct mdstat_ent
*mds
, *m
;
3328 mds
= mdstat_read(1, 0);
3329 for (m
= mds
; m
; m
= m
->next
)
3330 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3332 m
->percent
== RESYNC_DELAYED
)
3335 /* Haven't started the reshape thread
3342 if (delayed
== 1 && get_linux_version() < 3007000) {
3343 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3344 " You might experience problems until other reshapes complete.\n");
3348 mdstat_wait(30 - (delayed
-1) * 25);
3352 if (check_env("MDADM_GROW_VERIFY"))
3353 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3356 mlockall(MCL_FUTURE
);
3358 signal(SIGTERM
, catch_term
);
3360 if (st
->ss
->external
) {
3361 /* metadata handler takes it from here */
3362 done
= st
->ss
->manage_reshape(
3363 fd
, sra
, &reshape
, st
, blocks
,
3365 d
- odisks
, fdlist
+odisks
,
3368 done
= child_monitor(
3369 fd
, sra
, &reshape
, st
, blocks
,
3371 d
- odisks
, fdlist
+odisks
,
3377 if (backup_file
&& done
) {
3379 bul
= make_backup(sra
->sys_name
);
3382 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3390 unlink(backup_file
);
3397 if (!st
->ss
->external
&&
3398 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
3399 && info
->custom_array_size
) &&
3400 info
->new_level
== reshape
.level
&&
3402 /* no need to wait for the reshape to finish as
3403 * there is nothing more to do.
3410 if (st
->ss
->external
) {
3411 /* Re-load the metadata as much could have changed */
3412 int cfd
= open_dev(st
->container_devnm
);
3414 flush_mdmon(container
);
3415 st
->ss
->free_super(st
);
3416 st
->ss
->load_container(st
, cfd
, container
);
3421 /* set new array size if required customer_array_size is used
3424 if (reshape
.before
.data_disks
!=
3425 reshape
.after
.data_disks
&&
3426 info
->custom_array_size
)
3427 set_array_size(st
, info
, info
->text_version
);
3429 if (info
->new_level
!= reshape
.level
) {
3431 fd
= open(devname
, O_RDONLY
);
3432 impose_level(fd
, info
->new_level
, devname
, verbose
);
3434 if (info
->new_level
== 0)
3435 st
->update_tail
= NULL
;
3447 if (orig_level
!= UnSet
&& sra
) {
3448 c
= map_num(pers
, orig_level
);
3449 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3450 pr_err("aborting level change\n");
3458 /* mdfd handle is passed to be closed in child process (after fork).
3460 int reshape_container(char *container
, char *devname
,
3462 struct supertype
*st
,
3463 struct mdinfo
*info
,
3465 char *backup_file
, int verbose
,
3466 int forked
, int restart
, int freeze_reshape
)
3468 struct mdinfo
*cc
= NULL
;
3470 char last_devnm
[32] = "";
3472 /* component_size is not meaningful for a container,
3473 * so pass '0' meaning 'no change'
3476 reshape_super(st
, 0, info
->new_level
,
3477 info
->new_layout
, info
->new_chunk
,
3478 info
->array
.raid_disks
, info
->delta_disks
,
3479 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3487 /* ping monitor to be sure that update is on disk
3489 ping_monitor(container
);
3491 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3492 if (continue_via_systemd(container
))
3495 switch (forked
? 0 : fork()) {
3496 case -1: /* error */
3497 perror("Cannot fork to complete reshape\n");
3500 default: /* parent */
3501 if (!freeze_reshape
)
3502 printf("%s: multi-array reshape continues in background\n", Name
);
3509 /* close unused handle in child process
3515 /* For each member array with reshape_active,
3516 * we need to perform the reshape.
3517 * We pick the first array that needs reshaping and
3518 * reshape it. reshape_array() will re-read the metadata
3519 * so the next time through a different array should be
3520 * ready for reshape.
3521 * It is possible that the 'different' array will not
3522 * be assembled yet. In that case we simple exit.
3523 * When it is assembled, the mdadm which assembles it
3524 * will take over the reshape.
3526 struct mdinfo
*content
;
3528 struct mdstat_ent
*mdstat
;
3534 cc
= st
->ss
->container_content(st
, NULL
);
3536 for (content
= cc
; content
; content
= content
->next
) {
3538 if (!content
->reshape_active
)
3541 subarray
= strchr(content
->text_version
+1, '/')+1;
3542 mdstat
= mdstat_by_subdev(subarray
, container
);
3545 if (mdstat
->active
== 0) {
3546 pr_err("Skipping inactive array %s.\n",
3548 free_mdstat(mdstat
);
3557 devid
= devnm2devid(mdstat
->devnm
);
3558 adev
= map_dev(major(devid
), minor(devid
), 0);
3560 adev
= content
->text_version
;
3562 fd
= open_dev(mdstat
->devnm
);
3564 pr_err("Device %s cannot be opened for reshape.\n", adev
);
3568 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3569 /* Do not allow for multiple reshape_array() calls for
3571 * It can happen when reshape_array() returns without
3572 * error, when reshape is not finished (wrong reshape
3573 * starting/continuation conditions). Mdmon doesn't
3574 * switch to next array in container and reentry
3575 * conditions for the same array occur.
3576 * This is possibly interim until the behaviour of
3577 * reshape_array is resolved().
3579 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3583 strcpy(last_devnm
, mdstat
->devnm
);
3585 sysfs_init(content
, fd
, mdstat
->devnm
);
3587 if (mdmon_running(container
))
3588 flush_mdmon(container
);
3590 rv
= reshape_array(container
, fd
, adev
, st
,
3591 content
, force
, NULL
, INVALID_SECTORS
,
3592 backup_file
, verbose
, 1, restart
,
3596 if (freeze_reshape
) {
3605 if (mdmon_running(container
))
3606 flush_mdmon(container
);
3615 * We run a child process in the background which performs the following
3617 * - wait for resync to reach a certain point
3618 * - suspend io to the following section
3619 * - backup that section
3620 * - allow resync to proceed further
3622 * - discard the backup.
3624 * When are combined in slightly different ways in the three cases.
3626 * - suspend/backup/allow/wait/resume/discard
3628 * - allow/wait/suspend/backup/allow/wait/resume/discard
3630 * - wait/resume/discard/suspend/backup/allow
3632 * suspend/backup/allow always come together
3633 * wait/resume/discard do too.
3634 * For the same-size case we have two backups to improve flow.
3638 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3639 unsigned long long backup_point
,
3640 unsigned long long wait_point
,
3641 unsigned long long *suspend_point
,
3642 unsigned long long *reshape_completed
, int *frozen
)
3644 /* This function is called repeatedly by the reshape manager.
3645 * It determines how much progress can safely be made and allows
3647 * - 'info' identifies the array and particularly records in
3648 * ->reshape_progress the metadata's knowledge of progress
3649 * This is a sector offset from the start of the array
3650 * of the next array block to be relocated. This number
3651 * may increase from 0 or decrease from array_size, depending
3652 * on the type of reshape that is happening.
3653 * Note that in contrast, 'sync_completed' is a block count of the
3654 * reshape so far. It gives the distance between the start point
3655 * (head or tail of device) and the next place that data will be
3656 * written. It always increases.
3657 * - 'reshape' is the structure created by analyse_change
3658 * - 'backup_point' shows how much the metadata manager has backed-up
3659 * data. For reshapes with increasing progress, it is the next address
3660 * to be backed up, previous addresses have been backed-up. For
3661 * decreasing progress, it is the earliest address that has been
3662 * backed up - later address are also backed up.
3663 * So addresses between reshape_progress and backup_point are
3664 * backed up providing those are in the 'correct' order.
3665 * - 'wait_point' is an array address. When reshape_completed
3666 * passes this point, progress_reshape should return. It might
3667 * return earlier if it determines that ->reshape_progress needs
3668 * to be updated or further backup is needed.
3669 * - suspend_point is maintained by progress_reshape and the caller
3670 * should not touch it except to initialise to zero.
3671 * It is an array address and it only increases in 2.6.37 and earlier.
3672 * This makes it difficult to handle reducing reshapes with
3673 * external metadata.
3674 * However: it is similar to backup_point in that it records the
3675 * other end of a suspended region from reshape_progress.
3676 * it is moved to extend the region that is safe to backup and/or
3678 * - reshape_completed is read from sysfs and returned. The caller
3679 * should copy this into ->reshape_progress when it has reason to
3680 * believe that the metadata knows this, and any backup outside this
3684 * 1 if more data from backup_point - but only as far as suspend_point,
3685 * should be backed up
3686 * 0 if things are progressing smoothly
3687 * -1 if the reshape is finished because it is all done,
3688 * -2 if the reshape is finished due to an error.
3691 int advancing
= (reshape
->after
.data_disks
3692 >= reshape
->before
.data_disks
);
3693 unsigned long long need_backup
; /* All data between start of array and
3694 * here will at some point need to
3697 unsigned long long read_offset
, write_offset
;
3698 unsigned long long write_range
;
3699 unsigned long long max_progress
, target
, completed
;
3700 unsigned long long array_size
= (info
->component_size
3701 * reshape
->before
.data_disks
);
3705 /* First, we unsuspend any region that is now known to be safe.
3706 * If suspend_point is on the 'wrong' side of reshape_progress, then
3707 * we don't have or need suspension at the moment. This is true for
3708 * native metadata when we don't need to back-up.
3711 if (info
->reshape_progress
<= *suspend_point
)
3712 sysfs_set_num(info
, NULL
, "suspend_lo",
3713 info
->reshape_progress
);
3715 /* Note: this won't work in 2.6.37 and before.
3716 * Something somewhere should make sure we don't need it!
3718 if (info
->reshape_progress
>= *suspend_point
)
3719 sysfs_set_num(info
, NULL
, "suspend_hi",
3720 info
->reshape_progress
);
3723 /* Now work out how far it is safe to progress.
3724 * If the read_offset for ->reshape_progress is less than
3725 * 'blocks' beyond the write_offset, we can only progress as far
3727 * Otherwise we can progress until the write_offset for the new location
3728 * reaches (within 'blocks' of) the read_offset at the current location.
3729 * However that region must be suspended unless we are using native
3731 * If we need to suspend more, we limit it to 128M per device, which is
3732 * rather arbitrary and should be some time-based calculation.
3734 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3735 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3736 write_range
= info
->new_chunk
/512;
3737 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3738 need_backup
= array_size
;
3740 need_backup
= reshape
->backup_blocks
;
3742 if (read_offset
< write_offset
+ write_range
)
3743 max_progress
= backup_point
;
3747 reshape
->after
.data_disks
;
3749 if (read_offset
> write_offset
- write_range
)
3750 /* Can only progress as far as has been backed up,
3751 * which must be suspended */
3752 max_progress
= backup_point
;
3753 else if (info
->reshape_progress
<= need_backup
)
3754 max_progress
= backup_point
;
3756 if (info
->array
.major_version
>= 0)
3757 /* Can progress until backup is needed */
3758 max_progress
= need_backup
;
3760 /* Can progress until metadata update is required */
3763 reshape
->after
.data_disks
;
3764 /* but data must be suspended */
3765 if (max_progress
< *suspend_point
)
3766 max_progress
= *suspend_point
;
3771 /* We know it is safe to progress to 'max_progress' providing
3772 * it is suspended or we are using native metadata.
3773 * Consider extending suspend_point 128M per device if it
3774 * is less than 64M per device beyond reshape_progress.
3775 * But always do a multiple of 'blocks'
3776 * FIXME this is too big - it takes to long to complete
3779 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3780 reshape
->after
.data_disks
);
3781 target
/= reshape
->backup_blocks
;
3784 target
*= reshape
->backup_blocks
;
3786 /* For externally managed metadata we always need to suspend IO to
3787 * the area being reshaped so we regularly push suspend_point forward.
3788 * For native metadata we only need the suspend if we are going to do
3792 if ((need_backup
> info
->reshape_progress
3793 || info
->array
.major_version
< 0) &&
3794 *suspend_point
< info
->reshape_progress
+ target
) {
3795 if (need_backup
< *suspend_point
+ 2 * target
)
3796 *suspend_point
= need_backup
;
3797 else if (*suspend_point
+ 2 * target
< array_size
)
3798 *suspend_point
+= 2 * target
;
3800 *suspend_point
= array_size
;
3801 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
3802 if (max_progress
> *suspend_point
)
3803 max_progress
= *suspend_point
;
3806 if (info
->array
.major_version
>= 0) {
3807 /* Only need to suspend when about to backup */
3808 if (info
->reshape_progress
< need_backup
* 2 &&
3809 *suspend_point
> 0) {
3811 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
3812 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
3815 /* Need to suspend continually */
3816 if (info
->reshape_progress
< *suspend_point
)
3817 *suspend_point
= info
->reshape_progress
;
3818 if (*suspend_point
+ target
< info
->reshape_progress
)
3819 /* No need to move suspend region yet */;
3821 if (*suspend_point
>= 2 * target
)
3822 *suspend_point
-= 2 * target
;
3825 sysfs_set_num(info
, NULL
, "suspend_lo",
3828 if (max_progress
< *suspend_point
)
3829 max_progress
= *suspend_point
;
3833 /* now set sync_max to allow that progress. sync_max, like
3834 * sync_completed is a count of sectors written per device, so
3835 * we find the difference between max_progress and the start point,
3836 * and divide that by after.data_disks to get a sync_max
3838 * At the same time we convert wait_point to a similar number
3839 * for comparing against sync_completed.
3841 /* scale down max_progress to per_disk */
3842 max_progress
/= reshape
->after
.data_disks
;
3843 /* Round to chunk size as some kernels give an erroneously high number */
3844 max_progress
/= info
->new_chunk
/512;
3845 max_progress
*= info
->new_chunk
/512;
3846 /* And round to old chunk size as the kernel wants that */
3847 max_progress
/= info
->array
.chunk_size
/512;
3848 max_progress
*= info
->array
.chunk_size
/512;
3849 /* Limit progress to the whole device */
3850 if (max_progress
> info
->component_size
)
3851 max_progress
= info
->component_size
;
3852 wait_point
/= reshape
->after
.data_disks
;
3854 /* switch from 'device offset' to 'processed block count' */
3855 max_progress
= info
->component_size
- max_progress
;
3856 wait_point
= info
->component_size
- wait_point
;
3860 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
3862 /* Now wait. If we have already reached the point that we were
3863 * asked to wait to, don't wait at all, else wait for any change.
3864 * We need to select on 'sync_completed' as that is the place that
3865 * notifications happen, but we are really interested in
3866 * 'reshape_position'
3868 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
3870 goto check_progress
;
3872 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3873 goto check_progress
;
3875 while (completed
< max_progress
&& completed
< wait_point
) {
3876 /* Check that sync_action is still 'reshape' to avoid
3877 * waiting forever on a dead array
3880 if (sysfs_get_str(info
, NULL
, "sync_action",
3882 strncmp(action
, "reshape", 7) != 0)
3884 /* Some kernels reset 'sync_completed' to zero
3885 * before setting 'sync_action' to 'idle'.
3886 * So we need these extra tests.
3888 if (completed
== 0 && advancing
3889 && strncmp(action
, "idle", 4) == 0
3890 && info
->reshape_progress
> 0)
3892 if (completed
== 0 && !advancing
3893 && strncmp(action
, "idle", 4) == 0
3894 && info
->reshape_progress
< (info
->component_size
3895 * reshape
->after
.data_disks
))
3897 sysfs_wait(fd
, NULL
);
3898 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3899 goto check_progress
;
3901 /* Some kernels reset 'sync_completed' to zero,
3902 * we need to have real point we are in md.
3903 * So in that case, read 'reshape_position' from sysfs.
3905 if (completed
== 0) {
3906 unsigned long long reshapep
;
3908 if (sysfs_get_str(info
, NULL
, "sync_action",
3910 strncmp(action
, "idle", 4) == 0 &&
3911 sysfs_get_ll(info
, NULL
,
3912 "reshape_position", &reshapep
) == 0)
3913 *reshape_completed
= reshapep
;
3915 /* some kernels can give an incorrectly high
3916 * 'completed' number, so round down */
3917 completed
/= (info
->new_chunk
/512);
3918 completed
*= (info
->new_chunk
/512);
3919 /* Convert 'completed' back in to a 'progress' number */
3920 completed
*= reshape
->after
.data_disks
;
3922 completed
= (info
->component_size
3923 * reshape
->after
.data_disks
3925 *reshape_completed
= completed
;
3930 /* We return the need_backup flag. Caller will decide
3931 * how much - a multiple of ->backup_blocks up to *suspend_point
3934 return need_backup
> info
->reshape_progress
;
3936 return need_backup
>= info
->reshape_progress
;
3939 /* if we couldn't read a number from sync_completed, then
3940 * either the reshape did complete, or it aborted.
3941 * We can tell which by checking for 'none' in reshape_position.
3942 * If it did abort, then it might immediately restart if it
3943 * it was just a device failure that leaves us degraded but
3946 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
3947 || strncmp(buf
, "none", 4) != 0) {
3948 /* The abort might only be temporary. Wait up to 10
3949 * seconds for fd to contain a valid number again.
3953 unsigned long long new_sync_max
;
3954 while (fd
>= 0 && rv
< 0 && wait
> 0) {
3955 if (sysfs_wait(fd
, &wait
) != 1)
3957 switch (sysfs_fd_get_ll(fd
, &completed
)) {
3959 /* all good again */
3961 /* If "sync_max" is no longer max_progress
3962 * we need to freeze things
3964 sysfs_get_ll(info
, NULL
, "sync_max", &new_sync_max
);
3965 *frozen
= (new_sync_max
!= max_progress
);
3967 case -2: /* read error - abort */
3974 return rv
; /* abort */
3976 /* Maybe racing with array shutdown - check state */
3979 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
3980 || strncmp(buf
, "inactive", 8) == 0
3981 || strncmp(buf
, "clear",5) == 0)
3982 return -2; /* abort */
3983 return -1; /* complete */
3987 /* FIXME return status is never checked */
3988 static int grow_backup(struct mdinfo
*sra
,
3989 unsigned long long offset
, /* per device */
3990 unsigned long stripes
, /* per device, in old chunks */
3991 int *sources
, unsigned long long *offsets
,
3992 int disks
, int chunk
, int level
, int layout
,
3993 int dests
, int *destfd
, unsigned long long *destoffsets
,
3994 int part
, int *degraded
,
3997 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3998 * to storage 'destfd' (offset 'destoffsets'), after first
3999 * suspending IO. Then allow resync to continue
4000 * over the suspended section.
4001 * Use part 'part' of the backup-super-block.
4006 unsigned long long ll
;
4008 //printf("offset %llu\n", offset);
4014 /* Check that array hasn't become degraded, else we might backup the wrong data */
4015 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4016 return -1; /* FIXME this error is ignored */
4017 new_degraded
= (int)ll
;
4018 if (new_degraded
!= *degraded
) {
4019 /* check each device to ensure it is still working */
4021 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4022 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4024 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4026 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
4027 strstr(sbuf
, "faulty") ||
4028 strstr(sbuf
, "in_sync") == NULL
) {
4029 /* this device is dead */
4030 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4031 if (sd
->disk
.raid_disk
>= 0 &&
4032 sources
[sd
->disk
.raid_disk
] >= 0) {
4033 close(sources
[sd
->disk
.raid_disk
]);
4034 sources
[sd
->disk
.raid_disk
] = -1;
4039 *degraded
= new_degraded
;
4042 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4043 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4045 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4046 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4049 bsb
.magic
[15] = '2';
4050 for (i
= 0; i
< dests
; i
++)
4052 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
4054 lseek64(destfd
[i
], destoffsets
[i
], 0);
4056 rv
= save_stripes(sources
, offsets
,
4057 disks
, chunk
, level
, layout
,
4059 offset
*512*odata
, stripes
* chunk
* odata
,
4064 bsb
.mtime
= __cpu_to_le64(time(0));
4065 for (i
= 0; i
< dests
; i
++) {
4066 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4068 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4069 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4070 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4071 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4074 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
4075 != destoffsets
[i
] - 4096)
4077 if (write(destfd
[i
], &bsb
, 512) != 512)
4079 if (destoffsets
[i
] > 4096) {
4080 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4081 destoffsets
[i
]+stripes
*chunk
*odata
)
4083 if (write(destfd
[i
], &bsb
, 512) != 512)
4093 /* in 2.6.30, the value reported by sync_completed can be
4094 * less that it should be by one stripe.
4095 * This only happens when reshape hits sync_max and pauses.
4096 * So allow wait_backup to either extent sync_max further
4097 * than strictly necessary, or return before the
4098 * sync has got quite as far as we would really like.
4099 * This is what 'blocks2' is for.
4100 * The various caller give appropriate values so that
4103 /* FIXME return value is often ignored */
4104 static int forget_backup(int dests
, int *destfd
,
4105 unsigned long long *destoffsets
,
4109 * Erase backup 'part' (which is 0 or 1)
4115 bsb
.arraystart2
= __cpu_to_le64(0);
4116 bsb
.length2
= __cpu_to_le64(0);
4118 bsb
.arraystart
= __cpu_to_le64(0);
4119 bsb
.length
= __cpu_to_le64(0);
4121 bsb
.mtime
= __cpu_to_le64(time(0));
4123 for (i
= 0; i
< dests
; i
++) {
4124 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4125 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4126 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4127 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4128 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4129 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4130 destoffsets
[i
]-4096)
4133 write(destfd
[i
], &bsb
, 512) != 512)
4140 static void fail(char *msg
)
4143 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4144 rv
|= (write(2, "\n", 1) != 1);
4148 static char *abuf
, *bbuf
;
4149 static unsigned long long abuflen
;
4150 static void validate(int afd
, int bfd
, unsigned long long offset
)
4152 /* check that the data in the backup against the array.
4153 * This is only used for regression testing and should not
4154 * be used while the array is active
4158 lseek64(bfd
, offset
- 4096, 0);
4159 if (read(bfd
, &bsb2
, 512) != 512)
4160 fail("cannot read bsb");
4161 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4162 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4163 fail("first csum bad");
4164 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4165 fail("magic is bad");
4166 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4167 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4168 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4169 fail("second csum bad");
4171 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4172 fail("devstart is wrong");
4175 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4177 if (abuflen
< len
) {
4181 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4182 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4184 /* just stop validating on mem-alloc failure */
4189 lseek64(bfd
, offset
, 0);
4190 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4191 //printf("len %llu\n", len);
4192 fail("read first backup failed");
4194 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4195 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4196 fail("read first from array failed");
4197 if (memcmp(bbuf
, abuf
, len
) != 0) {
4200 printf("offset=%llu len=%llu\n",
4201 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4202 for (i
=0; i
<len
; i
++)
4203 if (bbuf
[i
] != abuf
[i
]) {
4204 printf("first diff byte %d\n", i
);
4208 fail("data1 compare failed");
4212 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4214 if (abuflen
< len
) {
4218 abuf
= xmalloc(abuflen
);
4219 bbuf
= xmalloc(abuflen
);
4222 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4223 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4224 fail("read second backup failed");
4225 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4226 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4227 fail("read second from array failed");
4228 if (memcmp(bbuf
, abuf
, len
) != 0)
4229 fail("data2 compare failed");
4233 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4234 struct supertype
*st
, unsigned long blocks
,
4235 int *fds
, unsigned long long *offsets
,
4236 int dests
, int *destfd
, unsigned long long *destoffsets
)
4238 /* Monitor a reshape where backup is being performed using
4239 * 'native' mechanism - either to a backup file, or
4240 * to some space in a spare.
4244 unsigned long long speed
;
4245 unsigned long long suspend_point
, array_size
;
4246 unsigned long long backup_point
, wait_point
;
4247 unsigned long long reshape_completed
;
4249 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
4250 int part
= 0; /* The next part of the backup area to fill. It may already
4251 * be full, so we need to check */
4252 int level
= reshape
->level
;
4253 int layout
= reshape
->before
.layout
;
4254 int data
= reshape
->before
.data_disks
;
4255 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4256 int chunk
= sra
->array
.chunk_size
;
4258 unsigned long stripes
;
4262 /* set up the backup-super-block. This requires the
4263 * uuid from the array.
4265 /* Find a superblock */
4266 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4270 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4272 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4273 devfd
= dev_open(dn
, O_RDONLY
);
4276 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4282 pr_err("Cannot find a superblock\n");
4286 memset(&bsb
, 0, 512);
4287 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4288 st
->ss
->uuid_from_super(st
, uuid
);
4289 memcpy(bsb
.set_uuid
, uuid
, 16);
4290 bsb
.mtime
= __cpu_to_le64(time(0));
4291 bsb
.devstart2
= blocks
;
4293 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4294 reshape
->before
.data_disks
;
4296 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4297 /* Don't start the 'reshape' */
4299 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4300 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4301 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4305 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4306 backup_point
= sra
->reshape_progress
;
4309 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4310 backup_point
= reshape
->backup_blocks
;
4311 suspend_point
= array_size
;
4317 /* Want to return as soon the oldest backup slot can
4318 * be released as that allows us to start backing up
4319 * some more, providing suspend_point has been
4320 * advanced, which it should have.
4323 wait_point
= array_size
;
4324 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4325 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4326 __le64_to_cpu(bsb
.length
));
4327 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4328 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4329 __le64_to_cpu(bsb
.length2
));
4332 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4333 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4334 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4335 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4338 reshape_completed
= sra
->reshape_progress
;
4339 rv
= progress_reshape(sra
, reshape
,
4340 backup_point
, wait_point
,
4341 &suspend_point
, &reshape_completed
,
4343 /* external metadata would need to ping_monitor here */
4344 sra
->reshape_progress
= reshape_completed
;
4346 /* Clear any backup region that is before 'here' */
4348 if (__le64_to_cpu(bsb
.length
) > 0 &&
4349 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4350 __le64_to_cpu(bsb
.length
)))
4351 forget_backup(dests
, destfd
,
4353 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4354 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4355 __le64_to_cpu(bsb
.length2
)))
4356 forget_backup(dests
, destfd
,
4359 if (__le64_to_cpu(bsb
.length
) > 0 &&
4360 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4361 forget_backup(dests
, destfd
,
4363 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4364 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4365 forget_backup(dests
, destfd
,
4375 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4376 /* No longer need to monitor this reshape */
4377 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4383 unsigned long long offset
;
4384 unsigned long actual_stripes
;
4385 /* Need to backup some data.
4386 * If 'part' is not used and the desired
4387 * backup size is suspended, do a backup,
4388 * then consider the next part.
4390 /* Check that 'part' is unused */
4391 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4393 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4396 offset
= backup_point
/ data
;
4397 actual_stripes
= stripes
;
4399 if (offset
+ actual_stripes
* (chunk
/512) >
4400 sra
->component_size
)
4401 actual_stripes
= ((sra
->component_size
- offset
)
4403 if (offset
+ actual_stripes
* (chunk
/512) >
4407 if (offset
< actual_stripes
* (chunk
/512))
4408 actual_stripes
= offset
/ (chunk
/512);
4409 offset
-= actual_stripes
* (chunk
/512);
4410 if (offset
< suspend_point
/data
)
4413 if (actual_stripes
== 0)
4415 grow_backup(sra
, offset
, actual_stripes
,
4417 disks
, chunk
, level
, layout
,
4418 dests
, destfd
, destoffsets
,
4419 part
, °raded
, buf
);
4420 validate(afd
, destfd
[0], destoffsets
[0]);
4421 /* record where 'part' is up to */
4424 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4426 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4430 /* FIXME maybe call progress_reshape one more time instead */
4431 /* remove any remaining suspension */
4432 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4433 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4434 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4435 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4437 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4438 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4444 * If any spare contains md_back_data-1 which is recent wrt mtime,
4445 * write that data into the array and update the super blocks with
4446 * the new reshape_progress
4448 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
4449 char *backup_file
, int verbose
)
4453 unsigned long long *offsets
;
4454 unsigned long long nstripe
, ostripe
;
4457 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4458 if (info
->array
.level
== 6) odata
--; /* number of data disks */
4459 ndata
= info
->array
.raid_disks
- 1;
4460 if (info
->new_level
== 6) ndata
--;
4462 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4464 if (info
->delta_disks
<= 0)
4465 /* Didn't grow, so the backup file must have
4469 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4470 struct mdinfo dinfo
;
4473 char *devname
, namebuf
[20];
4474 unsigned long long lo
, hi
;
4476 /* This was a spare and may have some saved data on it.
4477 * Load the superblock, find and load the
4478 * backup_super_block.
4479 * If either fail, go on to next device.
4480 * If the backup contains no new info, just return
4481 * else restore data and update all superblocks
4483 if (i
== old_disks
-1) {
4484 fd
= open(backup_file
, O_RDONLY
);
4486 pr_err("backup file %s inaccessible: %s\n",
4487 backup_file
, strerror(errno
));
4490 devname
= backup_file
;
4495 if (st
->ss
->load_super(st
, fd
, NULL
))
4498 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4499 st
->ss
->free_super(st
);
4502 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4504 pr_err("Cannot seek on device %d\n", i
);
4505 continue; /* Cannot seek */
4507 sprintf(namebuf
, "device-%d", i
);
4510 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4512 pr_err("Cannot read from %s\n", devname
);
4513 continue; /* Cannot read */
4515 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4516 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4518 pr_err("No backup metadata on %s\n", devname
);
4521 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4523 pr_err("Bad backup-metadata checksum on %s\n", devname
);
4524 continue; /* bad checksum */
4526 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4527 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4529 pr_err("Bad backup-metadata checksum2 on %s\n", devname
);
4530 continue; /* Bad second checksum */
4532 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4534 pr_err("Wrong uuid on backup-metadata on %s\n", devname
);
4535 continue; /* Wrong uuid */
4538 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4539 * sometimes they aren't... So allow considerable flexability in matching, and allow
4540 * this test to be overridden by an environment variable.
4542 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4543 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4544 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4545 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4546 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4547 (unsigned long)info
->array
.utime
);
4549 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4550 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4551 continue; /* time stamp is too bad */
4555 if (bsb
.magic
[15] == '1') {
4556 if (bsb
.length
== 0)
4558 if (info
->delta_disks
>= 0) {
4559 /* reshape_progress is increasing */
4560 if (__le64_to_cpu(bsb
.arraystart
)
4561 + __le64_to_cpu(bsb
.length
)
4562 < info
->reshape_progress
) {
4565 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4566 continue; /* No new data here */
4569 /* reshape_progress is decreasing */
4570 if (__le64_to_cpu(bsb
.arraystart
) >=
4571 info
->reshape_progress
)
4572 goto nonew
; /* No new data here */
4575 if (bsb
.length
== 0 && bsb
.length2
== 0)
4577 if (info
->delta_disks
>= 0) {
4578 /* reshape_progress is increasing */
4579 if ((__le64_to_cpu(bsb
.arraystart
)
4580 + __le64_to_cpu(bsb
.length
)
4581 < info
->reshape_progress
)
4583 (__le64_to_cpu(bsb
.arraystart2
)
4584 + __le64_to_cpu(bsb
.length2
)
4585 < info
->reshape_progress
))
4586 goto nonew
; /* No new data here */
4588 /* reshape_progress is decreasing */
4589 if (__le64_to_cpu(bsb
.arraystart
) >=
4590 info
->reshape_progress
&&
4591 __le64_to_cpu(bsb
.arraystart2
) >=
4592 info
->reshape_progress
)
4593 goto nonew
; /* No new data here */
4596 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4599 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4601 continue; /* Cannot seek */
4603 /* There should be a duplicate backup superblock 4k before here */
4604 if (lseek64(fd
, -4096, 1) < 0 ||
4605 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4606 goto second_fail
; /* Cannot find leading superblock */
4607 if (bsb
.magic
[15] == '1')
4608 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4610 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4611 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4612 goto second_fail
; /* Cannot find leading superblock */
4614 /* Now need the data offsets for all devices. */
4615 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4616 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4619 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4620 /* FIXME should be this be an error */
4622 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4623 st
->ss
->free_super(st
);
4624 offsets
[j
] = dinfo
.data_offset
* 512;
4626 printf("%s: restoring critical section\n", Name
);
4628 if (restore_stripes(fdlist
, offsets
,
4629 info
->array
.raid_disks
,
4633 fd
, __le64_to_cpu(bsb
.devstart
)*512,
4634 __le64_to_cpu(bsb
.arraystart
)*512,
4635 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4636 /* didn't succeed, so giveup */
4638 pr_err("Error restoring backup from %s\n",
4644 if (bsb
.magic
[15] == '2' &&
4645 restore_stripes(fdlist
, offsets
,
4646 info
->array
.raid_disks
,
4650 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
4651 __le64_to_cpu(bsb
.devstart2
)*512,
4652 __le64_to_cpu(bsb
.arraystart2
)*512,
4653 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4654 /* didn't succeed, so giveup */
4656 pr_err("Error restoring second backup from %s\n",
4664 /* Ok, so the data is restored. Let's update those superblocks. */
4668 lo
= __le64_to_cpu(bsb
.arraystart
);
4669 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4671 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4672 unsigned long long lo1
, hi1
;
4673 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4674 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4678 } else if (lo
< lo1
)
4684 (info
->reshape_progress
< lo
||
4685 info
->reshape_progress
> hi
))
4686 /* backup does not affect reshape_progress*/ ;
4687 else if (info
->delta_disks
>= 0) {
4688 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4689 __le64_to_cpu(bsb
.length
);
4690 if (bsb
.magic
[15] == '2') {
4691 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
4692 __le64_to_cpu(bsb
.length2
);
4693 if (p2
> info
->reshape_progress
)
4694 info
->reshape_progress
= p2
;
4697 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4698 if (bsb
.magic
[15] == '2') {
4699 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
4700 if (p2
< info
->reshape_progress
)
4701 info
->reshape_progress
= p2
;
4704 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4707 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4709 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4710 dinfo
.reshape_progress
= info
->reshape_progress
;
4711 st
->ss
->update_super(st
, &dinfo
,
4712 "_reshape_progress",
4714 st
->ss
->store_super(st
, fdlist
[j
]);
4715 st
->ss
->free_super(st
);
4719 /* Didn't find any backup data, try to see if any
4722 if (info
->delta_disks
< 0) {
4723 /* When shrinking, the critical section is at the end.
4724 * So see if we are before the critical section.
4726 unsigned long long first_block
;
4727 nstripe
= ostripe
= 0;
4729 while (ostripe
>= nstripe
) {
4730 ostripe
+= info
->array
.chunk_size
/ 512;
4731 first_block
= ostripe
* odata
;
4732 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4733 (info
->new_chunk
/512);
4736 if (info
->reshape_progress
>= first_block
)
4739 if (info
->delta_disks
> 0) {
4740 /* See if we are beyond the critical section. */
4741 unsigned long long last_block
;
4742 nstripe
= ostripe
= 0;
4744 while (nstripe
>= ostripe
) {
4745 nstripe
+= info
->new_chunk
/ 512;
4746 last_block
= nstripe
* ndata
;
4747 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4748 (info
->array
.chunk_size
/512);
4751 if (info
->reshape_progress
>= last_block
)
4754 /* needed to recover critical section! */
4756 pr_err("Failed to find backup of critical section\n");
4760 int Grow_continue_command(char *devname
, int fd
,
4761 char *backup_file
, int verbose
)
4764 struct supertype
*st
= NULL
;
4765 struct mdinfo
*content
= NULL
;
4766 struct mdinfo array
;
4767 char *subarray
= NULL
;
4768 struct mdinfo
*cc
= NULL
;
4769 struct mdstat_ent
*mdstat
= NULL
;
4773 dprintf("Grow continue from command line called for %s\n",
4776 st
= super_by_fd(fd
, &subarray
);
4777 if (!st
|| !st
->ss
) {
4778 pr_err("Unable to determine metadata format for %s\n",
4782 dprintf("Grow continue is run for ");
4783 if (st
->ss
->external
== 0) {
4785 dprintf_cont("native array (%s)\n", devname
);
4786 if (ioctl(fd
, GET_ARRAY_INFO
, &array
.array
) < 0) {
4787 pr_err("%s is not an active md array - aborting\n", devname
);
4789 goto Grow_continue_command_exit
;
4792 /* Need to load a superblock.
4793 * FIXME we should really get what we need from
4796 for (d
= 0; d
< MAX_DISKS
; d
++) {
4797 mdu_disk_info_t disk
;
4801 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
4803 if (disk
.major
== 0 && disk
.minor
== 0)
4805 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
4807 dv
= map_dev(disk
.major
, disk
.minor
, 1);
4810 fd2
= dev_open(dv
, O_RDONLY
);
4813 err
= st
->ss
->load_super(st
, fd2
, NULL
);
4819 if (d
== MAX_DISKS
) {
4820 pr_err("Unable to load metadata for %s\n",
4823 goto Grow_continue_command_exit
;
4825 st
->ss
->getinfo_super(st
, content
, NULL
);
4830 dprintf_cont("subarray (%s)\n", subarray
);
4831 container
= st
->container_devnm
;
4832 cfd
= open_dev_excl(st
->container_devnm
);
4834 container
= st
->devnm
;
4836 cfd
= open_dev_excl(st
->devnm
);
4837 dprintf_cont("container (%s)\n", container
);
4841 pr_err("Unable to open container for %s\n", devname
);
4843 goto Grow_continue_command_exit
;
4846 /* find in container array under reshape
4848 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
4850 pr_err("Cannot read superblock for %s\n",
4853 goto Grow_continue_command_exit
;
4856 cc
= st
->ss
->container_content(st
, subarray
);
4857 for (content
= cc
; content
; content
= content
->next
) {
4859 int allow_reshape
= 1;
4861 if (content
->reshape_active
== 0)
4863 /* The decision about array or container wide
4864 * reshape is taken in Grow_continue based
4865 * content->reshape_active state, therefore we
4866 * need to check_reshape based on
4867 * reshape_active and subarray name
4869 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
4871 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
4872 (content
->array
.state
4873 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
4876 if (!allow_reshape
) {
4877 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4878 devname
, container
);
4880 goto Grow_continue_command_exit
;
4883 array
= strchr(content
->text_version
+1, '/')+1;
4884 mdstat
= mdstat_by_subdev(array
, container
);
4887 if (mdstat
->active
== 0) {
4888 pr_err("Skipping inactive array %s.\n",
4890 free_mdstat(mdstat
);
4897 pr_err("Unable to determine reshaped array for %s\n", devname
);
4899 goto Grow_continue_command_exit
;
4901 fd2
= open_dev(mdstat
->devnm
);
4903 pr_err("cannot open (%s)\n", mdstat
->devnm
);
4905 goto Grow_continue_command_exit
;
4908 sysfs_init(content
, fd2
, mdstat
->devnm
);
4912 /* start mdmon in case it is not running
4914 if (!mdmon_running(container
))
4915 start_mdmon(container
);
4916 ping_monitor(container
);
4918 if (mdmon_running(container
))
4919 st
->update_tail
= &st
->updates
;
4921 pr_err("No mdmon found. Grow cannot continue.\n");
4923 goto Grow_continue_command_exit
;
4927 /* verify that array under reshape is started from
4930 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
4932 goto Grow_continue_command_exit
;
4937 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
4939 Grow_continue_command_exit
:
4942 st
->ss
->free_super(st
);
4943 free_mdstat(mdstat
);
4950 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
4951 char *backup_file
, int forked
, int freeze_reshape
)
4955 if (!info
->reshape_active
)
4958 if (st
->ss
->external
) {
4959 int cfd
= open_dev(st
->container_devnm
);
4964 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
4966 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
4967 st
, info
, 0, backup_file
,
4969 1 | info
->reshape_active
,
4972 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
4973 NULL
, INVALID_SECTORS
,
4974 backup_file
, 0, forked
,
4975 1 | info
->reshape_active
,
4981 char *make_backup(char *name
)
4983 char *base
= "backup_file-";
4987 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
4988 fname
= xmalloc(len
);
4989 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
4993 char *locate_backup(char *name
)
4995 char *fl
= make_backup(name
);
4998 if (stat(fl
, &stb
) == 0 &&
4999 S_ISREG(stb
.st_mode
))