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");
300 if (s
->bitmap_file
&& strcmp(s
->bitmap_file
, "clustered") == 0)
301 major
= BITMAP_MAJOR_CLUSTERED
;
303 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
305 pr_err("Memory allocation failure.\n");
307 pr_err("bitmaps not supported by this kernel.\n");
310 if (bmf
.pathname
[0]) {
311 if (strcmp(s
->bitmap_file
,"none")==0) {
312 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
313 pr_err("failed to remove bitmap %s\n",
319 pr_err("%s already has a bitmap (%s)\n",
320 devname
, bmf
.pathname
);
323 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
324 pr_err("cannot get array status for %s\n", devname
);
327 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
328 if (strcmp(s
->bitmap_file
, "none")==0) {
329 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
330 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
331 pr_err("failed to remove internal bitmap.\n");
336 pr_err("%s bitmap already present on %s\n", s
->bitmap_file
, devname
);
340 if (strcmp(s
->bitmap_file
, "none") == 0) {
341 pr_err("no bitmap found on %s\n", devname
);
344 if (array
.level
<= 0) {
345 pr_err("Bitmaps not meaningful with level %s\n",
346 map_num(pers
, array
.level
)?:"of this array");
349 bitmapsize
= array
.size
;
351 if (get_dev_size(fd
, NULL
, &array_size
) &&
352 array_size
> (0x7fffffffULL
<<9)) {
353 /* Array is big enough that we cannot trust array.size
354 * try other approaches
356 bitmapsize
= get_component_size(fd
);
358 if (bitmapsize
== 0) {
359 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
363 if (array
.level
== 10) {
364 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
365 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
368 st
= super_by_fd(fd
, &subarray
);
370 pr_err("Cannot understand version %d.%d\n",
371 array
.major_version
, array
.minor_version
);
375 pr_err("Cannot add bitmaps to sub-arrays yet\n");
380 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
381 strcmp(s
->bitmap_file
, "clustered") == 0) {
384 int offset_setable
= 0;
386 if (st
->ss
->add_internal_bitmap
== NULL
) {
387 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
390 st
->nodes
= c
->nodes
;
391 st
->cluster_name
= c
->homecluster
;
392 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
395 for (d
=0; d
< st
->max_devs
; d
++) {
396 mdu_disk_info_t disk
;
399 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
401 if (disk
.major
== 0 &&
404 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
406 dv
= map_dev(disk
.major
, disk
.minor
, 1);
408 int fd2
= dev_open(dv
, O_RDWR
);
411 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
412 if (st
->ss
->add_internal_bitmap(
414 &s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
415 bitmapsize
, offset_setable
,
418 st
->ss
->write_bitmap(st
, fd2
, NoUpdate
);
420 pr_err("failed to create internal bitmap - chunksize problem.\n");
428 if (offset_setable
) {
429 st
->ss
->getinfo_super(st
, mdi
, NULL
);
430 sysfs_init(mdi
, fd
, NULL
);
431 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
434 if (strcmp(s
->bitmap_file
, "clustered") == 0)
435 array
.state
|= (1<<MD_SB_CLUSTERED
);
436 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
437 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
441 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
442 pr_err("failed to set internal bitmap.\n");
449 int max_devs
= st
->max_devs
;
451 /* try to load a superblock */
452 for (d
= 0; d
< max_devs
; d
++) {
453 mdu_disk_info_t disk
;
457 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
459 if ((disk
.major
==0 && disk
.minor
==0) ||
460 (disk
.state
& (1<<MD_DISK_REMOVED
)))
462 dv
= map_dev(disk
.major
, disk
.minor
, 1);
465 fd2
= dev_open(dv
, O_RDONLY
);
467 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
469 st
->ss
->uuid_from_super(st
, uuid
);
476 pr_err("cannot find UUID for array!\n");
479 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
, s
->bitmap_chunk
,
480 c
->delay
, s
->write_behind
, bitmapsize
, major
)) {
483 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
485 pr_err("weird: %s cannot be opened\n",
489 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
492 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
493 pr_err("Cannot set bitmap file for %s: %s\n",
494 devname
, strerror(err
));
503 * When reshaping an array we might need to backup some data.
504 * This is written to all spares with a 'super_block' describing it.
505 * The superblock goes 4K from the end of the used space on the
507 * It if written after the backup is complete.
508 * It has the following structure.
511 static struct mdp_backup_super
{
512 char magic
[16]; /* md_backup_data-1 or -2 */
515 /* start/sizes in 512byte sectors */
516 __u64 devstart
; /* address on backup device/file of data */
519 __u32 sb_csum
; /* csum of preceeding bytes. */
521 __u64 devstart2
; /* offset in to data of second section */
524 __u32 sb_csum2
; /* csum of preceeding bytes. */
526 } __attribute__((aligned(512))) bsb
, bsb2
;
528 static __u32
bsb_csum(char *buf
, int len
)
532 for (i
= 0; i
< len
; i
++)
533 csum
= (csum
<<3) + buf
[0];
534 return __cpu_to_le32(csum
);
537 static int check_idle(struct supertype
*st
)
539 /* Check that all member arrays for this container, or the
540 * container of this array, are idle
542 char *container
= (st
->container_devnm
[0]
543 ? st
->container_devnm
: st
->devnm
);
544 struct mdstat_ent
*ent
, *e
;
547 ent
= mdstat_read(0, 0);
548 for (e
= ent
; e
; e
= e
->next
) {
549 if (!is_container_member(e
, container
))
551 if (e
->percent
>= 0) {
560 static int freeze_container(struct supertype
*st
)
562 char *container
= (st
->container_devnm
[0]
563 ? st
->container_devnm
: st
->devnm
);
568 if (block_monitor(container
, 1)) {
569 pr_err("failed to freeze container\n");
576 static void unfreeze_container(struct supertype
*st
)
578 char *container
= (st
->container_devnm
[0]
579 ? st
->container_devnm
: st
->devnm
);
581 unblock_monitor(container
, 1);
584 static int freeze(struct supertype
*st
)
586 /* Try to freeze resync/rebuild on this array/container.
587 * Return -1 if the array is busy,
588 * return -2 container cannot be frozen,
589 * return 0 if this kernel doesn't support 'frozen'
590 * return 1 if it worked.
592 if (st
->ss
->external
)
593 return freeze_container(st
);
595 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
601 /* Need to clear any 'read-auto' status */
602 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
603 strncmp(buf
, "read-auto", 9) == 0)
604 sysfs_set_str(sra
, NULL
, "array_state", "clean");
606 err
= sysfs_freeze_array(sra
);
612 static void unfreeze(struct supertype
*st
)
614 if (st
->ss
->external
)
615 return unfreeze_container(st
);
617 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
621 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0
622 && strcmp(buf
, "frozen\n") == 0)
623 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
628 static void wait_reshape(struct mdinfo
*sra
)
630 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
636 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
637 strncmp(action
, "reshape", 7) == 0)
638 sysfs_wait(fd
, NULL
);
642 static int reshape_super(struct supertype
*st
, unsigned long long size
,
643 int level
, int layout
, int chunksize
, int raid_disks
,
644 int delta_disks
, char *backup_file
, char *dev
,
645 int direction
, int verbose
)
647 /* nothing extra to check in the native case */
648 if (!st
->ss
->external
)
650 if (!st
->ss
->reshape_super
||
651 !st
->ss
->manage_reshape
) {
652 pr_err("%s metadata does not support reshape\n",
657 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
658 raid_disks
, delta_disks
, backup_file
, dev
,
662 static void sync_metadata(struct supertype
*st
)
664 if (st
->ss
->external
) {
665 if (st
->update_tail
) {
666 flush_metadata_updates(st
);
667 st
->update_tail
= &st
->updates
;
669 st
->ss
->sync_metadata(st
);
673 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
675 /* when dealing with external metadata subarrays we need to be
676 * prepared to handle EAGAIN. The kernel may need to wait for
677 * mdmon to mark the array active so the kernel can handle
678 * allocations/writeback when preparing the reshape action
679 * (md_allow_write()). We temporarily disable safe_mode_delay
680 * to close a race with the array_state going clean before the
681 * next write to raid_disks / stripe_cache_size
686 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
688 (strcmp(name
, "raid_disks") != 0 &&
689 strcmp(name
, "stripe_cache_size") != 0))
690 return sysfs_set_num(sra
, NULL
, name
, n
);
692 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
695 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
696 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
697 if (rc
< 0 && errno
== EAGAIN
) {
698 ping_monitor(container
);
699 /* if we get EAGAIN here then the monitor is not active
702 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
704 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
708 int start_reshape(struct mdinfo
*sra
, int already_running
,
709 int before_data_disks
, int data_disks
)
712 unsigned long long sync_max_to_set
;
714 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
715 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
716 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
717 sra
->reshape_progress
);
718 if (before_data_disks
<= data_disks
)
719 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
721 sync_max_to_set
= (sra
->component_size
* data_disks
722 - sra
->reshape_progress
) / data_disks
;
723 if (!already_running
)
724 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
725 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
726 if (!already_running
&& err
== 0) {
729 err
= sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
732 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
737 void abort_reshape(struct mdinfo
*sra
)
739 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
740 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
741 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
742 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
743 sysfs_set_num(sra
, NULL
, "sync_min", 0);
744 // It isn't safe to reset sync_max as we aren't monitoring.
745 // Array really should be stopped at this point.
748 int remove_disks_for_takeover(struct supertype
*st
,
753 struct mdinfo
*remaining
;
756 if (sra
->array
.level
== 10)
757 nr_of_copies
= layout
& 0xff;
758 else if (sra
->array
.level
== 1)
759 nr_of_copies
= sra
->array
.raid_disks
;
763 remaining
= sra
->devs
;
765 /* for each 'copy', select one device and remove from the list. */
766 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
767 struct mdinfo
**diskp
;
770 /* Find a working device to keep */
771 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
772 struct mdinfo
*disk
= *diskp
;
774 if (disk
->disk
.raid_disk
< slot
)
776 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
778 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
780 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
782 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
785 /* We have found a good disk to use! */
787 disk
->next
= sra
->devs
;
796 if (slot
< sra
->array
.raid_disks
) {
797 /* didn't find all slots */
803 sra
->devs
= remaining
;
807 /* Remove all 'remaining' devices from the array */
809 struct mdinfo
*sd
= remaining
;
810 remaining
= sd
->next
;
812 sysfs_set_str(sra
, sd
, "state", "faulty");
813 sysfs_set_str(sra
, sd
, "slot", "none");
814 /* for external metadata disks should be removed in mdmon */
815 if (!st
->ss
->external
)
816 sysfs_set_str(sra
, sd
, "state", "remove");
817 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
818 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
819 sd
->next
= sra
->devs
;
825 void reshape_free_fdlist(int *fdlist
,
826 unsigned long long *offsets
,
831 for (i
= 0; i
< size
; i
++)
839 int reshape_prepare_fdlist(char *devname
,
843 unsigned long blocks
,
846 unsigned long long *offsets
)
852 for (d
= 0; d
<= nrdisks
; d
++)
855 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
856 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
858 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
859 sd
->disk
.raid_disk
< raid_disks
) {
860 char *dn
= map_dev(sd
->disk
.major
,
862 fdlist
[sd
->disk
.raid_disk
]
863 = dev_open(dn
, O_RDONLY
);
864 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
865 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
866 pr_err("%s: cannot open component %s\n",
867 devname
, dn
? dn
: "-unknown-");
871 } else if (backup_file
== NULL
) {
873 char *dn
= map_dev(sd
->disk
.major
,
875 fdlist
[d
] = dev_open(dn
, O_RDWR
);
876 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
878 pr_err("%s: cannot open component %s\n",
879 devname
, dn
? dn
: "-unknown-");
890 int reshape_open_backup_file(char *backup_file
,
895 unsigned long long *offsets
,
899 /* Return 1 on success, 0 on any form of failure */
900 /* need to check backup file is large enough */
906 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
910 pr_err("%s: cannot create backup file %s: %s\n",
911 devname
, backup_file
, strerror(errno
));
914 /* Guard against backup file being on array device.
915 * If array is partitioned or if LVM etc is in the
916 * way this will not notice, but it is better than
919 fstat(*fdlist
, &stb
);
922 if (stb
.st_rdev
== dev
) {
923 pr_err("backup file must NOT be on the array being reshaped.\n");
929 for (i
=0; i
< blocks
+ 8 ; i
++) {
930 if (write(*fdlist
, buf
, 512) != 512) {
931 pr_err("%s: cannot create backup file %s: %s\n",
932 devname
, backup_file
, strerror(errno
));
936 if (fsync(*fdlist
) != 0) {
937 pr_err("%s: cannot create backup file %s: %s\n",
938 devname
, backup_file
, strerror(errno
));
942 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
943 char *bu
= make_backup(sys_name
);
944 if (symlink(backup_file
, bu
))
945 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
953 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
954 unsigned int ndata
, unsigned int odata
)
956 unsigned long a
, b
, blocks
;
957 /* So how much do we need to backup.
958 * We need an amount of data which is both a whole number of
959 * old stripes and a whole number of new stripes.
960 * So LCM for (chunksize*datadisks).
962 a
= (ochunk
/512) * odata
;
963 b
= (nchunk
/512) * ndata
;
966 /* LCM == product / GCD */
967 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
972 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
974 /* Based on the current array state in info->array and
975 * the changes in info->new_* etc, determine:
976 * - whether the change is possible
977 * - Intermediate level/raid_disks/layout
978 * - whether a restriping reshape is needed
979 * - number of sectors in minimum change unit. This
980 * will cover a whole number of stripes in 'before' and
983 * Return message if the change should be rejected
984 * NULL if the change can be achieved
986 * This can be called as part of starting a reshape, or
987 * when assembling an array that is undergoing reshape.
989 int near
, far
, offset
, copies
;
991 int old_chunk
, new_chunk
;
992 /* delta_parity records change in number of devices
993 * caused by level change
995 int delta_parity
= 0;
997 memset(re
, 0, sizeof(*re
));
999 /* If a new level not explicitly given, we assume no-change */
1000 if (info
->new_level
== UnSet
)
1001 info
->new_level
= info
->array
.level
;
1003 if (info
->new_chunk
)
1004 switch (info
->new_level
) {
1010 /* chunk size is meaningful, must divide component_size
1013 if (info
->component_size
% (info
->new_chunk
/512)) {
1014 unsigned long long shrink
= info
->component_size
;
1015 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1016 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1017 info
->new_chunk
/1024, info
->component_size
/2);
1018 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1020 pr_err("will shrink the array so the given chunk size would work.\n");
1025 return "chunk size not meaningful for this level";
1028 info
->new_chunk
= info
->array
.chunk_size
;
1030 switch (info
->array
.level
) {
1032 return "No reshape is possibly for this RAID level";
1034 if (info
->delta_disks
!= UnSet
)
1035 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1037 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1039 /* RAID1 can convert to RAID1 with different disks, or
1040 * raid5 with 2 disks, or
1043 if (info
->new_level
> 1 &&
1044 (info
->component_size
& 7))
1045 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1046 if (info
->new_level
== 0) {
1047 if (info
->delta_disks
!= UnSet
&&
1048 info
->delta_disks
!= 0)
1049 return "Cannot change number of disks with RAID1->RAID0 conversion";
1051 re
->before
.data_disks
= 1;
1052 re
->after
.data_disks
= 1;
1055 if (info
->new_level
== 1) {
1056 if (info
->delta_disks
== UnSet
)
1057 /* Don't know what to do */
1058 return "no change requested for Growing RAID1";
1062 if (info
->array
.raid_disks
== 2 &&
1063 info
->new_level
== 5) {
1066 re
->before
.data_disks
= 1;
1067 if (info
->delta_disks
!= UnSet
&&
1068 info
->delta_disks
!= 0)
1069 re
->after
.data_disks
= 1 + info
->delta_disks
;
1071 re
->after
.data_disks
= 1;
1072 if (re
->after
.data_disks
< 1)
1073 return "Number of disks too small for RAID5";
1075 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1076 info
->array
.chunk_size
= 65536;
1079 /* Could do some multi-stage conversions, but leave that to
1082 return "Impossibly level change request for RAID1";
1085 /* RAID10 can be converted from near mode to
1086 * RAID0 by removing some devices.
1087 * It can also be reshaped if the kernel supports
1090 switch (info
->new_level
) {
1092 if ((info
->array
.layout
& ~0xff) != 0x100)
1093 return "Cannot Grow RAID10 with far/offset layout";
1094 /* number of devices must be multiple of number of copies */
1095 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1096 return "RAID10 layout too complex for Grow operation";
1098 new_disks
= (info
->array
.raid_disks
1099 / (info
->array
.layout
& 0xff));
1100 if (info
->delta_disks
== UnSet
)
1101 info
->delta_disks
= (new_disks
1102 - info
->array
.raid_disks
);
1104 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1105 return "New number of raid-devices impossible for RAID10";
1106 if (info
->new_chunk
&&
1107 info
->new_chunk
!= info
->array
.chunk_size
)
1108 return "Cannot change chunk-size with RAID10 Grow";
1112 re
->before
.data_disks
= new_disks
;
1113 re
->after
.data_disks
= re
->before
.data_disks
;
1117 near
= info
->array
.layout
& 0xff;
1118 far
= (info
->array
.layout
>> 8) & 0xff;
1119 offset
= info
->array
.layout
& 0x10000;
1120 if (far
> 1 && !offset
)
1121 return "Cannot reshape RAID10 in far-mode";
1122 copies
= near
* far
;
1124 old_chunk
= info
->array
.chunk_size
* far
;
1126 if (info
->new_layout
== UnSet
)
1127 info
->new_layout
= info
->array
.layout
;
1129 near
= info
->new_layout
& 0xff;
1130 far
= (info
->new_layout
>> 8) & 0xff;
1131 offset
= info
->new_layout
& 0x10000;
1132 if (far
> 1 && !offset
)
1133 return "Cannot reshape RAID10 to far-mode";
1134 if (near
* far
!= copies
)
1135 return "Cannot change number of copies when reshaping RAID10";
1137 if (info
->delta_disks
== UnSet
)
1138 info
->delta_disks
= 0;
1139 new_disks
= (info
->array
.raid_disks
+
1142 new_chunk
= info
->new_chunk
* far
;
1145 re
->before
.layout
= info
->array
.layout
;
1146 re
->before
.data_disks
= info
->array
.raid_disks
;
1147 re
->after
.layout
= info
->new_layout
;
1148 re
->after
.data_disks
= new_disks
;
1149 /* For RAID10 we don't do backup but do allow reshape,
1150 * so set backup_blocks to INVALID_SECTORS rather than
1152 * And there is no need to synchronise stripes on both
1153 * 'old' and 'new'. So the important
1154 * number is the minimum data_offset difference
1155 * which is the larger of (offset copies * chunk).
1157 re
->backup_blocks
= INVALID_SECTORS
;
1158 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1159 if (new_disks
< re
->before
.data_disks
&&
1160 info
->space_after
< re
->min_offset_change
)
1161 /* Reduce component size by one chunk */
1162 re
->new_size
= (info
->component_size
-
1163 re
->min_offset_change
);
1165 re
->new_size
= info
->component_size
;
1166 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1170 return "RAID10 can only be changed to RAID0";
1173 /* RAID0 can be converted to RAID10, or to RAID456 */
1174 if (info
->new_level
== 10) {
1175 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1176 /* Assume near=2 layout */
1177 info
->new_layout
= 0x102;
1178 info
->delta_disks
= info
->array
.raid_disks
;
1180 if (info
->new_layout
== UnSet
) {
1181 int copies
= 1 + (info
->delta_disks
1182 / info
->array
.raid_disks
);
1183 if (info
->array
.raid_disks
* (copies
-1)
1184 != info
->delta_disks
)
1185 return "Impossible number of devices for RAID0->RAID10";
1186 info
->new_layout
= 0x100 + copies
;
1188 if (info
->delta_disks
== UnSet
) {
1189 int copies
= info
->new_layout
& 0xff;
1190 if (info
->new_layout
!= 0x100 + copies
)
1191 return "New layout impossible for RAID0->RAID10";;
1192 info
->delta_disks
= (copies
- 1) *
1193 info
->array
.raid_disks
;
1195 if (info
->new_chunk
&&
1196 info
->new_chunk
!= info
->array
.chunk_size
)
1197 return "Cannot change chunk-size with RAID0->RAID10";
1200 re
->before
.data_disks
= (info
->array
.raid_disks
+
1202 re
->after
.data_disks
= re
->before
.data_disks
;
1203 re
->before
.layout
= info
->new_layout
;
1207 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1208 * a raid4 style layout of the final level.
1210 switch (info
->new_level
) {
1215 re
->before
.layout
= 0;
1220 re
->before
.layout
= ALGORITHM_PARITY_N
;
1221 if (info
->new_layout
== UnSet
)
1222 info
->new_layout
= map_name(r5layout
, "default");
1227 re
->before
.layout
= ALGORITHM_PARITY_N
;
1228 if (info
->new_layout
== UnSet
)
1229 info
->new_layout
= map_name(r6layout
, "default");
1232 return "Impossible level change requested";
1234 re
->before
.data_disks
= info
->array
.raid_disks
;
1235 /* determining 'after' layout happens outside this 'switch' */
1239 info
->array
.layout
= ALGORITHM_PARITY_N
;
1241 switch (info
->new_level
) {
1245 re
->level
= info
->array
.level
;
1246 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1247 re
->before
.layout
= info
->array
.layout
;
1251 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1252 re
->before
.layout
= info
->array
.layout
;
1257 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1258 switch (info
->array
.layout
) {
1259 case ALGORITHM_LEFT_ASYMMETRIC
:
1260 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1262 case ALGORITHM_RIGHT_ASYMMETRIC
:
1263 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1265 case ALGORITHM_LEFT_SYMMETRIC
:
1266 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1268 case ALGORITHM_RIGHT_SYMMETRIC
:
1269 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1271 case ALGORITHM_PARITY_0
:
1272 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1274 case ALGORITHM_PARITY_N
:
1275 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1278 return "Cannot convert an array with this layout";
1282 if (info
->array
.raid_disks
!= 2)
1283 return "Can only convert a 2-device array to RAID1";
1284 if (info
->delta_disks
!= UnSet
&&
1285 info
->delta_disks
!= 0)
1286 return "Cannot set raid_disk when converting RAID5->RAID1";
1288 info
->new_chunk
= 0;
1291 return "Impossible level change requested";
1295 switch (info
->new_level
) {
1301 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1302 re
->before
.layout
= info
->array
.layout
;
1305 return "Impossible level change requested";
1310 /* If we reached here then it looks like a re-stripe is
1311 * happening. We have determined the intermediate level
1312 * and initial raid_disks/layout and stored these in 're'.
1314 * We need to deduce the final layout that can be atomically
1315 * converted to the end state.
1317 switch (info
->new_level
) {
1319 /* We can only get to RAID0 from RAID4 or RAID5
1320 * with appropriate layout and one extra device
1322 if (re
->level
!= 4 && re
->level
!= 5)
1323 return "Cannot covert to RAID0 from this level";
1325 switch (re
->level
) {
1327 re
->before
.layout
= 0;
1328 re
->after
.layout
= 0;
1331 re
->after
.layout
= ALGORITHM_PARITY_N
;
1337 /* We can only get to RAID4 from RAID5 */
1338 if (re
->level
!= 4 && re
->level
!= 5)
1339 return "Cannot convert to RAID4 from this level";
1341 switch (re
->level
) {
1343 re
->after
.layout
= 0;
1346 re
->after
.layout
= ALGORITHM_PARITY_N
;
1352 /* We get to RAID5 from RAID5 or RAID6 */
1353 if (re
->level
!= 5 && re
->level
!= 6)
1354 return "Cannot convert to RAID5 from this level";
1356 switch (re
->level
) {
1358 if (info
->new_layout
== UnSet
)
1359 re
->after
.layout
= re
->before
.layout
;
1361 re
->after
.layout
= info
->new_layout
;
1364 if (info
->new_layout
== UnSet
)
1365 info
->new_layout
= re
->before
.layout
;
1367 /* after.layout needs to be raid6 version of new_layout */
1368 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1369 re
->after
.layout
= ALGORITHM_PARITY_N
;
1372 char *ls
= map_num(r5layout
, info
->new_layout
);
1375 /* Current RAID6 layout has a RAID5
1378 strcat(strcpy(layout
, ls
), "-6");
1379 l
= map_name(r6layout
, layout
);
1381 return "Cannot find RAID6 layout to convert to";
1383 /* Current RAID6 has no equivalent.
1384 * If it is already a '-6' layout we
1385 * can leave it unchanged, else we must
1388 ls
= map_num(r6layout
, info
->new_layout
);
1390 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1391 return "Please specify new layout";
1392 l
= info
->new_layout
;
1394 re
->after
.layout
= l
;
1400 /* We must already be at level 6 */
1402 return "Impossible level change";
1403 if (info
->new_layout
== UnSet
)
1404 re
->after
.layout
= info
->array
.layout
;
1406 re
->after
.layout
= info
->new_layout
;
1409 return "Impossible level change requested";
1411 if (info
->delta_disks
== UnSet
)
1412 info
->delta_disks
= delta_parity
;
1414 re
->after
.data_disks
= (re
->before
.data_disks
1417 switch (re
->level
) {
1418 case 6: re
->parity
= 2;
1421 case 5: re
->parity
= 1;
1423 default: re
->parity
= 0;
1426 /* So we have a restripe operation, we need to calculate the number
1427 * of blocks per reshape operation.
1429 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1430 if (info
->new_chunk
== 0)
1431 info
->new_chunk
= info
->array
.chunk_size
;
1432 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1433 re
->after
.layout
== re
->before
.layout
&&
1434 info
->new_chunk
== info
->array
.chunk_size
) {
1435 /* Nothing to change, can change level immediately. */
1436 re
->level
= info
->new_level
;
1437 re
->backup_blocks
= 0;
1440 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1441 /* chunk and layout changes make no difference */
1442 re
->level
= info
->new_level
;
1443 re
->backup_blocks
= 0;
1447 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1448 get_linux_version() < 2006032)
1449 return "in-place reshape is not safe before 2.6.32 - sorry.";
1451 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1452 get_linux_version() < 2006030)
1453 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1455 re
->backup_blocks
= compute_backup_blocks(
1456 info
->new_chunk
, info
->array
.chunk_size
,
1457 re
->after
.data_disks
,
1458 re
->before
.data_disks
);
1459 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1461 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1465 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1468 struct mdinfo
*info
;
1472 if ((st
== NULL
) || (sra
== NULL
))
1475 if (text_version
== NULL
)
1476 text_version
= sra
->text_version
;
1477 subarray
= strchr(text_version
+1, '/')+1;
1478 info
= st
->ss
->container_content(st
, subarray
);
1480 unsigned long long current_size
= 0;
1481 unsigned long long new_size
=
1482 info
->custom_array_size
/2;
1484 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1485 new_size
> current_size
) {
1486 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1488 dprintf("Error: Cannot set array size");
1491 dprintf("Array size changed");
1493 dprintf_cont(" from %llu to %llu.\n",
1494 current_size
, new_size
);
1498 dprintf("Error: set_array_size(): info pointer in NULL\n");
1503 static int reshape_array(char *container
, int fd
, char *devname
,
1504 struct supertype
*st
, struct mdinfo
*info
,
1505 int force
, struct mddev_dev
*devlist
,
1506 unsigned long long data_offset
,
1507 char *backup_file
, int verbose
, int forked
,
1508 int restart
, int freeze_reshape
);
1509 static int reshape_container(char *container
, char *devname
,
1511 struct supertype
*st
,
1512 struct mdinfo
*info
,
1514 char *backup_file
, int verbose
,
1515 int forked
, int restart
, int freeze_reshape
);
1517 int Grow_reshape(char *devname
, int fd
,
1518 struct mddev_dev
*devlist
,
1519 unsigned long long data_offset
,
1520 struct context
*c
, struct shape
*s
)
1522 /* Make some changes in the shape of an array.
1523 * The kernel must support the change.
1525 * There are three different changes. Each can trigger
1526 * a resync or recovery so we freeze that until we have
1527 * requested everything (if kernel supports freezing - 2.6.30).
1529 * - change size (i.e. component_size)
1531 * - change layout/chunksize/ndisks
1533 * The last can require a reshape. It is different on different
1534 * levels so we need to check the level before actioning it.
1535 * Some times the level change needs to be requested after the
1536 * reshape (e.g. raid6->raid5, raid5->raid0)
1539 struct mdu_array_info_s array
;
1541 struct supertype
*st
;
1542 char *subarray
= NULL
;
1546 char *container
= NULL
;
1549 struct mddev_dev
*dv
;
1555 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1556 pr_err("%s is not an active md array - aborting\n",
1560 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10
1561 && (array
.level
< 4 || array
.level
> 6)) {
1562 pr_err("--grow --data-offset not yet supported\n");
1567 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1568 pr_err("cannot change component size at the same time as other changes.\n"
1569 " Change size first, then check data is intact before making other changes.\n");
1573 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&& array
.level
> 1 &&
1574 get_linux_version() < 2006032 &&
1575 !check_env("MDADM_FORCE_FEWER")) {
1576 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1577 " Please use a newer kernel\n");
1581 st
= super_by_fd(fd
, &subarray
);
1583 pr_err("Unable to determine metadata format for %s\n", devname
);
1586 if (s
->raiddisks
> st
->max_devs
) {
1587 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1591 /* in the external case we need to check that the requested reshape is
1592 * supported, and perform an initial check that the container holds the
1593 * pre-requisite spare devices (mdmon owns final validation)
1595 if (st
->ss
->external
) {
1599 container
= st
->container_devnm
;
1600 cfd
= open_dev_excl(st
->container_devnm
);
1602 container
= st
->devnm
;
1604 cfd
= open_dev_excl(st
->devnm
);
1608 pr_err("Unable to open container for %s\n",
1614 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1617 pr_err("Cannot read superblock for %s\n",
1623 /* check if operation is supported for metadata handler */
1624 if (st
->ss
->container_content
) {
1625 struct mdinfo
*cc
= NULL
;
1626 struct mdinfo
*content
= NULL
;
1628 cc
= st
->ss
->container_content(st
, subarray
);
1629 for (content
= cc
; content
; content
= content
->next
) {
1630 int allow_reshape
= 1;
1632 /* check if reshape is allowed based on metadata
1633 * indications stored in content.array.status
1635 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
1637 if (content
->array
.state
1638 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))
1640 if (!allow_reshape
) {
1641 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1642 devname
, container
);
1650 if (mdmon_running(container
))
1651 st
->update_tail
= &st
->updates
;
1655 for (dv
= devlist
; dv
; dv
= dv
->next
)
1657 if (s
->raiddisks
> array
.raid_disks
&&
1658 array
.spare_disks
+added_disks
< (s
->raiddisks
- array
.raid_disks
) &&
1660 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1661 " Use --force to over-ride this check.\n",
1662 s
->raiddisks
- array
.raid_disks
,
1663 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1664 array
.spare_disks
+ added_disks
);
1668 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
1669 | GET_STATE
| GET_VERSION
);
1671 if (st
->ss
->external
&& subarray
== NULL
) {
1672 array
.level
= LEVEL_CONTAINER
;
1673 sra
->array
.level
= LEVEL_CONTAINER
;
1676 pr_err("failed to read sysfs parameters for %s\n",
1680 frozen
= freeze(st
);
1682 /* freeze() already spewed the reason */
1685 } else if (frozen
< 0) {
1686 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1691 /* ========= set size =============== */
1692 if (s
->size
> 0 && (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1693 unsigned long long orig_size
= get_component_size(fd
)/2;
1694 unsigned long long min_csize
;
1696 int raid0_takeover
= 0;
1699 orig_size
= (unsigned) array
.size
;
1701 if (orig_size
== 0) {
1702 pr_err("Cannot set device size in this type of array.\n");
1707 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1708 devname
, APPLY_METADATA_CHANGES
, c
->verbose
> 0)) {
1713 if (st
->ss
->external
) {
1714 /* metadata can have size limitation
1715 * update size value according to metadata information
1717 struct mdinfo
*sizeinfo
=
1718 st
->ss
->container_content(st
, subarray
);
1720 unsigned long long new_size
=
1721 sizeinfo
->custom_array_size
/2;
1722 int data_disks
= get_data_disks(
1723 sizeinfo
->array
.level
,
1724 sizeinfo
->array
.layout
,
1725 sizeinfo
->array
.raid_disks
);
1726 new_size
/= data_disks
;
1727 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size
, new_size
,
1728 new_size
* data_disks
);
1730 sysfs_free(sizeinfo
);
1734 /* Update the size of each member device in case
1735 * they have been resized. This will never reduce
1736 * below the current used-size. The "size" attribute
1737 * understands '0' to mean 'max'.
1741 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
1742 if (sysfs_set_num(sra
, mdi
, "size",
1743 s
->size
== MAX_SIZE
? 0 : s
->size
) < 0) {
1744 /* Probably kernel refusing to let us
1745 * reduce the size - not an error.
1749 if (array
.not_persistent
== 0 &&
1750 array
.major_version
== 0 &&
1751 get_linux_version() < 3001000) {
1752 /* Dangerous to allow size to exceed 2TB */
1753 unsigned long long csize
;
1754 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
1755 if (csize
>= 2ULL*1024*1024*1024)
1756 csize
= 2ULL*1024*1024*1024;
1757 if ((min_csize
== 0 || (min_csize
1764 pr_err("Cannot set size on array members.\n");
1765 goto size_change_error
;
1767 if (min_csize
&& s
->size
> min_csize
) {
1768 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1770 goto size_change_error
;
1772 if (min_csize
&& s
->size
== MAX_SIZE
) {
1773 /* Don't let the kernel choose a size - it will get
1776 pr_err("Limited v0.90 array to 2TB per device\n");
1777 s
->size
= min_csize
;
1779 if (st
->ss
->external
) {
1780 if (sra
->array
.level
== 0) {
1781 rv
= sysfs_set_str(sra
, NULL
, "level",
1785 /* get array parameters after takeover
1786 * to change one parameter at time only
1788 rv
= ioctl(fd
, GET_ARRAY_INFO
, &array
);
1791 /* make sure mdmon is
1792 * aware of the new level */
1793 if (!mdmon_running(st
->container_devnm
))
1794 start_mdmon(st
->container_devnm
);
1795 ping_monitor(container
);
1796 if (mdmon_running(st
->container_devnm
) &&
1797 st
->update_tail
== NULL
)
1798 st
->update_tail
= &st
->updates
;
1801 if (s
->size
== MAX_SIZE
)
1803 array
.size
= s
->size
;
1804 if (s
->size
& ~INT32_MAX
) {
1805 /* got truncated to 32bit, write to
1806 * component_size instead
1809 rv
= sysfs_set_num(sra
, NULL
,
1810 "component_size", s
->size
);
1814 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1816 /* manage array size when it is managed externally
1818 if ((rv
== 0) && st
->ss
->external
)
1819 rv
= set_array_size(st
, sra
, sra
->text_version
);
1822 if (raid0_takeover
) {
1823 /* do not recync non-existing parity,
1824 * we will drop it anyway
1826 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
1827 /* go back to raid0, drop parity disk
1829 sysfs_set_str(sra
, NULL
, "level", "raid0");
1830 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1837 /* restore metadata */
1838 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1839 UnSet
, NULL
, devname
,
1840 ROLLBACK_METADATA_CHANGES
,
1843 pr_err("Cannot set device size for %s: %s\n",
1844 devname
, strerror(err
));
1846 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1847 cont_err("Bitmap must be removed before size can be changed\n");
1851 if (s
->assume_clean
) {
1852 /* This will fail on kernels older than 3.0 unless
1853 * a backport has been arranged.
1856 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
1857 pr_err("--assume-clean not supported with --grow on this kernel\n");
1859 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1860 s
->size
= get_component_size(fd
)/2;
1862 s
->size
= array
.size
;
1863 if (c
->verbose
>= 0) {
1864 if (s
->size
== orig_size
)
1865 pr_err("component size of %s unchanged at %lluK\n",
1868 pr_err("component size of %s has been set to %lluK\n",
1872 } else if (array
.level
!= LEVEL_CONTAINER
) {
1873 s
->size
= get_component_size(fd
)/2;
1875 s
->size
= array
.size
;
1878 /* See if there is anything else to do */
1879 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
1880 (s
->layout_str
== NULL
) &&
1881 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
1882 data_offset
== INVALID_SECTORS
&&
1883 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
1884 /* Nothing more to do */
1885 if (!changed
&& c
->verbose
>= 0)
1886 pr_err("%s: no change requested\n",
1891 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1892 * current implementation assumes that following conditions must be met:
1895 * - near_copies == 2
1897 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
1898 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1899 (s
->level
== 0 && array
.level
== 1 && sra
)) {
1901 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1903 dprintf("Array cannot be reshaped\n");
1909 /* Make sure mdmon has seen the device removal
1910 * and updated metadata before we continue with
1914 ping_monitor(container
);
1917 memset(&info
, 0, sizeof(info
));
1919 sysfs_init(&info
, fd
, NULL
);
1920 strcpy(info
.text_version
, sra
->text_version
);
1921 info
.component_size
= s
->size
*2;
1922 info
.new_level
= s
->level
;
1923 info
.new_chunk
= s
->chunk
* 1024;
1924 if (info
.array
.level
== LEVEL_CONTAINER
) {
1925 info
.delta_disks
= UnSet
;
1926 info
.array
.raid_disks
= s
->raiddisks
;
1927 } else if (s
->raiddisks
)
1928 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
1930 info
.delta_disks
= UnSet
;
1931 if (s
->layout_str
== NULL
) {
1932 info
.new_layout
= UnSet
;
1933 if (info
.array
.level
== 6 &&
1934 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1935 info
.array
.layout
>= 16) {
1936 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
1937 cont_err("during the reshape, please specify --layout=preserve\n");
1938 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1942 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
1943 strcmp(s
->layout_str
, "normalize") == 0) {
1944 /* If we have a -6 RAID6 layout, remove the '-6'. */
1945 info
.new_layout
= UnSet
;
1946 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1948 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1949 h
= strrchr(l
, '-');
1950 if (h
&& strcmp(h
, "-6") == 0) {
1952 info
.new_layout
= map_name(r6layout
, l
);
1955 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
1959 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
1960 /* This means that a non-standard RAID6 layout
1963 * - When reshape a RAID6 (e.g. adding a device)
1964 * which is in a non-standard layout, it is OK
1965 * to preserve that layout.
1966 * - When converting a RAID5 to RAID6, leave it in
1967 * the XXX-6 layout, don't re-layout.
1969 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
1970 info
.new_layout
= info
.array
.layout
;
1971 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
1973 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
1975 info
.new_layout
= map_name(r6layout
, l
);
1977 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
1982 int l
= info
.new_level
;
1984 l
= info
.array
.level
;
1987 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
1990 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
1993 info
.new_layout
= parse_layout_10(s
->layout_str
);
1996 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
1999 pr_err("layout not meaningful with this level\n");
2003 if (info
.new_layout
== UnSet
) {
2004 pr_err("layout %s not understood for this level\n",
2011 if (array
.level
== LEVEL_FAULTY
) {
2012 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2013 pr_err("cannot change level of Faulty device\n");
2017 pr_err("cannot set chunksize of Faulty device\n");
2020 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2021 pr_err("cannot set raid_disks of Faulty device\n");
2024 if (s
->layout_str
) {
2025 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2026 dprintf("Cannot get array information.\n");
2029 array
.layout
= info
.new_layout
;
2030 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2031 pr_err("failed to set new layout\n");
2033 } else if (c
->verbose
>= 0)
2034 printf("layout for %s set to %d\n",
2035 devname
, array
.layout
);
2037 } else if (array
.level
== LEVEL_CONTAINER
) {
2038 /* This change is to be applied to every array in the
2039 * container. This is only needed when the metadata imposes
2040 * restraints of the various arrays in the container.
2041 * Currently we only know that IMSM requires all arrays
2042 * to have the same number of devices so changing the
2043 * number of devices (On-Line Capacity Expansion) must be
2044 * performed at the level of the container
2046 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2047 c
->force
, c
->backup_file
, c
->verbose
, 0, 0, 0);
2050 /* get spare devices from external metadata
2052 if (st
->ss
->external
) {
2053 struct mdinfo
*info2
;
2055 info2
= st
->ss
->container_content(st
, subarray
);
2057 info
.array
.spare_disks
=
2058 info2
->array
.spare_disks
;
2063 /* Impose these changes on a single array. First
2064 * check that the metadata is OK with the change. */
2066 if (reshape_super(st
, 0, info
.new_level
,
2067 info
.new_layout
, info
.new_chunk
,
2068 info
.array
.raid_disks
, info
.delta_disks
,
2069 c
->backup_file
, devname
, APPLY_METADATA_CHANGES
,
2075 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2076 devlist
, data_offset
, c
->backup_file
, c
->verbose
,
2087 /* verify_reshape_position()
2088 * Function checks if reshape position in metadata is not farther
2089 * than position in md.
2091 * 0 : not valid sysfs entry
2092 * it can be caused by not started reshape, it should be started
2093 * by reshape array or raid0 array is before takeover
2094 * -1 : error, reshape position is obviously wrong
2095 * 1 : success, reshape progress correct or updated
2097 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2103 /* read sync_max, failure can mean raid0 array */
2104 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2108 unsigned long long position
= strtoull(buf
, &ep
, 0);
2110 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2111 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2112 position
*= get_data_disks(level
,
2114 info
->array
.raid_disks
);
2115 if (info
->reshape_progress
< position
) {
2116 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2117 info
->reshape_progress
, position
);
2118 info
->reshape_progress
= position
;
2120 } else if (info
->reshape_progress
> position
) {
2121 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2122 position
, info
->reshape_progress
);
2125 dprintf("Reshape position in md and metadata are the same;");
2129 } else if (rv
== 0) {
2130 /* for valid sysfs entry, 0-length content
2131 * should be indicated as error
2139 static unsigned long long choose_offset(unsigned long long lo
,
2140 unsigned long long hi
,
2141 unsigned long long min
,
2142 unsigned long long max
)
2144 /* Choose a new offset between hi and lo.
2145 * It must be between min and max, but
2146 * we would prefer something near the middle of hi/lo, and also
2147 * prefer to be aligned to a big power of 2.
2149 * So we start with the middle, then for each bit,
2150 * starting at '1' and increasing, if it is set, we either
2151 * add it or subtract it if possible, preferring the option
2152 * which is furthest from the boundary.
2154 * We stop once we get a 1MB alignment. As units are in sectors,
2155 * 1MB = 2*1024 sectors.
2157 unsigned long long choice
= (lo
+ hi
) / 2;
2158 unsigned long long bit
= 1;
2160 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2161 unsigned long long bigger
, smaller
;
2162 if (! (bit
& choice
))
2164 bigger
= choice
+ bit
;
2165 smaller
= choice
- bit
;
2166 if (bigger
> max
&& smaller
< min
)
2170 else if (smaller
< min
)
2172 else if (hi
- bigger
> smaller
- lo
)
2180 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2181 char *devname
, int delta_disks
,
2182 unsigned long long data_offset
,
2183 unsigned long long min
,
2189 unsigned long long before
, after
;
2191 /* Need to find min space before and after so same is used
2194 before
= UINT64_MAX
;
2196 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2200 struct supertype
*st2
;
2201 struct mdinfo info2
;
2203 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2205 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2206 dfd
= dev_open(dn
, O_RDONLY
);
2208 pr_err("%s: cannot open component %s\n",
2209 devname
, dn
? dn
: "-unknown-");
2212 st2
= dup_super(st
);
2213 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2217 pr_err("%s: cannot get superblock from %s\n",
2221 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2222 st2
->ss
->free_super(st2
);
2224 if (info2
.space_before
== 0 &&
2225 info2
.space_after
== 0) {
2226 /* Metadata doesn't support data_offset changes */
2228 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2232 if (before
> info2
.space_before
)
2233 before
= info2
.space_before
;
2234 if (after
> info2
.space_after
)
2235 after
= info2
.space_after
;
2237 if (data_offset
!= INVALID_SECTORS
) {
2239 if (info2
.data_offset
== data_offset
) {
2240 pr_err("%s: already has that data_offset\n",
2244 if (data_offset
< info2
.data_offset
)
2248 } else if ((data_offset
<= info2
.data_offset
&& dir
== 1) ||
2249 (data_offset
>= info2
.data_offset
&& dir
== -1)) {
2250 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2256 if (before
== UINT64_MAX
)
2257 /* impossible really, there must be no devices */
2260 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2261 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2262 unsigned long long new_data_offset
;
2264 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2266 if (delta_disks
< 0) {
2267 /* Don't need any space as array is shrinking
2268 * just move data_offset up by min
2270 if (data_offset
== INVALID_SECTORS
)
2271 new_data_offset
= sd
->data_offset
+ min
;
2273 if (data_offset
< sd
->data_offset
+ min
) {
2274 pr_err("--data-offset too small for %s\n",
2278 new_data_offset
= data_offset
;
2280 } else if (delta_disks
> 0) {
2281 /* need space before */
2285 pr_err("Insufficient head-space for reshape on %s\n",
2289 if (data_offset
== INVALID_SECTORS
)
2290 new_data_offset
= sd
->data_offset
- min
;
2292 if (data_offset
> sd
->data_offset
- min
) {
2293 pr_err("--data-offset too large for %s\n",
2297 new_data_offset
= data_offset
;
2301 /* can move up or down. If 'data_offset'
2302 * was set we would have already decided,
2303 * so just choose direction with most space.
2310 sysfs_set_str(sra
, NULL
, "reshape_direction",
2311 dir
== 1 ? "backwards" : "forwards");
2313 /* Increase data offset */
2317 pr_err("Insufficient tail-space for reshape on %s\n",
2321 if (data_offset
!= INVALID_SECTORS
&&
2322 data_offset
< sd
->data_offset
+ min
) {
2323 pr_err("--data-offset too small on %s\n",
2327 if (data_offset
!= INVALID_SECTORS
)
2328 new_data_offset
= data_offset
;
2330 new_data_offset
= choose_offset(sd
->data_offset
,
2331 sd
->data_offset
+ after
,
2332 sd
->data_offset
+ min
,
2333 sd
->data_offset
+ after
);
2335 /* Decrease data offset */
2339 pr_err("insufficient head-room on %s\n",
2343 if (data_offset
!= INVALID_SECTORS
&&
2344 data_offset
< sd
->data_offset
- min
) {
2345 pr_err("--data-offset too small on %s\n",
2349 if (data_offset
!= INVALID_SECTORS
)
2350 new_data_offset
= data_offset
;
2352 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2354 sd
->data_offset
- before
,
2355 sd
->data_offset
- min
);
2358 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2359 if (err
< 0 && errno
== E2BIG
) {
2360 /* try again after increasing data size to max */
2361 err
= sysfs_set_num(sra
, sd
, "size", 0);
2362 if (err
< 0 && errno
== EINVAL
&&
2363 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2364 /* some kernels have a bug where you cannot
2365 * use '0' on spare devices. */
2366 sysfs_set_num(sra
, sd
, "size",
2367 (sra
->component_size
+ after
)/2);
2369 err
= sysfs_set_num(sra
, sd
, "new_offset",
2373 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2374 pr_err("data-offset is too big for %s\n",
2378 if (sd
== sra
->devs
&&
2379 (errno
== ENOENT
|| errno
== E2BIG
))
2380 /* Early kernel, no 'new_offset' file,
2381 * or kernel doesn't like us.
2382 * For RAID5/6 this is not fatal
2385 pr_err("Cannot set new_offset for %s\n",
2394 /* Just use a backup file */
2398 static int raid10_reshape(char *container
, int fd
, char *devname
,
2399 struct supertype
*st
, struct mdinfo
*info
,
2400 struct reshape
*reshape
,
2401 unsigned long long data_offset
,
2402 int force
, int verbose
)
2404 /* Changing raid_disks, layout, chunksize or possibly
2405 * just data_offset for a RAID10.
2406 * We must always change data_offset. We change by at least
2407 * ->min_offset_change which is the largest of the old and new
2409 * If raid_disks is increasing, then data_offset must decrease
2410 * by at least this copy size.
2411 * If raid_disks is unchanged, data_offset must increase or
2412 * decrease by at least min_offset_change but preferably by much more.
2413 * We choose half of the available space.
2414 * If raid_disks is decreasing, data_offset must increase by
2415 * at least min_offset_change. To allow of this, component_size
2416 * must be decreased by the same amount.
2418 * So we calculate the required minimum and direction, possibly
2419 * reduce the component_size, then iterate through the devices
2420 * and set the new_data_offset.
2421 * If that all works, we set chunk_size, layout, raid_disks, and start
2425 unsigned long long min
;
2428 sra
= sysfs_read(fd
, NULL
,
2429 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2432 pr_err("%s: Cannot get array details from sysfs\n",
2436 min
= reshape
->min_offset_change
;
2438 if (info
->delta_disks
)
2439 sysfs_set_str(sra
, NULL
, "reshape_direction",
2440 info
->delta_disks
< 0 ? "backwards" : "forwards");
2441 if (info
->delta_disks
< 0 &&
2442 info
->space_after
< min
) {
2443 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2444 (sra
->component_size
-
2447 pr_err("cannot reduce component size\n");
2451 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
, data_offset
,
2454 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2455 cont_err("supported on this kernel\n");
2461 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2463 if (!err
&& sysfs_set_num(sra
, NULL
, "layout", reshape
->after
.layout
) < 0)
2465 if (!err
&& sysfs_set_num(sra
, NULL
, "raid_disks",
2466 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2468 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2471 pr_err("Cannot set array shape for %s\n",
2474 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2475 cont_err(" Bitmap must be removed before shape can be changed\n");
2485 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2487 struct mdinfo
*sra
, *sd
;
2488 /* Initialisation to silence compiler warning */
2489 unsigned long long min_space_before
= 0, min_space_after
= 0;
2492 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2495 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2498 struct supertype
*st2
;
2499 struct mdinfo info2
;
2501 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2503 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2504 dfd
= dev_open(dn
, O_RDONLY
);
2507 st2
= dup_super(st
);
2508 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2514 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2515 st2
->ss
->free_super(st2
);
2518 min_space_before
> info2
.space_before
)
2519 min_space_before
= info2
.space_before
;
2521 min_space_after
> info2
.space_after
)
2522 min_space_after
= info2
.space_after
;
2525 if (sd
== NULL
&& !first
) {
2526 info
->space_after
= min_space_after
;
2527 info
->space_before
= min_space_before
;
2532 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2533 struct mdinfo
*info
,
2534 int disks
, unsigned long long blocks
)
2536 /* Check that the internal stripe cache is
2537 * large enough, or it won't work.
2538 * It must hold at least 4 stripes of the larger
2541 unsigned long cache
;
2542 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2543 cache
*= 4; /* 4 stripes minimum */
2544 cache
/= 512; /* convert to sectors */
2545 /* make sure there is room for 'blocks' with a bit to spare */
2546 if (cache
< 16 + blocks
/ disks
)
2547 cache
= 16 + blocks
/ disks
;
2548 cache
/= (4096/512); /* Convert from sectors to pages */
2550 if (sra
->cache_size
< cache
)
2551 subarray_set_num(container
, sra
, "stripe_cache_size",
2555 static int impose_reshape(struct mdinfo
*sra
,
2556 struct mdinfo
*info
,
2557 struct supertype
*st
,
2560 char *devname
, char *container
,
2561 struct reshape
*reshape
)
2563 struct mdu_array_info_s array
;
2565 sra
->new_chunk
= info
->new_chunk
;
2568 /* for external metadata checkpoint saved by mdmon can be lost
2569 * or missed /due to e.g. crash/. Check if md is not during
2570 * restart farther than metadata points to.
2571 * If so, this means metadata information is obsolete.
2573 if (st
->ss
->external
)
2574 verify_reshape_position(info
, reshape
->level
);
2575 sra
->reshape_progress
= info
->reshape_progress
;
2577 sra
->reshape_progress
= 0;
2578 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2579 /* start from the end of the new array */
2580 sra
->reshape_progress
= (sra
->component_size
2581 * reshape
->after
.data_disks
);
2584 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2585 if (info
->array
.chunk_size
== info
->new_chunk
&&
2586 reshape
->before
.layout
== reshape
->after
.layout
&&
2587 st
->ss
->external
== 0) {
2588 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2589 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2591 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2594 pr_err("Cannot set device shape for %s: %s\n",
2595 devname
, strerror(errno
));
2598 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2599 cont_err("Bitmap must be removed before shape can be changed\n");
2603 } else if (!restart
) {
2604 /* set them all just in case some old 'new_*' value
2605 * persists from some earlier problem.
2608 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2610 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2611 reshape
->after
.layout
) < 0)
2613 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2614 reshape
->after
.data_disks
+
2615 reshape
->parity
) < 0)
2618 pr_err("Cannot set device shape for %s\n",
2622 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2623 cont_err("Bitmap must be removed before shape can be changed\n");
2632 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2635 struct mdu_array_info_s array
;
2637 sysfs_init(&info
, fd
, NULL
);
2639 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2641 (array
.level
>= 4 && array
.level
<= 6)) {
2642 /* To convert to RAID0 we need to fail and
2643 * remove any non-data devices. */
2646 int data_disks
= array
.raid_disks
- 1;
2647 if (array
.level
== 6)
2649 if (array
.level
== 5 &&
2650 array
.layout
!= ALGORITHM_PARITY_N
)
2652 if (array
.level
== 6 &&
2653 array
.layout
!= ALGORITHM_PARITY_N_6
)
2655 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2656 /* First remove any spares so no recovery starts */
2657 for (d
= 0, found
= 0;
2658 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2660 mdu_disk_info_t disk
;
2662 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2664 if (disk
.major
== 0 && disk
.minor
== 0)
2667 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2668 && disk
.raid_disk
< data_disks
)
2671 ioctl(fd
, HOT_REMOVE_DISK
,
2672 makedev(disk
.major
, disk
.minor
));
2674 /* Now fail anything left */
2675 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2676 for (d
= 0, found
= 0;
2677 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2680 mdu_disk_info_t disk
;
2682 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
2684 if (disk
.major
== 0 && disk
.minor
== 0)
2687 if ((disk
.state
& (1 << MD_DISK_ACTIVE
))
2688 && disk
.raid_disk
< data_disks
)
2691 ioctl(fd
, SET_DISK_FAULTY
,
2692 makedev(disk
.major
, disk
.minor
));
2694 while (ioctl(fd
, HOT_REMOVE_DISK
,
2695 makedev(disk
.major
, disk
.minor
)) < 0
2702 c
= map_num(pers
, level
);
2704 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2707 pr_err("%s: could not set level to %s\n",
2710 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2711 cont_err("Bitmap must be removed before level can be changed\n");
2715 pr_err("level of %s changed to %s\n",
2722 static void catch_term(int sig
)
2727 static int continue_via_systemd(char *devnm
)
2729 int skipped
, i
, pid
, status
;
2731 /* In a systemd/udev world, it is best to get systemd to
2732 * run "mdadm --grow --continue" rather than running in the
2737 /* FIXME yuk. CLOSE_EXEC?? */
2739 for (i
= 3; skipped
< 20; i
++)
2745 /* Don't want to see error messages from
2746 * systemctl. If the service doesn't exist,
2747 * we fork ourselves.
2750 open("/dev/null", O_WRONLY
);
2751 snprintf(pathbuf
, sizeof(pathbuf
), "mdadm-grow-continue@%s.service",
2753 status
= execl("/usr/bin/systemctl", "systemctl",
2756 status
= execl("/bin/systemctl", "systemctl", "start",
2759 case -1: /* Just do it ourselves. */
2761 default: /* parent - good */
2762 pid
= wait(&status
);
2763 if (pid
>= 0 && status
== 0)
2769 static int reshape_array(char *container
, int fd
, char *devname
,
2770 struct supertype
*st
, struct mdinfo
*info
,
2771 int force
, struct mddev_dev
*devlist
,
2772 unsigned long long data_offset
,
2773 char *backup_file
, int verbose
, int forked
,
2774 int restart
, int freeze_reshape
)
2776 struct reshape reshape
;
2779 int orig_level
= UnSet
;
2783 struct mdu_array_info_s array
;
2786 struct mddev_dev
*dv
;
2790 unsigned long long *offsets
= NULL
;
2794 unsigned long blocks
;
2795 unsigned long long array_size
;
2797 struct mdinfo
*sra
= NULL
;
2800 /* when reshaping a RAID0, the component_size might be zero.
2801 * So try to fix that up.
2803 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2804 dprintf("Cannot get array information.\n");
2807 if (array
.level
== 0 && info
->component_size
== 0) {
2808 get_dev_size(fd
, NULL
, &array_size
);
2809 info
->component_size
= array_size
/ array
.raid_disks
;
2812 if (array
.level
== 10)
2813 /* Need space_after info */
2814 get_space_after(fd
, st
, info
);
2816 if (info
->reshape_active
) {
2817 int new_level
= info
->new_level
;
2818 info
->new_level
= UnSet
;
2819 if (info
->delta_disks
> 0)
2820 info
->array
.raid_disks
-= info
->delta_disks
;
2821 msg
= analyse_change(devname
, info
, &reshape
);
2822 info
->new_level
= new_level
;
2823 if (info
->delta_disks
> 0)
2824 info
->array
.raid_disks
+= info
->delta_disks
;
2826 /* Make sure the array isn't read-only */
2827 ioctl(fd
, RESTART_ARRAY_RW
, 0);
2829 msg
= analyse_change(devname
, info
, &reshape
);
2831 /* if msg == "", error has already been printed */
2833 pr_err("%s\n", msg
);
2837 (reshape
.level
!= info
->array
.level
||
2838 reshape
.before
.layout
!= info
->array
.layout
||
2839 reshape
.before
.data_disks
+ reshape
.parity
2840 != info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
2841 pr_err("reshape info is not in native format - cannot continue.\n");
2845 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
2846 !((sysfs_get_str(info
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0) &&
2847 (strncmp(buf
, "reshape", 7) == 0))) {
2848 /* When reshape is restarted from '0', very begin of array
2849 * it is possible that for external metadata reshape and array
2850 * configuration doesn't happen.
2851 * Check if md has the same opinion, and reshape is restarted
2852 * from 0. If so, this is regular reshape start after reshape
2853 * switch in metadata to next array only.
2855 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
2856 (info
->reshape_progress
== 0))
2860 /* reshape already started. just skip to monitoring the reshape */
2861 if (reshape
.backup_blocks
== 0)
2863 if (restart
& RESHAPE_NO_BACKUP
)
2866 /* Need 'sra' down at 'started:' */
2867 sra
= sysfs_read(fd
, NULL
,
2868 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
2871 pr_err("%s: Cannot get array details from sysfs\n",
2877 backup_file
= locate_backup(sra
->sys_name
);
2881 /* The container is frozen but the array may not be.
2882 * So freeze the array so spares don't get put to the wrong use
2883 * FIXME there should probably be a cleaner separation between
2884 * freeze_array and freeze_container.
2886 sysfs_freeze_array(info
);
2887 /* Check we have enough spares to not be degraded */
2889 for (dv
= devlist
; dv
; dv
=dv
->next
)
2891 spares_needed
= max(reshape
.before
.data_disks
,
2892 reshape
.after
.data_disks
)
2893 + reshape
.parity
- array
.raid_disks
;
2896 info
->new_level
> 1 && info
->array
.level
> 1 &&
2897 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
2898 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2899 " Use --force to over-ride this check.\n",
2901 spares_needed
== 1 ? "" : "s",
2902 info
->array
.spare_disks
+ added_disks
);
2905 /* Check we have enough spares to not fail */
2906 spares_needed
= max(reshape
.before
.data_disks
,
2907 reshape
.after
.data_disks
)
2909 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
2910 spares_needed
> info
->array
.spare_disks
+added_disks
) {
2911 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2913 spares_needed
== 1 ? "" : "s",
2914 info
->array
.spare_disks
+ added_disks
);
2918 if (reshape
.level
!= array
.level
) {
2919 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
2922 info
->new_layout
= UnSet
; /* after level change,
2923 * layout is meaningless */
2924 orig_level
= array
.level
;
2925 sysfs_freeze_array(info
);
2927 if (reshape
.level
> 0 && st
->ss
->external
) {
2928 /* make sure mdmon is aware of the new level */
2929 if (mdmon_running(container
))
2930 flush_mdmon(container
);
2932 if (!mdmon_running(container
))
2933 start_mdmon(container
);
2934 ping_monitor(container
);
2935 if (mdmon_running(container
) &&
2936 st
->update_tail
== NULL
)
2937 st
->update_tail
= &st
->updates
;
2940 /* ->reshape_super might have chosen some spares from the
2941 * container that it wants to be part of the new array.
2942 * We can collect them with ->container_content and give
2943 * them to the kernel.
2945 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
2946 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2947 struct mdinfo
*info2
=
2948 st
->ss
->container_content(st
, subarray
);
2952 sysfs_init(info2
, fd
, st
->devnm
);
2953 /* When increasing number of devices, we need to set
2954 * new raid_disks before adding these, or they might
2957 if (reshape
.backup_blocks
&&
2958 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
2959 subarray_set_num(container
, info2
, "raid_disks",
2960 reshape
.after
.data_disks
+
2962 for (d
= info2
->devs
; d
; d
= d
->next
) {
2963 if (d
->disk
.state
== 0 &&
2964 d
->disk
.raid_disk
>= 0) {
2965 /* This is a spare that wants to
2966 * be part of the array.
2968 add_disk(fd
, st
, info2
, d
);
2974 /* We might have been given some devices to add to the
2975 * array. Now that the array has been changed to the right
2976 * level and frozen, we can safely add them.
2979 Manage_subdevs(devname
, fd
, devlist
, verbose
,
2982 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
2983 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
2984 if (reshape
.backup_blocks
== 0) {
2985 /* No restriping needed, but we might need to impose
2986 * some more changes: layout, raid_disks, chunk_size
2988 /* read current array info */
2989 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2990 dprintf("Cannot get array information.\n");
2993 /* compare current array info with new values and if
2994 * it is different update them to new */
2995 if (info
->new_layout
!= UnSet
&&
2996 info
->new_layout
!= array
.layout
) {
2997 array
.layout
= info
->new_layout
;
2998 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2999 pr_err("failed to set new layout\n");
3001 } else if (verbose
>= 0)
3002 printf("layout for %s set to %d\n",
3003 devname
, array
.layout
);
3005 if (info
->delta_disks
!= UnSet
&&
3006 info
->delta_disks
!= 0 &&
3007 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
3008 array
.raid_disks
+= info
->delta_disks
;
3009 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
3010 pr_err("failed to set raid disks\n");
3012 } else if (verbose
>= 0) {
3013 printf("raid_disks for %s set to %d\n",
3014 devname
, array
.raid_disks
);
3017 if (info
->new_chunk
!= 0 &&
3018 info
->new_chunk
!= array
.chunk_size
) {
3019 if (sysfs_set_num(info
, NULL
,
3020 "chunk_size", info
->new_chunk
) != 0) {
3021 pr_err("failed to set chunk size\n");
3023 } else if (verbose
>= 0)
3024 printf("chunk size for %s set to %d\n",
3025 devname
, array
.chunk_size
);
3032 * There are three possibilities.
3033 * 1/ The array will shrink.
3034 * We need to ensure the reshape will pause before reaching
3035 * the 'critical section'. We also need to fork and wait for
3036 * that to happen. When it does we
3037 * suspend/backup/complete/unfreeze
3039 * 2/ The array will not change size.
3040 * This requires that we keep a backup of a sliding window
3041 * so that we can restore data after a crash. So we need
3042 * to fork and monitor progress.
3043 * In future we will allow the data_offset to change, so
3044 * a sliding backup becomes unnecessary.
3046 * 3/ The array will grow. This is relatively easy.
3047 * However the kernel's restripe routines will cheerfully
3048 * overwrite some early data before it is safe. So we
3049 * need to make a backup of the early parts of the array
3050 * and be ready to restore it if rebuild aborts very early.
3051 * For externally managed metadata, we still need a forked
3052 * child to monitor the reshape and suspend IO over the region
3053 * that is being reshaped.
3055 * We backup data by writing it to one spare, or to a
3056 * file which was given on command line.
3058 * In each case, we first make sure that storage is available
3059 * for the required backup.
3061 * - request the shape change.
3062 * - fork to handle backup etc.
3064 /* Check that we can hold all the data */
3065 get_dev_size(fd
, NULL
, &array_size
);
3066 if (reshape
.new_size
< (array_size
/512)) {
3067 pr_err("this change will reduce the size of the array.\n"
3068 " use --grow --array-size first to truncate array.\n"
3069 " e.g. mdadm --grow %s --array-size %llu\n",
3070 devname
, reshape
.new_size
/2);
3074 if (array
.level
== 10) {
3075 /* Reshaping RAID10 does not require any data backup by
3076 * user-space. Instead it requires that the data_offset
3077 * is changed to avoid the need for backup.
3078 * So this is handled very separately
3081 /* Nothing to do. */
3083 return raid10_reshape(container
, fd
, devname
, st
, info
,
3084 &reshape
, data_offset
,
3087 sra
= sysfs_read(fd
, NULL
,
3088 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3091 pr_err("%s: Cannot get array details from sysfs\n",
3097 switch(set_new_data_offset(sra
, st
, devname
,
3098 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3100 reshape
.min_offset_change
, 1)) {
3104 /* Updated data_offset, so it's easy now */
3105 update_cache_size(container
, sra
, info
,
3106 min(reshape
.before
.data_disks
,
3107 reshape
.after
.data_disks
),
3108 reshape
.backup_blocks
);
3110 /* Right, everything seems fine. Let's kick things off.
3114 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3115 devname
, container
, &reshape
) < 0)
3117 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3119 if (errno
!= EINVAL
) {
3120 pr_err("Failed to initiate reshape!\n");
3123 /* revert data_offset and try the old way */
3124 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3125 sysfs_set_num(sra
, sd
, "new_offset",
3127 sysfs_set_str(sra
, NULL
, "reshape_direction",
3132 if (info
->new_level
== reshape
.level
)
3134 /* need to adjust level when reshape completes */
3136 case -1: /* ignore error, but don't wait */
3138 default: /* parent */
3146 fd
= open_dev(sra
->sys_name
);
3148 impose_level(fd
, info
->new_level
, devname
, verbose
);
3150 case 1: /* Couldn't set data_offset, try the old way */
3151 if (data_offset
!= INVALID_SECTORS
) {
3152 pr_err("Cannot update data_offset on this array\n");
3159 /* Decide how many blocks (sectors) for a reshape
3160 * unit. The number we have so far is just a minimum
3162 blocks
= reshape
.backup_blocks
;
3163 if (reshape
.before
.data_disks
==
3164 reshape
.after
.data_disks
) {
3165 /* Make 'blocks' bigger for better throughput, but
3166 * not so big that we reject it below.
3167 * Try for 16 megabytes
3169 while (blocks
* 32 < sra
->component_size
&&
3173 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3175 if (blocks
>= sra
->component_size
/2) {
3176 pr_err("%s: Something wrong - reshape aborted\n",
3181 /* Now we need to open all these devices so we can read/write.
3183 nrdisks
= max(reshape
.before
.data_disks
,
3184 reshape
.after
.data_disks
) + reshape
.parity
3185 + sra
->array
.spare_disks
;
3186 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3187 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3189 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3190 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
3191 nrdisks
, blocks
, backup_file
,
3196 if ((st
->ss
->manage_reshape
== NULL
) ||
3197 (st
->ss
->recover_backup
== NULL
)) {
3198 if (backup_file
== NULL
) {
3199 if (reshape
.after
.data_disks
<=
3200 reshape
.before
.data_disks
) {
3201 pr_err("%s: Cannot grow - need backup-file\n",
3203 pr_err(" Please provide one with \"--backup=...\"\n");
3205 } else if (d
== odisks
) {
3206 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3210 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3212 fdlist
+d
, offsets
+d
,
3221 update_cache_size(container
, sra
, info
,
3222 min(reshape
.before
.data_disks
, reshape
.after
.data_disks
),
3225 /* Right, everything seems fine. Let's kick things off.
3226 * If only changing raid_disks, use ioctl, else use
3231 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3232 devname
, container
, &reshape
) < 0)
3235 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3236 reshape
.after
.data_disks
);
3238 pr_err("Cannot %s reshape for %s\n",
3239 restart
? "continue" : "start",
3244 sysfs_set_str(sra
, NULL
, "array_state", "active");
3245 if (freeze_reshape
) {
3249 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3250 sra
->reshape_progress
);
3254 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3255 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3262 /* Now we just need to kick off the reshape and watch, while
3263 * handling backups of the data...
3264 * This is all done by a forked background process.
3266 switch(forked
? 0 : fork()) {
3268 pr_err("Cannot run child to monitor reshape: %s\n",
3282 /* If another array on the same devices is busy, the
3283 * reshape will wait for them. This would mean that
3284 * the first section that we suspend will stay suspended
3285 * for a long time. So check on that possibility
3286 * by looking for "DELAYED" in /proc/mdstat, and if found,
3290 struct mdstat_ent
*mds
, *m
;
3292 mds
= mdstat_read(1, 0);
3293 for (m
= mds
; m
; m
= m
->next
)
3294 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3296 m
->percent
== RESYNC_DELAYED
)
3299 /* Haven't started the reshape thread
3306 if (delayed
== 1 && get_linux_version() < 3007000) {
3307 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3308 " You might experience problems until other reshapes complete.\n");
3312 mdstat_wait(30 - (delayed
-1) * 25);
3316 if (check_env("MDADM_GROW_VERIFY"))
3317 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3320 mlockall(MCL_FUTURE
);
3322 signal(SIGTERM
, catch_term
);
3324 if (st
->ss
->external
) {
3325 /* metadata handler takes it from here */
3326 done
= st
->ss
->manage_reshape(
3327 fd
, sra
, &reshape
, st
, blocks
,
3329 d
- odisks
, fdlist
+odisks
,
3332 done
= child_monitor(
3333 fd
, sra
, &reshape
, st
, blocks
,
3335 d
- odisks
, fdlist
+odisks
,
3341 if (backup_file
&& done
) {
3343 bul
= make_backup(sra
->sys_name
);
3346 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3354 unlink(backup_file
);
3361 if (!st
->ss
->external
&&
3362 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
3363 && info
->custom_array_size
) &&
3364 info
->new_level
== reshape
.level
&&
3366 /* no need to wait for the reshape to finish as
3367 * there is nothing more to do.
3374 if (st
->ss
->external
) {
3375 /* Re-load the metadata as much could have changed */
3376 int cfd
= open_dev(st
->container_devnm
);
3378 flush_mdmon(container
);
3379 st
->ss
->free_super(st
);
3380 st
->ss
->load_container(st
, cfd
, container
);
3385 /* set new array size if required customer_array_size is used
3388 if (reshape
.before
.data_disks
!=
3389 reshape
.after
.data_disks
&&
3390 info
->custom_array_size
)
3391 set_array_size(st
, info
, info
->text_version
);
3393 if (info
->new_level
!= reshape
.level
) {
3395 fd
= open(devname
, O_RDONLY
);
3396 impose_level(fd
, info
->new_level
, devname
, verbose
);
3398 if (info
->new_level
== 0)
3399 st
->update_tail
= NULL
;
3411 if (orig_level
!= UnSet
&& sra
) {
3412 c
= map_num(pers
, orig_level
);
3413 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3414 pr_err("aborting level change\n");
3422 /* mdfd handle is passed to be closed in child process (after fork).
3424 int reshape_container(char *container
, char *devname
,
3426 struct supertype
*st
,
3427 struct mdinfo
*info
,
3429 char *backup_file
, int verbose
,
3430 int forked
, int restart
, int freeze_reshape
)
3432 struct mdinfo
*cc
= NULL
;
3434 char last_devnm
[32] = "";
3436 /* component_size is not meaningful for a container,
3437 * so pass '0' meaning 'no change'
3440 reshape_super(st
, 0, info
->new_level
,
3441 info
->new_layout
, info
->new_chunk
,
3442 info
->array
.raid_disks
, info
->delta_disks
,
3443 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3451 /* ping monitor to be sure that update is on disk
3453 ping_monitor(container
);
3455 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3456 if (continue_via_systemd(container
))
3459 switch (forked
? 0 : fork()) {
3460 case -1: /* error */
3461 perror("Cannot fork to complete reshape\n");
3464 default: /* parent */
3465 if (!freeze_reshape
)
3466 printf("%s: multi-array reshape continues in background\n", Name
);
3473 /* close unused handle in child process
3479 /* For each member array with reshape_active,
3480 * we need to perform the reshape.
3481 * We pick the first array that needs reshaping and
3482 * reshape it. reshape_array() will re-read the metadata
3483 * so the next time through a different array should be
3484 * ready for reshape.
3485 * It is possible that the 'different' array will not
3486 * be assembled yet. In that case we simple exit.
3487 * When it is assembled, the mdadm which assembles it
3488 * will take over the reshape.
3490 struct mdinfo
*content
;
3492 struct mdstat_ent
*mdstat
;
3498 cc
= st
->ss
->container_content(st
, NULL
);
3500 for (content
= cc
; content
; content
= content
->next
) {
3502 if (!content
->reshape_active
)
3505 subarray
= strchr(content
->text_version
+1, '/')+1;
3506 mdstat
= mdstat_by_subdev(subarray
, container
);
3509 if (mdstat
->active
== 0) {
3510 pr_err("Skipping inactive array %s.\n",
3512 free_mdstat(mdstat
);
3521 devid
= devnm2devid(mdstat
->devnm
);
3522 adev
= map_dev(major(devid
), minor(devid
), 0);
3524 adev
= content
->text_version
;
3526 fd
= open_dev(mdstat
->devnm
);
3528 pr_err("Device %s cannot be opened for reshape.\n", adev
);
3532 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3533 /* Do not allow for multiple reshape_array() calls for
3535 * It can happen when reshape_array() returns without
3536 * error, when reshape is not finished (wrong reshape
3537 * starting/continuation conditions). Mdmon doesn't
3538 * switch to next array in container and reentry
3539 * conditions for the same array occur.
3540 * This is possibly interim until the behaviour of
3541 * reshape_array is resolved().
3543 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3547 strcpy(last_devnm
, mdstat
->devnm
);
3549 sysfs_init(content
, fd
, mdstat
->devnm
);
3551 if (mdmon_running(container
))
3552 flush_mdmon(container
);
3554 rv
= reshape_array(container
, fd
, adev
, st
,
3555 content
, force
, NULL
, INVALID_SECTORS
,
3556 backup_file
, verbose
, 1, restart
,
3560 if (freeze_reshape
) {
3569 if (mdmon_running(container
))
3570 flush_mdmon(container
);
3579 * We run a child process in the background which performs the following
3581 * - wait for resync to reach a certain point
3582 * - suspend io to the following section
3583 * - backup that section
3584 * - allow resync to proceed further
3586 * - discard the backup.
3588 * When are combined in slightly different ways in the three cases.
3590 * - suspend/backup/allow/wait/resume/discard
3592 * - allow/wait/suspend/backup/allow/wait/resume/discard
3594 * - wait/resume/discard/suspend/backup/allow
3596 * suspend/backup/allow always come together
3597 * wait/resume/discard do too.
3598 * For the same-size case we have two backups to improve flow.
3602 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3603 unsigned long long backup_point
,
3604 unsigned long long wait_point
,
3605 unsigned long long *suspend_point
,
3606 unsigned long long *reshape_completed
, int *frozen
)
3608 /* This function is called repeatedly by the reshape manager.
3609 * It determines how much progress can safely be made and allows
3611 * - 'info' identifies the array and particularly records in
3612 * ->reshape_progress the metadata's knowledge of progress
3613 * This is a sector offset from the start of the array
3614 * of the next array block to be relocated. This number
3615 * may increase from 0 or decrease from array_size, depending
3616 * on the type of reshape that is happening.
3617 * Note that in contrast, 'sync_completed' is a block count of the
3618 * reshape so far. It gives the distance between the start point
3619 * (head or tail of device) and the next place that data will be
3620 * written. It always increases.
3621 * - 'reshape' is the structure created by analyse_change
3622 * - 'backup_point' shows how much the metadata manager has backed-up
3623 * data. For reshapes with increasing progress, it is the next address
3624 * to be backed up, previous addresses have been backed-up. For
3625 * decreasing progress, it is the earliest address that has been
3626 * backed up - later address are also backed up.
3627 * So addresses between reshape_progress and backup_point are
3628 * backed up providing those are in the 'correct' order.
3629 * - 'wait_point' is an array address. When reshape_completed
3630 * passes this point, progress_reshape should return. It might
3631 * return earlier if it determines that ->reshape_progress needs
3632 * to be updated or further backup is needed.
3633 * - suspend_point is maintained by progress_reshape and the caller
3634 * should not touch it except to initialise to zero.
3635 * It is an array address and it only increases in 2.6.37 and earlier.
3636 * This makes it difficult to handle reducing reshapes with
3637 * external metadata.
3638 * However: it is similar to backup_point in that it records the
3639 * other end of a suspended region from reshape_progress.
3640 * it is moved to extend the region that is safe to backup and/or
3642 * - reshape_completed is read from sysfs and returned. The caller
3643 * should copy this into ->reshape_progress when it has reason to
3644 * believe that the metadata knows this, and any backup outside this
3648 * 1 if more data from backup_point - but only as far as suspend_point,
3649 * should be backed up
3650 * 0 if things are progressing smoothly
3651 * -1 if the reshape is finished because it is all done,
3652 * -2 if the reshape is finished due to an error.
3655 int advancing
= (reshape
->after
.data_disks
3656 >= reshape
->before
.data_disks
);
3657 unsigned long long need_backup
; /* All data between start of array and
3658 * here will at some point need to
3661 unsigned long long read_offset
, write_offset
;
3662 unsigned long long write_range
;
3663 unsigned long long max_progress
, target
, completed
;
3664 unsigned long long array_size
= (info
->component_size
3665 * reshape
->before
.data_disks
);
3669 /* First, we unsuspend any region that is now known to be safe.
3670 * If suspend_point is on the 'wrong' side of reshape_progress, then
3671 * we don't have or need suspension at the moment. This is true for
3672 * native metadata when we don't need to back-up.
3675 if (info
->reshape_progress
<= *suspend_point
)
3676 sysfs_set_num(info
, NULL
, "suspend_lo",
3677 info
->reshape_progress
);
3679 /* Note: this won't work in 2.6.37 and before.
3680 * Something somewhere should make sure we don't need it!
3682 if (info
->reshape_progress
>= *suspend_point
)
3683 sysfs_set_num(info
, NULL
, "suspend_hi",
3684 info
->reshape_progress
);
3687 /* Now work out how far it is safe to progress.
3688 * If the read_offset for ->reshape_progress is less than
3689 * 'blocks' beyond the write_offset, we can only progress as far
3691 * Otherwise we can progress until the write_offset for the new location
3692 * reaches (within 'blocks' of) the read_offset at the current location.
3693 * However that region must be suspended unless we are using native
3695 * If we need to suspend more, we limit it to 128M per device, which is
3696 * rather arbitrary and should be some time-based calculation.
3698 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3699 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3700 write_range
= info
->new_chunk
/512;
3701 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3702 need_backup
= array_size
;
3704 need_backup
= reshape
->backup_blocks
;
3706 if (read_offset
< write_offset
+ write_range
)
3707 max_progress
= backup_point
;
3711 reshape
->after
.data_disks
;
3713 if (read_offset
> write_offset
- write_range
)
3714 /* Can only progress as far as has been backed up,
3715 * which must be suspended */
3716 max_progress
= backup_point
;
3717 else if (info
->reshape_progress
<= need_backup
)
3718 max_progress
= backup_point
;
3720 if (info
->array
.major_version
>= 0)
3721 /* Can progress until backup is needed */
3722 max_progress
= need_backup
;
3724 /* Can progress until metadata update is required */
3727 reshape
->after
.data_disks
;
3728 /* but data must be suspended */
3729 if (max_progress
< *suspend_point
)
3730 max_progress
= *suspend_point
;
3735 /* We know it is safe to progress to 'max_progress' providing
3736 * it is suspended or we are using native metadata.
3737 * Consider extending suspend_point 128M per device if it
3738 * is less than 64M per device beyond reshape_progress.
3739 * But always do a multiple of 'blocks'
3740 * FIXME this is too big - it takes to long to complete
3743 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3744 reshape
->after
.data_disks
);
3745 target
/= reshape
->backup_blocks
;
3748 target
*= reshape
->backup_blocks
;
3750 /* For externally managed metadata we always need to suspend IO to
3751 * the area being reshaped so we regularly push suspend_point forward.
3752 * For native metadata we only need the suspend if we are going to do
3756 if ((need_backup
> info
->reshape_progress
3757 || info
->array
.major_version
< 0) &&
3758 *suspend_point
< info
->reshape_progress
+ target
) {
3759 if (need_backup
< *suspend_point
+ 2 * target
)
3760 *suspend_point
= need_backup
;
3761 else if (*suspend_point
+ 2 * target
< array_size
)
3762 *suspend_point
+= 2 * target
;
3764 *suspend_point
= array_size
;
3765 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
3766 if (max_progress
> *suspend_point
)
3767 max_progress
= *suspend_point
;
3770 if (info
->array
.major_version
>= 0) {
3771 /* Only need to suspend when about to backup */
3772 if (info
->reshape_progress
< need_backup
* 2 &&
3773 *suspend_point
> 0) {
3775 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
3776 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
3779 /* Need to suspend continually */
3780 if (info
->reshape_progress
< *suspend_point
)
3781 *suspend_point
= info
->reshape_progress
;
3782 if (*suspend_point
+ target
< info
->reshape_progress
)
3783 /* No need to move suspend region yet */;
3785 if (*suspend_point
>= 2 * target
)
3786 *suspend_point
-= 2 * target
;
3789 sysfs_set_num(info
, NULL
, "suspend_lo",
3792 if (max_progress
< *suspend_point
)
3793 max_progress
= *suspend_point
;
3797 /* now set sync_max to allow that progress. sync_max, like
3798 * sync_completed is a count of sectors written per device, so
3799 * we find the difference between max_progress and the start point,
3800 * and divide that by after.data_disks to get a sync_max
3802 * At the same time we convert wait_point to a similar number
3803 * for comparing against sync_completed.
3805 /* scale down max_progress to per_disk */
3806 max_progress
/= reshape
->after
.data_disks
;
3807 /* Round to chunk size as some kernels give an erroneously high number */
3808 max_progress
/= info
->new_chunk
/512;
3809 max_progress
*= info
->new_chunk
/512;
3810 /* And round to old chunk size as the kernel wants that */
3811 max_progress
/= info
->array
.chunk_size
/512;
3812 max_progress
*= info
->array
.chunk_size
/512;
3813 /* Limit progress to the whole device */
3814 if (max_progress
> info
->component_size
)
3815 max_progress
= info
->component_size
;
3816 wait_point
/= reshape
->after
.data_disks
;
3818 /* switch from 'device offset' to 'processed block count' */
3819 max_progress
= info
->component_size
- max_progress
;
3820 wait_point
= info
->component_size
- wait_point
;
3824 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
3826 /* Now wait. If we have already reached the point that we were
3827 * asked to wait to, don't wait at all, else wait for any change.
3828 * We need to select on 'sync_completed' as that is the place that
3829 * notifications happen, but we are really interested in
3830 * 'reshape_position'
3832 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
3834 goto check_progress
;
3836 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3837 goto check_progress
;
3839 while (completed
< max_progress
&& completed
< wait_point
) {
3840 /* Check that sync_action is still 'reshape' to avoid
3841 * waiting forever on a dead array
3844 if (sysfs_get_str(info
, NULL
, "sync_action",
3846 strncmp(action
, "reshape", 7) != 0)
3848 /* Some kernels reset 'sync_completed' to zero
3849 * before setting 'sync_action' to 'idle'.
3850 * So we need these extra tests.
3852 if (completed
== 0 && advancing
3853 && strncmp(action
, "idle", 4) == 0
3854 && info
->reshape_progress
> 0)
3856 if (completed
== 0 && !advancing
3857 && strncmp(action
, "idle", 4) == 0
3858 && info
->reshape_progress
< (info
->component_size
3859 * reshape
->after
.data_disks
))
3861 sysfs_wait(fd
, NULL
);
3862 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
3863 goto check_progress
;
3865 /* Some kernels reset 'sync_completed' to zero,
3866 * we need to have real point we are in md.
3867 * So in that case, read 'reshape_position' from sysfs.
3869 if (completed
== 0) {
3870 unsigned long long reshapep
;
3872 if (sysfs_get_str(info
, NULL
, "sync_action",
3874 strncmp(action
, "idle", 4) == 0 &&
3875 sysfs_get_ll(info
, NULL
,
3876 "reshape_position", &reshapep
) == 0)
3877 *reshape_completed
= reshapep
;
3879 /* some kernels can give an incorrectly high
3880 * 'completed' number, so round down */
3881 completed
/= (info
->new_chunk
/512);
3882 completed
*= (info
->new_chunk
/512);
3883 /* Convert 'completed' back in to a 'progress' number */
3884 completed
*= reshape
->after
.data_disks
;
3886 completed
= (info
->component_size
3887 * reshape
->after
.data_disks
3889 *reshape_completed
= completed
;
3894 /* We return the need_backup flag. Caller will decide
3895 * how much - a multiple of ->backup_blocks up to *suspend_point
3898 return need_backup
> info
->reshape_progress
;
3900 return need_backup
>= info
->reshape_progress
;
3903 /* if we couldn't read a number from sync_completed, then
3904 * either the reshape did complete, or it aborted.
3905 * We can tell which by checking for 'none' in reshape_position.
3906 * If it did abort, then it might immediately restart if it
3907 * it was just a device failure that leaves us degraded but
3910 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
3911 || strncmp(buf
, "none", 4) != 0) {
3912 /* The abort might only be temporary. Wait up to 10
3913 * seconds for fd to contain a valid number again.
3917 unsigned long long new_sync_max
;
3918 while (fd
>= 0 && rv
< 0 && wait
> 0) {
3919 if (sysfs_wait(fd
, &wait
) != 1)
3921 switch (sysfs_fd_get_ll(fd
, &completed
)) {
3923 /* all good again */
3925 /* If "sync_max" is no longer max_progress
3926 * we need to freeze things
3928 sysfs_get_ll(info
, NULL
, "sync_max", &new_sync_max
);
3929 *frozen
= (new_sync_max
!= max_progress
);
3931 case -2: /* read error - abort */
3938 return rv
; /* abort */
3940 /* Maybe racing with array shutdown - check state */
3943 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
3944 || strncmp(buf
, "inactive", 8) == 0
3945 || strncmp(buf
, "clear",5) == 0)
3946 return -2; /* abort */
3947 return -1; /* complete */
3951 /* FIXME return status is never checked */
3952 static int grow_backup(struct mdinfo
*sra
,
3953 unsigned long long offset
, /* per device */
3954 unsigned long stripes
, /* per device, in old chunks */
3955 int *sources
, unsigned long long *offsets
,
3956 int disks
, int chunk
, int level
, int layout
,
3957 int dests
, int *destfd
, unsigned long long *destoffsets
,
3958 int part
, int *degraded
,
3961 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3962 * to storage 'destfd' (offset 'destoffsets'), after first
3963 * suspending IO. Then allow resync to continue
3964 * over the suspended section.
3965 * Use part 'part' of the backup-super-block.
3970 unsigned long long ll
;
3972 //printf("offset %llu\n", offset);
3978 /* Check that array hasn't become degraded, else we might backup the wrong data */
3979 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
3980 return -1; /* FIXME this error is ignored */
3981 new_degraded
= (int)ll
;
3982 if (new_degraded
!= *degraded
) {
3983 /* check each device to ensure it is still working */
3985 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3986 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
3988 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
3990 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
3991 strstr(sbuf
, "faulty") ||
3992 strstr(sbuf
, "in_sync") == NULL
) {
3993 /* this device is dead */
3994 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
3995 if (sd
->disk
.raid_disk
>= 0 &&
3996 sources
[sd
->disk
.raid_disk
] >= 0) {
3997 close(sources
[sd
->disk
.raid_disk
]);
3998 sources
[sd
->disk
.raid_disk
] = -1;
4003 *degraded
= new_degraded
;
4006 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4007 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4009 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4010 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4013 bsb
.magic
[15] = '2';
4014 for (i
= 0; i
< dests
; i
++)
4016 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
4018 lseek64(destfd
[i
], destoffsets
[i
], 0);
4020 rv
= save_stripes(sources
, offsets
,
4021 disks
, chunk
, level
, layout
,
4023 offset
*512*odata
, stripes
* chunk
* odata
,
4028 bsb
.mtime
= __cpu_to_le64(time(0));
4029 for (i
= 0; i
< dests
; i
++) {
4030 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4032 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4033 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4034 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4035 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4038 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
4039 != destoffsets
[i
] - 4096)
4041 if (write(destfd
[i
], &bsb
, 512) != 512)
4043 if (destoffsets
[i
] > 4096) {
4044 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4045 destoffsets
[i
]+stripes
*chunk
*odata
)
4047 if (write(destfd
[i
], &bsb
, 512) != 512)
4057 /* in 2.6.30, the value reported by sync_completed can be
4058 * less that it should be by one stripe.
4059 * This only happens when reshape hits sync_max and pauses.
4060 * So allow wait_backup to either extent sync_max further
4061 * than strictly necessary, or return before the
4062 * sync has got quite as far as we would really like.
4063 * This is what 'blocks2' is for.
4064 * The various caller give appropriate values so that
4067 /* FIXME return value is often ignored */
4068 static int forget_backup(int dests
, int *destfd
,
4069 unsigned long long *destoffsets
,
4073 * Erase backup 'part' (which is 0 or 1)
4079 bsb
.arraystart2
= __cpu_to_le64(0);
4080 bsb
.length2
= __cpu_to_le64(0);
4082 bsb
.arraystart
= __cpu_to_le64(0);
4083 bsb
.length
= __cpu_to_le64(0);
4085 bsb
.mtime
= __cpu_to_le64(time(0));
4087 for (i
= 0; i
< dests
; i
++) {
4088 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4089 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4090 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4091 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4092 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4093 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4094 destoffsets
[i
]-4096)
4097 write(destfd
[i
], &bsb
, 512) != 512)
4104 static void fail(char *msg
)
4107 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4108 rv
|= (write(2, "\n", 1) != 1);
4112 static char *abuf
, *bbuf
;
4113 static unsigned long long abuflen
;
4114 static void validate(int afd
, int bfd
, unsigned long long offset
)
4116 /* check that the data in the backup against the array.
4117 * This is only used for regression testing and should not
4118 * be used while the array is active
4122 lseek64(bfd
, offset
- 4096, 0);
4123 if (read(bfd
, &bsb2
, 512) != 512)
4124 fail("cannot read bsb");
4125 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4126 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4127 fail("first csum bad");
4128 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4129 fail("magic is bad");
4130 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4131 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4132 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4133 fail("second csum bad");
4135 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4136 fail("devstart is wrong");
4139 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4141 if (abuflen
< len
) {
4145 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4146 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4148 /* just stop validating on mem-alloc failure */
4153 lseek64(bfd
, offset
, 0);
4154 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4155 //printf("len %llu\n", len);
4156 fail("read first backup failed");
4158 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4159 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4160 fail("read first from array failed");
4161 if (memcmp(bbuf
, abuf
, len
) != 0) {
4164 printf("offset=%llu len=%llu\n",
4165 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4166 for (i
=0; i
<len
; i
++)
4167 if (bbuf
[i
] != abuf
[i
]) {
4168 printf("first diff byte %d\n", i
);
4172 fail("data1 compare failed");
4176 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4178 if (abuflen
< len
) {
4182 abuf
= xmalloc(abuflen
);
4183 bbuf
= xmalloc(abuflen
);
4186 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4187 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4188 fail("read second backup failed");
4189 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4190 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4191 fail("read second from array failed");
4192 if (memcmp(bbuf
, abuf
, len
) != 0)
4193 fail("data2 compare failed");
4197 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4198 struct supertype
*st
, unsigned long blocks
,
4199 int *fds
, unsigned long long *offsets
,
4200 int dests
, int *destfd
, unsigned long long *destoffsets
)
4202 /* Monitor a reshape where backup is being performed using
4203 * 'native' mechanism - either to a backup file, or
4204 * to some space in a spare.
4208 unsigned long long speed
;
4209 unsigned long long suspend_point
, array_size
;
4210 unsigned long long backup_point
, wait_point
;
4211 unsigned long long reshape_completed
;
4213 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
4214 int part
= 0; /* The next part of the backup area to fill. It may already
4215 * be full, so we need to check */
4216 int level
= reshape
->level
;
4217 int layout
= reshape
->before
.layout
;
4218 int data
= reshape
->before
.data_disks
;
4219 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4220 int chunk
= sra
->array
.chunk_size
;
4222 unsigned long stripes
;
4226 /* set up the backup-super-block. This requires the
4227 * uuid from the array.
4229 /* Find a superblock */
4230 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4234 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4236 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4237 devfd
= dev_open(dn
, O_RDONLY
);
4240 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4246 pr_err("Cannot find a superblock\n");
4250 memset(&bsb
, 0, 512);
4251 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4252 st
->ss
->uuid_from_super(st
, uuid
);
4253 memcpy(bsb
.set_uuid
, uuid
, 16);
4254 bsb
.mtime
= __cpu_to_le64(time(0));
4255 bsb
.devstart2
= blocks
;
4257 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4258 reshape
->before
.data_disks
;
4260 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4261 /* Don't start the 'reshape' */
4263 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4264 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4265 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4269 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4270 backup_point
= sra
->reshape_progress
;
4273 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4274 backup_point
= reshape
->backup_blocks
;
4275 suspend_point
= array_size
;
4281 /* Want to return as soon the oldest backup slot can
4282 * be released as that allows us to start backing up
4283 * some more, providing suspend_point has been
4284 * advanced, which it should have.
4287 wait_point
= array_size
;
4288 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4289 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4290 __le64_to_cpu(bsb
.length
));
4291 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4292 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4293 __le64_to_cpu(bsb
.length2
));
4296 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4297 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4298 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4299 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4302 reshape_completed
= sra
->reshape_progress
;
4303 rv
= progress_reshape(sra
, reshape
,
4304 backup_point
, wait_point
,
4305 &suspend_point
, &reshape_completed
,
4307 /* external metadata would need to ping_monitor here */
4308 sra
->reshape_progress
= reshape_completed
;
4310 /* Clear any backup region that is before 'here' */
4312 if (__le64_to_cpu(bsb
.length
) > 0 &&
4313 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4314 __le64_to_cpu(bsb
.length
)))
4315 forget_backup(dests
, destfd
,
4317 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4318 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4319 __le64_to_cpu(bsb
.length2
)))
4320 forget_backup(dests
, destfd
,
4323 if (__le64_to_cpu(bsb
.length
) > 0 &&
4324 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4325 forget_backup(dests
, destfd
,
4327 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4328 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4329 forget_backup(dests
, destfd
,
4339 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4340 /* No longer need to monitor this reshape */
4341 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4347 unsigned long long offset
;
4348 unsigned long actual_stripes
;
4349 /* Need to backup some data.
4350 * If 'part' is not used and the desired
4351 * backup size is suspended, do a backup,
4352 * then consider the next part.
4354 /* Check that 'part' is unused */
4355 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4357 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4360 offset
= backup_point
/ data
;
4361 actual_stripes
= stripes
;
4363 if (offset
+ actual_stripes
* (chunk
/512) >
4364 sra
->component_size
)
4365 actual_stripes
= ((sra
->component_size
- offset
)
4367 if (offset
+ actual_stripes
* (chunk
/512) >
4371 if (offset
< actual_stripes
* (chunk
/512))
4372 actual_stripes
= offset
/ (chunk
/512);
4373 offset
-= actual_stripes
* (chunk
/512);
4374 if (offset
< suspend_point
/data
)
4377 if (actual_stripes
== 0)
4379 grow_backup(sra
, offset
, actual_stripes
,
4381 disks
, chunk
, level
, layout
,
4382 dests
, destfd
, destoffsets
,
4383 part
, °raded
, buf
);
4384 validate(afd
, destfd
[0], destoffsets
[0]);
4385 /* record where 'part' is up to */
4388 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4390 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4394 /* FIXME maybe call progress_reshape one more time instead */
4395 /* remove any remaining suspension */
4396 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4397 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4398 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4399 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4401 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4402 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4408 * If any spare contains md_back_data-1 which is recent wrt mtime,
4409 * write that data into the array and update the super blocks with
4410 * the new reshape_progress
4412 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
4413 char *backup_file
, int verbose
)
4417 unsigned long long *offsets
;
4418 unsigned long long nstripe
, ostripe
;
4421 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4422 if (info
->array
.level
== 6) odata
--; /* number of data disks */
4423 ndata
= info
->array
.raid_disks
- 1;
4424 if (info
->new_level
== 6) ndata
--;
4426 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4428 if (info
->delta_disks
<= 0)
4429 /* Didn't grow, so the backup file must have
4433 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4434 struct mdinfo dinfo
;
4437 char *devname
, namebuf
[20];
4438 unsigned long long lo
, hi
;
4440 /* This was a spare and may have some saved data on it.
4441 * Load the superblock, find and load the
4442 * backup_super_block.
4443 * If either fail, go on to next device.
4444 * If the backup contains no new info, just return
4445 * else restore data and update all superblocks
4447 if (i
== old_disks
-1) {
4448 fd
= open(backup_file
, O_RDONLY
);
4450 pr_err("backup file %s inaccessible: %s\n",
4451 backup_file
, strerror(errno
));
4454 devname
= backup_file
;
4459 if (st
->ss
->load_super(st
, fd
, NULL
))
4462 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4463 st
->ss
->free_super(st
);
4466 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4468 pr_err("Cannot seek on device %d\n", i
);
4469 continue; /* Cannot seek */
4471 sprintf(namebuf
, "device-%d", i
);
4474 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4476 pr_err("Cannot read from %s\n", devname
);
4477 continue; /* Cannot read */
4479 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4480 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4482 pr_err("No backup metadata on %s\n", devname
);
4485 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4487 pr_err("Bad backup-metadata checksum on %s\n", devname
);
4488 continue; /* bad checksum */
4490 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4491 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4493 pr_err("Bad backup-metadata checksum2 on %s\n", devname
);
4494 continue; /* Bad second checksum */
4496 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4498 pr_err("Wrong uuid on backup-metadata on %s\n", devname
);
4499 continue; /* Wrong uuid */
4502 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4503 * sometimes they aren't... So allow considerable flexability in matching, and allow
4504 * this test to be overridden by an environment variable.
4506 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
4507 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
4508 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4509 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4510 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4511 (unsigned long)info
->array
.utime
);
4513 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4514 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4515 continue; /* time stamp is too bad */
4519 if (bsb
.magic
[15] == '1') {
4520 if (bsb
.length
== 0)
4522 if (info
->delta_disks
>= 0) {
4523 /* reshape_progress is increasing */
4524 if (__le64_to_cpu(bsb
.arraystart
)
4525 + __le64_to_cpu(bsb
.length
)
4526 < info
->reshape_progress
) {
4529 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4530 continue; /* No new data here */
4533 /* reshape_progress is decreasing */
4534 if (__le64_to_cpu(bsb
.arraystart
) >=
4535 info
->reshape_progress
)
4536 goto nonew
; /* No new data here */
4539 if (bsb
.length
== 0 && bsb
.length2
== 0)
4541 if (info
->delta_disks
>= 0) {
4542 /* reshape_progress is increasing */
4543 if ((__le64_to_cpu(bsb
.arraystart
)
4544 + __le64_to_cpu(bsb
.length
)
4545 < info
->reshape_progress
)
4547 (__le64_to_cpu(bsb
.arraystart2
)
4548 + __le64_to_cpu(bsb
.length2
)
4549 < info
->reshape_progress
))
4550 goto nonew
; /* No new data here */
4552 /* reshape_progress is decreasing */
4553 if (__le64_to_cpu(bsb
.arraystart
) >=
4554 info
->reshape_progress
&&
4555 __le64_to_cpu(bsb
.arraystart2
) >=
4556 info
->reshape_progress
)
4557 goto nonew
; /* No new data here */
4560 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4563 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4565 continue; /* Cannot seek */
4567 /* There should be a duplicate backup superblock 4k before here */
4568 if (lseek64(fd
, -4096, 1) < 0 ||
4569 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4570 goto second_fail
; /* Cannot find leading superblock */
4571 if (bsb
.magic
[15] == '1')
4572 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4574 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4575 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4576 goto second_fail
; /* Cannot find leading superblock */
4578 /* Now need the data offsets for all devices. */
4579 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4580 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4583 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4584 /* FIXME should be this be an error */
4586 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4587 st
->ss
->free_super(st
);
4588 offsets
[j
] = dinfo
.data_offset
* 512;
4590 printf("%s: restoring critical section\n", Name
);
4592 if (restore_stripes(fdlist
, offsets
,
4593 info
->array
.raid_disks
,
4597 fd
, __le64_to_cpu(bsb
.devstart
)*512,
4598 __le64_to_cpu(bsb
.arraystart
)*512,
4599 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4600 /* didn't succeed, so giveup */
4602 pr_err("Error restoring backup from %s\n",
4608 if (bsb
.magic
[15] == '2' &&
4609 restore_stripes(fdlist
, offsets
,
4610 info
->array
.raid_disks
,
4614 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
4615 __le64_to_cpu(bsb
.devstart2
)*512,
4616 __le64_to_cpu(bsb
.arraystart2
)*512,
4617 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4618 /* didn't succeed, so giveup */
4620 pr_err("Error restoring second backup from %s\n",
4628 /* Ok, so the data is restored. Let's update those superblocks. */
4632 lo
= __le64_to_cpu(bsb
.arraystart
);
4633 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4635 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4636 unsigned long long lo1
, hi1
;
4637 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4638 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4642 } else if (lo
< lo1
)
4648 (info
->reshape_progress
< lo
||
4649 info
->reshape_progress
> hi
))
4650 /* backup does not affect reshape_progress*/ ;
4651 else if (info
->delta_disks
>= 0) {
4652 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4653 __le64_to_cpu(bsb
.length
);
4654 if (bsb
.magic
[15] == '2') {
4655 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
4656 __le64_to_cpu(bsb
.length2
);
4657 if (p2
> info
->reshape_progress
)
4658 info
->reshape_progress
= p2
;
4661 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4662 if (bsb
.magic
[15] == '2') {
4663 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
4664 if (p2
< info
->reshape_progress
)
4665 info
->reshape_progress
= p2
;
4668 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4671 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4673 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4674 dinfo
.reshape_progress
= info
->reshape_progress
;
4675 st
->ss
->update_super(st
, &dinfo
,
4676 "_reshape_progress",
4678 st
->ss
->store_super(st
, fdlist
[j
]);
4679 st
->ss
->free_super(st
);
4683 /* Didn't find any backup data, try to see if any
4686 if (info
->delta_disks
< 0) {
4687 /* When shrinking, the critical section is at the end.
4688 * So see if we are before the critical section.
4690 unsigned long long first_block
;
4691 nstripe
= ostripe
= 0;
4693 while (ostripe
>= nstripe
) {
4694 ostripe
+= info
->array
.chunk_size
/ 512;
4695 first_block
= ostripe
* odata
;
4696 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4697 (info
->new_chunk
/512);
4700 if (info
->reshape_progress
>= first_block
)
4703 if (info
->delta_disks
> 0) {
4704 /* See if we are beyond the critical section. */
4705 unsigned long long last_block
;
4706 nstripe
= ostripe
= 0;
4708 while (nstripe
>= ostripe
) {
4709 nstripe
+= info
->new_chunk
/ 512;
4710 last_block
= nstripe
* ndata
;
4711 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4712 (info
->array
.chunk_size
/512);
4715 if (info
->reshape_progress
>= last_block
)
4718 /* needed to recover critical section! */
4720 pr_err("Failed to find backup of critical section\n");
4724 int Grow_continue_command(char *devname
, int fd
,
4725 char *backup_file
, int verbose
)
4728 struct supertype
*st
= NULL
;
4729 struct mdinfo
*content
= NULL
;
4730 struct mdinfo array
;
4731 char *subarray
= NULL
;
4732 struct mdinfo
*cc
= NULL
;
4733 struct mdstat_ent
*mdstat
= NULL
;
4737 dprintf("Grow continue from command line called for %s\n",
4740 st
= super_by_fd(fd
, &subarray
);
4741 if (!st
|| !st
->ss
) {
4742 pr_err("Unable to determine metadata format for %s\n",
4746 dprintf("Grow continue is run for ");
4747 if (st
->ss
->external
== 0) {
4749 dprintf_cont("native array (%s)\n", devname
);
4750 if (ioctl(fd
, GET_ARRAY_INFO
, &array
.array
) < 0) {
4751 pr_err("%s is not an active md array - aborting\n", devname
);
4753 goto Grow_continue_command_exit
;
4756 /* Need to load a superblock.
4757 * FIXME we should really get what we need from
4760 for (d
= 0; d
< MAX_DISKS
; d
++) {
4761 mdu_disk_info_t disk
;
4765 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
4767 if (disk
.major
== 0 && disk
.minor
== 0)
4769 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
4771 dv
= map_dev(disk
.major
, disk
.minor
, 1);
4774 fd2
= dev_open(dv
, O_RDONLY
);
4777 err
= st
->ss
->load_super(st
, fd2
, NULL
);
4779 /* invalidate fd2 to avoid possible double close() */
4785 if (d
== MAX_DISKS
) {
4786 pr_err("Unable to load metadata for %s\n",
4789 goto Grow_continue_command_exit
;
4791 st
->ss
->getinfo_super(st
, content
, NULL
);
4796 dprintf_cont("subarray (%s)\n", subarray
);
4797 container
= st
->container_devnm
;
4798 cfd
= open_dev_excl(st
->container_devnm
);
4800 container
= st
->devnm
;
4802 cfd
= open_dev_excl(st
->devnm
);
4803 dprintf_cont("container (%s)\n", container
);
4807 pr_err("Unable to open container for %s\n", devname
);
4809 goto Grow_continue_command_exit
;
4812 /* find in container array under reshape
4814 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
4816 pr_err("Cannot read superblock for %s\n",
4819 goto Grow_continue_command_exit
;
4822 cc
= st
->ss
->container_content(st
, subarray
);
4823 for (content
= cc
; content
; content
= content
->next
) {
4825 int allow_reshape
= 1;
4827 if (content
->reshape_active
== 0)
4829 /* The decision about array or container wide
4830 * reshape is taken in Grow_continue based
4831 * content->reshape_active state, therefore we
4832 * need to check_reshape based on
4833 * reshape_active and subarray name
4835 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
4837 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
4838 (content
->array
.state
4839 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
4842 if (!allow_reshape
) {
4843 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4844 devname
, container
);
4846 goto Grow_continue_command_exit
;
4849 array
= strchr(content
->text_version
+1, '/')+1;
4850 mdstat
= mdstat_by_subdev(array
, container
);
4853 if (mdstat
->active
== 0) {
4854 pr_err("Skipping inactive array %s.\n",
4856 free_mdstat(mdstat
);
4863 pr_err("Unable to determine reshaped array for %s\n", devname
);
4865 goto Grow_continue_command_exit
;
4867 fd2
= open_dev(mdstat
->devnm
);
4869 pr_err("cannot open (%s)\n", mdstat
->devnm
);
4871 goto Grow_continue_command_exit
;
4874 sysfs_init(content
, fd2
, mdstat
->devnm
);
4876 /* start mdmon in case it is not running
4878 if (!mdmon_running(container
))
4879 start_mdmon(container
);
4880 ping_monitor(container
);
4882 if (mdmon_running(container
))
4883 st
->update_tail
= &st
->updates
;
4885 pr_err("No mdmon found. Grow cannot continue.\n");
4887 goto Grow_continue_command_exit
;
4891 /* verify that array under reshape is started from
4894 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
4896 goto Grow_continue_command_exit
;
4901 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
4903 Grow_continue_command_exit
:
4908 st
->ss
->free_super(st
);
4909 free_mdstat(mdstat
);
4916 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
4917 char *backup_file
, int forked
, int freeze_reshape
)
4921 if (!info
->reshape_active
)
4924 if (st
->ss
->external
) {
4925 int cfd
= open_dev(st
->container_devnm
);
4930 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
4932 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
4933 st
, info
, 0, backup_file
,
4935 1 | info
->reshape_active
,
4938 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
4939 NULL
, INVALID_SECTORS
,
4940 backup_file
, 0, forked
,
4941 1 | info
->reshape_active
,
4947 char *make_backup(char *name
)
4949 char *base
= "backup_file-";
4953 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
4954 fname
= xmalloc(len
);
4955 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
4959 char *locate_backup(char *name
)
4961 char *fl
= make_backup(name
);
4964 if (stat(fl
, &stb
) == 0 &&
4965 S_ISREG(stb
.st_mode
))