2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
28 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
29 #error no endian defined
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
38 int restore_backup(struct supertype
*st
,
39 struct mdinfo
*content
,
49 int disk_count
= next_spare
+ working_disks
;
51 dprintf("Called restore_backup()\n");
52 fdlist
= malloc(sizeof(int) * disk_count
);
55 Name
": cannot allocate memory for disk list\n");
58 for (i
= 0; i
< next_spare
; i
++)
60 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
66 fd
= dev_open(buf
, O_RDWR
);
68 if (dev
->disk
.raid_disk
>= 0)
69 fdlist
[dev
->disk
.raid_disk
] = fd
;
71 fdlist
[next_spare
++] = fd
;
74 if (st
->ss
->external
&& st
->ss
->recover_backup
)
75 err
= st
->ss
->recover_backup(st
, content
);
77 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
78 backup_file
, verbose
> 0);
80 while (next_spare
> 0) {
82 if (fdlist
[next_spare
] >= 0)
83 close(fdlist
[next_spare
]);
87 fprintf(stderr
, Name
": Failed to restore critical"
88 " section for reshape - sorry.\n");
90 fprintf(stderr
, Name
": Possibly you need"
91 " to specify a --backup-file\n");
95 dprintf("restore_backup() returns status OK.\n");
99 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
101 /* Add a device to an active array.
102 * Currently, just extend a linear array.
103 * This requires writing a new superblock on the
104 * new device, calling the kernel to add the device,
105 * and if that succeeds, update the superblock on
107 * This means that we need to *find* all other devices.
114 struct supertype
*st
= NULL
;
115 char *subarray
= NULL
;
117 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
118 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
122 if (info
.array
.level
!= -1) {
123 fprintf(stderr
, Name
": can only add devices to linear arrays\n");
127 st
= super_by_fd(fd
, &subarray
);
129 fprintf(stderr
, Name
": cannot handle arrays with superblock version %d\n", info
.array
.major_version
);
134 fprintf(stderr
, Name
": Cannot grow linear sub-arrays yet\n");
140 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
142 fprintf(stderr
, Name
": cannot open %s\n", newdev
);
147 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
148 fprintf(stderr
, Name
": %s is not a block device!\n", newdev
);
153 /* now check out all the devices and make sure we can read the superblock */
154 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
155 mdu_disk_info_t disk
;
158 st
->ss
->free_super(st
);
161 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
162 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
168 dv
= map_dev(disk
.major
, disk
.minor
, 1);
170 fprintf(stderr
, Name
": cannot find device file for device %d\n",
176 fd2
= dev_open(dv
, O_RDWR
);
178 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
184 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
185 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
193 /* Ok, looks good. Lets update the superblock and write it out to
197 info
.disk
.number
= d
;
198 info
.disk
.major
= major(stb
.st_rdev
);
199 info
.disk
.minor
= minor(stb
.st_rdev
);
200 info
.disk
.raid_disk
= d
;
201 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
202 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
205 if (st
->ss
->store_super(st
, nfd
)) {
206 fprintf(stderr
, Name
": Cannot store new superblock on %s\n",
213 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
214 fprintf(stderr
, Name
": Cannot add new disk to this array\n");
217 /* Well, that seems to have worked.
218 * Now go through and update all superblocks
221 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
222 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
227 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
228 mdu_disk_info_t disk
;
232 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
233 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
237 dv
= map_dev(disk
.major
, disk
.minor
, 1);
239 fprintf(stderr
, Name
": cannot find device file for device %d\n",
243 fd2
= dev_open(dv
, O_RDWR
);
245 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
248 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
249 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
253 info
.array
.raid_disks
= nd
+1;
254 info
.array
.nr_disks
= nd
+1;
255 info
.array
.active_disks
= nd
+1;
256 info
.array
.working_disks
= nd
+1;
258 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
261 if (st
->ss
->store_super(st
, fd2
)) {
262 fprintf(stderr
, Name
": Cannot store new superblock on %s\n", dv
);
272 int Grow_addbitmap(char *devname
, int fd
, char *file
, int chunk
, int delay
, int write_behind
, int force
)
275 * First check that array doesn't have a bitmap
276 * Then create the bitmap
279 * For internal bitmaps, we need to check the version,
280 * find all the active devices, and write the bitmap block
283 mdu_bitmap_file_t bmf
;
284 mdu_array_info_t array
;
285 struct supertype
*st
;
286 char *subarray
= NULL
;
287 int major
= BITMAP_MAJOR_HI
;
288 int vers
= md_get_version(fd
);
289 unsigned long long bitmapsize
, array_size
;
292 major
= BITMAP_MAJOR_HOSTENDIAN
;
293 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel"
294 " are not portable\n"
295 " between different architectures. Consider upgrading"
296 " the Linux kernel.\n");
299 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
301 fprintf(stderr
, Name
": Memory allocation failure.\n");
303 fprintf(stderr
, Name
": bitmaps not supported by this kernel.\n");
306 if (bmf
.pathname
[0]) {
307 if (strcmp(file
,"none")==0) {
308 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
309 fprintf(stderr
, Name
": failed to remove bitmap %s\n",
315 fprintf(stderr
, Name
": %s already has a bitmap (%s)\n",
316 devname
, bmf
.pathname
);
319 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
320 fprintf(stderr
, Name
": cannot get array status for %s\n", devname
);
323 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
324 if (strcmp(file
, "none")==0) {
325 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
326 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
327 fprintf(stderr
, Name
": failed to remove internal bitmap.\n");
332 fprintf(stderr
, Name
": Internal bitmap already present on %s\n",
337 if (strcmp(file
, "none") == 0) {
338 fprintf(stderr
, Name
": no bitmap found on %s\n", devname
);
341 if (array
.level
<= 0) {
342 fprintf(stderr
, Name
": Bitmaps not meaningful with level %s\n",
343 map_num(pers
, array
.level
)?:"of this array");
346 bitmapsize
= array
.size
;
348 if (get_dev_size(fd
, NULL
, &array_size
) &&
349 array_size
> (0x7fffffffULL
<<9)) {
350 /* Array is big enough that we cannot trust array.size
351 * try other approaches
353 bitmapsize
= get_component_size(fd
);
355 if (bitmapsize
== 0) {
356 fprintf(stderr
, Name
": Cannot reliably determine size of array to create bitmap - sorry.\n");
360 if (array
.level
== 10) {
361 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
362 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
365 st
= super_by_fd(fd
, &subarray
);
367 fprintf(stderr
, Name
": Cannot understand version %d.%d\n",
368 array
.major_version
, array
.minor_version
);
372 fprintf(stderr
, Name
": Cannot add bitmaps to sub-arrays yet\n");
377 if (strcmp(file
, "internal") == 0) {
379 if (st
->ss
->add_internal_bitmap
== NULL
) {
380 fprintf(stderr
, Name
": Internal bitmaps not supported "
381 "with %s metadata\n", st
->ss
->name
);
384 for (d
=0; d
< st
->max_devs
; d
++) {
385 mdu_disk_info_t disk
;
388 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
390 if (disk
.major
== 0 &&
393 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
395 dv
= map_dev(disk
.major
, disk
.minor
, 1);
397 int fd2
= dev_open(dv
, O_RDWR
);
400 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
401 if (st
->ss
->add_internal_bitmap(
403 &chunk
, delay
, write_behind
,
404 bitmapsize
, 0, major
)
406 st
->ss
->write_bitmap(st
, fd2
);
408 fprintf(stderr
, Name
": failed to create internal bitmap - chunksize problem.\n");
416 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
417 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
420 ": Cannot add bitmap while array is"
421 " resyncing or reshaping etc.\n");
422 fprintf(stderr
, Name
": failed to set internal bitmap.\n");
429 int max_devs
= st
->max_devs
;
431 /* try to load a superblock */
432 for (d
=0; d
<max_devs
; d
++) {
433 mdu_disk_info_t disk
;
437 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
439 if ((disk
.major
==0 && disk
.minor
==0) ||
440 (disk
.state
& (1<<MD_DISK_REMOVED
)))
442 dv
= map_dev(disk
.major
, disk
.minor
, 1);
444 fd2
= dev_open(dv
, O_RDONLY
);
446 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
448 st
->ss
->uuid_from_super(st
, uuid
);
455 fprintf(stderr
, Name
": cannot find UUID for array!\n");
458 if (CreateBitmap(file
, force
, (char*)uuid
, chunk
,
459 delay
, write_behind
, bitmapsize
, major
)) {
462 bitmap_fd
= open(file
, O_RDWR
);
464 fprintf(stderr
, Name
": weird: %s cannot be opened\n",
468 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
472 ": Cannot add bitmap while array is"
473 " resyncing or reshaping etc.\n");
474 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
475 devname
, strerror(err
));
485 * When reshaping an array we might need to backup some data.
486 * This is written to all spares with a 'super_block' describing it.
487 * The superblock goes 4K from the end of the used space on the
489 * It if written after the backup is complete.
490 * It has the following structure.
493 static struct mdp_backup_super
{
494 char magic
[16]; /* md_backup_data-1 or -2 */
497 /* start/sizes in 512byte sectors */
498 __u64 devstart
; /* address on backup device/file of data */
501 __u32 sb_csum
; /* csum of preceeding bytes. */
503 __u64 devstart2
; /* offset in to data of second section */
506 __u32 sb_csum2
; /* csum of preceeding bytes. */
508 } __attribute__((aligned(512))) bsb
, bsb2
;
510 static __u32
bsb_csum(char *buf
, int len
)
514 for (i
=0; i
<len
; i
++)
515 csum
= (csum
<<3) + buf
[0];
516 return __cpu_to_le32(csum
);
519 static int check_idle(struct supertype
*st
)
521 /* Check that all member arrays for this container, or the
522 * container of this array, are idle
524 int container_dev
= (st
->container_dev
!= NoMdDev
525 ? st
->container_dev
: st
->devnum
);
527 struct mdstat_ent
*ent
, *e
;
530 fmt_devname(container
, container_dev
);
531 ent
= mdstat_read(0, 0);
532 for (e
= ent
; e
; e
= e
->next
) {
533 if (!is_container_member(e
, container
))
535 if (e
->percent
>= 0) {
544 static int freeze_container(struct supertype
*st
)
546 int container_dev
= (st
->container_dev
!= NoMdDev
547 ? st
->container_dev
: st
->devnum
);
553 fmt_devname(container
, container_dev
);
555 if (block_monitor(container
, 1)) {
556 fprintf(stderr
, Name
": failed to freeze container\n");
563 static void unfreeze_container(struct supertype
*st
)
565 int container_dev
= (st
->container_dev
!= NoMdDev
566 ? st
->container_dev
: st
->devnum
);
569 fmt_devname(container
, container_dev
);
571 unblock_monitor(container
, 1);
574 static int freeze(struct supertype
*st
)
576 /* Try to freeze resync/rebuild on this array/container.
577 * Return -1 if the array is busy,
578 * return -2 container cannot be frozen,
579 * return 0 if this kernel doesn't support 'frozen'
580 * return 1 if it worked.
582 if (st
->ss
->external
)
583 return freeze_container(st
);
585 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
591 /* Need to clear any 'read-auto' status */
592 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
593 strncmp(buf
, "read-auto", 9) == 0)
594 sysfs_set_str(sra
, NULL
, "array_state", "clean");
596 err
= sysfs_freeze_array(sra
);
602 static void unfreeze(struct supertype
*st
)
604 if (st
->ss
->external
)
605 return unfreeze_container(st
);
607 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
610 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
612 fprintf(stderr
, Name
": failed to unfreeze array\n");
617 static void wait_reshape(struct mdinfo
*sra
)
619 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
625 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
626 strncmp(action
, "reshape", 7) == 0) {
630 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
635 static int reshape_super(struct supertype
*st
, long long size
, int level
,
636 int layout
, int chunksize
, int raid_disks
,
637 int delta_disks
, char *backup_file
, char *dev
,
640 /* nothing extra to check in the native case */
641 if (!st
->ss
->external
)
643 if (!st
->ss
->reshape_super
||
644 !st
->ss
->manage_reshape
) {
645 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
650 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
651 raid_disks
, delta_disks
, backup_file
, dev
,
655 static void sync_metadata(struct supertype
*st
)
657 if (st
->ss
->external
) {
658 if (st
->update_tail
) {
659 flush_metadata_updates(st
);
660 st
->update_tail
= &st
->updates
;
662 st
->ss
->sync_metadata(st
);
666 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
668 /* when dealing with external metadata subarrays we need to be
669 * prepared to handle EAGAIN. The kernel may need to wait for
670 * mdmon to mark the array active so the kernel can handle
671 * allocations/writeback when preparing the reshape action
672 * (md_allow_write()). We temporarily disable safe_mode_delay
673 * to close a race with the array_state going clean before the
674 * next write to raid_disks / stripe_cache_size
679 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
681 (strcmp(name
, "raid_disks") != 0 &&
682 strcmp(name
, "stripe_cache_size") != 0))
683 return sysfs_set_num(sra
, NULL
, name
, n
);
685 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
688 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
689 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
690 if (rc
< 0 && errno
== EAGAIN
) {
691 ping_monitor(container
);
692 /* if we get EAGAIN here then the monitor is not active
695 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
697 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
701 int start_reshape(struct mdinfo
*sra
, int already_running
, int data_disks
)
704 unsigned long long sync_max_to_set
;
706 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
707 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
708 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
709 sra
->reshape_progress
);
710 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
711 if (!already_running
)
712 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
713 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
714 if (!already_running
)
715 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
720 void abort_reshape(struct mdinfo
*sra
)
722 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
723 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
724 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
725 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
726 sysfs_set_num(sra
, NULL
, "sync_min", 0);
727 sysfs_set_str(sra
, NULL
, "sync_max", "max");
730 int remove_disks_for_takeover(struct supertype
*st
,
735 struct mdinfo
*remaining
;
738 if (sra
->array
.level
== 10)
739 nr_of_copies
= layout
& 0xff;
740 else if (sra
->array
.level
== 1)
741 nr_of_copies
= sra
->array
.raid_disks
;
745 remaining
= sra
->devs
;
747 /* for each 'copy', select one device and remove from the list. */
748 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
749 struct mdinfo
**diskp
;
752 /* Find a working device to keep */
753 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
754 struct mdinfo
*disk
= *diskp
;
756 if (disk
->disk
.raid_disk
< slot
)
758 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
760 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
762 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
764 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
767 /* We have found a good disk to use! */
769 disk
->next
= sra
->devs
;
778 if (slot
< sra
->array
.raid_disks
) {
779 /* didn't find all slots */
785 sra
->devs
= remaining
;
789 /* Remove all 'remaining' devices from the array */
791 struct mdinfo
*sd
= remaining
;
792 remaining
= sd
->next
;
794 sysfs_set_str(sra
, sd
, "state", "faulty");
795 sysfs_set_str(sra
, sd
, "slot", "none");
796 /* for external metadata disks should be removed in mdmon */
797 if (!st
->ss
->external
)
798 sysfs_set_str(sra
, sd
, "state", "remove");
799 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
800 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
801 sd
->next
= sra
->devs
;
807 void reshape_free_fdlist(int *fdlist
,
808 unsigned long long *offsets
,
813 for (i
= 0; i
< size
; i
++)
821 int reshape_prepare_fdlist(char *devname
,
825 unsigned long blocks
,
828 unsigned long long *offsets
)
833 for (d
= 0; d
<= nrdisks
; d
++)
836 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
837 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
839 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
840 char *dn
= map_dev(sd
->disk
.major
,
842 fdlist
[sd
->disk
.raid_disk
]
843 = dev_open(dn
, O_RDONLY
);
844 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
845 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
847 Name
": %s: cannot open component %s\n",
848 devname
, dn
? dn
: "-unknown-");
852 } else if (backup_file
== NULL
) {
854 char *dn
= map_dev(sd
->disk
.major
,
856 fdlist
[d
] = dev_open(dn
, O_RDWR
);
857 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
859 fprintf(stderr
, Name
": %s: cannot open component %s\n",
860 devname
, dn
? dn
: "-unknown-");
871 int reshape_open_backup_file(char *backup_file
,
876 unsigned long long *offsets
,
879 /* Return 1 on success, 0 on any form of failure */
880 /* need to check backup file is large enough */
886 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
890 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
891 devname
, backup_file
, strerror(errno
));
894 /* Guard against backup file being on array device.
895 * If array is partitioned or if LVM etc is in the
896 * way this will not notice, but it is better than
899 fstat(*fdlist
, &stb
);
902 if (stb
.st_rdev
== dev
) {
903 fprintf(stderr
, Name
": backup file must NOT be"
904 " on the array being reshaped.\n");
910 for (i
=0; i
< blocks
+ 8 ; i
++) {
911 if (write(*fdlist
, buf
, 512) != 512) {
912 fprintf(stderr
, Name
": %s: cannot create"
913 " backup file %s: %s\n",
914 devname
, backup_file
, strerror(errno
));
918 if (fsync(*fdlist
) != 0) {
919 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
920 devname
, backup_file
, strerror(errno
));
927 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
928 unsigned int ndata
, unsigned int odata
)
930 unsigned long a
, b
, blocks
;
931 /* So how much do we need to backup.
932 * We need an amount of data which is both a whole number of
933 * old stripes and a whole number of new stripes.
934 * So LCM for (chunksize*datadisks).
936 a
= (ochunk
/512) * odata
;
937 b
= (nchunk
/512) * ndata
;
945 /* LCM == product / GCD */
946 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
951 char *analyse_change(struct mdinfo
*info
, struct reshape
*re
)
953 /* Based on the current array state in info->array and
954 * the changes in info->new_* etc, determine:
955 * - whether the change is possible
956 * - Intermediate level/raid_disks/layout
957 * - whether a restriping reshape is needed
958 * - number of sectors in minimum change unit. This
959 * will cover a whole number of stripes in 'before' and
962 * Return message if the change should be rejected
963 * NULL if the change can be achieved
965 * This can be called as part of starting a reshape, or
966 * when assembling an array that is undergoing reshape.
969 /* delta_parity records change in number of devices
970 * caused by level change
972 int delta_parity
= 0;
974 /* If a new level not explicitly given, we assume no-change */
975 if (info
->new_level
== UnSet
)
976 info
->new_level
= info
->array
.level
;
979 switch (info
->new_level
) {
985 /* chunk size is meaningful, must divide component_size
988 if (info
->component_size
% (info
->new_chunk
/512))
989 return "New chunk size does not"
990 " divide component size";
993 return "chunk size not meaningful for this level";
996 info
->new_chunk
= info
->array
.chunk_size
;
998 switch (info
->array
.level
) {
1000 return "Cannot understand this RAID level";
1002 /* RAID1 can convert to RAID1 with different disks, or
1003 * raid5 with 2 disks, or
1006 if (info
->new_level
== 0) {
1007 if (info
->delta_disks
!= UnSet
&&
1008 info
->delta_disks
!= 0)
1009 return "Cannot change number of disks "
1010 "with RAID1->RAID0 conversion";
1012 re
->before
.data_disks
= 1;
1013 re
->after
.data_disks
= 1;
1014 re
->before
.layout
= 0;
1015 re
->backup_blocks
= 0;
1019 if (info
->new_level
== 1) {
1020 if (info
->delta_disks
== UnSet
)
1021 /* Don't know what to do */
1022 return "no change requested for Growing RAID1";
1024 re
->backup_blocks
= 0;
1028 if (info
->array
.raid_disks
== 2 &&
1029 info
->new_level
== 5) {
1032 re
->before
.data_disks
= 1;
1033 if (info
->delta_disks
!= UnSet
&&
1034 info
->delta_disks
!= 0)
1035 re
->after
.data_disks
= 1 + info
->delta_disks
;
1037 re
->after
.data_disks
= 1;
1038 if (re
->after
.data_disks
< 1)
1039 return "Number of disks too small for RAID5";
1041 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1042 info
->array
.chunk_size
= 65536;
1045 /* Could do some multi-stage conversions, but leave that to
1048 return "Impossibly level change request for RAID1";
1051 /* RAID10 can only be converted from near mode to
1052 * RAID0 by removing some devices
1054 if ((info
->array
.layout
& ~0xff) != 0x100)
1055 return "Cannot Grow RAID10 with far/offset layout";
1056 /* number of devices must be multiple of number of copies */
1057 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1058 return "RAID10 layout too complex for Grow operation";
1060 if (info
->new_level
!= 0)
1061 return "RAID10 can only be changed to RAID0";
1062 new_disks
= (info
->array
.raid_disks
1063 / (info
->array
.layout
& 0xff));
1064 if (info
->delta_disks
== UnSet
)
1065 info
->delta_disks
= (new_disks
1066 - info
->array
.raid_disks
);
1068 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1069 return "New number of raid-devices impossible for RAID10";
1070 if (info
->new_chunk
&&
1071 info
->new_chunk
!= info
->array
.chunk_size
)
1072 return "Cannot change chunk-size with RAID10 Grow";
1077 re
->before
.data_disks
= new_disks
;
1078 re
->after
.data_disks
= re
->before
.data_disks
;
1079 re
->before
.layout
= 0;
1080 re
->backup_blocks
= 0;
1084 /* RAID0 can be converted to RAID10, or to RAID456 */
1085 if (info
->new_level
== 10) {
1086 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1087 /* Assume near=2 layout */
1088 info
->new_layout
= 0x102;
1089 info
->delta_disks
= info
->array
.raid_disks
;
1091 if (info
->new_layout
== UnSet
) {
1092 int copies
= 1 + (info
->delta_disks
1093 / info
->array
.raid_disks
);
1094 if (info
->array
.raid_disks
* (copies
-1)
1095 != info
->delta_disks
)
1096 return "Impossible number of devices"
1097 " for RAID0->RAID10";
1098 info
->new_layout
= 0x100 + copies
;
1100 if (info
->delta_disks
== UnSet
) {
1101 int copies
= info
->new_layout
& 0xff;
1102 if (info
->new_layout
!= 0x100 + copies
)
1103 return "New layout impossible"
1104 " for RAID0->RAID10";;
1105 info
->delta_disks
= (copies
- 1) *
1106 info
->array
.raid_disks
;
1108 if (info
->new_chunk
&&
1109 info
->new_chunk
!= info
->array
.chunk_size
)
1110 return "Cannot change chunk-size with RAID0->RAID10";
1114 re
->before
.data_disks
= (info
->array
.raid_disks
+
1116 re
->after
.data_disks
= re
->before
.data_disks
;
1117 re
->before
.layout
= info
->new_layout
;
1118 re
->backup_blocks
= 0;
1122 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1123 * a raid4 style layout of the final level.
1125 switch (info
->new_level
) {
1130 re
->before
.layout
= 0;
1135 re
->before
.layout
= ALGORITHM_PARITY_N
;
1140 re
->before
.layout
= ALGORITHM_PARITY_N
;
1143 return "Impossible level change requested";
1145 re
->before
.data_disks
= info
->array
.raid_disks
;
1146 /* determining 'after' layout happens outside this 'switch' */
1150 info
->array
.layout
= ALGORITHM_PARITY_N
;
1152 switch (info
->new_level
) {
1156 re
->level
= info
->array
.level
;
1157 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1158 re
->before
.layout
= info
->array
.layout
;
1162 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1163 re
->before
.layout
= info
->array
.layout
;
1168 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1169 switch (info
->array
.layout
) {
1170 case ALGORITHM_LEFT_ASYMMETRIC
:
1171 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1173 case ALGORITHM_RIGHT_ASYMMETRIC
:
1174 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1176 case ALGORITHM_LEFT_SYMMETRIC
:
1177 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1179 case ALGORITHM_RIGHT_SYMMETRIC
:
1180 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1182 case ALGORITHM_PARITY_0
:
1183 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1185 case ALGORITHM_PARITY_N
:
1186 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1189 return "Cannot convert an array with this layout";
1193 if (info
->array
.raid_disks
!= 2)
1194 return "Can only convert a 2-device array to RAID1";
1195 if (info
->delta_disks
!= UnSet
&&
1196 info
->delta_disks
!= 0)
1197 return "Cannot set raid_disk when "
1198 "converting RAID5->RAID1";
1202 return "Impossible level change requested";
1206 switch (info
->new_level
) {
1212 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1213 re
->before
.layout
= info
->array
.layout
;
1216 return "Impossible level change requested";
1221 /* If we reached here then it looks like a re-stripe is
1222 * happening. We have determined the intermediate level
1223 * and initial raid_disks/layout and stored these in 're'.
1225 * We need to deduce the final layout that can be atomically
1226 * converted to the end state.
1228 switch (info
->new_level
) {
1230 /* We can only get to RAID0 from RAID4 or RAID5
1231 * with appropriate layout and one extra device
1233 if (re
->level
!= 4 && re
->level
!= 5)
1234 return "Cannot covert to RAID0 from this level";
1236 switch (re
->level
) {
1238 re
->after
.layout
= 0 ; break;
1240 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1245 /* We can only get to RAID4 from RAID5 */
1246 if (re
->level
!= 4 && re
->level
!= 5)
1247 return "Cannot convert to RAID4 from this level";
1249 switch (re
->level
) {
1251 re
->after
.layout
= 0 ; break;
1253 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1258 /* We get to RAID5 for RAID5 or RAID6 */
1259 if (re
->level
!= 5 && re
->level
!= 6)
1260 return "Cannot convert to RAID5 from this level";
1262 switch (re
->level
) {
1264 if (info
->new_layout
== UnSet
)
1265 re
->after
.layout
= re
->before
.layout
;
1267 re
->after
.layout
= info
->new_layout
;
1270 if (info
->new_layout
== UnSet
)
1271 info
->new_layout
= re
->before
.layout
;
1273 /* after.layout needs to be raid6 version of new_layout */
1274 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1275 re
->after
.layout
= ALGORITHM_PARITY_N
;
1278 char *ls
= map_num(r5layout
, info
->new_layout
);
1280 strcat(strcpy(layout
, ls
), "-6");
1281 l
= map_name(r6layout
, layout
);
1283 return "Cannot find RAID6 layout"
1285 re
->after
.layout
= l
;
1291 /* We must already be at level 6 */
1293 return "Impossible level change";
1294 if (info
->new_layout
== UnSet
)
1295 re
->after
.layout
= info
->array
.layout
;
1297 re
->after
.layout
= info
->new_layout
;
1300 return "Impossible level change requested";
1302 if (info
->delta_disks
== UnSet
)
1303 info
->delta_disks
= delta_parity
;
1305 re
->after
.data_disks
= (re
->before
.data_disks
1308 switch (re
->level
) {
1309 case 6: re
->parity
= 2; break;
1311 case 5: re
->parity
= 1; break;
1312 default: re
->parity
= 0; break;
1314 /* So we have a restripe operation, we need to calculate the number
1315 * of blocks per reshape operation.
1317 if (info
->new_chunk
== 0)
1318 info
->new_chunk
= info
->array
.chunk_size
;
1319 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1320 re
->after
.layout
== re
->before
.layout
&&
1321 info
->new_chunk
== info
->array
.chunk_size
) {
1322 /* Nothing to change */
1323 re
->backup_blocks
= 0;
1326 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1327 /* chunk and layout changes make no difference */
1328 re
->backup_blocks
= 0;
1332 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1333 get_linux_version() < 2006032)
1334 return "in-place reshape is not safe before 2.6.32 - sorry.";
1336 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1337 get_linux_version() < 2006030)
1338 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1340 re
->backup_blocks
= compute_backup_blocks(
1341 info
->new_chunk
, info
->array
.chunk_size
,
1342 re
->after
.data_disks
,
1343 re
->before
.data_disks
);
1345 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1349 static int reshape_array(char *container
, int fd
, char *devname
,
1350 struct supertype
*st
, struct mdinfo
*info
,
1351 int force
, struct mddev_dev
*devlist
,
1352 char *backup_file
, int quiet
, int forked
,
1353 int restart
, int freeze_reshape
);
1354 static int reshape_container(char *container
, char *devname
,
1356 struct supertype
*st
,
1357 struct mdinfo
*info
,
1360 int quiet
, int restart
, int freeze_reshape
);
1362 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1364 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1365 struct mddev_dev
*devlist
,
1366 int assume_clean
, int force
)
1368 /* Make some changes in the shape of an array.
1369 * The kernel must support the change.
1371 * There are three different changes. Each can trigger
1372 * a resync or recovery so we freeze that until we have
1373 * requested everything (if kernel supports freezing - 2.6.30).
1375 * - change size (i.e. component_size)
1377 * - change layout/chunksize/ndisks
1379 * The last can require a reshape. It is different on different
1380 * levels so we need to check the level before actioning it.
1381 * Some times the level change needs to be requested after the
1382 * reshape (e.g. raid6->raid5, raid5->raid0)
1385 struct mdu_array_info_s array
;
1387 struct supertype
*st
;
1388 char *subarray
= NULL
;
1392 char *container
= NULL
;
1393 char container_buf
[20];
1396 struct mddev_dev
*dv
;
1402 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1403 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1409 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1410 fprintf(stderr
, Name
": cannot change component size at the same time "
1411 "as other changes.\n"
1412 " Change size first, then check data is intact before "
1413 "making other changes.\n");
1417 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1418 get_linux_version() < 2006032 &&
1419 !check_env("MDADM_FORCE_FEWER")) {
1420 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1421 " Please use a newer kernel\n");
1425 st
= super_by_fd(fd
, &subarray
);
1427 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1430 if (raid_disks
> st
->max_devs
) {
1431 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1432 " beyond %d\n", st
->max_devs
);
1436 /* in the external case we need to check that the requested reshape is
1437 * supported, and perform an initial check that the container holds the
1438 * pre-requisite spare devices (mdmon owns final validation)
1440 if (st
->ss
->external
) {
1445 container_dev
= st
->container_dev
;
1446 cfd
= open_dev_excl(st
->container_dev
);
1448 container_dev
= st
->devnum
;
1450 cfd
= open_dev_excl(st
->devnum
);
1454 fprintf(stderr
, Name
": Unable to open container for %s\n",
1460 fmt_devname(container_buf
, container_dev
);
1461 container
= container_buf
;
1463 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1466 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1472 /* check if operation is supported for metadata handler */
1473 if (st
->ss
->container_content
) {
1474 struct mdinfo
*cc
= NULL
;
1475 struct mdinfo
*content
= NULL
;
1477 cc
= st
->ss
->container_content(st
, subarray
);
1478 for (content
= cc
; content
; content
= content
->next
) {
1479 int allow_reshape
= 1;
1481 /* check if reshape is allowed based on metadata
1482 * indications stored in content.array.status
1484 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
1486 if (content
->array
.state
1487 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))
1489 if (!allow_reshape
) {
1490 fprintf(stderr
, Name
1491 " cannot reshape arrays in"
1492 " container with unsupported"
1493 " metadata: %s(%s)\n",
1494 devname
, container_buf
);
1502 if (mdmon_running(container_dev
))
1503 st
->update_tail
= &st
->updates
;
1507 for (dv
= devlist
; dv
; dv
= dv
->next
)
1509 if (raid_disks
> array
.raid_disks
&&
1510 array
.spare_disks
+added_disks
< (raid_disks
- array
.raid_disks
) &&
1513 Name
": Need %d spare%s to avoid degraded array,"
1514 " and only have %d.\n"
1515 " Use --force to over-ride this check.\n",
1516 raid_disks
- array
.raid_disks
,
1517 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1518 array
.spare_disks
+ added_disks
);
1522 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
1523 | GET_STATE
| GET_VERSION
);
1525 if (st
->ss
->external
&& subarray
== NULL
) {
1526 array
.level
= LEVEL_CONTAINER
;
1527 sra
->array
.level
= LEVEL_CONTAINER
;
1530 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1534 frozen
= freeze(st
);
1536 /* freeze() already spewed the reason */
1538 } else if (frozen
< 0) {
1539 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1540 " be reshaped\n", devname
);
1544 /* ========= set size =============== */
1545 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1546 long long orig_size
= get_component_size(fd
)/2;
1547 long long min_csize
;
1551 orig_size
= array
.size
;
1553 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1560 /* Update the size of each member device in case
1561 * they have been resized. This will never reduce
1562 * below the current used-size. The "size" attribute
1563 * understands '0' to mean 'max'.
1566 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
1567 if (sysfs_set_num(sra
, mdi
, "size", size
) < 0)
1569 if (array
.not_persistent
== 0 &&
1570 array
.major_version
== 0 &&
1571 get_linux_version() < 3001000) {
1572 /* Dangerous to allow size to exceed 2TB */
1573 unsigned long long csize
;
1574 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
1575 if (csize
>= 2ULL*1024*1024*1024)
1576 csize
= 2ULL*1024*1024*1024;
1577 if ((min_csize
== 0 || (min_csize
1578 > (long long)csize
)))
1583 if (min_csize
&& size
> min_csize
) {
1584 fprintf(stderr
, Name
": Cannot safely make this array "
1585 "use more than 2TB per device on this kernel.\n");
1589 if (min_csize
&& size
== 0) {
1590 /* Don't let the kernel choose a size - it will get
1593 fprintf(stderr
, Name
": Limited v0.90 array to "
1594 "2TB per device\n");
1599 if (array
.size
!= size
) {
1600 /* got truncated to 32bit, write to
1601 * component_size instead
1604 rv
= sysfs_set_num(sra
, NULL
,
1605 "component_size", size
);
1609 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1613 /* restore metadata */
1614 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1615 UnSet
, NULL
, devname
, !quiet
) == 0)
1617 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1618 devname
, strerror(err
));
1620 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1621 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1626 /* This will fail on kernels newer than 3.0 unless
1627 * a backport has been arranged.
1630 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
1631 fprintf(stderr
, Name
": --assume-clean not support with --grow on this kernel\n");
1633 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1634 size
= get_component_size(fd
)/2;
1638 if (size
== orig_size
)
1639 fprintf(stderr
, Name
": component size of %s "
1640 "unchanged at %lluK\n",
1643 fprintf(stderr
, Name
": component size of %s "
1644 "has been set to %lluK\n",
1648 } else if (array
.level
!= LEVEL_CONTAINER
) {
1649 size
= get_component_size(fd
)/2;
1654 /* See if there is anything else to do */
1655 if ((level
== UnSet
|| level
== array
.level
) &&
1656 (layout_str
== NULL
) &&
1657 (chunksize
== 0 || chunksize
== array
.chunk_size
) &&
1658 (raid_disks
== 0 || raid_disks
== array
.raid_disks
)) {
1659 /* Nothing more to do */
1660 if (!changed
&& !quiet
)
1661 fprintf(stderr
, Name
": %s: no change requested\n",
1666 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1667 * current implementation assumes that following conditions must be met:
1670 * - near_copies == 2
1672 if ((level
== 0 && array
.level
== 10 && sra
&&
1673 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1674 (level
== 0 && array
.level
== 1 && sra
)) {
1676 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1678 dprintf(Name
": Array cannot be reshaped\n");
1684 /* Make sure mdmon has seen the device removal
1685 * and updated metadata before we continue with
1689 ping_monitor(container
);
1692 memset(&info
, 0, sizeof(info
));
1694 sysfs_init(&info
, fd
, NoMdDev
);
1695 strcpy(info
.text_version
, sra
->text_version
);
1696 info
.component_size
= size
*2;
1697 info
.new_level
= level
;
1698 info
.new_chunk
= chunksize
* 1024;
1699 if (info
.array
.level
== LEVEL_CONTAINER
) {
1700 info
.delta_disks
= UnSet
;
1701 info
.array
.raid_disks
= raid_disks
;
1702 } else if (raid_disks
)
1703 info
.delta_disks
= raid_disks
- info
.array
.raid_disks
;
1705 info
.delta_disks
= UnSet
;
1706 if (layout_str
== NULL
) {
1707 info
.new_layout
= UnSet
;
1708 if (info
.array
.level
== 6 &&
1709 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1710 info
.array
.layout
>= 16) {
1711 fprintf(stderr
, Name
1712 ": %s has a non-standard layout. If you"
1713 " wish to preserve this\n"
1714 " during the reshape, please specify"
1715 " --layout=preserve\n"
1716 " If you want to change it, specify a"
1717 " layout or use --layout=normalise\n",
1722 } else if (strcmp(layout_str
, "normalise") == 0 ||
1723 strcmp(layout_str
, "normalize") == 0) {
1724 /* If we have a -6 RAID6 layout, remove the '-6'. */
1725 info
.new_layout
= UnSet
;
1726 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1728 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1729 h
= strrchr(l
, '-');
1730 if (h
&& strcmp(h
, "-6") == 0) {
1732 info
.new_layout
= map_name(r6layout
, l
);
1735 } else if (strcmp(layout_str
, "preserve") == 0) {
1736 info
.new_layout
= UnSet
;
1738 int l
= info
.new_level
;
1740 l
= info
.array
.level
;
1743 info
.new_layout
= map_name(r5layout
, layout_str
);
1746 info
.new_layout
= map_name(r6layout
, layout_str
);
1749 info
.new_layout
= parse_layout_10(layout_str
);
1752 info
.new_layout
= parse_layout_faulty(layout_str
);
1755 fprintf(stderr
, Name
": layout not meaningful"
1756 " with this level\n");
1760 if (info
.new_layout
== UnSet
) {
1761 fprintf(stderr
, Name
": layout %s not understood"
1762 " for this level\n",
1769 if (array
.level
== LEVEL_FAULTY
) {
1770 if (level
!= UnSet
&& level
!= array
.level
) {
1771 fprintf(stderr
, Name
": cannot change level of Faulty device\n");
1775 fprintf(stderr
, Name
": cannot set chunksize of Faulty device\n");
1778 if (raid_disks
&& raid_disks
!= 1) {
1779 fprintf(stderr
, Name
": cannot set raid_disks of Faulty device\n");
1783 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1784 dprintf("Cannot get array information.\n");
1787 array
.layout
= info
.new_layout
;
1788 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1789 fprintf(stderr
, Name
": failed to set new layout\n");
1792 printf("layout for %s set to %d\n",
1793 devname
, array
.layout
);
1795 } else if (array
.level
== LEVEL_CONTAINER
) {
1796 /* This change is to be applied to every array in the
1797 * container. This is only needed when the metadata imposes
1798 * restraints of the various arrays in the container.
1799 * Currently we only know that IMSM requires all arrays
1800 * to have the same number of devices so changing the
1801 * number of devices (On-Line Capacity Expansion) must be
1802 * performed at the level of the container
1804 rv
= reshape_container(container
, devname
, -1, st
, &info
,
1805 force
, backup_file
, quiet
, 0, 0);
1808 /* get spare devices from external metadata
1810 if (st
->ss
->external
) {
1811 struct mdinfo
*info2
;
1813 info2
= st
->ss
->container_content(st
, subarray
);
1815 info
.array
.spare_disks
=
1816 info2
->array
.spare_disks
;
1821 /* Impose these changes on a single array. First
1822 * check that the metadata is OK with the change. */
1824 if (reshape_super(st
, info
.component_size
, info
.new_level
,
1825 info
.new_layout
, info
.new_chunk
,
1826 info
.array
.raid_disks
, info
.delta_disks
,
1827 backup_file
, devname
, quiet
)) {
1832 rv
= reshape_array(container
, fd
, devname
, st
, &info
, force
,
1833 devlist
, backup_file
, quiet
, 0, 0, 0);
1842 static int reshape_array(char *container
, int fd
, char *devname
,
1843 struct supertype
*st
, struct mdinfo
*info
,
1844 int force
, struct mddev_dev
*devlist
,
1845 char *backup_file
, int quiet
, int forked
,
1846 int restart
, int freeze_reshape
)
1848 struct reshape reshape
;
1851 int orig_level
= UnSet
;
1854 struct mdu_array_info_s array
;
1857 struct mddev_dev
*dv
;
1861 unsigned long long *offsets
= NULL
;
1865 unsigned long blocks
;
1866 unsigned long cache
;
1867 unsigned long long array_size
;
1869 struct mdinfo
*sra
= NULL
;
1871 /* when reshaping a RAID0, the component_size might be zero.
1872 * So try to fix that up.
1874 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1875 dprintf("Cannot get array information.\n");
1878 if (array
.level
== 0 && info
->component_size
== 0) {
1879 get_dev_size(fd
, NULL
, &array_size
);
1880 info
->component_size
= array_size
/ array
.raid_disks
;
1883 if (info
->reshape_active
) {
1884 int new_level
= info
->new_level
;
1885 info
->new_level
= UnSet
;
1886 info
->array
.raid_disks
-= info
->delta_disks
;
1887 msg
= analyse_change(info
, &reshape
);
1888 info
->new_level
= new_level
;
1889 info
->array
.raid_disks
+= info
->delta_disks
;
1891 /* Make sure the array isn't read-only */
1892 ioctl(fd
, RESTART_ARRAY_RW
, 0);
1894 msg
= analyse_change(info
, &reshape
);
1896 fprintf(stderr
, Name
": %s\n", msg
);
1900 (reshape
.level
!= info
->array
.level
||
1901 reshape
.before
.layout
!= info
->array
.layout
||
1902 reshape
.before
.data_disks
+ reshape
.parity
1903 != info
->array
.raid_disks
- info
->delta_disks
)) {
1904 fprintf(stderr
, Name
": reshape info is not in native format -"
1905 " cannot continue.\n");
1910 /* reshape already started. just skip to monitoring the reshape */
1911 if (reshape
.backup_blocks
== 0)
1915 /* The container is frozen but the array may not be.
1916 * So freeze the array so spares don't get put to the wrong use
1917 * FIXME there should probably be a cleaner separation between
1918 * freeze_array and freeze_container.
1920 sysfs_freeze_array(info
);
1921 /* Check we have enough spares to not be degraded */
1923 for (dv
= devlist
; dv
; dv
=dv
->next
)
1925 spares_needed
= max(reshape
.before
.data_disks
,
1926 reshape
.after
.data_disks
)
1927 + reshape
.parity
- array
.raid_disks
;
1930 info
->new_level
> 1 && info
->array
.level
> 1 &&
1931 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
1933 Name
": Need %d spare%s to avoid degraded array,"
1934 " and only have %d.\n"
1935 " Use --force to over-ride this check.\n",
1937 spares_needed
== 1 ? "" : "s",
1938 info
->array
.spare_disks
+ added_disks
);
1941 /* Check we have enough spares to not fail */
1942 spares_needed
= max(reshape
.before
.data_disks
,
1943 reshape
.after
.data_disks
)
1945 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
1946 spares_needed
> info
->array
.spare_disks
+added_disks
) {
1948 Name
": Need %d spare%s to create working array,"
1949 " and only have %d.\n",
1951 spares_needed
== 1 ? "" : "s",
1952 info
->array
.spare_disks
+ added_disks
);
1956 if (reshape
.level
!= array
.level
) {
1957 char *c
= map_num(pers
, reshape
.level
);
1962 err
= sysfs_set_str(info
, NULL
, "level", c
);
1965 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1968 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1969 fprintf(stderr
, " Bitmap must be removed"
1970 " before level can be changed\n");
1974 fprintf(stderr
, Name
": level of %s changed to %s\n",
1976 orig_level
= array
.level
;
1977 sysfs_freeze_array(info
);
1979 if (reshape
.level
> 0 && st
->ss
->external
) {
1980 /* make sure mdmon is aware of the new level */
1981 if (!mdmon_running(st
->container_dev
))
1982 start_mdmon(st
->container_dev
);
1983 ping_monitor(container
);
1984 if (mdmon_running(st
->container_dev
) &&
1985 st
->update_tail
== NULL
)
1986 st
->update_tail
= &st
->updates
;
1989 /* ->reshape_super might have chosen some spares from the
1990 * container that it wants to be part of the new array.
1991 * We can collect them with ->container_content and give
1992 * them to the kernel.
1994 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
1995 char *subarray
= strchr(info
->text_version
+1, '/')+1;
1996 struct mdinfo
*info2
=
1997 st
->ss
->container_content(st
, subarray
);
2001 sysfs_init(info2
, fd
, st
->devnum
);
2002 /* When increasing number of devices, we need to set
2003 * new raid_disks before adding these, or they might
2006 if (reshape
.backup_blocks
&&
2007 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
2008 subarray_set_num(container
, info2
, "raid_disks",
2009 reshape
.after
.data_disks
+
2011 for (d
= info2
->devs
; d
; d
= d
->next
) {
2012 if (d
->disk
.state
== 0 &&
2013 d
->disk
.raid_disk
>= 0) {
2014 /* This is a spare that wants to
2015 * be part of the array.
2017 add_disk(fd
, st
, info2
, d
);
2023 /* We might have been given some devices to add to the
2024 * array. Now that the array has been changed to the right
2025 * level and frozen, we can safely add them.
2028 Manage_subdevs(devname
, fd
, devlist
, !quiet
,
2031 if (reshape
.backup_blocks
== 0) {
2032 /* No restriping needed, but we might need to impose
2033 * some more changes: layout, raid_disks, chunk_size
2035 /* read current array info */
2036 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2037 dprintf("Cannot get array information.\n");
2040 /* compare current array info with new values and if
2041 * it is different update them to new */
2042 if (info
->new_layout
!= UnSet
&&
2043 info
->new_layout
!= array
.layout
) {
2044 array
.layout
= info
->new_layout
;
2045 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2046 fprintf(stderr
, Name
": failed to set new layout\n");
2049 printf("layout for %s set to %d\n",
2050 devname
, array
.layout
);
2052 if (info
->delta_disks
!= UnSet
&&
2053 info
->delta_disks
!= 0 &&
2054 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
2055 array
.raid_disks
+= info
->delta_disks
;
2056 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2057 fprintf(stderr
, Name
": failed to set raid disks\n");
2059 } else if (!quiet
) {
2060 printf("raid_disks for %s set to %d\n",
2061 devname
, array
.raid_disks
);
2064 if (info
->new_chunk
!= 0 &&
2065 info
->new_chunk
!= array
.chunk_size
) {
2066 if (sysfs_set_num(info
, NULL
,
2067 "chunk_size", info
->new_chunk
) != 0) {
2068 fprintf(stderr
, Name
": failed to set chunk size\n");
2071 printf("chunk size for %s set to %d\n",
2072 devname
, array
.chunk_size
);
2079 * There are three possibilities.
2080 * 1/ The array will shrink.
2081 * We need to ensure the reshape will pause before reaching
2082 * the 'critical section'. We also need to fork and wait for
2083 * that to happen. When it does we
2084 * suspend/backup/complete/unfreeze
2086 * 2/ The array will not change size.
2087 * This requires that we keep a backup of a sliding window
2088 * so that we can restore data after a crash. So we need
2089 * to fork and monitor progress.
2090 * In future we will allow the data_offset to change, so
2091 * a sliding backup becomes unnecessary.
2093 * 3/ The array will grow. This is relatively easy.
2094 * However the kernel's restripe routines will cheerfully
2095 * overwrite some early data before it is safe. So we
2096 * need to make a backup of the early parts of the array
2097 * and be ready to restore it if rebuild aborts very early.
2098 * For externally managed metadata, we still need a forked
2099 * child to monitor the reshape and suspend IO over the region
2100 * that is being reshaped.
2102 * We backup data by writing it to one spare, or to a
2103 * file which was given on command line.
2105 * In each case, we first make sure that storage is available
2106 * for the required backup.
2108 * - request the shape change.
2109 * - fork to handle backup etc.
2112 /* Check that we can hold all the data */
2113 get_dev_size(fd
, NULL
, &array_size
);
2114 if (reshape
.new_size
< (array_size
/512)) {
2116 Name
": this change will reduce the size of the array.\n"
2117 " use --grow --array-size first to truncate array.\n"
2118 " e.g. mdadm --grow %s --array-size %llu\n",
2119 devname
, reshape
.new_size
/2);
2123 sra
= sysfs_read(fd
, 0,
2124 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
2127 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
2132 /* Decide how many blocks (sectors) for a reshape
2133 * unit. The number we have so far is just a minimum
2135 blocks
= reshape
.backup_blocks
;
2136 if (reshape
.before
.data_disks
==
2137 reshape
.after
.data_disks
) {
2138 /* Make 'blocks' bigger for better throughput, but
2139 * not so big that we reject it below.
2140 * Try for 16 megabytes
2142 while (blocks
* 32 < sra
->component_size
&&
2146 fprintf(stderr
, Name
": Need to backup %luK of critical "
2147 "section..\n", blocks
/2);
2149 if (blocks
>= sra
->component_size
/2) {
2150 fprintf(stderr
, Name
": %s: Something wrong"
2151 " - reshape aborted\n",
2156 /* Now we need to open all these devices so we can read/write.
2158 nrdisks
= max(reshape
.before
.data_disks
,
2159 reshape
.after
.data_disks
) + reshape
.parity
2160 + sra
->array
.spare_disks
;
2161 fdlist
= malloc((1+nrdisks
) * sizeof(int));
2162 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
2163 if (!fdlist
|| !offsets
) {
2164 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
2168 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
2169 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
2170 nrdisks
, blocks
, backup_file
,
2175 if ((st
->ss
->manage_reshape
== NULL
) ||
2176 (st
->ss
->recover_backup
== NULL
)) {
2177 if (backup_file
== NULL
) {
2178 if (reshape
.after
.data_disks
<=
2179 reshape
.before
.data_disks
) {
2180 fprintf(stderr
, Name
": %s: Cannot grow - "
2181 "need backup-file\n", devname
);
2183 } else if (sra
->array
.spare_disks
== 0) {
2184 fprintf(stderr
, Name
": %s: Cannot grow - "
2185 "need a spare or backup-file to backup "
2186 "critical section\n", devname
);
2190 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
2192 fdlist
+d
, offsets
+d
,
2200 /* lastly, check that the internal stripe cache is
2201 * large enough, or it won't work.
2202 * It must hold at least 4 stripes of the larger
2205 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2206 cache
*= 4; /* 4 stripes minimum */
2207 cache
/= 512; /* convert to sectors */
2208 disks
= min(reshape
.before
.data_disks
, reshape
.after
.data_disks
);
2209 /* make sure there is room for 'blocks' with a bit to spare */
2210 if (cache
< 16 + blocks
/ disks
)
2211 cache
= 16 + blocks
/ disks
;
2212 cache
/= (4096/512); /* Covert from sectors to pages */
2214 if (sra
->cache_size
< cache
)
2215 subarray_set_num(container
, sra
, "stripe_cache_size",
2218 /* Right, everything seems fine. Let's kick things off.
2219 * If only changing raid_disks, use ioctl, else use
2224 sra
->new_chunk
= info
->new_chunk
;
2227 sra
->reshape_progress
= info
->reshape_progress
;
2229 sra
->reshape_progress
= 0;
2230 if (reshape
.after
.data_disks
< reshape
.before
.data_disks
)
2231 /* start from the end of the new array */
2232 sra
->reshape_progress
= (sra
->component_size
2233 * reshape
.after
.data_disks
);
2236 if (info
->array
.chunk_size
== info
->new_chunk
&&
2237 reshape
.before
.layout
== reshape
.after
.layout
&&
2238 st
->ss
->external
== 0) {
2239 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2240 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2241 array
.raid_disks
= reshape
.after
.data_disks
+ reshape
.parity
;
2243 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2247 Name
": Cannot set device shape for %s: %s\n",
2248 devname
, strerror(errno
));
2251 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2253 " Bitmap must be removed before"
2254 " shape can be changed\n");
2258 } else if (!restart
) {
2259 /* set them all just in case some old 'new_*' value
2260 * persists from some earlier problem.
2263 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2265 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2266 reshape
.after
.layout
) < 0)
2268 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2269 reshape
.after
.data_disks
+
2270 reshape
.parity
) < 0)
2273 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
2277 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2279 " Bitmap must be removed before"
2280 " shape can be changed\n");
2285 err
= start_reshape(sra
, restart
,
2286 info
->array
.raid_disks
- reshape
.parity
);
2289 Name
": Cannot %s reshape for %s\n",
2290 restart
? "continue" : "start",
2295 sysfs_set_str(sra
, NULL
, "array_state", "active");
2296 if (freeze_reshape
) {
2300 fprintf(stderr
, Name
": Reshape has to be continued from"
2301 " location %llu when root fileststem has been mounted\n",
2302 sra
->reshape_progress
);
2306 /* Now we just need to kick off the reshape and watch, while
2307 * handling backups of the data...
2308 * This is all done by a forked background process.
2310 switch(forked
? 0 : fork()) {
2312 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
2327 if (check_env("MDADM_GROW_VERIFY"))
2328 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
2331 mlockall(MCL_FUTURE
);
2333 if (st
->ss
->external
) {
2334 /* metadata handler takes it from here */
2335 done
= st
->ss
->manage_reshape(
2336 fd
, sra
, &reshape
, st
, blocks
,
2338 d
- odisks
, fdlist
+odisks
,
2341 done
= child_monitor(
2342 fd
, sra
, &reshape
, st
, blocks
,
2344 d
- odisks
, fdlist
+odisks
,
2350 if (backup_file
&& done
)
2351 unlink(backup_file
);
2357 if (!st
->ss
->external
&&
2358 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
2359 && info
->custom_array_size
) &&
2360 info
->new_level
== reshape
.level
&&
2362 /* no need to wait for the reshape to finish as
2363 * there is nothing more to do.
2370 if (st
->ss
->external
) {
2371 /* Re-load the metadata as much could have changed */
2372 int cfd
= open_dev(st
->container_dev
);
2374 ping_monitor(container
);
2375 st
->ss
->free_super(st
);
2376 st
->ss
->load_container(st
, cfd
, container
);
2381 /* set new array size if required customer_array_size is used
2384 if (reshape
.before
.data_disks
!=
2385 reshape
.after
.data_disks
&&
2386 info
->custom_array_size
) {
2387 struct mdinfo
*info2
;
2388 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2390 info2
= st
->ss
->container_content(st
, subarray
);
2392 unsigned long long current_size
= 0;
2393 unsigned long long new_size
=
2394 info2
->custom_array_size
/2;
2396 if (sysfs_get_ll(sra
,
2399 ¤t_size
) == 0 &&
2400 new_size
> current_size
) {
2401 if (sysfs_set_num(sra
, NULL
,
2402 "array_size", new_size
)
2404 dprintf("Error: Cannot"
2407 dprintf("Array size "
2409 dprintf(" from %llu to %llu.\n",
2410 current_size
, new_size
);
2416 if (info
->new_level
!= reshape
.level
) {
2418 c
= map_num(pers
, info
->new_level
);
2420 err
= sysfs_set_str(sra
, NULL
, "level", c
);
2422 fprintf(stderr
, Name\
2423 ": %s: could not set level "
2424 "to %s\n", devname
, c
);
2426 if (info
->new_level
== 0)
2427 st
->update_tail
= NULL
;
2439 if (orig_level
!= UnSet
&& sra
) {
2440 c
= map_num(pers
, orig_level
);
2441 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
2442 fprintf(stderr
, Name
": aborting level change\n");
2450 /* mdfd handle is passed to be closed in child process (after fork).
2452 int reshape_container(char *container
, char *devname
,
2454 struct supertype
*st
,
2455 struct mdinfo
*info
,
2458 int quiet
, int restart
, int freeze_reshape
)
2460 struct mdinfo
*cc
= NULL
;
2463 /* component_size is not meaningful for a container,
2464 * so pass '-1' meaning 'no change'
2467 reshape_super(st
, -1, info
->new_level
,
2468 info
->new_layout
, info
->new_chunk
,
2469 info
->array
.raid_disks
, info
->delta_disks
,
2470 backup_file
, devname
, quiet
)) {
2477 /* ping monitor to be sure that update is on disk
2479 ping_monitor(container
);
2482 case -1: /* error */
2483 perror("Cannot fork to complete reshape\n");
2486 default: /* parent */
2487 if (!freeze_reshape
)
2488 printf(Name
": multi-array reshape continues"
2489 " in background\n");
2496 /* close unused handle in child process
2502 /* For each member array with reshape_active,
2503 * we need to perform the reshape.
2504 * We pick the first array that needs reshaping and
2505 * reshape it. reshape_array() will re-read the metadata
2506 * so the next time through a different array should be
2507 * ready for reshape.
2508 * It is possible that the 'different' array will not
2509 * be assembled yet. In that case we simple exit.
2510 * When it is assembled, the mdadm which assembles it
2511 * will take over the reshape.
2513 struct mdinfo
*content
;
2515 struct mdstat_ent
*mdstat
;
2520 cc
= st
->ss
->container_content(st
, NULL
);
2522 for (content
= cc
; content
; content
= content
->next
) {
2524 if (!content
->reshape_active
)
2527 subarray
= strchr(content
->text_version
+1, '/')+1;
2528 mdstat
= mdstat_by_subdev(subarray
,
2529 devname2devnum(container
));
2537 fd
= open_dev(mdstat
->devnum
);
2540 adev
= map_dev(dev2major(mdstat
->devnum
),
2541 dev2minor(mdstat
->devnum
),
2544 adev
= content
->text_version
;
2546 sysfs_init(content
, fd
, mdstat
->devnum
);
2548 rv
= reshape_array(container
, fd
, adev
, st
,
2549 content
, force
, NULL
,
2550 backup_file
, quiet
, 1, restart
,
2554 if (freeze_reshape
) {
2570 * We run a child process in the background which performs the following
2572 * - wait for resync to reach a certain point
2573 * - suspend io to the following section
2574 * - backup that section
2575 * - allow resync to proceed further
2577 * - discard the backup.
2579 * When are combined in slightly different ways in the three cases.
2581 * - suspend/backup/allow/wait/resume/discard
2583 * - allow/wait/suspend/backup/allow/wait/resume/discard
2585 * - wait/resume/discard/suspend/backup/allow
2587 * suspend/backup/allow always come together
2588 * wait/resume/discard do too.
2589 * For the same-size case we have two backups to improve flow.
2593 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
2594 unsigned long long backup_point
,
2595 unsigned long long wait_point
,
2596 unsigned long long *suspend_point
,
2597 unsigned long long *reshape_completed
)
2599 /* This function is called repeatedly by the reshape manager.
2600 * It determines how much progress can safely be made and allows
2602 * - 'info' identifies the array and particularly records in
2603 * ->reshape_progress the metadata's knowledge of progress
2604 * This is a sector offset from the start of the array
2605 * of the next array block to be relocated. This number
2606 * may increase from 0 or decrease from array_size, depending
2607 * on the type of reshape that is happening.
2608 * Note that in contrast, 'sync_completed' is a block count of the
2609 * reshape so far. It gives the distance between the start point
2610 * (head or tail of device) and the next place that data will be
2611 * written. It always increases.
2612 * - 'reshape' is the structure created by analyse_change
2613 * - 'backup_point' shows how much the metadata manager has backed-up
2614 * data. For reshapes with increasing progress, it is the next address
2615 * to be backed up, previous addresses have been backed-up. For
2616 * decreasing progress, it is the earliest address that has been
2617 * backed up - later address are also backed up.
2618 * So addresses between reshape_progress and backup_point are
2619 * backed up providing those are in the 'correct' order.
2620 * - 'wait_point' is an array address. When reshape_completed
2621 * passes this point, progress_reshape should return. It might
2622 * return earlier if it determines that ->reshape_progress needs
2623 * to be updated or further backup is needed.
2624 * - suspend_point is maintained by progress_reshape and the caller
2625 * should not touch it except to initialise to zero.
2626 * It is an array address and it only increases in 2.6.37 and earlier.
2627 * This makes it difficult to handle reducing reshapes with
2628 * external metadata.
2629 * However: it is similar to backup_point in that it records the
2630 * other end of a suspended region from reshape_progress.
2631 * it is moved to extend the region that is safe to backup and/or
2633 * - reshape_completed is read from sysfs and returned. The caller
2634 * should copy this into ->reshape_progress when it has reason to
2635 * believe that the metadata knows this, and any backup outside this
2639 * 1 if more data from backup_point - but only as far as suspend_point,
2640 * should be backed up
2641 * 0 if things are progressing smoothly
2642 * -1 if the reshape is finished because it is all done,
2643 * -2 if the reshape is finished due to an error.
2646 int advancing
= (reshape
->after
.data_disks
2647 >= reshape
->before
.data_disks
);
2648 unsigned long long need_backup
; /* All data between start of array and
2649 * here will at some point need to
2652 unsigned long long read_offset
, write_offset
;
2653 unsigned long long write_range
;
2654 unsigned long long max_progress
, target
, completed
;
2655 unsigned long long array_size
= (info
->component_size
2656 * reshape
->before
.data_disks
);
2660 /* First, we unsuspend any region that is now known to be safe.
2661 * If suspend_point is on the 'wrong' side of reshape_progress, then
2662 * we don't have or need suspension at the moment. This is true for
2663 * native metadata when we don't need to back-up.
2666 if (info
->reshape_progress
<= *suspend_point
)
2667 sysfs_set_num(info
, NULL
, "suspend_lo",
2668 info
->reshape_progress
);
2670 /* Note: this won't work in 2.6.37 and before.
2671 * Something somewhere should make sure we don't need it!
2673 if (info
->reshape_progress
>= *suspend_point
)
2674 sysfs_set_num(info
, NULL
, "suspend_hi",
2675 info
->reshape_progress
);
2678 /* Now work out how far it is safe to progress.
2679 * If the read_offset for ->reshape_progress is less than
2680 * 'blocks' beyond the write_offset, we can only progress as far
2682 * Otherwise we can progress until the write_offset for the new location
2683 * reaches (within 'blocks' of) the read_offset at the current location.
2684 * However that region must be suspended unless we are using native
2686 * If we need to suspend more, we limit it to 128M per device, which is
2687 * rather arbitrary and should be some time-based calculation.
2689 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
2690 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
2691 write_range
= info
->new_chunk
/512;
2692 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2693 need_backup
= array_size
;
2695 need_backup
= reshape
->backup_blocks
;
2697 if (read_offset
< write_offset
+ write_range
)
2698 max_progress
= backup_point
;
2702 reshape
->after
.data_disks
;
2704 if (read_offset
> write_offset
- write_range
)
2705 /* Can only progress as far as has been backed up,
2706 * which must be suspended */
2707 max_progress
= backup_point
;
2708 else if (info
->reshape_progress
<= need_backup
)
2709 max_progress
= backup_point
;
2711 if (info
->array
.major_version
>= 0)
2712 /* Can progress until backup is needed */
2713 max_progress
= need_backup
;
2715 /* Can progress until metadata update is required */
2718 reshape
->after
.data_disks
;
2719 /* but data must be suspended */
2720 if (max_progress
< *suspend_point
)
2721 max_progress
= *suspend_point
;
2726 /* We know it is safe to progress to 'max_progress' providing
2727 * it is suspended or we are using native metadata.
2728 * Consider extending suspend_point 128M per device if it
2729 * is less than 64M per device beyond reshape_progress.
2730 * But always do a multiple of 'blocks'
2731 * FIXME this is too big - it takes to long to complete
2734 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
2735 reshape
->after
.data_disks
);
2736 target
/= reshape
->backup_blocks
;
2739 target
*= reshape
->backup_blocks
;
2741 /* For externally managed metadata we always need to suspend IO to
2742 * the area being reshaped so we regularly push suspend_point forward.
2743 * For native metadata we only need the suspend if we are going to do
2747 if ((need_backup
> info
->reshape_progress
2748 || info
->array
.major_version
< 0) &&
2749 *suspend_point
< info
->reshape_progress
+ target
) {
2750 if (need_backup
< *suspend_point
+ 2 * target
)
2751 *suspend_point
= need_backup
;
2752 else if (*suspend_point
+ 2 * target
< array_size
)
2753 *suspend_point
+= 2 * target
;
2755 *suspend_point
= array_size
;
2756 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
2757 if (max_progress
> *suspend_point
)
2758 max_progress
= *suspend_point
;
2761 if (info
->array
.major_version
>= 0) {
2762 /* Only need to suspend when about to backup */
2763 if (info
->reshape_progress
< need_backup
* 2 &&
2764 *suspend_point
> 0) {
2766 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
2767 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
2770 /* Need to suspend continually */
2771 if (info
->reshape_progress
< *suspend_point
)
2772 *suspend_point
= info
->reshape_progress
;
2773 if (*suspend_point
+ target
< info
->reshape_progress
)
2774 /* No need to move suspend region yet */;
2776 if (*suspend_point
>= 2 * target
)
2777 *suspend_point
-= 2 * target
;
2780 sysfs_set_num(info
, NULL
, "suspend_lo",
2783 if (max_progress
< *suspend_point
)
2784 max_progress
= *suspend_point
;
2788 /* now set sync_max to allow that progress. sync_max, like
2789 * sync_completed is a count of sectors written per device, so
2790 * we find the difference between max_progress and the start point,
2791 * and divide that by after.data_disks to get a sync_max
2793 * At the same time we convert wait_point to a similar number
2794 * for comparing against sync_completed.
2796 /* scale down max_progress to per_disk */
2797 max_progress
/= reshape
->after
.data_disks
;
2798 /* Round to chunk size as some kernels give an erroneously high number */
2799 max_progress
/= info
->new_chunk
/512;
2800 max_progress
*= info
->new_chunk
/512;
2801 /* And round to old chunk size as the kernel wants that */
2802 max_progress
/= info
->array
.chunk_size
/512;
2803 max_progress
*= info
->array
.chunk_size
/512;
2804 /* Limit progress to the whole device */
2805 if (max_progress
> info
->component_size
)
2806 max_progress
= info
->component_size
;
2807 wait_point
/= reshape
->after
.data_disks
;
2809 /* switch from 'device offset' to 'processed block count' */
2810 max_progress
= info
->component_size
- max_progress
;
2811 wait_point
= info
->component_size
- wait_point
;
2814 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
2816 /* Now wait. If we have already reached the point that we were
2817 * asked to wait to, don't wait at all, else wait for any change.
2818 * We need to select on 'sync_completed' as that is the place that
2819 * notifications happen, but we are really interested in
2820 * 'reshape_position'
2822 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
2824 goto check_progress
;
2826 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
2827 goto check_progress
;
2829 while (completed
< max_progress
&& completed
< wait_point
) {
2830 /* Check that sync_action is still 'reshape' to avoid
2831 * waiting forever on a dead array
2835 if (sysfs_get_str(info
, NULL
, "sync_action",
2837 strncmp(action
, "reshape", 7) != 0)
2839 /* Some kernels reset 'sync_completed' to zero
2840 * before setting 'sync_action' to 'idle'.
2841 * So we need these extra tests.
2843 if (completed
== 0 && advancing
2844 && info
->reshape_progress
> 0)
2846 if (completed
== 0 && !advancing
2847 && info
->reshape_progress
< (info
->component_size
2848 * reshape
->after
.data_disks
))
2852 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2853 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
2854 goto check_progress
;
2856 /* Some kernels reset 'sync_completed' to zero,
2857 * we need to have real point we are in md
2860 completed
= max_progress
;
2862 /* some kernels can give an incorrectly high 'completed' number */
2863 completed
/= (info
->new_chunk
/512);
2864 completed
*= (info
->new_chunk
/512);
2865 /* Convert 'completed' back in to a 'progress' number */
2866 completed
*= reshape
->after
.data_disks
;
2868 completed
= info
->component_size
* reshape
->after
.data_disks
2871 *reshape_completed
= completed
;
2875 /* We return the need_backup flag. Caller will decide
2876 * how much - a multiple of ->backup_blocks up to *suspend_point
2879 return need_backup
> info
->reshape_progress
;
2881 return need_backup
>= info
->reshape_progress
;
2884 /* if we couldn't read a number from sync_completed, then
2885 * either the reshape did complete, or it aborted.
2886 * We can tell which by checking for 'none' in reshape_position.
2887 * If it did abort, then it might immediately restart if it
2888 * it was just a device failure that leaves us degraded but
2892 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
2893 || strncmp(buf
, "none", 4) != 0) {
2894 /* The abort might only be temporary. Wait up to 10
2895 * seconds for fd to contain a valid number again.
2901 while (fd
>= 0 && rv
< 0 && tv
.tv_sec
> 0) {
2905 if (select(fd
+1, NULL
, NULL
, &rfds
, &tv
) != 1)
2907 switch (sysfs_fd_get_ll(fd
, &completed
)) {
2909 /* all good again */
2912 case -2: /* read error - abort */
2919 return rv
; /* abort */
2921 /* Maybe racing with array shutdown - check state */
2924 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
2925 || strncmp(buf
, "inactive", 8) == 0
2926 || strncmp(buf
, "clear",5) == 0)
2927 return -2; /* abort */
2928 return -1; /* complete */
2933 /* FIXME return status is never checked */
2934 static int grow_backup(struct mdinfo
*sra
,
2935 unsigned long long offset
, /* per device */
2936 unsigned long stripes
, /* per device, in old chunks */
2937 int *sources
, unsigned long long *offsets
,
2938 int disks
, int chunk
, int level
, int layout
,
2939 int dests
, int *destfd
, unsigned long long *destoffsets
,
2940 int part
, int *degraded
,
2943 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2944 * to storage 'destfd' (offset 'destoffsets'), after first
2945 * suspending IO. Then allow resync to continue
2946 * over the suspended section.
2947 * Use part 'part' of the backup-super-block.
2952 unsigned long long ll
;
2954 //printf("offset %llu\n", offset);
2960 /* Check that array hasn't become degraded, else we might backup the wrong data */
2961 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
2962 return -1; /* FIXME this error is ignored */
2963 new_degraded
= (int)ll
;
2964 if (new_degraded
!= *degraded
) {
2965 /* check each device to ensure it is still working */
2967 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2968 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2970 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2972 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2973 strstr(sbuf
, "faulty") ||
2974 strstr(sbuf
, "in_sync") == NULL
) {
2975 /* this device is dead */
2976 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2977 if (sd
->disk
.raid_disk
>= 0 &&
2978 sources
[sd
->disk
.raid_disk
] >= 0) {
2979 close(sources
[sd
->disk
.raid_disk
]);
2980 sources
[sd
->disk
.raid_disk
] = -1;
2985 *degraded
= new_degraded
;
2988 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2989 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2991 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2992 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2995 bsb
.magic
[15] = '2';
2996 for (i
= 0; i
< dests
; i
++)
2998 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
3000 lseek64(destfd
[i
], destoffsets
[i
], 0);
3002 rv
= save_stripes(sources
, offsets
,
3003 disks
, chunk
, level
, layout
,
3005 offset
*512*odata
, stripes
* chunk
* odata
,
3010 bsb
.mtime
= __cpu_to_le64(time(0));
3011 for (i
= 0; i
< dests
; i
++) {
3012 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
3014 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
3015 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
3016 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
3017 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
3020 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
3021 != destoffsets
[i
] - 4096)
3023 if (write(destfd
[i
], &bsb
, 512) != 512)
3025 if (destoffsets
[i
] > 4096) {
3026 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
3027 destoffsets
[i
]+stripes
*chunk
*odata
)
3029 if (write(destfd
[i
], &bsb
, 512) != 512)
3039 /* in 2.6.30, the value reported by sync_completed can be
3040 * less that it should be by one stripe.
3041 * This only happens when reshape hits sync_max and pauses.
3042 * So allow wait_backup to either extent sync_max further
3043 * than strictly necessary, or return before the
3044 * sync has got quite as far as we would really like.
3045 * This is what 'blocks2' is for.
3046 * The various caller give appropriate values so that
3049 /* FIXME return value is often ignored */
3050 static int forget_backup(
3051 int dests
, int *destfd
, unsigned long long *destoffsets
,
3055 * Erase backup 'part' (which is 0 or 1)
3061 bsb
.arraystart2
= __cpu_to_le64(0);
3062 bsb
.length2
= __cpu_to_le64(0);
3064 bsb
.arraystart
= __cpu_to_le64(0);
3065 bsb
.length
= __cpu_to_le64(0);
3067 bsb
.mtime
= __cpu_to_le64(time(0));
3069 for (i
= 0; i
< dests
; i
++) {
3070 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
3071 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
3072 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
3073 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
3074 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
3075 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
3076 destoffsets
[i
]-4096)
3079 write(destfd
[i
], &bsb
, 512) != 512)
3086 static void fail(char *msg
)
3089 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
3090 rv
|= (write(2, "\n", 1) != 1);
3094 static char *abuf
, *bbuf
;
3095 static unsigned long long abuflen
;
3096 static void validate(int afd
, int bfd
, unsigned long long offset
)
3098 /* check that the data in the backup against the array.
3099 * This is only used for regression testing and should not
3100 * be used while the array is active
3104 lseek64(bfd
, offset
- 4096, 0);
3105 if (read(bfd
, &bsb2
, 512) != 512)
3106 fail("cannot read bsb");
3107 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
3108 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
3109 fail("first csum bad");
3110 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
3111 fail("magic is bad");
3112 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
3113 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
3114 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
3115 fail("second csum bad");
3117 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
3118 fail("devstart is wrong");
3121 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
3123 if (abuflen
< len
) {
3127 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
3128 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
3130 /* just stop validating on mem-alloc failure */
3135 lseek64(bfd
, offset
, 0);
3136 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
3137 //printf("len %llu\n", len);
3138 fail("read first backup failed");
3140 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
3141 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
3142 fail("read first from array failed");
3143 if (memcmp(bbuf
, abuf
, len
) != 0) {
3146 printf("offset=%llu len=%llu\n",
3147 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
3148 for (i
=0; i
<len
; i
++)
3149 if (bbuf
[i
] != abuf
[i
]) {
3150 printf("first diff byte %d\n", i
);
3154 fail("data1 compare failed");
3158 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
3160 if (abuflen
< len
) {
3164 abuf
= malloc(abuflen
);
3165 bbuf
= malloc(abuflen
);
3168 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
3169 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
3170 fail("read second backup failed");
3171 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
3172 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
3173 fail("read second from array failed");
3174 if (memcmp(bbuf
, abuf
, len
) != 0)
3175 fail("data2 compare failed");
3179 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
3180 struct supertype
*st
, unsigned long blocks
,
3181 int *fds
, unsigned long long *offsets
,
3182 int dests
, int *destfd
, unsigned long long *destoffsets
)
3184 /* Monitor a reshape where backup is being performed using
3185 * 'native' mechanism - either to a backup file, or
3186 * to some space in a spare.
3190 unsigned long long speed
;
3191 unsigned long long suspend_point
, array_size
;
3192 unsigned long long backup_point
, wait_point
;
3193 unsigned long long reshape_completed
;
3195 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
3196 int part
= 0; /* The next part of the backup area to fill. It may already
3197 * be full, so we need to check */
3198 int level
= reshape
->level
;
3199 int layout
= reshape
->before
.layout
;
3200 int data
= reshape
->before
.data_disks
;
3201 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
3202 int chunk
= sra
->array
.chunk_size
;
3204 unsigned long stripes
;
3207 /* set up the backup-super-block. This requires the
3208 * uuid from the array.
3210 /* Find a superblock */
3211 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3215 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
3217 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
3218 devfd
= dev_open(dn
, O_RDONLY
);
3221 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
3227 fprintf(stderr
, Name
": Cannot find a superblock\n");
3231 memset(&bsb
, 0, 512);
3232 memcpy(bsb
.magic
, "md_backup_data-1", 16);
3233 st
->ss
->uuid_from_super(st
, uuid
);
3234 memcpy(bsb
.set_uuid
, uuid
, 16);
3235 bsb
.mtime
= __cpu_to_le64(time(0));
3236 bsb
.devstart2
= blocks
;
3238 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
3239 reshape
->before
.data_disks
;
3241 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
3242 /* Don't start the 'reshape' */
3244 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
3245 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
3246 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
3250 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
3251 backup_point
= sra
->reshape_progress
;
3254 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
3255 backup_point
= reshape
->backup_blocks
;
3256 suspend_point
= array_size
;
3262 /* Want to return as soon the oldest backup slot can
3263 * be released as that allows us to start backing up
3264 * some more, providing suspend_point has been
3265 * advanced, which it should have.
3268 wait_point
= array_size
;
3269 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
3270 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
3271 __le64_to_cpu(bsb
.length
));
3272 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
3273 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
3274 __le64_to_cpu(bsb
.length2
));
3277 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
3278 wait_point
= __le64_to_cpu(bsb
.arraystart
);
3279 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
3280 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
3283 rv
= progress_reshape(sra
, reshape
,
3284 backup_point
, wait_point
,
3285 &suspend_point
, &reshape_completed
);
3286 /* external metadata would need to ping_monitor here */
3287 sra
->reshape_progress
= reshape_completed
;
3289 /* Clear any backup region that is before 'here' */
3291 if (__le64_to_cpu(bsb
.length
) > 0 &&
3292 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
3293 __le64_to_cpu(bsb
.length
)))
3294 forget_backup(dests
, destfd
,
3296 if (__le64_to_cpu(bsb
.length2
) > 0 &&
3297 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
3298 __le64_to_cpu(bsb
.length2
)))
3299 forget_backup(dests
, destfd
,
3302 if (__le64_to_cpu(bsb
.length
) > 0 &&
3303 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
3304 forget_backup(dests
, destfd
,
3306 if (__le64_to_cpu(bsb
.length2
) > 0 &&
3307 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
3308 forget_backup(dests
, destfd
,
3317 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
3318 /* No longer need to monitor this reshape */
3324 unsigned long long offset
;
3325 unsigned long actual_stripes
;
3326 /* Need to backup some data.
3327 * If 'part' is not used and the desired
3328 * backup size is suspended, do a backup,
3329 * then consider the next part.
3331 /* Check that 'part' is unused */
3332 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
3334 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
3337 offset
= backup_point
/ data
;
3338 actual_stripes
= stripes
;
3340 if (offset
+ actual_stripes
* (chunk
/512) >
3341 sra
->component_size
)
3342 actual_stripes
= ((sra
->component_size
- offset
)
3344 if (offset
+ actual_stripes
* (chunk
/512) >
3348 if (offset
< actual_stripes
* (chunk
/512))
3349 actual_stripes
= offset
/ (chunk
/512);
3350 offset
-= actual_stripes
* (chunk
/512);
3351 if (offset
< suspend_point
/data
)
3354 if (actual_stripes
== 0)
3356 grow_backup(sra
, offset
, actual_stripes
,
3358 disks
, chunk
, level
, layout
,
3359 dests
, destfd
, destoffsets
,
3360 part
, °raded
, buf
);
3361 validate(afd
, destfd
[0], destoffsets
[0]);
3362 /* record where 'part' is up to */
3365 backup_point
+= actual_stripes
* (chunk
/512) * data
;
3367 backup_point
-= actual_stripes
* (chunk
/512) * data
;
3371 /* FIXME maybe call progress_reshape one more time instead */
3372 abort_reshape(sra
); /* remove any remaining suspension */
3373 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3374 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
3380 * If any spare contains md_back_data-1 which is recent wrt mtime,
3381 * write that data into the array and update the super blocks with
3382 * the new reshape_progress
3384 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
3385 char *backup_file
, int verbose
)
3389 unsigned long long *offsets
;
3390 unsigned long long nstripe
, ostripe
;
3393 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
3394 if (info
->array
.level
== 6) odata
--; /* number of data disks */
3395 ndata
= info
->array
.raid_disks
- 1;
3396 if (info
->new_level
== 6) ndata
--;
3398 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
3400 if (info
->delta_disks
<= 0)
3401 /* Didn't grow, so the backup file must have
3405 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
3406 struct mdinfo dinfo
;
3409 char *devname
, namebuf
[20];
3410 unsigned long long lo
, hi
;
3412 /* This was a spare and may have some saved data on it.
3413 * Load the superblock, find and load the
3414 * backup_super_block.
3415 * If either fail, go on to next device.
3416 * If the backup contains no new info, just return
3417 * else restore data and update all superblocks
3419 if (i
== old_disks
-1) {
3420 fd
= open(backup_file
, O_RDONLY
);
3422 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
3423 backup_file
, strerror(errno
));
3426 devname
= backup_file
;
3431 if (st
->ss
->load_super(st
, fd
, NULL
))
3434 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3435 st
->ss
->free_super(st
);
3438 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
3440 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
3441 continue; /* Cannot seek */
3443 sprintf(namebuf
, "device-%d", i
);
3446 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
3448 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
3449 continue; /* Cannot read */
3451 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
3452 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
3454 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
3457 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
3459 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
3460 continue; /* bad checksum */
3462 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
3463 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
3465 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
3466 continue; /* Bad second checksum */
3468 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
3470 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
3471 continue; /* Wrong uuid */
3474 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3475 * sometimes they aren't... So allow considerable flexability in matching, and allow
3476 * this test to be overridden by an environment variable.
3478 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
3479 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
3480 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3481 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
3482 "for array with timestamp %lu\n",
3483 (unsigned long)__le64_to_cpu(bsb
.mtime
),
3484 (unsigned long)info
->array
.utime
);
3487 fprintf(stderr
, Name
": too-old timestamp on "
3488 "backup-metadata on %s\n", devname
);
3489 continue; /* time stamp is too bad */
3493 if (bsb
.magic
[15] == '1') {
3494 if (bsb
.length
== 0)
3496 if (info
->delta_disks
>= 0) {
3497 /* reshape_progress is increasing */
3498 if (__le64_to_cpu(bsb
.arraystart
)
3499 + __le64_to_cpu(bsb
.length
)
3500 < info
->reshape_progress
) {
3503 fprintf(stderr
, Name
3504 ": backup-metadata found on %s but is not needed\n", devname
);
3505 continue; /* No new data here */
3508 /* reshape_progress is decreasing */
3509 if (__le64_to_cpu(bsb
.arraystart
) >=
3510 info
->reshape_progress
)
3511 goto nonew
; /* No new data here */
3514 if (bsb
.length
== 0 && bsb
.length2
== 0)
3516 if (info
->delta_disks
>= 0) {
3517 /* reshape_progress is increasing */
3518 if ((__le64_to_cpu(bsb
.arraystart
)
3519 + __le64_to_cpu(bsb
.length
)
3520 < info
->reshape_progress
)
3522 (__le64_to_cpu(bsb
.arraystart2
)
3523 + __le64_to_cpu(bsb
.length2
)
3524 < info
->reshape_progress
))
3525 goto nonew
; /* No new data here */
3527 /* reshape_progress is decreasing */
3528 if (__le64_to_cpu(bsb
.arraystart
) >=
3529 info
->reshape_progress
&&
3530 __le64_to_cpu(bsb
.arraystart2
) >=
3531 info
->reshape_progress
)
3532 goto nonew
; /* No new data here */
3535 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
3538 fprintf(stderr
, Name
3539 ": Failed to verify secondary backup-metadata block on %s\n",
3541 continue; /* Cannot seek */
3543 /* There should be a duplicate backup superblock 4k before here */
3544 if (lseek64(fd
, -4096, 1) < 0 ||
3545 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
3546 goto second_fail
; /* Cannot find leading superblock */
3547 if (bsb
.magic
[15] == '1')
3548 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
3550 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
3551 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
3552 goto second_fail
; /* Cannot find leading superblock */
3554 /* Now need the data offsets for all devices. */
3555 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
3556 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
3559 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3560 /* FIXME should be this be an error */
3562 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3563 st
->ss
->free_super(st
);
3564 offsets
[j
] = dinfo
.data_offset
* 512;
3566 printf(Name
": restoring critical section\n");
3568 if (restore_stripes(fdlist
, offsets
,
3569 info
->array
.raid_disks
,
3573 fd
, __le64_to_cpu(bsb
.devstart
)*512,
3574 __le64_to_cpu(bsb
.arraystart
)*512,
3575 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
3576 /* didn't succeed, so giveup */
3578 fprintf(stderr
, Name
": Error restoring backup from %s\n",
3584 if (bsb
.magic
[15] == '2' &&
3585 restore_stripes(fdlist
, offsets
,
3586 info
->array
.raid_disks
,
3590 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
3591 __le64_to_cpu(bsb
.devstart2
)*512,
3592 __le64_to_cpu(bsb
.arraystart2
)*512,
3593 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
3594 /* didn't succeed, so giveup */
3596 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
3604 /* Ok, so the data is restored. Let's update those superblocks. */
3608 lo
= __le64_to_cpu(bsb
.arraystart
);
3609 hi
= lo
+ __le64_to_cpu(bsb
.length
);
3611 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
3612 unsigned long long lo1
, hi1
;
3613 lo1
= __le64_to_cpu(bsb
.arraystart2
);
3614 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
3618 } else if (lo
< lo1
)
3624 (info
->reshape_progress
< lo
||
3625 info
->reshape_progress
> hi
))
3626 /* backup does not affect reshape_progress*/ ;
3627 else if (info
->delta_disks
>= 0) {
3628 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
3629 __le64_to_cpu(bsb
.length
);
3630 if (bsb
.magic
[15] == '2') {
3631 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
3632 __le64_to_cpu(bsb
.length2
);
3633 if (p2
> info
->reshape_progress
)
3634 info
->reshape_progress
= p2
;
3637 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
3638 if (bsb
.magic
[15] == '2') {
3639 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
3640 if (p2
< info
->reshape_progress
)
3641 info
->reshape_progress
= p2
;
3644 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
3645 if (fdlist
[j
] < 0) continue;
3646 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3648 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3649 dinfo
.reshape_progress
= info
->reshape_progress
;
3650 st
->ss
->update_super(st
, &dinfo
,
3651 "_reshape_progress",
3653 st
->ss
->store_super(st
, fdlist
[j
]);
3654 st
->ss
->free_super(st
);
3658 /* Didn't find any backup data, try to see if any
3661 if (info
->delta_disks
< 0) {
3662 /* When shrinking, the critical section is at the end.
3663 * So see if we are before the critical section.
3665 unsigned long long first_block
;
3666 nstripe
= ostripe
= 0;
3668 while (ostripe
>= nstripe
) {
3669 ostripe
+= info
->array
.chunk_size
/ 512;
3670 first_block
= ostripe
* odata
;
3671 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
3672 (info
->new_chunk
/512);
3675 if (info
->reshape_progress
>= first_block
)
3678 if (info
->delta_disks
> 0) {
3679 /* See if we are beyond the critical section. */
3680 unsigned long long last_block
;
3681 nstripe
= ostripe
= 0;
3683 while (nstripe
>= ostripe
) {
3684 nstripe
+= info
->new_chunk
/ 512;
3685 last_block
= nstripe
* ndata
;
3686 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
3687 (info
->array
.chunk_size
/512);
3690 if (info
->reshape_progress
>= last_block
)
3693 /* needed to recover critical section! */
3695 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
3699 int Grow_continue_command(char *devname
, int fd
,
3700 char *backup_file
, int verbose
)
3703 struct supertype
*st
= NULL
;
3704 struct mdinfo
*content
= NULL
;
3705 struct mdinfo array
;
3706 char *subarray
= NULL
;
3707 struct mdinfo
*cc
= NULL
;
3708 struct mdstat_ent
*mdstat
= NULL
;
3713 unsigned long long position
;
3715 dprintf("Grow continue from command line called for %s\n",
3718 st
= super_by_fd(fd
, &subarray
);
3719 if (!st
|| !st
->ss
) {
3721 Name
": Unable to determine metadata format for %s\n",
3725 dprintf("Grow continue is run for ");
3726 if (st
->ss
->external
== 0) {
3727 dprintf("native array (%s)\n", devname
);
3728 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
3729 fprintf(stderr
, Name
": %s is not an active md array -"
3730 " aborting\n", devname
);
3732 goto Grow_continue_command_exit
;
3735 sysfs_init(content
, fd
, st
->devnum
);
3740 dprintf("subarray (%s)\n", subarray
);
3741 container_dev
= st
->container_dev
;
3742 cfd
= open_dev_excl(st
->container_dev
);
3744 container_dev
= st
->devnum
;
3746 cfd
= open_dev_excl(st
->devnum
);
3747 dprintf("container (%i)\n", container_dev
);
3751 fprintf(stderr
, Name
": Unable to open container "
3752 "for %s\n", devname
);
3754 goto Grow_continue_command_exit
;
3756 fmt_devname(buf
, container_dev
);
3758 /* find in container array under reshape
3760 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
3763 Name
": Cannot read superblock for %s\n",
3766 goto Grow_continue_command_exit
;
3769 cc
= st
->ss
->container_content(st
, subarray
);
3770 for (content
= cc
; content
; content
= content
->next
) {
3772 int allow_reshape
= 1;
3774 if (content
->reshape_active
== 0)
3776 /* The decision about array or container wide
3777 * reshape is taken in Grow_continue based
3778 * content->reshape_active state, therefore we
3779 * need to check_reshape based on
3780 * reshape_active and subarray name
3782 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
3784 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
3785 (content
->array
.state
3786 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
3789 if (!allow_reshape
) {
3790 fprintf(stderr
, Name
3791 ": cannot continue reshape of an array"
3792 " in container with unsupported"
3793 " metadata: %s(%s)\n",
3796 goto Grow_continue_command_exit
;
3799 array
= strchr(content
->text_version
+1, '/')+1;
3800 mdstat
= mdstat_by_subdev(array
, container_dev
);
3807 Name
": Unable to determine reshaped "
3808 "array for %s\n", devname
);
3810 goto Grow_continue_command_exit
;
3812 fd2
= open_dev(mdstat
->devnum
);
3814 fprintf(stderr
, Name
": cannot open (md%i)\n",
3817 goto Grow_continue_command_exit
;
3820 sysfs_init(content
, fd2
, mdstat
->devnum
);
3822 /* start mdmon in case it is not running
3824 if (!mdmon_running(container_dev
))
3825 start_mdmon(container_dev
);
3828 if (mdmon_running(container_dev
))
3829 st
->update_tail
= &st
->updates
;
3831 fprintf(stderr
, Name
": No mdmon found. "
3832 "Grow cannot continue.\n");
3834 goto Grow_continue_command_exit
;
3838 /* verify that array under reshape is started from
3841 ret_val
= sysfs_get_str(content
, NULL
, "sync_max", buf
, 40);
3843 fprintf(stderr
, Name
3844 ": cannot open verify reshape progress for %s (%i)\n",
3845 content
->sys_name
, ret_val
);
3847 goto Grow_continue_command_exit
;
3849 dprintf(Name
": Read sync_max sysfs entry is: %s\n", buf
);
3850 position
= strtoull(buf
, &ep
, 0);
3851 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' ')) {
3852 fprintf(stderr
, Name
": Fatal error: array reshape was"
3853 " not properly frozen\n");
3855 goto Grow_continue_command_exit
;
3857 position
*= get_data_disks(map_name(pers
, mdstat
->level
),
3858 content
->new_layout
,
3859 content
->array
.raid_disks
);
3860 if (position
!= content
->reshape_progress
) {
3861 fprintf(stderr
, Name
": Fatal error: array reshape was"
3862 " not properly frozen.\n");
3864 goto Grow_continue_command_exit
;
3869 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 0);
3871 Grow_continue_command_exit
:
3876 st
->ss
->free_super(st
);
3877 free_mdstat(mdstat
);
3884 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
3885 char *backup_file
, int freeze_reshape
)
3889 if (!info
->reshape_active
)
3892 if (st
->ss
->external
) {
3894 int cfd
= open_dev(st
->container_dev
);
3899 fmt_devname(container
, st
->container_dev
);
3900 st
->ss
->load_container(st
, cfd
, container
);
3902 ret_val
= reshape_container(container
, NULL
, mdfd
,
3903 st
, info
, 0, backup_file
,
3904 0, 1, freeze_reshape
);
3906 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
3907 NULL
, backup_file
, 0, 0, 1,