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");
139 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
141 fprintf(stderr
, Name
": cannot open %s\n", newdev
);
146 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
147 fprintf(stderr
, Name
": %s is not a block device!\n", newdev
);
152 /* now check out all the devices and make sure we can read the superblock */
153 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
154 mdu_disk_info_t disk
;
157 st
->ss
->free_super(st
);
160 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
161 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
167 dv
= map_dev(disk
.major
, disk
.minor
, 1);
169 fprintf(stderr
, Name
": cannot find device file for device %d\n",
175 fd2
= dev_open(dv
, O_RDWR
);
177 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
183 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
184 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
192 /* Ok, looks good. Lets update the superblock and write it out to
196 info
.disk
.number
= d
;
197 info
.disk
.major
= major(stb
.st_rdev
);
198 info
.disk
.minor
= minor(stb
.st_rdev
);
199 info
.disk
.raid_disk
= d
;
200 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
201 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
204 if (st
->ss
->store_super(st
, nfd
)) {
205 fprintf(stderr
, Name
": Cannot store new superblock on %s\n",
212 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
213 fprintf(stderr
, Name
": Cannot add new disk to this array\n");
216 /* Well, that seems to have worked.
217 * Now go through and update all superblocks
220 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
221 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
226 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
227 mdu_disk_info_t disk
;
231 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
232 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
236 dv
= map_dev(disk
.major
, disk
.minor
, 1);
238 fprintf(stderr
, Name
": cannot find device file for device %d\n",
242 fd2
= dev_open(dv
, O_RDWR
);
244 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
247 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
248 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
252 info
.array
.raid_disks
= nd
+1;
253 info
.array
.nr_disks
= nd
+1;
254 info
.array
.active_disks
= nd
+1;
255 info
.array
.working_disks
= nd
+1;
257 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
260 if (st
->ss
->store_super(st
, fd2
)) {
261 fprintf(stderr
, Name
": Cannot store new superblock on %s\n", dv
);
271 int Grow_addbitmap(char *devname
, int fd
, char *file
, int chunk
, int delay
, int write_behind
, int force
)
274 * First check that array doesn't have a bitmap
275 * Then create the bitmap
278 * For internal bitmaps, we need to check the version,
279 * find all the active devices, and write the bitmap block
282 mdu_bitmap_file_t bmf
;
283 mdu_array_info_t array
;
284 struct supertype
*st
;
285 char *subarray
= NULL
;
286 int major
= BITMAP_MAJOR_HI
;
287 int vers
= md_get_version(fd
);
288 unsigned long long bitmapsize
, array_size
;
291 major
= BITMAP_MAJOR_HOSTENDIAN
;
292 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel"
293 " are not portable\n"
294 " between different architectures. Consider upgrading"
295 " the Linux kernel.\n");
298 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
300 fprintf(stderr
, Name
": Memory allocation failure.\n");
302 fprintf(stderr
, Name
": bitmaps not supported by this kernel.\n");
305 if (bmf
.pathname
[0]) {
306 if (strcmp(file
,"none")==0) {
307 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
308 fprintf(stderr
, Name
": failed to remove bitmap %s\n",
314 fprintf(stderr
, Name
": %s already has a bitmap (%s)\n",
315 devname
, bmf
.pathname
);
318 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
319 fprintf(stderr
, Name
": cannot get array status for %s\n", devname
);
322 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
323 if (strcmp(file
, "none")==0) {
324 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
325 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
326 fprintf(stderr
, Name
": failed to remove internal bitmap.\n");
331 fprintf(stderr
, Name
": Internal bitmap already present on %s\n",
336 if (strcmp(file
, "none") == 0) {
337 fprintf(stderr
, Name
": no bitmap found on %s\n", devname
);
340 if (array
.level
<= 0) {
341 fprintf(stderr
, Name
": Bitmaps not meaningful with level %s\n",
342 map_num(pers
, array
.level
)?:"of this array");
345 bitmapsize
= array
.size
;
347 if (get_dev_size(fd
, NULL
, &array_size
) &&
348 array_size
> (0x7fffffffULL
<<9)) {
349 /* Array is big enough that we cannot trust array.size
350 * try other approaches
352 bitmapsize
= get_component_size(fd
);
354 if (bitmapsize
== 0) {
355 fprintf(stderr
, Name
": Cannot reliably determine size of array to create bitmap - sorry.\n");
359 if (array
.level
== 10) {
360 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
361 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
364 st
= super_by_fd(fd
, &subarray
);
366 fprintf(stderr
, Name
": Cannot understand version %d.%d\n",
367 array
.major_version
, array
.minor_version
);
371 fprintf(stderr
, Name
": Cannot add bitmaps to sub-arrays yet\n");
376 if (strcmp(file
, "internal") == 0) {
378 if (st
->ss
->add_internal_bitmap
== NULL
) {
379 fprintf(stderr
, Name
": Internal bitmaps not supported "
380 "with %s metadata\n", st
->ss
->name
);
383 for (d
=0; d
< st
->max_devs
; d
++) {
384 mdu_disk_info_t disk
;
387 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
389 if (disk
.major
== 0 &&
392 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
394 dv
= map_dev(disk
.major
, disk
.minor
, 1);
396 int fd2
= dev_open(dv
, O_RDWR
);
399 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
400 if (st
->ss
->add_internal_bitmap(
402 &chunk
, delay
, write_behind
,
403 bitmapsize
, 0, major
)
405 st
->ss
->write_bitmap(st
, fd2
);
407 fprintf(stderr
, Name
": failed to create internal bitmap - chunksize problem.\n");
415 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
416 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
419 ": Cannot add bitmap while array is"
420 " resyncing or reshaping etc.\n");
421 fprintf(stderr
, Name
": failed to set internal bitmap.\n");
428 int max_devs
= st
->max_devs
;
430 /* try to load a superblock */
431 for (d
=0; d
<max_devs
; d
++) {
432 mdu_disk_info_t disk
;
436 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
438 if ((disk
.major
==0 && disk
.minor
==0) ||
439 (disk
.state
& (1<<MD_DISK_REMOVED
)))
441 dv
= map_dev(disk
.major
, disk
.minor
, 1);
443 fd2
= dev_open(dv
, O_RDONLY
);
445 st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
447 st
->ss
->uuid_from_super(st
, uuid
);
453 fprintf(stderr
, Name
": cannot find UUID for array!\n");
456 if (CreateBitmap(file
, force
, (char*)uuid
, chunk
,
457 delay
, write_behind
, bitmapsize
, major
)) {
460 bitmap_fd
= open(file
, O_RDWR
);
462 fprintf(stderr
, Name
": weird: %s cannot be opened\n",
466 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
470 ": Cannot add bitmap while array is"
471 " resyncing or reshaping etc.\n");
472 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
473 devname
, strerror(err
));
483 * When reshaping an array we might need to backup some data.
484 * This is written to all spares with a 'super_block' describing it.
485 * The superblock goes 4K from the end of the used space on the
487 * It if written after the backup is complete.
488 * It has the following structure.
491 static struct mdp_backup_super
{
492 char magic
[16]; /* md_backup_data-1 or -2 */
495 /* start/sizes in 512byte sectors */
496 __u64 devstart
; /* address on backup device/file of data */
499 __u32 sb_csum
; /* csum of preceeding bytes. */
501 __u64 devstart2
; /* offset in to data of second section */
504 __u32 sb_csum2
; /* csum of preceeding bytes. */
506 } __attribute__((aligned(512))) bsb
, bsb2
;
508 static __u32
bsb_csum(char *buf
, int len
)
512 for (i
=0; i
<len
; i
++)
513 csum
= (csum
<<3) + buf
[0];
514 return __cpu_to_le32(csum
);
517 static int check_idle(struct supertype
*st
)
519 /* Check that all member arrays for this container, or the
520 * container of this array, are idle
522 int container_dev
= (st
->container_dev
!= NoMdDev
523 ? st
->container_dev
: st
->devnum
);
525 struct mdstat_ent
*ent
, *e
;
528 fmt_devname(container
, container_dev
);
529 ent
= mdstat_read(0, 0);
530 for (e
= ent
; e
; e
= e
->next
) {
531 if (!is_container_member(e
, container
))
533 if (e
->percent
>= 0) {
542 static int freeze_container(struct supertype
*st
)
544 int container_dev
= (st
->container_dev
!= NoMdDev
545 ? st
->container_dev
: st
->devnum
);
551 fmt_devname(container
, container_dev
);
553 if (block_monitor(container
, 1)) {
554 fprintf(stderr
, Name
": failed to freeze container\n");
561 static void unfreeze_container(struct supertype
*st
)
563 int container_dev
= (st
->container_dev
!= NoMdDev
564 ? st
->container_dev
: st
->devnum
);
567 fmt_devname(container
, container_dev
);
569 unblock_monitor(container
, 1);
572 static int freeze(struct supertype
*st
)
574 /* Try to freeze resync/rebuild on this array/container.
575 * Return -1 if the array is busy,
576 * return -2 container cannot be frozen,
577 * return 0 if this kernel doesn't support 'frozen'
578 * return 1 if it worked.
580 if (st
->ss
->external
)
581 return freeze_container(st
);
583 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
589 /* Need to clear any 'read-auto' status */
590 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
591 strncmp(buf
, "read-auto", 9) == 0)
592 sysfs_set_str(sra
, NULL
, "array_state", "clean");
594 err
= sysfs_freeze_array(sra
);
600 static void unfreeze(struct supertype
*st
)
602 if (st
->ss
->external
)
603 return unfreeze_container(st
);
605 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
608 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
610 fprintf(stderr
, Name
": failed to unfreeze array\n");
615 static void wait_reshape(struct mdinfo
*sra
)
617 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
623 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
624 strncmp(action
, "reshape", 7) == 0) {
628 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
633 static int reshape_super(struct supertype
*st
, long long size
, int level
,
634 int layout
, int chunksize
, int raid_disks
,
635 int delta_disks
, char *backup_file
, char *dev
,
638 /* nothing extra to check in the native case */
639 if (!st
->ss
->external
)
641 if (!st
->ss
->reshape_super
||
642 !st
->ss
->manage_reshape
) {
643 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
648 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
649 raid_disks
, delta_disks
, backup_file
, dev
,
653 static void sync_metadata(struct supertype
*st
)
655 if (st
->ss
->external
) {
656 if (st
->update_tail
) {
657 flush_metadata_updates(st
);
658 st
->update_tail
= &st
->updates
;
660 st
->ss
->sync_metadata(st
);
664 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
666 /* when dealing with external metadata subarrays we need to be
667 * prepared to handle EAGAIN. The kernel may need to wait for
668 * mdmon to mark the array active so the kernel can handle
669 * allocations/writeback when preparing the reshape action
670 * (md_allow_write()). We temporarily disable safe_mode_delay
671 * to close a race with the array_state going clean before the
672 * next write to raid_disks / stripe_cache_size
677 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
679 (strcmp(name
, "raid_disks") != 0 &&
680 strcmp(name
, "stripe_cache_size") != 0))
681 return sysfs_set_num(sra
, NULL
, name
, n
);
683 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
686 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
687 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
688 if (rc
< 0 && errno
== EAGAIN
) {
689 ping_monitor(container
);
690 /* if we get EAGAIN here then the monitor is not active
693 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
695 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
699 int start_reshape(struct mdinfo
*sra
, int already_running
, int data_disks
)
702 unsigned long long sync_max_to_set
;
704 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
705 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
706 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
707 sra
->reshape_progress
);
708 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
709 if (!already_running
)
710 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
711 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
712 if (!already_running
)
713 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
718 void abort_reshape(struct mdinfo
*sra
)
720 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
721 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
722 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
723 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
724 sysfs_set_num(sra
, NULL
, "sync_min", 0);
725 sysfs_set_str(sra
, NULL
, "sync_max", "max");
728 int remove_disks_for_takeover(struct supertype
*st
,
733 struct mdinfo
*remaining
;
736 if (sra
->array
.level
== 10)
737 nr_of_copies
= layout
& 0xff;
738 else if (sra
->array
.level
== 1)
739 nr_of_copies
= sra
->array
.raid_disks
;
743 remaining
= sra
->devs
;
745 /* for each 'copy', select one device and remove from the list. */
746 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
747 struct mdinfo
**diskp
;
750 /* Find a working device to keep */
751 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
752 struct mdinfo
*disk
= *diskp
;
754 if (disk
->disk
.raid_disk
< slot
)
756 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
758 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
760 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
762 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
765 /* We have found a good disk to use! */
767 disk
->next
= sra
->devs
;
776 if (slot
< sra
->array
.raid_disks
) {
777 /* didn't find all slots */
783 sra
->devs
= remaining
;
787 /* Remove all 'remaining' devices from the array */
789 struct mdinfo
*sd
= remaining
;
790 remaining
= sd
->next
;
792 sysfs_set_str(sra
, sd
, "state", "faulty");
793 sysfs_set_str(sra
, sd
, "slot", "none");
794 /* for external metadata disks should be removed in mdmon */
795 if (!st
->ss
->external
)
796 sysfs_set_str(sra
, sd
, "state", "remove");
797 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
798 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
799 sd
->next
= sra
->devs
;
805 void reshape_free_fdlist(int *fdlist
,
806 unsigned long long *offsets
,
811 for (i
= 0; i
< size
; i
++)
819 int reshape_prepare_fdlist(char *devname
,
823 unsigned long blocks
,
826 unsigned long long *offsets
)
831 for (d
= 0; d
<= nrdisks
; d
++)
834 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
835 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
837 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
838 char *dn
= map_dev(sd
->disk
.major
,
840 fdlist
[sd
->disk
.raid_disk
]
841 = dev_open(dn
, O_RDONLY
);
842 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
843 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
845 Name
": %s: cannot open component %s\n",
846 devname
, dn
? dn
: "-unknown-");
850 } else if (backup_file
== NULL
) {
852 char *dn
= map_dev(sd
->disk
.major
,
854 fdlist
[d
] = dev_open(dn
, O_RDWR
);
855 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
857 fprintf(stderr
, Name
": %s: cannot open component %s\n",
858 devname
, dn
? dn
: "-unknown-");
869 int reshape_open_backup_file(char *backup_file
,
874 unsigned long long *offsets
,
877 /* Return 1 on success, 0 on any form of failure */
878 /* need to check backup file is large enough */
884 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
888 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
889 devname
, backup_file
, strerror(errno
));
892 /* Guard against backup file being on array device.
893 * If array is partitioned or if LVM etc is in the
894 * way this will not notice, but it is better than
897 fstat(*fdlist
, &stb
);
900 if (stb
.st_rdev
== dev
) {
901 fprintf(stderr
, Name
": backup file must NOT be"
902 " on the array being reshaped.\n");
908 for (i
=0; i
< blocks
+ 8 ; i
++) {
909 if (write(*fdlist
, buf
, 512) != 512) {
910 fprintf(stderr
, Name
": %s: cannot create"
911 " backup file %s: %s\n",
912 devname
, backup_file
, strerror(errno
));
916 if (fsync(*fdlist
) != 0) {
917 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
918 devname
, backup_file
, strerror(errno
));
925 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
926 unsigned int ndata
, unsigned int odata
)
928 unsigned long a
, b
, blocks
;
929 /* So how much do we need to backup.
930 * We need an amount of data which is both a whole number of
931 * old stripes and a whole number of new stripes.
932 * So LCM for (chunksize*datadisks).
934 a
= (ochunk
/512) * odata
;
935 b
= (nchunk
/512) * ndata
;
943 /* LCM == product / GCD */
944 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
949 char *analyse_change(struct mdinfo
*info
, struct reshape
*re
)
951 /* Based on the current array state in info->array and
952 * the changes in info->new_* etc, determine:
953 * - whether the change is possible
954 * - Intermediate level/raid_disks/layout
955 * - whether a restriping reshape is needed
956 * - number of sectors in minimum change unit. This
957 * will cover a whole number of stripes in 'before' and
960 * Return message if the change should be rejected
961 * NULL if the change can be achieved
963 * This can be called as part of starting a reshape, or
964 * when assembling an array that is undergoing reshape.
967 /* delta_parity records change in number of devices
968 * caused by level change
970 int delta_parity
= 0;
972 /* If a new level not explicitly given, we assume no-change */
973 if (info
->new_level
== UnSet
)
974 info
->new_level
= info
->array
.level
;
977 switch (info
->new_level
) {
983 /* chunk size is meaningful, must divide component_size
986 if (info
->component_size
% (info
->new_chunk
/512))
987 return "New chunk size does not"
988 " divide component size";
991 return "chunk size not meaningful for this level";
994 info
->new_chunk
= info
->array
.chunk_size
;
996 switch (info
->array
.level
) {
998 return "Cannot understand this RAID level";
1000 /* RAID1 can convert to RAID1 with different disks, or
1001 * raid5 with 2 disks, or
1004 if (info
->new_level
== 0) {
1005 if (info
->delta_disks
!= UnSet
&&
1006 info
->delta_disks
!= 0)
1007 return "Cannot change number of disks "
1008 "with RAID1->RAID0 conversion";
1010 re
->before
.data_disks
= 1;
1011 re
->after
.data_disks
= 1;
1012 re
->before
.layout
= 0;
1013 re
->backup_blocks
= 0;
1017 if (info
->new_level
== 1) {
1018 if (info
->delta_disks
== UnSet
)
1019 /* Don't know what to do */
1020 return "no change requested for Growing RAID1";
1022 re
->backup_blocks
= 0;
1026 if (info
->array
.raid_disks
== 2 &&
1027 info
->new_level
== 5) {
1030 re
->before
.data_disks
= 1;
1031 if (info
->delta_disks
!= UnSet
&&
1032 info
->delta_disks
!= 0)
1033 re
->after
.data_disks
= 1 + info
->delta_disks
;
1035 re
->after
.data_disks
= 1;
1036 if (re
->after
.data_disks
< 1)
1037 return "Number of disks too small for RAID5";
1039 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1040 info
->array
.chunk_size
= 65536;
1043 /* Could do some multi-stage conversions, but leave that to
1046 return "Impossibly level change request for RAID1";
1049 /* RAID10 can only be converted from near mode to
1050 * RAID0 by removing some devices
1052 if ((info
->array
.layout
& ~0xff) != 0x100)
1053 return "Cannot Grow RAID10 with far/offset layout";
1054 /* number of devices must be multiple of number of copies */
1055 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1056 return "RAID10 layout too complex for Grow operation";
1058 if (info
->new_level
!= 0)
1059 return "RAID10 can only be changed to RAID0";
1060 new_disks
= (info
->array
.raid_disks
1061 / (info
->array
.layout
& 0xff));
1062 if (info
->delta_disks
== UnSet
)
1063 info
->delta_disks
= (new_disks
1064 - info
->array
.raid_disks
);
1066 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1067 return "New number of raid-devices impossible for RAID10";
1068 if (info
->new_chunk
&&
1069 info
->new_chunk
!= info
->array
.chunk_size
)
1070 return "Cannot change chunk-size with RAID10 Grow";
1075 re
->before
.data_disks
= new_disks
;
1076 re
->after
.data_disks
= re
->before
.data_disks
;
1077 re
->before
.layout
= 0;
1078 re
->backup_blocks
= 0;
1082 /* RAID0 can be converted to RAID10, or to RAID456 */
1083 if (info
->new_level
== 10) {
1084 if (info
->new_layout
== UnSet
&& info
->delta_disks
== UnSet
) {
1085 /* Assume near=2 layout */
1086 info
->new_layout
= 0x102;
1087 info
->delta_disks
= info
->array
.raid_disks
;
1089 if (info
->new_layout
== UnSet
) {
1090 int copies
= 1 + (info
->delta_disks
1091 / info
->array
.raid_disks
);
1092 if (info
->array
.raid_disks
* (copies
-1)
1093 != info
->delta_disks
)
1094 return "Impossible number of devices"
1095 " for RAID0->RAID10";
1096 info
->new_layout
= 0x100 + copies
;
1098 if (info
->delta_disks
== UnSet
) {
1099 int copies
= info
->new_layout
& 0xff;
1100 if (info
->new_layout
!= 0x100 + copies
)
1101 return "New layout impossible"
1102 " for RAID0->RAID10";;
1103 info
->delta_disks
= (copies
- 1) *
1104 info
->array
.raid_disks
;
1106 if (info
->new_chunk
&&
1107 info
->new_chunk
!= info
->array
.chunk_size
)
1108 return "Cannot change chunk-size with RAID0->RAID10";
1112 re
->before
.data_disks
= (info
->array
.raid_disks
+
1114 re
->after
.data_disks
= re
->before
.data_disks
;
1115 re
->before
.layout
= info
->new_layout
;
1116 re
->backup_blocks
= 0;
1120 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1121 * a raid4 style layout of the final level.
1123 switch (info
->new_level
) {
1128 re
->before
.layout
= 0;
1133 re
->before
.layout
= ALGORITHM_PARITY_N
;
1138 re
->before
.layout
= ALGORITHM_PARITY_N
;
1141 return "Impossible level change requested";
1143 re
->before
.data_disks
= info
->array
.raid_disks
;
1144 /* determining 'after' layout happens outside this 'switch' */
1148 info
->array
.layout
= ALGORITHM_PARITY_N
;
1150 switch (info
->new_level
) {
1154 re
->level
= info
->array
.level
;
1155 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1156 re
->before
.layout
= info
->array
.layout
;
1160 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1161 re
->before
.layout
= info
->array
.layout
;
1166 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1167 switch (info
->array
.layout
) {
1168 case ALGORITHM_LEFT_ASYMMETRIC
:
1169 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1171 case ALGORITHM_RIGHT_ASYMMETRIC
:
1172 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1174 case ALGORITHM_LEFT_SYMMETRIC
:
1175 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1177 case ALGORITHM_RIGHT_SYMMETRIC
:
1178 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1180 case ALGORITHM_PARITY_0
:
1181 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1183 case ALGORITHM_PARITY_N
:
1184 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1187 return "Cannot convert an array with this layout";
1191 if (info
->array
.raid_disks
!= 2)
1192 return "Can only convert a 2-device array to RAID1";
1193 if (info
->delta_disks
!= UnSet
&&
1194 info
->delta_disks
!= 0)
1195 return "Cannot set raid_disk when "
1196 "converting RAID5->RAID1";
1200 return "Impossible level change requested";
1204 switch (info
->new_level
) {
1210 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1211 re
->before
.layout
= info
->array
.layout
;
1214 return "Impossible level change requested";
1219 /* If we reached here then it looks like a re-stripe is
1220 * happening. We have determined the intermediate level
1221 * and initial raid_disks/layout and stored these in 're'.
1223 * We need to deduce the final layout that can be atomically
1224 * converted to the end state.
1226 switch (info
->new_level
) {
1228 /* We can only get to RAID0 from RAID4 or RAID5
1229 * with appropriate layout and one extra device
1231 if (re
->level
!= 4 && re
->level
!= 5)
1232 return "Cannot covert to RAID0 from this level";
1234 switch (re
->level
) {
1236 re
->after
.layout
= 0 ; break;
1238 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1243 /* We can only get to RAID4 from RAID5 */
1244 if (re
->level
!= 4 && re
->level
!= 5)
1245 return "Cannot convert to RAID4 from this level";
1247 switch (re
->level
) {
1249 re
->after
.layout
= 0 ; break;
1251 re
->after
.layout
= ALGORITHM_PARITY_N
; break;
1256 /* We get to RAID5 for RAID5 or RAID6 */
1257 if (re
->level
!= 5 && re
->level
!= 6)
1258 return "Cannot convert to RAID5 from this level";
1260 switch (re
->level
) {
1262 if (info
->new_layout
== UnSet
)
1263 re
->after
.layout
= re
->before
.layout
;
1265 re
->after
.layout
= info
->new_layout
;
1268 if (info
->new_layout
== UnSet
)
1269 info
->new_layout
= re
->before
.layout
;
1271 /* after.layout needs to be raid6 version of new_layout */
1272 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1273 re
->after
.layout
= ALGORITHM_PARITY_N
;
1276 char *ls
= map_num(r5layout
, info
->new_layout
);
1278 strcat(strcpy(layout
, ls
), "-6");
1279 l
= map_name(r6layout
, layout
);
1281 return "Cannot find RAID6 layout"
1283 re
->after
.layout
= l
;
1289 /* We must already be at level 6 */
1291 return "Impossible level change";
1292 if (info
->new_layout
== UnSet
)
1293 re
->after
.layout
= info
->array
.layout
;
1295 re
->after
.layout
= info
->new_layout
;
1298 return "Impossible level change requested";
1300 if (info
->delta_disks
== UnSet
)
1301 info
->delta_disks
= delta_parity
;
1303 re
->after
.data_disks
= (re
->before
.data_disks
1306 switch (re
->level
) {
1307 case 6: re
->parity
= 2; break;
1309 case 5: re
->parity
= 1; break;
1310 default: re
->parity
= 0; break;
1312 /* So we have a restripe operation, we need to calculate the number
1313 * of blocks per reshape operation.
1315 if (info
->new_chunk
== 0)
1316 info
->new_chunk
= info
->array
.chunk_size
;
1317 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1318 re
->after
.layout
== re
->before
.layout
&&
1319 info
->new_chunk
== info
->array
.chunk_size
) {
1320 /* Nothing to change */
1321 re
->backup_blocks
= 0;
1324 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1325 /* chunk and layout changes make no difference */
1326 re
->backup_blocks
= 0;
1330 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1331 get_linux_version() < 2006032)
1332 return "in-place reshape is not safe before 2.6.32 - sorry.";
1334 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1335 get_linux_version() < 2006030)
1336 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1338 re
->backup_blocks
= compute_backup_blocks(
1339 info
->new_chunk
, info
->array
.chunk_size
,
1340 re
->after
.data_disks
,
1341 re
->before
.data_disks
);
1343 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1347 static int reshape_array(char *container
, int fd
, char *devname
,
1348 struct supertype
*st
, struct mdinfo
*info
,
1349 int force
, struct mddev_dev
*devlist
,
1350 char *backup_file
, int quiet
, int forked
,
1351 int restart
, int freeze_reshape
);
1352 static int reshape_container(char *container
, char *devname
,
1354 struct supertype
*st
,
1355 struct mdinfo
*info
,
1358 int quiet
, int restart
, int freeze_reshape
);
1360 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
1362 int level
, char *layout_str
, int chunksize
, int raid_disks
,
1363 struct mddev_dev
*devlist
,
1364 int assume_clean
, int force
)
1366 /* Make some changes in the shape of an array.
1367 * The kernel must support the change.
1369 * There are three different changes. Each can trigger
1370 * a resync or recovery so we freeze that until we have
1371 * requested everything (if kernel supports freezing - 2.6.30).
1373 * - change size (i.e. component_size)
1375 * - change layout/chunksize/ndisks
1377 * The last can require a reshape. It is different on different
1378 * levels so we need to check the level before actioning it.
1379 * Some times the level change needs to be requested after the
1380 * reshape (e.g. raid6->raid5, raid5->raid0)
1383 struct mdu_array_info_s array
;
1385 struct supertype
*st
;
1386 char *subarray
= NULL
;
1390 char *container
= NULL
;
1391 char container_buf
[20];
1394 struct mddev_dev
*dv
;
1400 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
1401 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
1407 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
1408 fprintf(stderr
, Name
": cannot change component size at the same time "
1409 "as other changes.\n"
1410 " Change size first, then check data is intact before "
1411 "making other changes.\n");
1415 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
1416 get_linux_version() < 2006032 &&
1417 !check_env("MDADM_FORCE_FEWER")) {
1418 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
1419 " Please use a newer kernel\n");
1423 st
= super_by_fd(fd
, &subarray
);
1425 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
1428 if (raid_disks
> st
->max_devs
) {
1429 fprintf(stderr
, Name
": Cannot increase raid-disks on this array"
1430 " beyond %d\n", st
->max_devs
);
1434 /* in the external case we need to check that the requested reshape is
1435 * supported, and perform an initial check that the container holds the
1436 * pre-requisite spare devices (mdmon owns final validation)
1438 if (st
->ss
->external
) {
1443 container_dev
= st
->container_dev
;
1444 cfd
= open_dev_excl(st
->container_dev
);
1446 container_dev
= st
->devnum
;
1448 cfd
= open_dev_excl(st
->devnum
);
1452 fprintf(stderr
, Name
": Unable to open container for %s\n",
1458 fmt_devname(container_buf
, container_dev
);
1459 container
= container_buf
;
1461 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
1464 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
1470 if (mdmon_running(container_dev
))
1471 st
->update_tail
= &st
->updates
;
1475 for (dv
= devlist
; dv
; dv
= dv
->next
)
1477 if (raid_disks
> array
.raid_disks
&&
1478 array
.spare_disks
+added_disks
< (raid_disks
- array
.raid_disks
) &&
1481 Name
": Need %d spare%s to avoid degraded array,"
1482 " and only have %d.\n"
1483 " Use --force to over-ride this check.\n",
1484 raid_disks
- array
.raid_disks
,
1485 raid_disks
- array
.raid_disks
== 1 ? "" : "s",
1486 array
.spare_disks
+ added_disks
);
1490 sra
= sysfs_read(fd
, 0, GET_LEVEL
| GET_DISKS
| GET_DEVS
1491 | GET_STATE
| GET_VERSION
);
1493 if (st
->ss
->external
&& subarray
== NULL
) {
1494 array
.level
= LEVEL_CONTAINER
;
1495 sra
->array
.level
= LEVEL_CONTAINER
;
1498 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
1502 frozen
= freeze(st
);
1504 /* freeze() already spewed the reason */
1506 } else if (frozen
< 0) {
1507 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
1508 " be reshaped\n", devname
);
1512 /* ========= set size =============== */
1513 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
1514 long long orig_size
= get_component_size(fd
)/2;
1515 long long min_csize
;
1519 orig_size
= array
.size
;
1521 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1528 /* Update the size of each member device in case
1529 * they have been resized. This will never reduce
1530 * below the current used-size. The "size" attribute
1531 * understands '0' to mean 'max'.
1534 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
1535 if (sysfs_set_num(sra
, mdi
, "size", size
) < 0)
1537 if (array
.not_persistent
== 0 &&
1538 array
.major_version
== 0 &&
1539 get_linux_version() < 3001000) {
1540 /* Dangerous to allow size to exceed 2TB */
1541 unsigned long long csize
;
1542 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
1543 if (csize
>= 2ULL*1024*1024*1024)
1544 csize
= 2ULL*1024*1024*1024;
1545 if ((min_csize
== 0 || (min_csize
1546 > (long long)csize
)))
1551 if (min_csize
&& size
> min_csize
) {
1552 fprintf(stderr
, Name
": Cannot safely make this array "
1553 "use more than 2TB per device on this kernel.\n");
1557 if (min_csize
&& size
== 0) {
1558 /* Don't let the kernel choose a size - it will get
1561 fprintf(stderr
, Name
": Limited v0.90 array to "
1562 "2TB per device\n");
1567 if (array
.size
!= size
) {
1568 /* got truncated to 32bit, write to
1569 * component_size instead
1572 rv
= sysfs_set_num(sra
, NULL
,
1573 "component_size", size
);
1577 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
1581 /* restore metadata */
1582 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
1583 UnSet
, NULL
, devname
, !quiet
) == 0)
1585 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
1586 devname
, strerror(err
));
1588 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1589 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
1594 /* This will fail on kernels newer than 3.0 unless
1595 * a backport has been arranged.
1598 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
1599 fprintf(stderr
, Name
": --assume-clean not support with --grow on this kernel\n");
1601 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1602 size
= get_component_size(fd
)/2;
1606 if (size
== orig_size
)
1607 fprintf(stderr
, Name
": component size of %s "
1608 "unchanged at %lluK\n",
1611 fprintf(stderr
, Name
": component size of %s "
1612 "has been set to %lluK\n",
1616 } else if (array
.level
!= LEVEL_CONTAINER
) {
1617 size
= get_component_size(fd
)/2;
1622 /* See if there is anything else to do */
1623 if ((level
== UnSet
|| level
== array
.level
) &&
1624 (layout_str
== NULL
) &&
1625 (chunksize
== 0 || chunksize
== array
.chunk_size
) &&
1626 (raid_disks
== 0 || raid_disks
== array
.raid_disks
)) {
1627 /* Nothing more to do */
1628 if (!changed
&& !quiet
)
1629 fprintf(stderr
, Name
": %s: no change requested\n",
1634 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1635 * current implementation assumes that following conditions must be met:
1638 * - near_copies == 2
1640 if ((level
== 0 && array
.level
== 10 && sra
&&
1641 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
1642 (level
== 0 && array
.level
== 1 && sra
)) {
1644 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
1646 dprintf(Name
": Array cannot be reshaped\n");
1652 /* Make sure mdmon has seen the device removal
1653 * and updated metadata before we continue with
1657 ping_monitor(container
);
1660 memset(&info
, 0, sizeof(info
));
1662 sysfs_init(&info
, fd
, NoMdDev
);
1663 strcpy(info
.text_version
, sra
->text_version
);
1664 info
.component_size
= size
*2;
1665 info
.new_level
= level
;
1666 info
.new_chunk
= chunksize
* 1024;
1667 if (info
.array
.level
== LEVEL_CONTAINER
) {
1668 info
.delta_disks
= UnSet
;
1669 info
.array
.raid_disks
= raid_disks
;
1670 } else if (raid_disks
)
1671 info
.delta_disks
= raid_disks
- info
.array
.raid_disks
;
1673 info
.delta_disks
= UnSet
;
1674 if (layout_str
== NULL
) {
1675 info
.new_layout
= UnSet
;
1676 if (info
.array
.level
== 6 &&
1677 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
1678 info
.array
.layout
>= 16) {
1679 fprintf(stderr
, Name
1680 ": %s has a non-standard layout. If you"
1681 " wish to preserve this\n"
1682 " during the reshape, please specify"
1683 " --layout=preserve\n"
1684 " If you want to change it, specify a"
1685 " layout or use --layout=normalise\n",
1690 } else if (strcmp(layout_str
, "normalise") == 0 ||
1691 strcmp(layout_str
, "normalize") == 0) {
1692 /* If we have a -6 RAID6 layout, remove the '-6'. */
1693 info
.new_layout
= UnSet
;
1694 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
1696 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
1697 h
= strrchr(l
, '-');
1698 if (h
&& strcmp(h
, "-6") == 0) {
1700 info
.new_layout
= map_name(r6layout
, l
);
1703 } else if (strcmp(layout_str
, "preserve") == 0) {
1704 info
.new_layout
= UnSet
;
1706 int l
= info
.new_level
;
1708 l
= info
.array
.level
;
1711 info
.new_layout
= map_name(r5layout
, layout_str
);
1714 info
.new_layout
= map_name(r6layout
, layout_str
);
1717 info
.new_layout
= parse_layout_10(layout_str
);
1720 info
.new_layout
= parse_layout_faulty(layout_str
);
1723 fprintf(stderr
, Name
": layout not meaningful"
1724 " with this level\n");
1728 if (info
.new_layout
== UnSet
) {
1729 fprintf(stderr
, Name
": layout %s not understood"
1730 " for this level\n",
1737 if (array
.level
== LEVEL_FAULTY
) {
1738 if (level
!= UnSet
&& level
!= array
.level
) {
1739 fprintf(stderr
, Name
": cannot change level of Faulty device\n");
1743 fprintf(stderr
, Name
": cannot set chunksize of Faulty device\n");
1746 if (raid_disks
&& raid_disks
!= 1) {
1747 fprintf(stderr
, Name
": cannot set raid_disks of Faulty device\n");
1751 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1752 dprintf("Cannot get array information.\n");
1755 array
.layout
= info
.new_layout
;
1756 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1757 fprintf(stderr
, Name
": failed to set new layout\n");
1760 printf("layout for %s set to %d\n",
1761 devname
, array
.layout
);
1763 } else if (array
.level
== LEVEL_CONTAINER
) {
1764 /* This change is to be applied to every array in the
1765 * container. This is only needed when the metadata imposes
1766 * restraints of the various arrays in the container.
1767 * Currently we only know that IMSM requires all arrays
1768 * to have the same number of devices so changing the
1769 * number of devices (On-Line Capacity Expansion) must be
1770 * performed at the level of the container
1772 rv
= reshape_container(container
, devname
, -1, st
, &info
,
1773 force
, backup_file
, quiet
, 0, 0);
1776 /* get spare devices from external metadata
1778 if (st
->ss
->external
) {
1779 struct mdinfo
*info2
;
1781 info2
= st
->ss
->container_content(st
, subarray
);
1783 info
.array
.spare_disks
=
1784 info2
->array
.spare_disks
;
1789 /* Impose these changes on a single array. First
1790 * check that the metadata is OK with the change. */
1792 if (reshape_super(st
, info
.component_size
, info
.new_level
,
1793 info
.new_layout
, info
.new_chunk
,
1794 info
.array
.raid_disks
, info
.delta_disks
,
1795 backup_file
, devname
, quiet
)) {
1800 rv
= reshape_array(container
, fd
, devname
, st
, &info
, force
,
1801 devlist
, backup_file
, quiet
, 0, 0, 0);
1810 static int reshape_array(char *container
, int fd
, char *devname
,
1811 struct supertype
*st
, struct mdinfo
*info
,
1812 int force
, struct mddev_dev
*devlist
,
1813 char *backup_file
, int quiet
, int forked
,
1814 int restart
, int freeze_reshape
)
1816 struct reshape reshape
;
1819 int orig_level
= UnSet
;
1822 struct mdu_array_info_s array
;
1825 struct mddev_dev
*dv
;
1829 unsigned long long *offsets
;
1833 unsigned long blocks
;
1834 unsigned long cache
;
1835 unsigned long long array_size
;
1837 struct mdinfo
*sra
= NULL
;
1839 /* when reshaping a RAID0, the component_size might be zero.
1840 * So try to fix that up.
1842 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
1843 dprintf("Cannot get array information.\n");
1846 if (array
.level
== 0 && info
->component_size
== 0) {
1847 get_dev_size(fd
, NULL
, &array_size
);
1848 info
->component_size
= array_size
/ array
.raid_disks
;
1851 if (info
->reshape_active
) {
1852 int new_level
= info
->new_level
;
1853 info
->new_level
= UnSet
;
1854 info
->array
.raid_disks
-= info
->delta_disks
;
1855 msg
= analyse_change(info
, &reshape
);
1856 info
->new_level
= new_level
;
1857 info
->array
.raid_disks
+= info
->delta_disks
;
1859 /* Make sure the array isn't read-only */
1860 ioctl(fd
, RESTART_ARRAY_RW
, 0);
1862 msg
= analyse_change(info
, &reshape
);
1864 fprintf(stderr
, Name
": %s\n", msg
);
1868 (reshape
.level
!= info
->array
.level
||
1869 reshape
.before
.layout
!= info
->array
.layout
||
1870 reshape
.before
.data_disks
+ reshape
.parity
1871 != info
->array
.raid_disks
- info
->delta_disks
)) {
1872 fprintf(stderr
, Name
": reshape info is not in native format -"
1873 " cannot continue.\n");
1878 /* reshape already started. just skip to monitoring the reshape */
1879 if (reshape
.backup_blocks
== 0)
1883 /* The container is frozen but the array may not be.
1884 * So freeze the array so spares don't get put to the wrong use
1885 * FIXME there should probably be a cleaner separation between
1886 * freeze_array and freeze_container.
1888 sysfs_freeze_array(info
);
1889 /* Check we have enough spares to not be degraded */
1891 for (dv
= devlist
; dv
; dv
=dv
->next
)
1893 spares_needed
= max(reshape
.before
.data_disks
,
1894 reshape
.after
.data_disks
)
1895 + reshape
.parity
- array
.raid_disks
;
1898 info
->new_level
> 1 && info
->array
.level
> 1 &&
1899 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
1901 Name
": Need %d spare%s to avoid degraded array,"
1902 " and only have %d.\n"
1903 " Use --force to over-ride this check.\n",
1905 spares_needed
== 1 ? "" : "s",
1906 info
->array
.spare_disks
+ added_disks
);
1909 /* Check we have enough spares to not fail */
1910 spares_needed
= max(reshape
.before
.data_disks
,
1911 reshape
.after
.data_disks
)
1913 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
1914 spares_needed
> info
->array
.spare_disks
+added_disks
) {
1916 Name
": Need %d spare%s to create working array,"
1917 " and only have %d.\n",
1919 spares_needed
== 1 ? "" : "s",
1920 info
->array
.spare_disks
+ added_disks
);
1924 if (reshape
.level
!= array
.level
) {
1925 char *c
= map_num(pers
, reshape
.level
);
1930 err
= sysfs_set_str(info
, NULL
, "level", c
);
1933 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1936 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1937 fprintf(stderr
, " Bitmap must be removed"
1938 " before level can be changed\n");
1942 fprintf(stderr
, Name
": level of %s changed to %s\n",
1944 orig_level
= array
.level
;
1945 sysfs_freeze_array(info
);
1947 if (reshape
.level
> 0 && st
->ss
->external
) {
1948 /* make sure mdmon is aware of the new level */
1949 if (!mdmon_running(st
->container_dev
))
1950 start_mdmon(st
->container_dev
);
1951 ping_monitor(container
);
1952 if (mdmon_running(st
->container_dev
) &&
1953 st
->update_tail
== NULL
)
1954 st
->update_tail
= &st
->updates
;
1957 /* ->reshape_super might have chosen some spares from the
1958 * container that it wants to be part of the new array.
1959 * We can collect them with ->container_content and give
1960 * them to the kernel.
1962 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
1963 char *subarray
= strchr(info
->text_version
+1, '/')+1;
1964 struct mdinfo
*info2
=
1965 st
->ss
->container_content(st
, subarray
);
1969 sysfs_init(info2
, fd
, st
->devnum
);
1970 /* When increasing number of devices, we need to set
1971 * new raid_disks before adding these, or they might
1974 if (reshape
.backup_blocks
&&
1975 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
1976 subarray_set_num(container
, info2
, "raid_disks",
1977 reshape
.after
.data_disks
+
1979 for (d
= info2
->devs
; d
; d
= d
->next
) {
1980 if (d
->disk
.state
== 0 &&
1981 d
->disk
.raid_disk
>= 0) {
1982 /* This is a spare that wants to
1983 * be part of the array.
1985 add_disk(fd
, st
, info2
, d
);
1991 /* We might have been given some devices to add to the
1992 * array. Now that the array has been changed to the right
1993 * level and frozen, we can safely add them.
1996 Manage_subdevs(devname
, fd
, devlist
, !quiet
,
1999 if (reshape
.backup_blocks
== 0) {
2000 /* No restriping needed, but we might need to impose
2001 * some more changes: layout, raid_disks, chunk_size
2003 /* read current array info */
2004 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
2005 dprintf("Cannot get array information.\n");
2008 /* compare current array info with new values and if
2009 * it is different update them to new */
2010 if (info
->new_layout
!= UnSet
&&
2011 info
->new_layout
!= array
.layout
) {
2012 array
.layout
= info
->new_layout
;
2013 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2014 fprintf(stderr
, Name
": failed to set new layout\n");
2017 printf("layout for %s set to %d\n",
2018 devname
, array
.layout
);
2020 if (info
->delta_disks
!= UnSet
&&
2021 info
->delta_disks
!= 0 &&
2022 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
2023 array
.raid_disks
+= info
->delta_disks
;
2024 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2025 fprintf(stderr
, Name
": failed to set raid disks\n");
2027 } else if (!quiet
) {
2028 printf("raid_disks for %s set to %d\n",
2029 devname
, array
.raid_disks
);
2032 if (info
->new_chunk
!= 0 &&
2033 info
->new_chunk
!= array
.chunk_size
) {
2034 if (sysfs_set_num(info
, NULL
,
2035 "chunk_size", info
->new_chunk
) != 0) {
2036 fprintf(stderr
, Name
": failed to set chunk size\n");
2039 printf("chunk size for %s set to %d\n",
2040 devname
, array
.chunk_size
);
2047 * There are three possibilities.
2048 * 1/ The array will shrink.
2049 * We need to ensure the reshape will pause before reaching
2050 * the 'critical section'. We also need to fork and wait for
2051 * that to happen. When it does we
2052 * suspend/backup/complete/unfreeze
2054 * 2/ The array will not change size.
2055 * This requires that we keep a backup of a sliding window
2056 * so that we can restore data after a crash. So we need
2057 * to fork and monitor progress.
2058 * In future we will allow the data_offset to change, so
2059 * a sliding backup becomes unnecessary.
2061 * 3/ The array will grow. This is relatively easy.
2062 * However the kernel's restripe routines will cheerfully
2063 * overwrite some early data before it is safe. So we
2064 * need to make a backup of the early parts of the array
2065 * and be ready to restore it if rebuild aborts very early.
2066 * For externally managed metadata, we still need a forked
2067 * child to monitor the reshape and suspend IO over the region
2068 * that is being reshaped.
2070 * We backup data by writing it to one spare, or to a
2071 * file which was given on command line.
2073 * In each case, we first make sure that storage is available
2074 * for the required backup.
2076 * - request the shape change.
2077 * - fork to handle backup etc.
2080 /* Check that we can hold all the data */
2081 get_dev_size(fd
, NULL
, &array_size
);
2082 if (reshape
.new_size
< (array_size
/512)) {
2084 Name
": this change will reduce the size of the array.\n"
2085 " use --grow --array-size first to truncate array.\n"
2086 " e.g. mdadm --grow %s --array-size %llu\n",
2087 devname
, reshape
.new_size
/2);
2091 sra
= sysfs_read(fd
, 0,
2092 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
2095 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
2100 /* Decide how many blocks (sectors) for a reshape
2101 * unit. The number we have so far is just a minimum
2103 blocks
= reshape
.backup_blocks
;
2104 if (reshape
.before
.data_disks
==
2105 reshape
.after
.data_disks
) {
2106 /* Make 'blocks' bigger for better throughput, but
2107 * not so big that we reject it below.
2108 * Try for 16 megabytes
2110 while (blocks
* 32 < sra
->component_size
&&
2114 fprintf(stderr
, Name
": Need to backup %luK of critical "
2115 "section..\n", blocks
/2);
2117 if (blocks
>= sra
->component_size
/2) {
2118 fprintf(stderr
, Name
": %s: Something wrong"
2119 " - reshape aborted\n",
2124 /* Now we need to open all these devices so we can read/write.
2126 nrdisks
= max(reshape
.before
.data_disks
,
2127 reshape
.after
.data_disks
) + reshape
.parity
2128 + sra
->array
.spare_disks
;
2129 fdlist
= malloc((1+nrdisks
) * sizeof(int));
2130 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
2131 if (!fdlist
|| !offsets
) {
2132 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
2136 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
2137 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
2138 nrdisks
, blocks
, backup_file
,
2143 if ((st
->ss
->manage_reshape
== NULL
) ||
2144 (st
->ss
->recover_backup
== NULL
)) {
2145 if (backup_file
== NULL
) {
2146 if (reshape
.after
.data_disks
<=
2147 reshape
.before
.data_disks
) {
2148 fprintf(stderr
, Name
": %s: Cannot grow - "
2149 "need backup-file\n", devname
);
2151 } else if (sra
->array
.spare_disks
== 0) {
2152 fprintf(stderr
, Name
": %s: Cannot grow - "
2153 "need a spare or backup-file to backup "
2154 "critical section\n", devname
);
2158 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
2160 fdlist
+d
, offsets
+d
,
2168 /* lastly, check that the internal stripe cache is
2169 * large enough, or it won't work.
2170 * It must hold at least 4 stripes of the larger
2173 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2174 cache
*= 4; /* 4 stripes minimum */
2175 cache
/= 512; /* convert to sectors */
2176 disks
= min(reshape
.before
.data_disks
, reshape
.after
.data_disks
);
2177 /* make sure there is room for 'blocks' with a bit to spare */
2178 if (cache
< 16 + blocks
/ disks
)
2179 cache
= 16 + blocks
/ disks
;
2180 cache
/= (4096/512); /* Covert from sectors to pages */
2182 if (sra
->cache_size
< cache
)
2183 subarray_set_num(container
, sra
, "stripe_cache_size",
2186 /* Right, everything seems fine. Let's kick things off.
2187 * If only changing raid_disks, use ioctl, else use
2192 sra
->new_chunk
= info
->new_chunk
;
2195 sra
->reshape_progress
= info
->reshape_progress
;
2197 sra
->reshape_progress
= 0;
2198 if (reshape
.after
.data_disks
< reshape
.before
.data_disks
)
2199 /* start from the end of the new array */
2200 sra
->reshape_progress
= (sra
->component_size
2201 * reshape
.after
.data_disks
);
2204 if (info
->array
.chunk_size
== info
->new_chunk
&&
2205 reshape
.before
.layout
== reshape
.after
.layout
&&
2206 st
->ss
->external
== 0) {
2207 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2208 ioctl(fd
, GET_ARRAY_INFO
, &array
);
2209 array
.raid_disks
= reshape
.after
.data_disks
+ reshape
.parity
;
2211 ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
2215 Name
": Cannot set device shape for %s: %s\n",
2216 devname
, strerror(errno
));
2219 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2221 " Bitmap must be removed before"
2222 " shape can be changed\n");
2226 } else if (!restart
) {
2227 /* set them all just in case some old 'new_*' value
2228 * persists from some earlier problem.
2231 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2233 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2234 reshape
.after
.layout
) < 0)
2236 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2237 reshape
.after
.data_disks
+
2238 reshape
.parity
) < 0)
2241 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
2245 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2247 " Bitmap must be removed before"
2248 " shape can be changed\n");
2253 err
= start_reshape(sra
, restart
,
2254 info
->array
.raid_disks
- reshape
.parity
);
2257 Name
": Cannot %s reshape for %s\n",
2258 restart
? "continue" : "start",
2263 sysfs_set_str(sra
, NULL
, "array_state", "active");
2264 if (freeze_reshape
) {
2268 fprintf(stderr
, Name
": Reshape has to be continued from"
2269 " location %llu when root fileststem has been mounted\n",
2270 sra
->reshape_progress
);
2274 /* Now we just need to kick off the reshape and watch, while
2275 * handling backups of the data...
2276 * This is all done by a forked background process.
2278 switch(forked
? 0 : fork()) {
2280 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
2292 if (check_env("MDADM_GROW_VERIFY"))
2293 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
2296 mlockall(MCL_FUTURE
);
2298 if (st
->ss
->external
) {
2299 /* metadata handler takes it from here */
2300 done
= st
->ss
->manage_reshape(
2301 fd
, sra
, &reshape
, st
, blocks
,
2303 d
- odisks
, fdlist
+odisks
,
2306 done
= child_monitor(
2307 fd
, sra
, &reshape
, st
, blocks
,
2309 d
- odisks
, fdlist
+odisks
,
2312 if (backup_file
&& done
)
2313 unlink(backup_file
);
2319 if (!st
->ss
->external
&&
2320 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
2321 && info
->custom_array_size
) &&
2322 info
->new_level
== reshape
.level
&&
2324 /* no need to wait for the reshape to finish as
2325 * there is nothing more to do.
2331 if (st
->ss
->external
) {
2332 /* Re-load the metadata as much could have changed */
2333 int cfd
= open_dev(st
->container_dev
);
2335 ping_monitor(container
);
2336 st
->ss
->free_super(st
);
2337 st
->ss
->load_container(st
, cfd
, container
);
2342 /* set new array size if required customer_array_size is used
2345 if (reshape
.before
.data_disks
!=
2346 reshape
.after
.data_disks
&&
2347 info
->custom_array_size
) {
2348 struct mdinfo
*info2
;
2349 char *subarray
= strchr(info
->text_version
+1, '/')+1;
2351 info2
= st
->ss
->container_content(st
, subarray
);
2353 unsigned long long current_size
= 0;
2354 unsigned long long new_size
=
2355 info2
->custom_array_size
/2;
2357 if (sysfs_get_ll(sra
,
2360 ¤t_size
) == 0 &&
2361 new_size
> current_size
) {
2362 if (sysfs_set_num(sra
, NULL
,
2363 "array_size", new_size
)
2365 dprintf("Error: Cannot"
2368 dprintf("Array size "
2370 dprintf(" from %llu to %llu.\n",
2371 current_size
, new_size
);
2377 if (info
->new_level
!= reshape
.level
) {
2379 c
= map_num(pers
, info
->new_level
);
2381 err
= sysfs_set_str(sra
, NULL
, "level", c
);
2383 fprintf(stderr
, Name\
2384 ": %s: could not set level "
2385 "to %s\n", devname
, c
);
2387 if (info
->new_level
== 0)
2388 st
->update_tail
= NULL
;
2397 if (orig_level
!= UnSet
&& sra
) {
2398 c
= map_num(pers
, orig_level
);
2399 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
2400 fprintf(stderr
, Name
": aborting level change\n");
2407 /* mdfd handle is passed to be closed in child process (after fork).
2409 int reshape_container(char *container
, char *devname
,
2411 struct supertype
*st
,
2412 struct mdinfo
*info
,
2415 int quiet
, int restart
, int freeze_reshape
)
2417 struct mdinfo
*cc
= NULL
;
2420 /* component_size is not meaningful for a container,
2421 * so pass '-1' meaning 'no change'
2424 reshape_super(st
, -1, info
->new_level
,
2425 info
->new_layout
, info
->new_chunk
,
2426 info
->array
.raid_disks
, info
->delta_disks
,
2427 backup_file
, devname
, quiet
)) {
2434 /* ping monitor to be sure that update is on disk
2436 ping_monitor(container
);
2439 case -1: /* error */
2440 perror("Cannot fork to complete reshape\n");
2443 default: /* parent */
2444 if (!freeze_reshape
)
2445 printf(Name
": multi-array reshape continues"
2446 " in background\n");
2453 /* close unused handle in child process
2459 /* For each member array with reshape_active,
2460 * we need to perform the reshape.
2461 * We pick the first array that needs reshaping and
2462 * reshape it. reshape_array() will re-read the metadata
2463 * so the next time through a different array should be
2464 * ready for reshape.
2465 * It is possible that the 'different' array will not
2466 * be assembled yet. In that case we simple exit.
2467 * When it is assembled, the mdadm which assembles it
2468 * will take over the reshape.
2470 struct mdinfo
*content
;
2472 struct mdstat_ent
*mdstat
;
2477 cc
= st
->ss
->container_content(st
, NULL
);
2479 for (content
= cc
; content
; content
= content
->next
) {
2481 if (!content
->reshape_active
)
2484 subarray
= strchr(content
->text_version
+1, '/')+1;
2485 mdstat
= mdstat_by_subdev(subarray
,
2486 devname2devnum(container
));
2494 fd
= open_dev(mdstat
->devnum
);
2497 adev
= map_dev(dev2major(mdstat
->devnum
),
2498 dev2minor(mdstat
->devnum
),
2501 adev
= content
->text_version
;
2503 sysfs_init(content
, fd
, mdstat
->devnum
);
2505 rv
= reshape_array(container
, fd
, adev
, st
,
2506 content
, force
, NULL
,
2507 backup_file
, quiet
, 1, restart
,
2511 if (freeze_reshape
) {
2527 * We run a child process in the background which performs the following
2529 * - wait for resync to reach a certain point
2530 * - suspend io to the following section
2531 * - backup that section
2532 * - allow resync to proceed further
2534 * - discard the backup.
2536 * When are combined in slightly different ways in the three cases.
2538 * - suspend/backup/allow/wait/resume/discard
2540 * - allow/wait/suspend/backup/allow/wait/resume/discard
2542 * - wait/resume/discard/suspend/backup/allow
2544 * suspend/backup/allow always come together
2545 * wait/resume/discard do too.
2546 * For the same-size case we have two backups to improve flow.
2550 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
2551 unsigned long long backup_point
,
2552 unsigned long long wait_point
,
2553 unsigned long long *suspend_point
,
2554 unsigned long long *reshape_completed
)
2556 /* This function is called repeatedly by the reshape manager.
2557 * It determines how much progress can safely be made and allows
2559 * - 'info' identifies the array and particularly records in
2560 * ->reshape_progress the metadata's knowledge of progress
2561 * This is a sector offset from the start of the array
2562 * of the next array block to be relocated. This number
2563 * may increase from 0 or decrease from array_size, depending
2564 * on the type of reshape that is happening.
2565 * Note that in contrast, 'sync_completed' is a block count of the
2566 * reshape so far. It gives the distance between the start point
2567 * (head or tail of device) and the next place that data will be
2568 * written. It always increases.
2569 * - 'reshape' is the structure created by analyse_change
2570 * - 'backup_point' shows how much the metadata manager has backed-up
2571 * data. For reshapes with increasing progress, it is the next address
2572 * to be backed up, previous addresses have been backed-up. For
2573 * decreasing progress, it is the earliest address that has been
2574 * backed up - later address are also backed up.
2575 * So addresses between reshape_progress and backup_point are
2576 * backed up providing those are in the 'correct' order.
2577 * - 'wait_point' is an array address. When reshape_completed
2578 * passes this point, progress_reshape should return. It might
2579 * return earlier if it determines that ->reshape_progress needs
2580 * to be updated or further backup is needed.
2581 * - suspend_point is maintained by progress_reshape and the caller
2582 * should not touch it except to initialise to zero.
2583 * It is an array address and it only increases in 2.6.37 and earlier.
2584 * This makes it difficult to handle reducing reshapes with
2585 * external metadata.
2586 * However: it is similar to backup_point in that it records the
2587 * other end of a suspended region from reshape_progress.
2588 * it is moved to extend the region that is safe to backup and/or
2590 * - reshape_completed is read from sysfs and returned. The caller
2591 * should copy this into ->reshape_progress when it has reason to
2592 * believe that the metadata knows this, and any backup outside this
2596 * 1 if more data from backup_point - but only as far as suspend_point,
2597 * should be backed up
2598 * 0 if things are progressing smoothly
2599 * -1 if the reshape is finished because it is all done,
2600 * -2 if the reshape is finished due to an error.
2603 int advancing
= (reshape
->after
.data_disks
2604 >= reshape
->before
.data_disks
);
2605 unsigned long long need_backup
; /* All data between start of array and
2606 * here will at some point need to
2609 unsigned long long read_offset
, write_offset
;
2610 unsigned long long write_range
;
2611 unsigned long long max_progress
, target
, completed
;
2612 unsigned long long array_size
= (info
->component_size
2613 * reshape
->before
.data_disks
);
2617 /* First, we unsuspend any region that is now known to be safe.
2618 * If suspend_point is on the 'wrong' side of reshape_progress, then
2619 * we don't have or need suspension at the moment. This is true for
2620 * native metadata when we don't need to back-up.
2623 if (info
->reshape_progress
<= *suspend_point
)
2624 sysfs_set_num(info
, NULL
, "suspend_lo",
2625 info
->reshape_progress
);
2627 /* Note: this won't work in 2.6.37 and before.
2628 * Something somewhere should make sure we don't need it!
2630 if (info
->reshape_progress
>= *suspend_point
)
2631 sysfs_set_num(info
, NULL
, "suspend_hi",
2632 info
->reshape_progress
);
2635 /* Now work out how far it is safe to progress.
2636 * If the read_offset for ->reshape_progress is less than
2637 * 'blocks' beyond the write_offset, we can only progress as far
2639 * Otherwise we can progress until the write_offset for the new location
2640 * reaches (within 'blocks' of) the read_offset at the current location.
2641 * However that region must be suspended unless we are using native
2643 * If we need to suspend more, we limit it to 128M per device, which is
2644 * rather arbitrary and should be some time-based calculation.
2646 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
2647 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
2648 write_range
= info
->new_chunk
/512;
2649 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
2650 need_backup
= array_size
;
2652 need_backup
= reshape
->backup_blocks
;
2654 if (read_offset
< write_offset
+ write_range
)
2655 max_progress
= backup_point
;
2659 reshape
->after
.data_disks
;
2661 if (read_offset
> write_offset
- write_range
)
2662 /* Can only progress as far as has been backed up,
2663 * which must be suspended */
2664 max_progress
= backup_point
;
2665 else if (info
->reshape_progress
<= need_backup
)
2666 max_progress
= backup_point
;
2668 if (info
->array
.major_version
>= 0)
2669 /* Can progress until backup is needed */
2670 max_progress
= need_backup
;
2672 /* Can progress until metadata update is required */
2675 reshape
->after
.data_disks
;
2676 /* but data must be suspended */
2677 if (max_progress
< *suspend_point
)
2678 max_progress
= *suspend_point
;
2683 /* We know it is safe to progress to 'max_progress' providing
2684 * it is suspended or we are using native metadata.
2685 * Consider extending suspend_point 128M per device if it
2686 * is less than 64M per device beyond reshape_progress.
2687 * But always do a multiple of 'blocks'
2688 * FIXME this is too big - it takes to long to complete
2691 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
2692 reshape
->after
.data_disks
);
2693 target
/= reshape
->backup_blocks
;
2696 target
*= reshape
->backup_blocks
;
2698 /* For externally managed metadata we always need to suspend IO to
2699 * the area being reshaped so we regularly push suspend_point forward.
2700 * For native metadata we only need the suspend if we are going to do
2704 if ((need_backup
> info
->reshape_progress
2705 || info
->array
.major_version
< 0) &&
2706 *suspend_point
< info
->reshape_progress
+ target
) {
2707 if (need_backup
< *suspend_point
+ 2 * target
)
2708 *suspend_point
= need_backup
;
2709 else if (*suspend_point
+ 2 * target
< array_size
)
2710 *suspend_point
+= 2 * target
;
2712 *suspend_point
= array_size
;
2713 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
2714 if (max_progress
> *suspend_point
)
2715 max_progress
= *suspend_point
;
2718 if (info
->array
.major_version
>= 0) {
2719 /* Only need to suspend when about to backup */
2720 if (info
->reshape_progress
< need_backup
* 2 &&
2721 *suspend_point
> 0) {
2723 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
2724 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
2727 /* Need to suspend continually */
2728 if (info
->reshape_progress
< *suspend_point
)
2729 *suspend_point
= info
->reshape_progress
;
2730 if (*suspend_point
+ target
< info
->reshape_progress
)
2731 /* No need to move suspend region yet */;
2733 if (*suspend_point
>= 2 * target
)
2734 *suspend_point
-= 2 * target
;
2737 sysfs_set_num(info
, NULL
, "suspend_lo",
2740 if (max_progress
< *suspend_point
)
2741 max_progress
= *suspend_point
;
2745 /* now set sync_max to allow that progress. sync_max, like
2746 * sync_completed is a count of sectors written per device, so
2747 * we find the difference between max_progress and the start point,
2748 * and divide that by after.data_disks to get a sync_max
2750 * At the same time we convert wait_point to a similar number
2751 * for comparing against sync_completed.
2753 /* scale down max_progress to per_disk */
2754 max_progress
/= reshape
->after
.data_disks
;
2755 /* Round to chunk size as some kernels give an erroneously high number */
2756 max_progress
/= info
->new_chunk
/512;
2757 max_progress
*= info
->new_chunk
/512;
2758 /* And round to old chunk size as the kernel wants that */
2759 max_progress
/= info
->array
.chunk_size
/512;
2760 max_progress
*= info
->array
.chunk_size
/512;
2761 /* Limit progress to the whole device */
2762 if (max_progress
> info
->component_size
)
2763 max_progress
= info
->component_size
;
2764 wait_point
/= reshape
->after
.data_disks
;
2766 /* switch from 'device offset' to 'processed block count' */
2767 max_progress
= info
->component_size
- max_progress
;
2768 wait_point
= info
->component_size
- wait_point
;
2771 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
2773 /* Now wait. If we have already reached the point that we were
2774 * asked to wait to, don't wait at all, else wait for any change.
2775 * We need to select on 'sync_completed' as that is the place that
2776 * notifications happen, but we are really interested in
2777 * 'reshape_position'
2779 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
2781 goto check_progress
;
2783 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
2784 goto check_progress
;
2786 while (completed
< max_progress
&& completed
< wait_point
) {
2787 /* Check that sync_action is still 'reshape' to avoid
2788 * waiting forever on a dead array
2792 if (sysfs_get_str(info
, NULL
, "sync_action",
2794 strncmp(action
, "reshape", 7) != 0)
2796 /* Some kernels reset 'sync_completed' to zero
2797 * before setting 'sync_action' to 'idle'.
2798 * So we need these extra tests.
2800 if (completed
== 0 && advancing
2801 && info
->reshape_progress
> 0)
2803 if (completed
== 0 && !advancing
2804 && info
->reshape_progress
< (info
->component_size
2805 * reshape
->after
.data_disks
))
2809 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
2810 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
2811 goto check_progress
;
2813 /* Some kernels reset 'sync_completed' to zero,
2814 * we need to have real point we are in md
2817 completed
= max_progress
;
2819 /* some kernels can give an incorrectly high 'completed' number */
2820 completed
/= (info
->new_chunk
/512);
2821 completed
*= (info
->new_chunk
/512);
2822 /* Convert 'completed' back in to a 'progress' number */
2823 completed
*= reshape
->after
.data_disks
;
2825 completed
= info
->component_size
* reshape
->after
.data_disks
2828 *reshape_completed
= completed
;
2832 /* We return the need_backup flag. Caller will decide
2833 * how much - a multiple of ->backup_blocks up to *suspend_point
2836 return need_backup
> info
->reshape_progress
;
2838 return need_backup
>= info
->reshape_progress
;
2841 /* if we couldn't read a number from sync_completed, then
2842 * either the reshape did complete, or it aborted.
2843 * We can tell which by checking for 'none' in reshape_position.
2844 * If it did abort, then it might immediately restart if it
2845 * it was just a device failure that leaves us degraded but
2849 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
, sizeof(buf
)) < 0
2850 || strncmp(buf
, "none", 4) != 0) {
2851 /* The abort might only be temporary. Wait up to 10
2852 * seconds for fd to contain a valid number again.
2858 while (fd
>= 0 && rv
< 0 && tv
.tv_sec
> 0) {
2862 if (select(fd
+1, NULL
, NULL
, &rfds
, &tv
) != 1)
2864 switch (sysfs_fd_get_ll(fd
, &completed
)) {
2866 /* all good again */
2869 case -2: /* read error - abort */
2876 return rv
; /* abort */
2878 /* Maybe racing with array shutdown - check state */
2881 if (sysfs_get_str(info
, NULL
, "array_state", buf
, sizeof(buf
)) < 0
2882 || strncmp(buf
, "inactive", 8) == 0
2883 || strncmp(buf
, "clear",5) == 0)
2884 return -2; /* abort */
2885 return -1; /* complete */
2890 /* FIXME return status is never checked */
2891 static int grow_backup(struct mdinfo
*sra
,
2892 unsigned long long offset
, /* per device */
2893 unsigned long stripes
, /* per device, in old chunks */
2894 int *sources
, unsigned long long *offsets
,
2895 int disks
, int chunk
, int level
, int layout
,
2896 int dests
, int *destfd
, unsigned long long *destoffsets
,
2897 int part
, int *degraded
,
2900 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2901 * to storage 'destfd' (offset 'destoffsets'), after first
2902 * suspending IO. Then allow resync to continue
2903 * over the suspended section.
2904 * Use part 'part' of the backup-super-block.
2909 unsigned long long ll
;
2911 //printf("offset %llu\n", offset);
2917 /* Check that array hasn't become degraded, else we might backup the wrong data */
2918 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
2919 return -1; /* FIXME this error is ignored */
2920 new_degraded
= (int)ll
;
2921 if (new_degraded
!= *degraded
) {
2922 /* check each device to ensure it is still working */
2924 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2925 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2927 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2929 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
2930 strstr(sbuf
, "faulty") ||
2931 strstr(sbuf
, "in_sync") == NULL
) {
2932 /* this device is dead */
2933 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
2934 if (sd
->disk
.raid_disk
>= 0 &&
2935 sources
[sd
->disk
.raid_disk
] >= 0) {
2936 close(sources
[sd
->disk
.raid_disk
]);
2937 sources
[sd
->disk
.raid_disk
] = -1;
2942 *degraded
= new_degraded
;
2945 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
2946 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2948 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
2949 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
2952 bsb
.magic
[15] = '2';
2953 for (i
= 0; i
< dests
; i
++)
2955 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
2957 lseek64(destfd
[i
], destoffsets
[i
], 0);
2959 rv
= save_stripes(sources
, offsets
,
2960 disks
, chunk
, level
, layout
,
2962 offset
*512*odata
, stripes
* chunk
* odata
,
2967 bsb
.mtime
= __cpu_to_le64(time(0));
2968 for (i
= 0; i
< dests
; i
++) {
2969 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
2971 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
2972 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
2973 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
2974 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
2977 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
2978 != destoffsets
[i
] - 4096)
2980 if (write(destfd
[i
], &bsb
, 512) != 512)
2982 if (destoffsets
[i
] > 4096) {
2983 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
2984 destoffsets
[i
]+stripes
*chunk
*odata
)
2986 if (write(destfd
[i
], &bsb
, 512) != 512)
2996 /* in 2.6.30, the value reported by sync_completed can be
2997 * less that it should be by one stripe.
2998 * This only happens when reshape hits sync_max and pauses.
2999 * So allow wait_backup to either extent sync_max further
3000 * than strictly necessary, or return before the
3001 * sync has got quite as far as we would really like.
3002 * This is what 'blocks2' is for.
3003 * The various caller give appropriate values so that
3006 /* FIXME return value is often ignored */
3007 static int forget_backup(
3008 int dests
, int *destfd
, unsigned long long *destoffsets
,
3012 * Erase backup 'part' (which is 0 or 1)
3018 bsb
.arraystart2
= __cpu_to_le64(0);
3019 bsb
.length2
= __cpu_to_le64(0);
3021 bsb
.arraystart
= __cpu_to_le64(0);
3022 bsb
.length
= __cpu_to_le64(0);
3024 bsb
.mtime
= __cpu_to_le64(time(0));
3026 for (i
= 0; i
< dests
; i
++) {
3027 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
3028 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
3029 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
3030 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
3031 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
3032 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
3033 destoffsets
[i
]-4096)
3036 write(destfd
[i
], &bsb
, 512) != 512)
3043 static void fail(char *msg
)
3046 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
3047 rv
|= (write(2, "\n", 1) != 1);
3051 static char *abuf
, *bbuf
;
3052 static unsigned long long abuflen
;
3053 static void validate(int afd
, int bfd
, unsigned long long offset
)
3055 /* check that the data in the backup against the array.
3056 * This is only used for regression testing and should not
3057 * be used while the array is active
3061 lseek64(bfd
, offset
- 4096, 0);
3062 if (read(bfd
, &bsb2
, 512) != 512)
3063 fail("cannot read bsb");
3064 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
3065 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
3066 fail("first csum bad");
3067 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
3068 fail("magic is bad");
3069 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
3070 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
3071 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
3072 fail("second csum bad");
3074 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
3075 fail("devstart is wrong");
3078 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
3080 if (abuflen
< len
) {
3084 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
3085 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
3087 /* just stop validating on mem-alloc failure */
3092 lseek64(bfd
, offset
, 0);
3093 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
3094 //printf("len %llu\n", len);
3095 fail("read first backup failed");
3097 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
3098 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
3099 fail("read first from array failed");
3100 if (memcmp(bbuf
, abuf
, len
) != 0) {
3103 printf("offset=%llu len=%llu\n",
3104 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
3105 for (i
=0; i
<len
; i
++)
3106 if (bbuf
[i
] != abuf
[i
]) {
3107 printf("first diff byte %d\n", i
);
3111 fail("data1 compare failed");
3115 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
3117 if (abuflen
< len
) {
3121 abuf
= malloc(abuflen
);
3122 bbuf
= malloc(abuflen
);
3125 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
3126 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
3127 fail("read second backup failed");
3128 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
3129 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
3130 fail("read second from array failed");
3131 if (memcmp(bbuf
, abuf
, len
) != 0)
3132 fail("data2 compare failed");
3136 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
3137 struct supertype
*st
, unsigned long blocks
,
3138 int *fds
, unsigned long long *offsets
,
3139 int dests
, int *destfd
, unsigned long long *destoffsets
)
3141 /* Monitor a reshape where backup is being performed using
3142 * 'native' mechanism - either to a backup file, or
3143 * to some space in a spare.
3147 unsigned long long speed
;
3148 unsigned long long suspend_point
, array_size
;
3149 unsigned long long backup_point
, wait_point
;
3150 unsigned long long reshape_completed
;
3152 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
3153 int part
= 0; /* The next part of the backup area to fill. It may already
3154 * be full, so we need to check */
3155 int level
= reshape
->level
;
3156 int layout
= reshape
->before
.layout
;
3157 int data
= reshape
->before
.data_disks
;
3158 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
3159 int chunk
= sra
->array
.chunk_size
;
3161 unsigned long stripes
;
3164 /* set up the backup-super-block. This requires the
3165 * uuid from the array.
3167 /* Find a superblock */
3168 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3172 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
3174 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
3175 devfd
= dev_open(dn
, O_RDONLY
);
3178 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
3184 fprintf(stderr
, Name
": Cannot find a superblock\n");
3188 memset(&bsb
, 0, 512);
3189 memcpy(bsb
.magic
, "md_backup_data-1", 16);
3190 st
->ss
->uuid_from_super(st
, uuid
);
3191 memcpy(bsb
.set_uuid
, uuid
, 16);
3192 bsb
.mtime
= __cpu_to_le64(time(0));
3193 bsb
.devstart2
= blocks
;
3195 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
3196 reshape
->before
.data_disks
;
3198 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
3199 /* Don't start the 'reshape' */
3201 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
3202 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
3203 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
3207 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
3208 backup_point
= sra
->reshape_progress
;
3211 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
3212 backup_point
= reshape
->backup_blocks
;
3213 suspend_point
= array_size
;
3219 /* Want to return as soon the oldest backup slot can
3220 * be released as that allows us to start backing up
3221 * some more, providing suspend_point has been
3222 * advanced, which it should have.
3225 wait_point
= array_size
;
3226 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
3227 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
3228 __le64_to_cpu(bsb
.length
));
3229 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
3230 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
3231 __le64_to_cpu(bsb
.length2
));
3234 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
3235 wait_point
= __le64_to_cpu(bsb
.arraystart
);
3236 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
3237 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
3240 rv
= progress_reshape(sra
, reshape
,
3241 backup_point
, wait_point
,
3242 &suspend_point
, &reshape_completed
);
3243 /* external metadata would need to ping_monitor here */
3244 sra
->reshape_progress
= reshape_completed
;
3246 /* Clear any backup region that is before 'here' */
3248 if (__le64_to_cpu(bsb
.length
) > 0 &&
3249 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
3250 __le64_to_cpu(bsb
.length
)))
3251 forget_backup(dests
, destfd
,
3253 if (__le64_to_cpu(bsb
.length2
) > 0 &&
3254 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
3255 __le64_to_cpu(bsb
.length2
)))
3256 forget_backup(dests
, destfd
,
3259 if (__le64_to_cpu(bsb
.length
) > 0 &&
3260 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
3261 forget_backup(dests
, destfd
,
3263 if (__le64_to_cpu(bsb
.length2
) > 0 &&
3264 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
3265 forget_backup(dests
, destfd
,
3274 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
3275 /* No longer need to monitor this reshape */
3281 unsigned long long offset
;
3282 unsigned long actual_stripes
;
3283 /* Need to backup some data.
3284 * If 'part' is not used and the desired
3285 * backup size is suspended, do a backup,
3286 * then consider the next part.
3288 /* Check that 'part' is unused */
3289 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
3291 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
3294 offset
= backup_point
/ data
;
3295 actual_stripes
= stripes
;
3297 if (offset
+ actual_stripes
* (chunk
/512) >
3298 sra
->component_size
)
3299 actual_stripes
= ((sra
->component_size
- offset
)
3301 if (offset
+ actual_stripes
* (chunk
/512) >
3305 if (offset
< actual_stripes
* (chunk
/512))
3306 actual_stripes
= offset
/ (chunk
/512);
3307 offset
-= actual_stripes
* (chunk
/512);
3308 if (offset
< suspend_point
/data
)
3311 if (actual_stripes
== 0)
3313 grow_backup(sra
, offset
, actual_stripes
,
3315 disks
, chunk
, level
, layout
,
3316 dests
, destfd
, destoffsets
,
3317 part
, °raded
, buf
);
3318 validate(afd
, destfd
[0], destoffsets
[0]);
3319 /* record where 'part' is up to */
3322 backup_point
+= actual_stripes
* (chunk
/512) * data
;
3324 backup_point
-= actual_stripes
* (chunk
/512) * data
;
3328 /* FIXME maybe call progress_reshape one more time instead */
3329 abort_reshape(sra
); /* remove any remaining suspension */
3330 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3331 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
3337 * If any spare contains md_back_data-1 which is recent wrt mtime,
3338 * write that data into the array and update the super blocks with
3339 * the new reshape_progress
3341 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
3342 char *backup_file
, int verbose
)
3346 unsigned long long *offsets
;
3347 unsigned long long nstripe
, ostripe
;
3350 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
3351 if (info
->array
.level
== 6) odata
--; /* number of data disks */
3352 ndata
= info
->array
.raid_disks
- 1;
3353 if (info
->new_level
== 6) ndata
--;
3355 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
3357 if (info
->delta_disks
<= 0)
3358 /* Didn't grow, so the backup file must have
3362 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
3363 struct mdinfo dinfo
;
3366 char *devname
, namebuf
[20];
3367 unsigned long long lo
, hi
;
3369 /* This was a spare and may have some saved data on it.
3370 * Load the superblock, find and load the
3371 * backup_super_block.
3372 * If either fail, go on to next device.
3373 * If the backup contains no new info, just return
3374 * else restore data and update all superblocks
3376 if (i
== old_disks
-1) {
3377 fd
= open(backup_file
, O_RDONLY
);
3379 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
3380 backup_file
, strerror(errno
));
3383 devname
= backup_file
;
3388 if (st
->ss
->load_super(st
, fd
, NULL
))
3391 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3392 st
->ss
->free_super(st
);
3395 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
3397 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
3398 continue; /* Cannot seek */
3400 sprintf(namebuf
, "device-%d", i
);
3403 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
3405 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
3406 continue; /* Cannot read */
3408 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
3409 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
3411 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
3414 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
3416 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
3417 continue; /* bad checksum */
3419 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
3420 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
3422 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
3423 continue; /* Bad second checksum */
3425 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
3427 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
3428 continue; /* Wrong uuid */
3431 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3432 * sometimes they aren't... So allow considerable flexability in matching, and allow
3433 * this test to be overridden by an environment variable.
3435 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
3436 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
3437 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3438 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
3439 "for array with timestamp %lu\n",
3440 (unsigned long)__le64_to_cpu(bsb
.mtime
),
3441 (unsigned long)info
->array
.utime
);
3444 fprintf(stderr
, Name
": too-old timestamp on "
3445 "backup-metadata on %s\n", devname
);
3446 continue; /* time stamp is too bad */
3450 if (bsb
.magic
[15] == '1') {
3451 if (bsb
.length
== 0)
3453 if (info
->delta_disks
>= 0) {
3454 /* reshape_progress is increasing */
3455 if (__le64_to_cpu(bsb
.arraystart
)
3456 + __le64_to_cpu(bsb
.length
)
3457 < info
->reshape_progress
) {
3460 fprintf(stderr
, Name
3461 ": backup-metadata found on %s but is not needed\n", devname
);
3462 continue; /* No new data here */
3465 /* reshape_progress is decreasing */
3466 if (__le64_to_cpu(bsb
.arraystart
) >=
3467 info
->reshape_progress
)
3468 goto nonew
; /* No new data here */
3471 if (bsb
.length
== 0 && bsb
.length2
== 0)
3473 if (info
->delta_disks
>= 0) {
3474 /* reshape_progress is increasing */
3475 if ((__le64_to_cpu(bsb
.arraystart
)
3476 + __le64_to_cpu(bsb
.length
)
3477 < info
->reshape_progress
)
3479 (__le64_to_cpu(bsb
.arraystart2
)
3480 + __le64_to_cpu(bsb
.length2
)
3481 < info
->reshape_progress
))
3482 goto nonew
; /* No new data here */
3484 /* reshape_progress is decreasing */
3485 if (__le64_to_cpu(bsb
.arraystart
) >=
3486 info
->reshape_progress
&&
3487 __le64_to_cpu(bsb
.arraystart2
) >=
3488 info
->reshape_progress
)
3489 goto nonew
; /* No new data here */
3492 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
3495 fprintf(stderr
, Name
3496 ": Failed to verify secondary backup-metadata block on %s\n",
3498 continue; /* Cannot seek */
3500 /* There should be a duplicate backup superblock 4k before here */
3501 if (lseek64(fd
, -4096, 1) < 0 ||
3502 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
3503 goto second_fail
; /* Cannot find leading superblock */
3504 if (bsb
.magic
[15] == '1')
3505 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
3507 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
3508 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
3509 goto second_fail
; /* Cannot find leading superblock */
3511 /* Now need the data offsets for all devices. */
3512 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
3513 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
3516 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3517 /* FIXME should be this be an error */
3519 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3520 st
->ss
->free_super(st
);
3521 offsets
[j
] = dinfo
.data_offset
* 512;
3523 printf(Name
": restoring critical section\n");
3525 if (restore_stripes(fdlist
, offsets
,
3526 info
->array
.raid_disks
,
3530 fd
, __le64_to_cpu(bsb
.devstart
)*512,
3531 __le64_to_cpu(bsb
.arraystart
)*512,
3532 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
3533 /* didn't succeed, so giveup */
3535 fprintf(stderr
, Name
": Error restoring backup from %s\n",
3540 if (bsb
.magic
[15] == '2' &&
3541 restore_stripes(fdlist
, offsets
,
3542 info
->array
.raid_disks
,
3546 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
3547 __le64_to_cpu(bsb
.devstart2
)*512,
3548 __le64_to_cpu(bsb
.arraystart2
)*512,
3549 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
3550 /* didn't succeed, so giveup */
3552 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
3558 /* Ok, so the data is restored. Let's update those superblocks. */
3562 lo
= __le64_to_cpu(bsb
.arraystart
);
3563 hi
= lo
+ __le64_to_cpu(bsb
.length
);
3565 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
3566 unsigned long long lo1
, hi1
;
3567 lo1
= __le64_to_cpu(bsb
.arraystart2
);
3568 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
3572 } else if (lo
< lo1
)
3578 (info
->reshape_progress
< lo
||
3579 info
->reshape_progress
> hi
))
3580 /* backup does not affect reshape_progress*/ ;
3581 else if (info
->delta_disks
>= 0) {
3582 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
3583 __le64_to_cpu(bsb
.length
);
3584 if (bsb
.magic
[15] == '2') {
3585 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
3586 __le64_to_cpu(bsb
.length2
);
3587 if (p2
> info
->reshape_progress
)
3588 info
->reshape_progress
= p2
;
3591 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
3592 if (bsb
.magic
[15] == '2') {
3593 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
3594 if (p2
< info
->reshape_progress
)
3595 info
->reshape_progress
= p2
;
3598 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
3599 if (fdlist
[j
] < 0) continue;
3600 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
3602 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
3603 dinfo
.reshape_progress
= info
->reshape_progress
;
3604 st
->ss
->update_super(st
, &dinfo
,
3605 "_reshape_progress",
3607 st
->ss
->store_super(st
, fdlist
[j
]);
3608 st
->ss
->free_super(st
);
3612 /* Didn't find any backup data, try to see if any
3615 if (info
->delta_disks
< 0) {
3616 /* When shrinking, the critical section is at the end.
3617 * So see if we are before the critical section.
3619 unsigned long long first_block
;
3620 nstripe
= ostripe
= 0;
3622 while (ostripe
>= nstripe
) {
3623 ostripe
+= info
->array
.chunk_size
/ 512;
3624 first_block
= ostripe
* odata
;
3625 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
3626 (info
->new_chunk
/512);
3629 if (info
->reshape_progress
>= first_block
)
3632 if (info
->delta_disks
> 0) {
3633 /* See if we are beyond the critical section. */
3634 unsigned long long last_block
;
3635 nstripe
= ostripe
= 0;
3637 while (nstripe
>= ostripe
) {
3638 nstripe
+= info
->new_chunk
/ 512;
3639 last_block
= nstripe
* ndata
;
3640 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
3641 (info
->array
.chunk_size
/512);
3644 if (info
->reshape_progress
>= last_block
)
3647 /* needed to recover critical section! */
3649 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
3653 int Grow_continue_command(char *devname
, int fd
,
3654 char *backup_file
, int verbose
)
3657 struct supertype
*st
= NULL
;
3658 struct mdinfo
*content
= NULL
;
3659 struct mdinfo array
;
3660 char *subarray
= NULL
;
3661 struct mdinfo
*cc
= NULL
;
3662 struct mdstat_ent
*mdstat
= NULL
;
3667 unsigned long long position
;
3669 dprintf("Grow continue from command line called for %s\n",
3672 st
= super_by_fd(fd
, &subarray
);
3673 if (!st
|| !st
->ss
) {
3675 Name
": Unable to determine metadata format for %s\n",
3679 dprintf("Grow continue is run for ");
3680 if (st
->ss
->external
== 0) {
3681 dprintf("native array (%s)\n", devname
);
3682 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
3683 fprintf(stderr
, Name
": %s is not an active md array -"
3684 " aborting\n", devname
);
3686 goto Grow_continue_command_exit
;
3689 sysfs_init(content
, fd
, st
->devnum
);
3694 dprintf("subarray (%s)\n", subarray
);
3695 container_dev
= st
->container_dev
;
3696 cfd
= open_dev_excl(st
->container_dev
);
3698 container_dev
= st
->devnum
;
3700 cfd
= open_dev_excl(st
->devnum
);
3701 dprintf("container (%i)\n", container_dev
);
3705 fprintf(stderr
, Name
": Unable to open container "
3706 "for %s\n", devname
);
3708 goto Grow_continue_command_exit
;
3710 fmt_devname(buf
, container_dev
);
3712 /* find in container array under reshape
3714 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
3717 Name
": Cannot read superblock for %s\n",
3720 goto Grow_continue_command_exit
;
3723 cc
= st
->ss
->container_content(st
, NULL
);
3724 for (content
= cc
; content
; content
= content
->next
) {
3727 if (content
->reshape_active
== 0)
3730 array
= strchr(content
->text_version
+1, '/')+1;
3731 mdstat
= mdstat_by_subdev(array
, container_dev
);
3738 Name
": Unable to determine reshaped "
3739 "array for %s\n", devname
);
3741 goto Grow_continue_command_exit
;
3743 fd2
= open_dev(mdstat
->devnum
);
3745 fprintf(stderr
, Name
": cannot open (md%i)\n",
3748 goto Grow_continue_command_exit
;
3751 sysfs_init(content
, fd2
, mdstat
->devnum
);
3753 /* start mdmon in case it is not running
3755 if (!mdmon_running(container_dev
))
3756 start_mdmon(container_dev
);
3759 if (mdmon_running(container_dev
))
3760 st
->update_tail
= &st
->updates
;
3762 fprintf(stderr
, Name
": No mdmon found. "
3763 "Grow cannot continue.\n");
3765 goto Grow_continue_command_exit
;
3769 /* verify that array under reshape is started from
3772 ret_val
= sysfs_get_str(content
, NULL
, "sync_max", buf
, 40);
3774 fprintf(stderr
, Name
3775 ": cannot open verify reshape progress for %s (%i)\n",
3776 content
->sys_name
, ret_val
);
3778 goto Grow_continue_command_exit
;
3780 dprintf(Name
": Read sync_max sysfs entry is: %s\n", buf
);
3781 position
= strtoull(buf
, &ep
, 0);
3782 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' ')) {
3783 fprintf(stderr
, Name
": Fatal error: array reshape was"
3784 " not properly frozen\n");
3786 goto Grow_continue_command_exit
;
3788 position
*= get_data_disks(map_name(pers
, mdstat
->level
),
3789 content
->new_layout
,
3790 content
->array
.raid_disks
);
3791 if (position
!= content
->reshape_progress
) {
3792 fprintf(stderr
, Name
": Fatal error: array reshape was"
3793 " not properly frozen.\n");
3795 goto Grow_continue_command_exit
;
3800 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 0);
3802 Grow_continue_command_exit
:
3807 st
->ss
->free_super(st
);
3808 free_mdstat(mdstat
);
3815 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
3816 char *backup_file
, int freeze_reshape
)
3820 if (!info
->reshape_active
)
3823 if (st
->ss
->external
) {
3825 int cfd
= open_dev(st
->container_dev
);
3830 fmt_devname(container
, st
->container_dev
);
3831 st
->ss
->load_container(st
, cfd
, container
);
3833 ret_val
= reshape_container(container
, NULL
, mdfd
,
3834 st
, info
, 0, backup_file
,
3835 0, 1, freeze_reshape
);
3837 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
3838 NULL
, backup_file
, 0, 0, 1,