2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
31 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
32 #error no endian defined
37 int restore_backup(struct supertype
*st
,
38 struct mdinfo
*content
,
48 int disk_count
= next_spare
+ working_disks
;
49 char *backup_file
= *backup_filep
;
51 dprintf("Called restore_backup()\n");
52 fdlist
= xmalloc(sizeof(int) * disk_count
);
54 enable_fds(next_spare
);
55 for (i
= 0; i
< next_spare
; i
++)
57 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
61 sprintf(buf
, "%d:%d", dev
->disk
.major
, dev
->disk
.minor
);
62 fd
= dev_open(buf
, O_RDWR
);
64 if (dev
->disk
.raid_disk
>= 0)
65 fdlist
[dev
->disk
.raid_disk
] = fd
;
67 fdlist
[next_spare
++] = fd
;
71 backup_file
= locate_backup(content
->sys_name
);
72 *backup_filep
= backup_file
;
75 if (st
->ss
->external
&& st
->ss
->recover_backup
)
76 err
= st
->ss
->recover_backup(st
, content
);
78 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
79 backup_file
, verbose
> 0);
81 while (next_spare
> 0) {
83 if (fdlist
[next_spare
] >= 0)
84 close(fdlist
[next_spare
]);
88 pr_err("Failed to restore critical section for reshape - sorry.\n");
90 pr_err("Possibly you need to specify a --backup-file\n");
94 dprintf("restore_backup() returns status OK.\n");
98 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
100 /* Add a device to an active array.
101 * Currently, just extend a linear array.
102 * This requires writing a new superblock on the
103 * new device, calling the kernel to add the device,
104 * and if that succeeds, update the superblock on
106 * This means that we need to *find* all other devices.
113 struct supertype
*st
= NULL
;
114 char *subarray
= NULL
;
116 if (md_get_array_info(fd
, &info
.array
) < 0) {
117 pr_err("cannot get array info for %s\n", devname
);
121 if (info
.array
.level
!= -1) {
122 pr_err("can only add devices to linear arrays\n");
126 st
= super_by_fd(fd
, &subarray
);
128 pr_err("cannot handle arrays with superblock version %d\n",
129 info
.array
.major_version
);
134 pr_err("Cannot grow linear sub-arrays yet\n");
140 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
142 pr_err("cannot open %s\n", newdev
);
146 if (!fstat_is_blkdev(nfd
, newdev
, &rdev
)) {
151 /* now check out all the devices and make sure we can read the
153 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
154 mdu_disk_info_t disk
;
157 st
->ss
->free_super(st
);
160 if (md_get_disk_info(fd
, &disk
) < 0) {
161 pr_err("cannot get device detail for device %d\n", d
);
166 dv
= map_dev(disk
.major
, disk
.minor
, 1);
168 pr_err("cannot find device file for device %d\n", d
);
173 fd2
= dev_open(dv
, O_RDWR
);
175 pr_err("cannot open device file %s\n", dv
);
181 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
182 pr_err("cannot find super block on %s\n", dv
);
190 /* Ok, looks good. Lets update the superblock and write it out to
194 info
.disk
.number
= d
;
195 info
.disk
.major
= major(rdev
);
196 info
.disk
.minor
= minor(rdev
);
197 info
.disk
.raid_disk
= d
;
198 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
199 if (st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
201 pr_err("Preparing new metadata failed on %s\n", newdev
);
206 if (st
->ss
->store_super(st
, nfd
)) {
207 pr_err("Cannot store new superblock on %s\n", newdev
);
213 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
214 pr_err("Cannot add new disk to this array\n");
217 /* Well, that seems to have worked.
218 * Now go through and update all superblocks
221 if (md_get_array_info(fd
, &info
.array
) < 0) {
222 pr_err("cannot get array info for %s\n", devname
);
227 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
228 mdu_disk_info_t disk
;
232 if (md_get_disk_info(fd
, &disk
) < 0) {
233 pr_err("cannot get device detail for device %d\n", d
);
236 dv
= map_dev(disk
.major
, disk
.minor
, 1);
238 pr_err("cannot find device file for device %d\n", d
);
241 fd2
= dev_open(dv
, O_RDWR
);
243 pr_err("cannot open device file %s\n", dv
);
246 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
247 pr_err("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 if (st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
259 pr_err("Updating metadata failed on %s\n", dv
);
264 if (st
->ss
->store_super(st
, fd2
)) {
265 pr_err("Cannot store new superblock on %s\n", dv
);
275 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
286 mdu_bitmap_file_t bmf
;
287 mdu_array_info_t array
;
288 struct supertype
*st
;
289 char *subarray
= NULL
;
290 int major
= BITMAP_MAJOR_HI
;
291 unsigned long long bitmapsize
, array_size
;
295 * We only ever get called if s->bitmap_file is != NULL, so this check
296 * is just here to quiet down static code checkers.
301 if (strcmp(s
->bitmap_file
, "clustered") == 0)
302 major
= BITMAP_MAJOR_CLUSTERED
;
304 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
306 pr_err("Memory allocation failure.\n");
308 pr_err("bitmaps not supported by this kernel.\n");
311 if (bmf
.pathname
[0]) {
312 if (strcmp(s
->bitmap_file
,"none") == 0) {
313 if (ioctl(fd
, SET_BITMAP_FILE
, -1) != 0) {
314 pr_err("failed to remove bitmap %s\n",
320 pr_err("%s already has a bitmap (%s)\n", devname
, bmf
.pathname
);
323 if (md_get_array_info(fd
, &array
) != 0) {
324 pr_err("cannot get array status for %s\n", devname
);
327 if (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) {
328 if (strcmp(s
->bitmap_file
, "none")==0) {
329 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
330 if (md_set_array_info(fd
, &array
) != 0) {
331 if (array
.state
& (1 << MD_SB_CLUSTERED
))
332 pr_err("failed to remove clustered bitmap.\n");
334 pr_err("failed to remove internal bitmap.\n");
339 pr_err("bitmap already present on %s\n", devname
);
343 if (strcmp(s
->bitmap_file
, "none") == 0) {
344 pr_err("no bitmap found on %s\n", devname
);
347 if (array
.level
<= 0) {
348 pr_err("Bitmaps not meaningful with level %s\n",
349 map_num(pers
, array
.level
)?:"of this array");
352 bitmapsize
= array
.size
;
354 if (get_dev_size(fd
, NULL
, &array_size
) &&
355 array_size
> (0x7fffffffULL
<< 9)) {
356 /* Array is big enough that we cannot trust array.size
357 * try other approaches
359 bitmapsize
= get_component_size(fd
);
361 if (bitmapsize
== 0) {
362 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
366 if (array
.level
== 10) {
369 ncopies
= (array
.layout
& 255) * ((array
.layout
>> 8) & 255);
370 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
372 if (strcmp(s
->bitmap_file
, "clustered") == 0 &&
373 !is_near_layout_10(array
.layout
)) {
374 pr_err("only near layout is supported with clustered raid10\n");
379 st
= super_by_fd(fd
, &subarray
);
381 pr_err("Cannot understand version %d.%d\n",
382 array
.major_version
, array
.minor_version
);
386 pr_err("Cannot add bitmaps to sub-arrays yet\n");
392 mdi
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
);
394 if (mdi
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
395 pr_err("Cannot add bitmap to array with PPL\n");
403 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
404 strcmp(s
->bitmap_file
, "clustered") == 0) {
407 int offset_setable
= 0;
408 if (st
->ss
->add_internal_bitmap
== NULL
) {
409 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
412 st
->nodes
= c
->nodes
;
413 st
->cluster_name
= c
->homecluster
;
414 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
417 for (d
= 0; d
< st
->max_devs
; d
++) {
418 mdu_disk_info_t disk
;
423 if (md_get_disk_info(fd
, &disk
) < 0)
425 if (disk
.major
== 0 && disk
.minor
== 0)
427 if ((disk
.state
& (1 << MD_DISK_SYNC
)) == 0)
429 dv
= map_dev(disk
.major
, disk
.minor
, 1);
432 if (((disk
.state
& (1 << MD_DISK_WRITEMOSTLY
)) == 0) &&
433 (strcmp(s
->bitmap_file
, "clustered") == 0)) {
434 pr_err("%s disks marked write-mostly are not supported with clustered bitmap\n",devname
);
437 fd2
= dev_open(dv
, O_RDWR
);
440 rv
= st
->ss
->load_super(st
, fd2
, NULL
);
442 rv
= st
->ss
->add_internal_bitmap(
443 st
, &s
->bitmap_chunk
, c
->delay
,
444 s
->write_behind
, bitmapsize
,
445 offset_setable
, major
);
447 st
->ss
->write_bitmap(st
, fd2
,
450 pr_err("failed to create internal bitmap - chunksize problem.\n");
453 pr_err("failed to load super-block.\n");
459 if (offset_setable
) {
460 st
->ss
->getinfo_super(st
, mdi
, NULL
);
461 if (sysfs_init(mdi
, fd
, NULL
)) {
462 pr_err("failed to initialize sysfs.\n");
465 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
469 if (strcmp(s
->bitmap_file
, "clustered") == 0)
470 array
.state
|= (1 << MD_SB_CLUSTERED
);
471 array
.state
|= (1 << MD_SB_BITMAP_PRESENT
);
472 rv
= md_set_array_info(fd
, &array
);
476 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
477 pr_err("failed to set internal bitmap.\n");
484 int max_devs
= st
->max_devs
;
486 /* try to load a superblock */
487 for (d
= 0; d
< max_devs
; d
++) {
488 mdu_disk_info_t disk
;
492 if (md_get_disk_info(fd
, &disk
) < 0)
494 if ((disk
.major
==0 && disk
.minor
== 0) ||
495 (disk
.state
& (1 << MD_DISK_REMOVED
)))
497 dv
= map_dev(disk
.major
, disk
.minor
, 1);
500 fd2
= dev_open(dv
, O_RDONLY
);
502 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
504 st
->ss
->uuid_from_super(st
, uuid
);
511 pr_err("cannot find UUID for array!\n");
514 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
,
515 s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
516 bitmapsize
, major
)) {
519 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
521 pr_err("weird: %s cannot be opened\n", s
->bitmap_file
);
524 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
527 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
528 pr_err("Cannot set bitmap file for %s: %s\n",
529 devname
, strerror(err
));
537 int Grow_consistency_policy(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
539 struct supertype
*st
;
542 char *subarray
= NULL
;
544 char container_dev
[PATH_MAX
];
547 if (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
548 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
549 pr_err("Operation not supported for consistency policy %s\n",
550 map_num_s(consistency_policies
, s
->consistency_policy
));
554 st
= super_by_fd(fd
, &subarray
);
558 sra
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
|GET_LEVEL
|
565 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
566 !st
->ss
->write_init_ppl
) {
567 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
572 if (sra
->array
.level
!= 5) {
573 pr_err("Operation not supported for array level %d\n",
579 if (sra
->consistency_policy
== (unsigned)s
->consistency_policy
) {
580 pr_err("Consistency policy is already %s\n",
581 map_num_s(consistency_policies
, s
->consistency_policy
));
584 } else if (sra
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
585 sra
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
586 pr_err("Current consistency policy is %s, cannot change to %s\n",
587 map_num_s(consistency_policies
, sra
->consistency_policy
),
588 map_num_s(consistency_policies
, s
->consistency_policy
));
593 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
594 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) <= 0) {
597 } else if (strcmp(buf
, "reshape\n") == 0) {
598 pr_err("PPL cannot be enabled when reshape is in progress\n");
607 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
612 sprintf(container_dev
, "/dev/%s", st
->container_devnm
);
614 ret
= Update_subarray(container_dev
, subarray
, update
, NULL
,
620 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
627 cfd
= open(container_dev
, O_RDWR
|O_EXCL
);
629 pr_err("Failed to open %s\n", container_dev
);
634 ret
= st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
638 pr_err("Cannot read superblock for %s\n",
643 mdi
= st
->ss
->container_content(st
, subarray
);
648 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
652 devpath
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
653 dfd
= dev_open(devpath
, O_RDWR
);
655 pr_err("Failed to open %s\n", devpath
);
661 ret
= st
->ss
->load_super(st
, dfd
, NULL
);
663 pr_err("Failed to load super-block.\n");
668 ret
= st
->ss
->update_super(st
, sra
, "ppl",
670 c
->verbose
, 0, NULL
);
673 st
->ss
->free_super(st
);
676 st
->ss
->getinfo_super(st
, &info
, NULL
);
679 ret
|= sysfs_set_num(sra
, sd
, "ppl_sector",
681 ret
|= sysfs_set_num(sra
, sd
, "ppl_size",
685 pr_err("Failed to set PPL attributes for %s\n",
688 st
->ss
->free_super(st
);
692 ret
= st
->ss
->write_init_ppl(st
, &info
, dfd
);
694 pr_err("Failed to write PPL\n");
699 st
->ss
->free_super(st
);
706 ret
= sysfs_set_str(sra
, NULL
, "consistency_policy",
707 map_num_s(consistency_policies
,
708 s
->consistency_policy
));
710 pr_err("Failed to change array consistency policy\n");
722 * When reshaping an array we might need to backup some data.
723 * This is written to all spares with a 'super_block' describing it.
724 * The superblock goes 4K from the end of the used space on the
726 * It if written after the backup is complete.
727 * It has the following structure.
730 static struct mdp_backup_super
{
731 char magic
[16]; /* md_backup_data-1 or -2 */
734 /* start/sizes in 512byte sectors */
735 __u64 devstart
; /* address on backup device/file of data */
738 __u32 sb_csum
; /* csum of preceeding bytes. */
740 __u64 devstart2
; /* offset in to data of second section */
743 __u32 sb_csum2
; /* csum of preceeding bytes. */
745 } __attribute__((aligned(512))) bsb
, bsb2
;
747 static __u32
bsb_csum(char *buf
, int len
)
751 for (i
= 0; i
< len
; i
++)
752 csum
= (csum
<<3) + buf
[0];
753 return __cpu_to_le32(csum
);
756 static int check_idle(struct supertype
*st
)
758 /* Check that all member arrays for this container, or the
759 * container of this array, are idle
761 char *container
= (st
->container_devnm
[0]
762 ? st
->container_devnm
: st
->devnm
);
763 struct mdstat_ent
*ent
, *e
;
766 ent
= mdstat_read(0, 0);
767 for (e
= ent
; e
; e
= e
->next
) {
768 if (!is_container_member(e
, container
))
770 /* frozen array is not idle*/
771 if (e
->percent
>= 0 || e
->metadata_version
[9] == '-') {
780 static int freeze_container(struct supertype
*st
)
782 char *container
= (st
->container_devnm
[0]
783 ? st
->container_devnm
: st
->devnm
);
788 if (block_monitor(container
, 1)) {
789 pr_err("failed to freeze container\n");
796 static void unfreeze_container(struct supertype
*st
)
798 char *container
= (st
->container_devnm
[0]
799 ? st
->container_devnm
: st
->devnm
);
801 unblock_monitor(container
, 1);
804 static int freeze(struct supertype
*st
)
806 /* Try to freeze resync/rebuild on this array/container.
807 * Return -1 if the array is busy,
808 * return -2 container cannot be frozen,
809 * return 0 if this kernel doesn't support 'frozen'
810 * return 1 if it worked.
812 if (st
->ss
->external
)
813 return freeze_container(st
);
815 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
821 /* Need to clear any 'read-auto' status */
822 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
823 strncmp(buf
, "read-auto", 9) == 0)
824 sysfs_set_str(sra
, NULL
, "array_state", "clean");
826 err
= sysfs_freeze_array(sra
);
832 static void unfreeze(struct supertype
*st
)
834 if (st
->ss
->external
)
835 return unfreeze_container(st
);
837 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
841 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
842 strcmp(buf
, "frozen\n") == 0)
843 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
848 static void wait_reshape(struct mdinfo
*sra
)
850 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
856 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
857 strncmp(action
, "reshape", 7) == 0)
858 sysfs_wait(fd
, NULL
);
862 static int reshape_super(struct supertype
*st
, unsigned long long size
,
863 int level
, int layout
, int chunksize
, int raid_disks
,
864 int delta_disks
, char *backup_file
, char *dev
,
865 int direction
, int verbose
)
867 /* nothing extra to check in the native case */
868 if (!st
->ss
->external
)
870 if (!st
->ss
->reshape_super
|| !st
->ss
->manage_reshape
) {
871 pr_err("%s metadata does not support reshape\n",
876 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
877 raid_disks
, delta_disks
, backup_file
, dev
,
881 static void sync_metadata(struct supertype
*st
)
883 if (st
->ss
->external
) {
884 if (st
->update_tail
) {
885 flush_metadata_updates(st
);
886 st
->update_tail
= &st
->updates
;
888 st
->ss
->sync_metadata(st
);
892 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
894 /* when dealing with external metadata subarrays we need to be
895 * prepared to handle EAGAIN. The kernel may need to wait for
896 * mdmon to mark the array active so the kernel can handle
897 * allocations/writeback when preparing the reshape action
898 * (md_allow_write()). We temporarily disable safe_mode_delay
899 * to close a race with the array_state going clean before the
900 * next write to raid_disks / stripe_cache_size
905 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
907 (strcmp(name
, "raid_disks") != 0 &&
908 strcmp(name
, "stripe_cache_size") != 0))
909 return sysfs_set_num(sra
, NULL
, name
, n
);
911 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
914 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
915 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
916 if (rc
< 0 && errno
== EAGAIN
) {
917 ping_monitor(container
);
918 /* if we get EAGAIN here then the monitor is not active
921 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
923 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
927 int start_reshape(struct mdinfo
*sra
, int already_running
,
928 int before_data_disks
, int data_disks
, struct supertype
*st
)
931 unsigned long long sync_max_to_set
;
933 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
934 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
935 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
936 sra
->reshape_progress
);
937 if (before_data_disks
<= data_disks
)
938 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
940 sync_max_to_set
= (sra
->component_size
* data_disks
941 - sra
->reshape_progress
) / data_disks
;
943 if (!already_running
)
944 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
946 if (st
->ss
->external
)
947 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
949 err
= err
?: sysfs_set_str(sra
, NULL
, "sync_max", "max");
951 if (!already_running
&& err
== 0) {
954 err
= sysfs_set_str(sra
, NULL
, "sync_action",
958 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
963 void abort_reshape(struct mdinfo
*sra
)
965 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
967 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
968 * suspend_hi to decrease as well as increase.")
969 * you could only increase suspend_{lo,hi} unless the region they
970 * covered was empty. So to reset to 0, you need to push suspend_lo
971 * up past suspend_hi first. So to maximize the chance of mdadm
972 * working on all kernels, we want to keep doing that.
974 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
975 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
976 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
977 sysfs_set_num(sra
, NULL
, "sync_min", 0);
978 // It isn't safe to reset sync_max as we aren't monitoring.
979 // Array really should be stopped at this point.
982 int remove_disks_for_takeover(struct supertype
*st
,
987 struct mdinfo
*remaining
;
990 if (st
->ss
->external
) {
992 struct mdinfo
*arrays
= st
->ss
->container_content(st
, NULL
);
994 * containter_content returns list of arrays in container
995 * If arrays->next is not NULL it means that there are
996 * 2 arrays in container and operation should be blocked
1003 pr_err("Error. Cannot perform operation on %s- for this operation "
1004 "it MUST be single array in container\n", st
->devnm
);
1010 if (sra
->array
.level
== 10)
1011 nr_of_copies
= layout
& 0xff;
1012 else if (sra
->array
.level
== 1)
1013 nr_of_copies
= sra
->array
.raid_disks
;
1017 remaining
= sra
->devs
;
1019 /* for each 'copy', select one device and remove from the list. */
1020 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
1021 struct mdinfo
**diskp
;
1024 /* Find a working device to keep */
1025 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
1026 struct mdinfo
*disk
= *diskp
;
1028 if (disk
->disk
.raid_disk
< slot
)
1030 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
1032 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
1034 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
1036 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
1039 /* We have found a good disk to use! */
1040 *diskp
= disk
->next
;
1041 disk
->next
= sra
->devs
;
1050 if (slot
< sra
->array
.raid_disks
) {
1051 /* didn't find all slots */
1057 sra
->devs
= remaining
;
1061 /* Remove all 'remaining' devices from the array */
1063 struct mdinfo
*sd
= remaining
;
1064 remaining
= sd
->next
;
1066 sysfs_set_str(sra
, sd
, "state", "faulty");
1067 sysfs_set_str(sra
, sd
, "slot", "none");
1068 /* for external metadata disks should be removed in mdmon */
1069 if (!st
->ss
->external
)
1070 sysfs_set_str(sra
, sd
, "state", "remove");
1071 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
1072 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
1073 sd
->next
= sra
->devs
;
1079 void reshape_free_fdlist(int *fdlist
,
1080 unsigned long long *offsets
,
1085 for (i
= 0; i
< size
; i
++)
1093 int reshape_prepare_fdlist(char *devname
,
1097 unsigned long blocks
,
1100 unsigned long long *offsets
)
1105 enable_fds(nrdisks
);
1106 for (d
= 0; d
<= nrdisks
; d
++)
1109 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1110 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1112 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
1113 sd
->disk
.raid_disk
< raid_disks
) {
1114 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1115 fdlist
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDONLY
);
1116 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1117 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1118 pr_err("%s: cannot open component %s\n",
1119 devname
, dn
? dn
: "-unknown-");
1123 } else if (backup_file
== NULL
) {
1125 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1126 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1127 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1128 if (fdlist
[d
] < 0) {
1129 pr_err("%s: cannot open component %s\n",
1130 devname
, dn
? dn
: "-unknown-");
1141 int reshape_open_backup_file(char *backup_file
,
1146 unsigned long long *offsets
,
1150 /* Return 1 on success, 0 on any form of failure */
1151 /* need to check backup file is large enough */
1157 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
1161 pr_err("%s: cannot create backup file %s: %s\n",
1162 devname
, backup_file
, strerror(errno
));
1165 /* Guard against backup file being on array device.
1166 * If array is partitioned or if LVM etc is in the
1167 * way this will not notice, but it is better than
1170 fstat(*fdlist
, &stb
);
1173 if (stb
.st_rdev
== dev
) {
1174 pr_err("backup file must NOT be on the array being reshaped.\n");
1179 memset(buf
, 0, 512);
1180 for (i
=0; i
< blocks
+ 8 ; i
++) {
1181 if (write(*fdlist
, buf
, 512) != 512) {
1182 pr_err("%s: cannot create backup file %s: %s\n",
1183 devname
, backup_file
, strerror(errno
));
1187 if (fsync(*fdlist
) != 0) {
1188 pr_err("%s: cannot create backup file %s: %s\n",
1189 devname
, backup_file
, strerror(errno
));
1193 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
1194 char *bu
= make_backup(sys_name
);
1195 if (symlink(backup_file
, bu
))
1196 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
1204 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
1205 unsigned int ndata
, unsigned int odata
)
1207 unsigned long a
, b
, blocks
;
1208 /* So how much do we need to backup.
1209 * We need an amount of data which is both a whole number of
1210 * old stripes and a whole number of new stripes.
1211 * So LCM for (chunksize*datadisks).
1213 a
= (ochunk
/512) * odata
;
1214 b
= (nchunk
/512) * ndata
;
1217 /* LCM == product / GCD */
1218 blocks
= (unsigned long)(ochunk
/512) * (unsigned long)(nchunk
/512) *
1224 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1226 /* Based on the current array state in info->array and
1227 * the changes in info->new_* etc, determine:
1228 * - whether the change is possible
1229 * - Intermediate level/raid_disks/layout
1230 * - whether a restriping reshape is needed
1231 * - number of sectors in minimum change unit. This
1232 * will cover a whole number of stripes in 'before' and
1235 * Return message if the change should be rejected
1236 * NULL if the change can be achieved
1238 * This can be called as part of starting a reshape, or
1239 * when assembling an array that is undergoing reshape.
1241 int near
, far
, offset
, copies
;
1243 int old_chunk
, new_chunk
;
1244 /* delta_parity records change in number of devices
1245 * caused by level change
1247 int delta_parity
= 0;
1249 memset(re
, 0, sizeof(*re
));
1251 /* If a new level not explicitly given, we assume no-change */
1252 if (info
->new_level
== UnSet
)
1253 info
->new_level
= info
->array
.level
;
1255 if (info
->new_chunk
)
1256 switch (info
->new_level
) {
1262 /* chunk size is meaningful, must divide component_size
1265 if (info
->component_size
% (info
->new_chunk
/512)) {
1266 unsigned long long shrink
= info
->component_size
;
1267 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1268 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1269 info
->new_chunk
/1024, info
->component_size
/2);
1270 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1272 pr_err("will shrink the array so the given chunk size would work.\n");
1277 return "chunk size not meaningful for this level";
1280 info
->new_chunk
= info
->array
.chunk_size
;
1282 switch (info
->array
.level
) {
1284 return "No reshape is possibly for this RAID level";
1286 if (info
->delta_disks
!= UnSet
)
1287 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1289 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1291 /* RAID1 can convert to RAID1 with different disks, or
1292 * raid5 with 2 disks, or
1295 if (info
->new_level
> 1 && (info
->component_size
& 7))
1296 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1297 if (info
->new_level
== 0) {
1298 if (info
->delta_disks
!= UnSet
&&
1299 info
->delta_disks
!= 0)
1300 return "Cannot change number of disks with RAID1->RAID0 conversion";
1302 re
->before
.data_disks
= 1;
1303 re
->after
.data_disks
= 1;
1306 if (info
->new_level
== 1) {
1307 if (info
->delta_disks
== UnSet
)
1308 /* Don't know what to do */
1309 return "no change requested for Growing RAID1";
1313 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1314 return "Can only convert a 2-device array to RAID5";
1315 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1317 re
->before
.data_disks
= 1;
1318 if (info
->delta_disks
!= UnSet
&&
1319 info
->delta_disks
!= 0)
1320 re
->after
.data_disks
= 1 + info
->delta_disks
;
1322 re
->after
.data_disks
= 1;
1323 if (re
->after
.data_disks
< 1)
1324 return "Number of disks too small for RAID5";
1326 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1327 info
->array
.chunk_size
= 65536;
1330 /* Could do some multi-stage conversions, but leave that to
1333 return "Impossibly level change request for RAID1";
1336 /* RAID10 can be converted from near mode to
1337 * RAID0 by removing some devices.
1338 * It can also be reshaped if the kernel supports
1341 switch (info
->new_level
) {
1343 if ((info
->array
.layout
& ~0xff) != 0x100)
1344 return "Cannot Grow RAID10 with far/offset layout";
1346 * number of devices must be multiple of
1349 if (info
->array
.raid_disks
%
1350 (info
->array
.layout
& 0xff))
1351 return "RAID10 layout too complex for Grow operation";
1353 new_disks
= (info
->array
.raid_disks
/
1354 (info
->array
.layout
& 0xff));
1355 if (info
->delta_disks
== UnSet
)
1356 info
->delta_disks
= (new_disks
1357 - info
->array
.raid_disks
);
1359 if (info
->delta_disks
!=
1360 new_disks
- info
->array
.raid_disks
)
1361 return "New number of raid-devices impossible for RAID10";
1362 if (info
->new_chunk
&&
1363 info
->new_chunk
!= info
->array
.chunk_size
)
1364 return "Cannot change chunk-size with RAID10 Grow";
1368 re
->before
.data_disks
= new_disks
;
1369 re
->after
.data_disks
= re
->before
.data_disks
;
1373 near
= info
->array
.layout
& 0xff;
1374 far
= (info
->array
.layout
>> 8) & 0xff;
1375 offset
= info
->array
.layout
& 0x10000;
1376 if (far
> 1 && !offset
)
1377 return "Cannot reshape RAID10 in far-mode";
1378 copies
= near
* far
;
1380 old_chunk
= info
->array
.chunk_size
* far
;
1382 if (info
->new_layout
== UnSet
)
1383 info
->new_layout
= info
->array
.layout
;
1385 near
= info
->new_layout
& 0xff;
1386 far
= (info
->new_layout
>> 8) & 0xff;
1387 offset
= info
->new_layout
& 0x10000;
1388 if (far
> 1 && !offset
)
1389 return "Cannot reshape RAID10 to far-mode";
1390 if (near
* far
!= copies
)
1391 return "Cannot change number of copies when reshaping RAID10";
1393 if (info
->delta_disks
== UnSet
)
1394 info
->delta_disks
= 0;
1395 new_disks
= (info
->array
.raid_disks
+
1398 new_chunk
= info
->new_chunk
* far
;
1401 re
->before
.layout
= info
->array
.layout
;
1402 re
->before
.data_disks
= info
->array
.raid_disks
;
1403 re
->after
.layout
= info
->new_layout
;
1404 re
->after
.data_disks
= new_disks
;
1405 /* For RAID10 we don't do backup but do allow reshape,
1406 * so set backup_blocks to INVALID_SECTORS rather than
1408 * And there is no need to synchronise stripes on both
1409 * 'old' and 'new'. So the important
1410 * number is the minimum data_offset difference
1411 * which is the larger of (offset copies * chunk).
1413 re
->backup_blocks
= INVALID_SECTORS
;
1414 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1415 if (new_disks
< re
->before
.data_disks
&&
1416 info
->space_after
< re
->min_offset_change
)
1417 /* Reduce component size by one chunk */
1418 re
->new_size
= (info
->component_size
-
1419 re
->min_offset_change
);
1421 re
->new_size
= info
->component_size
;
1422 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1426 return "RAID10 can only be changed to RAID0";
1429 /* RAID0 can be converted to RAID10, or to RAID456 */
1430 if (info
->new_level
== 10) {
1431 if (info
->new_layout
== UnSet
&&
1432 info
->delta_disks
== UnSet
) {
1433 /* Assume near=2 layout */
1434 info
->new_layout
= 0x102;
1435 info
->delta_disks
= info
->array
.raid_disks
;
1437 if (info
->new_layout
== UnSet
) {
1438 int copies
= 1 + (info
->delta_disks
1439 / info
->array
.raid_disks
);
1440 if (info
->array
.raid_disks
* (copies
-1) !=
1442 return "Impossible number of devices for RAID0->RAID10";
1443 info
->new_layout
= 0x100 + copies
;
1445 if (info
->delta_disks
== UnSet
) {
1446 int copies
= info
->new_layout
& 0xff;
1447 if (info
->new_layout
!= 0x100 + copies
)
1448 return "New layout impossible for RAID0->RAID10";;
1449 info
->delta_disks
= (copies
- 1) *
1450 info
->array
.raid_disks
;
1452 if (info
->new_chunk
&&
1453 info
->new_chunk
!= info
->array
.chunk_size
)
1454 return "Cannot change chunk-size with RAID0->RAID10";
1457 re
->before
.data_disks
= (info
->array
.raid_disks
+
1459 re
->after
.data_disks
= re
->before
.data_disks
;
1460 re
->before
.layout
= info
->new_layout
;
1464 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1465 * a raid4 style layout of the final level.
1467 switch (info
->new_level
) {
1472 re
->before
.layout
= 0;
1477 re
->before
.layout
= ALGORITHM_PARITY_N
;
1478 if (info
->new_layout
== UnSet
)
1479 info
->new_layout
= map_name(r5layout
, "default");
1484 re
->before
.layout
= ALGORITHM_PARITY_N
;
1485 if (info
->new_layout
== UnSet
)
1486 info
->new_layout
= map_name(r6layout
, "default");
1489 return "Impossible level change requested";
1491 re
->before
.data_disks
= info
->array
.raid_disks
;
1492 /* determining 'after' layout happens outside this 'switch' */
1496 info
->array
.layout
= ALGORITHM_PARITY_N
;
1498 switch (info
->new_level
) {
1502 re
->level
= info
->array
.level
;
1503 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1504 re
->before
.layout
= info
->array
.layout
;
1508 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1509 re
->before
.layout
= info
->array
.layout
;
1514 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1515 switch (info
->array
.layout
) {
1516 case ALGORITHM_LEFT_ASYMMETRIC
:
1517 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1519 case ALGORITHM_RIGHT_ASYMMETRIC
:
1520 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1522 case ALGORITHM_LEFT_SYMMETRIC
:
1523 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1525 case ALGORITHM_RIGHT_SYMMETRIC
:
1526 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1528 case ALGORITHM_PARITY_0
:
1529 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1531 case ALGORITHM_PARITY_N
:
1532 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1535 return "Cannot convert an array with this layout";
1539 if (info
->array
.raid_disks
!= 2)
1540 return "Can only convert a 2-device array to RAID1";
1541 if (info
->delta_disks
!= UnSet
&&
1542 info
->delta_disks
!= 0)
1543 return "Cannot set raid_disk when converting RAID5->RAID1";
1545 info
->new_chunk
= 0;
1548 return "Impossible level change requested";
1552 switch (info
->new_level
) {
1558 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1559 re
->before
.layout
= info
->array
.layout
;
1562 return "Impossible level change requested";
1567 /* If we reached here then it looks like a re-stripe is
1568 * happening. We have determined the intermediate level
1569 * and initial raid_disks/layout and stored these in 're'.
1571 * We need to deduce the final layout that can be atomically
1572 * converted to the end state.
1574 switch (info
->new_level
) {
1576 /* We can only get to RAID0 from RAID4 or RAID5
1577 * with appropriate layout and one extra device
1579 if (re
->level
!= 4 && re
->level
!= 5)
1580 return "Cannot covert to RAID0 from this level";
1582 switch (re
->level
) {
1584 re
->before
.layout
= 0;
1585 re
->after
.layout
= 0;
1588 re
->after
.layout
= ALGORITHM_PARITY_N
;
1594 /* We can only get to RAID4 from RAID5 */
1595 if (re
->level
!= 4 && re
->level
!= 5)
1596 return "Cannot convert to RAID4 from this level";
1598 switch (re
->level
) {
1600 re
->after
.layout
= 0;
1603 re
->after
.layout
= ALGORITHM_PARITY_N
;
1609 /* We get to RAID5 from RAID5 or RAID6 */
1610 if (re
->level
!= 5 && re
->level
!= 6)
1611 return "Cannot convert to RAID5 from this level";
1613 switch (re
->level
) {
1615 if (info
->new_layout
== UnSet
)
1616 re
->after
.layout
= re
->before
.layout
;
1618 re
->after
.layout
= info
->new_layout
;
1621 if (info
->new_layout
== UnSet
)
1622 info
->new_layout
= re
->before
.layout
;
1624 /* after.layout needs to be raid6 version of new_layout */
1625 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1626 re
->after
.layout
= ALGORITHM_PARITY_N
;
1629 char *ls
= map_num(r5layout
, info
->new_layout
);
1632 /* Current RAID6 layout has a RAID5
1635 strcat(strcpy(layout
, ls
), "-6");
1636 l
= map_name(r6layout
, layout
);
1638 return "Cannot find RAID6 layout to convert to";
1640 /* Current RAID6 has no equivalent.
1641 * If it is already a '-6' layout we
1642 * can leave it unchanged, else we must
1645 ls
= map_num(r6layout
,
1648 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1649 return "Please specify new layout";
1650 l
= info
->new_layout
;
1652 re
->after
.layout
= l
;
1658 /* We must already be at level 6 */
1660 return "Impossible level change";
1661 if (info
->new_layout
== UnSet
)
1662 re
->after
.layout
= info
->array
.layout
;
1664 re
->after
.layout
= info
->new_layout
;
1667 return "Impossible level change requested";
1669 if (info
->delta_disks
== UnSet
)
1670 info
->delta_disks
= delta_parity
;
1672 re
->after
.data_disks
=
1673 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1675 switch (re
->level
) {
1687 /* So we have a restripe operation, we need to calculate the number
1688 * of blocks per reshape operation.
1690 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1691 if (info
->new_chunk
== 0)
1692 info
->new_chunk
= info
->array
.chunk_size
;
1693 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1694 re
->after
.layout
== re
->before
.layout
&&
1695 info
->new_chunk
== info
->array
.chunk_size
) {
1696 /* Nothing to change, can change level immediately. */
1697 re
->level
= info
->new_level
;
1698 re
->backup_blocks
= 0;
1701 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1702 /* chunk and layout changes make no difference */
1703 re
->level
= info
->new_level
;
1704 re
->backup_blocks
= 0;
1708 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1709 get_linux_version() < 2006032)
1710 return "in-place reshape is not safe before 2.6.32 - sorry.";
1712 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1713 get_linux_version() < 2006030)
1714 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1716 re
->backup_blocks
= compute_backup_blocks(
1717 info
->new_chunk
, info
->array
.chunk_size
,
1718 re
->after
.data_disks
, re
->before
.data_disks
);
1719 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1721 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1725 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1728 struct mdinfo
*info
;
1732 if ((st
== NULL
) || (sra
== NULL
))
1735 if (text_version
== NULL
)
1736 text_version
= sra
->text_version
;
1737 subarray
= strchr(text_version
+ 1, '/')+1;
1738 info
= st
->ss
->container_content(st
, subarray
);
1740 unsigned long long current_size
= 0;
1741 unsigned long long new_size
= info
->custom_array_size
/2;
1743 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1744 new_size
> current_size
) {
1745 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1747 dprintf("Error: Cannot set array size");
1750 dprintf("Array size changed");
1752 dprintf_cont(" from %llu to %llu.\n",
1753 current_size
, new_size
);
1757 dprintf("Error: set_array_size(): info pointer in NULL\n");
1762 static int reshape_array(char *container
, int fd
, char *devname
,
1763 struct supertype
*st
, struct mdinfo
*info
,
1764 int force
, struct mddev_dev
*devlist
,
1765 unsigned long long data_offset
,
1766 char *backup_file
, int verbose
, int forked
,
1767 int restart
, int freeze_reshape
);
1768 static int reshape_container(char *container
, char *devname
,
1770 struct supertype
*st
,
1771 struct mdinfo
*info
,
1773 char *backup_file
, int verbose
,
1774 int forked
, int restart
, int freeze_reshape
);
1776 int Grow_reshape(char *devname
, int fd
,
1777 struct mddev_dev
*devlist
,
1778 unsigned long long data_offset
,
1779 struct context
*c
, struct shape
*s
)
1781 /* Make some changes in the shape of an array.
1782 * The kernel must support the change.
1784 * There are three different changes. Each can trigger
1785 * a resync or recovery so we freeze that until we have
1786 * requested everything (if kernel supports freezing - 2.6.30).
1788 * - change size (i.e. component_size)
1790 * - change layout/chunksize/ndisks
1792 * The last can require a reshape. It is different on different
1793 * levels so we need to check the level before actioning it.
1794 * Some times the level change needs to be requested after the
1795 * reshape (e.g. raid6->raid5, raid5->raid0)
1798 struct mdu_array_info_s array
;
1800 struct supertype
*st
;
1801 char *subarray
= NULL
;
1805 char *container
= NULL
;
1808 struct mddev_dev
*dv
;
1814 if (md_get_array_info(fd
, &array
) < 0) {
1815 pr_err("%s is not an active md array - aborting\n",
1819 if (s
->level
!= UnSet
&& s
->chunk
) {
1820 pr_err("Cannot change array level in the same operation as changing chunk size.\n");
1824 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1825 (array
.level
< 4 || array
.level
> 6)) {
1826 pr_err("--grow --data-offset not yet supported\n");
1831 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1832 pr_err("cannot change component size at the same time as other changes.\n"
1833 " Change size first, then check data is intact before making other changes.\n");
1837 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&&
1838 array
.level
> 1 && get_linux_version() < 2006032 &&
1839 !check_env("MDADM_FORCE_FEWER")) {
1840 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1841 " Please use a newer kernel\n");
1845 if (array
.level
> 1 && s
->size
> 1 &&
1846 (unsigned long long) (array
.chunk_size
/ 1024) > s
->size
) {
1847 pr_err("component size must be larger than chunk size.\n");
1851 st
= super_by_fd(fd
, &subarray
);
1853 pr_err("Unable to determine metadata format for %s\n", devname
);
1856 if (s
->raiddisks
> st
->max_devs
) {
1857 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1860 if (s
->level
== 0 && (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) &&
1861 !(array
.state
& (1 << MD_SB_CLUSTERED
)) && !st
->ss
->external
) {
1862 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
1863 if (md_set_array_info(fd
, &array
) != 0) {
1864 pr_err("failed to remove internal bitmap.\n");
1869 /* in the external case we need to check that the requested reshape is
1870 * supported, and perform an initial check that the container holds the
1871 * pre-requisite spare devices (mdmon owns final validation)
1873 if (st
->ss
->external
) {
1877 container
= st
->container_devnm
;
1878 cfd
= open_dev_excl(st
->container_devnm
);
1880 container
= st
->devnm
;
1882 cfd
= open_dev_excl(st
->devnm
);
1886 pr_err("Unable to open container for %s\n", devname
);
1891 retval
= st
->ss
->load_container(st
, cfd
, NULL
);
1894 pr_err("Cannot read superblock for %s\n", devname
);
1900 if (s
->raiddisks
&& subarray
) {
1901 pr_err("--raid-devices operation can be performed on a container only\n");
1907 /* check if operation is supported for metadata handler */
1908 if (st
->ss
->container_content
) {
1909 struct mdinfo
*cc
= NULL
;
1910 struct mdinfo
*content
= NULL
;
1912 cc
= st
->ss
->container_content(st
, subarray
);
1913 for (content
= cc
; content
; content
= content
->next
) {
1914 int allow_reshape
= 1;
1916 /* check if reshape is allowed based on metadata
1917 * indications stored in content.array.status
1919 if (content
->array
.state
&
1920 (1 << MD_SB_BLOCK_VOLUME
))
1922 if (content
->array
.state
&
1923 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE
))
1925 if (!allow_reshape
) {
1926 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1927 devname
, container
);
1932 if (content
->consistency_policy
==
1933 CONSISTENCY_POLICY_PPL
) {
1934 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1939 if (content
->consistency_policy
==
1940 CONSISTENCY_POLICY_BITMAP
) {
1941 pr_err("Operation not supported when write-intent bitmap is enabled\n");
1949 if (mdmon_running(container
))
1950 st
->update_tail
= &st
->updates
;
1954 for (dv
= devlist
; dv
; dv
= dv
->next
)
1956 if (s
->raiddisks
> array
.raid_disks
&&
1957 array
.spare_disks
+ added_disks
<
1958 (s
->raiddisks
- array
.raid_disks
) &&
1960 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1961 " Use --force to over-ride this check.\n",
1962 s
->raiddisks
- array
.raid_disks
,
1963 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1964 array
.spare_disks
+ added_disks
);
1968 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1969 GET_STATE
| GET_VERSION
);
1971 if (st
->ss
->external
&& subarray
== NULL
) {
1972 array
.level
= LEVEL_CONTAINER
;
1973 sra
->array
.level
= LEVEL_CONTAINER
;
1976 pr_err("failed to read sysfs parameters for %s\n",
1980 frozen
= freeze(st
);
1982 /* freeze() already spewed the reason */
1985 } else if (frozen
< 0) {
1986 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1991 /* ========= set size =============== */
1993 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1994 unsigned long long orig_size
= get_component_size(fd
)/2;
1995 unsigned long long min_csize
;
1997 int raid0_takeover
= 0;
2000 orig_size
= (unsigned) array
.size
;
2002 if (orig_size
== 0) {
2003 pr_err("Cannot set device size in this type of array.\n");
2008 if (array
.level
== 0) {
2009 pr_err("Component size change is not supported for RAID0\n");
2014 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
2015 devname
, APPLY_METADATA_CHANGES
,
2021 if (st
->ss
->external
) {
2022 /* metadata can have size limitation
2023 * update size value according to metadata information
2025 struct mdinfo
*sizeinfo
=
2026 st
->ss
->container_content(st
, subarray
);
2028 unsigned long long new_size
=
2029 sizeinfo
->custom_array_size
/2;
2030 int data_disks
= get_data_disks(
2031 sizeinfo
->array
.level
,
2032 sizeinfo
->array
.layout
,
2033 sizeinfo
->array
.raid_disks
);
2034 new_size
/= data_disks
;
2035 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
2036 orig_size
, new_size
,
2037 new_size
* data_disks
);
2039 sysfs_free(sizeinfo
);
2043 /* Update the size of each member device in case
2044 * they have been resized. This will never reduce
2045 * below the current used-size. The "size" attribute
2046 * understands '0' to mean 'max'.
2049 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2050 sysfs_set_num(sra
, mdi
, "size",
2051 s
->size
== MAX_SIZE
? 0 : s
->size
);
2052 if (array
.not_persistent
== 0 &&
2053 array
.major_version
== 0 &&
2054 get_linux_version() < 3001000) {
2055 /* Dangerous to allow size to exceed 2TB */
2056 unsigned long long csize
;
2057 if (sysfs_get_ll(sra
, mdi
, "size",
2059 if (csize
>= 2ULL*1024*1024*1024)
2060 csize
= 2ULL*1024*1024*1024;
2061 if ((min_csize
== 0 ||
2062 (min_csize
> csize
)))
2067 if (min_csize
&& s
->size
> min_csize
) {
2068 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2070 goto size_change_error
;
2072 if (min_csize
&& s
->size
== MAX_SIZE
) {
2073 /* Don't let the kernel choose a size - it will get
2076 pr_err("Limited v0.90 array to 2TB per device\n");
2077 s
->size
= min_csize
;
2079 if (st
->ss
->external
) {
2080 if (sra
->array
.level
== 0) {
2081 rv
= sysfs_set_str(sra
, NULL
, "level", "raid5");
2084 /* get array parameters after takeover
2085 * to change one parameter at time only
2087 rv
= md_get_array_info(fd
, &array
);
2090 /* make sure mdmon is
2091 * aware of the new level */
2092 if (!mdmon_running(st
->container_devnm
))
2093 start_mdmon(st
->container_devnm
);
2094 ping_monitor(container
);
2095 if (mdmon_running(st
->container_devnm
) &&
2096 st
->update_tail
== NULL
)
2097 st
->update_tail
= &st
->updates
;
2100 if (s
->size
== MAX_SIZE
)
2102 array
.size
= s
->size
;
2103 if (s
->size
& ~INT32_MAX
) {
2104 /* got truncated to 32bit, write to
2105 * component_size instead
2108 rv
= sysfs_set_num(sra
, NULL
,
2109 "component_size", s
->size
);
2113 rv
= md_set_array_info(fd
, &array
);
2115 /* manage array size when it is managed externally
2117 if ((rv
== 0) && st
->ss
->external
)
2118 rv
= set_array_size(st
, sra
, sra
->text_version
);
2121 if (raid0_takeover
) {
2122 /* do not recync non-existing parity,
2123 * we will drop it anyway
2125 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2126 /* go back to raid0, drop parity disk
2128 sysfs_set_str(sra
, NULL
, "level", "raid0");
2129 md_get_array_info(fd
, &array
);
2136 /* restore metadata */
2137 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2138 UnSet
, NULL
, devname
,
2139 ROLLBACK_METADATA_CHANGES
,
2142 pr_err("Cannot set device size for %s: %s\n",
2143 devname
, strerror(err
));
2145 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2146 cont_err("Bitmap must be removed before size can be changed\n");
2150 if (s
->assume_clean
) {
2151 /* This will fail on kernels older than 3.0 unless
2152 * a backport has been arranged.
2155 sysfs_set_str(sra
, NULL
, "resync_start",
2157 pr_err("--assume-clean not supported with --grow on this kernel\n");
2159 md_get_array_info(fd
, &array
);
2160 s
->size
= get_component_size(fd
)/2;
2162 s
->size
= array
.size
;
2163 if (c
->verbose
>= 0) {
2164 if (s
->size
== orig_size
)
2165 pr_err("component size of %s unchanged at %lluK\n",
2168 pr_err("component size of %s has been set to %lluK\n",
2172 } else if (array
.level
!= LEVEL_CONTAINER
) {
2173 s
->size
= get_component_size(fd
)/2;
2175 s
->size
= array
.size
;
2178 /* See if there is anything else to do */
2179 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2180 (s
->layout_str
== NULL
) &&
2181 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2182 data_offset
== INVALID_SECTORS
&&
2183 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2184 /* Nothing more to do */
2185 if (!changed
&& c
->verbose
>= 0)
2186 pr_err("%s: no change requested\n", devname
);
2190 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2191 * current implementation assumes that following conditions must be met:
2194 * - near_copies == 2
2196 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2197 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2198 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2201 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2203 dprintf("Array cannot be reshaped\n");
2209 /* Make sure mdmon has seen the device removal
2210 * and updated metadata before we continue with
2214 ping_monitor(container
);
2217 memset(&info
, 0, sizeof(info
));
2219 if (sysfs_init(&info
, fd
, NULL
)) {
2220 pr_err("failed to initialize sysfs.\n");
2224 strcpy(info
.text_version
, sra
->text_version
);
2225 info
.component_size
= s
->size
*2;
2226 info
.new_level
= s
->level
;
2227 info
.new_chunk
= s
->chunk
* 1024;
2228 if (info
.array
.level
== LEVEL_CONTAINER
) {
2229 info
.delta_disks
= UnSet
;
2230 info
.array
.raid_disks
= s
->raiddisks
;
2231 } else if (s
->raiddisks
)
2232 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2234 info
.delta_disks
= UnSet
;
2235 if (s
->layout_str
== NULL
) {
2236 info
.new_layout
= UnSet
;
2237 if (info
.array
.level
== 6 &&
2238 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2239 info
.array
.layout
>= 16) {
2240 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2241 cont_err("during the reshape, please specify --layout=preserve\n");
2242 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2246 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2247 strcmp(s
->layout_str
, "normalize") == 0) {
2248 /* If we have a -6 RAID6 layout, remove the '-6'. */
2249 info
.new_layout
= UnSet
;
2250 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2252 strcpy(l
, map_num_s(r6layout
, info
.array
.layout
));
2253 h
= strrchr(l
, '-');
2254 if (h
&& strcmp(h
, "-6") == 0) {
2256 info
.new_layout
= map_name(r6layout
, l
);
2259 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2263 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2264 /* This means that a non-standard RAID6 layout
2267 * - When reshape a RAID6 (e.g. adding a device)
2268 * which is in a non-standard layout, it is OK
2269 * to preserve that layout.
2270 * - When converting a RAID5 to RAID6, leave it in
2271 * the XXX-6 layout, don't re-layout.
2273 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2274 info
.new_layout
= info
.array
.layout
;
2275 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2277 strcpy(l
, map_num_s(r5layout
, info
.array
.layout
));
2279 info
.new_layout
= map_name(r6layout
, l
);
2281 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2286 int l
= info
.new_level
;
2288 l
= info
.array
.level
;
2291 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2294 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2297 info
.new_layout
= parse_layout_10(s
->layout_str
);
2300 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2303 pr_err("layout not meaningful with this level\n");
2307 if (info
.new_layout
== UnSet
) {
2308 pr_err("layout %s not understood for this level\n",
2315 if (array
.level
== LEVEL_FAULTY
) {
2316 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2317 pr_err("cannot change level of Faulty device\n");
2321 pr_err("cannot set chunksize of Faulty device\n");
2324 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2325 pr_err("cannot set raid_disks of Faulty device\n");
2328 if (s
->layout_str
) {
2329 if (md_get_array_info(fd
, &array
) != 0) {
2330 dprintf("Cannot get array information.\n");
2333 array
.layout
= info
.new_layout
;
2334 if (md_set_array_info(fd
, &array
) != 0) {
2335 pr_err("failed to set new layout\n");
2337 } else if (c
->verbose
>= 0)
2338 printf("layout for %s set to %d\n",
2339 devname
, array
.layout
);
2341 } else if (array
.level
== LEVEL_CONTAINER
) {
2342 /* This change is to be applied to every array in the
2343 * container. This is only needed when the metadata imposes
2344 * restraints of the various arrays in the container.
2345 * Currently we only know that IMSM requires all arrays
2346 * to have the same number of devices so changing the
2347 * number of devices (On-Line Capacity Expansion) must be
2348 * performed at the level of the container
2351 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2352 c
->force
, c
->backup_file
, c
->verbose
,
2356 /* get spare devices from external metadata
2358 if (st
->ss
->external
) {
2359 struct mdinfo
*info2
;
2361 info2
= st
->ss
->container_content(st
, subarray
);
2363 info
.array
.spare_disks
=
2364 info2
->array
.spare_disks
;
2369 /* Impose these changes on a single array. First
2370 * check that the metadata is OK with the change. */
2372 if (reshape_super(st
, 0, info
.new_level
,
2373 info
.new_layout
, info
.new_chunk
,
2374 info
.array
.raid_disks
, info
.delta_disks
,
2375 c
->backup_file
, devname
,
2376 APPLY_METADATA_CHANGES
, c
->verbose
)) {
2381 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2382 devlist
, data_offset
, c
->backup_file
,
2383 c
->verbose
, 0, 0, 0);
2393 /* verify_reshape_position()
2394 * Function checks if reshape position in metadata is not farther
2395 * than position in md.
2397 * 0 : not valid sysfs entry
2398 * it can be caused by not started reshape, it should be started
2399 * by reshape array or raid0 array is before takeover
2400 * -1 : error, reshape position is obviously wrong
2401 * 1 : success, reshape progress correct or updated
2403 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2409 /* read sync_max, failure can mean raid0 array */
2410 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2414 unsigned long long position
= strtoull(buf
, &ep
, 0);
2416 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2417 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2418 position
*= get_data_disks(level
,
2420 info
->array
.raid_disks
);
2421 if (info
->reshape_progress
< position
) {
2422 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2423 info
->reshape_progress
, position
);
2424 info
->reshape_progress
= position
;
2426 } else if (info
->reshape_progress
> position
) {
2427 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2428 position
, info
->reshape_progress
);
2431 dprintf("Reshape position in md and metadata are the same;");
2435 } else if (rv
== 0) {
2436 /* for valid sysfs entry, 0-length content
2437 * should be indicated as error
2445 static unsigned long long choose_offset(unsigned long long lo
,
2446 unsigned long long hi
,
2447 unsigned long long min
,
2448 unsigned long long max
)
2450 /* Choose a new offset between hi and lo.
2451 * It must be between min and max, but
2452 * we would prefer something near the middle of hi/lo, and also
2453 * prefer to be aligned to a big power of 2.
2455 * So we start with the middle, then for each bit,
2456 * starting at '1' and increasing, if it is set, we either
2457 * add it or subtract it if possible, preferring the option
2458 * which is furthest from the boundary.
2460 * We stop once we get a 1MB alignment. As units are in sectors,
2461 * 1MB = 2*1024 sectors.
2463 unsigned long long choice
= (lo
+ hi
) / 2;
2464 unsigned long long bit
= 1;
2466 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2467 unsigned long long bigger
, smaller
;
2468 if (! (bit
& choice
))
2470 bigger
= choice
+ bit
;
2471 smaller
= choice
- bit
;
2472 if (bigger
> max
&& smaller
< min
)
2476 else if (smaller
< min
)
2478 else if (hi
- bigger
> smaller
- lo
)
2486 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2487 char *devname
, int delta_disks
,
2488 unsigned long long data_offset
,
2489 unsigned long long min
,
2495 unsigned long long before
, after
;
2497 /* Need to find min space before and after so same is used
2500 before
= UINT64_MAX
;
2502 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2506 struct supertype
*st2
;
2507 struct mdinfo info2
;
2509 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2511 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2512 dfd
= dev_open(dn
, O_RDONLY
);
2514 pr_err("%s: cannot open component %s\n",
2515 devname
, dn
? dn
: "-unknown-");
2518 st2
= dup_super(st
);
2519 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2523 pr_err("%s: cannot get superblock from %s\n",
2527 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2528 st2
->ss
->free_super(st2
);
2530 if (info2
.space_before
== 0 &&
2531 info2
.space_after
== 0) {
2532 /* Metadata doesn't support data_offset changes */
2534 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2538 if (before
> info2
.space_before
)
2539 before
= info2
.space_before
;
2540 if (after
> info2
.space_after
)
2541 after
= info2
.space_after
;
2543 if (data_offset
!= INVALID_SECTORS
) {
2545 if (info2
.data_offset
== data_offset
) {
2546 pr_err("%s: already has that data_offset\n",
2550 if (data_offset
< info2
.data_offset
)
2554 } else if ((data_offset
<= info2
.data_offset
&&
2556 (data_offset
>= info2
.data_offset
&&
2558 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2564 if (before
== UINT64_MAX
)
2565 /* impossible really, there must be no devices */
2568 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2569 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2570 unsigned long long new_data_offset
;
2572 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2574 if (delta_disks
< 0) {
2575 /* Don't need any space as array is shrinking
2576 * just move data_offset up by min
2578 if (data_offset
== INVALID_SECTORS
)
2579 new_data_offset
= sd
->data_offset
+ min
;
2581 if (data_offset
< sd
->data_offset
+ min
) {
2582 pr_err("--data-offset too small for %s\n",
2586 new_data_offset
= data_offset
;
2588 } else if (delta_disks
> 0) {
2589 /* need space before */
2593 pr_err("Insufficient head-space for reshape on %s\n",
2597 if (data_offset
== INVALID_SECTORS
)
2598 new_data_offset
= sd
->data_offset
- min
;
2600 if (data_offset
> sd
->data_offset
- min
) {
2601 pr_err("--data-offset too large for %s\n",
2605 new_data_offset
= data_offset
;
2609 /* can move up or down. If 'data_offset'
2610 * was set we would have already decided,
2611 * so just choose direction with most space.
2618 sysfs_set_str(sra
, NULL
, "reshape_direction",
2619 dir
== 1 ? "backwards" : "forwards");
2621 /* Increase data offset */
2625 pr_err("Insufficient tail-space for reshape on %s\n",
2629 if (data_offset
!= INVALID_SECTORS
&&
2630 data_offset
< sd
->data_offset
+ min
) {
2631 pr_err("--data-offset too small on %s\n",
2635 if (data_offset
!= INVALID_SECTORS
)
2636 new_data_offset
= data_offset
;
2638 new_data_offset
= choose_offset(sd
->data_offset
,
2639 sd
->data_offset
+ after
,
2640 sd
->data_offset
+ min
,
2641 sd
->data_offset
+ after
);
2643 /* Decrease data offset */
2647 pr_err("insufficient head-room on %s\n",
2651 if (data_offset
!= INVALID_SECTORS
&&
2652 data_offset
> sd
->data_offset
- min
) {
2653 pr_err("--data-offset too large on %s\n",
2657 if (data_offset
!= INVALID_SECTORS
)
2658 new_data_offset
= data_offset
;
2660 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2662 sd
->data_offset
- before
,
2663 sd
->data_offset
- min
);
2666 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2667 if (err
< 0 && errno
== E2BIG
) {
2668 /* try again after increasing data size to max */
2669 err
= sysfs_set_num(sra
, sd
, "size", 0);
2670 if (err
< 0 && errno
== EINVAL
&&
2671 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2672 /* some kernels have a bug where you cannot
2673 * use '0' on spare devices. */
2674 sysfs_set_num(sra
, sd
, "size",
2675 (sra
->component_size
+ after
)/2);
2677 err
= sysfs_set_num(sra
, sd
, "new_offset",
2681 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2682 pr_err("data-offset is too big for %s\n", dn
);
2685 if (sd
== sra
->devs
&&
2686 (errno
== ENOENT
|| errno
== E2BIG
))
2687 /* Early kernel, no 'new_offset' file,
2688 * or kernel doesn't like us.
2689 * For RAID5/6 this is not fatal
2692 pr_err("Cannot set new_offset for %s\n", dn
);
2700 /* Just use a backup file */
2704 static int raid10_reshape(char *container
, int fd
, char *devname
,
2705 struct supertype
*st
, struct mdinfo
*info
,
2706 struct reshape
*reshape
,
2707 unsigned long long data_offset
,
2708 int force
, int verbose
)
2710 /* Changing raid_disks, layout, chunksize or possibly
2711 * just data_offset for a RAID10.
2712 * We must always change data_offset. We change by at least
2713 * ->min_offset_change which is the largest of the old and new
2715 * If raid_disks is increasing, then data_offset must decrease
2716 * by at least this copy size.
2717 * If raid_disks is unchanged, data_offset must increase or
2718 * decrease by at least min_offset_change but preferably by much more.
2719 * We choose half of the available space.
2720 * If raid_disks is decreasing, data_offset must increase by
2721 * at least min_offset_change. To allow of this, component_size
2722 * must be decreased by the same amount.
2724 * So we calculate the required minimum and direction, possibly
2725 * reduce the component_size, then iterate through the devices
2726 * and set the new_data_offset.
2727 * If that all works, we set chunk_size, layout, raid_disks, and start
2731 unsigned long long min
;
2734 sra
= sysfs_read(fd
, NULL
,
2735 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2738 pr_err("%s: Cannot get array details from sysfs\n", devname
);
2741 min
= reshape
->min_offset_change
;
2743 if (info
->delta_disks
)
2744 sysfs_set_str(sra
, NULL
, "reshape_direction",
2745 info
->delta_disks
< 0 ? "backwards" : "forwards");
2746 if (info
->delta_disks
< 0 && info
->space_after
< min
) {
2747 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2748 (sra
->component_size
- min
)/2);
2750 pr_err("cannot reduce component size\n");
2754 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
,
2755 data_offset
, min
, 0);
2757 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2758 cont_err("supported on this kernel\n");
2764 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2766 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2767 reshape
->after
.layout
) < 0)
2770 sysfs_set_num(sra
, NULL
, "raid_disks",
2771 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2773 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2776 pr_err("Cannot set array shape for %s\n",
2779 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2780 cont_err(" Bitmap must be removed before shape can be changed\n");
2790 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2792 struct mdinfo
*sra
, *sd
;
2793 /* Initialisation to silence compiler warning */
2794 unsigned long long min_space_before
= 0, min_space_after
= 0;
2797 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2800 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2803 struct supertype
*st2
;
2804 struct mdinfo info2
;
2806 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2808 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2809 dfd
= dev_open(dn
, O_RDONLY
);
2812 st2
= dup_super(st
);
2813 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2819 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2820 st2
->ss
->free_super(st2
);
2823 min_space_before
> info2
.space_before
)
2824 min_space_before
= info2
.space_before
;
2826 min_space_after
> info2
.space_after
)
2827 min_space_after
= info2
.space_after
;
2830 if (sd
== NULL
&& !first
) {
2831 info
->space_after
= min_space_after
;
2832 info
->space_before
= min_space_before
;
2837 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2838 struct mdinfo
*info
,
2839 int disks
, unsigned long long blocks
)
2841 /* Check that the internal stripe cache is
2842 * large enough, or it won't work.
2843 * It must hold at least 4 stripes of the larger
2846 unsigned long cache
;
2847 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2848 cache
*= 4; /* 4 stripes minimum */
2849 cache
/= 512; /* convert to sectors */
2850 /* make sure there is room for 'blocks' with a bit to spare */
2851 if (cache
< 16 + blocks
/ disks
)
2852 cache
= 16 + blocks
/ disks
;
2853 cache
/= (4096/512); /* Convert from sectors to pages */
2855 if (sra
->cache_size
< cache
)
2856 subarray_set_num(container
, sra
, "stripe_cache_size",
2860 static int impose_reshape(struct mdinfo
*sra
,
2861 struct mdinfo
*info
,
2862 struct supertype
*st
,
2865 char *devname
, char *container
,
2866 struct reshape
*reshape
)
2868 struct mdu_array_info_s array
;
2870 sra
->new_chunk
= info
->new_chunk
;
2873 /* for external metadata checkpoint saved by mdmon can be lost
2874 * or missed /due to e.g. crash/. Check if md is not during
2875 * restart farther than metadata points to.
2876 * If so, this means metadata information is obsolete.
2878 if (st
->ss
->external
)
2879 verify_reshape_position(info
, reshape
->level
);
2880 sra
->reshape_progress
= info
->reshape_progress
;
2882 sra
->reshape_progress
= 0;
2883 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2884 /* start from the end of the new array */
2885 sra
->reshape_progress
= (sra
->component_size
2886 * reshape
->after
.data_disks
);
2889 md_get_array_info(fd
, &array
);
2890 if (info
->array
.chunk_size
== info
->new_chunk
&&
2891 reshape
->before
.layout
== reshape
->after
.layout
&&
2892 st
->ss
->external
== 0) {
2893 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2894 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2895 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2898 pr_err("Cannot set device shape for %s: %s\n",
2899 devname
, strerror(errno
));
2902 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2903 cont_err("Bitmap must be removed before shape can be changed\n");
2907 } else if (!restart
) {
2908 /* set them all just in case some old 'new_*' value
2909 * persists from some earlier problem.
2912 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2914 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2915 reshape
->after
.layout
) < 0)
2917 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2918 reshape
->after
.data_disks
+
2919 reshape
->parity
) < 0)
2922 pr_err("Cannot set device shape for %s\n", devname
);
2925 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2926 cont_err("Bitmap must be removed before shape can be changed\n");
2935 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2938 struct mdu_array_info_s array
;
2941 if (sysfs_init(&info
, fd
, NULL
)) {
2942 pr_err("failed to initialize sysfs.\n");
2946 md_get_array_info(fd
, &array
);
2947 if (level
== 0 && is_level456(array
.level
)) {
2948 /* To convert to RAID0 we need to fail and
2949 * remove any non-data devices. */
2952 int data_disks
= array
.raid_disks
- 1;
2953 if (array
.level
== 6)
2955 if (array
.level
== 5 && array
.layout
!= ALGORITHM_PARITY_N
)
2957 if (array
.level
== 6 && array
.layout
!= ALGORITHM_PARITY_N_6
)
2959 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2960 /* First remove any spares so no recovery starts */
2961 for (d
= 0, found
= 0;
2962 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2963 mdu_disk_info_t disk
;
2965 if (md_get_disk_info(fd
, &disk
) < 0)
2967 if (disk
.major
== 0 && disk
.minor
== 0)
2970 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2971 disk
.raid_disk
< data_disks
)
2974 ioctl(fd
, HOT_REMOVE_DISK
,
2975 makedev(disk
.major
, disk
.minor
));
2977 /* Now fail anything left */
2978 md_get_array_info(fd
, &array
);
2979 for (d
= 0, found
= 0;
2980 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2981 mdu_disk_info_t disk
;
2983 if (md_get_disk_info(fd
, &disk
) < 0)
2985 if (disk
.major
== 0 && disk
.minor
== 0)
2988 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2989 disk
.raid_disk
< data_disks
)
2992 ioctl(fd
, SET_DISK_FAULTY
,
2993 makedev(disk
.major
, disk
.minor
));
2994 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2997 c
= map_num(pers
, level
);
2999 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
3002 pr_err("%s: could not set level to %s\n",
3005 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
3006 cont_err("Bitmap must be removed before level can be changed\n");
3010 pr_err("level of %s changed to %s\n", devname
, c
);
3016 static void catch_term(int sig
)
3021 static int reshape_array(char *container
, int fd
, char *devname
,
3022 struct supertype
*st
, struct mdinfo
*info
,
3023 int force
, struct mddev_dev
*devlist
,
3024 unsigned long long data_offset
,
3025 char *backup_file
, int verbose
, int forked
,
3026 int restart
, int freeze_reshape
)
3028 struct reshape reshape
;
3031 int orig_level
= UnSet
;
3035 struct mdu_array_info_s array
;
3038 struct mddev_dev
*dv
;
3042 unsigned long long *offsets
= NULL
;
3046 unsigned long blocks
;
3047 unsigned long long array_size
;
3049 struct mdinfo
*sra
= NULL
;
3052 /* when reshaping a RAID0, the component_size might be zero.
3053 * So try to fix that up.
3055 if (md_get_array_info(fd
, &array
) != 0) {
3056 dprintf("Cannot get array information.\n");
3059 if (array
.level
== 0 && info
->component_size
== 0) {
3060 get_dev_size(fd
, NULL
, &array_size
);
3061 info
->component_size
= array_size
/ array
.raid_disks
;
3064 if (array
.level
== 10)
3065 /* Need space_after info */
3066 get_space_after(fd
, st
, info
);
3068 if (info
->reshape_active
) {
3069 int new_level
= info
->new_level
;
3070 info
->new_level
= UnSet
;
3071 if (info
->delta_disks
> 0)
3072 info
->array
.raid_disks
-= info
->delta_disks
;
3073 msg
= analyse_change(devname
, info
, &reshape
);
3074 info
->new_level
= new_level
;
3075 if (info
->delta_disks
> 0)
3076 info
->array
.raid_disks
+= info
->delta_disks
;
3078 /* Make sure the array isn't read-only */
3079 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3081 msg
= analyse_change(devname
, info
, &reshape
);
3083 /* if msg == "", error has already been printed */
3085 pr_err("%s\n", msg
);
3088 if (restart
&& (reshape
.level
!= info
->array
.level
||
3089 reshape
.before
.layout
!= info
->array
.layout
||
3090 reshape
.before
.data_disks
+ reshape
.parity
!=
3091 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3092 pr_err("reshape info is not in native format - cannot continue.\n");
3096 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3097 !((sysfs_get_str(info
, NULL
, "sync_action",
3098 buf
, sizeof(buf
)) > 0) &&
3099 (strncmp(buf
, "reshape", 7) == 0))) {
3100 /* When reshape is restarted from '0', very begin of array
3101 * it is possible that for external metadata reshape and array
3102 * configuration doesn't happen.
3103 * Check if md has the same opinion, and reshape is restarted
3104 * from 0. If so, this is regular reshape start after reshape
3105 * switch in metadata to next array only.
3107 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3108 (info
->reshape_progress
== 0))
3113 * reshape already started. just skip to monitoring
3116 if (reshape
.backup_blocks
== 0)
3118 if (restart
& RESHAPE_NO_BACKUP
)
3121 /* Need 'sra' down at 'started:' */
3122 sra
= sysfs_read(fd
, NULL
,
3123 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
3124 GET_CHUNK
|GET_CACHE
);
3126 pr_err("%s: Cannot get array details from sysfs\n",
3132 backup_file
= locate_backup(sra
->sys_name
);
3136 /* The container is frozen but the array may not be.
3137 * So freeze the array so spares don't get put to the wrong use
3138 * FIXME there should probably be a cleaner separation between
3139 * freeze_array and freeze_container.
3141 sysfs_freeze_array(info
);
3142 /* Check we have enough spares to not be degraded */
3144 for (dv
= devlist
; dv
; dv
=dv
->next
)
3146 spares_needed
= max(reshape
.before
.data_disks
,
3147 reshape
.after
.data_disks
) +
3148 reshape
.parity
- array
.raid_disks
;
3150 if (!force
&& info
->new_level
> 1 && info
->array
.level
> 1 &&
3151 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3152 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3153 " Use --force to over-ride this check.\n",
3155 spares_needed
== 1 ? "" : "s",
3156 info
->array
.spare_disks
+ added_disks
);
3159 /* Check we have enough spares to not fail */
3160 spares_needed
= max(reshape
.before
.data_disks
,
3161 reshape
.after
.data_disks
)
3163 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3164 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3165 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3166 spares_needed
, spares_needed
== 1 ? "" : "s",
3167 info
->array
.spare_disks
+ added_disks
);
3171 if (reshape
.level
!= array
.level
) {
3172 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3175 info
->new_layout
= UnSet
; /* after level change,
3176 * layout is meaningless */
3177 orig_level
= array
.level
;
3178 sysfs_freeze_array(info
);
3180 if (reshape
.level
> 0 && st
->ss
->external
) {
3181 /* make sure mdmon is aware of the new level */
3182 if (mdmon_running(container
))
3183 flush_mdmon(container
);
3185 if (!mdmon_running(container
))
3186 start_mdmon(container
);
3187 ping_monitor(container
);
3188 if (mdmon_running(container
) && st
->update_tail
== NULL
)
3189 st
->update_tail
= &st
->updates
;
3192 /* ->reshape_super might have chosen some spares from the
3193 * container that it wants to be part of the new array.
3194 * We can collect them with ->container_content and give
3195 * them to the kernel.
3197 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3198 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3199 struct mdinfo
*info2
=
3200 st
->ss
->container_content(st
, subarray
);
3204 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3205 pr_err("unable to initialize sysfs for %s\n",
3210 /* When increasing number of devices, we need to set
3211 * new raid_disks before adding these, or they might
3214 if (reshape
.backup_blocks
&&
3215 reshape
.after
.data_disks
>
3216 reshape
.before
.data_disks
)
3217 subarray_set_num(container
, info2
, "raid_disks",
3218 reshape
.after
.data_disks
+
3220 for (d
= info2
->devs
; d
; d
= d
->next
) {
3221 if (d
->disk
.state
== 0 &&
3222 d
->disk
.raid_disk
>= 0) {
3223 /* This is a spare that wants to
3224 * be part of the array.
3226 add_disk(fd
, st
, info2
, d
);
3232 /* We might have been given some devices to add to the
3233 * array. Now that the array has been changed to the right
3234 * level and frozen, we can safely add them.
3237 if (Manage_subdevs(devname
, fd
, devlist
, verbose
, 0, NULL
, 0))
3241 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3242 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3243 if (reshape
.backup_blocks
== 0) {
3244 /* No restriping needed, but we might need to impose
3245 * some more changes: layout, raid_disks, chunk_size
3247 /* read current array info */
3248 if (md_get_array_info(fd
, &array
) != 0) {
3249 dprintf("Cannot get array information.\n");
3252 /* compare current array info with new values and if
3253 * it is different update them to new */
3254 if (info
->new_layout
!= UnSet
&&
3255 info
->new_layout
!= array
.layout
) {
3256 array
.layout
= info
->new_layout
;
3257 if (md_set_array_info(fd
, &array
) != 0) {
3258 pr_err("failed to set new layout\n");
3260 } else if (verbose
>= 0)
3261 printf("layout for %s set to %d\n",
3262 devname
, array
.layout
);
3264 if (info
->delta_disks
!= UnSet
&& info
->delta_disks
!= 0 &&
3266 (info
->array
.raid_disks
+ info
->delta_disks
)) {
3267 array
.raid_disks
+= info
->delta_disks
;
3268 if (md_set_array_info(fd
, &array
) != 0) {
3269 pr_err("failed to set raid disks\n");
3271 } else if (verbose
>= 0) {
3272 printf("raid_disks for %s set to %d\n",
3273 devname
, array
.raid_disks
);
3276 if (info
->new_chunk
!= 0 &&
3277 info
->new_chunk
!= array
.chunk_size
) {
3278 if (sysfs_set_num(info
, NULL
,
3279 "chunk_size", info
->new_chunk
) != 0) {
3280 pr_err("failed to set chunk size\n");
3282 } else if (verbose
>= 0)
3283 printf("chunk size for %s set to %d\n",
3284 devname
, info
->new_chunk
);
3291 * There are three possibilities.
3292 * 1/ The array will shrink.
3293 * We need to ensure the reshape will pause before reaching
3294 * the 'critical section'. We also need to fork and wait for
3295 * that to happen. When it does we
3296 * suspend/backup/complete/unfreeze
3298 * 2/ The array will not change size.
3299 * This requires that we keep a backup of a sliding window
3300 * so that we can restore data after a crash. So we need
3301 * to fork and monitor progress.
3302 * In future we will allow the data_offset to change, so
3303 * a sliding backup becomes unnecessary.
3305 * 3/ The array will grow. This is relatively easy.
3306 * However the kernel's restripe routines will cheerfully
3307 * overwrite some early data before it is safe. So we
3308 * need to make a backup of the early parts of the array
3309 * and be ready to restore it if rebuild aborts very early.
3310 * For externally managed metadata, we still need a forked
3311 * child to monitor the reshape and suspend IO over the region
3312 * that is being reshaped.
3314 * We backup data by writing it to one spare, or to a
3315 * file which was given on command line.
3317 * In each case, we first make sure that storage is available
3318 * for the required backup.
3320 * - request the shape change.
3321 * - fork to handle backup etc.
3323 /* Check that we can hold all the data */
3324 get_dev_size(fd
, NULL
, &array_size
);
3325 if (reshape
.new_size
< (array_size
/512)) {
3326 pr_err("this change will reduce the size of the array.\n"
3327 " use --grow --array-size first to truncate array.\n"
3328 " e.g. mdadm --grow %s --array-size %llu\n",
3329 devname
, reshape
.new_size
/2);
3333 if (array
.level
== 10) {
3334 /* Reshaping RAID10 does not require any data backup by
3335 * user-space. Instead it requires that the data_offset
3336 * is changed to avoid the need for backup.
3337 * So this is handled very separately
3340 /* Nothing to do. */
3342 return raid10_reshape(container
, fd
, devname
, st
, info
,
3343 &reshape
, data_offset
, force
, verbose
);
3345 sra
= sysfs_read(fd
, NULL
,
3346 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3349 pr_err("%s: Cannot get array details from sysfs\n",
3355 switch(set_new_data_offset(sra
, st
, devname
,
3356 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3358 reshape
.min_offset_change
, 1)) {
3362 /* Updated data_offset, so it's easy now */
3363 update_cache_size(container
, sra
, info
,
3364 min(reshape
.before
.data_disks
,
3365 reshape
.after
.data_disks
),
3366 reshape
.backup_blocks
);
3368 /* Right, everything seems fine. Let's kick things off.
3372 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3373 devname
, container
, &reshape
) < 0)
3375 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3377 if (errno
!= EINVAL
) {
3378 pr_err("Failed to initiate reshape!\n");
3381 /* revert data_offset and try the old way */
3382 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3383 sysfs_set_num(sra
, sd
, "new_offset",
3385 sysfs_set_str(sra
, NULL
, "reshape_direction",
3390 if (info
->new_level
== reshape
.level
)
3392 /* need to adjust level when reshape completes */
3394 case -1: /* ignore error, but don't wait */
3396 default: /* parent */
3405 fd
= open_dev(sra
->sys_name
);
3407 impose_level(fd
, info
->new_level
, devname
, verbose
);
3409 case 1: /* Couldn't set data_offset, try the old way */
3410 if (data_offset
!= INVALID_SECTORS
) {
3411 pr_err("Cannot update data_offset on this array\n");
3418 /* Decide how many blocks (sectors) for a reshape
3419 * unit. The number we have so far is just a minimum
3421 blocks
= reshape
.backup_blocks
;
3422 if (reshape
.before
.data_disks
==
3423 reshape
.after
.data_disks
) {
3424 /* Make 'blocks' bigger for better throughput, but
3425 * not so big that we reject it below.
3426 * Try for 16 megabytes
3428 while (blocks
* 32 < sra
->component_size
&& blocks
< 16*1024*2)
3431 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3433 if (blocks
>= sra
->component_size
/2) {
3434 pr_err("%s: Something wrong - reshape aborted\n", devname
);
3438 /* Now we need to open all these devices so we can read/write.
3440 nrdisks
= max(reshape
.before
.data_disks
,
3441 reshape
.after
.data_disks
) + reshape
.parity
3442 + sra
->array
.spare_disks
;
3443 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3444 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3446 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3447 d
= reshape_prepare_fdlist(devname
, sra
, odisks
, nrdisks
, blocks
,
3448 backup_file
, fdlist
, offsets
);
3452 if ((st
->ss
->manage_reshape
== NULL
) ||
3453 (st
->ss
->recover_backup
== NULL
)) {
3454 if (backup_file
== NULL
) {
3455 if (reshape
.after
.data_disks
<=
3456 reshape
.before
.data_disks
) {
3457 pr_err("%s: Cannot grow - need backup-file\n",
3459 pr_err(" Please provide one with \"--backup=...\"\n");
3461 } else if (d
== odisks
) {
3462 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3466 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3468 fdlist
+d
, offsets
+d
,
3469 sra
->sys_name
, restart
)) {
3476 update_cache_size(container
, sra
, info
,
3477 min(reshape
.before
.data_disks
,
3478 reshape
.after
.data_disks
), blocks
);
3480 /* Right, everything seems fine. Let's kick things off.
3481 * If only changing raid_disks, use ioctl, else use
3486 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3487 devname
, container
, &reshape
) < 0)
3490 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3491 reshape
.after
.data_disks
, st
);
3493 pr_err("Cannot %s reshape for %s\n",
3494 restart
? "continue" : "start", devname
);
3498 sysfs_set_str(sra
, NULL
, "array_state", "active");
3499 if (freeze_reshape
) {
3503 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3504 sra
->reshape_progress
);
3509 if (continue_via_systemd(container
?: sra
->sys_name
,
3517 /* Now we just need to kick off the reshape and watch, while
3518 * handling backups of the data...
3519 * This is all done by a forked background process.
3521 switch(forked
? 0 : fork()) {
3523 pr_err("Cannot run child to monitor reshape: %s\n",
3537 /* Close unused file descriptor in the forked process */
3540 /* If another array on the same devices is busy, the
3541 * reshape will wait for them. This would mean that
3542 * the first section that we suspend will stay suspended
3543 * for a long time. So check on that possibility
3544 * by looking for "DELAYED" in /proc/mdstat, and if found,
3548 struct mdstat_ent
*mds
, *m
;
3550 mds
= mdstat_read(1, 0);
3551 for (m
= mds
; m
; m
= m
->next
)
3552 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3553 if (m
->resync
&& m
->percent
== RESYNC_DELAYED
)
3556 /* Haven't started the reshape thread
3563 if (delayed
== 1 && get_linux_version() < 3007000) {
3564 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3565 " You might experience problems until other reshapes complete.\n");
3569 mdstat_wait(30 - (delayed
-1) * 25);
3572 if (check_env("MDADM_GROW_VERIFY"))
3573 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3576 mlockall(MCL_FUTURE
);
3578 if (signal_s(SIGTERM
, catch_term
) == SIG_ERR
)
3581 if (st
->ss
->external
) {
3582 /* metadata handler takes it from here */
3583 done
= st
->ss
->manage_reshape(
3584 fd
, sra
, &reshape
, st
, blocks
,
3585 fdlist
, offsets
, d
- odisks
, fdlist
+ odisks
,
3588 done
= child_monitor(
3589 fd
, sra
, &reshape
, st
, blocks
, fdlist
, offsets
,
3590 d
- odisks
, fdlist
+ odisks
, offsets
+ odisks
);
3595 if (backup_file
&& done
) {
3597 bul
= make_backup(sra
->sys_name
);
3600 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3608 unlink(backup_file
);
3615 if (!st
->ss
->external
&&
3616 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3617 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3619 /* no need to wait for the reshape to finish as
3620 * there is nothing more to do.
3627 if (st
->ss
->external
) {
3628 /* Re-load the metadata as much could have changed */
3629 int cfd
= open_dev(st
->container_devnm
);
3631 flush_mdmon(container
);
3632 st
->ss
->free_super(st
);
3633 st
->ss
->load_container(st
, cfd
, container
);
3638 /* set new array size if required customer_array_size is used
3641 if (reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3642 info
->custom_array_size
)
3643 set_array_size(st
, info
, info
->text_version
);
3645 if (info
->new_level
!= reshape
.level
) {
3647 fd
= open(devname
, O_RDONLY
);
3648 impose_level(fd
, info
->new_level
, devname
, verbose
);
3650 if (info
->new_level
== 0)
3651 st
->update_tail
= NULL
;
3663 if (orig_level
!= UnSet
&& sra
) {
3664 c
= map_num(pers
, orig_level
);
3665 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3666 pr_err("aborting level change\n");
3674 /* mdfd handle is passed to be closed in child process (after fork).
3676 int reshape_container(char *container
, char *devname
,
3678 struct supertype
*st
,
3679 struct mdinfo
*info
,
3681 char *backup_file
, int verbose
,
3682 int forked
, int restart
, int freeze_reshape
)
3684 struct mdinfo
*cc
= NULL
;
3686 char last_devnm
[32] = "";
3688 /* component_size is not meaningful for a container,
3689 * so pass '0' meaning 'no change'
3692 reshape_super(st
, 0, info
->new_level
,
3693 info
->new_layout
, info
->new_chunk
,
3694 info
->array
.raid_disks
, info
->delta_disks
,
3695 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3703 /* ping monitor to be sure that update is on disk
3705 ping_monitor(container
);
3707 if (!forked
&& !freeze_reshape
)
3708 if (continue_via_systemd(container
, GROW_SERVICE
))
3711 switch (forked
? 0 : fork()) {
3712 case -1: /* error */
3713 perror("Cannot fork to complete reshape\n");
3716 default: /* parent */
3717 if (!freeze_reshape
)
3718 printf("%s: multi-array reshape continues in background\n", Name
);
3726 /* close unused handle in child process
3732 /* For each member array with reshape_active,
3733 * we need to perform the reshape.
3734 * We pick the first array that needs reshaping and
3735 * reshape it. reshape_array() will re-read the metadata
3736 * so the next time through a different array should be
3737 * ready for reshape.
3738 * It is possible that the 'different' array will not
3739 * be assembled yet. In that case we simple exit.
3740 * When it is assembled, the mdadm which assembles it
3741 * will take over the reshape.
3743 struct mdinfo
*content
;
3745 struct mdstat_ent
*mdstat
;
3751 cc
= st
->ss
->container_content(st
, NULL
);
3753 for (content
= cc
; content
; content
= content
->next
) {
3755 if (!content
->reshape_active
)
3758 subarray
= strchr(content
->text_version
+1, '/')+1;
3759 mdstat
= mdstat_by_subdev(subarray
, container
);
3762 if (mdstat
->active
== 0) {
3763 pr_err("Skipping inactive array %s.\n",
3765 free_mdstat(mdstat
);
3774 devid
= devnm2devid(mdstat
->devnm
);
3775 adev
= map_dev(major(devid
), minor(devid
), 0);
3777 adev
= content
->text_version
;
3779 fd
= open_dev(mdstat
->devnm
);
3781 pr_err("Device %s cannot be opened for reshape.\n",
3786 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3787 /* Do not allow for multiple reshape_array() calls for
3789 * It can happen when reshape_array() returns without
3790 * error, when reshape is not finished (wrong reshape
3791 * starting/continuation conditions). Mdmon doesn't
3792 * switch to next array in container and reentry
3793 * conditions for the same array occur.
3794 * This is possibly interim until the behaviour of
3795 * reshape_array is resolved().
3797 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3801 strcpy(last_devnm
, mdstat
->devnm
);
3803 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3804 pr_err("Unable to initialize sysfs for %s\n",
3810 if (mdmon_running(container
))
3811 flush_mdmon(container
);
3813 rv
= reshape_array(container
, fd
, adev
, st
,
3814 content
, force
, NULL
, INVALID_SECTORS
,
3815 backup_file
, verbose
, 1, restart
,
3819 if (freeze_reshape
) {
3828 if (mdmon_running(container
))
3829 flush_mdmon(container
);
3838 * We run a child process in the background which performs the following
3840 * - wait for resync to reach a certain point
3841 * - suspend io to the following section
3842 * - backup that section
3843 * - allow resync to proceed further
3845 * - discard the backup.
3847 * When are combined in slightly different ways in the three cases.
3849 * - suspend/backup/allow/wait/resume/discard
3851 * - allow/wait/suspend/backup/allow/wait/resume/discard
3853 * - wait/resume/discard/suspend/backup/allow
3855 * suspend/backup/allow always come together
3856 * wait/resume/discard do too.
3857 * For the same-size case we have two backups to improve flow.
3861 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3862 unsigned long long backup_point
,
3863 unsigned long long wait_point
,
3864 unsigned long long *suspend_point
,
3865 unsigned long long *reshape_completed
, int *frozen
)
3867 /* This function is called repeatedly by the reshape manager.
3868 * It determines how much progress can safely be made and allows
3870 * - 'info' identifies the array and particularly records in
3871 * ->reshape_progress the metadata's knowledge of progress
3872 * This is a sector offset from the start of the array
3873 * of the next array block to be relocated. This number
3874 * may increase from 0 or decrease from array_size, depending
3875 * on the type of reshape that is happening.
3876 * Note that in contrast, 'sync_completed' is a block count of the
3877 * reshape so far. It gives the distance between the start point
3878 * (head or tail of device) and the next place that data will be
3879 * written. It always increases.
3880 * - 'reshape' is the structure created by analyse_change
3881 * - 'backup_point' shows how much the metadata manager has backed-up
3882 * data. For reshapes with increasing progress, it is the next address
3883 * to be backed up, previous addresses have been backed-up. For
3884 * decreasing progress, it is the earliest address that has been
3885 * backed up - later address are also backed up.
3886 * So addresses between reshape_progress and backup_point are
3887 * backed up providing those are in the 'correct' order.
3888 * - 'wait_point' is an array address. When reshape_completed
3889 * passes this point, progress_reshape should return. It might
3890 * return earlier if it determines that ->reshape_progress needs
3891 * to be updated or further backup is needed.
3892 * - suspend_point is maintained by progress_reshape and the caller
3893 * should not touch it except to initialise to zero.
3894 * It is an array address and it only increases in 2.6.37 and earlier.
3895 * This makes it difficult to handle reducing reshapes with
3896 * external metadata.
3897 * However: it is similar to backup_point in that it records the
3898 * other end of a suspended region from reshape_progress.
3899 * it is moved to extend the region that is safe to backup and/or
3901 * - reshape_completed is read from sysfs and returned. The caller
3902 * should copy this into ->reshape_progress when it has reason to
3903 * believe that the metadata knows this, and any backup outside this
3907 * 1 if more data from backup_point - but only as far as suspend_point,
3908 * should be backed up
3909 * 0 if things are progressing smoothly
3910 * -1 if the reshape is finished because it is all done,
3911 * -2 if the reshape is finished due to an error.
3914 int advancing
= (reshape
->after
.data_disks
3915 >= reshape
->before
.data_disks
);
3916 unsigned long long need_backup
; /* All data between start of array and
3917 * here will at some point need to
3920 unsigned long long read_offset
, write_offset
;
3921 unsigned long long write_range
;
3922 unsigned long long max_progress
, target
, completed
;
3923 unsigned long long array_size
= (info
->component_size
3924 * reshape
->before
.data_disks
);
3928 /* First, we unsuspend any region that is now known to be safe.
3929 * If suspend_point is on the 'wrong' side of reshape_progress, then
3930 * we don't have or need suspension at the moment. This is true for
3931 * native metadata when we don't need to back-up.
3934 if (info
->reshape_progress
<= *suspend_point
)
3935 sysfs_set_num(info
, NULL
, "suspend_lo",
3936 info
->reshape_progress
);
3938 /* Note: this won't work in 2.6.37 and before.
3939 * Something somewhere should make sure we don't need it!
3941 if (info
->reshape_progress
>= *suspend_point
)
3942 sysfs_set_num(info
, NULL
, "suspend_hi",
3943 info
->reshape_progress
);
3946 /* Now work out how far it is safe to progress.
3947 * If the read_offset for ->reshape_progress is less than
3948 * 'blocks' beyond the write_offset, we can only progress as far
3950 * Otherwise we can progress until the write_offset for the new location
3951 * reaches (within 'blocks' of) the read_offset at the current location.
3952 * However that region must be suspended unless we are using native
3954 * If we need to suspend more, we limit it to 128M per device, which is
3955 * rather arbitrary and should be some time-based calculation.
3957 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3958 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3959 write_range
= info
->new_chunk
/512;
3960 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3961 need_backup
= array_size
;
3963 need_backup
= reshape
->backup_blocks
;
3965 if (read_offset
< write_offset
+ write_range
)
3966 max_progress
= backup_point
;
3969 read_offset
* reshape
->after
.data_disks
;
3971 if (read_offset
> write_offset
- write_range
)
3972 /* Can only progress as far as has been backed up,
3973 * which must be suspended */
3974 max_progress
= backup_point
;
3975 else if (info
->reshape_progress
<= need_backup
)
3976 max_progress
= backup_point
;
3978 if (info
->array
.major_version
>= 0)
3979 /* Can progress until backup is needed */
3980 max_progress
= need_backup
;
3982 /* Can progress until metadata update is required */
3984 read_offset
* reshape
->after
.data_disks
;
3985 /* but data must be suspended */
3986 if (max_progress
< *suspend_point
)
3987 max_progress
= *suspend_point
;
3992 /* We know it is safe to progress to 'max_progress' providing
3993 * it is suspended or we are using native metadata.
3994 * Consider extending suspend_point 128M per device if it
3995 * is less than 64M per device beyond reshape_progress.
3996 * But always do a multiple of 'blocks'
3997 * FIXME this is too big - it takes to long to complete
4000 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
4001 reshape
->after
.data_disks
);
4002 target
/= reshape
->backup_blocks
;
4005 target
*= reshape
->backup_blocks
;
4007 /* For externally managed metadata we always need to suspend IO to
4008 * the area being reshaped so we regularly push suspend_point forward.
4009 * For native metadata we only need the suspend if we are going to do
4013 if ((need_backup
> info
->reshape_progress
||
4014 info
->array
.major_version
< 0) &&
4015 *suspend_point
< info
->reshape_progress
+ target
) {
4016 if (need_backup
< *suspend_point
+ 2 * target
)
4017 *suspend_point
= need_backup
;
4018 else if (*suspend_point
+ 2 * target
< array_size
)
4019 *suspend_point
+= 2 * target
;
4021 *suspend_point
= array_size
;
4022 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
4023 if (max_progress
> *suspend_point
)
4024 max_progress
= *suspend_point
;
4027 if (info
->array
.major_version
>= 0) {
4028 /* Only need to suspend when about to backup */
4029 if (info
->reshape_progress
< need_backup
* 2 &&
4030 *suspend_point
> 0) {
4032 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4033 sysfs_set_num(info
, NULL
, "suspend_hi",
4037 /* Need to suspend continually */
4038 if (info
->reshape_progress
< *suspend_point
)
4039 *suspend_point
= info
->reshape_progress
;
4040 if (*suspend_point
+ target
< info
->reshape_progress
)
4041 /* No need to move suspend region yet */;
4043 if (*suspend_point
>= 2 * target
)
4044 *suspend_point
-= 2 * target
;
4047 sysfs_set_num(info
, NULL
, "suspend_lo",
4050 if (max_progress
< *suspend_point
)
4051 max_progress
= *suspend_point
;
4055 /* now set sync_max to allow that progress. sync_max, like
4056 * sync_completed is a count of sectors written per device, so
4057 * we find the difference between max_progress and the start point,
4058 * and divide that by after.data_disks to get a sync_max
4060 * At the same time we convert wait_point to a similar number
4061 * for comparing against sync_completed.
4063 /* scale down max_progress to per_disk */
4064 max_progress
/= reshape
->after
.data_disks
;
4066 * Round to chunk size as some kernels give an erroneously
4069 max_progress
/= info
->new_chunk
/512;
4070 max_progress
*= info
->new_chunk
/512;
4071 /* And round to old chunk size as the kernel wants that */
4072 max_progress
/= info
->array
.chunk_size
/512;
4073 max_progress
*= info
->array
.chunk_size
/512;
4074 /* Limit progress to the whole device */
4075 if (max_progress
> info
->component_size
)
4076 max_progress
= info
->component_size
;
4077 wait_point
/= reshape
->after
.data_disks
;
4079 /* switch from 'device offset' to 'processed block count' */
4080 max_progress
= info
->component_size
- max_progress
;
4081 wait_point
= info
->component_size
- wait_point
;
4085 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4087 /* Now wait. If we have already reached the point that we were
4088 * asked to wait to, don't wait at all, else wait for any change.
4089 * We need to select on 'sync_completed' as that is the place that
4090 * notifications happen, but we are really interested in
4091 * 'reshape_position'
4093 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4095 goto check_progress
;
4097 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4098 goto check_progress
;
4100 while (completed
< max_progress
&& completed
< wait_point
) {
4101 /* Check that sync_action is still 'reshape' to avoid
4102 * waiting forever on a dead array
4105 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) <= 0 ||
4106 strncmp(action
, "reshape", 7) != 0)
4108 /* Some kernels reset 'sync_completed' to zero
4109 * before setting 'sync_action' to 'idle'.
4110 * So we need these extra tests.
4112 if (completed
== 0 && advancing
&&
4113 strncmp(action
, "idle", 4) == 0 &&
4114 info
->reshape_progress
> 0)
4116 if (completed
== 0 && !advancing
&&
4117 strncmp(action
, "idle", 4) == 0 &&
4118 info
->reshape_progress
<
4119 (info
->component_size
* reshape
->after
.data_disks
))
4121 sysfs_wait(fd
, NULL
);
4122 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4123 goto check_progress
;
4125 /* Some kernels reset 'sync_completed' to zero,
4126 * we need to have real point we are in md.
4127 * So in that case, read 'reshape_position' from sysfs.
4129 if (completed
== 0) {
4130 unsigned long long reshapep
;
4132 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) > 0 &&
4133 strncmp(action
, "idle", 4) == 0 &&
4134 sysfs_get_ll(info
, NULL
,
4135 "reshape_position", &reshapep
) == 0)
4136 *reshape_completed
= reshapep
;
4138 /* some kernels can give an incorrectly high
4139 * 'completed' number, so round down */
4140 completed
/= (info
->new_chunk
/512);
4141 completed
*= (info
->new_chunk
/512);
4142 /* Convert 'completed' back in to a 'progress' number */
4143 completed
*= reshape
->after
.data_disks
;
4145 completed
= (info
->component_size
4146 * reshape
->after
.data_disks
4148 *reshape_completed
= completed
;
4153 /* We return the need_backup flag. Caller will decide
4154 * how much - a multiple of ->backup_blocks up to *suspend_point
4157 return need_backup
> info
->reshape_progress
;
4159 return need_backup
>= info
->reshape_progress
;
4162 /* if we couldn't read a number from sync_completed, then
4163 * either the reshape did complete, or it aborted.
4164 * We can tell which by checking for 'none' in reshape_position.
4165 * If it did abort, then it might immediately restart if it
4166 * it was just a device failure that leaves us degraded but
4169 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
,
4170 sizeof(buf
)) < 0 || strncmp(buf
, "none", 4) != 0) {
4171 /* The abort might only be temporary. Wait up to 10
4172 * seconds for fd to contain a valid number again.
4176 unsigned long long new_sync_max
;
4177 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4178 if (sysfs_wait(fd
, &wait
) != 1)
4180 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4182 /* all good again */
4184 /* If "sync_max" is no longer max_progress
4185 * we need to freeze things
4187 sysfs_get_ll(info
, NULL
, "sync_max",
4189 *frozen
= (new_sync_max
!= max_progress
);
4191 case -2: /* read error - abort */
4198 return rv
; /* abort */
4200 /* Maybe racing with array shutdown - check state */
4203 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4205 strncmp(buf
, "inactive", 8) == 0 ||
4206 strncmp(buf
, "clear",5) == 0)
4207 return -2; /* abort */
4208 return -1; /* complete */
4212 /* FIXME return status is never checked */
4213 static int grow_backup(struct mdinfo
*sra
,
4214 unsigned long long offset
, /* per device */
4215 unsigned long stripes
, /* per device, in old chunks */
4216 int *sources
, unsigned long long *offsets
,
4217 int disks
, int chunk
, int level
, int layout
,
4218 int dests
, int *destfd
, unsigned long long *destoffsets
,
4219 int part
, int *degraded
,
4222 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4223 * to storage 'destfd' (offset 'destoffsets'), after first
4224 * suspending IO. Then allow resync to continue
4225 * over the suspended section.
4226 * Use part 'part' of the backup-super-block.
4231 unsigned long long ll
;
4233 //printf("offset %llu\n", offset);
4239 /* Check that array hasn't become degraded, else we might backup the wrong data */
4240 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4241 return -1; /* FIXME this error is ignored */
4242 new_degraded
= (int)ll
;
4243 if (new_degraded
!= *degraded
) {
4244 /* check each device to ensure it is still working */
4246 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4247 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4249 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4252 if (sysfs_get_str(sra
, sd
, "state",
4253 sbuf
, sizeof(sbuf
)) < 0 ||
4254 strstr(sbuf
, "faulty") ||
4255 strstr(sbuf
, "in_sync") == NULL
) {
4256 /* this device is dead */
4257 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4258 if (sd
->disk
.raid_disk
>= 0 &&
4259 sources
[sd
->disk
.raid_disk
] >= 0) {
4260 close(sources
[sd
->disk
.raid_disk
]);
4261 sources
[sd
->disk
.raid_disk
] = -1;
4266 *degraded
= new_degraded
;
4269 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4270 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4272 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4273 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4276 bsb
.magic
[15] = '2';
4277 for (i
= 0; i
< dests
; i
++)
4279 lseek64(destfd
[i
], destoffsets
[i
] +
4280 __le64_to_cpu(bsb
.devstart2
)*512, 0);
4282 lseek64(destfd
[i
], destoffsets
[i
], 0);
4284 rv
= save_stripes(sources
, offsets
, disks
, chunk
, level
, layout
,
4285 dests
, destfd
, offset
* 512 * odata
,
4286 stripes
* chunk
* odata
, buf
);
4290 bsb
.mtime
= __cpu_to_le64(time(0));
4291 for (i
= 0; i
< dests
; i
++) {
4292 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4294 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4295 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4296 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4297 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4298 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4301 if ((unsigned long long)lseek64(destfd
[i
],
4302 destoffsets
[i
] - 4096, 0) !=
4303 destoffsets
[i
] - 4096)
4305 if (write(destfd
[i
], &bsb
, 512) != 512)
4307 if (destoffsets
[i
] > 4096) {
4308 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4309 destoffsets
[i
]+stripes
*chunk
*odata
)
4311 if (write(destfd
[i
], &bsb
, 512) != 512)
4321 /* in 2.6.30, the value reported by sync_completed can be
4322 * less that it should be by one stripe.
4323 * This only happens when reshape hits sync_max and pauses.
4324 * So allow wait_backup to either extent sync_max further
4325 * than strictly necessary, or return before the
4326 * sync has got quite as far as we would really like.
4327 * This is what 'blocks2' is for.
4328 * The various caller give appropriate values so that
4331 /* FIXME return value is often ignored */
4332 static int forget_backup(int dests
, int *destfd
,
4333 unsigned long long *destoffsets
,
4337 * Erase backup 'part' (which is 0 or 1)
4343 bsb
.arraystart2
= __cpu_to_le64(0);
4344 bsb
.length2
= __cpu_to_le64(0);
4346 bsb
.arraystart
= __cpu_to_le64(0);
4347 bsb
.length
= __cpu_to_le64(0);
4349 bsb
.mtime
= __cpu_to_le64(time(0));
4351 for (i
= 0; i
< dests
; i
++) {
4352 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4353 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4354 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4355 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4356 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4357 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4358 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4359 destoffsets
[i
]-4096)
4361 if (rv
== 0 && write(destfd
[i
], &bsb
, 512) != 512)
4368 static void fail(char *msg
)
4371 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4372 rv
|= (write(2, "\n", 1) != 1);
4376 static char *abuf
, *bbuf
;
4377 static unsigned long long abuflen
;
4378 static void validate(int afd
, int bfd
, unsigned long long offset
)
4380 /* check that the data in the backup against the array.
4381 * This is only used for regression testing and should not
4382 * be used while the array is active
4386 lseek64(bfd
, offset
- 4096, 0);
4387 if (read(bfd
, &bsb2
, 512) != 512)
4388 fail("cannot read bsb");
4389 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4390 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4391 fail("first csum bad");
4392 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4393 fail("magic is bad");
4394 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4395 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4396 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4397 fail("second csum bad");
4399 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4400 fail("devstart is wrong");
4403 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4405 if (abuflen
< len
) {
4409 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4410 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4412 /* just stop validating on mem-alloc failure */
4417 lseek64(bfd
, offset
, 0);
4418 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4419 //printf("len %llu\n", len);
4420 fail("read first backup failed");
4422 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4423 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4424 fail("read first from array failed");
4425 if (memcmp(bbuf
, abuf
, len
) != 0) {
4428 printf("offset=%llu len=%llu\n",
4429 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4430 for (i
=0; i
<len
; i
++)
4431 if (bbuf
[i
] != abuf
[i
]) {
4432 printf("first diff byte %d\n", i
);
4436 fail("data1 compare failed");
4440 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4442 if (abuflen
< len
) {
4446 abuf
= xmalloc(abuflen
);
4447 bbuf
= xmalloc(abuflen
);
4450 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4451 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4452 fail("read second backup failed");
4453 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4454 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4455 fail("read second from array failed");
4456 if (memcmp(bbuf
, abuf
, len
) != 0)
4457 fail("data2 compare failed");
4461 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4462 struct supertype
*st
, unsigned long blocks
,
4463 int *fds
, unsigned long long *offsets
,
4464 int dests
, int *destfd
, unsigned long long *destoffsets
)
4466 /* Monitor a reshape where backup is being performed using
4467 * 'native' mechanism - either to a backup file, or
4468 * to some space in a spare.
4472 unsigned long long speed
;
4473 unsigned long long suspend_point
, array_size
;
4474 unsigned long long backup_point
, wait_point
;
4475 unsigned long long reshape_completed
;
4477 int increasing
= reshape
->after
.data_disks
>=
4478 reshape
->before
.data_disks
;
4479 int part
= 0; /* The next part of the backup area to fill. It
4480 * may already be full, so we need to check */
4481 int level
= reshape
->level
;
4482 int layout
= reshape
->before
.layout
;
4483 int data
= reshape
->before
.data_disks
;
4484 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4485 int chunk
= sra
->array
.chunk_size
;
4487 unsigned long stripes
;
4491 /* set up the backup-super-block. This requires the
4492 * uuid from the array.
4494 /* Find a superblock */
4495 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4499 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4501 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4502 devfd
= dev_open(dn
, O_RDONLY
);
4505 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4511 pr_err("Cannot find a superblock\n");
4515 memset(&bsb
, 0, 512);
4516 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4517 st
->ss
->uuid_from_super(st
, uuid
);
4518 memcpy(bsb
.set_uuid
, uuid
, 16);
4519 bsb
.mtime
= __cpu_to_le64(time(0));
4520 bsb
.devstart2
= blocks
;
4522 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4523 reshape
->before
.data_disks
;
4525 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4526 /* Don't start the 'reshape' */
4528 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4529 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4530 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4534 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4535 backup_point
= sra
->reshape_progress
;
4538 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4539 backup_point
= reshape
->backup_blocks
;
4540 suspend_point
= array_size
;
4546 /* Want to return as soon the oldest backup slot can
4547 * be released as that allows us to start backing up
4548 * some more, providing suspend_point has been
4549 * advanced, which it should have.
4552 wait_point
= array_size
;
4553 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4554 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4555 __le64_to_cpu(bsb
.length
));
4556 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4557 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4558 __le64_to_cpu(bsb
.length2
));
4561 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4562 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4563 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4564 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4567 reshape_completed
= sra
->reshape_progress
;
4568 rv
= progress_reshape(sra
, reshape
,
4569 backup_point
, wait_point
,
4570 &suspend_point
, &reshape_completed
,
4572 /* external metadata would need to ping_monitor here */
4573 sra
->reshape_progress
= reshape_completed
;
4575 /* Clear any backup region that is before 'here' */
4577 if (__le64_to_cpu(bsb
.length
) > 0 &&
4578 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4579 __le64_to_cpu(bsb
.length
)))
4580 forget_backup(dests
, destfd
,
4582 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4583 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4584 __le64_to_cpu(bsb
.length2
)))
4585 forget_backup(dests
, destfd
,
4588 if (__le64_to_cpu(bsb
.length
) > 0 &&
4589 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4590 forget_backup(dests
, destfd
,
4592 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4593 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4594 forget_backup(dests
, destfd
,
4604 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4605 /* No longer need to monitor this reshape */
4606 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4612 unsigned long long offset
;
4613 unsigned long actual_stripes
;
4614 /* Need to backup some data.
4615 * If 'part' is not used and the desired
4616 * backup size is suspended, do a backup,
4617 * then consider the next part.
4619 /* Check that 'part' is unused */
4620 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4622 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4625 offset
= backup_point
/ data
;
4626 actual_stripes
= stripes
;
4628 if (offset
+ actual_stripes
* (chunk
/512) >
4629 sra
->component_size
)
4630 actual_stripes
= ((sra
->component_size
- offset
)
4632 if (offset
+ actual_stripes
* (chunk
/512) >
4636 if (offset
< actual_stripes
* (chunk
/512))
4637 actual_stripes
= offset
/ (chunk
/512);
4638 offset
-= actual_stripes
* (chunk
/512);
4639 if (offset
< suspend_point
/data
)
4642 if (actual_stripes
== 0)
4644 grow_backup(sra
, offset
, actual_stripes
, fds
, offsets
,
4645 disks
, chunk
, level
, layout
, dests
, destfd
,
4646 destoffsets
, part
, °raded
, buf
);
4647 validate(afd
, destfd
[0], destoffsets
[0]);
4648 /* record where 'part' is up to */
4651 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4653 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4657 /* FIXME maybe call progress_reshape one more time instead */
4658 /* remove any remaining suspension */
4659 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4660 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4661 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4662 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4664 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4665 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4671 * If any spare contains md_back_data-1 which is recent wrt mtime,
4672 * write that data into the array and update the super blocks with
4673 * the new reshape_progress
4675 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
,
4676 int cnt
, char *backup_file
, int verbose
)
4680 unsigned long long *offsets
;
4681 unsigned long long nstripe
, ostripe
;
4684 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4685 if (info
->array
.level
== 6)
4686 odata
--; /* number of data disks */
4687 ndata
= info
->array
.raid_disks
- 1;
4688 if (info
->new_level
== 6)
4691 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4693 if (info
->delta_disks
<= 0)
4694 /* Didn't grow, so the backup file must have
4698 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4699 struct mdinfo dinfo
;
4702 char *devname
, namebuf
[20];
4703 unsigned long long lo
, hi
;
4705 /* This was a spare and may have some saved data on it.
4706 * Load the superblock, find and load the
4707 * backup_super_block.
4708 * If either fail, go on to next device.
4709 * If the backup contains no new info, just return
4710 * else restore data and update all superblocks
4712 if (i
== old_disks
-1) {
4713 fd
= open(backup_file
, O_RDONLY
);
4715 pr_err("backup file %s inaccessible: %s\n",
4716 backup_file
, strerror(errno
));
4719 devname
= backup_file
;
4724 if (st
->ss
->load_super(st
, fd
, NULL
))
4727 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4728 st
->ss
->free_super(st
);
4731 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4733 pr_err("Cannot seek on device %d\n", i
);
4734 continue; /* Cannot seek */
4736 sprintf(namebuf
, "device-%d", i
);
4739 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4741 pr_err("Cannot read from %s\n", devname
);
4742 continue; /* Cannot read */
4744 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4745 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4747 pr_err("No backup metadata on %s\n", devname
);
4750 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4752 pr_err("Bad backup-metadata checksum on %s\n",
4754 continue; /* bad checksum */
4756 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4757 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4759 pr_err("Bad backup-metadata checksum2 on %s\n",
4761 continue; /* Bad second checksum */
4763 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4765 pr_err("Wrong uuid on backup-metadata on %s\n",
4767 continue; /* Wrong uuid */
4771 * array utime and backup-mtime should be updated at
4772 * much the same time, but it seems that sometimes
4773 * they aren't... So allow considerable flexability in
4774 * matching, and allow this test to be overridden by
4775 * an environment variable.
4777 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4778 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4779 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4780 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4781 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4782 (unsigned long)info
->array
.utime
);
4784 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4785 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4786 continue; /* time stamp is too bad */
4790 if (bsb
.magic
[15] == '1') {
4791 if (bsb
.length
== 0)
4793 if (info
->delta_disks
>= 0) {
4794 /* reshape_progress is increasing */
4795 if (__le64_to_cpu(bsb
.arraystart
)
4796 + __le64_to_cpu(bsb
.length
)
4797 < info
->reshape_progress
) {
4800 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4801 continue; /* No new data here */
4804 /* reshape_progress is decreasing */
4805 if (__le64_to_cpu(bsb
.arraystart
) >=
4806 info
->reshape_progress
)
4807 goto nonew
; /* No new data here */
4810 if (bsb
.length
== 0 && bsb
.length2
== 0)
4812 if (info
->delta_disks
>= 0) {
4813 /* reshape_progress is increasing */
4814 if ((__le64_to_cpu(bsb
.arraystart
)
4815 + __le64_to_cpu(bsb
.length
)
4816 < info
->reshape_progress
) &&
4817 (__le64_to_cpu(bsb
.arraystart2
)
4818 + __le64_to_cpu(bsb
.length2
)
4819 < info
->reshape_progress
))
4820 goto nonew
; /* No new data here */
4822 /* reshape_progress is decreasing */
4823 if (__le64_to_cpu(bsb
.arraystart
) >=
4824 info
->reshape_progress
&&
4825 __le64_to_cpu(bsb
.arraystart2
) >=
4826 info
->reshape_progress
)
4827 goto nonew
; /* No new data here */
4830 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4833 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4835 continue; /* Cannot seek */
4837 /* There should be a duplicate backup superblock 4k before here */
4838 if (lseek64(fd
, -4096, 1) < 0 ||
4839 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4840 goto second_fail
; /* Cannot find leading superblock */
4841 if (bsb
.magic
[15] == '1')
4842 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4844 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4845 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4846 goto second_fail
; /* Cannot find leading superblock */
4848 /* Now need the data offsets for all devices. */
4849 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4850 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4853 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4854 /* FIXME should be this be an error */
4856 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4857 st
->ss
->free_super(st
);
4858 offsets
[j
] = dinfo
.data_offset
* 512;
4860 printf("%s: restoring critical section\n", Name
);
4862 if (restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4863 info
->new_chunk
, info
->new_level
,
4864 info
->new_layout
, fd
,
4865 __le64_to_cpu(bsb
.devstart
)*512,
4866 __le64_to_cpu(bsb
.arraystart
)*512,
4867 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4868 /* didn't succeed, so giveup */
4870 pr_err("Error restoring backup from %s\n",
4876 if (bsb
.magic
[15] == '2' &&
4877 restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4878 info
->new_chunk
, info
->new_level
,
4879 info
->new_layout
, fd
,
4880 __le64_to_cpu(bsb
.devstart
)*512 +
4881 __le64_to_cpu(bsb
.devstart2
)*512,
4882 __le64_to_cpu(bsb
.arraystart2
)*512,
4883 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4884 /* didn't succeed, so giveup */
4886 pr_err("Error restoring second backup from %s\n",
4894 /* Ok, so the data is restored. Let's update those superblocks. */
4898 lo
= __le64_to_cpu(bsb
.arraystart
);
4899 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4901 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4902 unsigned long long lo1
, hi1
;
4903 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4904 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4908 } else if (lo
< lo1
)
4913 if (lo
< hi
&& (info
->reshape_progress
< lo
||
4914 info
->reshape_progress
> hi
))
4915 /* backup does not affect reshape_progress*/ ;
4916 else if (info
->delta_disks
>= 0) {
4917 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4918 __le64_to_cpu(bsb
.length
);
4919 if (bsb
.magic
[15] == '2') {
4920 unsigned long long p2
;
4922 p2
= __le64_to_cpu(bsb
.arraystart2
) +
4923 __le64_to_cpu(bsb
.length2
);
4924 if (p2
> info
->reshape_progress
)
4925 info
->reshape_progress
= p2
;
4928 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4929 if (bsb
.magic
[15] == '2') {
4930 unsigned long long p2
;
4932 p2
= __le64_to_cpu(bsb
.arraystart2
);
4933 if (p2
< info
->reshape_progress
)
4934 info
->reshape_progress
= p2
;
4937 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4940 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4942 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4943 dinfo
.reshape_progress
= info
->reshape_progress
;
4944 st
->ss
->update_super(st
, &dinfo
, "_reshape_progress",
4946 st
->ss
->store_super(st
, fdlist
[j
]);
4947 st
->ss
->free_super(st
);
4951 /* Didn't find any backup data, try to see if any
4954 if (info
->delta_disks
< 0) {
4955 /* When shrinking, the critical section is at the end.
4956 * So see if we are before the critical section.
4958 unsigned long long first_block
;
4959 nstripe
= ostripe
= 0;
4961 while (ostripe
>= nstripe
) {
4962 ostripe
+= info
->array
.chunk_size
/ 512;
4963 first_block
= ostripe
* odata
;
4964 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4965 (info
->new_chunk
/512);
4968 if (info
->reshape_progress
>= first_block
)
4971 if (info
->delta_disks
> 0) {
4972 /* See if we are beyond the critical section. */
4973 unsigned long long last_block
;
4974 nstripe
= ostripe
= 0;
4976 while (nstripe
>= ostripe
) {
4977 nstripe
+= info
->new_chunk
/ 512;
4978 last_block
= nstripe
* ndata
;
4979 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4980 (info
->array
.chunk_size
/512);
4983 if (info
->reshape_progress
>= last_block
)
4986 /* needed to recover critical section! */
4988 pr_err("Failed to find backup of critical section\n");
4992 int Grow_continue_command(char *devname
, int fd
,
4993 char *backup_file
, int verbose
)
4996 struct supertype
*st
= NULL
;
4997 struct mdinfo
*content
= NULL
;
4998 struct mdinfo array
;
4999 char *subarray
= NULL
;
5000 struct mdinfo
*cc
= NULL
;
5001 struct mdstat_ent
*mdstat
= NULL
;
5005 dprintf("Grow continue from command line called for %s\n", devname
);
5007 st
= super_by_fd(fd
, &subarray
);
5008 if (!st
|| !st
->ss
) {
5009 pr_err("Unable to determine metadata format for %s\n", devname
);
5012 dprintf("Grow continue is run for ");
5013 if (st
->ss
->external
== 0) {
5016 dprintf_cont("native array (%s)\n", devname
);
5017 if (md_get_array_info(fd
, &array
.array
) < 0) {
5018 pr_err("%s is not an active md array - aborting\n",
5021 goto Grow_continue_command_exit
;
5024 sysfs_init(content
, fd
, NULL
);
5025 /* Need to load a superblock.
5026 * FIXME we should really get what we need from
5030 for (d
= 0; d
< MAX_DISKS
; d
++) {
5031 mdu_disk_info_t disk
;
5035 if (md_get_disk_info(fd
, &disk
) < 0)
5037 if (disk
.major
== 0 && disk
.minor
== 0)
5039 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5041 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5044 fd2
= dev_open(dv
, O_RDONLY
);
5047 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5053 if (d
== MAX_DISKS
) {
5054 pr_err("Unable to load metadata for %s\n",
5057 goto Grow_continue_command_exit
;
5059 st
->ss
->getinfo_super(st
, content
, NULL
);
5060 if (!content
->reshape_active
)
5064 } while (cnt
-- > 0);
5069 dprintf_cont("subarray (%s)\n", subarray
);
5070 container
= st
->container_devnm
;
5071 cfd
= open_dev_excl(st
->container_devnm
);
5073 container
= st
->devnm
;
5075 cfd
= open_dev_excl(st
->devnm
);
5076 dprintf_cont("container (%s)\n", container
);
5080 pr_err("Unable to open container for %s\n", devname
);
5082 goto Grow_continue_command_exit
;
5085 /* find in container array under reshape
5087 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5089 pr_err("Cannot read superblock for %s\n", devname
);
5091 goto Grow_continue_command_exit
;
5094 cc
= st
->ss
->container_content(st
, subarray
);
5095 for (content
= cc
; content
; content
= content
->next
) {
5097 int allow_reshape
= 1;
5099 if (content
->reshape_active
== 0)
5101 /* The decision about array or container wide
5102 * reshape is taken in Grow_continue based
5103 * content->reshape_active state, therefore we
5104 * need to check_reshape based on
5105 * reshape_active and subarray name
5107 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5109 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5110 (content
->array
.state
5111 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5114 if (!allow_reshape
) {
5115 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5116 devname
, container
);
5118 goto Grow_continue_command_exit
;
5121 array_name
= strchr(content
->text_version
+1, '/')+1;
5122 mdstat
= mdstat_by_subdev(array_name
, container
);
5125 if (mdstat
->active
== 0) {
5126 pr_err("Skipping inactive array %s.\n",
5128 free_mdstat(mdstat
);
5135 pr_err("Unable to determine reshaped array for %s\n", devname
);
5137 goto Grow_continue_command_exit
;
5139 fd2
= open_dev(mdstat
->devnm
);
5141 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5143 goto Grow_continue_command_exit
;
5146 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5147 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5151 goto Grow_continue_command_exit
;
5156 /* start mdmon in case it is not running
5158 if (!mdmon_running(container
))
5159 start_mdmon(container
);
5160 ping_monitor(container
);
5162 if (mdmon_running(container
))
5163 st
->update_tail
= &st
->updates
;
5165 pr_err("No mdmon found. Grow cannot continue.\n");
5167 goto Grow_continue_command_exit
;
5171 /* verify that array under reshape is started from
5174 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5176 goto Grow_continue_command_exit
;
5181 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5183 Grow_continue_command_exit
:
5186 st
->ss
->free_super(st
);
5187 free_mdstat(mdstat
);
5194 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5195 char *backup_file
, int forked
, int freeze_reshape
)
5199 if (!info
->reshape_active
)
5202 if (st
->ss
->external
) {
5203 int cfd
= open_dev(st
->container_devnm
);
5208 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5210 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5211 st
, info
, 0, backup_file
, 0,
5212 forked
, 1 | info
->reshape_active
,
5215 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5216 NULL
, INVALID_SECTORS
, backup_file
,
5217 0, forked
, 1 | info
->reshape_active
,
5223 char *make_backup(char *name
)
5225 char *base
= "backup_file-";
5229 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5230 fname
= xmalloc(len
);
5231 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5235 char *locate_backup(char *name
)
5237 char *fl
= make_backup(name
);
5240 if (stat(fl
, &stb
) == 0 && S_ISREG(stb
.st_mode
))