2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
38 int restore_backup(struct supertype
*st
,
39 struct mdinfo
*content
,
49 int disk_count
= next_spare
+ working_disks
;
50 char *backup_file
= *backup_filep
;
52 dprintf("Called restore_backup()\n");
53 fdlist
= xmalloc(sizeof(int) * disk_count
);
55 enable_fds(next_spare
);
56 for (i
= 0; i
< next_spare
; i
++)
58 for (dev
= content
->devs
; dev
; dev
= dev
->next
) {
62 sprintf(buf
, "%d:%d", dev
->disk
.major
, dev
->disk
.minor
);
63 fd
= dev_open(buf
, O_RDWR
);
65 if (dev
->disk
.raid_disk
>= 0)
66 fdlist
[dev
->disk
.raid_disk
] = fd
;
68 fdlist
[next_spare
++] = fd
;
72 backup_file
= locate_backup(content
->sys_name
);
73 *backup_filep
= backup_file
;
76 if (st
->ss
->external
&& st
->ss
->recover_backup
)
77 err
= st
->ss
->recover_backup(st
, content
);
79 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
80 backup_file
, verbose
> 0);
82 while (next_spare
> 0) {
84 if (fdlist
[next_spare
] >= 0)
85 close(fdlist
[next_spare
]);
89 pr_err("Failed to restore critical section for reshape - sorry.\n");
91 pr_err("Possibly you need to specify a --backup-file\n");
95 dprintf("restore_backup() returns status OK.\n");
99 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
101 /* Add a device to an active array.
102 * Currently, just extend a linear array.
103 * This requires writing a new superblock on the
104 * new device, calling the kernel to add the device,
105 * and if that succeeds, update the superblock on
107 * This means that we need to *find* all other devices.
114 struct supertype
*st
= NULL
;
115 char *subarray
= NULL
;
117 if (md_get_array_info(fd
, &info
.array
) < 0) {
118 pr_err("cannot get array info for %s\n", devname
);
122 if (info
.array
.level
!= -1) {
123 pr_err("can only add devices to linear arrays\n");
127 st
= super_by_fd(fd
, &subarray
);
129 pr_err("cannot handle arrays with superblock version %d\n",
130 info
.array
.major_version
);
135 pr_err("Cannot grow linear sub-arrays yet\n");
141 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
143 pr_err("cannot open %s\n", newdev
);
147 if (!fstat_is_blkdev(nfd
, newdev
, &rdev
)) {
152 /* now check out all the devices and make sure we can read the
154 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
155 mdu_disk_info_t disk
;
158 st
->ss
->free_super(st
);
161 if (md_get_disk_info(fd
, &disk
) < 0) {
162 pr_err("cannot get device detail for device %d\n", d
);
167 dv
= map_dev(disk
.major
, disk
.minor
, 1);
169 pr_err("cannot find device file for device %d\n", d
);
174 fd2
= dev_open(dv
, O_RDWR
);
176 pr_err("cannot open device file %s\n", dv
);
182 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
183 pr_err("cannot find super block on %s\n", dv
);
191 /* Ok, looks good. Lets update the superblock and write it out to
195 info
.disk
.number
= d
;
196 info
.disk
.major
= major(rdev
);
197 info
.disk
.minor
= minor(rdev
);
198 info
.disk
.raid_disk
= d
;
199 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
200 if (st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
202 pr_err("Preparing new metadata failed on %s\n", newdev
);
207 if (st
->ss
->store_super(st
, nfd
)) {
208 pr_err("Cannot store new superblock on %s\n", newdev
);
214 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
215 pr_err("Cannot add new disk to this array\n");
218 /* Well, that seems to have worked.
219 * Now go through and update all superblocks
222 if (md_get_array_info(fd
, &info
.array
) < 0) {
223 pr_err("cannot get array info for %s\n", devname
);
228 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
229 mdu_disk_info_t disk
;
233 if (md_get_disk_info(fd
, &disk
) < 0) {
234 pr_err("cannot get device detail for device %d\n", d
);
237 dv
= map_dev(disk
.major
, disk
.minor
, 1);
239 pr_err("cannot find device file for device %d\n", d
);
242 fd2
= dev_open(dv
, O_RDWR
);
244 pr_err("cannot open device file %s\n", dv
);
247 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
248 pr_err("cannot find super block on %s\n", dv
);
253 info
.array
.raid_disks
= nd
+1;
254 info
.array
.nr_disks
= nd
+1;
255 info
.array
.active_disks
= nd
+1;
256 info
.array
.working_disks
= nd
+1;
258 if (st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
260 pr_err("Updating metadata failed on %s\n", dv
);
265 if (st
->ss
->store_super(st
, fd2
)) {
266 pr_err("Cannot store new superblock on %s\n", dv
);
276 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
279 * First check that array doesn't have a bitmap
280 * Then create the bitmap
283 * For internal bitmaps, we need to check the version,
284 * find all the active devices, and write the bitmap block
287 mdu_bitmap_file_t bmf
;
288 mdu_array_info_t array
;
289 struct supertype
*st
;
290 char *subarray
= NULL
;
291 int major
= BITMAP_MAJOR_HI
;
292 unsigned long long bitmapsize
, array_size
;
296 * We only ever get called if s->bitmap_file is != NULL, so this check
297 * is just here to quiet down static code checkers.
302 if (strcmp(s
->bitmap_file
, "clustered") == 0)
303 major
= BITMAP_MAJOR_CLUSTERED
;
305 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
307 pr_err("Memory allocation failure.\n");
309 pr_err("bitmaps not supported by this kernel.\n");
312 if (bmf
.pathname
[0]) {
313 if (strcmp(s
->bitmap_file
,"none") == 0) {
314 if (ioctl(fd
, SET_BITMAP_FILE
, -1) != 0) {
315 pr_err("failed to remove bitmap %s\n",
321 pr_err("%s already has a bitmap (%s)\n", devname
, bmf
.pathname
);
324 if (md_get_array_info(fd
, &array
) != 0) {
325 pr_err("cannot get array status for %s\n", devname
);
328 if (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) {
329 if (strcmp(s
->bitmap_file
, "none")==0) {
330 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
331 if (md_set_array_info(fd
, &array
) != 0) {
332 if (array
.state
& (1 << MD_SB_CLUSTERED
))
333 pr_err("failed to remove clustered bitmap.\n");
335 pr_err("failed to remove internal bitmap.\n");
340 pr_err("bitmap already present on %s\n", devname
);
344 if (strcmp(s
->bitmap_file
, "none") == 0) {
345 pr_err("no bitmap found on %s\n", devname
);
348 if (array
.level
<= 0) {
349 pr_err("Bitmaps not meaningful with level %s\n",
350 map_num(pers
, array
.level
)?:"of this array");
353 bitmapsize
= array
.size
;
355 if (get_dev_size(fd
, NULL
, &array_size
) &&
356 array_size
> (0x7fffffffULL
<< 9)) {
357 /* Array is big enough that we cannot trust array.size
358 * try other approaches
360 bitmapsize
= get_component_size(fd
);
362 if (bitmapsize
== 0) {
363 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
367 if (array
.level
== 10) {
370 ncopies
= (array
.layout
& 255) * ((array
.layout
>> 8) & 255);
371 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
373 if (strcmp(s
->bitmap_file
, "clustered") == 0 &&
374 !is_near_layout_10(array
.layout
)) {
375 pr_err("only near layout is supported with clustered raid10\n");
380 st
= super_by_fd(fd
, &subarray
);
382 pr_err("Cannot understand version %d.%d\n",
383 array
.major_version
, array
.minor_version
);
387 pr_err("Cannot add bitmaps to sub-arrays yet\n");
393 mdi
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
);
395 if (mdi
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
396 pr_err("Cannot add bitmap to array with PPL\n");
404 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
405 strcmp(s
->bitmap_file
, "clustered") == 0) {
408 int offset_setable
= 0;
409 if (st
->ss
->add_internal_bitmap
== NULL
) {
410 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
413 st
->nodes
= c
->nodes
;
414 st
->cluster_name
= c
->homecluster
;
415 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
418 for (d
= 0; d
< st
->max_devs
; d
++) {
419 mdu_disk_info_t disk
;
424 if (md_get_disk_info(fd
, &disk
) < 0)
426 if (disk
.major
== 0 && disk
.minor
== 0)
428 if ((disk
.state
& (1 << MD_DISK_SYNC
)) == 0)
430 dv
= map_dev(disk
.major
, disk
.minor
, 1);
433 fd2
= dev_open(dv
, O_RDWR
);
436 rv
= st
->ss
->load_super(st
, fd2
, NULL
);
438 rv
= st
->ss
->add_internal_bitmap(
439 st
, &s
->bitmap_chunk
, c
->delay
,
440 s
->write_behind
, bitmapsize
,
441 offset_setable
, major
);
443 st
->ss
->write_bitmap(st
, fd2
,
446 pr_err("failed to create internal bitmap - chunksize problem.\n");
449 pr_err("failed to load super-block.\n");
455 if (offset_setable
) {
456 st
->ss
->getinfo_super(st
, mdi
, NULL
);
457 if (sysfs_init(mdi
, fd
, NULL
)) {
458 pr_err("failed to initialize sysfs.\n");
461 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
465 if (strcmp(s
->bitmap_file
, "clustered") == 0)
466 array
.state
|= (1 << MD_SB_CLUSTERED
);
467 array
.state
|= (1 << MD_SB_BITMAP_PRESENT
);
468 rv
= md_set_array_info(fd
, &array
);
472 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
473 pr_err("failed to set internal bitmap.\n");
480 int max_devs
= st
->max_devs
;
482 /* try to load a superblock */
483 for (d
= 0; d
< max_devs
; d
++) {
484 mdu_disk_info_t disk
;
488 if (md_get_disk_info(fd
, &disk
) < 0)
490 if ((disk
.major
==0 && disk
.minor
== 0) ||
491 (disk
.state
& (1 << MD_DISK_REMOVED
)))
493 dv
= map_dev(disk
.major
, disk
.minor
, 1);
496 fd2
= dev_open(dv
, O_RDONLY
);
498 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
500 st
->ss
->uuid_from_super(st
, uuid
);
507 pr_err("cannot find UUID for array!\n");
510 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
,
511 s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
512 bitmapsize
, major
)) {
515 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
517 pr_err("weird: %s cannot be opened\n", s
->bitmap_file
);
520 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
523 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
524 pr_err("Cannot set bitmap file for %s: %s\n",
525 devname
, strerror(err
));
533 int Grow_consistency_policy(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
535 struct supertype
*st
;
538 char *subarray
= NULL
;
540 char container_dev
[PATH_MAX
];
543 if (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
544 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
545 pr_err("Operation not supported for consistency policy %s\n",
546 map_num(consistency_policies
, s
->consistency_policy
));
550 st
= super_by_fd(fd
, &subarray
);
554 sra
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
|GET_LEVEL
|
561 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
562 !st
->ss
->write_init_ppl
) {
563 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
568 if (sra
->array
.level
!= 5) {
569 pr_err("Operation not supported for array level %d\n",
575 if (sra
->consistency_policy
== (unsigned)s
->consistency_policy
) {
576 pr_err("Consistency policy is already %s\n",
577 map_num(consistency_policies
, s
->consistency_policy
));
580 } else if (sra
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
581 sra
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
582 pr_err("Current consistency policy is %s, cannot change to %s\n",
583 map_num(consistency_policies
, sra
->consistency_policy
),
584 map_num(consistency_policies
, s
->consistency_policy
));
589 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
590 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) <= 0) {
593 } else if (strcmp(buf
, "reshape\n") == 0) {
594 pr_err("PPL cannot be enabled when reshape is in progress\n");
603 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
608 sprintf(container_dev
, "/dev/%s", st
->container_devnm
);
610 ret
= Update_subarray(container_dev
, subarray
, update
, NULL
,
616 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
623 cfd
= open(container_dev
, O_RDWR
|O_EXCL
);
625 pr_err("Failed to open %s\n", container_dev
);
630 ret
= st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
634 pr_err("Cannot read superblock for %s\n",
639 mdi
= st
->ss
->container_content(st
, subarray
);
644 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
648 devpath
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
649 dfd
= dev_open(devpath
, O_RDWR
);
651 pr_err("Failed to open %s\n", devpath
);
657 ret
= st
->ss
->load_super(st
, dfd
, NULL
);
659 pr_err("Failed to load super-block.\n");
664 ret
= st
->ss
->update_super(st
, sra
, "ppl",
666 c
->verbose
, 0, NULL
);
669 st
->ss
->free_super(st
);
672 st
->ss
->getinfo_super(st
, &info
, NULL
);
675 ret
|= sysfs_set_num(sra
, sd
, "ppl_sector",
677 ret
|= sysfs_set_num(sra
, sd
, "ppl_size",
681 pr_err("Failed to set PPL attributes for %s\n",
684 st
->ss
->free_super(st
);
688 ret
= st
->ss
->write_init_ppl(st
, &info
, dfd
);
690 pr_err("Failed to write PPL\n");
695 st
->ss
->free_super(st
);
702 ret
= sysfs_set_str(sra
, NULL
, "consistency_policy",
703 map_num(consistency_policies
,
704 s
->consistency_policy
));
706 pr_err("Failed to change array consistency policy\n");
718 * When reshaping an array we might need to backup some data.
719 * This is written to all spares with a 'super_block' describing it.
720 * The superblock goes 4K from the end of the used space on the
722 * It if written after the backup is complete.
723 * It has the following structure.
726 static struct mdp_backup_super
{
727 char magic
[16]; /* md_backup_data-1 or -2 */
730 /* start/sizes in 512byte sectors */
731 __u64 devstart
; /* address on backup device/file of data */
734 __u32 sb_csum
; /* csum of preceeding bytes. */
736 __u64 devstart2
; /* offset in to data of second section */
739 __u32 sb_csum2
; /* csum of preceeding bytes. */
741 } __attribute__((aligned(512))) bsb
, bsb2
;
743 static __u32
bsb_csum(char *buf
, int len
)
747 for (i
= 0; i
< len
; i
++)
748 csum
= (csum
<<3) + buf
[0];
749 return __cpu_to_le32(csum
);
752 static int check_idle(struct supertype
*st
)
754 /* Check that all member arrays for this container, or the
755 * container of this array, are idle
757 char *container
= (st
->container_devnm
[0]
758 ? st
->container_devnm
: st
->devnm
);
759 struct mdstat_ent
*ent
, *e
;
762 ent
= mdstat_read(0, 0);
763 for (e
= ent
; e
; e
= e
->next
) {
764 if (!is_container_member(e
, container
))
766 /* frozen array is not idle*/
767 if (e
->percent
>= 0 || e
->metadata_version
[9] == '-') {
776 static int freeze_container(struct supertype
*st
)
778 char *container
= (st
->container_devnm
[0]
779 ? st
->container_devnm
: st
->devnm
);
784 if (block_monitor(container
, 1)) {
785 pr_err("failed to freeze container\n");
792 static void unfreeze_container(struct supertype
*st
)
794 char *container
= (st
->container_devnm
[0]
795 ? st
->container_devnm
: st
->devnm
);
797 unblock_monitor(container
, 1);
800 static int freeze(struct supertype
*st
)
802 /* Try to freeze resync/rebuild on this array/container.
803 * Return -1 if the array is busy,
804 * return -2 container cannot be frozen,
805 * return 0 if this kernel doesn't support 'frozen'
806 * return 1 if it worked.
808 if (st
->ss
->external
)
809 return freeze_container(st
);
811 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
817 /* Need to clear any 'read-auto' status */
818 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
819 strncmp(buf
, "read-auto", 9) == 0)
820 sysfs_set_str(sra
, NULL
, "array_state", "clean");
822 err
= sysfs_freeze_array(sra
);
828 static void unfreeze(struct supertype
*st
)
830 if (st
->ss
->external
)
831 return unfreeze_container(st
);
833 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
837 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
838 strcmp(buf
, "frozen\n") == 0)
839 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
844 static void wait_reshape(struct mdinfo
*sra
)
846 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
852 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
853 strncmp(action
, "reshape", 7) == 0)
854 sysfs_wait(fd
, NULL
);
858 static int reshape_super(struct supertype
*st
, unsigned long long size
,
859 int level
, int layout
, int chunksize
, int raid_disks
,
860 int delta_disks
, char *backup_file
, char *dev
,
861 int direction
, int verbose
)
863 /* nothing extra to check in the native case */
864 if (!st
->ss
->external
)
866 if (!st
->ss
->reshape_super
|| !st
->ss
->manage_reshape
) {
867 pr_err("%s metadata does not support reshape\n",
872 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
873 raid_disks
, delta_disks
, backup_file
, dev
,
877 static void sync_metadata(struct supertype
*st
)
879 if (st
->ss
->external
) {
880 if (st
->update_tail
) {
881 flush_metadata_updates(st
);
882 st
->update_tail
= &st
->updates
;
884 st
->ss
->sync_metadata(st
);
888 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
890 /* when dealing with external metadata subarrays we need to be
891 * prepared to handle EAGAIN. The kernel may need to wait for
892 * mdmon to mark the array active so the kernel can handle
893 * allocations/writeback when preparing the reshape action
894 * (md_allow_write()). We temporarily disable safe_mode_delay
895 * to close a race with the array_state going clean before the
896 * next write to raid_disks / stripe_cache_size
901 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
903 (strcmp(name
, "raid_disks") != 0 &&
904 strcmp(name
, "stripe_cache_size") != 0))
905 return sysfs_set_num(sra
, NULL
, name
, n
);
907 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
910 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
911 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
912 if (rc
< 0 && errno
== EAGAIN
) {
913 ping_monitor(container
);
914 /* if we get EAGAIN here then the monitor is not active
917 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
919 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
923 int start_reshape(struct mdinfo
*sra
, int already_running
,
924 int before_data_disks
, int data_disks
)
927 unsigned long long sync_max_to_set
;
929 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
930 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
931 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
932 sra
->reshape_progress
);
933 if (before_data_disks
<= data_disks
)
934 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
936 sync_max_to_set
= (sra
->component_size
* data_disks
937 - sra
->reshape_progress
) / data_disks
;
938 if (!already_running
)
939 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
940 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
941 if (!already_running
&& err
== 0) {
945 err
= sysfs_set_str(sra
, NULL
, "sync_action",
947 err2
= sysfs_set_str(sra
, NULL
, "sync_max",
951 } while (err
&& err2
&& errno
== EBUSY
&& cnt
-- > 0);
956 void abort_reshape(struct mdinfo
*sra
)
958 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
960 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
961 * suspend_hi to decrease as well as increase.")
962 * you could only increase suspend_{lo,hi} unless the region they
963 * covered was empty. So to reset to 0, you need to push suspend_lo
964 * up past suspend_hi first. So to maximize the chance of mdadm
965 * working on all kernels, we want to keep doing that.
967 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
968 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
969 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
970 sysfs_set_num(sra
, NULL
, "sync_min", 0);
971 // It isn't safe to reset sync_max as we aren't monitoring.
972 // Array really should be stopped at this point.
975 int remove_disks_for_takeover(struct supertype
*st
,
980 struct mdinfo
*remaining
;
983 if (st
->ss
->external
) {
985 struct mdinfo
*arrays
= st
->ss
->container_content(st
, NULL
);
987 * containter_content returns list of arrays in container
988 * If arrays->next is not NULL it means that there are
989 * 2 arrays in container and operation should be blocked
996 pr_err("Error. Cannot perform operation on /dev/%s\n", st
->devnm
);
997 pr_err("For this operation it MUST be single array in container\n");
1003 if (sra
->array
.level
== 10)
1004 nr_of_copies
= layout
& 0xff;
1005 else if (sra
->array
.level
== 1)
1006 nr_of_copies
= sra
->array
.raid_disks
;
1010 remaining
= sra
->devs
;
1012 /* for each 'copy', select one device and remove from the list. */
1013 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
1014 struct mdinfo
**diskp
;
1017 /* Find a working device to keep */
1018 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
1019 struct mdinfo
*disk
= *diskp
;
1021 if (disk
->disk
.raid_disk
< slot
)
1023 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
1025 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
1027 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
1029 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
1032 /* We have found a good disk to use! */
1033 *diskp
= disk
->next
;
1034 disk
->next
= sra
->devs
;
1043 if (slot
< sra
->array
.raid_disks
) {
1044 /* didn't find all slots */
1050 sra
->devs
= remaining
;
1054 /* Remove all 'remaining' devices from the array */
1056 struct mdinfo
*sd
= remaining
;
1057 remaining
= sd
->next
;
1059 sysfs_set_str(sra
, sd
, "state", "faulty");
1060 sysfs_set_str(sra
, sd
, "slot", "none");
1061 /* for external metadata disks should be removed in mdmon */
1062 if (!st
->ss
->external
)
1063 sysfs_set_str(sra
, sd
, "state", "remove");
1064 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
1065 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
1066 sd
->next
= sra
->devs
;
1072 void reshape_free_fdlist(int *fdlist
,
1073 unsigned long long *offsets
,
1078 for (i
= 0; i
< size
; i
++)
1086 int reshape_prepare_fdlist(char *devname
,
1090 unsigned long blocks
,
1093 unsigned long long *offsets
)
1098 enable_fds(nrdisks
);
1099 for (d
= 0; d
<= nrdisks
; d
++)
1102 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1103 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1105 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
1106 sd
->disk
.raid_disk
< raid_disks
) {
1107 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1108 fdlist
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDONLY
);
1109 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1110 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1111 pr_err("%s: cannot open component %s\n",
1112 devname
, dn
? dn
: "-unknown-");
1116 } else if (backup_file
== NULL
) {
1118 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1119 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1120 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1121 if (fdlist
[d
] < 0) {
1122 pr_err("%s: cannot open component %s\n",
1123 devname
, dn
? dn
: "-unknown-");
1134 int reshape_open_backup_file(char *backup_file
,
1139 unsigned long long *offsets
,
1143 /* Return 1 on success, 0 on any form of failure */
1144 /* need to check backup file is large enough */
1150 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
1154 pr_err("%s: cannot create backup file %s: %s\n",
1155 devname
, backup_file
, strerror(errno
));
1158 /* Guard against backup file being on array device.
1159 * If array is partitioned or if LVM etc is in the
1160 * way this will not notice, but it is better than
1163 fstat(*fdlist
, &stb
);
1166 if (stb
.st_rdev
== dev
) {
1167 pr_err("backup file must NOT be on the array being reshaped.\n");
1172 memset(buf
, 0, 512);
1173 for (i
=0; i
< blocks
+ 8 ; i
++) {
1174 if (write(*fdlist
, buf
, 512) != 512) {
1175 pr_err("%s: cannot create backup file %s: %s\n",
1176 devname
, backup_file
, strerror(errno
));
1180 if (fsync(*fdlist
) != 0) {
1181 pr_err("%s: cannot create backup file %s: %s\n",
1182 devname
, backup_file
, strerror(errno
));
1186 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
1187 char *bu
= make_backup(sys_name
);
1188 if (symlink(backup_file
, bu
))
1189 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
1197 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
1198 unsigned int ndata
, unsigned int odata
)
1200 unsigned long a
, b
, blocks
;
1201 /* So how much do we need to backup.
1202 * We need an amount of data which is both a whole number of
1203 * old stripes and a whole number of new stripes.
1204 * So LCM for (chunksize*datadisks).
1206 a
= (ochunk
/512) * odata
;
1207 b
= (nchunk
/512) * ndata
;
1210 /* LCM == product / GCD */
1211 blocks
= (unsigned long)(ochunk
/512) * (unsigned long)(nchunk
/512) *
1217 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1219 /* Based on the current array state in info->array and
1220 * the changes in info->new_* etc, determine:
1221 * - whether the change is possible
1222 * - Intermediate level/raid_disks/layout
1223 * - whether a restriping reshape is needed
1224 * - number of sectors in minimum change unit. This
1225 * will cover a whole number of stripes in 'before' and
1228 * Return message if the change should be rejected
1229 * NULL if the change can be achieved
1231 * This can be called as part of starting a reshape, or
1232 * when assembling an array that is undergoing reshape.
1234 int near
, far
, offset
, copies
;
1236 int old_chunk
, new_chunk
;
1237 /* delta_parity records change in number of devices
1238 * caused by level change
1240 int delta_parity
= 0;
1242 memset(re
, 0, sizeof(*re
));
1244 /* If a new level not explicitly given, we assume no-change */
1245 if (info
->new_level
== UnSet
)
1246 info
->new_level
= info
->array
.level
;
1248 if (info
->new_chunk
)
1249 switch (info
->new_level
) {
1255 /* chunk size is meaningful, must divide component_size
1258 if (info
->component_size
% (info
->new_chunk
/512)) {
1259 unsigned long long shrink
= info
->component_size
;
1260 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1261 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1262 info
->new_chunk
/1024, info
->component_size
/2);
1263 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1265 pr_err("will shrink the array so the given chunk size would work.\n");
1270 return "chunk size not meaningful for this level";
1273 info
->new_chunk
= info
->array
.chunk_size
;
1275 switch (info
->array
.level
) {
1277 return "No reshape is possibly for this RAID level";
1279 if (info
->delta_disks
!= UnSet
)
1280 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1282 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1284 /* RAID1 can convert to RAID1 with different disks, or
1285 * raid5 with 2 disks, or
1288 if (info
->new_level
> 1 && (info
->component_size
& 7))
1289 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1290 if (info
->new_level
== 0) {
1291 if (info
->delta_disks
!= UnSet
&&
1292 info
->delta_disks
!= 0)
1293 return "Cannot change number of disks with RAID1->RAID0 conversion";
1295 re
->before
.data_disks
= 1;
1296 re
->after
.data_disks
= 1;
1299 if (info
->new_level
== 1) {
1300 if (info
->delta_disks
== UnSet
)
1301 /* Don't know what to do */
1302 return "no change requested for Growing RAID1";
1306 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1307 return "Can only convert a 2-device array to RAID5";
1308 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1310 re
->before
.data_disks
= 1;
1311 if (info
->delta_disks
!= UnSet
&&
1312 info
->delta_disks
!= 0)
1313 re
->after
.data_disks
= 1 + info
->delta_disks
;
1315 re
->after
.data_disks
= 1;
1316 if (re
->after
.data_disks
< 1)
1317 return "Number of disks too small for RAID5";
1319 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1320 info
->array
.chunk_size
= 65536;
1323 /* Could do some multi-stage conversions, but leave that to
1326 return "Impossibly level change request for RAID1";
1329 /* RAID10 can be converted from near mode to
1330 * RAID0 by removing some devices.
1331 * It can also be reshaped if the kernel supports
1334 switch (info
->new_level
) {
1336 if ((info
->array
.layout
& ~0xff) != 0x100)
1337 return "Cannot Grow RAID10 with far/offset layout";
1339 * number of devices must be multiple of
1342 if (info
->array
.raid_disks
%
1343 (info
->array
.layout
& 0xff))
1344 return "RAID10 layout too complex for Grow operation";
1346 new_disks
= (info
->array
.raid_disks
/
1347 (info
->array
.layout
& 0xff));
1348 if (info
->delta_disks
== UnSet
)
1349 info
->delta_disks
= (new_disks
1350 - info
->array
.raid_disks
);
1352 if (info
->delta_disks
!=
1353 new_disks
- info
->array
.raid_disks
)
1354 return "New number of raid-devices impossible for RAID10";
1355 if (info
->new_chunk
&&
1356 info
->new_chunk
!= info
->array
.chunk_size
)
1357 return "Cannot change chunk-size with RAID10 Grow";
1361 re
->before
.data_disks
= new_disks
;
1362 re
->after
.data_disks
= re
->before
.data_disks
;
1366 near
= info
->array
.layout
& 0xff;
1367 far
= (info
->array
.layout
>> 8) & 0xff;
1368 offset
= info
->array
.layout
& 0x10000;
1369 if (far
> 1 && !offset
)
1370 return "Cannot reshape RAID10 in far-mode";
1371 copies
= near
* far
;
1373 old_chunk
= info
->array
.chunk_size
* far
;
1375 if (info
->new_layout
== UnSet
)
1376 info
->new_layout
= info
->array
.layout
;
1378 near
= info
->new_layout
& 0xff;
1379 far
= (info
->new_layout
>> 8) & 0xff;
1380 offset
= info
->new_layout
& 0x10000;
1381 if (far
> 1 && !offset
)
1382 return "Cannot reshape RAID10 to far-mode";
1383 if (near
* far
!= copies
)
1384 return "Cannot change number of copies when reshaping RAID10";
1386 if (info
->delta_disks
== UnSet
)
1387 info
->delta_disks
= 0;
1388 new_disks
= (info
->array
.raid_disks
+
1391 new_chunk
= info
->new_chunk
* far
;
1394 re
->before
.layout
= info
->array
.layout
;
1395 re
->before
.data_disks
= info
->array
.raid_disks
;
1396 re
->after
.layout
= info
->new_layout
;
1397 re
->after
.data_disks
= new_disks
;
1398 /* For RAID10 we don't do backup but do allow reshape,
1399 * so set backup_blocks to INVALID_SECTORS rather than
1401 * And there is no need to synchronise stripes on both
1402 * 'old' and 'new'. So the important
1403 * number is the minimum data_offset difference
1404 * which is the larger of (offset copies * chunk).
1406 re
->backup_blocks
= INVALID_SECTORS
;
1407 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1408 if (new_disks
< re
->before
.data_disks
&&
1409 info
->space_after
< re
->min_offset_change
)
1410 /* Reduce component size by one chunk */
1411 re
->new_size
= (info
->component_size
-
1412 re
->min_offset_change
);
1414 re
->new_size
= info
->component_size
;
1415 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1419 return "RAID10 can only be changed to RAID0";
1422 /* RAID0 can be converted to RAID10, or to RAID456 */
1423 if (info
->new_level
== 10) {
1424 if (info
->new_layout
== UnSet
&&
1425 info
->delta_disks
== UnSet
) {
1426 /* Assume near=2 layout */
1427 info
->new_layout
= 0x102;
1428 info
->delta_disks
= info
->array
.raid_disks
;
1430 if (info
->new_layout
== UnSet
) {
1431 int copies
= 1 + (info
->delta_disks
1432 / info
->array
.raid_disks
);
1433 if (info
->array
.raid_disks
* (copies
-1) !=
1435 return "Impossible number of devices for RAID0->RAID10";
1436 info
->new_layout
= 0x100 + copies
;
1438 if (info
->delta_disks
== UnSet
) {
1439 int copies
= info
->new_layout
& 0xff;
1440 if (info
->new_layout
!= 0x100 + copies
)
1441 return "New layout impossible for RAID0->RAID10";;
1442 info
->delta_disks
= (copies
- 1) *
1443 info
->array
.raid_disks
;
1445 if (info
->new_chunk
&&
1446 info
->new_chunk
!= info
->array
.chunk_size
)
1447 return "Cannot change chunk-size with RAID0->RAID10";
1450 re
->before
.data_disks
= (info
->array
.raid_disks
+
1452 re
->after
.data_disks
= re
->before
.data_disks
;
1453 re
->before
.layout
= info
->new_layout
;
1457 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1458 * a raid4 style layout of the final level.
1460 switch (info
->new_level
) {
1465 re
->before
.layout
= 0;
1470 re
->before
.layout
= ALGORITHM_PARITY_N
;
1471 if (info
->new_layout
== UnSet
)
1472 info
->new_layout
= map_name(r5layout
, "default");
1477 re
->before
.layout
= ALGORITHM_PARITY_N
;
1478 if (info
->new_layout
== UnSet
)
1479 info
->new_layout
= map_name(r6layout
, "default");
1482 return "Impossible level change requested";
1484 re
->before
.data_disks
= info
->array
.raid_disks
;
1485 /* determining 'after' layout happens outside this 'switch' */
1489 info
->array
.layout
= ALGORITHM_PARITY_N
;
1491 switch (info
->new_level
) {
1495 re
->level
= info
->array
.level
;
1496 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1497 re
->before
.layout
= info
->array
.layout
;
1501 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1502 re
->before
.layout
= info
->array
.layout
;
1507 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1508 switch (info
->array
.layout
) {
1509 case ALGORITHM_LEFT_ASYMMETRIC
:
1510 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1512 case ALGORITHM_RIGHT_ASYMMETRIC
:
1513 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1515 case ALGORITHM_LEFT_SYMMETRIC
:
1516 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1518 case ALGORITHM_RIGHT_SYMMETRIC
:
1519 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1521 case ALGORITHM_PARITY_0
:
1522 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1524 case ALGORITHM_PARITY_N
:
1525 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1528 return "Cannot convert an array with this layout";
1532 if (info
->array
.raid_disks
!= 2)
1533 return "Can only convert a 2-device array to RAID1";
1534 if (info
->delta_disks
!= UnSet
&&
1535 info
->delta_disks
!= 0)
1536 return "Cannot set raid_disk when converting RAID5->RAID1";
1538 info
->new_chunk
= 0;
1541 return "Impossible level change requested";
1545 switch (info
->new_level
) {
1551 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1552 re
->before
.layout
= info
->array
.layout
;
1555 return "Impossible level change requested";
1560 /* If we reached here then it looks like a re-stripe is
1561 * happening. We have determined the intermediate level
1562 * and initial raid_disks/layout and stored these in 're'.
1564 * We need to deduce the final layout that can be atomically
1565 * converted to the end state.
1567 switch (info
->new_level
) {
1569 /* We can only get to RAID0 from RAID4 or RAID5
1570 * with appropriate layout and one extra device
1572 if (re
->level
!= 4 && re
->level
!= 5)
1573 return "Cannot covert to RAID0 from this level";
1575 switch (re
->level
) {
1577 re
->before
.layout
= 0;
1578 re
->after
.layout
= 0;
1581 re
->after
.layout
= ALGORITHM_PARITY_N
;
1587 /* We can only get to RAID4 from RAID5 */
1588 if (re
->level
!= 4 && re
->level
!= 5)
1589 return "Cannot convert to RAID4 from this level";
1591 switch (re
->level
) {
1593 re
->after
.layout
= 0;
1596 re
->after
.layout
= ALGORITHM_PARITY_N
;
1602 /* We get to RAID5 from RAID5 or RAID6 */
1603 if (re
->level
!= 5 && re
->level
!= 6)
1604 return "Cannot convert to RAID5 from this level";
1606 switch (re
->level
) {
1608 if (info
->new_layout
== UnSet
)
1609 re
->after
.layout
= re
->before
.layout
;
1611 re
->after
.layout
= info
->new_layout
;
1614 if (info
->new_layout
== UnSet
)
1615 info
->new_layout
= re
->before
.layout
;
1617 /* after.layout needs to be raid6 version of new_layout */
1618 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1619 re
->after
.layout
= ALGORITHM_PARITY_N
;
1622 char *ls
= map_num(r5layout
, info
->new_layout
);
1625 /* Current RAID6 layout has a RAID5
1628 strcat(strcpy(layout
, ls
), "-6");
1629 l
= map_name(r6layout
, layout
);
1631 return "Cannot find RAID6 layout to convert to";
1633 /* Current RAID6 has no equivalent.
1634 * If it is already a '-6' layout we
1635 * can leave it unchanged, else we must
1638 ls
= map_num(r6layout
,
1641 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1642 return "Please specify new layout";
1643 l
= info
->new_layout
;
1645 re
->after
.layout
= l
;
1651 /* We must already be at level 6 */
1653 return "Impossible level change";
1654 if (info
->new_layout
== UnSet
)
1655 re
->after
.layout
= info
->array
.layout
;
1657 re
->after
.layout
= info
->new_layout
;
1660 return "Impossible level change requested";
1662 if (info
->delta_disks
== UnSet
)
1663 info
->delta_disks
= delta_parity
;
1665 re
->after
.data_disks
=
1666 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1668 switch (re
->level
) {
1680 /* So we have a restripe operation, we need to calculate the number
1681 * of blocks per reshape operation.
1683 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1684 if (info
->new_chunk
== 0)
1685 info
->new_chunk
= info
->array
.chunk_size
;
1686 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1687 re
->after
.layout
== re
->before
.layout
&&
1688 info
->new_chunk
== info
->array
.chunk_size
) {
1689 /* Nothing to change, can change level immediately. */
1690 re
->level
= info
->new_level
;
1691 re
->backup_blocks
= 0;
1694 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1695 /* chunk and layout changes make no difference */
1696 re
->level
= info
->new_level
;
1697 re
->backup_blocks
= 0;
1701 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1702 get_linux_version() < 2006032)
1703 return "in-place reshape is not safe before 2.6.32 - sorry.";
1705 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1706 get_linux_version() < 2006030)
1707 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1709 re
->backup_blocks
= compute_backup_blocks(
1710 info
->new_chunk
, info
->array
.chunk_size
,
1711 re
->after
.data_disks
, re
->before
.data_disks
);
1712 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1714 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1718 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1721 struct mdinfo
*info
;
1725 if ((st
== NULL
) || (sra
== NULL
))
1728 if (text_version
== NULL
)
1729 text_version
= sra
->text_version
;
1730 subarray
= strchr(text_version
+ 1, '/')+1;
1731 info
= st
->ss
->container_content(st
, subarray
);
1733 unsigned long long current_size
= 0;
1734 unsigned long long new_size
= info
->custom_array_size
/2;
1736 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1737 new_size
> current_size
) {
1738 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1740 dprintf("Error: Cannot set array size");
1743 dprintf("Array size changed");
1745 dprintf_cont(" from %llu to %llu.\n",
1746 current_size
, new_size
);
1750 dprintf("Error: set_array_size(): info pointer in NULL\n");
1755 static int reshape_array(char *container
, int fd
, char *devname
,
1756 struct supertype
*st
, struct mdinfo
*info
,
1757 int force
, struct mddev_dev
*devlist
,
1758 unsigned long long data_offset
,
1759 char *backup_file
, int verbose
, int forked
,
1760 int restart
, int freeze_reshape
);
1761 static int reshape_container(char *container
, char *devname
,
1763 struct supertype
*st
,
1764 struct mdinfo
*info
,
1766 char *backup_file
, int verbose
,
1767 int forked
, int restart
, int freeze_reshape
);
1769 int Grow_reshape(char *devname
, int fd
,
1770 struct mddev_dev
*devlist
,
1771 unsigned long long data_offset
,
1772 struct context
*c
, struct shape
*s
)
1774 /* Make some changes in the shape of an array.
1775 * The kernel must support the change.
1777 * There are three different changes. Each can trigger
1778 * a resync or recovery so we freeze that until we have
1779 * requested everything (if kernel supports freezing - 2.6.30).
1781 * - change size (i.e. component_size)
1783 * - change layout/chunksize/ndisks
1785 * The last can require a reshape. It is different on different
1786 * levels so we need to check the level before actioning it.
1787 * Some times the level change needs to be requested after the
1788 * reshape (e.g. raid6->raid5, raid5->raid0)
1791 struct mdu_array_info_s array
;
1793 struct supertype
*st
;
1794 char *subarray
= NULL
;
1798 char *container
= NULL
;
1801 struct mddev_dev
*dv
;
1807 if (md_get_array_info(fd
, &array
) < 0) {
1808 pr_err("%s is not an active md array - aborting\n",
1812 if (s
->level
!= UnSet
&& s
->chunk
) {
1813 pr_err("Cannot change array level in the same operation as changing chunk size.\n");
1817 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1818 (array
.level
< 4 || array
.level
> 6)) {
1819 pr_err("--grow --data-offset not yet supported\n");
1824 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1825 pr_err("cannot change component size at the same time as other changes.\n"
1826 " Change size first, then check data is intact before making other changes.\n");
1830 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&&
1831 array
.level
> 1 && get_linux_version() < 2006032 &&
1832 !check_env("MDADM_FORCE_FEWER")) {
1833 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1834 " Please use a newer kernel\n");
1838 if (array
.level
> 1 && s
->size
> 1 &&
1839 (unsigned long long) (array
.chunk_size
/ 1024) > s
->size
) {
1840 pr_err("component size must be larger than chunk size.\n");
1844 st
= super_by_fd(fd
, &subarray
);
1846 pr_err("Unable to determine metadata format for %s\n", devname
);
1849 if (s
->raiddisks
> st
->max_devs
) {
1850 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1853 if (s
->level
== 0 &&
1854 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) &&
1855 !(array
.state
& (1<<MD_SB_CLUSTERED
))) {
1856 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
1857 if (md_set_array_info(fd
, &array
)!= 0) {
1858 pr_err("failed to remove internal bitmap.\n");
1863 /* in the external case we need to check that the requested reshape is
1864 * supported, and perform an initial check that the container holds the
1865 * pre-requisite spare devices (mdmon owns final validation)
1867 if (st
->ss
->external
) {
1871 container
= st
->container_devnm
;
1872 cfd
= open_dev_excl(st
->container_devnm
);
1874 container
= st
->devnm
;
1876 cfd
= open_dev_excl(st
->devnm
);
1880 pr_err("Unable to open container for %s\n", devname
);
1885 retval
= st
->ss
->load_container(st
, cfd
, NULL
);
1888 pr_err("Cannot read superblock for %s\n", devname
);
1893 /* check if operation is supported for metadata handler */
1894 if (st
->ss
->container_content
) {
1895 struct mdinfo
*cc
= NULL
;
1896 struct mdinfo
*content
= NULL
;
1898 cc
= st
->ss
->container_content(st
, subarray
);
1899 for (content
= cc
; content
; content
= content
->next
) {
1900 int allow_reshape
= 1;
1902 /* check if reshape is allowed based on metadata
1903 * indications stored in content.array.status
1905 if (content
->array
.state
&
1906 (1 << MD_SB_BLOCK_VOLUME
))
1908 if (content
->array
.state
&
1909 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE
))
1911 if (!allow_reshape
) {
1912 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1913 devname
, container
);
1918 if (content
->consistency_policy
==
1919 CONSISTENCY_POLICY_PPL
) {
1920 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1928 if (mdmon_running(container
))
1929 st
->update_tail
= &st
->updates
;
1933 for (dv
= devlist
; dv
; dv
= dv
->next
)
1935 if (s
->raiddisks
> array
.raid_disks
&&
1936 array
.spare_disks
+ added_disks
<
1937 (s
->raiddisks
- array
.raid_disks
) &&
1939 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1940 " Use --force to over-ride this check.\n",
1941 s
->raiddisks
- array
.raid_disks
,
1942 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1943 array
.spare_disks
+ added_disks
);
1947 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1948 GET_STATE
| GET_VERSION
);
1950 if (st
->ss
->external
&& subarray
== NULL
) {
1951 array
.level
= LEVEL_CONTAINER
;
1952 sra
->array
.level
= LEVEL_CONTAINER
;
1955 pr_err("failed to read sysfs parameters for %s\n",
1959 frozen
= freeze(st
);
1961 /* freeze() already spewed the reason */
1964 } else if (frozen
< 0) {
1965 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1970 /* ========= set size =============== */
1972 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1973 unsigned long long orig_size
= get_component_size(fd
)/2;
1974 unsigned long long min_csize
;
1976 int raid0_takeover
= 0;
1979 orig_size
= (unsigned) array
.size
;
1981 if (orig_size
== 0) {
1982 pr_err("Cannot set device size in this type of array.\n");
1987 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1988 devname
, APPLY_METADATA_CHANGES
,
1994 if (st
->ss
->external
) {
1995 /* metadata can have size limitation
1996 * update size value according to metadata information
1998 struct mdinfo
*sizeinfo
=
1999 st
->ss
->container_content(st
, subarray
);
2001 unsigned long long new_size
=
2002 sizeinfo
->custom_array_size
/2;
2003 int data_disks
= get_data_disks(
2004 sizeinfo
->array
.level
,
2005 sizeinfo
->array
.layout
,
2006 sizeinfo
->array
.raid_disks
);
2007 new_size
/= data_disks
;
2008 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
2009 orig_size
, new_size
,
2010 new_size
* data_disks
);
2012 sysfs_free(sizeinfo
);
2016 /* Update the size of each member device in case
2017 * they have been resized. This will never reduce
2018 * below the current used-size. The "size" attribute
2019 * understands '0' to mean 'max'.
2022 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2023 sysfs_set_num(sra
, mdi
, "size",
2024 s
->size
== MAX_SIZE
? 0 : s
->size
);
2025 if (array
.not_persistent
== 0 &&
2026 array
.major_version
== 0 &&
2027 get_linux_version() < 3001000) {
2028 /* Dangerous to allow size to exceed 2TB */
2029 unsigned long long csize
;
2030 if (sysfs_get_ll(sra
, mdi
, "size",
2032 if (csize
>= 2ULL*1024*1024*1024)
2033 csize
= 2ULL*1024*1024*1024;
2034 if ((min_csize
== 0 ||
2035 (min_csize
> csize
)))
2040 if (min_csize
&& s
->size
> min_csize
) {
2041 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2043 goto size_change_error
;
2045 if (min_csize
&& s
->size
== MAX_SIZE
) {
2046 /* Don't let the kernel choose a size - it will get
2049 pr_err("Limited v0.90 array to 2TB per device\n");
2050 s
->size
= min_csize
;
2052 if (st
->ss
->external
) {
2053 if (sra
->array
.level
== 0) {
2054 rv
= sysfs_set_str(sra
, NULL
, "level", "raid5");
2057 /* get array parameters after takeover
2058 * to change one parameter at time only
2060 rv
= md_get_array_info(fd
, &array
);
2063 /* make sure mdmon is
2064 * aware of the new level */
2065 if (!mdmon_running(st
->container_devnm
))
2066 start_mdmon(st
->container_devnm
);
2067 ping_monitor(container
);
2068 if (mdmon_running(st
->container_devnm
) &&
2069 st
->update_tail
== NULL
)
2070 st
->update_tail
= &st
->updates
;
2073 if (s
->size
== MAX_SIZE
)
2075 array
.size
= s
->size
;
2076 if (s
->size
& ~INT32_MAX
) {
2077 /* got truncated to 32bit, write to
2078 * component_size instead
2081 rv
= sysfs_set_num(sra
, NULL
,
2082 "component_size", s
->size
);
2086 rv
= md_set_array_info(fd
, &array
);
2088 /* manage array size when it is managed externally
2090 if ((rv
== 0) && st
->ss
->external
)
2091 rv
= set_array_size(st
, sra
, sra
->text_version
);
2094 if (raid0_takeover
) {
2095 /* do not recync non-existing parity,
2096 * we will drop it anyway
2098 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2099 /* go back to raid0, drop parity disk
2101 sysfs_set_str(sra
, NULL
, "level", "raid0");
2102 md_get_array_info(fd
, &array
);
2109 /* restore metadata */
2110 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2111 UnSet
, NULL
, devname
,
2112 ROLLBACK_METADATA_CHANGES
,
2115 pr_err("Cannot set device size for %s: %s\n",
2116 devname
, strerror(err
));
2118 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2119 cont_err("Bitmap must be removed before size can be changed\n");
2123 if (s
->assume_clean
) {
2124 /* This will fail on kernels older than 3.0 unless
2125 * a backport has been arranged.
2128 sysfs_set_str(sra
, NULL
, "resync_start",
2130 pr_err("--assume-clean not supported with --grow on this kernel\n");
2132 md_get_array_info(fd
, &array
);
2133 s
->size
= get_component_size(fd
)/2;
2135 s
->size
= array
.size
;
2136 if (c
->verbose
>= 0) {
2137 if (s
->size
== orig_size
)
2138 pr_err("component size of %s unchanged at %lluK\n",
2141 pr_err("component size of %s has been set to %lluK\n",
2145 } else if (array
.level
!= LEVEL_CONTAINER
) {
2146 s
->size
= get_component_size(fd
)/2;
2148 s
->size
= array
.size
;
2151 /* See if there is anything else to do */
2152 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2153 (s
->layout_str
== NULL
) &&
2154 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2155 data_offset
== INVALID_SECTORS
&&
2156 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2157 /* Nothing more to do */
2158 if (!changed
&& c
->verbose
>= 0)
2159 pr_err("%s: no change requested\n", devname
);
2163 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2164 * current implementation assumes that following conditions must be met:
2167 * - near_copies == 2
2169 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2170 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2171 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2174 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2176 dprintf("Array cannot be reshaped\n");
2182 /* Make sure mdmon has seen the device removal
2183 * and updated metadata before we continue with
2187 ping_monitor(container
);
2190 memset(&info
, 0, sizeof(info
));
2192 if (sysfs_init(&info
, fd
, NULL
)) {
2193 pr_err("failed to initialize sysfs.\n");
2197 strcpy(info
.text_version
, sra
->text_version
);
2198 info
.component_size
= s
->size
*2;
2199 info
.new_level
= s
->level
;
2200 info
.new_chunk
= s
->chunk
* 1024;
2201 if (info
.array
.level
== LEVEL_CONTAINER
) {
2202 info
.delta_disks
= UnSet
;
2203 info
.array
.raid_disks
= s
->raiddisks
;
2204 } else if (s
->raiddisks
)
2205 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2207 info
.delta_disks
= UnSet
;
2208 if (s
->layout_str
== NULL
) {
2209 info
.new_layout
= UnSet
;
2210 if (info
.array
.level
== 6 &&
2211 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2212 info
.array
.layout
>= 16) {
2213 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2214 cont_err("during the reshape, please specify --layout=preserve\n");
2215 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2219 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2220 strcmp(s
->layout_str
, "normalize") == 0) {
2221 /* If we have a -6 RAID6 layout, remove the '-6'. */
2222 info
.new_layout
= UnSet
;
2223 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2225 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
2226 h
= strrchr(l
, '-');
2227 if (h
&& strcmp(h
, "-6") == 0) {
2229 info
.new_layout
= map_name(r6layout
, l
);
2232 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2236 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2237 /* This means that a non-standard RAID6 layout
2240 * - When reshape a RAID6 (e.g. adding a device)
2241 * which is in a non-standard layout, it is OK
2242 * to preserve that layout.
2243 * - When converting a RAID5 to RAID6, leave it in
2244 * the XXX-6 layout, don't re-layout.
2246 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2247 info
.new_layout
= info
.array
.layout
;
2248 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2250 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
2252 info
.new_layout
= map_name(r6layout
, l
);
2254 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2259 int l
= info
.new_level
;
2261 l
= info
.array
.level
;
2264 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2267 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2270 info
.new_layout
= parse_layout_10(s
->layout_str
);
2273 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2276 pr_err("layout not meaningful with this level\n");
2280 if (info
.new_layout
== UnSet
) {
2281 pr_err("layout %s not understood for this level\n",
2288 if (array
.level
== LEVEL_FAULTY
) {
2289 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2290 pr_err("cannot change level of Faulty device\n");
2294 pr_err("cannot set chunksize of Faulty device\n");
2297 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2298 pr_err("cannot set raid_disks of Faulty device\n");
2301 if (s
->layout_str
) {
2302 if (md_get_array_info(fd
, &array
) != 0) {
2303 dprintf("Cannot get array information.\n");
2306 array
.layout
= info
.new_layout
;
2307 if (md_set_array_info(fd
, &array
) != 0) {
2308 pr_err("failed to set new layout\n");
2310 } else if (c
->verbose
>= 0)
2311 printf("layout for %s set to %d\n",
2312 devname
, array
.layout
);
2314 } else if (array
.level
== LEVEL_CONTAINER
) {
2315 /* This change is to be applied to every array in the
2316 * container. This is only needed when the metadata imposes
2317 * restraints of the various arrays in the container.
2318 * Currently we only know that IMSM requires all arrays
2319 * to have the same number of devices so changing the
2320 * number of devices (On-Line Capacity Expansion) must be
2321 * performed at the level of the container
2327 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2328 c
->force
, c
->backup_file
, c
->verbose
,
2332 /* get spare devices from external metadata
2334 if (st
->ss
->external
) {
2335 struct mdinfo
*info2
;
2337 info2
= st
->ss
->container_content(st
, subarray
);
2339 info
.array
.spare_disks
=
2340 info2
->array
.spare_disks
;
2345 /* Impose these changes on a single array. First
2346 * check that the metadata is OK with the change. */
2348 if (reshape_super(st
, 0, info
.new_level
,
2349 info
.new_layout
, info
.new_chunk
,
2350 info
.array
.raid_disks
, info
.delta_disks
,
2351 c
->backup_file
, devname
,
2352 APPLY_METADATA_CHANGES
, c
->verbose
)) {
2357 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2358 devlist
, data_offset
, c
->backup_file
,
2359 c
->verbose
, 0, 0, 0);
2369 /* verify_reshape_position()
2370 * Function checks if reshape position in metadata is not farther
2371 * than position in md.
2373 * 0 : not valid sysfs entry
2374 * it can be caused by not started reshape, it should be started
2375 * by reshape array or raid0 array is before takeover
2376 * -1 : error, reshape position is obviously wrong
2377 * 1 : success, reshape progress correct or updated
2379 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2385 /* read sync_max, failure can mean raid0 array */
2386 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2390 unsigned long long position
= strtoull(buf
, &ep
, 0);
2392 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2393 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2394 position
*= get_data_disks(level
,
2396 info
->array
.raid_disks
);
2397 if (info
->reshape_progress
< position
) {
2398 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2399 info
->reshape_progress
, position
);
2400 info
->reshape_progress
= position
;
2402 } else if (info
->reshape_progress
> position
) {
2403 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2404 position
, info
->reshape_progress
);
2407 dprintf("Reshape position in md and metadata are the same;");
2411 } else if (rv
== 0) {
2412 /* for valid sysfs entry, 0-length content
2413 * should be indicated as error
2421 static unsigned long long choose_offset(unsigned long long lo
,
2422 unsigned long long hi
,
2423 unsigned long long min
,
2424 unsigned long long max
)
2426 /* Choose a new offset between hi and lo.
2427 * It must be between min and max, but
2428 * we would prefer something near the middle of hi/lo, and also
2429 * prefer to be aligned to a big power of 2.
2431 * So we start with the middle, then for each bit,
2432 * starting at '1' and increasing, if it is set, we either
2433 * add it or subtract it if possible, preferring the option
2434 * which is furthest from the boundary.
2436 * We stop once we get a 1MB alignment. As units are in sectors,
2437 * 1MB = 2*1024 sectors.
2439 unsigned long long choice
= (lo
+ hi
) / 2;
2440 unsigned long long bit
= 1;
2442 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2443 unsigned long long bigger
, smaller
;
2444 if (! (bit
& choice
))
2446 bigger
= choice
+ bit
;
2447 smaller
= choice
- bit
;
2448 if (bigger
> max
&& smaller
< min
)
2452 else if (smaller
< min
)
2454 else if (hi
- bigger
> smaller
- lo
)
2462 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2463 char *devname
, int delta_disks
,
2464 unsigned long long data_offset
,
2465 unsigned long long min
,
2471 unsigned long long before
, after
;
2473 /* Need to find min space before and after so same is used
2476 before
= UINT64_MAX
;
2478 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2482 struct supertype
*st2
;
2483 struct mdinfo info2
;
2485 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2487 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2488 dfd
= dev_open(dn
, O_RDONLY
);
2490 pr_err("%s: cannot open component %s\n",
2491 devname
, dn
? dn
: "-unknown-");
2494 st2
= dup_super(st
);
2495 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2499 pr_err("%s: cannot get superblock from %s\n",
2503 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2504 st2
->ss
->free_super(st2
);
2506 if (info2
.space_before
== 0 &&
2507 info2
.space_after
== 0) {
2508 /* Metadata doesn't support data_offset changes */
2510 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2514 if (before
> info2
.space_before
)
2515 before
= info2
.space_before
;
2516 if (after
> info2
.space_after
)
2517 after
= info2
.space_after
;
2519 if (data_offset
!= INVALID_SECTORS
) {
2521 if (info2
.data_offset
== data_offset
) {
2522 pr_err("%s: already has that data_offset\n",
2526 if (data_offset
< info2
.data_offset
)
2530 } else if ((data_offset
<= info2
.data_offset
&&
2532 (data_offset
>= info2
.data_offset
&&
2534 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2540 if (before
== UINT64_MAX
)
2541 /* impossible really, there must be no devices */
2544 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2545 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2546 unsigned long long new_data_offset
;
2548 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2550 if (delta_disks
< 0) {
2551 /* Don't need any space as array is shrinking
2552 * just move data_offset up by min
2554 if (data_offset
== INVALID_SECTORS
)
2555 new_data_offset
= sd
->data_offset
+ min
;
2557 if (data_offset
< sd
->data_offset
+ min
) {
2558 pr_err("--data-offset too small for %s\n",
2562 new_data_offset
= data_offset
;
2564 } else if (delta_disks
> 0) {
2565 /* need space before */
2569 pr_err("Insufficient head-space for reshape on %s\n",
2573 if (data_offset
== INVALID_SECTORS
)
2574 new_data_offset
= sd
->data_offset
- min
;
2576 if (data_offset
> sd
->data_offset
- min
) {
2577 pr_err("--data-offset too large for %s\n",
2581 new_data_offset
= data_offset
;
2585 /* can move up or down. If 'data_offset'
2586 * was set we would have already decided,
2587 * so just choose direction with most space.
2594 sysfs_set_str(sra
, NULL
, "reshape_direction",
2595 dir
== 1 ? "backwards" : "forwards");
2597 /* Increase data offset */
2601 pr_err("Insufficient tail-space for reshape on %s\n",
2605 if (data_offset
!= INVALID_SECTORS
&&
2606 data_offset
< sd
->data_offset
+ min
) {
2607 pr_err("--data-offset too small on %s\n",
2611 if (data_offset
!= INVALID_SECTORS
)
2612 new_data_offset
= data_offset
;
2614 new_data_offset
= choose_offset(sd
->data_offset
,
2615 sd
->data_offset
+ after
,
2616 sd
->data_offset
+ min
,
2617 sd
->data_offset
+ after
);
2619 /* Decrease data offset */
2623 pr_err("insufficient head-room on %s\n",
2627 if (data_offset
!= INVALID_SECTORS
&&
2628 data_offset
> sd
->data_offset
- min
) {
2629 pr_err("--data-offset too large on %s\n",
2633 if (data_offset
!= INVALID_SECTORS
)
2634 new_data_offset
= data_offset
;
2636 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2638 sd
->data_offset
- before
,
2639 sd
->data_offset
- min
);
2642 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2643 if (err
< 0 && errno
== E2BIG
) {
2644 /* try again after increasing data size to max */
2645 err
= sysfs_set_num(sra
, sd
, "size", 0);
2646 if (err
< 0 && errno
== EINVAL
&&
2647 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2648 /* some kernels have a bug where you cannot
2649 * use '0' on spare devices. */
2650 sysfs_set_num(sra
, sd
, "size",
2651 (sra
->component_size
+ after
)/2);
2653 err
= sysfs_set_num(sra
, sd
, "new_offset",
2657 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2658 pr_err("data-offset is too big for %s\n", dn
);
2661 if (sd
== sra
->devs
&&
2662 (errno
== ENOENT
|| errno
== E2BIG
))
2663 /* Early kernel, no 'new_offset' file,
2664 * or kernel doesn't like us.
2665 * For RAID5/6 this is not fatal
2668 pr_err("Cannot set new_offset for %s\n", dn
);
2676 /* Just use a backup file */
2680 static int raid10_reshape(char *container
, int fd
, char *devname
,
2681 struct supertype
*st
, struct mdinfo
*info
,
2682 struct reshape
*reshape
,
2683 unsigned long long data_offset
,
2684 int force
, int verbose
)
2686 /* Changing raid_disks, layout, chunksize or possibly
2687 * just data_offset for a RAID10.
2688 * We must always change data_offset. We change by at least
2689 * ->min_offset_change which is the largest of the old and new
2691 * If raid_disks is increasing, then data_offset must decrease
2692 * by at least this copy size.
2693 * If raid_disks is unchanged, data_offset must increase or
2694 * decrease by at least min_offset_change but preferably by much more.
2695 * We choose half of the available space.
2696 * If raid_disks is decreasing, data_offset must increase by
2697 * at least min_offset_change. To allow of this, component_size
2698 * must be decreased by the same amount.
2700 * So we calculate the required minimum and direction, possibly
2701 * reduce the component_size, then iterate through the devices
2702 * and set the new_data_offset.
2703 * If that all works, we set chunk_size, layout, raid_disks, and start
2707 unsigned long long min
;
2710 sra
= sysfs_read(fd
, NULL
,
2711 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2714 pr_err("%s: Cannot get array details from sysfs\n", devname
);
2717 min
= reshape
->min_offset_change
;
2719 if (info
->delta_disks
)
2720 sysfs_set_str(sra
, NULL
, "reshape_direction",
2721 info
->delta_disks
< 0 ? "backwards" : "forwards");
2722 if (info
->delta_disks
< 0 && info
->space_after
< min
) {
2723 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2724 (sra
->component_size
- min
)/2);
2726 pr_err("cannot reduce component size\n");
2730 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
,
2731 data_offset
, min
, 0);
2733 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2734 cont_err("supported on this kernel\n");
2740 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2742 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2743 reshape
->after
.layout
) < 0)
2746 sysfs_set_num(sra
, NULL
, "raid_disks",
2747 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2749 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2752 pr_err("Cannot set array shape for %s\n",
2755 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2756 cont_err(" Bitmap must be removed before shape can be changed\n");
2766 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2768 struct mdinfo
*sra
, *sd
;
2769 /* Initialisation to silence compiler warning */
2770 unsigned long long min_space_before
= 0, min_space_after
= 0;
2773 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2776 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2779 struct supertype
*st2
;
2780 struct mdinfo info2
;
2782 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2784 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2785 dfd
= dev_open(dn
, O_RDONLY
);
2788 st2
= dup_super(st
);
2789 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2795 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2796 st2
->ss
->free_super(st2
);
2799 min_space_before
> info2
.space_before
)
2800 min_space_before
= info2
.space_before
;
2802 min_space_after
> info2
.space_after
)
2803 min_space_after
= info2
.space_after
;
2806 if (sd
== NULL
&& !first
) {
2807 info
->space_after
= min_space_after
;
2808 info
->space_before
= min_space_before
;
2813 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2814 struct mdinfo
*info
,
2815 int disks
, unsigned long long blocks
)
2817 /* Check that the internal stripe cache is
2818 * large enough, or it won't work.
2819 * It must hold at least 4 stripes of the larger
2822 unsigned long cache
;
2823 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2824 cache
*= 4; /* 4 stripes minimum */
2825 cache
/= 512; /* convert to sectors */
2826 /* make sure there is room for 'blocks' with a bit to spare */
2827 if (cache
< 16 + blocks
/ disks
)
2828 cache
= 16 + blocks
/ disks
;
2829 cache
/= (4096/512); /* Convert from sectors to pages */
2831 if (sra
->cache_size
< cache
)
2832 subarray_set_num(container
, sra
, "stripe_cache_size",
2836 static int impose_reshape(struct mdinfo
*sra
,
2837 struct mdinfo
*info
,
2838 struct supertype
*st
,
2841 char *devname
, char *container
,
2842 struct reshape
*reshape
)
2844 struct mdu_array_info_s array
;
2846 sra
->new_chunk
= info
->new_chunk
;
2849 /* for external metadata checkpoint saved by mdmon can be lost
2850 * or missed /due to e.g. crash/. Check if md is not during
2851 * restart farther than metadata points to.
2852 * If so, this means metadata information is obsolete.
2854 if (st
->ss
->external
)
2855 verify_reshape_position(info
, reshape
->level
);
2856 sra
->reshape_progress
= info
->reshape_progress
;
2858 sra
->reshape_progress
= 0;
2859 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2860 /* start from the end of the new array */
2861 sra
->reshape_progress
= (sra
->component_size
2862 * reshape
->after
.data_disks
);
2865 md_get_array_info(fd
, &array
);
2866 if (info
->array
.chunk_size
== info
->new_chunk
&&
2867 reshape
->before
.layout
== reshape
->after
.layout
&&
2868 st
->ss
->external
== 0) {
2869 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2870 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2871 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2874 pr_err("Cannot set device shape for %s: %s\n",
2875 devname
, strerror(errno
));
2878 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2879 cont_err("Bitmap must be removed before shape can be changed\n");
2883 } else if (!restart
) {
2884 /* set them all just in case some old 'new_*' value
2885 * persists from some earlier problem.
2888 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2890 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2891 reshape
->after
.layout
) < 0)
2893 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2894 reshape
->after
.data_disks
+
2895 reshape
->parity
) < 0)
2898 pr_err("Cannot set device shape for %s\n", devname
);
2901 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2902 cont_err("Bitmap must be removed before shape can be changed\n");
2911 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2914 struct mdu_array_info_s array
;
2917 if (sysfs_init(&info
, fd
, NULL
)) {
2918 pr_err("failed to initialize sysfs.\n");
2922 md_get_array_info(fd
, &array
);
2923 if (level
== 0 && (array
.level
>= 4 && array
.level
<= 6)) {
2924 /* To convert to RAID0 we need to fail and
2925 * remove any non-data devices. */
2928 int data_disks
= array
.raid_disks
- 1;
2929 if (array
.level
== 6)
2931 if (array
.level
== 5 && array
.layout
!= ALGORITHM_PARITY_N
)
2933 if (array
.level
== 6 && array
.layout
!= ALGORITHM_PARITY_N_6
)
2935 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2936 /* First remove any spares so no recovery starts */
2937 for (d
= 0, found
= 0;
2938 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2939 mdu_disk_info_t disk
;
2941 if (md_get_disk_info(fd
, &disk
) < 0)
2943 if (disk
.major
== 0 && disk
.minor
== 0)
2946 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2947 disk
.raid_disk
< data_disks
)
2950 ioctl(fd
, HOT_REMOVE_DISK
,
2951 makedev(disk
.major
, disk
.minor
));
2953 /* Now fail anything left */
2954 md_get_array_info(fd
, &array
);
2955 for (d
= 0, found
= 0;
2956 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2957 mdu_disk_info_t disk
;
2959 if (md_get_disk_info(fd
, &disk
) < 0)
2961 if (disk
.major
== 0 && disk
.minor
== 0)
2964 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2965 disk
.raid_disk
< data_disks
)
2968 ioctl(fd
, SET_DISK_FAULTY
,
2969 makedev(disk
.major
, disk
.minor
));
2970 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2973 c
= map_num(pers
, level
);
2975 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2978 pr_err("%s: could not set level to %s\n",
2981 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2982 cont_err("Bitmap must be removed before level can be changed\n");
2986 pr_err("level of %s changed to %s\n", devname
, c
);
2992 static void catch_term(int sig
)
2997 static int reshape_array(char *container
, int fd
, char *devname
,
2998 struct supertype
*st
, struct mdinfo
*info
,
2999 int force
, struct mddev_dev
*devlist
,
3000 unsigned long long data_offset
,
3001 char *backup_file
, int verbose
, int forked
,
3002 int restart
, int freeze_reshape
)
3004 struct reshape reshape
;
3007 int orig_level
= UnSet
;
3011 struct mdu_array_info_s array
;
3014 struct mddev_dev
*dv
;
3018 unsigned long long *offsets
= NULL
;
3022 unsigned long blocks
;
3023 unsigned long long array_size
;
3025 struct mdinfo
*sra
= NULL
;
3028 /* when reshaping a RAID0, the component_size might be zero.
3029 * So try to fix that up.
3031 if (md_get_array_info(fd
, &array
) != 0) {
3032 dprintf("Cannot get array information.\n");
3035 if (array
.level
== 0 && info
->component_size
== 0) {
3036 get_dev_size(fd
, NULL
, &array_size
);
3037 info
->component_size
= array_size
/ array
.raid_disks
;
3040 if (array
.level
== 10)
3041 /* Need space_after info */
3042 get_space_after(fd
, st
, info
);
3044 if (info
->reshape_active
) {
3045 int new_level
= info
->new_level
;
3046 info
->new_level
= UnSet
;
3047 if (info
->delta_disks
> 0)
3048 info
->array
.raid_disks
-= info
->delta_disks
;
3049 msg
= analyse_change(devname
, info
, &reshape
);
3050 info
->new_level
= new_level
;
3051 if (info
->delta_disks
> 0)
3052 info
->array
.raid_disks
+= info
->delta_disks
;
3054 /* Make sure the array isn't read-only */
3055 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3057 msg
= analyse_change(devname
, info
, &reshape
);
3059 /* if msg == "", error has already been printed */
3061 pr_err("%s\n", msg
);
3064 if (restart
&& (reshape
.level
!= info
->array
.level
||
3065 reshape
.before
.layout
!= info
->array
.layout
||
3066 reshape
.before
.data_disks
+ reshape
.parity
!=
3067 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3068 pr_err("reshape info is not in native format - cannot continue.\n");
3072 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3073 !((sysfs_get_str(info
, NULL
, "sync_action",
3074 buf
, sizeof(buf
)) > 0) &&
3075 (strncmp(buf
, "reshape", 7) == 0))) {
3076 /* When reshape is restarted from '0', very begin of array
3077 * it is possible that for external metadata reshape and array
3078 * configuration doesn't happen.
3079 * Check if md has the same opinion, and reshape is restarted
3080 * from 0. If so, this is regular reshape start after reshape
3081 * switch in metadata to next array only.
3083 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3084 (info
->reshape_progress
== 0))
3089 * reshape already started. just skip to monitoring
3092 if (reshape
.backup_blocks
== 0)
3094 if (restart
& RESHAPE_NO_BACKUP
)
3097 /* Need 'sra' down at 'started:' */
3098 sra
= sysfs_read(fd
, NULL
,
3099 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
3100 GET_CHUNK
|GET_CACHE
);
3102 pr_err("%s: Cannot get array details from sysfs\n",
3108 backup_file
= locate_backup(sra
->sys_name
);
3112 /* The container is frozen but the array may not be.
3113 * So freeze the array so spares don't get put to the wrong use
3114 * FIXME there should probably be a cleaner separation between
3115 * freeze_array and freeze_container.
3117 sysfs_freeze_array(info
);
3118 /* Check we have enough spares to not be degraded */
3120 for (dv
= devlist
; dv
; dv
=dv
->next
)
3122 spares_needed
= max(reshape
.before
.data_disks
,
3123 reshape
.after
.data_disks
) +
3124 reshape
.parity
- array
.raid_disks
;
3126 if (!force
&& info
->new_level
> 1 && info
->array
.level
> 1 &&
3127 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3128 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3129 " Use --force to over-ride this check.\n",
3131 spares_needed
== 1 ? "" : "s",
3132 info
->array
.spare_disks
+ added_disks
);
3135 /* Check we have enough spares to not fail */
3136 spares_needed
= max(reshape
.before
.data_disks
,
3137 reshape
.after
.data_disks
)
3139 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3140 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3141 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3142 spares_needed
, spares_needed
== 1 ? "" : "s",
3143 info
->array
.spare_disks
+ added_disks
);
3147 if (reshape
.level
!= array
.level
) {
3148 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3151 info
->new_layout
= UnSet
; /* after level change,
3152 * layout is meaningless */
3153 orig_level
= array
.level
;
3154 sysfs_freeze_array(info
);
3156 if (reshape
.level
> 0 && st
->ss
->external
) {
3157 /* make sure mdmon is aware of the new level */
3158 if (mdmon_running(container
))
3159 flush_mdmon(container
);
3161 if (!mdmon_running(container
))
3162 start_mdmon(container
);
3163 ping_monitor(container
);
3164 if (mdmon_running(container
) && st
->update_tail
== NULL
)
3165 st
->update_tail
= &st
->updates
;
3168 /* ->reshape_super might have chosen some spares from the
3169 * container that it wants to be part of the new array.
3170 * We can collect them with ->container_content and give
3171 * them to the kernel.
3173 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3174 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3175 struct mdinfo
*info2
=
3176 st
->ss
->container_content(st
, subarray
);
3180 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3181 pr_err("unable to initialize sysfs for %s\n",
3186 /* When increasing number of devices, we need to set
3187 * new raid_disks before adding these, or they might
3190 if (reshape
.backup_blocks
&&
3191 reshape
.after
.data_disks
>
3192 reshape
.before
.data_disks
)
3193 subarray_set_num(container
, info2
, "raid_disks",
3194 reshape
.after
.data_disks
+
3196 for (d
= info2
->devs
; d
; d
= d
->next
) {
3197 if (d
->disk
.state
== 0 &&
3198 d
->disk
.raid_disk
>= 0) {
3199 /* This is a spare that wants to
3200 * be part of the array.
3202 add_disk(fd
, st
, info2
, d
);
3208 /* We might have been given some devices to add to the
3209 * array. Now that the array has been changed to the right
3210 * level and frozen, we can safely add them.
3213 if (Manage_subdevs(devname
, fd
, devlist
, verbose
, 0, NULL
, 0))
3217 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3218 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3219 if (reshape
.backup_blocks
== 0) {
3220 /* No restriping needed, but we might need to impose
3221 * some more changes: layout, raid_disks, chunk_size
3223 /* read current array info */
3224 if (md_get_array_info(fd
, &array
) != 0) {
3225 dprintf("Cannot get array information.\n");
3228 /* compare current array info with new values and if
3229 * it is different update them to new */
3230 if (info
->new_layout
!= UnSet
&&
3231 info
->new_layout
!= array
.layout
) {
3232 array
.layout
= info
->new_layout
;
3233 if (md_set_array_info(fd
, &array
) != 0) {
3234 pr_err("failed to set new layout\n");
3236 } else if (verbose
>= 0)
3237 printf("layout for %s set to %d\n",
3238 devname
, array
.layout
);
3240 if (info
->delta_disks
!= UnSet
&& info
->delta_disks
!= 0 &&
3242 (info
->array
.raid_disks
+ info
->delta_disks
)) {
3243 array
.raid_disks
+= info
->delta_disks
;
3244 if (md_set_array_info(fd
, &array
) != 0) {
3245 pr_err("failed to set raid disks\n");
3247 } else if (verbose
>= 0) {
3248 printf("raid_disks for %s set to %d\n",
3249 devname
, array
.raid_disks
);
3252 if (info
->new_chunk
!= 0 &&
3253 info
->new_chunk
!= array
.chunk_size
) {
3254 if (sysfs_set_num(info
, NULL
,
3255 "chunk_size", info
->new_chunk
) != 0) {
3256 pr_err("failed to set chunk size\n");
3258 } else if (verbose
>= 0)
3259 printf("chunk size for %s set to %d\n",
3260 devname
, info
->new_chunk
);
3267 * There are three possibilities.
3268 * 1/ The array will shrink.
3269 * We need to ensure the reshape will pause before reaching
3270 * the 'critical section'. We also need to fork and wait for
3271 * that to happen. When it does we
3272 * suspend/backup/complete/unfreeze
3274 * 2/ The array will not change size.
3275 * This requires that we keep a backup of a sliding window
3276 * so that we can restore data after a crash. So we need
3277 * to fork and monitor progress.
3278 * In future we will allow the data_offset to change, so
3279 * a sliding backup becomes unnecessary.
3281 * 3/ The array will grow. This is relatively easy.
3282 * However the kernel's restripe routines will cheerfully
3283 * overwrite some early data before it is safe. So we
3284 * need to make a backup of the early parts of the array
3285 * and be ready to restore it if rebuild aborts very early.
3286 * For externally managed metadata, we still need a forked
3287 * child to monitor the reshape and suspend IO over the region
3288 * that is being reshaped.
3290 * We backup data by writing it to one spare, or to a
3291 * file which was given on command line.
3293 * In each case, we first make sure that storage is available
3294 * for the required backup.
3296 * - request the shape change.
3297 * - fork to handle backup etc.
3299 /* Check that we can hold all the data */
3300 get_dev_size(fd
, NULL
, &array_size
);
3301 if (reshape
.new_size
< (array_size
/512)) {
3302 pr_err("this change will reduce the size of the array.\n"
3303 " use --grow --array-size first to truncate array.\n"
3304 " e.g. mdadm --grow %s --array-size %llu\n",
3305 devname
, reshape
.new_size
/2);
3309 if (array
.level
== 10) {
3310 /* Reshaping RAID10 does not require any data backup by
3311 * user-space. Instead it requires that the data_offset
3312 * is changed to avoid the need for backup.
3313 * So this is handled very separately
3316 /* Nothing to do. */
3318 return raid10_reshape(container
, fd
, devname
, st
, info
,
3319 &reshape
, data_offset
, force
, verbose
);
3321 sra
= sysfs_read(fd
, NULL
,
3322 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3325 pr_err("%s: Cannot get array details from sysfs\n",
3331 switch(set_new_data_offset(sra
, st
, devname
,
3332 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3334 reshape
.min_offset_change
, 1)) {
3338 /* Updated data_offset, so it's easy now */
3339 update_cache_size(container
, sra
, info
,
3340 min(reshape
.before
.data_disks
,
3341 reshape
.after
.data_disks
),
3342 reshape
.backup_blocks
);
3344 /* Right, everything seems fine. Let's kick things off.
3348 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3349 devname
, container
, &reshape
) < 0)
3351 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3353 if (errno
!= EINVAL
) {
3354 pr_err("Failed to initiate reshape!\n");
3357 /* revert data_offset and try the old way */
3358 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3359 sysfs_set_num(sra
, sd
, "new_offset",
3361 sysfs_set_str(sra
, NULL
, "reshape_direction",
3366 if (info
->new_level
== reshape
.level
)
3368 /* need to adjust level when reshape completes */
3370 case -1: /* ignore error, but don't wait */
3372 default: /* parent */
3381 fd
= open_dev(sra
->sys_name
);
3383 impose_level(fd
, info
->new_level
, devname
, verbose
);
3385 case 1: /* Couldn't set data_offset, try the old way */
3386 if (data_offset
!= INVALID_SECTORS
) {
3387 pr_err("Cannot update data_offset on this array\n");
3394 /* Decide how many blocks (sectors) for a reshape
3395 * unit. The number we have so far is just a minimum
3397 blocks
= reshape
.backup_blocks
;
3398 if (reshape
.before
.data_disks
==
3399 reshape
.after
.data_disks
) {
3400 /* Make 'blocks' bigger for better throughput, but
3401 * not so big that we reject it below.
3402 * Try for 16 megabytes
3404 while (blocks
* 32 < sra
->component_size
&& blocks
< 16*1024*2)
3407 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3409 if (blocks
>= sra
->component_size
/2) {
3410 pr_err("%s: Something wrong - reshape aborted\n", devname
);
3414 /* Now we need to open all these devices so we can read/write.
3416 nrdisks
= max(reshape
.before
.data_disks
,
3417 reshape
.after
.data_disks
) + reshape
.parity
3418 + sra
->array
.spare_disks
;
3419 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3420 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3422 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3423 d
= reshape_prepare_fdlist(devname
, sra
, odisks
, nrdisks
, blocks
,
3424 backup_file
, fdlist
, offsets
);
3428 if ((st
->ss
->manage_reshape
== NULL
) ||
3429 (st
->ss
->recover_backup
== NULL
)) {
3430 if (backup_file
== NULL
) {
3431 if (reshape
.after
.data_disks
<=
3432 reshape
.before
.data_disks
) {
3433 pr_err("%s: Cannot grow - need backup-file\n",
3435 pr_err(" Please provide one with \"--backup=...\"\n");
3437 } else if (d
== odisks
) {
3438 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3442 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3444 fdlist
+d
, offsets
+d
,
3445 sra
->sys_name
, restart
)) {
3452 update_cache_size(container
, sra
, info
,
3453 min(reshape
.before
.data_disks
,
3454 reshape
.after
.data_disks
), blocks
);
3456 /* Right, everything seems fine. Let's kick things off.
3457 * If only changing raid_disks, use ioctl, else use
3462 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3463 devname
, container
, &reshape
) < 0)
3466 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3467 reshape
.after
.data_disks
);
3469 pr_err("Cannot %s reshape for %s\n",
3470 restart
? "continue" : "start", devname
);
3474 sysfs_set_str(sra
, NULL
, "array_state", "active");
3475 if (freeze_reshape
) {
3479 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3480 sra
->reshape_progress
);
3485 if (continue_via_systemd(container
?: sra
->sys_name
,
3494 /* Now we just need to kick off the reshape and watch, while
3495 * handling backups of the data...
3496 * This is all done by a forked background process.
3498 switch(forked
? 0 : fork()) {
3500 pr_err("Cannot run child to monitor reshape: %s\n",
3514 /* If another array on the same devices is busy, the
3515 * reshape will wait for them. This would mean that
3516 * the first section that we suspend will stay suspended
3517 * for a long time. So check on that possibility
3518 * by looking for "DELAYED" in /proc/mdstat, and if found,
3522 struct mdstat_ent
*mds
, *m
;
3524 mds
= mdstat_read(1, 0);
3525 for (m
= mds
; m
; m
= m
->next
)
3526 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3527 if (m
->resync
&& m
->percent
== RESYNC_DELAYED
)
3530 /* Haven't started the reshape thread
3537 if (delayed
== 1 && get_linux_version() < 3007000) {
3538 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3539 " You might experience problems until other reshapes complete.\n");
3543 mdstat_wait(30 - (delayed
-1) * 25);
3546 if (check_env("MDADM_GROW_VERIFY"))
3547 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3550 mlockall(MCL_FUTURE
);
3552 signal(SIGTERM
, catch_term
);
3554 if (st
->ss
->external
) {
3555 /* metadata handler takes it from here */
3556 done
= st
->ss
->manage_reshape(
3557 fd
, sra
, &reshape
, st
, blocks
,
3558 fdlist
, offsets
, d
- odisks
, fdlist
+ odisks
,
3561 done
= child_monitor(
3562 fd
, sra
, &reshape
, st
, blocks
, fdlist
, offsets
,
3563 d
- odisks
, fdlist
+ odisks
, offsets
+ odisks
);
3568 if (backup_file
&& done
) {
3570 bul
= make_backup(sra
->sys_name
);
3573 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3581 unlink(backup_file
);
3588 if (!st
->ss
->external
&&
3589 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3590 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3592 /* no need to wait for the reshape to finish as
3593 * there is nothing more to do.
3600 if (st
->ss
->external
) {
3601 /* Re-load the metadata as much could have changed */
3602 int cfd
= open_dev(st
->container_devnm
);
3604 flush_mdmon(container
);
3605 st
->ss
->free_super(st
);
3606 st
->ss
->load_container(st
, cfd
, container
);
3611 /* set new array size if required customer_array_size is used
3614 if (reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3615 info
->custom_array_size
)
3616 set_array_size(st
, info
, info
->text_version
);
3618 if (info
->new_level
!= reshape
.level
) {
3620 fd
= open(devname
, O_RDONLY
);
3621 impose_level(fd
, info
->new_level
, devname
, verbose
);
3623 if (info
->new_level
== 0)
3624 st
->update_tail
= NULL
;
3636 if (orig_level
!= UnSet
&& sra
) {
3637 c
= map_num(pers
, orig_level
);
3638 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3639 pr_err("aborting level change\n");
3647 /* mdfd handle is passed to be closed in child process (after fork).
3649 int reshape_container(char *container
, char *devname
,
3651 struct supertype
*st
,
3652 struct mdinfo
*info
,
3654 char *backup_file
, int verbose
,
3655 int forked
, int restart
, int freeze_reshape
)
3657 struct mdinfo
*cc
= NULL
;
3659 char last_devnm
[32] = "";
3661 /* component_size is not meaningful for a container,
3662 * so pass '0' meaning 'no change'
3665 reshape_super(st
, 0, info
->new_level
,
3666 info
->new_layout
, info
->new_chunk
,
3667 info
->array
.raid_disks
, info
->delta_disks
,
3668 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3676 /* ping monitor to be sure that update is on disk
3678 ping_monitor(container
);
3680 if (!forked
&& !freeze_reshape
)
3681 if (continue_via_systemd(container
, GROW_SERVICE
))
3684 switch (forked
? 0 : fork()) {
3685 case -1: /* error */
3686 perror("Cannot fork to complete reshape\n");
3689 default: /* parent */
3690 if (!freeze_reshape
)
3691 printf("%s: multi-array reshape continues in background\n", Name
);
3699 /* close unused handle in child process
3705 /* For each member array with reshape_active,
3706 * we need to perform the reshape.
3707 * We pick the first array that needs reshaping and
3708 * reshape it. reshape_array() will re-read the metadata
3709 * so the next time through a different array should be
3710 * ready for reshape.
3711 * It is possible that the 'different' array will not
3712 * be assembled yet. In that case we simple exit.
3713 * When it is assembled, the mdadm which assembles it
3714 * will take over the reshape.
3716 struct mdinfo
*content
;
3718 struct mdstat_ent
*mdstat
;
3724 cc
= st
->ss
->container_content(st
, NULL
);
3726 for (content
= cc
; content
; content
= content
->next
) {
3728 if (!content
->reshape_active
)
3731 subarray
= strchr(content
->text_version
+1, '/')+1;
3732 mdstat
= mdstat_by_subdev(subarray
, container
);
3735 if (mdstat
->active
== 0) {
3736 pr_err("Skipping inactive array %s.\n",
3738 free_mdstat(mdstat
);
3747 devid
= devnm2devid(mdstat
->devnm
);
3748 adev
= map_dev(major(devid
), minor(devid
), 0);
3750 adev
= content
->text_version
;
3752 fd
= open_dev(mdstat
->devnm
);
3754 pr_err("Device %s cannot be opened for reshape.\n",
3759 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3760 /* Do not allow for multiple reshape_array() calls for
3762 * It can happen when reshape_array() returns without
3763 * error, when reshape is not finished (wrong reshape
3764 * starting/continuation conditions). Mdmon doesn't
3765 * switch to next array in container and reentry
3766 * conditions for the same array occur.
3767 * This is possibly interim until the behaviour of
3768 * reshape_array is resolved().
3770 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3774 strcpy(last_devnm
, mdstat
->devnm
);
3776 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3777 pr_err("Unable to initialize sysfs for %s\n",
3783 if (mdmon_running(container
))
3784 flush_mdmon(container
);
3786 rv
= reshape_array(container
, fd
, adev
, st
,
3787 content
, force
, NULL
, INVALID_SECTORS
,
3788 backup_file
, verbose
, 1, restart
,
3792 if (freeze_reshape
) {
3801 if (mdmon_running(container
))
3802 flush_mdmon(container
);
3811 * We run a child process in the background which performs the following
3813 * - wait for resync to reach a certain point
3814 * - suspend io to the following section
3815 * - backup that section
3816 * - allow resync to proceed further
3818 * - discard the backup.
3820 * When are combined in slightly different ways in the three cases.
3822 * - suspend/backup/allow/wait/resume/discard
3824 * - allow/wait/suspend/backup/allow/wait/resume/discard
3826 * - wait/resume/discard/suspend/backup/allow
3828 * suspend/backup/allow always come together
3829 * wait/resume/discard do too.
3830 * For the same-size case we have two backups to improve flow.
3834 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3835 unsigned long long backup_point
,
3836 unsigned long long wait_point
,
3837 unsigned long long *suspend_point
,
3838 unsigned long long *reshape_completed
, int *frozen
)
3840 /* This function is called repeatedly by the reshape manager.
3841 * It determines how much progress can safely be made and allows
3843 * - 'info' identifies the array and particularly records in
3844 * ->reshape_progress the metadata's knowledge of progress
3845 * This is a sector offset from the start of the array
3846 * of the next array block to be relocated. This number
3847 * may increase from 0 or decrease from array_size, depending
3848 * on the type of reshape that is happening.
3849 * Note that in contrast, 'sync_completed' is a block count of the
3850 * reshape so far. It gives the distance between the start point
3851 * (head or tail of device) and the next place that data will be
3852 * written. It always increases.
3853 * - 'reshape' is the structure created by analyse_change
3854 * - 'backup_point' shows how much the metadata manager has backed-up
3855 * data. For reshapes with increasing progress, it is the next address
3856 * to be backed up, previous addresses have been backed-up. For
3857 * decreasing progress, it is the earliest address that has been
3858 * backed up - later address are also backed up.
3859 * So addresses between reshape_progress and backup_point are
3860 * backed up providing those are in the 'correct' order.
3861 * - 'wait_point' is an array address. When reshape_completed
3862 * passes this point, progress_reshape should return. It might
3863 * return earlier if it determines that ->reshape_progress needs
3864 * to be updated or further backup is needed.
3865 * - suspend_point is maintained by progress_reshape and the caller
3866 * should not touch it except to initialise to zero.
3867 * It is an array address and it only increases in 2.6.37 and earlier.
3868 * This makes it difficult to handle reducing reshapes with
3869 * external metadata.
3870 * However: it is similar to backup_point in that it records the
3871 * other end of a suspended region from reshape_progress.
3872 * it is moved to extend the region that is safe to backup and/or
3874 * - reshape_completed is read from sysfs and returned. The caller
3875 * should copy this into ->reshape_progress when it has reason to
3876 * believe that the metadata knows this, and any backup outside this
3880 * 1 if more data from backup_point - but only as far as suspend_point,
3881 * should be backed up
3882 * 0 if things are progressing smoothly
3883 * -1 if the reshape is finished because it is all done,
3884 * -2 if the reshape is finished due to an error.
3887 int advancing
= (reshape
->after
.data_disks
3888 >= reshape
->before
.data_disks
);
3889 unsigned long long need_backup
; /* All data between start of array and
3890 * here will at some point need to
3893 unsigned long long read_offset
, write_offset
;
3894 unsigned long long write_range
;
3895 unsigned long long max_progress
, target
, completed
;
3896 unsigned long long array_size
= (info
->component_size
3897 * reshape
->before
.data_disks
);
3901 /* First, we unsuspend any region that is now known to be safe.
3902 * If suspend_point is on the 'wrong' side of reshape_progress, then
3903 * we don't have or need suspension at the moment. This is true for
3904 * native metadata when we don't need to back-up.
3907 if (info
->reshape_progress
<= *suspend_point
)
3908 sysfs_set_num(info
, NULL
, "suspend_lo",
3909 info
->reshape_progress
);
3911 /* Note: this won't work in 2.6.37 and before.
3912 * Something somewhere should make sure we don't need it!
3914 if (info
->reshape_progress
>= *suspend_point
)
3915 sysfs_set_num(info
, NULL
, "suspend_hi",
3916 info
->reshape_progress
);
3919 /* Now work out how far it is safe to progress.
3920 * If the read_offset for ->reshape_progress is less than
3921 * 'blocks' beyond the write_offset, we can only progress as far
3923 * Otherwise we can progress until the write_offset for the new location
3924 * reaches (within 'blocks' of) the read_offset at the current location.
3925 * However that region must be suspended unless we are using native
3927 * If we need to suspend more, we limit it to 128M per device, which is
3928 * rather arbitrary and should be some time-based calculation.
3930 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3931 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3932 write_range
= info
->new_chunk
/512;
3933 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3934 need_backup
= array_size
;
3936 need_backup
= reshape
->backup_blocks
;
3938 if (read_offset
< write_offset
+ write_range
)
3939 max_progress
= backup_point
;
3942 read_offset
* reshape
->after
.data_disks
;
3944 if (read_offset
> write_offset
- write_range
)
3945 /* Can only progress as far as has been backed up,
3946 * which must be suspended */
3947 max_progress
= backup_point
;
3948 else if (info
->reshape_progress
<= need_backup
)
3949 max_progress
= backup_point
;
3951 if (info
->array
.major_version
>= 0)
3952 /* Can progress until backup is needed */
3953 max_progress
= need_backup
;
3955 /* Can progress until metadata update is required */
3957 read_offset
* reshape
->after
.data_disks
;
3958 /* but data must be suspended */
3959 if (max_progress
< *suspend_point
)
3960 max_progress
= *suspend_point
;
3965 /* We know it is safe to progress to 'max_progress' providing
3966 * it is suspended or we are using native metadata.
3967 * Consider extending suspend_point 128M per device if it
3968 * is less than 64M per device beyond reshape_progress.
3969 * But always do a multiple of 'blocks'
3970 * FIXME this is too big - it takes to long to complete
3973 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3974 reshape
->after
.data_disks
);
3975 target
/= reshape
->backup_blocks
;
3978 target
*= reshape
->backup_blocks
;
3980 /* For externally managed metadata we always need to suspend IO to
3981 * the area being reshaped so we regularly push suspend_point forward.
3982 * For native metadata we only need the suspend if we are going to do
3986 if ((need_backup
> info
->reshape_progress
||
3987 info
->array
.major_version
< 0) &&
3988 *suspend_point
< info
->reshape_progress
+ target
) {
3989 if (need_backup
< *suspend_point
+ 2 * target
)
3990 *suspend_point
= need_backup
;
3991 else if (*suspend_point
+ 2 * target
< array_size
)
3992 *suspend_point
+= 2 * target
;
3994 *suspend_point
= array_size
;
3995 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
3996 if (max_progress
> *suspend_point
)
3997 max_progress
= *suspend_point
;
4000 if (info
->array
.major_version
>= 0) {
4001 /* Only need to suspend when about to backup */
4002 if (info
->reshape_progress
< need_backup
* 2 &&
4003 *suspend_point
> 0) {
4005 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4006 sysfs_set_num(info
, NULL
, "suspend_hi",
4010 /* Need to suspend continually */
4011 if (info
->reshape_progress
< *suspend_point
)
4012 *suspend_point
= info
->reshape_progress
;
4013 if (*suspend_point
+ target
< info
->reshape_progress
)
4014 /* No need to move suspend region yet */;
4016 if (*suspend_point
>= 2 * target
)
4017 *suspend_point
-= 2 * target
;
4020 sysfs_set_num(info
, NULL
, "suspend_lo",
4023 if (max_progress
< *suspend_point
)
4024 max_progress
= *suspend_point
;
4028 /* now set sync_max to allow that progress. sync_max, like
4029 * sync_completed is a count of sectors written per device, so
4030 * we find the difference between max_progress and the start point,
4031 * and divide that by after.data_disks to get a sync_max
4033 * At the same time we convert wait_point to a similar number
4034 * for comparing against sync_completed.
4036 /* scale down max_progress to per_disk */
4037 max_progress
/= reshape
->after
.data_disks
;
4039 * Round to chunk size as some kernels give an erroneously
4042 max_progress
/= info
->new_chunk
/512;
4043 max_progress
*= info
->new_chunk
/512;
4044 /* And round to old chunk size as the kernel wants that */
4045 max_progress
/= info
->array
.chunk_size
/512;
4046 max_progress
*= info
->array
.chunk_size
/512;
4047 /* Limit progress to the whole device */
4048 if (max_progress
> info
->component_size
)
4049 max_progress
= info
->component_size
;
4050 wait_point
/= reshape
->after
.data_disks
;
4052 /* switch from 'device offset' to 'processed block count' */
4053 max_progress
= info
->component_size
- max_progress
;
4054 wait_point
= info
->component_size
- wait_point
;
4058 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4060 /* Now wait. If we have already reached the point that we were
4061 * asked to wait to, don't wait at all, else wait for any change.
4062 * We need to select on 'sync_completed' as that is the place that
4063 * notifications happen, but we are really interested in
4064 * 'reshape_position'
4066 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4068 goto check_progress
;
4070 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4071 goto check_progress
;
4073 while (completed
< max_progress
&& completed
< wait_point
) {
4074 /* Check that sync_action is still 'reshape' to avoid
4075 * waiting forever on a dead array
4078 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) <= 0 ||
4079 strncmp(action
, "reshape", 7) != 0)
4081 /* Some kernels reset 'sync_completed' to zero
4082 * before setting 'sync_action' to 'idle'.
4083 * So we need these extra tests.
4085 if (completed
== 0 && advancing
&&
4086 strncmp(action
, "idle", 4) == 0 &&
4087 info
->reshape_progress
> 0)
4089 if (completed
== 0 && !advancing
&&
4090 strncmp(action
, "idle", 4) == 0 &&
4091 info
->reshape_progress
<
4092 (info
->component_size
* reshape
->after
.data_disks
))
4094 sysfs_wait(fd
, NULL
);
4095 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4096 goto check_progress
;
4098 /* Some kernels reset 'sync_completed' to zero,
4099 * we need to have real point we are in md.
4100 * So in that case, read 'reshape_position' from sysfs.
4102 if (completed
== 0) {
4103 unsigned long long reshapep
;
4105 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) > 0 &&
4106 strncmp(action
, "idle", 4) == 0 &&
4107 sysfs_get_ll(info
, NULL
,
4108 "reshape_position", &reshapep
) == 0)
4109 *reshape_completed
= reshapep
;
4111 /* some kernels can give an incorrectly high
4112 * 'completed' number, so round down */
4113 completed
/= (info
->new_chunk
/512);
4114 completed
*= (info
->new_chunk
/512);
4115 /* Convert 'completed' back in to a 'progress' number */
4116 completed
*= reshape
->after
.data_disks
;
4118 completed
= (info
->component_size
4119 * reshape
->after
.data_disks
4121 *reshape_completed
= completed
;
4126 /* We return the need_backup flag. Caller will decide
4127 * how much - a multiple of ->backup_blocks up to *suspend_point
4130 return need_backup
> info
->reshape_progress
;
4132 return need_backup
>= info
->reshape_progress
;
4135 /* if we couldn't read a number from sync_completed, then
4136 * either the reshape did complete, or it aborted.
4137 * We can tell which by checking for 'none' in reshape_position.
4138 * If it did abort, then it might immediately restart if it
4139 * it was just a device failure that leaves us degraded but
4142 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
,
4143 sizeof(buf
)) < 0 || strncmp(buf
, "none", 4) != 0) {
4144 /* The abort might only be temporary. Wait up to 10
4145 * seconds for fd to contain a valid number again.
4149 unsigned long long new_sync_max
;
4150 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4151 if (sysfs_wait(fd
, &wait
) != 1)
4153 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4155 /* all good again */
4157 /* If "sync_max" is no longer max_progress
4158 * we need to freeze things
4160 sysfs_get_ll(info
, NULL
, "sync_max",
4162 *frozen
= (new_sync_max
!= max_progress
);
4164 case -2: /* read error - abort */
4171 return rv
; /* abort */
4173 /* Maybe racing with array shutdown - check state */
4176 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4178 strncmp(buf
, "inactive", 8) == 0 ||
4179 strncmp(buf
, "clear",5) == 0)
4180 return -2; /* abort */
4181 return -1; /* complete */
4185 /* FIXME return status is never checked */
4186 static int grow_backup(struct mdinfo
*sra
,
4187 unsigned long long offset
, /* per device */
4188 unsigned long stripes
, /* per device, in old chunks */
4189 int *sources
, unsigned long long *offsets
,
4190 int disks
, int chunk
, int level
, int layout
,
4191 int dests
, int *destfd
, unsigned long long *destoffsets
,
4192 int part
, int *degraded
,
4195 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4196 * to storage 'destfd' (offset 'destoffsets'), after first
4197 * suspending IO. Then allow resync to continue
4198 * over the suspended section.
4199 * Use part 'part' of the backup-super-block.
4204 unsigned long long ll
;
4206 //printf("offset %llu\n", offset);
4212 /* Check that array hasn't become degraded, else we might backup the wrong data */
4213 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4214 return -1; /* FIXME this error is ignored */
4215 new_degraded
= (int)ll
;
4216 if (new_degraded
!= *degraded
) {
4217 /* check each device to ensure it is still working */
4219 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4220 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4222 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4225 if (sysfs_get_str(sra
, sd
, "state",
4226 sbuf
, sizeof(sbuf
)) < 0 ||
4227 strstr(sbuf
, "faulty") ||
4228 strstr(sbuf
, "in_sync") == NULL
) {
4229 /* this device is dead */
4230 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4231 if (sd
->disk
.raid_disk
>= 0 &&
4232 sources
[sd
->disk
.raid_disk
] >= 0) {
4233 close(sources
[sd
->disk
.raid_disk
]);
4234 sources
[sd
->disk
.raid_disk
] = -1;
4239 *degraded
= new_degraded
;
4242 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4243 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4245 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4246 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4249 bsb
.magic
[15] = '2';
4250 for (i
= 0; i
< dests
; i
++)
4252 lseek64(destfd
[i
], destoffsets
[i
] +
4253 __le64_to_cpu(bsb
.devstart2
)*512, 0);
4255 lseek64(destfd
[i
], destoffsets
[i
], 0);
4257 rv
= save_stripes(sources
, offsets
, disks
, chunk
, level
, layout
,
4258 dests
, destfd
, offset
* 512 * odata
,
4259 stripes
* chunk
* odata
, buf
);
4263 bsb
.mtime
= __cpu_to_le64(time(0));
4264 for (i
= 0; i
< dests
; i
++) {
4265 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4267 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4268 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4269 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4270 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4271 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4274 if ((unsigned long long)lseek64(destfd
[i
],
4275 destoffsets
[i
] - 4096, 0) !=
4276 destoffsets
[i
] - 4096)
4278 if (write(destfd
[i
], &bsb
, 512) != 512)
4280 if (destoffsets
[i
] > 4096) {
4281 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4282 destoffsets
[i
]+stripes
*chunk
*odata
)
4284 if (write(destfd
[i
], &bsb
, 512) != 512)
4294 /* in 2.6.30, the value reported by sync_completed can be
4295 * less that it should be by one stripe.
4296 * This only happens when reshape hits sync_max and pauses.
4297 * So allow wait_backup to either extent sync_max further
4298 * than strictly necessary, or return before the
4299 * sync has got quite as far as we would really like.
4300 * This is what 'blocks2' is for.
4301 * The various caller give appropriate values so that
4304 /* FIXME return value is often ignored */
4305 static int forget_backup(int dests
, int *destfd
,
4306 unsigned long long *destoffsets
,
4310 * Erase backup 'part' (which is 0 or 1)
4316 bsb
.arraystart2
= __cpu_to_le64(0);
4317 bsb
.length2
= __cpu_to_le64(0);
4319 bsb
.arraystart
= __cpu_to_le64(0);
4320 bsb
.length
= __cpu_to_le64(0);
4322 bsb
.mtime
= __cpu_to_le64(time(0));
4324 for (i
= 0; i
< dests
; i
++) {
4325 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4326 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4327 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4328 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4329 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4330 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4331 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4332 destoffsets
[i
]-4096)
4334 if (rv
== 0 && write(destfd
[i
], &bsb
, 512) != 512)
4341 static void fail(char *msg
)
4344 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4345 rv
|= (write(2, "\n", 1) != 1);
4349 static char *abuf
, *bbuf
;
4350 static unsigned long long abuflen
;
4351 static void validate(int afd
, int bfd
, unsigned long long offset
)
4353 /* check that the data in the backup against the array.
4354 * This is only used for regression testing and should not
4355 * be used while the array is active
4359 lseek64(bfd
, offset
- 4096, 0);
4360 if (read(bfd
, &bsb2
, 512) != 512)
4361 fail("cannot read bsb");
4362 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4363 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4364 fail("first csum bad");
4365 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4366 fail("magic is bad");
4367 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4368 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4369 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4370 fail("second csum bad");
4372 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4373 fail("devstart is wrong");
4376 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4378 if (abuflen
< len
) {
4382 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4383 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4385 /* just stop validating on mem-alloc failure */
4390 lseek64(bfd
, offset
, 0);
4391 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4392 //printf("len %llu\n", len);
4393 fail("read first backup failed");
4395 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4396 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4397 fail("read first from array failed");
4398 if (memcmp(bbuf
, abuf
, len
) != 0) {
4401 printf("offset=%llu len=%llu\n",
4402 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4403 for (i
=0; i
<len
; i
++)
4404 if (bbuf
[i
] != abuf
[i
]) {
4405 printf("first diff byte %d\n", i
);
4409 fail("data1 compare failed");
4413 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4415 if (abuflen
< len
) {
4419 abuf
= xmalloc(abuflen
);
4420 bbuf
= xmalloc(abuflen
);
4423 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4424 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4425 fail("read second backup failed");
4426 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4427 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4428 fail("read second from array failed");
4429 if (memcmp(bbuf
, abuf
, len
) != 0)
4430 fail("data2 compare failed");
4434 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4435 struct supertype
*st
, unsigned long blocks
,
4436 int *fds
, unsigned long long *offsets
,
4437 int dests
, int *destfd
, unsigned long long *destoffsets
)
4439 /* Monitor a reshape where backup is being performed using
4440 * 'native' mechanism - either to a backup file, or
4441 * to some space in a spare.
4445 unsigned long long speed
;
4446 unsigned long long suspend_point
, array_size
;
4447 unsigned long long backup_point
, wait_point
;
4448 unsigned long long reshape_completed
;
4450 int increasing
= reshape
->after
.data_disks
>=
4451 reshape
->before
.data_disks
;
4452 int part
= 0; /* The next part of the backup area to fill. It
4453 * may already be full, so we need to check */
4454 int level
= reshape
->level
;
4455 int layout
= reshape
->before
.layout
;
4456 int data
= reshape
->before
.data_disks
;
4457 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4458 int chunk
= sra
->array
.chunk_size
;
4460 unsigned long stripes
;
4464 /* set up the backup-super-block. This requires the
4465 * uuid from the array.
4467 /* Find a superblock */
4468 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4472 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4474 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4475 devfd
= dev_open(dn
, O_RDONLY
);
4478 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4484 pr_err("Cannot find a superblock\n");
4488 memset(&bsb
, 0, 512);
4489 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4490 st
->ss
->uuid_from_super(st
, uuid
);
4491 memcpy(bsb
.set_uuid
, uuid
, 16);
4492 bsb
.mtime
= __cpu_to_le64(time(0));
4493 bsb
.devstart2
= blocks
;
4495 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4496 reshape
->before
.data_disks
;
4498 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4499 /* Don't start the 'reshape' */
4501 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4502 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4503 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4507 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4508 backup_point
= sra
->reshape_progress
;
4511 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4512 backup_point
= reshape
->backup_blocks
;
4513 suspend_point
= array_size
;
4519 /* Want to return as soon the oldest backup slot can
4520 * be released as that allows us to start backing up
4521 * some more, providing suspend_point has been
4522 * advanced, which it should have.
4525 wait_point
= array_size
;
4526 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4527 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4528 __le64_to_cpu(bsb
.length
));
4529 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4530 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4531 __le64_to_cpu(bsb
.length2
));
4534 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4535 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4536 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4537 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4540 reshape_completed
= sra
->reshape_progress
;
4541 rv
= progress_reshape(sra
, reshape
,
4542 backup_point
, wait_point
,
4543 &suspend_point
, &reshape_completed
,
4545 /* external metadata would need to ping_monitor here */
4546 sra
->reshape_progress
= reshape_completed
;
4548 /* Clear any backup region that is before 'here' */
4550 if (__le64_to_cpu(bsb
.length
) > 0 &&
4551 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4552 __le64_to_cpu(bsb
.length
)))
4553 forget_backup(dests
, destfd
,
4555 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4556 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4557 __le64_to_cpu(bsb
.length2
)))
4558 forget_backup(dests
, destfd
,
4561 if (__le64_to_cpu(bsb
.length
) > 0 &&
4562 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4563 forget_backup(dests
, destfd
,
4565 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4566 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4567 forget_backup(dests
, destfd
,
4577 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4578 /* No longer need to monitor this reshape */
4579 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4585 unsigned long long offset
;
4586 unsigned long actual_stripes
;
4587 /* Need to backup some data.
4588 * If 'part' is not used and the desired
4589 * backup size is suspended, do a backup,
4590 * then consider the next part.
4592 /* Check that 'part' is unused */
4593 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4595 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4598 offset
= backup_point
/ data
;
4599 actual_stripes
= stripes
;
4601 if (offset
+ actual_stripes
* (chunk
/512) >
4602 sra
->component_size
)
4603 actual_stripes
= ((sra
->component_size
- offset
)
4605 if (offset
+ actual_stripes
* (chunk
/512) >
4609 if (offset
< actual_stripes
* (chunk
/512))
4610 actual_stripes
= offset
/ (chunk
/512);
4611 offset
-= actual_stripes
* (chunk
/512);
4612 if (offset
< suspend_point
/data
)
4615 if (actual_stripes
== 0)
4617 grow_backup(sra
, offset
, actual_stripes
, fds
, offsets
,
4618 disks
, chunk
, level
, layout
, dests
, destfd
,
4619 destoffsets
, part
, °raded
, buf
);
4620 validate(afd
, destfd
[0], destoffsets
[0]);
4621 /* record where 'part' is up to */
4624 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4626 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4630 /* FIXME maybe call progress_reshape one more time instead */
4631 /* remove any remaining suspension */
4632 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4633 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4634 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4635 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4637 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4638 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4644 * If any spare contains md_back_data-1 which is recent wrt mtime,
4645 * write that data into the array and update the super blocks with
4646 * the new reshape_progress
4648 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
,
4649 int cnt
, char *backup_file
, int verbose
)
4653 unsigned long long *offsets
;
4654 unsigned long long nstripe
, ostripe
;
4657 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4658 if (info
->array
.level
== 6)
4659 odata
--; /* number of data disks */
4660 ndata
= info
->array
.raid_disks
- 1;
4661 if (info
->new_level
== 6)
4664 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4666 if (info
->delta_disks
<= 0)
4667 /* Didn't grow, so the backup file must have
4671 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4672 struct mdinfo dinfo
;
4675 char *devname
, namebuf
[20];
4676 unsigned long long lo
, hi
;
4678 /* This was a spare and may have some saved data on it.
4679 * Load the superblock, find and load the
4680 * backup_super_block.
4681 * If either fail, go on to next device.
4682 * If the backup contains no new info, just return
4683 * else restore data and update all superblocks
4685 if (i
== old_disks
-1) {
4686 fd
= open(backup_file
, O_RDONLY
);
4688 pr_err("backup file %s inaccessible: %s\n",
4689 backup_file
, strerror(errno
));
4692 devname
= backup_file
;
4697 if (st
->ss
->load_super(st
, fd
, NULL
))
4700 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4701 st
->ss
->free_super(st
);
4704 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4706 pr_err("Cannot seek on device %d\n", i
);
4707 continue; /* Cannot seek */
4709 sprintf(namebuf
, "device-%d", i
);
4712 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4714 pr_err("Cannot read from %s\n", devname
);
4715 continue; /* Cannot read */
4717 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4718 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4720 pr_err("No backup metadata on %s\n", devname
);
4723 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4725 pr_err("Bad backup-metadata checksum on %s\n",
4727 continue; /* bad checksum */
4729 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4730 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4732 pr_err("Bad backup-metadata checksum2 on %s\n",
4734 continue; /* Bad second checksum */
4736 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4738 pr_err("Wrong uuid on backup-metadata on %s\n",
4740 continue; /* Wrong uuid */
4744 * array utime and backup-mtime should be updated at
4745 * much the same time, but it seems that sometimes
4746 * they aren't... So allow considerable flexability in
4747 * matching, and allow this test to be overridden by
4748 * an environment variable.
4750 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4751 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4752 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4753 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4754 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4755 (unsigned long)info
->array
.utime
);
4757 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4758 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4759 continue; /* time stamp is too bad */
4763 if (bsb
.magic
[15] == '1') {
4764 if (bsb
.length
== 0)
4766 if (info
->delta_disks
>= 0) {
4767 /* reshape_progress is increasing */
4768 if (__le64_to_cpu(bsb
.arraystart
)
4769 + __le64_to_cpu(bsb
.length
)
4770 < info
->reshape_progress
) {
4773 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4774 continue; /* No new data here */
4777 /* reshape_progress is decreasing */
4778 if (__le64_to_cpu(bsb
.arraystart
) >=
4779 info
->reshape_progress
)
4780 goto nonew
; /* No new data here */
4783 if (bsb
.length
== 0 && bsb
.length2
== 0)
4785 if (info
->delta_disks
>= 0) {
4786 /* reshape_progress is increasing */
4787 if ((__le64_to_cpu(bsb
.arraystart
)
4788 + __le64_to_cpu(bsb
.length
)
4789 < info
->reshape_progress
) &&
4790 (__le64_to_cpu(bsb
.arraystart2
)
4791 + __le64_to_cpu(bsb
.length2
)
4792 < info
->reshape_progress
))
4793 goto nonew
; /* No new data here */
4795 /* reshape_progress is decreasing */
4796 if (__le64_to_cpu(bsb
.arraystart
) >=
4797 info
->reshape_progress
&&
4798 __le64_to_cpu(bsb
.arraystart2
) >=
4799 info
->reshape_progress
)
4800 goto nonew
; /* No new data here */
4803 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4806 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4808 continue; /* Cannot seek */
4810 /* There should be a duplicate backup superblock 4k before here */
4811 if (lseek64(fd
, -4096, 1) < 0 ||
4812 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4813 goto second_fail
; /* Cannot find leading superblock */
4814 if (bsb
.magic
[15] == '1')
4815 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4817 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4818 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4819 goto second_fail
; /* Cannot find leading superblock */
4821 /* Now need the data offsets for all devices. */
4822 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4823 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4826 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4827 /* FIXME should be this be an error */
4829 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4830 st
->ss
->free_super(st
);
4831 offsets
[j
] = dinfo
.data_offset
* 512;
4833 printf("%s: restoring critical section\n", Name
);
4835 if (restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4836 info
->new_chunk
, info
->new_level
,
4837 info
->new_layout
, fd
,
4838 __le64_to_cpu(bsb
.devstart
)*512,
4839 __le64_to_cpu(bsb
.arraystart
)*512,
4840 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4841 /* didn't succeed, so giveup */
4843 pr_err("Error restoring backup from %s\n",
4849 if (bsb
.magic
[15] == '2' &&
4850 restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4851 info
->new_chunk
, info
->new_level
,
4852 info
->new_layout
, fd
,
4853 __le64_to_cpu(bsb
.devstart
)*512 +
4854 __le64_to_cpu(bsb
.devstart2
)*512,
4855 __le64_to_cpu(bsb
.arraystart2
)*512,
4856 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4857 /* didn't succeed, so giveup */
4859 pr_err("Error restoring second backup from %s\n",
4867 /* Ok, so the data is restored. Let's update those superblocks. */
4871 lo
= __le64_to_cpu(bsb
.arraystart
);
4872 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4874 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4875 unsigned long long lo1
, hi1
;
4876 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4877 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4881 } else if (lo
< lo1
)
4886 if (lo
< hi
&& (info
->reshape_progress
< lo
||
4887 info
->reshape_progress
> hi
))
4888 /* backup does not affect reshape_progress*/ ;
4889 else if (info
->delta_disks
>= 0) {
4890 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4891 __le64_to_cpu(bsb
.length
);
4892 if (bsb
.magic
[15] == '2') {
4893 unsigned long long p2
;
4895 p2
= __le64_to_cpu(bsb
.arraystart2
) +
4896 __le64_to_cpu(bsb
.length2
);
4897 if (p2
> info
->reshape_progress
)
4898 info
->reshape_progress
= p2
;
4901 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4902 if (bsb
.magic
[15] == '2') {
4903 unsigned long long p2
;
4905 p2
= __le64_to_cpu(bsb
.arraystart2
);
4906 if (p2
< info
->reshape_progress
)
4907 info
->reshape_progress
= p2
;
4910 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4913 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4915 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4916 dinfo
.reshape_progress
= info
->reshape_progress
;
4917 st
->ss
->update_super(st
, &dinfo
, "_reshape_progress",
4919 st
->ss
->store_super(st
, fdlist
[j
]);
4920 st
->ss
->free_super(st
);
4924 /* Didn't find any backup data, try to see if any
4927 if (info
->delta_disks
< 0) {
4928 /* When shrinking, the critical section is at the end.
4929 * So see if we are before the critical section.
4931 unsigned long long first_block
;
4932 nstripe
= ostripe
= 0;
4934 while (ostripe
>= nstripe
) {
4935 ostripe
+= info
->array
.chunk_size
/ 512;
4936 first_block
= ostripe
* odata
;
4937 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4938 (info
->new_chunk
/512);
4941 if (info
->reshape_progress
>= first_block
)
4944 if (info
->delta_disks
> 0) {
4945 /* See if we are beyond the critical section. */
4946 unsigned long long last_block
;
4947 nstripe
= ostripe
= 0;
4949 while (nstripe
>= ostripe
) {
4950 nstripe
+= info
->new_chunk
/ 512;
4951 last_block
= nstripe
* ndata
;
4952 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4953 (info
->array
.chunk_size
/512);
4956 if (info
->reshape_progress
>= last_block
)
4959 /* needed to recover critical section! */
4961 pr_err("Failed to find backup of critical section\n");
4965 int Grow_continue_command(char *devname
, int fd
,
4966 char *backup_file
, int verbose
)
4969 struct supertype
*st
= NULL
;
4970 struct mdinfo
*content
= NULL
;
4971 struct mdinfo array
;
4972 char *subarray
= NULL
;
4973 struct mdinfo
*cc
= NULL
;
4974 struct mdstat_ent
*mdstat
= NULL
;
4978 dprintf("Grow continue from command line called for %s\n", devname
);
4980 st
= super_by_fd(fd
, &subarray
);
4981 if (!st
|| !st
->ss
) {
4982 pr_err("Unable to determine metadata format for %s\n", devname
);
4985 dprintf("Grow continue is run for ");
4986 if (st
->ss
->external
== 0) {
4989 dprintf_cont("native array (%s)\n", devname
);
4990 if (md_get_array_info(fd
, &array
.array
) < 0) {
4991 pr_err("%s is not an active md array - aborting\n",
4994 goto Grow_continue_command_exit
;
4997 sysfs_init(content
, fd
, NULL
);
4998 /* Need to load a superblock.
4999 * FIXME we should really get what we need from
5003 for (d
= 0; d
< MAX_DISKS
; d
++) {
5004 mdu_disk_info_t disk
;
5008 if (md_get_disk_info(fd
, &disk
) < 0)
5010 if (disk
.major
== 0 && disk
.minor
== 0)
5012 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5014 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5017 fd2
= dev_open(dv
, O_RDONLY
);
5020 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5026 if (d
== MAX_DISKS
) {
5027 pr_err("Unable to load metadata for %s\n",
5030 goto Grow_continue_command_exit
;
5032 st
->ss
->getinfo_super(st
, content
, NULL
);
5033 if (!content
->reshape_active
)
5037 } while (cnt
-- > 0);
5042 dprintf_cont("subarray (%s)\n", subarray
);
5043 container
= st
->container_devnm
;
5044 cfd
= open_dev_excl(st
->container_devnm
);
5046 container
= st
->devnm
;
5048 cfd
= open_dev_excl(st
->devnm
);
5049 dprintf_cont("container (%s)\n", container
);
5053 pr_err("Unable to open container for %s\n", devname
);
5055 goto Grow_continue_command_exit
;
5058 /* find in container array under reshape
5060 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5062 pr_err("Cannot read superblock for %s\n", devname
);
5064 goto Grow_continue_command_exit
;
5067 cc
= st
->ss
->container_content(st
, subarray
);
5068 for (content
= cc
; content
; content
= content
->next
) {
5070 int allow_reshape
= 1;
5072 if (content
->reshape_active
== 0)
5074 /* The decision about array or container wide
5075 * reshape is taken in Grow_continue based
5076 * content->reshape_active state, therefore we
5077 * need to check_reshape based on
5078 * reshape_active and subarray name
5080 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5082 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5083 (content
->array
.state
5084 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5087 if (!allow_reshape
) {
5088 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5089 devname
, container
);
5091 goto Grow_continue_command_exit
;
5094 array_name
= strchr(content
->text_version
+1, '/')+1;
5095 mdstat
= mdstat_by_subdev(array_name
, container
);
5098 if (mdstat
->active
== 0) {
5099 pr_err("Skipping inactive array %s.\n",
5101 free_mdstat(mdstat
);
5108 pr_err("Unable to determine reshaped array for %s\n", devname
);
5110 goto Grow_continue_command_exit
;
5112 fd2
= open_dev(mdstat
->devnm
);
5114 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5116 goto Grow_continue_command_exit
;
5119 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5120 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5124 goto Grow_continue_command_exit
;
5129 /* start mdmon in case it is not running
5131 if (!mdmon_running(container
))
5132 start_mdmon(container
);
5133 ping_monitor(container
);
5135 if (mdmon_running(container
))
5136 st
->update_tail
= &st
->updates
;
5138 pr_err("No mdmon found. Grow cannot continue.\n");
5140 goto Grow_continue_command_exit
;
5144 /* verify that array under reshape is started from
5147 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5149 goto Grow_continue_command_exit
;
5154 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5156 Grow_continue_command_exit
:
5159 st
->ss
->free_super(st
);
5160 free_mdstat(mdstat
);
5167 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5168 char *backup_file
, int forked
, int freeze_reshape
)
5172 if (!info
->reshape_active
)
5175 if (st
->ss
->external
) {
5176 int cfd
= open_dev(st
->container_devnm
);
5181 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5183 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5184 st
, info
, 0, backup_file
, 0,
5185 forked
, 1 | info
->reshape_active
,
5188 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5189 NULL
, INVALID_SECTORS
, backup_file
,
5190 0, forked
, 1 | info
->reshape_active
,
5196 char *make_backup(char *name
)
5198 char *base
= "backup_file-";
5202 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5203 fname
= xmalloc(len
);
5204 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5208 char *locate_backup(char *name
)
5210 char *fl
= make_backup(name
);
5213 if (stat(fl
, &stb
) == 0 && S_ISREG(stb
.st_mode
))