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 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
, 0, 0, NULL
);
202 if (st
->ss
->store_super(st
, nfd
)) {
203 pr_err("Cannot store new superblock on %s\n", newdev
);
209 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
210 pr_err("Cannot add new disk to this array\n");
213 /* Well, that seems to have worked.
214 * Now go through and update all superblocks
217 if (md_get_array_info(fd
, &info
.array
) < 0) {
218 pr_err("cannot get array info for %s\n", devname
);
223 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
224 mdu_disk_info_t disk
;
228 if (md_get_disk_info(fd
, &disk
) < 0) {
229 pr_err("cannot get device detail for device %d\n", d
);
232 dv
= map_dev(disk
.major
, disk
.minor
, 1);
234 pr_err("cannot find device file for device %d\n", d
);
237 fd2
= dev_open(dv
, O_RDWR
);
239 pr_err("cannot open device file %s\n", dv
);
242 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
243 pr_err("cannot find super block on %s\n", dv
);
247 info
.array
.raid_disks
= nd
+1;
248 info
.array
.nr_disks
= nd
+1;
249 info
.array
.active_disks
= nd
+1;
250 info
.array
.working_disks
= nd
+1;
252 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
255 if (st
->ss
->store_super(st
, fd2
)) {
256 pr_err("Cannot store new superblock on %s\n", dv
);
266 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
269 * First check that array doesn't have a bitmap
270 * Then create the bitmap
273 * For internal bitmaps, we need to check the version,
274 * find all the active devices, and write the bitmap block
277 mdu_bitmap_file_t bmf
;
278 mdu_array_info_t array
;
279 struct supertype
*st
;
280 char *subarray
= NULL
;
281 int major
= BITMAP_MAJOR_HI
;
282 unsigned long long bitmapsize
, array_size
;
286 * We only ever get called if s->bitmap_file is != NULL, so this check
287 * is just here to quiet down static code checkers.
292 if (strcmp(s
->bitmap_file
, "clustered") == 0)
293 major
= BITMAP_MAJOR_CLUSTERED
;
295 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
297 pr_err("Memory allocation failure.\n");
299 pr_err("bitmaps not supported by this kernel.\n");
302 if (bmf
.pathname
[0]) {
303 if (strcmp(s
->bitmap_file
,"none") == 0) {
304 if (ioctl(fd
, SET_BITMAP_FILE
, -1) != 0) {
305 pr_err("failed to remove bitmap %s\n",
311 pr_err("%s already has a bitmap (%s)\n", devname
, bmf
.pathname
);
314 if (md_get_array_info(fd
, &array
) != 0) {
315 pr_err("cannot get array status for %s\n", devname
);
318 if (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) {
319 if (strcmp(s
->bitmap_file
, "none")==0) {
320 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
321 if (md_set_array_info(fd
, &array
) != 0) {
322 if (array
.state
& (1 << MD_SB_CLUSTERED
))
323 pr_err("failed to remove clustered bitmap.\n");
325 pr_err("failed to remove internal bitmap.\n");
330 pr_err("bitmap already present on %s\n", devname
);
334 if (strcmp(s
->bitmap_file
, "none") == 0) {
335 pr_err("no bitmap found on %s\n", devname
);
338 if (array
.level
<= 0) {
339 pr_err("Bitmaps not meaningful with level %s\n",
340 map_num(pers
, array
.level
)?:"of this array");
343 bitmapsize
= array
.size
;
345 if (get_dev_size(fd
, NULL
, &array_size
) &&
346 array_size
> (0x7fffffffULL
<< 9)) {
347 /* Array is big enough that we cannot trust array.size
348 * try other approaches
350 bitmapsize
= get_component_size(fd
);
352 if (bitmapsize
== 0) {
353 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
357 if (array
.level
== 10) {
360 ncopies
= (array
.layout
& 255) * ((array
.layout
>> 8) & 255);
361 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
363 if (strcmp(s
->bitmap_file
, "clustered") == 0 &&
364 !is_near_layout_10(array
.layout
)) {
365 pr_err("only near layout is supported with clustered raid10\n");
370 st
= super_by_fd(fd
, &subarray
);
372 pr_err("Cannot understand version %d.%d\n",
373 array
.major_version
, array
.minor_version
);
377 pr_err("Cannot add bitmaps to sub-arrays yet\n");
383 mdi
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
);
385 if (mdi
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
386 pr_err("Cannot add bitmap to array with PPL\n");
394 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
395 strcmp(s
->bitmap_file
, "clustered") == 0) {
398 int offset_setable
= 0;
399 if (st
->ss
->add_internal_bitmap
== NULL
) {
400 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
403 st
->nodes
= c
->nodes
;
404 st
->cluster_name
= c
->homecluster
;
405 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
408 for (d
= 0; d
< st
->max_devs
; d
++) {
409 mdu_disk_info_t disk
;
414 if (md_get_disk_info(fd
, &disk
) < 0)
416 if (disk
.major
== 0 && disk
.minor
== 0)
418 if ((disk
.state
& (1 << MD_DISK_SYNC
)) == 0)
420 dv
= map_dev(disk
.major
, disk
.minor
, 1);
423 fd2
= dev_open(dv
, O_RDWR
);
426 rv
= st
->ss
->load_super(st
, fd2
, NULL
);
428 rv
= st
->ss
->add_internal_bitmap(
429 st
, &s
->bitmap_chunk
, c
->delay
,
430 s
->write_behind
, bitmapsize
,
431 offset_setable
, major
);
433 st
->ss
->write_bitmap(st
, fd2
,
436 pr_err("failed to create internal bitmap - chunksize problem.\n");
439 pr_err("failed to load super-block.\n");
445 if (offset_setable
) {
446 st
->ss
->getinfo_super(st
, mdi
, NULL
);
447 if (sysfs_init(mdi
, fd
, NULL
)) {
448 pr_err("failed to intialize sysfs.\n");
451 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
455 if (strcmp(s
->bitmap_file
, "clustered") == 0)
456 array
.state
|= (1 << MD_SB_CLUSTERED
);
457 array
.state
|= (1 << MD_SB_BITMAP_PRESENT
);
458 rv
= md_set_array_info(fd
, &array
);
462 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
463 pr_err("failed to set internal bitmap.\n");
470 int max_devs
= st
->max_devs
;
472 /* try to load a superblock */
473 for (d
= 0; d
< max_devs
; d
++) {
474 mdu_disk_info_t disk
;
478 if (md_get_disk_info(fd
, &disk
) < 0)
480 if ((disk
.major
==0 && disk
.minor
== 0) ||
481 (disk
.state
& (1 << MD_DISK_REMOVED
)))
483 dv
= map_dev(disk
.major
, disk
.minor
, 1);
486 fd2
= dev_open(dv
, O_RDONLY
);
488 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
490 st
->ss
->uuid_from_super(st
, uuid
);
497 pr_err("cannot find UUID for array!\n");
500 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
,
501 s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
502 bitmapsize
, major
)) {
505 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
507 pr_err("weird: %s cannot be opened\n", s
->bitmap_file
);
510 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
513 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
514 pr_err("Cannot set bitmap file for %s: %s\n",
515 devname
, strerror(err
));
523 int Grow_consistency_policy(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
525 struct supertype
*st
;
528 char *subarray
= NULL
;
530 char container_dev
[PATH_MAX
];
533 if (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
534 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
535 pr_err("Operation not supported for consistency policy %s\n",
536 map_num(consistency_policies
, s
->consistency_policy
));
540 st
= super_by_fd(fd
, &subarray
);
544 sra
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
|GET_LEVEL
|
551 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
552 !st
->ss
->write_init_ppl
) {
553 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
558 if (sra
->array
.level
!= 5) {
559 pr_err("Operation not supported for array level %d\n",
565 if (sra
->consistency_policy
== (unsigned)s
->consistency_policy
) {
566 pr_err("Consistency policy is already %s\n",
567 map_num(consistency_policies
, s
->consistency_policy
));
570 } else if (sra
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
571 sra
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
572 pr_err("Current consistency policy is %s, cannot change to %s\n",
573 map_num(consistency_policies
, sra
->consistency_policy
),
574 map_num(consistency_policies
, s
->consistency_policy
));
579 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
580 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) <= 0) {
583 } else if (strcmp(buf
, "reshape\n") == 0) {
584 pr_err("PPL cannot be enabled when reshape is in progress\n");
593 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
598 sprintf(container_dev
, "/dev/%s", st
->container_devnm
);
600 ret
= Update_subarray(container_dev
, subarray
, update
, NULL
,
606 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
613 cfd
= open(container_dev
, O_RDWR
|O_EXCL
);
615 pr_err("Failed to open %s\n", container_dev
);
620 ret
= st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
624 pr_err("Cannot read superblock for %s\n",
629 mdi
= st
->ss
->container_content(st
, subarray
);
634 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
638 devpath
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
639 dfd
= dev_open(devpath
, O_RDWR
);
641 pr_err("Failed to open %s\n", devpath
);
647 ret
= st
->ss
->load_super(st
, dfd
, NULL
);
649 pr_err("Failed to load super-block.\n");
654 ret
= st
->ss
->update_super(st
, sra
, "ppl",
656 c
->verbose
, 0, NULL
);
659 st
->ss
->free_super(st
);
662 st
->ss
->getinfo_super(st
, &info
, NULL
);
665 ret
|= sysfs_set_num(sra
, sd
, "ppl_sector",
667 ret
|= sysfs_set_num(sra
, sd
, "ppl_size",
671 pr_err("Failed to set PPL attributes for %s\n",
674 st
->ss
->free_super(st
);
678 ret
= st
->ss
->write_init_ppl(st
, &info
, dfd
);
680 pr_err("Failed to write PPL\n");
685 st
->ss
->free_super(st
);
692 ret
= sysfs_set_str(sra
, NULL
, "consistency_policy",
693 map_num(consistency_policies
,
694 s
->consistency_policy
));
696 pr_err("Failed to change array consistency policy\n");
708 * When reshaping an array we might need to backup some data.
709 * This is written to all spares with a 'super_block' describing it.
710 * The superblock goes 4K from the end of the used space on the
712 * It if written after the backup is complete.
713 * It has the following structure.
716 static struct mdp_backup_super
{
717 char magic
[16]; /* md_backup_data-1 or -2 */
720 /* start/sizes in 512byte sectors */
721 __u64 devstart
; /* address on backup device/file of data */
724 __u32 sb_csum
; /* csum of preceeding bytes. */
726 __u64 devstart2
; /* offset in to data of second section */
729 __u32 sb_csum2
; /* csum of preceeding bytes. */
731 } __attribute__((aligned(512))) bsb
, bsb2
;
733 static __u32
bsb_csum(char *buf
, int len
)
737 for (i
= 0; i
< len
; i
++)
738 csum
= (csum
<<3) + buf
[0];
739 return __cpu_to_le32(csum
);
742 static int check_idle(struct supertype
*st
)
744 /* Check that all member arrays for this container, or the
745 * container of this array, are idle
747 char *container
= (st
->container_devnm
[0]
748 ? st
->container_devnm
: st
->devnm
);
749 struct mdstat_ent
*ent
, *e
;
752 ent
= mdstat_read(0, 0);
753 for (e
= ent
; e
; e
= e
->next
) {
754 if (!is_container_member(e
, container
))
756 if (e
->percent
>= 0) {
765 static int freeze_container(struct supertype
*st
)
767 char *container
= (st
->container_devnm
[0]
768 ? st
->container_devnm
: st
->devnm
);
773 if (block_monitor(container
, 1)) {
774 pr_err("failed to freeze container\n");
781 static void unfreeze_container(struct supertype
*st
)
783 char *container
= (st
->container_devnm
[0]
784 ? st
->container_devnm
: st
->devnm
);
786 unblock_monitor(container
, 1);
789 static int freeze(struct supertype
*st
)
791 /* Try to freeze resync/rebuild on this array/container.
792 * Return -1 if the array is busy,
793 * return -2 container cannot be frozen,
794 * return 0 if this kernel doesn't support 'frozen'
795 * return 1 if it worked.
797 if (st
->ss
->external
)
798 return freeze_container(st
);
800 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
806 /* Need to clear any 'read-auto' status */
807 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
808 strncmp(buf
, "read-auto", 9) == 0)
809 sysfs_set_str(sra
, NULL
, "array_state", "clean");
811 err
= sysfs_freeze_array(sra
);
817 static void unfreeze(struct supertype
*st
)
819 if (st
->ss
->external
)
820 return unfreeze_container(st
);
822 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
826 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
827 strcmp(buf
, "frozen\n") == 0)
828 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
833 static void wait_reshape(struct mdinfo
*sra
)
835 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
841 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
842 strncmp(action
, "reshape", 7) == 0)
843 sysfs_wait(fd
, NULL
);
847 static int reshape_super(struct supertype
*st
, unsigned long long size
,
848 int level
, int layout
, int chunksize
, int raid_disks
,
849 int delta_disks
, char *backup_file
, char *dev
,
850 int direction
, int verbose
)
852 /* nothing extra to check in the native case */
853 if (!st
->ss
->external
)
855 if (!st
->ss
->reshape_super
|| !st
->ss
->manage_reshape
) {
856 pr_err("%s metadata does not support reshape\n",
861 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
862 raid_disks
, delta_disks
, backup_file
, dev
,
866 static void sync_metadata(struct supertype
*st
)
868 if (st
->ss
->external
) {
869 if (st
->update_tail
) {
870 flush_metadata_updates(st
);
871 st
->update_tail
= &st
->updates
;
873 st
->ss
->sync_metadata(st
);
877 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
879 /* when dealing with external metadata subarrays we need to be
880 * prepared to handle EAGAIN. The kernel may need to wait for
881 * mdmon to mark the array active so the kernel can handle
882 * allocations/writeback when preparing the reshape action
883 * (md_allow_write()). We temporarily disable safe_mode_delay
884 * to close a race with the array_state going clean before the
885 * next write to raid_disks / stripe_cache_size
890 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
892 (strcmp(name
, "raid_disks") != 0 &&
893 strcmp(name
, "stripe_cache_size") != 0))
894 return sysfs_set_num(sra
, NULL
, name
, n
);
896 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
899 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
900 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
901 if (rc
< 0 && errno
== EAGAIN
) {
902 ping_monitor(container
);
903 /* if we get EAGAIN here then the monitor is not active
906 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
908 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
912 int start_reshape(struct mdinfo
*sra
, int already_running
,
913 int before_data_disks
, int data_disks
)
916 unsigned long long sync_max_to_set
;
918 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
919 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
920 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
921 sra
->reshape_progress
);
922 if (before_data_disks
<= data_disks
)
923 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
925 sync_max_to_set
= (sra
->component_size
* data_disks
926 - sra
->reshape_progress
) / data_disks
;
927 if (!already_running
)
928 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
929 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
930 if (!already_running
&& err
== 0) {
933 err
= sysfs_set_str(sra
, NULL
, "sync_action",
937 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
942 void abort_reshape(struct mdinfo
*sra
)
944 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
946 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
947 * suspend_hi to decrease as well as increase.")
948 * you could only increase suspend_{lo,hi} unless the region they
949 * covered was empty. So to reset to 0, you need to push suspend_lo
950 * up past suspend_hi first. So to maximize the chance of mdadm
951 * working on all kernels, we want to keep doing that.
953 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
954 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
955 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
956 sysfs_set_num(sra
, NULL
, "sync_min", 0);
957 // It isn't safe to reset sync_max as we aren't monitoring.
958 // Array really should be stopped at this point.
961 int remove_disks_for_takeover(struct supertype
*st
,
966 struct mdinfo
*remaining
;
969 if (st
->ss
->external
) {
971 struct mdinfo
*arrays
= st
->ss
->container_content(st
, NULL
);
973 * containter_content returns list of arrays in container
974 * If arrays->next is not NULL it means that there are
975 * 2 arrays in container and operation should be blocked
982 pr_err("Error. Cannot perform operation on /dev/%s\n", st
->devnm
);
983 pr_err("For this operation it MUST be single array in container\n");
989 if (sra
->array
.level
== 10)
990 nr_of_copies
= layout
& 0xff;
991 else if (sra
->array
.level
== 1)
992 nr_of_copies
= sra
->array
.raid_disks
;
996 remaining
= sra
->devs
;
998 /* for each 'copy', select one device and remove from the list. */
999 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
1000 struct mdinfo
**diskp
;
1003 /* Find a working device to keep */
1004 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
1005 struct mdinfo
*disk
= *diskp
;
1007 if (disk
->disk
.raid_disk
< slot
)
1009 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
1011 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
1013 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
1015 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
1018 /* We have found a good disk to use! */
1019 *diskp
= disk
->next
;
1020 disk
->next
= sra
->devs
;
1029 if (slot
< sra
->array
.raid_disks
) {
1030 /* didn't find all slots */
1036 sra
->devs
= remaining
;
1040 /* Remove all 'remaining' devices from the array */
1042 struct mdinfo
*sd
= remaining
;
1043 remaining
= sd
->next
;
1045 sysfs_set_str(sra
, sd
, "state", "faulty");
1046 sysfs_set_str(sra
, sd
, "slot", "none");
1047 /* for external metadata disks should be removed in mdmon */
1048 if (!st
->ss
->external
)
1049 sysfs_set_str(sra
, sd
, "state", "remove");
1050 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
1051 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
1052 sd
->next
= sra
->devs
;
1058 void reshape_free_fdlist(int *fdlist
,
1059 unsigned long long *offsets
,
1064 for (i
= 0; i
< size
; i
++)
1072 int reshape_prepare_fdlist(char *devname
,
1076 unsigned long blocks
,
1079 unsigned long long *offsets
)
1084 enable_fds(nrdisks
);
1085 for (d
= 0; d
<= nrdisks
; d
++)
1088 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1089 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1091 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
1092 sd
->disk
.raid_disk
< raid_disks
) {
1093 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1094 fdlist
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDONLY
);
1095 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1096 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1097 pr_err("%s: cannot open component %s\n",
1098 devname
, dn
? dn
: "-unknown-");
1102 } else if (backup_file
== NULL
) {
1104 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1105 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1106 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1107 if (fdlist
[d
] < 0) {
1108 pr_err("%s: cannot open component %s\n",
1109 devname
, dn
? dn
: "-unknown-");
1120 int reshape_open_backup_file(char *backup_file
,
1125 unsigned long long *offsets
,
1129 /* Return 1 on success, 0 on any form of failure */
1130 /* need to check backup file is large enough */
1136 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
1140 pr_err("%s: cannot create backup file %s: %s\n",
1141 devname
, backup_file
, strerror(errno
));
1144 /* Guard against backup file being on array device.
1145 * If array is partitioned or if LVM etc is in the
1146 * way this will not notice, but it is better than
1149 fstat(*fdlist
, &stb
);
1152 if (stb
.st_rdev
== dev
) {
1153 pr_err("backup file must NOT be on the array being reshaped.\n");
1158 memset(buf
, 0, 512);
1159 for (i
=0; i
< blocks
+ 8 ; i
++) {
1160 if (write(*fdlist
, buf
, 512) != 512) {
1161 pr_err("%s: cannot create backup file %s: %s\n",
1162 devname
, backup_file
, strerror(errno
));
1166 if (fsync(*fdlist
) != 0) {
1167 pr_err("%s: cannot create backup file %s: %s\n",
1168 devname
, backup_file
, strerror(errno
));
1172 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
1173 char *bu
= make_backup(sys_name
);
1174 if (symlink(backup_file
, bu
))
1175 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
1183 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
1184 unsigned int ndata
, unsigned int odata
)
1186 unsigned long a
, b
, blocks
;
1187 /* So how much do we need to backup.
1188 * We need an amount of data which is both a whole number of
1189 * old stripes and a whole number of new stripes.
1190 * So LCM for (chunksize*datadisks).
1192 a
= (ochunk
/512) * odata
;
1193 b
= (nchunk
/512) * ndata
;
1196 /* LCM == product / GCD */
1197 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
1202 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1204 /* Based on the current array state in info->array and
1205 * the changes in info->new_* etc, determine:
1206 * - whether the change is possible
1207 * - Intermediate level/raid_disks/layout
1208 * - whether a restriping reshape is needed
1209 * - number of sectors in minimum change unit. This
1210 * will cover a whole number of stripes in 'before' and
1213 * Return message if the change should be rejected
1214 * NULL if the change can be achieved
1216 * This can be called as part of starting a reshape, or
1217 * when assembling an array that is undergoing reshape.
1219 int near
, far
, offset
, copies
;
1221 int old_chunk
, new_chunk
;
1222 /* delta_parity records change in number of devices
1223 * caused by level change
1225 int delta_parity
= 0;
1227 memset(re
, 0, sizeof(*re
));
1229 /* If a new level not explicitly given, we assume no-change */
1230 if (info
->new_level
== UnSet
)
1231 info
->new_level
= info
->array
.level
;
1233 if (info
->new_chunk
)
1234 switch (info
->new_level
) {
1240 /* chunk size is meaningful, must divide component_size
1243 if (info
->component_size
% (info
->new_chunk
/512)) {
1244 unsigned long long shrink
= info
->component_size
;
1245 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1246 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1247 info
->new_chunk
/1024, info
->component_size
/2);
1248 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1250 pr_err("will shrink the array so the given chunk size would work.\n");
1255 return "chunk size not meaningful for this level";
1258 info
->new_chunk
= info
->array
.chunk_size
;
1260 switch (info
->array
.level
) {
1262 return "No reshape is possibly for this RAID level";
1264 if (info
->delta_disks
!= UnSet
)
1265 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1267 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1269 /* RAID1 can convert to RAID1 with different disks, or
1270 * raid5 with 2 disks, or
1273 if (info
->new_level
> 1 && (info
->component_size
& 7))
1274 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1275 if (info
->new_level
== 0) {
1276 if (info
->delta_disks
!= UnSet
&&
1277 info
->delta_disks
!= 0)
1278 return "Cannot change number of disks with RAID1->RAID0 conversion";
1280 re
->before
.data_disks
= 1;
1281 re
->after
.data_disks
= 1;
1284 if (info
->new_level
== 1) {
1285 if (info
->delta_disks
== UnSet
)
1286 /* Don't know what to do */
1287 return "no change requested for Growing RAID1";
1291 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1292 return "Can only convert a 2-device array to RAID5";
1293 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1295 re
->before
.data_disks
= 1;
1296 if (info
->delta_disks
!= UnSet
&&
1297 info
->delta_disks
!= 0)
1298 re
->after
.data_disks
= 1 + info
->delta_disks
;
1300 re
->after
.data_disks
= 1;
1301 if (re
->after
.data_disks
< 1)
1302 return "Number of disks too small for RAID5";
1304 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1305 info
->array
.chunk_size
= 65536;
1308 /* Could do some multi-stage conversions, but leave that to
1311 return "Impossibly level change request for RAID1";
1314 /* RAID10 can be converted from near mode to
1315 * RAID0 by removing some devices.
1316 * It can also be reshaped if the kernel supports
1319 switch (info
->new_level
) {
1321 if ((info
->array
.layout
& ~0xff) != 0x100)
1322 return "Cannot Grow RAID10 with far/offset layout";
1324 * number of devices must be multiple of
1327 if (info
->array
.raid_disks
%
1328 (info
->array
.layout
& 0xff))
1329 return "RAID10 layout too complex for Grow operation";
1331 new_disks
= (info
->array
.raid_disks
/
1332 (info
->array
.layout
& 0xff));
1333 if (info
->delta_disks
== UnSet
)
1334 info
->delta_disks
= (new_disks
1335 - info
->array
.raid_disks
);
1337 if (info
->delta_disks
!=
1338 new_disks
- info
->array
.raid_disks
)
1339 return "New number of raid-devices impossible for RAID10";
1340 if (info
->new_chunk
&&
1341 info
->new_chunk
!= info
->array
.chunk_size
)
1342 return "Cannot change chunk-size with RAID10 Grow";
1346 re
->before
.data_disks
= new_disks
;
1347 re
->after
.data_disks
= re
->before
.data_disks
;
1351 near
= info
->array
.layout
& 0xff;
1352 far
= (info
->array
.layout
>> 8) & 0xff;
1353 offset
= info
->array
.layout
& 0x10000;
1354 if (far
> 1 && !offset
)
1355 return "Cannot reshape RAID10 in far-mode";
1356 copies
= near
* far
;
1358 old_chunk
= info
->array
.chunk_size
* far
;
1360 if (info
->new_layout
== UnSet
)
1361 info
->new_layout
= info
->array
.layout
;
1363 near
= info
->new_layout
& 0xff;
1364 far
= (info
->new_layout
>> 8) & 0xff;
1365 offset
= info
->new_layout
& 0x10000;
1366 if (far
> 1 && !offset
)
1367 return "Cannot reshape RAID10 to far-mode";
1368 if (near
* far
!= copies
)
1369 return "Cannot change number of copies when reshaping RAID10";
1371 if (info
->delta_disks
== UnSet
)
1372 info
->delta_disks
= 0;
1373 new_disks
= (info
->array
.raid_disks
+
1376 new_chunk
= info
->new_chunk
* far
;
1379 re
->before
.layout
= info
->array
.layout
;
1380 re
->before
.data_disks
= info
->array
.raid_disks
;
1381 re
->after
.layout
= info
->new_layout
;
1382 re
->after
.data_disks
= new_disks
;
1383 /* For RAID10 we don't do backup but do allow reshape,
1384 * so set backup_blocks to INVALID_SECTORS rather than
1386 * And there is no need to synchronise stripes on both
1387 * 'old' and 'new'. So the important
1388 * number is the minimum data_offset difference
1389 * which is the larger of (offset copies * chunk).
1391 re
->backup_blocks
= INVALID_SECTORS
;
1392 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1393 if (new_disks
< re
->before
.data_disks
&&
1394 info
->space_after
< re
->min_offset_change
)
1395 /* Reduce component size by one chunk */
1396 re
->new_size
= (info
->component_size
-
1397 re
->min_offset_change
);
1399 re
->new_size
= info
->component_size
;
1400 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1404 return "RAID10 can only be changed to RAID0";
1407 /* RAID0 can be converted to RAID10, or to RAID456 */
1408 if (info
->new_level
== 10) {
1409 if (info
->new_layout
== UnSet
&&
1410 info
->delta_disks
== UnSet
) {
1411 /* Assume near=2 layout */
1412 info
->new_layout
= 0x102;
1413 info
->delta_disks
= info
->array
.raid_disks
;
1415 if (info
->new_layout
== UnSet
) {
1416 int copies
= 1 + (info
->delta_disks
1417 / info
->array
.raid_disks
);
1418 if (info
->array
.raid_disks
* (copies
-1) !=
1420 return "Impossible number of devices for RAID0->RAID10";
1421 info
->new_layout
= 0x100 + copies
;
1423 if (info
->delta_disks
== UnSet
) {
1424 int copies
= info
->new_layout
& 0xff;
1425 if (info
->new_layout
!= 0x100 + copies
)
1426 return "New layout impossible for RAID0->RAID10";;
1427 info
->delta_disks
= (copies
- 1) *
1428 info
->array
.raid_disks
;
1430 if (info
->new_chunk
&&
1431 info
->new_chunk
!= info
->array
.chunk_size
)
1432 return "Cannot change chunk-size with RAID0->RAID10";
1435 re
->before
.data_disks
= (info
->array
.raid_disks
+
1437 re
->after
.data_disks
= re
->before
.data_disks
;
1438 re
->before
.layout
= info
->new_layout
;
1442 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1443 * a raid4 style layout of the final level.
1445 switch (info
->new_level
) {
1450 re
->before
.layout
= 0;
1455 re
->before
.layout
= ALGORITHM_PARITY_N
;
1456 if (info
->new_layout
== UnSet
)
1457 info
->new_layout
= map_name(r5layout
, "default");
1462 re
->before
.layout
= ALGORITHM_PARITY_N
;
1463 if (info
->new_layout
== UnSet
)
1464 info
->new_layout
= map_name(r6layout
, "default");
1467 return "Impossible level change requested";
1469 re
->before
.data_disks
= info
->array
.raid_disks
;
1470 /* determining 'after' layout happens outside this 'switch' */
1474 info
->array
.layout
= ALGORITHM_PARITY_N
;
1476 switch (info
->new_level
) {
1480 re
->level
= info
->array
.level
;
1481 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1482 re
->before
.layout
= info
->array
.layout
;
1486 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1487 re
->before
.layout
= info
->array
.layout
;
1492 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1493 switch (info
->array
.layout
) {
1494 case ALGORITHM_LEFT_ASYMMETRIC
:
1495 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1497 case ALGORITHM_RIGHT_ASYMMETRIC
:
1498 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1500 case ALGORITHM_LEFT_SYMMETRIC
:
1501 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1503 case ALGORITHM_RIGHT_SYMMETRIC
:
1504 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1506 case ALGORITHM_PARITY_0
:
1507 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1509 case ALGORITHM_PARITY_N
:
1510 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1513 return "Cannot convert an array with this layout";
1517 if (info
->array
.raid_disks
!= 2)
1518 return "Can only convert a 2-device array to RAID1";
1519 if (info
->delta_disks
!= UnSet
&&
1520 info
->delta_disks
!= 0)
1521 return "Cannot set raid_disk when converting RAID5->RAID1";
1523 info
->new_chunk
= 0;
1526 return "Impossible level change requested";
1530 switch (info
->new_level
) {
1536 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1537 re
->before
.layout
= info
->array
.layout
;
1540 return "Impossible level change requested";
1545 /* If we reached here then it looks like a re-stripe is
1546 * happening. We have determined the intermediate level
1547 * and initial raid_disks/layout and stored these in 're'.
1549 * We need to deduce the final layout that can be atomically
1550 * converted to the end state.
1552 switch (info
->new_level
) {
1554 /* We can only get to RAID0 from RAID4 or RAID5
1555 * with appropriate layout and one extra device
1557 if (re
->level
!= 4 && re
->level
!= 5)
1558 return "Cannot covert to RAID0 from this level";
1560 switch (re
->level
) {
1562 re
->before
.layout
= 0;
1563 re
->after
.layout
= 0;
1566 re
->after
.layout
= ALGORITHM_PARITY_N
;
1572 /* We can only get to RAID4 from RAID5 */
1573 if (re
->level
!= 4 && re
->level
!= 5)
1574 return "Cannot convert to RAID4 from this level";
1576 switch (re
->level
) {
1578 re
->after
.layout
= 0;
1581 re
->after
.layout
= ALGORITHM_PARITY_N
;
1587 /* We get to RAID5 from RAID5 or RAID6 */
1588 if (re
->level
!= 5 && re
->level
!= 6)
1589 return "Cannot convert to RAID5 from this level";
1591 switch (re
->level
) {
1593 if (info
->new_layout
== UnSet
)
1594 re
->after
.layout
= re
->before
.layout
;
1596 re
->after
.layout
= info
->new_layout
;
1599 if (info
->new_layout
== UnSet
)
1600 info
->new_layout
= re
->before
.layout
;
1602 /* after.layout needs to be raid6 version of new_layout */
1603 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1604 re
->after
.layout
= ALGORITHM_PARITY_N
;
1607 char *ls
= map_num(r5layout
, info
->new_layout
);
1610 /* Current RAID6 layout has a RAID5
1613 strcat(strcpy(layout
, ls
), "-6");
1614 l
= map_name(r6layout
, layout
);
1616 return "Cannot find RAID6 layout to convert to";
1618 /* Current RAID6 has no equivalent.
1619 * If it is already a '-6' layout we
1620 * can leave it unchanged, else we must
1623 ls
= map_num(r6layout
,
1626 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1627 return "Please specify new layout";
1628 l
= info
->new_layout
;
1630 re
->after
.layout
= l
;
1636 /* We must already be at level 6 */
1638 return "Impossible level change";
1639 if (info
->new_layout
== UnSet
)
1640 re
->after
.layout
= info
->array
.layout
;
1642 re
->after
.layout
= info
->new_layout
;
1645 return "Impossible level change requested";
1647 if (info
->delta_disks
== UnSet
)
1648 info
->delta_disks
= delta_parity
;
1650 re
->after
.data_disks
=
1651 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1653 switch (re
->level
) {
1665 /* So we have a restripe operation, we need to calculate the number
1666 * of blocks per reshape operation.
1668 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1669 if (info
->new_chunk
== 0)
1670 info
->new_chunk
= info
->array
.chunk_size
;
1671 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1672 re
->after
.layout
== re
->before
.layout
&&
1673 info
->new_chunk
== info
->array
.chunk_size
) {
1674 /* Nothing to change, can change level immediately. */
1675 re
->level
= info
->new_level
;
1676 re
->backup_blocks
= 0;
1679 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1680 /* chunk and layout changes make no difference */
1681 re
->level
= info
->new_level
;
1682 re
->backup_blocks
= 0;
1686 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1687 get_linux_version() < 2006032)
1688 return "in-place reshape is not safe before 2.6.32 - sorry.";
1690 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1691 get_linux_version() < 2006030)
1692 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1694 re
->backup_blocks
= compute_backup_blocks(
1695 info
->new_chunk
, info
->array
.chunk_size
,
1696 re
->after
.data_disks
, re
->before
.data_disks
);
1697 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1699 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1703 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1706 struct mdinfo
*info
;
1710 if ((st
== NULL
) || (sra
== NULL
))
1713 if (text_version
== NULL
)
1714 text_version
= sra
->text_version
;
1715 subarray
= strchr(text_version
+ 1, '/')+1;
1716 info
= st
->ss
->container_content(st
, subarray
);
1718 unsigned long long current_size
= 0;
1719 unsigned long long new_size
= info
->custom_array_size
/2;
1721 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1722 new_size
> current_size
) {
1723 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1725 dprintf("Error: Cannot set array size");
1728 dprintf("Array size changed");
1730 dprintf_cont(" from %llu to %llu.\n",
1731 current_size
, new_size
);
1735 dprintf("Error: set_array_size(): info pointer in NULL\n");
1740 static int reshape_array(char *container
, int fd
, char *devname
,
1741 struct supertype
*st
, struct mdinfo
*info
,
1742 int force
, struct mddev_dev
*devlist
,
1743 unsigned long long data_offset
,
1744 char *backup_file
, int verbose
, int forked
,
1745 int restart
, int freeze_reshape
);
1746 static int reshape_container(char *container
, char *devname
,
1748 struct supertype
*st
,
1749 struct mdinfo
*info
,
1751 char *backup_file
, int verbose
,
1752 int forked
, int restart
, int freeze_reshape
);
1754 int Grow_reshape(char *devname
, int fd
,
1755 struct mddev_dev
*devlist
,
1756 unsigned long long data_offset
,
1757 struct context
*c
, struct shape
*s
)
1759 /* Make some changes in the shape of an array.
1760 * The kernel must support the change.
1762 * There are three different changes. Each can trigger
1763 * a resync or recovery so we freeze that until we have
1764 * requested everything (if kernel supports freezing - 2.6.30).
1766 * - change size (i.e. component_size)
1768 * - change layout/chunksize/ndisks
1770 * The last can require a reshape. It is different on different
1771 * levels so we need to check the level before actioning it.
1772 * Some times the level change needs to be requested after the
1773 * reshape (e.g. raid6->raid5, raid5->raid0)
1776 struct mdu_array_info_s array
;
1778 struct supertype
*st
;
1779 char *subarray
= NULL
;
1783 char *container
= NULL
;
1786 struct mddev_dev
*dv
;
1792 if (md_get_array_info(fd
, &array
) < 0) {
1793 pr_err("%s is not an active md array - aborting\n",
1797 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1798 (array
.level
< 4 || array
.level
> 6)) {
1799 pr_err("--grow --data-offset not yet supported\n");
1804 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1805 pr_err("cannot change component size at the same time as other changes.\n"
1806 " Change size first, then check data is intact before making other changes.\n");
1810 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&&
1811 array
.level
> 1 && get_linux_version() < 2006032 &&
1812 !check_env("MDADM_FORCE_FEWER")) {
1813 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1814 " Please use a newer kernel\n");
1818 if (array
.level
> 1 && s
->size
> 1 &&
1819 (array
.chunk_size
/ 1024) > (int)s
->size
) {
1820 pr_err("component size must be larger than chunk size.\n");
1824 st
= super_by_fd(fd
, &subarray
);
1826 pr_err("Unable to determine metadata format for %s\n", devname
);
1829 if (s
->raiddisks
> st
->max_devs
) {
1830 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1833 if (s
->level
== 0 &&
1834 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) &&
1835 !(array
.state
& (1<<MD_SB_CLUSTERED
))) {
1836 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
1837 if (md_set_array_info(fd
, &array
)!= 0) {
1838 pr_err("failed to remove internal bitmap.\n");
1843 /* in the external case we need to check that the requested reshape is
1844 * supported, and perform an initial check that the container holds the
1845 * pre-requisite spare devices (mdmon owns final validation)
1847 if (st
->ss
->external
) {
1851 container
= st
->container_devnm
;
1852 cfd
= open_dev_excl(st
->container_devnm
);
1854 container
= st
->devnm
;
1856 cfd
= open_dev_excl(st
->devnm
);
1860 pr_err("Unable to open container for %s\n", devname
);
1865 retval
= st
->ss
->load_container(st
, cfd
, NULL
);
1868 pr_err("Cannot read superblock for %s\n", devname
);
1873 /* check if operation is supported for metadata handler */
1874 if (st
->ss
->container_content
) {
1875 struct mdinfo
*cc
= NULL
;
1876 struct mdinfo
*content
= NULL
;
1878 cc
= st
->ss
->container_content(st
, subarray
);
1879 for (content
= cc
; content
; content
= content
->next
) {
1880 int allow_reshape
= 1;
1882 /* check if reshape is allowed based on metadata
1883 * indications stored in content.array.status
1885 if (content
->array
.state
&
1886 (1 << MD_SB_BLOCK_VOLUME
))
1888 if (content
->array
.state
&
1889 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE
))
1891 if (!allow_reshape
) {
1892 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1893 devname
, container
);
1898 if (content
->consistency_policy
==
1899 CONSISTENCY_POLICY_PPL
) {
1900 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1908 if (mdmon_running(container
))
1909 st
->update_tail
= &st
->updates
;
1913 for (dv
= devlist
; dv
; dv
= dv
->next
)
1915 if (s
->raiddisks
> array
.raid_disks
&&
1916 array
.spare_disks
+ added_disks
<
1917 (s
->raiddisks
- array
.raid_disks
) &&
1919 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1920 " Use --force to over-ride this check.\n",
1921 s
->raiddisks
- array
.raid_disks
,
1922 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1923 array
.spare_disks
+ added_disks
);
1927 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1928 GET_STATE
| GET_VERSION
);
1930 if (st
->ss
->external
&& subarray
== NULL
) {
1931 array
.level
= LEVEL_CONTAINER
;
1932 sra
->array
.level
= LEVEL_CONTAINER
;
1935 pr_err("failed to read sysfs parameters for %s\n",
1939 frozen
= freeze(st
);
1941 /* freeze() already spewed the reason */
1944 } else if (frozen
< 0) {
1945 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1950 /* ========= set size =============== */
1952 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1953 unsigned long long orig_size
= get_component_size(fd
)/2;
1954 unsigned long long min_csize
;
1956 int raid0_takeover
= 0;
1959 orig_size
= (unsigned) array
.size
;
1961 if (orig_size
== 0) {
1962 pr_err("Cannot set device size in this type of array.\n");
1967 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1968 devname
, APPLY_METADATA_CHANGES
,
1974 if (st
->ss
->external
) {
1975 /* metadata can have size limitation
1976 * update size value according to metadata information
1978 struct mdinfo
*sizeinfo
=
1979 st
->ss
->container_content(st
, subarray
);
1981 unsigned long long new_size
=
1982 sizeinfo
->custom_array_size
/2;
1983 int data_disks
= get_data_disks(
1984 sizeinfo
->array
.level
,
1985 sizeinfo
->array
.layout
,
1986 sizeinfo
->array
.raid_disks
);
1987 new_size
/= data_disks
;
1988 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1989 orig_size
, new_size
,
1990 new_size
* data_disks
);
1992 sysfs_free(sizeinfo
);
1996 /* Update the size of each member device in case
1997 * they have been resized. This will never reduce
1998 * below the current used-size. The "size" attribute
1999 * understands '0' to mean 'max'.
2002 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2003 sysfs_set_num(sra
, mdi
, "size",
2004 s
->size
== MAX_SIZE
? 0 : s
->size
);
2005 if (array
.not_persistent
== 0 &&
2006 array
.major_version
== 0 &&
2007 get_linux_version() < 3001000) {
2008 /* Dangerous to allow size to exceed 2TB */
2009 unsigned long long csize
;
2010 if (sysfs_get_ll(sra
, mdi
, "size",
2012 if (csize
>= 2ULL*1024*1024*1024)
2013 csize
= 2ULL*1024*1024*1024;
2014 if ((min_csize
== 0 ||
2015 (min_csize
> csize
)))
2020 if (min_csize
&& s
->size
> min_csize
) {
2021 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2023 goto size_change_error
;
2025 if (min_csize
&& s
->size
== MAX_SIZE
) {
2026 /* Don't let the kernel choose a size - it will get
2029 pr_err("Limited v0.90 array to 2TB per device\n");
2030 s
->size
= min_csize
;
2032 if (st
->ss
->external
) {
2033 if (sra
->array
.level
== 0) {
2034 rv
= sysfs_set_str(sra
, NULL
, "level", "raid5");
2037 /* get array parameters after takeover
2038 * to change one parameter at time only
2040 rv
= md_get_array_info(fd
, &array
);
2043 /* make sure mdmon is
2044 * aware of the new level */
2045 if (!mdmon_running(st
->container_devnm
))
2046 start_mdmon(st
->container_devnm
);
2047 ping_monitor(container
);
2048 if (mdmon_running(st
->container_devnm
) &&
2049 st
->update_tail
== NULL
)
2050 st
->update_tail
= &st
->updates
;
2053 if (s
->size
== MAX_SIZE
)
2055 array
.size
= s
->size
;
2056 if (s
->size
& ~INT32_MAX
) {
2057 /* got truncated to 32bit, write to
2058 * component_size instead
2061 rv
= sysfs_set_num(sra
, NULL
,
2062 "component_size", s
->size
);
2066 rv
= md_set_array_info(fd
, &array
);
2068 /* manage array size when it is managed externally
2070 if ((rv
== 0) && st
->ss
->external
)
2071 rv
= set_array_size(st
, sra
, sra
->text_version
);
2074 if (raid0_takeover
) {
2075 /* do not recync non-existing parity,
2076 * we will drop it anyway
2078 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2079 /* go back to raid0, drop parity disk
2081 sysfs_set_str(sra
, NULL
, "level", "raid0");
2082 md_get_array_info(fd
, &array
);
2089 /* restore metadata */
2090 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2091 UnSet
, NULL
, devname
,
2092 ROLLBACK_METADATA_CHANGES
,
2095 pr_err("Cannot set device size for %s: %s\n",
2096 devname
, strerror(err
));
2098 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2099 cont_err("Bitmap must be removed before size can be changed\n");
2103 if (s
->assume_clean
) {
2104 /* This will fail on kernels older than 3.0 unless
2105 * a backport has been arranged.
2108 sysfs_set_str(sra
, NULL
, "resync_start",
2110 pr_err("--assume-clean not supported with --grow on this kernel\n");
2112 md_get_array_info(fd
, &array
);
2113 s
->size
= get_component_size(fd
)/2;
2115 s
->size
= array
.size
;
2116 if (c
->verbose
>= 0) {
2117 if (s
->size
== orig_size
)
2118 pr_err("component size of %s unchanged at %lluK\n",
2121 pr_err("component size of %s has been set to %lluK\n",
2125 } else if (array
.level
!= LEVEL_CONTAINER
) {
2126 s
->size
= get_component_size(fd
)/2;
2128 s
->size
= array
.size
;
2131 /* See if there is anything else to do */
2132 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2133 (s
->layout_str
== NULL
) &&
2134 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2135 data_offset
== INVALID_SECTORS
&&
2136 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2137 /* Nothing more to do */
2138 if (!changed
&& c
->verbose
>= 0)
2139 pr_err("%s: no change requested\n", devname
);
2143 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2144 * current implementation assumes that following conditions must be met:
2147 * - near_copies == 2
2149 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2150 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2151 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2154 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2156 dprintf("Array cannot be reshaped\n");
2162 /* Make sure mdmon has seen the device removal
2163 * and updated metadata before we continue with
2167 ping_monitor(container
);
2170 memset(&info
, 0, sizeof(info
));
2172 if (sysfs_init(&info
, fd
, NULL
)) {
2173 pr_err("failed to intialize sysfs.\n");
2177 strcpy(info
.text_version
, sra
->text_version
);
2178 info
.component_size
= s
->size
*2;
2179 info
.new_level
= s
->level
;
2180 info
.new_chunk
= s
->chunk
* 1024;
2181 if (info
.array
.level
== LEVEL_CONTAINER
) {
2182 info
.delta_disks
= UnSet
;
2183 info
.array
.raid_disks
= s
->raiddisks
;
2184 } else if (s
->raiddisks
)
2185 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2187 info
.delta_disks
= UnSet
;
2188 if (s
->layout_str
== NULL
) {
2189 info
.new_layout
= UnSet
;
2190 if (info
.array
.level
== 6 &&
2191 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2192 info
.array
.layout
>= 16) {
2193 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2194 cont_err("during the reshape, please specify --layout=preserve\n");
2195 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2199 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2200 strcmp(s
->layout_str
, "normalize") == 0) {
2201 /* If we have a -6 RAID6 layout, remove the '-6'. */
2202 info
.new_layout
= UnSet
;
2203 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2205 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
2206 h
= strrchr(l
, '-');
2207 if (h
&& strcmp(h
, "-6") == 0) {
2209 info
.new_layout
= map_name(r6layout
, l
);
2212 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2216 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2217 /* This means that a non-standard RAID6 layout
2220 * - When reshape a RAID6 (e.g. adding a device)
2221 * which is in a non-standard layout, it is OK
2222 * to preserve that layout.
2223 * - When converting a RAID5 to RAID6, leave it in
2224 * the XXX-6 layout, don't re-layout.
2226 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2227 info
.new_layout
= info
.array
.layout
;
2228 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2230 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
2232 info
.new_layout
= map_name(r6layout
, l
);
2234 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2239 int l
= info
.new_level
;
2241 l
= info
.array
.level
;
2244 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2247 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2250 info
.new_layout
= parse_layout_10(s
->layout_str
);
2253 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2256 pr_err("layout not meaningful with this level\n");
2260 if (info
.new_layout
== UnSet
) {
2261 pr_err("layout %s not understood for this level\n",
2268 if (array
.level
== LEVEL_FAULTY
) {
2269 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2270 pr_err("cannot change level of Faulty device\n");
2274 pr_err("cannot set chunksize of Faulty device\n");
2277 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2278 pr_err("cannot set raid_disks of Faulty device\n");
2281 if (s
->layout_str
) {
2282 if (md_get_array_info(fd
, &array
) != 0) {
2283 dprintf("Cannot get array information.\n");
2286 array
.layout
= info
.new_layout
;
2287 if (md_set_array_info(fd
, &array
) != 0) {
2288 pr_err("failed to set new layout\n");
2290 } else if (c
->verbose
>= 0)
2291 printf("layout for %s set to %d\n",
2292 devname
, array
.layout
);
2294 } else if (array
.level
== LEVEL_CONTAINER
) {
2295 /* This change is to be applied to every array in the
2296 * container. This is only needed when the metadata imposes
2297 * restraints of the various arrays in the container.
2298 * Currently we only know that IMSM requires all arrays
2299 * to have the same number of devices so changing the
2300 * number of devices (On-Line Capacity Expansion) must be
2301 * performed at the level of the container
2307 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2308 c
->force
, c
->backup_file
, c
->verbose
,
2312 /* get spare devices from external metadata
2314 if (st
->ss
->external
) {
2315 struct mdinfo
*info2
;
2317 info2
= st
->ss
->container_content(st
, subarray
);
2319 info
.array
.spare_disks
=
2320 info2
->array
.spare_disks
;
2325 /* Impose these changes on a single array. First
2326 * check that the metadata is OK with the change. */
2328 if (reshape_super(st
, 0, info
.new_level
,
2329 info
.new_layout
, info
.new_chunk
,
2330 info
.array
.raid_disks
, info
.delta_disks
,
2331 c
->backup_file
, devname
,
2332 APPLY_METADATA_CHANGES
, c
->verbose
)) {
2337 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2338 devlist
, data_offset
, c
->backup_file
,
2339 c
->verbose
, 0, 0, 0);
2349 /* verify_reshape_position()
2350 * Function checks if reshape position in metadata is not farther
2351 * than position in md.
2353 * 0 : not valid sysfs entry
2354 * it can be caused by not started reshape, it should be started
2355 * by reshape array or raid0 array is before takeover
2356 * -1 : error, reshape position is obviously wrong
2357 * 1 : success, reshape progress correct or updated
2359 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2365 /* read sync_max, failure can mean raid0 array */
2366 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2370 unsigned long long position
= strtoull(buf
, &ep
, 0);
2372 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2373 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2374 position
*= get_data_disks(level
,
2376 info
->array
.raid_disks
);
2377 if (info
->reshape_progress
< position
) {
2378 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2379 info
->reshape_progress
, position
);
2380 info
->reshape_progress
= position
;
2382 } else if (info
->reshape_progress
> position
) {
2383 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2384 position
, info
->reshape_progress
);
2387 dprintf("Reshape position in md and metadata are the same;");
2391 } else if (rv
== 0) {
2392 /* for valid sysfs entry, 0-length content
2393 * should be indicated as error
2401 static unsigned long long choose_offset(unsigned long long lo
,
2402 unsigned long long hi
,
2403 unsigned long long min
,
2404 unsigned long long max
)
2406 /* Choose a new offset between hi and lo.
2407 * It must be between min and max, but
2408 * we would prefer something near the middle of hi/lo, and also
2409 * prefer to be aligned to a big power of 2.
2411 * So we start with the middle, then for each bit,
2412 * starting at '1' and increasing, if it is set, we either
2413 * add it or subtract it if possible, preferring the option
2414 * which is furthest from the boundary.
2416 * We stop once we get a 1MB alignment. As units are in sectors,
2417 * 1MB = 2*1024 sectors.
2419 unsigned long long choice
= (lo
+ hi
) / 2;
2420 unsigned long long bit
= 1;
2422 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2423 unsigned long long bigger
, smaller
;
2424 if (! (bit
& choice
))
2426 bigger
= choice
+ bit
;
2427 smaller
= choice
- bit
;
2428 if (bigger
> max
&& smaller
< min
)
2432 else if (smaller
< min
)
2434 else if (hi
- bigger
> smaller
- lo
)
2442 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2443 char *devname
, int delta_disks
,
2444 unsigned long long data_offset
,
2445 unsigned long long min
,
2451 unsigned long long before
, after
;
2453 /* Need to find min space before and after so same is used
2456 before
= UINT64_MAX
;
2458 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2462 struct supertype
*st2
;
2463 struct mdinfo info2
;
2465 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2467 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2468 dfd
= dev_open(dn
, O_RDONLY
);
2470 pr_err("%s: cannot open component %s\n",
2471 devname
, dn
? dn
: "-unknown-");
2474 st2
= dup_super(st
);
2475 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2479 pr_err("%s: cannot get superblock from %s\n",
2483 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2484 st2
->ss
->free_super(st2
);
2486 if (info2
.space_before
== 0 &&
2487 info2
.space_after
== 0) {
2488 /* Metadata doesn't support data_offset changes */
2490 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2494 if (before
> info2
.space_before
)
2495 before
= info2
.space_before
;
2496 if (after
> info2
.space_after
)
2497 after
= info2
.space_after
;
2499 if (data_offset
!= INVALID_SECTORS
) {
2501 if (info2
.data_offset
== data_offset
) {
2502 pr_err("%s: already has that data_offset\n",
2506 if (data_offset
< info2
.data_offset
)
2510 } else if ((data_offset
<= info2
.data_offset
&&
2512 (data_offset
>= info2
.data_offset
&&
2514 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2520 if (before
== UINT64_MAX
)
2521 /* impossible really, there must be no devices */
2524 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2525 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2526 unsigned long long new_data_offset
;
2528 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2530 if (delta_disks
< 0) {
2531 /* Don't need any space as array is shrinking
2532 * just move data_offset up by min
2534 if (data_offset
== INVALID_SECTORS
)
2535 new_data_offset
= sd
->data_offset
+ min
;
2537 if (data_offset
< sd
->data_offset
+ min
) {
2538 pr_err("--data-offset too small for %s\n",
2542 new_data_offset
= data_offset
;
2544 } else if (delta_disks
> 0) {
2545 /* need space before */
2549 pr_err("Insufficient head-space for reshape on %s\n",
2553 if (data_offset
== INVALID_SECTORS
)
2554 new_data_offset
= sd
->data_offset
- min
;
2556 if (data_offset
> sd
->data_offset
- min
) {
2557 pr_err("--data-offset too large for %s\n",
2561 new_data_offset
= data_offset
;
2565 /* can move up or down. If 'data_offset'
2566 * was set we would have already decided,
2567 * so just choose direction with most space.
2574 sysfs_set_str(sra
, NULL
, "reshape_direction",
2575 dir
== 1 ? "backwards" : "forwards");
2577 /* Increase data offset */
2581 pr_err("Insufficient tail-space for reshape on %s\n",
2585 if (data_offset
!= INVALID_SECTORS
&&
2586 data_offset
< sd
->data_offset
+ min
) {
2587 pr_err("--data-offset too small on %s\n",
2591 if (data_offset
!= INVALID_SECTORS
)
2592 new_data_offset
= data_offset
;
2594 new_data_offset
= choose_offset(sd
->data_offset
,
2595 sd
->data_offset
+ after
,
2596 sd
->data_offset
+ min
,
2597 sd
->data_offset
+ after
);
2599 /* Decrease data offset */
2603 pr_err("insufficient head-room on %s\n",
2607 if (data_offset
!= INVALID_SECTORS
&&
2608 data_offset
< sd
->data_offset
- min
) {
2609 pr_err("--data-offset too small on %s\n",
2613 if (data_offset
!= INVALID_SECTORS
)
2614 new_data_offset
= data_offset
;
2616 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2618 sd
->data_offset
- before
,
2619 sd
->data_offset
- min
);
2622 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2623 if (err
< 0 && errno
== E2BIG
) {
2624 /* try again after increasing data size to max */
2625 err
= sysfs_set_num(sra
, sd
, "size", 0);
2626 if (err
< 0 && errno
== EINVAL
&&
2627 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2628 /* some kernels have a bug where you cannot
2629 * use '0' on spare devices. */
2630 sysfs_set_num(sra
, sd
, "size",
2631 (sra
->component_size
+ after
)/2);
2633 err
= sysfs_set_num(sra
, sd
, "new_offset",
2637 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2638 pr_err("data-offset is too big for %s\n", dn
);
2641 if (sd
== sra
->devs
&&
2642 (errno
== ENOENT
|| errno
== E2BIG
))
2643 /* Early kernel, no 'new_offset' file,
2644 * or kernel doesn't like us.
2645 * For RAID5/6 this is not fatal
2648 pr_err("Cannot set new_offset for %s\n", dn
);
2656 /* Just use a backup file */
2660 static int raid10_reshape(char *container
, int fd
, char *devname
,
2661 struct supertype
*st
, struct mdinfo
*info
,
2662 struct reshape
*reshape
,
2663 unsigned long long data_offset
,
2664 int force
, int verbose
)
2666 /* Changing raid_disks, layout, chunksize or possibly
2667 * just data_offset for a RAID10.
2668 * We must always change data_offset. We change by at least
2669 * ->min_offset_change which is the largest of the old and new
2671 * If raid_disks is increasing, then data_offset must decrease
2672 * by at least this copy size.
2673 * If raid_disks is unchanged, data_offset must increase or
2674 * decrease by at least min_offset_change but preferably by much more.
2675 * We choose half of the available space.
2676 * If raid_disks is decreasing, data_offset must increase by
2677 * at least min_offset_change. To allow of this, component_size
2678 * must be decreased by the same amount.
2680 * So we calculate the required minimum and direction, possibly
2681 * reduce the component_size, then iterate through the devices
2682 * and set the new_data_offset.
2683 * If that all works, we set chunk_size, layout, raid_disks, and start
2687 unsigned long long min
;
2690 sra
= sysfs_read(fd
, NULL
,
2691 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2694 pr_err("%s: Cannot get array details from sysfs\n", devname
);
2697 min
= reshape
->min_offset_change
;
2699 if (info
->delta_disks
)
2700 sysfs_set_str(sra
, NULL
, "reshape_direction",
2701 info
->delta_disks
< 0 ? "backwards" : "forwards");
2702 if (info
->delta_disks
< 0 && info
->space_after
< min
) {
2703 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2704 (sra
->component_size
- min
)/2);
2706 pr_err("cannot reduce component size\n");
2710 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
,
2711 data_offset
, min
, 0);
2713 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2714 cont_err("supported on this kernel\n");
2720 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2722 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2723 reshape
->after
.layout
) < 0)
2726 sysfs_set_num(sra
, NULL
, "raid_disks",
2727 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2729 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2732 pr_err("Cannot set array shape for %s\n",
2735 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2736 cont_err(" Bitmap must be removed before shape can be changed\n");
2746 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2748 struct mdinfo
*sra
, *sd
;
2749 /* Initialisation to silence compiler warning */
2750 unsigned long long min_space_before
= 0, min_space_after
= 0;
2753 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2756 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2759 struct supertype
*st2
;
2760 struct mdinfo info2
;
2762 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2764 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2765 dfd
= dev_open(dn
, O_RDONLY
);
2768 st2
= dup_super(st
);
2769 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2775 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2776 st2
->ss
->free_super(st2
);
2779 min_space_before
> info2
.space_before
)
2780 min_space_before
= info2
.space_before
;
2782 min_space_after
> info2
.space_after
)
2783 min_space_after
= info2
.space_after
;
2786 if (sd
== NULL
&& !first
) {
2787 info
->space_after
= min_space_after
;
2788 info
->space_before
= min_space_before
;
2793 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2794 struct mdinfo
*info
,
2795 int disks
, unsigned long long blocks
)
2797 /* Check that the internal stripe cache is
2798 * large enough, or it won't work.
2799 * It must hold at least 4 stripes of the larger
2802 unsigned long cache
;
2803 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2804 cache
*= 4; /* 4 stripes minimum */
2805 cache
/= 512; /* convert to sectors */
2806 /* make sure there is room for 'blocks' with a bit to spare */
2807 if (cache
< 16 + blocks
/ disks
)
2808 cache
= 16 + blocks
/ disks
;
2809 cache
/= (4096/512); /* Convert from sectors to pages */
2811 if (sra
->cache_size
< cache
)
2812 subarray_set_num(container
, sra
, "stripe_cache_size",
2816 static int impose_reshape(struct mdinfo
*sra
,
2817 struct mdinfo
*info
,
2818 struct supertype
*st
,
2821 char *devname
, char *container
,
2822 struct reshape
*reshape
)
2824 struct mdu_array_info_s array
;
2826 sra
->new_chunk
= info
->new_chunk
;
2829 /* for external metadata checkpoint saved by mdmon can be lost
2830 * or missed /due to e.g. crash/. Check if md is not during
2831 * restart farther than metadata points to.
2832 * If so, this means metadata information is obsolete.
2834 if (st
->ss
->external
)
2835 verify_reshape_position(info
, reshape
->level
);
2836 sra
->reshape_progress
= info
->reshape_progress
;
2838 sra
->reshape_progress
= 0;
2839 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2840 /* start from the end of the new array */
2841 sra
->reshape_progress
= (sra
->component_size
2842 * reshape
->after
.data_disks
);
2845 md_get_array_info(fd
, &array
);
2846 if (info
->array
.chunk_size
== info
->new_chunk
&&
2847 reshape
->before
.layout
== reshape
->after
.layout
&&
2848 st
->ss
->external
== 0) {
2849 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2850 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2851 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2854 pr_err("Cannot set device shape for %s: %s\n",
2855 devname
, strerror(errno
));
2858 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2859 cont_err("Bitmap must be removed before shape can be changed\n");
2863 } else if (!restart
) {
2864 /* set them all just in case some old 'new_*' value
2865 * persists from some earlier problem.
2868 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2870 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2871 reshape
->after
.layout
) < 0)
2873 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2874 reshape
->after
.data_disks
+
2875 reshape
->parity
) < 0)
2878 pr_err("Cannot set device shape for %s\n", devname
);
2881 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2882 cont_err("Bitmap must be removed before shape can be changed\n");
2891 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2894 struct mdu_array_info_s array
;
2897 if (sysfs_init(&info
, fd
, NULL
)) {
2898 pr_err("failed to intialize sysfs.\n");
2902 md_get_array_info(fd
, &array
);
2903 if (level
== 0 && (array
.level
>= 4 && array
.level
<= 6)) {
2904 /* To convert to RAID0 we need to fail and
2905 * remove any non-data devices. */
2908 int data_disks
= array
.raid_disks
- 1;
2909 if (array
.level
== 6)
2911 if (array
.level
== 5 && array
.layout
!= ALGORITHM_PARITY_N
)
2913 if (array
.level
== 6 && array
.layout
!= ALGORITHM_PARITY_N_6
)
2915 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2916 /* First remove any spares so no recovery starts */
2917 for (d
= 0, found
= 0;
2918 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2919 mdu_disk_info_t disk
;
2921 if (md_get_disk_info(fd
, &disk
) < 0)
2923 if (disk
.major
== 0 && disk
.minor
== 0)
2926 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2927 disk
.raid_disk
< data_disks
)
2930 ioctl(fd
, HOT_REMOVE_DISK
,
2931 makedev(disk
.major
, disk
.minor
));
2933 /* Now fail anything left */
2934 md_get_array_info(fd
, &array
);
2935 for (d
= 0, found
= 0;
2936 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2937 mdu_disk_info_t disk
;
2939 if (md_get_disk_info(fd
, &disk
) < 0)
2941 if (disk
.major
== 0 && disk
.minor
== 0)
2944 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2945 disk
.raid_disk
< data_disks
)
2948 ioctl(fd
, SET_DISK_FAULTY
,
2949 makedev(disk
.major
, disk
.minor
));
2950 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2953 c
= map_num(pers
, level
);
2955 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2958 pr_err("%s: could not set level to %s\n",
2961 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2962 cont_err("Bitmap must be removed before level can be changed\n");
2966 pr_err("level of %s changed to %s\n", devname
, c
);
2972 static void catch_term(int sig
)
2977 static int continue_via_systemd(char *devnm
)
2979 int skipped
, i
, pid
, status
;
2981 /* In a systemd/udev world, it is best to get systemd to
2982 * run "mdadm --grow --continue" rather than running in the
2987 /* FIXME yuk. CLOSE_EXEC?? */
2989 for (i
= 3; skipped
< 20; i
++)
2995 /* Don't want to see error messages from
2996 * systemctl. If the service doesn't exist,
2997 * we fork ourselves.
3000 open("/dev/null", O_WRONLY
);
3001 snprintf(pathbuf
, sizeof(pathbuf
),
3002 "mdadm-grow-continue@%s.service", devnm
);
3003 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
3005 status
= execl("/bin/systemctl", "systemctl", "restart",
3008 case -1: /* Just do it ourselves. */
3010 default: /* parent - good */
3011 pid
= wait(&status
);
3012 if (pid
>= 0 && status
== 0)
3018 static int reshape_array(char *container
, int fd
, char *devname
,
3019 struct supertype
*st
, struct mdinfo
*info
,
3020 int force
, struct mddev_dev
*devlist
,
3021 unsigned long long data_offset
,
3022 char *backup_file
, int verbose
, int forked
,
3023 int restart
, int freeze_reshape
)
3025 struct reshape reshape
;
3028 int orig_level
= UnSet
;
3032 struct mdu_array_info_s array
;
3035 struct mddev_dev
*dv
;
3039 unsigned long long *offsets
= NULL
;
3043 unsigned long blocks
;
3044 unsigned long long array_size
;
3046 struct mdinfo
*sra
= NULL
;
3049 /* when reshaping a RAID0, the component_size might be zero.
3050 * So try to fix that up.
3052 if (md_get_array_info(fd
, &array
) != 0) {
3053 dprintf("Cannot get array information.\n");
3056 if (array
.level
== 0 && info
->component_size
== 0) {
3057 get_dev_size(fd
, NULL
, &array_size
);
3058 info
->component_size
= array_size
/ array
.raid_disks
;
3061 if (array
.level
== 10)
3062 /* Need space_after info */
3063 get_space_after(fd
, st
, info
);
3065 if (info
->reshape_active
) {
3066 int new_level
= info
->new_level
;
3067 info
->new_level
= UnSet
;
3068 if (info
->delta_disks
> 0)
3069 info
->array
.raid_disks
-= info
->delta_disks
;
3070 msg
= analyse_change(devname
, info
, &reshape
);
3071 info
->new_level
= new_level
;
3072 if (info
->delta_disks
> 0)
3073 info
->array
.raid_disks
+= info
->delta_disks
;
3075 /* Make sure the array isn't read-only */
3076 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3078 msg
= analyse_change(devname
, info
, &reshape
);
3080 /* if msg == "", error has already been printed */
3082 pr_err("%s\n", msg
);
3085 if (restart
&& (reshape
.level
!= info
->array
.level
||
3086 reshape
.before
.layout
!= info
->array
.layout
||
3087 reshape
.before
.data_disks
+ reshape
.parity
!=
3088 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3089 pr_err("reshape info is not in native format - cannot continue.\n");
3093 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3094 !((sysfs_get_str(info
, NULL
, "sync_action",
3095 buf
, sizeof(buf
)) > 0) &&
3096 (strncmp(buf
, "reshape", 7) == 0))) {
3097 /* When reshape is restarted from '0', very begin of array
3098 * it is possible that for external metadata reshape and array
3099 * configuration doesn't happen.
3100 * Check if md has the same opinion, and reshape is restarted
3101 * from 0. If so, this is regular reshape start after reshape
3102 * switch in metadata to next array only.
3104 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3105 (info
->reshape_progress
== 0))
3110 * reshape already started. just skip to monitoring
3113 if (reshape
.backup_blocks
== 0)
3115 if (restart
& RESHAPE_NO_BACKUP
)
3118 /* Need 'sra' down at 'started:' */
3119 sra
= sysfs_read(fd
, NULL
,
3120 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
3121 GET_CHUNK
|GET_CACHE
);
3123 pr_err("%s: Cannot get array details from sysfs\n",
3129 backup_file
= locate_backup(sra
->sys_name
);
3133 /* The container is frozen but the array may not be.
3134 * So freeze the array so spares don't get put to the wrong use
3135 * FIXME there should probably be a cleaner separation between
3136 * freeze_array and freeze_container.
3138 sysfs_freeze_array(info
);
3139 /* Check we have enough spares to not be degraded */
3141 for (dv
= devlist
; dv
; dv
=dv
->next
)
3143 spares_needed
= max(reshape
.before
.data_disks
,
3144 reshape
.after
.data_disks
) +
3145 reshape
.parity
- array
.raid_disks
;
3147 if (!force
&& info
->new_level
> 1 && info
->array
.level
> 1 &&
3148 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3149 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3150 " Use --force to over-ride this check.\n",
3152 spares_needed
== 1 ? "" : "s",
3153 info
->array
.spare_disks
+ added_disks
);
3156 /* Check we have enough spares to not fail */
3157 spares_needed
= max(reshape
.before
.data_disks
,
3158 reshape
.after
.data_disks
)
3160 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3161 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3162 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3163 spares_needed
, spares_needed
== 1 ? "" : "s",
3164 info
->array
.spare_disks
+ added_disks
);
3168 if (reshape
.level
!= array
.level
) {
3169 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3172 info
->new_layout
= UnSet
; /* after level change,
3173 * layout is meaningless */
3174 orig_level
= array
.level
;
3175 sysfs_freeze_array(info
);
3177 if (reshape
.level
> 0 && st
->ss
->external
) {
3178 /* make sure mdmon is aware of the new level */
3179 if (mdmon_running(container
))
3180 flush_mdmon(container
);
3182 if (!mdmon_running(container
))
3183 start_mdmon(container
);
3184 ping_monitor(container
);
3185 if (mdmon_running(container
) && st
->update_tail
== NULL
)
3186 st
->update_tail
= &st
->updates
;
3189 /* ->reshape_super might have chosen some spares from the
3190 * container that it wants to be part of the new array.
3191 * We can collect them with ->container_content and give
3192 * them to the kernel.
3194 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3195 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3196 struct mdinfo
*info2
=
3197 st
->ss
->container_content(st
, subarray
);
3201 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3202 pr_err("unable to initialize sysfs for %s\n",
3207 /* When increasing number of devices, we need to set
3208 * new raid_disks before adding these, or they might
3211 if (reshape
.backup_blocks
&&
3212 reshape
.after
.data_disks
>
3213 reshape
.before
.data_disks
)
3214 subarray_set_num(container
, info2
, "raid_disks",
3215 reshape
.after
.data_disks
+
3217 for (d
= info2
->devs
; d
; d
= d
->next
) {
3218 if (d
->disk
.state
== 0 &&
3219 d
->disk
.raid_disk
>= 0) {
3220 /* This is a spare that wants to
3221 * be part of the array.
3223 add_disk(fd
, st
, info2
, d
);
3229 /* We might have been given some devices to add to the
3230 * array. Now that the array has been changed to the right
3231 * level and frozen, we can safely add them.
3234 if (Manage_subdevs(devname
, fd
, devlist
, verbose
, 0, NULL
, 0))
3238 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3239 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3240 if (reshape
.backup_blocks
== 0) {
3241 /* No restriping needed, but we might need to impose
3242 * some more changes: layout, raid_disks, chunk_size
3244 /* read current array info */
3245 if (md_get_array_info(fd
, &array
) != 0) {
3246 dprintf("Cannot get array information.\n");
3249 /* compare current array info with new values and if
3250 * it is different update them to new */
3251 if (info
->new_layout
!= UnSet
&&
3252 info
->new_layout
!= array
.layout
) {
3253 array
.layout
= info
->new_layout
;
3254 if (md_set_array_info(fd
, &array
) != 0) {
3255 pr_err("failed to set new layout\n");
3257 } else if (verbose
>= 0)
3258 printf("layout for %s set to %d\n",
3259 devname
, array
.layout
);
3261 if (info
->delta_disks
!= UnSet
&& info
->delta_disks
!= 0 &&
3263 (info
->array
.raid_disks
+ info
->delta_disks
)) {
3264 array
.raid_disks
+= info
->delta_disks
;
3265 if (md_set_array_info(fd
, &array
) != 0) {
3266 pr_err("failed to set raid disks\n");
3268 } else if (verbose
>= 0) {
3269 printf("raid_disks for %s set to %d\n",
3270 devname
, array
.raid_disks
);
3273 if (info
->new_chunk
!= 0 &&
3274 info
->new_chunk
!= array
.chunk_size
) {
3275 if (sysfs_set_num(info
, NULL
,
3276 "chunk_size", info
->new_chunk
) != 0) {
3277 pr_err("failed to set chunk size\n");
3279 } else if (verbose
>= 0)
3280 printf("chunk size for %s set to %d\n",
3281 devname
, array
.chunk_size
);
3288 * There are three possibilities.
3289 * 1/ The array will shrink.
3290 * We need to ensure the reshape will pause before reaching
3291 * the 'critical section'. We also need to fork and wait for
3292 * that to happen. When it does we
3293 * suspend/backup/complete/unfreeze
3295 * 2/ The array will not change size.
3296 * This requires that we keep a backup of a sliding window
3297 * so that we can restore data after a crash. So we need
3298 * to fork and monitor progress.
3299 * In future we will allow the data_offset to change, so
3300 * a sliding backup becomes unnecessary.
3302 * 3/ The array will grow. This is relatively easy.
3303 * However the kernel's restripe routines will cheerfully
3304 * overwrite some early data before it is safe. So we
3305 * need to make a backup of the early parts of the array
3306 * and be ready to restore it if rebuild aborts very early.
3307 * For externally managed metadata, we still need a forked
3308 * child to monitor the reshape and suspend IO over the region
3309 * that is being reshaped.
3311 * We backup data by writing it to one spare, or to a
3312 * file which was given on command line.
3314 * In each case, we first make sure that storage is available
3315 * for the required backup.
3317 * - request the shape change.
3318 * - fork to handle backup etc.
3320 /* Check that we can hold all the data */
3321 get_dev_size(fd
, NULL
, &array_size
);
3322 if (reshape
.new_size
< (array_size
/512)) {
3323 pr_err("this change will reduce the size of the array.\n"
3324 " use --grow --array-size first to truncate array.\n"
3325 " e.g. mdadm --grow %s --array-size %llu\n",
3326 devname
, reshape
.new_size
/2);
3330 if (array
.level
== 10) {
3331 /* Reshaping RAID10 does not require any data backup by
3332 * user-space. Instead it requires that the data_offset
3333 * is changed to avoid the need for backup.
3334 * So this is handled very separately
3337 /* Nothing to do. */
3339 return raid10_reshape(container
, fd
, devname
, st
, info
,
3340 &reshape
, data_offset
, force
, verbose
);
3342 sra
= sysfs_read(fd
, NULL
,
3343 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3346 pr_err("%s: Cannot get array details from sysfs\n",
3352 switch(set_new_data_offset(sra
, st
, devname
,
3353 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3355 reshape
.min_offset_change
, 1)) {
3359 /* Updated data_offset, so it's easy now */
3360 update_cache_size(container
, sra
, info
,
3361 min(reshape
.before
.data_disks
,
3362 reshape
.after
.data_disks
),
3363 reshape
.backup_blocks
);
3365 /* Right, everything seems fine. Let's kick things off.
3369 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3370 devname
, container
, &reshape
) < 0)
3372 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3374 if (errno
!= EINVAL
) {
3375 pr_err("Failed to initiate reshape!\n");
3378 /* revert data_offset and try the old way */
3379 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3380 sysfs_set_num(sra
, sd
, "new_offset",
3382 sysfs_set_str(sra
, NULL
, "reshape_direction",
3387 if (info
->new_level
== reshape
.level
)
3389 /* need to adjust level when reshape completes */
3391 case -1: /* ignore error, but don't wait */
3393 default: /* parent */
3401 fd
= open_dev(sra
->sys_name
);
3403 impose_level(fd
, info
->new_level
, devname
, verbose
);
3405 case 1: /* Couldn't set data_offset, try the old way */
3406 if (data_offset
!= INVALID_SECTORS
) {
3407 pr_err("Cannot update data_offset on this array\n");
3414 /* Decide how many blocks (sectors) for a reshape
3415 * unit. The number we have so far is just a minimum
3417 blocks
= reshape
.backup_blocks
;
3418 if (reshape
.before
.data_disks
==
3419 reshape
.after
.data_disks
) {
3420 /* Make 'blocks' bigger for better throughput, but
3421 * not so big that we reject it below.
3422 * Try for 16 megabytes
3424 while (blocks
* 32 < sra
->component_size
&& blocks
< 16*1024*2)
3427 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3429 if (blocks
>= sra
->component_size
/2) {
3430 pr_err("%s: Something wrong - reshape aborted\n", devname
);
3434 /* Now we need to open all these devices so we can read/write.
3436 nrdisks
= max(reshape
.before
.data_disks
,
3437 reshape
.after
.data_disks
) + reshape
.parity
3438 + sra
->array
.spare_disks
;
3439 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3440 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3442 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3443 d
= reshape_prepare_fdlist(devname
, sra
, odisks
, nrdisks
, blocks
,
3444 backup_file
, fdlist
, offsets
);
3448 if ((st
->ss
->manage_reshape
== NULL
) ||
3449 (st
->ss
->recover_backup
== NULL
)) {
3450 if (backup_file
== NULL
) {
3451 if (reshape
.after
.data_disks
<=
3452 reshape
.before
.data_disks
) {
3453 pr_err("%s: Cannot grow - need backup-file\n",
3455 pr_err(" Please provide one with \"--backup=...\"\n");
3457 } else if (d
== odisks
) {
3458 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3462 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3464 fdlist
+d
, offsets
+d
,
3465 sra
->sys_name
, restart
)) {
3472 update_cache_size(container
, sra
, info
,
3473 min(reshape
.before
.data_disks
,
3474 reshape
.after
.data_disks
), blocks
);
3476 /* Right, everything seems fine. Let's kick things off.
3477 * If only changing raid_disks, use ioctl, else use
3482 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3483 devname
, container
, &reshape
) < 0)
3486 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3487 reshape
.after
.data_disks
);
3489 pr_err("Cannot %s reshape for %s\n",
3490 restart
? "continue" : "start", devname
);
3494 sysfs_set_str(sra
, NULL
, "array_state", "active");
3495 if (freeze_reshape
) {
3499 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3500 sra
->reshape_progress
);
3504 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3505 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3512 /* Now we just need to kick off the reshape and watch, while
3513 * handling backups of the data...
3514 * This is all done by a forked background process.
3516 switch(forked
? 0 : fork()) {
3518 pr_err("Cannot run child to monitor reshape: %s\n",
3532 /* If another array on the same devices is busy, the
3533 * reshape will wait for them. This would mean that
3534 * the first section that we suspend will stay suspended
3535 * for a long time. So check on that possibility
3536 * by looking for "DELAYED" in /proc/mdstat, and if found,
3540 struct mdstat_ent
*mds
, *m
;
3542 mds
= mdstat_read(1, 0);
3543 for (m
= mds
; m
; m
= m
->next
)
3544 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3545 if (m
->resync
&& m
->percent
== RESYNC_DELAYED
)
3548 /* Haven't started the reshape thread
3555 if (delayed
== 1 && get_linux_version() < 3007000) {
3556 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3557 " You might experience problems until other reshapes complete.\n");
3561 mdstat_wait(30 - (delayed
-1) * 25);
3565 if (check_env("MDADM_GROW_VERIFY"))
3566 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3569 mlockall(MCL_FUTURE
);
3571 signal(SIGTERM
, catch_term
);
3573 if (st
->ss
->external
) {
3574 /* metadata handler takes it from here */
3575 done
= st
->ss
->manage_reshape(
3576 fd
, sra
, &reshape
, st
, blocks
,
3577 fdlist
, offsets
, d
- odisks
, fdlist
+ odisks
,
3580 done
= child_monitor(
3581 fd
, sra
, &reshape
, st
, blocks
, fdlist
, offsets
,
3582 d
- odisks
, fdlist
+ odisks
, offsets
+ odisks
);
3587 if (backup_file
&& done
) {
3589 bul
= make_backup(sra
->sys_name
);
3592 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3600 unlink(backup_file
);
3607 if (!st
->ss
->external
&&
3608 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3609 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3611 /* no need to wait for the reshape to finish as
3612 * there is nothing more to do.
3619 if (st
->ss
->external
) {
3620 /* Re-load the metadata as much could have changed */
3621 int cfd
= open_dev(st
->container_devnm
);
3623 flush_mdmon(container
);
3624 st
->ss
->free_super(st
);
3625 st
->ss
->load_container(st
, cfd
, container
);
3630 /* set new array size if required customer_array_size is used
3633 if (reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3634 info
->custom_array_size
)
3635 set_array_size(st
, info
, info
->text_version
);
3637 if (info
->new_level
!= reshape
.level
) {
3639 fd
= open(devname
, O_RDONLY
);
3640 impose_level(fd
, info
->new_level
, devname
, verbose
);
3642 if (info
->new_level
== 0)
3643 st
->update_tail
= NULL
;
3655 if (orig_level
!= UnSet
&& sra
) {
3656 c
= map_num(pers
, orig_level
);
3657 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3658 pr_err("aborting level change\n");
3666 /* mdfd handle is passed to be closed in child process (after fork).
3668 int reshape_container(char *container
, char *devname
,
3670 struct supertype
*st
,
3671 struct mdinfo
*info
,
3673 char *backup_file
, int verbose
,
3674 int forked
, int restart
, int freeze_reshape
)
3676 struct mdinfo
*cc
= NULL
;
3678 char last_devnm
[32] = "";
3680 /* component_size is not meaningful for a container,
3681 * so pass '0' meaning 'no change'
3684 reshape_super(st
, 0, info
->new_level
,
3685 info
->new_layout
, info
->new_chunk
,
3686 info
->array
.raid_disks
, info
->delta_disks
,
3687 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3695 /* ping monitor to be sure that update is on disk
3697 ping_monitor(container
);
3699 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3700 if (continue_via_systemd(container
))
3703 switch (forked
? 0 : fork()) {
3704 case -1: /* error */
3705 perror("Cannot fork to complete reshape\n");
3708 default: /* parent */
3709 if (!freeze_reshape
)
3710 printf("%s: multi-array reshape continues in background\n", Name
);
3717 /* close unused handle in child process
3723 /* For each member array with reshape_active,
3724 * we need to perform the reshape.
3725 * We pick the first array that needs reshaping and
3726 * reshape it. reshape_array() will re-read the metadata
3727 * so the next time through a different array should be
3728 * ready for reshape.
3729 * It is possible that the 'different' array will not
3730 * be assembled yet. In that case we simple exit.
3731 * When it is assembled, the mdadm which assembles it
3732 * will take over the reshape.
3734 struct mdinfo
*content
;
3736 struct mdstat_ent
*mdstat
;
3742 cc
= st
->ss
->container_content(st
, NULL
);
3744 for (content
= cc
; content
; content
= content
->next
) {
3746 if (!content
->reshape_active
)
3749 subarray
= strchr(content
->text_version
+1, '/')+1;
3750 mdstat
= mdstat_by_subdev(subarray
, container
);
3753 if (mdstat
->active
== 0) {
3754 pr_err("Skipping inactive array %s.\n",
3756 free_mdstat(mdstat
);
3765 devid
= devnm2devid(mdstat
->devnm
);
3766 adev
= map_dev(major(devid
), minor(devid
), 0);
3768 adev
= content
->text_version
;
3770 fd
= open_dev(mdstat
->devnm
);
3772 pr_err("Device %s cannot be opened for reshape.\n",
3777 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3778 /* Do not allow for multiple reshape_array() calls for
3780 * It can happen when reshape_array() returns without
3781 * error, when reshape is not finished (wrong reshape
3782 * starting/continuation conditions). Mdmon doesn't
3783 * switch to next array in container and reentry
3784 * conditions for the same array occur.
3785 * This is possibly interim until the behaviour of
3786 * reshape_array is resolved().
3788 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3792 strcpy(last_devnm
, mdstat
->devnm
);
3794 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3795 pr_err("Unable to initialize sysfs for %s\n",
3801 if (mdmon_running(container
))
3802 flush_mdmon(container
);
3804 rv
= reshape_array(container
, fd
, adev
, st
,
3805 content
, force
, NULL
, INVALID_SECTORS
,
3806 backup_file
, verbose
, 1, restart
,
3810 if (freeze_reshape
) {
3819 if (mdmon_running(container
))
3820 flush_mdmon(container
);
3829 * We run a child process in the background which performs the following
3831 * - wait for resync to reach a certain point
3832 * - suspend io to the following section
3833 * - backup that section
3834 * - allow resync to proceed further
3836 * - discard the backup.
3838 * When are combined in slightly different ways in the three cases.
3840 * - suspend/backup/allow/wait/resume/discard
3842 * - allow/wait/suspend/backup/allow/wait/resume/discard
3844 * - wait/resume/discard/suspend/backup/allow
3846 * suspend/backup/allow always come together
3847 * wait/resume/discard do too.
3848 * For the same-size case we have two backups to improve flow.
3852 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3853 unsigned long long backup_point
,
3854 unsigned long long wait_point
,
3855 unsigned long long *suspend_point
,
3856 unsigned long long *reshape_completed
, int *frozen
)
3858 /* This function is called repeatedly by the reshape manager.
3859 * It determines how much progress can safely be made and allows
3861 * - 'info' identifies the array and particularly records in
3862 * ->reshape_progress the metadata's knowledge of progress
3863 * This is a sector offset from the start of the array
3864 * of the next array block to be relocated. This number
3865 * may increase from 0 or decrease from array_size, depending
3866 * on the type of reshape that is happening.
3867 * Note that in contrast, 'sync_completed' is a block count of the
3868 * reshape so far. It gives the distance between the start point
3869 * (head or tail of device) and the next place that data will be
3870 * written. It always increases.
3871 * - 'reshape' is the structure created by analyse_change
3872 * - 'backup_point' shows how much the metadata manager has backed-up
3873 * data. For reshapes with increasing progress, it is the next address
3874 * to be backed up, previous addresses have been backed-up. For
3875 * decreasing progress, it is the earliest address that has been
3876 * backed up - later address are also backed up.
3877 * So addresses between reshape_progress and backup_point are
3878 * backed up providing those are in the 'correct' order.
3879 * - 'wait_point' is an array address. When reshape_completed
3880 * passes this point, progress_reshape should return. It might
3881 * return earlier if it determines that ->reshape_progress needs
3882 * to be updated or further backup is needed.
3883 * - suspend_point is maintained by progress_reshape and the caller
3884 * should not touch it except to initialise to zero.
3885 * It is an array address and it only increases in 2.6.37 and earlier.
3886 * This makes it difficult to handle reducing reshapes with
3887 * external metadata.
3888 * However: it is similar to backup_point in that it records the
3889 * other end of a suspended region from reshape_progress.
3890 * it is moved to extend the region that is safe to backup and/or
3892 * - reshape_completed is read from sysfs and returned. The caller
3893 * should copy this into ->reshape_progress when it has reason to
3894 * believe that the metadata knows this, and any backup outside this
3898 * 1 if more data from backup_point - but only as far as suspend_point,
3899 * should be backed up
3900 * 0 if things are progressing smoothly
3901 * -1 if the reshape is finished because it is all done,
3902 * -2 if the reshape is finished due to an error.
3905 int advancing
= (reshape
->after
.data_disks
3906 >= reshape
->before
.data_disks
);
3907 unsigned long long need_backup
; /* All data between start of array and
3908 * here will at some point need to
3911 unsigned long long read_offset
, write_offset
;
3912 unsigned long long write_range
;
3913 unsigned long long max_progress
, target
, completed
;
3914 unsigned long long array_size
= (info
->component_size
3915 * reshape
->before
.data_disks
);
3919 /* First, we unsuspend any region that is now known to be safe.
3920 * If suspend_point is on the 'wrong' side of reshape_progress, then
3921 * we don't have or need suspension at the moment. This is true for
3922 * native metadata when we don't need to back-up.
3925 if (info
->reshape_progress
<= *suspend_point
)
3926 sysfs_set_num(info
, NULL
, "suspend_lo",
3927 info
->reshape_progress
);
3929 /* Note: this won't work in 2.6.37 and before.
3930 * Something somewhere should make sure we don't need it!
3932 if (info
->reshape_progress
>= *suspend_point
)
3933 sysfs_set_num(info
, NULL
, "suspend_hi",
3934 info
->reshape_progress
);
3937 /* Now work out how far it is safe to progress.
3938 * If the read_offset for ->reshape_progress is less than
3939 * 'blocks' beyond the write_offset, we can only progress as far
3941 * Otherwise we can progress until the write_offset for the new location
3942 * reaches (within 'blocks' of) the read_offset at the current location.
3943 * However that region must be suspended unless we are using native
3945 * If we need to suspend more, we limit it to 128M per device, which is
3946 * rather arbitrary and should be some time-based calculation.
3948 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3949 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3950 write_range
= info
->new_chunk
/512;
3951 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3952 need_backup
= array_size
;
3954 need_backup
= reshape
->backup_blocks
;
3956 if (read_offset
< write_offset
+ write_range
)
3957 max_progress
= backup_point
;
3960 read_offset
* reshape
->after
.data_disks
;
3962 if (read_offset
> write_offset
- write_range
)
3963 /* Can only progress as far as has been backed up,
3964 * which must be suspended */
3965 max_progress
= backup_point
;
3966 else if (info
->reshape_progress
<= need_backup
)
3967 max_progress
= backup_point
;
3969 if (info
->array
.major_version
>= 0)
3970 /* Can progress until backup is needed */
3971 max_progress
= need_backup
;
3973 /* Can progress until metadata update is required */
3975 read_offset
* reshape
->after
.data_disks
;
3976 /* but data must be suspended */
3977 if (max_progress
< *suspend_point
)
3978 max_progress
= *suspend_point
;
3983 /* We know it is safe to progress to 'max_progress' providing
3984 * it is suspended or we are using native metadata.
3985 * Consider extending suspend_point 128M per device if it
3986 * is less than 64M per device beyond reshape_progress.
3987 * But always do a multiple of 'blocks'
3988 * FIXME this is too big - it takes to long to complete
3991 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3992 reshape
->after
.data_disks
);
3993 target
/= reshape
->backup_blocks
;
3996 target
*= reshape
->backup_blocks
;
3998 /* For externally managed metadata we always need to suspend IO to
3999 * the area being reshaped so we regularly push suspend_point forward.
4000 * For native metadata we only need the suspend if we are going to do
4004 if ((need_backup
> info
->reshape_progress
||
4005 info
->array
.major_version
< 0) &&
4006 *suspend_point
< info
->reshape_progress
+ target
) {
4007 if (need_backup
< *suspend_point
+ 2 * target
)
4008 *suspend_point
= need_backup
;
4009 else if (*suspend_point
+ 2 * target
< array_size
)
4010 *suspend_point
+= 2 * target
;
4012 *suspend_point
= array_size
;
4013 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
4014 if (max_progress
> *suspend_point
)
4015 max_progress
= *suspend_point
;
4018 if (info
->array
.major_version
>= 0) {
4019 /* Only need to suspend when about to backup */
4020 if (info
->reshape_progress
< need_backup
* 2 &&
4021 *suspend_point
> 0) {
4023 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4024 sysfs_set_num(info
, NULL
, "suspend_hi",
4028 /* Need to suspend continually */
4029 if (info
->reshape_progress
< *suspend_point
)
4030 *suspend_point
= info
->reshape_progress
;
4031 if (*suspend_point
+ target
< info
->reshape_progress
)
4032 /* No need to move suspend region yet */;
4034 if (*suspend_point
>= 2 * target
)
4035 *suspend_point
-= 2 * target
;
4038 sysfs_set_num(info
, NULL
, "suspend_lo",
4041 if (max_progress
< *suspend_point
)
4042 max_progress
= *suspend_point
;
4046 /* now set sync_max to allow that progress. sync_max, like
4047 * sync_completed is a count of sectors written per device, so
4048 * we find the difference between max_progress and the start point,
4049 * and divide that by after.data_disks to get a sync_max
4051 * At the same time we convert wait_point to a similar number
4052 * for comparing against sync_completed.
4054 /* scale down max_progress to per_disk */
4055 max_progress
/= reshape
->after
.data_disks
;
4057 * Round to chunk size as some kernels give an erroneously
4060 max_progress
/= info
->new_chunk
/512;
4061 max_progress
*= info
->new_chunk
/512;
4062 /* And round to old chunk size as the kernel wants that */
4063 max_progress
/= info
->array
.chunk_size
/512;
4064 max_progress
*= info
->array
.chunk_size
/512;
4065 /* Limit progress to the whole device */
4066 if (max_progress
> info
->component_size
)
4067 max_progress
= info
->component_size
;
4068 wait_point
/= reshape
->after
.data_disks
;
4070 /* switch from 'device offset' to 'processed block count' */
4071 max_progress
= info
->component_size
- max_progress
;
4072 wait_point
= info
->component_size
- wait_point
;
4076 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4078 /* Now wait. If we have already reached the point that we were
4079 * asked to wait to, don't wait at all, else wait for any change.
4080 * We need to select on 'sync_completed' as that is the place that
4081 * notifications happen, but we are really interested in
4082 * 'reshape_position'
4084 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4086 goto check_progress
;
4088 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4089 goto check_progress
;
4091 while (completed
< max_progress
&& completed
< wait_point
) {
4092 /* Check that sync_action is still 'reshape' to avoid
4093 * waiting forever on a dead array
4096 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) <= 0 ||
4097 strncmp(action
, "reshape", 7) != 0)
4099 /* Some kernels reset 'sync_completed' to zero
4100 * before setting 'sync_action' to 'idle'.
4101 * So we need these extra tests.
4103 if (completed
== 0 && advancing
&&
4104 strncmp(action
, "idle", 4) == 0 &&
4105 info
->reshape_progress
> 0)
4107 if (completed
== 0 && !advancing
&&
4108 strncmp(action
, "idle", 4) == 0 &&
4109 info
->reshape_progress
<
4110 (info
->component_size
* reshape
->after
.data_disks
))
4112 sysfs_wait(fd
, NULL
);
4113 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4114 goto check_progress
;
4116 /* Some kernels reset 'sync_completed' to zero,
4117 * we need to have real point we are in md.
4118 * So in that case, read 'reshape_position' from sysfs.
4120 if (completed
== 0) {
4121 unsigned long long reshapep
;
4123 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) > 0 &&
4124 strncmp(action
, "idle", 4) == 0 &&
4125 sysfs_get_ll(info
, NULL
,
4126 "reshape_position", &reshapep
) == 0)
4127 *reshape_completed
= reshapep
;
4129 /* some kernels can give an incorrectly high
4130 * 'completed' number, so round down */
4131 completed
/= (info
->new_chunk
/512);
4132 completed
*= (info
->new_chunk
/512);
4133 /* Convert 'completed' back in to a 'progress' number */
4134 completed
*= reshape
->after
.data_disks
;
4136 completed
= (info
->component_size
4137 * reshape
->after
.data_disks
4139 *reshape_completed
= completed
;
4144 /* We return the need_backup flag. Caller will decide
4145 * how much - a multiple of ->backup_blocks up to *suspend_point
4148 return need_backup
> info
->reshape_progress
;
4150 return need_backup
>= info
->reshape_progress
;
4153 /* if we couldn't read a number from sync_completed, then
4154 * either the reshape did complete, or it aborted.
4155 * We can tell which by checking for 'none' in reshape_position.
4156 * If it did abort, then it might immediately restart if it
4157 * it was just a device failure that leaves us degraded but
4160 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
,
4161 sizeof(buf
)) < 0 || strncmp(buf
, "none", 4) != 0) {
4162 /* The abort might only be temporary. Wait up to 10
4163 * seconds for fd to contain a valid number again.
4167 unsigned long long new_sync_max
;
4168 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4169 if (sysfs_wait(fd
, &wait
) != 1)
4171 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4173 /* all good again */
4175 /* If "sync_max" is no longer max_progress
4176 * we need to freeze things
4178 sysfs_get_ll(info
, NULL
, "sync_max",
4180 *frozen
= (new_sync_max
!= max_progress
);
4182 case -2: /* read error - abort */
4189 return rv
; /* abort */
4191 /* Maybe racing with array shutdown - check state */
4194 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4196 strncmp(buf
, "inactive", 8) == 0 ||
4197 strncmp(buf
, "clear",5) == 0)
4198 return -2; /* abort */
4199 return -1; /* complete */
4203 /* FIXME return status is never checked */
4204 static int grow_backup(struct mdinfo
*sra
,
4205 unsigned long long offset
, /* per device */
4206 unsigned long stripes
, /* per device, in old chunks */
4207 int *sources
, unsigned long long *offsets
,
4208 int disks
, int chunk
, int level
, int layout
,
4209 int dests
, int *destfd
, unsigned long long *destoffsets
,
4210 int part
, int *degraded
,
4213 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4214 * to storage 'destfd' (offset 'destoffsets'), after first
4215 * suspending IO. Then allow resync to continue
4216 * over the suspended section.
4217 * Use part 'part' of the backup-super-block.
4222 unsigned long long ll
;
4224 //printf("offset %llu\n", offset);
4230 /* Check that array hasn't become degraded, else we might backup the wrong data */
4231 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4232 return -1; /* FIXME this error is ignored */
4233 new_degraded
= (int)ll
;
4234 if (new_degraded
!= *degraded
) {
4235 /* check each device to ensure it is still working */
4237 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4238 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4240 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4243 if (sysfs_get_str(sra
, sd
, "state",
4244 sbuf
, sizeof(sbuf
)) < 0 ||
4245 strstr(sbuf
, "faulty") ||
4246 strstr(sbuf
, "in_sync") == NULL
) {
4247 /* this device is dead */
4248 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4249 if (sd
->disk
.raid_disk
>= 0 &&
4250 sources
[sd
->disk
.raid_disk
] >= 0) {
4251 close(sources
[sd
->disk
.raid_disk
]);
4252 sources
[sd
->disk
.raid_disk
] = -1;
4257 *degraded
= new_degraded
;
4260 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4261 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4263 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4264 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4267 bsb
.magic
[15] = '2';
4268 for (i
= 0; i
< dests
; i
++)
4270 lseek64(destfd
[i
], destoffsets
[i
] +
4271 __le64_to_cpu(bsb
.devstart2
)*512, 0);
4273 lseek64(destfd
[i
], destoffsets
[i
], 0);
4275 rv
= save_stripes(sources
, offsets
, disks
, chunk
, level
, layout
,
4276 dests
, destfd
, offset
* 512 * odata
,
4277 stripes
* chunk
* odata
, buf
);
4281 bsb
.mtime
= __cpu_to_le64(time(0));
4282 for (i
= 0; i
< dests
; i
++) {
4283 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4285 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4286 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4287 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4288 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4289 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4292 if ((unsigned long long)lseek64(destfd
[i
],
4293 destoffsets
[i
] - 4096, 0) !=
4294 destoffsets
[i
] - 4096)
4296 if (write(destfd
[i
], &bsb
, 512) != 512)
4298 if (destoffsets
[i
] > 4096) {
4299 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4300 destoffsets
[i
]+stripes
*chunk
*odata
)
4302 if (write(destfd
[i
], &bsb
, 512) != 512)
4312 /* in 2.6.30, the value reported by sync_completed can be
4313 * less that it should be by one stripe.
4314 * This only happens when reshape hits sync_max and pauses.
4315 * So allow wait_backup to either extent sync_max further
4316 * than strictly necessary, or return before the
4317 * sync has got quite as far as we would really like.
4318 * This is what 'blocks2' is for.
4319 * The various caller give appropriate values so that
4322 /* FIXME return value is often ignored */
4323 static int forget_backup(int dests
, int *destfd
,
4324 unsigned long long *destoffsets
,
4328 * Erase backup 'part' (which is 0 or 1)
4334 bsb
.arraystart2
= __cpu_to_le64(0);
4335 bsb
.length2
= __cpu_to_le64(0);
4337 bsb
.arraystart
= __cpu_to_le64(0);
4338 bsb
.length
= __cpu_to_le64(0);
4340 bsb
.mtime
= __cpu_to_le64(time(0));
4342 for (i
= 0; i
< dests
; i
++) {
4343 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4344 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4345 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4346 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4347 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4348 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4349 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4350 destoffsets
[i
]-4096)
4352 if (rv
== 0 && write(destfd
[i
], &bsb
, 512) != 512)
4359 static void fail(char *msg
)
4362 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4363 rv
|= (write(2, "\n", 1) != 1);
4367 static char *abuf
, *bbuf
;
4368 static unsigned long long abuflen
;
4369 static void validate(int afd
, int bfd
, unsigned long long offset
)
4371 /* check that the data in the backup against the array.
4372 * This is only used for regression testing and should not
4373 * be used while the array is active
4377 lseek64(bfd
, offset
- 4096, 0);
4378 if (read(bfd
, &bsb2
, 512) != 512)
4379 fail("cannot read bsb");
4380 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4381 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4382 fail("first csum bad");
4383 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4384 fail("magic is bad");
4385 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4386 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4387 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4388 fail("second csum bad");
4390 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4391 fail("devstart is wrong");
4394 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4396 if (abuflen
< len
) {
4400 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4401 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4403 /* just stop validating on mem-alloc failure */
4408 lseek64(bfd
, offset
, 0);
4409 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4410 //printf("len %llu\n", len);
4411 fail("read first backup failed");
4413 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4414 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4415 fail("read first from array failed");
4416 if (memcmp(bbuf
, abuf
, len
) != 0) {
4419 printf("offset=%llu len=%llu\n",
4420 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4421 for (i
=0; i
<len
; i
++)
4422 if (bbuf
[i
] != abuf
[i
]) {
4423 printf("first diff byte %d\n", i
);
4427 fail("data1 compare failed");
4431 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4433 if (abuflen
< len
) {
4437 abuf
= xmalloc(abuflen
);
4438 bbuf
= xmalloc(abuflen
);
4441 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4442 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4443 fail("read second backup failed");
4444 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4445 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4446 fail("read second from array failed");
4447 if (memcmp(bbuf
, abuf
, len
) != 0)
4448 fail("data2 compare failed");
4452 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4453 struct supertype
*st
, unsigned long blocks
,
4454 int *fds
, unsigned long long *offsets
,
4455 int dests
, int *destfd
, unsigned long long *destoffsets
)
4457 /* Monitor a reshape where backup is being performed using
4458 * 'native' mechanism - either to a backup file, or
4459 * to some space in a spare.
4463 unsigned long long speed
;
4464 unsigned long long suspend_point
, array_size
;
4465 unsigned long long backup_point
, wait_point
;
4466 unsigned long long reshape_completed
;
4468 int increasing
= reshape
->after
.data_disks
>=
4469 reshape
->before
.data_disks
;
4470 int part
= 0; /* The next part of the backup area to fill. It
4471 * may already be full, so we need to check */
4472 int level
= reshape
->level
;
4473 int layout
= reshape
->before
.layout
;
4474 int data
= reshape
->before
.data_disks
;
4475 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4476 int chunk
= sra
->array
.chunk_size
;
4478 unsigned long stripes
;
4482 /* set up the backup-super-block. This requires the
4483 * uuid from the array.
4485 /* Find a superblock */
4486 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4490 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4492 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4493 devfd
= dev_open(dn
, O_RDONLY
);
4496 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4502 pr_err("Cannot find a superblock\n");
4506 memset(&bsb
, 0, 512);
4507 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4508 st
->ss
->uuid_from_super(st
, uuid
);
4509 memcpy(bsb
.set_uuid
, uuid
, 16);
4510 bsb
.mtime
= __cpu_to_le64(time(0));
4511 bsb
.devstart2
= blocks
;
4513 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4514 reshape
->before
.data_disks
;
4516 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4517 /* Don't start the 'reshape' */
4519 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4520 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4521 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4525 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4526 backup_point
= sra
->reshape_progress
;
4529 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4530 backup_point
= reshape
->backup_blocks
;
4531 suspend_point
= array_size
;
4537 /* Want to return as soon the oldest backup slot can
4538 * be released as that allows us to start backing up
4539 * some more, providing suspend_point has been
4540 * advanced, which it should have.
4543 wait_point
= array_size
;
4544 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4545 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4546 __le64_to_cpu(bsb
.length
));
4547 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4548 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4549 __le64_to_cpu(bsb
.length2
));
4552 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4553 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4554 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4555 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4558 reshape_completed
= sra
->reshape_progress
;
4559 rv
= progress_reshape(sra
, reshape
,
4560 backup_point
, wait_point
,
4561 &suspend_point
, &reshape_completed
,
4563 /* external metadata would need to ping_monitor here */
4564 sra
->reshape_progress
= reshape_completed
;
4566 /* Clear any backup region that is before 'here' */
4568 if (__le64_to_cpu(bsb
.length
) > 0 &&
4569 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4570 __le64_to_cpu(bsb
.length
)))
4571 forget_backup(dests
, destfd
,
4573 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4574 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4575 __le64_to_cpu(bsb
.length2
)))
4576 forget_backup(dests
, destfd
,
4579 if (__le64_to_cpu(bsb
.length
) > 0 &&
4580 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4581 forget_backup(dests
, destfd
,
4583 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4584 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4585 forget_backup(dests
, destfd
,
4595 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4596 /* No longer need to monitor this reshape */
4597 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4603 unsigned long long offset
;
4604 unsigned long actual_stripes
;
4605 /* Need to backup some data.
4606 * If 'part' is not used and the desired
4607 * backup size is suspended, do a backup,
4608 * then consider the next part.
4610 /* Check that 'part' is unused */
4611 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4613 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4616 offset
= backup_point
/ data
;
4617 actual_stripes
= stripes
;
4619 if (offset
+ actual_stripes
* (chunk
/512) >
4620 sra
->component_size
)
4621 actual_stripes
= ((sra
->component_size
- offset
)
4623 if (offset
+ actual_stripes
* (chunk
/512) >
4627 if (offset
< actual_stripes
* (chunk
/512))
4628 actual_stripes
= offset
/ (chunk
/512);
4629 offset
-= actual_stripes
* (chunk
/512);
4630 if (offset
< suspend_point
/data
)
4633 if (actual_stripes
== 0)
4635 grow_backup(sra
, offset
, actual_stripes
, fds
, offsets
,
4636 disks
, chunk
, level
, layout
, dests
, destfd
,
4637 destoffsets
, part
, °raded
, buf
);
4638 validate(afd
, destfd
[0], destoffsets
[0]);
4639 /* record where 'part' is up to */
4642 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4644 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4648 /* FIXME maybe call progress_reshape one more time instead */
4649 /* remove any remaining suspension */
4650 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4651 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4652 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4653 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4655 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4656 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4662 * If any spare contains md_back_data-1 which is recent wrt mtime,
4663 * write that data into the array and update the super blocks with
4664 * the new reshape_progress
4666 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
,
4667 int cnt
, char *backup_file
, int verbose
)
4671 unsigned long long *offsets
;
4672 unsigned long long nstripe
, ostripe
;
4675 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4676 if (info
->array
.level
== 6)
4677 odata
--; /* number of data disks */
4678 ndata
= info
->array
.raid_disks
- 1;
4679 if (info
->new_level
== 6)
4682 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4684 if (info
->delta_disks
<= 0)
4685 /* Didn't grow, so the backup file must have
4689 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4690 struct mdinfo dinfo
;
4693 char *devname
, namebuf
[20];
4694 unsigned long long lo
, hi
;
4696 /* This was a spare and may have some saved data on it.
4697 * Load the superblock, find and load the
4698 * backup_super_block.
4699 * If either fail, go on to next device.
4700 * If the backup contains no new info, just return
4701 * else restore data and update all superblocks
4703 if (i
== old_disks
-1) {
4704 fd
= open(backup_file
, O_RDONLY
);
4706 pr_err("backup file %s inaccessible: %s\n",
4707 backup_file
, strerror(errno
));
4710 devname
= backup_file
;
4715 if (st
->ss
->load_super(st
, fd
, NULL
))
4718 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4719 st
->ss
->free_super(st
);
4722 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4724 pr_err("Cannot seek on device %d\n", i
);
4725 continue; /* Cannot seek */
4727 sprintf(namebuf
, "device-%d", i
);
4730 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4732 pr_err("Cannot read from %s\n", devname
);
4733 continue; /* Cannot read */
4735 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4736 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4738 pr_err("No backup metadata on %s\n", devname
);
4741 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4743 pr_err("Bad backup-metadata checksum on %s\n",
4745 continue; /* bad checksum */
4747 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4748 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4750 pr_err("Bad backup-metadata checksum2 on %s\n",
4752 continue; /* Bad second checksum */
4754 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4756 pr_err("Wrong uuid on backup-metadata on %s\n",
4758 continue; /* Wrong uuid */
4762 * array utime and backup-mtime should be updated at
4763 * much the same time, but it seems that sometimes
4764 * they aren't... So allow considerable flexability in
4765 * matching, and allow this test to be overridden by
4766 * an environment variable.
4768 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4769 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4770 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4771 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4772 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4773 (unsigned long)info
->array
.utime
);
4775 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4776 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4777 continue; /* time stamp is too bad */
4781 if (bsb
.magic
[15] == '1') {
4782 if (bsb
.length
== 0)
4784 if (info
->delta_disks
>= 0) {
4785 /* reshape_progress is increasing */
4786 if (__le64_to_cpu(bsb
.arraystart
)
4787 + __le64_to_cpu(bsb
.length
)
4788 < info
->reshape_progress
) {
4791 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4792 continue; /* No new data here */
4795 /* reshape_progress is decreasing */
4796 if (__le64_to_cpu(bsb
.arraystart
) >=
4797 info
->reshape_progress
)
4798 goto nonew
; /* No new data here */
4801 if (bsb
.length
== 0 && bsb
.length2
== 0)
4803 if (info
->delta_disks
>= 0) {
4804 /* reshape_progress is increasing */
4805 if ((__le64_to_cpu(bsb
.arraystart
)
4806 + __le64_to_cpu(bsb
.length
)
4807 < info
->reshape_progress
) &&
4808 (__le64_to_cpu(bsb
.arraystart2
)
4809 + __le64_to_cpu(bsb
.length2
)
4810 < info
->reshape_progress
))
4811 goto nonew
; /* No new data here */
4813 /* reshape_progress is decreasing */
4814 if (__le64_to_cpu(bsb
.arraystart
) >=
4815 info
->reshape_progress
&&
4816 __le64_to_cpu(bsb
.arraystart2
) >=
4817 info
->reshape_progress
)
4818 goto nonew
; /* No new data here */
4821 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4824 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4826 continue; /* Cannot seek */
4828 /* There should be a duplicate backup superblock 4k before here */
4829 if (lseek64(fd
, -4096, 1) < 0 ||
4830 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4831 goto second_fail
; /* Cannot find leading superblock */
4832 if (bsb
.magic
[15] == '1')
4833 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4835 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4836 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4837 goto second_fail
; /* Cannot find leading superblock */
4839 /* Now need the data offsets for all devices. */
4840 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4841 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4844 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4845 /* FIXME should be this be an error */
4847 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4848 st
->ss
->free_super(st
);
4849 offsets
[j
] = dinfo
.data_offset
* 512;
4851 printf("%s: restoring critical section\n", Name
);
4853 if (restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4854 info
->new_chunk
, info
->new_level
,
4855 info
->new_layout
, fd
,
4856 __le64_to_cpu(bsb
.devstart
)*512,
4857 __le64_to_cpu(bsb
.arraystart
)*512,
4858 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4859 /* didn't succeed, so giveup */
4861 pr_err("Error restoring backup from %s\n",
4867 if (bsb
.magic
[15] == '2' &&
4868 restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4869 info
->new_chunk
, info
->new_level
,
4870 info
->new_layout
, fd
,
4871 __le64_to_cpu(bsb
.devstart
)*512 +
4872 __le64_to_cpu(bsb
.devstart2
)*512,
4873 __le64_to_cpu(bsb
.arraystart2
)*512,
4874 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4875 /* didn't succeed, so giveup */
4877 pr_err("Error restoring second backup from %s\n",
4885 /* Ok, so the data is restored. Let's update those superblocks. */
4889 lo
= __le64_to_cpu(bsb
.arraystart
);
4890 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4892 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4893 unsigned long long lo1
, hi1
;
4894 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4895 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4899 } else if (lo
< lo1
)
4904 if (lo
< hi
&& (info
->reshape_progress
< lo
||
4905 info
->reshape_progress
> hi
))
4906 /* backup does not affect reshape_progress*/ ;
4907 else if (info
->delta_disks
>= 0) {
4908 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4909 __le64_to_cpu(bsb
.length
);
4910 if (bsb
.magic
[15] == '2') {
4911 unsigned long long p2
;
4913 p2
= __le64_to_cpu(bsb
.arraystart2
) +
4914 __le64_to_cpu(bsb
.length2
);
4915 if (p2
> info
->reshape_progress
)
4916 info
->reshape_progress
= p2
;
4919 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4920 if (bsb
.magic
[15] == '2') {
4921 unsigned long long p2
;
4923 p2
= __le64_to_cpu(bsb
.arraystart2
);
4924 if (p2
< info
->reshape_progress
)
4925 info
->reshape_progress
= p2
;
4928 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4931 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4933 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4934 dinfo
.reshape_progress
= info
->reshape_progress
;
4935 st
->ss
->update_super(st
, &dinfo
, "_reshape_progress",
4937 st
->ss
->store_super(st
, fdlist
[j
]);
4938 st
->ss
->free_super(st
);
4942 /* Didn't find any backup data, try to see if any
4945 if (info
->delta_disks
< 0) {
4946 /* When shrinking, the critical section is at the end.
4947 * So see if we are before the critical section.
4949 unsigned long long first_block
;
4950 nstripe
= ostripe
= 0;
4952 while (ostripe
>= nstripe
) {
4953 ostripe
+= info
->array
.chunk_size
/ 512;
4954 first_block
= ostripe
* odata
;
4955 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4956 (info
->new_chunk
/512);
4959 if (info
->reshape_progress
>= first_block
)
4962 if (info
->delta_disks
> 0) {
4963 /* See if we are beyond the critical section. */
4964 unsigned long long last_block
;
4965 nstripe
= ostripe
= 0;
4967 while (nstripe
>= ostripe
) {
4968 nstripe
+= info
->new_chunk
/ 512;
4969 last_block
= nstripe
* ndata
;
4970 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4971 (info
->array
.chunk_size
/512);
4974 if (info
->reshape_progress
>= last_block
)
4977 /* needed to recover critical section! */
4979 pr_err("Failed to find backup of critical section\n");
4983 int Grow_continue_command(char *devname
, int fd
,
4984 char *backup_file
, int verbose
)
4987 struct supertype
*st
= NULL
;
4988 struct mdinfo
*content
= NULL
;
4989 struct mdinfo array
;
4990 char *subarray
= NULL
;
4991 struct mdinfo
*cc
= NULL
;
4992 struct mdstat_ent
*mdstat
= NULL
;
4996 dprintf("Grow continue from command line called for %s\n", devname
);
4998 st
= super_by_fd(fd
, &subarray
);
4999 if (!st
|| !st
->ss
) {
5000 pr_err("Unable to determine metadata format for %s\n", devname
);
5003 dprintf("Grow continue is run for ");
5004 if (st
->ss
->external
== 0) {
5007 dprintf_cont("native array (%s)\n", devname
);
5008 if (md_get_array_info(fd
, &array
.array
) < 0) {
5009 pr_err("%s is not an active md array - aborting\n",
5012 goto Grow_continue_command_exit
;
5015 sysfs_init(content
, fd
, NULL
);
5016 /* Need to load a superblock.
5017 * FIXME we should really get what we need from
5021 for (d
= 0; d
< MAX_DISKS
; d
++) {
5022 mdu_disk_info_t disk
;
5026 if (md_get_disk_info(fd
, &disk
) < 0)
5028 if (disk
.major
== 0 && disk
.minor
== 0)
5030 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5032 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5035 fd2
= dev_open(dv
, O_RDONLY
);
5038 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5044 if (d
== MAX_DISKS
) {
5045 pr_err("Unable to load metadata for %s\n",
5048 goto Grow_continue_command_exit
;
5050 st
->ss
->getinfo_super(st
, content
, NULL
);
5051 if (!content
->reshape_active
)
5055 } while (cnt
-- > 0);
5060 dprintf_cont("subarray (%s)\n", subarray
);
5061 container
= st
->container_devnm
;
5062 cfd
= open_dev_excl(st
->container_devnm
);
5064 container
= st
->devnm
;
5066 cfd
= open_dev_excl(st
->devnm
);
5067 dprintf_cont("container (%s)\n", container
);
5071 pr_err("Unable to open container for %s\n", devname
);
5073 goto Grow_continue_command_exit
;
5076 /* find in container array under reshape
5078 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5080 pr_err("Cannot read superblock for %s\n", devname
);
5082 goto Grow_continue_command_exit
;
5085 cc
= st
->ss
->container_content(st
, subarray
);
5086 for (content
= cc
; content
; content
= content
->next
) {
5088 int allow_reshape
= 1;
5090 if (content
->reshape_active
== 0)
5092 /* The decision about array or container wide
5093 * reshape is taken in Grow_continue based
5094 * content->reshape_active state, therefore we
5095 * need to check_reshape based on
5096 * reshape_active and subarray name
5098 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5100 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5101 (content
->array
.state
5102 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5105 if (!allow_reshape
) {
5106 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5107 devname
, container
);
5109 goto Grow_continue_command_exit
;
5112 array_name
= strchr(content
->text_version
+1, '/')+1;
5113 mdstat
= mdstat_by_subdev(array_name
, container
);
5116 if (mdstat
->active
== 0) {
5117 pr_err("Skipping inactive array %s.\n",
5119 free_mdstat(mdstat
);
5126 pr_err("Unable to determine reshaped array for %s\n", devname
);
5128 goto Grow_continue_command_exit
;
5130 fd2
= open_dev(mdstat
->devnm
);
5132 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5134 goto Grow_continue_command_exit
;
5137 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5138 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5142 goto Grow_continue_command_exit
;
5147 /* start mdmon in case it is not running
5149 if (!mdmon_running(container
))
5150 start_mdmon(container
);
5151 ping_monitor(container
);
5153 if (mdmon_running(container
))
5154 st
->update_tail
= &st
->updates
;
5156 pr_err("No mdmon found. Grow cannot continue.\n");
5158 goto Grow_continue_command_exit
;
5162 /* verify that array under reshape is started from
5165 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5167 goto Grow_continue_command_exit
;
5172 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5174 Grow_continue_command_exit
:
5177 st
->ss
->free_super(st
);
5178 free_mdstat(mdstat
);
5185 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5186 char *backup_file
, int forked
, int freeze_reshape
)
5190 if (!info
->reshape_active
)
5193 if (st
->ss
->external
) {
5194 int cfd
= open_dev(st
->container_devnm
);
5199 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5201 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5202 st
, info
, 0, backup_file
, 0,
5203 forked
, 1 | info
->reshape_active
,
5206 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5207 NULL
, INVALID_SECTORS
, backup_file
,
5208 0, forked
, 1 | info
->reshape_active
,
5214 char *make_backup(char *name
)
5216 char *base
= "backup_file-";
5220 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5221 fname
= xmalloc(len
);
5222 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5226 char *locate_backup(char *name
)
5228 char *fl
= make_backup(name
);
5231 if (stat(fl
, &stb
) == 0 && S_ISREG(stb
.st_mode
))