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
);
248 info
.array
.raid_disks
= nd
+1;
249 info
.array
.nr_disks
= nd
+1;
250 info
.array
.active_disks
= nd
+1;
251 info
.array
.working_disks
= nd
+1;
253 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
256 if (st
->ss
->store_super(st
, fd2
)) {
257 pr_err("Cannot store new superblock on %s\n", dv
);
267 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
270 * First check that array doesn't have a bitmap
271 * Then create the bitmap
274 * For internal bitmaps, we need to check the version,
275 * find all the active devices, and write the bitmap block
278 mdu_bitmap_file_t bmf
;
279 mdu_array_info_t array
;
280 struct supertype
*st
;
281 char *subarray
= NULL
;
282 int major
= BITMAP_MAJOR_HI
;
283 unsigned long long bitmapsize
, array_size
;
287 * We only ever get called if s->bitmap_file is != NULL, so this check
288 * is just here to quiet down static code checkers.
293 if (strcmp(s
->bitmap_file
, "clustered") == 0)
294 major
= BITMAP_MAJOR_CLUSTERED
;
296 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
298 pr_err("Memory allocation failure.\n");
300 pr_err("bitmaps not supported by this kernel.\n");
303 if (bmf
.pathname
[0]) {
304 if (strcmp(s
->bitmap_file
,"none") == 0) {
305 if (ioctl(fd
, SET_BITMAP_FILE
, -1) != 0) {
306 pr_err("failed to remove bitmap %s\n",
312 pr_err("%s already has a bitmap (%s)\n", devname
, bmf
.pathname
);
315 if (md_get_array_info(fd
, &array
) != 0) {
316 pr_err("cannot get array status for %s\n", devname
);
319 if (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) {
320 if (strcmp(s
->bitmap_file
, "none")==0) {
321 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
322 if (md_set_array_info(fd
, &array
) != 0) {
323 if (array
.state
& (1 << MD_SB_CLUSTERED
))
324 pr_err("failed to remove clustered bitmap.\n");
326 pr_err("failed to remove internal bitmap.\n");
331 pr_err("bitmap already present on %s\n", devname
);
335 if (strcmp(s
->bitmap_file
, "none") == 0) {
336 pr_err("no bitmap found on %s\n", devname
);
339 if (array
.level
<= 0) {
340 pr_err("Bitmaps not meaningful with level %s\n",
341 map_num(pers
, array
.level
)?:"of this array");
344 bitmapsize
= array
.size
;
346 if (get_dev_size(fd
, NULL
, &array_size
) &&
347 array_size
> (0x7fffffffULL
<< 9)) {
348 /* Array is big enough that we cannot trust array.size
349 * try other approaches
351 bitmapsize
= get_component_size(fd
);
353 if (bitmapsize
== 0) {
354 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
358 if (array
.level
== 10) {
361 ncopies
= (array
.layout
& 255) * ((array
.layout
>> 8) & 255);
362 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
364 if (strcmp(s
->bitmap_file
, "clustered") == 0 &&
365 !is_near_layout_10(array
.layout
)) {
366 pr_err("only near layout is supported with clustered raid10\n");
371 st
= super_by_fd(fd
, &subarray
);
373 pr_err("Cannot understand version %d.%d\n",
374 array
.major_version
, array
.minor_version
);
378 pr_err("Cannot add bitmaps to sub-arrays yet\n");
384 mdi
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
);
386 if (mdi
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
387 pr_err("Cannot add bitmap to array with PPL\n");
395 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
396 strcmp(s
->bitmap_file
, "clustered") == 0) {
399 int offset_setable
= 0;
400 if (st
->ss
->add_internal_bitmap
== NULL
) {
401 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
404 st
->nodes
= c
->nodes
;
405 st
->cluster_name
= c
->homecluster
;
406 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
409 for (d
= 0; d
< st
->max_devs
; d
++) {
410 mdu_disk_info_t disk
;
415 if (md_get_disk_info(fd
, &disk
) < 0)
417 if (disk
.major
== 0 && disk
.minor
== 0)
419 if ((disk
.state
& (1 << MD_DISK_SYNC
)) == 0)
421 dv
= map_dev(disk
.major
, disk
.minor
, 1);
424 fd2
= dev_open(dv
, O_RDWR
);
427 rv
= st
->ss
->load_super(st
, fd2
, NULL
);
429 rv
= st
->ss
->add_internal_bitmap(
430 st
, &s
->bitmap_chunk
, c
->delay
,
431 s
->write_behind
, bitmapsize
,
432 offset_setable
, major
);
434 st
->ss
->write_bitmap(st
, fd2
,
437 pr_err("failed to create internal bitmap - chunksize problem.\n");
440 pr_err("failed to load super-block.\n");
446 if (offset_setable
) {
447 st
->ss
->getinfo_super(st
, mdi
, NULL
);
448 if (sysfs_init(mdi
, fd
, NULL
)) {
449 pr_err("failed to intialize sysfs.\n");
452 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
456 if (strcmp(s
->bitmap_file
, "clustered") == 0)
457 array
.state
|= (1 << MD_SB_CLUSTERED
);
458 array
.state
|= (1 << MD_SB_BITMAP_PRESENT
);
459 rv
= md_set_array_info(fd
, &array
);
463 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
464 pr_err("failed to set internal bitmap.\n");
471 int max_devs
= st
->max_devs
;
473 /* try to load a superblock */
474 for (d
= 0; d
< max_devs
; d
++) {
475 mdu_disk_info_t disk
;
479 if (md_get_disk_info(fd
, &disk
) < 0)
481 if ((disk
.major
==0 && disk
.minor
== 0) ||
482 (disk
.state
& (1 << MD_DISK_REMOVED
)))
484 dv
= map_dev(disk
.major
, disk
.minor
, 1);
487 fd2
= dev_open(dv
, O_RDONLY
);
489 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
491 st
->ss
->uuid_from_super(st
, uuid
);
498 pr_err("cannot find UUID for array!\n");
501 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
,
502 s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
503 bitmapsize
, major
)) {
506 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
508 pr_err("weird: %s cannot be opened\n", s
->bitmap_file
);
511 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
514 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
515 pr_err("Cannot set bitmap file for %s: %s\n",
516 devname
, strerror(err
));
524 int Grow_consistency_policy(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
526 struct supertype
*st
;
529 char *subarray
= NULL
;
531 char container_dev
[PATH_MAX
];
534 if (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
535 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
536 pr_err("Operation not supported for consistency policy %s\n",
537 map_num(consistency_policies
, s
->consistency_policy
));
541 st
= super_by_fd(fd
, &subarray
);
545 sra
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
|GET_LEVEL
|
552 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
553 !st
->ss
->write_init_ppl
) {
554 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
559 if (sra
->array
.level
!= 5) {
560 pr_err("Operation not supported for array level %d\n",
566 if (sra
->consistency_policy
== (unsigned)s
->consistency_policy
) {
567 pr_err("Consistency policy is already %s\n",
568 map_num(consistency_policies
, s
->consistency_policy
));
571 } else if (sra
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
572 sra
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
573 pr_err("Current consistency policy is %s, cannot change to %s\n",
574 map_num(consistency_policies
, sra
->consistency_policy
),
575 map_num(consistency_policies
, s
->consistency_policy
));
580 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
581 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) <= 0) {
584 } else if (strcmp(buf
, "reshape\n") == 0) {
585 pr_err("PPL cannot be enabled when reshape is in progress\n");
594 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
599 sprintf(container_dev
, "/dev/%s", st
->container_devnm
);
601 ret
= Update_subarray(container_dev
, subarray
, update
, NULL
,
607 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
614 cfd
= open(container_dev
, O_RDWR
|O_EXCL
);
616 pr_err("Failed to open %s\n", container_dev
);
621 ret
= st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
625 pr_err("Cannot read superblock for %s\n",
630 mdi
= st
->ss
->container_content(st
, subarray
);
635 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
639 devpath
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
640 dfd
= dev_open(devpath
, O_RDWR
);
642 pr_err("Failed to open %s\n", devpath
);
648 ret
= st
->ss
->load_super(st
, dfd
, NULL
);
650 pr_err("Failed to load super-block.\n");
655 ret
= st
->ss
->update_super(st
, sra
, "ppl",
657 c
->verbose
, 0, NULL
);
660 st
->ss
->free_super(st
);
663 st
->ss
->getinfo_super(st
, &info
, NULL
);
666 ret
|= sysfs_set_num(sra
, sd
, "ppl_sector",
668 ret
|= sysfs_set_num(sra
, sd
, "ppl_size",
672 pr_err("Failed to set PPL attributes for %s\n",
675 st
->ss
->free_super(st
);
679 ret
= st
->ss
->write_init_ppl(st
, &info
, dfd
);
681 pr_err("Failed to write PPL\n");
686 st
->ss
->free_super(st
);
693 ret
= sysfs_set_str(sra
, NULL
, "consistency_policy",
694 map_num(consistency_policies
,
695 s
->consistency_policy
));
697 pr_err("Failed to change array consistency policy\n");
709 * When reshaping an array we might need to backup some data.
710 * This is written to all spares with a 'super_block' describing it.
711 * The superblock goes 4K from the end of the used space on the
713 * It if written after the backup is complete.
714 * It has the following structure.
717 static struct mdp_backup_super
{
718 char magic
[16]; /* md_backup_data-1 or -2 */
721 /* start/sizes in 512byte sectors */
722 __u64 devstart
; /* address on backup device/file of data */
725 __u32 sb_csum
; /* csum of preceeding bytes. */
727 __u64 devstart2
; /* offset in to data of second section */
730 __u32 sb_csum2
; /* csum of preceeding bytes. */
732 } __attribute__((aligned(512))) bsb
, bsb2
;
734 static __u32
bsb_csum(char *buf
, int len
)
738 for (i
= 0; i
< len
; i
++)
739 csum
= (csum
<<3) + buf
[0];
740 return __cpu_to_le32(csum
);
743 static int check_idle(struct supertype
*st
)
745 /* Check that all member arrays for this container, or the
746 * container of this array, are idle
748 char *container
= (st
->container_devnm
[0]
749 ? st
->container_devnm
: st
->devnm
);
750 struct mdstat_ent
*ent
, *e
;
753 ent
= mdstat_read(0, 0);
754 for (e
= ent
; e
; e
= e
->next
) {
755 if (!is_container_member(e
, container
))
757 if (e
->percent
>= 0) {
766 static int freeze_container(struct supertype
*st
)
768 char *container
= (st
->container_devnm
[0]
769 ? st
->container_devnm
: st
->devnm
);
774 if (block_monitor(container
, 1)) {
775 pr_err("failed to freeze container\n");
782 static void unfreeze_container(struct supertype
*st
)
784 char *container
= (st
->container_devnm
[0]
785 ? st
->container_devnm
: st
->devnm
);
787 unblock_monitor(container
, 1);
790 static int freeze(struct supertype
*st
)
792 /* Try to freeze resync/rebuild on this array/container.
793 * Return -1 if the array is busy,
794 * return -2 container cannot be frozen,
795 * return 0 if this kernel doesn't support 'frozen'
796 * return 1 if it worked.
798 if (st
->ss
->external
)
799 return freeze_container(st
);
801 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
807 /* Need to clear any 'read-auto' status */
808 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
809 strncmp(buf
, "read-auto", 9) == 0)
810 sysfs_set_str(sra
, NULL
, "array_state", "clean");
812 err
= sysfs_freeze_array(sra
);
818 static void unfreeze(struct supertype
*st
)
820 if (st
->ss
->external
)
821 return unfreeze_container(st
);
823 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
827 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
828 strcmp(buf
, "frozen\n") == 0)
829 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
834 static void wait_reshape(struct mdinfo
*sra
)
836 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
842 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
843 strncmp(action
, "reshape", 7) == 0)
844 sysfs_wait(fd
, NULL
);
848 static int reshape_super(struct supertype
*st
, unsigned long long size
,
849 int level
, int layout
, int chunksize
, int raid_disks
,
850 int delta_disks
, char *backup_file
, char *dev
,
851 int direction
, int verbose
)
853 /* nothing extra to check in the native case */
854 if (!st
->ss
->external
)
856 if (!st
->ss
->reshape_super
|| !st
->ss
->manage_reshape
) {
857 pr_err("%s metadata does not support reshape\n",
862 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
863 raid_disks
, delta_disks
, backup_file
, dev
,
867 static void sync_metadata(struct supertype
*st
)
869 if (st
->ss
->external
) {
870 if (st
->update_tail
) {
871 flush_metadata_updates(st
);
872 st
->update_tail
= &st
->updates
;
874 st
->ss
->sync_metadata(st
);
878 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
880 /* when dealing with external metadata subarrays we need to be
881 * prepared to handle EAGAIN. The kernel may need to wait for
882 * mdmon to mark the array active so the kernel can handle
883 * allocations/writeback when preparing the reshape action
884 * (md_allow_write()). We temporarily disable safe_mode_delay
885 * to close a race with the array_state going clean before the
886 * next write to raid_disks / stripe_cache_size
891 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
893 (strcmp(name
, "raid_disks") != 0 &&
894 strcmp(name
, "stripe_cache_size") != 0))
895 return sysfs_set_num(sra
, NULL
, name
, n
);
897 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
900 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
901 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
902 if (rc
< 0 && errno
== EAGAIN
) {
903 ping_monitor(container
);
904 /* if we get EAGAIN here then the monitor is not active
907 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
909 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
913 int start_reshape(struct mdinfo
*sra
, int already_running
,
914 int before_data_disks
, int data_disks
)
917 unsigned long long sync_max_to_set
;
919 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
920 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
921 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
922 sra
->reshape_progress
);
923 if (before_data_disks
<= data_disks
)
924 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
926 sync_max_to_set
= (sra
->component_size
* data_disks
927 - sra
->reshape_progress
) / data_disks
;
928 if (!already_running
)
929 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
930 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
931 if (!already_running
&& err
== 0) {
934 err
= sysfs_set_str(sra
, NULL
, "sync_action",
938 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
943 void abort_reshape(struct mdinfo
*sra
)
945 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
947 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
948 * suspend_hi to decrease as well as increase.")
949 * you could only increase suspend_{lo,hi} unless the region they
950 * covered was empty. So to reset to 0, you need to push suspend_lo
951 * up past suspend_hi first. So to maximize the chance of mdadm
952 * working on all kernels, we want to keep doing that.
954 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
955 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
956 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
957 sysfs_set_num(sra
, NULL
, "sync_min", 0);
958 // It isn't safe to reset sync_max as we aren't monitoring.
959 // Array really should be stopped at this point.
962 int remove_disks_for_takeover(struct supertype
*st
,
967 struct mdinfo
*remaining
;
970 if (st
->ss
->external
) {
972 struct mdinfo
*arrays
= st
->ss
->container_content(st
, NULL
);
974 * containter_content returns list of arrays in container
975 * If arrays->next is not NULL it means that there are
976 * 2 arrays in container and operation should be blocked
983 pr_err("Error. Cannot perform operation on /dev/%s\n", st
->devnm
);
984 pr_err("For this operation it MUST be single array in container\n");
990 if (sra
->array
.level
== 10)
991 nr_of_copies
= layout
& 0xff;
992 else if (sra
->array
.level
== 1)
993 nr_of_copies
= sra
->array
.raid_disks
;
997 remaining
= sra
->devs
;
999 /* for each 'copy', select one device and remove from the list. */
1000 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
1001 struct mdinfo
**diskp
;
1004 /* Find a working device to keep */
1005 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
1006 struct mdinfo
*disk
= *diskp
;
1008 if (disk
->disk
.raid_disk
< slot
)
1010 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
1012 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
1014 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
1016 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
1019 /* We have found a good disk to use! */
1020 *diskp
= disk
->next
;
1021 disk
->next
= sra
->devs
;
1030 if (slot
< sra
->array
.raid_disks
) {
1031 /* didn't find all slots */
1037 sra
->devs
= remaining
;
1041 /* Remove all 'remaining' devices from the array */
1043 struct mdinfo
*sd
= remaining
;
1044 remaining
= sd
->next
;
1046 sysfs_set_str(sra
, sd
, "state", "faulty");
1047 sysfs_set_str(sra
, sd
, "slot", "none");
1048 /* for external metadata disks should be removed in mdmon */
1049 if (!st
->ss
->external
)
1050 sysfs_set_str(sra
, sd
, "state", "remove");
1051 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
1052 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
1053 sd
->next
= sra
->devs
;
1059 void reshape_free_fdlist(int *fdlist
,
1060 unsigned long long *offsets
,
1065 for (i
= 0; i
< size
; i
++)
1073 int reshape_prepare_fdlist(char *devname
,
1077 unsigned long blocks
,
1080 unsigned long long *offsets
)
1085 enable_fds(nrdisks
);
1086 for (d
= 0; d
<= nrdisks
; d
++)
1089 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1090 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1092 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
1093 sd
->disk
.raid_disk
< raid_disks
) {
1094 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1095 fdlist
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDONLY
);
1096 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1097 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1098 pr_err("%s: cannot open component %s\n",
1099 devname
, dn
? dn
: "-unknown-");
1103 } else if (backup_file
== NULL
) {
1105 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1106 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1107 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1108 if (fdlist
[d
] < 0) {
1109 pr_err("%s: cannot open component %s\n",
1110 devname
, dn
? dn
: "-unknown-");
1121 int reshape_open_backup_file(char *backup_file
,
1126 unsigned long long *offsets
,
1130 /* Return 1 on success, 0 on any form of failure */
1131 /* need to check backup file is large enough */
1137 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
1141 pr_err("%s: cannot create backup file %s: %s\n",
1142 devname
, backup_file
, strerror(errno
));
1145 /* Guard against backup file being on array device.
1146 * If array is partitioned or if LVM etc is in the
1147 * way this will not notice, but it is better than
1150 fstat(*fdlist
, &stb
);
1153 if (stb
.st_rdev
== dev
) {
1154 pr_err("backup file must NOT be on the array being reshaped.\n");
1159 memset(buf
, 0, 512);
1160 for (i
=0; i
< blocks
+ 8 ; i
++) {
1161 if (write(*fdlist
, buf
, 512) != 512) {
1162 pr_err("%s: cannot create backup file %s: %s\n",
1163 devname
, backup_file
, strerror(errno
));
1167 if (fsync(*fdlist
) != 0) {
1168 pr_err("%s: cannot create backup file %s: %s\n",
1169 devname
, backup_file
, strerror(errno
));
1173 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
1174 char *bu
= make_backup(sys_name
);
1175 if (symlink(backup_file
, bu
))
1176 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
1184 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
1185 unsigned int ndata
, unsigned int odata
)
1187 unsigned long a
, b
, blocks
;
1188 /* So how much do we need to backup.
1189 * We need an amount of data which is both a whole number of
1190 * old stripes and a whole number of new stripes.
1191 * So LCM for (chunksize*datadisks).
1193 a
= (ochunk
/512) * odata
;
1194 b
= (nchunk
/512) * ndata
;
1197 /* LCM == product / GCD */
1198 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
1203 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1205 /* Based on the current array state in info->array and
1206 * the changes in info->new_* etc, determine:
1207 * - whether the change is possible
1208 * - Intermediate level/raid_disks/layout
1209 * - whether a restriping reshape is needed
1210 * - number of sectors in minimum change unit. This
1211 * will cover a whole number of stripes in 'before' and
1214 * Return message if the change should be rejected
1215 * NULL if the change can be achieved
1217 * This can be called as part of starting a reshape, or
1218 * when assembling an array that is undergoing reshape.
1220 int near
, far
, offset
, copies
;
1222 int old_chunk
, new_chunk
;
1223 /* delta_parity records change in number of devices
1224 * caused by level change
1226 int delta_parity
= 0;
1228 memset(re
, 0, sizeof(*re
));
1230 /* If a new level not explicitly given, we assume no-change */
1231 if (info
->new_level
== UnSet
)
1232 info
->new_level
= info
->array
.level
;
1234 if (info
->new_chunk
)
1235 switch (info
->new_level
) {
1241 /* chunk size is meaningful, must divide component_size
1244 if (info
->component_size
% (info
->new_chunk
/512)) {
1245 unsigned long long shrink
= info
->component_size
;
1246 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1247 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1248 info
->new_chunk
/1024, info
->component_size
/2);
1249 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1251 pr_err("will shrink the array so the given chunk size would work.\n");
1256 return "chunk size not meaningful for this level";
1259 info
->new_chunk
= info
->array
.chunk_size
;
1261 switch (info
->array
.level
) {
1263 return "No reshape is possibly for this RAID level";
1265 if (info
->delta_disks
!= UnSet
)
1266 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1268 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1270 /* RAID1 can convert to RAID1 with different disks, or
1271 * raid5 with 2 disks, or
1274 if (info
->new_level
> 1 && (info
->component_size
& 7))
1275 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1276 if (info
->new_level
== 0) {
1277 if (info
->delta_disks
!= UnSet
&&
1278 info
->delta_disks
!= 0)
1279 return "Cannot change number of disks with RAID1->RAID0 conversion";
1281 re
->before
.data_disks
= 1;
1282 re
->after
.data_disks
= 1;
1285 if (info
->new_level
== 1) {
1286 if (info
->delta_disks
== UnSet
)
1287 /* Don't know what to do */
1288 return "no change requested for Growing RAID1";
1292 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1293 return "Can only convert a 2-device array to RAID5";
1294 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1296 re
->before
.data_disks
= 1;
1297 if (info
->delta_disks
!= UnSet
&&
1298 info
->delta_disks
!= 0)
1299 re
->after
.data_disks
= 1 + info
->delta_disks
;
1301 re
->after
.data_disks
= 1;
1302 if (re
->after
.data_disks
< 1)
1303 return "Number of disks too small for RAID5";
1305 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1306 info
->array
.chunk_size
= 65536;
1309 /* Could do some multi-stage conversions, but leave that to
1312 return "Impossibly level change request for RAID1";
1315 /* RAID10 can be converted from near mode to
1316 * RAID0 by removing some devices.
1317 * It can also be reshaped if the kernel supports
1320 switch (info
->new_level
) {
1322 if ((info
->array
.layout
& ~0xff) != 0x100)
1323 return "Cannot Grow RAID10 with far/offset layout";
1325 * number of devices must be multiple of
1328 if (info
->array
.raid_disks
%
1329 (info
->array
.layout
& 0xff))
1330 return "RAID10 layout too complex for Grow operation";
1332 new_disks
= (info
->array
.raid_disks
/
1333 (info
->array
.layout
& 0xff));
1334 if (info
->delta_disks
== UnSet
)
1335 info
->delta_disks
= (new_disks
1336 - info
->array
.raid_disks
);
1338 if (info
->delta_disks
!=
1339 new_disks
- info
->array
.raid_disks
)
1340 return "New number of raid-devices impossible for RAID10";
1341 if (info
->new_chunk
&&
1342 info
->new_chunk
!= info
->array
.chunk_size
)
1343 return "Cannot change chunk-size with RAID10 Grow";
1347 re
->before
.data_disks
= new_disks
;
1348 re
->after
.data_disks
= re
->before
.data_disks
;
1352 near
= info
->array
.layout
& 0xff;
1353 far
= (info
->array
.layout
>> 8) & 0xff;
1354 offset
= info
->array
.layout
& 0x10000;
1355 if (far
> 1 && !offset
)
1356 return "Cannot reshape RAID10 in far-mode";
1357 copies
= near
* far
;
1359 old_chunk
= info
->array
.chunk_size
* far
;
1361 if (info
->new_layout
== UnSet
)
1362 info
->new_layout
= info
->array
.layout
;
1364 near
= info
->new_layout
& 0xff;
1365 far
= (info
->new_layout
>> 8) & 0xff;
1366 offset
= info
->new_layout
& 0x10000;
1367 if (far
> 1 && !offset
)
1368 return "Cannot reshape RAID10 to far-mode";
1369 if (near
* far
!= copies
)
1370 return "Cannot change number of copies when reshaping RAID10";
1372 if (info
->delta_disks
== UnSet
)
1373 info
->delta_disks
= 0;
1374 new_disks
= (info
->array
.raid_disks
+
1377 new_chunk
= info
->new_chunk
* far
;
1380 re
->before
.layout
= info
->array
.layout
;
1381 re
->before
.data_disks
= info
->array
.raid_disks
;
1382 re
->after
.layout
= info
->new_layout
;
1383 re
->after
.data_disks
= new_disks
;
1384 /* For RAID10 we don't do backup but do allow reshape,
1385 * so set backup_blocks to INVALID_SECTORS rather than
1387 * And there is no need to synchronise stripes on both
1388 * 'old' and 'new'. So the important
1389 * number is the minimum data_offset difference
1390 * which is the larger of (offset copies * chunk).
1392 re
->backup_blocks
= INVALID_SECTORS
;
1393 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1394 if (new_disks
< re
->before
.data_disks
&&
1395 info
->space_after
< re
->min_offset_change
)
1396 /* Reduce component size by one chunk */
1397 re
->new_size
= (info
->component_size
-
1398 re
->min_offset_change
);
1400 re
->new_size
= info
->component_size
;
1401 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1405 return "RAID10 can only be changed to RAID0";
1408 /* RAID0 can be converted to RAID10, or to RAID456 */
1409 if (info
->new_level
== 10) {
1410 if (info
->new_layout
== UnSet
&&
1411 info
->delta_disks
== UnSet
) {
1412 /* Assume near=2 layout */
1413 info
->new_layout
= 0x102;
1414 info
->delta_disks
= info
->array
.raid_disks
;
1416 if (info
->new_layout
== UnSet
) {
1417 int copies
= 1 + (info
->delta_disks
1418 / info
->array
.raid_disks
);
1419 if (info
->array
.raid_disks
* (copies
-1) !=
1421 return "Impossible number of devices for RAID0->RAID10";
1422 info
->new_layout
= 0x100 + copies
;
1424 if (info
->delta_disks
== UnSet
) {
1425 int copies
= info
->new_layout
& 0xff;
1426 if (info
->new_layout
!= 0x100 + copies
)
1427 return "New layout impossible for RAID0->RAID10";;
1428 info
->delta_disks
= (copies
- 1) *
1429 info
->array
.raid_disks
;
1431 if (info
->new_chunk
&&
1432 info
->new_chunk
!= info
->array
.chunk_size
)
1433 return "Cannot change chunk-size with RAID0->RAID10";
1436 re
->before
.data_disks
= (info
->array
.raid_disks
+
1438 re
->after
.data_disks
= re
->before
.data_disks
;
1439 re
->before
.layout
= info
->new_layout
;
1443 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1444 * a raid4 style layout of the final level.
1446 switch (info
->new_level
) {
1451 re
->before
.layout
= 0;
1456 re
->before
.layout
= ALGORITHM_PARITY_N
;
1457 if (info
->new_layout
== UnSet
)
1458 info
->new_layout
= map_name(r5layout
, "default");
1463 re
->before
.layout
= ALGORITHM_PARITY_N
;
1464 if (info
->new_layout
== UnSet
)
1465 info
->new_layout
= map_name(r6layout
, "default");
1468 return "Impossible level change requested";
1470 re
->before
.data_disks
= info
->array
.raid_disks
;
1471 /* determining 'after' layout happens outside this 'switch' */
1475 info
->array
.layout
= ALGORITHM_PARITY_N
;
1477 switch (info
->new_level
) {
1481 re
->level
= info
->array
.level
;
1482 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1483 re
->before
.layout
= info
->array
.layout
;
1487 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1488 re
->before
.layout
= info
->array
.layout
;
1493 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1494 switch (info
->array
.layout
) {
1495 case ALGORITHM_LEFT_ASYMMETRIC
:
1496 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1498 case ALGORITHM_RIGHT_ASYMMETRIC
:
1499 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1501 case ALGORITHM_LEFT_SYMMETRIC
:
1502 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1504 case ALGORITHM_RIGHT_SYMMETRIC
:
1505 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1507 case ALGORITHM_PARITY_0
:
1508 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1510 case ALGORITHM_PARITY_N
:
1511 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1514 return "Cannot convert an array with this layout";
1518 if (info
->array
.raid_disks
!= 2)
1519 return "Can only convert a 2-device array to RAID1";
1520 if (info
->delta_disks
!= UnSet
&&
1521 info
->delta_disks
!= 0)
1522 return "Cannot set raid_disk when converting RAID5->RAID1";
1524 info
->new_chunk
= 0;
1527 return "Impossible level change requested";
1531 switch (info
->new_level
) {
1537 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1538 re
->before
.layout
= info
->array
.layout
;
1541 return "Impossible level change requested";
1546 /* If we reached here then it looks like a re-stripe is
1547 * happening. We have determined the intermediate level
1548 * and initial raid_disks/layout and stored these in 're'.
1550 * We need to deduce the final layout that can be atomically
1551 * converted to the end state.
1553 switch (info
->new_level
) {
1555 /* We can only get to RAID0 from RAID4 or RAID5
1556 * with appropriate layout and one extra device
1558 if (re
->level
!= 4 && re
->level
!= 5)
1559 return "Cannot covert to RAID0 from this level";
1561 switch (re
->level
) {
1563 re
->before
.layout
= 0;
1564 re
->after
.layout
= 0;
1567 re
->after
.layout
= ALGORITHM_PARITY_N
;
1573 /* We can only get to RAID4 from RAID5 */
1574 if (re
->level
!= 4 && re
->level
!= 5)
1575 return "Cannot convert to RAID4 from this level";
1577 switch (re
->level
) {
1579 re
->after
.layout
= 0;
1582 re
->after
.layout
= ALGORITHM_PARITY_N
;
1588 /* We get to RAID5 from RAID5 or RAID6 */
1589 if (re
->level
!= 5 && re
->level
!= 6)
1590 return "Cannot convert to RAID5 from this level";
1592 switch (re
->level
) {
1594 if (info
->new_layout
== UnSet
)
1595 re
->after
.layout
= re
->before
.layout
;
1597 re
->after
.layout
= info
->new_layout
;
1600 if (info
->new_layout
== UnSet
)
1601 info
->new_layout
= re
->before
.layout
;
1603 /* after.layout needs to be raid6 version of new_layout */
1604 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1605 re
->after
.layout
= ALGORITHM_PARITY_N
;
1608 char *ls
= map_num(r5layout
, info
->new_layout
);
1611 /* Current RAID6 layout has a RAID5
1614 strcat(strcpy(layout
, ls
), "-6");
1615 l
= map_name(r6layout
, layout
);
1617 return "Cannot find RAID6 layout to convert to";
1619 /* Current RAID6 has no equivalent.
1620 * If it is already a '-6' layout we
1621 * can leave it unchanged, else we must
1624 ls
= map_num(r6layout
,
1627 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1628 return "Please specify new layout";
1629 l
= info
->new_layout
;
1631 re
->after
.layout
= l
;
1637 /* We must already be at level 6 */
1639 return "Impossible level change";
1640 if (info
->new_layout
== UnSet
)
1641 re
->after
.layout
= info
->array
.layout
;
1643 re
->after
.layout
= info
->new_layout
;
1646 return "Impossible level change requested";
1648 if (info
->delta_disks
== UnSet
)
1649 info
->delta_disks
= delta_parity
;
1651 re
->after
.data_disks
=
1652 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1654 switch (re
->level
) {
1666 /* So we have a restripe operation, we need to calculate the number
1667 * of blocks per reshape operation.
1669 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1670 if (info
->new_chunk
== 0)
1671 info
->new_chunk
= info
->array
.chunk_size
;
1672 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1673 re
->after
.layout
== re
->before
.layout
&&
1674 info
->new_chunk
== info
->array
.chunk_size
) {
1675 /* Nothing to change, can change level immediately. */
1676 re
->level
= info
->new_level
;
1677 re
->backup_blocks
= 0;
1680 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1681 /* chunk and layout changes make no difference */
1682 re
->level
= info
->new_level
;
1683 re
->backup_blocks
= 0;
1687 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1688 get_linux_version() < 2006032)
1689 return "in-place reshape is not safe before 2.6.32 - sorry.";
1691 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1692 get_linux_version() < 2006030)
1693 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1695 re
->backup_blocks
= compute_backup_blocks(
1696 info
->new_chunk
, info
->array
.chunk_size
,
1697 re
->after
.data_disks
, re
->before
.data_disks
);
1698 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1700 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1704 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1707 struct mdinfo
*info
;
1711 if ((st
== NULL
) || (sra
== NULL
))
1714 if (text_version
== NULL
)
1715 text_version
= sra
->text_version
;
1716 subarray
= strchr(text_version
+ 1, '/')+1;
1717 info
= st
->ss
->container_content(st
, subarray
);
1719 unsigned long long current_size
= 0;
1720 unsigned long long new_size
= info
->custom_array_size
/2;
1722 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1723 new_size
> current_size
) {
1724 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1726 dprintf("Error: Cannot set array size");
1729 dprintf("Array size changed");
1731 dprintf_cont(" from %llu to %llu.\n",
1732 current_size
, new_size
);
1736 dprintf("Error: set_array_size(): info pointer in NULL\n");
1741 static int reshape_array(char *container
, int fd
, char *devname
,
1742 struct supertype
*st
, struct mdinfo
*info
,
1743 int force
, struct mddev_dev
*devlist
,
1744 unsigned long long data_offset
,
1745 char *backup_file
, int verbose
, int forked
,
1746 int restart
, int freeze_reshape
);
1747 static int reshape_container(char *container
, char *devname
,
1749 struct supertype
*st
,
1750 struct mdinfo
*info
,
1752 char *backup_file
, int verbose
,
1753 int forked
, int restart
, int freeze_reshape
);
1755 int Grow_reshape(char *devname
, int fd
,
1756 struct mddev_dev
*devlist
,
1757 unsigned long long data_offset
,
1758 struct context
*c
, struct shape
*s
)
1760 /* Make some changes in the shape of an array.
1761 * The kernel must support the change.
1763 * There are three different changes. Each can trigger
1764 * a resync or recovery so we freeze that until we have
1765 * requested everything (if kernel supports freezing - 2.6.30).
1767 * - change size (i.e. component_size)
1769 * - change layout/chunksize/ndisks
1771 * The last can require a reshape. It is different on different
1772 * levels so we need to check the level before actioning it.
1773 * Some times the level change needs to be requested after the
1774 * reshape (e.g. raid6->raid5, raid5->raid0)
1777 struct mdu_array_info_s array
;
1779 struct supertype
*st
;
1780 char *subarray
= NULL
;
1784 char *container
= NULL
;
1787 struct mddev_dev
*dv
;
1793 if (md_get_array_info(fd
, &array
) < 0) {
1794 pr_err("%s is not an active md array - aborting\n",
1798 if (s
->level
!= UnSet
&& s
->chunk
) {
1799 pr_err("Cannot change array level in the same operation as changing chunk size.\n");
1803 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1804 (array
.level
< 4 || array
.level
> 6)) {
1805 pr_err("--grow --data-offset not yet supported\n");
1810 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1811 pr_err("cannot change component size at the same time as other changes.\n"
1812 " Change size first, then check data is intact before making other changes.\n");
1816 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&&
1817 array
.level
> 1 && get_linux_version() < 2006032 &&
1818 !check_env("MDADM_FORCE_FEWER")) {
1819 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1820 " Please use a newer kernel\n");
1824 if (array
.level
> 1 && s
->size
> 1 &&
1825 (unsigned long long) (array
.chunk_size
/ 1024) > s
->size
) {
1826 pr_err("component size must be larger than chunk size.\n");
1830 st
= super_by_fd(fd
, &subarray
);
1832 pr_err("Unable to determine metadata format for %s\n", devname
);
1835 if (s
->raiddisks
> st
->max_devs
) {
1836 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1839 if (s
->level
== 0 &&
1840 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) &&
1841 !(array
.state
& (1<<MD_SB_CLUSTERED
))) {
1842 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
1843 if (md_set_array_info(fd
, &array
)!= 0) {
1844 pr_err("failed to remove internal bitmap.\n");
1849 /* in the external case we need to check that the requested reshape is
1850 * supported, and perform an initial check that the container holds the
1851 * pre-requisite spare devices (mdmon owns final validation)
1853 if (st
->ss
->external
) {
1857 container
= st
->container_devnm
;
1858 cfd
= open_dev_excl(st
->container_devnm
);
1860 container
= st
->devnm
;
1862 cfd
= open_dev_excl(st
->devnm
);
1866 pr_err("Unable to open container for %s\n", devname
);
1871 retval
= st
->ss
->load_container(st
, cfd
, NULL
);
1874 pr_err("Cannot read superblock for %s\n", devname
);
1879 /* check if operation is supported for metadata handler */
1880 if (st
->ss
->container_content
) {
1881 struct mdinfo
*cc
= NULL
;
1882 struct mdinfo
*content
= NULL
;
1884 cc
= st
->ss
->container_content(st
, subarray
);
1885 for (content
= cc
; content
; content
= content
->next
) {
1886 int allow_reshape
= 1;
1888 /* check if reshape is allowed based on metadata
1889 * indications stored in content.array.status
1891 if (content
->array
.state
&
1892 (1 << MD_SB_BLOCK_VOLUME
))
1894 if (content
->array
.state
&
1895 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE
))
1897 if (!allow_reshape
) {
1898 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1899 devname
, container
);
1904 if (content
->consistency_policy
==
1905 CONSISTENCY_POLICY_PPL
) {
1906 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1914 if (mdmon_running(container
))
1915 st
->update_tail
= &st
->updates
;
1919 for (dv
= devlist
; dv
; dv
= dv
->next
)
1921 if (s
->raiddisks
> array
.raid_disks
&&
1922 array
.spare_disks
+ added_disks
<
1923 (s
->raiddisks
- array
.raid_disks
) &&
1925 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1926 " Use --force to over-ride this check.\n",
1927 s
->raiddisks
- array
.raid_disks
,
1928 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1929 array
.spare_disks
+ added_disks
);
1933 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1934 GET_STATE
| GET_VERSION
);
1936 if (st
->ss
->external
&& subarray
== NULL
) {
1937 array
.level
= LEVEL_CONTAINER
;
1938 sra
->array
.level
= LEVEL_CONTAINER
;
1941 pr_err("failed to read sysfs parameters for %s\n",
1945 frozen
= freeze(st
);
1947 /* freeze() already spewed the reason */
1950 } else if (frozen
< 0) {
1951 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1956 /* ========= set size =============== */
1958 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1959 unsigned long long orig_size
= get_component_size(fd
)/2;
1960 unsigned long long min_csize
;
1962 int raid0_takeover
= 0;
1965 orig_size
= (unsigned) array
.size
;
1967 if (orig_size
== 0) {
1968 pr_err("Cannot set device size in this type of array.\n");
1973 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1974 devname
, APPLY_METADATA_CHANGES
,
1980 if (st
->ss
->external
) {
1981 /* metadata can have size limitation
1982 * update size value according to metadata information
1984 struct mdinfo
*sizeinfo
=
1985 st
->ss
->container_content(st
, subarray
);
1987 unsigned long long new_size
=
1988 sizeinfo
->custom_array_size
/2;
1989 int data_disks
= get_data_disks(
1990 sizeinfo
->array
.level
,
1991 sizeinfo
->array
.layout
,
1992 sizeinfo
->array
.raid_disks
);
1993 new_size
/= data_disks
;
1994 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1995 orig_size
, new_size
,
1996 new_size
* data_disks
);
1998 sysfs_free(sizeinfo
);
2002 /* Update the size of each member device in case
2003 * they have been resized. This will never reduce
2004 * below the current used-size. The "size" attribute
2005 * understands '0' to mean 'max'.
2008 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2009 sysfs_set_num(sra
, mdi
, "size",
2010 s
->size
== MAX_SIZE
? 0 : s
->size
);
2011 if (array
.not_persistent
== 0 &&
2012 array
.major_version
== 0 &&
2013 get_linux_version() < 3001000) {
2014 /* Dangerous to allow size to exceed 2TB */
2015 unsigned long long csize
;
2016 if (sysfs_get_ll(sra
, mdi
, "size",
2018 if (csize
>= 2ULL*1024*1024*1024)
2019 csize
= 2ULL*1024*1024*1024;
2020 if ((min_csize
== 0 ||
2021 (min_csize
> csize
)))
2026 if (min_csize
&& s
->size
> min_csize
) {
2027 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2029 goto size_change_error
;
2031 if (min_csize
&& s
->size
== MAX_SIZE
) {
2032 /* Don't let the kernel choose a size - it will get
2035 pr_err("Limited v0.90 array to 2TB per device\n");
2036 s
->size
= min_csize
;
2038 if (st
->ss
->external
) {
2039 if (sra
->array
.level
== 0) {
2040 rv
= sysfs_set_str(sra
, NULL
, "level", "raid5");
2043 /* get array parameters after takeover
2044 * to change one parameter at time only
2046 rv
= md_get_array_info(fd
, &array
);
2049 /* make sure mdmon is
2050 * aware of the new level */
2051 if (!mdmon_running(st
->container_devnm
))
2052 start_mdmon(st
->container_devnm
);
2053 ping_monitor(container
);
2054 if (mdmon_running(st
->container_devnm
) &&
2055 st
->update_tail
== NULL
)
2056 st
->update_tail
= &st
->updates
;
2059 if (s
->size
== MAX_SIZE
)
2061 array
.size
= s
->size
;
2062 if (s
->size
& ~INT32_MAX
) {
2063 /* got truncated to 32bit, write to
2064 * component_size instead
2067 rv
= sysfs_set_num(sra
, NULL
,
2068 "component_size", s
->size
);
2072 rv
= md_set_array_info(fd
, &array
);
2074 /* manage array size when it is managed externally
2076 if ((rv
== 0) && st
->ss
->external
)
2077 rv
= set_array_size(st
, sra
, sra
->text_version
);
2080 if (raid0_takeover
) {
2081 /* do not recync non-existing parity,
2082 * we will drop it anyway
2084 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2085 /* go back to raid0, drop parity disk
2087 sysfs_set_str(sra
, NULL
, "level", "raid0");
2088 md_get_array_info(fd
, &array
);
2095 /* restore metadata */
2096 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2097 UnSet
, NULL
, devname
,
2098 ROLLBACK_METADATA_CHANGES
,
2101 pr_err("Cannot set device size for %s: %s\n",
2102 devname
, strerror(err
));
2104 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2105 cont_err("Bitmap must be removed before size can be changed\n");
2109 if (s
->assume_clean
) {
2110 /* This will fail on kernels older than 3.0 unless
2111 * a backport has been arranged.
2114 sysfs_set_str(sra
, NULL
, "resync_start",
2116 pr_err("--assume-clean not supported with --grow on this kernel\n");
2118 md_get_array_info(fd
, &array
);
2119 s
->size
= get_component_size(fd
)/2;
2121 s
->size
= array
.size
;
2122 if (c
->verbose
>= 0) {
2123 if (s
->size
== orig_size
)
2124 pr_err("component size of %s unchanged at %lluK\n",
2127 pr_err("component size of %s has been set to %lluK\n",
2131 } else if (array
.level
!= LEVEL_CONTAINER
) {
2132 s
->size
= get_component_size(fd
)/2;
2134 s
->size
= array
.size
;
2137 /* See if there is anything else to do */
2138 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2139 (s
->layout_str
== NULL
) &&
2140 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2141 data_offset
== INVALID_SECTORS
&&
2142 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2143 /* Nothing more to do */
2144 if (!changed
&& c
->verbose
>= 0)
2145 pr_err("%s: no change requested\n", devname
);
2149 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2150 * current implementation assumes that following conditions must be met:
2153 * - near_copies == 2
2155 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2156 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2157 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2160 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2162 dprintf("Array cannot be reshaped\n");
2168 /* Make sure mdmon has seen the device removal
2169 * and updated metadata before we continue with
2173 ping_monitor(container
);
2176 memset(&info
, 0, sizeof(info
));
2178 if (sysfs_init(&info
, fd
, NULL
)) {
2179 pr_err("failed to intialize sysfs.\n");
2183 strcpy(info
.text_version
, sra
->text_version
);
2184 info
.component_size
= s
->size
*2;
2185 info
.new_level
= s
->level
;
2186 info
.new_chunk
= s
->chunk
* 1024;
2187 if (info
.array
.level
== LEVEL_CONTAINER
) {
2188 info
.delta_disks
= UnSet
;
2189 info
.array
.raid_disks
= s
->raiddisks
;
2190 } else if (s
->raiddisks
)
2191 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2193 info
.delta_disks
= UnSet
;
2194 if (s
->layout_str
== NULL
) {
2195 info
.new_layout
= UnSet
;
2196 if (info
.array
.level
== 6 &&
2197 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2198 info
.array
.layout
>= 16) {
2199 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2200 cont_err("during the reshape, please specify --layout=preserve\n");
2201 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2205 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2206 strcmp(s
->layout_str
, "normalize") == 0) {
2207 /* If we have a -6 RAID6 layout, remove the '-6'. */
2208 info
.new_layout
= UnSet
;
2209 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2211 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
2212 h
= strrchr(l
, '-');
2213 if (h
&& strcmp(h
, "-6") == 0) {
2215 info
.new_layout
= map_name(r6layout
, l
);
2218 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2222 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2223 /* This means that a non-standard RAID6 layout
2226 * - When reshape a RAID6 (e.g. adding a device)
2227 * which is in a non-standard layout, it is OK
2228 * to preserve that layout.
2229 * - When converting a RAID5 to RAID6, leave it in
2230 * the XXX-6 layout, don't re-layout.
2232 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2233 info
.new_layout
= info
.array
.layout
;
2234 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2236 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
2238 info
.new_layout
= map_name(r6layout
, l
);
2240 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2245 int l
= info
.new_level
;
2247 l
= info
.array
.level
;
2250 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2253 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2256 info
.new_layout
= parse_layout_10(s
->layout_str
);
2259 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2262 pr_err("layout not meaningful with this level\n");
2266 if (info
.new_layout
== UnSet
) {
2267 pr_err("layout %s not understood for this level\n",
2274 if (array
.level
== LEVEL_FAULTY
) {
2275 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2276 pr_err("cannot change level of Faulty device\n");
2280 pr_err("cannot set chunksize of Faulty device\n");
2283 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2284 pr_err("cannot set raid_disks of Faulty device\n");
2287 if (s
->layout_str
) {
2288 if (md_get_array_info(fd
, &array
) != 0) {
2289 dprintf("Cannot get array information.\n");
2292 array
.layout
= info
.new_layout
;
2293 if (md_set_array_info(fd
, &array
) != 0) {
2294 pr_err("failed to set new layout\n");
2296 } else if (c
->verbose
>= 0)
2297 printf("layout for %s set to %d\n",
2298 devname
, array
.layout
);
2300 } else if (array
.level
== LEVEL_CONTAINER
) {
2301 /* This change is to be applied to every array in the
2302 * container. This is only needed when the metadata imposes
2303 * restraints of the various arrays in the container.
2304 * Currently we only know that IMSM requires all arrays
2305 * to have the same number of devices so changing the
2306 * number of devices (On-Line Capacity Expansion) must be
2307 * performed at the level of the container
2313 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2314 c
->force
, c
->backup_file
, c
->verbose
,
2318 /* get spare devices from external metadata
2320 if (st
->ss
->external
) {
2321 struct mdinfo
*info2
;
2323 info2
= st
->ss
->container_content(st
, subarray
);
2325 info
.array
.spare_disks
=
2326 info2
->array
.spare_disks
;
2331 /* Impose these changes on a single array. First
2332 * check that the metadata is OK with the change. */
2334 if (reshape_super(st
, 0, info
.new_level
,
2335 info
.new_layout
, info
.new_chunk
,
2336 info
.array
.raid_disks
, info
.delta_disks
,
2337 c
->backup_file
, devname
,
2338 APPLY_METADATA_CHANGES
, c
->verbose
)) {
2343 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2344 devlist
, data_offset
, c
->backup_file
,
2345 c
->verbose
, 0, 0, 0);
2355 /* verify_reshape_position()
2356 * Function checks if reshape position in metadata is not farther
2357 * than position in md.
2359 * 0 : not valid sysfs entry
2360 * it can be caused by not started reshape, it should be started
2361 * by reshape array or raid0 array is before takeover
2362 * -1 : error, reshape position is obviously wrong
2363 * 1 : success, reshape progress correct or updated
2365 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2371 /* read sync_max, failure can mean raid0 array */
2372 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2376 unsigned long long position
= strtoull(buf
, &ep
, 0);
2378 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2379 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2380 position
*= get_data_disks(level
,
2382 info
->array
.raid_disks
);
2383 if (info
->reshape_progress
< position
) {
2384 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2385 info
->reshape_progress
, position
);
2386 info
->reshape_progress
= position
;
2388 } else if (info
->reshape_progress
> position
) {
2389 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2390 position
, info
->reshape_progress
);
2393 dprintf("Reshape position in md and metadata are the same;");
2397 } else if (rv
== 0) {
2398 /* for valid sysfs entry, 0-length content
2399 * should be indicated as error
2407 static unsigned long long choose_offset(unsigned long long lo
,
2408 unsigned long long hi
,
2409 unsigned long long min
,
2410 unsigned long long max
)
2412 /* Choose a new offset between hi and lo.
2413 * It must be between min and max, but
2414 * we would prefer something near the middle of hi/lo, and also
2415 * prefer to be aligned to a big power of 2.
2417 * So we start with the middle, then for each bit,
2418 * starting at '1' and increasing, if it is set, we either
2419 * add it or subtract it if possible, preferring the option
2420 * which is furthest from the boundary.
2422 * We stop once we get a 1MB alignment. As units are in sectors,
2423 * 1MB = 2*1024 sectors.
2425 unsigned long long choice
= (lo
+ hi
) / 2;
2426 unsigned long long bit
= 1;
2428 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2429 unsigned long long bigger
, smaller
;
2430 if (! (bit
& choice
))
2432 bigger
= choice
+ bit
;
2433 smaller
= choice
- bit
;
2434 if (bigger
> max
&& smaller
< min
)
2438 else if (smaller
< min
)
2440 else if (hi
- bigger
> smaller
- lo
)
2448 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2449 char *devname
, int delta_disks
,
2450 unsigned long long data_offset
,
2451 unsigned long long min
,
2457 unsigned long long before
, after
;
2459 /* Need to find min space before and after so same is used
2462 before
= UINT64_MAX
;
2464 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2468 struct supertype
*st2
;
2469 struct mdinfo info2
;
2471 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2473 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2474 dfd
= dev_open(dn
, O_RDONLY
);
2476 pr_err("%s: cannot open component %s\n",
2477 devname
, dn
? dn
: "-unknown-");
2480 st2
= dup_super(st
);
2481 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2485 pr_err("%s: cannot get superblock from %s\n",
2489 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2490 st2
->ss
->free_super(st2
);
2492 if (info2
.space_before
== 0 &&
2493 info2
.space_after
== 0) {
2494 /* Metadata doesn't support data_offset changes */
2496 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2500 if (before
> info2
.space_before
)
2501 before
= info2
.space_before
;
2502 if (after
> info2
.space_after
)
2503 after
= info2
.space_after
;
2505 if (data_offset
!= INVALID_SECTORS
) {
2507 if (info2
.data_offset
== data_offset
) {
2508 pr_err("%s: already has that data_offset\n",
2512 if (data_offset
< info2
.data_offset
)
2516 } else if ((data_offset
<= info2
.data_offset
&&
2518 (data_offset
>= info2
.data_offset
&&
2520 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2526 if (before
== UINT64_MAX
)
2527 /* impossible really, there must be no devices */
2530 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2531 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2532 unsigned long long new_data_offset
;
2534 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2536 if (delta_disks
< 0) {
2537 /* Don't need any space as array is shrinking
2538 * just move data_offset up by min
2540 if (data_offset
== INVALID_SECTORS
)
2541 new_data_offset
= sd
->data_offset
+ min
;
2543 if (data_offset
< sd
->data_offset
+ min
) {
2544 pr_err("--data-offset too small for %s\n",
2548 new_data_offset
= data_offset
;
2550 } else if (delta_disks
> 0) {
2551 /* need space before */
2555 pr_err("Insufficient head-space for reshape on %s\n",
2559 if (data_offset
== INVALID_SECTORS
)
2560 new_data_offset
= sd
->data_offset
- min
;
2562 if (data_offset
> sd
->data_offset
- min
) {
2563 pr_err("--data-offset too large for %s\n",
2567 new_data_offset
= data_offset
;
2571 /* can move up or down. If 'data_offset'
2572 * was set we would have already decided,
2573 * so just choose direction with most space.
2580 sysfs_set_str(sra
, NULL
, "reshape_direction",
2581 dir
== 1 ? "backwards" : "forwards");
2583 /* Increase data offset */
2587 pr_err("Insufficient tail-space for reshape on %s\n",
2591 if (data_offset
!= INVALID_SECTORS
&&
2592 data_offset
< sd
->data_offset
+ min
) {
2593 pr_err("--data-offset too small on %s\n",
2597 if (data_offset
!= INVALID_SECTORS
)
2598 new_data_offset
= data_offset
;
2600 new_data_offset
= choose_offset(sd
->data_offset
,
2601 sd
->data_offset
+ after
,
2602 sd
->data_offset
+ min
,
2603 sd
->data_offset
+ after
);
2605 /* Decrease data offset */
2609 pr_err("insufficient head-room on %s\n",
2613 if (data_offset
!= INVALID_SECTORS
&&
2614 data_offset
< sd
->data_offset
- min
) {
2615 pr_err("--data-offset too small on %s\n",
2619 if (data_offset
!= INVALID_SECTORS
)
2620 new_data_offset
= data_offset
;
2622 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2624 sd
->data_offset
- before
,
2625 sd
->data_offset
- min
);
2628 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2629 if (err
< 0 && errno
== E2BIG
) {
2630 /* try again after increasing data size to max */
2631 err
= sysfs_set_num(sra
, sd
, "size", 0);
2632 if (err
< 0 && errno
== EINVAL
&&
2633 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2634 /* some kernels have a bug where you cannot
2635 * use '0' on spare devices. */
2636 sysfs_set_num(sra
, sd
, "size",
2637 (sra
->component_size
+ after
)/2);
2639 err
= sysfs_set_num(sra
, sd
, "new_offset",
2643 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2644 pr_err("data-offset is too big for %s\n", dn
);
2647 if (sd
== sra
->devs
&&
2648 (errno
== ENOENT
|| errno
== E2BIG
))
2649 /* Early kernel, no 'new_offset' file,
2650 * or kernel doesn't like us.
2651 * For RAID5/6 this is not fatal
2654 pr_err("Cannot set new_offset for %s\n", dn
);
2662 /* Just use a backup file */
2666 static int raid10_reshape(char *container
, int fd
, char *devname
,
2667 struct supertype
*st
, struct mdinfo
*info
,
2668 struct reshape
*reshape
,
2669 unsigned long long data_offset
,
2670 int force
, int verbose
)
2672 /* Changing raid_disks, layout, chunksize or possibly
2673 * just data_offset for a RAID10.
2674 * We must always change data_offset. We change by at least
2675 * ->min_offset_change which is the largest of the old and new
2677 * If raid_disks is increasing, then data_offset must decrease
2678 * by at least this copy size.
2679 * If raid_disks is unchanged, data_offset must increase or
2680 * decrease by at least min_offset_change but preferably by much more.
2681 * We choose half of the available space.
2682 * If raid_disks is decreasing, data_offset must increase by
2683 * at least min_offset_change. To allow of this, component_size
2684 * must be decreased by the same amount.
2686 * So we calculate the required minimum and direction, possibly
2687 * reduce the component_size, then iterate through the devices
2688 * and set the new_data_offset.
2689 * If that all works, we set chunk_size, layout, raid_disks, and start
2693 unsigned long long min
;
2696 sra
= sysfs_read(fd
, NULL
,
2697 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2700 pr_err("%s: Cannot get array details from sysfs\n", devname
);
2703 min
= reshape
->min_offset_change
;
2705 if (info
->delta_disks
)
2706 sysfs_set_str(sra
, NULL
, "reshape_direction",
2707 info
->delta_disks
< 0 ? "backwards" : "forwards");
2708 if (info
->delta_disks
< 0 && info
->space_after
< min
) {
2709 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2710 (sra
->component_size
- min
)/2);
2712 pr_err("cannot reduce component size\n");
2716 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
,
2717 data_offset
, min
, 0);
2719 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2720 cont_err("supported on this kernel\n");
2726 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2728 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2729 reshape
->after
.layout
) < 0)
2732 sysfs_set_num(sra
, NULL
, "raid_disks",
2733 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2735 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2738 pr_err("Cannot set array shape for %s\n",
2741 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2742 cont_err(" Bitmap must be removed before shape can be changed\n");
2752 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2754 struct mdinfo
*sra
, *sd
;
2755 /* Initialisation to silence compiler warning */
2756 unsigned long long min_space_before
= 0, min_space_after
= 0;
2759 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2762 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2765 struct supertype
*st2
;
2766 struct mdinfo info2
;
2768 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2770 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2771 dfd
= dev_open(dn
, O_RDONLY
);
2774 st2
= dup_super(st
);
2775 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2781 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2782 st2
->ss
->free_super(st2
);
2785 min_space_before
> info2
.space_before
)
2786 min_space_before
= info2
.space_before
;
2788 min_space_after
> info2
.space_after
)
2789 min_space_after
= info2
.space_after
;
2792 if (sd
== NULL
&& !first
) {
2793 info
->space_after
= min_space_after
;
2794 info
->space_before
= min_space_before
;
2799 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2800 struct mdinfo
*info
,
2801 int disks
, unsigned long long blocks
)
2803 /* Check that the internal stripe cache is
2804 * large enough, or it won't work.
2805 * It must hold at least 4 stripes of the larger
2808 unsigned long cache
;
2809 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2810 cache
*= 4; /* 4 stripes minimum */
2811 cache
/= 512; /* convert to sectors */
2812 /* make sure there is room for 'blocks' with a bit to spare */
2813 if (cache
< 16 + blocks
/ disks
)
2814 cache
= 16 + blocks
/ disks
;
2815 cache
/= (4096/512); /* Convert from sectors to pages */
2817 if (sra
->cache_size
< cache
)
2818 subarray_set_num(container
, sra
, "stripe_cache_size",
2822 static int impose_reshape(struct mdinfo
*sra
,
2823 struct mdinfo
*info
,
2824 struct supertype
*st
,
2827 char *devname
, char *container
,
2828 struct reshape
*reshape
)
2830 struct mdu_array_info_s array
;
2832 sra
->new_chunk
= info
->new_chunk
;
2835 /* for external metadata checkpoint saved by mdmon can be lost
2836 * or missed /due to e.g. crash/. Check if md is not during
2837 * restart farther than metadata points to.
2838 * If so, this means metadata information is obsolete.
2840 if (st
->ss
->external
)
2841 verify_reshape_position(info
, reshape
->level
);
2842 sra
->reshape_progress
= info
->reshape_progress
;
2844 sra
->reshape_progress
= 0;
2845 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2846 /* start from the end of the new array */
2847 sra
->reshape_progress
= (sra
->component_size
2848 * reshape
->after
.data_disks
);
2851 md_get_array_info(fd
, &array
);
2852 if (info
->array
.chunk_size
== info
->new_chunk
&&
2853 reshape
->before
.layout
== reshape
->after
.layout
&&
2854 st
->ss
->external
== 0) {
2855 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2856 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2857 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2860 pr_err("Cannot set device shape for %s: %s\n",
2861 devname
, strerror(errno
));
2864 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2865 cont_err("Bitmap must be removed before shape can be changed\n");
2869 } else if (!restart
) {
2870 /* set them all just in case some old 'new_*' value
2871 * persists from some earlier problem.
2874 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2876 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2877 reshape
->after
.layout
) < 0)
2879 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2880 reshape
->after
.data_disks
+
2881 reshape
->parity
) < 0)
2884 pr_err("Cannot set device shape for %s\n", devname
);
2887 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2888 cont_err("Bitmap must be removed before shape can be changed\n");
2897 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2900 struct mdu_array_info_s array
;
2903 if (sysfs_init(&info
, fd
, NULL
)) {
2904 pr_err("failed to intialize sysfs.\n");
2908 md_get_array_info(fd
, &array
);
2909 if (level
== 0 && (array
.level
>= 4 && array
.level
<= 6)) {
2910 /* To convert to RAID0 we need to fail and
2911 * remove any non-data devices. */
2914 int data_disks
= array
.raid_disks
- 1;
2915 if (array
.level
== 6)
2917 if (array
.level
== 5 && array
.layout
!= ALGORITHM_PARITY_N
)
2919 if (array
.level
== 6 && array
.layout
!= ALGORITHM_PARITY_N_6
)
2921 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2922 /* First remove any spares so no recovery starts */
2923 for (d
= 0, found
= 0;
2924 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2925 mdu_disk_info_t disk
;
2927 if (md_get_disk_info(fd
, &disk
) < 0)
2929 if (disk
.major
== 0 && disk
.minor
== 0)
2932 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2933 disk
.raid_disk
< data_disks
)
2936 ioctl(fd
, HOT_REMOVE_DISK
,
2937 makedev(disk
.major
, disk
.minor
));
2939 /* Now fail anything left */
2940 md_get_array_info(fd
, &array
);
2941 for (d
= 0, found
= 0;
2942 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2943 mdu_disk_info_t disk
;
2945 if (md_get_disk_info(fd
, &disk
) < 0)
2947 if (disk
.major
== 0 && disk
.minor
== 0)
2950 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2951 disk
.raid_disk
< data_disks
)
2954 ioctl(fd
, SET_DISK_FAULTY
,
2955 makedev(disk
.major
, disk
.minor
));
2956 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2959 c
= map_num(pers
, level
);
2961 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2964 pr_err("%s: could not set level to %s\n",
2967 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2968 cont_err("Bitmap must be removed before level can be changed\n");
2972 pr_err("level of %s changed to %s\n", devname
, c
);
2978 static void catch_term(int sig
)
2983 static int continue_via_systemd(char *devnm
)
2985 int skipped
, i
, pid
, status
;
2987 /* In a systemd/udev world, it is best to get systemd to
2988 * run "mdadm --grow --continue" rather than running in the
2993 /* FIXME yuk. CLOSE_EXEC?? */
2995 for (i
= 3; skipped
< 20; i
++)
3001 /* Don't want to see error messages from
3002 * systemctl. If the service doesn't exist,
3003 * we fork ourselves.
3006 open("/dev/null", O_WRONLY
);
3007 snprintf(pathbuf
, sizeof(pathbuf
),
3008 "mdadm-grow-continue@%s.service", devnm
);
3009 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
3011 status
= execl("/bin/systemctl", "systemctl", "restart",
3014 case -1: /* Just do it ourselves. */
3016 default: /* parent - good */
3017 pid
= wait(&status
);
3018 if (pid
>= 0 && status
== 0)
3024 static int reshape_array(char *container
, int fd
, char *devname
,
3025 struct supertype
*st
, struct mdinfo
*info
,
3026 int force
, struct mddev_dev
*devlist
,
3027 unsigned long long data_offset
,
3028 char *backup_file
, int verbose
, int forked
,
3029 int restart
, int freeze_reshape
)
3031 struct reshape reshape
;
3034 int orig_level
= UnSet
;
3038 struct mdu_array_info_s array
;
3041 struct mddev_dev
*dv
;
3045 unsigned long long *offsets
= NULL
;
3049 unsigned long blocks
;
3050 unsigned long long array_size
;
3052 struct mdinfo
*sra
= NULL
;
3055 /* when reshaping a RAID0, the component_size might be zero.
3056 * So try to fix that up.
3058 if (md_get_array_info(fd
, &array
) != 0) {
3059 dprintf("Cannot get array information.\n");
3062 if (array
.level
== 0 && info
->component_size
== 0) {
3063 get_dev_size(fd
, NULL
, &array_size
);
3064 info
->component_size
= array_size
/ array
.raid_disks
;
3067 if (array
.level
== 10)
3068 /* Need space_after info */
3069 get_space_after(fd
, st
, info
);
3071 if (info
->reshape_active
) {
3072 int new_level
= info
->new_level
;
3073 info
->new_level
= UnSet
;
3074 if (info
->delta_disks
> 0)
3075 info
->array
.raid_disks
-= info
->delta_disks
;
3076 msg
= analyse_change(devname
, info
, &reshape
);
3077 info
->new_level
= new_level
;
3078 if (info
->delta_disks
> 0)
3079 info
->array
.raid_disks
+= info
->delta_disks
;
3081 /* Make sure the array isn't read-only */
3082 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3084 msg
= analyse_change(devname
, info
, &reshape
);
3086 /* if msg == "", error has already been printed */
3088 pr_err("%s\n", msg
);
3091 if (restart
&& (reshape
.level
!= info
->array
.level
||
3092 reshape
.before
.layout
!= info
->array
.layout
||
3093 reshape
.before
.data_disks
+ reshape
.parity
!=
3094 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3095 pr_err("reshape info is not in native format - cannot continue.\n");
3099 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3100 !((sysfs_get_str(info
, NULL
, "sync_action",
3101 buf
, sizeof(buf
)) > 0) &&
3102 (strncmp(buf
, "reshape", 7) == 0))) {
3103 /* When reshape is restarted from '0', very begin of array
3104 * it is possible that for external metadata reshape and array
3105 * configuration doesn't happen.
3106 * Check if md has the same opinion, and reshape is restarted
3107 * from 0. If so, this is regular reshape start after reshape
3108 * switch in metadata to next array only.
3110 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3111 (info
->reshape_progress
== 0))
3116 * reshape already started. just skip to monitoring
3119 if (reshape
.backup_blocks
== 0)
3121 if (restart
& RESHAPE_NO_BACKUP
)
3124 /* Need 'sra' down at 'started:' */
3125 sra
= sysfs_read(fd
, NULL
,
3126 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
3127 GET_CHUNK
|GET_CACHE
);
3129 pr_err("%s: Cannot get array details from sysfs\n",
3135 backup_file
= locate_backup(sra
->sys_name
);
3139 /* The container is frozen but the array may not be.
3140 * So freeze the array so spares don't get put to the wrong use
3141 * FIXME there should probably be a cleaner separation between
3142 * freeze_array and freeze_container.
3144 sysfs_freeze_array(info
);
3145 /* Check we have enough spares to not be degraded */
3147 for (dv
= devlist
; dv
; dv
=dv
->next
)
3149 spares_needed
= max(reshape
.before
.data_disks
,
3150 reshape
.after
.data_disks
) +
3151 reshape
.parity
- array
.raid_disks
;
3153 if (!force
&& info
->new_level
> 1 && info
->array
.level
> 1 &&
3154 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3155 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3156 " Use --force to over-ride this check.\n",
3158 spares_needed
== 1 ? "" : "s",
3159 info
->array
.spare_disks
+ added_disks
);
3162 /* Check we have enough spares to not fail */
3163 spares_needed
= max(reshape
.before
.data_disks
,
3164 reshape
.after
.data_disks
)
3166 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3167 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3168 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3169 spares_needed
, spares_needed
== 1 ? "" : "s",
3170 info
->array
.spare_disks
+ added_disks
);
3174 if (reshape
.level
!= array
.level
) {
3175 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3178 info
->new_layout
= UnSet
; /* after level change,
3179 * layout is meaningless */
3180 orig_level
= array
.level
;
3181 sysfs_freeze_array(info
);
3183 if (reshape
.level
> 0 && st
->ss
->external
) {
3184 /* make sure mdmon is aware of the new level */
3185 if (mdmon_running(container
))
3186 flush_mdmon(container
);
3188 if (!mdmon_running(container
))
3189 start_mdmon(container
);
3190 ping_monitor(container
);
3191 if (mdmon_running(container
) && st
->update_tail
== NULL
)
3192 st
->update_tail
= &st
->updates
;
3195 /* ->reshape_super might have chosen some spares from the
3196 * container that it wants to be part of the new array.
3197 * We can collect them with ->container_content and give
3198 * them to the kernel.
3200 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3201 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3202 struct mdinfo
*info2
=
3203 st
->ss
->container_content(st
, subarray
);
3207 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3208 pr_err("unable to initialize sysfs for %s\n",
3213 /* When increasing number of devices, we need to set
3214 * new raid_disks before adding these, or they might
3217 if (reshape
.backup_blocks
&&
3218 reshape
.after
.data_disks
>
3219 reshape
.before
.data_disks
)
3220 subarray_set_num(container
, info2
, "raid_disks",
3221 reshape
.after
.data_disks
+
3223 for (d
= info2
->devs
; d
; d
= d
->next
) {
3224 if (d
->disk
.state
== 0 &&
3225 d
->disk
.raid_disk
>= 0) {
3226 /* This is a spare that wants to
3227 * be part of the array.
3229 add_disk(fd
, st
, info2
, d
);
3235 /* We might have been given some devices to add to the
3236 * array. Now that the array has been changed to the right
3237 * level and frozen, we can safely add them.
3240 if (Manage_subdevs(devname
, fd
, devlist
, verbose
, 0, NULL
, 0))
3244 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3245 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3246 if (reshape
.backup_blocks
== 0) {
3247 /* No restriping needed, but we might need to impose
3248 * some more changes: layout, raid_disks, chunk_size
3250 /* read current array info */
3251 if (md_get_array_info(fd
, &array
) != 0) {
3252 dprintf("Cannot get array information.\n");
3255 /* compare current array info with new values and if
3256 * it is different update them to new */
3257 if (info
->new_layout
!= UnSet
&&
3258 info
->new_layout
!= array
.layout
) {
3259 array
.layout
= info
->new_layout
;
3260 if (md_set_array_info(fd
, &array
) != 0) {
3261 pr_err("failed to set new layout\n");
3263 } else if (verbose
>= 0)
3264 printf("layout for %s set to %d\n",
3265 devname
, array
.layout
);
3267 if (info
->delta_disks
!= UnSet
&& info
->delta_disks
!= 0 &&
3269 (info
->array
.raid_disks
+ info
->delta_disks
)) {
3270 array
.raid_disks
+= info
->delta_disks
;
3271 if (md_set_array_info(fd
, &array
) != 0) {
3272 pr_err("failed to set raid disks\n");
3274 } else if (verbose
>= 0) {
3275 printf("raid_disks for %s set to %d\n",
3276 devname
, array
.raid_disks
);
3279 if (info
->new_chunk
!= 0 &&
3280 info
->new_chunk
!= array
.chunk_size
) {
3281 if (sysfs_set_num(info
, NULL
,
3282 "chunk_size", info
->new_chunk
) != 0) {
3283 pr_err("failed to set chunk size\n");
3285 } else if (verbose
>= 0)
3286 printf("chunk size for %s set to %d\n",
3287 devname
, array
.chunk_size
);
3294 * There are three possibilities.
3295 * 1/ The array will shrink.
3296 * We need to ensure the reshape will pause before reaching
3297 * the 'critical section'. We also need to fork and wait for
3298 * that to happen. When it does we
3299 * suspend/backup/complete/unfreeze
3301 * 2/ The array will not change size.
3302 * This requires that we keep a backup of a sliding window
3303 * so that we can restore data after a crash. So we need
3304 * to fork and monitor progress.
3305 * In future we will allow the data_offset to change, so
3306 * a sliding backup becomes unnecessary.
3308 * 3/ The array will grow. This is relatively easy.
3309 * However the kernel's restripe routines will cheerfully
3310 * overwrite some early data before it is safe. So we
3311 * need to make a backup of the early parts of the array
3312 * and be ready to restore it if rebuild aborts very early.
3313 * For externally managed metadata, we still need a forked
3314 * child to monitor the reshape and suspend IO over the region
3315 * that is being reshaped.
3317 * We backup data by writing it to one spare, or to a
3318 * file which was given on command line.
3320 * In each case, we first make sure that storage is available
3321 * for the required backup.
3323 * - request the shape change.
3324 * - fork to handle backup etc.
3326 /* Check that we can hold all the data */
3327 get_dev_size(fd
, NULL
, &array_size
);
3328 if (reshape
.new_size
< (array_size
/512)) {
3329 pr_err("this change will reduce the size of the array.\n"
3330 " use --grow --array-size first to truncate array.\n"
3331 " e.g. mdadm --grow %s --array-size %llu\n",
3332 devname
, reshape
.new_size
/2);
3336 if (array
.level
== 10) {
3337 /* Reshaping RAID10 does not require any data backup by
3338 * user-space. Instead it requires that the data_offset
3339 * is changed to avoid the need for backup.
3340 * So this is handled very separately
3343 /* Nothing to do. */
3345 return raid10_reshape(container
, fd
, devname
, st
, info
,
3346 &reshape
, data_offset
, force
, verbose
);
3348 sra
= sysfs_read(fd
, NULL
,
3349 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3352 pr_err("%s: Cannot get array details from sysfs\n",
3358 switch(set_new_data_offset(sra
, st
, devname
,
3359 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3361 reshape
.min_offset_change
, 1)) {
3365 /* Updated data_offset, so it's easy now */
3366 update_cache_size(container
, sra
, info
,
3367 min(reshape
.before
.data_disks
,
3368 reshape
.after
.data_disks
),
3369 reshape
.backup_blocks
);
3371 /* Right, everything seems fine. Let's kick things off.
3375 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3376 devname
, container
, &reshape
) < 0)
3378 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3380 if (errno
!= EINVAL
) {
3381 pr_err("Failed to initiate reshape!\n");
3384 /* revert data_offset and try the old way */
3385 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3386 sysfs_set_num(sra
, sd
, "new_offset",
3388 sysfs_set_str(sra
, NULL
, "reshape_direction",
3393 if (info
->new_level
== reshape
.level
)
3395 /* need to adjust level when reshape completes */
3397 case -1: /* ignore error, but don't wait */
3399 default: /* parent */
3407 fd
= open_dev(sra
->sys_name
);
3409 impose_level(fd
, info
->new_level
, devname
, verbose
);
3411 case 1: /* Couldn't set data_offset, try the old way */
3412 if (data_offset
!= INVALID_SECTORS
) {
3413 pr_err("Cannot update data_offset on this array\n");
3420 /* Decide how many blocks (sectors) for a reshape
3421 * unit. The number we have so far is just a minimum
3423 blocks
= reshape
.backup_blocks
;
3424 if (reshape
.before
.data_disks
==
3425 reshape
.after
.data_disks
) {
3426 /* Make 'blocks' bigger for better throughput, but
3427 * not so big that we reject it below.
3428 * Try for 16 megabytes
3430 while (blocks
* 32 < sra
->component_size
&& blocks
< 16*1024*2)
3433 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3435 if (blocks
>= sra
->component_size
/2) {
3436 pr_err("%s: Something wrong - reshape aborted\n", devname
);
3440 /* Now we need to open all these devices so we can read/write.
3442 nrdisks
= max(reshape
.before
.data_disks
,
3443 reshape
.after
.data_disks
) + reshape
.parity
3444 + sra
->array
.spare_disks
;
3445 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3446 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3448 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3449 d
= reshape_prepare_fdlist(devname
, sra
, odisks
, nrdisks
, blocks
,
3450 backup_file
, fdlist
, offsets
);
3454 if ((st
->ss
->manage_reshape
== NULL
) ||
3455 (st
->ss
->recover_backup
== NULL
)) {
3456 if (backup_file
== NULL
) {
3457 if (reshape
.after
.data_disks
<=
3458 reshape
.before
.data_disks
) {
3459 pr_err("%s: Cannot grow - need backup-file\n",
3461 pr_err(" Please provide one with \"--backup=...\"\n");
3463 } else if (d
== odisks
) {
3464 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3468 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3470 fdlist
+d
, offsets
+d
,
3471 sra
->sys_name
, restart
)) {
3478 update_cache_size(container
, sra
, info
,
3479 min(reshape
.before
.data_disks
,
3480 reshape
.after
.data_disks
), blocks
);
3482 /* Right, everything seems fine. Let's kick things off.
3483 * If only changing raid_disks, use ioctl, else use
3488 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3489 devname
, container
, &reshape
) < 0)
3492 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3493 reshape
.after
.data_disks
);
3495 pr_err("Cannot %s reshape for %s\n",
3496 restart
? "continue" : "start", devname
);
3500 sysfs_set_str(sra
, NULL
, "array_state", "active");
3501 if (freeze_reshape
) {
3505 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3506 sra
->reshape_progress
);
3510 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3511 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3518 /* Now we just need to kick off the reshape and watch, while
3519 * handling backups of the data...
3520 * This is all done by a forked background process.
3522 switch(forked
? 0 : fork()) {
3524 pr_err("Cannot run child to monitor reshape: %s\n",
3538 /* If another array on the same devices is busy, the
3539 * reshape will wait for them. This would mean that
3540 * the first section that we suspend will stay suspended
3541 * for a long time. So check on that possibility
3542 * by looking for "DELAYED" in /proc/mdstat, and if found,
3546 struct mdstat_ent
*mds
, *m
;
3548 mds
= mdstat_read(1, 0);
3549 for (m
= mds
; m
; m
= m
->next
)
3550 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3551 if (m
->resync
&& m
->percent
== RESYNC_DELAYED
)
3554 /* Haven't started the reshape thread
3561 if (delayed
== 1 && get_linux_version() < 3007000) {
3562 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3563 " You might experience problems until other reshapes complete.\n");
3567 mdstat_wait(30 - (delayed
-1) * 25);
3571 if (check_env("MDADM_GROW_VERIFY"))
3572 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3575 mlockall(MCL_FUTURE
);
3577 signal(SIGTERM
, catch_term
);
3579 if (st
->ss
->external
) {
3580 /* metadata handler takes it from here */
3581 done
= st
->ss
->manage_reshape(
3582 fd
, sra
, &reshape
, st
, blocks
,
3583 fdlist
, offsets
, d
- odisks
, fdlist
+ odisks
,
3586 done
= child_monitor(
3587 fd
, sra
, &reshape
, st
, blocks
, fdlist
, offsets
,
3588 d
- odisks
, fdlist
+ odisks
, offsets
+ odisks
);
3593 if (backup_file
&& done
) {
3595 bul
= make_backup(sra
->sys_name
);
3598 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3606 unlink(backup_file
);
3613 if (!st
->ss
->external
&&
3614 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3615 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3617 /* no need to wait for the reshape to finish as
3618 * there is nothing more to do.
3625 if (st
->ss
->external
) {
3626 /* Re-load the metadata as much could have changed */
3627 int cfd
= open_dev(st
->container_devnm
);
3629 flush_mdmon(container
);
3630 st
->ss
->free_super(st
);
3631 st
->ss
->load_container(st
, cfd
, container
);
3636 /* set new array size if required customer_array_size is used
3639 if (reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3640 info
->custom_array_size
)
3641 set_array_size(st
, info
, info
->text_version
);
3643 if (info
->new_level
!= reshape
.level
) {
3645 fd
= open(devname
, O_RDONLY
);
3646 impose_level(fd
, info
->new_level
, devname
, verbose
);
3648 if (info
->new_level
== 0)
3649 st
->update_tail
= NULL
;
3661 if (orig_level
!= UnSet
&& sra
) {
3662 c
= map_num(pers
, orig_level
);
3663 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3664 pr_err("aborting level change\n");
3672 /* mdfd handle is passed to be closed in child process (after fork).
3674 int reshape_container(char *container
, char *devname
,
3676 struct supertype
*st
,
3677 struct mdinfo
*info
,
3679 char *backup_file
, int verbose
,
3680 int forked
, int restart
, int freeze_reshape
)
3682 struct mdinfo
*cc
= NULL
;
3684 char last_devnm
[32] = "";
3686 /* component_size is not meaningful for a container,
3687 * so pass '0' meaning 'no change'
3690 reshape_super(st
, 0, info
->new_level
,
3691 info
->new_layout
, info
->new_chunk
,
3692 info
->array
.raid_disks
, info
->delta_disks
,
3693 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3701 /* ping monitor to be sure that update is on disk
3703 ping_monitor(container
);
3705 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3706 if (continue_via_systemd(container
))
3709 switch (forked
? 0 : fork()) {
3710 case -1: /* error */
3711 perror("Cannot fork to complete reshape\n");
3714 default: /* parent */
3715 if (!freeze_reshape
)
3716 printf("%s: multi-array reshape continues in background\n", Name
);
3723 /* close unused handle in child process
3729 /* For each member array with reshape_active,
3730 * we need to perform the reshape.
3731 * We pick the first array that needs reshaping and
3732 * reshape it. reshape_array() will re-read the metadata
3733 * so the next time through a different array should be
3734 * ready for reshape.
3735 * It is possible that the 'different' array will not
3736 * be assembled yet. In that case we simple exit.
3737 * When it is assembled, the mdadm which assembles it
3738 * will take over the reshape.
3740 struct mdinfo
*content
;
3742 struct mdstat_ent
*mdstat
;
3748 cc
= st
->ss
->container_content(st
, NULL
);
3750 for (content
= cc
; content
; content
= content
->next
) {
3752 if (!content
->reshape_active
)
3755 subarray
= strchr(content
->text_version
+1, '/')+1;
3756 mdstat
= mdstat_by_subdev(subarray
, container
);
3759 if (mdstat
->active
== 0) {
3760 pr_err("Skipping inactive array %s.\n",
3762 free_mdstat(mdstat
);
3771 devid
= devnm2devid(mdstat
->devnm
);
3772 adev
= map_dev(major(devid
), minor(devid
), 0);
3774 adev
= content
->text_version
;
3776 fd
= open_dev(mdstat
->devnm
);
3778 pr_err("Device %s cannot be opened for reshape.\n",
3783 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3784 /* Do not allow for multiple reshape_array() calls for
3786 * It can happen when reshape_array() returns without
3787 * error, when reshape is not finished (wrong reshape
3788 * starting/continuation conditions). Mdmon doesn't
3789 * switch to next array in container and reentry
3790 * conditions for the same array occur.
3791 * This is possibly interim until the behaviour of
3792 * reshape_array is resolved().
3794 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3798 strcpy(last_devnm
, mdstat
->devnm
);
3800 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3801 pr_err("Unable to initialize sysfs for %s\n",
3807 if (mdmon_running(container
))
3808 flush_mdmon(container
);
3810 rv
= reshape_array(container
, fd
, adev
, st
,
3811 content
, force
, NULL
, INVALID_SECTORS
,
3812 backup_file
, verbose
, 1, restart
,
3816 if (freeze_reshape
) {
3825 if (mdmon_running(container
))
3826 flush_mdmon(container
);
3835 * We run a child process in the background which performs the following
3837 * - wait for resync to reach a certain point
3838 * - suspend io to the following section
3839 * - backup that section
3840 * - allow resync to proceed further
3842 * - discard the backup.
3844 * When are combined in slightly different ways in the three cases.
3846 * - suspend/backup/allow/wait/resume/discard
3848 * - allow/wait/suspend/backup/allow/wait/resume/discard
3850 * - wait/resume/discard/suspend/backup/allow
3852 * suspend/backup/allow always come together
3853 * wait/resume/discard do too.
3854 * For the same-size case we have two backups to improve flow.
3858 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3859 unsigned long long backup_point
,
3860 unsigned long long wait_point
,
3861 unsigned long long *suspend_point
,
3862 unsigned long long *reshape_completed
, int *frozen
)
3864 /* This function is called repeatedly by the reshape manager.
3865 * It determines how much progress can safely be made and allows
3867 * - 'info' identifies the array and particularly records in
3868 * ->reshape_progress the metadata's knowledge of progress
3869 * This is a sector offset from the start of the array
3870 * of the next array block to be relocated. This number
3871 * may increase from 0 or decrease from array_size, depending
3872 * on the type of reshape that is happening.
3873 * Note that in contrast, 'sync_completed' is a block count of the
3874 * reshape so far. It gives the distance between the start point
3875 * (head or tail of device) and the next place that data will be
3876 * written. It always increases.
3877 * - 'reshape' is the structure created by analyse_change
3878 * - 'backup_point' shows how much the metadata manager has backed-up
3879 * data. For reshapes with increasing progress, it is the next address
3880 * to be backed up, previous addresses have been backed-up. For
3881 * decreasing progress, it is the earliest address that has been
3882 * backed up - later address are also backed up.
3883 * So addresses between reshape_progress and backup_point are
3884 * backed up providing those are in the 'correct' order.
3885 * - 'wait_point' is an array address. When reshape_completed
3886 * passes this point, progress_reshape should return. It might
3887 * return earlier if it determines that ->reshape_progress needs
3888 * to be updated or further backup is needed.
3889 * - suspend_point is maintained by progress_reshape and the caller
3890 * should not touch it except to initialise to zero.
3891 * It is an array address and it only increases in 2.6.37 and earlier.
3892 * This makes it difficult to handle reducing reshapes with
3893 * external metadata.
3894 * However: it is similar to backup_point in that it records the
3895 * other end of a suspended region from reshape_progress.
3896 * it is moved to extend the region that is safe to backup and/or
3898 * - reshape_completed is read from sysfs and returned. The caller
3899 * should copy this into ->reshape_progress when it has reason to
3900 * believe that the metadata knows this, and any backup outside this
3904 * 1 if more data from backup_point - but only as far as suspend_point,
3905 * should be backed up
3906 * 0 if things are progressing smoothly
3907 * -1 if the reshape is finished because it is all done,
3908 * -2 if the reshape is finished due to an error.
3911 int advancing
= (reshape
->after
.data_disks
3912 >= reshape
->before
.data_disks
);
3913 unsigned long long need_backup
; /* All data between start of array and
3914 * here will at some point need to
3917 unsigned long long read_offset
, write_offset
;
3918 unsigned long long write_range
;
3919 unsigned long long max_progress
, target
, completed
;
3920 unsigned long long array_size
= (info
->component_size
3921 * reshape
->before
.data_disks
);
3925 /* First, we unsuspend any region that is now known to be safe.
3926 * If suspend_point is on the 'wrong' side of reshape_progress, then
3927 * we don't have or need suspension at the moment. This is true for
3928 * native metadata when we don't need to back-up.
3931 if (info
->reshape_progress
<= *suspend_point
)
3932 sysfs_set_num(info
, NULL
, "suspend_lo",
3933 info
->reshape_progress
);
3935 /* Note: this won't work in 2.6.37 and before.
3936 * Something somewhere should make sure we don't need it!
3938 if (info
->reshape_progress
>= *suspend_point
)
3939 sysfs_set_num(info
, NULL
, "suspend_hi",
3940 info
->reshape_progress
);
3943 /* Now work out how far it is safe to progress.
3944 * If the read_offset for ->reshape_progress is less than
3945 * 'blocks' beyond the write_offset, we can only progress as far
3947 * Otherwise we can progress until the write_offset for the new location
3948 * reaches (within 'blocks' of) the read_offset at the current location.
3949 * However that region must be suspended unless we are using native
3951 * If we need to suspend more, we limit it to 128M per device, which is
3952 * rather arbitrary and should be some time-based calculation.
3954 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3955 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3956 write_range
= info
->new_chunk
/512;
3957 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3958 need_backup
= array_size
;
3960 need_backup
= reshape
->backup_blocks
;
3962 if (read_offset
< write_offset
+ write_range
)
3963 max_progress
= backup_point
;
3966 read_offset
* reshape
->after
.data_disks
;
3968 if (read_offset
> write_offset
- write_range
)
3969 /* Can only progress as far as has been backed up,
3970 * which must be suspended */
3971 max_progress
= backup_point
;
3972 else if (info
->reshape_progress
<= need_backup
)
3973 max_progress
= backup_point
;
3975 if (info
->array
.major_version
>= 0)
3976 /* Can progress until backup is needed */
3977 max_progress
= need_backup
;
3979 /* Can progress until metadata update is required */
3981 read_offset
* reshape
->after
.data_disks
;
3982 /* but data must be suspended */
3983 if (max_progress
< *suspend_point
)
3984 max_progress
= *suspend_point
;
3989 /* We know it is safe to progress to 'max_progress' providing
3990 * it is suspended or we are using native metadata.
3991 * Consider extending suspend_point 128M per device if it
3992 * is less than 64M per device beyond reshape_progress.
3993 * But always do a multiple of 'blocks'
3994 * FIXME this is too big - it takes to long to complete
3997 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3998 reshape
->after
.data_disks
);
3999 target
/= reshape
->backup_blocks
;
4002 target
*= reshape
->backup_blocks
;
4004 /* For externally managed metadata we always need to suspend IO to
4005 * the area being reshaped so we regularly push suspend_point forward.
4006 * For native metadata we only need the suspend if we are going to do
4010 if ((need_backup
> info
->reshape_progress
||
4011 info
->array
.major_version
< 0) &&
4012 *suspend_point
< info
->reshape_progress
+ target
) {
4013 if (need_backup
< *suspend_point
+ 2 * target
)
4014 *suspend_point
= need_backup
;
4015 else if (*suspend_point
+ 2 * target
< array_size
)
4016 *suspend_point
+= 2 * target
;
4018 *suspend_point
= array_size
;
4019 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
4020 if (max_progress
> *suspend_point
)
4021 max_progress
= *suspend_point
;
4024 if (info
->array
.major_version
>= 0) {
4025 /* Only need to suspend when about to backup */
4026 if (info
->reshape_progress
< need_backup
* 2 &&
4027 *suspend_point
> 0) {
4029 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4030 sysfs_set_num(info
, NULL
, "suspend_hi",
4034 /* Need to suspend continually */
4035 if (info
->reshape_progress
< *suspend_point
)
4036 *suspend_point
= info
->reshape_progress
;
4037 if (*suspend_point
+ target
< info
->reshape_progress
)
4038 /* No need to move suspend region yet */;
4040 if (*suspend_point
>= 2 * target
)
4041 *suspend_point
-= 2 * target
;
4044 sysfs_set_num(info
, NULL
, "suspend_lo",
4047 if (max_progress
< *suspend_point
)
4048 max_progress
= *suspend_point
;
4052 /* now set sync_max to allow that progress. sync_max, like
4053 * sync_completed is a count of sectors written per device, so
4054 * we find the difference between max_progress and the start point,
4055 * and divide that by after.data_disks to get a sync_max
4057 * At the same time we convert wait_point to a similar number
4058 * for comparing against sync_completed.
4060 /* scale down max_progress to per_disk */
4061 max_progress
/= reshape
->after
.data_disks
;
4063 * Round to chunk size as some kernels give an erroneously
4066 max_progress
/= info
->new_chunk
/512;
4067 max_progress
*= info
->new_chunk
/512;
4068 /* And round to old chunk size as the kernel wants that */
4069 max_progress
/= info
->array
.chunk_size
/512;
4070 max_progress
*= info
->array
.chunk_size
/512;
4071 /* Limit progress to the whole device */
4072 if (max_progress
> info
->component_size
)
4073 max_progress
= info
->component_size
;
4074 wait_point
/= reshape
->after
.data_disks
;
4076 /* switch from 'device offset' to 'processed block count' */
4077 max_progress
= info
->component_size
- max_progress
;
4078 wait_point
= info
->component_size
- wait_point
;
4082 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4084 /* Now wait. If we have already reached the point that we were
4085 * asked to wait to, don't wait at all, else wait for any change.
4086 * We need to select on 'sync_completed' as that is the place that
4087 * notifications happen, but we are really interested in
4088 * 'reshape_position'
4090 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4092 goto check_progress
;
4094 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4095 goto check_progress
;
4097 while (completed
< max_progress
&& completed
< wait_point
) {
4098 /* Check that sync_action is still 'reshape' to avoid
4099 * waiting forever on a dead array
4102 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) <= 0 ||
4103 strncmp(action
, "reshape", 7) != 0)
4105 /* Some kernels reset 'sync_completed' to zero
4106 * before setting 'sync_action' to 'idle'.
4107 * So we need these extra tests.
4109 if (completed
== 0 && advancing
&&
4110 strncmp(action
, "idle", 4) == 0 &&
4111 info
->reshape_progress
> 0)
4113 if (completed
== 0 && !advancing
&&
4114 strncmp(action
, "idle", 4) == 0 &&
4115 info
->reshape_progress
<
4116 (info
->component_size
* reshape
->after
.data_disks
))
4118 sysfs_wait(fd
, NULL
);
4119 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4120 goto check_progress
;
4122 /* Some kernels reset 'sync_completed' to zero,
4123 * we need to have real point we are in md.
4124 * So in that case, read 'reshape_position' from sysfs.
4126 if (completed
== 0) {
4127 unsigned long long reshapep
;
4129 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) > 0 &&
4130 strncmp(action
, "idle", 4) == 0 &&
4131 sysfs_get_ll(info
, NULL
,
4132 "reshape_position", &reshapep
) == 0)
4133 *reshape_completed
= reshapep
;
4135 /* some kernels can give an incorrectly high
4136 * 'completed' number, so round down */
4137 completed
/= (info
->new_chunk
/512);
4138 completed
*= (info
->new_chunk
/512);
4139 /* Convert 'completed' back in to a 'progress' number */
4140 completed
*= reshape
->after
.data_disks
;
4142 completed
= (info
->component_size
4143 * reshape
->after
.data_disks
4145 *reshape_completed
= completed
;
4150 /* We return the need_backup flag. Caller will decide
4151 * how much - a multiple of ->backup_blocks up to *suspend_point
4154 return need_backup
> info
->reshape_progress
;
4156 return need_backup
>= info
->reshape_progress
;
4159 /* if we couldn't read a number from sync_completed, then
4160 * either the reshape did complete, or it aborted.
4161 * We can tell which by checking for 'none' in reshape_position.
4162 * If it did abort, then it might immediately restart if it
4163 * it was just a device failure that leaves us degraded but
4166 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
,
4167 sizeof(buf
)) < 0 || strncmp(buf
, "none", 4) != 0) {
4168 /* The abort might only be temporary. Wait up to 10
4169 * seconds for fd to contain a valid number again.
4173 unsigned long long new_sync_max
;
4174 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4175 if (sysfs_wait(fd
, &wait
) != 1)
4177 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4179 /* all good again */
4181 /* If "sync_max" is no longer max_progress
4182 * we need to freeze things
4184 sysfs_get_ll(info
, NULL
, "sync_max",
4186 *frozen
= (new_sync_max
!= max_progress
);
4188 case -2: /* read error - abort */
4195 return rv
; /* abort */
4197 /* Maybe racing with array shutdown - check state */
4200 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4202 strncmp(buf
, "inactive", 8) == 0 ||
4203 strncmp(buf
, "clear",5) == 0)
4204 return -2; /* abort */
4205 return -1; /* complete */
4209 /* FIXME return status is never checked */
4210 static int grow_backup(struct mdinfo
*sra
,
4211 unsigned long long offset
, /* per device */
4212 unsigned long stripes
, /* per device, in old chunks */
4213 int *sources
, unsigned long long *offsets
,
4214 int disks
, int chunk
, int level
, int layout
,
4215 int dests
, int *destfd
, unsigned long long *destoffsets
,
4216 int part
, int *degraded
,
4219 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4220 * to storage 'destfd' (offset 'destoffsets'), after first
4221 * suspending IO. Then allow resync to continue
4222 * over the suspended section.
4223 * Use part 'part' of the backup-super-block.
4228 unsigned long long ll
;
4230 //printf("offset %llu\n", offset);
4236 /* Check that array hasn't become degraded, else we might backup the wrong data */
4237 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4238 return -1; /* FIXME this error is ignored */
4239 new_degraded
= (int)ll
;
4240 if (new_degraded
!= *degraded
) {
4241 /* check each device to ensure it is still working */
4243 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4244 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4246 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4249 if (sysfs_get_str(sra
, sd
, "state",
4250 sbuf
, sizeof(sbuf
)) < 0 ||
4251 strstr(sbuf
, "faulty") ||
4252 strstr(sbuf
, "in_sync") == NULL
) {
4253 /* this device is dead */
4254 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4255 if (sd
->disk
.raid_disk
>= 0 &&
4256 sources
[sd
->disk
.raid_disk
] >= 0) {
4257 close(sources
[sd
->disk
.raid_disk
]);
4258 sources
[sd
->disk
.raid_disk
] = -1;
4263 *degraded
= new_degraded
;
4266 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4267 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4269 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4270 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4273 bsb
.magic
[15] = '2';
4274 for (i
= 0; i
< dests
; i
++)
4276 lseek64(destfd
[i
], destoffsets
[i
] +
4277 __le64_to_cpu(bsb
.devstart2
)*512, 0);
4279 lseek64(destfd
[i
], destoffsets
[i
], 0);
4281 rv
= save_stripes(sources
, offsets
, disks
, chunk
, level
, layout
,
4282 dests
, destfd
, offset
* 512 * odata
,
4283 stripes
* chunk
* odata
, buf
);
4287 bsb
.mtime
= __cpu_to_le64(time(0));
4288 for (i
= 0; i
< dests
; i
++) {
4289 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4291 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4292 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4293 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4294 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4295 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4298 if ((unsigned long long)lseek64(destfd
[i
],
4299 destoffsets
[i
] - 4096, 0) !=
4300 destoffsets
[i
] - 4096)
4302 if (write(destfd
[i
], &bsb
, 512) != 512)
4304 if (destoffsets
[i
] > 4096) {
4305 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4306 destoffsets
[i
]+stripes
*chunk
*odata
)
4308 if (write(destfd
[i
], &bsb
, 512) != 512)
4318 /* in 2.6.30, the value reported by sync_completed can be
4319 * less that it should be by one stripe.
4320 * This only happens when reshape hits sync_max and pauses.
4321 * So allow wait_backup to either extent sync_max further
4322 * than strictly necessary, or return before the
4323 * sync has got quite as far as we would really like.
4324 * This is what 'blocks2' is for.
4325 * The various caller give appropriate values so that
4328 /* FIXME return value is often ignored */
4329 static int forget_backup(int dests
, int *destfd
,
4330 unsigned long long *destoffsets
,
4334 * Erase backup 'part' (which is 0 or 1)
4340 bsb
.arraystart2
= __cpu_to_le64(0);
4341 bsb
.length2
= __cpu_to_le64(0);
4343 bsb
.arraystart
= __cpu_to_le64(0);
4344 bsb
.length
= __cpu_to_le64(0);
4346 bsb
.mtime
= __cpu_to_le64(time(0));
4348 for (i
= 0; i
< dests
; i
++) {
4349 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4350 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4351 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4352 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4353 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4354 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4355 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4356 destoffsets
[i
]-4096)
4358 if (rv
== 0 && write(destfd
[i
], &bsb
, 512) != 512)
4365 static void fail(char *msg
)
4368 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4369 rv
|= (write(2, "\n", 1) != 1);
4373 static char *abuf
, *bbuf
;
4374 static unsigned long long abuflen
;
4375 static void validate(int afd
, int bfd
, unsigned long long offset
)
4377 /* check that the data in the backup against the array.
4378 * This is only used for regression testing and should not
4379 * be used while the array is active
4383 lseek64(bfd
, offset
- 4096, 0);
4384 if (read(bfd
, &bsb2
, 512) != 512)
4385 fail("cannot read bsb");
4386 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4387 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4388 fail("first csum bad");
4389 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4390 fail("magic is bad");
4391 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4392 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4393 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4394 fail("second csum bad");
4396 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4397 fail("devstart is wrong");
4400 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4402 if (abuflen
< len
) {
4406 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4407 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4409 /* just stop validating on mem-alloc failure */
4414 lseek64(bfd
, offset
, 0);
4415 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4416 //printf("len %llu\n", len);
4417 fail("read first backup failed");
4419 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4420 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4421 fail("read first from array failed");
4422 if (memcmp(bbuf
, abuf
, len
) != 0) {
4425 printf("offset=%llu len=%llu\n",
4426 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4427 for (i
=0; i
<len
; i
++)
4428 if (bbuf
[i
] != abuf
[i
]) {
4429 printf("first diff byte %d\n", i
);
4433 fail("data1 compare failed");
4437 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4439 if (abuflen
< len
) {
4443 abuf
= xmalloc(abuflen
);
4444 bbuf
= xmalloc(abuflen
);
4447 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4448 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4449 fail("read second backup failed");
4450 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4451 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4452 fail("read second from array failed");
4453 if (memcmp(bbuf
, abuf
, len
) != 0)
4454 fail("data2 compare failed");
4458 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4459 struct supertype
*st
, unsigned long blocks
,
4460 int *fds
, unsigned long long *offsets
,
4461 int dests
, int *destfd
, unsigned long long *destoffsets
)
4463 /* Monitor a reshape where backup is being performed using
4464 * 'native' mechanism - either to a backup file, or
4465 * to some space in a spare.
4469 unsigned long long speed
;
4470 unsigned long long suspend_point
, array_size
;
4471 unsigned long long backup_point
, wait_point
;
4472 unsigned long long reshape_completed
;
4474 int increasing
= reshape
->after
.data_disks
>=
4475 reshape
->before
.data_disks
;
4476 int part
= 0; /* The next part of the backup area to fill. It
4477 * may already be full, so we need to check */
4478 int level
= reshape
->level
;
4479 int layout
= reshape
->before
.layout
;
4480 int data
= reshape
->before
.data_disks
;
4481 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4482 int chunk
= sra
->array
.chunk_size
;
4484 unsigned long stripes
;
4488 /* set up the backup-super-block. This requires the
4489 * uuid from the array.
4491 /* Find a superblock */
4492 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4496 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4498 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4499 devfd
= dev_open(dn
, O_RDONLY
);
4502 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4508 pr_err("Cannot find a superblock\n");
4512 memset(&bsb
, 0, 512);
4513 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4514 st
->ss
->uuid_from_super(st
, uuid
);
4515 memcpy(bsb
.set_uuid
, uuid
, 16);
4516 bsb
.mtime
= __cpu_to_le64(time(0));
4517 bsb
.devstart2
= blocks
;
4519 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4520 reshape
->before
.data_disks
;
4522 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4523 /* Don't start the 'reshape' */
4525 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4526 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4527 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4531 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4532 backup_point
= sra
->reshape_progress
;
4535 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4536 backup_point
= reshape
->backup_blocks
;
4537 suspend_point
= array_size
;
4543 /* Want to return as soon the oldest backup slot can
4544 * be released as that allows us to start backing up
4545 * some more, providing suspend_point has been
4546 * advanced, which it should have.
4549 wait_point
= array_size
;
4550 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4551 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4552 __le64_to_cpu(bsb
.length
));
4553 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4554 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4555 __le64_to_cpu(bsb
.length2
));
4558 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4559 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4560 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4561 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4564 reshape_completed
= sra
->reshape_progress
;
4565 rv
= progress_reshape(sra
, reshape
,
4566 backup_point
, wait_point
,
4567 &suspend_point
, &reshape_completed
,
4569 /* external metadata would need to ping_monitor here */
4570 sra
->reshape_progress
= reshape_completed
;
4572 /* Clear any backup region that is before 'here' */
4574 if (__le64_to_cpu(bsb
.length
) > 0 &&
4575 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4576 __le64_to_cpu(bsb
.length
)))
4577 forget_backup(dests
, destfd
,
4579 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4580 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4581 __le64_to_cpu(bsb
.length2
)))
4582 forget_backup(dests
, destfd
,
4585 if (__le64_to_cpu(bsb
.length
) > 0 &&
4586 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4587 forget_backup(dests
, destfd
,
4589 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4590 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4591 forget_backup(dests
, destfd
,
4601 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4602 /* No longer need to monitor this reshape */
4603 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4609 unsigned long long offset
;
4610 unsigned long actual_stripes
;
4611 /* Need to backup some data.
4612 * If 'part' is not used and the desired
4613 * backup size is suspended, do a backup,
4614 * then consider the next part.
4616 /* Check that 'part' is unused */
4617 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4619 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4622 offset
= backup_point
/ data
;
4623 actual_stripes
= stripes
;
4625 if (offset
+ actual_stripes
* (chunk
/512) >
4626 sra
->component_size
)
4627 actual_stripes
= ((sra
->component_size
- offset
)
4629 if (offset
+ actual_stripes
* (chunk
/512) >
4633 if (offset
< actual_stripes
* (chunk
/512))
4634 actual_stripes
= offset
/ (chunk
/512);
4635 offset
-= actual_stripes
* (chunk
/512);
4636 if (offset
< suspend_point
/data
)
4639 if (actual_stripes
== 0)
4641 grow_backup(sra
, offset
, actual_stripes
, fds
, offsets
,
4642 disks
, chunk
, level
, layout
, dests
, destfd
,
4643 destoffsets
, part
, °raded
, buf
);
4644 validate(afd
, destfd
[0], destoffsets
[0]);
4645 /* record where 'part' is up to */
4648 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4650 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4654 /* FIXME maybe call progress_reshape one more time instead */
4655 /* remove any remaining suspension */
4656 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4657 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4658 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4659 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4661 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4662 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4668 * If any spare contains md_back_data-1 which is recent wrt mtime,
4669 * write that data into the array and update the super blocks with
4670 * the new reshape_progress
4672 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
,
4673 int cnt
, char *backup_file
, int verbose
)
4677 unsigned long long *offsets
;
4678 unsigned long long nstripe
, ostripe
;
4681 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4682 if (info
->array
.level
== 6)
4683 odata
--; /* number of data disks */
4684 ndata
= info
->array
.raid_disks
- 1;
4685 if (info
->new_level
== 6)
4688 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4690 if (info
->delta_disks
<= 0)
4691 /* Didn't grow, so the backup file must have
4695 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4696 struct mdinfo dinfo
;
4699 char *devname
, namebuf
[20];
4700 unsigned long long lo
, hi
;
4702 /* This was a spare and may have some saved data on it.
4703 * Load the superblock, find and load the
4704 * backup_super_block.
4705 * If either fail, go on to next device.
4706 * If the backup contains no new info, just return
4707 * else restore data and update all superblocks
4709 if (i
== old_disks
-1) {
4710 fd
= open(backup_file
, O_RDONLY
);
4712 pr_err("backup file %s inaccessible: %s\n",
4713 backup_file
, strerror(errno
));
4716 devname
= backup_file
;
4721 if (st
->ss
->load_super(st
, fd
, NULL
))
4724 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4725 st
->ss
->free_super(st
);
4728 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4730 pr_err("Cannot seek on device %d\n", i
);
4731 continue; /* Cannot seek */
4733 sprintf(namebuf
, "device-%d", i
);
4736 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4738 pr_err("Cannot read from %s\n", devname
);
4739 continue; /* Cannot read */
4741 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4742 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4744 pr_err("No backup metadata on %s\n", devname
);
4747 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4749 pr_err("Bad backup-metadata checksum on %s\n",
4751 continue; /* bad checksum */
4753 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4754 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4756 pr_err("Bad backup-metadata checksum2 on %s\n",
4758 continue; /* Bad second checksum */
4760 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4762 pr_err("Wrong uuid on backup-metadata on %s\n",
4764 continue; /* Wrong uuid */
4768 * array utime and backup-mtime should be updated at
4769 * much the same time, but it seems that sometimes
4770 * they aren't... So allow considerable flexability in
4771 * matching, and allow this test to be overridden by
4772 * an environment variable.
4774 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4775 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4776 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4777 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4778 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4779 (unsigned long)info
->array
.utime
);
4781 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4782 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4783 continue; /* time stamp is too bad */
4787 if (bsb
.magic
[15] == '1') {
4788 if (bsb
.length
== 0)
4790 if (info
->delta_disks
>= 0) {
4791 /* reshape_progress is increasing */
4792 if (__le64_to_cpu(bsb
.arraystart
)
4793 + __le64_to_cpu(bsb
.length
)
4794 < info
->reshape_progress
) {
4797 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4798 continue; /* No new data here */
4801 /* reshape_progress is decreasing */
4802 if (__le64_to_cpu(bsb
.arraystart
) >=
4803 info
->reshape_progress
)
4804 goto nonew
; /* No new data here */
4807 if (bsb
.length
== 0 && bsb
.length2
== 0)
4809 if (info
->delta_disks
>= 0) {
4810 /* reshape_progress is increasing */
4811 if ((__le64_to_cpu(bsb
.arraystart
)
4812 + __le64_to_cpu(bsb
.length
)
4813 < info
->reshape_progress
) &&
4814 (__le64_to_cpu(bsb
.arraystart2
)
4815 + __le64_to_cpu(bsb
.length2
)
4816 < info
->reshape_progress
))
4817 goto nonew
; /* No new data here */
4819 /* reshape_progress is decreasing */
4820 if (__le64_to_cpu(bsb
.arraystart
) >=
4821 info
->reshape_progress
&&
4822 __le64_to_cpu(bsb
.arraystart2
) >=
4823 info
->reshape_progress
)
4824 goto nonew
; /* No new data here */
4827 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4830 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4832 continue; /* Cannot seek */
4834 /* There should be a duplicate backup superblock 4k before here */
4835 if (lseek64(fd
, -4096, 1) < 0 ||
4836 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4837 goto second_fail
; /* Cannot find leading superblock */
4838 if (bsb
.magic
[15] == '1')
4839 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4841 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4842 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4843 goto second_fail
; /* Cannot find leading superblock */
4845 /* Now need the data offsets for all devices. */
4846 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4847 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4850 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4851 /* FIXME should be this be an error */
4853 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4854 st
->ss
->free_super(st
);
4855 offsets
[j
] = dinfo
.data_offset
* 512;
4857 printf("%s: restoring critical section\n", Name
);
4859 if (restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4860 info
->new_chunk
, info
->new_level
,
4861 info
->new_layout
, fd
,
4862 __le64_to_cpu(bsb
.devstart
)*512,
4863 __le64_to_cpu(bsb
.arraystart
)*512,
4864 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4865 /* didn't succeed, so giveup */
4867 pr_err("Error restoring backup from %s\n",
4873 if (bsb
.magic
[15] == '2' &&
4874 restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4875 info
->new_chunk
, info
->new_level
,
4876 info
->new_layout
, fd
,
4877 __le64_to_cpu(bsb
.devstart
)*512 +
4878 __le64_to_cpu(bsb
.devstart2
)*512,
4879 __le64_to_cpu(bsb
.arraystart2
)*512,
4880 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4881 /* didn't succeed, so giveup */
4883 pr_err("Error restoring second backup from %s\n",
4891 /* Ok, so the data is restored. Let's update those superblocks. */
4895 lo
= __le64_to_cpu(bsb
.arraystart
);
4896 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4898 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4899 unsigned long long lo1
, hi1
;
4900 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4901 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4905 } else if (lo
< lo1
)
4910 if (lo
< hi
&& (info
->reshape_progress
< lo
||
4911 info
->reshape_progress
> hi
))
4912 /* backup does not affect reshape_progress*/ ;
4913 else if (info
->delta_disks
>= 0) {
4914 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4915 __le64_to_cpu(bsb
.length
);
4916 if (bsb
.magic
[15] == '2') {
4917 unsigned long long p2
;
4919 p2
= __le64_to_cpu(bsb
.arraystart2
) +
4920 __le64_to_cpu(bsb
.length2
);
4921 if (p2
> info
->reshape_progress
)
4922 info
->reshape_progress
= p2
;
4925 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4926 if (bsb
.magic
[15] == '2') {
4927 unsigned long long p2
;
4929 p2
= __le64_to_cpu(bsb
.arraystart2
);
4930 if (p2
< info
->reshape_progress
)
4931 info
->reshape_progress
= p2
;
4934 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4937 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4939 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4940 dinfo
.reshape_progress
= info
->reshape_progress
;
4941 st
->ss
->update_super(st
, &dinfo
, "_reshape_progress",
4943 st
->ss
->store_super(st
, fdlist
[j
]);
4944 st
->ss
->free_super(st
);
4948 /* Didn't find any backup data, try to see if any
4951 if (info
->delta_disks
< 0) {
4952 /* When shrinking, the critical section is at the end.
4953 * So see if we are before the critical section.
4955 unsigned long long first_block
;
4956 nstripe
= ostripe
= 0;
4958 while (ostripe
>= nstripe
) {
4959 ostripe
+= info
->array
.chunk_size
/ 512;
4960 first_block
= ostripe
* odata
;
4961 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4962 (info
->new_chunk
/512);
4965 if (info
->reshape_progress
>= first_block
)
4968 if (info
->delta_disks
> 0) {
4969 /* See if we are beyond the critical section. */
4970 unsigned long long last_block
;
4971 nstripe
= ostripe
= 0;
4973 while (nstripe
>= ostripe
) {
4974 nstripe
+= info
->new_chunk
/ 512;
4975 last_block
= nstripe
* ndata
;
4976 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4977 (info
->array
.chunk_size
/512);
4980 if (info
->reshape_progress
>= last_block
)
4983 /* needed to recover critical section! */
4985 pr_err("Failed to find backup of critical section\n");
4989 int Grow_continue_command(char *devname
, int fd
,
4990 char *backup_file
, int verbose
)
4993 struct supertype
*st
= NULL
;
4994 struct mdinfo
*content
= NULL
;
4995 struct mdinfo array
;
4996 char *subarray
= NULL
;
4997 struct mdinfo
*cc
= NULL
;
4998 struct mdstat_ent
*mdstat
= NULL
;
5002 dprintf("Grow continue from command line called for %s\n", devname
);
5004 st
= super_by_fd(fd
, &subarray
);
5005 if (!st
|| !st
->ss
) {
5006 pr_err("Unable to determine metadata format for %s\n", devname
);
5009 dprintf("Grow continue is run for ");
5010 if (st
->ss
->external
== 0) {
5013 dprintf_cont("native array (%s)\n", devname
);
5014 if (md_get_array_info(fd
, &array
.array
) < 0) {
5015 pr_err("%s is not an active md array - aborting\n",
5018 goto Grow_continue_command_exit
;
5021 sysfs_init(content
, fd
, NULL
);
5022 /* Need to load a superblock.
5023 * FIXME we should really get what we need from
5027 for (d
= 0; d
< MAX_DISKS
; d
++) {
5028 mdu_disk_info_t disk
;
5032 if (md_get_disk_info(fd
, &disk
) < 0)
5034 if (disk
.major
== 0 && disk
.minor
== 0)
5036 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5038 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5041 fd2
= dev_open(dv
, O_RDONLY
);
5044 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5050 if (d
== MAX_DISKS
) {
5051 pr_err("Unable to load metadata for %s\n",
5054 goto Grow_continue_command_exit
;
5056 st
->ss
->getinfo_super(st
, content
, NULL
);
5057 if (!content
->reshape_active
)
5061 } while (cnt
-- > 0);
5066 dprintf_cont("subarray (%s)\n", subarray
);
5067 container
= st
->container_devnm
;
5068 cfd
= open_dev_excl(st
->container_devnm
);
5070 container
= st
->devnm
;
5072 cfd
= open_dev_excl(st
->devnm
);
5073 dprintf_cont("container (%s)\n", container
);
5077 pr_err("Unable to open container for %s\n", devname
);
5079 goto Grow_continue_command_exit
;
5082 /* find in container array under reshape
5084 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5086 pr_err("Cannot read superblock for %s\n", devname
);
5088 goto Grow_continue_command_exit
;
5091 cc
= st
->ss
->container_content(st
, subarray
);
5092 for (content
= cc
; content
; content
= content
->next
) {
5094 int allow_reshape
= 1;
5096 if (content
->reshape_active
== 0)
5098 /* The decision about array or container wide
5099 * reshape is taken in Grow_continue based
5100 * content->reshape_active state, therefore we
5101 * need to check_reshape based on
5102 * reshape_active and subarray name
5104 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5106 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5107 (content
->array
.state
5108 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5111 if (!allow_reshape
) {
5112 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5113 devname
, container
);
5115 goto Grow_continue_command_exit
;
5118 array_name
= strchr(content
->text_version
+1, '/')+1;
5119 mdstat
= mdstat_by_subdev(array_name
, container
);
5122 if (mdstat
->active
== 0) {
5123 pr_err("Skipping inactive array %s.\n",
5125 free_mdstat(mdstat
);
5132 pr_err("Unable to determine reshaped array for %s\n", devname
);
5134 goto Grow_continue_command_exit
;
5136 fd2
= open_dev(mdstat
->devnm
);
5138 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5140 goto Grow_continue_command_exit
;
5143 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5144 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5148 goto Grow_continue_command_exit
;
5153 /* start mdmon in case it is not running
5155 if (!mdmon_running(container
))
5156 start_mdmon(container
);
5157 ping_monitor(container
);
5159 if (mdmon_running(container
))
5160 st
->update_tail
= &st
->updates
;
5162 pr_err("No mdmon found. Grow cannot continue.\n");
5164 goto Grow_continue_command_exit
;
5168 /* verify that array under reshape is started from
5171 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5173 goto Grow_continue_command_exit
;
5178 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5180 Grow_continue_command_exit
:
5183 st
->ss
->free_super(st
);
5184 free_mdstat(mdstat
);
5191 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5192 char *backup_file
, int forked
, int freeze_reshape
)
5196 if (!info
->reshape_active
)
5199 if (st
->ss
->external
) {
5200 int cfd
= open_dev(st
->container_devnm
);
5205 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5207 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5208 st
, info
, 0, backup_file
, 0,
5209 forked
, 1 | info
->reshape_active
,
5212 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5213 NULL
, INVALID_SECTORS
, backup_file
,
5214 0, forked
, 1 | info
->reshape_active
,
5220 char *make_backup(char *name
)
5222 char *base
= "backup_file-";
5226 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5227 fname
= xmalloc(len
);
5228 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5232 char *locate_backup(char *name
)
5234 char *fl
= make_backup(name
);
5237 if (stat(fl
, &stb
) == 0 && S_ISREG(stb
.st_mode
))