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 /* frozen array is not idle*/
758 if (e
->percent
>= 0 || e
->metadata_version
[9] == '-') {
767 static int freeze_container(struct supertype
*st
)
769 char *container
= (st
->container_devnm
[0]
770 ? st
->container_devnm
: st
->devnm
);
775 if (block_monitor(container
, 1)) {
776 pr_err("failed to freeze container\n");
783 static void unfreeze_container(struct supertype
*st
)
785 char *container
= (st
->container_devnm
[0]
786 ? st
->container_devnm
: st
->devnm
);
788 unblock_monitor(container
, 1);
791 static int freeze(struct supertype
*st
)
793 /* Try to freeze resync/rebuild on this array/container.
794 * Return -1 if the array is busy,
795 * return -2 container cannot be frozen,
796 * return 0 if this kernel doesn't support 'frozen'
797 * return 1 if it worked.
799 if (st
->ss
->external
)
800 return freeze_container(st
);
802 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
808 /* Need to clear any 'read-auto' status */
809 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
810 strncmp(buf
, "read-auto", 9) == 0)
811 sysfs_set_str(sra
, NULL
, "array_state", "clean");
813 err
= sysfs_freeze_array(sra
);
819 static void unfreeze(struct supertype
*st
)
821 if (st
->ss
->external
)
822 return unfreeze_container(st
);
824 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
828 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
829 strcmp(buf
, "frozen\n") == 0)
830 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
835 static void wait_reshape(struct mdinfo
*sra
)
837 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
843 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
844 strncmp(action
, "reshape", 7) == 0)
845 sysfs_wait(fd
, NULL
);
849 static int reshape_super(struct supertype
*st
, unsigned long long size
,
850 int level
, int layout
, int chunksize
, int raid_disks
,
851 int delta_disks
, char *backup_file
, char *dev
,
852 int direction
, int verbose
)
854 /* nothing extra to check in the native case */
855 if (!st
->ss
->external
)
857 if (!st
->ss
->reshape_super
|| !st
->ss
->manage_reshape
) {
858 pr_err("%s metadata does not support reshape\n",
863 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
864 raid_disks
, delta_disks
, backup_file
, dev
,
868 static void sync_metadata(struct supertype
*st
)
870 if (st
->ss
->external
) {
871 if (st
->update_tail
) {
872 flush_metadata_updates(st
);
873 st
->update_tail
= &st
->updates
;
875 st
->ss
->sync_metadata(st
);
879 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
881 /* when dealing with external metadata subarrays we need to be
882 * prepared to handle EAGAIN. The kernel may need to wait for
883 * mdmon to mark the array active so the kernel can handle
884 * allocations/writeback when preparing the reshape action
885 * (md_allow_write()). We temporarily disable safe_mode_delay
886 * to close a race with the array_state going clean before the
887 * next write to raid_disks / stripe_cache_size
892 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
894 (strcmp(name
, "raid_disks") != 0 &&
895 strcmp(name
, "stripe_cache_size") != 0))
896 return sysfs_set_num(sra
, NULL
, name
, n
);
898 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
901 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
902 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
903 if (rc
< 0 && errno
== EAGAIN
) {
904 ping_monitor(container
);
905 /* if we get EAGAIN here then the monitor is not active
908 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
910 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
914 int start_reshape(struct mdinfo
*sra
, int already_running
,
915 int before_data_disks
, int data_disks
)
918 unsigned long long sync_max_to_set
;
920 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
921 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
922 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
923 sra
->reshape_progress
);
924 if (before_data_disks
<= data_disks
)
925 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
927 sync_max_to_set
= (sra
->component_size
* data_disks
928 - sra
->reshape_progress
) / data_disks
;
929 if (!already_running
)
930 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
931 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
932 if (!already_running
&& err
== 0) {
935 err
= sysfs_set_str(sra
, NULL
, "sync_action",
939 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
944 void abort_reshape(struct mdinfo
*sra
)
946 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
948 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
949 * suspend_hi to decrease as well as increase.")
950 * you could only increase suspend_{lo,hi} unless the region they
951 * covered was empty. So to reset to 0, you need to push suspend_lo
952 * up past suspend_hi first. So to maximize the chance of mdadm
953 * working on all kernels, we want to keep doing that.
955 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
956 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
957 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
958 sysfs_set_num(sra
, NULL
, "sync_min", 0);
959 // It isn't safe to reset sync_max as we aren't monitoring.
960 // Array really should be stopped at this point.
963 int remove_disks_for_takeover(struct supertype
*st
,
968 struct mdinfo
*remaining
;
971 if (st
->ss
->external
) {
973 struct mdinfo
*arrays
= st
->ss
->container_content(st
, NULL
);
975 * containter_content returns list of arrays in container
976 * If arrays->next is not NULL it means that there are
977 * 2 arrays in container and operation should be blocked
984 pr_err("Error. Cannot perform operation on /dev/%s\n", st
->devnm
);
985 pr_err("For this operation it MUST be single array in container\n");
991 if (sra
->array
.level
== 10)
992 nr_of_copies
= layout
& 0xff;
993 else if (sra
->array
.level
== 1)
994 nr_of_copies
= sra
->array
.raid_disks
;
998 remaining
= sra
->devs
;
1000 /* for each 'copy', select one device and remove from the list. */
1001 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
1002 struct mdinfo
**diskp
;
1005 /* Find a working device to keep */
1006 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
1007 struct mdinfo
*disk
= *diskp
;
1009 if (disk
->disk
.raid_disk
< slot
)
1011 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
1013 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
1015 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
1017 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
1020 /* We have found a good disk to use! */
1021 *diskp
= disk
->next
;
1022 disk
->next
= sra
->devs
;
1031 if (slot
< sra
->array
.raid_disks
) {
1032 /* didn't find all slots */
1038 sra
->devs
= remaining
;
1042 /* Remove all 'remaining' devices from the array */
1044 struct mdinfo
*sd
= remaining
;
1045 remaining
= sd
->next
;
1047 sysfs_set_str(sra
, sd
, "state", "faulty");
1048 sysfs_set_str(sra
, sd
, "slot", "none");
1049 /* for external metadata disks should be removed in mdmon */
1050 if (!st
->ss
->external
)
1051 sysfs_set_str(sra
, sd
, "state", "remove");
1052 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
1053 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
1054 sd
->next
= sra
->devs
;
1060 void reshape_free_fdlist(int *fdlist
,
1061 unsigned long long *offsets
,
1066 for (i
= 0; i
< size
; i
++)
1074 int reshape_prepare_fdlist(char *devname
,
1078 unsigned long blocks
,
1081 unsigned long long *offsets
)
1086 enable_fds(nrdisks
);
1087 for (d
= 0; d
<= nrdisks
; d
++)
1090 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1091 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1093 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
1094 sd
->disk
.raid_disk
< raid_disks
) {
1095 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1096 fdlist
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDONLY
);
1097 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1098 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1099 pr_err("%s: cannot open component %s\n",
1100 devname
, dn
? dn
: "-unknown-");
1104 } else if (backup_file
== NULL
) {
1106 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1107 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1108 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1109 if (fdlist
[d
] < 0) {
1110 pr_err("%s: cannot open component %s\n",
1111 devname
, dn
? dn
: "-unknown-");
1122 int reshape_open_backup_file(char *backup_file
,
1127 unsigned long long *offsets
,
1131 /* Return 1 on success, 0 on any form of failure */
1132 /* need to check backup file is large enough */
1138 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
1142 pr_err("%s: cannot create backup file %s: %s\n",
1143 devname
, backup_file
, strerror(errno
));
1146 /* Guard against backup file being on array device.
1147 * If array is partitioned or if LVM etc is in the
1148 * way this will not notice, but it is better than
1151 fstat(*fdlist
, &stb
);
1154 if (stb
.st_rdev
== dev
) {
1155 pr_err("backup file must NOT be on the array being reshaped.\n");
1160 memset(buf
, 0, 512);
1161 for (i
=0; i
< blocks
+ 8 ; i
++) {
1162 if (write(*fdlist
, buf
, 512) != 512) {
1163 pr_err("%s: cannot create backup file %s: %s\n",
1164 devname
, backup_file
, strerror(errno
));
1168 if (fsync(*fdlist
) != 0) {
1169 pr_err("%s: cannot create backup file %s: %s\n",
1170 devname
, backup_file
, strerror(errno
));
1174 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
1175 char *bu
= make_backup(sys_name
);
1176 if (symlink(backup_file
, bu
))
1177 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
1185 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
1186 unsigned int ndata
, unsigned int odata
)
1188 unsigned long a
, b
, blocks
;
1189 /* So how much do we need to backup.
1190 * We need an amount of data which is both a whole number of
1191 * old stripes and a whole number of new stripes.
1192 * So LCM for (chunksize*datadisks).
1194 a
= (ochunk
/512) * odata
;
1195 b
= (nchunk
/512) * ndata
;
1198 /* LCM == product / GCD */
1199 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
1204 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1206 /* Based on the current array state in info->array and
1207 * the changes in info->new_* etc, determine:
1208 * - whether the change is possible
1209 * - Intermediate level/raid_disks/layout
1210 * - whether a restriping reshape is needed
1211 * - number of sectors in minimum change unit. This
1212 * will cover a whole number of stripes in 'before' and
1215 * Return message if the change should be rejected
1216 * NULL if the change can be achieved
1218 * This can be called as part of starting a reshape, or
1219 * when assembling an array that is undergoing reshape.
1221 int near
, far
, offset
, copies
;
1223 int old_chunk
, new_chunk
;
1224 /* delta_parity records change in number of devices
1225 * caused by level change
1227 int delta_parity
= 0;
1229 memset(re
, 0, sizeof(*re
));
1231 /* If a new level not explicitly given, we assume no-change */
1232 if (info
->new_level
== UnSet
)
1233 info
->new_level
= info
->array
.level
;
1235 if (info
->new_chunk
)
1236 switch (info
->new_level
) {
1242 /* chunk size is meaningful, must divide component_size
1245 if (info
->component_size
% (info
->new_chunk
/512)) {
1246 unsigned long long shrink
= info
->component_size
;
1247 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1248 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1249 info
->new_chunk
/1024, info
->component_size
/2);
1250 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1252 pr_err("will shrink the array so the given chunk size would work.\n");
1257 return "chunk size not meaningful for this level";
1260 info
->new_chunk
= info
->array
.chunk_size
;
1262 switch (info
->array
.level
) {
1264 return "No reshape is possibly for this RAID level";
1266 if (info
->delta_disks
!= UnSet
)
1267 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1269 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1271 /* RAID1 can convert to RAID1 with different disks, or
1272 * raid5 with 2 disks, or
1275 if (info
->new_level
> 1 && (info
->component_size
& 7))
1276 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1277 if (info
->new_level
== 0) {
1278 if (info
->delta_disks
!= UnSet
&&
1279 info
->delta_disks
!= 0)
1280 return "Cannot change number of disks with RAID1->RAID0 conversion";
1282 re
->before
.data_disks
= 1;
1283 re
->after
.data_disks
= 1;
1286 if (info
->new_level
== 1) {
1287 if (info
->delta_disks
== UnSet
)
1288 /* Don't know what to do */
1289 return "no change requested for Growing RAID1";
1293 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1294 return "Can only convert a 2-device array to RAID5";
1295 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1297 re
->before
.data_disks
= 1;
1298 if (info
->delta_disks
!= UnSet
&&
1299 info
->delta_disks
!= 0)
1300 re
->after
.data_disks
= 1 + info
->delta_disks
;
1302 re
->after
.data_disks
= 1;
1303 if (re
->after
.data_disks
< 1)
1304 return "Number of disks too small for RAID5";
1306 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1307 info
->array
.chunk_size
= 65536;
1310 /* Could do some multi-stage conversions, but leave that to
1313 return "Impossibly level change request for RAID1";
1316 /* RAID10 can be converted from near mode to
1317 * RAID0 by removing some devices.
1318 * It can also be reshaped if the kernel supports
1321 switch (info
->new_level
) {
1323 if ((info
->array
.layout
& ~0xff) != 0x100)
1324 return "Cannot Grow RAID10 with far/offset layout";
1326 * number of devices must be multiple of
1329 if (info
->array
.raid_disks
%
1330 (info
->array
.layout
& 0xff))
1331 return "RAID10 layout too complex for Grow operation";
1333 new_disks
= (info
->array
.raid_disks
/
1334 (info
->array
.layout
& 0xff));
1335 if (info
->delta_disks
== UnSet
)
1336 info
->delta_disks
= (new_disks
1337 - info
->array
.raid_disks
);
1339 if (info
->delta_disks
!=
1340 new_disks
- info
->array
.raid_disks
)
1341 return "New number of raid-devices impossible for RAID10";
1342 if (info
->new_chunk
&&
1343 info
->new_chunk
!= info
->array
.chunk_size
)
1344 return "Cannot change chunk-size with RAID10 Grow";
1348 re
->before
.data_disks
= new_disks
;
1349 re
->after
.data_disks
= re
->before
.data_disks
;
1353 near
= info
->array
.layout
& 0xff;
1354 far
= (info
->array
.layout
>> 8) & 0xff;
1355 offset
= info
->array
.layout
& 0x10000;
1356 if (far
> 1 && !offset
)
1357 return "Cannot reshape RAID10 in far-mode";
1358 copies
= near
* far
;
1360 old_chunk
= info
->array
.chunk_size
* far
;
1362 if (info
->new_layout
== UnSet
)
1363 info
->new_layout
= info
->array
.layout
;
1365 near
= info
->new_layout
& 0xff;
1366 far
= (info
->new_layout
>> 8) & 0xff;
1367 offset
= info
->new_layout
& 0x10000;
1368 if (far
> 1 && !offset
)
1369 return "Cannot reshape RAID10 to far-mode";
1370 if (near
* far
!= copies
)
1371 return "Cannot change number of copies when reshaping RAID10";
1373 if (info
->delta_disks
== UnSet
)
1374 info
->delta_disks
= 0;
1375 new_disks
= (info
->array
.raid_disks
+
1378 new_chunk
= info
->new_chunk
* far
;
1381 re
->before
.layout
= info
->array
.layout
;
1382 re
->before
.data_disks
= info
->array
.raid_disks
;
1383 re
->after
.layout
= info
->new_layout
;
1384 re
->after
.data_disks
= new_disks
;
1385 /* For RAID10 we don't do backup but do allow reshape,
1386 * so set backup_blocks to INVALID_SECTORS rather than
1388 * And there is no need to synchronise stripes on both
1389 * 'old' and 'new'. So the important
1390 * number is the minimum data_offset difference
1391 * which is the larger of (offset copies * chunk).
1393 re
->backup_blocks
= INVALID_SECTORS
;
1394 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1395 if (new_disks
< re
->before
.data_disks
&&
1396 info
->space_after
< re
->min_offset_change
)
1397 /* Reduce component size by one chunk */
1398 re
->new_size
= (info
->component_size
-
1399 re
->min_offset_change
);
1401 re
->new_size
= info
->component_size
;
1402 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1406 return "RAID10 can only be changed to RAID0";
1409 /* RAID0 can be converted to RAID10, or to RAID456 */
1410 if (info
->new_level
== 10) {
1411 if (info
->new_layout
== UnSet
&&
1412 info
->delta_disks
== UnSet
) {
1413 /* Assume near=2 layout */
1414 info
->new_layout
= 0x102;
1415 info
->delta_disks
= info
->array
.raid_disks
;
1417 if (info
->new_layout
== UnSet
) {
1418 int copies
= 1 + (info
->delta_disks
1419 / info
->array
.raid_disks
);
1420 if (info
->array
.raid_disks
* (copies
-1) !=
1422 return "Impossible number of devices for RAID0->RAID10";
1423 info
->new_layout
= 0x100 + copies
;
1425 if (info
->delta_disks
== UnSet
) {
1426 int copies
= info
->new_layout
& 0xff;
1427 if (info
->new_layout
!= 0x100 + copies
)
1428 return "New layout impossible for RAID0->RAID10";;
1429 info
->delta_disks
= (copies
- 1) *
1430 info
->array
.raid_disks
;
1432 if (info
->new_chunk
&&
1433 info
->new_chunk
!= info
->array
.chunk_size
)
1434 return "Cannot change chunk-size with RAID0->RAID10";
1437 re
->before
.data_disks
= (info
->array
.raid_disks
+
1439 re
->after
.data_disks
= re
->before
.data_disks
;
1440 re
->before
.layout
= info
->new_layout
;
1444 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1445 * a raid4 style layout of the final level.
1447 switch (info
->new_level
) {
1452 re
->before
.layout
= 0;
1457 re
->before
.layout
= ALGORITHM_PARITY_N
;
1458 if (info
->new_layout
== UnSet
)
1459 info
->new_layout
= map_name(r5layout
, "default");
1464 re
->before
.layout
= ALGORITHM_PARITY_N
;
1465 if (info
->new_layout
== UnSet
)
1466 info
->new_layout
= map_name(r6layout
, "default");
1469 return "Impossible level change requested";
1471 re
->before
.data_disks
= info
->array
.raid_disks
;
1472 /* determining 'after' layout happens outside this 'switch' */
1476 info
->array
.layout
= ALGORITHM_PARITY_N
;
1478 switch (info
->new_level
) {
1482 re
->level
= info
->array
.level
;
1483 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1484 re
->before
.layout
= info
->array
.layout
;
1488 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1489 re
->before
.layout
= info
->array
.layout
;
1494 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1495 switch (info
->array
.layout
) {
1496 case ALGORITHM_LEFT_ASYMMETRIC
:
1497 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1499 case ALGORITHM_RIGHT_ASYMMETRIC
:
1500 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1502 case ALGORITHM_LEFT_SYMMETRIC
:
1503 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1505 case ALGORITHM_RIGHT_SYMMETRIC
:
1506 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1508 case ALGORITHM_PARITY_0
:
1509 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1511 case ALGORITHM_PARITY_N
:
1512 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1515 return "Cannot convert an array with this layout";
1519 if (info
->array
.raid_disks
!= 2)
1520 return "Can only convert a 2-device array to RAID1";
1521 if (info
->delta_disks
!= UnSet
&&
1522 info
->delta_disks
!= 0)
1523 return "Cannot set raid_disk when converting RAID5->RAID1";
1525 info
->new_chunk
= 0;
1528 return "Impossible level change requested";
1532 switch (info
->new_level
) {
1538 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1539 re
->before
.layout
= info
->array
.layout
;
1542 return "Impossible level change requested";
1547 /* If we reached here then it looks like a re-stripe is
1548 * happening. We have determined the intermediate level
1549 * and initial raid_disks/layout and stored these in 're'.
1551 * We need to deduce the final layout that can be atomically
1552 * converted to the end state.
1554 switch (info
->new_level
) {
1556 /* We can only get to RAID0 from RAID4 or RAID5
1557 * with appropriate layout and one extra device
1559 if (re
->level
!= 4 && re
->level
!= 5)
1560 return "Cannot covert to RAID0 from this level";
1562 switch (re
->level
) {
1564 re
->before
.layout
= 0;
1565 re
->after
.layout
= 0;
1568 re
->after
.layout
= ALGORITHM_PARITY_N
;
1574 /* We can only get to RAID4 from RAID5 */
1575 if (re
->level
!= 4 && re
->level
!= 5)
1576 return "Cannot convert to RAID4 from this level";
1578 switch (re
->level
) {
1580 re
->after
.layout
= 0;
1583 re
->after
.layout
= ALGORITHM_PARITY_N
;
1589 /* We get to RAID5 from RAID5 or RAID6 */
1590 if (re
->level
!= 5 && re
->level
!= 6)
1591 return "Cannot convert to RAID5 from this level";
1593 switch (re
->level
) {
1595 if (info
->new_layout
== UnSet
)
1596 re
->after
.layout
= re
->before
.layout
;
1598 re
->after
.layout
= info
->new_layout
;
1601 if (info
->new_layout
== UnSet
)
1602 info
->new_layout
= re
->before
.layout
;
1604 /* after.layout needs to be raid6 version of new_layout */
1605 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1606 re
->after
.layout
= ALGORITHM_PARITY_N
;
1609 char *ls
= map_num(r5layout
, info
->new_layout
);
1612 /* Current RAID6 layout has a RAID5
1615 strcat(strcpy(layout
, ls
), "-6");
1616 l
= map_name(r6layout
, layout
);
1618 return "Cannot find RAID6 layout to convert to";
1620 /* Current RAID6 has no equivalent.
1621 * If it is already a '-6' layout we
1622 * can leave it unchanged, else we must
1625 ls
= map_num(r6layout
,
1628 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1629 return "Please specify new layout";
1630 l
= info
->new_layout
;
1632 re
->after
.layout
= l
;
1638 /* We must already be at level 6 */
1640 return "Impossible level change";
1641 if (info
->new_layout
== UnSet
)
1642 re
->after
.layout
= info
->array
.layout
;
1644 re
->after
.layout
= info
->new_layout
;
1647 return "Impossible level change requested";
1649 if (info
->delta_disks
== UnSet
)
1650 info
->delta_disks
= delta_parity
;
1652 re
->after
.data_disks
=
1653 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1655 switch (re
->level
) {
1667 /* So we have a restripe operation, we need to calculate the number
1668 * of blocks per reshape operation.
1670 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1671 if (info
->new_chunk
== 0)
1672 info
->new_chunk
= info
->array
.chunk_size
;
1673 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1674 re
->after
.layout
== re
->before
.layout
&&
1675 info
->new_chunk
== info
->array
.chunk_size
) {
1676 /* Nothing to change, can change level immediately. */
1677 re
->level
= info
->new_level
;
1678 re
->backup_blocks
= 0;
1681 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1682 /* chunk and layout changes make no difference */
1683 re
->level
= info
->new_level
;
1684 re
->backup_blocks
= 0;
1688 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1689 get_linux_version() < 2006032)
1690 return "in-place reshape is not safe before 2.6.32 - sorry.";
1692 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1693 get_linux_version() < 2006030)
1694 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1696 re
->backup_blocks
= compute_backup_blocks(
1697 info
->new_chunk
, info
->array
.chunk_size
,
1698 re
->after
.data_disks
, re
->before
.data_disks
);
1699 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1701 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1705 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1708 struct mdinfo
*info
;
1712 if ((st
== NULL
) || (sra
== NULL
))
1715 if (text_version
== NULL
)
1716 text_version
= sra
->text_version
;
1717 subarray
= strchr(text_version
+ 1, '/')+1;
1718 info
= st
->ss
->container_content(st
, subarray
);
1720 unsigned long long current_size
= 0;
1721 unsigned long long new_size
= info
->custom_array_size
/2;
1723 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1724 new_size
> current_size
) {
1725 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1727 dprintf("Error: Cannot set array size");
1730 dprintf("Array size changed");
1732 dprintf_cont(" from %llu to %llu.\n",
1733 current_size
, new_size
);
1737 dprintf("Error: set_array_size(): info pointer in NULL\n");
1742 static int reshape_array(char *container
, int fd
, char *devname
,
1743 struct supertype
*st
, struct mdinfo
*info
,
1744 int force
, struct mddev_dev
*devlist
,
1745 unsigned long long data_offset
,
1746 char *backup_file
, int verbose
, int forked
,
1747 int restart
, int freeze_reshape
);
1748 static int reshape_container(char *container
, char *devname
,
1750 struct supertype
*st
,
1751 struct mdinfo
*info
,
1753 char *backup_file
, int verbose
,
1754 int forked
, int restart
, int freeze_reshape
);
1756 int Grow_reshape(char *devname
, int fd
,
1757 struct mddev_dev
*devlist
,
1758 unsigned long long data_offset
,
1759 struct context
*c
, struct shape
*s
)
1761 /* Make some changes in the shape of an array.
1762 * The kernel must support the change.
1764 * There are three different changes. Each can trigger
1765 * a resync or recovery so we freeze that until we have
1766 * requested everything (if kernel supports freezing - 2.6.30).
1768 * - change size (i.e. component_size)
1770 * - change layout/chunksize/ndisks
1772 * The last can require a reshape. It is different on different
1773 * levels so we need to check the level before actioning it.
1774 * Some times the level change needs to be requested after the
1775 * reshape (e.g. raid6->raid5, raid5->raid0)
1778 struct mdu_array_info_s array
;
1780 struct supertype
*st
;
1781 char *subarray
= NULL
;
1785 char *container
= NULL
;
1788 struct mddev_dev
*dv
;
1794 if (md_get_array_info(fd
, &array
) < 0) {
1795 pr_err("%s is not an active md array - aborting\n",
1799 if (s
->level
!= UnSet
&& s
->chunk
) {
1800 pr_err("Cannot change array level in the same operation as changing chunk size.\n");
1804 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1805 (array
.level
< 4 || array
.level
> 6)) {
1806 pr_err("--grow --data-offset not yet supported\n");
1811 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1812 pr_err("cannot change component size at the same time as other changes.\n"
1813 " Change size first, then check data is intact before making other changes.\n");
1817 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&&
1818 array
.level
> 1 && get_linux_version() < 2006032 &&
1819 !check_env("MDADM_FORCE_FEWER")) {
1820 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1821 " Please use a newer kernel\n");
1825 if (array
.level
> 1 && s
->size
> 1 &&
1826 (unsigned long long) (array
.chunk_size
/ 1024) > s
->size
) {
1827 pr_err("component size must be larger than chunk size.\n");
1831 st
= super_by_fd(fd
, &subarray
);
1833 pr_err("Unable to determine metadata format for %s\n", devname
);
1836 if (s
->raiddisks
> st
->max_devs
) {
1837 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1840 if (s
->level
== 0 &&
1841 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) &&
1842 !(array
.state
& (1<<MD_SB_CLUSTERED
))) {
1843 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
1844 if (md_set_array_info(fd
, &array
)!= 0) {
1845 pr_err("failed to remove internal bitmap.\n");
1850 /* in the external case we need to check that the requested reshape is
1851 * supported, and perform an initial check that the container holds the
1852 * pre-requisite spare devices (mdmon owns final validation)
1854 if (st
->ss
->external
) {
1858 container
= st
->container_devnm
;
1859 cfd
= open_dev_excl(st
->container_devnm
);
1861 container
= st
->devnm
;
1863 cfd
= open_dev_excl(st
->devnm
);
1867 pr_err("Unable to open container for %s\n", devname
);
1872 retval
= st
->ss
->load_container(st
, cfd
, NULL
);
1875 pr_err("Cannot read superblock for %s\n", devname
);
1880 /* check if operation is supported for metadata handler */
1881 if (st
->ss
->container_content
) {
1882 struct mdinfo
*cc
= NULL
;
1883 struct mdinfo
*content
= NULL
;
1885 cc
= st
->ss
->container_content(st
, subarray
);
1886 for (content
= cc
; content
; content
= content
->next
) {
1887 int allow_reshape
= 1;
1889 /* check if reshape is allowed based on metadata
1890 * indications stored in content.array.status
1892 if (content
->array
.state
&
1893 (1 << MD_SB_BLOCK_VOLUME
))
1895 if (content
->array
.state
&
1896 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE
))
1898 if (!allow_reshape
) {
1899 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1900 devname
, container
);
1905 if (content
->consistency_policy
==
1906 CONSISTENCY_POLICY_PPL
) {
1907 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1915 if (mdmon_running(container
))
1916 st
->update_tail
= &st
->updates
;
1920 for (dv
= devlist
; dv
; dv
= dv
->next
)
1922 if (s
->raiddisks
> array
.raid_disks
&&
1923 array
.spare_disks
+ added_disks
<
1924 (s
->raiddisks
- array
.raid_disks
) &&
1926 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1927 " Use --force to over-ride this check.\n",
1928 s
->raiddisks
- array
.raid_disks
,
1929 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1930 array
.spare_disks
+ added_disks
);
1934 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1935 GET_STATE
| GET_VERSION
);
1937 if (st
->ss
->external
&& subarray
== NULL
) {
1938 array
.level
= LEVEL_CONTAINER
;
1939 sra
->array
.level
= LEVEL_CONTAINER
;
1942 pr_err("failed to read sysfs parameters for %s\n",
1946 frozen
= freeze(st
);
1948 /* freeze() already spewed the reason */
1951 } else if (frozen
< 0) {
1952 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1957 /* ========= set size =============== */
1959 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1960 unsigned long long orig_size
= get_component_size(fd
)/2;
1961 unsigned long long min_csize
;
1963 int raid0_takeover
= 0;
1966 orig_size
= (unsigned) array
.size
;
1968 if (orig_size
== 0) {
1969 pr_err("Cannot set device size in this type of array.\n");
1974 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1975 devname
, APPLY_METADATA_CHANGES
,
1981 if (st
->ss
->external
) {
1982 /* metadata can have size limitation
1983 * update size value according to metadata information
1985 struct mdinfo
*sizeinfo
=
1986 st
->ss
->container_content(st
, subarray
);
1988 unsigned long long new_size
=
1989 sizeinfo
->custom_array_size
/2;
1990 int data_disks
= get_data_disks(
1991 sizeinfo
->array
.level
,
1992 sizeinfo
->array
.layout
,
1993 sizeinfo
->array
.raid_disks
);
1994 new_size
/= data_disks
;
1995 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1996 orig_size
, new_size
,
1997 new_size
* data_disks
);
1999 sysfs_free(sizeinfo
);
2003 /* Update the size of each member device in case
2004 * they have been resized. This will never reduce
2005 * below the current used-size. The "size" attribute
2006 * understands '0' to mean 'max'.
2009 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2010 sysfs_set_num(sra
, mdi
, "size",
2011 s
->size
== MAX_SIZE
? 0 : s
->size
);
2012 if (array
.not_persistent
== 0 &&
2013 array
.major_version
== 0 &&
2014 get_linux_version() < 3001000) {
2015 /* Dangerous to allow size to exceed 2TB */
2016 unsigned long long csize
;
2017 if (sysfs_get_ll(sra
, mdi
, "size",
2019 if (csize
>= 2ULL*1024*1024*1024)
2020 csize
= 2ULL*1024*1024*1024;
2021 if ((min_csize
== 0 ||
2022 (min_csize
> csize
)))
2027 if (min_csize
&& s
->size
> min_csize
) {
2028 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2030 goto size_change_error
;
2032 if (min_csize
&& s
->size
== MAX_SIZE
) {
2033 /* Don't let the kernel choose a size - it will get
2036 pr_err("Limited v0.90 array to 2TB per device\n");
2037 s
->size
= min_csize
;
2039 if (st
->ss
->external
) {
2040 if (sra
->array
.level
== 0) {
2041 rv
= sysfs_set_str(sra
, NULL
, "level", "raid5");
2044 /* get array parameters after takeover
2045 * to change one parameter at time only
2047 rv
= md_get_array_info(fd
, &array
);
2050 /* make sure mdmon is
2051 * aware of the new level */
2052 if (!mdmon_running(st
->container_devnm
))
2053 start_mdmon(st
->container_devnm
);
2054 ping_monitor(container
);
2055 if (mdmon_running(st
->container_devnm
) &&
2056 st
->update_tail
== NULL
)
2057 st
->update_tail
= &st
->updates
;
2060 if (s
->size
== MAX_SIZE
)
2062 array
.size
= s
->size
;
2063 if (s
->size
& ~INT32_MAX
) {
2064 /* got truncated to 32bit, write to
2065 * component_size instead
2068 rv
= sysfs_set_num(sra
, NULL
,
2069 "component_size", s
->size
);
2073 rv
= md_set_array_info(fd
, &array
);
2075 /* manage array size when it is managed externally
2077 if ((rv
== 0) && st
->ss
->external
)
2078 rv
= set_array_size(st
, sra
, sra
->text_version
);
2081 if (raid0_takeover
) {
2082 /* do not recync non-existing parity,
2083 * we will drop it anyway
2085 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2086 /* go back to raid0, drop parity disk
2088 sysfs_set_str(sra
, NULL
, "level", "raid0");
2089 md_get_array_info(fd
, &array
);
2096 /* restore metadata */
2097 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2098 UnSet
, NULL
, devname
,
2099 ROLLBACK_METADATA_CHANGES
,
2102 pr_err("Cannot set device size for %s: %s\n",
2103 devname
, strerror(err
));
2105 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2106 cont_err("Bitmap must be removed before size can be changed\n");
2110 if (s
->assume_clean
) {
2111 /* This will fail on kernels older than 3.0 unless
2112 * a backport has been arranged.
2115 sysfs_set_str(sra
, NULL
, "resync_start",
2117 pr_err("--assume-clean not supported with --grow on this kernel\n");
2119 md_get_array_info(fd
, &array
);
2120 s
->size
= get_component_size(fd
)/2;
2122 s
->size
= array
.size
;
2123 if (c
->verbose
>= 0) {
2124 if (s
->size
== orig_size
)
2125 pr_err("component size of %s unchanged at %lluK\n",
2128 pr_err("component size of %s has been set to %lluK\n",
2132 } else if (array
.level
!= LEVEL_CONTAINER
) {
2133 s
->size
= get_component_size(fd
)/2;
2135 s
->size
= array
.size
;
2138 /* See if there is anything else to do */
2139 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2140 (s
->layout_str
== NULL
) &&
2141 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2142 data_offset
== INVALID_SECTORS
&&
2143 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2144 /* Nothing more to do */
2145 if (!changed
&& c
->verbose
>= 0)
2146 pr_err("%s: no change requested\n", devname
);
2150 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2151 * current implementation assumes that following conditions must be met:
2154 * - near_copies == 2
2156 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2157 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2158 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2161 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2163 dprintf("Array cannot be reshaped\n");
2169 /* Make sure mdmon has seen the device removal
2170 * and updated metadata before we continue with
2174 ping_monitor(container
);
2177 memset(&info
, 0, sizeof(info
));
2179 if (sysfs_init(&info
, fd
, NULL
)) {
2180 pr_err("failed to intialize sysfs.\n");
2184 strcpy(info
.text_version
, sra
->text_version
);
2185 info
.component_size
= s
->size
*2;
2186 info
.new_level
= s
->level
;
2187 info
.new_chunk
= s
->chunk
* 1024;
2188 if (info
.array
.level
== LEVEL_CONTAINER
) {
2189 info
.delta_disks
= UnSet
;
2190 info
.array
.raid_disks
= s
->raiddisks
;
2191 } else if (s
->raiddisks
)
2192 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2194 info
.delta_disks
= UnSet
;
2195 if (s
->layout_str
== NULL
) {
2196 info
.new_layout
= UnSet
;
2197 if (info
.array
.level
== 6 &&
2198 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2199 info
.array
.layout
>= 16) {
2200 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2201 cont_err("during the reshape, please specify --layout=preserve\n");
2202 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2206 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2207 strcmp(s
->layout_str
, "normalize") == 0) {
2208 /* If we have a -6 RAID6 layout, remove the '-6'. */
2209 info
.new_layout
= UnSet
;
2210 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2212 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
2213 h
= strrchr(l
, '-');
2214 if (h
&& strcmp(h
, "-6") == 0) {
2216 info
.new_layout
= map_name(r6layout
, l
);
2219 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2223 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2224 /* This means that a non-standard RAID6 layout
2227 * - When reshape a RAID6 (e.g. adding a device)
2228 * which is in a non-standard layout, it is OK
2229 * to preserve that layout.
2230 * - When converting a RAID5 to RAID6, leave it in
2231 * the XXX-6 layout, don't re-layout.
2233 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2234 info
.new_layout
= info
.array
.layout
;
2235 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2237 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
2239 info
.new_layout
= map_name(r6layout
, l
);
2241 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2246 int l
= info
.new_level
;
2248 l
= info
.array
.level
;
2251 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2254 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2257 info
.new_layout
= parse_layout_10(s
->layout_str
);
2260 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2263 pr_err("layout not meaningful with this level\n");
2267 if (info
.new_layout
== UnSet
) {
2268 pr_err("layout %s not understood for this level\n",
2275 if (array
.level
== LEVEL_FAULTY
) {
2276 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2277 pr_err("cannot change level of Faulty device\n");
2281 pr_err("cannot set chunksize of Faulty device\n");
2284 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2285 pr_err("cannot set raid_disks of Faulty device\n");
2288 if (s
->layout_str
) {
2289 if (md_get_array_info(fd
, &array
) != 0) {
2290 dprintf("Cannot get array information.\n");
2293 array
.layout
= info
.new_layout
;
2294 if (md_set_array_info(fd
, &array
) != 0) {
2295 pr_err("failed to set new layout\n");
2297 } else if (c
->verbose
>= 0)
2298 printf("layout for %s set to %d\n",
2299 devname
, array
.layout
);
2301 } else if (array
.level
== LEVEL_CONTAINER
) {
2302 /* This change is to be applied to every array in the
2303 * container. This is only needed when the metadata imposes
2304 * restraints of the various arrays in the container.
2305 * Currently we only know that IMSM requires all arrays
2306 * to have the same number of devices so changing the
2307 * number of devices (On-Line Capacity Expansion) must be
2308 * performed at the level of the container
2314 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2315 c
->force
, c
->backup_file
, c
->verbose
,
2319 /* get spare devices from external metadata
2321 if (st
->ss
->external
) {
2322 struct mdinfo
*info2
;
2324 info2
= st
->ss
->container_content(st
, subarray
);
2326 info
.array
.spare_disks
=
2327 info2
->array
.spare_disks
;
2332 /* Impose these changes on a single array. First
2333 * check that the metadata is OK with the change. */
2335 if (reshape_super(st
, 0, info
.new_level
,
2336 info
.new_layout
, info
.new_chunk
,
2337 info
.array
.raid_disks
, info
.delta_disks
,
2338 c
->backup_file
, devname
,
2339 APPLY_METADATA_CHANGES
, c
->verbose
)) {
2344 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2345 devlist
, data_offset
, c
->backup_file
,
2346 c
->verbose
, 0, 0, 0);
2356 /* verify_reshape_position()
2357 * Function checks if reshape position in metadata is not farther
2358 * than position in md.
2360 * 0 : not valid sysfs entry
2361 * it can be caused by not started reshape, it should be started
2362 * by reshape array or raid0 array is before takeover
2363 * -1 : error, reshape position is obviously wrong
2364 * 1 : success, reshape progress correct or updated
2366 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2372 /* read sync_max, failure can mean raid0 array */
2373 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2377 unsigned long long position
= strtoull(buf
, &ep
, 0);
2379 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2380 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2381 position
*= get_data_disks(level
,
2383 info
->array
.raid_disks
);
2384 if (info
->reshape_progress
< position
) {
2385 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2386 info
->reshape_progress
, position
);
2387 info
->reshape_progress
= position
;
2389 } else if (info
->reshape_progress
> position
) {
2390 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2391 position
, info
->reshape_progress
);
2394 dprintf("Reshape position in md and metadata are the same;");
2398 } else if (rv
== 0) {
2399 /* for valid sysfs entry, 0-length content
2400 * should be indicated as error
2408 static unsigned long long choose_offset(unsigned long long lo
,
2409 unsigned long long hi
,
2410 unsigned long long min
,
2411 unsigned long long max
)
2413 /* Choose a new offset between hi and lo.
2414 * It must be between min and max, but
2415 * we would prefer something near the middle of hi/lo, and also
2416 * prefer to be aligned to a big power of 2.
2418 * So we start with the middle, then for each bit,
2419 * starting at '1' and increasing, if it is set, we either
2420 * add it or subtract it if possible, preferring the option
2421 * which is furthest from the boundary.
2423 * We stop once we get a 1MB alignment. As units are in sectors,
2424 * 1MB = 2*1024 sectors.
2426 unsigned long long choice
= (lo
+ hi
) / 2;
2427 unsigned long long bit
= 1;
2429 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2430 unsigned long long bigger
, smaller
;
2431 if (! (bit
& choice
))
2433 bigger
= choice
+ bit
;
2434 smaller
= choice
- bit
;
2435 if (bigger
> max
&& smaller
< min
)
2439 else if (smaller
< min
)
2441 else if (hi
- bigger
> smaller
- lo
)
2449 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2450 char *devname
, int delta_disks
,
2451 unsigned long long data_offset
,
2452 unsigned long long min
,
2458 unsigned long long before
, after
;
2460 /* Need to find min space before and after so same is used
2463 before
= UINT64_MAX
;
2465 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2469 struct supertype
*st2
;
2470 struct mdinfo info2
;
2472 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2474 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2475 dfd
= dev_open(dn
, O_RDONLY
);
2477 pr_err("%s: cannot open component %s\n",
2478 devname
, dn
? dn
: "-unknown-");
2481 st2
= dup_super(st
);
2482 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2486 pr_err("%s: cannot get superblock from %s\n",
2490 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2491 st2
->ss
->free_super(st2
);
2493 if (info2
.space_before
== 0 &&
2494 info2
.space_after
== 0) {
2495 /* Metadata doesn't support data_offset changes */
2497 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2501 if (before
> info2
.space_before
)
2502 before
= info2
.space_before
;
2503 if (after
> info2
.space_after
)
2504 after
= info2
.space_after
;
2506 if (data_offset
!= INVALID_SECTORS
) {
2508 if (info2
.data_offset
== data_offset
) {
2509 pr_err("%s: already has that data_offset\n",
2513 if (data_offset
< info2
.data_offset
)
2517 } else if ((data_offset
<= info2
.data_offset
&&
2519 (data_offset
>= info2
.data_offset
&&
2521 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2527 if (before
== UINT64_MAX
)
2528 /* impossible really, there must be no devices */
2531 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2532 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2533 unsigned long long new_data_offset
;
2535 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2537 if (delta_disks
< 0) {
2538 /* Don't need any space as array is shrinking
2539 * just move data_offset up by min
2541 if (data_offset
== INVALID_SECTORS
)
2542 new_data_offset
= sd
->data_offset
+ min
;
2544 if (data_offset
< sd
->data_offset
+ min
) {
2545 pr_err("--data-offset too small for %s\n",
2549 new_data_offset
= data_offset
;
2551 } else if (delta_disks
> 0) {
2552 /* need space before */
2556 pr_err("Insufficient head-space for reshape on %s\n",
2560 if (data_offset
== INVALID_SECTORS
)
2561 new_data_offset
= sd
->data_offset
- min
;
2563 if (data_offset
> sd
->data_offset
- min
) {
2564 pr_err("--data-offset too large for %s\n",
2568 new_data_offset
= data_offset
;
2572 /* can move up or down. If 'data_offset'
2573 * was set we would have already decided,
2574 * so just choose direction with most space.
2581 sysfs_set_str(sra
, NULL
, "reshape_direction",
2582 dir
== 1 ? "backwards" : "forwards");
2584 /* Increase data offset */
2588 pr_err("Insufficient tail-space for reshape on %s\n",
2592 if (data_offset
!= INVALID_SECTORS
&&
2593 data_offset
< sd
->data_offset
+ min
) {
2594 pr_err("--data-offset too small on %s\n",
2598 if (data_offset
!= INVALID_SECTORS
)
2599 new_data_offset
= data_offset
;
2601 new_data_offset
= choose_offset(sd
->data_offset
,
2602 sd
->data_offset
+ after
,
2603 sd
->data_offset
+ min
,
2604 sd
->data_offset
+ after
);
2606 /* Decrease data offset */
2610 pr_err("insufficient head-room on %s\n",
2614 if (data_offset
!= INVALID_SECTORS
&&
2615 data_offset
< sd
->data_offset
- min
) {
2616 pr_err("--data-offset too small on %s\n",
2620 if (data_offset
!= INVALID_SECTORS
)
2621 new_data_offset
= data_offset
;
2623 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2625 sd
->data_offset
- before
,
2626 sd
->data_offset
- min
);
2629 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2630 if (err
< 0 && errno
== E2BIG
) {
2631 /* try again after increasing data size to max */
2632 err
= sysfs_set_num(sra
, sd
, "size", 0);
2633 if (err
< 0 && errno
== EINVAL
&&
2634 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2635 /* some kernels have a bug where you cannot
2636 * use '0' on spare devices. */
2637 sysfs_set_num(sra
, sd
, "size",
2638 (sra
->component_size
+ after
)/2);
2640 err
= sysfs_set_num(sra
, sd
, "new_offset",
2644 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2645 pr_err("data-offset is too big for %s\n", dn
);
2648 if (sd
== sra
->devs
&&
2649 (errno
== ENOENT
|| errno
== E2BIG
))
2650 /* Early kernel, no 'new_offset' file,
2651 * or kernel doesn't like us.
2652 * For RAID5/6 this is not fatal
2655 pr_err("Cannot set new_offset for %s\n", dn
);
2663 /* Just use a backup file */
2667 static int raid10_reshape(char *container
, int fd
, char *devname
,
2668 struct supertype
*st
, struct mdinfo
*info
,
2669 struct reshape
*reshape
,
2670 unsigned long long data_offset
,
2671 int force
, int verbose
)
2673 /* Changing raid_disks, layout, chunksize or possibly
2674 * just data_offset for a RAID10.
2675 * We must always change data_offset. We change by at least
2676 * ->min_offset_change which is the largest of the old and new
2678 * If raid_disks is increasing, then data_offset must decrease
2679 * by at least this copy size.
2680 * If raid_disks is unchanged, data_offset must increase or
2681 * decrease by at least min_offset_change but preferably by much more.
2682 * We choose half of the available space.
2683 * If raid_disks is decreasing, data_offset must increase by
2684 * at least min_offset_change. To allow of this, component_size
2685 * must be decreased by the same amount.
2687 * So we calculate the required minimum and direction, possibly
2688 * reduce the component_size, then iterate through the devices
2689 * and set the new_data_offset.
2690 * If that all works, we set chunk_size, layout, raid_disks, and start
2694 unsigned long long min
;
2697 sra
= sysfs_read(fd
, NULL
,
2698 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2701 pr_err("%s: Cannot get array details from sysfs\n", devname
);
2704 min
= reshape
->min_offset_change
;
2706 if (info
->delta_disks
)
2707 sysfs_set_str(sra
, NULL
, "reshape_direction",
2708 info
->delta_disks
< 0 ? "backwards" : "forwards");
2709 if (info
->delta_disks
< 0 && info
->space_after
< min
) {
2710 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2711 (sra
->component_size
- min
)/2);
2713 pr_err("cannot reduce component size\n");
2717 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
,
2718 data_offset
, min
, 0);
2720 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2721 cont_err("supported on this kernel\n");
2727 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2729 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2730 reshape
->after
.layout
) < 0)
2733 sysfs_set_num(sra
, NULL
, "raid_disks",
2734 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2736 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2739 pr_err("Cannot set array shape for %s\n",
2742 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2743 cont_err(" Bitmap must be removed before shape can be changed\n");
2753 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2755 struct mdinfo
*sra
, *sd
;
2756 /* Initialisation to silence compiler warning */
2757 unsigned long long min_space_before
= 0, min_space_after
= 0;
2760 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2763 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2766 struct supertype
*st2
;
2767 struct mdinfo info2
;
2769 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2771 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2772 dfd
= dev_open(dn
, O_RDONLY
);
2775 st2
= dup_super(st
);
2776 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2782 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2783 st2
->ss
->free_super(st2
);
2786 min_space_before
> info2
.space_before
)
2787 min_space_before
= info2
.space_before
;
2789 min_space_after
> info2
.space_after
)
2790 min_space_after
= info2
.space_after
;
2793 if (sd
== NULL
&& !first
) {
2794 info
->space_after
= min_space_after
;
2795 info
->space_before
= min_space_before
;
2800 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2801 struct mdinfo
*info
,
2802 int disks
, unsigned long long blocks
)
2804 /* Check that the internal stripe cache is
2805 * large enough, or it won't work.
2806 * It must hold at least 4 stripes of the larger
2809 unsigned long cache
;
2810 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2811 cache
*= 4; /* 4 stripes minimum */
2812 cache
/= 512; /* convert to sectors */
2813 /* make sure there is room for 'blocks' with a bit to spare */
2814 if (cache
< 16 + blocks
/ disks
)
2815 cache
= 16 + blocks
/ disks
;
2816 cache
/= (4096/512); /* Convert from sectors to pages */
2818 if (sra
->cache_size
< cache
)
2819 subarray_set_num(container
, sra
, "stripe_cache_size",
2823 static int impose_reshape(struct mdinfo
*sra
,
2824 struct mdinfo
*info
,
2825 struct supertype
*st
,
2828 char *devname
, char *container
,
2829 struct reshape
*reshape
)
2831 struct mdu_array_info_s array
;
2833 sra
->new_chunk
= info
->new_chunk
;
2836 /* for external metadata checkpoint saved by mdmon can be lost
2837 * or missed /due to e.g. crash/. Check if md is not during
2838 * restart farther than metadata points to.
2839 * If so, this means metadata information is obsolete.
2841 if (st
->ss
->external
)
2842 verify_reshape_position(info
, reshape
->level
);
2843 sra
->reshape_progress
= info
->reshape_progress
;
2845 sra
->reshape_progress
= 0;
2846 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2847 /* start from the end of the new array */
2848 sra
->reshape_progress
= (sra
->component_size
2849 * reshape
->after
.data_disks
);
2852 md_get_array_info(fd
, &array
);
2853 if (info
->array
.chunk_size
== info
->new_chunk
&&
2854 reshape
->before
.layout
== reshape
->after
.layout
&&
2855 st
->ss
->external
== 0) {
2856 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2857 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2858 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2861 pr_err("Cannot set device shape for %s: %s\n",
2862 devname
, strerror(errno
));
2865 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2866 cont_err("Bitmap must be removed before shape can be changed\n");
2870 } else if (!restart
) {
2871 /* set them all just in case some old 'new_*' value
2872 * persists from some earlier problem.
2875 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2877 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2878 reshape
->after
.layout
) < 0)
2880 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2881 reshape
->after
.data_disks
+
2882 reshape
->parity
) < 0)
2885 pr_err("Cannot set device shape for %s\n", devname
);
2888 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2889 cont_err("Bitmap must be removed before shape can be changed\n");
2898 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2901 struct mdu_array_info_s array
;
2904 if (sysfs_init(&info
, fd
, NULL
)) {
2905 pr_err("failed to intialize sysfs.\n");
2909 md_get_array_info(fd
, &array
);
2910 if (level
== 0 && (array
.level
>= 4 && array
.level
<= 6)) {
2911 /* To convert to RAID0 we need to fail and
2912 * remove any non-data devices. */
2915 int data_disks
= array
.raid_disks
- 1;
2916 if (array
.level
== 6)
2918 if (array
.level
== 5 && array
.layout
!= ALGORITHM_PARITY_N
)
2920 if (array
.level
== 6 && array
.layout
!= ALGORITHM_PARITY_N_6
)
2922 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2923 /* First remove any spares so no recovery starts */
2924 for (d
= 0, found
= 0;
2925 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2926 mdu_disk_info_t disk
;
2928 if (md_get_disk_info(fd
, &disk
) < 0)
2930 if (disk
.major
== 0 && disk
.minor
== 0)
2933 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2934 disk
.raid_disk
< data_disks
)
2937 ioctl(fd
, HOT_REMOVE_DISK
,
2938 makedev(disk
.major
, disk
.minor
));
2940 /* Now fail anything left */
2941 md_get_array_info(fd
, &array
);
2942 for (d
= 0, found
= 0;
2943 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2944 mdu_disk_info_t disk
;
2946 if (md_get_disk_info(fd
, &disk
) < 0)
2948 if (disk
.major
== 0 && disk
.minor
== 0)
2951 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2952 disk
.raid_disk
< data_disks
)
2955 ioctl(fd
, SET_DISK_FAULTY
,
2956 makedev(disk
.major
, disk
.minor
));
2957 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2960 c
= map_num(pers
, level
);
2962 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2965 pr_err("%s: could not set level to %s\n",
2968 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2969 cont_err("Bitmap must be removed before level can be changed\n");
2973 pr_err("level of %s changed to %s\n", devname
, c
);
2979 static void catch_term(int sig
)
2984 static int continue_via_systemd(char *devnm
)
2986 int skipped
, i
, pid
, status
;
2988 /* In a systemd/udev world, it is best to get systemd to
2989 * run "mdadm --grow --continue" rather than running in the
2994 /* FIXME yuk. CLOSE_EXEC?? */
2996 for (i
= 3; skipped
< 20; i
++)
3002 /* Don't want to see error messages from
3003 * systemctl. If the service doesn't exist,
3004 * we fork ourselves.
3007 open("/dev/null", O_WRONLY
);
3008 snprintf(pathbuf
, sizeof(pathbuf
),
3009 "mdadm-grow-continue@%s.service", devnm
);
3010 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
3012 status
= execl("/bin/systemctl", "systemctl", "restart",
3015 case -1: /* Just do it ourselves. */
3017 default: /* parent - good */
3018 pid
= wait(&status
);
3019 if (pid
>= 0 && status
== 0)
3025 static int reshape_array(char *container
, int fd
, char *devname
,
3026 struct supertype
*st
, struct mdinfo
*info
,
3027 int force
, struct mddev_dev
*devlist
,
3028 unsigned long long data_offset
,
3029 char *backup_file
, int verbose
, int forked
,
3030 int restart
, int freeze_reshape
)
3032 struct reshape reshape
;
3035 int orig_level
= UnSet
;
3039 struct mdu_array_info_s array
;
3042 struct mddev_dev
*dv
;
3046 unsigned long long *offsets
= NULL
;
3050 unsigned long blocks
;
3051 unsigned long long array_size
;
3053 struct mdinfo
*sra
= NULL
;
3056 /* when reshaping a RAID0, the component_size might be zero.
3057 * So try to fix that up.
3059 if (md_get_array_info(fd
, &array
) != 0) {
3060 dprintf("Cannot get array information.\n");
3063 if (array
.level
== 0 && info
->component_size
== 0) {
3064 get_dev_size(fd
, NULL
, &array_size
);
3065 info
->component_size
= array_size
/ array
.raid_disks
;
3068 if (array
.level
== 10)
3069 /* Need space_after info */
3070 get_space_after(fd
, st
, info
);
3072 if (info
->reshape_active
) {
3073 int new_level
= info
->new_level
;
3074 info
->new_level
= UnSet
;
3075 if (info
->delta_disks
> 0)
3076 info
->array
.raid_disks
-= info
->delta_disks
;
3077 msg
= analyse_change(devname
, info
, &reshape
);
3078 info
->new_level
= new_level
;
3079 if (info
->delta_disks
> 0)
3080 info
->array
.raid_disks
+= info
->delta_disks
;
3082 /* Make sure the array isn't read-only */
3083 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3085 msg
= analyse_change(devname
, info
, &reshape
);
3087 /* if msg == "", error has already been printed */
3089 pr_err("%s\n", msg
);
3092 if (restart
&& (reshape
.level
!= info
->array
.level
||
3093 reshape
.before
.layout
!= info
->array
.layout
||
3094 reshape
.before
.data_disks
+ reshape
.parity
!=
3095 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3096 pr_err("reshape info is not in native format - cannot continue.\n");
3100 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3101 !((sysfs_get_str(info
, NULL
, "sync_action",
3102 buf
, sizeof(buf
)) > 0) &&
3103 (strncmp(buf
, "reshape", 7) == 0))) {
3104 /* When reshape is restarted from '0', very begin of array
3105 * it is possible that for external metadata reshape and array
3106 * configuration doesn't happen.
3107 * Check if md has the same opinion, and reshape is restarted
3108 * from 0. If so, this is regular reshape start after reshape
3109 * switch in metadata to next array only.
3111 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3112 (info
->reshape_progress
== 0))
3117 * reshape already started. just skip to monitoring
3120 if (reshape
.backup_blocks
== 0)
3122 if (restart
& RESHAPE_NO_BACKUP
)
3125 /* Need 'sra' down at 'started:' */
3126 sra
= sysfs_read(fd
, NULL
,
3127 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
3128 GET_CHUNK
|GET_CACHE
);
3130 pr_err("%s: Cannot get array details from sysfs\n",
3136 backup_file
= locate_backup(sra
->sys_name
);
3140 /* The container is frozen but the array may not be.
3141 * So freeze the array so spares don't get put to the wrong use
3142 * FIXME there should probably be a cleaner separation between
3143 * freeze_array and freeze_container.
3145 sysfs_freeze_array(info
);
3146 /* Check we have enough spares to not be degraded */
3148 for (dv
= devlist
; dv
; dv
=dv
->next
)
3150 spares_needed
= max(reshape
.before
.data_disks
,
3151 reshape
.after
.data_disks
) +
3152 reshape
.parity
- array
.raid_disks
;
3154 if (!force
&& info
->new_level
> 1 && info
->array
.level
> 1 &&
3155 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3156 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3157 " Use --force to over-ride this check.\n",
3159 spares_needed
== 1 ? "" : "s",
3160 info
->array
.spare_disks
+ added_disks
);
3163 /* Check we have enough spares to not fail */
3164 spares_needed
= max(reshape
.before
.data_disks
,
3165 reshape
.after
.data_disks
)
3167 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3168 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3169 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3170 spares_needed
, spares_needed
== 1 ? "" : "s",
3171 info
->array
.spare_disks
+ added_disks
);
3175 if (reshape
.level
!= array
.level
) {
3176 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3179 info
->new_layout
= UnSet
; /* after level change,
3180 * layout is meaningless */
3181 orig_level
= array
.level
;
3182 sysfs_freeze_array(info
);
3184 if (reshape
.level
> 0 && st
->ss
->external
) {
3185 /* make sure mdmon is aware of the new level */
3186 if (mdmon_running(container
))
3187 flush_mdmon(container
);
3189 if (!mdmon_running(container
))
3190 start_mdmon(container
);
3191 ping_monitor(container
);
3192 if (mdmon_running(container
) && st
->update_tail
== NULL
)
3193 st
->update_tail
= &st
->updates
;
3196 /* ->reshape_super might have chosen some spares from the
3197 * container that it wants to be part of the new array.
3198 * We can collect them with ->container_content and give
3199 * them to the kernel.
3201 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3202 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3203 struct mdinfo
*info2
=
3204 st
->ss
->container_content(st
, subarray
);
3208 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3209 pr_err("unable to initialize sysfs for %s\n",
3214 /* When increasing number of devices, we need to set
3215 * new raid_disks before adding these, or they might
3218 if (reshape
.backup_blocks
&&
3219 reshape
.after
.data_disks
>
3220 reshape
.before
.data_disks
)
3221 subarray_set_num(container
, info2
, "raid_disks",
3222 reshape
.after
.data_disks
+
3224 for (d
= info2
->devs
; d
; d
= d
->next
) {
3225 if (d
->disk
.state
== 0 &&
3226 d
->disk
.raid_disk
>= 0) {
3227 /* This is a spare that wants to
3228 * be part of the array.
3230 add_disk(fd
, st
, info2
, d
);
3236 /* We might have been given some devices to add to the
3237 * array. Now that the array has been changed to the right
3238 * level and frozen, we can safely add them.
3241 if (Manage_subdevs(devname
, fd
, devlist
, verbose
, 0, NULL
, 0))
3245 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3246 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3247 if (reshape
.backup_blocks
== 0) {
3248 /* No restriping needed, but we might need to impose
3249 * some more changes: layout, raid_disks, chunk_size
3251 /* read current array info */
3252 if (md_get_array_info(fd
, &array
) != 0) {
3253 dprintf("Cannot get array information.\n");
3256 /* compare current array info with new values and if
3257 * it is different update them to new */
3258 if (info
->new_layout
!= UnSet
&&
3259 info
->new_layout
!= array
.layout
) {
3260 array
.layout
= info
->new_layout
;
3261 if (md_set_array_info(fd
, &array
) != 0) {
3262 pr_err("failed to set new layout\n");
3264 } else if (verbose
>= 0)
3265 printf("layout for %s set to %d\n",
3266 devname
, array
.layout
);
3268 if (info
->delta_disks
!= UnSet
&& info
->delta_disks
!= 0 &&
3270 (info
->array
.raid_disks
+ info
->delta_disks
)) {
3271 array
.raid_disks
+= info
->delta_disks
;
3272 if (md_set_array_info(fd
, &array
) != 0) {
3273 pr_err("failed to set raid disks\n");
3275 } else if (verbose
>= 0) {
3276 printf("raid_disks for %s set to %d\n",
3277 devname
, array
.raid_disks
);
3280 if (info
->new_chunk
!= 0 &&
3281 info
->new_chunk
!= array
.chunk_size
) {
3282 if (sysfs_set_num(info
, NULL
,
3283 "chunk_size", info
->new_chunk
) != 0) {
3284 pr_err("failed to set chunk size\n");
3286 } else if (verbose
>= 0)
3287 printf("chunk size for %s set to %d\n",
3288 devname
, array
.chunk_size
);
3295 * There are three possibilities.
3296 * 1/ The array will shrink.
3297 * We need to ensure the reshape will pause before reaching
3298 * the 'critical section'. We also need to fork and wait for
3299 * that to happen. When it does we
3300 * suspend/backup/complete/unfreeze
3302 * 2/ The array will not change size.
3303 * This requires that we keep a backup of a sliding window
3304 * so that we can restore data after a crash. So we need
3305 * to fork and monitor progress.
3306 * In future we will allow the data_offset to change, so
3307 * a sliding backup becomes unnecessary.
3309 * 3/ The array will grow. This is relatively easy.
3310 * However the kernel's restripe routines will cheerfully
3311 * overwrite some early data before it is safe. So we
3312 * need to make a backup of the early parts of the array
3313 * and be ready to restore it if rebuild aborts very early.
3314 * For externally managed metadata, we still need a forked
3315 * child to monitor the reshape and suspend IO over the region
3316 * that is being reshaped.
3318 * We backup data by writing it to one spare, or to a
3319 * file which was given on command line.
3321 * In each case, we first make sure that storage is available
3322 * for the required backup.
3324 * - request the shape change.
3325 * - fork to handle backup etc.
3327 /* Check that we can hold all the data */
3328 get_dev_size(fd
, NULL
, &array_size
);
3329 if (reshape
.new_size
< (array_size
/512)) {
3330 pr_err("this change will reduce the size of the array.\n"
3331 " use --grow --array-size first to truncate array.\n"
3332 " e.g. mdadm --grow %s --array-size %llu\n",
3333 devname
, reshape
.new_size
/2);
3337 if (array
.level
== 10) {
3338 /* Reshaping RAID10 does not require any data backup by
3339 * user-space. Instead it requires that the data_offset
3340 * is changed to avoid the need for backup.
3341 * So this is handled very separately
3344 /* Nothing to do. */
3346 return raid10_reshape(container
, fd
, devname
, st
, info
,
3347 &reshape
, data_offset
, force
, verbose
);
3349 sra
= sysfs_read(fd
, NULL
,
3350 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3353 pr_err("%s: Cannot get array details from sysfs\n",
3359 switch(set_new_data_offset(sra
, st
, devname
,
3360 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3362 reshape
.min_offset_change
, 1)) {
3366 /* Updated data_offset, so it's easy now */
3367 update_cache_size(container
, sra
, info
,
3368 min(reshape
.before
.data_disks
,
3369 reshape
.after
.data_disks
),
3370 reshape
.backup_blocks
);
3372 /* Right, everything seems fine. Let's kick things off.
3376 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3377 devname
, container
, &reshape
) < 0)
3379 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3381 if (errno
!= EINVAL
) {
3382 pr_err("Failed to initiate reshape!\n");
3385 /* revert data_offset and try the old way */
3386 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3387 sysfs_set_num(sra
, sd
, "new_offset",
3389 sysfs_set_str(sra
, NULL
, "reshape_direction",
3394 if (info
->new_level
== reshape
.level
)
3396 /* need to adjust level when reshape completes */
3398 case -1: /* ignore error, but don't wait */
3400 default: /* parent */
3408 fd
= open_dev(sra
->sys_name
);
3410 impose_level(fd
, info
->new_level
, devname
, verbose
);
3412 case 1: /* Couldn't set data_offset, try the old way */
3413 if (data_offset
!= INVALID_SECTORS
) {
3414 pr_err("Cannot update data_offset on this array\n");
3421 /* Decide how many blocks (sectors) for a reshape
3422 * unit. The number we have so far is just a minimum
3424 blocks
= reshape
.backup_blocks
;
3425 if (reshape
.before
.data_disks
==
3426 reshape
.after
.data_disks
) {
3427 /* Make 'blocks' bigger for better throughput, but
3428 * not so big that we reject it below.
3429 * Try for 16 megabytes
3431 while (blocks
* 32 < sra
->component_size
&& blocks
< 16*1024*2)
3434 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3436 if (blocks
>= sra
->component_size
/2) {
3437 pr_err("%s: Something wrong - reshape aborted\n", devname
);
3441 /* Now we need to open all these devices so we can read/write.
3443 nrdisks
= max(reshape
.before
.data_disks
,
3444 reshape
.after
.data_disks
) + reshape
.parity
3445 + sra
->array
.spare_disks
;
3446 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3447 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3449 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3450 d
= reshape_prepare_fdlist(devname
, sra
, odisks
, nrdisks
, blocks
,
3451 backup_file
, fdlist
, offsets
);
3455 if ((st
->ss
->manage_reshape
== NULL
) ||
3456 (st
->ss
->recover_backup
== NULL
)) {
3457 if (backup_file
== NULL
) {
3458 if (reshape
.after
.data_disks
<=
3459 reshape
.before
.data_disks
) {
3460 pr_err("%s: Cannot grow - need backup-file\n",
3462 pr_err(" Please provide one with \"--backup=...\"\n");
3464 } else if (d
== odisks
) {
3465 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3469 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3471 fdlist
+d
, offsets
+d
,
3472 sra
->sys_name
, restart
)) {
3479 update_cache_size(container
, sra
, info
,
3480 min(reshape
.before
.data_disks
,
3481 reshape
.after
.data_disks
), blocks
);
3483 /* Right, everything seems fine. Let's kick things off.
3484 * If only changing raid_disks, use ioctl, else use
3489 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3490 devname
, container
, &reshape
) < 0)
3493 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3494 reshape
.after
.data_disks
);
3496 pr_err("Cannot %s reshape for %s\n",
3497 restart
? "continue" : "start", devname
);
3501 sysfs_set_str(sra
, NULL
, "array_state", "active");
3502 if (freeze_reshape
) {
3506 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3507 sra
->reshape_progress
);
3511 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3512 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3519 /* Now we just need to kick off the reshape and watch, while
3520 * handling backups of the data...
3521 * This is all done by a forked background process.
3523 switch(forked
? 0 : fork()) {
3525 pr_err("Cannot run child to monitor reshape: %s\n",
3539 /* If another array on the same devices is busy, the
3540 * reshape will wait for them. This would mean that
3541 * the first section that we suspend will stay suspended
3542 * for a long time. So check on that possibility
3543 * by looking for "DELAYED" in /proc/mdstat, and if found,
3547 struct mdstat_ent
*mds
, *m
;
3549 mds
= mdstat_read(1, 0);
3550 for (m
= mds
; m
; m
= m
->next
)
3551 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3552 if (m
->resync
&& m
->percent
== RESYNC_DELAYED
)
3555 /* Haven't started the reshape thread
3562 if (delayed
== 1 && get_linux_version() < 3007000) {
3563 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3564 " You might experience problems until other reshapes complete.\n");
3568 mdstat_wait(30 - (delayed
-1) * 25);
3572 if (check_env("MDADM_GROW_VERIFY"))
3573 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3576 mlockall(MCL_FUTURE
);
3578 signal(SIGTERM
, catch_term
);
3580 if (st
->ss
->external
) {
3581 /* metadata handler takes it from here */
3582 done
= st
->ss
->manage_reshape(
3583 fd
, sra
, &reshape
, st
, blocks
,
3584 fdlist
, offsets
, d
- odisks
, fdlist
+ odisks
,
3587 done
= child_monitor(
3588 fd
, sra
, &reshape
, st
, blocks
, fdlist
, offsets
,
3589 d
- odisks
, fdlist
+ odisks
, offsets
+ odisks
);
3594 if (backup_file
&& done
) {
3596 bul
= make_backup(sra
->sys_name
);
3599 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3607 unlink(backup_file
);
3614 if (!st
->ss
->external
&&
3615 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3616 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3618 /* no need to wait for the reshape to finish as
3619 * there is nothing more to do.
3626 if (st
->ss
->external
) {
3627 /* Re-load the metadata as much could have changed */
3628 int cfd
= open_dev(st
->container_devnm
);
3630 flush_mdmon(container
);
3631 st
->ss
->free_super(st
);
3632 st
->ss
->load_container(st
, cfd
, container
);
3637 /* set new array size if required customer_array_size is used
3640 if (reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3641 info
->custom_array_size
)
3642 set_array_size(st
, info
, info
->text_version
);
3644 if (info
->new_level
!= reshape
.level
) {
3646 fd
= open(devname
, O_RDONLY
);
3647 impose_level(fd
, info
->new_level
, devname
, verbose
);
3649 if (info
->new_level
== 0)
3650 st
->update_tail
= NULL
;
3662 if (orig_level
!= UnSet
&& sra
) {
3663 c
= map_num(pers
, orig_level
);
3664 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3665 pr_err("aborting level change\n");
3673 /* mdfd handle is passed to be closed in child process (after fork).
3675 int reshape_container(char *container
, char *devname
,
3677 struct supertype
*st
,
3678 struct mdinfo
*info
,
3680 char *backup_file
, int verbose
,
3681 int forked
, int restart
, int freeze_reshape
)
3683 struct mdinfo
*cc
= NULL
;
3685 char last_devnm
[32] = "";
3687 /* component_size is not meaningful for a container,
3688 * so pass '0' meaning 'no change'
3691 reshape_super(st
, 0, info
->new_level
,
3692 info
->new_layout
, info
->new_chunk
,
3693 info
->array
.raid_disks
, info
->delta_disks
,
3694 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3702 /* ping monitor to be sure that update is on disk
3704 ping_monitor(container
);
3706 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3707 if (continue_via_systemd(container
))
3710 switch (forked
? 0 : fork()) {
3711 case -1: /* error */
3712 perror("Cannot fork to complete reshape\n");
3715 default: /* parent */
3716 if (!freeze_reshape
)
3717 printf("%s: multi-array reshape continues in background\n", Name
);
3724 /* close unused handle in child process
3730 /* For each member array with reshape_active,
3731 * we need to perform the reshape.
3732 * We pick the first array that needs reshaping and
3733 * reshape it. reshape_array() will re-read the metadata
3734 * so the next time through a different array should be
3735 * ready for reshape.
3736 * It is possible that the 'different' array will not
3737 * be assembled yet. In that case we simple exit.
3738 * When it is assembled, the mdadm which assembles it
3739 * will take over the reshape.
3741 struct mdinfo
*content
;
3743 struct mdstat_ent
*mdstat
;
3749 cc
= st
->ss
->container_content(st
, NULL
);
3751 for (content
= cc
; content
; content
= content
->next
) {
3753 if (!content
->reshape_active
)
3756 subarray
= strchr(content
->text_version
+1, '/')+1;
3757 mdstat
= mdstat_by_subdev(subarray
, container
);
3760 if (mdstat
->active
== 0) {
3761 pr_err("Skipping inactive array %s.\n",
3763 free_mdstat(mdstat
);
3772 devid
= devnm2devid(mdstat
->devnm
);
3773 adev
= map_dev(major(devid
), minor(devid
), 0);
3775 adev
= content
->text_version
;
3777 fd
= open_dev(mdstat
->devnm
);
3779 pr_err("Device %s cannot be opened for reshape.\n",
3784 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3785 /* Do not allow for multiple reshape_array() calls for
3787 * It can happen when reshape_array() returns without
3788 * error, when reshape is not finished (wrong reshape
3789 * starting/continuation conditions). Mdmon doesn't
3790 * switch to next array in container and reentry
3791 * conditions for the same array occur.
3792 * This is possibly interim until the behaviour of
3793 * reshape_array is resolved().
3795 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3799 strcpy(last_devnm
, mdstat
->devnm
);
3801 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3802 pr_err("Unable to initialize sysfs for %s\n",
3808 if (mdmon_running(container
))
3809 flush_mdmon(container
);
3811 rv
= reshape_array(container
, fd
, adev
, st
,
3812 content
, force
, NULL
, INVALID_SECTORS
,
3813 backup_file
, verbose
, 1, restart
,
3817 if (freeze_reshape
) {
3826 if (mdmon_running(container
))
3827 flush_mdmon(container
);
3836 * We run a child process in the background which performs the following
3838 * - wait for resync to reach a certain point
3839 * - suspend io to the following section
3840 * - backup that section
3841 * - allow resync to proceed further
3843 * - discard the backup.
3845 * When are combined in slightly different ways in the three cases.
3847 * - suspend/backup/allow/wait/resume/discard
3849 * - allow/wait/suspend/backup/allow/wait/resume/discard
3851 * - wait/resume/discard/suspend/backup/allow
3853 * suspend/backup/allow always come together
3854 * wait/resume/discard do too.
3855 * For the same-size case we have two backups to improve flow.
3859 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3860 unsigned long long backup_point
,
3861 unsigned long long wait_point
,
3862 unsigned long long *suspend_point
,
3863 unsigned long long *reshape_completed
, int *frozen
)
3865 /* This function is called repeatedly by the reshape manager.
3866 * It determines how much progress can safely be made and allows
3868 * - 'info' identifies the array and particularly records in
3869 * ->reshape_progress the metadata's knowledge of progress
3870 * This is a sector offset from the start of the array
3871 * of the next array block to be relocated. This number
3872 * may increase from 0 or decrease from array_size, depending
3873 * on the type of reshape that is happening.
3874 * Note that in contrast, 'sync_completed' is a block count of the
3875 * reshape so far. It gives the distance between the start point
3876 * (head or tail of device) and the next place that data will be
3877 * written. It always increases.
3878 * - 'reshape' is the structure created by analyse_change
3879 * - 'backup_point' shows how much the metadata manager has backed-up
3880 * data. For reshapes with increasing progress, it is the next address
3881 * to be backed up, previous addresses have been backed-up. For
3882 * decreasing progress, it is the earliest address that has been
3883 * backed up - later address are also backed up.
3884 * So addresses between reshape_progress and backup_point are
3885 * backed up providing those are in the 'correct' order.
3886 * - 'wait_point' is an array address. When reshape_completed
3887 * passes this point, progress_reshape should return. It might
3888 * return earlier if it determines that ->reshape_progress needs
3889 * to be updated or further backup is needed.
3890 * - suspend_point is maintained by progress_reshape and the caller
3891 * should not touch it except to initialise to zero.
3892 * It is an array address and it only increases in 2.6.37 and earlier.
3893 * This makes it difficult to handle reducing reshapes with
3894 * external metadata.
3895 * However: it is similar to backup_point in that it records the
3896 * other end of a suspended region from reshape_progress.
3897 * it is moved to extend the region that is safe to backup and/or
3899 * - reshape_completed is read from sysfs and returned. The caller
3900 * should copy this into ->reshape_progress when it has reason to
3901 * believe that the metadata knows this, and any backup outside this
3905 * 1 if more data from backup_point - but only as far as suspend_point,
3906 * should be backed up
3907 * 0 if things are progressing smoothly
3908 * -1 if the reshape is finished because it is all done,
3909 * -2 if the reshape is finished due to an error.
3912 int advancing
= (reshape
->after
.data_disks
3913 >= reshape
->before
.data_disks
);
3914 unsigned long long need_backup
; /* All data between start of array and
3915 * here will at some point need to
3918 unsigned long long read_offset
, write_offset
;
3919 unsigned long long write_range
;
3920 unsigned long long max_progress
, target
, completed
;
3921 unsigned long long array_size
= (info
->component_size
3922 * reshape
->before
.data_disks
);
3926 /* First, we unsuspend any region that is now known to be safe.
3927 * If suspend_point is on the 'wrong' side of reshape_progress, then
3928 * we don't have or need suspension at the moment. This is true for
3929 * native metadata when we don't need to back-up.
3932 if (info
->reshape_progress
<= *suspend_point
)
3933 sysfs_set_num(info
, NULL
, "suspend_lo",
3934 info
->reshape_progress
);
3936 /* Note: this won't work in 2.6.37 and before.
3937 * Something somewhere should make sure we don't need it!
3939 if (info
->reshape_progress
>= *suspend_point
)
3940 sysfs_set_num(info
, NULL
, "suspend_hi",
3941 info
->reshape_progress
);
3944 /* Now work out how far it is safe to progress.
3945 * If the read_offset for ->reshape_progress is less than
3946 * 'blocks' beyond the write_offset, we can only progress as far
3948 * Otherwise we can progress until the write_offset for the new location
3949 * reaches (within 'blocks' of) the read_offset at the current location.
3950 * However that region must be suspended unless we are using native
3952 * If we need to suspend more, we limit it to 128M per device, which is
3953 * rather arbitrary and should be some time-based calculation.
3955 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3956 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3957 write_range
= info
->new_chunk
/512;
3958 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3959 need_backup
= array_size
;
3961 need_backup
= reshape
->backup_blocks
;
3963 if (read_offset
< write_offset
+ write_range
)
3964 max_progress
= backup_point
;
3967 read_offset
* reshape
->after
.data_disks
;
3969 if (read_offset
> write_offset
- write_range
)
3970 /* Can only progress as far as has been backed up,
3971 * which must be suspended */
3972 max_progress
= backup_point
;
3973 else if (info
->reshape_progress
<= need_backup
)
3974 max_progress
= backup_point
;
3976 if (info
->array
.major_version
>= 0)
3977 /* Can progress until backup is needed */
3978 max_progress
= need_backup
;
3980 /* Can progress until metadata update is required */
3982 read_offset
* reshape
->after
.data_disks
;
3983 /* but data must be suspended */
3984 if (max_progress
< *suspend_point
)
3985 max_progress
= *suspend_point
;
3990 /* We know it is safe to progress to 'max_progress' providing
3991 * it is suspended or we are using native metadata.
3992 * Consider extending suspend_point 128M per device if it
3993 * is less than 64M per device beyond reshape_progress.
3994 * But always do a multiple of 'blocks'
3995 * FIXME this is too big - it takes to long to complete
3998 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
3999 reshape
->after
.data_disks
);
4000 target
/= reshape
->backup_blocks
;
4003 target
*= reshape
->backup_blocks
;
4005 /* For externally managed metadata we always need to suspend IO to
4006 * the area being reshaped so we regularly push suspend_point forward.
4007 * For native metadata we only need the suspend if we are going to do
4011 if ((need_backup
> info
->reshape_progress
||
4012 info
->array
.major_version
< 0) &&
4013 *suspend_point
< info
->reshape_progress
+ target
) {
4014 if (need_backup
< *suspend_point
+ 2 * target
)
4015 *suspend_point
= need_backup
;
4016 else if (*suspend_point
+ 2 * target
< array_size
)
4017 *suspend_point
+= 2 * target
;
4019 *suspend_point
= array_size
;
4020 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
4021 if (max_progress
> *suspend_point
)
4022 max_progress
= *suspend_point
;
4025 if (info
->array
.major_version
>= 0) {
4026 /* Only need to suspend when about to backup */
4027 if (info
->reshape_progress
< need_backup
* 2 &&
4028 *suspend_point
> 0) {
4030 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4031 sysfs_set_num(info
, NULL
, "suspend_hi",
4035 /* Need to suspend continually */
4036 if (info
->reshape_progress
< *suspend_point
)
4037 *suspend_point
= info
->reshape_progress
;
4038 if (*suspend_point
+ target
< info
->reshape_progress
)
4039 /* No need to move suspend region yet */;
4041 if (*suspend_point
>= 2 * target
)
4042 *suspend_point
-= 2 * target
;
4045 sysfs_set_num(info
, NULL
, "suspend_lo",
4048 if (max_progress
< *suspend_point
)
4049 max_progress
= *suspend_point
;
4053 /* now set sync_max to allow that progress. sync_max, like
4054 * sync_completed is a count of sectors written per device, so
4055 * we find the difference between max_progress and the start point,
4056 * and divide that by after.data_disks to get a sync_max
4058 * At the same time we convert wait_point to a similar number
4059 * for comparing against sync_completed.
4061 /* scale down max_progress to per_disk */
4062 max_progress
/= reshape
->after
.data_disks
;
4064 * Round to chunk size as some kernels give an erroneously
4067 max_progress
/= info
->new_chunk
/512;
4068 max_progress
*= info
->new_chunk
/512;
4069 /* And round to old chunk size as the kernel wants that */
4070 max_progress
/= info
->array
.chunk_size
/512;
4071 max_progress
*= info
->array
.chunk_size
/512;
4072 /* Limit progress to the whole device */
4073 if (max_progress
> info
->component_size
)
4074 max_progress
= info
->component_size
;
4075 wait_point
/= reshape
->after
.data_disks
;
4077 /* switch from 'device offset' to 'processed block count' */
4078 max_progress
= info
->component_size
- max_progress
;
4079 wait_point
= info
->component_size
- wait_point
;
4083 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4085 /* Now wait. If we have already reached the point that we were
4086 * asked to wait to, don't wait at all, else wait for any change.
4087 * We need to select on 'sync_completed' as that is the place that
4088 * notifications happen, but we are really interested in
4089 * 'reshape_position'
4091 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4093 goto check_progress
;
4095 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4096 goto check_progress
;
4098 while (completed
< max_progress
&& completed
< wait_point
) {
4099 /* Check that sync_action is still 'reshape' to avoid
4100 * waiting forever on a dead array
4103 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) <= 0 ||
4104 strncmp(action
, "reshape", 7) != 0)
4106 /* Some kernels reset 'sync_completed' to zero
4107 * before setting 'sync_action' to 'idle'.
4108 * So we need these extra tests.
4110 if (completed
== 0 && advancing
&&
4111 strncmp(action
, "idle", 4) == 0 &&
4112 info
->reshape_progress
> 0)
4114 if (completed
== 0 && !advancing
&&
4115 strncmp(action
, "idle", 4) == 0 &&
4116 info
->reshape_progress
<
4117 (info
->component_size
* reshape
->after
.data_disks
))
4119 sysfs_wait(fd
, NULL
);
4120 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4121 goto check_progress
;
4123 /* Some kernels reset 'sync_completed' to zero,
4124 * we need to have real point we are in md.
4125 * So in that case, read 'reshape_position' from sysfs.
4127 if (completed
== 0) {
4128 unsigned long long reshapep
;
4130 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) > 0 &&
4131 strncmp(action
, "idle", 4) == 0 &&
4132 sysfs_get_ll(info
, NULL
,
4133 "reshape_position", &reshapep
) == 0)
4134 *reshape_completed
= reshapep
;
4136 /* some kernels can give an incorrectly high
4137 * 'completed' number, so round down */
4138 completed
/= (info
->new_chunk
/512);
4139 completed
*= (info
->new_chunk
/512);
4140 /* Convert 'completed' back in to a 'progress' number */
4141 completed
*= reshape
->after
.data_disks
;
4143 completed
= (info
->component_size
4144 * reshape
->after
.data_disks
4146 *reshape_completed
= completed
;
4151 /* We return the need_backup flag. Caller will decide
4152 * how much - a multiple of ->backup_blocks up to *suspend_point
4155 return need_backup
> info
->reshape_progress
;
4157 return need_backup
>= info
->reshape_progress
;
4160 /* if we couldn't read a number from sync_completed, then
4161 * either the reshape did complete, or it aborted.
4162 * We can tell which by checking for 'none' in reshape_position.
4163 * If it did abort, then it might immediately restart if it
4164 * it was just a device failure that leaves us degraded but
4167 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
,
4168 sizeof(buf
)) < 0 || strncmp(buf
, "none", 4) != 0) {
4169 /* The abort might only be temporary. Wait up to 10
4170 * seconds for fd to contain a valid number again.
4174 unsigned long long new_sync_max
;
4175 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4176 if (sysfs_wait(fd
, &wait
) != 1)
4178 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4180 /* all good again */
4182 /* If "sync_max" is no longer max_progress
4183 * we need to freeze things
4185 sysfs_get_ll(info
, NULL
, "sync_max",
4187 *frozen
= (new_sync_max
!= max_progress
);
4189 case -2: /* read error - abort */
4196 return rv
; /* abort */
4198 /* Maybe racing with array shutdown - check state */
4201 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4203 strncmp(buf
, "inactive", 8) == 0 ||
4204 strncmp(buf
, "clear",5) == 0)
4205 return -2; /* abort */
4206 return -1; /* complete */
4210 /* FIXME return status is never checked */
4211 static int grow_backup(struct mdinfo
*sra
,
4212 unsigned long long offset
, /* per device */
4213 unsigned long stripes
, /* per device, in old chunks */
4214 int *sources
, unsigned long long *offsets
,
4215 int disks
, int chunk
, int level
, int layout
,
4216 int dests
, int *destfd
, unsigned long long *destoffsets
,
4217 int part
, int *degraded
,
4220 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4221 * to storage 'destfd' (offset 'destoffsets'), after first
4222 * suspending IO. Then allow resync to continue
4223 * over the suspended section.
4224 * Use part 'part' of the backup-super-block.
4229 unsigned long long ll
;
4231 //printf("offset %llu\n", offset);
4237 /* Check that array hasn't become degraded, else we might backup the wrong data */
4238 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4239 return -1; /* FIXME this error is ignored */
4240 new_degraded
= (int)ll
;
4241 if (new_degraded
!= *degraded
) {
4242 /* check each device to ensure it is still working */
4244 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4245 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4247 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4250 if (sysfs_get_str(sra
, sd
, "state",
4251 sbuf
, sizeof(sbuf
)) < 0 ||
4252 strstr(sbuf
, "faulty") ||
4253 strstr(sbuf
, "in_sync") == NULL
) {
4254 /* this device is dead */
4255 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4256 if (sd
->disk
.raid_disk
>= 0 &&
4257 sources
[sd
->disk
.raid_disk
] >= 0) {
4258 close(sources
[sd
->disk
.raid_disk
]);
4259 sources
[sd
->disk
.raid_disk
] = -1;
4264 *degraded
= new_degraded
;
4267 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4268 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4270 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4271 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4274 bsb
.magic
[15] = '2';
4275 for (i
= 0; i
< dests
; i
++)
4277 lseek64(destfd
[i
], destoffsets
[i
] +
4278 __le64_to_cpu(bsb
.devstart2
)*512, 0);
4280 lseek64(destfd
[i
], destoffsets
[i
], 0);
4282 rv
= save_stripes(sources
, offsets
, disks
, chunk
, level
, layout
,
4283 dests
, destfd
, offset
* 512 * odata
,
4284 stripes
* chunk
* odata
, buf
);
4288 bsb
.mtime
= __cpu_to_le64(time(0));
4289 for (i
= 0; i
< dests
; i
++) {
4290 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4292 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4293 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4294 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4295 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4296 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4299 if ((unsigned long long)lseek64(destfd
[i
],
4300 destoffsets
[i
] - 4096, 0) !=
4301 destoffsets
[i
] - 4096)
4303 if (write(destfd
[i
], &bsb
, 512) != 512)
4305 if (destoffsets
[i
] > 4096) {
4306 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4307 destoffsets
[i
]+stripes
*chunk
*odata
)
4309 if (write(destfd
[i
], &bsb
, 512) != 512)
4319 /* in 2.6.30, the value reported by sync_completed can be
4320 * less that it should be by one stripe.
4321 * This only happens when reshape hits sync_max and pauses.
4322 * So allow wait_backup to either extent sync_max further
4323 * than strictly necessary, or return before the
4324 * sync has got quite as far as we would really like.
4325 * This is what 'blocks2' is for.
4326 * The various caller give appropriate values so that
4329 /* FIXME return value is often ignored */
4330 static int forget_backup(int dests
, int *destfd
,
4331 unsigned long long *destoffsets
,
4335 * Erase backup 'part' (which is 0 or 1)
4341 bsb
.arraystart2
= __cpu_to_le64(0);
4342 bsb
.length2
= __cpu_to_le64(0);
4344 bsb
.arraystart
= __cpu_to_le64(0);
4345 bsb
.length
= __cpu_to_le64(0);
4347 bsb
.mtime
= __cpu_to_le64(time(0));
4349 for (i
= 0; i
< dests
; i
++) {
4350 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4351 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4352 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4353 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4354 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4355 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4356 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4357 destoffsets
[i
]-4096)
4359 if (rv
== 0 && write(destfd
[i
], &bsb
, 512) != 512)
4366 static void fail(char *msg
)
4369 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4370 rv
|= (write(2, "\n", 1) != 1);
4374 static char *abuf
, *bbuf
;
4375 static unsigned long long abuflen
;
4376 static void validate(int afd
, int bfd
, unsigned long long offset
)
4378 /* check that the data in the backup against the array.
4379 * This is only used for regression testing and should not
4380 * be used while the array is active
4384 lseek64(bfd
, offset
- 4096, 0);
4385 if (read(bfd
, &bsb2
, 512) != 512)
4386 fail("cannot read bsb");
4387 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4388 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4389 fail("first csum bad");
4390 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4391 fail("magic is bad");
4392 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4393 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4394 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4395 fail("second csum bad");
4397 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4398 fail("devstart is wrong");
4401 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4403 if (abuflen
< len
) {
4407 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4408 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4410 /* just stop validating on mem-alloc failure */
4415 lseek64(bfd
, offset
, 0);
4416 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4417 //printf("len %llu\n", len);
4418 fail("read first backup failed");
4420 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4421 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4422 fail("read first from array failed");
4423 if (memcmp(bbuf
, abuf
, len
) != 0) {
4426 printf("offset=%llu len=%llu\n",
4427 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4428 for (i
=0; i
<len
; i
++)
4429 if (bbuf
[i
] != abuf
[i
]) {
4430 printf("first diff byte %d\n", i
);
4434 fail("data1 compare failed");
4438 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4440 if (abuflen
< len
) {
4444 abuf
= xmalloc(abuflen
);
4445 bbuf
= xmalloc(abuflen
);
4448 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4449 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4450 fail("read second backup failed");
4451 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4452 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4453 fail("read second from array failed");
4454 if (memcmp(bbuf
, abuf
, len
) != 0)
4455 fail("data2 compare failed");
4459 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4460 struct supertype
*st
, unsigned long blocks
,
4461 int *fds
, unsigned long long *offsets
,
4462 int dests
, int *destfd
, unsigned long long *destoffsets
)
4464 /* Monitor a reshape where backup is being performed using
4465 * 'native' mechanism - either to a backup file, or
4466 * to some space in a spare.
4470 unsigned long long speed
;
4471 unsigned long long suspend_point
, array_size
;
4472 unsigned long long backup_point
, wait_point
;
4473 unsigned long long reshape_completed
;
4475 int increasing
= reshape
->after
.data_disks
>=
4476 reshape
->before
.data_disks
;
4477 int part
= 0; /* The next part of the backup area to fill. It
4478 * may already be full, so we need to check */
4479 int level
= reshape
->level
;
4480 int layout
= reshape
->before
.layout
;
4481 int data
= reshape
->before
.data_disks
;
4482 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4483 int chunk
= sra
->array
.chunk_size
;
4485 unsigned long stripes
;
4489 /* set up the backup-super-block. This requires the
4490 * uuid from the array.
4492 /* Find a superblock */
4493 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4497 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4499 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4500 devfd
= dev_open(dn
, O_RDONLY
);
4503 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4509 pr_err("Cannot find a superblock\n");
4513 memset(&bsb
, 0, 512);
4514 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4515 st
->ss
->uuid_from_super(st
, uuid
);
4516 memcpy(bsb
.set_uuid
, uuid
, 16);
4517 bsb
.mtime
= __cpu_to_le64(time(0));
4518 bsb
.devstart2
= blocks
;
4520 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4521 reshape
->before
.data_disks
;
4523 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4524 /* Don't start the 'reshape' */
4526 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4527 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4528 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4532 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4533 backup_point
= sra
->reshape_progress
;
4536 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4537 backup_point
= reshape
->backup_blocks
;
4538 suspend_point
= array_size
;
4544 /* Want to return as soon the oldest backup slot can
4545 * be released as that allows us to start backing up
4546 * some more, providing suspend_point has been
4547 * advanced, which it should have.
4550 wait_point
= array_size
;
4551 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4552 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4553 __le64_to_cpu(bsb
.length
));
4554 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4555 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4556 __le64_to_cpu(bsb
.length2
));
4559 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4560 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4561 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4562 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4565 reshape_completed
= sra
->reshape_progress
;
4566 rv
= progress_reshape(sra
, reshape
,
4567 backup_point
, wait_point
,
4568 &suspend_point
, &reshape_completed
,
4570 /* external metadata would need to ping_monitor here */
4571 sra
->reshape_progress
= reshape_completed
;
4573 /* Clear any backup region that is before 'here' */
4575 if (__le64_to_cpu(bsb
.length
) > 0 &&
4576 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4577 __le64_to_cpu(bsb
.length
)))
4578 forget_backup(dests
, destfd
,
4580 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4581 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4582 __le64_to_cpu(bsb
.length2
)))
4583 forget_backup(dests
, destfd
,
4586 if (__le64_to_cpu(bsb
.length
) > 0 &&
4587 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4588 forget_backup(dests
, destfd
,
4590 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4591 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4592 forget_backup(dests
, destfd
,
4602 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4603 /* No longer need to monitor this reshape */
4604 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4610 unsigned long long offset
;
4611 unsigned long actual_stripes
;
4612 /* Need to backup some data.
4613 * If 'part' is not used and the desired
4614 * backup size is suspended, do a backup,
4615 * then consider the next part.
4617 /* Check that 'part' is unused */
4618 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4620 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4623 offset
= backup_point
/ data
;
4624 actual_stripes
= stripes
;
4626 if (offset
+ actual_stripes
* (chunk
/512) >
4627 sra
->component_size
)
4628 actual_stripes
= ((sra
->component_size
- offset
)
4630 if (offset
+ actual_stripes
* (chunk
/512) >
4634 if (offset
< actual_stripes
* (chunk
/512))
4635 actual_stripes
= offset
/ (chunk
/512);
4636 offset
-= actual_stripes
* (chunk
/512);
4637 if (offset
< suspend_point
/data
)
4640 if (actual_stripes
== 0)
4642 grow_backup(sra
, offset
, actual_stripes
, fds
, offsets
,
4643 disks
, chunk
, level
, layout
, dests
, destfd
,
4644 destoffsets
, part
, °raded
, buf
);
4645 validate(afd
, destfd
[0], destoffsets
[0]);
4646 /* record where 'part' is up to */
4649 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4651 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4655 /* FIXME maybe call progress_reshape one more time instead */
4656 /* remove any remaining suspension */
4657 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4658 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4659 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4660 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4662 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4663 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4669 * If any spare contains md_back_data-1 which is recent wrt mtime,
4670 * write that data into the array and update the super blocks with
4671 * the new reshape_progress
4673 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
,
4674 int cnt
, char *backup_file
, int verbose
)
4678 unsigned long long *offsets
;
4679 unsigned long long nstripe
, ostripe
;
4682 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4683 if (info
->array
.level
== 6)
4684 odata
--; /* number of data disks */
4685 ndata
= info
->array
.raid_disks
- 1;
4686 if (info
->new_level
== 6)
4689 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4691 if (info
->delta_disks
<= 0)
4692 /* Didn't grow, so the backup file must have
4696 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4697 struct mdinfo dinfo
;
4700 char *devname
, namebuf
[20];
4701 unsigned long long lo
, hi
;
4703 /* This was a spare and may have some saved data on it.
4704 * Load the superblock, find and load the
4705 * backup_super_block.
4706 * If either fail, go on to next device.
4707 * If the backup contains no new info, just return
4708 * else restore data and update all superblocks
4710 if (i
== old_disks
-1) {
4711 fd
= open(backup_file
, O_RDONLY
);
4713 pr_err("backup file %s inaccessible: %s\n",
4714 backup_file
, strerror(errno
));
4717 devname
= backup_file
;
4722 if (st
->ss
->load_super(st
, fd
, NULL
))
4725 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4726 st
->ss
->free_super(st
);
4729 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4731 pr_err("Cannot seek on device %d\n", i
);
4732 continue; /* Cannot seek */
4734 sprintf(namebuf
, "device-%d", i
);
4737 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4739 pr_err("Cannot read from %s\n", devname
);
4740 continue; /* Cannot read */
4742 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4743 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4745 pr_err("No backup metadata on %s\n", devname
);
4748 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4750 pr_err("Bad backup-metadata checksum on %s\n",
4752 continue; /* bad checksum */
4754 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4755 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4757 pr_err("Bad backup-metadata checksum2 on %s\n",
4759 continue; /* Bad second checksum */
4761 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4763 pr_err("Wrong uuid on backup-metadata on %s\n",
4765 continue; /* Wrong uuid */
4769 * array utime and backup-mtime should be updated at
4770 * much the same time, but it seems that sometimes
4771 * they aren't... So allow considerable flexability in
4772 * matching, and allow this test to be overridden by
4773 * an environment variable.
4775 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4776 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4777 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4778 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4779 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4780 (unsigned long)info
->array
.utime
);
4782 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4783 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4784 continue; /* time stamp is too bad */
4788 if (bsb
.magic
[15] == '1') {
4789 if (bsb
.length
== 0)
4791 if (info
->delta_disks
>= 0) {
4792 /* reshape_progress is increasing */
4793 if (__le64_to_cpu(bsb
.arraystart
)
4794 + __le64_to_cpu(bsb
.length
)
4795 < info
->reshape_progress
) {
4798 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4799 continue; /* No new data here */
4802 /* reshape_progress is decreasing */
4803 if (__le64_to_cpu(bsb
.arraystart
) >=
4804 info
->reshape_progress
)
4805 goto nonew
; /* No new data here */
4808 if (bsb
.length
== 0 && bsb
.length2
== 0)
4810 if (info
->delta_disks
>= 0) {
4811 /* reshape_progress is increasing */
4812 if ((__le64_to_cpu(bsb
.arraystart
)
4813 + __le64_to_cpu(bsb
.length
)
4814 < info
->reshape_progress
) &&
4815 (__le64_to_cpu(bsb
.arraystart2
)
4816 + __le64_to_cpu(bsb
.length2
)
4817 < info
->reshape_progress
))
4818 goto nonew
; /* No new data here */
4820 /* reshape_progress is decreasing */
4821 if (__le64_to_cpu(bsb
.arraystart
) >=
4822 info
->reshape_progress
&&
4823 __le64_to_cpu(bsb
.arraystart2
) >=
4824 info
->reshape_progress
)
4825 goto nonew
; /* No new data here */
4828 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4831 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4833 continue; /* Cannot seek */
4835 /* There should be a duplicate backup superblock 4k before here */
4836 if (lseek64(fd
, -4096, 1) < 0 ||
4837 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4838 goto second_fail
; /* Cannot find leading superblock */
4839 if (bsb
.magic
[15] == '1')
4840 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4842 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4843 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4844 goto second_fail
; /* Cannot find leading superblock */
4846 /* Now need the data offsets for all devices. */
4847 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4848 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4851 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4852 /* FIXME should be this be an error */
4854 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4855 st
->ss
->free_super(st
);
4856 offsets
[j
] = dinfo
.data_offset
* 512;
4858 printf("%s: restoring critical section\n", Name
);
4860 if (restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4861 info
->new_chunk
, info
->new_level
,
4862 info
->new_layout
, fd
,
4863 __le64_to_cpu(bsb
.devstart
)*512,
4864 __le64_to_cpu(bsb
.arraystart
)*512,
4865 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4866 /* didn't succeed, so giveup */
4868 pr_err("Error restoring backup from %s\n",
4874 if (bsb
.magic
[15] == '2' &&
4875 restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4876 info
->new_chunk
, info
->new_level
,
4877 info
->new_layout
, fd
,
4878 __le64_to_cpu(bsb
.devstart
)*512 +
4879 __le64_to_cpu(bsb
.devstart2
)*512,
4880 __le64_to_cpu(bsb
.arraystart2
)*512,
4881 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4882 /* didn't succeed, so giveup */
4884 pr_err("Error restoring second backup from %s\n",
4892 /* Ok, so the data is restored. Let's update those superblocks. */
4896 lo
= __le64_to_cpu(bsb
.arraystart
);
4897 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4899 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4900 unsigned long long lo1
, hi1
;
4901 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4902 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4906 } else if (lo
< lo1
)
4911 if (lo
< hi
&& (info
->reshape_progress
< lo
||
4912 info
->reshape_progress
> hi
))
4913 /* backup does not affect reshape_progress*/ ;
4914 else if (info
->delta_disks
>= 0) {
4915 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4916 __le64_to_cpu(bsb
.length
);
4917 if (bsb
.magic
[15] == '2') {
4918 unsigned long long p2
;
4920 p2
= __le64_to_cpu(bsb
.arraystart2
) +
4921 __le64_to_cpu(bsb
.length2
);
4922 if (p2
> info
->reshape_progress
)
4923 info
->reshape_progress
= p2
;
4926 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4927 if (bsb
.magic
[15] == '2') {
4928 unsigned long long p2
;
4930 p2
= __le64_to_cpu(bsb
.arraystart2
);
4931 if (p2
< info
->reshape_progress
)
4932 info
->reshape_progress
= p2
;
4935 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4938 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4940 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4941 dinfo
.reshape_progress
= info
->reshape_progress
;
4942 st
->ss
->update_super(st
, &dinfo
, "_reshape_progress",
4944 st
->ss
->store_super(st
, fdlist
[j
]);
4945 st
->ss
->free_super(st
);
4949 /* Didn't find any backup data, try to see if any
4952 if (info
->delta_disks
< 0) {
4953 /* When shrinking, the critical section is at the end.
4954 * So see if we are before the critical section.
4956 unsigned long long first_block
;
4957 nstripe
= ostripe
= 0;
4959 while (ostripe
>= nstripe
) {
4960 ostripe
+= info
->array
.chunk_size
/ 512;
4961 first_block
= ostripe
* odata
;
4962 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4963 (info
->new_chunk
/512);
4966 if (info
->reshape_progress
>= first_block
)
4969 if (info
->delta_disks
> 0) {
4970 /* See if we are beyond the critical section. */
4971 unsigned long long last_block
;
4972 nstripe
= ostripe
= 0;
4974 while (nstripe
>= ostripe
) {
4975 nstripe
+= info
->new_chunk
/ 512;
4976 last_block
= nstripe
* ndata
;
4977 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4978 (info
->array
.chunk_size
/512);
4981 if (info
->reshape_progress
>= last_block
)
4984 /* needed to recover critical section! */
4986 pr_err("Failed to find backup of critical section\n");
4990 int Grow_continue_command(char *devname
, int fd
,
4991 char *backup_file
, int verbose
)
4994 struct supertype
*st
= NULL
;
4995 struct mdinfo
*content
= NULL
;
4996 struct mdinfo array
;
4997 char *subarray
= NULL
;
4998 struct mdinfo
*cc
= NULL
;
4999 struct mdstat_ent
*mdstat
= NULL
;
5003 dprintf("Grow continue from command line called for %s\n", devname
);
5005 st
= super_by_fd(fd
, &subarray
);
5006 if (!st
|| !st
->ss
) {
5007 pr_err("Unable to determine metadata format for %s\n", devname
);
5010 dprintf("Grow continue is run for ");
5011 if (st
->ss
->external
== 0) {
5014 dprintf_cont("native array (%s)\n", devname
);
5015 if (md_get_array_info(fd
, &array
.array
) < 0) {
5016 pr_err("%s is not an active md array - aborting\n",
5019 goto Grow_continue_command_exit
;
5022 sysfs_init(content
, fd
, NULL
);
5023 /* Need to load a superblock.
5024 * FIXME we should really get what we need from
5028 for (d
= 0; d
< MAX_DISKS
; d
++) {
5029 mdu_disk_info_t disk
;
5033 if (md_get_disk_info(fd
, &disk
) < 0)
5035 if (disk
.major
== 0 && disk
.minor
== 0)
5037 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5039 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5042 fd2
= dev_open(dv
, O_RDONLY
);
5045 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5051 if (d
== MAX_DISKS
) {
5052 pr_err("Unable to load metadata for %s\n",
5055 goto Grow_continue_command_exit
;
5057 st
->ss
->getinfo_super(st
, content
, NULL
);
5058 if (!content
->reshape_active
)
5062 } while (cnt
-- > 0);
5067 dprintf_cont("subarray (%s)\n", subarray
);
5068 container
= st
->container_devnm
;
5069 cfd
= open_dev_excl(st
->container_devnm
);
5071 container
= st
->devnm
;
5073 cfd
= open_dev_excl(st
->devnm
);
5074 dprintf_cont("container (%s)\n", container
);
5078 pr_err("Unable to open container for %s\n", devname
);
5080 goto Grow_continue_command_exit
;
5083 /* find in container array under reshape
5085 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5087 pr_err("Cannot read superblock for %s\n", devname
);
5089 goto Grow_continue_command_exit
;
5092 cc
= st
->ss
->container_content(st
, subarray
);
5093 for (content
= cc
; content
; content
= content
->next
) {
5095 int allow_reshape
= 1;
5097 if (content
->reshape_active
== 0)
5099 /* The decision about array or container wide
5100 * reshape is taken in Grow_continue based
5101 * content->reshape_active state, therefore we
5102 * need to check_reshape based on
5103 * reshape_active and subarray name
5105 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5107 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5108 (content
->array
.state
5109 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5112 if (!allow_reshape
) {
5113 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5114 devname
, container
);
5116 goto Grow_continue_command_exit
;
5119 array_name
= strchr(content
->text_version
+1, '/')+1;
5120 mdstat
= mdstat_by_subdev(array_name
, container
);
5123 if (mdstat
->active
== 0) {
5124 pr_err("Skipping inactive array %s.\n",
5126 free_mdstat(mdstat
);
5133 pr_err("Unable to determine reshaped array for %s\n", devname
);
5135 goto Grow_continue_command_exit
;
5137 fd2
= open_dev(mdstat
->devnm
);
5139 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5141 goto Grow_continue_command_exit
;
5144 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5145 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5149 goto Grow_continue_command_exit
;
5154 /* start mdmon in case it is not running
5156 if (!mdmon_running(container
))
5157 start_mdmon(container
);
5158 ping_monitor(container
);
5160 if (mdmon_running(container
))
5161 st
->update_tail
= &st
->updates
;
5163 pr_err("No mdmon found. Grow cannot continue.\n");
5165 goto Grow_continue_command_exit
;
5169 /* verify that array under reshape is started from
5172 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5174 goto Grow_continue_command_exit
;
5179 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5181 Grow_continue_command_exit
:
5184 st
->ss
->free_super(st
);
5185 free_mdstat(mdstat
);
5192 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5193 char *backup_file
, int forked
, int freeze_reshape
)
5197 if (!info
->reshape_active
)
5200 if (st
->ss
->external
) {
5201 int cfd
= open_dev(st
->container_devnm
);
5206 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5208 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5209 st
, info
, 0, backup_file
, 0,
5210 forked
, 1 | info
->reshape_active
,
5213 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5214 NULL
, INVALID_SECTORS
, backup_file
,
5215 0, forked
, 1 | info
->reshape_active
,
5221 char *make_backup(char *name
)
5223 char *base
= "backup_file-";
5227 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5228 fname
= xmalloc(len
);
5229 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5233 char *locate_backup(char *name
)
5235 char *fl
= make_backup(name
);
5238 if (stat(fl
, &stb
) == 0 && S_ISREG(stb
.st_mode
))