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 initialize 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
= (unsigned long)(ochunk
/512) * (unsigned long)(nchunk
/512) *
1205 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1207 /* Based on the current array state in info->array and
1208 * the changes in info->new_* etc, determine:
1209 * - whether the change is possible
1210 * - Intermediate level/raid_disks/layout
1211 * - whether a restriping reshape is needed
1212 * - number of sectors in minimum change unit. This
1213 * will cover a whole number of stripes in 'before' and
1216 * Return message if the change should be rejected
1217 * NULL if the change can be achieved
1219 * This can be called as part of starting a reshape, or
1220 * when assembling an array that is undergoing reshape.
1222 int near
, far
, offset
, copies
;
1224 int old_chunk
, new_chunk
;
1225 /* delta_parity records change in number of devices
1226 * caused by level change
1228 int delta_parity
= 0;
1230 memset(re
, 0, sizeof(*re
));
1232 /* If a new level not explicitly given, we assume no-change */
1233 if (info
->new_level
== UnSet
)
1234 info
->new_level
= info
->array
.level
;
1236 if (info
->new_chunk
)
1237 switch (info
->new_level
) {
1243 /* chunk size is meaningful, must divide component_size
1246 if (info
->component_size
% (info
->new_chunk
/512)) {
1247 unsigned long long shrink
= info
->component_size
;
1248 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1249 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1250 info
->new_chunk
/1024, info
->component_size
/2);
1251 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1253 pr_err("will shrink the array so the given chunk size would work.\n");
1258 return "chunk size not meaningful for this level";
1261 info
->new_chunk
= info
->array
.chunk_size
;
1263 switch (info
->array
.level
) {
1265 return "No reshape is possibly for this RAID level";
1267 if (info
->delta_disks
!= UnSet
)
1268 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1270 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1272 /* RAID1 can convert to RAID1 with different disks, or
1273 * raid5 with 2 disks, or
1276 if (info
->new_level
> 1 && (info
->component_size
& 7))
1277 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1278 if (info
->new_level
== 0) {
1279 if (info
->delta_disks
!= UnSet
&&
1280 info
->delta_disks
!= 0)
1281 return "Cannot change number of disks with RAID1->RAID0 conversion";
1283 re
->before
.data_disks
= 1;
1284 re
->after
.data_disks
= 1;
1287 if (info
->new_level
== 1) {
1288 if (info
->delta_disks
== UnSet
)
1289 /* Don't know what to do */
1290 return "no change requested for Growing RAID1";
1294 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1295 return "Can only convert a 2-device array to RAID5";
1296 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1298 re
->before
.data_disks
= 1;
1299 if (info
->delta_disks
!= UnSet
&&
1300 info
->delta_disks
!= 0)
1301 re
->after
.data_disks
= 1 + info
->delta_disks
;
1303 re
->after
.data_disks
= 1;
1304 if (re
->after
.data_disks
< 1)
1305 return "Number of disks too small for RAID5";
1307 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1308 info
->array
.chunk_size
= 65536;
1311 /* Could do some multi-stage conversions, but leave that to
1314 return "Impossibly level change request for RAID1";
1317 /* RAID10 can be converted from near mode to
1318 * RAID0 by removing some devices.
1319 * It can also be reshaped if the kernel supports
1322 switch (info
->new_level
) {
1324 if ((info
->array
.layout
& ~0xff) != 0x100)
1325 return "Cannot Grow RAID10 with far/offset layout";
1327 * number of devices must be multiple of
1330 if (info
->array
.raid_disks
%
1331 (info
->array
.layout
& 0xff))
1332 return "RAID10 layout too complex for Grow operation";
1334 new_disks
= (info
->array
.raid_disks
/
1335 (info
->array
.layout
& 0xff));
1336 if (info
->delta_disks
== UnSet
)
1337 info
->delta_disks
= (new_disks
1338 - info
->array
.raid_disks
);
1340 if (info
->delta_disks
!=
1341 new_disks
- info
->array
.raid_disks
)
1342 return "New number of raid-devices impossible for RAID10";
1343 if (info
->new_chunk
&&
1344 info
->new_chunk
!= info
->array
.chunk_size
)
1345 return "Cannot change chunk-size with RAID10 Grow";
1349 re
->before
.data_disks
= new_disks
;
1350 re
->after
.data_disks
= re
->before
.data_disks
;
1354 near
= info
->array
.layout
& 0xff;
1355 far
= (info
->array
.layout
>> 8) & 0xff;
1356 offset
= info
->array
.layout
& 0x10000;
1357 if (far
> 1 && !offset
)
1358 return "Cannot reshape RAID10 in far-mode";
1359 copies
= near
* far
;
1361 old_chunk
= info
->array
.chunk_size
* far
;
1363 if (info
->new_layout
== UnSet
)
1364 info
->new_layout
= info
->array
.layout
;
1366 near
= info
->new_layout
& 0xff;
1367 far
= (info
->new_layout
>> 8) & 0xff;
1368 offset
= info
->new_layout
& 0x10000;
1369 if (far
> 1 && !offset
)
1370 return "Cannot reshape RAID10 to far-mode";
1371 if (near
* far
!= copies
)
1372 return "Cannot change number of copies when reshaping RAID10";
1374 if (info
->delta_disks
== UnSet
)
1375 info
->delta_disks
= 0;
1376 new_disks
= (info
->array
.raid_disks
+
1379 new_chunk
= info
->new_chunk
* far
;
1382 re
->before
.layout
= info
->array
.layout
;
1383 re
->before
.data_disks
= info
->array
.raid_disks
;
1384 re
->after
.layout
= info
->new_layout
;
1385 re
->after
.data_disks
= new_disks
;
1386 /* For RAID10 we don't do backup but do allow reshape,
1387 * so set backup_blocks to INVALID_SECTORS rather than
1389 * And there is no need to synchronise stripes on both
1390 * 'old' and 'new'. So the important
1391 * number is the minimum data_offset difference
1392 * which is the larger of (offset copies * chunk).
1394 re
->backup_blocks
= INVALID_SECTORS
;
1395 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1396 if (new_disks
< re
->before
.data_disks
&&
1397 info
->space_after
< re
->min_offset_change
)
1398 /* Reduce component size by one chunk */
1399 re
->new_size
= (info
->component_size
-
1400 re
->min_offset_change
);
1402 re
->new_size
= info
->component_size
;
1403 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1407 return "RAID10 can only be changed to RAID0";
1410 /* RAID0 can be converted to RAID10, or to RAID456 */
1411 if (info
->new_level
== 10) {
1412 if (info
->new_layout
== UnSet
&&
1413 info
->delta_disks
== UnSet
) {
1414 /* Assume near=2 layout */
1415 info
->new_layout
= 0x102;
1416 info
->delta_disks
= info
->array
.raid_disks
;
1418 if (info
->new_layout
== UnSet
) {
1419 int copies
= 1 + (info
->delta_disks
1420 / info
->array
.raid_disks
);
1421 if (info
->array
.raid_disks
* (copies
-1) !=
1423 return "Impossible number of devices for RAID0->RAID10";
1424 info
->new_layout
= 0x100 + copies
;
1426 if (info
->delta_disks
== UnSet
) {
1427 int copies
= info
->new_layout
& 0xff;
1428 if (info
->new_layout
!= 0x100 + copies
)
1429 return "New layout impossible for RAID0->RAID10";;
1430 info
->delta_disks
= (copies
- 1) *
1431 info
->array
.raid_disks
;
1433 if (info
->new_chunk
&&
1434 info
->new_chunk
!= info
->array
.chunk_size
)
1435 return "Cannot change chunk-size with RAID0->RAID10";
1438 re
->before
.data_disks
= (info
->array
.raid_disks
+
1440 re
->after
.data_disks
= re
->before
.data_disks
;
1441 re
->before
.layout
= info
->new_layout
;
1445 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1446 * a raid4 style layout of the final level.
1448 switch (info
->new_level
) {
1453 re
->before
.layout
= 0;
1458 re
->before
.layout
= ALGORITHM_PARITY_N
;
1459 if (info
->new_layout
== UnSet
)
1460 info
->new_layout
= map_name(r5layout
, "default");
1465 re
->before
.layout
= ALGORITHM_PARITY_N
;
1466 if (info
->new_layout
== UnSet
)
1467 info
->new_layout
= map_name(r6layout
, "default");
1470 return "Impossible level change requested";
1472 re
->before
.data_disks
= info
->array
.raid_disks
;
1473 /* determining 'after' layout happens outside this 'switch' */
1477 info
->array
.layout
= ALGORITHM_PARITY_N
;
1479 switch (info
->new_level
) {
1483 re
->level
= info
->array
.level
;
1484 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1485 re
->before
.layout
= info
->array
.layout
;
1489 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1490 re
->before
.layout
= info
->array
.layout
;
1495 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1496 switch (info
->array
.layout
) {
1497 case ALGORITHM_LEFT_ASYMMETRIC
:
1498 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1500 case ALGORITHM_RIGHT_ASYMMETRIC
:
1501 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1503 case ALGORITHM_LEFT_SYMMETRIC
:
1504 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1506 case ALGORITHM_RIGHT_SYMMETRIC
:
1507 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1509 case ALGORITHM_PARITY_0
:
1510 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1512 case ALGORITHM_PARITY_N
:
1513 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1516 return "Cannot convert an array with this layout";
1520 if (info
->array
.raid_disks
!= 2)
1521 return "Can only convert a 2-device array to RAID1";
1522 if (info
->delta_disks
!= UnSet
&&
1523 info
->delta_disks
!= 0)
1524 return "Cannot set raid_disk when converting RAID5->RAID1";
1526 info
->new_chunk
= 0;
1529 return "Impossible level change requested";
1533 switch (info
->new_level
) {
1539 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1540 re
->before
.layout
= info
->array
.layout
;
1543 return "Impossible level change requested";
1548 /* If we reached here then it looks like a re-stripe is
1549 * happening. We have determined the intermediate level
1550 * and initial raid_disks/layout and stored these in 're'.
1552 * We need to deduce the final layout that can be atomically
1553 * converted to the end state.
1555 switch (info
->new_level
) {
1557 /* We can only get to RAID0 from RAID4 or RAID5
1558 * with appropriate layout and one extra device
1560 if (re
->level
!= 4 && re
->level
!= 5)
1561 return "Cannot covert to RAID0 from this level";
1563 switch (re
->level
) {
1565 re
->before
.layout
= 0;
1566 re
->after
.layout
= 0;
1569 re
->after
.layout
= ALGORITHM_PARITY_N
;
1575 /* We can only get to RAID4 from RAID5 */
1576 if (re
->level
!= 4 && re
->level
!= 5)
1577 return "Cannot convert to RAID4 from this level";
1579 switch (re
->level
) {
1581 re
->after
.layout
= 0;
1584 re
->after
.layout
= ALGORITHM_PARITY_N
;
1590 /* We get to RAID5 from RAID5 or RAID6 */
1591 if (re
->level
!= 5 && re
->level
!= 6)
1592 return "Cannot convert to RAID5 from this level";
1594 switch (re
->level
) {
1596 if (info
->new_layout
== UnSet
)
1597 re
->after
.layout
= re
->before
.layout
;
1599 re
->after
.layout
= info
->new_layout
;
1602 if (info
->new_layout
== UnSet
)
1603 info
->new_layout
= re
->before
.layout
;
1605 /* after.layout needs to be raid6 version of new_layout */
1606 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1607 re
->after
.layout
= ALGORITHM_PARITY_N
;
1610 char *ls
= map_num(r5layout
, info
->new_layout
);
1613 /* Current RAID6 layout has a RAID5
1616 strcat(strcpy(layout
, ls
), "-6");
1617 l
= map_name(r6layout
, layout
);
1619 return "Cannot find RAID6 layout to convert to";
1621 /* Current RAID6 has no equivalent.
1622 * If it is already a '-6' layout we
1623 * can leave it unchanged, else we must
1626 ls
= map_num(r6layout
,
1629 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1630 return "Please specify new layout";
1631 l
= info
->new_layout
;
1633 re
->after
.layout
= l
;
1639 /* We must already be at level 6 */
1641 return "Impossible level change";
1642 if (info
->new_layout
== UnSet
)
1643 re
->after
.layout
= info
->array
.layout
;
1645 re
->after
.layout
= info
->new_layout
;
1648 return "Impossible level change requested";
1650 if (info
->delta_disks
== UnSet
)
1651 info
->delta_disks
= delta_parity
;
1653 re
->after
.data_disks
=
1654 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1656 switch (re
->level
) {
1668 /* So we have a restripe operation, we need to calculate the number
1669 * of blocks per reshape operation.
1671 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1672 if (info
->new_chunk
== 0)
1673 info
->new_chunk
= info
->array
.chunk_size
;
1674 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1675 re
->after
.layout
== re
->before
.layout
&&
1676 info
->new_chunk
== info
->array
.chunk_size
) {
1677 /* Nothing to change, can change level immediately. */
1678 re
->level
= info
->new_level
;
1679 re
->backup_blocks
= 0;
1682 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1683 /* chunk and layout changes make no difference */
1684 re
->level
= info
->new_level
;
1685 re
->backup_blocks
= 0;
1689 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1690 get_linux_version() < 2006032)
1691 return "in-place reshape is not safe before 2.6.32 - sorry.";
1693 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1694 get_linux_version() < 2006030)
1695 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1697 re
->backup_blocks
= compute_backup_blocks(
1698 info
->new_chunk
, info
->array
.chunk_size
,
1699 re
->after
.data_disks
, re
->before
.data_disks
);
1700 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1702 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1706 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1709 struct mdinfo
*info
;
1713 if ((st
== NULL
) || (sra
== NULL
))
1716 if (text_version
== NULL
)
1717 text_version
= sra
->text_version
;
1718 subarray
= strchr(text_version
+ 1, '/')+1;
1719 info
= st
->ss
->container_content(st
, subarray
);
1721 unsigned long long current_size
= 0;
1722 unsigned long long new_size
= info
->custom_array_size
/2;
1724 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1725 new_size
> current_size
) {
1726 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1728 dprintf("Error: Cannot set array size");
1731 dprintf("Array size changed");
1733 dprintf_cont(" from %llu to %llu.\n",
1734 current_size
, new_size
);
1738 dprintf("Error: set_array_size(): info pointer in NULL\n");
1743 static int reshape_array(char *container
, int fd
, char *devname
,
1744 struct supertype
*st
, struct mdinfo
*info
,
1745 int force
, struct mddev_dev
*devlist
,
1746 unsigned long long data_offset
,
1747 char *backup_file
, int verbose
, int forked
,
1748 int restart
, int freeze_reshape
);
1749 static int reshape_container(char *container
, char *devname
,
1751 struct supertype
*st
,
1752 struct mdinfo
*info
,
1754 char *backup_file
, int verbose
,
1755 int forked
, int restart
, int freeze_reshape
);
1757 int Grow_reshape(char *devname
, int fd
,
1758 struct mddev_dev
*devlist
,
1759 unsigned long long data_offset
,
1760 struct context
*c
, struct shape
*s
)
1762 /* Make some changes in the shape of an array.
1763 * The kernel must support the change.
1765 * There are three different changes. Each can trigger
1766 * a resync or recovery so we freeze that until we have
1767 * requested everything (if kernel supports freezing - 2.6.30).
1769 * - change size (i.e. component_size)
1771 * - change layout/chunksize/ndisks
1773 * The last can require a reshape. It is different on different
1774 * levels so we need to check the level before actioning it.
1775 * Some times the level change needs to be requested after the
1776 * reshape (e.g. raid6->raid5, raid5->raid0)
1779 struct mdu_array_info_s array
;
1781 struct supertype
*st
;
1782 char *subarray
= NULL
;
1786 char *container
= NULL
;
1789 struct mddev_dev
*dv
;
1795 if (md_get_array_info(fd
, &array
) < 0) {
1796 pr_err("%s is not an active md array - aborting\n",
1800 if (s
->level
!= UnSet
&& s
->chunk
) {
1801 pr_err("Cannot change array level in the same operation as changing chunk size.\n");
1805 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1806 (array
.level
< 4 || array
.level
> 6)) {
1807 pr_err("--grow --data-offset not yet supported\n");
1812 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1813 pr_err("cannot change component size at the same time as other changes.\n"
1814 " Change size first, then check data is intact before making other changes.\n");
1818 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&&
1819 array
.level
> 1 && get_linux_version() < 2006032 &&
1820 !check_env("MDADM_FORCE_FEWER")) {
1821 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1822 " Please use a newer kernel\n");
1826 if (array
.level
> 1 && s
->size
> 1 &&
1827 (unsigned long long) (array
.chunk_size
/ 1024) > s
->size
) {
1828 pr_err("component size must be larger than chunk size.\n");
1832 st
= super_by_fd(fd
, &subarray
);
1834 pr_err("Unable to determine metadata format for %s\n", devname
);
1837 if (s
->raiddisks
> st
->max_devs
) {
1838 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1841 if (s
->level
== 0 &&
1842 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) &&
1843 !(array
.state
& (1<<MD_SB_CLUSTERED
))) {
1844 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
1845 if (md_set_array_info(fd
, &array
)!= 0) {
1846 pr_err("failed to remove internal bitmap.\n");
1851 /* in the external case we need to check that the requested reshape is
1852 * supported, and perform an initial check that the container holds the
1853 * pre-requisite spare devices (mdmon owns final validation)
1855 if (st
->ss
->external
) {
1859 container
= st
->container_devnm
;
1860 cfd
= open_dev_excl(st
->container_devnm
);
1862 container
= st
->devnm
;
1864 cfd
= open_dev_excl(st
->devnm
);
1868 pr_err("Unable to open container for %s\n", devname
);
1873 retval
= st
->ss
->load_container(st
, cfd
, NULL
);
1876 pr_err("Cannot read superblock for %s\n", devname
);
1881 /* check if operation is supported for metadata handler */
1882 if (st
->ss
->container_content
) {
1883 struct mdinfo
*cc
= NULL
;
1884 struct mdinfo
*content
= NULL
;
1886 cc
= st
->ss
->container_content(st
, subarray
);
1887 for (content
= cc
; content
; content
= content
->next
) {
1888 int allow_reshape
= 1;
1890 /* check if reshape is allowed based on metadata
1891 * indications stored in content.array.status
1893 if (content
->array
.state
&
1894 (1 << MD_SB_BLOCK_VOLUME
))
1896 if (content
->array
.state
&
1897 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE
))
1899 if (!allow_reshape
) {
1900 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1901 devname
, container
);
1906 if (content
->consistency_policy
==
1907 CONSISTENCY_POLICY_PPL
) {
1908 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1916 if (mdmon_running(container
))
1917 st
->update_tail
= &st
->updates
;
1921 for (dv
= devlist
; dv
; dv
= dv
->next
)
1923 if (s
->raiddisks
> array
.raid_disks
&&
1924 array
.spare_disks
+ added_disks
<
1925 (s
->raiddisks
- array
.raid_disks
) &&
1927 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1928 " Use --force to over-ride this check.\n",
1929 s
->raiddisks
- array
.raid_disks
,
1930 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1931 array
.spare_disks
+ added_disks
);
1935 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1936 GET_STATE
| GET_VERSION
);
1938 if (st
->ss
->external
&& subarray
== NULL
) {
1939 array
.level
= LEVEL_CONTAINER
;
1940 sra
->array
.level
= LEVEL_CONTAINER
;
1943 pr_err("failed to read sysfs parameters for %s\n",
1947 frozen
= freeze(st
);
1949 /* freeze() already spewed the reason */
1952 } else if (frozen
< 0) {
1953 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1958 /* ========= set size =============== */
1960 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1961 unsigned long long orig_size
= get_component_size(fd
)/2;
1962 unsigned long long min_csize
;
1964 int raid0_takeover
= 0;
1967 orig_size
= (unsigned) array
.size
;
1969 if (orig_size
== 0) {
1970 pr_err("Cannot set device size in this type of array.\n");
1975 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1976 devname
, APPLY_METADATA_CHANGES
,
1982 if (st
->ss
->external
) {
1983 /* metadata can have size limitation
1984 * update size value according to metadata information
1986 struct mdinfo
*sizeinfo
=
1987 st
->ss
->container_content(st
, subarray
);
1989 unsigned long long new_size
=
1990 sizeinfo
->custom_array_size
/2;
1991 int data_disks
= get_data_disks(
1992 sizeinfo
->array
.level
,
1993 sizeinfo
->array
.layout
,
1994 sizeinfo
->array
.raid_disks
);
1995 new_size
/= data_disks
;
1996 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1997 orig_size
, new_size
,
1998 new_size
* data_disks
);
2000 sysfs_free(sizeinfo
);
2004 /* Update the size of each member device in case
2005 * they have been resized. This will never reduce
2006 * below the current used-size. The "size" attribute
2007 * understands '0' to mean 'max'.
2010 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2011 sysfs_set_num(sra
, mdi
, "size",
2012 s
->size
== MAX_SIZE
? 0 : s
->size
);
2013 if (array
.not_persistent
== 0 &&
2014 array
.major_version
== 0 &&
2015 get_linux_version() < 3001000) {
2016 /* Dangerous to allow size to exceed 2TB */
2017 unsigned long long csize
;
2018 if (sysfs_get_ll(sra
, mdi
, "size",
2020 if (csize
>= 2ULL*1024*1024*1024)
2021 csize
= 2ULL*1024*1024*1024;
2022 if ((min_csize
== 0 ||
2023 (min_csize
> csize
)))
2028 if (min_csize
&& s
->size
> min_csize
) {
2029 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2031 goto size_change_error
;
2033 if (min_csize
&& s
->size
== MAX_SIZE
) {
2034 /* Don't let the kernel choose a size - it will get
2037 pr_err("Limited v0.90 array to 2TB per device\n");
2038 s
->size
= min_csize
;
2040 if (st
->ss
->external
) {
2041 if (sra
->array
.level
== 0) {
2042 rv
= sysfs_set_str(sra
, NULL
, "level", "raid5");
2045 /* get array parameters after takeover
2046 * to change one parameter at time only
2048 rv
= md_get_array_info(fd
, &array
);
2051 /* make sure mdmon is
2052 * aware of the new level */
2053 if (!mdmon_running(st
->container_devnm
))
2054 start_mdmon(st
->container_devnm
);
2055 ping_monitor(container
);
2056 if (mdmon_running(st
->container_devnm
) &&
2057 st
->update_tail
== NULL
)
2058 st
->update_tail
= &st
->updates
;
2061 if (s
->size
== MAX_SIZE
)
2063 array
.size
= s
->size
;
2064 if (s
->size
& ~INT32_MAX
) {
2065 /* got truncated to 32bit, write to
2066 * component_size instead
2069 rv
= sysfs_set_num(sra
, NULL
,
2070 "component_size", s
->size
);
2074 rv
= md_set_array_info(fd
, &array
);
2076 /* manage array size when it is managed externally
2078 if ((rv
== 0) && st
->ss
->external
)
2079 rv
= set_array_size(st
, sra
, sra
->text_version
);
2082 if (raid0_takeover
) {
2083 /* do not recync non-existing parity,
2084 * we will drop it anyway
2086 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2087 /* go back to raid0, drop parity disk
2089 sysfs_set_str(sra
, NULL
, "level", "raid0");
2090 md_get_array_info(fd
, &array
);
2097 /* restore metadata */
2098 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2099 UnSet
, NULL
, devname
,
2100 ROLLBACK_METADATA_CHANGES
,
2103 pr_err("Cannot set device size for %s: %s\n",
2104 devname
, strerror(err
));
2106 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2107 cont_err("Bitmap must be removed before size can be changed\n");
2111 if (s
->assume_clean
) {
2112 /* This will fail on kernels older than 3.0 unless
2113 * a backport has been arranged.
2116 sysfs_set_str(sra
, NULL
, "resync_start",
2118 pr_err("--assume-clean not supported with --grow on this kernel\n");
2120 md_get_array_info(fd
, &array
);
2121 s
->size
= get_component_size(fd
)/2;
2123 s
->size
= array
.size
;
2124 if (c
->verbose
>= 0) {
2125 if (s
->size
== orig_size
)
2126 pr_err("component size of %s unchanged at %lluK\n",
2129 pr_err("component size of %s has been set to %lluK\n",
2133 } else if (array
.level
!= LEVEL_CONTAINER
) {
2134 s
->size
= get_component_size(fd
)/2;
2136 s
->size
= array
.size
;
2139 /* See if there is anything else to do */
2140 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2141 (s
->layout_str
== NULL
) &&
2142 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2143 data_offset
== INVALID_SECTORS
&&
2144 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2145 /* Nothing more to do */
2146 if (!changed
&& c
->verbose
>= 0)
2147 pr_err("%s: no change requested\n", devname
);
2151 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2152 * current implementation assumes that following conditions must be met:
2155 * - near_copies == 2
2157 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2158 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2159 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2162 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2164 dprintf("Array cannot be reshaped\n");
2170 /* Make sure mdmon has seen the device removal
2171 * and updated metadata before we continue with
2175 ping_monitor(container
);
2178 memset(&info
, 0, sizeof(info
));
2180 if (sysfs_init(&info
, fd
, NULL
)) {
2181 pr_err("failed to initialize sysfs.\n");
2185 strcpy(info
.text_version
, sra
->text_version
);
2186 info
.component_size
= s
->size
*2;
2187 info
.new_level
= s
->level
;
2188 info
.new_chunk
= s
->chunk
* 1024;
2189 if (info
.array
.level
== LEVEL_CONTAINER
) {
2190 info
.delta_disks
= UnSet
;
2191 info
.array
.raid_disks
= s
->raiddisks
;
2192 } else if (s
->raiddisks
)
2193 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2195 info
.delta_disks
= UnSet
;
2196 if (s
->layout_str
== NULL
) {
2197 info
.new_layout
= UnSet
;
2198 if (info
.array
.level
== 6 &&
2199 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2200 info
.array
.layout
>= 16) {
2201 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2202 cont_err("during the reshape, please specify --layout=preserve\n");
2203 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2207 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2208 strcmp(s
->layout_str
, "normalize") == 0) {
2209 /* If we have a -6 RAID6 layout, remove the '-6'. */
2210 info
.new_layout
= UnSet
;
2211 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2213 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
2214 h
= strrchr(l
, '-');
2215 if (h
&& strcmp(h
, "-6") == 0) {
2217 info
.new_layout
= map_name(r6layout
, l
);
2220 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2224 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2225 /* This means that a non-standard RAID6 layout
2228 * - When reshape a RAID6 (e.g. adding a device)
2229 * which is in a non-standard layout, it is OK
2230 * to preserve that layout.
2231 * - When converting a RAID5 to RAID6, leave it in
2232 * the XXX-6 layout, don't re-layout.
2234 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2235 info
.new_layout
= info
.array
.layout
;
2236 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2238 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
2240 info
.new_layout
= map_name(r6layout
, l
);
2242 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2247 int l
= info
.new_level
;
2249 l
= info
.array
.level
;
2252 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2255 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2258 info
.new_layout
= parse_layout_10(s
->layout_str
);
2261 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2264 pr_err("layout not meaningful with this level\n");
2268 if (info
.new_layout
== UnSet
) {
2269 pr_err("layout %s not understood for this level\n",
2276 if (array
.level
== LEVEL_FAULTY
) {
2277 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2278 pr_err("cannot change level of Faulty device\n");
2282 pr_err("cannot set chunksize of Faulty device\n");
2285 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2286 pr_err("cannot set raid_disks of Faulty device\n");
2289 if (s
->layout_str
) {
2290 if (md_get_array_info(fd
, &array
) != 0) {
2291 dprintf("Cannot get array information.\n");
2294 array
.layout
= info
.new_layout
;
2295 if (md_set_array_info(fd
, &array
) != 0) {
2296 pr_err("failed to set new layout\n");
2298 } else if (c
->verbose
>= 0)
2299 printf("layout for %s set to %d\n",
2300 devname
, array
.layout
);
2302 } else if (array
.level
== LEVEL_CONTAINER
) {
2303 /* This change is to be applied to every array in the
2304 * container. This is only needed when the metadata imposes
2305 * restraints of the various arrays in the container.
2306 * Currently we only know that IMSM requires all arrays
2307 * to have the same number of devices so changing the
2308 * number of devices (On-Line Capacity Expansion) must be
2309 * performed at the level of the container
2315 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2316 c
->force
, c
->backup_file
, c
->verbose
,
2320 /* get spare devices from external metadata
2322 if (st
->ss
->external
) {
2323 struct mdinfo
*info2
;
2325 info2
= st
->ss
->container_content(st
, subarray
);
2327 info
.array
.spare_disks
=
2328 info2
->array
.spare_disks
;
2333 /* Impose these changes on a single array. First
2334 * check that the metadata is OK with the change. */
2336 if (reshape_super(st
, 0, info
.new_level
,
2337 info
.new_layout
, info
.new_chunk
,
2338 info
.array
.raid_disks
, info
.delta_disks
,
2339 c
->backup_file
, devname
,
2340 APPLY_METADATA_CHANGES
, c
->verbose
)) {
2345 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2346 devlist
, data_offset
, c
->backup_file
,
2347 c
->verbose
, 0, 0, 0);
2357 /* verify_reshape_position()
2358 * Function checks if reshape position in metadata is not farther
2359 * than position in md.
2361 * 0 : not valid sysfs entry
2362 * it can be caused by not started reshape, it should be started
2363 * by reshape array or raid0 array is before takeover
2364 * -1 : error, reshape position is obviously wrong
2365 * 1 : success, reshape progress correct or updated
2367 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2373 /* read sync_max, failure can mean raid0 array */
2374 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2378 unsigned long long position
= strtoull(buf
, &ep
, 0);
2380 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2381 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2382 position
*= get_data_disks(level
,
2384 info
->array
.raid_disks
);
2385 if (info
->reshape_progress
< position
) {
2386 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2387 info
->reshape_progress
, position
);
2388 info
->reshape_progress
= position
;
2390 } else if (info
->reshape_progress
> position
) {
2391 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2392 position
, info
->reshape_progress
);
2395 dprintf("Reshape position in md and metadata are the same;");
2399 } else if (rv
== 0) {
2400 /* for valid sysfs entry, 0-length content
2401 * should be indicated as error
2409 static unsigned long long choose_offset(unsigned long long lo
,
2410 unsigned long long hi
,
2411 unsigned long long min
,
2412 unsigned long long max
)
2414 /* Choose a new offset between hi and lo.
2415 * It must be between min and max, but
2416 * we would prefer something near the middle of hi/lo, and also
2417 * prefer to be aligned to a big power of 2.
2419 * So we start with the middle, then for each bit,
2420 * starting at '1' and increasing, if it is set, we either
2421 * add it or subtract it if possible, preferring the option
2422 * which is furthest from the boundary.
2424 * We stop once we get a 1MB alignment. As units are in sectors,
2425 * 1MB = 2*1024 sectors.
2427 unsigned long long choice
= (lo
+ hi
) / 2;
2428 unsigned long long bit
= 1;
2430 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2431 unsigned long long bigger
, smaller
;
2432 if (! (bit
& choice
))
2434 bigger
= choice
+ bit
;
2435 smaller
= choice
- bit
;
2436 if (bigger
> max
&& smaller
< min
)
2440 else if (smaller
< min
)
2442 else if (hi
- bigger
> smaller
- lo
)
2450 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2451 char *devname
, int delta_disks
,
2452 unsigned long long data_offset
,
2453 unsigned long long min
,
2459 unsigned long long before
, after
;
2461 /* Need to find min space before and after so same is used
2464 before
= UINT64_MAX
;
2466 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2470 struct supertype
*st2
;
2471 struct mdinfo info2
;
2473 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2475 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2476 dfd
= dev_open(dn
, O_RDONLY
);
2478 pr_err("%s: cannot open component %s\n",
2479 devname
, dn
? dn
: "-unknown-");
2482 st2
= dup_super(st
);
2483 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2487 pr_err("%s: cannot get superblock from %s\n",
2491 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2492 st2
->ss
->free_super(st2
);
2494 if (info2
.space_before
== 0 &&
2495 info2
.space_after
== 0) {
2496 /* Metadata doesn't support data_offset changes */
2498 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2502 if (before
> info2
.space_before
)
2503 before
= info2
.space_before
;
2504 if (after
> info2
.space_after
)
2505 after
= info2
.space_after
;
2507 if (data_offset
!= INVALID_SECTORS
) {
2509 if (info2
.data_offset
== data_offset
) {
2510 pr_err("%s: already has that data_offset\n",
2514 if (data_offset
< info2
.data_offset
)
2518 } else if ((data_offset
<= info2
.data_offset
&&
2520 (data_offset
>= info2
.data_offset
&&
2522 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2528 if (before
== UINT64_MAX
)
2529 /* impossible really, there must be no devices */
2532 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2533 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2534 unsigned long long new_data_offset
;
2536 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2538 if (delta_disks
< 0) {
2539 /* Don't need any space as array is shrinking
2540 * just move data_offset up by min
2542 if (data_offset
== INVALID_SECTORS
)
2543 new_data_offset
= sd
->data_offset
+ min
;
2545 if (data_offset
< sd
->data_offset
+ min
) {
2546 pr_err("--data-offset too small for %s\n",
2550 new_data_offset
= data_offset
;
2552 } else if (delta_disks
> 0) {
2553 /* need space before */
2557 pr_err("Insufficient head-space for reshape on %s\n",
2561 if (data_offset
== INVALID_SECTORS
)
2562 new_data_offset
= sd
->data_offset
- min
;
2564 if (data_offset
> sd
->data_offset
- min
) {
2565 pr_err("--data-offset too large for %s\n",
2569 new_data_offset
= data_offset
;
2573 /* can move up or down. If 'data_offset'
2574 * was set we would have already decided,
2575 * so just choose direction with most space.
2582 sysfs_set_str(sra
, NULL
, "reshape_direction",
2583 dir
== 1 ? "backwards" : "forwards");
2585 /* Increase data offset */
2589 pr_err("Insufficient tail-space for reshape on %s\n",
2593 if (data_offset
!= INVALID_SECTORS
&&
2594 data_offset
< sd
->data_offset
+ min
) {
2595 pr_err("--data-offset too small on %s\n",
2599 if (data_offset
!= INVALID_SECTORS
)
2600 new_data_offset
= data_offset
;
2602 new_data_offset
= choose_offset(sd
->data_offset
,
2603 sd
->data_offset
+ after
,
2604 sd
->data_offset
+ min
,
2605 sd
->data_offset
+ after
);
2607 /* Decrease data offset */
2611 pr_err("insufficient head-room on %s\n",
2615 if (data_offset
!= INVALID_SECTORS
&&
2616 data_offset
> sd
->data_offset
- min
) {
2617 pr_err("--data-offset too large on %s\n",
2621 if (data_offset
!= INVALID_SECTORS
)
2622 new_data_offset
= data_offset
;
2624 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2626 sd
->data_offset
- before
,
2627 sd
->data_offset
- min
);
2630 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2631 if (err
< 0 && errno
== E2BIG
) {
2632 /* try again after increasing data size to max */
2633 err
= sysfs_set_num(sra
, sd
, "size", 0);
2634 if (err
< 0 && errno
== EINVAL
&&
2635 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2636 /* some kernels have a bug where you cannot
2637 * use '0' on spare devices. */
2638 sysfs_set_num(sra
, sd
, "size",
2639 (sra
->component_size
+ after
)/2);
2641 err
= sysfs_set_num(sra
, sd
, "new_offset",
2645 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2646 pr_err("data-offset is too big for %s\n", dn
);
2649 if (sd
== sra
->devs
&&
2650 (errno
== ENOENT
|| errno
== E2BIG
))
2651 /* Early kernel, no 'new_offset' file,
2652 * or kernel doesn't like us.
2653 * For RAID5/6 this is not fatal
2656 pr_err("Cannot set new_offset for %s\n", dn
);
2664 /* Just use a backup file */
2668 static int raid10_reshape(char *container
, int fd
, char *devname
,
2669 struct supertype
*st
, struct mdinfo
*info
,
2670 struct reshape
*reshape
,
2671 unsigned long long data_offset
,
2672 int force
, int verbose
)
2674 /* Changing raid_disks, layout, chunksize or possibly
2675 * just data_offset for a RAID10.
2676 * We must always change data_offset. We change by at least
2677 * ->min_offset_change which is the largest of the old and new
2679 * If raid_disks is increasing, then data_offset must decrease
2680 * by at least this copy size.
2681 * If raid_disks is unchanged, data_offset must increase or
2682 * decrease by at least min_offset_change but preferably by much more.
2683 * We choose half of the available space.
2684 * If raid_disks is decreasing, data_offset must increase by
2685 * at least min_offset_change. To allow of this, component_size
2686 * must be decreased by the same amount.
2688 * So we calculate the required minimum and direction, possibly
2689 * reduce the component_size, then iterate through the devices
2690 * and set the new_data_offset.
2691 * If that all works, we set chunk_size, layout, raid_disks, and start
2695 unsigned long long min
;
2698 sra
= sysfs_read(fd
, NULL
,
2699 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2702 pr_err("%s: Cannot get array details from sysfs\n", devname
);
2705 min
= reshape
->min_offset_change
;
2707 if (info
->delta_disks
)
2708 sysfs_set_str(sra
, NULL
, "reshape_direction",
2709 info
->delta_disks
< 0 ? "backwards" : "forwards");
2710 if (info
->delta_disks
< 0 && info
->space_after
< min
) {
2711 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2712 (sra
->component_size
- min
)/2);
2714 pr_err("cannot reduce component size\n");
2718 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
,
2719 data_offset
, min
, 0);
2721 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2722 cont_err("supported on this kernel\n");
2728 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2730 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2731 reshape
->after
.layout
) < 0)
2734 sysfs_set_num(sra
, NULL
, "raid_disks",
2735 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2737 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2740 pr_err("Cannot set array shape for %s\n",
2743 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2744 cont_err(" Bitmap must be removed before shape can be changed\n");
2754 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2756 struct mdinfo
*sra
, *sd
;
2757 /* Initialisation to silence compiler warning */
2758 unsigned long long min_space_before
= 0, min_space_after
= 0;
2761 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2764 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2767 struct supertype
*st2
;
2768 struct mdinfo info2
;
2770 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2772 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2773 dfd
= dev_open(dn
, O_RDONLY
);
2776 st2
= dup_super(st
);
2777 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2783 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2784 st2
->ss
->free_super(st2
);
2787 min_space_before
> info2
.space_before
)
2788 min_space_before
= info2
.space_before
;
2790 min_space_after
> info2
.space_after
)
2791 min_space_after
= info2
.space_after
;
2794 if (sd
== NULL
&& !first
) {
2795 info
->space_after
= min_space_after
;
2796 info
->space_before
= min_space_before
;
2801 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2802 struct mdinfo
*info
,
2803 int disks
, unsigned long long blocks
)
2805 /* Check that the internal stripe cache is
2806 * large enough, or it won't work.
2807 * It must hold at least 4 stripes of the larger
2810 unsigned long cache
;
2811 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2812 cache
*= 4; /* 4 stripes minimum */
2813 cache
/= 512; /* convert to sectors */
2814 /* make sure there is room for 'blocks' with a bit to spare */
2815 if (cache
< 16 + blocks
/ disks
)
2816 cache
= 16 + blocks
/ disks
;
2817 cache
/= (4096/512); /* Convert from sectors to pages */
2819 if (sra
->cache_size
< cache
)
2820 subarray_set_num(container
, sra
, "stripe_cache_size",
2824 static int impose_reshape(struct mdinfo
*sra
,
2825 struct mdinfo
*info
,
2826 struct supertype
*st
,
2829 char *devname
, char *container
,
2830 struct reshape
*reshape
)
2832 struct mdu_array_info_s array
;
2834 sra
->new_chunk
= info
->new_chunk
;
2837 /* for external metadata checkpoint saved by mdmon can be lost
2838 * or missed /due to e.g. crash/. Check if md is not during
2839 * restart farther than metadata points to.
2840 * If so, this means metadata information is obsolete.
2842 if (st
->ss
->external
)
2843 verify_reshape_position(info
, reshape
->level
);
2844 sra
->reshape_progress
= info
->reshape_progress
;
2846 sra
->reshape_progress
= 0;
2847 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2848 /* start from the end of the new array */
2849 sra
->reshape_progress
= (sra
->component_size
2850 * reshape
->after
.data_disks
);
2853 md_get_array_info(fd
, &array
);
2854 if (info
->array
.chunk_size
== info
->new_chunk
&&
2855 reshape
->before
.layout
== reshape
->after
.layout
&&
2856 st
->ss
->external
== 0) {
2857 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2858 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2859 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2862 pr_err("Cannot set device shape for %s: %s\n",
2863 devname
, strerror(errno
));
2866 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2867 cont_err("Bitmap must be removed before shape can be changed\n");
2871 } else if (!restart
) {
2872 /* set them all just in case some old 'new_*' value
2873 * persists from some earlier problem.
2876 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2878 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2879 reshape
->after
.layout
) < 0)
2881 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2882 reshape
->after
.data_disks
+
2883 reshape
->parity
) < 0)
2886 pr_err("Cannot set device shape for %s\n", devname
);
2889 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2890 cont_err("Bitmap must be removed before shape can be changed\n");
2899 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2902 struct mdu_array_info_s array
;
2905 if (sysfs_init(&info
, fd
, NULL
)) {
2906 pr_err("failed to initialize sysfs.\n");
2910 md_get_array_info(fd
, &array
);
2911 if (level
== 0 && (array
.level
>= 4 && array
.level
<= 6)) {
2912 /* To convert to RAID0 we need to fail and
2913 * remove any non-data devices. */
2916 int data_disks
= array
.raid_disks
- 1;
2917 if (array
.level
== 6)
2919 if (array
.level
== 5 && array
.layout
!= ALGORITHM_PARITY_N
)
2921 if (array
.level
== 6 && array
.layout
!= ALGORITHM_PARITY_N_6
)
2923 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2924 /* First remove any spares so no recovery starts */
2925 for (d
= 0, found
= 0;
2926 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2927 mdu_disk_info_t disk
;
2929 if (md_get_disk_info(fd
, &disk
) < 0)
2931 if (disk
.major
== 0 && disk
.minor
== 0)
2934 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2935 disk
.raid_disk
< data_disks
)
2938 ioctl(fd
, HOT_REMOVE_DISK
,
2939 makedev(disk
.major
, disk
.minor
));
2941 /* Now fail anything left */
2942 md_get_array_info(fd
, &array
);
2943 for (d
= 0, found
= 0;
2944 d
< MAX_DISKS
&& found
< array
.nr_disks
; d
++) {
2945 mdu_disk_info_t disk
;
2947 if (md_get_disk_info(fd
, &disk
) < 0)
2949 if (disk
.major
== 0 && disk
.minor
== 0)
2952 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2953 disk
.raid_disk
< data_disks
)
2956 ioctl(fd
, SET_DISK_FAULTY
,
2957 makedev(disk
.major
, disk
.minor
));
2958 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2961 c
= map_num(pers
, level
);
2963 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2966 pr_err("%s: could not set level to %s\n",
2969 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2970 cont_err("Bitmap must be removed before level can be changed\n");
2974 pr_err("level of %s changed to %s\n", devname
, c
);
2980 static void catch_term(int sig
)
2985 static int continue_via_systemd(char *devnm
)
2987 int skipped
, i
, pid
, status
;
2989 /* In a systemd/udev world, it is best to get systemd to
2990 * run "mdadm --grow --continue" rather than running in the
2995 /* FIXME yuk. CLOSE_EXEC?? */
2997 for (i
= 3; skipped
< 20; i
++)
3003 /* Don't want to see error messages from
3004 * systemctl. If the service doesn't exist,
3005 * we fork ourselves.
3008 open("/dev/null", O_WRONLY
);
3009 snprintf(pathbuf
, sizeof(pathbuf
),
3010 "mdadm-grow-continue@%s.service", devnm
);
3011 status
= execl("/usr/bin/systemctl", "systemctl", "restart",
3013 status
= execl("/bin/systemctl", "systemctl", "restart",
3016 case -1: /* Just do it ourselves. */
3018 default: /* parent - good */
3019 pid
= wait(&status
);
3020 if (pid
>= 0 && status
== 0)
3026 static int reshape_array(char *container
, int fd
, char *devname
,
3027 struct supertype
*st
, struct mdinfo
*info
,
3028 int force
, struct mddev_dev
*devlist
,
3029 unsigned long long data_offset
,
3030 char *backup_file
, int verbose
, int forked
,
3031 int restart
, int freeze_reshape
)
3033 struct reshape reshape
;
3036 int orig_level
= UnSet
;
3040 struct mdu_array_info_s array
;
3043 struct mddev_dev
*dv
;
3047 unsigned long long *offsets
= NULL
;
3051 unsigned long blocks
;
3052 unsigned long long array_size
;
3054 struct mdinfo
*sra
= NULL
;
3057 /* when reshaping a RAID0, the component_size might be zero.
3058 * So try to fix that up.
3060 if (md_get_array_info(fd
, &array
) != 0) {
3061 dprintf("Cannot get array information.\n");
3064 if (array
.level
== 0 && info
->component_size
== 0) {
3065 get_dev_size(fd
, NULL
, &array_size
);
3066 info
->component_size
= array_size
/ array
.raid_disks
;
3069 if (array
.level
== 10)
3070 /* Need space_after info */
3071 get_space_after(fd
, st
, info
);
3073 if (info
->reshape_active
) {
3074 int new_level
= info
->new_level
;
3075 info
->new_level
= UnSet
;
3076 if (info
->delta_disks
> 0)
3077 info
->array
.raid_disks
-= info
->delta_disks
;
3078 msg
= analyse_change(devname
, info
, &reshape
);
3079 info
->new_level
= new_level
;
3080 if (info
->delta_disks
> 0)
3081 info
->array
.raid_disks
+= info
->delta_disks
;
3083 /* Make sure the array isn't read-only */
3084 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3086 msg
= analyse_change(devname
, info
, &reshape
);
3088 /* if msg == "", error has already been printed */
3090 pr_err("%s\n", msg
);
3093 if (restart
&& (reshape
.level
!= info
->array
.level
||
3094 reshape
.before
.layout
!= info
->array
.layout
||
3095 reshape
.before
.data_disks
+ reshape
.parity
!=
3096 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3097 pr_err("reshape info is not in native format - cannot continue.\n");
3101 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3102 !((sysfs_get_str(info
, NULL
, "sync_action",
3103 buf
, sizeof(buf
)) > 0) &&
3104 (strncmp(buf
, "reshape", 7) == 0))) {
3105 /* When reshape is restarted from '0', very begin of array
3106 * it is possible that for external metadata reshape and array
3107 * configuration doesn't happen.
3108 * Check if md has the same opinion, and reshape is restarted
3109 * from 0. If so, this is regular reshape start after reshape
3110 * switch in metadata to next array only.
3112 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3113 (info
->reshape_progress
== 0))
3118 * reshape already started. just skip to monitoring
3121 if (reshape
.backup_blocks
== 0)
3123 if (restart
& RESHAPE_NO_BACKUP
)
3126 /* Need 'sra' down at 'started:' */
3127 sra
= sysfs_read(fd
, NULL
,
3128 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
3129 GET_CHUNK
|GET_CACHE
);
3131 pr_err("%s: Cannot get array details from sysfs\n",
3137 backup_file
= locate_backup(sra
->sys_name
);
3141 /* The container is frozen but the array may not be.
3142 * So freeze the array so spares don't get put to the wrong use
3143 * FIXME there should probably be a cleaner separation between
3144 * freeze_array and freeze_container.
3146 sysfs_freeze_array(info
);
3147 /* Check we have enough spares to not be degraded */
3149 for (dv
= devlist
; dv
; dv
=dv
->next
)
3151 spares_needed
= max(reshape
.before
.data_disks
,
3152 reshape
.after
.data_disks
) +
3153 reshape
.parity
- array
.raid_disks
;
3155 if (!force
&& info
->new_level
> 1 && info
->array
.level
> 1 &&
3156 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3157 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3158 " Use --force to over-ride this check.\n",
3160 spares_needed
== 1 ? "" : "s",
3161 info
->array
.spare_disks
+ added_disks
);
3164 /* Check we have enough spares to not fail */
3165 spares_needed
= max(reshape
.before
.data_disks
,
3166 reshape
.after
.data_disks
)
3168 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3169 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3170 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3171 spares_needed
, spares_needed
== 1 ? "" : "s",
3172 info
->array
.spare_disks
+ added_disks
);
3176 if (reshape
.level
!= array
.level
) {
3177 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3180 info
->new_layout
= UnSet
; /* after level change,
3181 * layout is meaningless */
3182 orig_level
= array
.level
;
3183 sysfs_freeze_array(info
);
3185 if (reshape
.level
> 0 && st
->ss
->external
) {
3186 /* make sure mdmon is aware of the new level */
3187 if (mdmon_running(container
))
3188 flush_mdmon(container
);
3190 if (!mdmon_running(container
))
3191 start_mdmon(container
);
3192 ping_monitor(container
);
3193 if (mdmon_running(container
) && st
->update_tail
== NULL
)
3194 st
->update_tail
= &st
->updates
;
3197 /* ->reshape_super might have chosen some spares from the
3198 * container that it wants to be part of the new array.
3199 * We can collect them with ->container_content and give
3200 * them to the kernel.
3202 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3203 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3204 struct mdinfo
*info2
=
3205 st
->ss
->container_content(st
, subarray
);
3209 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3210 pr_err("unable to initialize sysfs for %s\n",
3215 /* When increasing number of devices, we need to set
3216 * new raid_disks before adding these, or they might
3219 if (reshape
.backup_blocks
&&
3220 reshape
.after
.data_disks
>
3221 reshape
.before
.data_disks
)
3222 subarray_set_num(container
, info2
, "raid_disks",
3223 reshape
.after
.data_disks
+
3225 for (d
= info2
->devs
; d
; d
= d
->next
) {
3226 if (d
->disk
.state
== 0 &&
3227 d
->disk
.raid_disk
>= 0) {
3228 /* This is a spare that wants to
3229 * be part of the array.
3231 add_disk(fd
, st
, info2
, d
);
3237 /* We might have been given some devices to add to the
3238 * array. Now that the array has been changed to the right
3239 * level and frozen, we can safely add them.
3242 if (Manage_subdevs(devname
, fd
, devlist
, verbose
, 0, NULL
, 0))
3246 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3247 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3248 if (reshape
.backup_blocks
== 0) {
3249 /* No restriping needed, but we might need to impose
3250 * some more changes: layout, raid_disks, chunk_size
3252 /* read current array info */
3253 if (md_get_array_info(fd
, &array
) != 0) {
3254 dprintf("Cannot get array information.\n");
3257 /* compare current array info with new values and if
3258 * it is different update them to new */
3259 if (info
->new_layout
!= UnSet
&&
3260 info
->new_layout
!= array
.layout
) {
3261 array
.layout
= info
->new_layout
;
3262 if (md_set_array_info(fd
, &array
) != 0) {
3263 pr_err("failed to set new layout\n");
3265 } else if (verbose
>= 0)
3266 printf("layout for %s set to %d\n",
3267 devname
, array
.layout
);
3269 if (info
->delta_disks
!= UnSet
&& info
->delta_disks
!= 0 &&
3271 (info
->array
.raid_disks
+ info
->delta_disks
)) {
3272 array
.raid_disks
+= info
->delta_disks
;
3273 if (md_set_array_info(fd
, &array
) != 0) {
3274 pr_err("failed to set raid disks\n");
3276 } else if (verbose
>= 0) {
3277 printf("raid_disks for %s set to %d\n",
3278 devname
, array
.raid_disks
);
3281 if (info
->new_chunk
!= 0 &&
3282 info
->new_chunk
!= array
.chunk_size
) {
3283 if (sysfs_set_num(info
, NULL
,
3284 "chunk_size", info
->new_chunk
) != 0) {
3285 pr_err("failed to set chunk size\n");
3287 } else if (verbose
>= 0)
3288 printf("chunk size for %s set to %d\n",
3289 devname
, info
->new_chunk
);
3296 * There are three possibilities.
3297 * 1/ The array will shrink.
3298 * We need to ensure the reshape will pause before reaching
3299 * the 'critical section'. We also need to fork and wait for
3300 * that to happen. When it does we
3301 * suspend/backup/complete/unfreeze
3303 * 2/ The array will not change size.
3304 * This requires that we keep a backup of a sliding window
3305 * so that we can restore data after a crash. So we need
3306 * to fork and monitor progress.
3307 * In future we will allow the data_offset to change, so
3308 * a sliding backup becomes unnecessary.
3310 * 3/ The array will grow. This is relatively easy.
3311 * However the kernel's restripe routines will cheerfully
3312 * overwrite some early data before it is safe. So we
3313 * need to make a backup of the early parts of the array
3314 * and be ready to restore it if rebuild aborts very early.
3315 * For externally managed metadata, we still need a forked
3316 * child to monitor the reshape and suspend IO over the region
3317 * that is being reshaped.
3319 * We backup data by writing it to one spare, or to a
3320 * file which was given on command line.
3322 * In each case, we first make sure that storage is available
3323 * for the required backup.
3325 * - request the shape change.
3326 * - fork to handle backup etc.
3328 /* Check that we can hold all the data */
3329 get_dev_size(fd
, NULL
, &array_size
);
3330 if (reshape
.new_size
< (array_size
/512)) {
3331 pr_err("this change will reduce the size of the array.\n"
3332 " use --grow --array-size first to truncate array.\n"
3333 " e.g. mdadm --grow %s --array-size %llu\n",
3334 devname
, reshape
.new_size
/2);
3338 if (array
.level
== 10) {
3339 /* Reshaping RAID10 does not require any data backup by
3340 * user-space. Instead it requires that the data_offset
3341 * is changed to avoid the need for backup.
3342 * So this is handled very separately
3345 /* Nothing to do. */
3347 return raid10_reshape(container
, fd
, devname
, st
, info
,
3348 &reshape
, data_offset
, force
, verbose
);
3350 sra
= sysfs_read(fd
, NULL
,
3351 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3354 pr_err("%s: Cannot get array details from sysfs\n",
3360 switch(set_new_data_offset(sra
, st
, devname
,
3361 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3363 reshape
.min_offset_change
, 1)) {
3367 /* Updated data_offset, so it's easy now */
3368 update_cache_size(container
, sra
, info
,
3369 min(reshape
.before
.data_disks
,
3370 reshape
.after
.data_disks
),
3371 reshape
.backup_blocks
);
3373 /* Right, everything seems fine. Let's kick things off.
3377 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3378 devname
, container
, &reshape
) < 0)
3380 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3382 if (errno
!= EINVAL
) {
3383 pr_err("Failed to initiate reshape!\n");
3386 /* revert data_offset and try the old way */
3387 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3388 sysfs_set_num(sra
, sd
, "new_offset",
3390 sysfs_set_str(sra
, NULL
, "reshape_direction",
3395 if (info
->new_level
== reshape
.level
)
3397 /* need to adjust level when reshape completes */
3399 case -1: /* ignore error, but don't wait */
3401 default: /* parent */
3409 fd
= open_dev(sra
->sys_name
);
3411 impose_level(fd
, info
->new_level
, devname
, verbose
);
3413 case 1: /* Couldn't set data_offset, try the old way */
3414 if (data_offset
!= INVALID_SECTORS
) {
3415 pr_err("Cannot update data_offset on this array\n");
3422 /* Decide how many blocks (sectors) for a reshape
3423 * unit. The number we have so far is just a minimum
3425 blocks
= reshape
.backup_blocks
;
3426 if (reshape
.before
.data_disks
==
3427 reshape
.after
.data_disks
) {
3428 /* Make 'blocks' bigger for better throughput, but
3429 * not so big that we reject it below.
3430 * Try for 16 megabytes
3432 while (blocks
* 32 < sra
->component_size
&& blocks
< 16*1024*2)
3435 pr_err("Need to backup %luK of critical section..\n", blocks
/2);
3437 if (blocks
>= sra
->component_size
/2) {
3438 pr_err("%s: Something wrong - reshape aborted\n", devname
);
3442 /* Now we need to open all these devices so we can read/write.
3444 nrdisks
= max(reshape
.before
.data_disks
,
3445 reshape
.after
.data_disks
) + reshape
.parity
3446 + sra
->array
.spare_disks
;
3447 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3448 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3450 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3451 d
= reshape_prepare_fdlist(devname
, sra
, odisks
, nrdisks
, blocks
,
3452 backup_file
, fdlist
, offsets
);
3456 if ((st
->ss
->manage_reshape
== NULL
) ||
3457 (st
->ss
->recover_backup
== NULL
)) {
3458 if (backup_file
== NULL
) {
3459 if (reshape
.after
.data_disks
<=
3460 reshape
.before
.data_disks
) {
3461 pr_err("%s: Cannot grow - need backup-file\n",
3463 pr_err(" Please provide one with \"--backup=...\"\n");
3465 } else if (d
== odisks
) {
3466 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3470 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3472 fdlist
+d
, offsets
+d
,
3473 sra
->sys_name
, restart
)) {
3480 update_cache_size(container
, sra
, info
,
3481 min(reshape
.before
.data_disks
,
3482 reshape
.after
.data_disks
), blocks
);
3484 /* Right, everything seems fine. Let's kick things off.
3485 * If only changing raid_disks, use ioctl, else use
3490 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3491 devname
, container
, &reshape
) < 0)
3494 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3495 reshape
.after
.data_disks
);
3497 pr_err("Cannot %s reshape for %s\n",
3498 restart
? "continue" : "start", devname
);
3502 sysfs_set_str(sra
, NULL
, "array_state", "active");
3503 if (freeze_reshape
) {
3507 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3508 sra
->reshape_progress
);
3512 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3513 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3521 /* Now we just need to kick off the reshape and watch, while
3522 * handling backups of the data...
3523 * This is all done by a forked background process.
3525 switch(forked
? 0 : fork()) {
3527 pr_err("Cannot run child to monitor reshape: %s\n",
3541 /* If another array on the same devices is busy, the
3542 * reshape will wait for them. This would mean that
3543 * the first section that we suspend will stay suspended
3544 * for a long time. So check on that possibility
3545 * by looking for "DELAYED" in /proc/mdstat, and if found,
3549 struct mdstat_ent
*mds
, *m
;
3551 mds
= mdstat_read(1, 0);
3552 for (m
= mds
; m
; m
= m
->next
)
3553 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3554 if (m
->resync
&& m
->percent
== RESYNC_DELAYED
)
3557 /* Haven't started the reshape thread
3564 if (delayed
== 1 && get_linux_version() < 3007000) {
3565 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3566 " You might experience problems until other reshapes complete.\n");
3570 mdstat_wait(30 - (delayed
-1) * 25);
3573 if (check_env("MDADM_GROW_VERIFY"))
3574 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3577 mlockall(MCL_FUTURE
);
3579 signal(SIGTERM
, catch_term
);
3581 if (st
->ss
->external
) {
3582 /* metadata handler takes it from here */
3583 done
= st
->ss
->manage_reshape(
3584 fd
, sra
, &reshape
, st
, blocks
,
3585 fdlist
, offsets
, d
- odisks
, fdlist
+ odisks
,
3588 done
= child_monitor(
3589 fd
, sra
, &reshape
, st
, blocks
, fdlist
, offsets
,
3590 d
- odisks
, fdlist
+ odisks
, offsets
+ odisks
);
3595 if (backup_file
&& done
) {
3597 bul
= make_backup(sra
->sys_name
);
3600 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3608 unlink(backup_file
);
3615 if (!st
->ss
->external
&&
3616 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3617 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3619 /* no need to wait for the reshape to finish as
3620 * there is nothing more to do.
3627 if (st
->ss
->external
) {
3628 /* Re-load the metadata as much could have changed */
3629 int cfd
= open_dev(st
->container_devnm
);
3631 flush_mdmon(container
);
3632 st
->ss
->free_super(st
);
3633 st
->ss
->load_container(st
, cfd
, container
);
3638 /* set new array size if required customer_array_size is used
3641 if (reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3642 info
->custom_array_size
)
3643 set_array_size(st
, info
, info
->text_version
);
3645 if (info
->new_level
!= reshape
.level
) {
3647 fd
= open(devname
, O_RDONLY
);
3648 impose_level(fd
, info
->new_level
, devname
, verbose
);
3650 if (info
->new_level
== 0)
3651 st
->update_tail
= NULL
;
3663 if (orig_level
!= UnSet
&& sra
) {
3664 c
= map_num(pers
, orig_level
);
3665 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3666 pr_err("aborting level change\n");
3674 /* mdfd handle is passed to be closed in child process (after fork).
3676 int reshape_container(char *container
, char *devname
,
3678 struct supertype
*st
,
3679 struct mdinfo
*info
,
3681 char *backup_file
, int verbose
,
3682 int forked
, int restart
, int freeze_reshape
)
3684 struct mdinfo
*cc
= NULL
;
3686 char last_devnm
[32] = "";
3688 /* component_size is not meaningful for a container,
3689 * so pass '0' meaning 'no change'
3692 reshape_super(st
, 0, info
->new_level
,
3693 info
->new_layout
, info
->new_chunk
,
3694 info
->array
.raid_disks
, info
->delta_disks
,
3695 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3703 /* ping monitor to be sure that update is on disk
3705 ping_monitor(container
);
3707 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3708 if (continue_via_systemd(container
))
3711 switch (forked
? 0 : fork()) {
3712 case -1: /* error */
3713 perror("Cannot fork to complete reshape\n");
3716 default: /* parent */
3717 if (!freeze_reshape
)
3718 printf("%s: multi-array reshape continues in background\n", Name
);
3725 /* close unused handle in child process
3731 /* For each member array with reshape_active,
3732 * we need to perform the reshape.
3733 * We pick the first array that needs reshaping and
3734 * reshape it. reshape_array() will re-read the metadata
3735 * so the next time through a different array should be
3736 * ready for reshape.
3737 * It is possible that the 'different' array will not
3738 * be assembled yet. In that case we simple exit.
3739 * When it is assembled, the mdadm which assembles it
3740 * will take over the reshape.
3742 struct mdinfo
*content
;
3744 struct mdstat_ent
*mdstat
;
3750 cc
= st
->ss
->container_content(st
, NULL
);
3752 for (content
= cc
; content
; content
= content
->next
) {
3754 if (!content
->reshape_active
)
3757 subarray
= strchr(content
->text_version
+1, '/')+1;
3758 mdstat
= mdstat_by_subdev(subarray
, container
);
3761 if (mdstat
->active
== 0) {
3762 pr_err("Skipping inactive array %s.\n",
3764 free_mdstat(mdstat
);
3773 devid
= devnm2devid(mdstat
->devnm
);
3774 adev
= map_dev(major(devid
), minor(devid
), 0);
3776 adev
= content
->text_version
;
3778 fd
= open_dev(mdstat
->devnm
);
3780 pr_err("Device %s cannot be opened for reshape.\n",
3785 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3786 /* Do not allow for multiple reshape_array() calls for
3788 * It can happen when reshape_array() returns without
3789 * error, when reshape is not finished (wrong reshape
3790 * starting/continuation conditions). Mdmon doesn't
3791 * switch to next array in container and reentry
3792 * conditions for the same array occur.
3793 * This is possibly interim until the behaviour of
3794 * reshape_array is resolved().
3796 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3800 strcpy(last_devnm
, mdstat
->devnm
);
3802 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3803 pr_err("Unable to initialize sysfs for %s\n",
3809 if (mdmon_running(container
))
3810 flush_mdmon(container
);
3812 rv
= reshape_array(container
, fd
, adev
, st
,
3813 content
, force
, NULL
, INVALID_SECTORS
,
3814 backup_file
, verbose
, 1, restart
,
3818 if (freeze_reshape
) {
3827 if (mdmon_running(container
))
3828 flush_mdmon(container
);
3837 * We run a child process in the background which performs the following
3839 * - wait for resync to reach a certain point
3840 * - suspend io to the following section
3841 * - backup that section
3842 * - allow resync to proceed further
3844 * - discard the backup.
3846 * When are combined in slightly different ways in the three cases.
3848 * - suspend/backup/allow/wait/resume/discard
3850 * - allow/wait/suspend/backup/allow/wait/resume/discard
3852 * - wait/resume/discard/suspend/backup/allow
3854 * suspend/backup/allow always come together
3855 * wait/resume/discard do too.
3856 * For the same-size case we have two backups to improve flow.
3860 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3861 unsigned long long backup_point
,
3862 unsigned long long wait_point
,
3863 unsigned long long *suspend_point
,
3864 unsigned long long *reshape_completed
, int *frozen
)
3866 /* This function is called repeatedly by the reshape manager.
3867 * It determines how much progress can safely be made and allows
3869 * - 'info' identifies the array and particularly records in
3870 * ->reshape_progress the metadata's knowledge of progress
3871 * This is a sector offset from the start of the array
3872 * of the next array block to be relocated. This number
3873 * may increase from 0 or decrease from array_size, depending
3874 * on the type of reshape that is happening.
3875 * Note that in contrast, 'sync_completed' is a block count of the
3876 * reshape so far. It gives the distance between the start point
3877 * (head or tail of device) and the next place that data will be
3878 * written. It always increases.
3879 * - 'reshape' is the structure created by analyse_change
3880 * - 'backup_point' shows how much the metadata manager has backed-up
3881 * data. For reshapes with increasing progress, it is the next address
3882 * to be backed up, previous addresses have been backed-up. For
3883 * decreasing progress, it is the earliest address that has been
3884 * backed up - later address are also backed up.
3885 * So addresses between reshape_progress and backup_point are
3886 * backed up providing those are in the 'correct' order.
3887 * - 'wait_point' is an array address. When reshape_completed
3888 * passes this point, progress_reshape should return. It might
3889 * return earlier if it determines that ->reshape_progress needs
3890 * to be updated or further backup is needed.
3891 * - suspend_point is maintained by progress_reshape and the caller
3892 * should not touch it except to initialise to zero.
3893 * It is an array address and it only increases in 2.6.37 and earlier.
3894 * This makes it difficult to handle reducing reshapes with
3895 * external metadata.
3896 * However: it is similar to backup_point in that it records the
3897 * other end of a suspended region from reshape_progress.
3898 * it is moved to extend the region that is safe to backup and/or
3900 * - reshape_completed is read from sysfs and returned. The caller
3901 * should copy this into ->reshape_progress when it has reason to
3902 * believe that the metadata knows this, and any backup outside this
3906 * 1 if more data from backup_point - but only as far as suspend_point,
3907 * should be backed up
3908 * 0 if things are progressing smoothly
3909 * -1 if the reshape is finished because it is all done,
3910 * -2 if the reshape is finished due to an error.
3913 int advancing
= (reshape
->after
.data_disks
3914 >= reshape
->before
.data_disks
);
3915 unsigned long long need_backup
; /* All data between start of array and
3916 * here will at some point need to
3919 unsigned long long read_offset
, write_offset
;
3920 unsigned long long write_range
;
3921 unsigned long long max_progress
, target
, completed
;
3922 unsigned long long array_size
= (info
->component_size
3923 * reshape
->before
.data_disks
);
3927 /* First, we unsuspend any region that is now known to be safe.
3928 * If suspend_point is on the 'wrong' side of reshape_progress, then
3929 * we don't have or need suspension at the moment. This is true for
3930 * native metadata when we don't need to back-up.
3933 if (info
->reshape_progress
<= *suspend_point
)
3934 sysfs_set_num(info
, NULL
, "suspend_lo",
3935 info
->reshape_progress
);
3937 /* Note: this won't work in 2.6.37 and before.
3938 * Something somewhere should make sure we don't need it!
3940 if (info
->reshape_progress
>= *suspend_point
)
3941 sysfs_set_num(info
, NULL
, "suspend_hi",
3942 info
->reshape_progress
);
3945 /* Now work out how far it is safe to progress.
3946 * If the read_offset for ->reshape_progress is less than
3947 * 'blocks' beyond the write_offset, we can only progress as far
3949 * Otherwise we can progress until the write_offset for the new location
3950 * reaches (within 'blocks' of) the read_offset at the current location.
3951 * However that region must be suspended unless we are using native
3953 * If we need to suspend more, we limit it to 128M per device, which is
3954 * rather arbitrary and should be some time-based calculation.
3956 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3957 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3958 write_range
= info
->new_chunk
/512;
3959 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3960 need_backup
= array_size
;
3962 need_backup
= reshape
->backup_blocks
;
3964 if (read_offset
< write_offset
+ write_range
)
3965 max_progress
= backup_point
;
3968 read_offset
* reshape
->after
.data_disks
;
3970 if (read_offset
> write_offset
- write_range
)
3971 /* Can only progress as far as has been backed up,
3972 * which must be suspended */
3973 max_progress
= backup_point
;
3974 else if (info
->reshape_progress
<= need_backup
)
3975 max_progress
= backup_point
;
3977 if (info
->array
.major_version
>= 0)
3978 /* Can progress until backup is needed */
3979 max_progress
= need_backup
;
3981 /* Can progress until metadata update is required */
3983 read_offset
* reshape
->after
.data_disks
;
3984 /* but data must be suspended */
3985 if (max_progress
< *suspend_point
)
3986 max_progress
= *suspend_point
;
3991 /* We know it is safe to progress to 'max_progress' providing
3992 * it is suspended or we are using native metadata.
3993 * Consider extending suspend_point 128M per device if it
3994 * is less than 64M per device beyond reshape_progress.
3995 * But always do a multiple of 'blocks'
3996 * FIXME this is too big - it takes to long to complete
3999 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
4000 reshape
->after
.data_disks
);
4001 target
/= reshape
->backup_blocks
;
4004 target
*= reshape
->backup_blocks
;
4006 /* For externally managed metadata we always need to suspend IO to
4007 * the area being reshaped so we regularly push suspend_point forward.
4008 * For native metadata we only need the suspend if we are going to do
4012 if ((need_backup
> info
->reshape_progress
||
4013 info
->array
.major_version
< 0) &&
4014 *suspend_point
< info
->reshape_progress
+ target
) {
4015 if (need_backup
< *suspend_point
+ 2 * target
)
4016 *suspend_point
= need_backup
;
4017 else if (*suspend_point
+ 2 * target
< array_size
)
4018 *suspend_point
+= 2 * target
;
4020 *suspend_point
= array_size
;
4021 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
4022 if (max_progress
> *suspend_point
)
4023 max_progress
= *suspend_point
;
4026 if (info
->array
.major_version
>= 0) {
4027 /* Only need to suspend when about to backup */
4028 if (info
->reshape_progress
< need_backup
* 2 &&
4029 *suspend_point
> 0) {
4031 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4032 sysfs_set_num(info
, NULL
, "suspend_hi",
4036 /* Need to suspend continually */
4037 if (info
->reshape_progress
< *suspend_point
)
4038 *suspend_point
= info
->reshape_progress
;
4039 if (*suspend_point
+ target
< info
->reshape_progress
)
4040 /* No need to move suspend region yet */;
4042 if (*suspend_point
>= 2 * target
)
4043 *suspend_point
-= 2 * target
;
4046 sysfs_set_num(info
, NULL
, "suspend_lo",
4049 if (max_progress
< *suspend_point
)
4050 max_progress
= *suspend_point
;
4054 /* now set sync_max to allow that progress. sync_max, like
4055 * sync_completed is a count of sectors written per device, so
4056 * we find the difference between max_progress and the start point,
4057 * and divide that by after.data_disks to get a sync_max
4059 * At the same time we convert wait_point to a similar number
4060 * for comparing against sync_completed.
4062 /* scale down max_progress to per_disk */
4063 max_progress
/= reshape
->after
.data_disks
;
4065 * Round to chunk size as some kernels give an erroneously
4068 max_progress
/= info
->new_chunk
/512;
4069 max_progress
*= info
->new_chunk
/512;
4070 /* And round to old chunk size as the kernel wants that */
4071 max_progress
/= info
->array
.chunk_size
/512;
4072 max_progress
*= info
->array
.chunk_size
/512;
4073 /* Limit progress to the whole device */
4074 if (max_progress
> info
->component_size
)
4075 max_progress
= info
->component_size
;
4076 wait_point
/= reshape
->after
.data_disks
;
4078 /* switch from 'device offset' to 'processed block count' */
4079 max_progress
= info
->component_size
- max_progress
;
4080 wait_point
= info
->component_size
- wait_point
;
4084 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4086 /* Now wait. If we have already reached the point that we were
4087 * asked to wait to, don't wait at all, else wait for any change.
4088 * We need to select on 'sync_completed' as that is the place that
4089 * notifications happen, but we are really interested in
4090 * 'reshape_position'
4092 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4094 goto check_progress
;
4096 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4097 goto check_progress
;
4099 while (completed
< max_progress
&& completed
< wait_point
) {
4100 /* Check that sync_action is still 'reshape' to avoid
4101 * waiting forever on a dead array
4104 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) <= 0 ||
4105 strncmp(action
, "reshape", 7) != 0)
4107 /* Some kernels reset 'sync_completed' to zero
4108 * before setting 'sync_action' to 'idle'.
4109 * So we need these extra tests.
4111 if (completed
== 0 && advancing
&&
4112 strncmp(action
, "idle", 4) == 0 &&
4113 info
->reshape_progress
> 0)
4115 if (completed
== 0 && !advancing
&&
4116 strncmp(action
, "idle", 4) == 0 &&
4117 info
->reshape_progress
<
4118 (info
->component_size
* reshape
->after
.data_disks
))
4120 sysfs_wait(fd
, NULL
);
4121 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4122 goto check_progress
;
4124 /* Some kernels reset 'sync_completed' to zero,
4125 * we need to have real point we are in md.
4126 * So in that case, read 'reshape_position' from sysfs.
4128 if (completed
== 0) {
4129 unsigned long long reshapep
;
4131 if (sysfs_get_str(info
, NULL
, "sync_action", action
, 20) > 0 &&
4132 strncmp(action
, "idle", 4) == 0 &&
4133 sysfs_get_ll(info
, NULL
,
4134 "reshape_position", &reshapep
) == 0)
4135 *reshape_completed
= reshapep
;
4137 /* some kernels can give an incorrectly high
4138 * 'completed' number, so round down */
4139 completed
/= (info
->new_chunk
/512);
4140 completed
*= (info
->new_chunk
/512);
4141 /* Convert 'completed' back in to a 'progress' number */
4142 completed
*= reshape
->after
.data_disks
;
4144 completed
= (info
->component_size
4145 * reshape
->after
.data_disks
4147 *reshape_completed
= completed
;
4152 /* We return the need_backup flag. Caller will decide
4153 * how much - a multiple of ->backup_blocks up to *suspend_point
4156 return need_backup
> info
->reshape_progress
;
4158 return need_backup
>= info
->reshape_progress
;
4161 /* if we couldn't read a number from sync_completed, then
4162 * either the reshape did complete, or it aborted.
4163 * We can tell which by checking for 'none' in reshape_position.
4164 * If it did abort, then it might immediately restart if it
4165 * it was just a device failure that leaves us degraded but
4168 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
,
4169 sizeof(buf
)) < 0 || strncmp(buf
, "none", 4) != 0) {
4170 /* The abort might only be temporary. Wait up to 10
4171 * seconds for fd to contain a valid number again.
4175 unsigned long long new_sync_max
;
4176 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4177 if (sysfs_wait(fd
, &wait
) != 1)
4179 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4181 /* all good again */
4183 /* If "sync_max" is no longer max_progress
4184 * we need to freeze things
4186 sysfs_get_ll(info
, NULL
, "sync_max",
4188 *frozen
= (new_sync_max
!= max_progress
);
4190 case -2: /* read error - abort */
4197 return rv
; /* abort */
4199 /* Maybe racing with array shutdown - check state */
4202 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4204 strncmp(buf
, "inactive", 8) == 0 ||
4205 strncmp(buf
, "clear",5) == 0)
4206 return -2; /* abort */
4207 return -1; /* complete */
4211 /* FIXME return status is never checked */
4212 static int grow_backup(struct mdinfo
*sra
,
4213 unsigned long long offset
, /* per device */
4214 unsigned long stripes
, /* per device, in old chunks */
4215 int *sources
, unsigned long long *offsets
,
4216 int disks
, int chunk
, int level
, int layout
,
4217 int dests
, int *destfd
, unsigned long long *destoffsets
,
4218 int part
, int *degraded
,
4221 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4222 * to storage 'destfd' (offset 'destoffsets'), after first
4223 * suspending IO. Then allow resync to continue
4224 * over the suspended section.
4225 * Use part 'part' of the backup-super-block.
4230 unsigned long long ll
;
4232 //printf("offset %llu\n", offset);
4238 /* Check that array hasn't become degraded, else we might backup the wrong data */
4239 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4240 return -1; /* FIXME this error is ignored */
4241 new_degraded
= (int)ll
;
4242 if (new_degraded
!= *degraded
) {
4243 /* check each device to ensure it is still working */
4245 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4246 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4248 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4251 if (sysfs_get_str(sra
, sd
, "state",
4252 sbuf
, sizeof(sbuf
)) < 0 ||
4253 strstr(sbuf
, "faulty") ||
4254 strstr(sbuf
, "in_sync") == NULL
) {
4255 /* this device is dead */
4256 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4257 if (sd
->disk
.raid_disk
>= 0 &&
4258 sources
[sd
->disk
.raid_disk
] >= 0) {
4259 close(sources
[sd
->disk
.raid_disk
]);
4260 sources
[sd
->disk
.raid_disk
] = -1;
4265 *degraded
= new_degraded
;
4268 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4269 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4271 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4272 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4275 bsb
.magic
[15] = '2';
4276 for (i
= 0; i
< dests
; i
++)
4278 lseek64(destfd
[i
], destoffsets
[i
] +
4279 __le64_to_cpu(bsb
.devstart2
)*512, 0);
4281 lseek64(destfd
[i
], destoffsets
[i
], 0);
4283 rv
= save_stripes(sources
, offsets
, disks
, chunk
, level
, layout
,
4284 dests
, destfd
, offset
* 512 * odata
,
4285 stripes
* chunk
* odata
, buf
);
4289 bsb
.mtime
= __cpu_to_le64(time(0));
4290 for (i
= 0; i
< dests
; i
++) {
4291 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4293 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4294 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4295 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4296 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4297 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4300 if ((unsigned long long)lseek64(destfd
[i
],
4301 destoffsets
[i
] - 4096, 0) !=
4302 destoffsets
[i
] - 4096)
4304 if (write(destfd
[i
], &bsb
, 512) != 512)
4306 if (destoffsets
[i
] > 4096) {
4307 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4308 destoffsets
[i
]+stripes
*chunk
*odata
)
4310 if (write(destfd
[i
], &bsb
, 512) != 512)
4320 /* in 2.6.30, the value reported by sync_completed can be
4321 * less that it should be by one stripe.
4322 * This only happens when reshape hits sync_max and pauses.
4323 * So allow wait_backup to either extent sync_max further
4324 * than strictly necessary, or return before the
4325 * sync has got quite as far as we would really like.
4326 * This is what 'blocks2' is for.
4327 * The various caller give appropriate values so that
4330 /* FIXME return value is often ignored */
4331 static int forget_backup(int dests
, int *destfd
,
4332 unsigned long long *destoffsets
,
4336 * Erase backup 'part' (which is 0 or 1)
4342 bsb
.arraystart2
= __cpu_to_le64(0);
4343 bsb
.length2
= __cpu_to_le64(0);
4345 bsb
.arraystart
= __cpu_to_le64(0);
4346 bsb
.length
= __cpu_to_le64(0);
4348 bsb
.mtime
= __cpu_to_le64(time(0));
4350 for (i
= 0; i
< dests
; i
++) {
4351 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4352 bsb
.sb_csum
= bsb_csum((char*)&bsb
,
4353 ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4354 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4355 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4356 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4357 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4358 destoffsets
[i
]-4096)
4360 if (rv
== 0 && write(destfd
[i
], &bsb
, 512) != 512)
4367 static void fail(char *msg
)
4370 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4371 rv
|= (write(2, "\n", 1) != 1);
4375 static char *abuf
, *bbuf
;
4376 static unsigned long long abuflen
;
4377 static void validate(int afd
, int bfd
, unsigned long long offset
)
4379 /* check that the data in the backup against the array.
4380 * This is only used for regression testing and should not
4381 * be used while the array is active
4385 lseek64(bfd
, offset
- 4096, 0);
4386 if (read(bfd
, &bsb2
, 512) != 512)
4387 fail("cannot read bsb");
4388 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4389 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4390 fail("first csum bad");
4391 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4392 fail("magic is bad");
4393 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4394 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4395 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4396 fail("second csum bad");
4398 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4399 fail("devstart is wrong");
4402 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4404 if (abuflen
< len
) {
4408 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4409 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4411 /* just stop validating on mem-alloc failure */
4416 lseek64(bfd
, offset
, 0);
4417 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4418 //printf("len %llu\n", len);
4419 fail("read first backup failed");
4421 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4422 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4423 fail("read first from array failed");
4424 if (memcmp(bbuf
, abuf
, len
) != 0) {
4427 printf("offset=%llu len=%llu\n",
4428 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4429 for (i
=0; i
<len
; i
++)
4430 if (bbuf
[i
] != abuf
[i
]) {
4431 printf("first diff byte %d\n", i
);
4435 fail("data1 compare failed");
4439 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4441 if (abuflen
< len
) {
4445 abuf
= xmalloc(abuflen
);
4446 bbuf
= xmalloc(abuflen
);
4449 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4450 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4451 fail("read second backup failed");
4452 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4453 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4454 fail("read second from array failed");
4455 if (memcmp(bbuf
, abuf
, len
) != 0)
4456 fail("data2 compare failed");
4460 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4461 struct supertype
*st
, unsigned long blocks
,
4462 int *fds
, unsigned long long *offsets
,
4463 int dests
, int *destfd
, unsigned long long *destoffsets
)
4465 /* Monitor a reshape where backup is being performed using
4466 * 'native' mechanism - either to a backup file, or
4467 * to some space in a spare.
4471 unsigned long long speed
;
4472 unsigned long long suspend_point
, array_size
;
4473 unsigned long long backup_point
, wait_point
;
4474 unsigned long long reshape_completed
;
4476 int increasing
= reshape
->after
.data_disks
>=
4477 reshape
->before
.data_disks
;
4478 int part
= 0; /* The next part of the backup area to fill. It
4479 * may already be full, so we need to check */
4480 int level
= reshape
->level
;
4481 int layout
= reshape
->before
.layout
;
4482 int data
= reshape
->before
.data_disks
;
4483 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4484 int chunk
= sra
->array
.chunk_size
;
4486 unsigned long stripes
;
4490 /* set up the backup-super-block. This requires the
4491 * uuid from the array.
4493 /* Find a superblock */
4494 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4498 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4500 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4501 devfd
= dev_open(dn
, O_RDONLY
);
4504 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4510 pr_err("Cannot find a superblock\n");
4514 memset(&bsb
, 0, 512);
4515 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4516 st
->ss
->uuid_from_super(st
, uuid
);
4517 memcpy(bsb
.set_uuid
, uuid
, 16);
4518 bsb
.mtime
= __cpu_to_le64(time(0));
4519 bsb
.devstart2
= blocks
;
4521 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4522 reshape
->before
.data_disks
;
4524 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4525 /* Don't start the 'reshape' */
4527 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4528 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4529 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4533 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4534 backup_point
= sra
->reshape_progress
;
4537 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4538 backup_point
= reshape
->backup_blocks
;
4539 suspend_point
= array_size
;
4545 /* Want to return as soon the oldest backup slot can
4546 * be released as that allows us to start backing up
4547 * some more, providing suspend_point has been
4548 * advanced, which it should have.
4551 wait_point
= array_size
;
4552 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4553 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4554 __le64_to_cpu(bsb
.length
));
4555 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4556 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4557 __le64_to_cpu(bsb
.length2
));
4560 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4561 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4562 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4563 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4566 reshape_completed
= sra
->reshape_progress
;
4567 rv
= progress_reshape(sra
, reshape
,
4568 backup_point
, wait_point
,
4569 &suspend_point
, &reshape_completed
,
4571 /* external metadata would need to ping_monitor here */
4572 sra
->reshape_progress
= reshape_completed
;
4574 /* Clear any backup region that is before 'here' */
4576 if (__le64_to_cpu(bsb
.length
) > 0 &&
4577 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4578 __le64_to_cpu(bsb
.length
)))
4579 forget_backup(dests
, destfd
,
4581 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4582 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4583 __le64_to_cpu(bsb
.length2
)))
4584 forget_backup(dests
, destfd
,
4587 if (__le64_to_cpu(bsb
.length
) > 0 &&
4588 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4589 forget_backup(dests
, destfd
,
4591 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4592 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4593 forget_backup(dests
, destfd
,
4603 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4604 /* No longer need to monitor this reshape */
4605 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4611 unsigned long long offset
;
4612 unsigned long actual_stripes
;
4613 /* Need to backup some data.
4614 * If 'part' is not used and the desired
4615 * backup size is suspended, do a backup,
4616 * then consider the next part.
4618 /* Check that 'part' is unused */
4619 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4621 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4624 offset
= backup_point
/ data
;
4625 actual_stripes
= stripes
;
4627 if (offset
+ actual_stripes
* (chunk
/512) >
4628 sra
->component_size
)
4629 actual_stripes
= ((sra
->component_size
- offset
)
4631 if (offset
+ actual_stripes
* (chunk
/512) >
4635 if (offset
< actual_stripes
* (chunk
/512))
4636 actual_stripes
= offset
/ (chunk
/512);
4637 offset
-= actual_stripes
* (chunk
/512);
4638 if (offset
< suspend_point
/data
)
4641 if (actual_stripes
== 0)
4643 grow_backup(sra
, offset
, actual_stripes
, fds
, offsets
,
4644 disks
, chunk
, level
, layout
, dests
, destfd
,
4645 destoffsets
, part
, °raded
, buf
);
4646 validate(afd
, destfd
[0], destoffsets
[0]);
4647 /* record where 'part' is up to */
4650 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4652 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4656 /* FIXME maybe call progress_reshape one more time instead */
4657 /* remove any remaining suspension */
4658 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4659 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4660 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4661 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4663 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4664 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4670 * If any spare contains md_back_data-1 which is recent wrt mtime,
4671 * write that data into the array and update the super blocks with
4672 * the new reshape_progress
4674 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
,
4675 int cnt
, char *backup_file
, int verbose
)
4679 unsigned long long *offsets
;
4680 unsigned long long nstripe
, ostripe
;
4683 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4684 if (info
->array
.level
== 6)
4685 odata
--; /* number of data disks */
4686 ndata
= info
->array
.raid_disks
- 1;
4687 if (info
->new_level
== 6)
4690 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4692 if (info
->delta_disks
<= 0)
4693 /* Didn't grow, so the backup file must have
4697 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4698 struct mdinfo dinfo
;
4701 char *devname
, namebuf
[20];
4702 unsigned long long lo
, hi
;
4704 /* This was a spare and may have some saved data on it.
4705 * Load the superblock, find and load the
4706 * backup_super_block.
4707 * If either fail, go on to next device.
4708 * If the backup contains no new info, just return
4709 * else restore data and update all superblocks
4711 if (i
== old_disks
-1) {
4712 fd
= open(backup_file
, O_RDONLY
);
4714 pr_err("backup file %s inaccessible: %s\n",
4715 backup_file
, strerror(errno
));
4718 devname
= backup_file
;
4723 if (st
->ss
->load_super(st
, fd
, NULL
))
4726 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4727 st
->ss
->free_super(st
);
4730 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4732 pr_err("Cannot seek on device %d\n", i
);
4733 continue; /* Cannot seek */
4735 sprintf(namebuf
, "device-%d", i
);
4738 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4740 pr_err("Cannot read from %s\n", devname
);
4741 continue; /* Cannot read */
4743 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4744 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4746 pr_err("No backup metadata on %s\n", devname
);
4749 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4751 pr_err("Bad backup-metadata checksum on %s\n",
4753 continue; /* bad checksum */
4755 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4756 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4758 pr_err("Bad backup-metadata checksum2 on %s\n",
4760 continue; /* Bad second checksum */
4762 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4764 pr_err("Wrong uuid on backup-metadata on %s\n",
4766 continue; /* Wrong uuid */
4770 * array utime and backup-mtime should be updated at
4771 * much the same time, but it seems that sometimes
4772 * they aren't... So allow considerable flexability in
4773 * matching, and allow this test to be overridden by
4774 * an environment variable.
4776 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4777 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4778 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4779 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4780 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4781 (unsigned long)info
->array
.utime
);
4783 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4784 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4785 continue; /* time stamp is too bad */
4789 if (bsb
.magic
[15] == '1') {
4790 if (bsb
.length
== 0)
4792 if (info
->delta_disks
>= 0) {
4793 /* reshape_progress is increasing */
4794 if (__le64_to_cpu(bsb
.arraystart
)
4795 + __le64_to_cpu(bsb
.length
)
4796 < info
->reshape_progress
) {
4799 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4800 continue; /* No new data here */
4803 /* reshape_progress is decreasing */
4804 if (__le64_to_cpu(bsb
.arraystart
) >=
4805 info
->reshape_progress
)
4806 goto nonew
; /* No new data here */
4809 if (bsb
.length
== 0 && bsb
.length2
== 0)
4811 if (info
->delta_disks
>= 0) {
4812 /* reshape_progress is increasing */
4813 if ((__le64_to_cpu(bsb
.arraystart
)
4814 + __le64_to_cpu(bsb
.length
)
4815 < info
->reshape_progress
) &&
4816 (__le64_to_cpu(bsb
.arraystart2
)
4817 + __le64_to_cpu(bsb
.length2
)
4818 < info
->reshape_progress
))
4819 goto nonew
; /* No new data here */
4821 /* reshape_progress is decreasing */
4822 if (__le64_to_cpu(bsb
.arraystart
) >=
4823 info
->reshape_progress
&&
4824 __le64_to_cpu(bsb
.arraystart2
) >=
4825 info
->reshape_progress
)
4826 goto nonew
; /* No new data here */
4829 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4832 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4834 continue; /* Cannot seek */
4836 /* There should be a duplicate backup superblock 4k before here */
4837 if (lseek64(fd
, -4096, 1) < 0 ||
4838 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4839 goto second_fail
; /* Cannot find leading superblock */
4840 if (bsb
.magic
[15] == '1')
4841 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4843 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4844 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4845 goto second_fail
; /* Cannot find leading superblock */
4847 /* Now need the data offsets for all devices. */
4848 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4849 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4852 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4853 /* FIXME should be this be an error */
4855 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4856 st
->ss
->free_super(st
);
4857 offsets
[j
] = dinfo
.data_offset
* 512;
4859 printf("%s: restoring critical section\n", Name
);
4861 if (restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4862 info
->new_chunk
, info
->new_level
,
4863 info
->new_layout
, fd
,
4864 __le64_to_cpu(bsb
.devstart
)*512,
4865 __le64_to_cpu(bsb
.arraystart
)*512,
4866 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4867 /* didn't succeed, so giveup */
4869 pr_err("Error restoring backup from %s\n",
4875 if (bsb
.magic
[15] == '2' &&
4876 restore_stripes(fdlist
, offsets
, info
->array
.raid_disks
,
4877 info
->new_chunk
, info
->new_level
,
4878 info
->new_layout
, fd
,
4879 __le64_to_cpu(bsb
.devstart
)*512 +
4880 __le64_to_cpu(bsb
.devstart2
)*512,
4881 __le64_to_cpu(bsb
.arraystart2
)*512,
4882 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4883 /* didn't succeed, so giveup */
4885 pr_err("Error restoring second backup from %s\n",
4893 /* Ok, so the data is restored. Let's update those superblocks. */
4897 lo
= __le64_to_cpu(bsb
.arraystart
);
4898 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4900 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4901 unsigned long long lo1
, hi1
;
4902 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4903 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4907 } else if (lo
< lo1
)
4912 if (lo
< hi
&& (info
->reshape_progress
< lo
||
4913 info
->reshape_progress
> hi
))
4914 /* backup does not affect reshape_progress*/ ;
4915 else if (info
->delta_disks
>= 0) {
4916 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4917 __le64_to_cpu(bsb
.length
);
4918 if (bsb
.magic
[15] == '2') {
4919 unsigned long long p2
;
4921 p2
= __le64_to_cpu(bsb
.arraystart2
) +
4922 __le64_to_cpu(bsb
.length2
);
4923 if (p2
> info
->reshape_progress
)
4924 info
->reshape_progress
= p2
;
4927 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4928 if (bsb
.magic
[15] == '2') {
4929 unsigned long long p2
;
4931 p2
= __le64_to_cpu(bsb
.arraystart2
);
4932 if (p2
< info
->reshape_progress
)
4933 info
->reshape_progress
= p2
;
4936 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4939 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4941 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4942 dinfo
.reshape_progress
= info
->reshape_progress
;
4943 st
->ss
->update_super(st
, &dinfo
, "_reshape_progress",
4945 st
->ss
->store_super(st
, fdlist
[j
]);
4946 st
->ss
->free_super(st
);
4950 /* Didn't find any backup data, try to see if any
4953 if (info
->delta_disks
< 0) {
4954 /* When shrinking, the critical section is at the end.
4955 * So see if we are before the critical section.
4957 unsigned long long first_block
;
4958 nstripe
= ostripe
= 0;
4960 while (ostripe
>= nstripe
) {
4961 ostripe
+= info
->array
.chunk_size
/ 512;
4962 first_block
= ostripe
* odata
;
4963 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4964 (info
->new_chunk
/512);
4967 if (info
->reshape_progress
>= first_block
)
4970 if (info
->delta_disks
> 0) {
4971 /* See if we are beyond the critical section. */
4972 unsigned long long last_block
;
4973 nstripe
= ostripe
= 0;
4975 while (nstripe
>= ostripe
) {
4976 nstripe
+= info
->new_chunk
/ 512;
4977 last_block
= nstripe
* ndata
;
4978 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4979 (info
->array
.chunk_size
/512);
4982 if (info
->reshape_progress
>= last_block
)
4985 /* needed to recover critical section! */
4987 pr_err("Failed to find backup of critical section\n");
4991 int Grow_continue_command(char *devname
, int fd
,
4992 char *backup_file
, int verbose
)
4995 struct supertype
*st
= NULL
;
4996 struct mdinfo
*content
= NULL
;
4997 struct mdinfo array
;
4998 char *subarray
= NULL
;
4999 struct mdinfo
*cc
= NULL
;
5000 struct mdstat_ent
*mdstat
= NULL
;
5004 dprintf("Grow continue from command line called for %s\n", devname
);
5006 st
= super_by_fd(fd
, &subarray
);
5007 if (!st
|| !st
->ss
) {
5008 pr_err("Unable to determine metadata format for %s\n", devname
);
5011 dprintf("Grow continue is run for ");
5012 if (st
->ss
->external
== 0) {
5015 dprintf_cont("native array (%s)\n", devname
);
5016 if (md_get_array_info(fd
, &array
.array
) < 0) {
5017 pr_err("%s is not an active md array - aborting\n",
5020 goto Grow_continue_command_exit
;
5023 sysfs_init(content
, fd
, NULL
);
5024 /* Need to load a superblock.
5025 * FIXME we should really get what we need from
5029 for (d
= 0; d
< MAX_DISKS
; d
++) {
5030 mdu_disk_info_t disk
;
5034 if (md_get_disk_info(fd
, &disk
) < 0)
5036 if (disk
.major
== 0 && disk
.minor
== 0)
5038 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5040 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5043 fd2
= dev_open(dv
, O_RDONLY
);
5046 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5052 if (d
== MAX_DISKS
) {
5053 pr_err("Unable to load metadata for %s\n",
5056 goto Grow_continue_command_exit
;
5058 st
->ss
->getinfo_super(st
, content
, NULL
);
5059 if (!content
->reshape_active
)
5063 } while (cnt
-- > 0);
5068 dprintf_cont("subarray (%s)\n", subarray
);
5069 container
= st
->container_devnm
;
5070 cfd
= open_dev_excl(st
->container_devnm
);
5072 container
= st
->devnm
;
5074 cfd
= open_dev_excl(st
->devnm
);
5075 dprintf_cont("container (%s)\n", container
);
5079 pr_err("Unable to open container for %s\n", devname
);
5081 goto Grow_continue_command_exit
;
5084 /* find in container array under reshape
5086 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5088 pr_err("Cannot read superblock for %s\n", devname
);
5090 goto Grow_continue_command_exit
;
5093 cc
= st
->ss
->container_content(st
, subarray
);
5094 for (content
= cc
; content
; content
= content
->next
) {
5096 int allow_reshape
= 1;
5098 if (content
->reshape_active
== 0)
5100 /* The decision about array or container wide
5101 * reshape is taken in Grow_continue based
5102 * content->reshape_active state, therefore we
5103 * need to check_reshape based on
5104 * reshape_active and subarray name
5106 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5108 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5109 (content
->array
.state
5110 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5113 if (!allow_reshape
) {
5114 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5115 devname
, container
);
5117 goto Grow_continue_command_exit
;
5120 array_name
= strchr(content
->text_version
+1, '/')+1;
5121 mdstat
= mdstat_by_subdev(array_name
, container
);
5124 if (mdstat
->active
== 0) {
5125 pr_err("Skipping inactive array %s.\n",
5127 free_mdstat(mdstat
);
5134 pr_err("Unable to determine reshaped array for %s\n", devname
);
5136 goto Grow_continue_command_exit
;
5138 fd2
= open_dev(mdstat
->devnm
);
5140 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5142 goto Grow_continue_command_exit
;
5145 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5146 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5150 goto Grow_continue_command_exit
;
5155 /* start mdmon in case it is not running
5157 if (!mdmon_running(container
))
5158 start_mdmon(container
);
5159 ping_monitor(container
);
5161 if (mdmon_running(container
))
5162 st
->update_tail
= &st
->updates
;
5164 pr_err("No mdmon found. Grow cannot continue.\n");
5166 goto Grow_continue_command_exit
;
5170 /* verify that array under reshape is started from
5173 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5175 goto Grow_continue_command_exit
;
5180 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5182 Grow_continue_command_exit
:
5185 st
->ss
->free_super(st
);
5186 free_mdstat(mdstat
);
5193 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5194 char *backup_file
, int forked
, int freeze_reshape
)
5198 if (!info
->reshape_active
)
5201 if (st
->ss
->external
) {
5202 int cfd
= open_dev(st
->container_devnm
);
5207 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5209 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5210 st
, info
, 0, backup_file
, 0,
5211 forked
, 1 | info
->reshape_active
,
5214 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5215 NULL
, INVALID_SECTORS
, backup_file
,
5216 0, forked
, 1 | info
->reshape_active
,
5222 char *make_backup(char *name
)
5224 char *base
= "backup_file-";
5228 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5229 fname
= xmalloc(len
);
5230 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5234 char *locate_backup(char *name
)
5236 char *fl
= make_backup(name
);
5239 if (stat(fl
, &stb
) == 0 && S_ISREG(stb
.st_mode
))