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
) {
64 fd
= dev_open(buf
, O_RDWR
);
66 if (dev
->disk
.raid_disk
>= 0)
67 fdlist
[dev
->disk
.raid_disk
] = fd
;
69 fdlist
[next_spare
++] = fd
;
73 backup_file
= locate_backup(content
->sys_name
);
74 *backup_filep
= backup_file
;
77 if (st
->ss
->external
&& st
->ss
->recover_backup
)
78 err
= st
->ss
->recover_backup(st
, content
);
80 err
= Grow_restart(st
, content
, fdlist
, next_spare
,
81 backup_file
, verbose
> 0);
83 while (next_spare
> 0) {
85 if (fdlist
[next_spare
] >= 0)
86 close(fdlist
[next_spare
]);
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
92 pr_err("Possibly you need to specify a --backup-file\n");
96 dprintf("restore_backup() returns status OK.\n");
100 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
108 * This means that we need to *find* all other devices.
115 struct supertype
*st
= NULL
;
116 char *subarray
= NULL
;
118 if (md_get_array_info(fd
, &info
.array
) < 0) {
119 pr_err("cannot get array info for %s\n", devname
);
123 if (info
.array
.level
!= -1) {
124 pr_err("can only add devices to linear arrays\n");
128 st
= super_by_fd(fd
, &subarray
);
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info
.array
.major_version
);
136 pr_err("Cannot grow linear sub-arrays yet\n");
142 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
144 pr_err("cannot open %s\n", newdev
);
148 if (!fstat_is_blkdev(nfd
, newdev
, &rdev
)) {
153 /* now check out all the devices and make sure we can read the
155 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
156 mdu_disk_info_t disk
;
159 st
->ss
->free_super(st
);
162 if (md_get_disk_info(fd
, &disk
) < 0) {
163 pr_err("cannot get device detail for device %d\n",
169 dv
= map_dev(disk
.major
, disk
.minor
, 1);
171 pr_err("cannot find device file for device %d\n",
177 fd2
= dev_open(dv
, O_RDWR
);
179 pr_err("cannot open device file %s\n", dv
);
185 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
186 pr_err("cannot find super block on %s\n", dv
);
194 /* Ok, looks good. Lets update the superblock and write it out to
198 info
.disk
.number
= d
;
199 info
.disk
.major
= major(rdev
);
200 info
.disk
.minor
= minor(rdev
);
201 info
.disk
.raid_disk
= d
;
202 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
203 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
206 if (st
->ss
->store_super(st
, nfd
)) {
207 pr_err("Cannot store new superblock on %s\n",
214 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
215 pr_err("Cannot add new disk to this array\n");
218 /* Well, that seems to have worked.
219 * Now go through and update all superblocks
222 if (md_get_array_info(fd
, &info
.array
) < 0) {
223 pr_err("cannot get array info for %s\n", devname
);
228 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
229 mdu_disk_info_t disk
;
233 if (md_get_disk_info(fd
, &disk
) < 0) {
234 pr_err("cannot get device detail for device %d\n",
238 dv
= map_dev(disk
.major
, disk
.minor
, 1);
240 pr_err("cannot find device file for device %d\n",
244 fd2
= dev_open(dv
, O_RDWR
);
246 pr_err("cannot open device file %s\n", dv
);
249 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
250 pr_err("cannot find super block on %s\n", dv
);
254 info
.array
.raid_disks
= nd
+1;
255 info
.array
.nr_disks
= nd
+1;
256 info
.array
.active_disks
= nd
+1;
257 info
.array
.working_disks
= nd
+1;
259 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
262 if (st
->ss
->store_super(st
, fd2
)) {
263 pr_err("Cannot store new superblock on %s\n", dv
);
273 int Grow_addbitmap(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
276 * First check that array doesn't have a bitmap
277 * Then create the bitmap
280 * For internal bitmaps, we need to check the version,
281 * find all the active devices, and write the bitmap block
284 mdu_bitmap_file_t bmf
;
285 mdu_array_info_t array
;
286 struct supertype
*st
;
287 char *subarray
= NULL
;
288 int major
= BITMAP_MAJOR_HI
;
289 unsigned long long bitmapsize
, array_size
;
293 * We only ever get called if s->bitmap_file is != NULL, so this check
294 * is just here to quiet down static code checkers.
299 if (strcmp(s
->bitmap_file
, "clustered") == 0)
300 major
= BITMAP_MAJOR_CLUSTERED
;
302 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
304 pr_err("Memory allocation failure.\n");
306 pr_err("bitmaps not supported by this kernel.\n");
309 if (bmf
.pathname
[0]) {
310 if (strcmp(s
->bitmap_file
,"none") == 0) {
311 if (ioctl(fd
, SET_BITMAP_FILE
, -1) != 0) {
312 pr_err("failed to remove bitmap %s\n",
318 pr_err("%s already has a bitmap (%s)\n",
319 devname
, bmf
.pathname
);
322 if (md_get_array_info(fd
, &array
) != 0) {
323 pr_err("cannot get array status for %s\n", devname
);
326 if (array
.state
& (1 << MD_SB_BITMAP_PRESENT
)) {
327 if (strcmp(s
->bitmap_file
, "none")==0) {
328 array
.state
&= ~(1 << MD_SB_BITMAP_PRESENT
);
329 if (md_set_array_info(fd
, &array
) != 0) {
330 if (array
.state
& (1 << MD_SB_CLUSTERED
))
331 pr_err("failed to remove clustered bitmap.\n");
333 pr_err("failed to remove internal bitmap.\n");
338 pr_err("bitmap already present on %s\n", devname
);
342 if (strcmp(s
->bitmap_file
, "none") == 0) {
343 pr_err("no bitmap found on %s\n", devname
);
346 if (array
.level
<= 0) {
347 pr_err("Bitmaps not meaningful with level %s\n",
348 map_num(pers
, array
.level
)?:"of this array");
351 bitmapsize
= array
.size
;
353 if (get_dev_size(fd
, NULL
, &array_size
) &&
354 array_size
> (0x7fffffffULL
<< 9)) {
355 /* Array is big enough that we cannot trust array.size
356 * try other approaches
358 bitmapsize
= get_component_size(fd
);
360 if (bitmapsize
== 0) {
361 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
365 if (array
.level
== 10) {
368 ncopies
= (array
.layout
& 255) * ((array
.layout
>> 8) & 255);
369 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
372 st
= super_by_fd(fd
, &subarray
);
374 pr_err("Cannot understand version %d.%d\n",
375 array
.major_version
, array
.minor_version
);
379 pr_err("Cannot add bitmaps to sub-arrays yet\n");
385 mdi
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
);
387 if (mdi
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
388 pr_err("Cannot add bitmap to array with PPL\n");
396 if (strcmp(s
->bitmap_file
, "internal") == 0 ||
397 strcmp(s
->bitmap_file
, "clustered") == 0) {
400 int offset_setable
= 0;
401 if (st
->ss
->add_internal_bitmap
== NULL
) {
402 pr_err("Internal bitmaps not supported with %s metadata\n", st
->ss
->name
);
405 st
->nodes
= c
->nodes
;
406 st
->cluster_name
= c
->homecluster
;
407 mdi
= sysfs_read(fd
, NULL
, GET_BITMAP_LOCATION
);
410 for (d
= 0; d
< st
->max_devs
; d
++) {
411 mdu_disk_info_t disk
;
416 if (md_get_disk_info(fd
, &disk
) < 0)
418 if (disk
.major
== 0 && disk
.minor
== 0)
420 if ((disk
.state
& (1 << MD_DISK_SYNC
)) == 0)
422 dv
= map_dev(disk
.major
, disk
.minor
, 1);
425 fd2
= dev_open(dv
, O_RDWR
);
428 rv
= st
->ss
->load_super(st
, fd2
, NULL
);
430 rv
= st
->ss
->add_internal_bitmap(
431 st
, &s
->bitmap_chunk
, c
->delay
,
432 s
->write_behind
, bitmapsize
,
433 offset_setable
, major
);
435 st
->ss
->write_bitmap(st
, fd2
,
438 pr_err("failed to create internal bitmap - chunksize problem.\n");
441 pr_err("failed to load super-block.\n");
447 if (offset_setable
) {
448 st
->ss
->getinfo_super(st
, mdi
, NULL
);
449 if (sysfs_init(mdi
, fd
, NULL
)) {
450 pr_err("failed to intialize sysfs.\n");
453 rv
= sysfs_set_num_signed(mdi
, NULL
, "bitmap/location",
457 if (strcmp(s
->bitmap_file
, "clustered") == 0)
458 array
.state
|= (1 << MD_SB_CLUSTERED
);
459 array
.state
|= (1 << MD_SB_BITMAP_PRESENT
);
460 rv
= md_set_array_info(fd
, &array
);
464 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
465 pr_err("failed to set internal bitmap.\n");
472 int max_devs
= st
->max_devs
;
474 /* try to load a superblock */
475 for (d
= 0; d
< max_devs
; d
++) {
476 mdu_disk_info_t disk
;
480 if (md_get_disk_info(fd
, &disk
) < 0)
482 if ((disk
.major
==0 && disk
.minor
== 0) ||
483 (disk
.state
& (1 << MD_DISK_REMOVED
)))
485 dv
= map_dev(disk
.major
, disk
.minor
, 1);
488 fd2
= dev_open(dv
, O_RDONLY
);
490 if (st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
492 st
->ss
->uuid_from_super(st
, uuid
);
499 pr_err("cannot find UUID for array!\n");
502 if (CreateBitmap(s
->bitmap_file
, c
->force
, (char*)uuid
,
503 s
->bitmap_chunk
, c
->delay
, s
->write_behind
,
504 bitmapsize
, major
)) {
507 bitmap_fd
= open(s
->bitmap_file
, O_RDWR
);
509 pr_err("weird: %s cannot be opened\n", s
->bitmap_file
);
512 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
515 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
516 pr_err("Cannot set bitmap file for %s: %s\n",
517 devname
, strerror(err
));
525 int Grow_consistency_policy(char *devname
, int fd
, struct context
*c
, struct shape
*s
)
527 struct supertype
*st
;
530 char *subarray
= NULL
;
532 char container_dev
[PATH_MAX
];
535 if (s
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
536 s
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
537 pr_err("Operation not supported for consistency policy %s\n",
538 map_num(consistency_policies
, s
->consistency_policy
));
542 st
= super_by_fd(fd
, &subarray
);
546 sra
= sysfs_read(fd
, NULL
, GET_CONSISTENCY_POLICY
|GET_LEVEL
|
553 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
&&
554 !st
->ss
->write_init_ppl
) {
555 pr_err("%s metadata does not support PPL\n", st
->ss
->name
);
560 if (sra
->array
.level
!= 5) {
561 pr_err("Operation not supported for array level %d\n",
567 if (sra
->consistency_policy
== (unsigned)s
->consistency_policy
) {
568 pr_err("Consistency policy is already %s\n",
569 map_num(consistency_policies
, s
->consistency_policy
));
572 } else if (sra
->consistency_policy
!= CONSISTENCY_POLICY_RESYNC
&&
573 sra
->consistency_policy
!= CONSISTENCY_POLICY_PPL
) {
574 pr_err("Current consistency policy is %s, cannot change to %s\n",
575 map_num(consistency_policies
, sra
->consistency_policy
),
576 map_num(consistency_policies
, s
->consistency_policy
));
581 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
582 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) <= 0) {
585 } else if (strcmp(buf
, "reshape\n") == 0) {
586 pr_err("PPL cannot be enabled when reshape is in progress\n");
595 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
)
600 sprintf(container_dev
, "/dev/%s", st
->container_devnm
);
602 ret
= Update_subarray(container_dev
, subarray
, update
, NULL
,
608 if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
615 cfd
= open(container_dev
, O_RDWR
|O_EXCL
);
617 pr_err("Failed to open %s\n", container_dev
);
622 ret
= st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
626 pr_err("Cannot read superblock for %s\n",
631 mdi
= st
->ss
->container_content(st
, subarray
);
636 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
640 devpath
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
641 dfd
= dev_open(devpath
, O_RDWR
);
643 pr_err("Failed to open %s\n", devpath
);
649 ret
= st
->ss
->load_super(st
, dfd
, NULL
);
651 pr_err("Failed to load super-block.\n");
656 ret
= st
->ss
->update_super(st
, sra
, "ppl", devname
,
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", info
.ppl_sector
);
667 ret
|= sysfs_set_num(sra
, sd
, "ppl_size", info
.ppl_size
);
670 pr_err("Failed to set PPL attributes for %s\n",
673 st
->ss
->free_super(st
);
677 ret
= st
->ss
->write_init_ppl(st
, &info
, dfd
);
679 pr_err("Failed to write PPL\n");
684 st
->ss
->free_super(st
);
691 ret
= sysfs_set_str(sra
, NULL
, "consistency_policy",
692 map_num(consistency_policies
,
693 s
->consistency_policy
));
695 pr_err("Failed to change array consistency policy\n");
707 * When reshaping an array we might need to backup some data.
708 * This is written to all spares with a 'super_block' describing it.
709 * The superblock goes 4K from the end of the used space on the
711 * It if written after the backup is complete.
712 * It has the following structure.
715 static struct mdp_backup_super
{
716 char magic
[16]; /* md_backup_data-1 or -2 */
719 /* start/sizes in 512byte sectors */
720 __u64 devstart
; /* address on backup device/file of data */
723 __u32 sb_csum
; /* csum of preceeding bytes. */
725 __u64 devstart2
; /* offset in to data of second section */
728 __u32 sb_csum2
; /* csum of preceeding bytes. */
730 } __attribute__((aligned(512))) bsb
, bsb2
;
732 static __u32
bsb_csum(char *buf
, int len
)
736 for (i
= 0; i
< len
; i
++)
737 csum
= (csum
<<3) + buf
[0];
738 return __cpu_to_le32(csum
);
741 static int check_idle(struct supertype
*st
)
743 /* Check that all member arrays for this container, or the
744 * container of this array, are idle
746 char *container
= (st
->container_devnm
[0]
747 ? st
->container_devnm
: st
->devnm
);
748 struct mdstat_ent
*ent
, *e
;
751 ent
= mdstat_read(0, 0);
752 for (e
= ent
; e
; e
= e
->next
) {
753 if (!is_container_member(e
, container
))
755 if (e
->percent
>= 0) {
764 static int freeze_container(struct supertype
*st
)
766 char *container
= (st
->container_devnm
[0]
767 ? st
->container_devnm
: st
->devnm
);
772 if (block_monitor(container
, 1)) {
773 pr_err("failed to freeze container\n");
780 static void unfreeze_container(struct supertype
*st
)
782 char *container
= (st
->container_devnm
[0]
783 ? st
->container_devnm
: st
->devnm
);
785 unblock_monitor(container
, 1);
788 static int freeze(struct supertype
*st
)
790 /* Try to freeze resync/rebuild on this array/container.
791 * Return -1 if the array is busy,
792 * return -2 container cannot be frozen,
793 * return 0 if this kernel doesn't support 'frozen'
794 * return 1 if it worked.
796 if (st
->ss
->external
)
797 return freeze_container(st
);
799 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
805 /* Need to clear any 'read-auto' status */
806 if (sysfs_get_str(sra
, NULL
, "array_state", buf
, 20) > 0 &&
807 strncmp(buf
, "read-auto", 9) == 0)
808 sysfs_set_str(sra
, NULL
, "array_state", "clean");
810 err
= sysfs_freeze_array(sra
);
816 static void unfreeze(struct supertype
*st
)
818 if (st
->ss
->external
)
819 return unfreeze_container(st
);
821 struct mdinfo
*sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
825 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
826 strcmp(buf
, "frozen\n") == 0)
827 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
832 static void wait_reshape(struct mdinfo
*sra
)
834 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
840 while (sysfs_fd_get_str(fd
, action
, 20) > 0 &&
841 strncmp(action
, "reshape", 7) == 0)
842 sysfs_wait(fd
, NULL
);
846 static int reshape_super(struct supertype
*st
, unsigned long long size
,
847 int level
, int layout
, int chunksize
, int raid_disks
,
848 int delta_disks
, char *backup_file
, char *dev
,
849 int direction
, int verbose
)
851 /* nothing extra to check in the native case */
852 if (!st
->ss
->external
)
854 if (!st
->ss
->reshape_super
||
855 !st
->ss
->manage_reshape
) {
856 pr_err("%s metadata does not support reshape\n",
861 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
862 raid_disks
, delta_disks
, backup_file
, dev
,
866 static void sync_metadata(struct supertype
*st
)
868 if (st
->ss
->external
) {
869 if (st
->update_tail
) {
870 flush_metadata_updates(st
);
871 st
->update_tail
= &st
->updates
;
873 st
->ss
->sync_metadata(st
);
877 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
879 /* when dealing with external metadata subarrays we need to be
880 * prepared to handle EAGAIN. The kernel may need to wait for
881 * mdmon to mark the array active so the kernel can handle
882 * allocations/writeback when preparing the reshape action
883 * (md_allow_write()). We temporarily disable safe_mode_delay
884 * to close a race with the array_state going clean before the
885 * next write to raid_disks / stripe_cache_size
890 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
892 (strcmp(name
, "raid_disks") != 0 &&
893 strcmp(name
, "stripe_cache_size") != 0))
894 return sysfs_set_num(sra
, NULL
, name
, n
);
896 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
899 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
900 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
901 if (rc
< 0 && errno
== EAGAIN
) {
902 ping_monitor(container
);
903 /* if we get EAGAIN here then the monitor is not active
906 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
908 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
912 int start_reshape(struct mdinfo
*sra
, int already_running
,
913 int before_data_disks
, int data_disks
)
916 unsigned long long sync_max_to_set
;
918 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
919 err
= sysfs_set_num(sra
, NULL
, "suspend_hi", sra
->reshape_progress
);
920 err
= err
?: sysfs_set_num(sra
, NULL
, "suspend_lo",
921 sra
->reshape_progress
);
922 if (before_data_disks
<= data_disks
)
923 sync_max_to_set
= sra
->reshape_progress
/ data_disks
;
925 sync_max_to_set
= (sra
->component_size
* data_disks
926 - sra
->reshape_progress
) / data_disks
;
927 if (!already_running
)
928 sysfs_set_num(sra
, NULL
, "sync_min", sync_max_to_set
);
929 err
= err
?: sysfs_set_num(sra
, NULL
, "sync_max", sync_max_to_set
);
930 if (!already_running
&& err
== 0) {
933 err
= sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
936 } while (err
&& errno
== EBUSY
&& cnt
-- > 0);
941 void abort_reshape(struct mdinfo
*sra
)
943 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
945 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
946 * suspend_hi to decrease as well as increase.")
947 * you could only increase suspend_{lo,hi} unless the region they
948 * covered was empty. So to reset to 0, you need to push suspend_lo
949 * up past suspend_hi first. So to maximize the chance of mdadm
950 * working on all kernels, we want to keep doing that.
952 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
953 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
954 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
955 sysfs_set_num(sra
, NULL
, "sync_min", 0);
956 // It isn't safe to reset sync_max as we aren't monitoring.
957 // Array really should be stopped at this point.
960 int remove_disks_for_takeover(struct supertype
*st
,
965 struct mdinfo
*remaining
;
968 if (st
->ss
->external
) {
970 struct mdinfo
*arrays
= st
->ss
->container_content(st
, NULL
);
972 * containter_content returns list of arrays in container
973 * If arrays->next is not NULL it means that there are
974 * 2 arrays in container and operation should be blocked
981 pr_err("Error. Cannot perform operation on /dev/%s\n", st
->devnm
);
982 pr_err("For this operation it MUST be single array in container\n");
988 if (sra
->array
.level
== 10)
989 nr_of_copies
= layout
& 0xff;
990 else if (sra
->array
.level
== 1)
991 nr_of_copies
= sra
->array
.raid_disks
;
995 remaining
= sra
->devs
;
997 /* for each 'copy', select one device and remove from the list. */
998 for (slot
= 0; slot
< sra
->array
.raid_disks
; slot
+= nr_of_copies
) {
999 struct mdinfo
**diskp
;
1002 /* Find a working device to keep */
1003 for (diskp
= &remaining
; *diskp
; diskp
= &(*diskp
)->next
) {
1004 struct mdinfo
*disk
= *diskp
;
1006 if (disk
->disk
.raid_disk
< slot
)
1008 if (disk
->disk
.raid_disk
>= slot
+ nr_of_copies
)
1010 if (disk
->disk
.state
& (1<<MD_DISK_REMOVED
))
1012 if (disk
->disk
.state
& (1<<MD_DISK_FAULTY
))
1014 if (!(disk
->disk
.state
& (1<<MD_DISK_SYNC
)))
1017 /* We have found a good disk to use! */
1018 *diskp
= disk
->next
;
1019 disk
->next
= sra
->devs
;
1028 if (slot
< sra
->array
.raid_disks
) {
1029 /* didn't find all slots */
1035 sra
->devs
= remaining
;
1039 /* Remove all 'remaining' devices from the array */
1041 struct mdinfo
*sd
= remaining
;
1042 remaining
= sd
->next
;
1044 sysfs_set_str(sra
, sd
, "state", "faulty");
1045 sysfs_set_str(sra
, sd
, "slot", "none");
1046 /* for external metadata disks should be removed in mdmon */
1047 if (!st
->ss
->external
)
1048 sysfs_set_str(sra
, sd
, "state", "remove");
1049 sd
->disk
.state
|= (1<<MD_DISK_REMOVED
);
1050 sd
->disk
.state
&= ~(1<<MD_DISK_SYNC
);
1051 sd
->next
= sra
->devs
;
1057 void reshape_free_fdlist(int *fdlist
,
1058 unsigned long long *offsets
,
1063 for (i
= 0; i
< size
; i
++)
1071 int reshape_prepare_fdlist(char *devname
,
1075 unsigned long blocks
,
1078 unsigned long long *offsets
)
1083 enable_fds(nrdisks
);
1084 for (d
= 0; d
<= nrdisks
; d
++)
1087 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1088 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1090 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
) &&
1091 sd
->disk
.raid_disk
< raid_disks
) {
1092 char *dn
= map_dev(sd
->disk
.major
,
1094 fdlist
[sd
->disk
.raid_disk
]
1095 = dev_open(dn
, O_RDONLY
);
1096 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1097 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1098 pr_err("%s: cannot open component %s\n",
1099 devname
, dn
? dn
: "-unknown-");
1103 } else if (backup_file
== NULL
) {
1105 char *dn
= map_dev(sd
->disk
.major
,
1107 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1108 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1109 if (fdlist
[d
] < 0) {
1110 pr_err("%s: cannot open component %s\n",
1111 devname
, dn
? dn
: "-unknown-");
1122 int reshape_open_backup_file(char *backup_file
,
1127 unsigned long long *offsets
,
1131 /* Return 1 on success, 0 on any form of failure */
1132 /* need to check backup file is large enough */
1138 *fdlist
= open(backup_file
, O_RDWR
|O_CREAT
|(restart
? O_TRUNC
: O_EXCL
),
1142 pr_err("%s: cannot create backup file %s: %s\n",
1143 devname
, backup_file
, strerror(errno
));
1146 /* Guard against backup file being on array device.
1147 * If array is partitioned or if LVM etc is in the
1148 * way this will not notice, but it is better than
1151 fstat(*fdlist
, &stb
);
1154 if (stb
.st_rdev
== dev
) {
1155 pr_err("backup file must NOT be on the array being reshaped.\n");
1160 memset(buf
, 0, 512);
1161 for (i
=0; i
< blocks
+ 8 ; i
++) {
1162 if (write(*fdlist
, buf
, 512) != 512) {
1163 pr_err("%s: cannot create backup file %s: %s\n",
1164 devname
, backup_file
, strerror(errno
));
1168 if (fsync(*fdlist
) != 0) {
1169 pr_err("%s: cannot create backup file %s: %s\n",
1170 devname
, backup_file
, strerror(errno
));
1174 if (!restart
&& strncmp(backup_file
, MAP_DIR
, strlen(MAP_DIR
)) != 0) {
1175 char *bu
= make_backup(sys_name
);
1176 if (symlink(backup_file
, bu
))
1177 pr_err("Recording backup file in " MAP_DIR
" failed: %s\n",
1185 unsigned long compute_backup_blocks(int nchunk
, int ochunk
,
1186 unsigned int ndata
, unsigned int odata
)
1188 unsigned long a
, b
, blocks
;
1189 /* So how much do we need to backup.
1190 * We need an amount of data which is both a whole number of
1191 * old stripes and a whole number of new stripes.
1192 * So LCM for (chunksize*datadisks).
1194 a
= (ochunk
/512) * odata
;
1195 b
= (nchunk
/512) * ndata
;
1198 /* LCM == product / GCD */
1199 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
1204 char *analyse_change(char *devname
, struct mdinfo
*info
, struct reshape
*re
)
1206 /* Based on the current array state in info->array and
1207 * the changes in info->new_* etc, determine:
1208 * - whether the change is possible
1209 * - Intermediate level/raid_disks/layout
1210 * - whether a restriping reshape is needed
1211 * - number of sectors in minimum change unit. This
1212 * will cover a whole number of stripes in 'before' and
1215 * Return message if the change should be rejected
1216 * NULL if the change can be achieved
1218 * This can be called as part of starting a reshape, or
1219 * when assembling an array that is undergoing reshape.
1221 int near
, far
, offset
, copies
;
1223 int old_chunk
, new_chunk
;
1224 /* delta_parity records change in number of devices
1225 * caused by level change
1227 int delta_parity
= 0;
1229 memset(re
, 0, sizeof(*re
));
1231 /* If a new level not explicitly given, we assume no-change */
1232 if (info
->new_level
== UnSet
)
1233 info
->new_level
= info
->array
.level
;
1235 if (info
->new_chunk
)
1236 switch (info
->new_level
) {
1242 /* chunk size is meaningful, must divide component_size
1245 if (info
->component_size
% (info
->new_chunk
/512)) {
1246 unsigned long long shrink
= info
->component_size
;
1247 shrink
&= ~(unsigned long long)(info
->new_chunk
/512-1);
1248 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1249 info
->new_chunk
/1024, info
->component_size
/2);
1250 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1252 pr_err("will shrink the array so the given chunk size would work.\n");
1257 return "chunk size not meaningful for this level";
1260 info
->new_chunk
= info
->array
.chunk_size
;
1262 switch (info
->array
.level
) {
1264 return "No reshape is possibly for this RAID level";
1266 if (info
->delta_disks
!= UnSet
)
1267 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1269 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1271 /* RAID1 can convert to RAID1 with different disks, or
1272 * raid5 with 2 disks, or
1275 if (info
->new_level
> 1 && (info
->component_size
& 7))
1276 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1277 if (info
->new_level
== 0) {
1278 if (info
->delta_disks
!= UnSet
&&
1279 info
->delta_disks
!= 0)
1280 return "Cannot change number of disks with RAID1->RAID0 conversion";
1282 re
->before
.data_disks
= 1;
1283 re
->after
.data_disks
= 1;
1286 if (info
->new_level
== 1) {
1287 if (info
->delta_disks
== UnSet
)
1288 /* Don't know what to do */
1289 return "no change requested for Growing RAID1";
1293 if (info
->array
.raid_disks
!= 2 && info
->new_level
== 5)
1294 return "Can only convert a 2-device array to RAID5";
1295 if (info
->array
.raid_disks
== 2 && info
->new_level
== 5) {
1297 re
->before
.data_disks
= 1;
1298 if (info
->delta_disks
!= UnSet
&&
1299 info
->delta_disks
!= 0)
1300 re
->after
.data_disks
= 1 + info
->delta_disks
;
1302 re
->after
.data_disks
= 1;
1303 if (re
->after
.data_disks
< 1)
1304 return "Number of disks too small for RAID5";
1306 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC
;
1307 info
->array
.chunk_size
= 65536;
1310 /* Could do some multi-stage conversions, but leave that to
1313 return "Impossibly level change request for RAID1";
1316 /* RAID10 can be converted from near mode to
1317 * RAID0 by removing some devices.
1318 * It can also be reshaped if the kernel supports
1321 switch (info
->new_level
) {
1323 if ((info
->array
.layout
& ~0xff) != 0x100)
1324 return "Cannot Grow RAID10 with far/offset layout";
1325 /* number of devices must be multiple of number of copies */
1326 if (info
->array
.raid_disks
% (info
->array
.layout
& 0xff))
1327 return "RAID10 layout too complex for Grow operation";
1329 new_disks
= (info
->array
.raid_disks
1330 / (info
->array
.layout
& 0xff));
1331 if (info
->delta_disks
== UnSet
)
1332 info
->delta_disks
= (new_disks
1333 - info
->array
.raid_disks
);
1335 if (info
->delta_disks
!= new_disks
- info
->array
.raid_disks
)
1336 return "New number of raid-devices impossible for RAID10";
1337 if (info
->new_chunk
&&
1338 info
->new_chunk
!= info
->array
.chunk_size
)
1339 return "Cannot change chunk-size with RAID10 Grow";
1343 re
->before
.data_disks
= new_disks
;
1344 re
->after
.data_disks
= re
->before
.data_disks
;
1348 near
= info
->array
.layout
& 0xff;
1349 far
= (info
->array
.layout
>> 8) & 0xff;
1350 offset
= info
->array
.layout
& 0x10000;
1351 if (far
> 1 && !offset
)
1352 return "Cannot reshape RAID10 in far-mode";
1353 copies
= near
* far
;
1355 old_chunk
= info
->array
.chunk_size
* far
;
1357 if (info
->new_layout
== UnSet
)
1358 info
->new_layout
= info
->array
.layout
;
1360 near
= info
->new_layout
& 0xff;
1361 far
= (info
->new_layout
>> 8) & 0xff;
1362 offset
= info
->new_layout
& 0x10000;
1363 if (far
> 1 && !offset
)
1364 return "Cannot reshape RAID10 to far-mode";
1365 if (near
* far
!= copies
)
1366 return "Cannot change number of copies when reshaping RAID10";
1368 if (info
->delta_disks
== UnSet
)
1369 info
->delta_disks
= 0;
1370 new_disks
= (info
->array
.raid_disks
+
1373 new_chunk
= info
->new_chunk
* far
;
1376 re
->before
.layout
= info
->array
.layout
;
1377 re
->before
.data_disks
= info
->array
.raid_disks
;
1378 re
->after
.layout
= info
->new_layout
;
1379 re
->after
.data_disks
= new_disks
;
1380 /* For RAID10 we don't do backup but do allow reshape,
1381 * so set backup_blocks to INVALID_SECTORS rather than
1383 * And there is no need to synchronise stripes on both
1384 * 'old' and 'new'. So the important
1385 * number is the minimum data_offset difference
1386 * which is the larger of (offset copies * chunk).
1388 re
->backup_blocks
= INVALID_SECTORS
;
1389 re
->min_offset_change
= max(old_chunk
, new_chunk
) / 512;
1390 if (new_disks
< re
->before
.data_disks
&&
1391 info
->space_after
< re
->min_offset_change
)
1392 /* Reduce component size by one chunk */
1393 re
->new_size
= (info
->component_size
-
1394 re
->min_offset_change
);
1396 re
->new_size
= info
->component_size
;
1397 re
->new_size
= re
->new_size
* new_disks
/ copies
;
1401 return "RAID10 can only be changed to RAID0";
1404 /* RAID0 can be converted to RAID10, or to RAID456 */
1405 if (info
->new_level
== 10) {
1406 if (info
->new_layout
== UnSet
&&
1407 info
->delta_disks
== UnSet
) {
1408 /* Assume near=2 layout */
1409 info
->new_layout
= 0x102;
1410 info
->delta_disks
= info
->array
.raid_disks
;
1412 if (info
->new_layout
== UnSet
) {
1413 int copies
= 1 + (info
->delta_disks
1414 / info
->array
.raid_disks
);
1415 if (info
->array
.raid_disks
* (copies
-1) !=
1417 return "Impossible number of devices for RAID0->RAID10";
1418 info
->new_layout
= 0x100 + copies
;
1420 if (info
->delta_disks
== UnSet
) {
1421 int copies
= info
->new_layout
& 0xff;
1422 if (info
->new_layout
!= 0x100 + copies
)
1423 return "New layout impossible for RAID0->RAID10";;
1424 info
->delta_disks
= (copies
- 1) *
1425 info
->array
.raid_disks
;
1427 if (info
->new_chunk
&&
1428 info
->new_chunk
!= info
->array
.chunk_size
)
1429 return "Cannot change chunk-size with RAID0->RAID10";
1432 re
->before
.data_disks
= (info
->array
.raid_disks
+
1434 re
->after
.data_disks
= re
->before
.data_disks
;
1435 re
->before
.layout
= info
->new_layout
;
1439 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1440 * a raid4 style layout of the final level.
1442 switch (info
->new_level
) {
1447 re
->before
.layout
= 0;
1452 re
->before
.layout
= ALGORITHM_PARITY_N
;
1453 if (info
->new_layout
== UnSet
)
1454 info
->new_layout
= map_name(r5layout
, "default");
1459 re
->before
.layout
= ALGORITHM_PARITY_N
;
1460 if (info
->new_layout
== UnSet
)
1461 info
->new_layout
= map_name(r6layout
, "default");
1464 return "Impossible level change requested";
1466 re
->before
.data_disks
= info
->array
.raid_disks
;
1467 /* determining 'after' layout happens outside this 'switch' */
1471 info
->array
.layout
= ALGORITHM_PARITY_N
;
1473 switch (info
->new_level
) {
1477 re
->level
= info
->array
.level
;
1478 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1479 re
->before
.layout
= info
->array
.layout
;
1483 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1484 re
->before
.layout
= info
->array
.layout
;
1489 re
->before
.data_disks
= info
->array
.raid_disks
- 1;
1490 switch (info
->array
.layout
) {
1491 case ALGORITHM_LEFT_ASYMMETRIC
:
1492 re
->before
.layout
= ALGORITHM_LEFT_ASYMMETRIC_6
;
1494 case ALGORITHM_RIGHT_ASYMMETRIC
:
1495 re
->before
.layout
= ALGORITHM_RIGHT_ASYMMETRIC_6
;
1497 case ALGORITHM_LEFT_SYMMETRIC
:
1498 re
->before
.layout
= ALGORITHM_LEFT_SYMMETRIC_6
;
1500 case ALGORITHM_RIGHT_SYMMETRIC
:
1501 re
->before
.layout
= ALGORITHM_RIGHT_SYMMETRIC_6
;
1503 case ALGORITHM_PARITY_0
:
1504 re
->before
.layout
= ALGORITHM_PARITY_0_6
;
1506 case ALGORITHM_PARITY_N
:
1507 re
->before
.layout
= ALGORITHM_PARITY_N_6
;
1510 return "Cannot convert an array with this layout";
1514 if (info
->array
.raid_disks
!= 2)
1515 return "Can only convert a 2-device array to RAID1";
1516 if (info
->delta_disks
!= UnSet
&&
1517 info
->delta_disks
!= 0)
1518 return "Cannot set raid_disk when converting RAID5->RAID1";
1520 info
->new_chunk
= 0;
1523 return "Impossible level change requested";
1527 switch (info
->new_level
) {
1533 re
->before
.data_disks
= info
->array
.raid_disks
- 2;
1534 re
->before
.layout
= info
->array
.layout
;
1537 return "Impossible level change requested";
1542 /* If we reached here then it looks like a re-stripe is
1543 * happening. We have determined the intermediate level
1544 * and initial raid_disks/layout and stored these in 're'.
1546 * We need to deduce the final layout that can be atomically
1547 * converted to the end state.
1549 switch (info
->new_level
) {
1551 /* We can only get to RAID0 from RAID4 or RAID5
1552 * with appropriate layout and one extra device
1554 if (re
->level
!= 4 && re
->level
!= 5)
1555 return "Cannot covert to RAID0 from this level";
1557 switch (re
->level
) {
1559 re
->before
.layout
= 0;
1560 re
->after
.layout
= 0;
1563 re
->after
.layout
= ALGORITHM_PARITY_N
;
1569 /* We can only get to RAID4 from RAID5 */
1570 if (re
->level
!= 4 && re
->level
!= 5)
1571 return "Cannot convert to RAID4 from this level";
1573 switch (re
->level
) {
1575 re
->after
.layout
= 0;
1578 re
->after
.layout
= ALGORITHM_PARITY_N
;
1584 /* We get to RAID5 from RAID5 or RAID6 */
1585 if (re
->level
!= 5 && re
->level
!= 6)
1586 return "Cannot convert to RAID5 from this level";
1588 switch (re
->level
) {
1590 if (info
->new_layout
== UnSet
)
1591 re
->after
.layout
= re
->before
.layout
;
1593 re
->after
.layout
= info
->new_layout
;
1596 if (info
->new_layout
== UnSet
)
1597 info
->new_layout
= re
->before
.layout
;
1599 /* after.layout needs to be raid6 version of new_layout */
1600 if (info
->new_layout
== ALGORITHM_PARITY_N
)
1601 re
->after
.layout
= ALGORITHM_PARITY_N
;
1604 char *ls
= map_num(r5layout
, info
->new_layout
);
1607 /* Current RAID6 layout has a RAID5
1610 strcat(strcpy(layout
, ls
), "-6");
1611 l
= map_name(r6layout
, layout
);
1613 return "Cannot find RAID6 layout to convert to";
1615 /* Current RAID6 has no equivalent.
1616 * If it is already a '-6' layout we
1617 * can leave it unchanged, else we must
1620 ls
= map_num(r6layout
, info
->new_layout
);
1622 strcmp(ls
+strlen(ls
)-2, "-6") != 0)
1623 return "Please specify new layout";
1624 l
= info
->new_layout
;
1626 re
->after
.layout
= l
;
1632 /* We must already be at level 6 */
1634 return "Impossible level change";
1635 if (info
->new_layout
== UnSet
)
1636 re
->after
.layout
= info
->array
.layout
;
1638 re
->after
.layout
= info
->new_layout
;
1641 return "Impossible level change requested";
1643 if (info
->delta_disks
== UnSet
)
1644 info
->delta_disks
= delta_parity
;
1646 re
->after
.data_disks
=
1647 (re
->before
.data_disks
+ info
->delta_disks
- delta_parity
);
1649 switch (re
->level
) {
1661 /* So we have a restripe operation, we need to calculate the number
1662 * of blocks per reshape operation.
1664 re
->new_size
= info
->component_size
* re
->before
.data_disks
;
1665 if (info
->new_chunk
== 0)
1666 info
->new_chunk
= info
->array
.chunk_size
;
1667 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1668 re
->after
.layout
== re
->before
.layout
&&
1669 info
->new_chunk
== info
->array
.chunk_size
) {
1670 /* Nothing to change, can change level immediately. */
1671 re
->level
= info
->new_level
;
1672 re
->backup_blocks
= 0;
1675 if (re
->after
.data_disks
== 1 && re
->before
.data_disks
== 1) {
1676 /* chunk and layout changes make no difference */
1677 re
->level
= info
->new_level
;
1678 re
->backup_blocks
= 0;
1682 if (re
->after
.data_disks
== re
->before
.data_disks
&&
1683 get_linux_version() < 2006032)
1684 return "in-place reshape is not safe before 2.6.32 - sorry.";
1686 if (re
->after
.data_disks
< re
->before
.data_disks
&&
1687 get_linux_version() < 2006030)
1688 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1690 re
->backup_blocks
= compute_backup_blocks(
1691 info
->new_chunk
, info
->array
.chunk_size
,
1692 re
->after
.data_disks
,
1693 re
->before
.data_disks
);
1694 re
->min_offset_change
= re
->backup_blocks
/ re
->before
.data_disks
;
1696 re
->new_size
= info
->component_size
* re
->after
.data_disks
;
1700 static int set_array_size(struct supertype
*st
, struct mdinfo
*sra
,
1703 struct mdinfo
*info
;
1707 if ((st
== NULL
) || (sra
== NULL
))
1710 if (text_version
== NULL
)
1711 text_version
= sra
->text_version
;
1712 subarray
= strchr(text_version
+ 1, '/')+1;
1713 info
= st
->ss
->container_content(st
, subarray
);
1715 unsigned long long current_size
= 0;
1716 unsigned long long new_size
=
1717 info
->custom_array_size
/2;
1719 if (sysfs_get_ll(sra
, NULL
, "array_size", ¤t_size
) == 0 &&
1720 new_size
> current_size
) {
1721 if (sysfs_set_num(sra
, NULL
, "array_size", new_size
)
1723 dprintf("Error: Cannot set array size");
1726 dprintf("Array size changed");
1728 dprintf_cont(" from %llu to %llu.\n",
1729 current_size
, new_size
);
1733 dprintf("Error: set_array_size(): info pointer in NULL\n");
1738 static int reshape_array(char *container
, int fd
, char *devname
,
1739 struct supertype
*st
, struct mdinfo
*info
,
1740 int force
, struct mddev_dev
*devlist
,
1741 unsigned long long data_offset
,
1742 char *backup_file
, int verbose
, int forked
,
1743 int restart
, int freeze_reshape
);
1744 static int reshape_container(char *container
, char *devname
,
1746 struct supertype
*st
,
1747 struct mdinfo
*info
,
1749 char *backup_file
, int verbose
,
1750 int forked
, int restart
, int freeze_reshape
);
1752 int Grow_reshape(char *devname
, int fd
,
1753 struct mddev_dev
*devlist
,
1754 unsigned long long data_offset
,
1755 struct context
*c
, struct shape
*s
)
1757 /* Make some changes in the shape of an array.
1758 * The kernel must support the change.
1760 * There are three different changes. Each can trigger
1761 * a resync or recovery so we freeze that until we have
1762 * requested everything (if kernel supports freezing - 2.6.30).
1764 * - change size (i.e. component_size)
1766 * - change layout/chunksize/ndisks
1768 * The last can require a reshape. It is different on different
1769 * levels so we need to check the level before actioning it.
1770 * Some times the level change needs to be requested after the
1771 * reshape (e.g. raid6->raid5, raid5->raid0)
1774 struct mdu_array_info_s array
;
1776 struct supertype
*st
;
1777 char *subarray
= NULL
;
1781 char *container
= NULL
;
1784 struct mddev_dev
*dv
;
1790 if (md_get_array_info(fd
, &array
) < 0) {
1791 pr_err("%s is not an active md array - aborting\n",
1795 if (data_offset
!= INVALID_SECTORS
&& array
.level
!= 10 &&
1796 (array
.level
< 4 || array
.level
> 6)) {
1797 pr_err("--grow --data-offset not yet supported\n");
1802 (s
->chunk
|| s
->level
!= UnSet
|| s
->layout_str
|| s
->raiddisks
)) {
1803 pr_err("cannot change component size at the same time as other changes.\n"
1804 " Change size first, then check data is intact before making other changes.\n");
1808 if (s
->raiddisks
&& s
->raiddisks
< array
.raid_disks
&&
1809 array
.level
> 1 && get_linux_version() < 2006032 &&
1810 !check_env("MDADM_FORCE_FEWER")) {
1811 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1812 " Please use a newer kernel\n");
1816 if (array
.level
> 1 &&
1817 (array
.chunk_size
/ 1024) > (int)s
->size
) {
1818 pr_err("component size must be larger than chunk size.\n");
1822 st
= super_by_fd(fd
, &subarray
);
1824 pr_err("Unable to determine metadata format for %s\n", devname
);
1827 if (s
->raiddisks
> st
->max_devs
) {
1828 pr_err("Cannot increase raid-disks on this array beyond %d\n", st
->max_devs
);
1831 if (s
->level
== 0 &&
1832 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) &&
1833 !(array
.state
& (1<<MD_SB_CLUSTERED
))) {
1834 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
1835 if (md_set_array_info(fd
, &array
)!= 0) {
1836 pr_err("failed to remove internal bitmap.\n");
1841 /* in the external case we need to check that the requested reshape is
1842 * supported, and perform an initial check that the container holds the
1843 * pre-requisite spare devices (mdmon owns final validation)
1845 if (st
->ss
->external
) {
1849 container
= st
->container_devnm
;
1850 cfd
= open_dev_excl(st
->container_devnm
);
1852 container
= st
->devnm
;
1854 cfd
= open_dev_excl(st
->devnm
);
1858 pr_err("Unable to open container for %s\n",
1864 retval
= st
->ss
->load_container(st
, cfd
, NULL
);
1867 pr_err("Cannot read superblock for %s\n",
1873 /* check if operation is supported for metadata handler */
1874 if (st
->ss
->container_content
) {
1875 struct mdinfo
*cc
= NULL
;
1876 struct mdinfo
*content
= NULL
;
1878 cc
= st
->ss
->container_content(st
, subarray
);
1879 for (content
= cc
; content
; content
= content
->next
) {
1880 int allow_reshape
= 1;
1882 /* check if reshape is allowed based on metadata
1883 * indications stored in content.array.status
1885 if (content
->array
.state
&
1886 (1 << MD_SB_BLOCK_VOLUME
))
1888 if (content
->array
.state
&
1889 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE
))
1891 if (!allow_reshape
) {
1892 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1893 devname
, container
);
1898 if (content
->consistency_policy
==
1899 CONSISTENCY_POLICY_PPL
) {
1900 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1908 if (mdmon_running(container
))
1909 st
->update_tail
= &st
->updates
;
1913 for (dv
= devlist
; dv
; dv
= dv
->next
)
1915 if (s
->raiddisks
> array
.raid_disks
&&
1916 array
.spare_disks
+ added_disks
< (s
->raiddisks
- array
.raid_disks
) &&
1918 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1919 " Use --force to over-ride this check.\n",
1920 s
->raiddisks
- array
.raid_disks
,
1921 s
->raiddisks
- array
.raid_disks
== 1 ? "" : "s",
1922 array
.spare_disks
+ added_disks
);
1926 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
1927 GET_STATE
| GET_VERSION
);
1929 if (st
->ss
->external
&& subarray
== NULL
) {
1930 array
.level
= LEVEL_CONTAINER
;
1931 sra
->array
.level
= LEVEL_CONTAINER
;
1934 pr_err("failed to read sysfs parameters for %s\n",
1938 frozen
= freeze(st
);
1940 /* freeze() already spewed the reason */
1943 } else if (frozen
< 0) {
1944 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname
);
1949 /* ========= set size =============== */
1951 (s
->size
== MAX_SIZE
|| s
->size
!= (unsigned)array
.size
)) {
1952 unsigned long long orig_size
= get_component_size(fd
)/2;
1953 unsigned long long min_csize
;
1955 int raid0_takeover
= 0;
1958 orig_size
= (unsigned) array
.size
;
1960 if (orig_size
== 0) {
1961 pr_err("Cannot set device size in this type of array.\n");
1966 if (reshape_super(st
, s
->size
, UnSet
, UnSet
, 0, 0, UnSet
, NULL
,
1967 devname
, APPLY_METADATA_CHANGES
,
1973 if (st
->ss
->external
) {
1974 /* metadata can have size limitation
1975 * update size value according to metadata information
1977 struct mdinfo
*sizeinfo
=
1978 st
->ss
->container_content(st
, subarray
);
1980 unsigned long long new_size
=
1981 sizeinfo
->custom_array_size
/2;
1982 int data_disks
= get_data_disks(
1983 sizeinfo
->array
.level
,
1984 sizeinfo
->array
.layout
,
1985 sizeinfo
->array
.raid_disks
);
1986 new_size
/= data_disks
;
1987 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1988 orig_size
, new_size
,
1989 new_size
* data_disks
);
1991 sysfs_free(sizeinfo
);
1995 /* Update the size of each member device in case
1996 * they have been resized. This will never reduce
1997 * below the current used-size. The "size" attribute
1998 * understands '0' to mean 'max'.
2001 for (mdi
= sra
->devs
; mdi
; mdi
= mdi
->next
) {
2002 sysfs_set_num(sra
, mdi
, "size", s
->size
== MAX_SIZE
? 0
2004 if (array
.not_persistent
== 0 &&
2005 array
.major_version
== 0 &&
2006 get_linux_version() < 3001000) {
2007 /* Dangerous to allow size to exceed 2TB */
2008 unsigned long long csize
;
2009 if (sysfs_get_ll(sra
, mdi
, "size", &csize
) == 0) {
2010 if (csize
>= 2ULL*1024*1024*1024)
2011 csize
= 2ULL*1024*1024*1024;
2012 if ((min_csize
== 0 || (min_csize
2018 if (min_csize
&& s
->size
> min_csize
) {
2019 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2021 goto size_change_error
;
2023 if (min_csize
&& s
->size
== MAX_SIZE
) {
2024 /* Don't let the kernel choose a size - it will get
2027 pr_err("Limited v0.90 array to 2TB per device\n");
2028 s
->size
= min_csize
;
2030 if (st
->ss
->external
) {
2031 if (sra
->array
.level
== 0) {
2032 rv
= sysfs_set_str(sra
, NULL
, "level",
2036 /* get array parameters after takeover
2037 * to change one parameter at time only
2039 rv
= md_get_array_info(fd
, &array
);
2042 /* make sure mdmon is
2043 * aware of the new level */
2044 if (!mdmon_running(st
->container_devnm
))
2045 start_mdmon(st
->container_devnm
);
2046 ping_monitor(container
);
2047 if (mdmon_running(st
->container_devnm
) &&
2048 st
->update_tail
== NULL
)
2049 st
->update_tail
= &st
->updates
;
2052 if (s
->size
== MAX_SIZE
)
2054 array
.size
= s
->size
;
2055 if (s
->size
& ~INT32_MAX
) {
2056 /* got truncated to 32bit, write to
2057 * component_size instead
2060 rv
= sysfs_set_num(sra
, NULL
,
2061 "component_size", s
->size
);
2065 rv
= md_set_array_info(fd
, &array
);
2067 /* manage array size when it is managed externally
2069 if ((rv
== 0) && st
->ss
->external
)
2070 rv
= set_array_size(st
, sra
, sra
->text_version
);
2073 if (raid0_takeover
) {
2074 /* do not recync non-existing parity,
2075 * we will drop it anyway
2077 sysfs_set_str(sra
, NULL
, "sync_action", "frozen");
2078 /* go back to raid0, drop parity disk
2080 sysfs_set_str(sra
, NULL
, "level", "raid0");
2081 md_get_array_info(fd
, &array
);
2088 /* restore metadata */
2089 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
2090 UnSet
, NULL
, devname
,
2091 ROLLBACK_METADATA_CHANGES
,
2094 pr_err("Cannot set device size for %s: %s\n",
2095 devname
, strerror(err
));
2097 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2098 cont_err("Bitmap must be removed before size can be changed\n");
2102 if (s
->assume_clean
) {
2103 /* This will fail on kernels older than 3.0 unless
2104 * a backport has been arranged.
2107 sysfs_set_str(sra
, NULL
, "resync_start", "none") < 0)
2108 pr_err("--assume-clean not supported with --grow on this kernel\n");
2110 md_get_array_info(fd
, &array
);
2111 s
->size
= get_component_size(fd
)/2;
2113 s
->size
= array
.size
;
2114 if (c
->verbose
>= 0) {
2115 if (s
->size
== orig_size
)
2116 pr_err("component size of %s unchanged at %lluK\n",
2119 pr_err("component size of %s has been set to %lluK\n",
2123 } else if (array
.level
!= LEVEL_CONTAINER
) {
2124 s
->size
= get_component_size(fd
)/2;
2126 s
->size
= array
.size
;
2129 /* See if there is anything else to do */
2130 if ((s
->level
== UnSet
|| s
->level
== array
.level
) &&
2131 (s
->layout_str
== NULL
) &&
2132 (s
->chunk
== 0 || s
->chunk
== array
.chunk_size
) &&
2133 data_offset
== INVALID_SECTORS
&&
2134 (s
->raiddisks
== 0 || s
->raiddisks
== array
.raid_disks
)) {
2135 /* Nothing more to do */
2136 if (!changed
&& c
->verbose
>= 0)
2137 pr_err("%s: no change requested\n",
2142 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2143 * current implementation assumes that following conditions must be met:
2146 * - near_copies == 2
2148 if ((s
->level
== 0 && array
.level
== 10 && sra
&&
2149 array
.layout
== ((1 << 8) + 2) && !(array
.raid_disks
& 1)) ||
2150 (s
->level
== 0 && array
.level
== 1 && sra
)) {
2152 err
= remove_disks_for_takeover(st
, sra
, array
.layout
);
2154 dprintf("Array cannot be reshaped\n");
2160 /* Make sure mdmon has seen the device removal
2161 * and updated metadata before we continue with
2165 ping_monitor(container
);
2168 memset(&info
, 0, sizeof(info
));
2170 if (sysfs_init(&info
, fd
, NULL
)) {
2171 pr_err("failed to intialize sysfs.\n");
2175 strcpy(info
.text_version
, sra
->text_version
);
2176 info
.component_size
= s
->size
*2;
2177 info
.new_level
= s
->level
;
2178 info
.new_chunk
= s
->chunk
* 1024;
2179 if (info
.array
.level
== LEVEL_CONTAINER
) {
2180 info
.delta_disks
= UnSet
;
2181 info
.array
.raid_disks
= s
->raiddisks
;
2182 } else if (s
->raiddisks
)
2183 info
.delta_disks
= s
->raiddisks
- info
.array
.raid_disks
;
2185 info
.delta_disks
= UnSet
;
2186 if (s
->layout_str
== NULL
) {
2187 info
.new_layout
= UnSet
;
2188 if (info
.array
.level
== 6 &&
2189 (info
.new_level
== 6 || info
.new_level
== UnSet
) &&
2190 info
.array
.layout
>= 16) {
2191 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname
);
2192 cont_err("during the reshape, please specify --layout=preserve\n");
2193 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2197 } else if (strcmp(s
->layout_str
, "normalise") == 0 ||
2198 strcmp(s
->layout_str
, "normalize") == 0) {
2199 /* If we have a -6 RAID6 layout, remove the '-6'. */
2200 info
.new_layout
= UnSet
;
2201 if (info
.array
.level
== 6 && info
.new_level
== UnSet
) {
2203 strcpy(l
, map_num(r6layout
, info
.array
.layout
));
2204 h
= strrchr(l
, '-');
2205 if (h
&& strcmp(h
, "-6") == 0) {
2207 info
.new_layout
= map_name(r6layout
, l
);
2210 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s
->layout_str
);
2214 } else if (strcmp(s
->layout_str
, "preserve") == 0) {
2215 /* This means that a non-standard RAID6 layout
2218 * - When reshape a RAID6 (e.g. adding a device)
2219 * which is in a non-standard layout, it is OK
2220 * to preserve that layout.
2221 * - When converting a RAID5 to RAID6, leave it in
2222 * the XXX-6 layout, don't re-layout.
2224 if (info
.array
.level
== 6 && info
.new_level
== UnSet
)
2225 info
.new_layout
= info
.array
.layout
;
2226 else if (info
.array
.level
== 5 && info
.new_level
== 6) {
2228 strcpy(l
, map_num(r5layout
, info
.array
.layout
));
2230 info
.new_layout
= map_name(r6layout
, l
);
2232 pr_err("%s in only meaningful when reshaping to RAID6\n", s
->layout_str
);
2237 int l
= info
.new_level
;
2239 l
= info
.array
.level
;
2242 info
.new_layout
= map_name(r5layout
, s
->layout_str
);
2245 info
.new_layout
= map_name(r6layout
, s
->layout_str
);
2248 info
.new_layout
= parse_layout_10(s
->layout_str
);
2251 info
.new_layout
= parse_layout_faulty(s
->layout_str
);
2254 pr_err("layout not meaningful with this level\n");
2258 if (info
.new_layout
== UnSet
) {
2259 pr_err("layout %s not understood for this level\n",
2266 if (array
.level
== LEVEL_FAULTY
) {
2267 if (s
->level
!= UnSet
&& s
->level
!= array
.level
) {
2268 pr_err("cannot change level of Faulty device\n");
2272 pr_err("cannot set chunksize of Faulty device\n");
2275 if (s
->raiddisks
&& s
->raiddisks
!= 1) {
2276 pr_err("cannot set raid_disks of Faulty device\n");
2279 if (s
->layout_str
) {
2280 if (md_get_array_info(fd
, &array
) != 0) {
2281 dprintf("Cannot get array information.\n");
2284 array
.layout
= info
.new_layout
;
2285 if (md_set_array_info(fd
, &array
) != 0) {
2286 pr_err("failed to set new layout\n");
2288 } else if (c
->verbose
>= 0)
2289 printf("layout for %s set to %d\n",
2290 devname
, array
.layout
);
2292 } else if (array
.level
== LEVEL_CONTAINER
) {
2293 /* This change is to be applied to every array in the
2294 * container. This is only needed when the metadata imposes
2295 * restraints of the various arrays in the container.
2296 * Currently we only know that IMSM requires all arrays
2297 * to have the same number of devices so changing the
2298 * number of devices (On-Line Capacity Expansion) must be
2299 * performed at the level of the container
2305 rv
= reshape_container(container
, devname
, -1, st
, &info
,
2306 c
->force
, c
->backup_file
, c
->verbose
, 0, 0, 0);
2309 /* get spare devices from external metadata
2311 if (st
->ss
->external
) {
2312 struct mdinfo
*info2
;
2314 info2
= st
->ss
->container_content(st
, subarray
);
2316 info
.array
.spare_disks
=
2317 info2
->array
.spare_disks
;
2322 /* Impose these changes on a single array. First
2323 * check that the metadata is OK with the change. */
2325 if (reshape_super(st
, 0, info
.new_level
,
2326 info
.new_layout
, info
.new_chunk
,
2327 info
.array
.raid_disks
, info
.delta_disks
,
2328 c
->backup_file
, devname
, APPLY_METADATA_CHANGES
,
2334 rv
= reshape_array(container
, fd
, devname
, st
, &info
, c
->force
,
2335 devlist
, data_offset
, c
->backup_file
, c
->verbose
,
2346 /* verify_reshape_position()
2347 * Function checks if reshape position in metadata is not farther
2348 * than position in md.
2350 * 0 : not valid sysfs entry
2351 * it can be caused by not started reshape, it should be started
2352 * by reshape array or raid0 array is before takeover
2353 * -1 : error, reshape position is obviously wrong
2354 * 1 : success, reshape progress correct or updated
2356 static int verify_reshape_position(struct mdinfo
*info
, int level
)
2362 /* read sync_max, failure can mean raid0 array */
2363 rv
= sysfs_get_str(info
, NULL
, "sync_max", buf
, 40);
2367 unsigned long long position
= strtoull(buf
, &ep
, 0);
2369 dprintf("Read sync_max sysfs entry is: %s\n", buf
);
2370 if (!(ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))) {
2371 position
*= get_data_disks(level
,
2373 info
->array
.raid_disks
);
2374 if (info
->reshape_progress
< position
) {
2375 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2376 info
->reshape_progress
, position
);
2377 info
->reshape_progress
= position
;
2379 } else if (info
->reshape_progress
> position
) {
2380 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2381 position
, info
->reshape_progress
);
2384 dprintf("Reshape position in md and metadata are the same;");
2388 } else if (rv
== 0) {
2389 /* for valid sysfs entry, 0-length content
2390 * should be indicated as error
2398 static unsigned long long choose_offset(unsigned long long lo
,
2399 unsigned long long hi
,
2400 unsigned long long min
,
2401 unsigned long long max
)
2403 /* Choose a new offset between hi and lo.
2404 * It must be between min and max, but
2405 * we would prefer something near the middle of hi/lo, and also
2406 * prefer to be aligned to a big power of 2.
2408 * So we start with the middle, then for each bit,
2409 * starting at '1' and increasing, if it is set, we either
2410 * add it or subtract it if possible, preferring the option
2411 * which is furthest from the boundary.
2413 * We stop once we get a 1MB alignment. As units are in sectors,
2414 * 1MB = 2*1024 sectors.
2416 unsigned long long choice
= (lo
+ hi
) / 2;
2417 unsigned long long bit
= 1;
2419 for (bit
= 1; bit
< 2*1024; bit
= bit
<< 1) {
2420 unsigned long long bigger
, smaller
;
2421 if (! (bit
& choice
))
2423 bigger
= choice
+ bit
;
2424 smaller
= choice
- bit
;
2425 if (bigger
> max
&& smaller
< min
)
2429 else if (smaller
< min
)
2431 else if (hi
- bigger
> smaller
- lo
)
2439 static int set_new_data_offset(struct mdinfo
*sra
, struct supertype
*st
,
2440 char *devname
, int delta_disks
,
2441 unsigned long long data_offset
,
2442 unsigned long long min
,
2448 unsigned long long before
, after
;
2450 /* Need to find min space before and after so same is used
2453 before
= UINT64_MAX
;
2455 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2459 struct supertype
*st2
;
2460 struct mdinfo info2
;
2462 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2464 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2465 dfd
= dev_open(dn
, O_RDONLY
);
2467 pr_err("%s: cannot open component %s\n",
2468 devname
, dn
? dn
: "-unknown-");
2471 st2
= dup_super(st
);
2472 rv
= st2
->ss
->load_super(st2
,dfd
, NULL
);
2476 pr_err("%s: cannot get superblock from %s\n",
2480 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2481 st2
->ss
->free_super(st2
);
2483 if (info2
.space_before
== 0 &&
2484 info2
.space_after
== 0) {
2485 /* Metadata doesn't support data_offset changes */
2487 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2491 if (before
> info2
.space_before
)
2492 before
= info2
.space_before
;
2493 if (after
> info2
.space_after
)
2494 after
= info2
.space_after
;
2496 if (data_offset
!= INVALID_SECTORS
) {
2498 if (info2
.data_offset
== data_offset
) {
2499 pr_err("%s: already has that data_offset\n",
2503 if (data_offset
< info2
.data_offset
)
2507 } else if ((data_offset
<= info2
.data_offset
&& dir
== 1) ||
2508 (data_offset
>= info2
.data_offset
&& dir
== -1)) {
2509 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2515 if (before
== UINT64_MAX
)
2516 /* impossible really, there must be no devices */
2519 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2520 char *dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2521 unsigned long long new_data_offset
;
2523 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2525 if (delta_disks
< 0) {
2526 /* Don't need any space as array is shrinking
2527 * just move data_offset up by min
2529 if (data_offset
== INVALID_SECTORS
)
2530 new_data_offset
= sd
->data_offset
+ min
;
2532 if (data_offset
< sd
->data_offset
+ min
) {
2533 pr_err("--data-offset too small for %s\n",
2537 new_data_offset
= data_offset
;
2539 } else if (delta_disks
> 0) {
2540 /* need space before */
2544 pr_err("Insufficient head-space for reshape on %s\n",
2548 if (data_offset
== INVALID_SECTORS
)
2549 new_data_offset
= sd
->data_offset
- min
;
2551 if (data_offset
> sd
->data_offset
- min
) {
2552 pr_err("--data-offset too large for %s\n",
2556 new_data_offset
= data_offset
;
2560 /* can move up or down. If 'data_offset'
2561 * was set we would have already decided,
2562 * so just choose direction with most space.
2569 sysfs_set_str(sra
, NULL
, "reshape_direction",
2570 dir
== 1 ? "backwards" : "forwards");
2572 /* Increase data offset */
2576 pr_err("Insufficient tail-space for reshape on %s\n",
2580 if (data_offset
!= INVALID_SECTORS
&&
2581 data_offset
< sd
->data_offset
+ min
) {
2582 pr_err("--data-offset too small on %s\n",
2586 if (data_offset
!= INVALID_SECTORS
)
2587 new_data_offset
= data_offset
;
2589 new_data_offset
= choose_offset(sd
->data_offset
,
2590 sd
->data_offset
+ after
,
2591 sd
->data_offset
+ min
,
2592 sd
->data_offset
+ after
);
2594 /* Decrease data offset */
2598 pr_err("insufficient head-room on %s\n",
2602 if (data_offset
!= INVALID_SECTORS
&&
2603 data_offset
< sd
->data_offset
- min
) {
2604 pr_err("--data-offset too small on %s\n",
2608 if (data_offset
!= INVALID_SECTORS
)
2609 new_data_offset
= data_offset
;
2611 new_data_offset
= choose_offset(sd
->data_offset
- before
,
2613 sd
->data_offset
- before
,
2614 sd
->data_offset
- min
);
2617 err
= sysfs_set_num(sra
, sd
, "new_offset", new_data_offset
);
2618 if (err
< 0 && errno
== E2BIG
) {
2619 /* try again after increasing data size to max */
2620 err
= sysfs_set_num(sra
, sd
, "size", 0);
2621 if (err
< 0 && errno
== EINVAL
&&
2622 !(sd
->disk
.state
& (1<<MD_DISK_SYNC
))) {
2623 /* some kernels have a bug where you cannot
2624 * use '0' on spare devices. */
2625 sysfs_set_num(sra
, sd
, "size",
2626 (sra
->component_size
+ after
)/2);
2628 err
= sysfs_set_num(sra
, sd
, "new_offset",
2632 if (errno
== E2BIG
&& data_offset
!= INVALID_SECTORS
) {
2633 pr_err("data-offset is too big for %s\n",
2637 if (sd
== sra
->devs
&&
2638 (errno
== ENOENT
|| errno
== E2BIG
))
2639 /* Early kernel, no 'new_offset' file,
2640 * or kernel doesn't like us.
2641 * For RAID5/6 this is not fatal
2644 pr_err("Cannot set new_offset for %s\n",
2653 /* Just use a backup file */
2657 static int raid10_reshape(char *container
, int fd
, char *devname
,
2658 struct supertype
*st
, struct mdinfo
*info
,
2659 struct reshape
*reshape
,
2660 unsigned long long data_offset
,
2661 int force
, int verbose
)
2663 /* Changing raid_disks, layout, chunksize or possibly
2664 * just data_offset for a RAID10.
2665 * We must always change data_offset. We change by at least
2666 * ->min_offset_change which is the largest of the old and new
2668 * If raid_disks is increasing, then data_offset must decrease
2669 * by at least this copy size.
2670 * If raid_disks is unchanged, data_offset must increase or
2671 * decrease by at least min_offset_change but preferably by much more.
2672 * We choose half of the available space.
2673 * If raid_disks is decreasing, data_offset must increase by
2674 * at least min_offset_change. To allow of this, component_size
2675 * must be decreased by the same amount.
2677 * So we calculate the required minimum and direction, possibly
2678 * reduce the component_size, then iterate through the devices
2679 * and set the new_data_offset.
2680 * If that all works, we set chunk_size, layout, raid_disks, and start
2684 unsigned long long min
;
2687 sra
= sysfs_read(fd
, NULL
,
2688 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
2691 pr_err("%s: Cannot get array details from sysfs\n",
2695 min
= reshape
->min_offset_change
;
2697 if (info
->delta_disks
)
2698 sysfs_set_str(sra
, NULL
, "reshape_direction",
2699 info
->delta_disks
< 0 ? "backwards" : "forwards");
2700 if (info
->delta_disks
< 0 &&
2701 info
->space_after
< min
) {
2702 int rv
= sysfs_set_num(sra
, NULL
, "component_size",
2703 (sra
->component_size
-
2706 pr_err("cannot reduce component size\n");
2710 err
= set_new_data_offset(sra
, st
, devname
, info
->delta_disks
, data_offset
,
2713 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2714 cont_err("supported on this kernel\n");
2720 if (!err
&& sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2722 if (!err
&& sysfs_set_num(sra
, NULL
, "layout", reshape
->after
.layout
) < 0)
2724 if (!err
&& sysfs_set_num(sra
, NULL
, "raid_disks",
2725 info
->array
.raid_disks
+ info
->delta_disks
) < 0)
2727 if (!err
&& sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0)
2730 pr_err("Cannot set array shape for %s\n",
2733 (info
->array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2734 cont_err(" Bitmap must be removed before shape can be changed\n");
2744 static void get_space_after(int fd
, struct supertype
*st
, struct mdinfo
*info
)
2746 struct mdinfo
*sra
, *sd
;
2747 /* Initialisation to silence compiler warning */
2748 unsigned long long min_space_before
= 0, min_space_after
= 0;
2751 sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
2754 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2757 struct supertype
*st2
;
2758 struct mdinfo info2
;
2760 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2762 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 0);
2763 dfd
= dev_open(dn
, O_RDONLY
);
2766 st2
= dup_super(st
);
2767 if (st2
->ss
->load_super(st2
,dfd
, NULL
)) {
2773 st2
->ss
->getinfo_super(st2
, &info2
, NULL
);
2774 st2
->ss
->free_super(st2
);
2777 min_space_before
> info2
.space_before
)
2778 min_space_before
= info2
.space_before
;
2780 min_space_after
> info2
.space_after
)
2781 min_space_after
= info2
.space_after
;
2784 if (sd
== NULL
&& !first
) {
2785 info
->space_after
= min_space_after
;
2786 info
->space_before
= min_space_before
;
2791 static void update_cache_size(char *container
, struct mdinfo
*sra
,
2792 struct mdinfo
*info
,
2793 int disks
, unsigned long long blocks
)
2795 /* Check that the internal stripe cache is
2796 * large enough, or it won't work.
2797 * It must hold at least 4 stripes of the larger
2800 unsigned long cache
;
2801 cache
= max(info
->array
.chunk_size
, info
->new_chunk
);
2802 cache
*= 4; /* 4 stripes minimum */
2803 cache
/= 512; /* convert to sectors */
2804 /* make sure there is room for 'blocks' with a bit to spare */
2805 if (cache
< 16 + blocks
/ disks
)
2806 cache
= 16 + blocks
/ disks
;
2807 cache
/= (4096/512); /* Convert from sectors to pages */
2809 if (sra
->cache_size
< cache
)
2810 subarray_set_num(container
, sra
, "stripe_cache_size",
2814 static int impose_reshape(struct mdinfo
*sra
,
2815 struct mdinfo
*info
,
2816 struct supertype
*st
,
2819 char *devname
, char *container
,
2820 struct reshape
*reshape
)
2822 struct mdu_array_info_s array
;
2824 sra
->new_chunk
= info
->new_chunk
;
2827 /* for external metadata checkpoint saved by mdmon can be lost
2828 * or missed /due to e.g. crash/. Check if md is not during
2829 * restart farther than metadata points to.
2830 * If so, this means metadata information is obsolete.
2832 if (st
->ss
->external
)
2833 verify_reshape_position(info
, reshape
->level
);
2834 sra
->reshape_progress
= info
->reshape_progress
;
2836 sra
->reshape_progress
= 0;
2837 if (reshape
->after
.data_disks
< reshape
->before
.data_disks
)
2838 /* start from the end of the new array */
2839 sra
->reshape_progress
= (sra
->component_size
2840 * reshape
->after
.data_disks
);
2843 md_get_array_info(fd
, &array
);
2844 if (info
->array
.chunk_size
== info
->new_chunk
&&
2845 reshape
->before
.layout
== reshape
->after
.layout
&&
2846 st
->ss
->external
== 0) {
2847 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2848 array
.raid_disks
= reshape
->after
.data_disks
+ reshape
->parity
;
2849 if (!restart
&& md_set_array_info(fd
, &array
) != 0) {
2852 pr_err("Cannot set device shape for %s: %s\n",
2853 devname
, strerror(errno
));
2856 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2857 cont_err("Bitmap must be removed before shape can be changed\n");
2861 } else if (!restart
) {
2862 /* set them all just in case some old 'new_*' value
2863 * persists from some earlier problem.
2866 if (sysfs_set_num(sra
, NULL
, "chunk_size", info
->new_chunk
) < 0)
2868 if (!err
&& sysfs_set_num(sra
, NULL
, "layout",
2869 reshape
->after
.layout
) < 0)
2871 if (!err
&& subarray_set_num(container
, sra
, "raid_disks",
2872 reshape
->after
.data_disks
+
2873 reshape
->parity
) < 0)
2876 pr_err("Cannot set device shape for %s\n",
2880 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2881 cont_err("Bitmap must be removed before shape can be changed\n");
2890 static int impose_level(int fd
, int level
, char *devname
, int verbose
)
2893 struct mdu_array_info_s array
;
2896 if (sysfs_init(&info
, fd
, NULL
)) {
2897 pr_err("failed to intialize sysfs.\n");
2901 md_get_array_info(fd
, &array
);
2903 (array
.level
>= 4 && array
.level
<= 6)) {
2904 /* To convert to RAID0 we need to fail and
2905 * remove any non-data devices. */
2908 int data_disks
= array
.raid_disks
- 1;
2909 if (array
.level
== 6)
2911 if (array
.level
== 5 &&
2912 array
.layout
!= ALGORITHM_PARITY_N
)
2914 if (array
.level
== 6 &&
2915 array
.layout
!= ALGORITHM_PARITY_N_6
)
2917 sysfs_set_str(&info
, NULL
,"sync_action", "idle");
2918 /* First remove any spares so no recovery starts */
2919 for (d
= 0, found
= 0;
2920 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2922 mdu_disk_info_t disk
;
2924 if (md_get_disk_info(fd
, &disk
) < 0)
2926 if (disk
.major
== 0 && disk
.minor
== 0)
2929 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2930 disk
.raid_disk
< data_disks
)
2933 ioctl(fd
, HOT_REMOVE_DISK
,
2934 makedev(disk
.major
, disk
.minor
));
2936 /* Now fail anything left */
2937 md_get_array_info(fd
, &array
);
2938 for (d
= 0, found
= 0;
2939 d
< MAX_DISKS
&& found
< array
.nr_disks
;
2941 mdu_disk_info_t disk
;
2943 if (md_get_disk_info(fd
, &disk
) < 0)
2945 if (disk
.major
== 0 && disk
.minor
== 0)
2948 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) &&
2949 disk
.raid_disk
< data_disks
)
2952 ioctl(fd
, SET_DISK_FAULTY
,
2953 makedev(disk
.major
, disk
.minor
));
2954 hot_remove_disk(fd
, makedev(disk
.major
, disk
.minor
), 1);
2957 c
= map_num(pers
, level
);
2959 int err
= sysfs_set_str(&info
, NULL
, "level", c
);
2962 pr_err("%s: could not set level to %s\n",
2965 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
2966 cont_err("Bitmap must be removed before level can be changed\n");
2970 pr_err("level of %s changed to %s\n",
2977 static void catch_term(int sig
)
2982 static int continue_via_systemd(char *devnm
)
2984 int skipped
, i
, pid
, status
;
2986 /* In a systemd/udev world, it is best to get systemd to
2987 * run "mdadm --grow --continue" rather than running in the
2992 /* FIXME yuk. CLOSE_EXEC?? */
2994 for (i
= 3; skipped
< 20; i
++)
3000 /* Don't want to see error messages from
3001 * systemctl. If the service doesn't exist,
3002 * we fork ourselves.
3005 open("/dev/null", O_WRONLY
);
3006 snprintf(pathbuf
, sizeof(pathbuf
), "mdadm-grow-continue@%s.service",
3008 status
= execl("/usr/bin/systemctl", "systemctl",
3011 status
= execl("/bin/systemctl", "systemctl", "restart",
3014 case -1: /* Just do it ourselves. */
3016 default: /* parent - good */
3017 pid
= wait(&status
);
3018 if (pid
>= 0 && status
== 0)
3024 static int reshape_array(char *container
, int fd
, char *devname
,
3025 struct supertype
*st
, struct mdinfo
*info
,
3026 int force
, struct mddev_dev
*devlist
,
3027 unsigned long long data_offset
,
3028 char *backup_file
, int verbose
, int forked
,
3029 int restart
, int freeze_reshape
)
3031 struct reshape reshape
;
3034 int orig_level
= UnSet
;
3038 struct mdu_array_info_s array
;
3041 struct mddev_dev
*dv
;
3045 unsigned long long *offsets
= NULL
;
3049 unsigned long blocks
;
3050 unsigned long long array_size
;
3052 struct mdinfo
*sra
= NULL
;
3055 /* when reshaping a RAID0, the component_size might be zero.
3056 * So try to fix that up.
3058 if (md_get_array_info(fd
, &array
) != 0) {
3059 dprintf("Cannot get array information.\n");
3062 if (array
.level
== 0 && info
->component_size
== 0) {
3063 get_dev_size(fd
, NULL
, &array_size
);
3064 info
->component_size
= array_size
/ array
.raid_disks
;
3067 if (array
.level
== 10)
3068 /* Need space_after info */
3069 get_space_after(fd
, st
, info
);
3071 if (info
->reshape_active
) {
3072 int new_level
= info
->new_level
;
3073 info
->new_level
= UnSet
;
3074 if (info
->delta_disks
> 0)
3075 info
->array
.raid_disks
-= info
->delta_disks
;
3076 msg
= analyse_change(devname
, info
, &reshape
);
3077 info
->new_level
= new_level
;
3078 if (info
->delta_disks
> 0)
3079 info
->array
.raid_disks
+= info
->delta_disks
;
3081 /* Make sure the array isn't read-only */
3082 ioctl(fd
, RESTART_ARRAY_RW
, 0);
3084 msg
= analyse_change(devname
, info
, &reshape
);
3086 /* if msg == "", error has already been printed */
3088 pr_err("%s\n", msg
);
3092 (reshape
.level
!= info
->array
.level
||
3093 reshape
.before
.layout
!= info
->array
.layout
||
3094 reshape
.before
.data_disks
+ reshape
.parity
!=
3095 info
->array
.raid_disks
- max(0, info
->delta_disks
))) {
3096 pr_err("reshape info is not in native format - cannot continue.\n");
3100 if (st
->ss
->external
&& restart
&& (info
->reshape_progress
== 0) &&
3101 !((sysfs_get_str(info
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0) &&
3102 (strncmp(buf
, "reshape", 7) == 0))) {
3103 /* When reshape is restarted from '0', very begin of array
3104 * it is possible that for external metadata reshape and array
3105 * configuration doesn't happen.
3106 * Check if md has the same opinion, and reshape is restarted
3107 * from 0. If so, this is regular reshape start after reshape
3108 * switch in metadata to next array only.
3110 if ((verify_reshape_position(info
, reshape
.level
) >= 0) &&
3111 (info
->reshape_progress
== 0))
3115 /* reshape already started. just skip to monitoring the reshape */
3116 if (reshape
.backup_blocks
== 0)
3118 if (restart
& RESHAPE_NO_BACKUP
)
3121 /* Need 'sra' down at 'started:' */
3122 sra
= sysfs_read(fd
, NULL
,
3123 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3126 pr_err("%s: Cannot get array details from sysfs\n",
3132 backup_file
= locate_backup(sra
->sys_name
);
3136 /* The container is frozen but the array may not be.
3137 * So freeze the array so spares don't get put to the wrong use
3138 * FIXME there should probably be a cleaner separation between
3139 * freeze_array and freeze_container.
3141 sysfs_freeze_array(info
);
3142 /* Check we have enough spares to not be degraded */
3144 for (dv
= devlist
; dv
; dv
=dv
->next
)
3146 spares_needed
= max(reshape
.before
.data_disks
,
3147 reshape
.after
.data_disks
)
3148 + reshape
.parity
- array
.raid_disks
;
3151 info
->new_level
> 1 && info
->array
.level
> 1 &&
3152 spares_needed
> info
->array
.spare_disks
+ added_disks
) {
3153 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3154 " Use --force to over-ride this check.\n",
3156 spares_needed
== 1 ? "" : "s",
3157 info
->array
.spare_disks
+ added_disks
);
3160 /* Check we have enough spares to not fail */
3161 spares_needed
= max(reshape
.before
.data_disks
,
3162 reshape
.after
.data_disks
)
3164 if ((info
->new_level
> 1 || info
->new_level
== 0) &&
3165 spares_needed
> info
->array
.spare_disks
+added_disks
) {
3166 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3168 spares_needed
== 1 ? "" : "s",
3169 info
->array
.spare_disks
+ added_disks
);
3173 if (reshape
.level
!= array
.level
) {
3174 int err
= impose_level(fd
, reshape
.level
, devname
, verbose
);
3177 info
->new_layout
= UnSet
; /* after level change,
3178 * layout is meaningless */
3179 orig_level
= array
.level
;
3180 sysfs_freeze_array(info
);
3182 if (reshape
.level
> 0 && st
->ss
->external
) {
3183 /* make sure mdmon is aware of the new level */
3184 if (mdmon_running(container
))
3185 flush_mdmon(container
);
3187 if (!mdmon_running(container
))
3188 start_mdmon(container
);
3189 ping_monitor(container
);
3190 if (mdmon_running(container
) &&
3191 st
->update_tail
== NULL
)
3192 st
->update_tail
= &st
->updates
;
3195 /* ->reshape_super might have chosen some spares from the
3196 * container that it wants to be part of the new array.
3197 * We can collect them with ->container_content and give
3198 * them to the kernel.
3200 if (st
->ss
->reshape_super
&& st
->ss
->container_content
) {
3201 char *subarray
= strchr(info
->text_version
+1, '/')+1;
3202 struct mdinfo
*info2
=
3203 st
->ss
->container_content(st
, subarray
);
3207 if (sysfs_init(info2
, fd
, st
->devnm
)) {
3208 pr_err("unable to initialize sysfs for %s\n",
3213 /* When increasing number of devices, we need to set
3214 * new raid_disks before adding these, or they might
3217 if (reshape
.backup_blocks
&&
3218 reshape
.after
.data_disks
> reshape
.before
.data_disks
)
3219 subarray_set_num(container
, info2
, "raid_disks",
3220 reshape
.after
.data_disks
+
3222 for (d
= info2
->devs
; d
; d
= d
->next
) {
3223 if (d
->disk
.state
== 0 &&
3224 d
->disk
.raid_disk
>= 0) {
3225 /* This is a spare that wants to
3226 * be part of the array.
3228 add_disk(fd
, st
, info2
, d
);
3234 /* We might have been given some devices to add to the
3235 * array. Now that the array has been changed to the right
3236 * level and frozen, we can safely add them.
3239 if (Manage_subdevs(devname
, fd
, devlist
, verbose
,
3244 if (reshape
.backup_blocks
== 0 && data_offset
!= INVALID_SECTORS
)
3245 reshape
.backup_blocks
= reshape
.before
.data_disks
* info
->array
.chunk_size
/512;
3246 if (reshape
.backup_blocks
== 0) {
3247 /* No restriping needed, but we might need to impose
3248 * some more changes: layout, raid_disks, chunk_size
3250 /* read current array info */
3251 if (md_get_array_info(fd
, &array
) != 0) {
3252 dprintf("Cannot get array information.\n");
3255 /* compare current array info with new values and if
3256 * it is different update them to new */
3257 if (info
->new_layout
!= UnSet
&&
3258 info
->new_layout
!= array
.layout
) {
3259 array
.layout
= info
->new_layout
;
3260 if (md_set_array_info(fd
, &array
) != 0) {
3261 pr_err("failed to set new layout\n");
3263 } else if (verbose
>= 0)
3264 printf("layout for %s set to %d\n",
3265 devname
, array
.layout
);
3267 if (info
->delta_disks
!= UnSet
&&
3268 info
->delta_disks
!= 0 &&
3269 array
.raid_disks
!= (info
->array
.raid_disks
+ info
->delta_disks
)) {
3270 array
.raid_disks
+= info
->delta_disks
;
3271 if (md_set_array_info(fd
, &array
) != 0) {
3272 pr_err("failed to set raid disks\n");
3274 } else if (verbose
>= 0) {
3275 printf("raid_disks for %s set to %d\n",
3276 devname
, array
.raid_disks
);
3279 if (info
->new_chunk
!= 0 &&
3280 info
->new_chunk
!= array
.chunk_size
) {
3281 if (sysfs_set_num(info
, NULL
,
3282 "chunk_size", info
->new_chunk
) != 0) {
3283 pr_err("failed to set chunk size\n");
3285 } else if (verbose
>= 0)
3286 printf("chunk size for %s set to %d\n",
3287 devname
, array
.chunk_size
);
3294 * There are three possibilities.
3295 * 1/ The array will shrink.
3296 * We need to ensure the reshape will pause before reaching
3297 * the 'critical section'. We also need to fork and wait for
3298 * that to happen. When it does we
3299 * suspend/backup/complete/unfreeze
3301 * 2/ The array will not change size.
3302 * This requires that we keep a backup of a sliding window
3303 * so that we can restore data after a crash. So we need
3304 * to fork and monitor progress.
3305 * In future we will allow the data_offset to change, so
3306 * a sliding backup becomes unnecessary.
3308 * 3/ The array will grow. This is relatively easy.
3309 * However the kernel's restripe routines will cheerfully
3310 * overwrite some early data before it is safe. So we
3311 * need to make a backup of the early parts of the array
3312 * and be ready to restore it if rebuild aborts very early.
3313 * For externally managed metadata, we still need a forked
3314 * child to monitor the reshape and suspend IO over the region
3315 * that is being reshaped.
3317 * We backup data by writing it to one spare, or to a
3318 * file which was given on command line.
3320 * In each case, we first make sure that storage is available
3321 * for the required backup.
3323 * - request the shape change.
3324 * - fork to handle backup etc.
3326 /* Check that we can hold all the data */
3327 get_dev_size(fd
, NULL
, &array_size
);
3328 if (reshape
.new_size
< (array_size
/512)) {
3329 pr_err("this change will reduce the size of the array.\n"
3330 " use --grow --array-size first to truncate array.\n"
3331 " e.g. mdadm --grow %s --array-size %llu\n",
3332 devname
, reshape
.new_size
/2);
3336 if (array
.level
== 10) {
3337 /* Reshaping RAID10 does not require any data backup by
3338 * user-space. Instead it requires that the data_offset
3339 * is changed to avoid the need for backup.
3340 * So this is handled very separately
3343 /* Nothing to do. */
3345 return raid10_reshape(container
, fd
, devname
, st
, info
,
3346 &reshape
, data_offset
,
3349 sra
= sysfs_read(fd
, NULL
,
3350 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|GET_CHUNK
|
3353 pr_err("%s: Cannot get array details from sysfs\n",
3359 switch(set_new_data_offset(sra
, st
, devname
,
3360 reshape
.after
.data_disks
- reshape
.before
.data_disks
,
3362 reshape
.min_offset_change
, 1)) {
3366 /* Updated data_offset, so it's easy now */
3367 update_cache_size(container
, sra
, info
,
3368 min(reshape
.before
.data_disks
,
3369 reshape
.after
.data_disks
),
3370 reshape
.backup_blocks
);
3372 /* Right, everything seems fine. Let's kick things off.
3376 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3377 devname
, container
, &reshape
) < 0)
3379 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
3381 if (errno
!= EINVAL
) {
3382 pr_err("Failed to initiate reshape!\n");
3385 /* revert data_offset and try the old way */
3386 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3387 sysfs_set_num(sra
, sd
, "new_offset",
3389 sysfs_set_str(sra
, NULL
, "reshape_direction",
3394 if (info
->new_level
== reshape
.level
)
3396 /* need to adjust level when reshape completes */
3398 case -1: /* ignore error, but don't wait */
3400 default: /* parent */
3408 fd
= open_dev(sra
->sys_name
);
3410 impose_level(fd
, info
->new_level
, devname
, verbose
);
3412 case 1: /* Couldn't set data_offset, try the old way */
3413 if (data_offset
!= INVALID_SECTORS
) {
3414 pr_err("Cannot update data_offset on this array\n");
3421 /* Decide how many blocks (sectors) for a reshape
3422 * unit. The number we have so far is just a minimum
3424 blocks
= reshape
.backup_blocks
;
3425 if (reshape
.before
.data_disks
==
3426 reshape
.after
.data_disks
) {
3427 /* Make 'blocks' bigger for better throughput, but
3428 * not so big that we reject it below.
3429 * Try for 16 megabytes
3431 while (blocks
* 32 < sra
->component_size
&&
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",
3443 /* Now we need to open all these devices so we can read/write.
3445 nrdisks
= max(reshape
.before
.data_disks
,
3446 reshape
.after
.data_disks
) + reshape
.parity
3447 + sra
->array
.spare_disks
;
3448 fdlist
= xcalloc((1+nrdisks
), sizeof(int));
3449 offsets
= xcalloc((1+nrdisks
), sizeof(offsets
[0]));
3451 odisks
= reshape
.before
.data_disks
+ reshape
.parity
;
3452 d
= reshape_prepare_fdlist(devname
, sra
, odisks
,
3453 nrdisks
, blocks
, backup_file
,
3458 if ((st
->ss
->manage_reshape
== NULL
) ||
3459 (st
->ss
->recover_backup
== NULL
)) {
3460 if (backup_file
== NULL
) {
3461 if (reshape
.after
.data_disks
<=
3462 reshape
.before
.data_disks
) {
3463 pr_err("%s: Cannot grow - need backup-file\n",
3465 pr_err(" Please provide one with \"--backup=...\"\n");
3467 } else if (d
== odisks
) {
3468 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname
);
3472 if (!reshape_open_backup_file(backup_file
, fd
, devname
,
3474 fdlist
+d
, offsets
+d
,
3483 update_cache_size(container
, sra
, info
,
3484 min(reshape
.before
.data_disks
, reshape
.after
.data_disks
),
3487 /* Right, everything seems fine. Let's kick things off.
3488 * If only changing raid_disks, use ioctl, else use
3493 if (impose_reshape(sra
, info
, st
, fd
, restart
,
3494 devname
, container
, &reshape
) < 0)
3497 err
= start_reshape(sra
, restart
, reshape
.before
.data_disks
,
3498 reshape
.after
.data_disks
);
3500 pr_err("Cannot %s reshape for %s\n",
3501 restart
? "continue" : "start",
3506 sysfs_set_str(sra
, NULL
, "array_state", "active");
3507 if (freeze_reshape
) {
3511 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3512 sra
->reshape_progress
);
3516 if (!forked
&& !check_env("MDADM_NO_SYSTEMCTL"))
3517 if (continue_via_systemd(container
?: sra
->sys_name
)) {
3524 /* Now we just need to kick off the reshape and watch, while
3525 * handling backups of the data...
3526 * This is all done by a forked background process.
3528 switch(forked
? 0 : fork()) {
3530 pr_err("Cannot run child to monitor reshape: %s\n",
3544 /* If another array on the same devices is busy, the
3545 * reshape will wait for them. This would mean that
3546 * the first section that we suspend will stay suspended
3547 * for a long time. So check on that possibility
3548 * by looking for "DELAYED" in /proc/mdstat, and if found,
3552 struct mdstat_ent
*mds
, *m
;
3554 mds
= mdstat_read(1, 0);
3555 for (m
= mds
; m
; m
= m
->next
)
3556 if (strcmp(m
->devnm
, sra
->sys_name
) == 0) {
3558 m
->percent
== RESYNC_DELAYED
)
3561 /* Haven't started the reshape thread
3568 if (delayed
== 1 && get_linux_version() < 3007000) {
3569 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3570 " You might experience problems until other reshapes complete.\n");
3574 mdstat_wait(30 - (delayed
-1) * 25);
3578 if (check_env("MDADM_GROW_VERIFY"))
3579 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
3582 mlockall(MCL_FUTURE
);
3584 signal(SIGTERM
, catch_term
);
3586 if (st
->ss
->external
) {
3587 /* metadata handler takes it from here */
3588 done
= st
->ss
->manage_reshape(
3589 fd
, sra
, &reshape
, st
, blocks
,
3591 d
- odisks
, fdlist
+odisks
,
3594 done
= child_monitor(
3595 fd
, sra
, &reshape
, st
, blocks
,
3597 d
- odisks
, fdlist
+odisks
,
3603 if (backup_file
&& done
) {
3605 bul
= make_backup(sra
->sys_name
);
3608 int l
= readlink(bul
, buf
, sizeof(buf
) - 1);
3616 unlink(backup_file
);
3623 if (!st
->ss
->external
&&
3624 !(reshape
.before
.data_disks
!= reshape
.after
.data_disks
&&
3625 info
->custom_array_size
) && info
->new_level
== reshape
.level
&&
3627 /* no need to wait for the reshape to finish as
3628 * there is nothing more to do.
3635 if (st
->ss
->external
) {
3636 /* Re-load the metadata as much could have changed */
3637 int cfd
= open_dev(st
->container_devnm
);
3639 flush_mdmon(container
);
3640 st
->ss
->free_super(st
);
3641 st
->ss
->load_container(st
, cfd
, container
);
3646 /* set new array size if required customer_array_size is used
3649 if (reshape
.before
.data_disks
!=
3650 reshape
.after
.data_disks
&&
3651 info
->custom_array_size
)
3652 set_array_size(st
, info
, info
->text_version
);
3654 if (info
->new_level
!= reshape
.level
) {
3656 fd
= open(devname
, O_RDONLY
);
3657 impose_level(fd
, info
->new_level
, devname
, verbose
);
3659 if (info
->new_level
== 0)
3660 st
->update_tail
= NULL
;
3672 if (orig_level
!= UnSet
&& sra
) {
3673 c
= map_num(pers
, orig_level
);
3674 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
3675 pr_err("aborting level change\n");
3683 /* mdfd handle is passed to be closed in child process (after fork).
3685 int reshape_container(char *container
, char *devname
,
3687 struct supertype
*st
,
3688 struct mdinfo
*info
,
3690 char *backup_file
, int verbose
,
3691 int forked
, int restart
, int freeze_reshape
)
3693 struct mdinfo
*cc
= NULL
;
3695 char last_devnm
[32] = "";
3697 /* component_size is not meaningful for a container,
3698 * so pass '0' meaning 'no change'
3701 reshape_super(st
, 0, info
->new_level
,
3702 info
->new_layout
, info
->new_chunk
,
3703 info
->array
.raid_disks
, info
->delta_disks
,
3704 backup_file
, devname
, APPLY_METADATA_CHANGES
,
3712 /* ping monitor to be sure that update is on disk
3714 ping_monitor(container
);
3716 if (!forked
&& !freeze_reshape
&& !check_env("MDADM_NO_SYSTEMCTL"))
3717 if (continue_via_systemd(container
))
3720 switch (forked
? 0 : fork()) {
3721 case -1: /* error */
3722 perror("Cannot fork to complete reshape\n");
3725 default: /* parent */
3726 if (!freeze_reshape
)
3727 printf("%s: multi-array reshape continues in background\n", Name
);
3734 /* close unused handle in child process
3740 /* For each member array with reshape_active,
3741 * we need to perform the reshape.
3742 * We pick the first array that needs reshaping and
3743 * reshape it. reshape_array() will re-read the metadata
3744 * so the next time through a different array should be
3745 * ready for reshape.
3746 * It is possible that the 'different' array will not
3747 * be assembled yet. In that case we simple exit.
3748 * When it is assembled, the mdadm which assembles it
3749 * will take over the reshape.
3751 struct mdinfo
*content
;
3753 struct mdstat_ent
*mdstat
;
3759 cc
= st
->ss
->container_content(st
, NULL
);
3761 for (content
= cc
; content
; content
= content
->next
) {
3763 if (!content
->reshape_active
)
3766 subarray
= strchr(content
->text_version
+1, '/')+1;
3767 mdstat
= mdstat_by_subdev(subarray
, container
);
3770 if (mdstat
->active
== 0) {
3771 pr_err("Skipping inactive array %s.\n",
3773 free_mdstat(mdstat
);
3782 devid
= devnm2devid(mdstat
->devnm
);
3783 adev
= map_dev(major(devid
), minor(devid
), 0);
3785 adev
= content
->text_version
;
3787 fd
= open_dev(mdstat
->devnm
);
3789 pr_err("Device %s cannot be opened for reshape.\n", adev
);
3793 if (strcmp(last_devnm
, mdstat
->devnm
) == 0) {
3794 /* Do not allow for multiple reshape_array() calls for
3796 * It can happen when reshape_array() returns without
3797 * error, when reshape is not finished (wrong reshape
3798 * starting/continuation conditions). Mdmon doesn't
3799 * switch to next array in container and reentry
3800 * conditions for the same array occur.
3801 * This is possibly interim until the behaviour of
3802 * reshape_array is resolved().
3804 printf("%s: Multiple reshape execution detected for device %s.\n", Name
, adev
);
3808 strcpy(last_devnm
, mdstat
->devnm
);
3810 if (sysfs_init(content
, fd
, mdstat
->devnm
)) {
3811 pr_err("Unable to initialize sysfs for %s\n",
3817 if (mdmon_running(container
))
3818 flush_mdmon(container
);
3820 rv
= reshape_array(container
, fd
, adev
, st
,
3821 content
, force
, NULL
, INVALID_SECTORS
,
3822 backup_file
, verbose
, 1, restart
,
3826 if (freeze_reshape
) {
3835 if (mdmon_running(container
))
3836 flush_mdmon(container
);
3845 * We run a child process in the background which performs the following
3847 * - wait for resync to reach a certain point
3848 * - suspend io to the following section
3849 * - backup that section
3850 * - allow resync to proceed further
3852 * - discard the backup.
3854 * When are combined in slightly different ways in the three cases.
3856 * - suspend/backup/allow/wait/resume/discard
3858 * - allow/wait/suspend/backup/allow/wait/resume/discard
3860 * - wait/resume/discard/suspend/backup/allow
3862 * suspend/backup/allow always come together
3863 * wait/resume/discard do too.
3864 * For the same-size case we have two backups to improve flow.
3868 int progress_reshape(struct mdinfo
*info
, struct reshape
*reshape
,
3869 unsigned long long backup_point
,
3870 unsigned long long wait_point
,
3871 unsigned long long *suspend_point
,
3872 unsigned long long *reshape_completed
, int *frozen
)
3874 /* This function is called repeatedly by the reshape manager.
3875 * It determines how much progress can safely be made and allows
3877 * - 'info' identifies the array and particularly records in
3878 * ->reshape_progress the metadata's knowledge of progress
3879 * This is a sector offset from the start of the array
3880 * of the next array block to be relocated. This number
3881 * may increase from 0 or decrease from array_size, depending
3882 * on the type of reshape that is happening.
3883 * Note that in contrast, 'sync_completed' is a block count of the
3884 * reshape so far. It gives the distance between the start point
3885 * (head or tail of device) and the next place that data will be
3886 * written. It always increases.
3887 * - 'reshape' is the structure created by analyse_change
3888 * - 'backup_point' shows how much the metadata manager has backed-up
3889 * data. For reshapes with increasing progress, it is the next address
3890 * to be backed up, previous addresses have been backed-up. For
3891 * decreasing progress, it is the earliest address that has been
3892 * backed up - later address are also backed up.
3893 * So addresses between reshape_progress and backup_point are
3894 * backed up providing those are in the 'correct' order.
3895 * - 'wait_point' is an array address. When reshape_completed
3896 * passes this point, progress_reshape should return. It might
3897 * return earlier if it determines that ->reshape_progress needs
3898 * to be updated or further backup is needed.
3899 * - suspend_point is maintained by progress_reshape and the caller
3900 * should not touch it except to initialise to zero.
3901 * It is an array address and it only increases in 2.6.37 and earlier.
3902 * This makes it difficult to handle reducing reshapes with
3903 * external metadata.
3904 * However: it is similar to backup_point in that it records the
3905 * other end of a suspended region from reshape_progress.
3906 * it is moved to extend the region that is safe to backup and/or
3908 * - reshape_completed is read from sysfs and returned. The caller
3909 * should copy this into ->reshape_progress when it has reason to
3910 * believe that the metadata knows this, and any backup outside this
3914 * 1 if more data from backup_point - but only as far as suspend_point,
3915 * should be backed up
3916 * 0 if things are progressing smoothly
3917 * -1 if the reshape is finished because it is all done,
3918 * -2 if the reshape is finished due to an error.
3921 int advancing
= (reshape
->after
.data_disks
3922 >= reshape
->before
.data_disks
);
3923 unsigned long long need_backup
; /* All data between start of array and
3924 * here will at some point need to
3927 unsigned long long read_offset
, write_offset
;
3928 unsigned long long write_range
;
3929 unsigned long long max_progress
, target
, completed
;
3930 unsigned long long array_size
= (info
->component_size
3931 * reshape
->before
.data_disks
);
3935 /* First, we unsuspend any region that is now known to be safe.
3936 * If suspend_point is on the 'wrong' side of reshape_progress, then
3937 * we don't have or need suspension at the moment. This is true for
3938 * native metadata when we don't need to back-up.
3941 if (info
->reshape_progress
<= *suspend_point
)
3942 sysfs_set_num(info
, NULL
, "suspend_lo",
3943 info
->reshape_progress
);
3945 /* Note: this won't work in 2.6.37 and before.
3946 * Something somewhere should make sure we don't need it!
3948 if (info
->reshape_progress
>= *suspend_point
)
3949 sysfs_set_num(info
, NULL
, "suspend_hi",
3950 info
->reshape_progress
);
3953 /* Now work out how far it is safe to progress.
3954 * If the read_offset for ->reshape_progress is less than
3955 * 'blocks' beyond the write_offset, we can only progress as far
3957 * Otherwise we can progress until the write_offset for the new location
3958 * reaches (within 'blocks' of) the read_offset at the current location.
3959 * However that region must be suspended unless we are using native
3961 * If we need to suspend more, we limit it to 128M per device, which is
3962 * rather arbitrary and should be some time-based calculation.
3964 read_offset
= info
->reshape_progress
/ reshape
->before
.data_disks
;
3965 write_offset
= info
->reshape_progress
/ reshape
->after
.data_disks
;
3966 write_range
= info
->new_chunk
/512;
3967 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
3968 need_backup
= array_size
;
3970 need_backup
= reshape
->backup_blocks
;
3972 if (read_offset
< write_offset
+ write_range
)
3973 max_progress
= backup_point
;
3977 reshape
->after
.data_disks
;
3979 if (read_offset
> write_offset
- write_range
)
3980 /* Can only progress as far as has been backed up,
3981 * which must be suspended */
3982 max_progress
= backup_point
;
3983 else if (info
->reshape_progress
<= need_backup
)
3984 max_progress
= backup_point
;
3986 if (info
->array
.major_version
>= 0)
3987 /* Can progress until backup is needed */
3988 max_progress
= need_backup
;
3990 /* Can progress until metadata update is required */
3993 reshape
->after
.data_disks
;
3994 /* but data must be suspended */
3995 if (max_progress
< *suspend_point
)
3996 max_progress
= *suspend_point
;
4001 /* We know it is safe to progress to 'max_progress' providing
4002 * it is suspended or we are using native metadata.
4003 * Consider extending suspend_point 128M per device if it
4004 * is less than 64M per device beyond reshape_progress.
4005 * But always do a multiple of 'blocks'
4006 * FIXME this is too big - it takes to long to complete
4009 target
= 64*1024*2 * min(reshape
->before
.data_disks
,
4010 reshape
->after
.data_disks
);
4011 target
/= reshape
->backup_blocks
;
4014 target
*= reshape
->backup_blocks
;
4016 /* For externally managed metadata we always need to suspend IO to
4017 * the area being reshaped so we regularly push suspend_point forward.
4018 * For native metadata we only need the suspend if we are going to do
4022 if ((need_backup
> info
->reshape_progress
||
4023 info
->array
.major_version
< 0) &&
4024 *suspend_point
< info
->reshape_progress
+ target
) {
4025 if (need_backup
< *suspend_point
+ 2 * target
)
4026 *suspend_point
= need_backup
;
4027 else if (*suspend_point
+ 2 * target
< array_size
)
4028 *suspend_point
+= 2 * target
;
4030 *suspend_point
= array_size
;
4031 sysfs_set_num(info
, NULL
, "suspend_hi", *suspend_point
);
4032 if (max_progress
> *suspend_point
)
4033 max_progress
= *suspend_point
;
4036 if (info
->array
.major_version
>= 0) {
4037 /* Only need to suspend when about to backup */
4038 if (info
->reshape_progress
< need_backup
* 2 &&
4039 *suspend_point
> 0) {
4041 sysfs_set_num(info
, NULL
, "suspend_lo", 0);
4042 sysfs_set_num(info
, NULL
, "suspend_hi", need_backup
);
4045 /* Need to suspend continually */
4046 if (info
->reshape_progress
< *suspend_point
)
4047 *suspend_point
= info
->reshape_progress
;
4048 if (*suspend_point
+ target
< info
->reshape_progress
)
4049 /* No need to move suspend region yet */;
4051 if (*suspend_point
>= 2 * target
)
4052 *suspend_point
-= 2 * target
;
4055 sysfs_set_num(info
, NULL
, "suspend_lo",
4058 if (max_progress
< *suspend_point
)
4059 max_progress
= *suspend_point
;
4063 /* now set sync_max to allow that progress. sync_max, like
4064 * sync_completed is a count of sectors written per device, so
4065 * we find the difference between max_progress and the start point,
4066 * and divide that by after.data_disks to get a sync_max
4068 * At the same time we convert wait_point to a similar number
4069 * for comparing against sync_completed.
4071 /* scale down max_progress to per_disk */
4072 max_progress
/= reshape
->after
.data_disks
;
4073 /* Round to chunk size as some kernels give an erroneously high number */
4074 max_progress
/= info
->new_chunk
/512;
4075 max_progress
*= info
->new_chunk
/512;
4076 /* And round to old chunk size as the kernel wants that */
4077 max_progress
/= info
->array
.chunk_size
/512;
4078 max_progress
*= info
->array
.chunk_size
/512;
4079 /* Limit progress to the whole device */
4080 if (max_progress
> info
->component_size
)
4081 max_progress
= info
->component_size
;
4082 wait_point
/= reshape
->after
.data_disks
;
4084 /* switch from 'device offset' to 'processed block count' */
4085 max_progress
= info
->component_size
- max_progress
;
4086 wait_point
= info
->component_size
- wait_point
;
4090 sysfs_set_num(info
, NULL
, "sync_max", max_progress
);
4092 /* Now wait. If we have already reached the point that we were
4093 * asked to wait to, don't wait at all, else wait for any change.
4094 * We need to select on 'sync_completed' as that is the place that
4095 * notifications happen, but we are really interested in
4096 * 'reshape_position'
4098 fd
= sysfs_get_fd(info
, NULL
, "sync_completed");
4100 goto check_progress
;
4102 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4103 goto check_progress
;
4105 while (completed
< max_progress
&& completed
< wait_point
) {
4106 /* Check that sync_action is still 'reshape' to avoid
4107 * waiting forever on a dead array
4110 if (sysfs_get_str(info
, NULL
, "sync_action",
4112 strncmp(action
, "reshape", 7) != 0)
4114 /* Some kernels reset 'sync_completed' to zero
4115 * before setting 'sync_action' to 'idle'.
4116 * So we need these extra tests.
4118 if (completed
== 0 && advancing
&&
4119 strncmp(action
, "idle", 4) == 0 &&
4120 info
->reshape_progress
> 0)
4122 if (completed
== 0 && !advancing
&&
4123 strncmp(action
, "idle", 4) == 0 &&
4124 info
->reshape_progress
< (info
->component_size
4125 * reshape
->after
.data_disks
))
4127 sysfs_wait(fd
, NULL
);
4128 if (sysfs_fd_get_ll(fd
, &completed
) < 0)
4129 goto check_progress
;
4131 /* Some kernels reset 'sync_completed' to zero,
4132 * we need to have real point we are in md.
4133 * So in that case, read 'reshape_position' from sysfs.
4135 if (completed
== 0) {
4136 unsigned long long reshapep
;
4138 if (sysfs_get_str(info
, NULL
, "sync_action",
4140 strncmp(action
, "idle", 4) == 0 &&
4141 sysfs_get_ll(info
, NULL
,
4142 "reshape_position", &reshapep
) == 0)
4143 *reshape_completed
= reshapep
;
4145 /* some kernels can give an incorrectly high
4146 * 'completed' number, so round down */
4147 completed
/= (info
->new_chunk
/512);
4148 completed
*= (info
->new_chunk
/512);
4149 /* Convert 'completed' back in to a 'progress' number */
4150 completed
*= reshape
->after
.data_disks
;
4152 completed
= (info
->component_size
4153 * reshape
->after
.data_disks
4155 *reshape_completed
= completed
;
4160 /* We return the need_backup flag. Caller will decide
4161 * how much - a multiple of ->backup_blocks up to *suspend_point
4164 return need_backup
> info
->reshape_progress
;
4166 return need_backup
>= info
->reshape_progress
;
4169 /* if we couldn't read a number from sync_completed, then
4170 * either the reshape did complete, or it aborted.
4171 * We can tell which by checking for 'none' in reshape_position.
4172 * If it did abort, then it might immediately restart if it
4173 * it was just a device failure that leaves us degraded but
4176 if (sysfs_get_str(info
, NULL
, "reshape_position", buf
,
4178 strncmp(buf
, "none", 4) != 0) {
4179 /* The abort might only be temporary. Wait up to 10
4180 * seconds for fd to contain a valid number again.
4184 unsigned long long new_sync_max
;
4185 while (fd
>= 0 && rv
< 0 && wait
> 0) {
4186 if (sysfs_wait(fd
, &wait
) != 1)
4188 switch (sysfs_fd_get_ll(fd
, &completed
)) {
4190 /* all good again */
4192 /* If "sync_max" is no longer max_progress
4193 * we need to freeze things
4195 sysfs_get_ll(info
, NULL
, "sync_max", &new_sync_max
);
4196 *frozen
= (new_sync_max
!= max_progress
);
4198 case -2: /* read error - abort */
4205 return rv
; /* abort */
4207 /* Maybe racing with array shutdown - check state */
4210 if (sysfs_get_str(info
, NULL
, "array_state", buf
,
4212 strncmp(buf
, "inactive", 8) == 0 ||
4213 strncmp(buf
, "clear",5) == 0)
4214 return -2; /* abort */
4215 return -1; /* complete */
4219 /* FIXME return status is never checked */
4220 static int grow_backup(struct mdinfo
*sra
,
4221 unsigned long long offset
, /* per device */
4222 unsigned long stripes
, /* per device, in old chunks */
4223 int *sources
, unsigned long long *offsets
,
4224 int disks
, int chunk
, int level
, int layout
,
4225 int dests
, int *destfd
, unsigned long long *destoffsets
,
4226 int part
, int *degraded
,
4229 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4230 * to storage 'destfd' (offset 'destoffsets'), after first
4231 * suspending IO. Then allow resync to continue
4232 * over the suspended section.
4233 * Use part 'part' of the backup-super-block.
4238 unsigned long long ll
;
4240 //printf("offset %llu\n", offset);
4246 /* Check that array hasn't become degraded, else we might backup the wrong data */
4247 if (sysfs_get_ll(sra
, NULL
, "degraded", &ll
) < 0)
4248 return -1; /* FIXME this error is ignored */
4249 new_degraded
= (int)ll
;
4250 if (new_degraded
!= *degraded
) {
4251 /* check each device to ensure it is still working */
4253 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4254 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4256 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
4259 if (sysfs_get_str(sra
, sd
, "state",
4260 sbuf
, sizeof(sbuf
)) < 0 ||
4261 strstr(sbuf
, "faulty") ||
4262 strstr(sbuf
, "in_sync") == NULL
) {
4263 /* this device is dead */
4264 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
4265 if (sd
->disk
.raid_disk
>= 0 &&
4266 sources
[sd
->disk
.raid_disk
] >= 0) {
4267 close(sources
[sd
->disk
.raid_disk
]);
4268 sources
[sd
->disk
.raid_disk
] = -1;
4273 *degraded
= new_degraded
;
4276 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
4277 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4279 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
4280 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
4283 bsb
.magic
[15] = '2';
4284 for (i
= 0; i
< dests
; i
++)
4286 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
4288 lseek64(destfd
[i
], destoffsets
[i
], 0);
4290 rv
= save_stripes(sources
, offsets
,
4291 disks
, chunk
, level
, layout
,
4293 offset
*512*odata
, stripes
* chunk
* odata
,
4298 bsb
.mtime
= __cpu_to_le64(time(0));
4299 for (i
= 0; i
< dests
; i
++) {
4300 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4302 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4303 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4304 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4305 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4308 if ((unsigned long long)lseek64(destfd
[i
],
4309 destoffsets
[i
] - 4096, 0) !=
4310 destoffsets
[i
] - 4096)
4312 if (write(destfd
[i
], &bsb
, 512) != 512)
4314 if (destoffsets
[i
] > 4096) {
4315 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
4316 destoffsets
[i
]+stripes
*chunk
*odata
)
4318 if (write(destfd
[i
], &bsb
, 512) != 512)
4328 /* in 2.6.30, the value reported by sync_completed can be
4329 * less that it should be by one stripe.
4330 * This only happens when reshape hits sync_max and pauses.
4331 * So allow wait_backup to either extent sync_max further
4332 * than strictly necessary, or return before the
4333 * sync has got quite as far as we would really like.
4334 * This is what 'blocks2' is for.
4335 * The various caller give appropriate values so that
4338 /* FIXME return value is often ignored */
4339 static int forget_backup(int dests
, int *destfd
,
4340 unsigned long long *destoffsets
,
4344 * Erase backup 'part' (which is 0 or 1)
4350 bsb
.arraystart2
= __cpu_to_le64(0);
4351 bsb
.length2
= __cpu_to_le64(0);
4353 bsb
.arraystart
= __cpu_to_le64(0);
4354 bsb
.length
= __cpu_to_le64(0);
4356 bsb
.mtime
= __cpu_to_le64(time(0));
4358 for (i
= 0; i
< dests
; i
++) {
4359 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
4360 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
4361 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
4362 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
4363 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
4364 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
4365 destoffsets
[i
]-4096)
4368 write(destfd
[i
], &bsb
, 512) != 512)
4375 static void fail(char *msg
)
4378 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
4379 rv
|= (write(2, "\n", 1) != 1);
4383 static char *abuf
, *bbuf
;
4384 static unsigned long long abuflen
;
4385 static void validate(int afd
, int bfd
, unsigned long long offset
)
4387 /* check that the data in the backup against the array.
4388 * This is only used for regression testing and should not
4389 * be used while the array is active
4393 lseek64(bfd
, offset
- 4096, 0);
4394 if (read(bfd
, &bsb2
, 512) != 512)
4395 fail("cannot read bsb");
4396 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
4397 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
4398 fail("first csum bad");
4399 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
4400 fail("magic is bad");
4401 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
4402 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
4403 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
4404 fail("second csum bad");
4406 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
4407 fail("devstart is wrong");
4410 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
4412 if (abuflen
< len
) {
4416 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
4417 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
4419 /* just stop validating on mem-alloc failure */
4424 lseek64(bfd
, offset
, 0);
4425 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
4426 //printf("len %llu\n", len);
4427 fail("read first backup failed");
4429 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
4430 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4431 fail("read first from array failed");
4432 if (memcmp(bbuf
, abuf
, len
) != 0) {
4435 printf("offset=%llu len=%llu\n",
4436 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
4437 for (i
=0; i
<len
; i
++)
4438 if (bbuf
[i
] != abuf
[i
]) {
4439 printf("first diff byte %d\n", i
);
4443 fail("data1 compare failed");
4447 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
4449 if (abuflen
< len
) {
4453 abuf
= xmalloc(abuflen
);
4454 bbuf
= xmalloc(abuflen
);
4457 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
4458 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
4459 fail("read second backup failed");
4460 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
4461 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
4462 fail("read second from array failed");
4463 if (memcmp(bbuf
, abuf
, len
) != 0)
4464 fail("data2 compare failed");
4468 int child_monitor(int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
4469 struct supertype
*st
, unsigned long blocks
,
4470 int *fds
, unsigned long long *offsets
,
4471 int dests
, int *destfd
, unsigned long long *destoffsets
)
4473 /* Monitor a reshape where backup is being performed using
4474 * 'native' mechanism - either to a backup file, or
4475 * to some space in a spare.
4479 unsigned long long speed
;
4480 unsigned long long suspend_point
, array_size
;
4481 unsigned long long backup_point
, wait_point
;
4482 unsigned long long reshape_completed
;
4484 int increasing
= reshape
->after
.data_disks
>= reshape
->before
.data_disks
;
4485 int part
= 0; /* The next part of the backup area to fill. It may already
4486 * be full, so we need to check */
4487 int level
= reshape
->level
;
4488 int layout
= reshape
->before
.layout
;
4489 int data
= reshape
->before
.data_disks
;
4490 int disks
= reshape
->before
.data_disks
+ reshape
->parity
;
4491 int chunk
= sra
->array
.chunk_size
;
4493 unsigned long stripes
;
4497 /* set up the backup-super-block. This requires the
4498 * uuid from the array.
4500 /* Find a superblock */
4501 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
4505 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
4507 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
4508 devfd
= dev_open(dn
, O_RDONLY
);
4511 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
4517 pr_err("Cannot find a superblock\n");
4521 memset(&bsb
, 0, 512);
4522 memcpy(bsb
.magic
, "md_backup_data-1", 16);
4523 st
->ss
->uuid_from_super(st
, uuid
);
4524 memcpy(bsb
.set_uuid
, uuid
, 16);
4525 bsb
.mtime
= __cpu_to_le64(time(0));
4526 bsb
.devstart2
= blocks
;
4528 stripes
= blocks
/ (sra
->array
.chunk_size
/512) /
4529 reshape
->before
.data_disks
;
4531 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
4532 /* Don't start the 'reshape' */
4534 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
) {
4535 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
4536 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
4540 array_size
= sra
->component_size
* reshape
->after
.data_disks
;
4541 backup_point
= sra
->reshape_progress
;
4544 array_size
= sra
->component_size
* reshape
->before
.data_disks
;
4545 backup_point
= reshape
->backup_blocks
;
4546 suspend_point
= array_size
;
4552 /* Want to return as soon the oldest backup slot can
4553 * be released as that allows us to start backing up
4554 * some more, providing suspend_point has been
4555 * advanced, which it should have.
4558 wait_point
= array_size
;
4559 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4560 wait_point
= (__le64_to_cpu(bsb
.arraystart
) +
4561 __le64_to_cpu(bsb
.length
));
4562 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4563 wait_point
= (__le64_to_cpu(bsb
.arraystart2
) +
4564 __le64_to_cpu(bsb
.length2
));
4567 if (part
== 0 && __le64_to_cpu(bsb
.length
) > 0)
4568 wait_point
= __le64_to_cpu(bsb
.arraystart
);
4569 if (part
== 1 && __le64_to_cpu(bsb
.length2
) > 0)
4570 wait_point
= __le64_to_cpu(bsb
.arraystart2
);
4573 reshape_completed
= sra
->reshape_progress
;
4574 rv
= progress_reshape(sra
, reshape
,
4575 backup_point
, wait_point
,
4576 &suspend_point
, &reshape_completed
,
4578 /* external metadata would need to ping_monitor here */
4579 sra
->reshape_progress
= reshape_completed
;
4581 /* Clear any backup region that is before 'here' */
4583 if (__le64_to_cpu(bsb
.length
) > 0 &&
4584 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart
) +
4585 __le64_to_cpu(bsb
.length
)))
4586 forget_backup(dests
, destfd
,
4588 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4589 reshape_completed
>= (__le64_to_cpu(bsb
.arraystart2
) +
4590 __le64_to_cpu(bsb
.length2
)))
4591 forget_backup(dests
, destfd
,
4594 if (__le64_to_cpu(bsb
.length
) > 0 &&
4595 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart
)))
4596 forget_backup(dests
, destfd
,
4598 if (__le64_to_cpu(bsb
.length2
) > 0 &&
4599 reshape_completed
<= (__le64_to_cpu(bsb
.arraystart2
)))
4600 forget_backup(dests
, destfd
,
4610 if (rv
== 0 && increasing
&& !st
->ss
->external
) {
4611 /* No longer need to monitor this reshape */
4612 sysfs_set_str(sra
, NULL
, "sync_max", "max");
4618 unsigned long long offset
;
4619 unsigned long actual_stripes
;
4620 /* Need to backup some data.
4621 * If 'part' is not used and the desired
4622 * backup size is suspended, do a backup,
4623 * then consider the next part.
4625 /* Check that 'part' is unused */
4626 if (part
== 0 && __le64_to_cpu(bsb
.length
) != 0)
4628 if (part
== 1 && __le64_to_cpu(bsb
.length2
) != 0)
4631 offset
= backup_point
/ data
;
4632 actual_stripes
= stripes
;
4634 if (offset
+ actual_stripes
* (chunk
/512) >
4635 sra
->component_size
)
4636 actual_stripes
= ((sra
->component_size
- offset
)
4638 if (offset
+ actual_stripes
* (chunk
/512) >
4642 if (offset
< actual_stripes
* (chunk
/512))
4643 actual_stripes
= offset
/ (chunk
/512);
4644 offset
-= actual_stripes
* (chunk
/512);
4645 if (offset
< suspend_point
/data
)
4648 if (actual_stripes
== 0)
4650 grow_backup(sra
, offset
, actual_stripes
,
4652 disks
, chunk
, level
, layout
,
4653 dests
, destfd
, destoffsets
,
4654 part
, °raded
, buf
);
4655 validate(afd
, destfd
[0], destoffsets
[0]);
4656 /* record where 'part' is up to */
4659 backup_point
+= actual_stripes
* (chunk
/512) * data
;
4661 backup_point
-= actual_stripes
* (chunk
/512) * data
;
4665 /* FIXME maybe call progress_reshape one more time instead */
4666 /* remove any remaining suspension */
4667 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
4668 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
4669 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
4670 sysfs_set_num(sra
, NULL
, "sync_min", 0);
4672 if (reshape
->before
.data_disks
== reshape
->after
.data_disks
)
4673 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
4679 * If any spare contains md_back_data-1 which is recent wrt mtime,
4680 * write that data into the array and update the super blocks with
4681 * the new reshape_progress
4683 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
4684 char *backup_file
, int verbose
)
4688 unsigned long long *offsets
;
4689 unsigned long long nstripe
, ostripe
;
4692 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
4693 if (info
->array
.level
== 6) odata
--; /* number of data disks */
4694 ndata
= info
->array
.raid_disks
- 1;
4695 if (info
->new_level
== 6) ndata
--;
4697 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
4699 if (info
->delta_disks
<= 0)
4700 /* Didn't grow, so the backup file must have
4704 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
4705 struct mdinfo dinfo
;
4708 char *devname
, namebuf
[20];
4709 unsigned long long lo
, hi
;
4711 /* This was a spare and may have some saved data on it.
4712 * Load the superblock, find and load the
4713 * backup_super_block.
4714 * If either fail, go on to next device.
4715 * If the backup contains no new info, just return
4716 * else restore data and update all superblocks
4718 if (i
== old_disks
-1) {
4719 fd
= open(backup_file
, O_RDONLY
);
4721 pr_err("backup file %s inaccessible: %s\n",
4722 backup_file
, strerror(errno
));
4725 devname
= backup_file
;
4730 if (st
->ss
->load_super(st
, fd
, NULL
))
4733 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4734 st
->ss
->free_super(st
);
4737 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
4739 pr_err("Cannot seek on device %d\n", i
);
4740 continue; /* Cannot seek */
4742 sprintf(namebuf
, "device-%d", i
);
4745 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
4747 pr_err("Cannot read from %s\n", devname
);
4748 continue; /* Cannot read */
4750 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
4751 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
4753 pr_err("No backup metadata on %s\n", devname
);
4756 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
4758 pr_err("Bad backup-metadata checksum on %s\n", devname
);
4759 continue; /* bad checksum */
4761 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
4762 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
4764 pr_err("Bad backup-metadata checksum2 on %s\n", devname
);
4765 continue; /* Bad second checksum */
4767 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
4769 pr_err("Wrong uuid on backup-metadata on %s\n", devname
);
4770 continue; /* Wrong uuid */
4773 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4774 * sometimes they aren't... So allow considerable flexability in matching, and allow
4775 * this test to be overridden by an environment variable.
4777 if(time_after(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) + 2*60*60) ||
4778 time_before(info
->array
.utime
, (unsigned int)__le64_to_cpu(bsb
.mtime
) - 10*60)) {
4779 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4780 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4781 (unsigned long)__le64_to_cpu(bsb
.mtime
),
4782 (unsigned long)info
->array
.utime
);
4784 pr_err("too-old timestamp on backup-metadata on %s\n", devname
);
4785 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4786 continue; /* time stamp is too bad */
4790 if (bsb
.magic
[15] == '1') {
4791 if (bsb
.length
== 0)
4793 if (info
->delta_disks
>= 0) {
4794 /* reshape_progress is increasing */
4795 if (__le64_to_cpu(bsb
.arraystart
)
4796 + __le64_to_cpu(bsb
.length
)
4797 < info
->reshape_progress
) {
4800 pr_err("backup-metadata found on %s but is not needed\n", devname
);
4801 continue; /* No new data here */
4804 /* reshape_progress is decreasing */
4805 if (__le64_to_cpu(bsb
.arraystart
) >=
4806 info
->reshape_progress
)
4807 goto nonew
; /* No new data here */
4810 if (bsb
.length
== 0 && bsb
.length2
== 0)
4812 if (info
->delta_disks
>= 0) {
4813 /* reshape_progress is increasing */
4814 if ((__le64_to_cpu(bsb
.arraystart
)
4815 + __le64_to_cpu(bsb
.length
)
4816 < info
->reshape_progress
) &&
4817 (__le64_to_cpu(bsb
.arraystart2
)
4818 + __le64_to_cpu(bsb
.length2
)
4819 < info
->reshape_progress
))
4820 goto nonew
; /* No new data here */
4822 /* reshape_progress is decreasing */
4823 if (__le64_to_cpu(bsb
.arraystart
) >=
4824 info
->reshape_progress
&&
4825 __le64_to_cpu(bsb
.arraystart2
) >=
4826 info
->reshape_progress
)
4827 goto nonew
; /* No new data here */
4830 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
4833 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4835 continue; /* Cannot seek */
4837 /* There should be a duplicate backup superblock 4k before here */
4838 if (lseek64(fd
, -4096, 1) < 0 ||
4839 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
4840 goto second_fail
; /* Cannot find leading superblock */
4841 if (bsb
.magic
[15] == '1')
4842 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
4844 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
4845 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
4846 goto second_fail
; /* Cannot find leading superblock */
4848 /* Now need the data offsets for all devices. */
4849 offsets
= xmalloc(sizeof(*offsets
)*info
->array
.raid_disks
);
4850 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
4853 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4854 /* FIXME should be this be an error */
4856 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4857 st
->ss
->free_super(st
);
4858 offsets
[j
] = dinfo
.data_offset
* 512;
4860 printf("%s: restoring critical section\n", Name
);
4862 if (restore_stripes(fdlist
, offsets
,
4863 info
->array
.raid_disks
,
4867 fd
, __le64_to_cpu(bsb
.devstart
)*512,
4868 __le64_to_cpu(bsb
.arraystart
)*512,
4869 __le64_to_cpu(bsb
.length
)*512, NULL
)) {
4870 /* didn't succeed, so giveup */
4872 pr_err("Error restoring backup from %s\n",
4878 if (bsb
.magic
[15] == '2' &&
4879 restore_stripes(fdlist
, offsets
,
4880 info
->array
.raid_disks
,
4884 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
4885 __le64_to_cpu(bsb
.devstart2
)*512,
4886 __le64_to_cpu(bsb
.arraystart2
)*512,
4887 __le64_to_cpu(bsb
.length2
)*512, NULL
)) {
4888 /* didn't succeed, so giveup */
4890 pr_err("Error restoring second backup from %s\n",
4898 /* Ok, so the data is restored. Let's update those superblocks. */
4902 lo
= __le64_to_cpu(bsb
.arraystart
);
4903 hi
= lo
+ __le64_to_cpu(bsb
.length
);
4905 if (bsb
.magic
[15] == '2' && bsb
.length2
) {
4906 unsigned long long lo1
, hi1
;
4907 lo1
= __le64_to_cpu(bsb
.arraystart2
);
4908 hi1
= lo1
+ __le64_to_cpu(bsb
.length2
);
4912 } else if (lo
< lo1
)
4918 (info
->reshape_progress
< lo
||
4919 info
->reshape_progress
> hi
))
4920 /* backup does not affect reshape_progress*/ ;
4921 else if (info
->delta_disks
>= 0) {
4922 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
4923 __le64_to_cpu(bsb
.length
);
4924 if (bsb
.magic
[15] == '2') {
4925 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
4926 __le64_to_cpu(bsb
.length2
);
4927 if (p2
> info
->reshape_progress
)
4928 info
->reshape_progress
= p2
;
4931 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
4932 if (bsb
.magic
[15] == '2') {
4933 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
4934 if (p2
< info
->reshape_progress
)
4935 info
->reshape_progress
= p2
;
4938 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
4941 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
4943 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
4944 dinfo
.reshape_progress
= info
->reshape_progress
;
4945 st
->ss
->update_super(st
, &dinfo
,
4946 "_reshape_progress",
4948 st
->ss
->store_super(st
, fdlist
[j
]);
4949 st
->ss
->free_super(st
);
4953 /* Didn't find any backup data, try to see if any
4956 if (info
->delta_disks
< 0) {
4957 /* When shrinking, the critical section is at the end.
4958 * So see if we are before the critical section.
4960 unsigned long long first_block
;
4961 nstripe
= ostripe
= 0;
4963 while (ostripe
>= nstripe
) {
4964 ostripe
+= info
->array
.chunk_size
/ 512;
4965 first_block
= ostripe
* odata
;
4966 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
4967 (info
->new_chunk
/512);
4970 if (info
->reshape_progress
>= first_block
)
4973 if (info
->delta_disks
> 0) {
4974 /* See if we are beyond the critical section. */
4975 unsigned long long last_block
;
4976 nstripe
= ostripe
= 0;
4978 while (nstripe
>= ostripe
) {
4979 nstripe
+= info
->new_chunk
/ 512;
4980 last_block
= nstripe
* ndata
;
4981 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
4982 (info
->array
.chunk_size
/512);
4985 if (info
->reshape_progress
>= last_block
)
4988 /* needed to recover critical section! */
4990 pr_err("Failed to find backup of critical section\n");
4994 int Grow_continue_command(char *devname
, int fd
,
4995 char *backup_file
, int verbose
)
4998 struct supertype
*st
= NULL
;
4999 struct mdinfo
*content
= NULL
;
5000 struct mdinfo array
;
5001 char *subarray
= NULL
;
5002 struct mdinfo
*cc
= NULL
;
5003 struct mdstat_ent
*mdstat
= NULL
;
5007 dprintf("Grow continue from command line called for %s\n",
5010 st
= super_by_fd(fd
, &subarray
);
5011 if (!st
|| !st
->ss
) {
5012 pr_err("Unable to determine metadata format for %s\n",
5016 dprintf("Grow continue is run for ");
5017 if (st
->ss
->external
== 0) {
5020 dprintf_cont("native array (%s)\n", devname
);
5021 if (md_get_array_info(fd
, &array
.array
) < 0) {
5022 pr_err("%s is not an active md array - aborting\n",
5025 goto Grow_continue_command_exit
;
5028 sysfs_init(content
, fd
, NULL
);
5029 /* Need to load a superblock.
5030 * FIXME we should really get what we need from
5034 for (d
= 0; d
< MAX_DISKS
; d
++) {
5035 mdu_disk_info_t disk
;
5039 if (md_get_disk_info(fd
, &disk
) < 0)
5041 if (disk
.major
== 0 && disk
.minor
== 0)
5043 if ((disk
.state
& (1 << MD_DISK_ACTIVE
)) == 0)
5045 dv
= map_dev(disk
.major
, disk
.minor
, 1);
5048 fd2
= dev_open(dv
, O_RDONLY
);
5051 err
= st
->ss
->load_super(st
, fd2
, NULL
);
5057 if (d
== MAX_DISKS
) {
5058 pr_err("Unable to load metadata for %s\n",
5061 goto Grow_continue_command_exit
;
5063 st
->ss
->getinfo_super(st
, content
, NULL
);
5064 if (!content
->reshape_active
)
5068 } while (cnt
-- > 0);
5073 dprintf_cont("subarray (%s)\n", subarray
);
5074 container
= st
->container_devnm
;
5075 cfd
= open_dev_excl(st
->container_devnm
);
5077 container
= st
->devnm
;
5079 cfd
= open_dev_excl(st
->devnm
);
5080 dprintf_cont("container (%s)\n", container
);
5084 pr_err("Unable to open container for %s\n", devname
);
5086 goto Grow_continue_command_exit
;
5089 /* find in container array under reshape
5091 ret_val
= st
->ss
->load_container(st
, cfd
, NULL
);
5093 pr_err("Cannot read superblock for %s\n",
5096 goto Grow_continue_command_exit
;
5099 cc
= st
->ss
->container_content(st
, subarray
);
5100 for (content
= cc
; content
; content
= content
->next
) {
5102 int allow_reshape
= 1;
5104 if (content
->reshape_active
== 0)
5106 /* The decision about array or container wide
5107 * reshape is taken in Grow_continue based
5108 * content->reshape_active state, therefore we
5109 * need to check_reshape based on
5110 * reshape_active and subarray name
5112 if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
))
5114 if (content
->reshape_active
== CONTAINER_RESHAPE
&&
5115 (content
->array
.state
5116 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
)))
5119 if (!allow_reshape
) {
5120 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5121 devname
, container
);
5123 goto Grow_continue_command_exit
;
5126 array_name
= strchr(content
->text_version
+1, '/')+1;
5127 mdstat
= mdstat_by_subdev(array_name
, container
);
5130 if (mdstat
->active
== 0) {
5131 pr_err("Skipping inactive array %s.\n",
5133 free_mdstat(mdstat
);
5140 pr_err("Unable to determine reshaped array for %s\n", devname
);
5142 goto Grow_continue_command_exit
;
5144 fd2
= open_dev(mdstat
->devnm
);
5146 pr_err("cannot open (%s)\n", mdstat
->devnm
);
5148 goto Grow_continue_command_exit
;
5151 if (sysfs_init(content
, fd2
, mdstat
->devnm
)) {
5152 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5156 goto Grow_continue_command_exit
;
5161 /* start mdmon in case it is not running
5163 if (!mdmon_running(container
))
5164 start_mdmon(container
);
5165 ping_monitor(container
);
5167 if (mdmon_running(container
))
5168 st
->update_tail
= &st
->updates
;
5170 pr_err("No mdmon found. Grow cannot continue.\n");
5172 goto Grow_continue_command_exit
;
5176 /* verify that array under reshape is started from
5179 if (verify_reshape_position(content
, content
->array
.level
) < 0) {
5181 goto Grow_continue_command_exit
;
5186 ret_val
= Grow_continue(fd
, st
, content
, backup_file
, 1, 0);
5188 Grow_continue_command_exit
:
5191 st
->ss
->free_super(st
);
5192 free_mdstat(mdstat
);
5199 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
5200 char *backup_file
, int forked
, int freeze_reshape
)
5204 if (!info
->reshape_active
)
5207 if (st
->ss
->external
) {
5208 int cfd
= open_dev(st
->container_devnm
);
5213 st
->ss
->load_container(st
, cfd
, st
->container_devnm
);
5215 ret_val
= reshape_container(st
->container_devnm
, NULL
, mdfd
,
5216 st
, info
, 0, backup_file
,
5218 1 | info
->reshape_active
,
5221 ret_val
= reshape_array(NULL
, mdfd
, "array", st
, info
, 1,
5222 NULL
, INVALID_SECTORS
,
5223 backup_file
, 0, forked
,
5224 1 | info
->reshape_active
,
5230 char *make_backup(char *name
)
5232 char *base
= "backup_file-";
5236 len
= strlen(MAP_DIR
) + 1 + strlen(base
) + strlen(name
)+1;
5237 fname
= xmalloc(len
);
5238 sprintf(fname
, "%s/%s%s", MAP_DIR
, base
, name
);
5242 char *locate_backup(char *name
)
5244 char *fl
= make_backup(name
);
5247 if (stat(fl
, &stb
) == 0 &&
5248 S_ISREG(stb
.st_mode
))