2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 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>
28 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
29 #error no endian defined
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
38 int Grow_Add_device(char *devname
, int fd
, char *newdev
)
40 /* Add a device to an active array.
41 * Currently, just extend a linear array.
42 * This requires writing a new superblock on the
43 * new device, calling the kernel to add the device,
44 * and if that succeeds, update the superblock on
46 * This means that we need to *find* all other devices.
53 struct supertype
*st
= NULL
;
54 char *subarray
= NULL
;
56 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
57 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
61 if (info
.array
.level
!= -1) {
62 fprintf(stderr
, Name
": can only add devices to linear arrays\n");
66 st
= super_by_fd(fd
, &subarray
);
68 fprintf(stderr
, Name
": cannot handle arrays with superblock version %d\n", info
.array
.major_version
);
73 fprintf(stderr
, Name
": Cannot grow linear sub-arrays yet\n");
78 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
80 fprintf(stderr
, Name
": cannot open %s\n", newdev
);
85 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
86 fprintf(stderr
, Name
": %s is not a block device!\n", newdev
);
91 /* now check out all the devices and make sure we can read the superblock */
92 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
96 st
->ss
->free_super(st
);
99 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
100 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
106 dv
= map_dev(disk
.major
, disk
.minor
, 1);
108 fprintf(stderr
, Name
": cannot find device file for device %d\n",
114 fd2
= dev_open(dv
, O_RDWR
);
116 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
122 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
123 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
131 /* Ok, looks good. Lets update the superblock and write it out to
135 info
.disk
.number
= d
;
136 info
.disk
.major
= major(stb
.st_rdev
);
137 info
.disk
.minor
= minor(stb
.st_rdev
);
138 info
.disk
.raid_disk
= d
;
139 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
140 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
143 if (st
->ss
->store_super(st
, nfd
)) {
144 fprintf(stderr
, Name
": Cannot store new superblock on %s\n",
151 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
152 fprintf(stderr
, Name
": Cannot add new disk to this array\n");
155 /* Well, that seems to have worked.
156 * Now go through and update all superblocks
159 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
160 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
165 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
166 mdu_disk_info_t disk
;
170 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
171 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
175 dv
= map_dev(disk
.major
, disk
.minor
, 1);
177 fprintf(stderr
, Name
": cannot find device file for device %d\n",
181 fd2
= dev_open(dv
, O_RDWR
);
183 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
186 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
187 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
191 info
.array
.raid_disks
= nd
+1;
192 info
.array
.nr_disks
= nd
+1;
193 info
.array
.active_disks
= nd
+1;
194 info
.array
.working_disks
= nd
+1;
196 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
199 if (st
->ss
->store_super(st
, fd2
)) {
200 fprintf(stderr
, Name
": Cannot store new superblock on %s\n", dv
);
210 int Grow_addbitmap(char *devname
, int fd
, char *file
, int chunk
, int delay
, int write_behind
, int force
)
213 * First check that array doesn't have a bitmap
214 * Then create the bitmap
217 * For internal bitmaps, we need to check the version,
218 * find all the active devices, and write the bitmap block
221 mdu_bitmap_file_t bmf
;
222 mdu_array_info_t array
;
223 struct supertype
*st
;
224 char *subarray
= NULL
;
225 int major
= BITMAP_MAJOR_HI
;
226 int vers
= md_get_version(fd
);
227 unsigned long long bitmapsize
, array_size
;
230 major
= BITMAP_MAJOR_HOSTENDIAN
;
232 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
233 " between different architectured. Consider upgrading the Linux kernel.\n");
237 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
239 fprintf(stderr
, Name
": Memory allocation failure.\n");
241 fprintf(stderr
, Name
": bitmaps not supported by this kernel.\n");
244 if (bmf
.pathname
[0]) {
245 if (strcmp(file
,"none")==0) {
246 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
247 fprintf(stderr
, Name
": failed to remove bitmap %s\n",
253 fprintf(stderr
, Name
": %s already has a bitmap (%s)\n",
254 devname
, bmf
.pathname
);
257 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
258 fprintf(stderr
, Name
": cannot get array status for %s\n", devname
);
261 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
262 if (strcmp(file
, "none")==0) {
263 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
264 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
265 fprintf(stderr
, Name
": failed to remove internal bitmap.\n");
270 fprintf(stderr
, Name
": Internal bitmap already present on %s\n",
275 if (strcmp(file
, "none") == 0) {
276 fprintf(stderr
, Name
": no bitmap found on %s\n", devname
);
279 if (array
.level
<= 0) {
280 fprintf(stderr
, Name
": Bitmaps not meaningful with level %s\n",
281 map_num(pers
, array
.level
)?:"of this array");
284 bitmapsize
= array
.size
;
286 if (get_dev_size(fd
, NULL
, &array_size
) &&
287 array_size
> (0x7fffffffULL
<<9)) {
288 /* Array is big enough that we cannot trust array.size
289 * try other approaches
291 bitmapsize
= get_component_size(fd
);
293 if (bitmapsize
== 0) {
294 fprintf(stderr
, Name
": Cannot reliably determine size of array to create bitmap - sorry.\n");
298 if (array
.level
== 10) {
299 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
300 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
303 st
= super_by_fd(fd
, &subarray
);
305 fprintf(stderr
, Name
": Cannot understand version %d.%d\n",
306 array
.major_version
, array
.minor_version
);
310 fprintf(stderr
, Name
": Cannot add bitmaps to sub-arrays yet\n");
315 if (strcmp(file
, "internal") == 0) {
317 if (st
->ss
->add_internal_bitmap
== NULL
) {
318 fprintf(stderr
, Name
": Internal bitmaps not supported "
319 "with %s metadata\n", st
->ss
->name
);
322 for (d
=0; d
< st
->max_devs
; d
++) {
323 mdu_disk_info_t disk
;
326 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
328 if (disk
.major
== 0 &&
331 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
333 dv
= map_dev(disk
.major
, disk
.minor
, 1);
335 int fd2
= dev_open(dv
, O_RDWR
);
338 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
339 if (st
->ss
->add_internal_bitmap(
341 &chunk
, delay
, write_behind
,
342 bitmapsize
, 0, major
)
344 st
->ss
->write_bitmap(st
, fd2
);
346 fprintf(stderr
, Name
": failed to create internal bitmap - chunksize problem.\n");
354 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
355 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
356 fprintf(stderr
, Name
": failed to set internal bitmap.\n");
363 int max_devs
= st
->max_devs
;
365 /* try to load a superblock */
366 for (d
=0; d
<max_devs
; d
++) {
367 mdu_disk_info_t disk
;
371 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
373 if ((disk
.major
==0 && disk
.minor
==0) ||
374 (disk
.state
& (1<<MD_DISK_REMOVED
)))
376 dv
= map_dev(disk
.major
, disk
.minor
, 1);
378 fd2
= dev_open(dv
, O_RDONLY
);
380 st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
382 st
->ss
->uuid_from_super(st
, uuid
);
388 fprintf(stderr
, Name
": cannot find UUID for array!\n");
391 if (CreateBitmap(file
, force
, (char*)uuid
, chunk
,
392 delay
, write_behind
, bitmapsize
, major
)) {
395 bitmap_fd
= open(file
, O_RDWR
);
397 fprintf(stderr
, Name
": weird: %s cannot be opened\n",
401 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
402 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
403 devname
, strerror(errno
));
413 * When reshaping an array we might need to backup some data.
414 * This is written to all spares with a 'super_block' describing it.
415 * The superblock goes 4K from the end of the used space on the
417 * It if written after the backup is complete.
418 * It has the following structure.
421 static struct mdp_backup_super
{
422 char magic
[16]; /* md_backup_data-1 or -2 */
425 /* start/sizes in 512byte sectors */
426 __u64 devstart
; /* address on backup device/file of data */
429 __u32 sb_csum
; /* csum of preceeding bytes. */
431 __u64 devstart2
; /* offset in to data of second section */
434 __u32 sb_csum2
; /* csum of preceeding bytes. */
436 } __attribute__((aligned(512))) bsb
, bsb2
;
438 static __u32
bsb_csum(char *buf
, int len
)
442 for (i
=0; i
<len
; i
++)
443 csum
= (csum
<<3) + buf
[0];
444 return __cpu_to_le32(csum
);
447 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
448 int *fds
, unsigned long long *offsets
,
449 int disks
, int chunk
, int level
, int layout
, int data
,
450 int dests
, int *destfd
, unsigned long long *destoffsets
);
451 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
452 int *fds
, unsigned long long *offsets
,
453 int disks
, int chunk
, int level
, int layout
, int data
,
454 int dests
, int *destfd
, unsigned long long *destoffsets
);
455 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
456 int *fds
, unsigned long long *offsets
,
457 unsigned long long start
,
458 int disks
, int chunk
, int level
, int layout
, int data
,
459 int dests
, int *destfd
, unsigned long long *destoffsets
);
461 static int freeze_container(struct supertype
*st
)
463 int container_dev
= (st
->container_dev
!= NoMdDev
464 ? st
->container_dev
: st
->devnum
);
465 char *container
= devnum2devname(container_dev
);
469 ": could not determine container name, freeze aborted\n");
473 if (block_monitor(container
, 1)) {
474 fprintf(stderr
, Name
": failed to freeze container\n");
481 static void unfreeze_container(struct supertype
*st
)
483 int container_dev
= (st
->container_dev
!= NoMdDev
484 ? st
->container_dev
: st
->devnum
);
485 char *container
= devnum2devname(container_dev
);
489 ": could not determine container name, unfreeze aborted\n");
493 unblock_monitor(container
, 1);
496 static int freeze(struct supertype
*st
)
498 /* Try to freeze resync/rebuild on this array/container.
499 * Return -1 if the array is busy,
500 * return -2 container cannot be frozen,
501 * return 0 if this kernel doesn't support 'frozen'
502 * return 1 if it worked.
504 if (st
->ss
->external
)
505 return freeze_container(st
);
507 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
512 err
= sysfs_freeze_array(sra
);
518 static void unfreeze(struct supertype
*st
, int frozen
)
520 /* If 'frozen' is 1, unfreeze the array */
524 if (st
->ss
->external
)
525 return unfreeze_container(st
);
527 struct mdinfo
*sra
= sysfs_read(-1, st
->devnum
, GET_VERSION
);
530 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
532 fprintf(stderr
, Name
": failed to unfreeze array\n");
537 static void wait_reshape(struct mdinfo
*sra
)
539 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
546 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
548 if (sysfs_fd_get_str(fd
, action
, 20) < 0) {
552 } while (strncmp(action
, "reshape", 7) == 0);
555 static int reshape_super(struct supertype
*st
, long long size
, int level
,
556 int layout
, int chunksize
, int raid_disks
,
557 char *backup_file
, char *dev
, int verbose
)
559 /* nothing extra to check in the native case */
560 if (!st
->ss
->external
)
562 if (!st
->ss
->reshape_super
||
563 !st
->ss
->manage_reshape
) {
564 fprintf(stderr
, Name
": %s metadata does not support reshape\n",
569 return st
->ss
->reshape_super(st
, size
, level
, layout
, chunksize
,
570 raid_disks
, backup_file
, dev
, verbose
);
573 static void sync_metadata(struct supertype
*st
)
575 if (st
->ss
->external
) {
577 flush_metadata_updates(st
);
579 st
->ss
->sync_metadata(st
);
583 static int subarray_set_num(char *container
, struct mdinfo
*sra
, char *name
, int n
)
585 /* when dealing with external metadata subarrays we need to be
586 * prepared to handle EAGAIN. The kernel may need to wait for
587 * mdmon to mark the array active so the kernel can handle
588 * allocations/writeback when preparing the reshape action
589 * (md_allow_write()). We temporarily disable safe_mode_delay
590 * to close a race with the array_state going clean before the
591 * next write to raid_disks / stripe_cache_size
596 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
597 if (strcmp(name
, "raid_disks") != 0 &&
598 strcmp(name
, "stripe_cache_size") != 0)
599 return sysfs_set_num(sra
, NULL
, name
, n
);
601 rc
= sysfs_get_str(sra
, NULL
, "safe_mode_delay", safe
, sizeof(safe
));
604 sysfs_set_num(sra
, NULL
, "safe_mode_delay", 0);
605 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
606 if (rc
< 0 && errno
== EAGAIN
) {
607 ping_monitor(container
);
608 /* if we get EAGAIN here then the monitor is not active
611 rc
= sysfs_set_num(sra
, NULL
, name
, n
);
613 sysfs_set_str(sra
, NULL
, "safe_mode_delay", safe
);
617 static int reshape_container_raid_disks(char *container
, int raid_disks
)
619 /* for each subarray switch to a raid level that can
620 * support the reshape, and set raid disks
622 struct mdstat_ent
*ent
, *e
;
623 int changed
= 0, rv
= 0, err
= 0;
625 ent
= mdstat_read(1, 0);
627 fprintf(stderr
, Name
": unable to read /proc/mdstat\n");
632 for (e
= ent
; e
; e
= e
->next
) {
635 int level
, takeover_delta
= 0;
637 if (!is_container_member(e
, container
))
640 level
= map_name(pers
, e
->level
);
642 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
645 /* metadata records 'orig_level' */
646 rv
= sysfs_set_num(sub
, NULL
, "level", 4);
651 /* we want spares to be used for capacity
652 * expansion, not rebuild
668 sub
= sysfs_read(-1, e
->devnum
, GET_CHUNK
|GET_CACHE
);
671 cache
= (sub
->array
.chunk_size
/ 4096) * 4;
672 if (cache
> sub
->cache_size
)
673 rv
= subarray_set_num(container
, sub
,
674 "stripe_cache_size", cache
);
682 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
686 rv
= subarray_set_num(container
, sub
, "raid_disks",
687 raid_disks
+ takeover_delta
);
701 ": failed to initiate container reshape%s%s\n",
702 err
? ": " : "", err
? strerror(err
) : "");
709 static void revert_container_raid_disks(struct supertype
*st
, int fd
, char *container
)
711 /* we failed to prepare all subarrays in the container for
712 * reshape, so cancel the changes and restore the nominal raid
715 struct mdstat_ent
*ent
, *e
;
717 ent
= mdstat_read(0, 0);
720 ": failed to read /proc/mdstat while aborting reshape\n");
724 if (st
->ss
->load_container(st
, fd
, NULL
)) {
726 ": failed read metadata while aborting reshape\n");
731 for (e
= ent
; e
; e
= e
->next
) {
732 int level_fixed
= 0, disks_fixed
= 0;
733 struct mdinfo
*sub
, *prev
;
736 if (!is_container_member(e
, container
))
739 subarray
= to_subarray(e
, container
);
740 prev
= st
->ss
->container_content(st
, subarray
);
742 /* changing level might change raid_disks so we do it
743 * first and then check if raid_disks still needs fixing
745 if (map_name(pers
, e
->level
) != prev
->array
.level
) {
746 sub
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
748 !sysfs_set_num(sub
, NULL
, "level", prev
->array
.level
))
754 sub
= sysfs_read(-1, e
->devnum
, GET_DISKS
);
755 if (sub
&& sub
->array
.raid_disks
!= prev
->array
.raid_disks
) {
756 if (!subarray_set_num(container
, sub
, "raid_disks",
757 prev
->array
.raid_disks
))
763 if (!disks_fixed
|| !level_fixed
)
765 ": failed to restore %s to a %d-disk %s array\n",
766 e
->dev
, prev
->array
.raid_disks
,
767 map_num(pers
, prev
->array
.level
));
770 st
->ss
->free_super(st
);
774 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
776 int level
, char *layout_str
, int chunksize
, int raid_disks
)
778 /* Make some changes in the shape of an array.
779 * The kernel must support the change.
781 * There are three different changes. Each can trigger
782 * a resync or recovery so we freeze that until we have
783 * requested everything (if kernel supports freezing - 2.6.30).
785 * - change size (i.e. component_size)
787 * - change layout/chunksize/ndisks
789 * The last can require a reshape. It is different on different
790 * levels so we need to check the level before actioning it.
791 * Some times the level change needs to be requested after the
792 * reshape (e.g. raid6->raid5, raid5->raid0)
795 struct mdu_array_info_s array
, orig
;
798 struct supertype
*st
;
799 char *subarray
= NULL
;
802 int nlayout
, olayout
;
804 unsigned int ndata
, odata
;
805 int orig_level
= UnSet
;
808 unsigned long long *offsets
;
813 unsigned long a
,b
, blocks
, stripes
;
815 unsigned long long array_size
;
817 char *container
= NULL
;
824 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
825 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
831 (chunksize
|| level
!= UnSet
|| layout_str
|| raid_disks
)) {
832 fprintf(stderr
, Name
": cannot change component size at the same time "
833 "as other changes.\n"
834 " Change size first, then check data is intact before "
835 "making other changes.\n");
839 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
840 get_linux_version() < 2006032 &&
841 !check_env("MDADM_FORCE_FEWER")) {
842 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
843 " Please use a newer kernel\n");
847 st
= super_by_fd(fd
, &subarray
);
849 fprintf(stderr
, Name
": Unable to determine metadata format for %s\n", devname
);
853 /* in the external case we need to check that the requested reshape is
854 * supported, and perform an initial check that the container holds the
855 * pre-requisite spare devices (mdmon owns final validation)
857 if (st
->ss
->external
) {
862 container_dev
= st
->container_dev
;
863 cfd
= open_dev_excl(st
->container_dev
);
864 } else if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 ||
867 Name
": %s is a container, only 'raid-devices' can be changed\n",
871 container_dev
= st
->devnum
;
873 cfd
= open_dev_excl(st
->devnum
);
877 fprintf(stderr
, Name
": Unable to open container for %s\n",
883 container
= devnum2devname(st
->devnum
);
885 fprintf(stderr
, Name
": Could not determine container name\n");
891 rv
= st
->ss
->load_container(st
, cfd
, NULL
);
893 rv
= st
->ss
->load_super(st
, cfd
, NULL
);
895 fprintf(stderr
, Name
": Cannot read superblock for %s\n",
901 if (mdmon_running(container_dev
))
902 st
->update_tail
= &st
->updates
;
905 sra
= sysfs_read(fd
, 0, GET_LEVEL
);
907 if (st
->ss
->external
&& subarray
== NULL
) {
908 array
.level
= LEVEL_CONTAINER
;
909 sra
->array
.level
= LEVEL_CONTAINER
;
912 fprintf(stderr
, Name
": failed to read sysfs parameters for %s\n",
918 /* freeze() already spewed the reason */
920 } else if (frozen
< 0) {
921 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
922 " be reshaped\n", devname
);
926 /* ========= set size =============== */
927 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
928 long long orig_size
= array
.size
;
930 if (reshape_super(st
, size
, UnSet
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
936 if (array
.size
!= size
) {
937 /* got truncated to 32bit, write to
938 * component_size instead
941 rv
= sysfs_set_num(sra
, NULL
,
942 "component_size", size
);
946 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
950 /* restore metadata */
951 if (reshape_super(st
, orig_size
, UnSet
, UnSet
, 0, 0,
952 NULL
, devname
, !quiet
) == 0)
954 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
955 devname
, strerror(err
));
957 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
958 fprintf(stderr
, " Bitmap must be removed before size can be changed\n");
962 ioctl(fd
, GET_ARRAY_INFO
, &array
);
963 size
= get_component_size(fd
)/2;
967 fprintf(stderr
, Name
": component size of %s has been set to %lluK\n",
970 } else if (array
.level
!= LEVEL_CONTAINER
) {
971 size
= get_component_size(fd
)/2;
976 /* ======= set level =========== */
977 if (level
!= UnSet
&& level
!= array
.level
) {
978 /* Trying to change the level.
979 * We might need to change layout first and schedule a
980 * level change for later.
981 * Level changes that can happen immediately are:
982 * 0->4,5,6 1->5 4->5,6 5->1,6
983 * Level changes that need a layout change first are:
984 * 6->5,4,0 : need a -6 layout, or parity-last
985 * 5->4,0 : need parity-last
987 if ((array
.level
== 6 || array
.level
== 5) &&
988 (level
== 5 || level
== 4 || level
== 0)) {
989 /* Don't change level yet, but choose intermediate
993 if (layout_str
== NULL
)
994 switch (array
.layout
) {
995 case ALGORITHM_LEFT_ASYMMETRIC
:
996 case ALGORITHM_LEFT_ASYMMETRIC_6
:
997 case ALGORITHM_ROTATING_N_RESTART
:
998 layout_str
= "left-asymmetric-6";
1000 case ALGORITHM_LEFT_SYMMETRIC
:
1001 case ALGORITHM_LEFT_SYMMETRIC_6
:
1002 case ALGORITHM_ROTATING_N_CONTINUE
:
1003 layout_str
= "left-symmetric-6";
1005 case ALGORITHM_RIGHT_ASYMMETRIC
:
1006 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
1007 case ALGORITHM_ROTATING_ZERO_RESTART
:
1008 layout_str
= "right-asymmetric-6";
1010 case ALGORITHM_RIGHT_SYMMETRIC
:
1011 case ALGORITHM_RIGHT_SYMMETRIC_6
:
1012 layout_str
= "right-symmetric-6";
1014 case ALGORITHM_PARITY_0
:
1015 case ALGORITHM_PARITY_0_6
:
1016 layout_str
= "parity-first-6";
1018 case ALGORITHM_PARITY_N
:
1019 layout_str
= "parity-last";
1022 fprintf(stderr
, Name
": %s: cannot"
1023 "convert layout to RAID5 equivalent\n",
1029 int l
= map_name(r5layout
, layout_str
);
1031 fprintf(stderr
, Name
": %s: layout '%s' not recognised\n",
1032 devname
, layout_str
);
1036 if (l
!= ALGORITHM_PARITY_N
) {
1037 /* need the -6 version */
1038 char *ls
= map_num(r5layout
, l
);
1039 strcat(strcpy(alt_layout
, ls
),
1041 layout_str
= alt_layout
;
1045 /* The final raid6->raid5 conversion
1046 * will reduce the number of disks,
1047 * so now we need to aim higher
1051 layout_str
= "parity-last";
1053 /* Level change is a simple takeover. In the external
1054 * case we don't check with the metadata handler until
1055 * we establish what the final layout will be. If the
1056 * level change is disallowed we will revert to
1057 * orig_level without disturbing the metadata, otherwise
1058 * we will send an update.
1060 c
= map_num(pers
, level
);
1062 rv
= 1;/* not possible */
1065 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1068 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1071 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1072 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1077 orig_level
= orig
.level
;
1078 ioctl(fd
, GET_ARRAY_INFO
, &array
);
1079 if (layout_str
== NULL
&&
1080 orig
.level
== 5 && level
== 6 &&
1081 array
.layout
!= orig
.layout
)
1082 layout_str
= map_num(r5layout
, orig
.layout
);
1084 fprintf(stderr
, Name
" level of %s changed to %s\n",
1090 /* ========= set shape (chunk_size / layout / ndisks) ============== */
1091 /* Check if layout change is a no-op */
1092 switch (array
.level
) {
1094 if (layout_str
&& array
.layout
== map_name(r5layout
, layout_str
))
1098 if (layout_str
== NULL
&&
1099 ((chunksize
&& chunksize
* 1024 != array
.chunk_size
) ||
1100 (raid_disks
&& raid_disks
!= array
.raid_disks
)) &&
1101 array
.layout
>= 16) {
1102 fprintf(stderr
, Name
1103 ": %s has a non-standard layout. If you wish to preserve this\n"
1104 " during the reshape, please specify --layout=preserve\n"
1105 " If you want to change it, specify a layout or use --layout=normalise\n",
1111 (strcmp(layout_str
, "normalise") == 0 ||
1112 strcmp(layout_str
, "normalize") == 0)) {
1114 strcpy(alt_layout
, map_num(r6layout
, array
.layout
));
1115 hyphen
= strrchr(alt_layout
, '-');
1116 if (hyphen
&& strcmp(hyphen
, "-6") == 0) {
1118 layout_str
= alt_layout
;
1122 if (layout_str
&& array
.layout
== map_name(r6layout
, layout_str
))
1124 if (layout_str
&& strcmp(layout_str
, "preserve") == 0)
1128 if (layout_str
== NULL
1129 && (chunksize
== 0 || chunksize
*1024 == array
.chunk_size
)
1130 && (raid_disks
== 0 || raid_disks
== array
.raid_disks
)) {
1131 if (reshape_super(st
, -1, level
, UnSet
, 0, 0, NULL
, devname
, !quiet
)) {
1137 if (level
!= UnSet
&& level
!= array
.level
) {
1138 /* Looks like this level change doesn't need
1139 * a reshape after all.
1141 c
= map_num(pers
, level
);
1143 rv
= sysfs_set_str(sra
, NULL
, "level", c
);
1146 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1149 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1150 fprintf(stderr
, " Bitmap must be removed before level can be changed\n");
1154 } else if (!changed
&& !quiet
)
1155 fprintf(stderr
, Name
": %s: no change requested\n",
1158 if (st
->ss
->external
&& !mdmon_running(st
->container_dev
) &&
1160 start_mdmon(st
->container_dev
);
1161 ping_monitor(container
);
1166 c
= map_num(pers
, array
.level
);
1167 if (c
== NULL
) c
= "-unknown-";
1168 switch (array
.level
) {
1169 default: /* raid0, linear, multipath cannot be reconfigured */
1170 fprintf(stderr
, Name
": %s array %s cannot be reshaped.\n",
1172 /* TODO raid0 raiddisks can be reshaped via raid4 */
1175 case LEVEL_CONTAINER
: {
1178 /* double check that we are not changing anything but raid_disks */
1179 if (size
>= 0 || layout_str
!= NULL
|| chunksize
!= 0 || level
!= UnSet
) {
1181 Name
": %s is a container, only 'raid-devices' can be changed\n",
1187 st
->update_tail
= &st
->updates
;
1188 if (reshape_super(st
, -1, UnSet
, UnSet
, 0, raid_disks
,
1189 backup_file
, devname
, !quiet
)) {
1194 count
= reshape_container_raid_disks(container
, raid_disks
);
1196 revert_container_raid_disks(st
, fd
, container
);
1199 } else if (count
== 0) {
1201 fprintf(stderr
, Name
1202 ": no active subarrays to reshape\n");
1206 if (!mdmon_running(st
->devnum
)) {
1207 start_mdmon(st
->devnum
);
1208 ping_monitor(container
);
1212 /* give mdmon a chance to allocate spares */
1213 ping_manager(container
);
1215 /* manage_reshape takes care of releasing the array(s) */
1216 st
->ss
->manage_reshape(st
, backup_file
);
1220 case LEVEL_FAULTY
: /* only 'layout' change is permitted */
1222 if (chunksize
|| raid_disks
) {
1223 fprintf(stderr
, Name
": %s: Cannot change chunksize or disks of a 'faulty' array\n",
1228 if (layout_str
== NULL
)
1229 break; /* nothing to do.... */
1231 array
.layout
= parse_layout_faulty(layout_str
);
1232 if (array
.layout
< 0) {
1233 fprintf(stderr
, Name
": %s: layout %s not understood for 'faulty' array\n",
1234 devname
, layout_str
);
1238 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1239 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
1240 devname
, strerror(errno
));
1243 printf("layout for %s set to %d\n", devname
, array
.layout
);
1246 case 1: /* only raid_disks can each be changed. */
1248 if (chunksize
|| layout_str
!= NULL
) {
1249 fprintf(stderr
, Name
": %s: Cannot change chunk size or layout for a RAID1 array.\n",
1254 if (raid_disks
> 0) {
1255 if (reshape_super(st
, -1, UnSet
, UnSet
, 0, raid_disks
,
1256 NULL
, devname
, !quiet
)) {
1261 array
.raid_disks
= raid_disks
;
1262 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1263 fprintf(stderr
, Name
": Cannot set raid-devices for %s: %s\n",
1264 devname
, strerror(errno
));
1275 * layout/chunksize/raid_disks can be changed
1276 * though the kernel may not support it all.
1279 fprintf(stderr
, Name
": Cannot reshape subarrays yet\n");
1284 * There are three possibilities.
1285 * 1/ The array will shrink.
1286 * We need to ensure the reshape will pause before reaching
1287 * the 'critical section'. We also need to fork and wait for
1288 * that to happen. When it does we
1289 * suspend/backup/complete/unfreeze
1291 * 2/ The array will not change size.
1292 * This requires that we keep a backup of a sliding window
1293 * so that we can restore data after a crash. So we need
1294 * to fork and monitor progress.
1296 * 3/ The array will grow. This is relatively easy.
1297 * However the kernel's restripe routines will cheerfully
1298 * overwrite some early data before it is safe. So we
1299 * need to make a backup of the early parts of the array
1300 * and be ready to restore it if rebuild aborts very early.
1302 * We backup data by writing it to one spare, or to a
1303 * file which was given on command line.
1305 * [FOLLOWING IS OLD AND PARTLY WRONG]
1306 * So: we enumerate the devices in the array and
1307 * make sure we can open all of them.
1308 * Then we freeze the early part of the array and
1309 * backup to the various spares.
1310 * Then we request changes and start the reshape.
1311 * Monitor progress until it has passed the danger zone.
1312 * and finally invalidate the copied data and unfreeze the
1313 * start of the array.
1315 * In each case, we first make sure that storage is available
1316 * for the required backup.
1318 * - request the shape change.
1319 * - for to handle backup etc.
1321 nchunk
= ochunk
= array
.chunk_size
;
1322 nlayout
= olayout
= array
.layout
;
1323 ndisks
= odisks
= array
.raid_disks
;
1326 nchunk
= chunksize
* 1024;
1327 if (size
% chunksize
) {
1328 fprintf(stderr
, Name
": component size %lluK is not"
1329 " a multiple of chunksize %dK\n",
1334 if (layout_str
!= NULL
)
1335 switch(array
.level
) {
1336 case 4: /* ignore layout */
1339 nlayout
= map_name(r5layout
, layout_str
);
1340 if (nlayout
== UnSet
) {
1341 fprintf(stderr
, Name
": layout %s not understood for raid5.\n",
1349 nlayout
= map_name(r6layout
, layout_str
);
1350 if (nlayout
== UnSet
) {
1351 fprintf(stderr
, Name
": layout %s not understood for raid6.\n",
1358 if (raid_disks
) ndisks
= raid_disks
;
1362 if (array
.level
== 6) {
1363 odata
--; /* number of data disks */
1367 if (odata
== ndata
&&
1368 get_linux_version() < 2006032) {
1369 fprintf(stderr
, Name
": in-place reshape is not safe before 2.6.32, sorry.\n");
1373 /* Check that we can hold all the data */
1374 get_dev_size(fd
, NULL
, &array_size
);
1375 if (ndata
* (unsigned long long)size
< (array_size
/1024)) {
1376 fprintf(stderr
, Name
": this change will reduce the size of the array.\n"
1377 " use --grow --array-size first to truncate array.\n"
1378 " e.g. mdadm --grow %s --array-size %llu\n",
1379 devname
, ndata
* size
);
1384 /* So how much do we need to backup.
1385 * We need an amount of data which is both a whole number of
1386 * old stripes and a whole number of new stripes.
1387 * So LCM for (chunksize*datadisks).
1389 a
= (ochunk
/512) * odata
;
1390 b
= (nchunk
/512) * ndata
;
1398 /* LCM == product / GCD */
1399 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
1402 sra
= sysfs_read(fd
, 0,
1403 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
1407 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
1413 if (ndata
== odata
) {
1414 /* Make 'blocks' bigger for better throughput, but
1415 * not so big that we reject it below.
1416 * Try for 16 megabytes
1418 while (blocks
* 32 < sra
->component_size
&&
1422 fprintf(stderr
, Name
": Need to backup %luK of critical "
1423 "section..\n", blocks
/2);
1425 if (blocks
>= sra
->component_size
/2) {
1426 fprintf(stderr
, Name
": %s: Something wrong - reshape aborted\n",
1431 nrdisks
= array
.raid_disks
+ sra
->array
.spare_disks
;
1432 /* Now we need to open all these devices so we can read/write.
1434 fdlist
= malloc((1+nrdisks
) * sizeof(int));
1435 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
1436 if (!fdlist
|| !offsets
) {
1437 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
1441 for (d
=0; d
<= nrdisks
; d
++)
1443 d
= array
.raid_disks
;
1444 for (sd
= sra
->devs
; sd
; sd
=sd
->next
) {
1445 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1447 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
1448 char *dn
= map_dev(sd
->disk
.major
,
1450 fdlist
[sd
->disk
.raid_disk
]
1451 = dev_open(dn
, O_RDONLY
);
1452 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1453 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
1454 fprintf(stderr
, Name
": %s: cannot open component %s\n",
1455 devname
, dn
?dn
:"-unknown-");
1459 } else if (backup_file
== NULL
) {
1461 char *dn
= map_dev(sd
->disk
.major
,
1463 fdlist
[d
] = dev_open(dn
, O_RDWR
);
1464 offsets
[d
] = (sd
->data_offset
+ sra
->component_size
- blocks
- 8)*512;
1466 fprintf(stderr
, Name
": %s: cannot open component %s\n",
1467 devname
, dn
?dn
:"-unknown");
1474 if (backup_file
== NULL
) {
1475 if (st
->ss
->external
&& !st
->ss
->manage_reshape
) {
1476 fprintf(stderr
, Name
": %s Grow operation not supported by %s metadata\n",
1477 devname
, st
->ss
->name
);
1481 if (ndata
<= odata
) {
1482 fprintf(stderr
, Name
": %s: Cannot grow - need backup-file\n",
1486 } else if (sra
->array
.spare_disks
== 0) {
1487 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
1488 "backup-file to backup critical section\n",
1493 if (d
== array
.raid_disks
) {
1494 fprintf(stderr
, Name
": %s: No spare device for backup\n",
1500 /* need to check backup file is large enough */
1502 fdlist
[d
] = open(backup_file
, O_RDWR
|O_CREAT
|O_EXCL
,
1504 offsets
[d
] = 8 * 512;
1505 if (fdlist
[d
] < 0) {
1506 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1507 devname
, backup_file
, strerror(errno
));
1511 memset(buf
, 0, 512);
1512 for (i
=0; i
< (signed)blocks
+ 1 ; i
++) {
1513 if (write(fdlist
[d
], buf
, 512) != 512) {
1514 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1515 devname
, backup_file
, strerror(errno
));
1520 if (fsync(fdlist
[d
]) != 0) {
1521 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1522 devname
, backup_file
, strerror(errno
));
1529 /* check that the operation is supported by the metadata */
1530 if (reshape_super(st
, -1, level
, nlayout
, nchunk
, ndisks
,
1531 backup_file
, devname
, !quiet
)) {
1536 /* lastly, check that the internal stripe cache is
1537 * large enough, or it won't work.
1540 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
1541 cache
= cache
* 4 / 4096;
1542 if (cache
< blocks
/ 8 / odisks
+ 16)
1543 /* Make it big enough to hold 'blocks' */
1544 cache
= blocks
/ 8 / odisks
+ 16;
1545 if (sra
->cache_size
< cache
)
1546 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
1548 /* Right, everything seems fine. Let's kick things off.
1549 * If only changing raid_disks, use ioctl, else use
1553 if (ochunk
== nchunk
&& olayout
== nlayout
) {
1554 array
.raid_disks
= ndisks
;
1555 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1558 fprintf(stderr
, Name
": Cannot set device shape for %s: %s\n",
1559 devname
, strerror(errno
));
1560 if (ndisks
< odisks
&&
1561 get_linux_version() < 2006030)
1562 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1564 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1565 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1570 /* set them all just in case some old 'new_*' value
1571 * persists from some earlier problem
1573 int err
= err
; /* only used if rv==1, and always set if
1574 * rv==1, so initialisation not needed,
1575 * despite gcc warning
1577 if (sysfs_set_num(sra
, NULL
, "chunk_size", nchunk
) < 0)
1578 rv
= 1, err
= errno
;
1579 if (!rv
&& sysfs_set_num(sra
, NULL
, "layout", nlayout
) < 0)
1580 rv
= 1, err
= errno
;
1581 if (!rv
&& sysfs_set_num(sra
, NULL
, "raid_disks", ndisks
) < 0)
1582 rv
= 1, err
= errno
;
1584 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1586 if (get_linux_version() < 2006030)
1587 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1589 (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)))
1590 fprintf(stderr
, " Bitmap must be removed before shape can be changed\n");
1595 if (ndisks
== 2 && odisks
== 2) {
1596 /* No reshape is needed in this trivial case */
1601 if (st
->ss
->external
) {
1602 /* metadata handler takes it from here */
1603 ping_manager(container
);
1604 st
->ss
->manage_reshape(st
, backup_file
);
1609 /* set up the backup-super-block. This requires the
1610 * uuid from the array.
1612 /* Find a superblock */
1613 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1617 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1619 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1620 devfd
= dev_open(dn
, O_RDONLY
);
1623 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
1629 fprintf(stderr
, Name
": %s: Cannot find a superblock\n",
1635 memset(&bsb
, 0, 512);
1636 memcpy(bsb
.magic
, "md_backup_data-1", 16);
1637 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
1638 bsb
.mtime
= __cpu_to_le64(time(0));
1639 bsb
.devstart2
= blocks
;
1640 stripes
= blocks
/ (ochunk
/512) / odata
;
1641 /* Now we just need to kick off the reshape and watch, while
1642 * handling backups of the data...
1643 * This is all done by a forked background process.
1648 if (check_env("MDADM_GROW_VERIFY"))
1649 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
1652 mlockall(MCL_FUTURE
);
1655 done
= child_grow(fd
, sra
, stripes
,
1657 odisks
, ochunk
, array
.level
, olayout
, odata
,
1658 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1659 else if (odata
> ndata
)
1660 done
= child_shrink(fd
, sra
, stripes
,
1662 odisks
, ochunk
, array
.level
, olayout
, odata
,
1663 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1665 done
= child_same_size(fd
, sra
, stripes
,
1668 odisks
, ochunk
, array
.level
, olayout
, odata
,
1669 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1670 if (backup_file
&& done
)
1671 unlink(backup_file
);
1672 if (level
!= UnSet
&& level
!= array
.level
) {
1673 /* We need to wait for the reshape to finish
1674 * (which will have happened unless odata < ndata)
1675 * and then set the level
1678 c
= map_num(pers
, level
);
1680 exit(0);/* not possible */
1684 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1686 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1691 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
1696 /* The child will take care of unfreezing the array */
1705 if (rv
&& orig_level
!= UnSet
&& sra
) {
1706 c
= map_num(pers
, orig_level
);
1707 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
1708 fprintf(stderr
, Name
": aborting level change\n");
1710 unfreeze(st
, frozen
);
1715 * We run a child process in the background which performs the following
1717 * - wait for resync to reach a certain point
1718 * - suspend io to the following section
1719 * - backup that section
1720 * - allow resync to proceed further
1722 * - discard the backup.
1724 * When are combined in slightly different ways in the three cases.
1726 * - suspend/backup/allow/wait/resume/discard
1728 * - allow/wait/suspend/backup/allow/wait/resume/discard
1730 * - wait/resume/discard/suspend/backup/allow
1732 * suspend/backup/allow always come together
1733 * wait/resume/discard do too.
1734 * For the same-size case we have two backups to improve flow.
1738 /* FIXME return status is never checked */
1739 static int grow_backup(struct mdinfo
*sra
,
1740 unsigned long long offset
, /* per device */
1741 unsigned long stripes
, /* per device */
1742 int *sources
, unsigned long long *offsets
,
1743 int disks
, int chunk
, int level
, int layout
,
1744 int dests
, int *destfd
, unsigned long long *destoffsets
,
1745 int part
, int *degraded
,
1748 /* Backup 'blocks' sectors at 'offset' on each device of the array,
1749 * to storage 'destfd' (offset 'destoffsets'), after first
1750 * suspending IO. Then allow resync to continue
1751 * over the suspended section.
1752 * Use part 'part' of the backup-super-block.
1757 unsigned long long ll
;
1759 //printf("offset %llu\n", offset);
1764 sysfs_set_num(sra
, NULL
, "suspend_hi", (offset
+ stripes
* (chunk
/512)) * odata
);
1765 /* Check that array hasn't become degraded, else we might backup the wrong data */
1766 sysfs_get_ll(sra
, NULL
, "degraded", &ll
);
1767 new_degraded
= (int)ll
;
1768 if (new_degraded
!= *degraded
) {
1769 /* check each device to ensure it is still working */
1771 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1772 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1774 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
1776 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
1777 strstr(sbuf
, "faulty") ||
1778 strstr(sbuf
, "in_sync") == NULL
) {
1779 /* this device is dead */
1780 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
1781 if (sd
->disk
.raid_disk
>= 0 &&
1782 sources
[sd
->disk
.raid_disk
] >= 0) {
1783 close(sources
[sd
->disk
.raid_disk
]);
1784 sources
[sd
->disk
.raid_disk
] = -1;
1789 *degraded
= new_degraded
;
1792 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
1793 bsb
.length2
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
1795 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
1796 bsb
.length
= __cpu_to_le64(stripes
* (chunk
/512) * odata
);
1799 bsb
.magic
[15] = '2';
1800 for (i
= 0; i
< dests
; i
++)
1802 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
1804 lseek64(destfd
[i
], destoffsets
[i
], 0);
1806 rv
= save_stripes(sources
, offsets
,
1807 disks
, chunk
, level
, layout
,
1809 offset
*512*odata
, stripes
* chunk
* odata
,
1814 bsb
.mtime
= __cpu_to_le64(time(0));
1815 for (i
= 0; i
< dests
; i
++) {
1816 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
1818 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
1819 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
1820 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
1821 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
1824 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0)
1825 != destoffsets
[i
] - 4096)
1827 if (write(destfd
[i
], &bsb
, 512) != 512)
1829 if (destoffsets
[i
] > 4096) {
1830 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]+stripes
*chunk
*odata
, 0) !=
1831 destoffsets
[i
]+stripes
*chunk
*odata
)
1833 if (write(destfd
[i
], &bsb
, 512) != 512)
1843 /* in 2.6.30, the value reported by sync_completed can be
1844 * less that it should be by one stripe.
1845 * This only happens when reshape hits sync_max and pauses.
1846 * So allow wait_backup to either extent sync_max further
1847 * than strictly necessary, or return before the
1848 * sync has got quite as far as we would really like.
1849 * This is what 'blocks2' is for.
1850 * The various caller give appropriate values so that
1853 /* FIXME return value is often ignored */
1854 static int wait_backup(struct mdinfo
*sra
,
1855 unsigned long long offset
, /* per device */
1856 unsigned long long blocks
, /* per device */
1857 unsigned long long blocks2
, /* per device - hack */
1858 int dests
, int *destfd
, unsigned long long *destoffsets
,
1861 /* Wait for resync to pass the section that was backed up
1862 * then erase the backup and allow IO
1864 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
1865 unsigned long long completed
;
1871 sysfs_set_num(sra
, NULL
, "sync_max", offset
+ blocks
+ blocks2
);
1873 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
1879 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
1880 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
1884 if (sysfs_get_str(sra
, NULL
, "sync_action",
1886 strncmp(action
, "reshape", 7) != 0)
1888 } while (completed
< offset
+ blocks
);
1892 bsb
.arraystart2
= __cpu_to_le64(0);
1893 bsb
.length2
= __cpu_to_le64(0);
1895 bsb
.arraystart
= __cpu_to_le64(0);
1896 bsb
.length
= __cpu_to_le64(0);
1898 bsb
.mtime
= __cpu_to_le64(time(0));
1900 for (i
= 0; i
< dests
; i
++) {
1901 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
1902 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
1903 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
1904 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
1905 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
1906 if ((unsigned long long)lseek64(destfd
[i
], destoffsets
[i
]-4096, 0) !=
1907 destoffsets
[i
]-4096)
1910 write(destfd
[i
], &bsb
, 512) != 512)
1917 static void fail(char *msg
)
1920 rv
= (write(2, msg
, strlen(msg
)) != (int)strlen(msg
));
1921 rv
|= (write(2, "\n", 1) != 1);
1925 static char *abuf
, *bbuf
;
1926 static unsigned long long abuflen
;
1927 static void validate(int afd
, int bfd
, unsigned long long offset
)
1929 /* check that the data in the backup against the array.
1930 * This is only used for regression testing and should not
1931 * be used while the array is active
1935 lseek64(bfd
, offset
- 4096, 0);
1936 if (read(bfd
, &bsb2
, 512) != 512)
1937 fail("cannot read bsb");
1938 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
1939 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
1940 fail("first csum bad");
1941 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
1942 fail("magic is bad");
1943 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
1944 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
1945 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
1946 fail("second csum bad");
1948 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
1949 fail("devstart is wrong");
1952 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
1954 if (abuflen
< len
) {
1958 if (posix_memalign((void**)&abuf
, 4096, abuflen
) ||
1959 posix_memalign((void**)&bbuf
, 4096, abuflen
)) {
1961 /* just stop validating on mem-alloc failure */
1966 lseek64(bfd
, offset
, 0);
1967 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
) {
1968 //printf("len %llu\n", len);
1969 fail("read first backup failed");
1971 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
1972 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
1973 fail("read first from array failed");
1974 if (memcmp(bbuf
, abuf
, len
) != 0) {
1977 printf("offset=%llu len=%llu\n",
1978 (unsigned long long)__le64_to_cpu(bsb2
.arraystart
)*512, len
);
1979 for (i
=0; i
<len
; i
++)
1980 if (bbuf
[i
] != abuf
[i
]) {
1981 printf("first diff byte %d\n", i
);
1985 fail("data1 compare failed");
1989 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
1991 if (abuflen
< len
) {
1995 abuf
= malloc(abuflen
);
1996 bbuf
= malloc(abuflen
);
1999 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
2000 if ((unsigned long long)read(bfd
, bbuf
, len
) != len
)
2001 fail("read second backup failed");
2002 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
2003 if ((unsigned long long)read(afd
, abuf
, len
) != len
)
2004 fail("read second from array failed");
2005 if (memcmp(bbuf
, abuf
, len
) != 0)
2006 fail("data2 compare failed");
2010 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2011 int *fds
, unsigned long long *offsets
,
2012 int disks
, int chunk
, int level
, int layout
, int data
,
2013 int dests
, int *destfd
, unsigned long long *destoffsets
)
2018 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2019 /* Don't start the 'reshape' */
2021 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2022 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2023 grow_backup(sra
, 0, stripes
,
2024 fds
, offsets
, disks
, chunk
, level
, layout
,
2025 dests
, destfd
, destoffsets
,
2027 validate(afd
, destfd
[0], destoffsets
[0]);
2028 wait_backup(sra
, 0, stripes
* (chunk
/ 512), stripes
* (chunk
/ 512),
2029 dests
, destfd
, destoffsets
,
2031 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2033 /* FIXME this should probably be numeric */
2034 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2038 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2039 int *fds
, unsigned long long *offsets
,
2040 int disks
, int chunk
, int level
, int layout
, int data
,
2041 int dests
, int *destfd
, unsigned long long *destoffsets
)
2044 unsigned long long start
;
2048 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2050 start
= sra
->component_size
- stripes
* (chunk
/512);
2051 sysfs_set_num(sra
, NULL
, "sync_max", start
);
2052 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
2053 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2054 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2055 rv
= wait_backup(sra
, 0, start
- stripes
* (chunk
/512), stripes
* (chunk
/512),
2056 dests
, destfd
, destoffsets
, 0);
2059 grow_backup(sra
, 0, stripes
,
2061 disks
, chunk
, level
, layout
,
2062 dests
, destfd
, destoffsets
,
2064 validate(afd
, destfd
[0], destoffsets
[0]);
2065 wait_backup(sra
, start
, stripes
*(chunk
/512), 0,
2066 dests
, destfd
, destoffsets
, 0);
2067 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* (chunk
/512)) * data
);
2069 /* FIXME this should probably be numeric */
2070 sysfs_set_str(sra
, NULL
, "sync_max", "max");
2074 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
2075 int *fds
, unsigned long long *offsets
,
2076 unsigned long long start
,
2077 int disks
, int chunk
, int level
, int layout
, int data
,
2078 int dests
, int *destfd
, unsigned long long *destoffsets
)
2080 unsigned long long size
;
2081 unsigned long tailstripes
= stripes
;
2084 unsigned long long speed
;
2088 if (posix_memalign((void**)&buf
, 4096, disks
* chunk
))
2091 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
2092 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
2094 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
2095 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
2097 grow_backup(sra
, start
, stripes
,
2099 disks
, chunk
, level
, layout
,
2100 dests
, destfd
, destoffsets
,
2102 grow_backup(sra
, (start
+ stripes
) * (chunk
/512), stripes
,
2104 disks
, chunk
, level
, layout
,
2105 dests
, destfd
, destoffsets
,
2107 validate(afd
, destfd
[0], destoffsets
[0]);
2109 start
+= stripes
* 2; /* where to read next */
2110 size
= sra
->component_size
/ (chunk
/512);
2111 while (start
< size
) {
2112 if (wait_backup(sra
, (start
-stripes
*2)*(chunk
/512),
2113 stripes
*(chunk
/512), 0,
2114 dests
, destfd
, destoffsets
,
2117 sysfs_set_num(sra
, NULL
, "suspend_lo", start
*(chunk
/512) * data
);
2118 if (start
+ stripes
> size
)
2119 tailstripes
= (size
- start
);
2121 grow_backup(sra
, start
*(chunk
/512), tailstripes
,
2123 disks
, chunk
, level
, layout
,
2124 dests
, destfd
, destoffsets
,
2125 part
, °raded
, buf
);
2128 validate(afd
, destfd
[0], destoffsets
[0]);
2130 if (wait_backup(sra
, (start
-stripes
*2) * (chunk
/512), stripes
* (chunk
/512), 0,
2131 dests
, destfd
, destoffsets
,
2134 sysfs_set_num(sra
, NULL
, "suspend_lo", ((start
-stripes
)*(chunk
/512)) * data
);
2135 wait_backup(sra
, (start
-stripes
) * (chunk
/512), tailstripes
* (chunk
/512), 0,
2136 dests
, destfd
, destoffsets
,
2138 sysfs_set_num(sra
, NULL
, "suspend_lo", (size
*(chunk
/512)) * data
);
2139 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
2145 * If any spare contains md_back_data-1 which is recent wrt mtime,
2146 * write that data into the array and update the super blocks with
2147 * the new reshape_progress
2149 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
2150 char *backup_file
, int verbose
)
2154 unsigned long long *offsets
;
2155 unsigned long long nstripe
, ostripe
;
2158 if (info
->new_level
!= info
->array
.level
)
2159 return 1; /* Cannot handle level changes (they are instantaneous) */
2161 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
2162 if (info
->array
.level
== 6) odata
--; /* number of data disks */
2163 ndata
= info
->array
.raid_disks
- 1;
2164 if (info
->new_level
== 6) ndata
--;
2166 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
2168 if (info
->delta_disks
<= 0)
2169 /* Didn't grow, so the backup file must have
2173 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
2174 struct mdinfo dinfo
;
2177 char *devname
, namebuf
[20];
2179 /* This was a spare and may have some saved data on it.
2180 * Load the superblock, find and load the
2181 * backup_super_block.
2182 * If either fail, go on to next device.
2183 * If the backup contains no new info, just return
2184 * else restore data and update all superblocks
2186 if (i
== old_disks
-1) {
2187 fd
= open(backup_file
, O_RDONLY
);
2189 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
2190 backup_file
, strerror(errno
));
2193 devname
= backup_file
;
2198 if (st
->ss
->load_super(st
, fd
, NULL
))
2201 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2202 st
->ss
->free_super(st
);
2205 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
2207 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
2208 continue; /* Cannot seek */
2210 sprintf(namebuf
, "device-%d", i
);
2213 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
2215 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
2216 continue; /* Cannot read */
2218 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
2219 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
2221 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
2224 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
2226 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
2227 continue; /* bad checksum */
2229 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
2230 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
2232 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
2233 continue; /* Bad second checksum */
2235 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
2237 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
2238 continue; /* Wrong uuid */
2241 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2242 * sometimes they aren't... So allow considerable flexability in matching, and allow
2243 * this test to be overridden by an environment variable.
2245 if (info
->array
.utime
> (int)__le64_to_cpu(bsb
.mtime
) + 2*60*60 ||
2246 info
->array
.utime
< (int)__le64_to_cpu(bsb
.mtime
) - 10*60) {
2247 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2248 fprintf(stderr
, Name
": accepting backup with timestamp %lu "
2249 "for array with timestamp %lu\n",
2250 (unsigned long)__le64_to_cpu(bsb
.mtime
),
2251 (unsigned long)info
->array
.utime
);
2254 fprintf(stderr
, Name
": too-old timestamp on "
2255 "backup-metadata on %s\n", devname
);
2256 continue; /* time stamp is too bad */
2260 if (bsb
.magic
[15] == '1') {
2261 if (info
->delta_disks
>= 0) {
2262 /* reshape_progress is increasing */
2263 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2264 info
->reshape_progress
) {
2267 fprintf(stderr
, Name
": backup-metadata found on %s but is not needed\n", devname
);
2268 continue; /* No new data here */
2271 /* reshape_progress is decreasing */
2272 if (__le64_to_cpu(bsb
.arraystart
) >=
2273 info
->reshape_progress
)
2274 goto nonew
; /* No new data here */
2277 if (info
->delta_disks
>= 0) {
2278 /* reshape_progress is increasing */
2279 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
2280 info
->reshape_progress
&&
2281 __le64_to_cpu(bsb
.arraystart2
) + __le64_to_cpu(bsb
.length2
) <
2282 info
->reshape_progress
)
2283 goto nonew
; /* No new data here */
2285 /* reshape_progress is decreasing */
2286 if (__le64_to_cpu(bsb
.arraystart
) >=
2287 info
->reshape_progress
&&
2288 __le64_to_cpu(bsb
.arraystart2
) >=
2289 info
->reshape_progress
)
2290 goto nonew
; /* No new data here */
2293 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
2296 fprintf(stderr
, Name
": Failed to verify secondary backup-metadata block on %s\n",
2298 continue; /* Cannot seek */
2300 /* There should be a duplicate backup superblock 4k before here */
2301 if (lseek64(fd
, -4096, 1) < 0 ||
2302 read(fd
, &bsb2
, sizeof(bsb2
)) != sizeof(bsb2
))
2303 goto second_fail
; /* Cannot find leading superblock */
2304 if (bsb
.magic
[15] == '1')
2305 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
2307 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
2308 if (memcmp(&bsb2
, &bsb
, bsbsize
) != 0)
2309 goto second_fail
; /* Cannot find leading superblock */
2311 /* Now need the data offsets for all devices. */
2312 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
2313 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
2316 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2317 /* FIXME should be this be an error */
2319 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2320 st
->ss
->free_super(st
);
2321 offsets
[j
] = dinfo
.data_offset
* 512;
2323 printf(Name
": restoring critical section\n");
2325 if (restore_stripes(fdlist
, offsets
,
2326 info
->array
.raid_disks
,
2330 fd
, __le64_to_cpu(bsb
.devstart
)*512,
2331 __le64_to_cpu(bsb
.arraystart
)*512,
2332 __le64_to_cpu(bsb
.length
)*512)) {
2333 /* didn't succeed, so giveup */
2335 fprintf(stderr
, Name
": Error restoring backup from %s\n",
2340 if (bsb
.magic
[15] == '2' &&
2341 restore_stripes(fdlist
, offsets
,
2342 info
->array
.raid_disks
,
2346 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
2347 __le64_to_cpu(bsb
.devstart2
)*512,
2348 __le64_to_cpu(bsb
.arraystart2
)*512,
2349 __le64_to_cpu(bsb
.length2
)*512)) {
2350 /* didn't succeed, so giveup */
2352 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
2358 /* Ok, so the data is restored. Let's update those superblocks. */
2360 if (info
->delta_disks
>= 0) {
2361 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
2362 __le64_to_cpu(bsb
.length
);
2363 if (bsb
.magic
[15] == '2') {
2364 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
2365 __le64_to_cpu(bsb
.length2
);
2366 if (p2
> info
->reshape_progress
)
2367 info
->reshape_progress
= p2
;
2370 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
2371 if (bsb
.magic
[15] == '2') {
2372 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
2373 if (p2
< info
->reshape_progress
)
2374 info
->reshape_progress
= p2
;
2377 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
2378 if (fdlist
[j
] < 0) continue;
2379 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
2381 st
->ss
->getinfo_super(st
, &dinfo
, NULL
);
2382 dinfo
.reshape_progress
= info
->reshape_progress
;
2383 st
->ss
->update_super(st
, &dinfo
,
2384 "_reshape_progress",
2386 st
->ss
->store_super(st
, fdlist
[j
]);
2387 st
->ss
->free_super(st
);
2391 /* Didn't find any backup data, try to see if any
2394 if (info
->delta_disks
< 0) {
2395 /* When shrinking, the critical section is at the end.
2396 * So see if we are before the critical section.
2398 unsigned long long first_block
;
2399 nstripe
= ostripe
= 0;
2401 while (ostripe
>= nstripe
) {
2402 ostripe
+= info
->array
.chunk_size
/ 512;
2403 first_block
= ostripe
* odata
;
2404 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
2405 (info
->new_chunk
/512);
2408 if (info
->reshape_progress
>= first_block
)
2411 if (info
->delta_disks
> 0) {
2412 /* See if we are beyond the critical section. */
2413 unsigned long long last_block
;
2414 nstripe
= ostripe
= 0;
2416 while (nstripe
>= ostripe
) {
2417 nstripe
+= info
->new_chunk
/ 512;
2418 last_block
= nstripe
* ndata
;
2419 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
2420 (info
->array
.chunk_size
/512);
2423 if (info
->reshape_progress
>= last_block
)
2426 /* needed to recover critical section! */
2428 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
2432 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
2435 /* Array is assembled and ready to be started, but
2436 * monitoring is probably required.
2439 * - set upper bound for resync
2440 * - initialise the 'suspend' boundaries
2441 * - switch to read-write
2442 * - fork and continue monitoring
2446 unsigned long long backup_offsets
[1];
2447 int odisks
, ndisks
, ochunk
, nchunk
,odata
,ndata
;
2448 unsigned long a
,b
,blocks
,stripes
;
2451 unsigned long long *offsets
;
2453 struct mdinfo
*sra
, *sd
;
2455 unsigned long cache
;
2458 err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
2462 /* make sure reshape doesn't progress until we are ready */
2463 sysfs_set_str(info
, NULL
, "sync_max", "0");
2464 sysfs_set_str(info
, NULL
, "array_state", "active"); /* FIXME or clean */
2466 sra
= sysfs_read(-1, devname2devnum(info
->sys_name
),
2467 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
2472 /* ndisks is not growing, so raid_disks is old and +delta is new */
2473 odisks
= info
->array
.raid_disks
;
2474 ndisks
= odisks
+ info
->delta_disks
;
2477 if (info
->array
.level
== 6) {
2481 ochunk
= info
->array
.chunk_size
;
2482 nchunk
= info
->new_chunk
;
2484 a
= (ochunk
/512) * odata
;
2485 b
= (nchunk
/512) * ndata
;
2493 /* LCM == product / GCD */
2494 blocks
= (ochunk
/512) * (nchunk
/512) * odata
* ndata
/ a
;
2497 while (blocks
* 32 < sra
->component_size
&&
2500 stripes
= blocks
/ (info
->array
.chunk_size
/512) / odata
;
2502 /* check that the internal stripe cache is
2503 * large enough, or it won't work.
2505 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
2506 cache
= cache
* 4 / 4096;
2507 if (cache
< blocks
/ 8 / odisks
+ 16)
2508 /* Make it big enough to hold 'blocks' */
2509 cache
= blocks
/ 8 / odisks
+ 16;
2510 if (sra
->cache_size
< cache
)
2511 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
2514 memset(&bsb
, 0, 512);
2515 memcpy(bsb
.magic
, "md_backup_data-1", 16);
2516 memcpy(&bsb
.set_uuid
, info
->uuid
, 16);
2517 bsb
.mtime
= __cpu_to_le64(time(0));
2518 bsb
.devstart2
= blocks
;
2520 backup_fd
= open(backup_file
, O_RDWR
|O_CREAT
, S_IRUSR
| S_IWUSR
);
2521 backup_list
[0] = backup_fd
;
2522 backup_offsets
[0] = 8 * 512;
2523 fds
= malloc(odisks
* sizeof(fds
[0]));
2524 offsets
= malloc(odisks
* sizeof(offsets
[0]));
2525 for (d
=0; d
<odisks
; d
++)
2528 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2529 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
2531 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
2532 char *dn
= map_dev(sd
->disk
.major
,
2534 fds
[sd
->disk
.raid_disk
]
2535 = dev_open(dn
, O_RDONLY
);
2536 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
2537 if (fds
[sd
->disk
.raid_disk
] < 0) {
2538 fprintf(stderr
, Name
": %s: cannot open component %s\n",
2539 info
->sys_name
, dn
?dn
:"-unknown-");
2550 mlockall(MCL_FUTURE
);
2551 if (info
->delta_disks
< 0)
2552 done
= child_shrink(-1, info
, stripes
,
2554 info
->array
.raid_disks
,
2555 info
->array
.chunk_size
,
2556 info
->array
.level
, info
->array
.layout
,
2558 1, backup_list
, backup_offsets
);
2559 else if (info
->delta_disks
== 0) {
2560 /* The 'start' is a per-device stripe number.
2561 * reshape_progress is a per-array sector number.
2562 * So divide by ndata * chunk_size
2564 unsigned long long start
= info
->reshape_progress
/ ndata
;
2565 start
/= (info
->array
.chunk_size
/512);
2566 done
= child_same_size(-1, info
, stripes
,
2569 info
->array
.raid_disks
,
2570 info
->array
.chunk_size
,
2571 info
->array
.level
, info
->array
.layout
,
2573 1, backup_list
, backup_offsets
);
2575 if (backup_file
&& done
)
2576 unlink(backup_file
);
2577 /* FIXME should I intuit a level change */
2580 fprintf(stderr
, Name
": Cannot run child to continue monitoring reshape: %s\n",