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
;
56 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
57 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
63 fprintf(stderr
, Name
": cannot handle arrays with superblock version %d\n", info
.array
.major_version
);
67 if (info
.array
.level
!= -1) {
68 fprintf(stderr
, Name
": can only add devices to linear arrays\n");
72 nfd
= open(newdev
, O_RDWR
|O_EXCL
|O_DIRECT
);
74 fprintf(stderr
, Name
": cannot open %s\n", newdev
);
78 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
79 fprintf(stderr
, Name
": %s is not a block device!\n", newdev
);
83 /* now check out all the devices and make sure we can read the superblock */
84 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
89 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
90 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
94 dv
= map_dev(disk
.major
, disk
.minor
, 1);
96 fprintf(stderr
, Name
": cannot find device file for device %d\n",
100 fd2
= dev_open(dv
, O_RDWR
);
102 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
105 st
->ss
->free_super(st
);
107 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
108 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
114 /* Ok, looks good. Lets update the superblock and write it out to
118 info
.disk
.number
= d
;
119 info
.disk
.major
= major(stb
.st_rdev
);
120 info
.disk
.minor
= minor(stb
.st_rdev
);
121 info
.disk
.raid_disk
= d
;
122 info
.disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
123 st
->ss
->update_super(st
, &info
, "linear-grow-new", newdev
,
126 if (st
->ss
->store_super(st
, nfd
)) {
127 fprintf(stderr
, Name
": Cannot store new superblock on %s\n",
134 if (ioctl(fd
, ADD_NEW_DISK
, &info
.disk
) != 0) {
135 fprintf(stderr
, Name
": Cannot add new disk to this array\n");
138 /* Well, that seems to have worked.
139 * Now go through and update all superblocks
142 if (ioctl(fd
, GET_ARRAY_INFO
, &info
.array
) < 0) {
143 fprintf(stderr
, Name
": cannot get array info for %s\n", devname
);
148 for (d
=0 ; d
< info
.array
.raid_disks
; d
++) {
149 mdu_disk_info_t disk
;
153 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0) {
154 fprintf(stderr
, Name
": cannot get device detail for device %d\n",
158 dv
= map_dev(disk
.major
, disk
.minor
, 1);
160 fprintf(stderr
, Name
": cannot find device file for device %d\n",
164 fd2
= dev_open(dv
, O_RDWR
);
166 fprintf(stderr
, Name
": cannot open device file %s\n", dv
);
169 if (st
->ss
->load_super(st
, fd2
, NULL
)) {
170 fprintf(stderr
, Name
": cannot find super block on %s\n", dv
);
174 info
.array
.raid_disks
= nd
+1;
175 info
.array
.nr_disks
= nd
+1;
176 info
.array
.active_disks
= nd
+1;
177 info
.array
.working_disks
= nd
+1;
179 st
->ss
->update_super(st
, &info
, "linear-grow-update", dv
,
182 if (st
->ss
->store_super(st
, fd2
)) {
183 fprintf(stderr
, Name
": Cannot store new superblock on %s\n", dv
);
193 int Grow_addbitmap(char *devname
, int fd
, char *file
, int chunk
, int delay
, int write_behind
, int force
)
196 * First check that array doesn't have a bitmap
197 * Then create the bitmap
200 * For internal bitmaps, we need to check the version,
201 * find all the active devices, and write the bitmap block
204 mdu_bitmap_file_t bmf
;
205 mdu_array_info_t array
;
206 struct supertype
*st
;
207 int major
= BITMAP_MAJOR_HI
;
208 int vers
= md_get_version(fd
);
209 unsigned long long bitmapsize
, array_size
;
212 major
= BITMAP_MAJOR_HOSTENDIAN
;
214 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
215 " between different architectured. Consider upgrading the Linux kernel.\n");
219 if (ioctl(fd
, GET_BITMAP_FILE
, &bmf
) != 0) {
221 fprintf(stderr
, Name
": Memory allocation failure.\n");
223 fprintf(stderr
, Name
": bitmaps not supported by this kernel.\n");
226 if (bmf
.pathname
[0]) {
227 if (strcmp(file
,"none")==0) {
228 if (ioctl(fd
, SET_BITMAP_FILE
, -1)!= 0) {
229 fprintf(stderr
, Name
": failed to remove bitmap %s\n",
235 fprintf(stderr
, Name
": %s already has a bitmap (%s)\n",
236 devname
, bmf
.pathname
);
239 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0) {
240 fprintf(stderr
, Name
": cannot get array status for %s\n", devname
);
243 if (array
.state
& (1<<MD_SB_BITMAP_PRESENT
)) {
244 if (strcmp(file
, "none")==0) {
245 array
.state
&= ~(1<<MD_SB_BITMAP_PRESENT
);
246 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
247 fprintf(stderr
, Name
": failed to remove internal bitmap.\n");
252 fprintf(stderr
, Name
": Internal bitmap already present on %s\n",
256 if (array
.level
<= 0) {
257 fprintf(stderr
, Name
": Bitmaps not meaningful with level %s\n",
258 map_num(pers
, array
.level
)?:"of this array");
261 bitmapsize
= array
.size
;
263 if (get_dev_size(fd
, NULL
, &array_size
) &&
264 array_size
> (0x7fffffffULL
<<9)) {
265 /* Array is big enough that we cannot trust array.size
266 * try other approaches
268 bitmapsize
= get_component_size(fd
);
270 if (bitmapsize
== 0) {
271 fprintf(stderr
, Name
": Cannot reliably determine size of array to create bitmap - sorry.\n");
275 if (array
.level
== 10) {
276 int ncopies
= (array
.layout
&255)*((array
.layout
>>8)&255);
277 bitmapsize
= bitmapsize
* array
.raid_disks
/ ncopies
;
280 st
= super_by_fd(fd
);
282 fprintf(stderr
, Name
": Cannot understand version %d.%d\n",
283 array
.major_version
, array
.minor_version
);
286 if (strcmp(file
, "none") == 0) {
287 fprintf(stderr
, Name
": no bitmap found on %s\n", devname
);
289 } else if (strcmp(file
, "internal") == 0) {
291 for (d
=0; d
< st
->max_devs
; d
++) {
292 mdu_disk_info_t disk
;
295 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
297 if (disk
.major
== 0 &&
300 if ((disk
.state
& (1<<MD_DISK_SYNC
))==0)
302 dv
= map_dev(disk
.major
, disk
.minor
, 1);
304 int fd2
= dev_open(dv
, O_RDWR
);
307 if (st
->ss
->load_super(st
, fd2
, NULL
)==0) {
308 if (st
->ss
->add_internal_bitmap(
310 &chunk
, delay
, write_behind
,
311 bitmapsize
, 0, major
)
313 st
->ss
->write_bitmap(st
, fd2
);
315 fprintf(stderr
, Name
": failed to create internal bitmap - chunksize problem.\n");
323 array
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
324 if (ioctl(fd
, SET_ARRAY_INFO
, &array
)!= 0) {
325 fprintf(stderr
, Name
": failed to set internal bitmap.\n");
332 int max_devs
= st
->max_devs
;
334 /* try to load a superblock */
335 for (d
=0; d
<max_devs
; d
++) {
336 mdu_disk_info_t disk
;
340 if (ioctl(fd
, GET_DISK_INFO
, &disk
) < 0)
342 if ((disk
.major
==0 && disk
.minor
==0) ||
343 (disk
.state
& (1<<MD_DISK_REMOVED
)))
345 dv
= map_dev(disk
.major
, disk
.minor
, 1);
347 fd2
= dev_open(dv
, O_RDONLY
);
349 st
->ss
->load_super(st
, fd2
, NULL
) == 0) {
351 st
->ss
->uuid_from_super(st
, uuid
);
357 fprintf(stderr
, Name
": cannot find UUID for array!\n");
360 if (CreateBitmap(file
, force
, (char*)uuid
, chunk
,
361 delay
, write_behind
, bitmapsize
, major
)) {
364 bitmap_fd
= open(file
, O_RDWR
);
366 fprintf(stderr
, Name
": weird: %s cannot be opened\n",
370 if (ioctl(fd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
371 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
372 devname
, strerror(errno
));
382 * When reshaping an array we might need to backup some data.
383 * This is written to all spares with a 'super_block' describing it.
384 * The superblock goes 1K form the end of the used space on the
386 * It if written after the backup is complete.
387 * It has the following structure.
390 static struct mdp_backup_super
{
391 char magic
[16]; /* md_backup_data-1 or -2 */
394 /* start/sizes in 512byte sectors */
395 __u64 devstart
; /* address on backup device/file of data */
398 __u32 sb_csum
; /* csum of preceeding bytes. */
400 __u64 devstart2
; /* offset in to data of second section */
403 __u32 sb_csum2
; /* csum of preceeding bytes. */
405 } __attribute__((aligned(512))) bsb
, bsb2
;
407 int bsb_csum(char *buf
, int len
)
411 for (i
=0; i
<len
; i
++)
412 csum
= (csum
<<3) + buf
[0];
413 return __cpu_to_le32(csum
);
416 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
417 int *fds
, unsigned long long *offsets
,
418 int disks
, int chunk
, int level
, int layout
, int data
,
419 int dests
, int *destfd
, unsigned long long *destoffsets
);
420 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
421 int *fds
, unsigned long long *offsets
,
422 int disks
, int chunk
, int level
, int layout
, int data
,
423 int dests
, int *destfd
, unsigned long long *destoffsets
);
424 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long blocks
,
425 int *fds
, unsigned long long *offsets
,
426 unsigned long long start
,
427 int disks
, int chunk
, int level
, int layout
, int data
,
428 int dests
, int *destfd
, unsigned long long *destoffsets
);
430 int freeze_array(struct mdinfo
*sra
)
432 /* Try to freeze resync on this array.
433 * Return -1 if the array is busy,
434 * return 0 if this kernel doesn't support 'frozen'
435 * return 1 if it worked.
438 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) <= 0)
440 if (strcmp(buf
, "idle\n") != 0 &&
441 strcmp(buf
, "frozen\n") != 0)
443 if (sysfs_set_str(sra
, NULL
, "sync_action", "frozen") < 0)
448 void unfreeze_array(struct mdinfo
*sra
, int frozen
)
450 /* If 'frozen' is 1, unfreeze the array */
452 sysfs_set_str(sra
, NULL
, "sync_action", "idle");
455 void wait_reshape(struct mdinfo
*sra
)
457 int fd
= sysfs_get_fd(sra
, NULL
, "sync_action");
464 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
466 if (sysfs_fd_get_str(fd
, action
, 20) < 0) {
470 } while (strncmp(action
, "reshape", 7) == 0);
474 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
476 int level
, char *layout_str
, int chunksize
, int raid_disks
)
478 /* Make some changes in the shape of an array.
479 * The kernel must support the change.
481 * There are three different changes. Each can trigger
482 * a resync or recovery so we freeze that until we have
483 * requested everything (if kernel supports freezing - 2.6.30).
485 * - change size (i.e. component_size)
487 * - change layout/chunksize/ndisks
489 * The last can require a reshape. It is different on different
490 * levels so we need to check the level before actioning it.
491 * Some times the level change needs to be requested after the
492 * reshape (e.g. raid6->raid5, raid5->raid0)
495 struct mdu_array_info_s array
, orig
;
498 struct supertype
*st
;
501 int nlayout
, olayout
;
504 int orig_level
= UnSet
;
507 unsigned long long *offsets
;
512 unsigned long a
,b
, blocks
, stripes
;
514 unsigned long long array_size
;
521 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
522 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
527 if (raid_disks
&& raid_disks
< array
.raid_disks
&& array
.level
> 1 &&
528 get_linux_version() < 2006032 &&
529 !check_env("MDADM_FORCE_FEWER")) {
530 fprintf(stderr
, Name
": reducing the number of devices is not safe before Linux 2.6.32\n"
531 " Please use a newer kernel\n");
534 sra
= sysfs_read(fd
, 0, GET_LEVEL
);
535 frozen
= freeze_array(sra
);
537 fprintf(stderr
, Name
": %s is performing resync/recovery and cannot"
538 " be reshaped\n", devname
);
542 /* ========= set size =============== */
543 if (size
>= 0 && (size
== 0 || size
!= array
.size
)) {
545 if (array
.size
!= size
) {
546 /* got truncated to 32bit, write to
547 * component_size instead
550 rv
= sysfs_set_num(sra
, NULL
,
551 "component_size", size
);
555 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
557 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
558 devname
, strerror(errno
));
562 ioctl(fd
, GET_ARRAY_INFO
, &array
);
564 fprintf(stderr
, Name
": component size of %s has been set to %dK\n",
565 devname
, array
.size
);
569 /* ======= set level =========== */
570 if (level
!= UnSet
&& level
!= array
.level
) {
571 /* Trying to change the level.
572 * We might need to change layout first and schedule a
573 * level change for later.
574 * Level changes that can happen immediately are:
575 * 0->4,5,6 1->5 4->5,6 5->1,6
576 * Level changes that need a layout change first are:
577 * 6->5,4,0 : need a -6 layout, or parity-last
578 * 5->4,0 : need parity-last
580 if ((array
.level
== 6 || array
.level
== 5) &&
581 (level
== 5 || level
== 4 || level
== 0)) {
582 /* Don't change level yet, but choose intermediate
586 if (layout_str
== NULL
)
587 switch (array
.layout
) {
588 case ALGORITHM_LEFT_ASYMMETRIC
:
589 case ALGORITHM_LEFT_ASYMMETRIC_6
:
590 case ALGORITHM_ROTATING_N_RESTART
:
591 layout_str
= "left-asymmetric-6";
593 case ALGORITHM_LEFT_SYMMETRIC
:
594 case ALGORITHM_LEFT_SYMMETRIC_6
:
595 case ALGORITHM_ROTATING_N_CONTINUE
:
596 layout_str
= "left-symmetric-6";
598 case ALGORITHM_RIGHT_ASYMMETRIC
:
599 case ALGORITHM_RIGHT_ASYMMETRIC_6
:
600 case ALGORITHM_ROTATING_ZERO_RESTART
:
601 layout_str
= "right-asymmetric-6";
603 case ALGORITHM_RIGHT_SYMMETRIC
:
604 case ALGORITHM_RIGHT_SYMMETRIC_6
:
605 layout_str
= "right-symmetric-6";
607 case ALGORITHM_PARITY_0
:
608 case ALGORITHM_PARITY_0_6
:
609 layout_str
= "parity-first-6";
611 case ALGORITHM_PARITY_N
:
612 layout_str
= "parity-last";
615 fprintf(stderr
, Name
": %s: cannot"
616 "convert layout to RAID5 equivalent\n",
622 int l
= map_name(r5layout
, layout_str
);
624 fprintf(stderr
, Name
": %s: layout '%s' not recognised\n",
625 devname
, layout_str
);
629 if (l
!= ALGORITHM_PARITY_N
) {
630 /* need the -6 version */
631 char *ls
= map_num(r5layout
, l
);
632 strcat(strcpy(alt_layout
, ls
),
634 layout_str
= alt_layout
;
638 /* The final raid6->raid5 conversion
639 * will reduce the number of disks,
640 * so now we need to aim higher
644 layout_str
= "parity-last";
646 c
= map_num(pers
, level
);
648 return 1;/* not possible */
649 err
= sysfs_set_str(sra
, NULL
, "level", c
);
651 fprintf(stderr
, Name
": %s: could not set level to %s\n",
657 orig_level
= orig
.level
;
658 ioctl(fd
, GET_ARRAY_INFO
, &array
);
659 if (layout_str
== NULL
&&
660 orig
.level
== 5 && level
== 6 &&
661 array
.layout
!= orig
.layout
)
662 layout_str
= map_num(r5layout
, orig
.layout
);
664 fprintf(stderr
, Name
" level of %s changed to %s\n",
670 /* ========= set shape (chunk_size / layout / ndisks) ============== */
671 /* Check if layout change is a no-op */
672 if (layout_str
) switch(array
.level
) {
674 if (array
.layout
== map_name(r5layout
, layout_str
))
678 if (layout_str
== NULL
&&
679 ((chunksize
&& chunksize
* 1024 != array
.chunk_size
) ||
680 (raid_disks
&& raid_disks
!= array
.raid_disks
)) &&
681 array
.layout
>= 16) {
683 ": %s has a non-standard layout. If you wish to preserve this\n"
684 " during the reshape, please specify --layout=preserve\n"
685 " If you want to change it, specify a layout or use --layout=normalise\n",
690 if (strcmp(layout_str
, "normalise") == 0 ||
691 strcmp(layout_str
, "normalize") == 0) {
693 strcpy(alt_layout
, map_num(r6layout
, array
.layout
));
694 hyphen
= strrchr(alt_layout
, '-');
695 if (hyphen
&& strcmp(hyphen
, "-6") == 0) {
697 layout_str
= alt_layout
;
701 if (array
.layout
== map_name(r6layout
, layout_str
))
703 if (layout_str
&& strcmp(layout_str
, "preserve") == 0)
707 if (layout_str
== NULL
708 && (chunksize
== 0 || chunksize
*1024 == array
.chunk_size
)
709 && (raid_disks
== 0 || raid_disks
== array
.raid_disks
)) {
711 if (level
!= UnSet
&& level
!= array
.level
) {
712 /* Looks like this level change doesn't need
713 * a reshape after all.
715 c
= map_num(pers
, level
);
717 rv
= sysfs_set_str(sra
, NULL
, "level", c
);
719 fprintf(stderr
, Name
": %s: could not set level to %s\n",
722 } else if (!changed
&& !quiet
)
723 fprintf(stderr
, Name
": %s: no change requested\n",
728 c
= map_num(pers
, array
.level
);
729 if (c
== NULL
) c
= "-unknown-";
730 switch(array
.level
) {
731 default: /* raid0, linear, multipath cannot be reconfigured */
732 fprintf(stderr
, Name
": %s array %s cannot be reshaped.\n",
737 case LEVEL_FAULTY
: /* only 'layout' change is permitted */
739 if (chunksize
|| raid_disks
) {
740 fprintf(stderr
, Name
": %s: Cannot change chunksize or disks of a 'faulty' array\n",
745 if (layout_str
== NULL
)
746 break; /* nothing to do.... */
748 array
.layout
= parse_layout_faulty(layout_str
);
749 if (array
.layout
< 0) {
751 fprintf(stderr
, Name
": %s: layout %s not understood for 'faulty' array\n",
752 devname
, layout_str
);
756 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
757 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
758 devname
, strerror(errno
));
761 printf("layout for %s set to %d\n", devname
, array
.layout
);
764 case 1: /* only raid_disks can each be changed. */
766 if (chunksize
|| layout_str
!= NULL
) {
767 fprintf(stderr
, Name
": %s: Cannot change chunk size or layout for a RAID1 array.\n",
772 if (raid_disks
> 0) {
773 array
.raid_disks
= raid_disks
;
774 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
775 fprintf(stderr
, Name
": Cannot set raid-devices for %s: %s\n",
776 devname
, strerror(errno
));
787 * layout/chunksize/raid_disks can be changed
788 * though the kernel may not support it all.
790 st
= super_by_fd(fd
);
793 * There are three possibilities.
794 * 1/ The array will shrink.
795 * We need to ensure the reshape will pause before reaching
796 * the 'critical section'. We also need to fork and wait for
797 * that to happen. When it does we
798 * suspend/backup/complete/unfreeze
800 * 2/ The array will not change size.
801 * This requires that we keep a backup of a sliding window
802 * so that we can restore data after a crash. So we need
803 * to fork and monitor progress.
805 * 3/ The array will grow. This is relatively easy.
806 * However the kernel's restripe routines will cheerfully
807 * overwrite some early data before it is safe. So we
808 * need to make a backup of the early parts of the array
809 * and be ready to restore it if rebuild aborts very early.
811 * We backup data by writing it to one spare, or to a
812 * file which was given on command line.
814 * [FOLLOWING IS OLD AND PARTLY WRONG]
815 * So: we enumerate the devices in the array and
816 * make sure we can open all of them.
817 * Then we freeze the early part of the array and
818 * backup to the various spares.
819 * Then we request changes and start the reshape.
820 * Monitor progress until it has passed the danger zone.
821 * and finally invalidate the copied data and unfreeze the
822 * start of the array.
824 * In each case, we first make sure that storage is available
825 * for the required backup.
827 * - request the shape change.
828 * - for to handle backup etc.
830 nchunk
= ochunk
= array
.chunk_size
;
831 nlayout
= olayout
= array
.layout
;
832 ndisks
= odisks
= array
.raid_disks
;
835 nchunk
= chunksize
* 1024;
836 if (array
.size
% chunksize
) {
837 fprintf(stderr
, Name
": component size %dK is not"
838 " a multiple of chunksize %dK\n",
839 array
.size
, chunksize
);
843 if (layout_str
!= NULL
)
844 switch(array
.level
) {
845 case 4: /* ignore layout */
848 nlayout
= map_name(r5layout
, layout_str
);
849 if (nlayout
== UnSet
) {
850 fprintf(stderr
, Name
": layout %s not understood for raid5.\n",
857 nlayout
= map_name(r6layout
, layout_str
);
858 if (nlayout
== UnSet
) {
859 fprintf(stderr
, Name
": layout %s not understood for raid6.\n",
865 if (raid_disks
) ndisks
= raid_disks
;
869 if (array
.level
== 6) {
870 odata
--; /* number of data disks */
874 /* Check that we can hold all the data */
875 size
= ndata
* array
.size
;
876 get_dev_size(fd
, NULL
, &array_size
);
877 if (size
< (array_size
/1024)) {
878 fprintf(stderr
, Name
": this change will reduce the size of the array.\n"
879 " use --grow --array-size first to truncate array.\n"
880 " e.g. mdadm --grow %s --array-size %llu\n",
886 /* So how much do we need to backup.
887 * We need an amount of data which is both a whole number of
888 * old stripes and a whole number of new stripes.
889 * So LCM for (chunksize*datadisks).
891 a
= ochunk
/512 * odata
;
892 b
= nchunk
/512 * ndata
;
900 /* LCM == product / GCD */
901 blocks
= ochunk
/512 * nchunk
/512 * odata
* ndata
/ a
;
904 sra
= sysfs_read(fd
, 0,
905 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
908 if (ndata
== odata
) {
909 /* Make 'blocks' bigger for better throughput, but
910 * not so big that we reject it below.
912 if (blocks
* 32 < sra
->component_size
)
915 fprintf(stderr
, Name
": Need to backup %luK of critical "
916 "section..\n", blocks
/2);
919 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
925 if (blocks
>= sra
->component_size
/2) {
926 fprintf(stderr
, Name
": %s: Something wrong - reshape aborted\n",
931 nrdisks
= array
.nr_disks
+ sra
->array
.spare_disks
;
932 /* Now we need to open all these devices so we can read/write.
934 fdlist
= malloc((1+nrdisks
) * sizeof(int));
935 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
936 if (!fdlist
|| !offsets
) {
937 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
941 for (d
=0; d
<= nrdisks
; d
++)
943 d
= array
.raid_disks
;
944 for (sd
= sra
->devs
; sd
; sd
=sd
->next
) {
945 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
947 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
948 char *dn
= map_dev(sd
->disk
.major
,
950 fdlist
[sd
->disk
.raid_disk
]
951 = dev_open(dn
, O_RDONLY
);
952 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
953 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
954 fprintf(stderr
, Name
": %s: cannot open component %s\n",
955 devname
, dn
?dn
:"-unknown-");
959 } else if (backup_file
== NULL
) {
961 char *dn
= map_dev(sd
->disk
.major
,
963 fdlist
[d
] = dev_open(dn
, O_RDWR
);
964 offsets
[d
] = (sra
->component_size
- blocks
- 8)*512;
966 fprintf(stderr
, Name
": %s: cannot open component %s\n",
967 devname
, dn
?dn
:"-unknown");
974 if (backup_file
== NULL
) {
975 if (ndata
<= odata
) {
976 fprintf(stderr
, Name
": %s: Cannot grow - need backup-file\n",
980 } else if (sra
->array
.spare_disks
== 0) {
981 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or "
982 "backup-file to backup critical section\n",
987 if (d
== array
.raid_disks
) {
988 fprintf(stderr
, Name
": %s: No spare device for backup\n",
994 /* need to check backup file is large enough */
996 fdlist
[d
] = open(backup_file
, O_RDWR
|O_CREAT
|O_EXCL
,
998 offsets
[d
] = 8 * 512;
1000 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1001 devname
, backup_file
, strerror(errno
));
1005 memset(buf
, 0, 512);
1006 for (i
=0; i
< blocks
+ 1 ; i
++) {
1007 if (write(fdlist
[d
], buf
, 512) != 512) {
1008 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1009 devname
, backup_file
, strerror(errno
));
1014 if (fsync(fdlist
[d
]) != 0) {
1015 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
1016 devname
, backup_file
, strerror(errno
));
1023 /* lastly, check that the internal stripe cache is
1024 * large enough, or it won't work.
1027 cache
= (nchunk
< ochunk
) ? ochunk
: nchunk
;
1028 cache
= cache
* 4 / 4096;
1029 if (sra
->cache_size
< cache
)
1030 sysfs_set_num(sra
, NULL
, "stripe_cache_size",
1032 /* Right, everything seems fine. Let's kick things off.
1033 * If only changing raid_disks, use ioctl, else use
1036 if (ochunk
== nchunk
&& olayout
== nlayout
) {
1037 array
.raid_disks
= ndisks
;
1038 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
1040 fprintf(stderr
, Name
": Cannot set device shape for %s: %s\n",
1041 devname
, strerror(errno
));
1042 if (ndisks
< odisks
&&
1043 get_linux_version() < 2006030)
1044 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1049 /* set them all just in case some old 'new_*' value
1050 * persists from some earlier problem
1052 if (sysfs_set_num(sra
, NULL
, "chunk_size", nchunk
) < 0)
1054 if (sysfs_set_num(sra
, NULL
, "layout", nlayout
) < 0)
1056 if (sysfs_set_num(sra
, NULL
, "raid_disks", ndisks
) < 0)
1059 fprintf(stderr
, Name
": Cannot set device shape for %s\n",
1061 if (get_linux_version() < 2006030)
1062 fprintf(stderr
, Name
": linux 2.6.30 or later required\n");
1067 if (ndisks
== 2 && odisks
== 2) {
1068 /* No reshape is needed in this trivial case */
1073 /* set up the backup-super-block. This requires the
1074 * uuid from the array.
1076 /* Find a superblock */
1077 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1081 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1083 dn
= map_dev(sd
->disk
.major
, sd
->disk
.minor
, 1);
1084 devfd
= dev_open(dn
, O_RDONLY
);
1087 ok
= st
->ss
->load_super(st
, devfd
, NULL
);
1093 fprintf(stderr
, Name
": %s: Cannot find a superblock\n",
1099 memset(&bsb
, 0, 512);
1100 memcpy(bsb
.magic
, "md_backup_data-1", 16);
1101 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
1102 bsb
.mtime
= __cpu_to_le64(time(0));
1103 bsb
.devstart2
= blocks
;
1104 stripes
= blocks
/ (ochunk
/512) / odata
;
1105 /* Now we just need to kick off the reshape and watch, while
1106 * handling backups of the data...
1107 * This is all done by a forked background process.
1112 if (check_env("MDADM_GROW_VERIFY"))
1113 fd
= open(devname
, O_RDONLY
| O_DIRECT
);
1116 mlockall(MCL_FUTURE
);
1119 done
= child_grow(fd
, sra
, stripes
,
1121 odisks
, ochunk
, array
.level
, olayout
, odata
,
1122 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1123 else if (odata
> ndata
)
1124 done
= child_shrink(fd
, sra
, stripes
,
1126 odisks
, ochunk
, array
.level
, olayout
, odata
,
1127 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1129 done
= child_same_size(fd
, sra
, stripes
,
1132 odisks
, ochunk
, array
.level
, olayout
, odata
,
1133 d
- odisks
, fdlist
+odisks
, offsets
+odisks
);
1134 if (backup_file
&& done
)
1135 unlink(backup_file
);
1136 if (level
!= UnSet
&& level
!= array
.level
) {
1137 /* We need to wait for the reshape to finish
1138 * (which will have happened unless odata < ndata)
1139 * and then set the level
1142 c
= map_num(pers
, level
);
1144 exit(0);/* not possible */
1148 err
= sysfs_set_str(sra
, NULL
, "level", c
);
1150 fprintf(stderr
, Name
": %s: could not set level to %s\n",
1155 fprintf(stderr
, Name
": Cannot run child to monitor reshape: %s\n",
1160 /* The child will take care of unfreezing the array */
1169 if (rv
&& orig_level
!= UnSet
&& sra
) {
1170 c
= map_num(pers
, orig_level
);
1171 if (c
&& sysfs_set_str(sra
, NULL
, "level", c
) == 0)
1172 fprintf(stderr
, Name
": aborting level change\n");
1175 unfreeze_array(sra
, frozen
);
1180 * We run a child process in the background which performs the following
1182 * - wait for resync to reach a certain point
1183 * - suspend io to the following section
1184 * - backup that section
1185 * - allow resync to proceed further
1187 * - discard the backup.
1189 * When are combined in slightly different ways in the three cases.
1191 * - suspend/backup/allow/wait/resume/discard
1193 * - allow/wait/suspend/backup/allow/wait/resume/discard
1195 * - wait/resume/discard/suspend/backup/allow
1197 * suspend/backup/allow always come together
1198 * wait/resume/discard do too.
1199 * For the same-size case we have two backups to improve flow.
1203 int grow_backup(struct mdinfo
*sra
,
1204 unsigned long long offset
, /* per device */
1205 unsigned long stripes
, /* per device */
1206 int *sources
, unsigned long long *offsets
,
1207 int disks
, int chunk
, int level
, int layout
,
1208 int dests
, int *destfd
, unsigned long long *destoffsets
,
1209 int part
, int *degraded
,
1212 /* Backup 'blocks' sectors at 'offset' on each device of the array,
1213 * to storage 'destfd' (offset 'destoffsets'), after first
1214 * suspending IO. Then allow resync to continue
1215 * over the suspended section.
1216 * Use part 'part' of the backup-super-block.
1221 unsigned long long new_degraded
;
1222 //printf("offset %llu\n", offset);
1227 sysfs_set_num(sra
, NULL
, "suspend_hi", (offset
+ stripes
* chunk
/512) * odata
);
1228 /* Check that array hasn't become degraded, else we might backup the wrong data */
1229 sysfs_get_ll(sra
, NULL
, "degraded", &new_degraded
);
1230 if (new_degraded
!= *degraded
) {
1231 /* check each device to ensure it is still working */
1233 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1234 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1236 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
1238 if (sysfs_get_str(sra
, sd
, "state", sbuf
, 20) < 0 ||
1239 strstr(sbuf
, "faulty") ||
1240 strstr(sbuf
, "in_sync") == NULL
) {
1241 /* this device is dead */
1242 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
1243 if (sd
->disk
.raid_disk
>= 0 &&
1244 sources
[sd
->disk
.raid_disk
] >= 0) {
1245 close(sources
[sd
->disk
.raid_disk
]);
1246 sources
[sd
->disk
.raid_disk
] = -1;
1251 *degraded
= new_degraded
;
1254 bsb
.arraystart2
= __cpu_to_le64(offset
* odata
);
1255 bsb
.length2
= __cpu_to_le64(stripes
* chunk
/512 * odata
);
1257 bsb
.arraystart
= __cpu_to_le64(offset
* odata
);
1258 bsb
.length
= __cpu_to_le64(stripes
* chunk
/512 * odata
);
1261 bsb
.magic
[15] = '2';
1262 for (i
= 0; i
< dests
; i
++)
1264 lseek64(destfd
[i
], destoffsets
[i
] + __le64_to_cpu(bsb
.devstart2
)*512, 0);
1266 lseek64(destfd
[i
], destoffsets
[i
], 0);
1268 rv
= save_stripes(sources
, offsets
,
1269 disks
, chunk
, level
, layout
,
1271 offset
*512*odata
, stripes
* chunk
* odata
,
1276 bsb
.mtime
= __cpu_to_le64(time(0));
1277 for (i
= 0; i
< dests
; i
++) {
1278 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
1280 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
1281 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
1282 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
1283 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
1285 lseek64(destfd
[i
], destoffsets
[i
] - 4096, 0);
1286 write(destfd
[i
], &bsb
, 512);
1293 /* in 2.6.30, the value reported by sync_completed can be
1294 * less that it should be by one stripe.
1295 * This only happens when reshape hits sync_max and pauses.
1296 * So allow wait_backup to either extent sync_max further
1297 * than strictly necessary, or return before the
1298 * sync has got quite as far as we would really like.
1299 * This is what 'blocks2' is for.
1300 * The various caller give appropriate values so that
1303 int wait_backup(struct mdinfo
*sra
,
1304 unsigned long long offset
, /* per device */
1305 unsigned long long blocks
, /* per device */
1306 unsigned long long blocks2
, /* per device - hack */
1307 int dests
, int *destfd
, unsigned long long *destoffsets
,
1310 /* Wait for resync to pass the section that was backed up
1311 * then erase the backup and allow IO
1313 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
1314 unsigned long long completed
;
1319 sysfs_set_num(sra
, NULL
, "sync_max", offset
+ blocks
+ blocks2
);
1321 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
1327 select(fd
+1, NULL
, NULL
, &rfds
, NULL
);
1328 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
1332 if (sysfs_get_str(sra
, NULL
, "sync_action",
1334 strncmp(action
, "reshape", 7) != 0)
1336 } while (completed
< offset
+ blocks
);
1340 bsb
.arraystart2
= __cpu_to_le64(0);
1341 bsb
.length2
= __cpu_to_le64(0);
1343 bsb
.arraystart
= __cpu_to_le64(0);
1344 bsb
.length
= __cpu_to_le64(0);
1346 bsb
.mtime
= __cpu_to_le64(time(0));
1347 for (i
= 0; i
< dests
; i
++) {
1348 bsb
.devstart
= __cpu_to_le64(destoffsets
[i
]/512);
1349 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
1350 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0)
1351 bsb
.sb_csum2
= bsb_csum((char*)&bsb
,
1352 ((char*)&bsb
.sb_csum2
)-((char*)&bsb
));
1353 lseek64(destfd
[i
], destoffsets
[i
]-4096, 0);
1354 write(destfd
[i
], &bsb
, 512);
1360 static void fail(char *msg
)
1362 write(2, msg
, strlen(msg
));
1367 static char *abuf
, *bbuf
;
1369 static void validate(int afd
, int bfd
, unsigned long long offset
)
1371 /* check that the data in the backup against the array.
1372 * This is only used for regression testing and should not
1373 * be used while the array is active
1377 lseek64(bfd
, offset
- 4096, 0);
1378 if (read(bfd
, &bsb2
, 512) != 512)
1379 fail("cannot read bsb");
1380 if (bsb2
.sb_csum
!= bsb_csum((char*)&bsb2
,
1381 ((char*)&bsb2
.sb_csum
)-((char*)&bsb2
)))
1382 fail("first csum bad");
1383 if (memcmp(bsb2
.magic
, "md_backup_data", 14) != 0)
1384 fail("magic is bad");
1385 if (memcmp(bsb2
.magic
, "md_backup_data-2", 16) == 0 &&
1386 bsb2
.sb_csum2
!= bsb_csum((char*)&bsb2
,
1387 ((char*)&bsb2
.sb_csum2
)-((char*)&bsb2
)))
1388 fail("second csum bad");
1390 if (__le64_to_cpu(bsb2
.devstart
)*512 != offset
)
1391 fail("devstart is wrong");
1394 unsigned long long len
= __le64_to_cpu(bsb2
.length
)*512;
1396 if (abuflen
< len
) {
1400 posix_memalign((void**)&abuf
, 4096, abuflen
);
1401 posix_memalign((void**)&bbuf
, 4096, abuflen
);
1404 lseek64(bfd
, offset
, 0);
1405 if (read(bfd
, bbuf
, len
) != len
) {
1406 printf("len %llu\n", len
);
1407 fail("read first backup failed");
1409 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart
)*512, 0);
1410 if (read(afd
, abuf
, len
) != len
)
1411 fail("read first from array failed");
1412 if (memcmp(bbuf
, abuf
, len
) != 0) {
1414 printf("offset=%llu len=%llu\n",
1415 __le64_to_cpu(bsb2
.arraystart
)*512, len
);
1416 for (i
=0; i
<len
; i
++)
1417 if (bbuf
[i
] != abuf
[i
]) {
1418 printf("first diff byte %d\n", i
);
1421 fail("data1 compare failed");
1425 unsigned long long len
= __le64_to_cpu(bsb2
.length2
)*512;
1427 if (abuflen
< len
) {
1431 abuf
= malloc(abuflen
);
1432 bbuf
= malloc(abuflen
);
1435 lseek64(bfd
, offset
+__le64_to_cpu(bsb2
.devstart2
)*512, 0);
1436 if (read(bfd
, bbuf
, len
) != len
)
1437 fail("read second backup failed");
1438 lseek64(afd
, __le64_to_cpu(bsb2
.arraystart2
)*512, 0);
1439 if (read(afd
, abuf
, len
) != len
)
1440 fail("read second from array failed");
1441 if (memcmp(bbuf
, abuf
, len
) != 0)
1442 fail("data2 compare failed");
1446 static int child_grow(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
1447 int *fds
, unsigned long long *offsets
,
1448 int disks
, int chunk
, int level
, int layout
, int data
,
1449 int dests
, int *destfd
, unsigned long long *destoffsets
)
1454 posix_memalign((void**)&buf
, 4096, disks
* chunk
);
1455 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
1456 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
1457 grow_backup(sra
, 0, stripes
,
1458 fds
, offsets
, disks
, chunk
, level
, layout
,
1459 dests
, destfd
, destoffsets
,
1461 validate(afd
, destfd
[0], destoffsets
[0]);
1462 wait_backup(sra
, 0, stripes
* chunk
/ 512, stripes
* chunk
/ 512,
1463 dests
, destfd
, destoffsets
,
1465 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* chunk
/512) * data
);
1467 /* FIXME this should probably be numeric */
1468 sysfs_set_str(sra
, NULL
, "sync_max", "max");
1472 static int child_shrink(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
1473 int *fds
, unsigned long long *offsets
,
1474 int disks
, int chunk
, int level
, int layout
, int data
,
1475 int dests
, int *destfd
, unsigned long long *destoffsets
)
1478 unsigned long long start
;
1482 posix_memalign((void**)&buf
, 4096, disks
* chunk
);
1483 start
= sra
->component_size
- stripes
* chunk
/512;
1484 sysfs_set_num(sra
, NULL
, "sync_max", start
);
1485 sysfs_set_str(sra
, NULL
, "sync_action", "reshape");
1486 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
1487 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
1488 rv
= wait_backup(sra
, 0, start
- stripes
* chunk
/512, stripes
* chunk
/512,
1489 dests
, destfd
, destoffsets
, 0);
1492 grow_backup(sra
, 0, stripes
,
1494 disks
, chunk
, level
, layout
,
1495 dests
, destfd
, destoffsets
,
1497 validate(afd
, destfd
[0], destoffsets
[0]);
1498 wait_backup(sra
, start
, stripes
*chunk
/512, 0,
1499 dests
, destfd
, destoffsets
, 0);
1500 sysfs_set_num(sra
, NULL
, "suspend_lo", (stripes
* chunk
/512) * data
);
1502 /* FIXME this should probably be numeric */
1503 sysfs_set_str(sra
, NULL
, "sync_max", "max");
1507 static int child_same_size(int afd
, struct mdinfo
*sra
, unsigned long stripes
,
1508 int *fds
, unsigned long long *offsets
,
1509 unsigned long long start
,
1510 int disks
, int chunk
, int level
, int layout
, int data
,
1511 int dests
, int *destfd
, unsigned long long *destoffsets
)
1513 unsigned long long size
;
1514 unsigned long tailstripes
= stripes
;
1517 unsigned long long speed
;
1521 posix_memalign((void**)&buf
, 4096, disks
* chunk
);
1523 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
1524 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
1526 sysfs_get_ll(sra
, NULL
, "sync_speed_min", &speed
);
1527 sysfs_set_num(sra
, NULL
, "sync_speed_min", 200000);
1529 grow_backup(sra
, start
, stripes
,
1531 disks
, chunk
, level
, layout
,
1532 dests
, destfd
, destoffsets
,
1534 grow_backup(sra
, (start
+ stripes
) * chunk
/512, stripes
,
1536 disks
, chunk
, level
, layout
,
1537 dests
, destfd
, destoffsets
,
1539 validate(afd
, destfd
[0], destoffsets
[0]);
1541 start
+= stripes
* 2; /* where to read next */
1542 size
= sra
->component_size
/ (chunk
/512);
1543 while (start
< size
) {
1544 if (wait_backup(sra
, (start
-stripes
*2)*chunk
/512,
1545 stripes
*chunk
/512, 0,
1546 dests
, destfd
, destoffsets
,
1549 sysfs_set_num(sra
, NULL
, "suspend_lo", start
*chunk
/512 * data
);
1550 if (start
+ stripes
> size
)
1551 tailstripes
= (size
- start
);
1553 grow_backup(sra
, start
*chunk
/512, tailstripes
,
1555 disks
, chunk
, level
, layout
,
1556 dests
, destfd
, destoffsets
,
1557 part
, °raded
, buf
);
1560 validate(afd
, destfd
[0], destoffsets
[0]);
1562 if (wait_backup(sra
, (start
-stripes
*2) * chunk
/512, stripes
* chunk
/512, 0,
1563 dests
, destfd
, destoffsets
,
1566 sysfs_set_num(sra
, NULL
, "suspend_lo", ((start
-stripes
)*chunk
/512) * data
);
1567 wait_backup(sra
, (start
-stripes
) * chunk
/512, tailstripes
* chunk
/512, 0,
1568 dests
, destfd
, destoffsets
,
1570 sysfs_set_num(sra
, NULL
, "suspend_lo", (size
*chunk
/512) * data
);
1571 sysfs_set_num(sra
, NULL
, "sync_speed_min", speed
);
1577 * If any spare contains md_back_data-1 which is recent wrt mtime,
1578 * write that data into the array and update the super blocks with
1579 * the new reshape_progress
1581 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
,
1582 char *backup_file
, int verbose
)
1586 unsigned long long *offsets
;
1587 unsigned long long nstripe
, ostripe
;
1590 if (info
->new_level
!= info
->array
.level
)
1591 return 1; /* Cannot handle level changes (they are instantaneous) */
1593 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
1594 if (info
->array
.level
== 6) odata
--; /* number of data disks */
1595 ndata
= info
->array
.raid_disks
- 1;
1596 if (info
->new_level
== 6) ndata
--;
1598 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
1600 if (info
->delta_disks
<= 0)
1601 /* Didn't grow, so the backup file must have
1605 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
1606 struct mdinfo dinfo
;
1610 char *devname
, namebuf
[20];
1612 /* This was a spare and may have some saved data on it.
1613 * Load the superblock, find and load the
1614 * backup_super_block.
1615 * If either fail, go on to next device.
1616 * If the backup contains no new info, just return
1617 * else restore data and update all superblocks
1619 if (i
== old_disks
-1) {
1620 fd
= open(backup_file
, O_RDONLY
);
1622 fprintf(stderr
, Name
": backup file %s inaccessible: %s\n",
1623 backup_file
, strerror(errno
));
1626 devname
= backup_file
;
1631 if (st
->ss
->load_super(st
, fd
, NULL
))
1634 st
->ss
->getinfo_super(st
, &dinfo
);
1635 st
->ss
->free_super(st
);
1638 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
1640 fprintf(stderr
, Name
": Cannot seek on device %d\n", i
);
1641 continue; /* Cannot seek */
1643 sprintf(namebuf
, "device-%d", i
);
1646 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
)) {
1648 fprintf(stderr
, Name
": Cannot read from %s\n", devname
);
1649 continue; /* Cannot read */
1651 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0 &&
1652 memcmp(bsb
.magic
, "md_backup_data-2", 16) != 0) {
1654 fprintf(stderr
, Name
": No backup metadata on %s\n", devname
);
1657 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
))) {
1659 fprintf(stderr
, Name
": Bad backup-metadata checksum on %s\n", devname
);
1660 continue; /* bad checksum */
1662 if (memcmp(bsb
.magic
, "md_backup_data-2", 16) == 0 &&
1663 bsb
.sb_csum2
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum2
)-((char*)&bsb
))) {
1665 fprintf(stderr
, Name
": Bad backup-metadata checksum2 on %s\n", devname
);
1666 continue; /* Bad second checksum */
1668 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0) {
1670 fprintf(stderr
, Name
": Wrong uuid on backup-metadata on %s\n", devname
);
1671 continue; /* Wrong uuid */
1674 if (info
->array
.utime
> __le64_to_cpu(bsb
.mtime
) + 10*60 ||
1675 info
->array
.utime
< __le64_to_cpu(bsb
.mtime
) - 10*60) {
1677 fprintf(stderr
, Name
": too-old timestamp on backup-metadata on %s\n", devname
);
1678 continue; /* time stamp is too bad */
1681 if (bsb
.magic
[15] == '1') {
1682 if (info
->delta_disks
>= 0) {
1683 /* reshape_progress is increasing */
1684 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
1685 info
->reshape_progress
) {
1688 fprintf(stderr
, Name
": backup-metadata found on %s but is not needed\n", devname
);
1689 continue; /* No new data here */
1692 /* reshape_progress is decreasing */
1693 if (__le64_to_cpu(bsb
.arraystart
) >=
1694 info
->reshape_progress
)
1695 goto nonew
; /* No new data here */
1698 if (info
->delta_disks
>= 0) {
1699 /* reshape_progress is increasing */
1700 if (__le64_to_cpu(bsb
.arraystart
) + __le64_to_cpu(bsb
.length
) <
1701 info
->reshape_progress
&&
1702 __le64_to_cpu(bsb
.arraystart2
) + __le64_to_cpu(bsb
.length2
) <
1703 info
->reshape_progress
)
1704 goto nonew
; /* No new data here */
1706 /* reshape_progress is decreasing */
1707 if (__le64_to_cpu(bsb
.arraystart
) >=
1708 info
->reshape_progress
&&
1709 __le64_to_cpu(bsb
.arraystart2
) >=
1710 info
->reshape_progress
)
1711 goto nonew
; /* No new data here */
1714 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0) {
1717 fprintf(stderr
, Name
": Failed to verify secondary backup-metadata block on %s\n",
1719 continue; /* Cannot seek */
1721 /* There should be a duplicate backup superblock 4k before here */
1722 if (lseek64(fd
, -4096, 1) < 0 ||
1723 read(fd
, buf
, 4096) != 4096)
1724 goto second_fail
; /* Cannot find leading superblock */
1725 if (bsb
.magic
[15] == '1')
1726 bsbsize
= offsetof(struct mdp_backup_super
, pad1
);
1728 bsbsize
= offsetof(struct mdp_backup_super
, pad
);
1729 if (memcmp(buf
, &bsb
, bsbsize
) != 0)
1730 goto second_fail
; /* Cannot find leading superblock */
1732 /* Now need the data offsets for all devices. */
1733 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
1734 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
1737 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
1738 /* FIXME should be this be an error */
1740 st
->ss
->getinfo_super(st
, &dinfo
);
1741 st
->ss
->free_super(st
);
1742 offsets
[j
] = dinfo
.data_offset
;
1744 printf(Name
": restoring critical section\n");
1746 if (restore_stripes(fdlist
, offsets
,
1747 info
->array
.raid_disks
,
1751 fd
, __le64_to_cpu(bsb
.devstart
)*512,
1752 __le64_to_cpu(bsb
.arraystart
)*512,
1753 __le64_to_cpu(bsb
.length
)*512)) {
1754 /* didn't succeed, so giveup */
1756 fprintf(stderr
, Name
": Error restoring backup from %s\n",
1761 if (bsb
.magic
[15] == '2' &&
1762 restore_stripes(fdlist
, offsets
,
1763 info
->array
.raid_disks
,
1767 fd
, __le64_to_cpu(bsb
.devstart
)*512 +
1768 __le64_to_cpu(bsb
.devstart2
)*512,
1769 __le64_to_cpu(bsb
.arraystart2
)*512,
1770 __le64_to_cpu(bsb
.length2
)*512)) {
1771 /* didn't succeed, so giveup */
1773 fprintf(stderr
, Name
": Error restoring second backup from %s\n",
1779 /* Ok, so the data is restored. Let's update those superblocks. */
1781 if (info
->delta_disks
>= 0) {
1782 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
) +
1783 __le64_to_cpu(bsb
.length
);
1784 if (bsb
.magic
[15] == '2') {
1785 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
) +
1786 __le64_to_cpu(bsb
.length2
);
1787 if (p2
> info
->reshape_progress
)
1788 info
->reshape_progress
= p2
;
1791 info
->reshape_progress
= __le64_to_cpu(bsb
.arraystart
);
1792 if (bsb
.magic
[15] == '2') {
1793 unsigned long long p2
= __le64_to_cpu(bsb
.arraystart2
);
1794 if (p2
< info
->reshape_progress
)
1795 info
->reshape_progress
= p2
;
1798 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
1799 if (fdlist
[j
] < 0) continue;
1800 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
1802 st
->ss
->getinfo_super(st
, &dinfo
);
1803 dinfo
.reshape_progress
= info
->reshape_progress
;
1804 st
->ss
->update_super(st
, &dinfo
,
1805 "_reshape_progress",
1807 st
->ss
->store_super(st
, fdlist
[j
]);
1808 st
->ss
->free_super(st
);
1812 /* Didn't find any backup data, try to see if any
1815 if (info
->delta_disks
< 0) {
1816 /* When shrinking, the critical section is at the end.
1817 * So see if we are before the critical section.
1819 unsigned long long first_block
;
1820 nstripe
= ostripe
= 0;
1822 while (ostripe
>= nstripe
) {
1823 ostripe
+= info
->array
.chunk_size
/ 512;
1824 first_block
= ostripe
* odata
;
1825 nstripe
= first_block
/ ndata
/ (info
->new_chunk
/512) *
1826 (info
->new_chunk
/512);
1829 if (info
->reshape_progress
>= first_block
)
1832 if (info
->delta_disks
> 0) {
1833 /* See if we are beyond the critical section. */
1834 unsigned long long last_block
;
1835 nstripe
= ostripe
= 0;
1837 while (nstripe
>= ostripe
) {
1838 nstripe
+= info
->new_chunk
/ 512;
1839 last_block
= nstripe
* ndata
;
1840 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
1841 (info
->array
.chunk_size
/512);
1844 if (info
->reshape_progress
>= last_block
)
1847 /* needed to recover critical section! */
1849 fprintf(stderr
, Name
": Failed to find backup of critical section\n");
1853 int Grow_continue(int mdfd
, struct supertype
*st
, struct mdinfo
*info
,
1856 /* Array is assembled and ready to be started, but
1857 * monitoring is probably required.
1860 * - set upper bound for resync
1861 * - initialise the 'suspend' boundaries
1862 * - switch to read-write
1863 * - fork and continue monitoring
1867 unsigned long long backup_offsets
[1];
1868 int odisks
, ndisks
, ochunk
, nchunk
,odata
,ndata
;
1869 unsigned long a
,b
,blocks
,stripes
;
1872 unsigned long long *offsets
;
1874 struct mdinfo
*sra
, *sd
;
1878 err
= sysfs_set_str(info
, NULL
, "array_state", "readonly");
1882 /* make sure reshape doesn't progress until we are ready */
1883 sysfs_set_str(info
, NULL
, "sync_max", "0");
1884 sysfs_set_str(info
, NULL
, "array_state", "active"); /* FIXME or clean */
1886 /* ndisks is not growing, so raid_disks is old and +delta is new */
1887 odisks
= info
->array
.raid_disks
;
1888 ndisks
= odisks
+ info
->delta_disks
;
1891 if (info
->array
.level
== 6) {
1895 ochunk
= info
->array
.chunk_size
;
1896 nchunk
= info
->new_chunk
;
1899 a
= ochunk
/512 * odata
;
1900 b
= nchunk
/512 * ndata
;
1908 /* LCM == product / GCD */
1909 blocks
= ochunk
/512 * nchunk
/512 * odata
* ndata
/ a
;
1913 stripes
= blocks
/ (info
->array
.chunk_size
/512) / odata
;
1916 memset(&bsb
, 0, 512);
1917 memcpy(bsb
.magic
, "md_backup_data-1", 16);
1918 memcpy(&bsb
.set_uuid
, info
->uuid
, 16);
1919 bsb
.mtime
= __cpu_to_le64(time(0));
1920 bsb
.devstart2
= blocks
;
1922 backup_fd
= open(backup_file
, O_RDWR
|O_CREAT
, S_IRUSR
| S_IWUSR
);
1923 backup_list
[0] = backup_fd
;
1924 backup_offsets
[0] = 8 * 512;
1925 fds
= malloc(odisks
* sizeof(fds
[0]));
1926 offsets
= malloc(odisks
* sizeof(offsets
[0]));
1927 for (d
=0; d
<odisks
; d
++)
1930 sra
= sysfs_read(-1, devname2devnum(info
->sys_name
),
1931 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
1934 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1935 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
1937 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
1938 char *dn
= map_dev(sd
->disk
.major
,
1940 fds
[sd
->disk
.raid_disk
]
1941 = dev_open(dn
, O_RDONLY
);
1942 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
*512;
1943 if (fds
[sd
->disk
.raid_disk
] < 0) {
1944 fprintf(stderr
, Name
": %s: cannot open component %s\n",
1945 info
->sys_name
, dn
?dn
:"-unknown-");
1956 mlockall(MCL_FUTURE
);
1957 if (info
->delta_disks
< 0)
1958 done
= child_shrink(-1, info
, stripes
,
1960 info
->array
.raid_disks
,
1961 info
->array
.chunk_size
,
1962 info
->array
.level
, info
->array
.layout
,
1964 1, backup_list
, backup_offsets
);
1965 else if (info
->delta_disks
== 0) {
1966 /* The 'start' is a per-device stripe number.
1967 * reshape_progress is a per-array sector number.
1968 * So divide by ndata * chunk_size
1970 unsigned long long start
= info
->reshape_progress
/ ndata
;
1971 start
/= (info
->array
.chunk_size
/512);
1972 done
= child_same_size(-1, info
, stripes
,
1975 info
->array
.raid_disks
,
1976 info
->array
.chunk_size
,
1977 info
->array
.level
, info
->array
.layout
,
1979 1, backup_list
, backup_offsets
);
1981 if (backup_file
&& done
)
1982 unlink(backup_file
);
1983 /* FIXME should I intuit a level change */
1986 fprintf(stderr
, Name
": Cannot run child to continue monitoring reshape: %s\n",