]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
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 struct mdp_backup_super
{
391 char magic
[16]; /* md_backup_data-1 */
394 /* start/sizes in 512byte sectors */
398 __u32 sb_csum
; /* csum of preceeding bytes. */
400 } __attribute__((aligned(512))) bsb
;
402 int bsb_csum(char *buf
, int len
)
406 for (i
=0; i
<len
; i
++)
407 csum
= (csum
<<3) + buf
[0];
408 return __cpu_to_le32(csum
);
411 int Grow_reshape(char *devname
, int fd
, int quiet
, char *backup_file
,
413 int level
, char *layout_str
, int chunksize
, int raid_disks
)
415 /* Make some changes in the shape of an array.
416 * The kernel must support the change.
417 * Different reshapes have subtly different meaning for different
418 * levels, so we need to check the current state of the array
421 struct mdu_array_info_s array
;
424 struct supertype
*st
;
428 int nlayout
, olayout
;
431 unsigned long long nstripe
, ostripe
, last_block
;
433 unsigned long long *offsets
;
442 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) < 0) {
443 fprintf(stderr
, Name
": %s is not an active md array - aborting\n",
447 c
= map_num(pers
, array
.level
);
448 if (c
== NULL
) c
= "-unknown-";
449 switch(array
.level
) {
450 default: /* raid0, linear, multipath cannot be reconfigured */
451 fprintf(stderr
, Name
": %s array %s cannot be reshaped.\n",
455 case LEVEL_FAULTY
: /* only 'layout' change is permitted */
458 fprintf(stderr
, Name
": %s: Cannot change size of a 'faulty' array\n",
462 if (level
!= UnSet
&& level
!= LEVEL_FAULTY
) {
463 fprintf(stderr
, Name
": %s: Cannot change RAID level of a 'faulty' array\n",
467 if (chunksize
|| raid_disks
) {
468 fprintf(stderr
, Name
": %s: Cannot change chunksize or disks of a 'faulty' array\n",
472 if (layout_str
== NULL
)
473 return 0; /* nothing to do.... */
475 array
.layout
= parse_layout_faulty(layout_str
);
476 if (array
.layout
< 0) {
477 fprintf(stderr
, Name
": %s: layout %s not understood for 'faulty' array\n",
478 devname
, layout_str
);
481 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
482 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
483 devname
, strerror(errno
));
487 printf("layout for %s set to %d\n", devname
, array
.layout
);
490 case 1: /* raid_disks and size can each be changed. They are independant */
492 if (level
!= UnSet
&& level
!= 1) {
493 fprintf(stderr
, Name
": %s: Cannot change RAID level of a RAID1 array.\n",
497 if (chunksize
|| layout_str
!= NULL
) {
498 fprintf(stderr
, Name
": %s: Cannot change chunk size of layout for a RAID1 array.\n",
503 /* Each can trigger a resync/recovery which will block the
504 * other from happening. Later we could block
505 * resync for the duration via 'sync_action'...
507 if (raid_disks
> 0) {
508 array
.raid_disks
= raid_disks
;
509 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
510 fprintf(stderr
, Name
": Cannot set raid-devices for %s: %s\n",
511 devname
, strerror(errno
));
518 if (array
.size
!= size
) {
519 /* got truncated to 32bit, write to
520 * component_size instead
522 sra
= sysfs_read(fd
, 0, 0);
524 rv
= sysfs_set_num(sra
, NULL
,
525 "component_size", size
);
529 rv
= ioctl(fd
, SET_ARRAY_INFO
, &array
);
531 fprintf(stderr
, Name
": Cannot set device size for %s: %s\n",
532 devname
, strerror(errno
));
541 st
= super_by_fd(fd
);
543 /* size can be changed independently.
544 * layout/chunksize/raid_disks/level can be changed
545 * though the kernel may not support it all.
546 * If 'suspend_lo' is not present in devfs, then
547 * these cannot be changed.
550 /* Cannot change other details as well.. */
551 if (layout_str
!= NULL
||
555 fprintf(stderr
, Name
": %s: Cannot change shape as well as size of a %s array.\n",
560 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
561 fprintf(stderr
, Name
": Cannot set device size/shape for %s: %s\n",
562 devname
, strerror(errno
));
567 /* Ok, just change the shape. This can be awkward.
568 * There are three possibilities.
569 * 1/ The array will shrink. We don't support this
570 * possibility. Maybe one day...
571 * 2/ The array will not change size. This is easy enough
572 * to do, but not reliably. If the process is aborted
573 * the array *will* be corrupted. So maybe we can allow
574 * this but only if the user is really certain. e.g.
575 * --really-risk-everything
576 * 3/ The array will grow. This can be reliably achieved.
577 * However the kernel's restripe routines will cheerfully
578 * overwrite some early data before it is safe. So we
579 * need to make a backup of the early parts of the array
580 * and be ready to restore it if rebuild aborts very early.
582 * We backup data by writing it to all spares (there must be
583 * at least 1, so even raid6->raid5 requires a spare to be
586 * So: we enumerate the devices in the array and
587 * make sure we can open all of them.
588 * Then we freeze the early part of the array and
589 * backup to the various spares.
590 * Then we request changes and start the reshape.
591 * Monitor progress until it has passed the danger zone.
592 * and finally invalidate the copied data and unfreeze the
593 * start of the array.
595 * Before we can do this we need to decide:
596 * - will the array grow? Just calculate size
597 * - how much needs to be saved: count stripes.
598 * - where to save data... good question.
601 nlevel
= olevel
= array
.level
;
602 nchunk
= ochunk
= array
.chunk_size
;
603 nlayout
= olayout
= array
.layout
;
604 ndisks
= odisks
= array
.raid_disks
;
606 if (level
!= UnSet
) nlevel
= level
;
607 if (chunksize
) nchunk
= chunksize
;
608 if (layout_str
!= NULL
)
610 case 4: /* ignore layout */
613 nlayout
= map_name(r5layout
, layout_str
);
614 if (nlayout
== UnSet
) {
615 fprintf(stderr
, Name
": layout %s not understood for raid5.\n",
622 nlayout
= map_name(r6layout
, layout_str
);
623 if (nlayout
== UnSet
) {
624 fprintf(stderr
, Name
": layout %s not understood for raid6.\n",
630 if (raid_disks
) ndisks
= raid_disks
;
633 if (olevel
== 6) odata
--; /* number of data disks */
635 if (nlevel
== 6) ndata
--;
638 fprintf(stderr
, Name
": %s: Cannot reduce number of data disks (yet).\n",
642 if (ndata
== odata
) {
643 fprintf(stderr
, Name
": %s: Cannot reshape array without increasing size (yet).\n",
647 /* Well, it is growing... so how much do we need to backup.
648 * Need to backup a full number of new-stripes, such that the
649 * last one does not over-write any place that it would be read
652 nstripe
= ostripe
= 0;
653 while (nstripe
>= ostripe
) {
654 nstripe
+= nchunk
/512;
655 last_block
= nstripe
* ndata
;
656 ostripe
= last_block
/ odata
/ (ochunk
/512) * (ochunk
/512);
658 fprintf(stderr
, Name
": Need to backup %lluK of critical "
659 "section..\n", last_block
/2);
661 sra
= sysfs_read(fd
, 0,
662 GET_COMPONENT
|GET_DEVS
|GET_OFFSET
|GET_STATE
|
665 fprintf(stderr
, Name
": %s: Cannot get array details from sysfs\n",
670 if (last_block
>= sra
->component_size
/2) {
671 fprintf(stderr
, Name
": %s: Something wrong - reshape aborted\n",
675 if (sra
->array
.spare_disks
== 0 && backup_file
== NULL
) {
676 fprintf(stderr
, Name
": %s: Cannot grow - need a spare or backup-file to backup critical section\n",
681 nrdisks
= array
.nr_disks
+ sra
->array
.spare_disks
;
682 /* Now we need to open all these devices so we can read/write.
684 fdlist
= malloc((1+nrdisks
) * sizeof(int));
685 offsets
= malloc((1+nrdisks
) * sizeof(offsets
[0]));
686 if (!fdlist
|| !offsets
) {
687 fprintf(stderr
, Name
": malloc failed: grow aborted\n");
690 for (d
=0; d
<= nrdisks
; d
++)
692 d
= array
.raid_disks
;
693 for (sd
= sra
->devs
; sd
; sd
=sd
->next
) {
694 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
696 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
697 char *dn
= map_dev(sd
->disk
.major
,
699 fdlist
[sd
->disk
.raid_disk
]
700 = dev_open(dn
, O_RDONLY
);
701 offsets
[sd
->disk
.raid_disk
] = sd
->data_offset
;
702 if (fdlist
[sd
->disk
.raid_disk
] < 0) {
703 fprintf(stderr
, Name
": %s: cannot open component %s\n",
704 devname
, dn
?dn
:"-unknown-");
709 char *dn
= map_dev(sd
->disk
.major
,
711 fdlist
[d
] = dev_open(dn
, O_RDWR
);
712 offsets
[d
] = sd
->data_offset
;
714 fprintf(stderr
, Name
": %s: cannot open component %s\n",
715 devname
, dn
?dn
:"-unknown");
721 for (i
=0 ; i
<array
.raid_disks
; i
++)
723 fprintf(stderr
, Name
": %s: failed to find device %d. Array might be degraded.\n"
724 " --grow aborted\n", devname
, i
);
727 spares
= sra
->array
.spare_disks
;
729 fdlist
[d
] = open(backup_file
, O_RDWR
|O_CREAT
|O_EXCL
, S_IRUSR
| S_IWUSR
);
731 fprintf(stderr
, Name
": %s: cannot create backup file %s: %s\n",
732 devname
, backup_file
, strerror(errno
));
739 if (fdlist
[array
.raid_disks
] < 0) {
740 fprintf(stderr
, Name
": %s: failed to find a spare and no backup-file given - --grow aborted\n",
745 /* Find a superblock */
746 if (st
->ss
->load_super(st
, fdlist
[0], NULL
)) {
747 fprintf(stderr
, Name
": %s: Cannot find a superblock\n",
753 memcpy(bsb
.magic
, "md_backup_data-1", 16);
754 st
->ss
->uuid_from_super(st
, (int*)&bsb
.set_uuid
);
755 bsb
.mtime
= __cpu_to_le64(time(0));
757 bsb
.length
= __cpu_to_le64(last_block
);
759 /* Decide offset for the backup, llseek the spares, and write
760 * a leading superblock 4K earlier.
762 for (i
=array
.raid_disks
; i
<d
; i
++) {
764 char *buf
= (char*)(((unsigned long)abuf
+511)& ~511);
765 if (i
==d
-1 && backup_file
) {
766 /* This is the backup file */
769 offsets
[i
] += sra
->component_size
- last_block
- 8;
770 if (lseek64(fdlist
[i
], (offsets
[i
]<<9) - 4096, 0)
771 != (offsets
[i
]<<9) - 4096) {
772 fprintf(stderr
, Name
": could not seek...\n");
775 memset(buf
, 0, 4096);
776 bsb
.devstart
= __cpu_to_le64(offsets
[i
]);
777 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
778 memcpy(buf
, &bsb
, sizeof(bsb
));
779 if (write(fdlist
[i
], buf
, 4096) != 4096) {
780 fprintf(stderr
, Name
": could not write leading superblock\n");
784 array
.level
= nlevel
;
785 array
.raid_disks
= ndisks
;
786 array
.chunk_size
= nchunk
;
787 array
.layout
= nlayout
;
788 if (ioctl(fd
, SET_ARRAY_INFO
, &array
) != 0) {
789 if (errno
== ENOSPC
) {
790 /* stripe cache is not big enough.
791 * It needs to be 4 times chunksize_size,
792 * and we assume pagesize is 4K
794 if (sra
->cache_size
< 4 * (nchunk
/4096)) {
795 sysfs_set_num(sra
, NULL
,
797 4 * (nchunk
/4096) +1);
798 if (ioctl(fd
, SET_ARRAY_INFO
,
803 fprintf(stderr
, Name
": Cannot set device size/shape for %s: %s\n",
804 devname
, strerror(errno
));
809 /* suspend the relevant region */
810 sysfs_set_num(sra
, NULL
, "suspend_hi", 0); /* just in case */
811 if (sysfs_set_num(sra
, NULL
, "suspend_lo", 0) < 0 ||
812 sysfs_set_num(sra
, NULL
, "suspend_hi", last_block
) < 0) {
813 fprintf(stderr
, Name
": %s: failed to suspend device.\n",
818 buf
= malloc(odisks
* ochunk
);
820 err
= save_stripes(fdlist
, offsets
,
821 odisks
, ochunk
, olevel
, olayout
,
822 spares
, fdlist
+odisks
,
823 0ULL, last_block
*512, buf
);
825 /* abort if there was an error */
827 fprintf(stderr
, Name
": %s: failed to save critical region\n",
832 for (i
=odisks
; i
<d
; i
++) {
833 bsb
.devstart
= __cpu_to_le64(offsets
[i
]);
834 bsb
.sb_csum
= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
));
835 if (lseek64(fdlist
[i
], (offsets
[i
]+last_block
)<<9, 0) < 0 ||
836 write(fdlist
[i
], &bsb
, sizeof(bsb
)) != sizeof(bsb
) ||
837 fsync(fdlist
[i
]) != 0) {
838 fprintf(stderr
, Name
": %s: failed to save metadata for critical region backups.\n",
844 /* start the reshape happening */
845 if (sysfs_set_str(sra
, NULL
, "sync_action", "reshape") < 0) {
846 fprintf(stderr
, Name
": %s: failed to initiate reshape\n",
850 /* wait for reshape to pass the critical region */
852 unsigned long long comp
;
854 if (sysfs_get_ll(sra
, NULL
, "sync_completed", &comp
)<0) {
861 /* Maybe it finished already */
863 if (sysfs_get_str(sra
, NULL
, "sync_action",
865 strncmp(action
, "reshape", 7) != 0)
871 /* invalidate superblocks */
872 memset(&bsb
, 0, sizeof(bsb
));
873 for (i
=odisks
; i
<d
; i
++) {
874 lseek64(fdlist
[i
], (offsets
[i
]+last_block
)<<9, 0);
875 if (write(fdlist
[i
], &bsb
, sizeof(bsb
)) < 0) {
876 fprintf(stderr
, Name
": %s: failed to invalidate metadata for raid disk %d\n",
882 sysfs_set_num(sra
, NULL
, "suspend_lo", last_block
);
892 fprintf(stderr
, Name
": ... critical section passed.\n");
899 sysfs_set_num(sra
, NULL
, "suspend_lo", last_block
);
901 for (i
=0; i
<array
.nr_disks
; i
++)
913 * If any spare contains md_back_data-1 which is recent wrt mtime,
914 * write that data into the array and update the super blocks with
915 * the new reshape_progress
917 int Grow_restart(struct supertype
*st
, struct mdinfo
*info
, int *fdlist
, int cnt
, char *backup_file
)
921 unsigned long long *offsets
;
922 unsigned long long nstripe
, ostripe
, last_block
;
925 if (info
->delta_disks
< 0)
926 return 1; /* cannot handle a shrink */
927 if (info
->new_level
!= info
->array
.level
||
928 info
->new_layout
!= info
->array
.layout
||
929 info
->new_chunk
!= info
->array
.chunk_size
)
930 return 1; /* Can only handle change in disks */
932 old_disks
= info
->array
.raid_disks
- info
->delta_disks
;
934 for (i
=old_disks
-(backup_file
?1:0); i
<cnt
; i
++) {
939 /* This was a spare and may have some saved data on it.
940 * Load the superblock, find and load the
941 * backup_super_block.
942 * If either fail, go on to next device.
943 * If the backup contains no new info, just return
944 * else restore data and update all superblocks
946 if (i
== old_disks
-1) {
947 fd
= open(backup_file
, O_RDONLY
);
954 if (st
->ss
->load_super(st
, fd
, NULL
))
957 st
->ss
->getinfo_super(st
, &dinfo
);
958 st
->ss
->free_super(st
);
961 (dinfo
.data_offset
+ dinfo
.component_size
- 8) <<9,
963 continue; /* Cannot seek */
965 if (read(fd
, &bsb
, sizeof(bsb
)) != sizeof(bsb
))
966 continue; /* Cannot read */
967 if (memcmp(bsb
.magic
, "md_backup_data-1", 16) != 0)
969 if (bsb
.sb_csum
!= bsb_csum((char*)&bsb
, ((char*)&bsb
.sb_csum
)-((char*)&bsb
)))
970 continue; /* bad checksum */
971 if (memcmp(bsb
.set_uuid
,info
->uuid
, 16) != 0)
972 continue; /* Wrong uuid */
974 if (info
->array
.utime
> __le64_to_cpu(bsb
.mtime
) + 3600 ||
975 info
->array
.utime
< __le64_to_cpu(bsb
.mtime
))
976 continue; /* time stamp is too bad */
978 if (__le64_to_cpu(bsb
.arraystart
) != 0)
979 continue; /* Can only handle backup from start of array */
980 if (__le64_to_cpu(bsb
.length
) <
981 info
->reshape_progress
)
982 continue; /* No new data here */
984 if (lseek64(fd
, __le64_to_cpu(bsb
.devstart
)*512, 0)< 0)
985 continue; /* Cannot seek */
986 /* There should be a duplicate backup superblock 4k before here */
987 if (lseek64(fd
, -4096, 1) < 0 ||
988 read(fd
, buf
, 4096) != 4096 ||
989 memcmp(buf
, &bsb
, sizeof(bsb
)) != 0)
990 continue; /* Cannot find leading superblock */
992 /* Now need the data offsets for all devices. */
993 offsets
= malloc(sizeof(*offsets
)*info
->array
.raid_disks
);
994 for(j
=0; j
<info
->array
.raid_disks
; j
++) {
997 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
998 /* FIXME should be this be an error */
1000 st
->ss
->getinfo_super(st
, &dinfo
);
1001 st
->ss
->free_super(st
);
1002 offsets
[j
] = dinfo
.data_offset
;
1004 printf(Name
": restoring critical section\n");
1006 if (restore_stripes(fdlist
, offsets
,
1007 info
->array
.raid_disks
,
1011 fd
, __le64_to_cpu(bsb
.devstart
)*512,
1012 0, __le64_to_cpu(bsb
.length
)*512)) {
1013 /* didn't succeed, so giveup */
1017 /* Ok, so the data is restored. Let's update those superblocks. */
1019 for (j
=0; j
<info
->array
.raid_disks
; j
++) {
1020 if (fdlist
[j
] < 0) continue;
1021 if (st
->ss
->load_super(st
, fdlist
[j
], NULL
))
1023 st
->ss
->getinfo_super(st
, &dinfo
);
1024 dinfo
.reshape_progress
= __le64_to_cpu(bsb
.length
);
1025 st
->ss
->update_super(st
, &dinfo
,
1026 "_reshape_progress",
1028 st
->ss
->store_super(st
, fdlist
[j
]);
1029 st
->ss
->free_super(st
);
1032 /* And we are done! */
1035 /* Didn't find any backup data, try to see if any
1038 nstripe
= ostripe
= 0;
1039 odata
= info
->array
.raid_disks
- info
->delta_disks
- 1;
1040 if (info
->array
.level
== 6) odata
--; /* number of data disks */
1041 ndata
= info
->array
.raid_disks
- 1;
1042 if (info
->new_level
== 6) ndata
--;
1044 while (nstripe
>= ostripe
) {
1045 nstripe
+= info
->new_chunk
/ 512;
1046 last_block
= nstripe
* ndata
;
1047 ostripe
= last_block
/ odata
/ (info
->array
.chunk_size
/512) *
1048 (info
->array
.chunk_size
/512);
1051 if (info
->reshape_progress
>= last_block
)
1053 /* needed to recover critical section! */