2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
15 * flush after percent set rather than just time based. (maybe both).
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
33 static inline char *bmname(struct bitmap
*bitmap
)
35 return bitmap
->mddev
? mdname(bitmap
->mddev
) : "mdX";
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
48 static int bitmap_checkpage(struct bitmap_counts
*bitmap
,
49 unsigned long page
, int create
)
50 __releases(bitmap
->lock
)
51 __acquires(bitmap
->lock
)
53 unsigned char *mappage
;
55 if (page
>= bitmap
->pages
) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
63 if (bitmap
->bp
[page
].hijacked
) /* it's hijacked, don't try to alloc */
66 if (bitmap
->bp
[page
].map
) /* page is already allocated, just return */
72 /* this page has not been allocated yet */
74 spin_unlock_irq(&bitmap
->lock
);
75 mappage
= kzalloc(PAGE_SIZE
, GFP_NOIO
);
76 spin_lock_irq(&bitmap
->lock
);
78 if (mappage
== NULL
) {
79 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
80 /* failed - set the hijacked flag so that we can use the
81 * pointer as a counter */
82 if (!bitmap
->bp
[page
].map
)
83 bitmap
->bp
[page
].hijacked
= 1;
84 } else if (bitmap
->bp
[page
].map
||
85 bitmap
->bp
[page
].hijacked
) {
86 /* somebody beat us to getting the page */
91 /* no page was in place and we have one, so install it */
93 bitmap
->bp
[page
].map
= mappage
;
94 bitmap
->missing_pages
--;
99 /* if page is completely empty, put it back on the free list, or dealloc it */
100 /* if page was hijacked, unmark the flag so it might get alloced next time */
101 /* Note: lock should be held when calling this */
102 static void bitmap_checkfree(struct bitmap_counts
*bitmap
, unsigned long page
)
106 if (bitmap
->bp
[page
].count
) /* page is still busy */
109 /* page is no longer in use, it can be released */
111 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
112 bitmap
->bp
[page
].hijacked
= 0;
113 bitmap
->bp
[page
].map
= NULL
;
115 /* normal case, free the page */
116 ptr
= bitmap
->bp
[page
].map
;
117 bitmap
->bp
[page
].map
= NULL
;
118 bitmap
->missing_pages
++;
124 * bitmap file handling - read and write the bitmap file and its superblock
128 * basic page I/O operations
131 /* IO operations when bitmap is stored near all superblocks */
132 static int read_sb_page(struct mddev
*mddev
, loff_t offset
,
134 unsigned long index
, int size
)
136 /* choose a good rdev and read the page from there */
138 struct md_rdev
*rdev
;
141 rdev_for_each(rdev
, mddev
) {
142 if (! test_bit(In_sync
, &rdev
->flags
)
143 || test_bit(Faulty
, &rdev
->flags
))
146 target
= offset
+ index
* (PAGE_SIZE
/512);
148 if (sync_page_io(rdev
, target
,
149 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
158 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
160 /* Iterate the disks of an mddev, using rcu to protect access to the
161 * linked list, and raising the refcount of devices we return to ensure
162 * they don't disappear while in use.
163 * As devices are only added or removed when raid_disk is < 0 and
164 * nr_pending is 0 and In_sync is clear, the entries we return will
165 * still be in the same position on the list when we re-enter
166 * list_for_each_entry_continue_rcu.
170 /* start at the beginning */
171 rdev
= list_entry_rcu(&mddev
->disks
, struct md_rdev
, same_set
);
173 /* release the previous rdev and start from there. */
174 rdev_dec_pending(rdev
, mddev
);
176 list_for_each_entry_continue_rcu(rdev
, &mddev
->disks
, same_set
) {
177 if (rdev
->raid_disk
>= 0 &&
178 !test_bit(Faulty
, &rdev
->flags
)) {
179 /* this is a usable devices */
180 atomic_inc(&rdev
->nr_pending
);
189 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
191 struct md_rdev
*rdev
= NULL
;
192 struct block_device
*bdev
;
193 struct mddev
*mddev
= bitmap
->mddev
;
194 struct bitmap_storage
*store
= &bitmap
->storage
;
196 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
197 int size
= PAGE_SIZE
;
198 loff_t offset
= mddev
->bitmap_info
.offset
;
200 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
202 if (page
->index
== store
->file_pages
-1) {
203 int last_page_size
= store
->bytes
& (PAGE_SIZE
-1);
204 if (last_page_size
== 0)
205 last_page_size
= PAGE_SIZE
;
206 size
= roundup(last_page_size
,
207 bdev_logical_block_size(bdev
));
209 /* Just make sure we aren't corrupting data or
212 if (mddev
->external
) {
213 /* Bitmap could be anywhere. */
214 if (rdev
->sb_start
+ offset
+ (page
->index
218 rdev
->sb_start
+ offset
219 < (rdev
->data_offset
+ mddev
->dev_sectors
222 } else if (offset
< 0) {
223 /* DATA BITMAP METADATA */
225 + (long)(page
->index
* (PAGE_SIZE
/512))
227 /* bitmap runs in to metadata */
229 if (rdev
->data_offset
+ mddev
->dev_sectors
230 > rdev
->sb_start
+ offset
)
231 /* data runs in to bitmap */
233 } else if (rdev
->sb_start
< rdev
->data_offset
) {
234 /* METADATA BITMAP DATA */
237 + page
->index
*(PAGE_SIZE
/512) + size
/512
239 /* bitmap runs in to data */
242 /* DATA METADATA BITMAP - no problems */
244 md_super_write(mddev
, rdev
,
245 rdev
->sb_start
+ offset
246 + page
->index
* (PAGE_SIZE
/512),
252 md_super_wait(mddev
);
259 static void bitmap_file_kick(struct bitmap
*bitmap
);
261 * write out a page to a file
263 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
265 struct buffer_head
*bh
;
267 if (bitmap
->storage
.file
== NULL
) {
268 switch (write_sb_page(bitmap
, page
, wait
)) {
270 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
274 bh
= page_buffers(page
);
276 while (bh
&& bh
->b_blocknr
) {
277 atomic_inc(&bitmap
->pending_writes
);
278 set_buffer_locked(bh
);
279 set_buffer_mapped(bh
);
280 submit_bh(WRITE
| REQ_SYNC
, bh
);
281 bh
= bh
->b_this_page
;
285 wait_event(bitmap
->write_wait
,
286 atomic_read(&bitmap
->pending_writes
)==0);
288 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
289 bitmap_file_kick(bitmap
);
292 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
294 struct bitmap
*bitmap
= bh
->b_private
;
297 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
298 if (atomic_dec_and_test(&bitmap
->pending_writes
))
299 wake_up(&bitmap
->write_wait
);
302 /* copied from buffer.c */
304 __clear_page_buffers(struct page
*page
)
306 ClearPagePrivate(page
);
307 set_page_private(page
, 0);
308 page_cache_release(page
);
310 static void free_buffers(struct page
*page
)
312 struct buffer_head
*bh
;
314 if (!PagePrivate(page
))
317 bh
= page_buffers(page
);
319 struct buffer_head
*next
= bh
->b_this_page
;
320 free_buffer_head(bh
);
323 __clear_page_buffers(page
);
327 /* read a page from a file.
328 * We both read the page, and attach buffers to the page to record the
329 * address of each block (using bmap). These addresses will be used
330 * to write the block later, completely bypassing the filesystem.
331 * This usage is similar to how swap files are handled, and allows us
332 * to write to a file with no concerns of memory allocation failing.
334 static int read_page(struct file
*file
, unsigned long index
,
335 struct bitmap
*bitmap
,
340 struct inode
*inode
= file_inode(file
);
341 struct buffer_head
*bh
;
344 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
345 (unsigned long long)index
<< PAGE_SHIFT
);
347 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
352 attach_page_buffers(page
, bh
);
353 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
358 bh
->b_blocknr
= bmap(inode
, block
);
359 if (bh
->b_blocknr
== 0) {
360 /* Cannot use this file! */
364 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
365 if (count
< (1<<inode
->i_blkbits
))
368 count
-= (1<<inode
->i_blkbits
);
370 bh
->b_end_io
= end_bitmap_write
;
371 bh
->b_private
= bitmap
;
372 atomic_inc(&bitmap
->pending_writes
);
373 set_buffer_locked(bh
);
374 set_buffer_mapped(bh
);
378 bh
= bh
->b_this_page
;
382 wait_event(bitmap
->write_wait
,
383 atomic_read(&bitmap
->pending_writes
)==0);
384 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
388 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %d\n",
390 (unsigned long long)index
<< PAGE_SHIFT
,
396 * bitmap file superblock operations
399 /* update the event counter and sync the superblock to disk */
400 void bitmap_update_sb(struct bitmap
*bitmap
)
404 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
406 if (bitmap
->mddev
->bitmap_info
.external
)
408 if (!bitmap
->storage
.sb_page
) /* no superblock */
410 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
411 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
412 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
413 /* rocking back to read-only */
414 bitmap
->events_cleared
= bitmap
->mddev
->events
;
415 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
416 sb
->state
= cpu_to_le32(bitmap
->flags
);
417 /* Just in case these have been changed via sysfs: */
418 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
419 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
420 /* This might have been changed by a reshape */
421 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
422 sb
->chunksize
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.chunksize
);
423 sb
->sectors_reserved
= cpu_to_le32(bitmap
->mddev
->
426 write_page(bitmap
, bitmap
->storage
.sb_page
, 1);
429 /* print out the bitmap file superblock */
430 void bitmap_print_sb(struct bitmap
*bitmap
)
434 if (!bitmap
|| !bitmap
->storage
.sb_page
)
436 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
437 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
438 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
439 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
440 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
441 *(__u32
*)(sb
->uuid
+0),
442 *(__u32
*)(sb
->uuid
+4),
443 *(__u32
*)(sb
->uuid
+8),
444 *(__u32
*)(sb
->uuid
+12));
445 printk(KERN_DEBUG
" events: %llu\n",
446 (unsigned long long) le64_to_cpu(sb
->events
));
447 printk(KERN_DEBUG
"events cleared: %llu\n",
448 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
449 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
450 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
451 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
452 printk(KERN_DEBUG
" sync size: %llu KB\n",
453 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
454 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
462 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
463 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
464 * This function verifies 'bitmap_info' and populates the on-disk bitmap
465 * structure, which is to be written to disk.
467 * Returns: 0 on success, -Exxx on error
469 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
472 unsigned long chunksize
, daemon_sleep
, write_behind
;
474 bitmap
->storage
.sb_page
= alloc_page(GFP_KERNEL
);
475 if (bitmap
->storage
.sb_page
== NULL
)
477 bitmap
->storage
.sb_page
->index
= 0;
479 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
481 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
482 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
484 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
486 if (!is_power_of_2(chunksize
)) {
488 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
491 sb
->chunksize
= cpu_to_le32(chunksize
);
493 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
495 (daemon_sleep
< 1) || (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
496 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
497 daemon_sleep
= 5 * HZ
;
499 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
500 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
503 * FIXME: write_behind for RAID1. If not specified, what
504 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
506 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
507 if (write_behind
> COUNTER_MAX
)
508 write_behind
= COUNTER_MAX
/ 2;
509 sb
->write_behind
= cpu_to_le32(write_behind
);
510 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
512 /* keep the array size field of the bitmap superblock up to date */
513 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
515 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
517 set_bit(BITMAP_STALE
, &bitmap
->flags
);
518 sb
->state
= cpu_to_le32(bitmap
->flags
);
519 bitmap
->events_cleared
= bitmap
->mddev
->events
;
520 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
527 /* read the superblock from the bitmap file and initialize some bitmap fields */
528 static int bitmap_read_sb(struct bitmap
*bitmap
)
532 unsigned long chunksize
, daemon_sleep
, write_behind
;
533 unsigned long long events
;
534 unsigned long sectors_reserved
= 0;
536 struct page
*sb_page
;
538 if (!bitmap
->storage
.file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
539 chunksize
= 128 * 1024 * 1024;
540 daemon_sleep
= 5 * HZ
;
542 set_bit(BITMAP_STALE
, &bitmap
->flags
);
546 /* page 0 is the superblock, read it... */
547 sb_page
= alloc_page(GFP_KERNEL
);
550 bitmap
->storage
.sb_page
= sb_page
;
552 if (bitmap
->storage
.file
) {
553 loff_t isize
= i_size_read(bitmap
->storage
.file
->f_mapping
->host
);
554 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
556 err
= read_page(bitmap
->storage
.file
, 0,
557 bitmap
, bytes
, sb_page
);
559 err
= read_sb_page(bitmap
->mddev
,
560 bitmap
->mddev
->bitmap_info
.offset
,
562 0, sizeof(bitmap_super_t
));
567 sb
= kmap_atomic(sb_page
);
569 chunksize
= le32_to_cpu(sb
->chunksize
);
570 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
571 write_behind
= le32_to_cpu(sb
->write_behind
);
572 sectors_reserved
= le32_to_cpu(sb
->sectors_reserved
);
574 /* verify that the bitmap-specific fields are valid */
575 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
576 reason
= "bad magic";
577 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
578 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
579 reason
= "unrecognized superblock version";
580 else if (chunksize
< 512)
581 reason
= "bitmap chunksize too small";
582 else if (!is_power_of_2(chunksize
))
583 reason
= "bitmap chunksize not a power of 2";
584 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
585 reason
= "daemon sleep period out of range";
586 else if (write_behind
> COUNTER_MAX
)
587 reason
= "write-behind limit out of range (0 - 16383)";
589 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
590 bmname(bitmap
), reason
);
594 /* keep the array size field of the bitmap superblock up to date */
595 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
597 if (bitmap
->mddev
->persistent
) {
599 * We have a persistent array superblock, so compare the
600 * bitmap's UUID and event counter to the mddev's
602 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
604 "%s: bitmap superblock UUID mismatch\n",
608 events
= le64_to_cpu(sb
->events
);
609 if (events
< bitmap
->mddev
->events
) {
611 "%s: bitmap file is out of date (%llu < %llu) "
612 "-- forcing full recovery\n",
613 bmname(bitmap
), events
,
614 (unsigned long long) bitmap
->mddev
->events
);
615 set_bit(BITMAP_STALE
, &bitmap
->flags
);
619 /* assign fields using values from superblock */
620 bitmap
->flags
|= le32_to_cpu(sb
->state
);
621 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
622 set_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
);
623 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
628 if (test_bit(BITMAP_STALE
, &bitmap
->flags
))
629 bitmap
->events_cleared
= bitmap
->mddev
->events
;
630 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
631 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
632 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
633 if (bitmap
->mddev
->bitmap_info
.space
== 0 ||
634 bitmap
->mddev
->bitmap_info
.space
> sectors_reserved
)
635 bitmap
->mddev
->bitmap_info
.space
= sectors_reserved
;
637 bitmap_print_sb(bitmap
);
642 * general bitmap file operations
648 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
649 * file a page at a time. There's a superblock at the start of the file.
651 /* calculate the index of the page that contains this bit */
652 static inline unsigned long file_page_index(struct bitmap_storage
*store
,
656 chunk
+= sizeof(bitmap_super_t
) << 3;
657 return chunk
>> PAGE_BIT_SHIFT
;
660 /* calculate the (bit) offset of this bit within a page */
661 static inline unsigned long file_page_offset(struct bitmap_storage
*store
,
665 chunk
+= sizeof(bitmap_super_t
) << 3;
666 return chunk
& (PAGE_BITS
- 1);
670 * return a pointer to the page in the filemap that contains the given bit
673 static inline struct page
*filemap_get_page(struct bitmap_storage
*store
,
676 if (file_page_index(store
, chunk
) >= store
->file_pages
)
678 return store
->filemap
[file_page_index(store
, chunk
)];
681 static int bitmap_storage_alloc(struct bitmap_storage
*store
,
682 unsigned long chunks
, int with_super
)
685 unsigned long num_pages
;
688 bytes
= DIV_ROUND_UP(chunks
, 8);
690 bytes
+= sizeof(bitmap_super_t
);
692 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
694 store
->filemap
= kmalloc(sizeof(struct page
*)
695 * num_pages
, GFP_KERNEL
);
699 if (with_super
&& !store
->sb_page
) {
700 store
->sb_page
= alloc_page(GFP_KERNEL
|__GFP_ZERO
);
701 if (store
->sb_page
== NULL
)
703 store
->sb_page
->index
= 0;
706 if (store
->sb_page
) {
707 store
->filemap
[0] = store
->sb_page
;
710 for ( ; pnum
< num_pages
; pnum
++) {
711 store
->filemap
[pnum
] = alloc_page(GFP_KERNEL
|__GFP_ZERO
);
712 if (!store
->filemap
[pnum
]) {
713 store
->file_pages
= pnum
;
716 store
->filemap
[pnum
]->index
= pnum
;
718 store
->file_pages
= pnum
;
720 /* We need 4 bits per page, rounded up to a multiple
721 * of sizeof(unsigned long) */
722 store
->filemap_attr
= kzalloc(
723 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
725 if (!store
->filemap_attr
)
728 store
->bytes
= bytes
;
733 static void bitmap_file_unmap(struct bitmap_storage
*store
)
735 struct page
**map
, *sb_page
;
740 map
= store
->filemap
;
741 pages
= store
->file_pages
;
742 sb_page
= store
->sb_page
;
745 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
746 free_buffers(map
[pages
]);
748 kfree(store
->filemap_attr
);
751 free_buffers(sb_page
);
754 struct inode
*inode
= file_inode(file
);
755 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
761 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
762 * then it is no longer reliable, so we stop using it and we mark the file
763 * as failed in the superblock
765 static void bitmap_file_kick(struct bitmap
*bitmap
)
767 char *path
, *ptr
= NULL
;
769 if (!test_and_set_bit(BITMAP_STALE
, &bitmap
->flags
)) {
770 bitmap_update_sb(bitmap
);
772 if (bitmap
->storage
.file
) {
773 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
775 ptr
= d_path(&bitmap
->storage
.file
->f_path
,
779 "%s: kicking failed bitmap file %s from array!\n",
780 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
785 "%s: disabling internal bitmap due to errors\n",
790 enum bitmap_page_attr
{
791 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
792 BITMAP_PAGE_PENDING
= 1, /* there are bits that are being cleaned.
793 * i.e. counter is 1 or 2. */
794 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
797 static inline void set_page_attr(struct bitmap
*bitmap
, int pnum
,
798 enum bitmap_page_attr attr
)
800 set_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
803 static inline void clear_page_attr(struct bitmap
*bitmap
, int pnum
,
804 enum bitmap_page_attr attr
)
806 clear_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
809 static inline int test_page_attr(struct bitmap
*bitmap
, int pnum
,
810 enum bitmap_page_attr attr
)
812 return test_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
815 static inline int test_and_clear_page_attr(struct bitmap
*bitmap
, int pnum
,
816 enum bitmap_page_attr attr
)
818 return test_and_clear_bit((pnum
<<2) + attr
,
819 bitmap
->storage
.filemap_attr
);
822 * bitmap_file_set_bit -- called before performing a write to the md device
823 * to set (and eventually sync) a particular bit in the bitmap file
825 * we set the bit immediately, then we record the page number so that
826 * when an unplug occurs, we can flush the dirty pages out to disk
828 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
833 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
835 page
= filemap_get_page(&bitmap
->storage
, chunk
);
838 bit
= file_page_offset(&bitmap
->storage
, chunk
);
841 kaddr
= kmap_atomic(page
);
842 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
845 set_bit_le(bit
, kaddr
);
846 kunmap_atomic(kaddr
);
847 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
848 /* record page number so it gets flushed to disk when unplug occurs */
849 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_DIRTY
);
852 static void bitmap_file_clear_bit(struct bitmap
*bitmap
, sector_t block
)
857 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
859 page
= filemap_get_page(&bitmap
->storage
, chunk
);
862 bit
= file_page_offset(&bitmap
->storage
, chunk
);
863 paddr
= kmap_atomic(page
);
864 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
865 clear_bit(bit
, paddr
);
867 clear_bit_le(bit
, paddr
);
868 kunmap_atomic(paddr
);
869 if (!test_page_attr(bitmap
, page
->index
, BITMAP_PAGE_NEEDWRITE
)) {
870 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_PENDING
);
871 bitmap
->allclean
= 0;
875 /* this gets called when the md device is ready to unplug its underlying
876 * (slave) device queues -- before we let any writes go down, we need to
877 * sync the dirty pages of the bitmap file to disk */
878 void bitmap_unplug(struct bitmap
*bitmap
)
881 int dirty
, need_write
;
883 if (!bitmap
|| !bitmap
->storage
.filemap
||
884 test_bit(BITMAP_STALE
, &bitmap
->flags
))
887 /* look at each page to see if there are any set bits that need to be
888 * flushed out to disk */
889 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++) {
890 if (!bitmap
->storage
.filemap
)
892 dirty
= test_and_clear_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
893 need_write
= test_and_clear_page_attr(bitmap
, i
,
894 BITMAP_PAGE_NEEDWRITE
);
895 if (dirty
|| need_write
) {
896 clear_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
);
897 write_page(bitmap
, bitmap
->storage
.filemap
[i
], 0);
900 if (bitmap
->storage
.file
)
901 wait_event(bitmap
->write_wait
,
902 atomic_read(&bitmap
->pending_writes
)==0);
904 md_super_wait(bitmap
->mddev
);
906 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
907 bitmap_file_kick(bitmap
);
909 EXPORT_SYMBOL(bitmap_unplug
);
911 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
912 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
913 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
914 * memory mapping of the bitmap file
916 * if there's no bitmap file, or if the bitmap file had been
917 * previously kicked from the array, we mark all the bits as
918 * 1's in order to cause a full resync.
920 * We ignore all bits for sectors that end earlier than 'start'.
921 * This is used when reading an out-of-date bitmap...
923 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
925 unsigned long i
, chunks
, index
, oldindex
, bit
;
926 struct page
*page
= NULL
;
927 unsigned long bit_cnt
= 0;
929 unsigned long offset
;
933 struct bitmap_storage
*store
= &bitmap
->storage
;
935 chunks
= bitmap
->counts
.chunks
;
938 if (!file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
939 /* No permanent bitmap - fill with '1s'. */
940 store
->filemap
= NULL
;
941 store
->file_pages
= 0;
942 for (i
= 0; i
< chunks
; i
++) {
943 /* if the disk bit is set, set the memory bit */
944 int needed
= ((sector_t
)(i
+1) << (bitmap
->counts
.chunkshift
)
946 bitmap_set_memory_bits(bitmap
,
947 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
953 outofdate
= test_bit(BITMAP_STALE
, &bitmap
->flags
);
955 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
956 "recovery\n", bmname(bitmap
));
958 if (file
&& i_size_read(file
->f_mapping
->host
) < store
->bytes
) {
959 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
961 (unsigned long) i_size_read(file
->f_mapping
->host
),
968 if (!bitmap
->mddev
->bitmap_info
.external
)
969 offset
= sizeof(bitmap_super_t
);
971 for (i
= 0; i
< chunks
; i
++) {
973 index
= file_page_index(&bitmap
->storage
, i
);
974 bit
= file_page_offset(&bitmap
->storage
, i
);
975 if (index
!= oldindex
) { /* this is a new page, read it in */
977 /* unmap the old page, we're done with it */
978 if (index
== store
->file_pages
-1)
979 count
= store
->bytes
- index
* PAGE_SIZE
;
982 page
= store
->filemap
[index
];
984 ret
= read_page(file
, index
, bitmap
,
989 bitmap
->mddev
->bitmap_info
.offset
,
1000 * if bitmap is out of date, dirty the
1001 * whole page and write it out
1003 paddr
= kmap_atomic(page
);
1004 memset(paddr
+ offset
, 0xff,
1005 PAGE_SIZE
- offset
);
1006 kunmap_atomic(paddr
);
1007 write_page(bitmap
, page
, 1);
1010 if (test_bit(BITMAP_WRITE_ERROR
,
1015 paddr
= kmap_atomic(page
);
1016 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
1017 b
= test_bit(bit
, paddr
);
1019 b
= test_bit_le(bit
, paddr
);
1020 kunmap_atomic(paddr
);
1022 /* if the disk bit is set, set the memory bit */
1023 int needed
= ((sector_t
)(i
+1) << bitmap
->counts
.chunkshift
1025 bitmap_set_memory_bits(bitmap
,
1026 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1033 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1034 "read %lu pages, set %lu of %lu bits\n",
1035 bmname(bitmap
), store
->file_pages
,
1041 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1042 bmname(bitmap
), ret
);
1046 void bitmap_write_all(struct bitmap
*bitmap
)
1048 /* We don't actually write all bitmap blocks here,
1049 * just flag them as needing to be written
1053 if (!bitmap
|| !bitmap
->storage
.filemap
)
1055 if (bitmap
->storage
.file
)
1056 /* Only one copy, so nothing needed */
1059 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1060 set_page_attr(bitmap
, i
,
1061 BITMAP_PAGE_NEEDWRITE
);
1062 bitmap
->allclean
= 0;
1065 static void bitmap_count_page(struct bitmap_counts
*bitmap
,
1066 sector_t offset
, int inc
)
1068 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1069 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1070 bitmap
->bp
[page
].count
+= inc
;
1071 bitmap_checkfree(bitmap
, page
);
1074 static void bitmap_set_pending(struct bitmap_counts
*bitmap
, sector_t offset
)
1076 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1077 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1078 struct bitmap_page
*bp
= &bitmap
->bp
[page
];
1084 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1085 sector_t offset
, sector_t
*blocks
,
1089 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1093 void bitmap_daemon_work(struct mddev
*mddev
)
1095 struct bitmap
*bitmap
;
1097 unsigned long nextpage
;
1099 struct bitmap_counts
*counts
;
1101 /* Use a mutex to guard daemon_work against
1104 mutex_lock(&mddev
->bitmap_info
.mutex
);
1105 bitmap
= mddev
->bitmap
;
1106 if (bitmap
== NULL
) {
1107 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1110 if (time_before(jiffies
, bitmap
->daemon_lastrun
1111 + mddev
->bitmap_info
.daemon_sleep
))
1114 bitmap
->daemon_lastrun
= jiffies
;
1115 if (bitmap
->allclean
) {
1116 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1119 bitmap
->allclean
= 1;
1121 /* Any file-page which is PENDING now needs to be written.
1122 * So set NEEDWRITE now, then after we make any last-minute changes
1125 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++)
1126 if (test_and_clear_page_attr(bitmap
, j
,
1127 BITMAP_PAGE_PENDING
))
1128 set_page_attr(bitmap
, j
,
1129 BITMAP_PAGE_NEEDWRITE
);
1131 if (bitmap
->need_sync
&&
1132 mddev
->bitmap_info
.external
== 0) {
1133 /* Arrange for superblock update as well as
1136 bitmap
->need_sync
= 0;
1137 if (bitmap
->storage
.filemap
) {
1138 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
1139 sb
->events_cleared
=
1140 cpu_to_le64(bitmap
->events_cleared
);
1142 set_page_attr(bitmap
, 0,
1143 BITMAP_PAGE_NEEDWRITE
);
1146 /* Now look at the bitmap counters and if any are '2' or '1',
1147 * decrement and handle accordingly.
1149 counts
= &bitmap
->counts
;
1150 spin_lock_irq(&counts
->lock
);
1152 for (j
= 0; j
< counts
->chunks
; j
++) {
1153 bitmap_counter_t
*bmc
;
1154 sector_t block
= (sector_t
)j
<< counts
->chunkshift
;
1156 if (j
== nextpage
) {
1157 nextpage
+= PAGE_COUNTER_RATIO
;
1158 if (!counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
) {
1159 j
|= PAGE_COUNTER_MASK
;
1162 counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
= 0;
1164 bmc
= bitmap_get_counter(counts
,
1169 j
|= PAGE_COUNTER_MASK
;
1172 if (*bmc
== 1 && !bitmap
->need_sync
) {
1173 /* We can clear the bit */
1175 bitmap_count_page(counts
, block
, -1);
1176 bitmap_file_clear_bit(bitmap
, block
);
1177 } else if (*bmc
&& *bmc
<= 2) {
1179 bitmap_set_pending(counts
, block
);
1180 bitmap
->allclean
= 0;
1183 spin_unlock_irq(&counts
->lock
);
1185 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1186 * DIRTY pages need to be written by bitmap_unplug so it can wait
1188 * If we find any DIRTY page we stop there and let bitmap_unplug
1189 * handle all the rest. This is important in the case where
1190 * the first blocking holds the superblock and it has been updated.
1191 * We mustn't write any other blocks before the superblock.
1194 j
< bitmap
->storage
.file_pages
1195 && !test_bit(BITMAP_STALE
, &bitmap
->flags
);
1198 if (test_page_attr(bitmap
, j
,
1200 /* bitmap_unplug will handle the rest */
1202 if (test_and_clear_page_attr(bitmap
, j
,
1203 BITMAP_PAGE_NEEDWRITE
)) {
1204 write_page(bitmap
, bitmap
->storage
.filemap
[j
], 0);
1209 if (bitmap
->allclean
== 0)
1210 mddev
->thread
->timeout
=
1211 mddev
->bitmap_info
.daemon_sleep
;
1212 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1215 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1216 sector_t offset
, sector_t
*blocks
,
1218 __releases(bitmap
->lock
)
1219 __acquires(bitmap
->lock
)
1221 /* If 'create', we might release the lock and reclaim it.
1222 * The lock must have been taken with interrupts enabled.
1223 * If !create, we don't release the lock.
1225 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1226 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1227 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1231 err
= bitmap_checkpage(bitmap
, page
, create
);
1233 if (bitmap
->bp
[page
].hijacked
||
1234 bitmap
->bp
[page
].map
== NULL
)
1235 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1236 PAGE_COUNTER_SHIFT
- 1);
1238 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1239 *blocks
= csize
- (offset
& (csize
- 1));
1244 /* now locked ... */
1246 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1247 /* should we use the first or second counter field
1248 * of the hijacked pointer? */
1249 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1250 return &((bitmap_counter_t
*)
1251 &bitmap
->bp
[page
].map
)[hi
];
1252 } else /* page is allocated */
1253 return (bitmap_counter_t
*)
1254 &(bitmap
->bp
[page
].map
[pageoff
]);
1257 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1264 atomic_inc(&bitmap
->behind_writes
);
1265 bw
= atomic_read(&bitmap
->behind_writes
);
1266 if (bw
> bitmap
->behind_writes_used
)
1267 bitmap
->behind_writes_used
= bw
;
1269 pr_debug("inc write-behind count %d/%lu\n",
1270 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1275 bitmap_counter_t
*bmc
;
1277 spin_lock_irq(&bitmap
->counts
.lock
);
1278 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 1);
1280 spin_unlock_irq(&bitmap
->counts
.lock
);
1284 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1285 DEFINE_WAIT(__wait
);
1286 /* note that it is safe to do the prepare_to_wait
1287 * after the test as long as we do it before dropping
1290 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1291 TASK_UNINTERRUPTIBLE
);
1292 spin_unlock_irq(&bitmap
->counts
.lock
);
1294 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1300 bitmap_file_set_bit(bitmap
, offset
);
1301 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1309 spin_unlock_irq(&bitmap
->counts
.lock
);
1312 if (sectors
> blocks
)
1319 EXPORT_SYMBOL(bitmap_startwrite
);
1321 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1322 int success
, int behind
)
1327 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1328 wake_up(&bitmap
->behind_wait
);
1329 pr_debug("dec write-behind count %d/%lu\n",
1330 atomic_read(&bitmap
->behind_writes
),
1331 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1336 unsigned long flags
;
1337 bitmap_counter_t
*bmc
;
1339 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1340 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 0);
1342 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1346 if (success
&& !bitmap
->mddev
->degraded
&&
1347 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1348 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1349 bitmap
->need_sync
= 1;
1350 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1353 if (!success
&& !NEEDED(*bmc
))
1354 *bmc
|= NEEDED_MASK
;
1356 if (COUNTER(*bmc
) == COUNTER_MAX
)
1357 wake_up(&bitmap
->overflow_wait
);
1361 bitmap_set_pending(&bitmap
->counts
, offset
);
1362 bitmap
->allclean
= 0;
1364 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1366 if (sectors
> blocks
)
1372 EXPORT_SYMBOL(bitmap_endwrite
);
1374 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1377 bitmap_counter_t
*bmc
;
1379 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1381 return 1; /* always resync if no bitmap */
1383 spin_lock_irq(&bitmap
->counts
.lock
);
1384 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1390 else if (NEEDED(*bmc
)) {
1392 if (!degraded
) { /* don't set/clear bits if degraded */
1393 *bmc
|= RESYNC_MASK
;
1394 *bmc
&= ~NEEDED_MASK
;
1398 spin_unlock_irq(&bitmap
->counts
.lock
);
1402 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1405 /* bitmap_start_sync must always report on multiples of whole
1406 * pages, otherwise resync (which is very PAGE_SIZE based) will
1408 * So call __bitmap_start_sync repeatedly (if needed) until
1409 * At least PAGE_SIZE>>9 blocks are covered.
1410 * Return the 'or' of the result.
1416 while (*blocks
< (PAGE_SIZE
>>9)) {
1417 rv
|= __bitmap_start_sync(bitmap
, offset
,
1418 &blocks1
, degraded
);
1424 EXPORT_SYMBOL(bitmap_start_sync
);
1426 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1428 bitmap_counter_t
*bmc
;
1429 unsigned long flags
;
1431 if (bitmap
== NULL
) {
1435 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1436 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1441 *bmc
&= ~RESYNC_MASK
;
1443 if (!NEEDED(*bmc
) && aborted
)
1444 *bmc
|= NEEDED_MASK
;
1447 bitmap_set_pending(&bitmap
->counts
, offset
);
1448 bitmap
->allclean
= 0;
1453 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1455 EXPORT_SYMBOL(bitmap_end_sync
);
1457 void bitmap_close_sync(struct bitmap
*bitmap
)
1459 /* Sync has finished, and any bitmap chunks that weren't synced
1460 * properly have been aborted. It remains to us to clear the
1461 * RESYNC bit wherever it is still on
1463 sector_t sector
= 0;
1467 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1468 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1472 EXPORT_SYMBOL(bitmap_close_sync
);
1474 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
)
1482 bitmap
->last_end_sync
= jiffies
;
1485 if (time_before(jiffies
, (bitmap
->last_end_sync
1486 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1488 wait_event(bitmap
->mddev
->recovery_wait
,
1489 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1491 bitmap
->mddev
->curr_resync_completed
= sector
;
1492 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1493 sector
&= ~((1ULL << bitmap
->counts
.chunkshift
) - 1);
1495 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1496 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1499 bitmap
->last_end_sync
= jiffies
;
1500 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1502 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1504 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1506 /* For each chunk covered by any of these sectors, set the
1507 * counter to 2 and possibly set resync_needed. They should all
1508 * be 0 at this point
1512 bitmap_counter_t
*bmc
;
1513 spin_lock_irq(&bitmap
->counts
.lock
);
1514 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &secs
, 1);
1516 spin_unlock_irq(&bitmap
->counts
.lock
);
1520 *bmc
= 2 | (needed
? NEEDED_MASK
: 0);
1521 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1522 bitmap_set_pending(&bitmap
->counts
, offset
);
1523 bitmap
->allclean
= 0;
1525 spin_unlock_irq(&bitmap
->counts
.lock
);
1528 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1529 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1531 unsigned long chunk
;
1533 for (chunk
= s
; chunk
<= e
; chunk
++) {
1534 sector_t sec
= (sector_t
)chunk
<< bitmap
->counts
.chunkshift
;
1535 bitmap_set_memory_bits(bitmap
, sec
, 1);
1536 bitmap_file_set_bit(bitmap
, sec
);
1537 if (sec
< bitmap
->mddev
->recovery_cp
)
1538 /* We are asserting that the array is dirty,
1539 * so move the recovery_cp address back so
1540 * that it is obvious that it is dirty
1542 bitmap
->mddev
->recovery_cp
= sec
;
1547 * flush out any pending updates
1549 void bitmap_flush(struct mddev
*mddev
)
1551 struct bitmap
*bitmap
= mddev
->bitmap
;
1554 if (!bitmap
) /* there was no bitmap */
1557 /* run the daemon_work three time to ensure everything is flushed
1560 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1561 bitmap
->daemon_lastrun
-= sleep
;
1562 bitmap_daemon_work(mddev
);
1563 bitmap
->daemon_lastrun
-= sleep
;
1564 bitmap_daemon_work(mddev
);
1565 bitmap
->daemon_lastrun
-= sleep
;
1566 bitmap_daemon_work(mddev
);
1567 bitmap_update_sb(bitmap
);
1571 * free memory that was allocated
1573 static void bitmap_free(struct bitmap
*bitmap
)
1575 unsigned long k
, pages
;
1576 struct bitmap_page
*bp
;
1578 if (!bitmap
) /* there was no bitmap */
1581 /* Shouldn't be needed - but just in case.... */
1582 wait_event(bitmap
->write_wait
,
1583 atomic_read(&bitmap
->pending_writes
) == 0);
1585 /* release the bitmap file */
1586 bitmap_file_unmap(&bitmap
->storage
);
1588 bp
= bitmap
->counts
.bp
;
1589 pages
= bitmap
->counts
.pages
;
1591 /* free all allocated memory */
1593 if (bp
) /* deallocate the page memory */
1594 for (k
= 0; k
< pages
; k
++)
1595 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1601 void bitmap_destroy(struct mddev
*mddev
)
1603 struct bitmap
*bitmap
= mddev
->bitmap
;
1605 if (!bitmap
) /* there was no bitmap */
1608 mutex_lock(&mddev
->bitmap_info
.mutex
);
1609 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1610 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1612 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1614 if (bitmap
->sysfs_can_clear
)
1615 sysfs_put(bitmap
->sysfs_can_clear
);
1617 bitmap_free(bitmap
);
1621 * initialize the bitmap structure
1622 * if this returns an error, bitmap_destroy must be called to do clean up
1624 int bitmap_create(struct mddev
*mddev
)
1626 struct bitmap
*bitmap
;
1627 sector_t blocks
= mddev
->resync_max_sectors
;
1628 struct file
*file
= mddev
->bitmap_info
.file
;
1630 struct kernfs_node
*bm
= NULL
;
1632 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1634 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1636 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1640 spin_lock_init(&bitmap
->counts
.lock
);
1641 atomic_set(&bitmap
->pending_writes
, 0);
1642 init_waitqueue_head(&bitmap
->write_wait
);
1643 init_waitqueue_head(&bitmap
->overflow_wait
);
1644 init_waitqueue_head(&bitmap
->behind_wait
);
1646 bitmap
->mddev
= mddev
;
1649 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, "bitmap");
1651 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, "can_clear");
1654 bitmap
->sysfs_can_clear
= NULL
;
1656 bitmap
->storage
.file
= file
;
1659 /* As future accesses to this file will use bmap,
1660 * and bypass the page cache, we must sync the file
1665 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1666 if (!mddev
->bitmap_info
.external
) {
1668 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1669 * instructing us to create a new on-disk bitmap instance.
1671 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1672 err
= bitmap_new_disk_sb(bitmap
);
1674 err
= bitmap_read_sb(bitmap
);
1677 if (mddev
->bitmap_info
.chunksize
== 0 ||
1678 mddev
->bitmap_info
.daemon_sleep
== 0)
1679 /* chunksize and time_base need to be
1686 bitmap
->daemon_lastrun
= jiffies
;
1687 err
= bitmap_resize(bitmap
, blocks
, mddev
->bitmap_info
.chunksize
, 1);
1691 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1692 bitmap
->counts
.pages
, bmname(bitmap
));
1694 mddev
->bitmap
= bitmap
;
1695 return test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
) ? -EIO
: 0;
1698 bitmap_free(bitmap
);
1702 int bitmap_load(struct mddev
*mddev
)
1706 sector_t sector
= 0;
1707 struct bitmap
*bitmap
= mddev
->bitmap
;
1712 /* Clear out old bitmap info first: Either there is none, or we
1713 * are resuming after someone else has possibly changed things,
1714 * so we should forget old cached info.
1715 * All chunks should be clean, but some might need_sync.
1717 while (sector
< mddev
->resync_max_sectors
) {
1719 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1722 bitmap_close_sync(bitmap
);
1724 if (mddev
->degraded
== 0
1725 || bitmap
->events_cleared
== mddev
->events
)
1726 /* no need to keep dirty bits to optimise a
1727 * re-add of a missing device */
1728 start
= mddev
->recovery_cp
;
1730 mutex_lock(&mddev
->bitmap_info
.mutex
);
1731 err
= bitmap_init_from_disk(bitmap
, start
);
1732 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1736 clear_bit(BITMAP_STALE
, &bitmap
->flags
);
1738 /* Kick recovery in case any bits were set */
1739 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1741 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1742 md_wakeup_thread(mddev
->thread
);
1744 bitmap_update_sb(bitmap
);
1746 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1751 EXPORT_SYMBOL_GPL(bitmap_load
);
1753 void bitmap_status(struct seq_file
*seq
, struct bitmap
*bitmap
)
1755 unsigned long chunk_kb
;
1756 struct bitmap_counts
*counts
;
1761 counts
= &bitmap
->counts
;
1763 chunk_kb
= bitmap
->mddev
->bitmap_info
.chunksize
>> 10;
1764 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
1766 counts
->pages
- counts
->missing_pages
,
1768 (counts
->pages
- counts
->missing_pages
)
1769 << (PAGE_SHIFT
- 10),
1770 chunk_kb
? chunk_kb
: bitmap
->mddev
->bitmap_info
.chunksize
,
1771 chunk_kb
? "KB" : "B");
1772 if (bitmap
->storage
.file
) {
1773 seq_printf(seq
, ", file: ");
1774 seq_path(seq
, &bitmap
->storage
.file
->f_path
, " \t\n");
1777 seq_printf(seq
, "\n");
1780 int bitmap_resize(struct bitmap
*bitmap
, sector_t blocks
,
1781 int chunksize
, int init
)
1783 /* If chunk_size is 0, choose an appropriate chunk size.
1784 * Then possibly allocate new storage space.
1785 * Then quiesce, copy bits, replace bitmap, and re-start
1787 * This function is called both to set up the initial bitmap
1788 * and to resize the bitmap while the array is active.
1789 * If this happens as a result of the array being resized,
1790 * chunksize will be zero, and we need to choose a suitable
1791 * chunksize, otherwise we use what we are given.
1793 struct bitmap_storage store
;
1794 struct bitmap_counts old_counts
;
1795 unsigned long chunks
;
1797 sector_t old_blocks
, new_blocks
;
1801 struct bitmap_page
*new_bp
;
1803 if (chunksize
== 0) {
1804 /* If there is enough space, leave the chunk size unchanged,
1805 * else increase by factor of two until there is enough space.
1808 long space
= bitmap
->mddev
->bitmap_info
.space
;
1811 /* We don't know how much space there is, so limit
1812 * to current size - in sectors.
1814 bytes
= DIV_ROUND_UP(bitmap
->counts
.chunks
, 8);
1815 if (!bitmap
->mddev
->bitmap_info
.external
)
1816 bytes
+= sizeof(bitmap_super_t
);
1817 space
= DIV_ROUND_UP(bytes
, 512);
1818 bitmap
->mddev
->bitmap_info
.space
= space
;
1820 chunkshift
= bitmap
->counts
.chunkshift
;
1823 /* 'chunkshift' is shift from block size to chunk size */
1825 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
1826 bytes
= DIV_ROUND_UP(chunks
, 8);
1827 if (!bitmap
->mddev
->bitmap_info
.external
)
1828 bytes
+= sizeof(bitmap_super_t
);
1829 } while (bytes
> (space
<< 9));
1831 chunkshift
= ffz(~chunksize
) - BITMAP_BLOCK_SHIFT
;
1833 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
1834 memset(&store
, 0, sizeof(store
));
1835 if (bitmap
->mddev
->bitmap_info
.offset
|| bitmap
->mddev
->bitmap_info
.file
)
1836 ret
= bitmap_storage_alloc(&store
, chunks
,
1837 !bitmap
->mddev
->bitmap_info
.external
);
1841 pages
= DIV_ROUND_UP(chunks
, PAGE_COUNTER_RATIO
);
1843 new_bp
= kzalloc(pages
* sizeof(*new_bp
), GFP_KERNEL
);
1846 bitmap_file_unmap(&store
);
1851 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 1);
1853 store
.file
= bitmap
->storage
.file
;
1854 bitmap
->storage
.file
= NULL
;
1856 if (store
.sb_page
&& bitmap
->storage
.sb_page
)
1857 memcpy(page_address(store
.sb_page
),
1858 page_address(bitmap
->storage
.sb_page
),
1859 sizeof(bitmap_super_t
));
1860 bitmap_file_unmap(&bitmap
->storage
);
1861 bitmap
->storage
= store
;
1863 old_counts
= bitmap
->counts
;
1864 bitmap
->counts
.bp
= new_bp
;
1865 bitmap
->counts
.pages
= pages
;
1866 bitmap
->counts
.missing_pages
= pages
;
1867 bitmap
->counts
.chunkshift
= chunkshift
;
1868 bitmap
->counts
.chunks
= chunks
;
1869 bitmap
->mddev
->bitmap_info
.chunksize
= 1 << (chunkshift
+
1870 BITMAP_BLOCK_SHIFT
);
1872 blocks
= min(old_counts
.chunks
<< old_counts
.chunkshift
,
1873 chunks
<< chunkshift
);
1875 spin_lock_irq(&bitmap
->counts
.lock
);
1876 for (block
= 0; block
< blocks
; ) {
1877 bitmap_counter_t
*bmc_old
, *bmc_new
;
1880 bmc_old
= bitmap_get_counter(&old_counts
, block
,
1882 set
= bmc_old
&& NEEDED(*bmc_old
);
1885 bmc_new
= bitmap_get_counter(&bitmap
->counts
, block
,
1887 if (*bmc_new
== 0) {
1888 /* need to set on-disk bits too. */
1889 sector_t end
= block
+ new_blocks
;
1890 sector_t start
= block
>> chunkshift
;
1891 start
<<= chunkshift
;
1892 while (start
< end
) {
1893 bitmap_file_set_bit(bitmap
, block
);
1894 start
+= 1 << chunkshift
;
1897 bitmap_count_page(&bitmap
->counts
,
1899 bitmap_set_pending(&bitmap
->counts
,
1902 *bmc_new
|= NEEDED_MASK
;
1903 if (new_blocks
< old_blocks
)
1904 old_blocks
= new_blocks
;
1906 block
+= old_blocks
;
1911 while (block
< (chunks
<< chunkshift
)) {
1912 bitmap_counter_t
*bmc
;
1913 bmc
= bitmap_get_counter(&bitmap
->counts
, block
,
1916 /* new space. It needs to be resynced, so
1917 * we set NEEDED_MASK.
1920 *bmc
= NEEDED_MASK
| 2;
1921 bitmap_count_page(&bitmap
->counts
,
1923 bitmap_set_pending(&bitmap
->counts
,
1927 block
+= new_blocks
;
1929 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1930 set_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
1932 spin_unlock_irq(&bitmap
->counts
.lock
);
1935 bitmap_unplug(bitmap
);
1936 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 0);
1942 EXPORT_SYMBOL_GPL(bitmap_resize
);
1945 location_show(struct mddev
*mddev
, char *page
)
1948 if (mddev
->bitmap_info
.file
)
1949 len
= sprintf(page
, "file");
1950 else if (mddev
->bitmap_info
.offset
)
1951 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
1953 len
= sprintf(page
, "none");
1954 len
+= sprintf(page
+len
, "\n");
1959 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1963 if (!mddev
->pers
->quiesce
)
1965 if (mddev
->recovery
|| mddev
->sync_thread
)
1969 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
1970 mddev
->bitmap_info
.offset
) {
1971 /* bitmap already configured. Only option is to clear it */
1972 if (strncmp(buf
, "none", 4) != 0)
1975 mddev
->pers
->quiesce(mddev
, 1);
1976 bitmap_destroy(mddev
);
1977 mddev
->pers
->quiesce(mddev
, 0);
1979 mddev
->bitmap_info
.offset
= 0;
1980 if (mddev
->bitmap_info
.file
) {
1981 struct file
*f
= mddev
->bitmap_info
.file
;
1982 mddev
->bitmap_info
.file
= NULL
;
1986 /* No bitmap, OK to set a location */
1988 if (strncmp(buf
, "none", 4) == 0)
1989 /* nothing to be done */;
1990 else if (strncmp(buf
, "file:", 5) == 0) {
1991 /* Not supported yet */
1996 rv
= kstrtoll(buf
+1, 10, &offset
);
1998 rv
= kstrtoll(buf
, 10, &offset
);
2003 if (mddev
->bitmap_info
.external
== 0 &&
2004 mddev
->major_version
== 0 &&
2005 offset
!= mddev
->bitmap_info
.default_offset
)
2007 mddev
->bitmap_info
.offset
= offset
;
2009 mddev
->pers
->quiesce(mddev
, 1);
2010 rv
= bitmap_create(mddev
);
2012 rv
= bitmap_load(mddev
);
2014 bitmap_destroy(mddev
);
2015 mddev
->bitmap_info
.offset
= 0;
2017 mddev
->pers
->quiesce(mddev
, 0);
2023 if (!mddev
->external
) {
2024 /* Ensure new bitmap info is stored in
2025 * metadata promptly.
2027 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2028 md_wakeup_thread(mddev
->thread
);
2033 static struct md_sysfs_entry bitmap_location
=
2034 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
2036 /* 'bitmap/space' is the space available at 'location' for the
2037 * bitmap. This allows the kernel to know when it is safe to
2038 * resize the bitmap to match a resized array.
2041 space_show(struct mddev
*mddev
, char *page
)
2043 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.space
);
2047 space_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2049 unsigned long sectors
;
2052 rv
= kstrtoul(buf
, 10, §ors
);
2059 if (mddev
->bitmap
&&
2060 sectors
< (mddev
->bitmap
->storage
.bytes
+ 511) >> 9)
2061 return -EFBIG
; /* Bitmap is too big for this small space */
2063 /* could make sure it isn't too big, but that isn't really
2064 * needed - user-space should be careful.
2066 mddev
->bitmap_info
.space
= sectors
;
2070 static struct md_sysfs_entry bitmap_space
=
2071 __ATTR(space
, S_IRUGO
|S_IWUSR
, space_show
, space_store
);
2074 timeout_show(struct mddev
*mddev
, char *page
)
2077 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
2078 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
2080 len
= sprintf(page
, "%lu", secs
);
2082 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
2083 len
+= sprintf(page
+len
, "\n");
2088 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2090 /* timeout can be set at any time */
2091 unsigned long timeout
;
2092 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
2096 /* just to make sure we don't overflow... */
2097 if (timeout
>= LONG_MAX
/ HZ
)
2100 timeout
= timeout
* HZ
/ 10000;
2102 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
2103 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2106 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2107 if (mddev
->thread
) {
2108 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2109 * the bitmap is all clean and we don't need to
2110 * adjust the timeout right now
2112 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2113 mddev
->thread
->timeout
= timeout
;
2114 md_wakeup_thread(mddev
->thread
);
2120 static struct md_sysfs_entry bitmap_timeout
=
2121 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2124 backlog_show(struct mddev
*mddev
, char *page
)
2126 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2130 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2132 unsigned long backlog
;
2133 int rv
= kstrtoul(buf
, 10, &backlog
);
2136 if (backlog
> COUNTER_MAX
)
2138 mddev
->bitmap_info
.max_write_behind
= backlog
;
2142 static struct md_sysfs_entry bitmap_backlog
=
2143 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2146 chunksize_show(struct mddev
*mddev
, char *page
)
2148 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2152 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2154 /* Can only be changed when no bitmap is active */
2156 unsigned long csize
;
2159 rv
= kstrtoul(buf
, 10, &csize
);
2163 !is_power_of_2(csize
))
2165 mddev
->bitmap_info
.chunksize
= csize
;
2169 static struct md_sysfs_entry bitmap_chunksize
=
2170 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2172 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2174 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2175 ? "external" : "internal"));
2178 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2180 if (mddev
->bitmap
||
2181 mddev
->bitmap_info
.file
||
2182 mddev
->bitmap_info
.offset
)
2184 if (strncmp(buf
, "external", 8) == 0)
2185 mddev
->bitmap_info
.external
= 1;
2186 else if (strncmp(buf
, "internal", 8) == 0)
2187 mddev
->bitmap_info
.external
= 0;
2193 static struct md_sysfs_entry bitmap_metadata
=
2194 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2196 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2200 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2203 len
= sprintf(page
, "\n");
2207 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2209 if (mddev
->bitmap
== NULL
)
2211 if (strncmp(buf
, "false", 5) == 0)
2212 mddev
->bitmap
->need_sync
= 1;
2213 else if (strncmp(buf
, "true", 4) == 0) {
2214 if (mddev
->degraded
)
2216 mddev
->bitmap
->need_sync
= 0;
2222 static struct md_sysfs_entry bitmap_can_clear
=
2223 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2226 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2228 if (mddev
->bitmap
== NULL
)
2229 return sprintf(page
, "0\n");
2230 return sprintf(page
, "%lu\n",
2231 mddev
->bitmap
->behind_writes_used
);
2235 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2238 mddev
->bitmap
->behind_writes_used
= 0;
2242 static struct md_sysfs_entry max_backlog_used
=
2243 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2244 behind_writes_used_show
, behind_writes_used_reset
);
2246 static struct attribute
*md_bitmap_attrs
[] = {
2247 &bitmap_location
.attr
,
2249 &bitmap_timeout
.attr
,
2250 &bitmap_backlog
.attr
,
2251 &bitmap_chunksize
.attr
,
2252 &bitmap_metadata
.attr
,
2253 &bitmap_can_clear
.attr
,
2254 &max_backlog_used
.attr
,
2257 struct attribute_group md_bitmap_group
= {
2259 .attrs
= md_bitmap_attrs
,