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
*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("%s: bitmap map page allocation failed, hijacking\n",
81 /* failed - set the hijacked flag so that we can use the
82 * pointer as a counter */
83 if (!bitmap
->bp
[page
].map
)
84 bitmap
->bp
[page
].hijacked
= 1;
85 } else if (bitmap
->bp
[page
].map
||
86 bitmap
->bp
[page
].hijacked
) {
87 /* somebody beat us to getting the page */
92 /* no page was in place and we have one, so install it */
94 bitmap
->bp
[page
].map
= mappage
;
95 bitmap
->missing_pages
--;
100 /* if page is completely empty, put it back on the free list, or dealloc it */
101 /* if page was hijacked, unmark the flag so it might get alloced next time */
102 /* Note: lock should be held when calling this */
103 static void bitmap_checkfree(struct bitmap
*bitmap
, unsigned long page
)
107 if (bitmap
->bp
[page
].count
) /* page is still busy */
110 /* page is no longer in use, it can be released */
112 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
113 bitmap
->bp
[page
].hijacked
= 0;
114 bitmap
->bp
[page
].map
= NULL
;
116 /* normal case, free the page */
117 ptr
= bitmap
->bp
[page
].map
;
118 bitmap
->bp
[page
].map
= NULL
;
119 bitmap
->missing_pages
++;
125 * bitmap file handling - read and write the bitmap file and its superblock
129 * basic page I/O operations
132 /* IO operations when bitmap is stored near all superblocks */
133 static struct page
*read_sb_page(struct mddev
*mddev
, loff_t offset
,
135 unsigned long index
, int size
)
137 /* choose a good rdev and read the page from there */
139 struct md_rdev
*rdev
;
144 page
= alloc_page(GFP_KERNEL
);
146 return ERR_PTR(-ENOMEM
);
150 rdev_for_each(rdev
, mddev
) {
151 if (! test_bit(In_sync
, &rdev
->flags
)
152 || test_bit(Faulty
, &rdev
->flags
))
155 target
= offset
+ index
* (PAGE_SIZE
/512);
157 if (sync_page_io(rdev
, target
,
158 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
161 attach_page_buffers(page
, NULL
); /* so that free_buffer will
168 return ERR_PTR(-EIO
);
172 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
174 /* Iterate the disks of an mddev, using rcu to protect access to the
175 * linked list, and raising the refcount of devices we return to ensure
176 * they don't disappear while in use.
177 * As devices are only added or removed when raid_disk is < 0 and
178 * nr_pending is 0 and In_sync is clear, the entries we return will
179 * still be in the same position on the list when we re-enter
180 * list_for_each_continue_rcu.
182 struct list_head
*pos
;
185 /* start at the beginning */
188 /* release the previous rdev and start from there. */
189 rdev_dec_pending(rdev
, mddev
);
190 pos
= &rdev
->same_set
;
192 list_for_each_continue_rcu(pos
, &mddev
->disks
) {
193 rdev
= list_entry(pos
, struct md_rdev
, same_set
);
194 if (rdev
->raid_disk
>= 0 &&
195 !test_bit(Faulty
, &rdev
->flags
)) {
196 /* this is a usable devices */
197 atomic_inc(&rdev
->nr_pending
);
206 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
208 struct md_rdev
*rdev
= NULL
;
209 struct block_device
*bdev
;
210 struct mddev
*mddev
= bitmap
->mddev
;
212 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
213 int size
= PAGE_SIZE
;
214 loff_t offset
= mddev
->bitmap_info
.offset
;
216 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
218 if (page
->index
== bitmap
->file_pages
-1)
219 size
= roundup(bitmap
->last_page_size
,
220 bdev_logical_block_size(bdev
));
221 /* Just make sure we aren't corrupting data or
224 if (mddev
->external
) {
225 /* Bitmap could be anywhere. */
226 if (rdev
->sb_start
+ offset
+ (page
->index
230 rdev
->sb_start
+ offset
231 < (rdev
->data_offset
+ mddev
->dev_sectors
234 } else if (offset
< 0) {
235 /* DATA BITMAP METADATA */
237 + (long)(page
->index
* (PAGE_SIZE
/512))
239 /* bitmap runs in to metadata */
241 if (rdev
->data_offset
+ mddev
->dev_sectors
242 > rdev
->sb_start
+ offset
)
243 /* data runs in to bitmap */
245 } else if (rdev
->sb_start
< rdev
->data_offset
) {
246 /* METADATA BITMAP DATA */
249 + page
->index
*(PAGE_SIZE
/512) + size
/512
251 /* bitmap runs in to data */
254 /* DATA METADATA BITMAP - no problems */
256 md_super_write(mddev
, rdev
,
257 rdev
->sb_start
+ offset
258 + page
->index
* (PAGE_SIZE
/512),
264 md_super_wait(mddev
);
271 static void bitmap_file_kick(struct bitmap
*bitmap
);
273 * write out a page to a file
275 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
277 struct buffer_head
*bh
;
279 if (bitmap
->file
== NULL
) {
280 switch (write_sb_page(bitmap
, page
, wait
)) {
282 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
286 bh
= page_buffers(page
);
288 while (bh
&& bh
->b_blocknr
) {
289 atomic_inc(&bitmap
->pending_writes
);
290 set_buffer_locked(bh
);
291 set_buffer_mapped(bh
);
292 submit_bh(WRITE
| REQ_SYNC
, bh
);
293 bh
= bh
->b_this_page
;
297 wait_event(bitmap
->write_wait
,
298 atomic_read(&bitmap
->pending_writes
)==0);
300 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
301 bitmap_file_kick(bitmap
);
304 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
306 struct bitmap
*bitmap
= bh
->b_private
;
310 spin_lock_irqsave(&bitmap
->lock
, flags
);
311 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
312 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
314 if (atomic_dec_and_test(&bitmap
->pending_writes
))
315 wake_up(&bitmap
->write_wait
);
318 /* copied from buffer.c */
320 __clear_page_buffers(struct page
*page
)
322 ClearPagePrivate(page
);
323 set_page_private(page
, 0);
324 page_cache_release(page
);
326 static void free_buffers(struct page
*page
)
328 struct buffer_head
*bh
= page_buffers(page
);
331 struct buffer_head
*next
= bh
->b_this_page
;
332 free_buffer_head(bh
);
335 __clear_page_buffers(page
);
339 /* read a page from a file.
340 * We both read the page, and attach buffers to the page to record the
341 * address of each block (using bmap). These addresses will be used
342 * to write the block later, completely bypassing the filesystem.
343 * This usage is similar to how swap files are handled, and allows us
344 * to write to a file with no concerns of memory allocation failing.
346 static struct page
*read_page(struct file
*file
, unsigned long index
,
347 struct bitmap
*bitmap
,
350 struct page
*page
= NULL
;
351 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
352 struct buffer_head
*bh
;
355 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
356 (unsigned long long)index
<< PAGE_SHIFT
);
358 page
= alloc_page(GFP_KERNEL
);
360 page
= ERR_PTR(-ENOMEM
);
364 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
367 page
= ERR_PTR(-ENOMEM
);
370 attach_page_buffers(page
, bh
);
371 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
376 bh
->b_blocknr
= bmap(inode
, block
);
377 if (bh
->b_blocknr
== 0) {
378 /* Cannot use this file! */
380 page
= ERR_PTR(-EINVAL
);
383 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
384 if (count
< (1<<inode
->i_blkbits
))
387 count
-= (1<<inode
->i_blkbits
);
389 bh
->b_end_io
= end_bitmap_write
;
390 bh
->b_private
= bitmap
;
391 atomic_inc(&bitmap
->pending_writes
);
392 set_buffer_locked(bh
);
393 set_buffer_mapped(bh
);
397 bh
= bh
->b_this_page
;
401 wait_event(bitmap
->write_wait
,
402 atomic_read(&bitmap
->pending_writes
)==0);
403 if (bitmap
->flags
& BITMAP_WRITE_ERROR
) {
405 page
= ERR_PTR(-EIO
);
409 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %ld\n",
411 (unsigned long long)index
<< PAGE_SHIFT
,
417 * bitmap file superblock operations
420 /* update the event counter and sync the superblock to disk */
421 void bitmap_update_sb(struct bitmap
*bitmap
)
425 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
427 if (bitmap
->mddev
->bitmap_info
.external
)
429 if (!bitmap
->sb_page
) /* no superblock */
431 sb
= kmap_atomic(bitmap
->sb_page
);
432 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
433 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
434 /* rocking back to read-only */
435 bitmap
->events_cleared
= bitmap
->mddev
->events
;
436 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
437 sb
->state
= cpu_to_le32(bitmap
->flags
);
438 /* Just in case these have been changed via sysfs: */
439 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
440 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
442 write_page(bitmap
, bitmap
->sb_page
, 1);
445 /* print out the bitmap file superblock */
446 void bitmap_print_sb(struct bitmap
*bitmap
)
450 if (!bitmap
|| !bitmap
->sb_page
)
452 sb
= kmap_atomic(bitmap
->sb_page
);
453 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
454 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
455 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
456 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
457 *(__u32
*)(sb
->uuid
+0),
458 *(__u32
*)(sb
->uuid
+4),
459 *(__u32
*)(sb
->uuid
+8),
460 *(__u32
*)(sb
->uuid
+12));
461 printk(KERN_DEBUG
" events: %llu\n",
462 (unsigned long long) le64_to_cpu(sb
->events
));
463 printk(KERN_DEBUG
"events cleared: %llu\n",
464 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
465 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
466 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
467 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
468 printk(KERN_DEBUG
" sync size: %llu KB\n",
469 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
470 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
478 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
479 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
480 * This function verifies 'bitmap_info' and populates the on-disk bitmap
481 * structure, which is to be written to disk.
483 * Returns: 0 on success, -Exxx on error
485 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
488 unsigned long chunksize
, daemon_sleep
, write_behind
;
491 bitmap
->sb_page
= alloc_page(GFP_KERNEL
);
492 if (IS_ERR(bitmap
->sb_page
)) {
493 err
= PTR_ERR(bitmap
->sb_page
);
494 bitmap
->sb_page
= NULL
;
497 bitmap
->sb_page
->index
= 0;
499 sb
= kmap_atomic(bitmap
->sb_page
);
501 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
502 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
504 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
506 if (!is_power_of_2(chunksize
)) {
508 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
511 sb
->chunksize
= cpu_to_le32(chunksize
);
513 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
515 (daemon_sleep
< 1) || (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
516 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
517 daemon_sleep
= 5 * HZ
;
519 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
520 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
523 * FIXME: write_behind for RAID1. If not specified, what
524 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
526 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
527 if (write_behind
> COUNTER_MAX
)
528 write_behind
= COUNTER_MAX
/ 2;
529 sb
->write_behind
= cpu_to_le32(write_behind
);
530 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
532 /* keep the array size field of the bitmap superblock up to date */
533 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
535 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
537 bitmap
->flags
|= BITMAP_STALE
;
538 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
539 bitmap
->events_cleared
= bitmap
->mddev
->events
;
540 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
542 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
543 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HOSTENDIAN
);
550 /* read the superblock from the bitmap file and initialize some bitmap fields */
551 static int bitmap_read_sb(struct bitmap
*bitmap
)
555 unsigned long chunksize
, daemon_sleep
, write_behind
;
556 unsigned long long events
;
559 /* page 0 is the superblock, read it... */
561 loff_t isize
= i_size_read(bitmap
->file
->f_mapping
->host
);
562 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
564 bitmap
->sb_page
= read_page(bitmap
->file
, 0, bitmap
, bytes
);
566 bitmap
->sb_page
= read_sb_page(bitmap
->mddev
,
567 bitmap
->mddev
->bitmap_info
.offset
,
569 0, sizeof(bitmap_super_t
));
571 if (IS_ERR(bitmap
->sb_page
)) {
572 err
= PTR_ERR(bitmap
->sb_page
);
573 bitmap
->sb_page
= NULL
;
577 sb
= kmap_atomic(bitmap
->sb_page
);
579 chunksize
= le32_to_cpu(sb
->chunksize
);
580 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
581 write_behind
= le32_to_cpu(sb
->write_behind
);
583 /* verify that the bitmap-specific fields are valid */
584 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
585 reason
= "bad magic";
586 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
587 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
588 reason
= "unrecognized superblock version";
589 else if (chunksize
< 512)
590 reason
= "bitmap chunksize too small";
591 else if (!is_power_of_2(chunksize
))
592 reason
= "bitmap chunksize not a power of 2";
593 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
594 reason
= "daemon sleep period out of range";
595 else if (write_behind
> COUNTER_MAX
)
596 reason
= "write-behind limit out of range (0 - 16383)";
598 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
599 bmname(bitmap
), reason
);
603 /* keep the array size field of the bitmap superblock up to date */
604 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
606 if (bitmap
->mddev
->persistent
) {
608 * We have a persistent array superblock, so compare the
609 * bitmap's UUID and event counter to the mddev's
611 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
613 "%s: bitmap superblock UUID mismatch\n",
617 events
= le64_to_cpu(sb
->events
);
618 if (events
< bitmap
->mddev
->events
) {
620 "%s: bitmap file is out of date (%llu < %llu) "
621 "-- forcing full recovery\n",
622 bmname(bitmap
), events
,
623 (unsigned long long) bitmap
->mddev
->events
);
624 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
628 /* assign fields using values from superblock */
629 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
630 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
631 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
632 bitmap
->flags
|= le32_to_cpu(sb
->state
);
633 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
634 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
635 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
636 if (bitmap
->flags
& BITMAP_STALE
)
637 bitmap
->events_cleared
= bitmap
->mddev
->events
;
642 bitmap_print_sb(bitmap
);
646 enum bitmap_mask_op
{
651 /* record the state of the bitmap in the superblock. Return the old value */
652 static int bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
653 enum bitmap_mask_op op
)
658 if (!bitmap
->sb_page
) /* can't set the state */
660 sb
= kmap_atomic(bitmap
->sb_page
);
661 old
= le32_to_cpu(sb
->state
) & bits
;
664 sb
->state
|= cpu_to_le32(bits
);
665 bitmap
->flags
|= bits
;
668 sb
->state
&= cpu_to_le32(~bits
);
669 bitmap
->flags
&= ~bits
;
679 * general bitmap file operations
685 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
686 * file a page at a time. There's a superblock at the start of the file.
688 /* calculate the index of the page that contains this bit */
689 static inline unsigned long file_page_index(struct bitmap
*bitmap
, unsigned long chunk
)
691 if (!bitmap
->mddev
->bitmap_info
.external
)
692 chunk
+= sizeof(bitmap_super_t
) << 3;
693 return chunk
>> PAGE_BIT_SHIFT
;
696 /* calculate the (bit) offset of this bit within a page */
697 static inline unsigned long file_page_offset(struct bitmap
*bitmap
, unsigned long chunk
)
699 if (!bitmap
->mddev
->bitmap_info
.external
)
700 chunk
+= sizeof(bitmap_super_t
) << 3;
701 return chunk
& (PAGE_BITS
- 1);
705 * return a pointer to the page in the filemap that contains the given bit
707 * this lookup is complicated by the fact that the bitmap sb might be exactly
708 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
711 static inline struct page
*filemap_get_page(struct bitmap
*bitmap
,
714 if (file_page_index(bitmap
, chunk
) >= bitmap
->file_pages
)
716 return bitmap
->filemap
[file_page_index(bitmap
, chunk
)
717 - file_page_index(bitmap
, 0)];
720 static void bitmap_file_unmap(struct bitmap
*bitmap
)
722 struct page
**map
, *sb_page
;
727 spin_lock_irqsave(&bitmap
->lock
, flags
);
728 map
= bitmap
->filemap
;
729 bitmap
->filemap
= NULL
;
730 attr
= bitmap
->filemap_attr
;
731 bitmap
->filemap_attr
= NULL
;
732 pages
= bitmap
->file_pages
;
733 bitmap
->file_pages
= 0;
734 sb_page
= bitmap
->sb_page
;
735 bitmap
->sb_page
= NULL
;
736 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
739 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
740 free_buffers(map
[pages
]);
745 free_buffers(sb_page
);
748 static void bitmap_file_put(struct bitmap
*bitmap
)
753 spin_lock_irqsave(&bitmap
->lock
, flags
);
756 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
759 wait_event(bitmap
->write_wait
,
760 atomic_read(&bitmap
->pending_writes
)==0);
761 bitmap_file_unmap(bitmap
);
764 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
765 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
771 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
772 * then it is no longer reliable, so we stop using it and we mark the file
773 * as failed in the superblock
775 static void bitmap_file_kick(struct bitmap
*bitmap
)
777 char *path
, *ptr
= NULL
;
779 if (bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_SET
) == 0) {
780 bitmap_update_sb(bitmap
);
783 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
785 ptr
= d_path(&bitmap
->file
->f_path
, path
,
789 "%s: kicking failed bitmap file %s from array!\n",
790 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
795 "%s: disabling internal bitmap due to errors\n",
799 bitmap_file_put(bitmap
);
804 enum bitmap_page_attr
{
805 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
806 BITMAP_PAGE_PENDING
= 1, /* there are bits that are being cleaned.
807 * i.e. counter is 1 or 2. */
808 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
811 static inline void set_page_attr(struct bitmap
*bitmap
, struct page
*page
,
812 enum bitmap_page_attr attr
)
814 __set_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
817 static inline void clear_page_attr(struct bitmap
*bitmap
, struct page
*page
,
818 enum bitmap_page_attr attr
)
820 __clear_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
823 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, struct page
*page
,
824 enum bitmap_page_attr attr
)
826 return test_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
830 * bitmap_file_set_bit -- called before performing a write to the md device
831 * to set (and eventually sync) a particular bit in the bitmap file
833 * we set the bit immediately, then we record the page number so that
834 * when an unplug occurs, we can flush the dirty pages out to disk
836 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
841 unsigned long chunk
= block
>> bitmap
->chunkshift
;
843 if (!bitmap
->filemap
)
846 page
= filemap_get_page(bitmap
, chunk
);
849 bit
= file_page_offset(bitmap
, chunk
);
852 kaddr
= kmap_atomic(page
);
853 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
856 __set_bit_le(bit
, kaddr
);
857 kunmap_atomic(kaddr
);
858 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
859 /* record page number so it gets flushed to disk when unplug occurs */
860 set_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
863 /* this gets called when the md device is ready to unplug its underlying
864 * (slave) device queues -- before we let any writes go down, we need to
865 * sync the dirty pages of the bitmap file to disk */
866 void bitmap_unplug(struct bitmap
*bitmap
)
868 unsigned long i
, flags
;
869 int dirty
, need_write
;
876 /* look at each page to see if there are any set bits that need to be
877 * flushed out to disk */
878 for (i
= 0; i
< bitmap
->file_pages
; i
++) {
879 spin_lock_irqsave(&bitmap
->lock
, flags
);
880 if (!bitmap
->filemap
) {
881 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
884 page
= bitmap
->filemap
[i
];
885 dirty
= test_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
886 need_write
= test_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
887 clear_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
888 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
891 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
893 if (dirty
|| need_write
)
894 write_page(bitmap
, page
, 0);
896 if (wait
) { /* if any writes were performed, we need to wait on them */
898 wait_event(bitmap
->write_wait
,
899 atomic_read(&bitmap
->pending_writes
)==0);
901 md_super_wait(bitmap
->mddev
);
903 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
904 bitmap_file_kick(bitmap
);
906 EXPORT_SYMBOL(bitmap_unplug
);
908 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
909 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
910 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
911 * memory mapping of the bitmap file
913 * if there's no bitmap file, or if the bitmap file had been
914 * previously kicked from the array, we mark all the bits as
915 * 1's in order to cause a full resync.
917 * We ignore all bits for sectors that end earlier than 'start'.
918 * This is used when reading an out-of-date bitmap...
920 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
922 unsigned long i
, chunks
, index
, oldindex
, bit
;
923 struct page
*page
= NULL
, *oldpage
= NULL
;
924 unsigned long num_pages
, bit_cnt
= 0;
926 unsigned long bytes
, offset
;
931 chunks
= bitmap
->chunks
;
934 BUG_ON(!file
&& !bitmap
->mddev
->bitmap_info
.offset
);
936 outofdate
= bitmap
->flags
& BITMAP_STALE
;
938 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
939 "recovery\n", bmname(bitmap
));
941 bytes
= DIV_ROUND_UP(bitmap
->chunks
, 8);
942 if (!bitmap
->mddev
->bitmap_info
.external
)
943 bytes
+= sizeof(bitmap_super_t
);
945 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
947 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
) {
948 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
950 (unsigned long) i_size_read(file
->f_mapping
->host
),
957 bitmap
->filemap
= kmalloc(sizeof(struct page
*) * num_pages
, GFP_KERNEL
);
958 if (!bitmap
->filemap
)
961 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
962 bitmap
->filemap_attr
= kzalloc(
963 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
965 if (!bitmap
->filemap_attr
)
970 for (i
= 0; i
< chunks
; i
++) {
972 index
= file_page_index(bitmap
, i
);
973 bit
= file_page_offset(bitmap
, i
);
974 if (index
!= oldindex
) { /* this is a new page, read it in */
976 /* unmap the old page, we're done with it */
977 if (index
== num_pages
-1)
978 count
= bytes
- index
* PAGE_SIZE
;
981 if (index
== 0 && bitmap
->sb_page
) {
983 * if we're here then the superblock page
984 * contains some bits (PAGE_SIZE != sizeof sb)
985 * we've already read it in, so just use it
987 page
= bitmap
->sb_page
;
988 offset
= sizeof(bitmap_super_t
);
992 bitmap
->mddev
->bitmap_info
.offset
,
996 page
= read_page(file
, index
, bitmap
, count
);
999 page
= read_sb_page(bitmap
->mddev
,
1000 bitmap
->mddev
->bitmap_info
.offset
,
1005 if (IS_ERR(page
)) { /* read error */
1006 ret
= PTR_ERR(page
);
1013 bitmap
->filemap
[bitmap
->file_pages
++] = page
;
1014 bitmap
->last_page_size
= count
;
1018 * if bitmap is out of date, dirty the
1019 * whole page and write it out
1021 paddr
= kmap_atomic(page
);
1022 memset(paddr
+ offset
, 0xff,
1023 PAGE_SIZE
- offset
);
1024 kunmap_atomic(paddr
);
1025 write_page(bitmap
, page
, 1);
1028 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1032 paddr
= kmap_atomic(page
);
1033 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1034 b
= test_bit(bit
, paddr
);
1036 b
= test_bit_le(bit
, paddr
);
1037 kunmap_atomic(paddr
);
1039 /* if the disk bit is set, set the memory bit */
1040 int needed
= ((sector_t
)(i
+1) << bitmap
->chunkshift
1042 bitmap_set_memory_bits(bitmap
,
1043 (sector_t
)i
<< bitmap
->chunkshift
,
1049 /* everything went OK */
1051 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
1053 if (bit_cnt
) { /* Kick recovery if any bits were set */
1054 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1055 md_wakeup_thread(bitmap
->mddev
->thread
);
1058 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1059 "read %lu/%lu pages, set %lu of %lu bits\n",
1060 bmname(bitmap
), bitmap
->file_pages
, num_pages
, bit_cnt
, chunks
);
1065 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1066 bmname(bitmap
), ret
);
1070 void bitmap_write_all(struct bitmap
*bitmap
)
1072 /* We don't actually write all bitmap blocks here,
1073 * just flag them as needing to be written
1077 spin_lock_irq(&bitmap
->lock
);
1078 for (i
= 0; i
< bitmap
->file_pages
; i
++)
1079 set_page_attr(bitmap
, bitmap
->filemap
[i
],
1080 BITMAP_PAGE_NEEDWRITE
);
1081 bitmap
->allclean
= 0;
1082 spin_unlock_irq(&bitmap
->lock
);
1085 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
1087 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1088 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1089 bitmap
->bp
[page
].count
+= inc
;
1090 bitmap_checkfree(bitmap
, page
);
1092 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1093 sector_t offset
, sector_t
*blocks
,
1097 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1101 void bitmap_daemon_work(struct mddev
*mddev
)
1103 struct bitmap
*bitmap
;
1105 unsigned long flags
;
1106 struct page
*page
= NULL
, *lastpage
= NULL
;
1110 /* Use a mutex to guard daemon_work against
1113 mutex_lock(&mddev
->bitmap_info
.mutex
);
1114 bitmap
= mddev
->bitmap
;
1115 if (bitmap
== NULL
) {
1116 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1119 if (time_before(jiffies
, bitmap
->daemon_lastrun
1120 + mddev
->bitmap_info
.daemon_sleep
))
1123 bitmap
->daemon_lastrun
= jiffies
;
1124 if (bitmap
->allclean
) {
1125 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1128 bitmap
->allclean
= 1;
1130 spin_lock_irqsave(&bitmap
->lock
, flags
);
1131 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1132 bitmap_counter_t
*bmc
;
1133 if (!bitmap
->filemap
)
1134 /* error or shutdown */
1137 page
= filemap_get_page(bitmap
, j
);
1139 if (page
!= lastpage
) {
1140 /* skip this page unless it's marked as needing cleaning */
1141 if (!test_page_attr(bitmap
, page
, BITMAP_PAGE_PENDING
)) {
1142 int need_write
= test_page_attr(bitmap
, page
,
1143 BITMAP_PAGE_NEEDWRITE
);
1145 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
1147 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1149 write_page(bitmap
, page
, 0);
1150 spin_lock_irqsave(&bitmap
->lock
, flags
);
1151 j
|= (PAGE_BITS
- 1);
1155 /* grab the new page, sync and release the old */
1156 if (lastpage
!= NULL
) {
1157 if (test_page_attr(bitmap
, lastpage
,
1158 BITMAP_PAGE_NEEDWRITE
)) {
1159 clear_page_attr(bitmap
, lastpage
,
1160 BITMAP_PAGE_NEEDWRITE
);
1161 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1162 write_page(bitmap
, lastpage
, 0);
1164 set_page_attr(bitmap
, lastpage
,
1165 BITMAP_PAGE_NEEDWRITE
);
1166 bitmap
->allclean
= 0;
1167 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1170 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1173 /* We are possibly going to clear some bits, so make
1174 * sure that events_cleared is up-to-date.
1176 if (bitmap
->need_sync
&&
1177 mddev
->bitmap_info
.external
== 0) {
1179 bitmap
->need_sync
= 0;
1180 sb
= kmap_atomic(bitmap
->sb_page
);
1181 sb
->events_cleared
=
1182 cpu_to_le64(bitmap
->events_cleared
);
1184 write_page(bitmap
, bitmap
->sb_page
, 1);
1186 spin_lock_irqsave(&bitmap
->lock
, flags
);
1187 if (!bitmap
->need_sync
)
1188 clear_page_attr(bitmap
, page
, BITMAP_PAGE_PENDING
);
1190 bitmap
->allclean
= 0;
1192 bmc
= bitmap_get_counter(bitmap
,
1193 (sector_t
)j
<< bitmap
->chunkshift
,
1196 j
|= PAGE_COUNTER_MASK
;
1198 if (*bmc
== 1 && !bitmap
->need_sync
) {
1199 /* we can clear the bit */
1201 bitmap_count_page(bitmap
,
1202 (sector_t
)j
<< bitmap
->chunkshift
,
1206 paddr
= kmap_atomic(page
);
1207 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1208 clear_bit(file_page_offset(bitmap
, j
),
1212 file_page_offset(bitmap
,
1215 kunmap_atomic(paddr
);
1216 } else if (*bmc
<= 2) {
1217 *bmc
= 1; /* maybe clear the bit next time */
1218 set_page_attr(bitmap
, page
, BITMAP_PAGE_PENDING
);
1219 bitmap
->allclean
= 0;
1223 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1225 /* now sync the final page */
1226 if (lastpage
!= NULL
) {
1227 spin_lock_irqsave(&bitmap
->lock
, flags
);
1228 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1229 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1230 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1231 write_page(bitmap
, lastpage
, 0);
1233 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1234 bitmap
->allclean
= 0;
1235 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1240 if (bitmap
->allclean
== 0)
1241 mddev
->thread
->timeout
=
1242 mddev
->bitmap_info
.daemon_sleep
;
1243 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1246 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1247 sector_t offset
, sector_t
*blocks
,
1249 __releases(bitmap
->lock
)
1250 __acquires(bitmap
->lock
)
1252 /* If 'create', we might release the lock and reclaim it.
1253 * The lock must have been taken with interrupts enabled.
1254 * If !create, we don't release the lock.
1256 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1257 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1258 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1262 err
= bitmap_checkpage(bitmap
, page
, create
);
1264 if (bitmap
->bp
[page
].hijacked
||
1265 bitmap
->bp
[page
].map
== NULL
)
1266 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1267 PAGE_COUNTER_SHIFT
- 1);
1269 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1270 *blocks
= csize
- (offset
& (csize
- 1));
1275 /* now locked ... */
1277 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1278 /* should we use the first or second counter field
1279 * of the hijacked pointer? */
1280 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1281 return &((bitmap_counter_t
*)
1282 &bitmap
->bp
[page
].map
)[hi
];
1283 } else /* page is allocated */
1284 return (bitmap_counter_t
*)
1285 &(bitmap
->bp
[page
].map
[pageoff
]);
1288 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1295 atomic_inc(&bitmap
->behind_writes
);
1296 bw
= atomic_read(&bitmap
->behind_writes
);
1297 if (bw
> bitmap
->behind_writes_used
)
1298 bitmap
->behind_writes_used
= bw
;
1300 pr_debug("inc write-behind count %d/%lu\n",
1301 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1306 bitmap_counter_t
*bmc
;
1308 spin_lock_irq(&bitmap
->lock
);
1309 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1311 spin_unlock_irq(&bitmap
->lock
);
1315 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1316 DEFINE_WAIT(__wait
);
1317 /* note that it is safe to do the prepare_to_wait
1318 * after the test as long as we do it before dropping
1321 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1322 TASK_UNINTERRUPTIBLE
);
1323 spin_unlock_irq(&bitmap
->lock
);
1325 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1331 bitmap_file_set_bit(bitmap
, offset
);
1332 bitmap_count_page(bitmap
, offset
, 1);
1340 spin_unlock_irq(&bitmap
->lock
);
1343 if (sectors
> blocks
)
1350 EXPORT_SYMBOL(bitmap_startwrite
);
1352 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1353 int success
, int behind
)
1358 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1359 wake_up(&bitmap
->behind_wait
);
1360 pr_debug("dec write-behind count %d/%lu\n",
1361 atomic_read(&bitmap
->behind_writes
),
1362 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1367 unsigned long flags
;
1368 bitmap_counter_t
*bmc
;
1370 spin_lock_irqsave(&bitmap
->lock
, flags
);
1371 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1373 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1377 if (success
&& !bitmap
->mddev
->degraded
&&
1378 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1379 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1380 bitmap
->need_sync
= 1;
1381 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1384 if (!success
&& !NEEDED(*bmc
))
1385 *bmc
|= NEEDED_MASK
;
1387 if (COUNTER(*bmc
) == COUNTER_MAX
)
1388 wake_up(&bitmap
->overflow_wait
);
1392 set_page_attr(bitmap
,
1395 offset
>> bitmap
->chunkshift
),
1396 BITMAP_PAGE_PENDING
);
1397 bitmap
->allclean
= 0;
1399 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1401 if (sectors
> blocks
)
1407 EXPORT_SYMBOL(bitmap_endwrite
);
1409 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1412 bitmap_counter_t
*bmc
;
1414 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1416 return 1; /* always resync if no bitmap */
1418 spin_lock_irq(&bitmap
->lock
);
1419 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1425 else if (NEEDED(*bmc
)) {
1427 if (!degraded
) { /* don't set/clear bits if degraded */
1428 *bmc
|= RESYNC_MASK
;
1429 *bmc
&= ~NEEDED_MASK
;
1433 spin_unlock_irq(&bitmap
->lock
);
1437 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1440 /* bitmap_start_sync must always report on multiples of whole
1441 * pages, otherwise resync (which is very PAGE_SIZE based) will
1443 * So call __bitmap_start_sync repeatedly (if needed) until
1444 * At least PAGE_SIZE>>9 blocks are covered.
1445 * Return the 'or' of the result.
1451 while (*blocks
< (PAGE_SIZE
>>9)) {
1452 rv
|= __bitmap_start_sync(bitmap
, offset
,
1453 &blocks1
, degraded
);
1459 EXPORT_SYMBOL(bitmap_start_sync
);
1461 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1463 bitmap_counter_t
*bmc
;
1464 unsigned long flags
;
1466 if (bitmap
== NULL
) {
1470 spin_lock_irqsave(&bitmap
->lock
, flags
);
1471 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1476 *bmc
&= ~RESYNC_MASK
;
1478 if (!NEEDED(*bmc
) && aborted
)
1479 *bmc
|= NEEDED_MASK
;
1482 set_page_attr(bitmap
,
1483 filemap_get_page(bitmap
, offset
>> bitmap
->chunkshift
),
1484 BITMAP_PAGE_PENDING
);
1485 bitmap
->allclean
= 0;
1490 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1492 EXPORT_SYMBOL(bitmap_end_sync
);
1494 void bitmap_close_sync(struct bitmap
*bitmap
)
1496 /* Sync has finished, and any bitmap chunks that weren't synced
1497 * properly have been aborted. It remains to us to clear the
1498 * RESYNC bit wherever it is still on
1500 sector_t sector
= 0;
1504 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1505 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1509 EXPORT_SYMBOL(bitmap_close_sync
);
1511 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
)
1519 bitmap
->last_end_sync
= jiffies
;
1522 if (time_before(jiffies
, (bitmap
->last_end_sync
1523 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1525 wait_event(bitmap
->mddev
->recovery_wait
,
1526 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1528 bitmap
->mddev
->curr_resync_completed
= sector
;
1529 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1530 sector
&= ~((1ULL << bitmap
->chunkshift
) - 1);
1532 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1533 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1536 bitmap
->last_end_sync
= jiffies
;
1537 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1539 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1541 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1543 /* For each chunk covered by any of these sectors, set the
1544 * counter to 1 and set resync_needed. They should all
1545 * be 0 at this point
1549 bitmap_counter_t
*bmc
;
1550 spin_lock_irq(&bitmap
->lock
);
1551 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1553 spin_unlock_irq(&bitmap
->lock
);
1558 *bmc
= 2 | (needed
? NEEDED_MASK
: 0);
1559 bitmap_count_page(bitmap
, offset
, 1);
1560 page
= filemap_get_page(bitmap
, offset
>> bitmap
->chunkshift
);
1561 set_page_attr(bitmap
, page
, BITMAP_PAGE_PENDING
);
1562 bitmap
->allclean
= 0;
1564 spin_unlock_irq(&bitmap
->lock
);
1567 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1568 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1570 unsigned long chunk
;
1572 for (chunk
= s
; chunk
<= e
; chunk
++) {
1573 sector_t sec
= (sector_t
)chunk
<< bitmap
->chunkshift
;
1574 bitmap_set_memory_bits(bitmap
, sec
, 1);
1575 spin_lock_irq(&bitmap
->lock
);
1576 bitmap_file_set_bit(bitmap
, sec
);
1577 spin_unlock_irq(&bitmap
->lock
);
1578 if (sec
< bitmap
->mddev
->recovery_cp
)
1579 /* We are asserting that the array is dirty,
1580 * so move the recovery_cp address back so
1581 * that it is obvious that it is dirty
1583 bitmap
->mddev
->recovery_cp
= sec
;
1588 * flush out any pending updates
1590 void bitmap_flush(struct mddev
*mddev
)
1592 struct bitmap
*bitmap
= mddev
->bitmap
;
1595 if (!bitmap
) /* there was no bitmap */
1598 /* run the daemon_work three time to ensure everything is flushed
1601 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1602 bitmap
->daemon_lastrun
-= sleep
;
1603 bitmap_daemon_work(mddev
);
1604 bitmap
->daemon_lastrun
-= sleep
;
1605 bitmap_daemon_work(mddev
);
1606 bitmap
->daemon_lastrun
-= sleep
;
1607 bitmap_daemon_work(mddev
);
1608 bitmap_update_sb(bitmap
);
1612 * free memory that was allocated
1614 static void bitmap_free(struct bitmap
*bitmap
)
1616 unsigned long k
, pages
;
1617 struct bitmap_page
*bp
;
1619 if (!bitmap
) /* there was no bitmap */
1622 /* release the bitmap file and kill the daemon */
1623 bitmap_file_put(bitmap
);
1626 pages
= bitmap
->pages
;
1628 /* free all allocated memory */
1630 if (bp
) /* deallocate the page memory */
1631 for (k
= 0; k
< pages
; k
++)
1632 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1638 void bitmap_destroy(struct mddev
*mddev
)
1640 struct bitmap
*bitmap
= mddev
->bitmap
;
1642 if (!bitmap
) /* there was no bitmap */
1645 mutex_lock(&mddev
->bitmap_info
.mutex
);
1646 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1647 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1649 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1651 if (bitmap
->sysfs_can_clear
)
1652 sysfs_put(bitmap
->sysfs_can_clear
);
1654 bitmap_free(bitmap
);
1658 * initialize the bitmap structure
1659 * if this returns an error, bitmap_destroy must be called to do clean up
1661 int bitmap_create(struct mddev
*mddev
)
1663 struct bitmap
*bitmap
;
1664 sector_t blocks
= mddev
->resync_max_sectors
;
1665 unsigned long chunks
;
1666 unsigned long pages
;
1667 struct file
*file
= mddev
->bitmap_info
.file
;
1669 struct sysfs_dirent
*bm
= NULL
;
1671 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1674 && !mddev
->bitmap_info
.offset
) /* bitmap disabled, nothing to do */
1677 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1679 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1683 spin_lock_init(&bitmap
->lock
);
1684 atomic_set(&bitmap
->pending_writes
, 0);
1685 init_waitqueue_head(&bitmap
->write_wait
);
1686 init_waitqueue_head(&bitmap
->overflow_wait
);
1687 init_waitqueue_head(&bitmap
->behind_wait
);
1689 bitmap
->mddev
= mddev
;
1692 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "bitmap");
1694 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, NULL
, "can_clear");
1697 bitmap
->sysfs_can_clear
= NULL
;
1699 bitmap
->file
= file
;
1702 /* As future accesses to this file will use bmap,
1703 * and bypass the page cache, we must sync the file
1708 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1709 if (!mddev
->bitmap_info
.external
) {
1711 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1712 * instructing us to create a new on-disk bitmap instance.
1714 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1715 err
= bitmap_new_disk_sb(bitmap
);
1717 err
= bitmap_read_sb(bitmap
);
1720 if (mddev
->bitmap_info
.chunksize
== 0 ||
1721 mddev
->bitmap_info
.daemon_sleep
== 0)
1722 /* chunksize and time_base need to be
1729 bitmap
->daemon_lastrun
= jiffies
;
1730 bitmap
->chunkshift
= (ffz(~mddev
->bitmap_info
.chunksize
)
1731 - BITMAP_BLOCK_SHIFT
);
1733 /* now that chunksize and chunkshift are set, we can use these macros */
1734 chunks
= (blocks
+ bitmap
->chunkshift
- 1) >>
1736 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1740 bitmap
->chunks
= chunks
;
1741 bitmap
->pages
= pages
;
1742 bitmap
->missing_pages
= pages
;
1744 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1750 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1751 pages
, bmname(bitmap
));
1753 mddev
->bitmap
= bitmap
;
1756 return (bitmap
->flags
& BITMAP_WRITE_ERROR
) ? -EIO
: 0;
1759 bitmap_free(bitmap
);
1763 int bitmap_load(struct mddev
*mddev
)
1767 sector_t sector
= 0;
1768 struct bitmap
*bitmap
= mddev
->bitmap
;
1773 /* Clear out old bitmap info first: Either there is none, or we
1774 * are resuming after someone else has possibly changed things,
1775 * so we should forget old cached info.
1776 * All chunks should be clean, but some might need_sync.
1778 while (sector
< mddev
->resync_max_sectors
) {
1780 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1783 bitmap_close_sync(bitmap
);
1785 if (mddev
->degraded
== 0
1786 || bitmap
->events_cleared
== mddev
->events
)
1787 /* no need to keep dirty bits to optimise a
1788 * re-add of a missing device */
1789 start
= mddev
->recovery_cp
;
1791 err
= bitmap_init_from_disk(bitmap
, start
);
1796 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1797 md_wakeup_thread(mddev
->thread
);
1799 bitmap_update_sb(bitmap
);
1801 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1806 EXPORT_SYMBOL_GPL(bitmap_load
);
1808 void bitmap_status(struct seq_file
*seq
, struct bitmap
*bitmap
)
1810 unsigned long chunk_kb
;
1811 unsigned long flags
;
1816 spin_lock_irqsave(&bitmap
->lock
, flags
);
1817 chunk_kb
= bitmap
->mddev
->bitmap_info
.chunksize
>> 10;
1818 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
1820 bitmap
->pages
- bitmap
->missing_pages
,
1822 (bitmap
->pages
- bitmap
->missing_pages
)
1823 << (PAGE_SHIFT
- 10),
1824 chunk_kb
? chunk_kb
: bitmap
->mddev
->bitmap_info
.chunksize
,
1825 chunk_kb
? "KB" : "B");
1827 seq_printf(seq
, ", file: ");
1828 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
1831 seq_printf(seq
, "\n");
1832 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1836 location_show(struct mddev
*mddev
, char *page
)
1839 if (mddev
->bitmap_info
.file
)
1840 len
= sprintf(page
, "file");
1841 else if (mddev
->bitmap_info
.offset
)
1842 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
1844 len
= sprintf(page
, "none");
1845 len
+= sprintf(page
+len
, "\n");
1850 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1854 if (!mddev
->pers
->quiesce
)
1856 if (mddev
->recovery
|| mddev
->sync_thread
)
1860 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
1861 mddev
->bitmap_info
.offset
) {
1862 /* bitmap already configured. Only option is to clear it */
1863 if (strncmp(buf
, "none", 4) != 0)
1866 mddev
->pers
->quiesce(mddev
, 1);
1867 bitmap_destroy(mddev
);
1868 mddev
->pers
->quiesce(mddev
, 0);
1870 mddev
->bitmap_info
.offset
= 0;
1871 if (mddev
->bitmap_info
.file
) {
1872 struct file
*f
= mddev
->bitmap_info
.file
;
1873 mddev
->bitmap_info
.file
= NULL
;
1874 restore_bitmap_write_access(f
);
1878 /* No bitmap, OK to set a location */
1880 if (strncmp(buf
, "none", 4) == 0)
1881 /* nothing to be done */;
1882 else if (strncmp(buf
, "file:", 5) == 0) {
1883 /* Not supported yet */
1888 rv
= strict_strtoll(buf
+1, 10, &offset
);
1890 rv
= strict_strtoll(buf
, 10, &offset
);
1895 if (mddev
->bitmap_info
.external
== 0 &&
1896 mddev
->major_version
== 0 &&
1897 offset
!= mddev
->bitmap_info
.default_offset
)
1899 mddev
->bitmap_info
.offset
= offset
;
1901 mddev
->pers
->quiesce(mddev
, 1);
1902 rv
= bitmap_create(mddev
);
1904 rv
= bitmap_load(mddev
);
1906 bitmap_destroy(mddev
);
1907 mddev
->bitmap_info
.offset
= 0;
1909 mddev
->pers
->quiesce(mddev
, 0);
1915 if (!mddev
->external
) {
1916 /* Ensure new bitmap info is stored in
1917 * metadata promptly.
1919 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1920 md_wakeup_thread(mddev
->thread
);
1925 static struct md_sysfs_entry bitmap_location
=
1926 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
1929 timeout_show(struct mddev
*mddev
, char *page
)
1932 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
1933 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
1935 len
= sprintf(page
, "%lu", secs
);
1937 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
1938 len
+= sprintf(page
+len
, "\n");
1943 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1945 /* timeout can be set at any time */
1946 unsigned long timeout
;
1947 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
1951 /* just to make sure we don't overflow... */
1952 if (timeout
>= LONG_MAX
/ HZ
)
1955 timeout
= timeout
* HZ
/ 10000;
1957 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
1958 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
1961 mddev
->bitmap_info
.daemon_sleep
= timeout
;
1962 if (mddev
->thread
) {
1963 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1964 * the bitmap is all clean and we don't need to
1965 * adjust the timeout right now
1967 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
1968 mddev
->thread
->timeout
= timeout
;
1969 md_wakeup_thread(mddev
->thread
);
1975 static struct md_sysfs_entry bitmap_timeout
=
1976 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
1979 backlog_show(struct mddev
*mddev
, char *page
)
1981 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
1985 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1987 unsigned long backlog
;
1988 int rv
= strict_strtoul(buf
, 10, &backlog
);
1991 if (backlog
> COUNTER_MAX
)
1993 mddev
->bitmap_info
.max_write_behind
= backlog
;
1997 static struct md_sysfs_entry bitmap_backlog
=
1998 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2001 chunksize_show(struct mddev
*mddev
, char *page
)
2003 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2007 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2009 /* Can only be changed when no bitmap is active */
2011 unsigned long csize
;
2014 rv
= strict_strtoul(buf
, 10, &csize
);
2018 !is_power_of_2(csize
))
2020 mddev
->bitmap_info
.chunksize
= csize
;
2024 static struct md_sysfs_entry bitmap_chunksize
=
2025 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2027 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2029 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2030 ? "external" : "internal"));
2033 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2035 if (mddev
->bitmap
||
2036 mddev
->bitmap_info
.file
||
2037 mddev
->bitmap_info
.offset
)
2039 if (strncmp(buf
, "external", 8) == 0)
2040 mddev
->bitmap_info
.external
= 1;
2041 else if (strncmp(buf
, "internal", 8) == 0)
2042 mddev
->bitmap_info
.external
= 0;
2048 static struct md_sysfs_entry bitmap_metadata
=
2049 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2051 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2055 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2058 len
= sprintf(page
, "\n");
2062 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2064 if (mddev
->bitmap
== NULL
)
2066 if (strncmp(buf
, "false", 5) == 0)
2067 mddev
->bitmap
->need_sync
= 1;
2068 else if (strncmp(buf
, "true", 4) == 0) {
2069 if (mddev
->degraded
)
2071 mddev
->bitmap
->need_sync
= 0;
2077 static struct md_sysfs_entry bitmap_can_clear
=
2078 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2081 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2083 if (mddev
->bitmap
== NULL
)
2084 return sprintf(page
, "0\n");
2085 return sprintf(page
, "%lu\n",
2086 mddev
->bitmap
->behind_writes_used
);
2090 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2093 mddev
->bitmap
->behind_writes_used
= 0;
2097 static struct md_sysfs_entry max_backlog_used
=
2098 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2099 behind_writes_used_show
, behind_writes_used_reset
);
2101 static struct attribute
*md_bitmap_attrs
[] = {
2102 &bitmap_location
.attr
,
2103 &bitmap_timeout
.attr
,
2104 &bitmap_backlog
.attr
,
2105 &bitmap_chunksize
.attr
,
2106 &bitmap_metadata
.attr
,
2107 &bitmap_can_clear
.attr
,
2108 &max_backlog_used
.attr
,
2111 struct attribute_group md_bitmap_group
= {
2113 .attrs
= md_bitmap_attrs
,