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 int 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
;
142 rdev_for_each(rdev
, mddev
) {
143 if (! test_bit(In_sync
, &rdev
->flags
)
144 || test_bit(Faulty
, &rdev
->flags
))
147 target
= offset
+ index
* (PAGE_SIZE
/512);
149 if (sync_page_io(rdev
, target
,
150 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
159 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
161 /* Iterate the disks of an mddev, using rcu to protect access to the
162 * linked list, and raising the refcount of devices we return to ensure
163 * they don't disappear while in use.
164 * As devices are only added or removed when raid_disk is < 0 and
165 * nr_pending is 0 and In_sync is clear, the entries we return will
166 * still be in the same position on the list when we re-enter
167 * list_for_each_continue_rcu.
169 struct list_head
*pos
;
172 /* start at the beginning */
175 /* release the previous rdev and start from there. */
176 rdev_dec_pending(rdev
, mddev
);
177 pos
= &rdev
->same_set
;
179 list_for_each_continue_rcu(pos
, &mddev
->disks
) {
180 rdev
= list_entry(pos
, struct md_rdev
, same_set
);
181 if (rdev
->raid_disk
>= 0 &&
182 !test_bit(Faulty
, &rdev
->flags
)) {
183 /* this is a usable devices */
184 atomic_inc(&rdev
->nr_pending
);
193 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
195 struct md_rdev
*rdev
= NULL
;
196 struct block_device
*bdev
;
197 struct mddev
*mddev
= bitmap
->mddev
;
198 struct bitmap_storage
*store
= &bitmap
->storage
;
200 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
201 int size
= PAGE_SIZE
;
202 loff_t offset
= mddev
->bitmap_info
.offset
;
204 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
206 if (page
->index
== store
->file_pages
-1) {
207 int last_page_size
= store
->bytes
& (PAGE_SIZE
-1);
208 if (last_page_size
== 0)
209 last_page_size
= PAGE_SIZE
;
210 size
= roundup(last_page_size
,
211 bdev_logical_block_size(bdev
));
213 /* Just make sure we aren't corrupting data or
216 if (mddev
->external
) {
217 /* Bitmap could be anywhere. */
218 if (rdev
->sb_start
+ offset
+ (page
->index
222 rdev
->sb_start
+ offset
223 < (rdev
->data_offset
+ mddev
->dev_sectors
226 } else if (offset
< 0) {
227 /* DATA BITMAP METADATA */
229 + (long)(page
->index
* (PAGE_SIZE
/512))
231 /* bitmap runs in to metadata */
233 if (rdev
->data_offset
+ mddev
->dev_sectors
234 > rdev
->sb_start
+ offset
)
235 /* data runs in to bitmap */
237 } else if (rdev
->sb_start
< rdev
->data_offset
) {
238 /* METADATA BITMAP DATA */
241 + page
->index
*(PAGE_SIZE
/512) + size
/512
243 /* bitmap runs in to data */
246 /* DATA METADATA BITMAP - no problems */
248 md_super_write(mddev
, rdev
,
249 rdev
->sb_start
+ offset
250 + page
->index
* (PAGE_SIZE
/512),
256 md_super_wait(mddev
);
263 static void bitmap_file_kick(struct bitmap
*bitmap
);
265 * write out a page to a file
267 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
269 struct buffer_head
*bh
;
271 if (bitmap
->storage
.file
== NULL
) {
272 switch (write_sb_page(bitmap
, page
, wait
)) {
274 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
278 bh
= page_buffers(page
);
280 while (bh
&& bh
->b_blocknr
) {
281 atomic_inc(&bitmap
->pending_writes
);
282 set_buffer_locked(bh
);
283 set_buffer_mapped(bh
);
284 submit_bh(WRITE
| REQ_SYNC
, bh
);
285 bh
= bh
->b_this_page
;
289 wait_event(bitmap
->write_wait
,
290 atomic_read(&bitmap
->pending_writes
)==0);
292 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
293 bitmap_file_kick(bitmap
);
296 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
298 struct bitmap
*bitmap
= bh
->b_private
;
302 spin_lock_irqsave(&bitmap
->lock
, flags
);
303 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
304 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
306 if (atomic_dec_and_test(&bitmap
->pending_writes
))
307 wake_up(&bitmap
->write_wait
);
310 /* copied from buffer.c */
312 __clear_page_buffers(struct page
*page
)
314 ClearPagePrivate(page
);
315 set_page_private(page
, 0);
316 page_cache_release(page
);
318 static void free_buffers(struct page
*page
)
320 struct buffer_head
*bh
;
322 if (!PagePrivate(page
))
325 bh
= page_buffers(page
);
327 struct buffer_head
*next
= bh
->b_this_page
;
328 free_buffer_head(bh
);
331 __clear_page_buffers(page
);
335 /* read a page from a file.
336 * We both read the page, and attach buffers to the page to record the
337 * address of each block (using bmap). These addresses will be used
338 * to write the block later, completely bypassing the filesystem.
339 * This usage is similar to how swap files are handled, and allows us
340 * to write to a file with no concerns of memory allocation failing.
342 static int read_page(struct file
*file
, unsigned long index
,
343 struct bitmap
*bitmap
,
348 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
349 struct buffer_head
*bh
;
352 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
353 (unsigned long long)index
<< PAGE_SHIFT
);
355 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
360 attach_page_buffers(page
, bh
);
361 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
366 bh
->b_blocknr
= bmap(inode
, block
);
367 if (bh
->b_blocknr
== 0) {
368 /* Cannot use this file! */
372 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
373 if (count
< (1<<inode
->i_blkbits
))
376 count
-= (1<<inode
->i_blkbits
);
378 bh
->b_end_io
= end_bitmap_write
;
379 bh
->b_private
= bitmap
;
380 atomic_inc(&bitmap
->pending_writes
);
381 set_buffer_locked(bh
);
382 set_buffer_mapped(bh
);
386 bh
= bh
->b_this_page
;
390 wait_event(bitmap
->write_wait
,
391 atomic_read(&bitmap
->pending_writes
)==0);
392 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
396 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %d\n",
398 (unsigned long long)index
<< PAGE_SHIFT
,
404 * bitmap file superblock operations
407 /* update the event counter and sync the superblock to disk */
408 void bitmap_update_sb(struct bitmap
*bitmap
)
412 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
414 if (bitmap
->mddev
->bitmap_info
.external
)
416 if (!bitmap
->storage
.sb_page
) /* no superblock */
418 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
419 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
420 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
421 /* rocking back to read-only */
422 bitmap
->events_cleared
= bitmap
->mddev
->events
;
423 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
424 sb
->state
= cpu_to_le32(bitmap
->flags
);
425 /* Just in case these have been changed via sysfs: */
426 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
427 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
429 write_page(bitmap
, bitmap
->storage
.sb_page
, 1);
432 /* print out the bitmap file superblock */
433 void bitmap_print_sb(struct bitmap
*bitmap
)
437 if (!bitmap
|| !bitmap
->storage
.sb_page
)
439 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
440 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
441 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
442 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
443 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
444 *(__u32
*)(sb
->uuid
+0),
445 *(__u32
*)(sb
->uuid
+4),
446 *(__u32
*)(sb
->uuid
+8),
447 *(__u32
*)(sb
->uuid
+12));
448 printk(KERN_DEBUG
" events: %llu\n",
449 (unsigned long long) le64_to_cpu(sb
->events
));
450 printk(KERN_DEBUG
"events cleared: %llu\n",
451 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
452 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
453 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
454 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
455 printk(KERN_DEBUG
" sync size: %llu KB\n",
456 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
457 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
465 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
466 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
467 * This function verifies 'bitmap_info' and populates the on-disk bitmap
468 * structure, which is to be written to disk.
470 * Returns: 0 on success, -Exxx on error
472 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
475 unsigned long chunksize
, daemon_sleep
, write_behind
;
478 bitmap
->storage
.sb_page
= alloc_page(GFP_KERNEL
);
479 if (IS_ERR(bitmap
->storage
.sb_page
)) {
480 err
= PTR_ERR(bitmap
->storage
.sb_page
);
481 bitmap
->storage
.sb_page
= NULL
;
484 bitmap
->storage
.sb_page
->index
= 0;
486 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
488 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
489 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
491 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
493 if (!is_power_of_2(chunksize
)) {
495 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
498 sb
->chunksize
= cpu_to_le32(chunksize
);
500 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
502 (daemon_sleep
< 1) || (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
503 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
504 daemon_sleep
= 5 * HZ
;
506 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
507 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
510 * FIXME: write_behind for RAID1. If not specified, what
511 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
513 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
514 if (write_behind
> COUNTER_MAX
)
515 write_behind
= COUNTER_MAX
/ 2;
516 sb
->write_behind
= cpu_to_le32(write_behind
);
517 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
519 /* keep the array size field of the bitmap superblock up to date */
520 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
522 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
524 bitmap
->flags
|= BITMAP_STALE
;
525 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
526 bitmap
->events_cleared
= bitmap
->mddev
->events
;
527 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
534 /* read the superblock from the bitmap file and initialize some bitmap fields */
535 static int bitmap_read_sb(struct bitmap
*bitmap
)
539 unsigned long chunksize
, daemon_sleep
, write_behind
;
540 unsigned long long events
;
542 struct page
*sb_page
;
544 if (!bitmap
->storage
.file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
545 chunksize
= 128 * 1024 * 1024;
546 daemon_sleep
= 5 * HZ
;
548 bitmap
->flags
= BITMAP_STALE
;
552 /* page 0 is the superblock, read it... */
553 sb_page
= alloc_page(GFP_KERNEL
);
556 bitmap
->storage
.sb_page
= sb_page
;
558 if (bitmap
->storage
.file
) {
559 loff_t isize
= i_size_read(bitmap
->storage
.file
->f_mapping
->host
);
560 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
562 err
= read_page(bitmap
->storage
.file
, 0,
563 bitmap
, bytes
, sb_page
);
565 err
= read_sb_page(bitmap
->mddev
,
566 bitmap
->mddev
->bitmap_info
.offset
,
568 0, sizeof(bitmap_super_t
));
573 sb
= kmap_atomic(sb_page
);
575 chunksize
= le32_to_cpu(sb
->chunksize
);
576 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
577 write_behind
= le32_to_cpu(sb
->write_behind
);
579 /* verify that the bitmap-specific fields are valid */
580 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
581 reason
= "bad magic";
582 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
583 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
584 reason
= "unrecognized superblock version";
585 else if (chunksize
< 512)
586 reason
= "bitmap chunksize too small";
587 else if (!is_power_of_2(chunksize
))
588 reason
= "bitmap chunksize not a power of 2";
589 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
590 reason
= "daemon sleep period out of range";
591 else if (write_behind
> COUNTER_MAX
)
592 reason
= "write-behind limit out of range (0 - 16383)";
594 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
595 bmname(bitmap
), reason
);
599 /* keep the array size field of the bitmap superblock up to date */
600 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
602 if (bitmap
->mddev
->persistent
) {
604 * We have a persistent array superblock, so compare the
605 * bitmap's UUID and event counter to the mddev's
607 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
609 "%s: bitmap superblock UUID mismatch\n",
613 events
= le64_to_cpu(sb
->events
);
614 if (events
< bitmap
->mddev
->events
) {
616 "%s: bitmap file is out of date (%llu < %llu) "
617 "-- forcing full recovery\n",
618 bmname(bitmap
), events
,
619 (unsigned long long) bitmap
->mddev
->events
);
620 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
624 /* assign fields using values from superblock */
625 bitmap
->flags
|= le32_to_cpu(sb
->state
);
626 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
627 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
628 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
633 if (bitmap
->flags
& BITMAP_STALE
)
634 bitmap
->events_cleared
= bitmap
->mddev
->events
;
635 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
636 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
637 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
639 bitmap_print_sb(bitmap
);
643 enum bitmap_mask_op
{
648 /* record the state of the bitmap in the superblock. Return the old value */
649 static int bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
650 enum bitmap_mask_op op
)
655 if (!bitmap
->storage
.sb_page
) /* can't set the state */
657 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
658 old
= le32_to_cpu(sb
->state
) & bits
;
661 sb
->state
|= cpu_to_le32(bits
);
662 bitmap
->flags
|= bits
;
665 sb
->state
&= cpu_to_le32(~bits
);
666 bitmap
->flags
&= ~bits
;
676 * general bitmap file operations
682 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
683 * file a page at a time. There's a superblock at the start of the file.
685 /* calculate the index of the page that contains this bit */
686 static inline unsigned long file_page_index(struct bitmap_storage
*store
,
690 chunk
+= sizeof(bitmap_super_t
) << 3;
691 return chunk
>> PAGE_BIT_SHIFT
;
694 /* calculate the (bit) offset of this bit within a page */
695 static inline unsigned long file_page_offset(struct bitmap_storage
*store
,
699 chunk
+= sizeof(bitmap_super_t
) << 3;
700 return chunk
& (PAGE_BITS
- 1);
704 * return a pointer to the page in the filemap that contains the given bit
706 * this lookup is complicated by the fact that the bitmap sb might be exactly
707 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
710 static inline struct page
*filemap_get_page(struct bitmap_storage
*store
,
713 if (file_page_index(store
, chunk
) >= store
->file_pages
)
715 return store
->filemap
[file_page_index(store
, chunk
)
716 - file_page_index(store
, 0)];
719 static void bitmap_file_unmap(struct bitmap
*bitmap
)
721 struct page
**map
, *sb_page
;
725 struct bitmap_storage
*store
= &bitmap
->storage
;
727 spin_lock_irqsave(&bitmap
->lock
, flags
);
728 map
= store
->filemap
;
729 store
->filemap
= NULL
;
730 attr
= store
->filemap_attr
;
731 store
->filemap_attr
= NULL
;
732 pages
= store
->file_pages
;
733 store
->file_pages
= 0;
734 sb_page
= store
->sb_page
;
735 store
->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
);
754 file
= bitmap
->storage
.file
;
755 bitmap
->storage
.file
= NULL
;
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
);
782 if (bitmap
->storage
.file
) {
783 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
785 ptr
= d_path(&bitmap
->storage
.file
->f_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
, int pnum
,
812 enum bitmap_page_attr attr
)
814 __set_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
817 static inline void clear_page_attr(struct bitmap
*bitmap
, int pnum
,
818 enum bitmap_page_attr attr
)
820 __clear_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
823 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, int pnum
,
824 enum bitmap_page_attr attr
)
826 return test_bit((pnum
<<2) + attr
, bitmap
->storage
.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 page
= filemap_get_page(&bitmap
->storage
, chunk
);
846 bit
= file_page_offset(&bitmap
->storage
, chunk
);
849 kaddr
= kmap_atomic(page
);
850 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
853 __set_bit_le(bit
, kaddr
);
854 kunmap_atomic(kaddr
);
855 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
856 /* record page number so it gets flushed to disk when unplug occurs */
857 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_DIRTY
);
860 static void bitmap_file_clear_bit(struct bitmap
*bitmap
, sector_t block
)
865 unsigned long chunk
= block
>> bitmap
->chunkshift
;
867 page
= filemap_get_page(&bitmap
->storage
, chunk
);
870 bit
= file_page_offset(&bitmap
->storage
, chunk
);
871 paddr
= kmap_atomic(page
);
872 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
873 clear_bit(bit
, paddr
);
875 __clear_bit_le(bit
, paddr
);
876 kunmap_atomic(paddr
);
877 if (!test_page_attr(bitmap
, page
->index
, BITMAP_PAGE_NEEDWRITE
)) {
878 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_PENDING
);
879 bitmap
->allclean
= 0;
883 /* this gets called when the md device is ready to unplug its underlying
884 * (slave) device queues -- before we let any writes go down, we need to
885 * sync the dirty pages of the bitmap file to disk */
886 void bitmap_unplug(struct bitmap
*bitmap
)
888 unsigned long i
, flags
;
889 int dirty
, need_write
;
892 if (!bitmap
|| !bitmap
->storage
.filemap
)
895 /* look at each page to see if there are any set bits that need to be
896 * flushed out to disk */
897 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++) {
898 spin_lock_irqsave(&bitmap
->lock
, flags
);
899 if (!bitmap
->storage
.filemap
) {
900 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
903 dirty
= test_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
904 need_write
= test_page_attr(bitmap
, i
, BITMAP_PAGE_NEEDWRITE
);
905 clear_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
906 clear_page_attr(bitmap
, i
, BITMAP_PAGE_NEEDWRITE
);
907 if (dirty
|| need_write
)
908 clear_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
);
911 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
913 if (dirty
|| need_write
)
914 write_page(bitmap
, bitmap
->storage
.filemap
[i
], 0);
916 if (wait
) { /* if any writes were performed, we need to wait on them */
917 if (bitmap
->storage
.file
)
918 wait_event(bitmap
->write_wait
,
919 atomic_read(&bitmap
->pending_writes
)==0);
921 md_super_wait(bitmap
->mddev
);
923 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
924 bitmap_file_kick(bitmap
);
926 EXPORT_SYMBOL(bitmap_unplug
);
928 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
929 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
930 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
931 * memory mapping of the bitmap file
933 * if there's no bitmap file, or if the bitmap file had been
934 * previously kicked from the array, we mark all the bits as
935 * 1's in order to cause a full resync.
937 * We ignore all bits for sectors that end earlier than 'start'.
938 * This is used when reading an out-of-date bitmap...
940 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
942 unsigned long i
, chunks
, index
, oldindex
, bit
;
944 struct page
*page
= NULL
;
945 unsigned long num_pages
, bit_cnt
= 0;
947 unsigned long bytes
, offset
;
951 struct bitmap_storage
*store
= &bitmap
->storage
;
953 chunks
= bitmap
->chunks
;
956 if (!file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
957 /* No permanent bitmap - fill with '1s'. */
958 store
->filemap
= NULL
;
959 store
->file_pages
= 0;
960 for (i
= 0; i
< chunks
; i
++) {
961 /* if the disk bit is set, set the memory bit */
962 int needed
= ((sector_t
)(i
+1) << (bitmap
->chunkshift
)
964 bitmap_set_memory_bits(bitmap
,
965 (sector_t
)i
<< bitmap
->chunkshift
,
971 outofdate
= bitmap
->flags
& BITMAP_STALE
;
973 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
974 "recovery\n", bmname(bitmap
));
976 bytes
= DIV_ROUND_UP(bitmap
->chunks
, 8);
977 if (!bitmap
->mddev
->bitmap_info
.external
)
978 bytes
+= sizeof(bitmap_super_t
);
980 store
->bytes
= bytes
;
982 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
984 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
) {
985 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
987 (unsigned long) i_size_read(file
->f_mapping
->host
),
994 store
->filemap
= kmalloc(sizeof(struct page
*)
995 * num_pages
, GFP_KERNEL
);
1001 if (store
->sb_page
) {
1002 store
->filemap
[0] = store
->sb_page
;
1004 offset
= sizeof(bitmap_super_t
);
1006 for ( ; pnum
< num_pages
; pnum
++) {
1007 store
->filemap
[pnum
] = alloc_page(GFP_KERNEL
);
1008 if (!store
->filemap
[pnum
]) {
1009 store
->file_pages
= pnum
;
1013 store
->file_pages
= pnum
;
1015 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
1016 store
->filemap_attr
= kzalloc(
1017 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
1019 if (!store
->filemap_attr
)
1024 for (i
= 0; i
< chunks
; i
++) {
1026 index
= file_page_index(&bitmap
->storage
, i
);
1027 bit
= file_page_offset(&bitmap
->storage
, i
);
1028 if (index
!= oldindex
) { /* this is a new page, read it in */
1030 /* unmap the old page, we're done with it */
1031 if (index
== num_pages
-1)
1032 count
= bytes
- index
* PAGE_SIZE
;
1035 page
= store
->filemap
[index
];
1037 ret
= read_page(file
, index
, bitmap
,
1042 bitmap
->mddev
->bitmap_info
.offset
,
1053 * if bitmap is out of date, dirty the
1054 * whole page and write it out
1056 paddr
= kmap_atomic(page
);
1057 memset(paddr
+ offset
, 0xff,
1058 PAGE_SIZE
- offset
);
1059 kunmap_atomic(paddr
);
1060 write_page(bitmap
, page
, 1);
1063 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1067 paddr
= kmap_atomic(page
);
1068 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1069 b
= test_bit(bit
, paddr
);
1071 b
= test_bit_le(bit
, paddr
);
1072 kunmap_atomic(paddr
);
1074 /* if the disk bit is set, set the memory bit */
1075 int needed
= ((sector_t
)(i
+1) << bitmap
->chunkshift
1077 bitmap_set_memory_bits(bitmap
,
1078 (sector_t
)i
<< bitmap
->chunkshift
,
1085 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1086 "read %lu/%lu pages, set %lu of %lu bits\n",
1087 bmname(bitmap
), store
->file_pages
,
1088 num_pages
, bit_cnt
, chunks
);
1093 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1094 bmname(bitmap
), ret
);
1098 void bitmap_write_all(struct bitmap
*bitmap
)
1100 /* We don't actually write all bitmap blocks here,
1101 * just flag them as needing to be written
1105 if (!bitmap
|| !bitmap
->storage
.filemap
)
1107 if (bitmap
->storage
.file
)
1108 /* Only one copy, so nothing needed */
1111 spin_lock_irq(&bitmap
->lock
);
1112 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1113 set_page_attr(bitmap
, i
,
1114 BITMAP_PAGE_NEEDWRITE
);
1115 bitmap
->allclean
= 0;
1116 spin_unlock_irq(&bitmap
->lock
);
1119 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
1121 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1122 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1123 bitmap
->bp
[page
].count
+= inc
;
1124 bitmap_checkfree(bitmap
, page
);
1127 static void bitmap_set_pending(struct bitmap
*bitmap
, sector_t offset
)
1129 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1130 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1131 struct bitmap_page
*bp
= &bitmap
->bp
[page
];
1137 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1138 sector_t offset
, sector_t
*blocks
,
1142 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1146 void bitmap_daemon_work(struct mddev
*mddev
)
1148 struct bitmap
*bitmap
;
1150 unsigned long nextpage
;
1151 unsigned long flags
;
1154 /* Use a mutex to guard daemon_work against
1157 mutex_lock(&mddev
->bitmap_info
.mutex
);
1158 bitmap
= mddev
->bitmap
;
1159 if (bitmap
== NULL
) {
1160 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1163 if (time_before(jiffies
, bitmap
->daemon_lastrun
1164 + mddev
->bitmap_info
.daemon_sleep
))
1167 bitmap
->daemon_lastrun
= jiffies
;
1168 if (bitmap
->allclean
) {
1169 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1172 bitmap
->allclean
= 1;
1174 /* Any file-page which is PENDING now needs to be written.
1175 * So set NEEDWRITE now, then after we make any last-minute changes
1178 spin_lock_irqsave(&bitmap
->lock
, flags
);
1179 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++)
1180 if (test_page_attr(bitmap
, j
,
1181 BITMAP_PAGE_PENDING
)) {
1182 set_page_attr(bitmap
, j
,
1183 BITMAP_PAGE_NEEDWRITE
);
1184 clear_page_attr(bitmap
, j
,
1185 BITMAP_PAGE_PENDING
);
1188 if (bitmap
->need_sync
&&
1189 mddev
->bitmap_info
.external
== 0) {
1190 /* Arrange for superblock update as well as
1193 bitmap
->need_sync
= 0;
1194 if (bitmap
->storage
.filemap
) {
1195 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
1196 sb
->events_cleared
=
1197 cpu_to_le64(bitmap
->events_cleared
);
1199 set_page_attr(bitmap
, 0,
1200 BITMAP_PAGE_NEEDWRITE
);
1203 /* Now look at the bitmap counters and if any are '2' or '1',
1204 * decrement and handle accordingly.
1207 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1208 bitmap_counter_t
*bmc
;
1209 sector_t block
= (sector_t
)j
<< bitmap
->chunkshift
;
1211 if (j
== nextpage
) {
1212 nextpage
+= PAGE_COUNTER_RATIO
;
1213 if (!bitmap
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
) {
1214 j
|= PAGE_COUNTER_MASK
;
1217 bitmap
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
= 0;
1219 bmc
= bitmap_get_counter(bitmap
,
1224 j
|= PAGE_COUNTER_MASK
;
1227 if (*bmc
== 1 && !bitmap
->need_sync
) {
1228 /* We can clear the bit */
1230 bitmap_count_page(bitmap
, block
, -1);
1231 bitmap_file_clear_bit(bitmap
, block
);
1232 } else if (*bmc
&& *bmc
<= 2) {
1234 bitmap_set_pending(bitmap
, block
);
1235 bitmap
->allclean
= 0;
1239 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1240 * DIRTY pages need to be written by bitmap_unplug so it can wait
1242 * If we find any DIRTY page we stop there and let bitmap_unplug
1243 * handle all the rest. This is important in the case where
1244 * the first blocking holds the superblock and it has been updated.
1245 * We mustn't write any other blocks before the superblock.
1247 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++) {
1249 if (test_page_attr(bitmap
, j
,
1251 /* bitmap_unplug will handle the rest */
1253 if (test_page_attr(bitmap
, j
,
1254 BITMAP_PAGE_NEEDWRITE
)) {
1255 clear_page_attr(bitmap
, j
,
1256 BITMAP_PAGE_NEEDWRITE
);
1257 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1258 write_page(bitmap
, bitmap
->storage
.filemap
[j
], 0);
1259 spin_lock_irqsave(&bitmap
->lock
, flags
);
1260 if (!bitmap
->storage
.filemap
)
1264 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1267 if (bitmap
->allclean
== 0)
1268 mddev
->thread
->timeout
=
1269 mddev
->bitmap_info
.daemon_sleep
;
1270 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1273 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1274 sector_t offset
, sector_t
*blocks
,
1276 __releases(bitmap
->lock
)
1277 __acquires(bitmap
->lock
)
1279 /* If 'create', we might release the lock and reclaim it.
1280 * The lock must have been taken with interrupts enabled.
1281 * If !create, we don't release the lock.
1283 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1284 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1285 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1289 err
= bitmap_checkpage(bitmap
, page
, create
);
1291 if (bitmap
->bp
[page
].hijacked
||
1292 bitmap
->bp
[page
].map
== NULL
)
1293 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1294 PAGE_COUNTER_SHIFT
- 1);
1296 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1297 *blocks
= csize
- (offset
& (csize
- 1));
1302 /* now locked ... */
1304 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1305 /* should we use the first or second counter field
1306 * of the hijacked pointer? */
1307 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1308 return &((bitmap_counter_t
*)
1309 &bitmap
->bp
[page
].map
)[hi
];
1310 } else /* page is allocated */
1311 return (bitmap_counter_t
*)
1312 &(bitmap
->bp
[page
].map
[pageoff
]);
1315 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1322 atomic_inc(&bitmap
->behind_writes
);
1323 bw
= atomic_read(&bitmap
->behind_writes
);
1324 if (bw
> bitmap
->behind_writes_used
)
1325 bitmap
->behind_writes_used
= bw
;
1327 pr_debug("inc write-behind count %d/%lu\n",
1328 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1333 bitmap_counter_t
*bmc
;
1335 spin_lock_irq(&bitmap
->lock
);
1336 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1338 spin_unlock_irq(&bitmap
->lock
);
1342 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1343 DEFINE_WAIT(__wait
);
1344 /* note that it is safe to do the prepare_to_wait
1345 * after the test as long as we do it before dropping
1348 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1349 TASK_UNINTERRUPTIBLE
);
1350 spin_unlock_irq(&bitmap
->lock
);
1352 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1358 bitmap_file_set_bit(bitmap
, offset
);
1359 bitmap_count_page(bitmap
, offset
, 1);
1367 spin_unlock_irq(&bitmap
->lock
);
1370 if (sectors
> blocks
)
1377 EXPORT_SYMBOL(bitmap_startwrite
);
1379 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1380 int success
, int behind
)
1385 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1386 wake_up(&bitmap
->behind_wait
);
1387 pr_debug("dec write-behind count %d/%lu\n",
1388 atomic_read(&bitmap
->behind_writes
),
1389 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1394 unsigned long flags
;
1395 bitmap_counter_t
*bmc
;
1397 spin_lock_irqsave(&bitmap
->lock
, flags
);
1398 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1400 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1404 if (success
&& !bitmap
->mddev
->degraded
&&
1405 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1406 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1407 bitmap
->need_sync
= 1;
1408 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1411 if (!success
&& !NEEDED(*bmc
))
1412 *bmc
|= NEEDED_MASK
;
1414 if (COUNTER(*bmc
) == COUNTER_MAX
)
1415 wake_up(&bitmap
->overflow_wait
);
1419 bitmap_set_pending(bitmap
, offset
);
1420 bitmap
->allclean
= 0;
1422 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1424 if (sectors
> blocks
)
1430 EXPORT_SYMBOL(bitmap_endwrite
);
1432 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1435 bitmap_counter_t
*bmc
;
1437 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1439 return 1; /* always resync if no bitmap */
1441 spin_lock_irq(&bitmap
->lock
);
1442 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1448 else if (NEEDED(*bmc
)) {
1450 if (!degraded
) { /* don't set/clear bits if degraded */
1451 *bmc
|= RESYNC_MASK
;
1452 *bmc
&= ~NEEDED_MASK
;
1456 spin_unlock_irq(&bitmap
->lock
);
1460 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1463 /* bitmap_start_sync must always report on multiples of whole
1464 * pages, otherwise resync (which is very PAGE_SIZE based) will
1466 * So call __bitmap_start_sync repeatedly (if needed) until
1467 * At least PAGE_SIZE>>9 blocks are covered.
1468 * Return the 'or' of the result.
1474 while (*blocks
< (PAGE_SIZE
>>9)) {
1475 rv
|= __bitmap_start_sync(bitmap
, offset
,
1476 &blocks1
, degraded
);
1482 EXPORT_SYMBOL(bitmap_start_sync
);
1484 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1486 bitmap_counter_t
*bmc
;
1487 unsigned long flags
;
1489 if (bitmap
== NULL
) {
1493 spin_lock_irqsave(&bitmap
->lock
, flags
);
1494 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1499 *bmc
&= ~RESYNC_MASK
;
1501 if (!NEEDED(*bmc
) && aborted
)
1502 *bmc
|= NEEDED_MASK
;
1505 bitmap_set_pending(bitmap
, offset
);
1506 bitmap
->allclean
= 0;
1511 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1513 EXPORT_SYMBOL(bitmap_end_sync
);
1515 void bitmap_close_sync(struct bitmap
*bitmap
)
1517 /* Sync has finished, and any bitmap chunks that weren't synced
1518 * properly have been aborted. It remains to us to clear the
1519 * RESYNC bit wherever it is still on
1521 sector_t sector
= 0;
1525 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1526 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1530 EXPORT_SYMBOL(bitmap_close_sync
);
1532 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
)
1540 bitmap
->last_end_sync
= jiffies
;
1543 if (time_before(jiffies
, (bitmap
->last_end_sync
1544 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1546 wait_event(bitmap
->mddev
->recovery_wait
,
1547 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1549 bitmap
->mddev
->curr_resync_completed
= sector
;
1550 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1551 sector
&= ~((1ULL << bitmap
->chunkshift
) - 1);
1553 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1554 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1557 bitmap
->last_end_sync
= jiffies
;
1558 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1560 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1562 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1564 /* For each chunk covered by any of these sectors, set the
1565 * counter to 2 and possibly set resync_needed. They should all
1566 * be 0 at this point
1570 bitmap_counter_t
*bmc
;
1571 spin_lock_irq(&bitmap
->lock
);
1572 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1574 spin_unlock_irq(&bitmap
->lock
);
1578 *bmc
= 2 | (needed
? NEEDED_MASK
: 0);
1579 bitmap_count_page(bitmap
, offset
, 1);
1580 bitmap_set_pending(bitmap
, offset
);
1581 bitmap
->allclean
= 0;
1583 spin_unlock_irq(&bitmap
->lock
);
1586 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1587 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1589 unsigned long chunk
;
1591 for (chunk
= s
; chunk
<= e
; chunk
++) {
1592 sector_t sec
= (sector_t
)chunk
<< bitmap
->chunkshift
;
1593 bitmap_set_memory_bits(bitmap
, sec
, 1);
1594 spin_lock_irq(&bitmap
->lock
);
1595 bitmap_file_set_bit(bitmap
, sec
);
1596 spin_unlock_irq(&bitmap
->lock
);
1597 if (sec
< bitmap
->mddev
->recovery_cp
)
1598 /* We are asserting that the array is dirty,
1599 * so move the recovery_cp address back so
1600 * that it is obvious that it is dirty
1602 bitmap
->mddev
->recovery_cp
= sec
;
1607 * flush out any pending updates
1609 void bitmap_flush(struct mddev
*mddev
)
1611 struct bitmap
*bitmap
= mddev
->bitmap
;
1614 if (!bitmap
) /* there was no bitmap */
1617 /* run the daemon_work three time to ensure everything is flushed
1620 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1621 bitmap
->daemon_lastrun
-= sleep
;
1622 bitmap_daemon_work(mddev
);
1623 bitmap
->daemon_lastrun
-= sleep
;
1624 bitmap_daemon_work(mddev
);
1625 bitmap
->daemon_lastrun
-= sleep
;
1626 bitmap_daemon_work(mddev
);
1627 bitmap_update_sb(bitmap
);
1631 * free memory that was allocated
1633 static void bitmap_free(struct bitmap
*bitmap
)
1635 unsigned long k
, pages
;
1636 struct bitmap_page
*bp
;
1638 if (!bitmap
) /* there was no bitmap */
1641 /* release the bitmap file and kill the daemon */
1642 bitmap_file_put(bitmap
);
1645 pages
= bitmap
->pages
;
1647 /* free all allocated memory */
1649 if (bp
) /* deallocate the page memory */
1650 for (k
= 0; k
< pages
; k
++)
1651 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1657 void bitmap_destroy(struct mddev
*mddev
)
1659 struct bitmap
*bitmap
= mddev
->bitmap
;
1661 if (!bitmap
) /* there was no bitmap */
1664 mutex_lock(&mddev
->bitmap_info
.mutex
);
1665 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1666 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1668 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1670 if (bitmap
->sysfs_can_clear
)
1671 sysfs_put(bitmap
->sysfs_can_clear
);
1673 bitmap_free(bitmap
);
1677 * initialize the bitmap structure
1678 * if this returns an error, bitmap_destroy must be called to do clean up
1680 int bitmap_create(struct mddev
*mddev
)
1682 struct bitmap
*bitmap
;
1683 sector_t blocks
= mddev
->resync_max_sectors
;
1684 unsigned long chunks
;
1685 unsigned long pages
;
1686 struct file
*file
= mddev
->bitmap_info
.file
;
1688 struct sysfs_dirent
*bm
= NULL
;
1690 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1692 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1694 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1698 spin_lock_init(&bitmap
->lock
);
1699 atomic_set(&bitmap
->pending_writes
, 0);
1700 init_waitqueue_head(&bitmap
->write_wait
);
1701 init_waitqueue_head(&bitmap
->overflow_wait
);
1702 init_waitqueue_head(&bitmap
->behind_wait
);
1704 bitmap
->mddev
= mddev
;
1707 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "bitmap");
1709 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, NULL
, "can_clear");
1712 bitmap
->sysfs_can_clear
= NULL
;
1714 bitmap
->storage
.file
= file
;
1717 /* As future accesses to this file will use bmap,
1718 * and bypass the page cache, we must sync the file
1723 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1724 if (!mddev
->bitmap_info
.external
) {
1726 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1727 * instructing us to create a new on-disk bitmap instance.
1729 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1730 err
= bitmap_new_disk_sb(bitmap
);
1732 err
= bitmap_read_sb(bitmap
);
1735 if (mddev
->bitmap_info
.chunksize
== 0 ||
1736 mddev
->bitmap_info
.daemon_sleep
== 0)
1737 /* chunksize and time_base need to be
1744 bitmap
->daemon_lastrun
= jiffies
;
1745 bitmap
->chunkshift
= (ffz(~mddev
->bitmap_info
.chunksize
)
1746 - BITMAP_BLOCK_SHIFT
);
1748 chunks
= (blocks
+ (1 << bitmap
->chunkshift
) - 1) >>
1750 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1754 bitmap
->chunks
= chunks
;
1755 bitmap
->pages
= pages
;
1756 bitmap
->missing_pages
= pages
;
1758 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1764 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1765 pages
, bmname(bitmap
));
1767 mddev
->bitmap
= bitmap
;
1770 return (bitmap
->flags
& BITMAP_WRITE_ERROR
) ? -EIO
: 0;
1773 bitmap_free(bitmap
);
1777 int bitmap_load(struct mddev
*mddev
)
1781 sector_t sector
= 0;
1782 struct bitmap
*bitmap
= mddev
->bitmap
;
1787 /* Clear out old bitmap info first: Either there is none, or we
1788 * are resuming after someone else has possibly changed things,
1789 * so we should forget old cached info.
1790 * All chunks should be clean, but some might need_sync.
1792 while (sector
< mddev
->resync_max_sectors
) {
1794 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1797 bitmap_close_sync(bitmap
);
1799 if (mddev
->degraded
== 0
1800 || bitmap
->events_cleared
== mddev
->events
)
1801 /* no need to keep dirty bits to optimise a
1802 * re-add of a missing device */
1803 start
= mddev
->recovery_cp
;
1805 mutex_lock(&mddev
->bitmap_info
.mutex
);
1806 err
= bitmap_init_from_disk(bitmap
, start
);
1807 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1811 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
1813 /* Kick recovery in case any bits were set */
1814 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1816 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1817 md_wakeup_thread(mddev
->thread
);
1819 bitmap_update_sb(bitmap
);
1821 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1826 EXPORT_SYMBOL_GPL(bitmap_load
);
1828 void bitmap_status(struct seq_file
*seq
, struct bitmap
*bitmap
)
1830 unsigned long chunk_kb
;
1831 unsigned long flags
;
1836 spin_lock_irqsave(&bitmap
->lock
, flags
);
1837 chunk_kb
= bitmap
->mddev
->bitmap_info
.chunksize
>> 10;
1838 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
1840 bitmap
->pages
- bitmap
->missing_pages
,
1842 (bitmap
->pages
- bitmap
->missing_pages
)
1843 << (PAGE_SHIFT
- 10),
1844 chunk_kb
? chunk_kb
: bitmap
->mddev
->bitmap_info
.chunksize
,
1845 chunk_kb
? "KB" : "B");
1846 if (bitmap
->storage
.file
) {
1847 seq_printf(seq
, ", file: ");
1848 seq_path(seq
, &bitmap
->storage
.file
->f_path
, " \t\n");
1851 seq_printf(seq
, "\n");
1852 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1856 location_show(struct mddev
*mddev
, char *page
)
1859 if (mddev
->bitmap_info
.file
)
1860 len
= sprintf(page
, "file");
1861 else if (mddev
->bitmap_info
.offset
)
1862 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
1864 len
= sprintf(page
, "none");
1865 len
+= sprintf(page
+len
, "\n");
1870 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1874 if (!mddev
->pers
->quiesce
)
1876 if (mddev
->recovery
|| mddev
->sync_thread
)
1880 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
1881 mddev
->bitmap_info
.offset
) {
1882 /* bitmap already configured. Only option is to clear it */
1883 if (strncmp(buf
, "none", 4) != 0)
1886 mddev
->pers
->quiesce(mddev
, 1);
1887 bitmap_destroy(mddev
);
1888 mddev
->pers
->quiesce(mddev
, 0);
1890 mddev
->bitmap_info
.offset
= 0;
1891 if (mddev
->bitmap_info
.file
) {
1892 struct file
*f
= mddev
->bitmap_info
.file
;
1893 mddev
->bitmap_info
.file
= NULL
;
1894 restore_bitmap_write_access(f
);
1898 /* No bitmap, OK to set a location */
1900 if (strncmp(buf
, "none", 4) == 0)
1901 /* nothing to be done */;
1902 else if (strncmp(buf
, "file:", 5) == 0) {
1903 /* Not supported yet */
1908 rv
= strict_strtoll(buf
+1, 10, &offset
);
1910 rv
= strict_strtoll(buf
, 10, &offset
);
1915 if (mddev
->bitmap_info
.external
== 0 &&
1916 mddev
->major_version
== 0 &&
1917 offset
!= mddev
->bitmap_info
.default_offset
)
1919 mddev
->bitmap_info
.offset
= offset
;
1921 mddev
->pers
->quiesce(mddev
, 1);
1922 rv
= bitmap_create(mddev
);
1924 rv
= bitmap_load(mddev
);
1926 bitmap_destroy(mddev
);
1927 mddev
->bitmap_info
.offset
= 0;
1929 mddev
->pers
->quiesce(mddev
, 0);
1935 if (!mddev
->external
) {
1936 /* Ensure new bitmap info is stored in
1937 * metadata promptly.
1939 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1940 md_wakeup_thread(mddev
->thread
);
1945 static struct md_sysfs_entry bitmap_location
=
1946 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
1948 /* 'bitmap/space' is the space available at 'location' for the
1949 * bitmap. This allows the kernel to know when it is safe to
1950 * resize the bitmap to match a resized array.
1953 space_show(struct mddev
*mddev
, char *page
)
1955 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.space
);
1959 space_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1961 unsigned long sectors
;
1964 rv
= kstrtoul(buf
, 10, §ors
);
1971 if (mddev
->bitmap
&&
1972 sectors
< (mddev
->bitmap
->storage
.bytes
+ 511) >> 9)
1973 return -EFBIG
; /* Bitmap is too big for this small space */
1975 /* could make sure it isn't too big, but that isn't really
1976 * needed - user-space should be careful.
1978 mddev
->bitmap_info
.space
= sectors
;
1982 static struct md_sysfs_entry bitmap_space
=
1983 __ATTR(space
, S_IRUGO
|S_IWUSR
, space_show
, space_store
);
1986 timeout_show(struct mddev
*mddev
, char *page
)
1989 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
1990 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
1992 len
= sprintf(page
, "%lu", secs
);
1994 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
1995 len
+= sprintf(page
+len
, "\n");
2000 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2002 /* timeout can be set at any time */
2003 unsigned long timeout
;
2004 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
2008 /* just to make sure we don't overflow... */
2009 if (timeout
>= LONG_MAX
/ HZ
)
2012 timeout
= timeout
* HZ
/ 10000;
2014 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
2015 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2018 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2019 if (mddev
->thread
) {
2020 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2021 * the bitmap is all clean and we don't need to
2022 * adjust the timeout right now
2024 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2025 mddev
->thread
->timeout
= timeout
;
2026 md_wakeup_thread(mddev
->thread
);
2032 static struct md_sysfs_entry bitmap_timeout
=
2033 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2036 backlog_show(struct mddev
*mddev
, char *page
)
2038 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2042 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2044 unsigned long backlog
;
2045 int rv
= strict_strtoul(buf
, 10, &backlog
);
2048 if (backlog
> COUNTER_MAX
)
2050 mddev
->bitmap_info
.max_write_behind
= backlog
;
2054 static struct md_sysfs_entry bitmap_backlog
=
2055 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2058 chunksize_show(struct mddev
*mddev
, char *page
)
2060 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2064 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2066 /* Can only be changed when no bitmap is active */
2068 unsigned long csize
;
2071 rv
= strict_strtoul(buf
, 10, &csize
);
2075 !is_power_of_2(csize
))
2077 mddev
->bitmap_info
.chunksize
= csize
;
2081 static struct md_sysfs_entry bitmap_chunksize
=
2082 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2084 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2086 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2087 ? "external" : "internal"));
2090 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2092 if (mddev
->bitmap
||
2093 mddev
->bitmap_info
.file
||
2094 mddev
->bitmap_info
.offset
)
2096 if (strncmp(buf
, "external", 8) == 0)
2097 mddev
->bitmap_info
.external
= 1;
2098 else if (strncmp(buf
, "internal", 8) == 0)
2099 mddev
->bitmap_info
.external
= 0;
2105 static struct md_sysfs_entry bitmap_metadata
=
2106 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2108 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2112 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2115 len
= sprintf(page
, "\n");
2119 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2121 if (mddev
->bitmap
== NULL
)
2123 if (strncmp(buf
, "false", 5) == 0)
2124 mddev
->bitmap
->need_sync
= 1;
2125 else if (strncmp(buf
, "true", 4) == 0) {
2126 if (mddev
->degraded
)
2128 mddev
->bitmap
->need_sync
= 0;
2134 static struct md_sysfs_entry bitmap_can_clear
=
2135 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2138 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2140 if (mddev
->bitmap
== NULL
)
2141 return sprintf(page
, "0\n");
2142 return sprintf(page
, "%lu\n",
2143 mddev
->bitmap
->behind_writes_used
);
2147 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2150 mddev
->bitmap
->behind_writes_used
= 0;
2154 static struct md_sysfs_entry max_backlog_used
=
2155 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2156 behind_writes_used_show
, behind_writes_used_reset
);
2158 static struct attribute
*md_bitmap_attrs
[] = {
2159 &bitmap_location
.attr
,
2161 &bitmap_timeout
.attr
,
2162 &bitmap_backlog
.attr
,
2163 &bitmap_chunksize
.attr
,
2164 &bitmap_metadata
.attr
,
2165 &bitmap_can_clear
.attr
,
2166 &max_backlog_used
.attr
,
2169 struct attribute_group md_bitmap_group
= {
2171 .attrs
= md_bitmap_attrs
,