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>
32 #include <linux/dm-dirty-log.h>
38 /* these are for debugging purposes only! */
40 /* define one and only one of these */
41 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
42 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
43 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
44 #define INJECT_FAULTS_4 0 /* undef */
45 #define INJECT_FAULTS_5 0 /* undef */
46 #define INJECT_FAULTS_6 0
48 /* if these are defined, the driver will fail! debug only */
49 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
50 #define INJECT_FATAL_FAULT_2 0 /* undef */
51 #define INJECT_FATAL_FAULT_3 0 /* undef */
56 # define PRINTK(x...) printk(KERN_DEBUG x)
62 static inline char *bmname(struct bitmap
*bitmap
)
64 return bitmap
->mddev
? mdname(bitmap
->mddev
) : "mdX";
68 * just a placeholder - calls kmalloc for bitmap pages
70 static unsigned char *bitmap_alloc_page(struct bitmap
*bitmap
)
74 #ifdef INJECT_FAULTS_1
77 page
= kzalloc(PAGE_SIZE
, GFP_NOIO
);
80 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap
));
82 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
83 bmname(bitmap
), page
);
88 * for now just a placeholder -- just calls kfree for bitmap pages
90 static void bitmap_free_page(struct bitmap
*bitmap
, unsigned char *page
)
92 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap
), page
);
97 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
99 * 1) check to see if this page is allocated, if it's not then try to alloc
100 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
101 * page pointer directly as a counter
103 * if we find our page, we increment the page's refcount so that it stays
104 * allocated while we're using it
106 static int bitmap_checkpage(struct bitmap
*bitmap
,
107 unsigned long page
, int create
)
108 __releases(bitmap
->lock
)
109 __acquires(bitmap
->lock
)
111 unsigned char *mappage
;
113 if (page
>= bitmap
->pages
) {
114 /* This can happen if bitmap_start_sync goes beyond
115 * End-of-device while looking for a whole page.
121 if (bitmap
->bp
[page
].hijacked
) /* it's hijacked, don't try to alloc */
124 if (bitmap
->bp
[page
].map
) /* page is already allocated, just return */
130 /* this page has not been allocated yet */
132 spin_unlock_irq(&bitmap
->lock
);
133 mappage
= bitmap_alloc_page(bitmap
);
134 spin_lock_irq(&bitmap
->lock
);
136 if (mappage
== NULL
) {
137 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
139 /* failed - set the hijacked flag so that we can use the
140 * pointer as a counter */
141 if (!bitmap
->bp
[page
].map
)
142 bitmap
->bp
[page
].hijacked
= 1;
143 } else if (bitmap
->bp
[page
].map
||
144 bitmap
->bp
[page
].hijacked
) {
145 /* somebody beat us to getting the page */
146 bitmap_free_page(bitmap
, mappage
);
150 /* no page was in place and we have one, so install it */
152 bitmap
->bp
[page
].map
= mappage
;
153 bitmap
->missing_pages
--;
158 /* if page is completely empty, put it back on the free list, or dealloc it */
159 /* if page was hijacked, unmark the flag so it might get alloced next time */
160 /* Note: lock should be held when calling this */
161 static void bitmap_checkfree(struct bitmap
*bitmap
, unsigned long page
)
165 if (bitmap
->bp
[page
].count
) /* page is still busy */
168 /* page is no longer in use, it can be released */
170 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
171 bitmap
->bp
[page
].hijacked
= 0;
172 bitmap
->bp
[page
].map
= NULL
;
174 /* normal case, free the page */
175 ptr
= bitmap
->bp
[page
].map
;
176 bitmap
->bp
[page
].map
= NULL
;
177 bitmap
->missing_pages
++;
178 bitmap_free_page(bitmap
, ptr
);
183 * bitmap file handling - read and write the bitmap file and its superblock
187 * basic page I/O operations
190 /* IO operations when bitmap is stored near all superblocks */
191 static struct page
*read_sb_page(mddev_t
*mddev
, loff_t offset
,
193 unsigned long index
, int size
)
195 /* choose a good rdev and read the page from there */
202 page
= alloc_page(GFP_KERNEL
);
204 return ERR_PTR(-ENOMEM
);
208 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
209 if (! test_bit(In_sync
, &rdev
->flags
)
210 || test_bit(Faulty
, &rdev
->flags
))
213 target
= offset
+ index
* (PAGE_SIZE
/512);
215 if (sync_page_io(rdev
, target
,
216 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
219 attach_page_buffers(page
, NULL
); /* so that free_buffer will
226 return ERR_PTR(-EIO
);
230 static mdk_rdev_t
*next_active_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
232 /* Iterate the disks of an mddev, using rcu to protect access to the
233 * linked list, and raising the refcount of devices we return to ensure
234 * they don't disappear while in use.
235 * As devices are only added or removed when raid_disk is < 0 and
236 * nr_pending is 0 and In_sync is clear, the entries we return will
237 * still be in the same position on the list when we re-enter
238 * list_for_each_continue_rcu.
240 struct list_head
*pos
;
243 /* start at the beginning */
246 /* release the previous rdev and start from there. */
247 rdev_dec_pending(rdev
, mddev
);
248 pos
= &rdev
->same_set
;
250 list_for_each_continue_rcu(pos
, &mddev
->disks
) {
251 rdev
= list_entry(pos
, mdk_rdev_t
, same_set
);
252 if (rdev
->raid_disk
>= 0 &&
253 !test_bit(Faulty
, &rdev
->flags
)) {
254 /* this is a usable devices */
255 atomic_inc(&rdev
->nr_pending
);
264 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
266 mdk_rdev_t
*rdev
= NULL
;
267 struct block_device
*bdev
;
268 mddev_t
*mddev
= bitmap
->mddev
;
270 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
271 int size
= PAGE_SIZE
;
272 loff_t offset
= mddev
->bitmap_info
.offset
;
274 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
276 if (page
->index
== bitmap
->file_pages
-1)
277 size
= roundup(bitmap
->last_page_size
,
278 bdev_logical_block_size(bdev
));
279 /* Just make sure we aren't corrupting data or
282 if (mddev
->external
) {
283 /* Bitmap could be anywhere. */
284 if (rdev
->sb_start
+ offset
+ (page
->index
288 rdev
->sb_start
+ offset
289 < (rdev
->data_offset
+ mddev
->dev_sectors
292 } else if (offset
< 0) {
293 /* DATA BITMAP METADATA */
295 + (long)(page
->index
* (PAGE_SIZE
/512))
297 /* bitmap runs in to metadata */
299 if (rdev
->data_offset
+ mddev
->dev_sectors
300 > rdev
->sb_start
+ offset
)
301 /* data runs in to bitmap */
303 } else if (rdev
->sb_start
< rdev
->data_offset
) {
304 /* METADATA BITMAP DATA */
307 + page
->index
*(PAGE_SIZE
/512) + size
/512
309 /* bitmap runs in to data */
312 /* DATA METADATA BITMAP - no problems */
314 md_super_write(mddev
, rdev
,
315 rdev
->sb_start
+ offset
316 + page
->index
* (PAGE_SIZE
/512),
322 md_super_wait(mddev
);
329 static void bitmap_file_kick(struct bitmap
*bitmap
);
331 * write out a page to a file
333 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
335 struct buffer_head
*bh
;
337 if (bitmap
->file
== NULL
) {
338 switch (write_sb_page(bitmap
, page
, wait
)) {
340 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
344 bh
= page_buffers(page
);
346 while (bh
&& bh
->b_blocknr
) {
347 atomic_inc(&bitmap
->pending_writes
);
348 set_buffer_locked(bh
);
349 set_buffer_mapped(bh
);
350 submit_bh(WRITE
| REQ_SYNC
, bh
);
351 bh
= bh
->b_this_page
;
355 wait_event(bitmap
->write_wait
,
356 atomic_read(&bitmap
->pending_writes
)==0);
358 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
359 bitmap_file_kick(bitmap
);
362 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
364 struct bitmap
*bitmap
= bh
->b_private
;
368 spin_lock_irqsave(&bitmap
->lock
, flags
);
369 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
370 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
372 if (atomic_dec_and_test(&bitmap
->pending_writes
))
373 wake_up(&bitmap
->write_wait
);
376 /* copied from buffer.c */
378 __clear_page_buffers(struct page
*page
)
380 ClearPagePrivate(page
);
381 set_page_private(page
, 0);
382 page_cache_release(page
);
384 static void free_buffers(struct page
*page
)
386 struct buffer_head
*bh
= page_buffers(page
);
389 struct buffer_head
*next
= bh
->b_this_page
;
390 free_buffer_head(bh
);
393 __clear_page_buffers(page
);
397 /* read a page from a file.
398 * We both read the page, and attach buffers to the page to record the
399 * address of each block (using bmap). These addresses will be used
400 * to write the block later, completely bypassing the filesystem.
401 * This usage is similar to how swap files are handled, and allows us
402 * to write to a file with no concerns of memory allocation failing.
404 static struct page
*read_page(struct file
*file
, unsigned long index
,
405 struct bitmap
*bitmap
,
408 struct page
*page
= NULL
;
409 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
410 struct buffer_head
*bh
;
413 PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
414 (unsigned long long)index
<< PAGE_SHIFT
);
416 page
= alloc_page(GFP_KERNEL
);
418 page
= ERR_PTR(-ENOMEM
);
422 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
425 page
= ERR_PTR(-ENOMEM
);
428 attach_page_buffers(page
, bh
);
429 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
434 bh
->b_blocknr
= bmap(inode
, block
);
435 if (bh
->b_blocknr
== 0) {
436 /* Cannot use this file! */
438 page
= ERR_PTR(-EINVAL
);
441 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
442 if (count
< (1<<inode
->i_blkbits
))
445 count
-= (1<<inode
->i_blkbits
);
447 bh
->b_end_io
= end_bitmap_write
;
448 bh
->b_private
= bitmap
;
449 atomic_inc(&bitmap
->pending_writes
);
450 set_buffer_locked(bh
);
451 set_buffer_mapped(bh
);
455 bh
= bh
->b_this_page
;
459 wait_event(bitmap
->write_wait
,
460 atomic_read(&bitmap
->pending_writes
)==0);
461 if (bitmap
->flags
& BITMAP_WRITE_ERROR
) {
463 page
= ERR_PTR(-EIO
);
467 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %ld\n",
469 (unsigned long long)index
<< PAGE_SHIFT
,
475 * bitmap file superblock operations
478 /* update the event counter and sync the superblock to disk */
479 void bitmap_update_sb(struct bitmap
*bitmap
)
484 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
486 if (bitmap
->mddev
->bitmap_info
.external
)
488 spin_lock_irqsave(&bitmap
->lock
, flags
);
489 if (!bitmap
->sb_page
) { /* no superblock */
490 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
493 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
494 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
495 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
496 if (bitmap
->mddev
->events
< bitmap
->events_cleared
) {
497 /* rocking back to read-only */
498 bitmap
->events_cleared
= bitmap
->mddev
->events
;
499 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
501 /* Just in case these have been changed via sysfs: */
502 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
503 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
504 kunmap_atomic(sb
, KM_USER0
);
505 write_page(bitmap
, bitmap
->sb_page
, 1);
508 /* print out the bitmap file superblock */
509 void bitmap_print_sb(struct bitmap
*bitmap
)
513 if (!bitmap
|| !bitmap
->sb_page
)
515 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
516 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
517 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
518 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
519 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
520 *(__u32
*)(sb
->uuid
+0),
521 *(__u32
*)(sb
->uuid
+4),
522 *(__u32
*)(sb
->uuid
+8),
523 *(__u32
*)(sb
->uuid
+12));
524 printk(KERN_DEBUG
" events: %llu\n",
525 (unsigned long long) le64_to_cpu(sb
->events
));
526 printk(KERN_DEBUG
"events cleared: %llu\n",
527 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
528 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
529 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
530 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
531 printk(KERN_DEBUG
" sync size: %llu KB\n",
532 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
533 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
534 kunmap_atomic(sb
, KM_USER0
);
537 /* read the superblock from the bitmap file and initialize some bitmap fields */
538 static int bitmap_read_sb(struct bitmap
*bitmap
)
542 unsigned long chunksize
, daemon_sleep
, write_behind
;
543 unsigned long long events
;
546 /* page 0 is the superblock, read it... */
548 loff_t isize
= i_size_read(bitmap
->file
->f_mapping
->host
);
549 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
551 bitmap
->sb_page
= read_page(bitmap
->file
, 0, bitmap
, bytes
);
553 bitmap
->sb_page
= read_sb_page(bitmap
->mddev
,
554 bitmap
->mddev
->bitmap_info
.offset
,
556 0, sizeof(bitmap_super_t
));
558 if (IS_ERR(bitmap
->sb_page
)) {
559 err
= PTR_ERR(bitmap
->sb_page
);
560 bitmap
->sb_page
= NULL
;
564 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
566 chunksize
= le32_to_cpu(sb
->chunksize
);
567 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
568 write_behind
= le32_to_cpu(sb
->write_behind
);
570 /* verify that the bitmap-specific fields are valid */
571 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
572 reason
= "bad magic";
573 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
574 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
575 reason
= "unrecognized superblock version";
576 else if (chunksize
< 512)
577 reason
= "bitmap chunksize too small";
578 else if ((1 << ffz(~chunksize
)) != chunksize
)
579 reason
= "bitmap chunksize not a power of 2";
580 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
581 reason
= "daemon sleep period out of range";
582 else if (write_behind
> COUNTER_MAX
)
583 reason
= "write-behind limit out of range (0 - 16383)";
585 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
586 bmname(bitmap
), reason
);
590 /* keep the array size field of the bitmap superblock up to date */
591 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
593 if (!bitmap
->mddev
->persistent
)
597 * if we have a persistent array superblock, compare the
598 * bitmap's UUID and event counter to the mddev's
600 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
601 printk(KERN_INFO
"%s: bitmap superblock UUID mismatch\n",
605 events
= le64_to_cpu(sb
->events
);
606 if (events
< bitmap
->mddev
->events
) {
607 printk(KERN_INFO
"%s: bitmap file is out of date (%llu < %llu) "
608 "-- forcing full recovery\n", bmname(bitmap
), events
,
609 (unsigned long long) bitmap
->mddev
->events
);
610 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
613 /* assign fields using values from superblock */
614 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
615 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
616 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
617 bitmap
->flags
|= le32_to_cpu(sb
->state
);
618 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
619 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
620 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
621 if (sb
->state
& cpu_to_le32(BITMAP_STALE
))
622 bitmap
->events_cleared
= bitmap
->mddev
->events
;
625 kunmap_atomic(sb
, KM_USER0
);
627 bitmap_print_sb(bitmap
);
631 enum bitmap_mask_op
{
636 /* record the state of the bitmap in the superblock. Return the old value */
637 static int bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
638 enum bitmap_mask_op op
)
644 spin_lock_irqsave(&bitmap
->lock
, flags
);
645 if (!bitmap
->sb_page
) { /* can't set the state */
646 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
649 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
650 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
651 old
= le32_to_cpu(sb
->state
) & bits
;
654 sb
->state
|= cpu_to_le32(bits
);
657 sb
->state
&= cpu_to_le32(~bits
);
662 kunmap_atomic(sb
, KM_USER0
);
667 * general bitmap file operations
673 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
674 * file a page at a time. There's a superblock at the start of the file.
676 /* calculate the index of the page that contains this bit */
677 static inline unsigned long file_page_index(struct bitmap
*bitmap
, unsigned long chunk
)
679 if (!bitmap
->mddev
->bitmap_info
.external
)
680 chunk
+= sizeof(bitmap_super_t
) << 3;
681 return chunk
>> PAGE_BIT_SHIFT
;
684 /* calculate the (bit) offset of this bit within a page */
685 static inline unsigned long file_page_offset(struct bitmap
*bitmap
, unsigned long chunk
)
687 if (!bitmap
->mddev
->bitmap_info
.external
)
688 chunk
+= sizeof(bitmap_super_t
) << 3;
689 return chunk
& (PAGE_BITS
- 1);
693 * return a pointer to the page in the filemap that contains the given bit
695 * this lookup is complicated by the fact that the bitmap sb might be exactly
696 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
699 static inline struct page
*filemap_get_page(struct bitmap
*bitmap
,
702 if (bitmap
->filemap
== NULL
)
704 if (file_page_index(bitmap
, chunk
) >= bitmap
->file_pages
)
706 return bitmap
->filemap
[file_page_index(bitmap
, chunk
)
707 - file_page_index(bitmap
, 0)];
710 static void bitmap_file_unmap(struct bitmap
*bitmap
)
712 struct page
**map
, *sb_page
;
717 spin_lock_irqsave(&bitmap
->lock
, flags
);
718 map
= bitmap
->filemap
;
719 bitmap
->filemap
= NULL
;
720 attr
= bitmap
->filemap_attr
;
721 bitmap
->filemap_attr
= NULL
;
722 pages
= bitmap
->file_pages
;
723 bitmap
->file_pages
= 0;
724 sb_page
= bitmap
->sb_page
;
725 bitmap
->sb_page
= NULL
;
726 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
729 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
730 free_buffers(map
[pages
]);
735 free_buffers(sb_page
);
738 static void bitmap_file_put(struct bitmap
*bitmap
)
743 spin_lock_irqsave(&bitmap
->lock
, flags
);
746 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
749 wait_event(bitmap
->write_wait
,
750 atomic_read(&bitmap
->pending_writes
)==0);
751 bitmap_file_unmap(bitmap
);
754 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
755 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
761 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
762 * then it is no longer reliable, so we stop using it and we mark the file
763 * as failed in the superblock
765 static void bitmap_file_kick(struct bitmap
*bitmap
)
767 char *path
, *ptr
= NULL
;
769 if (bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_SET
) == 0) {
770 bitmap_update_sb(bitmap
);
773 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
775 ptr
= d_path(&bitmap
->file
->f_path
, path
,
779 "%s: kicking failed bitmap file %s from array!\n",
780 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
785 "%s: disabling internal bitmap due to errors\n",
789 bitmap_file_put(bitmap
);
794 enum bitmap_page_attr
{
795 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
796 BITMAP_PAGE_CLEAN
= 1, /* there are bits that might need to be cleared */
797 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
800 static inline void set_page_attr(struct bitmap
*bitmap
, struct page
*page
,
801 enum bitmap_page_attr attr
)
804 __set_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
806 __set_bit(attr
, &bitmap
->logattrs
);
809 static inline void clear_page_attr(struct bitmap
*bitmap
, struct page
*page
,
810 enum bitmap_page_attr attr
)
813 __clear_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
815 __clear_bit(attr
, &bitmap
->logattrs
);
818 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, struct page
*page
,
819 enum bitmap_page_attr attr
)
822 return test_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
824 return test_bit(attr
, &bitmap
->logattrs
);
828 * bitmap_file_set_bit -- called before performing a write to the md device
829 * to set (and eventually sync) a particular bit in the bitmap file
831 * we set the bit immediately, then we record the page number so that
832 * when an unplug occurs, we can flush the dirty pages out to disk
834 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
837 struct page
*page
= NULL
;
839 unsigned long chunk
= block
>> CHUNK_BLOCK_SHIFT(bitmap
);
841 if (!bitmap
->filemap
) {
842 struct dm_dirty_log
*log
= bitmap
->mddev
->bitmap_info
.log
;
844 log
->type
->mark_region(log
, chunk
);
847 page
= filemap_get_page(bitmap
, chunk
);
850 bit
= file_page_offset(bitmap
, chunk
);
853 kaddr
= kmap_atomic(page
, KM_USER0
);
854 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
857 __test_and_set_bit_le(bit
, kaddr
);
858 kunmap_atomic(kaddr
, KM_USER0
);
859 PRINTK("set file bit %lu page %lu\n", bit
, page
->index
);
861 /* record page number so it gets flushed to disk when unplug occurs */
862 set_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
865 /* this gets called when the md device is ready to unplug its underlying
866 * (slave) device queues -- before we let any writes go down, we need to
867 * sync the dirty pages of the bitmap file to disk */
868 void bitmap_unplug(struct bitmap
*bitmap
)
870 unsigned long i
, flags
;
871 int dirty
, need_write
;
877 if (!bitmap
->filemap
) {
878 /* Must be using a dirty_log */
879 struct dm_dirty_log
*log
= bitmap
->mddev
->bitmap_info
.log
;
880 dirty
= test_and_clear_bit(BITMAP_PAGE_DIRTY
, &bitmap
->logattrs
);
881 need_write
= test_and_clear_bit(BITMAP_PAGE_NEEDWRITE
, &bitmap
->logattrs
);
882 if (dirty
|| need_write
)
883 if (log
->type
->flush(log
))
884 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
888 /* look at each page to see if there are any set bits that need to be
889 * flushed out to disk */
890 for (i
= 0; i
< bitmap
->file_pages
; i
++) {
891 spin_lock_irqsave(&bitmap
->lock
, flags
);
892 if (!bitmap
->filemap
) {
893 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
896 page
= bitmap
->filemap
[i
];
897 dirty
= test_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
898 need_write
= test_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
899 clear_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
900 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
903 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
905 if (dirty
|| need_write
)
906 write_page(bitmap
, page
, 0);
908 if (wait
) { /* if any writes were performed, we need to wait on them */
910 wait_event(bitmap
->write_wait
,
911 atomic_read(&bitmap
->pending_writes
)==0);
913 md_super_wait(bitmap
->mddev
);
916 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
917 bitmap_file_kick(bitmap
);
919 EXPORT_SYMBOL(bitmap_unplug
);
921 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
922 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
923 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
924 * memory mapping of the bitmap file
926 * if there's no bitmap file, or if the bitmap file had been
927 * previously kicked from the array, we mark all the bits as
928 * 1's in order to cause a full resync.
930 * We ignore all bits for sectors that end earlier than 'start'.
931 * This is used when reading an out-of-date bitmap...
933 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
935 unsigned long i
, chunks
, index
, oldindex
, bit
;
936 struct page
*page
= NULL
, *oldpage
= NULL
;
937 unsigned long num_pages
, bit_cnt
= 0;
939 unsigned long bytes
, offset
;
944 chunks
= bitmap
->chunks
;
947 BUG_ON(!file
&& !bitmap
->mddev
->bitmap_info
.offset
);
949 #ifdef INJECT_FAULTS_3
952 outofdate
= bitmap
->flags
& BITMAP_STALE
;
955 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
956 "recovery\n", bmname(bitmap
));
958 bytes
= DIV_ROUND_UP(bitmap
->chunks
, 8);
959 if (!bitmap
->mddev
->bitmap_info
.external
)
960 bytes
+= sizeof(bitmap_super_t
);
962 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
964 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
) {
965 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
967 (unsigned long) i_size_read(file
->f_mapping
->host
),
974 bitmap
->filemap
= kmalloc(sizeof(struct page
*) * num_pages
, GFP_KERNEL
);
975 if (!bitmap
->filemap
)
978 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
979 bitmap
->filemap_attr
= kzalloc(
980 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
982 if (!bitmap
->filemap_attr
)
987 for (i
= 0; i
< chunks
; i
++) {
989 index
= file_page_index(bitmap
, i
);
990 bit
= file_page_offset(bitmap
, i
);
991 if (index
!= oldindex
) { /* this is a new page, read it in */
993 /* unmap the old page, we're done with it */
994 if (index
== num_pages
-1)
995 count
= bytes
- index
* PAGE_SIZE
;
998 if (index
== 0 && bitmap
->sb_page
) {
1000 * if we're here then the superblock page
1001 * contains some bits (PAGE_SIZE != sizeof sb)
1002 * we've already read it in, so just use it
1004 page
= bitmap
->sb_page
;
1005 offset
= sizeof(bitmap_super_t
);
1007 page
= read_sb_page(
1009 bitmap
->mddev
->bitmap_info
.offset
,
1013 page
= read_page(file
, index
, bitmap
, count
);
1016 page
= read_sb_page(bitmap
->mddev
,
1017 bitmap
->mddev
->bitmap_info
.offset
,
1022 if (IS_ERR(page
)) { /* read error */
1023 ret
= PTR_ERR(page
);
1030 bitmap
->filemap
[bitmap
->file_pages
++] = page
;
1031 bitmap
->last_page_size
= count
;
1035 * if bitmap is out of date, dirty the
1036 * whole page and write it out
1038 paddr
= kmap_atomic(page
, KM_USER0
);
1039 memset(paddr
+ offset
, 0xff,
1040 PAGE_SIZE
- offset
);
1041 kunmap_atomic(paddr
, KM_USER0
);
1042 write_page(bitmap
, page
, 1);
1045 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1049 paddr
= kmap_atomic(page
, KM_USER0
);
1050 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1051 b
= test_bit(bit
, paddr
);
1053 b
= test_bit_le(bit
, paddr
);
1054 kunmap_atomic(paddr
, KM_USER0
);
1056 /* if the disk bit is set, set the memory bit */
1057 int needed
= ((sector_t
)(i
+1) << (CHUNK_BLOCK_SHIFT(bitmap
))
1059 bitmap_set_memory_bits(bitmap
,
1060 (sector_t
)i
<< CHUNK_BLOCK_SHIFT(bitmap
),
1063 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1067 /* everything went OK */
1069 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
1071 if (bit_cnt
) { /* Kick recovery if any bits were set */
1072 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1073 md_wakeup_thread(bitmap
->mddev
->thread
);
1076 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1077 "read %lu/%lu pages, set %lu bits\n",
1078 bmname(bitmap
), bitmap
->file_pages
, num_pages
, bit_cnt
);
1083 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1084 bmname(bitmap
), ret
);
1088 void bitmap_write_all(struct bitmap
*bitmap
)
1090 /* We don't actually write all bitmap blocks here,
1091 * just flag them as needing to be written
1095 for (i
= 0; i
< bitmap
->file_pages
; i
++)
1096 set_page_attr(bitmap
, bitmap
->filemap
[i
],
1097 BITMAP_PAGE_NEEDWRITE
);
1100 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
1102 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1103 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1104 bitmap
->bp
[page
].count
+= inc
;
1105 bitmap_checkfree(bitmap
, page
);
1107 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1108 sector_t offset
, sector_t
*blocks
,
1112 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1116 void bitmap_daemon_work(mddev_t
*mddev
)
1118 struct bitmap
*bitmap
;
1120 unsigned long flags
;
1121 struct page
*page
= NULL
, *lastpage
= NULL
;
1124 struct dm_dirty_log
*log
= mddev
->bitmap_info
.log
;
1126 /* Use a mutex to guard daemon_work against
1129 mutex_lock(&mddev
->bitmap_info
.mutex
);
1130 bitmap
= mddev
->bitmap
;
1131 if (bitmap
== NULL
) {
1132 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1135 if (time_before(jiffies
, bitmap
->daemon_lastrun
1136 + bitmap
->mddev
->bitmap_info
.daemon_sleep
))
1139 bitmap
->daemon_lastrun
= jiffies
;
1140 if (bitmap
->allclean
) {
1141 bitmap
->mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1144 bitmap
->allclean
= 1;
1146 spin_lock_irqsave(&bitmap
->lock
, flags
);
1147 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1148 bitmap_counter_t
*bmc
;
1149 if (!bitmap
->filemap
) {
1151 /* error or shutdown */
1154 page
= filemap_get_page(bitmap
, j
);
1156 if (page
!= lastpage
) {
1157 /* skip this page unless it's marked as needing cleaning */
1158 if (!test_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
)) {
1159 int need_write
= test_page_attr(bitmap
, page
,
1160 BITMAP_PAGE_NEEDWRITE
);
1162 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
1164 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1166 write_page(bitmap
, page
, 0);
1167 bitmap
->allclean
= 0;
1169 spin_lock_irqsave(&bitmap
->lock
, flags
);
1170 j
|= (PAGE_BITS
- 1);
1174 /* grab the new page, sync and release the old */
1175 if (lastpage
!= NULL
) {
1176 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1177 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1178 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1179 write_page(bitmap
, lastpage
, 0);
1181 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1182 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1185 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1188 /* We are possibly going to clear some bits, so make
1189 * sure that events_cleared is up-to-date.
1191 if (bitmap
->need_sync
&&
1192 bitmap
->mddev
->bitmap_info
.external
== 0) {
1194 bitmap
->need_sync
= 0;
1195 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
1196 sb
->events_cleared
=
1197 cpu_to_le64(bitmap
->events_cleared
);
1198 kunmap_atomic(sb
, KM_USER0
);
1199 write_page(bitmap
, bitmap
->sb_page
, 1);
1201 spin_lock_irqsave(&bitmap
->lock
, flags
);
1202 if (!bitmap
->need_sync
)
1203 clear_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1205 bmc
= bitmap_get_counter(bitmap
,
1206 (sector_t
)j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1210 bitmap
->allclean
= 0;
1213 *bmc
= 1; /* maybe clear the bit next time */
1214 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1215 } else if (*bmc
== 1 && !bitmap
->need_sync
) {
1216 /* we can clear the bit */
1218 bitmap_count_page(bitmap
,
1219 (sector_t
)j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1224 paddr
= kmap_atomic(page
, KM_USER0
);
1225 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1226 clear_bit(file_page_offset(bitmap
, j
),
1229 __test_and_clear_bit_le(file_page_offset(bitmap
, j
),
1231 kunmap_atomic(paddr
, KM_USER0
);
1233 log
->type
->clear_region(log
, j
);
1236 j
|= PAGE_COUNTER_MASK
;
1238 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1240 /* now sync the final page */
1241 if (lastpage
!= NULL
|| log
!= NULL
) {
1242 spin_lock_irqsave(&bitmap
->lock
, flags
);
1243 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1244 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1245 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1247 write_page(bitmap
, lastpage
, 0);
1249 if (log
->type
->flush(log
))
1250 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
1252 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1253 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1258 if (bitmap
->allclean
== 0)
1259 bitmap
->mddev
->thread
->timeout
=
1260 bitmap
->mddev
->bitmap_info
.daemon_sleep
;
1261 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1264 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1265 sector_t offset
, sector_t
*blocks
,
1267 __releases(bitmap
->lock
)
1268 __acquires(bitmap
->lock
)
1270 /* If 'create', we might release the lock and reclaim it.
1271 * The lock must have been taken with interrupts enabled.
1272 * If !create, we don't release the lock.
1274 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1275 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1276 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1280 err
= bitmap_checkpage(bitmap
, page
, create
);
1282 if (bitmap
->bp
[page
].hijacked
||
1283 bitmap
->bp
[page
].map
== NULL
)
1284 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
) +
1285 PAGE_COUNTER_SHIFT
- 1);
1287 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1288 *blocks
= csize
- (offset
& (csize
- 1));
1293 /* now locked ... */
1295 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1296 /* should we use the first or second counter field
1297 * of the hijacked pointer? */
1298 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1299 return &((bitmap_counter_t
*)
1300 &bitmap
->bp
[page
].map
)[hi
];
1301 } else /* page is allocated */
1302 return (bitmap_counter_t
*)
1303 &(bitmap
->bp
[page
].map
[pageoff
]);
1306 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1313 atomic_inc(&bitmap
->behind_writes
);
1314 bw
= atomic_read(&bitmap
->behind_writes
);
1315 if (bw
> bitmap
->behind_writes_used
)
1316 bitmap
->behind_writes_used
= bw
;
1318 PRINTK(KERN_DEBUG
"inc write-behind count %d/%d\n",
1319 bw
, bitmap
->max_write_behind
);
1324 bitmap_counter_t
*bmc
;
1326 spin_lock_irq(&bitmap
->lock
);
1327 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1329 spin_unlock_irq(&bitmap
->lock
);
1333 if (unlikely((*bmc
& COUNTER_MAX
) == COUNTER_MAX
)) {
1334 DEFINE_WAIT(__wait
);
1335 /* note that it is safe to do the prepare_to_wait
1336 * after the test as long as we do it before dropping
1339 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1340 TASK_UNINTERRUPTIBLE
);
1341 spin_unlock_irq(&bitmap
->lock
);
1343 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1349 bitmap_file_set_bit(bitmap
, offset
);
1350 bitmap_count_page(bitmap
, offset
, 1);
1358 spin_unlock_irq(&bitmap
->lock
);
1361 if (sectors
> blocks
)
1366 bitmap
->allclean
= 0;
1369 EXPORT_SYMBOL(bitmap_startwrite
);
1371 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1372 int success
, int behind
)
1377 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1378 wake_up(&bitmap
->behind_wait
);
1379 PRINTK(KERN_DEBUG
"dec write-behind count %d/%d\n",
1380 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1382 if (bitmap
->mddev
->degraded
)
1383 /* Never clear bits or update events_cleared when degraded */
1388 unsigned long flags
;
1389 bitmap_counter_t
*bmc
;
1391 spin_lock_irqsave(&bitmap
->lock
, flags
);
1392 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1394 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1399 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1400 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1401 bitmap
->need_sync
= 1;
1402 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1405 if (!success
&& ! (*bmc
& NEEDED_MASK
))
1406 *bmc
|= NEEDED_MASK
;
1408 if ((*bmc
& COUNTER_MAX
) == COUNTER_MAX
)
1409 wake_up(&bitmap
->overflow_wait
);
1413 set_page_attr(bitmap
,
1416 offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1419 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1421 if (sectors
> blocks
)
1427 EXPORT_SYMBOL(bitmap_endwrite
);
1429 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1432 bitmap_counter_t
*bmc
;
1434 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1436 return 1; /* always resync if no bitmap */
1438 spin_lock_irq(&bitmap
->lock
);
1439 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1445 else if (NEEDED(*bmc
)) {
1447 if (!degraded
) { /* don't set/clear bits if degraded */
1448 *bmc
|= RESYNC_MASK
;
1449 *bmc
&= ~NEEDED_MASK
;
1453 spin_unlock_irq(&bitmap
->lock
);
1454 bitmap
->allclean
= 0;
1458 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1461 /* bitmap_start_sync must always report on multiples of whole
1462 * pages, otherwise resync (which is very PAGE_SIZE based) will
1464 * So call __bitmap_start_sync repeatedly (if needed) until
1465 * At least PAGE_SIZE>>9 blocks are covered.
1466 * Return the 'or' of the result.
1472 while (*blocks
< (PAGE_SIZE
>>9)) {
1473 rv
|= __bitmap_start_sync(bitmap
, offset
,
1474 &blocks1
, degraded
);
1480 EXPORT_SYMBOL(bitmap_start_sync
);
1482 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1484 bitmap_counter_t
*bmc
;
1485 unsigned long flags
;
1487 if (bitmap
== NULL
) {
1491 spin_lock_irqsave(&bitmap
->lock
, flags
);
1492 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1497 *bmc
&= ~RESYNC_MASK
;
1499 if (!NEEDED(*bmc
) && aborted
)
1500 *bmc
|= NEEDED_MASK
;
1503 set_page_attr(bitmap
,
1504 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1509 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1510 bitmap
->allclean
= 0;
1512 EXPORT_SYMBOL(bitmap_end_sync
);
1514 void bitmap_close_sync(struct bitmap
*bitmap
)
1516 /* Sync has finished, and any bitmap chunks that weren't synced
1517 * properly have been aborted. It remains to us to clear the
1518 * RESYNC bit wherever it is still on
1520 sector_t sector
= 0;
1524 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1525 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1529 EXPORT_SYMBOL(bitmap_close_sync
);
1531 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
)
1539 bitmap
->last_end_sync
= jiffies
;
1542 if (time_before(jiffies
, (bitmap
->last_end_sync
1543 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1545 wait_event(bitmap
->mddev
->recovery_wait
,
1546 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1548 bitmap
->mddev
->curr_resync_completed
= sector
;
1549 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1550 sector
&= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap
)) - 1);
1552 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1553 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1556 bitmap
->last_end_sync
= jiffies
;
1557 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1559 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1561 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1563 /* For each chunk covered by any of these sectors, set the
1564 * counter to 1 and set resync_needed. They should all
1565 * be 0 at this point
1569 bitmap_counter_t
*bmc
;
1570 spin_lock_irq(&bitmap
->lock
);
1571 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1573 spin_unlock_irq(&bitmap
->lock
);
1578 *bmc
= 1 | (needed
? NEEDED_MASK
: 0);
1579 bitmap_count_page(bitmap
, offset
, 1);
1580 page
= filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
));
1581 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1583 spin_unlock_irq(&bitmap
->lock
);
1584 bitmap
->allclean
= 0;
1587 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1588 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1590 unsigned long chunk
;
1592 for (chunk
= s
; chunk
<= e
; chunk
++) {
1593 sector_t sec
= (sector_t
)chunk
<< CHUNK_BLOCK_SHIFT(bitmap
);
1594 bitmap_set_memory_bits(bitmap
, sec
, 1);
1595 bitmap_file_set_bit(bitmap
, sec
);
1596 if (sec
< bitmap
->mddev
->recovery_cp
)
1597 /* We are asserting that the array is dirty,
1598 * so move the recovery_cp address back so
1599 * that it is obvious that it is dirty
1601 bitmap
->mddev
->recovery_cp
= sec
;
1606 * flush out any pending updates
1608 void bitmap_flush(mddev_t
*mddev
)
1610 struct bitmap
*bitmap
= mddev
->bitmap
;
1613 if (!bitmap
) /* there was no bitmap */
1616 /* run the daemon_work three time to ensure everything is flushed
1619 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1620 bitmap
->daemon_lastrun
-= sleep
;
1621 bitmap_daemon_work(mddev
);
1622 bitmap
->daemon_lastrun
-= sleep
;
1623 bitmap_daemon_work(mddev
);
1624 bitmap
->daemon_lastrun
-= sleep
;
1625 bitmap_daemon_work(mddev
);
1626 bitmap_update_sb(bitmap
);
1630 * free memory that was allocated
1632 static void bitmap_free(struct bitmap
*bitmap
)
1634 unsigned long k
, pages
;
1635 struct bitmap_page
*bp
;
1637 if (!bitmap
) /* there was no bitmap */
1640 /* release the bitmap file and kill the daemon */
1641 bitmap_file_put(bitmap
);
1644 pages
= bitmap
->pages
;
1646 /* free all allocated memory */
1648 if (bp
) /* deallocate the page memory */
1649 for (k
= 0; k
< pages
; k
++)
1650 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1656 void bitmap_destroy(mddev_t
*mddev
)
1658 struct bitmap
*bitmap
= mddev
->bitmap
;
1660 if (!bitmap
) /* there was no bitmap */
1663 mutex_lock(&mddev
->bitmap_info
.mutex
);
1664 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1665 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1667 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1669 if (bitmap
->sysfs_can_clear
)
1670 sysfs_put(bitmap
->sysfs_can_clear
);
1672 bitmap_free(bitmap
);
1676 * initialize the bitmap structure
1677 * if this returns an error, bitmap_destroy must be called to do clean up
1679 int bitmap_create(mddev_t
*mddev
)
1681 struct bitmap
*bitmap
;
1682 sector_t blocks
= mddev
->resync_max_sectors
;
1683 unsigned long chunks
;
1684 unsigned long pages
;
1685 struct file
*file
= mddev
->bitmap_info
.file
;
1687 struct sysfs_dirent
*bm
= NULL
;
1689 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1692 && !mddev
->bitmap_info
.offset
1693 && !mddev
->bitmap_info
.log
) /* bitmap disabled, nothing to do */
1696 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1697 BUG_ON(mddev
->bitmap_info
.offset
&& mddev
->bitmap_info
.log
);
1699 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1703 spin_lock_init(&bitmap
->lock
);
1704 atomic_set(&bitmap
->pending_writes
, 0);
1705 init_waitqueue_head(&bitmap
->write_wait
);
1706 init_waitqueue_head(&bitmap
->overflow_wait
);
1707 init_waitqueue_head(&bitmap
->behind_wait
);
1709 bitmap
->mddev
= mddev
;
1712 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "bitmap");
1714 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, NULL
, "can_clear");
1717 bitmap
->sysfs_can_clear
= NULL
;
1719 bitmap
->file
= file
;
1722 /* As future accesses to this file will use bmap,
1723 * and bypass the page cache, we must sync the file
1728 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1729 if (!mddev
->bitmap_info
.external
)
1730 err
= bitmap_read_sb(bitmap
);
1733 if (mddev
->bitmap_info
.chunksize
== 0 ||
1734 mddev
->bitmap_info
.daemon_sleep
== 0)
1735 /* chunksize and time_base need to be
1742 bitmap
->daemon_lastrun
= jiffies
;
1743 bitmap
->chunkshift
= ffz(~mddev
->bitmap_info
.chunksize
);
1745 /* now that chunksize and chunkshift are set, we can use these macros */
1746 chunks
= (blocks
+ CHUNK_BLOCK_RATIO(bitmap
) - 1) >>
1747 CHUNK_BLOCK_SHIFT(bitmap
);
1748 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1752 bitmap
->chunks
= chunks
;
1753 bitmap
->pages
= pages
;
1754 bitmap
->missing_pages
= pages
;
1755 bitmap
->counter_bits
= COUNTER_BITS
;
1757 bitmap
->syncchunk
= ~0UL;
1759 #ifdef INJECT_FATAL_FAULT_1
1762 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1768 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1769 pages
, bmname(bitmap
));
1771 mddev
->bitmap
= bitmap
;
1774 return (bitmap
->flags
& BITMAP_WRITE_ERROR
) ? -EIO
: 0;
1777 bitmap_free(bitmap
);
1781 int bitmap_load(mddev_t
*mddev
)
1784 sector_t sector
= 0;
1785 struct bitmap
*bitmap
= mddev
->bitmap
;
1790 /* Clear out old bitmap info first: Either there is none, or we
1791 * are resuming after someone else has possibly changed things,
1792 * so we should forget old cached info.
1793 * All chunks should be clean, but some might need_sync.
1795 while (sector
< mddev
->resync_max_sectors
) {
1797 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1800 bitmap_close_sync(bitmap
);
1802 if (mddev
->bitmap_info
.log
) {
1804 struct dm_dirty_log
*log
= mddev
->bitmap_info
.log
;
1805 for (i
= 0; i
< bitmap
->chunks
; i
++)
1806 if (!log
->type
->in_sync(log
, i
, 1))
1807 bitmap_set_memory_bits(bitmap
,
1808 (sector_t
)i
<< CHUNK_BLOCK_SHIFT(bitmap
),
1812 if (mddev
->degraded
== 0
1813 || bitmap
->events_cleared
== mddev
->events
)
1814 /* no need to keep dirty bits to optimise a
1815 * re-add of a missing device */
1816 start
= mddev
->recovery_cp
;
1818 err
= bitmap_init_from_disk(bitmap
, start
);
1823 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1824 md_wakeup_thread(mddev
->thread
);
1826 bitmap_update_sb(bitmap
);
1828 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1833 EXPORT_SYMBOL_GPL(bitmap_load
);
1836 location_show(mddev_t
*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(mddev_t
*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 bitmap_destroy(mddev
);
1905 mddev
->bitmap_info
.offset
= 0;
1907 mddev
->pers
->quiesce(mddev
, 0);
1913 if (!mddev
->external
) {
1914 /* Ensure new bitmap info is stored in
1915 * metadata promptly.
1917 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1918 md_wakeup_thread(mddev
->thread
);
1923 static struct md_sysfs_entry bitmap_location
=
1924 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
1927 timeout_show(mddev_t
*mddev
, char *page
)
1930 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
1931 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
1933 len
= sprintf(page
, "%lu", secs
);
1935 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
1936 len
+= sprintf(page
+len
, "\n");
1941 timeout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1943 /* timeout can be set at any time */
1944 unsigned long timeout
;
1945 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
1949 /* just to make sure we don't overflow... */
1950 if (timeout
>= LONG_MAX
/ HZ
)
1953 timeout
= timeout
* HZ
/ 10000;
1955 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
1956 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
1959 mddev
->bitmap_info
.daemon_sleep
= timeout
;
1960 if (mddev
->thread
) {
1961 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1962 * the bitmap is all clean and we don't need to
1963 * adjust the timeout right now
1965 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
1966 mddev
->thread
->timeout
= timeout
;
1967 md_wakeup_thread(mddev
->thread
);
1973 static struct md_sysfs_entry bitmap_timeout
=
1974 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
1977 backlog_show(mddev_t
*mddev
, char *page
)
1979 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
1983 backlog_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1985 unsigned long backlog
;
1986 int rv
= strict_strtoul(buf
, 10, &backlog
);
1989 if (backlog
> COUNTER_MAX
)
1991 mddev
->bitmap_info
.max_write_behind
= backlog
;
1995 static struct md_sysfs_entry bitmap_backlog
=
1996 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
1999 chunksize_show(mddev_t
*mddev
, char *page
)
2001 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2005 chunksize_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2007 /* Can only be changed when no bitmap is active */
2009 unsigned long csize
;
2012 rv
= strict_strtoul(buf
, 10, &csize
);
2016 !is_power_of_2(csize
))
2018 mddev
->bitmap_info
.chunksize
= csize
;
2022 static struct md_sysfs_entry bitmap_chunksize
=
2023 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2025 static ssize_t
metadata_show(mddev_t
*mddev
, char *page
)
2027 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2028 ? "external" : "internal"));
2031 static ssize_t
metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2033 if (mddev
->bitmap
||
2034 mddev
->bitmap_info
.file
||
2035 mddev
->bitmap_info
.offset
)
2037 if (strncmp(buf
, "external", 8) == 0)
2038 mddev
->bitmap_info
.external
= 1;
2039 else if (strncmp(buf
, "internal", 8) == 0)
2040 mddev
->bitmap_info
.external
= 0;
2046 static struct md_sysfs_entry bitmap_metadata
=
2047 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2049 static ssize_t
can_clear_show(mddev_t
*mddev
, char *page
)
2053 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2056 len
= sprintf(page
, "\n");
2060 static ssize_t
can_clear_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2062 if (mddev
->bitmap
== NULL
)
2064 if (strncmp(buf
, "false", 5) == 0)
2065 mddev
->bitmap
->need_sync
= 1;
2066 else if (strncmp(buf
, "true", 4) == 0) {
2067 if (mddev
->degraded
)
2069 mddev
->bitmap
->need_sync
= 0;
2075 static struct md_sysfs_entry bitmap_can_clear
=
2076 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2079 behind_writes_used_show(mddev_t
*mddev
, char *page
)
2081 if (mddev
->bitmap
== NULL
)
2082 return sprintf(page
, "0\n");
2083 return sprintf(page
, "%lu\n",
2084 mddev
->bitmap
->behind_writes_used
);
2088 behind_writes_used_reset(mddev_t
*mddev
, const char *buf
, size_t len
)
2091 mddev
->bitmap
->behind_writes_used
= 0;
2095 static struct md_sysfs_entry max_backlog_used
=
2096 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2097 behind_writes_used_show
, behind_writes_used_reset
);
2099 static struct attribute
*md_bitmap_attrs
[] = {
2100 &bitmap_location
.attr
,
2101 &bitmap_timeout
.attr
,
2102 &bitmap_backlog
.attr
,
2103 &bitmap_chunksize
.attr
,
2104 &bitmap_metadata
.attr
,
2105 &bitmap_can_clear
.attr
,
2106 &max_backlog_used
.attr
,
2109 struct attribute_group md_bitmap_group
= {
2111 .attrs
= md_bitmap_attrs
,