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
= rdev
->sb_start
+ 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 mddev_t
*mddev
= bitmap
->mddev
;
269 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
270 int size
= PAGE_SIZE
;
271 loff_t offset
= mddev
->bitmap_info
.offset
;
272 if (page
->index
== bitmap
->file_pages
-1)
273 size
= roundup(bitmap
->last_page_size
,
274 bdev_logical_block_size(rdev
->bdev
));
275 /* Just make sure we aren't corrupting data or
278 if (mddev
->external
) {
279 /* Bitmap could be anywhere. */
280 if (rdev
->sb_start
+ offset
+ (page
->index
284 rdev
->sb_start
+ offset
285 < (rdev
->data_offset
+ mddev
->dev_sectors
288 } else if (offset
< 0) {
289 /* DATA BITMAP METADATA */
291 + (long)(page
->index
* (PAGE_SIZE
/512))
293 /* bitmap runs in to metadata */
295 if (rdev
->data_offset
+ mddev
->dev_sectors
296 > rdev
->sb_start
+ offset
)
297 /* data runs in to bitmap */
299 } else if (rdev
->sb_start
< rdev
->data_offset
) {
300 /* METADATA BITMAP DATA */
303 + page
->index
*(PAGE_SIZE
/512) + size
/512
305 /* bitmap runs in to data */
308 /* DATA METADATA BITMAP - no problems */
310 md_super_write(mddev
, rdev
,
311 rdev
->sb_start
+ offset
312 + page
->index
* (PAGE_SIZE
/512),
318 md_super_wait(mddev
);
325 static void bitmap_file_kick(struct bitmap
*bitmap
);
327 * write out a page to a file
329 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
331 struct buffer_head
*bh
;
333 if (bitmap
->file
== NULL
) {
334 switch (write_sb_page(bitmap
, page
, wait
)) {
336 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
340 bh
= page_buffers(page
);
342 while (bh
&& bh
->b_blocknr
) {
343 atomic_inc(&bitmap
->pending_writes
);
344 set_buffer_locked(bh
);
345 set_buffer_mapped(bh
);
346 submit_bh(WRITE
, bh
);
347 bh
= bh
->b_this_page
;
351 wait_event(bitmap
->write_wait
,
352 atomic_read(&bitmap
->pending_writes
)==0);
354 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
355 bitmap_file_kick(bitmap
);
358 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
360 struct bitmap
*bitmap
= bh
->b_private
;
364 spin_lock_irqsave(&bitmap
->lock
, flags
);
365 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
366 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
368 if (atomic_dec_and_test(&bitmap
->pending_writes
))
369 wake_up(&bitmap
->write_wait
);
372 /* copied from buffer.c */
374 __clear_page_buffers(struct page
*page
)
376 ClearPagePrivate(page
);
377 set_page_private(page
, 0);
378 page_cache_release(page
);
380 static void free_buffers(struct page
*page
)
382 struct buffer_head
*bh
= page_buffers(page
);
385 struct buffer_head
*next
= bh
->b_this_page
;
386 free_buffer_head(bh
);
389 __clear_page_buffers(page
);
393 /* read a page from a file.
394 * We both read the page, and attach buffers to the page to record the
395 * address of each block (using bmap). These addresses will be used
396 * to write the block later, completely bypassing the filesystem.
397 * This usage is similar to how swap files are handled, and allows us
398 * to write to a file with no concerns of memory allocation failing.
400 static struct page
*read_page(struct file
*file
, unsigned long index
,
401 struct bitmap
*bitmap
,
404 struct page
*page
= NULL
;
405 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
406 struct buffer_head
*bh
;
409 PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
410 (unsigned long long)index
<< PAGE_SHIFT
);
412 page
= alloc_page(GFP_KERNEL
);
414 page
= ERR_PTR(-ENOMEM
);
418 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
421 page
= ERR_PTR(-ENOMEM
);
424 attach_page_buffers(page
, bh
);
425 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
430 bh
->b_blocknr
= bmap(inode
, block
);
431 if (bh
->b_blocknr
== 0) {
432 /* Cannot use this file! */
434 page
= ERR_PTR(-EINVAL
);
437 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
438 if (count
< (1<<inode
->i_blkbits
))
441 count
-= (1<<inode
->i_blkbits
);
443 bh
->b_end_io
= end_bitmap_write
;
444 bh
->b_private
= bitmap
;
445 atomic_inc(&bitmap
->pending_writes
);
446 set_buffer_locked(bh
);
447 set_buffer_mapped(bh
);
451 bh
= bh
->b_this_page
;
455 wait_event(bitmap
->write_wait
,
456 atomic_read(&bitmap
->pending_writes
)==0);
457 if (bitmap
->flags
& BITMAP_WRITE_ERROR
) {
459 page
= ERR_PTR(-EIO
);
463 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %ld\n",
465 (unsigned long long)index
<< PAGE_SHIFT
,
471 * bitmap file superblock operations
474 /* update the event counter and sync the superblock to disk */
475 void bitmap_update_sb(struct bitmap
*bitmap
)
480 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
482 if (bitmap
->mddev
->bitmap_info
.external
)
484 spin_lock_irqsave(&bitmap
->lock
, flags
);
485 if (!bitmap
->sb_page
) { /* no superblock */
486 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
489 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
490 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
491 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
492 if (bitmap
->mddev
->events
< bitmap
->events_cleared
) {
493 /* rocking back to read-only */
494 bitmap
->events_cleared
= bitmap
->mddev
->events
;
495 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
497 /* Just in case these have been changed via sysfs: */
498 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
499 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
500 kunmap_atomic(sb
, KM_USER0
);
501 write_page(bitmap
, bitmap
->sb_page
, 1);
504 /* print out the bitmap file superblock */
505 void bitmap_print_sb(struct bitmap
*bitmap
)
509 if (!bitmap
|| !bitmap
->sb_page
)
511 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
512 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
513 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
514 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
515 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
516 *(__u32
*)(sb
->uuid
+0),
517 *(__u32
*)(sb
->uuid
+4),
518 *(__u32
*)(sb
->uuid
+8),
519 *(__u32
*)(sb
->uuid
+12));
520 printk(KERN_DEBUG
" events: %llu\n",
521 (unsigned long long) le64_to_cpu(sb
->events
));
522 printk(KERN_DEBUG
"events cleared: %llu\n",
523 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
524 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
525 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
526 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
527 printk(KERN_DEBUG
" sync size: %llu KB\n",
528 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
529 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
530 kunmap_atomic(sb
, KM_USER0
);
533 /* read the superblock from the bitmap file and initialize some bitmap fields */
534 static int bitmap_read_sb(struct bitmap
*bitmap
)
538 unsigned long chunksize
, daemon_sleep
, write_behind
;
539 unsigned long long events
;
542 /* page 0 is the superblock, read it... */
544 loff_t isize
= i_size_read(bitmap
->file
->f_mapping
->host
);
545 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
547 bitmap
->sb_page
= read_page(bitmap
->file
, 0, bitmap
, bytes
);
549 bitmap
->sb_page
= read_sb_page(bitmap
->mddev
,
550 bitmap
->mddev
->bitmap_info
.offset
,
552 0, sizeof(bitmap_super_t
));
554 if (IS_ERR(bitmap
->sb_page
)) {
555 err
= PTR_ERR(bitmap
->sb_page
);
556 bitmap
->sb_page
= NULL
;
560 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
562 chunksize
= le32_to_cpu(sb
->chunksize
);
563 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
564 write_behind
= le32_to_cpu(sb
->write_behind
);
566 /* verify that the bitmap-specific fields are valid */
567 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
568 reason
= "bad magic";
569 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
570 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
571 reason
= "unrecognized superblock version";
572 else if (chunksize
< 512)
573 reason
= "bitmap chunksize too small";
574 else if ((1 << ffz(~chunksize
)) != chunksize
)
575 reason
= "bitmap chunksize not a power of 2";
576 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
577 reason
= "daemon sleep period out of range";
578 else if (write_behind
> COUNTER_MAX
)
579 reason
= "write-behind limit out of range (0 - 16383)";
581 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
582 bmname(bitmap
), reason
);
586 /* keep the array size field of the bitmap superblock up to date */
587 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
589 if (!bitmap
->mddev
->persistent
)
593 * if we have a persistent array superblock, compare the
594 * bitmap's UUID and event counter to the mddev's
596 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
597 printk(KERN_INFO
"%s: bitmap superblock UUID mismatch\n",
601 events
= le64_to_cpu(sb
->events
);
602 if (events
< bitmap
->mddev
->events
) {
603 printk(KERN_INFO
"%s: bitmap file is out of date (%llu < %llu) "
604 "-- forcing full recovery\n", bmname(bitmap
), events
,
605 (unsigned long long) bitmap
->mddev
->events
);
606 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
609 /* assign fields using values from superblock */
610 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
611 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
612 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
613 bitmap
->flags
|= le32_to_cpu(sb
->state
);
614 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
615 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
616 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
617 if (sb
->state
& cpu_to_le32(BITMAP_STALE
))
618 bitmap
->events_cleared
= bitmap
->mddev
->events
;
621 kunmap_atomic(sb
, KM_USER0
);
623 bitmap_print_sb(bitmap
);
627 enum bitmap_mask_op
{
632 /* record the state of the bitmap in the superblock. Return the old value */
633 static int bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
634 enum bitmap_mask_op op
)
640 spin_lock_irqsave(&bitmap
->lock
, flags
);
641 if (!bitmap
->sb_page
) { /* can't set the state */
642 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
645 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
646 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
647 old
= le32_to_cpu(sb
->state
) & bits
;
650 sb
->state
|= cpu_to_le32(bits
);
653 sb
->state
&= cpu_to_le32(~bits
);
658 kunmap_atomic(sb
, KM_USER0
);
663 * general bitmap file operations
669 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
670 * file a page at a time. There's a superblock at the start of the file.
672 /* calculate the index of the page that contains this bit */
673 static inline unsigned long file_page_index(struct bitmap
*bitmap
, unsigned long chunk
)
675 if (!bitmap
->mddev
->bitmap_info
.external
)
676 chunk
+= sizeof(bitmap_super_t
) << 3;
677 return chunk
>> PAGE_BIT_SHIFT
;
680 /* calculate the (bit) offset of this bit within a page */
681 static inline unsigned long file_page_offset(struct bitmap
*bitmap
, unsigned long chunk
)
683 if (!bitmap
->mddev
->bitmap_info
.external
)
684 chunk
+= sizeof(bitmap_super_t
) << 3;
685 return chunk
& (PAGE_BITS
- 1);
689 * return a pointer to the page in the filemap that contains the given bit
691 * this lookup is complicated by the fact that the bitmap sb might be exactly
692 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
695 static inline struct page
*filemap_get_page(struct bitmap
*bitmap
,
698 if (bitmap
->filemap
== NULL
)
700 if (file_page_index(bitmap
, chunk
) >= bitmap
->file_pages
)
702 return bitmap
->filemap
[file_page_index(bitmap
, chunk
)
703 - file_page_index(bitmap
, 0)];
706 static void bitmap_file_unmap(struct bitmap
*bitmap
)
708 struct page
**map
, *sb_page
;
713 spin_lock_irqsave(&bitmap
->lock
, flags
);
714 map
= bitmap
->filemap
;
715 bitmap
->filemap
= NULL
;
716 attr
= bitmap
->filemap_attr
;
717 bitmap
->filemap_attr
= NULL
;
718 pages
= bitmap
->file_pages
;
719 bitmap
->file_pages
= 0;
720 sb_page
= bitmap
->sb_page
;
721 bitmap
->sb_page
= NULL
;
722 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
725 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
726 free_buffers(map
[pages
]);
731 free_buffers(sb_page
);
734 static void bitmap_file_put(struct bitmap
*bitmap
)
739 spin_lock_irqsave(&bitmap
->lock
, flags
);
742 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
745 wait_event(bitmap
->write_wait
,
746 atomic_read(&bitmap
->pending_writes
)==0);
747 bitmap_file_unmap(bitmap
);
750 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
751 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
757 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
758 * then it is no longer reliable, so we stop using it and we mark the file
759 * as failed in the superblock
761 static void bitmap_file_kick(struct bitmap
*bitmap
)
763 char *path
, *ptr
= NULL
;
765 if (bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_SET
) == 0) {
766 bitmap_update_sb(bitmap
);
769 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
771 ptr
= d_path(&bitmap
->file
->f_path
, path
,
775 "%s: kicking failed bitmap file %s from array!\n",
776 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
781 "%s: disabling internal bitmap due to errors\n",
785 bitmap_file_put(bitmap
);
790 enum bitmap_page_attr
{
791 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
792 BITMAP_PAGE_CLEAN
= 1, /* there are bits that might need to be cleared */
793 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
796 static inline void set_page_attr(struct bitmap
*bitmap
, struct page
*page
,
797 enum bitmap_page_attr attr
)
800 __set_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
802 __set_bit(attr
, &bitmap
->logattrs
);
805 static inline void clear_page_attr(struct bitmap
*bitmap
, struct page
*page
,
806 enum bitmap_page_attr attr
)
809 __clear_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
811 __clear_bit(attr
, &bitmap
->logattrs
);
814 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, struct page
*page
,
815 enum bitmap_page_attr attr
)
818 return test_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
820 return test_bit(attr
, &bitmap
->logattrs
);
824 * bitmap_file_set_bit -- called before performing a write to the md device
825 * to set (and eventually sync) a particular bit in the bitmap file
827 * we set the bit immediately, then we record the page number so that
828 * when an unplug occurs, we can flush the dirty pages out to disk
830 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
833 struct page
*page
= NULL
;
835 unsigned long chunk
= block
>> CHUNK_BLOCK_SHIFT(bitmap
);
837 if (!bitmap
->filemap
) {
838 struct dm_dirty_log
*log
= bitmap
->mddev
->bitmap_info
.log
;
840 log
->type
->mark_region(log
, chunk
);
843 page
= filemap_get_page(bitmap
, chunk
);
846 bit
= file_page_offset(bitmap
, chunk
);
849 kaddr
= kmap_atomic(page
, KM_USER0
);
850 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
853 ext2_set_bit(bit
, kaddr
);
854 kunmap_atomic(kaddr
, KM_USER0
);
855 PRINTK("set file bit %lu page %lu\n", bit
, page
->index
);
857 /* record page number so it gets flushed to disk when unplug occurs */
858 set_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
861 /* this gets called when the md device is ready to unplug its underlying
862 * (slave) device queues -- before we let any writes go down, we need to
863 * sync the dirty pages of the bitmap file to disk */
864 void bitmap_unplug(struct bitmap
*bitmap
)
866 unsigned long i
, flags
;
867 int dirty
, need_write
;
873 if (!bitmap
->filemap
) {
874 /* Must be using a dirty_log */
875 struct dm_dirty_log
*log
= bitmap
->mddev
->bitmap_info
.log
;
876 dirty
= test_and_clear_bit(BITMAP_PAGE_DIRTY
, &bitmap
->logattrs
);
877 need_write
= test_and_clear_bit(BITMAP_PAGE_NEEDWRITE
, &bitmap
->logattrs
);
878 if (dirty
|| need_write
)
879 if (log
->type
->flush(log
))
880 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
884 /* look at each page to see if there are any set bits that need to be
885 * flushed out to disk */
886 for (i
= 0; i
< bitmap
->file_pages
; i
++) {
887 spin_lock_irqsave(&bitmap
->lock
, flags
);
888 if (!bitmap
->filemap
) {
889 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
892 page
= bitmap
->filemap
[i
];
893 dirty
= test_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
894 need_write
= test_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
895 clear_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
896 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
899 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
901 if (dirty
|| need_write
)
902 write_page(bitmap
, page
, 0);
904 if (wait
) { /* if any writes were performed, we need to wait on them */
906 wait_event(bitmap
->write_wait
,
907 atomic_read(&bitmap
->pending_writes
)==0);
909 md_super_wait(bitmap
->mddev
);
912 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
913 bitmap_file_kick(bitmap
);
915 EXPORT_SYMBOL(bitmap_unplug
);
917 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
918 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
919 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
920 * memory mapping of the bitmap file
922 * if there's no bitmap file, or if the bitmap file had been
923 * previously kicked from the array, we mark all the bits as
924 * 1's in order to cause a full resync.
926 * We ignore all bits for sectors that end earlier than 'start'.
927 * This is used when reading an out-of-date bitmap...
929 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
931 unsigned long i
, chunks
, index
, oldindex
, bit
;
932 struct page
*page
= NULL
, *oldpage
= NULL
;
933 unsigned long num_pages
, bit_cnt
= 0;
935 unsigned long bytes
, offset
;
940 chunks
= bitmap
->chunks
;
943 BUG_ON(!file
&& !bitmap
->mddev
->bitmap_info
.offset
);
945 #ifdef INJECT_FAULTS_3
948 outofdate
= bitmap
->flags
& BITMAP_STALE
;
951 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
952 "recovery\n", bmname(bitmap
));
954 bytes
= DIV_ROUND_UP(bitmap
->chunks
, 8);
955 if (!bitmap
->mddev
->bitmap_info
.external
)
956 bytes
+= sizeof(bitmap_super_t
);
958 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
960 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
) {
961 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
963 (unsigned long) i_size_read(file
->f_mapping
->host
),
970 bitmap
->filemap
= kmalloc(sizeof(struct page
*) * num_pages
, GFP_KERNEL
);
971 if (!bitmap
->filemap
)
974 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
975 bitmap
->filemap_attr
= kzalloc(
976 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
978 if (!bitmap
->filemap_attr
)
983 for (i
= 0; i
< chunks
; i
++) {
985 index
= file_page_index(bitmap
, i
);
986 bit
= file_page_offset(bitmap
, i
);
987 if (index
!= oldindex
) { /* this is a new page, read it in */
989 /* unmap the old page, we're done with it */
990 if (index
== num_pages
-1)
991 count
= bytes
- index
* PAGE_SIZE
;
994 if (index
== 0 && bitmap
->sb_page
) {
996 * if we're here then the superblock page
997 * contains some bits (PAGE_SIZE != sizeof sb)
998 * we've already read it in, so just use it
1000 page
= bitmap
->sb_page
;
1001 offset
= sizeof(bitmap_super_t
);
1003 page
= read_sb_page(
1005 bitmap
->mddev
->bitmap_info
.offset
,
1009 page
= read_page(file
, index
, bitmap
, count
);
1012 page
= read_sb_page(bitmap
->mddev
,
1013 bitmap
->mddev
->bitmap_info
.offset
,
1018 if (IS_ERR(page
)) { /* read error */
1019 ret
= PTR_ERR(page
);
1026 bitmap
->filemap
[bitmap
->file_pages
++] = page
;
1027 bitmap
->last_page_size
= count
;
1031 * if bitmap is out of date, dirty the
1032 * whole page and write it out
1034 paddr
= kmap_atomic(page
, KM_USER0
);
1035 memset(paddr
+ offset
, 0xff,
1036 PAGE_SIZE
- offset
);
1037 kunmap_atomic(paddr
, KM_USER0
);
1038 write_page(bitmap
, page
, 1);
1041 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1045 paddr
= kmap_atomic(page
, KM_USER0
);
1046 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1047 b
= test_bit(bit
, paddr
);
1049 b
= ext2_test_bit(bit
, paddr
);
1050 kunmap_atomic(paddr
, KM_USER0
);
1052 /* if the disk bit is set, set the memory bit */
1053 int needed
= ((sector_t
)(i
+1) << (CHUNK_BLOCK_SHIFT(bitmap
))
1055 bitmap_set_memory_bits(bitmap
,
1056 (sector_t
)i
<< CHUNK_BLOCK_SHIFT(bitmap
),
1059 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1063 /* everything went OK */
1065 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
1067 if (bit_cnt
) { /* Kick recovery if any bits were set */
1068 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1069 md_wakeup_thread(bitmap
->mddev
->thread
);
1072 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1073 "read %lu/%lu pages, set %lu bits\n",
1074 bmname(bitmap
), bitmap
->file_pages
, num_pages
, bit_cnt
);
1079 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1080 bmname(bitmap
), ret
);
1084 void bitmap_write_all(struct bitmap
*bitmap
)
1086 /* We don't actually write all bitmap blocks here,
1087 * just flag them as needing to be written
1091 for (i
= 0; i
< bitmap
->file_pages
; i
++)
1092 set_page_attr(bitmap
, bitmap
->filemap
[i
],
1093 BITMAP_PAGE_NEEDWRITE
);
1096 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
1098 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1099 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1100 bitmap
->bp
[page
].count
+= inc
;
1101 bitmap_checkfree(bitmap
, page
);
1103 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1104 sector_t offset
, sector_t
*blocks
,
1108 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1112 void bitmap_daemon_work(mddev_t
*mddev
)
1114 struct bitmap
*bitmap
;
1116 unsigned long flags
;
1117 struct page
*page
= NULL
, *lastpage
= NULL
;
1120 struct dm_dirty_log
*log
= mddev
->bitmap_info
.log
;
1122 /* Use a mutex to guard daemon_work against
1125 mutex_lock(&mddev
->bitmap_info
.mutex
);
1126 bitmap
= mddev
->bitmap
;
1127 if (bitmap
== NULL
) {
1128 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1131 if (time_before(jiffies
, bitmap
->daemon_lastrun
1132 + bitmap
->mddev
->bitmap_info
.daemon_sleep
))
1135 bitmap
->daemon_lastrun
= jiffies
;
1136 if (bitmap
->allclean
) {
1137 bitmap
->mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1140 bitmap
->allclean
= 1;
1142 spin_lock_irqsave(&bitmap
->lock
, flags
);
1143 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1144 bitmap_counter_t
*bmc
;
1145 if (!bitmap
->filemap
) {
1147 /* error or shutdown */
1150 page
= filemap_get_page(bitmap
, j
);
1152 if (page
!= lastpage
) {
1153 /* skip this page unless it's marked as needing cleaning */
1154 if (!test_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
)) {
1155 int need_write
= test_page_attr(bitmap
, page
,
1156 BITMAP_PAGE_NEEDWRITE
);
1158 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
1160 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1162 write_page(bitmap
, page
, 0);
1163 bitmap
->allclean
= 0;
1165 spin_lock_irqsave(&bitmap
->lock
, flags
);
1166 j
|= (PAGE_BITS
- 1);
1170 /* grab the new page, sync and release the old */
1171 if (lastpage
!= NULL
) {
1172 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1173 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1174 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1175 write_page(bitmap
, lastpage
, 0);
1177 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1178 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1181 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1184 /* We are possibly going to clear some bits, so make
1185 * sure that events_cleared is up-to-date.
1187 if (bitmap
->need_sync
&&
1188 bitmap
->mddev
->bitmap_info
.external
== 0) {
1190 bitmap
->need_sync
= 0;
1191 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
1192 sb
->events_cleared
=
1193 cpu_to_le64(bitmap
->events_cleared
);
1194 kunmap_atomic(sb
, KM_USER0
);
1195 write_page(bitmap
, bitmap
->sb_page
, 1);
1197 spin_lock_irqsave(&bitmap
->lock
, flags
);
1198 if (!bitmap
->need_sync
)
1199 clear_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1201 bmc
= bitmap_get_counter(bitmap
,
1202 (sector_t
)j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1206 bitmap
->allclean
= 0;
1209 *bmc
= 1; /* maybe clear the bit next time */
1210 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1211 } else if (*bmc
== 1 && !bitmap
->need_sync
) {
1212 /* we can clear the bit */
1214 bitmap_count_page(bitmap
,
1215 (sector_t
)j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1220 paddr
= kmap_atomic(page
, KM_USER0
);
1221 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1222 clear_bit(file_page_offset(bitmap
, j
),
1225 ext2_clear_bit(file_page_offset(bitmap
, j
),
1227 kunmap_atomic(paddr
, KM_USER0
);
1229 log
->type
->clear_region(log
, j
);
1232 j
|= PAGE_COUNTER_MASK
;
1234 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1236 /* now sync the final page */
1237 if (lastpage
!= NULL
|| log
!= NULL
) {
1238 spin_lock_irqsave(&bitmap
->lock
, flags
);
1239 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1240 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1241 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1243 write_page(bitmap
, lastpage
, 0);
1245 if (log
->type
->flush(log
))
1246 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
1248 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1249 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1254 if (bitmap
->allclean
== 0)
1255 bitmap
->mddev
->thread
->timeout
=
1256 bitmap
->mddev
->bitmap_info
.daemon_sleep
;
1257 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1260 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1261 sector_t offset
, sector_t
*blocks
,
1263 __releases(bitmap
->lock
)
1264 __acquires(bitmap
->lock
)
1266 /* If 'create', we might release the lock and reclaim it.
1267 * The lock must have been taken with interrupts enabled.
1268 * If !create, we don't release the lock.
1270 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1271 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1272 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1276 err
= bitmap_checkpage(bitmap
, page
, create
);
1278 if (bitmap
->bp
[page
].hijacked
||
1279 bitmap
->bp
[page
].map
== NULL
)
1280 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
) +
1281 PAGE_COUNTER_SHIFT
- 1);
1283 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1284 *blocks
= csize
- (offset
& (csize
- 1));
1289 /* now locked ... */
1291 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1292 /* should we use the first or second counter field
1293 * of the hijacked pointer? */
1294 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1295 return &((bitmap_counter_t
*)
1296 &bitmap
->bp
[page
].map
)[hi
];
1297 } else /* page is allocated */
1298 return (bitmap_counter_t
*)
1299 &(bitmap
->bp
[page
].map
[pageoff
]);
1302 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1309 atomic_inc(&bitmap
->behind_writes
);
1310 bw
= atomic_read(&bitmap
->behind_writes
);
1311 if (bw
> bitmap
->behind_writes_used
)
1312 bitmap
->behind_writes_used
= bw
;
1314 PRINTK(KERN_DEBUG
"inc write-behind count %d/%d\n",
1315 bw
, bitmap
->max_write_behind
);
1320 bitmap_counter_t
*bmc
;
1322 spin_lock_irq(&bitmap
->lock
);
1323 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1325 spin_unlock_irq(&bitmap
->lock
);
1329 if (unlikely((*bmc
& COUNTER_MAX
) == COUNTER_MAX
)) {
1330 DEFINE_WAIT(__wait
);
1331 /* note that it is safe to do the prepare_to_wait
1332 * after the test as long as we do it before dropping
1335 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1336 TASK_UNINTERRUPTIBLE
);
1337 spin_unlock_irq(&bitmap
->lock
);
1338 md_unplug(bitmap
->mddev
);
1340 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1346 bitmap_file_set_bit(bitmap
, offset
);
1347 bitmap_count_page(bitmap
, offset
, 1);
1355 spin_unlock_irq(&bitmap
->lock
);
1358 if (sectors
> blocks
)
1363 bitmap
->allclean
= 0;
1366 EXPORT_SYMBOL(bitmap_startwrite
);
1368 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1369 int success
, int behind
)
1374 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1375 wake_up(&bitmap
->behind_wait
);
1376 PRINTK(KERN_DEBUG
"dec write-behind count %d/%d\n",
1377 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1379 if (bitmap
->mddev
->degraded
)
1380 /* Never clear bits or update events_cleared when degraded */
1385 unsigned long flags
;
1386 bitmap_counter_t
*bmc
;
1388 spin_lock_irqsave(&bitmap
->lock
, flags
);
1389 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1391 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1396 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1397 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1398 bitmap
->need_sync
= 1;
1399 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1402 if (!success
&& ! (*bmc
& NEEDED_MASK
))
1403 *bmc
|= NEEDED_MASK
;
1405 if ((*bmc
& COUNTER_MAX
) == COUNTER_MAX
)
1406 wake_up(&bitmap
->overflow_wait
);
1410 set_page_attr(bitmap
,
1413 offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1416 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1418 if (sectors
> blocks
)
1424 EXPORT_SYMBOL(bitmap_endwrite
);
1426 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1429 bitmap_counter_t
*bmc
;
1431 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1433 return 1; /* always resync if no bitmap */
1435 spin_lock_irq(&bitmap
->lock
);
1436 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1442 else if (NEEDED(*bmc
)) {
1444 if (!degraded
) { /* don't set/clear bits if degraded */
1445 *bmc
|= RESYNC_MASK
;
1446 *bmc
&= ~NEEDED_MASK
;
1450 spin_unlock_irq(&bitmap
->lock
);
1451 bitmap
->allclean
= 0;
1455 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1458 /* bitmap_start_sync must always report on multiples of whole
1459 * pages, otherwise resync (which is very PAGE_SIZE based) will
1461 * So call __bitmap_start_sync repeatedly (if needed) until
1462 * At least PAGE_SIZE>>9 blocks are covered.
1463 * Return the 'or' of the result.
1469 while (*blocks
< (PAGE_SIZE
>>9)) {
1470 rv
|= __bitmap_start_sync(bitmap
, offset
,
1471 &blocks1
, degraded
);
1477 EXPORT_SYMBOL(bitmap_start_sync
);
1479 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1481 bitmap_counter_t
*bmc
;
1482 unsigned long flags
;
1484 if (bitmap
== NULL
) {
1488 spin_lock_irqsave(&bitmap
->lock
, flags
);
1489 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1494 *bmc
&= ~RESYNC_MASK
;
1496 if (!NEEDED(*bmc
) && aborted
)
1497 *bmc
|= NEEDED_MASK
;
1500 set_page_attr(bitmap
,
1501 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1506 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1507 bitmap
->allclean
= 0;
1509 EXPORT_SYMBOL(bitmap_end_sync
);
1511 void bitmap_close_sync(struct bitmap
*bitmap
)
1513 /* Sync has finished, and any bitmap chunks that weren't synced
1514 * properly have been aborted. It remains to us to clear the
1515 * RESYNC bit wherever it is still on
1517 sector_t sector
= 0;
1521 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1522 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1526 EXPORT_SYMBOL(bitmap_close_sync
);
1528 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
)
1536 bitmap
->last_end_sync
= jiffies
;
1539 if (time_before(jiffies
, (bitmap
->last_end_sync
1540 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1542 wait_event(bitmap
->mddev
->recovery_wait
,
1543 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1545 bitmap
->mddev
->curr_resync_completed
= bitmap
->mddev
->curr_resync
;
1546 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1547 sector
&= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap
)) - 1);
1549 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1550 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1553 bitmap
->last_end_sync
= jiffies
;
1554 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1556 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1558 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1560 /* For each chunk covered by any of these sectors, set the
1561 * counter to 1 and set resync_needed. They should all
1562 * be 0 at this point
1566 bitmap_counter_t
*bmc
;
1567 spin_lock_irq(&bitmap
->lock
);
1568 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1570 spin_unlock_irq(&bitmap
->lock
);
1575 *bmc
= 1 | (needed
? NEEDED_MASK
: 0);
1576 bitmap_count_page(bitmap
, offset
, 1);
1577 page
= filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
));
1578 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1580 spin_unlock_irq(&bitmap
->lock
);
1581 bitmap
->allclean
= 0;
1584 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1585 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1587 unsigned long chunk
;
1589 for (chunk
= s
; chunk
<= e
; chunk
++) {
1590 sector_t sec
= (sector_t
)chunk
<< CHUNK_BLOCK_SHIFT(bitmap
);
1591 bitmap_set_memory_bits(bitmap
, sec
, 1);
1592 bitmap_file_set_bit(bitmap
, sec
);
1593 if (sec
< bitmap
->mddev
->recovery_cp
)
1594 /* We are asserting that the array is dirty,
1595 * so move the recovery_cp address back so
1596 * that it is obvious that it is dirty
1598 bitmap
->mddev
->recovery_cp
= sec
;
1603 * flush out any pending updates
1605 void bitmap_flush(mddev_t
*mddev
)
1607 struct bitmap
*bitmap
= mddev
->bitmap
;
1610 if (!bitmap
) /* there was no bitmap */
1613 /* run the daemon_work three time to ensure everything is flushed
1616 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1617 bitmap
->daemon_lastrun
-= sleep
;
1618 bitmap_daemon_work(mddev
);
1619 bitmap
->daemon_lastrun
-= sleep
;
1620 bitmap_daemon_work(mddev
);
1621 bitmap
->daemon_lastrun
-= sleep
;
1622 bitmap_daemon_work(mddev
);
1623 bitmap_update_sb(bitmap
);
1627 * free memory that was allocated
1629 static void bitmap_free(struct bitmap
*bitmap
)
1631 unsigned long k
, pages
;
1632 struct bitmap_page
*bp
;
1634 if (!bitmap
) /* there was no bitmap */
1637 /* release the bitmap file and kill the daemon */
1638 bitmap_file_put(bitmap
);
1641 pages
= bitmap
->pages
;
1643 /* free all allocated memory */
1645 if (bp
) /* deallocate the page memory */
1646 for (k
= 0; k
< pages
; k
++)
1647 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1653 void bitmap_destroy(mddev_t
*mddev
)
1655 struct bitmap
*bitmap
= mddev
->bitmap
;
1657 if (!bitmap
) /* there was no bitmap */
1660 mutex_lock(&mddev
->bitmap_info
.mutex
);
1661 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1662 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1664 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1666 if (bitmap
->sysfs_can_clear
)
1667 sysfs_put(bitmap
->sysfs_can_clear
);
1669 bitmap_free(bitmap
);
1673 * initialize the bitmap structure
1674 * if this returns an error, bitmap_destroy must be called to do clean up
1676 int bitmap_create(mddev_t
*mddev
)
1678 struct bitmap
*bitmap
;
1679 sector_t blocks
= mddev
->resync_max_sectors
;
1680 unsigned long chunks
;
1681 unsigned long pages
;
1682 struct file
*file
= mddev
->bitmap_info
.file
;
1684 struct sysfs_dirent
*bm
= NULL
;
1686 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1689 && !mddev
->bitmap_info
.offset
1690 && !mddev
->bitmap_info
.log
) /* bitmap disabled, nothing to do */
1693 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1694 BUG_ON(mddev
->bitmap_info
.offset
&& mddev
->bitmap_info
.log
);
1696 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1700 spin_lock_init(&bitmap
->lock
);
1701 atomic_set(&bitmap
->pending_writes
, 0);
1702 init_waitqueue_head(&bitmap
->write_wait
);
1703 init_waitqueue_head(&bitmap
->overflow_wait
);
1704 init_waitqueue_head(&bitmap
->behind_wait
);
1706 bitmap
->mddev
= mddev
;
1709 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "bitmap");
1711 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, NULL
, "can_clear");
1714 bitmap
->sysfs_can_clear
= NULL
;
1716 bitmap
->file
= file
;
1719 /* As future accesses to this file will use bmap,
1720 * and bypass the page cache, we must sync the file
1725 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1726 if (!mddev
->bitmap_info
.external
)
1727 err
= bitmap_read_sb(bitmap
);
1730 if (mddev
->bitmap_info
.chunksize
== 0 ||
1731 mddev
->bitmap_info
.daemon_sleep
== 0)
1732 /* chunksize and time_base need to be
1739 bitmap
->daemon_lastrun
= jiffies
;
1740 bitmap
->chunkshift
= ffz(~mddev
->bitmap_info
.chunksize
);
1742 /* now that chunksize and chunkshift are set, we can use these macros */
1743 chunks
= (blocks
+ CHUNK_BLOCK_RATIO(bitmap
) - 1) >>
1744 CHUNK_BLOCK_SHIFT(bitmap
);
1745 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1749 bitmap
->chunks
= chunks
;
1750 bitmap
->pages
= pages
;
1751 bitmap
->missing_pages
= pages
;
1752 bitmap
->counter_bits
= COUNTER_BITS
;
1754 bitmap
->syncchunk
= ~0UL;
1756 #ifdef INJECT_FATAL_FAULT_1
1759 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1765 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1766 pages
, bmname(bitmap
));
1768 mddev
->bitmap
= bitmap
;
1771 return (bitmap
->flags
& BITMAP_WRITE_ERROR
) ? -EIO
: 0;
1774 bitmap_free(bitmap
);
1778 int bitmap_load(mddev_t
*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
->bitmap_info
.log
) {
1801 struct dm_dirty_log
*log
= mddev
->bitmap_info
.log
;
1802 for (i
= 0; i
< bitmap
->chunks
; i
++)
1803 if (!log
->type
->in_sync(log
, i
, 1))
1804 bitmap_set_memory_bits(bitmap
,
1805 (sector_t
)i
<< CHUNK_BLOCK_SHIFT(bitmap
),
1809 if (mddev
->degraded
== 0
1810 || bitmap
->events_cleared
== mddev
->events
)
1811 /* no need to keep dirty bits to optimise a
1812 * re-add of a missing device */
1813 start
= mddev
->recovery_cp
;
1815 err
= bitmap_init_from_disk(bitmap
, start
);
1820 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1821 md_wakeup_thread(mddev
->thread
);
1823 bitmap_update_sb(bitmap
);
1825 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1830 EXPORT_SYMBOL_GPL(bitmap_load
);
1833 location_show(mddev_t
*mddev
, char *page
)
1836 if (mddev
->bitmap_info
.file
)
1837 len
= sprintf(page
, "file");
1838 else if (mddev
->bitmap_info
.offset
)
1839 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
1841 len
= sprintf(page
, "none");
1842 len
+= sprintf(page
+len
, "\n");
1847 location_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1851 if (!mddev
->pers
->quiesce
)
1853 if (mddev
->recovery
|| mddev
->sync_thread
)
1857 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
1858 mddev
->bitmap_info
.offset
) {
1859 /* bitmap already configured. Only option is to clear it */
1860 if (strncmp(buf
, "none", 4) != 0)
1863 mddev
->pers
->quiesce(mddev
, 1);
1864 bitmap_destroy(mddev
);
1865 mddev
->pers
->quiesce(mddev
, 0);
1867 mddev
->bitmap_info
.offset
= 0;
1868 if (mddev
->bitmap_info
.file
) {
1869 struct file
*f
= mddev
->bitmap_info
.file
;
1870 mddev
->bitmap_info
.file
= NULL
;
1871 restore_bitmap_write_access(f
);
1875 /* No bitmap, OK to set a location */
1877 if (strncmp(buf
, "none", 4) == 0)
1878 /* nothing to be done */;
1879 else if (strncmp(buf
, "file:", 5) == 0) {
1880 /* Not supported yet */
1885 rv
= strict_strtoll(buf
+1, 10, &offset
);
1887 rv
= strict_strtoll(buf
, 10, &offset
);
1892 if (mddev
->bitmap_info
.external
== 0 &&
1893 mddev
->major_version
== 0 &&
1894 offset
!= mddev
->bitmap_info
.default_offset
)
1896 mddev
->bitmap_info
.offset
= offset
;
1898 mddev
->pers
->quiesce(mddev
, 1);
1899 rv
= bitmap_create(mddev
);
1901 bitmap_destroy(mddev
);
1902 mddev
->bitmap_info
.offset
= 0;
1904 mddev
->pers
->quiesce(mddev
, 0);
1910 if (!mddev
->external
) {
1911 /* Ensure new bitmap info is stored in
1912 * metadata promptly.
1914 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1915 md_wakeup_thread(mddev
->thread
);
1920 static struct md_sysfs_entry bitmap_location
=
1921 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
1924 timeout_show(mddev_t
*mddev
, char *page
)
1927 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
1928 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
1930 len
= sprintf(page
, "%lu", secs
);
1932 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
1933 len
+= sprintf(page
+len
, "\n");
1938 timeout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1940 /* timeout can be set at any time */
1941 unsigned long timeout
;
1942 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
1946 /* just to make sure we don't overflow... */
1947 if (timeout
>= LONG_MAX
/ HZ
)
1950 timeout
= timeout
* HZ
/ 10000;
1952 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
1953 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
1956 mddev
->bitmap_info
.daemon_sleep
= timeout
;
1957 if (mddev
->thread
) {
1958 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1959 * the bitmap is all clean and we don't need to
1960 * adjust the timeout right now
1962 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
1963 mddev
->thread
->timeout
= timeout
;
1964 md_wakeup_thread(mddev
->thread
);
1970 static struct md_sysfs_entry bitmap_timeout
=
1971 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
1974 backlog_show(mddev_t
*mddev
, char *page
)
1976 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
1980 backlog_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1982 unsigned long backlog
;
1983 int rv
= strict_strtoul(buf
, 10, &backlog
);
1986 if (backlog
> COUNTER_MAX
)
1988 mddev
->bitmap_info
.max_write_behind
= backlog
;
1992 static struct md_sysfs_entry bitmap_backlog
=
1993 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
1996 chunksize_show(mddev_t
*mddev
, char *page
)
1998 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2002 chunksize_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2004 /* Can only be changed when no bitmap is active */
2006 unsigned long csize
;
2009 rv
= strict_strtoul(buf
, 10, &csize
);
2013 !is_power_of_2(csize
))
2015 mddev
->bitmap_info
.chunksize
= csize
;
2019 static struct md_sysfs_entry bitmap_chunksize
=
2020 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2022 static ssize_t
metadata_show(mddev_t
*mddev
, char *page
)
2024 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2025 ? "external" : "internal"));
2028 static ssize_t
metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2030 if (mddev
->bitmap
||
2031 mddev
->bitmap_info
.file
||
2032 mddev
->bitmap_info
.offset
)
2034 if (strncmp(buf
, "external", 8) == 0)
2035 mddev
->bitmap_info
.external
= 1;
2036 else if (strncmp(buf
, "internal", 8) == 0)
2037 mddev
->bitmap_info
.external
= 0;
2043 static struct md_sysfs_entry bitmap_metadata
=
2044 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2046 static ssize_t
can_clear_show(mddev_t
*mddev
, char *page
)
2050 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2053 len
= sprintf(page
, "\n");
2057 static ssize_t
can_clear_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2059 if (mddev
->bitmap
== NULL
)
2061 if (strncmp(buf
, "false", 5) == 0)
2062 mddev
->bitmap
->need_sync
= 1;
2063 else if (strncmp(buf
, "true", 4) == 0) {
2064 if (mddev
->degraded
)
2066 mddev
->bitmap
->need_sync
= 0;
2072 static struct md_sysfs_entry bitmap_can_clear
=
2073 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2076 behind_writes_used_show(mddev_t
*mddev
, char *page
)
2078 if (mddev
->bitmap
== NULL
)
2079 return sprintf(page
, "0\n");
2080 return sprintf(page
, "%lu\n",
2081 mddev
->bitmap
->behind_writes_used
);
2085 behind_writes_used_reset(mddev_t
*mddev
, const char *buf
, size_t len
)
2088 mddev
->bitmap
->behind_writes_used
= 0;
2092 static struct md_sysfs_entry max_backlog_used
=
2093 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2094 behind_writes_used_show
, behind_writes_used_reset
);
2096 static struct attribute
*md_bitmap_attrs
[] = {
2097 &bitmap_location
.attr
,
2098 &bitmap_timeout
.attr
,
2099 &bitmap_backlog
.attr
,
2100 &bitmap_chunksize
.attr
,
2101 &bitmap_metadata
.attr
,
2102 &bitmap_can_clear
.attr
,
2103 &max_backlog_used
.attr
,
2106 struct attribute_group md_bitmap_group
= {
2108 .attrs
= md_bitmap_attrs
,