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
);
500 sb
->state
= cpu_to_le32(bitmap
->flags
);
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
);
541 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
542 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
543 * This function verifies 'bitmap_info' and populates the on-disk bitmap
544 * structure, which is to be written to disk.
546 * Returns: 0 on success, -Exxx on error
548 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
551 unsigned long chunksize
, daemon_sleep
, write_behind
;
554 bitmap
->sb_page
= alloc_page(GFP_KERNEL
);
555 if (IS_ERR(bitmap
->sb_page
)) {
556 err
= PTR_ERR(bitmap
->sb_page
);
557 bitmap
->sb_page
= NULL
;
560 bitmap
->sb_page
->index
= 0;
562 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
564 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
565 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
567 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
569 if (!is_power_of_2(chunksize
)) {
570 kunmap_atomic(sb
, KM_USER0
);
571 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
574 sb
->chunksize
= cpu_to_le32(chunksize
);
576 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
578 (daemon_sleep
< 1) || (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
579 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
580 daemon_sleep
= 5 * HZ
;
582 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
583 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
586 * FIXME: write_behind for RAID1. If not specified, what
587 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
589 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
590 if (write_behind
> COUNTER_MAX
)
591 write_behind
= COUNTER_MAX
/ 2;
592 sb
->write_behind
= cpu_to_le32(write_behind
);
593 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
595 /* keep the array size field of the bitmap superblock up to date */
596 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
598 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
600 bitmap
->flags
|= BITMAP_STALE
;
601 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
602 bitmap
->events_cleared
= bitmap
->mddev
->events
;
603 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
605 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
606 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HOSTENDIAN
);
608 kunmap_atomic(sb
, KM_USER0
);
613 /* read the superblock from the bitmap file and initialize some bitmap fields */
614 static int bitmap_read_sb(struct bitmap
*bitmap
)
618 unsigned long chunksize
, daemon_sleep
, write_behind
;
619 unsigned long long events
;
622 /* page 0 is the superblock, read it... */
624 loff_t isize
= i_size_read(bitmap
->file
->f_mapping
->host
);
625 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
627 bitmap
->sb_page
= read_page(bitmap
->file
, 0, bitmap
, bytes
);
629 bitmap
->sb_page
= read_sb_page(bitmap
->mddev
,
630 bitmap
->mddev
->bitmap_info
.offset
,
632 0, sizeof(bitmap_super_t
));
634 if (IS_ERR(bitmap
->sb_page
)) {
635 err
= PTR_ERR(bitmap
->sb_page
);
636 bitmap
->sb_page
= NULL
;
640 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
642 chunksize
= le32_to_cpu(sb
->chunksize
);
643 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
644 write_behind
= le32_to_cpu(sb
->write_behind
);
646 /* verify that the bitmap-specific fields are valid */
647 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
648 reason
= "bad magic";
649 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
650 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
651 reason
= "unrecognized superblock version";
652 else if (chunksize
< 512)
653 reason
= "bitmap chunksize too small";
654 else if (!is_power_of_2(chunksize
))
655 reason
= "bitmap chunksize not a power of 2";
656 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
657 reason
= "daemon sleep period out of range";
658 else if (write_behind
> COUNTER_MAX
)
659 reason
= "write-behind limit out of range (0 - 16383)";
661 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
662 bmname(bitmap
), reason
);
666 /* keep the array size field of the bitmap superblock up to date */
667 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
669 if (!bitmap
->mddev
->persistent
)
673 * if we have a persistent array superblock, compare the
674 * bitmap's UUID and event counter to the mddev's
676 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
677 printk(KERN_INFO
"%s: bitmap superblock UUID mismatch\n",
681 events
= le64_to_cpu(sb
->events
);
682 if (events
< bitmap
->mddev
->events
) {
683 printk(KERN_INFO
"%s: bitmap file is out of date (%llu < %llu) "
684 "-- forcing full recovery\n", bmname(bitmap
), events
,
685 (unsigned long long) bitmap
->mddev
->events
);
686 sb
->state
|= cpu_to_le32(BITMAP_STALE
);
689 /* assign fields using values from superblock */
690 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
691 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
692 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
693 bitmap
->flags
|= le32_to_cpu(sb
->state
);
694 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
695 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
696 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
697 if (bitmap
->flags
& BITMAP_STALE
)
698 bitmap
->events_cleared
= bitmap
->mddev
->events
;
701 kunmap_atomic(sb
, KM_USER0
);
703 bitmap_print_sb(bitmap
);
707 enum bitmap_mask_op
{
712 /* record the state of the bitmap in the superblock. Return the old value */
713 static int bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
714 enum bitmap_mask_op op
)
720 spin_lock_irqsave(&bitmap
->lock
, flags
);
721 if (!bitmap
->sb_page
) { /* can't set the state */
722 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
725 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
726 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
727 old
= le32_to_cpu(sb
->state
) & bits
;
730 sb
->state
|= cpu_to_le32(bits
);
731 bitmap
->flags
|= bits
;
734 sb
->state
&= cpu_to_le32(~bits
);
735 bitmap
->flags
&= ~bits
;
740 kunmap_atomic(sb
, KM_USER0
);
745 * general bitmap file operations
751 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
752 * file a page at a time. There's a superblock at the start of the file.
754 /* calculate the index of the page that contains this bit */
755 static inline unsigned long file_page_index(struct bitmap
*bitmap
, unsigned long chunk
)
757 if (!bitmap
->mddev
->bitmap_info
.external
)
758 chunk
+= sizeof(bitmap_super_t
) << 3;
759 return chunk
>> PAGE_BIT_SHIFT
;
762 /* calculate the (bit) offset of this bit within a page */
763 static inline unsigned long file_page_offset(struct bitmap
*bitmap
, unsigned long chunk
)
765 if (!bitmap
->mddev
->bitmap_info
.external
)
766 chunk
+= sizeof(bitmap_super_t
) << 3;
767 return chunk
& (PAGE_BITS
- 1);
771 * return a pointer to the page in the filemap that contains the given bit
773 * this lookup is complicated by the fact that the bitmap sb might be exactly
774 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
777 static inline struct page
*filemap_get_page(struct bitmap
*bitmap
,
780 if (bitmap
->filemap
== NULL
)
782 if (file_page_index(bitmap
, chunk
) >= bitmap
->file_pages
)
784 return bitmap
->filemap
[file_page_index(bitmap
, chunk
)
785 - file_page_index(bitmap
, 0)];
788 static void bitmap_file_unmap(struct bitmap
*bitmap
)
790 struct page
**map
, *sb_page
;
795 spin_lock_irqsave(&bitmap
->lock
, flags
);
796 map
= bitmap
->filemap
;
797 bitmap
->filemap
= NULL
;
798 attr
= bitmap
->filemap_attr
;
799 bitmap
->filemap_attr
= NULL
;
800 pages
= bitmap
->file_pages
;
801 bitmap
->file_pages
= 0;
802 sb_page
= bitmap
->sb_page
;
803 bitmap
->sb_page
= NULL
;
804 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
807 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
808 free_buffers(map
[pages
]);
813 free_buffers(sb_page
);
816 static void bitmap_file_put(struct bitmap
*bitmap
)
821 spin_lock_irqsave(&bitmap
->lock
, flags
);
824 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
827 wait_event(bitmap
->write_wait
,
828 atomic_read(&bitmap
->pending_writes
)==0);
829 bitmap_file_unmap(bitmap
);
832 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
833 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
839 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
840 * then it is no longer reliable, so we stop using it and we mark the file
841 * as failed in the superblock
843 static void bitmap_file_kick(struct bitmap
*bitmap
)
845 char *path
, *ptr
= NULL
;
847 if (bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_SET
) == 0) {
848 bitmap_update_sb(bitmap
);
851 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
853 ptr
= d_path(&bitmap
->file
->f_path
, path
,
857 "%s: kicking failed bitmap file %s from array!\n",
858 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
863 "%s: disabling internal bitmap due to errors\n",
867 bitmap_file_put(bitmap
);
872 enum bitmap_page_attr
{
873 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
874 BITMAP_PAGE_CLEAN
= 1, /* there are bits that might need to be cleared */
875 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
878 static inline void set_page_attr(struct bitmap
*bitmap
, struct page
*page
,
879 enum bitmap_page_attr attr
)
882 __set_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
884 __set_bit(attr
, &bitmap
->logattrs
);
887 static inline void clear_page_attr(struct bitmap
*bitmap
, struct page
*page
,
888 enum bitmap_page_attr attr
)
891 __clear_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
893 __clear_bit(attr
, &bitmap
->logattrs
);
896 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, struct page
*page
,
897 enum bitmap_page_attr attr
)
900 return test_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
902 return test_bit(attr
, &bitmap
->logattrs
);
906 * bitmap_file_set_bit -- called before performing a write to the md device
907 * to set (and eventually sync) a particular bit in the bitmap file
909 * we set the bit immediately, then we record the page number so that
910 * when an unplug occurs, we can flush the dirty pages out to disk
912 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
915 struct page
*page
= NULL
;
917 unsigned long chunk
= block
>> CHUNK_BLOCK_SHIFT(bitmap
);
919 if (!bitmap
->filemap
) {
920 struct dm_dirty_log
*log
= bitmap
->mddev
->bitmap_info
.log
;
922 log
->type
->mark_region(log
, chunk
);
925 page
= filemap_get_page(bitmap
, chunk
);
928 bit
= file_page_offset(bitmap
, chunk
);
931 kaddr
= kmap_atomic(page
, KM_USER0
);
932 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
935 __test_and_set_bit_le(bit
, kaddr
);
936 kunmap_atomic(kaddr
, KM_USER0
);
937 PRINTK("set file bit %lu page %lu\n", bit
, page
->index
);
939 /* record page number so it gets flushed to disk when unplug occurs */
940 set_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
943 /* this gets called when the md device is ready to unplug its underlying
944 * (slave) device queues -- before we let any writes go down, we need to
945 * sync the dirty pages of the bitmap file to disk */
946 void bitmap_unplug(struct bitmap
*bitmap
)
948 unsigned long i
, flags
;
949 int dirty
, need_write
;
955 if (!bitmap
->filemap
) {
956 /* Must be using a dirty_log */
957 struct dm_dirty_log
*log
= bitmap
->mddev
->bitmap_info
.log
;
958 dirty
= test_and_clear_bit(BITMAP_PAGE_DIRTY
, &bitmap
->logattrs
);
959 need_write
= test_and_clear_bit(BITMAP_PAGE_NEEDWRITE
, &bitmap
->logattrs
);
960 if (dirty
|| need_write
)
961 if (log
->type
->flush(log
))
962 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
966 /* look at each page to see if there are any set bits that need to be
967 * flushed out to disk */
968 for (i
= 0; i
< bitmap
->file_pages
; i
++) {
969 spin_lock_irqsave(&bitmap
->lock
, flags
);
970 if (!bitmap
->filemap
) {
971 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
974 page
= bitmap
->filemap
[i
];
975 dirty
= test_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
976 need_write
= test_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
977 clear_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
978 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
981 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
983 if (dirty
|| need_write
)
984 write_page(bitmap
, page
, 0);
986 if (wait
) { /* if any writes were performed, we need to wait on them */
988 wait_event(bitmap
->write_wait
,
989 atomic_read(&bitmap
->pending_writes
)==0);
991 md_super_wait(bitmap
->mddev
);
994 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
995 bitmap_file_kick(bitmap
);
997 EXPORT_SYMBOL(bitmap_unplug
);
999 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
1000 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1001 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1002 * memory mapping of the bitmap file
1004 * if there's no bitmap file, or if the bitmap file had been
1005 * previously kicked from the array, we mark all the bits as
1006 * 1's in order to cause a full resync.
1008 * We ignore all bits for sectors that end earlier than 'start'.
1009 * This is used when reading an out-of-date bitmap...
1011 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
1013 unsigned long i
, chunks
, index
, oldindex
, bit
;
1014 struct page
*page
= NULL
, *oldpage
= NULL
;
1015 unsigned long num_pages
, bit_cnt
= 0;
1017 unsigned long bytes
, offset
;
1022 chunks
= bitmap
->chunks
;
1023 file
= bitmap
->file
;
1025 BUG_ON(!file
&& !bitmap
->mddev
->bitmap_info
.offset
);
1027 #ifdef INJECT_FAULTS_3
1030 outofdate
= bitmap
->flags
& BITMAP_STALE
;
1033 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
1034 "recovery\n", bmname(bitmap
));
1036 bytes
= DIV_ROUND_UP(bitmap
->chunks
, 8);
1037 if (!bitmap
->mddev
->bitmap_info
.external
)
1038 bytes
+= sizeof(bitmap_super_t
);
1040 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
1042 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
) {
1043 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
1045 (unsigned long) i_size_read(file
->f_mapping
->host
),
1052 bitmap
->filemap
= kmalloc(sizeof(struct page
*) * num_pages
, GFP_KERNEL
);
1053 if (!bitmap
->filemap
)
1056 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
1057 bitmap
->filemap_attr
= kzalloc(
1058 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
1060 if (!bitmap
->filemap_attr
)
1065 for (i
= 0; i
< chunks
; i
++) {
1067 index
= file_page_index(bitmap
, i
);
1068 bit
= file_page_offset(bitmap
, i
);
1069 if (index
!= oldindex
) { /* this is a new page, read it in */
1071 /* unmap the old page, we're done with it */
1072 if (index
== num_pages
-1)
1073 count
= bytes
- index
* PAGE_SIZE
;
1076 if (index
== 0 && bitmap
->sb_page
) {
1078 * if we're here then the superblock page
1079 * contains some bits (PAGE_SIZE != sizeof sb)
1080 * we've already read it in, so just use it
1082 page
= bitmap
->sb_page
;
1083 offset
= sizeof(bitmap_super_t
);
1085 page
= read_sb_page(
1087 bitmap
->mddev
->bitmap_info
.offset
,
1091 page
= read_page(file
, index
, bitmap
, count
);
1094 page
= read_sb_page(bitmap
->mddev
,
1095 bitmap
->mddev
->bitmap_info
.offset
,
1100 if (IS_ERR(page
)) { /* read error */
1101 ret
= PTR_ERR(page
);
1108 bitmap
->filemap
[bitmap
->file_pages
++] = page
;
1109 bitmap
->last_page_size
= count
;
1113 * if bitmap is out of date, dirty the
1114 * whole page and write it out
1116 paddr
= kmap_atomic(page
, KM_USER0
);
1117 memset(paddr
+ offset
, 0xff,
1118 PAGE_SIZE
- offset
);
1119 kunmap_atomic(paddr
, KM_USER0
);
1120 write_page(bitmap
, page
, 1);
1123 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1127 paddr
= kmap_atomic(page
, KM_USER0
);
1128 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1129 b
= test_bit(bit
, paddr
);
1131 b
= test_bit_le(bit
, paddr
);
1132 kunmap_atomic(paddr
, KM_USER0
);
1134 /* if the disk bit is set, set the memory bit */
1135 int needed
= ((sector_t
)(i
+1) << (CHUNK_BLOCK_SHIFT(bitmap
))
1137 bitmap_set_memory_bits(bitmap
,
1138 (sector_t
)i
<< CHUNK_BLOCK_SHIFT(bitmap
),
1141 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1145 /* everything went OK */
1147 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
1149 if (bit_cnt
) { /* Kick recovery if any bits were set */
1150 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1151 md_wakeup_thread(bitmap
->mddev
->thread
);
1154 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1155 "read %lu/%lu pages, set %lu of %lu bits\n",
1156 bmname(bitmap
), bitmap
->file_pages
, num_pages
, bit_cnt
, chunks
);
1161 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1162 bmname(bitmap
), ret
);
1166 void bitmap_write_all(struct bitmap
*bitmap
)
1168 /* We don't actually write all bitmap blocks here,
1169 * just flag them as needing to be written
1173 for (i
= 0; i
< bitmap
->file_pages
; i
++)
1174 set_page_attr(bitmap
, bitmap
->filemap
[i
],
1175 BITMAP_PAGE_NEEDWRITE
);
1178 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
1180 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1181 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1182 bitmap
->bp
[page
].count
+= inc
;
1183 bitmap_checkfree(bitmap
, page
);
1185 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1186 sector_t offset
, sector_t
*blocks
,
1190 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1194 void bitmap_daemon_work(mddev_t
*mddev
)
1196 struct bitmap
*bitmap
;
1198 unsigned long flags
;
1199 struct page
*page
= NULL
, *lastpage
= NULL
;
1202 struct dm_dirty_log
*log
= mddev
->bitmap_info
.log
;
1204 /* Use a mutex to guard daemon_work against
1207 mutex_lock(&mddev
->bitmap_info
.mutex
);
1208 bitmap
= mddev
->bitmap
;
1209 if (bitmap
== NULL
) {
1210 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1213 if (time_before(jiffies
, bitmap
->daemon_lastrun
1214 + bitmap
->mddev
->bitmap_info
.daemon_sleep
))
1217 bitmap
->daemon_lastrun
= jiffies
;
1218 if (bitmap
->allclean
) {
1219 bitmap
->mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1222 bitmap
->allclean
= 1;
1224 spin_lock_irqsave(&bitmap
->lock
, flags
);
1225 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1226 bitmap_counter_t
*bmc
;
1227 if (!bitmap
->filemap
) {
1229 /* error or shutdown */
1232 page
= filemap_get_page(bitmap
, j
);
1234 if (page
!= lastpage
) {
1235 /* skip this page unless it's marked as needing cleaning */
1236 if (!test_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
)) {
1237 int need_write
= test_page_attr(bitmap
, page
,
1238 BITMAP_PAGE_NEEDWRITE
);
1240 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
1242 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1244 write_page(bitmap
, page
, 0);
1245 bitmap
->allclean
= 0;
1247 spin_lock_irqsave(&bitmap
->lock
, flags
);
1248 j
|= (PAGE_BITS
- 1);
1252 /* grab the new page, sync and release the old */
1253 if (lastpage
!= NULL
) {
1254 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1255 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1256 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1257 write_page(bitmap
, lastpage
, 0);
1259 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1260 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1263 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1266 /* We are possibly going to clear some bits, so make
1267 * sure that events_cleared is up-to-date.
1269 if (bitmap
->need_sync
&&
1270 bitmap
->mddev
->bitmap_info
.external
== 0) {
1272 bitmap
->need_sync
= 0;
1273 sb
= kmap_atomic(bitmap
->sb_page
, KM_USER0
);
1274 sb
->events_cleared
=
1275 cpu_to_le64(bitmap
->events_cleared
);
1276 kunmap_atomic(sb
, KM_USER0
);
1277 write_page(bitmap
, bitmap
->sb_page
, 1);
1279 spin_lock_irqsave(&bitmap
->lock
, flags
);
1280 if (!bitmap
->need_sync
)
1281 clear_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1283 bmc
= bitmap_get_counter(bitmap
,
1284 (sector_t
)j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1288 bitmap
->allclean
= 0;
1291 *bmc
= 1; /* maybe clear the bit next time */
1292 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1293 } else if (*bmc
== 1 && !bitmap
->need_sync
) {
1294 /* we can clear the bit */
1296 bitmap_count_page(bitmap
,
1297 (sector_t
)j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1302 paddr
= kmap_atomic(page
, KM_USER0
);
1303 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1304 clear_bit(file_page_offset(bitmap
, j
),
1307 __test_and_clear_bit_le(file_page_offset(bitmap
, j
),
1309 kunmap_atomic(paddr
, KM_USER0
);
1311 log
->type
->clear_region(log
, j
);
1314 j
|= PAGE_COUNTER_MASK
;
1316 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1318 /* now sync the final page */
1319 if (lastpage
!= NULL
|| log
!= NULL
) {
1320 spin_lock_irqsave(&bitmap
->lock
, flags
);
1321 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1322 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1323 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1325 write_page(bitmap
, lastpage
, 0);
1327 if (log
->type
->flush(log
))
1328 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
1330 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1331 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1336 if (bitmap
->allclean
== 0)
1337 bitmap
->mddev
->thread
->timeout
=
1338 bitmap
->mddev
->bitmap_info
.daemon_sleep
;
1339 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1342 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1343 sector_t offset
, sector_t
*blocks
,
1345 __releases(bitmap
->lock
)
1346 __acquires(bitmap
->lock
)
1348 /* If 'create', we might release the lock and reclaim it.
1349 * The lock must have been taken with interrupts enabled.
1350 * If !create, we don't release the lock.
1352 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1353 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1354 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1358 err
= bitmap_checkpage(bitmap
, page
, create
);
1360 if (bitmap
->bp
[page
].hijacked
||
1361 bitmap
->bp
[page
].map
== NULL
)
1362 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
) +
1363 PAGE_COUNTER_SHIFT
- 1);
1365 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1366 *blocks
= csize
- (offset
& (csize
- 1));
1371 /* now locked ... */
1373 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1374 /* should we use the first or second counter field
1375 * of the hijacked pointer? */
1376 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1377 return &((bitmap_counter_t
*)
1378 &bitmap
->bp
[page
].map
)[hi
];
1379 } else /* page is allocated */
1380 return (bitmap_counter_t
*)
1381 &(bitmap
->bp
[page
].map
[pageoff
]);
1384 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1391 atomic_inc(&bitmap
->behind_writes
);
1392 bw
= atomic_read(&bitmap
->behind_writes
);
1393 if (bw
> bitmap
->behind_writes_used
)
1394 bitmap
->behind_writes_used
= bw
;
1396 PRINTK(KERN_DEBUG
"inc write-behind count %d/%d\n",
1397 bw
, bitmap
->max_write_behind
);
1402 bitmap_counter_t
*bmc
;
1404 spin_lock_irq(&bitmap
->lock
);
1405 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1407 spin_unlock_irq(&bitmap
->lock
);
1411 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1412 DEFINE_WAIT(__wait
);
1413 /* note that it is safe to do the prepare_to_wait
1414 * after the test as long as we do it before dropping
1417 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1418 TASK_UNINTERRUPTIBLE
);
1419 spin_unlock_irq(&bitmap
->lock
);
1421 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1427 bitmap_file_set_bit(bitmap
, offset
);
1428 bitmap_count_page(bitmap
, offset
, 1);
1436 spin_unlock_irq(&bitmap
->lock
);
1439 if (sectors
> blocks
)
1444 bitmap
->allclean
= 0;
1447 EXPORT_SYMBOL(bitmap_startwrite
);
1449 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1450 int success
, int behind
)
1455 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1456 wake_up(&bitmap
->behind_wait
);
1457 PRINTK(KERN_DEBUG
"dec write-behind count %d/%d\n",
1458 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1460 if (bitmap
->mddev
->degraded
)
1461 /* Never clear bits or update events_cleared when degraded */
1466 unsigned long flags
;
1467 bitmap_counter_t
*bmc
;
1469 spin_lock_irqsave(&bitmap
->lock
, flags
);
1470 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1472 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1477 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1478 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1479 bitmap
->need_sync
= 1;
1480 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1483 if (!success
&& !NEEDED(*bmc
))
1484 *bmc
|= NEEDED_MASK
;
1486 if (COUNTER(*bmc
) == COUNTER_MAX
)
1487 wake_up(&bitmap
->overflow_wait
);
1491 set_page_attr(bitmap
,
1494 offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1497 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1499 if (sectors
> blocks
)
1505 EXPORT_SYMBOL(bitmap_endwrite
);
1507 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1510 bitmap_counter_t
*bmc
;
1512 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1514 return 1; /* always resync if no bitmap */
1516 spin_lock_irq(&bitmap
->lock
);
1517 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1523 else if (NEEDED(*bmc
)) {
1525 if (!degraded
) { /* don't set/clear bits if degraded */
1526 *bmc
|= RESYNC_MASK
;
1527 *bmc
&= ~NEEDED_MASK
;
1531 spin_unlock_irq(&bitmap
->lock
);
1532 bitmap
->allclean
= 0;
1536 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1539 /* bitmap_start_sync must always report on multiples of whole
1540 * pages, otherwise resync (which is very PAGE_SIZE based) will
1542 * So call __bitmap_start_sync repeatedly (if needed) until
1543 * At least PAGE_SIZE>>9 blocks are covered.
1544 * Return the 'or' of the result.
1550 while (*blocks
< (PAGE_SIZE
>>9)) {
1551 rv
|= __bitmap_start_sync(bitmap
, offset
,
1552 &blocks1
, degraded
);
1558 EXPORT_SYMBOL(bitmap_start_sync
);
1560 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1562 bitmap_counter_t
*bmc
;
1563 unsigned long flags
;
1565 if (bitmap
== NULL
) {
1569 spin_lock_irqsave(&bitmap
->lock
, flags
);
1570 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1575 *bmc
&= ~RESYNC_MASK
;
1577 if (!NEEDED(*bmc
) && aborted
)
1578 *bmc
|= NEEDED_MASK
;
1581 set_page_attr(bitmap
,
1582 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1587 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1588 bitmap
->allclean
= 0;
1590 EXPORT_SYMBOL(bitmap_end_sync
);
1592 void bitmap_close_sync(struct bitmap
*bitmap
)
1594 /* Sync has finished, and any bitmap chunks that weren't synced
1595 * properly have been aborted. It remains to us to clear the
1596 * RESYNC bit wherever it is still on
1598 sector_t sector
= 0;
1602 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1603 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1607 EXPORT_SYMBOL(bitmap_close_sync
);
1609 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
)
1617 bitmap
->last_end_sync
= jiffies
;
1620 if (time_before(jiffies
, (bitmap
->last_end_sync
1621 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1623 wait_event(bitmap
->mddev
->recovery_wait
,
1624 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1626 bitmap
->mddev
->curr_resync_completed
= sector
;
1627 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1628 sector
&= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap
)) - 1);
1630 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1631 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1634 bitmap
->last_end_sync
= jiffies
;
1635 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1637 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1639 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1641 /* For each chunk covered by any of these sectors, set the
1642 * counter to 1 and set resync_needed. They should all
1643 * be 0 at this point
1647 bitmap_counter_t
*bmc
;
1648 spin_lock_irq(&bitmap
->lock
);
1649 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1651 spin_unlock_irq(&bitmap
->lock
);
1656 *bmc
= 1 | (needed
? NEEDED_MASK
: 0);
1657 bitmap_count_page(bitmap
, offset
, 1);
1658 page
= filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
));
1659 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1661 spin_unlock_irq(&bitmap
->lock
);
1662 bitmap
->allclean
= 0;
1665 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1666 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1668 unsigned long chunk
;
1670 for (chunk
= s
; chunk
<= e
; chunk
++) {
1671 sector_t sec
= (sector_t
)chunk
<< CHUNK_BLOCK_SHIFT(bitmap
);
1672 bitmap_set_memory_bits(bitmap
, sec
, 1);
1673 bitmap_file_set_bit(bitmap
, sec
);
1674 if (sec
< bitmap
->mddev
->recovery_cp
)
1675 /* We are asserting that the array is dirty,
1676 * so move the recovery_cp address back so
1677 * that it is obvious that it is dirty
1679 bitmap
->mddev
->recovery_cp
= sec
;
1684 * flush out any pending updates
1686 void bitmap_flush(mddev_t
*mddev
)
1688 struct bitmap
*bitmap
= mddev
->bitmap
;
1691 if (!bitmap
) /* there was no bitmap */
1694 /* run the daemon_work three time to ensure everything is flushed
1697 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1698 bitmap
->daemon_lastrun
-= sleep
;
1699 bitmap_daemon_work(mddev
);
1700 bitmap
->daemon_lastrun
-= sleep
;
1701 bitmap_daemon_work(mddev
);
1702 bitmap
->daemon_lastrun
-= sleep
;
1703 bitmap_daemon_work(mddev
);
1704 bitmap_update_sb(bitmap
);
1708 * free memory that was allocated
1710 static void bitmap_free(struct bitmap
*bitmap
)
1712 unsigned long k
, pages
;
1713 struct bitmap_page
*bp
;
1715 if (!bitmap
) /* there was no bitmap */
1718 /* release the bitmap file and kill the daemon */
1719 bitmap_file_put(bitmap
);
1722 pages
= bitmap
->pages
;
1724 /* free all allocated memory */
1726 if (bp
) /* deallocate the page memory */
1727 for (k
= 0; k
< pages
; k
++)
1728 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1734 void bitmap_destroy(mddev_t
*mddev
)
1736 struct bitmap
*bitmap
= mddev
->bitmap
;
1738 if (!bitmap
) /* there was no bitmap */
1741 mutex_lock(&mddev
->bitmap_info
.mutex
);
1742 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1743 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1745 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1747 if (bitmap
->sysfs_can_clear
)
1748 sysfs_put(bitmap
->sysfs_can_clear
);
1750 bitmap_free(bitmap
);
1754 * initialize the bitmap structure
1755 * if this returns an error, bitmap_destroy must be called to do clean up
1757 int bitmap_create(mddev_t
*mddev
)
1759 struct bitmap
*bitmap
;
1760 sector_t blocks
= mddev
->resync_max_sectors
;
1761 unsigned long chunks
;
1762 unsigned long pages
;
1763 struct file
*file
= mddev
->bitmap_info
.file
;
1765 struct sysfs_dirent
*bm
= NULL
;
1767 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1770 && !mddev
->bitmap_info
.offset
1771 && !mddev
->bitmap_info
.log
) /* bitmap disabled, nothing to do */
1774 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1775 BUG_ON(mddev
->bitmap_info
.offset
&& mddev
->bitmap_info
.log
);
1777 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1781 spin_lock_init(&bitmap
->lock
);
1782 atomic_set(&bitmap
->pending_writes
, 0);
1783 init_waitqueue_head(&bitmap
->write_wait
);
1784 init_waitqueue_head(&bitmap
->overflow_wait
);
1785 init_waitqueue_head(&bitmap
->behind_wait
);
1787 bitmap
->mddev
= mddev
;
1790 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "bitmap");
1792 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, NULL
, "can_clear");
1795 bitmap
->sysfs_can_clear
= NULL
;
1797 bitmap
->file
= file
;
1800 /* As future accesses to this file will use bmap,
1801 * and bypass the page cache, we must sync the file
1806 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1807 if (!mddev
->bitmap_info
.external
) {
1809 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1810 * instructing us to create a new on-disk bitmap instance.
1812 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1813 err
= bitmap_new_disk_sb(bitmap
);
1815 err
= bitmap_read_sb(bitmap
);
1818 if (mddev
->bitmap_info
.chunksize
== 0 ||
1819 mddev
->bitmap_info
.daemon_sleep
== 0)
1820 /* chunksize and time_base need to be
1827 bitmap
->daemon_lastrun
= jiffies
;
1828 bitmap
->chunkshift
= ffz(~mddev
->bitmap_info
.chunksize
);
1830 /* now that chunksize and chunkshift are set, we can use these macros */
1831 chunks
= (blocks
+ CHUNK_BLOCK_RATIO(bitmap
) - 1) >>
1832 CHUNK_BLOCK_SHIFT(bitmap
);
1833 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1837 bitmap
->chunks
= chunks
;
1838 bitmap
->pages
= pages
;
1839 bitmap
->missing_pages
= pages
;
1841 #ifdef INJECT_FATAL_FAULT_1
1844 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1850 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1851 pages
, bmname(bitmap
));
1853 mddev
->bitmap
= bitmap
;
1856 return (bitmap
->flags
& BITMAP_WRITE_ERROR
) ? -EIO
: 0;
1859 bitmap_free(bitmap
);
1863 int bitmap_load(mddev_t
*mddev
)
1866 sector_t sector
= 0;
1867 struct bitmap
*bitmap
= mddev
->bitmap
;
1872 /* Clear out old bitmap info first: Either there is none, or we
1873 * are resuming after someone else has possibly changed things,
1874 * so we should forget old cached info.
1875 * All chunks should be clean, but some might need_sync.
1877 while (sector
< mddev
->resync_max_sectors
) {
1879 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1882 bitmap_close_sync(bitmap
);
1884 if (mddev
->bitmap_info
.log
) {
1886 struct dm_dirty_log
*log
= mddev
->bitmap_info
.log
;
1887 for (i
= 0; i
< bitmap
->chunks
; i
++)
1888 if (!log
->type
->in_sync(log
, i
, 1))
1889 bitmap_set_memory_bits(bitmap
,
1890 (sector_t
)i
<< CHUNK_BLOCK_SHIFT(bitmap
),
1894 if (mddev
->degraded
== 0
1895 || bitmap
->events_cleared
== mddev
->events
)
1896 /* no need to keep dirty bits to optimise a
1897 * re-add of a missing device */
1898 start
= mddev
->recovery_cp
;
1900 err
= bitmap_init_from_disk(bitmap
, start
);
1905 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1906 md_wakeup_thread(mddev
->thread
);
1908 bitmap_update_sb(bitmap
);
1910 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
1915 EXPORT_SYMBOL_GPL(bitmap_load
);
1918 location_show(mddev_t
*mddev
, char *page
)
1921 if (mddev
->bitmap_info
.file
)
1922 len
= sprintf(page
, "file");
1923 else if (mddev
->bitmap_info
.offset
)
1924 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
1926 len
= sprintf(page
, "none");
1927 len
+= sprintf(page
+len
, "\n");
1932 location_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1936 if (!mddev
->pers
->quiesce
)
1938 if (mddev
->recovery
|| mddev
->sync_thread
)
1942 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
1943 mddev
->bitmap_info
.offset
) {
1944 /* bitmap already configured. Only option is to clear it */
1945 if (strncmp(buf
, "none", 4) != 0)
1948 mddev
->pers
->quiesce(mddev
, 1);
1949 bitmap_destroy(mddev
);
1950 mddev
->pers
->quiesce(mddev
, 0);
1952 mddev
->bitmap_info
.offset
= 0;
1953 if (mddev
->bitmap_info
.file
) {
1954 struct file
*f
= mddev
->bitmap_info
.file
;
1955 mddev
->bitmap_info
.file
= NULL
;
1956 restore_bitmap_write_access(f
);
1960 /* No bitmap, OK to set a location */
1962 if (strncmp(buf
, "none", 4) == 0)
1963 /* nothing to be done */;
1964 else if (strncmp(buf
, "file:", 5) == 0) {
1965 /* Not supported yet */
1970 rv
= strict_strtoll(buf
+1, 10, &offset
);
1972 rv
= strict_strtoll(buf
, 10, &offset
);
1977 if (mddev
->bitmap_info
.external
== 0 &&
1978 mddev
->major_version
== 0 &&
1979 offset
!= mddev
->bitmap_info
.default_offset
)
1981 mddev
->bitmap_info
.offset
= offset
;
1983 mddev
->pers
->quiesce(mddev
, 1);
1984 rv
= bitmap_create(mddev
);
1986 bitmap_destroy(mddev
);
1987 mddev
->bitmap_info
.offset
= 0;
1989 mddev
->pers
->quiesce(mddev
, 0);
1995 if (!mddev
->external
) {
1996 /* Ensure new bitmap info is stored in
1997 * metadata promptly.
1999 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2000 md_wakeup_thread(mddev
->thread
);
2005 static struct md_sysfs_entry bitmap_location
=
2006 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
2009 timeout_show(mddev_t
*mddev
, char *page
)
2012 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
2013 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
2015 len
= sprintf(page
, "%lu", secs
);
2017 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
2018 len
+= sprintf(page
+len
, "\n");
2023 timeout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2025 /* timeout can be set at any time */
2026 unsigned long timeout
;
2027 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
2031 /* just to make sure we don't overflow... */
2032 if (timeout
>= LONG_MAX
/ HZ
)
2035 timeout
= timeout
* HZ
/ 10000;
2037 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
2038 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2041 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2042 if (mddev
->thread
) {
2043 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2044 * the bitmap is all clean and we don't need to
2045 * adjust the timeout right now
2047 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2048 mddev
->thread
->timeout
= timeout
;
2049 md_wakeup_thread(mddev
->thread
);
2055 static struct md_sysfs_entry bitmap_timeout
=
2056 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2059 backlog_show(mddev_t
*mddev
, char *page
)
2061 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2065 backlog_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2067 unsigned long backlog
;
2068 int rv
= strict_strtoul(buf
, 10, &backlog
);
2071 if (backlog
> COUNTER_MAX
)
2073 mddev
->bitmap_info
.max_write_behind
= backlog
;
2077 static struct md_sysfs_entry bitmap_backlog
=
2078 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2081 chunksize_show(mddev_t
*mddev
, char *page
)
2083 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2087 chunksize_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2089 /* Can only be changed when no bitmap is active */
2091 unsigned long csize
;
2094 rv
= strict_strtoul(buf
, 10, &csize
);
2098 !is_power_of_2(csize
))
2100 mddev
->bitmap_info
.chunksize
= csize
;
2104 static struct md_sysfs_entry bitmap_chunksize
=
2105 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2107 static ssize_t
metadata_show(mddev_t
*mddev
, char *page
)
2109 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2110 ? "external" : "internal"));
2113 static ssize_t
metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2115 if (mddev
->bitmap
||
2116 mddev
->bitmap_info
.file
||
2117 mddev
->bitmap_info
.offset
)
2119 if (strncmp(buf
, "external", 8) == 0)
2120 mddev
->bitmap_info
.external
= 1;
2121 else if (strncmp(buf
, "internal", 8) == 0)
2122 mddev
->bitmap_info
.external
= 0;
2128 static struct md_sysfs_entry bitmap_metadata
=
2129 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2131 static ssize_t
can_clear_show(mddev_t
*mddev
, char *page
)
2135 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2138 len
= sprintf(page
, "\n");
2142 static ssize_t
can_clear_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2144 if (mddev
->bitmap
== NULL
)
2146 if (strncmp(buf
, "false", 5) == 0)
2147 mddev
->bitmap
->need_sync
= 1;
2148 else if (strncmp(buf
, "true", 4) == 0) {
2149 if (mddev
->degraded
)
2151 mddev
->bitmap
->need_sync
= 0;
2157 static struct md_sysfs_entry bitmap_can_clear
=
2158 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2161 behind_writes_used_show(mddev_t
*mddev
, char *page
)
2163 if (mddev
->bitmap
== NULL
)
2164 return sprintf(page
, "0\n");
2165 return sprintf(page
, "%lu\n",
2166 mddev
->bitmap
->behind_writes_used
);
2170 behind_writes_used_reset(mddev_t
*mddev
, const char *buf
, size_t len
)
2173 mddev
->bitmap
->behind_writes_used
= 0;
2177 static struct md_sysfs_entry max_backlog_used
=
2178 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2179 behind_writes_used_show
, behind_writes_used_reset
);
2181 static struct attribute
*md_bitmap_attrs
[] = {
2182 &bitmap_location
.attr
,
2183 &bitmap_timeout
.attr
,
2184 &bitmap_backlog
.attr
,
2185 &bitmap_chunksize
.attr
,
2186 &bitmap_metadata
.attr
,
2187 &bitmap_can_clear
.attr
,
2188 &max_backlog_used
.attr
,
2191 struct attribute_group md_bitmap_group
= {
2193 .attrs
= md_bitmap_attrs
,