2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
15 * flush after percent set rather than just time based. (maybe both).
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
33 static inline char *bmname(struct bitmap
*bitmap
)
35 return bitmap
->mddev
? mdname(bitmap
->mddev
) : "mdX";
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
48 static int bitmap_checkpage(struct bitmap
*bitmap
,
49 unsigned long page
, int create
)
50 __releases(bitmap
->lock
)
51 __acquires(bitmap
->lock
)
53 unsigned char *mappage
;
55 if (page
>= bitmap
->pages
) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
63 if (bitmap
->bp
[page
].hijacked
) /* it's hijacked, don't try to alloc */
66 if (bitmap
->bp
[page
].map
) /* page is already allocated, just return */
72 /* this page has not been allocated yet */
74 spin_unlock_irq(&bitmap
->lock
);
75 mappage
= kzalloc(PAGE_SIZE
, GFP_NOIO
);
76 spin_lock_irq(&bitmap
->lock
);
78 if (mappage
== NULL
) {
79 pr_debug("%s: bitmap map page allocation failed, hijacking\n",
81 /* failed - set the hijacked flag so that we can use the
82 * pointer as a counter */
83 if (!bitmap
->bp
[page
].map
)
84 bitmap
->bp
[page
].hijacked
= 1;
85 } else if (bitmap
->bp
[page
].map
||
86 bitmap
->bp
[page
].hijacked
) {
87 /* somebody beat us to getting the page */
92 /* no page was in place and we have one, so install it */
94 bitmap
->bp
[page
].map
= mappage
;
95 bitmap
->missing_pages
--;
100 /* if page is completely empty, put it back on the free list, or dealloc it */
101 /* if page was hijacked, unmark the flag so it might get alloced next time */
102 /* Note: lock should be held when calling this */
103 static void bitmap_checkfree(struct bitmap
*bitmap
, unsigned long page
)
107 if (bitmap
->bp
[page
].count
) /* page is still busy */
110 /* page is no longer in use, it can be released */
112 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
113 bitmap
->bp
[page
].hijacked
= 0;
114 bitmap
->bp
[page
].map
= NULL
;
116 /* normal case, free the page */
117 ptr
= bitmap
->bp
[page
].map
;
118 bitmap
->bp
[page
].map
= NULL
;
119 bitmap
->missing_pages
++;
125 * bitmap file handling - read and write the bitmap file and its superblock
129 * basic page I/O operations
132 /* IO operations when bitmap is stored near all superblocks */
133 static int read_sb_page(struct mddev
*mddev
, loff_t offset
,
135 unsigned long index
, int size
)
137 /* choose a good rdev and read the page from there */
139 struct md_rdev
*rdev
;
142 rdev_for_each(rdev
, mddev
) {
143 if (! test_bit(In_sync
, &rdev
->flags
)
144 || test_bit(Faulty
, &rdev
->flags
))
147 target
= offset
+ index
* (PAGE_SIZE
/512);
149 if (sync_page_io(rdev
, target
,
150 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
159 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
161 /* Iterate the disks of an mddev, using rcu to protect access to the
162 * linked list, and raising the refcount of devices we return to ensure
163 * they don't disappear while in use.
164 * As devices are only added or removed when raid_disk is < 0 and
165 * nr_pending is 0 and In_sync is clear, the entries we return will
166 * still be in the same position on the list when we re-enter
167 * list_for_each_continue_rcu.
169 struct list_head
*pos
;
172 /* start at the beginning */
175 /* release the previous rdev and start from there. */
176 rdev_dec_pending(rdev
, mddev
);
177 pos
= &rdev
->same_set
;
179 list_for_each_continue_rcu(pos
, &mddev
->disks
) {
180 rdev
= list_entry(pos
, struct md_rdev
, same_set
);
181 if (rdev
->raid_disk
>= 0 &&
182 !test_bit(Faulty
, &rdev
->flags
)) {
183 /* this is a usable devices */
184 atomic_inc(&rdev
->nr_pending
);
193 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
195 struct md_rdev
*rdev
= NULL
;
196 struct block_device
*bdev
;
197 struct mddev
*mddev
= bitmap
->mddev
;
198 struct bitmap_storage
*store
= &bitmap
->storage
;
200 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
201 int size
= PAGE_SIZE
;
202 loff_t offset
= mddev
->bitmap_info
.offset
;
204 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
206 if (page
->index
== store
->file_pages
-1) {
207 int last_page_size
= store
->bytes
& (PAGE_SIZE
-1);
208 if (last_page_size
== 0)
209 last_page_size
= PAGE_SIZE
;
210 size
= roundup(last_page_size
,
211 bdev_logical_block_size(bdev
));
213 /* Just make sure we aren't corrupting data or
216 if (mddev
->external
) {
217 /* Bitmap could be anywhere. */
218 if (rdev
->sb_start
+ offset
+ (page
->index
222 rdev
->sb_start
+ offset
223 < (rdev
->data_offset
+ mddev
->dev_sectors
226 } else if (offset
< 0) {
227 /* DATA BITMAP METADATA */
229 + (long)(page
->index
* (PAGE_SIZE
/512))
231 /* bitmap runs in to metadata */
233 if (rdev
->data_offset
+ mddev
->dev_sectors
234 > rdev
->sb_start
+ offset
)
235 /* data runs in to bitmap */
237 } else if (rdev
->sb_start
< rdev
->data_offset
) {
238 /* METADATA BITMAP DATA */
241 + page
->index
*(PAGE_SIZE
/512) + size
/512
243 /* bitmap runs in to data */
246 /* DATA METADATA BITMAP - no problems */
248 md_super_write(mddev
, rdev
,
249 rdev
->sb_start
+ offset
250 + page
->index
* (PAGE_SIZE
/512),
256 md_super_wait(mddev
);
263 static void bitmap_file_kick(struct bitmap
*bitmap
);
265 * write out a page to a file
267 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
269 struct buffer_head
*bh
;
271 if (bitmap
->storage
.file
== NULL
) {
272 switch (write_sb_page(bitmap
, page
, wait
)) {
274 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
278 bh
= page_buffers(page
);
280 while (bh
&& bh
->b_blocknr
) {
281 atomic_inc(&bitmap
->pending_writes
);
282 set_buffer_locked(bh
);
283 set_buffer_mapped(bh
);
284 submit_bh(WRITE
| REQ_SYNC
, bh
);
285 bh
= bh
->b_this_page
;
289 wait_event(bitmap
->write_wait
,
290 atomic_read(&bitmap
->pending_writes
)==0);
292 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
293 bitmap_file_kick(bitmap
);
296 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
298 struct bitmap
*bitmap
= bh
->b_private
;
301 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
302 if (atomic_dec_and_test(&bitmap
->pending_writes
))
303 wake_up(&bitmap
->write_wait
);
306 /* copied from buffer.c */
308 __clear_page_buffers(struct page
*page
)
310 ClearPagePrivate(page
);
311 set_page_private(page
, 0);
312 page_cache_release(page
);
314 static void free_buffers(struct page
*page
)
316 struct buffer_head
*bh
;
318 if (!PagePrivate(page
))
321 bh
= page_buffers(page
);
323 struct buffer_head
*next
= bh
->b_this_page
;
324 free_buffer_head(bh
);
327 __clear_page_buffers(page
);
331 /* read a page from a file.
332 * We both read the page, and attach buffers to the page to record the
333 * address of each block (using bmap). These addresses will be used
334 * to write the block later, completely bypassing the filesystem.
335 * This usage is similar to how swap files are handled, and allows us
336 * to write to a file with no concerns of memory allocation failing.
338 static int read_page(struct file
*file
, unsigned long index
,
339 struct bitmap
*bitmap
,
344 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
345 struct buffer_head
*bh
;
348 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
349 (unsigned long long)index
<< PAGE_SHIFT
);
351 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
356 attach_page_buffers(page
, bh
);
357 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
362 bh
->b_blocknr
= bmap(inode
, block
);
363 if (bh
->b_blocknr
== 0) {
364 /* Cannot use this file! */
368 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
369 if (count
< (1<<inode
->i_blkbits
))
372 count
-= (1<<inode
->i_blkbits
);
374 bh
->b_end_io
= end_bitmap_write
;
375 bh
->b_private
= bitmap
;
376 atomic_inc(&bitmap
->pending_writes
);
377 set_buffer_locked(bh
);
378 set_buffer_mapped(bh
);
382 bh
= bh
->b_this_page
;
386 wait_event(bitmap
->write_wait
,
387 atomic_read(&bitmap
->pending_writes
)==0);
388 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
392 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %d\n",
394 (unsigned long long)index
<< PAGE_SHIFT
,
400 * bitmap file superblock operations
403 /* update the event counter and sync the superblock to disk */
404 void bitmap_update_sb(struct bitmap
*bitmap
)
408 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
410 if (bitmap
->mddev
->bitmap_info
.external
)
412 if (!bitmap
->storage
.sb_page
) /* no superblock */
414 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
415 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
416 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
417 /* rocking back to read-only */
418 bitmap
->events_cleared
= bitmap
->mddev
->events
;
419 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
420 sb
->state
= cpu_to_le32(bitmap
->flags
);
421 /* Just in case these have been changed via sysfs: */
422 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
423 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
425 write_page(bitmap
, bitmap
->storage
.sb_page
, 1);
428 /* print out the bitmap file superblock */
429 void bitmap_print_sb(struct bitmap
*bitmap
)
433 if (!bitmap
|| !bitmap
->storage
.sb_page
)
435 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
436 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
437 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
438 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
439 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
440 *(__u32
*)(sb
->uuid
+0),
441 *(__u32
*)(sb
->uuid
+4),
442 *(__u32
*)(sb
->uuid
+8),
443 *(__u32
*)(sb
->uuid
+12));
444 printk(KERN_DEBUG
" events: %llu\n",
445 (unsigned long long) le64_to_cpu(sb
->events
));
446 printk(KERN_DEBUG
"events cleared: %llu\n",
447 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
448 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
449 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
450 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
451 printk(KERN_DEBUG
" sync size: %llu KB\n",
452 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
453 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
461 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
462 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
463 * This function verifies 'bitmap_info' and populates the on-disk bitmap
464 * structure, which is to be written to disk.
466 * Returns: 0 on success, -Exxx on error
468 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
471 unsigned long chunksize
, daemon_sleep
, write_behind
;
474 bitmap
->storage
.sb_page
= alloc_page(GFP_KERNEL
);
475 if (IS_ERR(bitmap
->storage
.sb_page
)) {
476 err
= PTR_ERR(bitmap
->storage
.sb_page
);
477 bitmap
->storage
.sb_page
= NULL
;
480 bitmap
->storage
.sb_page
->index
= 0;
482 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
484 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
485 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
487 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
489 if (!is_power_of_2(chunksize
)) {
491 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
494 sb
->chunksize
= cpu_to_le32(chunksize
);
496 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
498 (daemon_sleep
< 1) || (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
499 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
500 daemon_sleep
= 5 * HZ
;
502 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
503 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
506 * FIXME: write_behind for RAID1. If not specified, what
507 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
509 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
510 if (write_behind
> COUNTER_MAX
)
511 write_behind
= COUNTER_MAX
/ 2;
512 sb
->write_behind
= cpu_to_le32(write_behind
);
513 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
515 /* keep the array size field of the bitmap superblock up to date */
516 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
518 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
520 set_bit(BITMAP_STALE
, &bitmap
->flags
);
521 sb
->state
= cpu_to_le32(bitmap
->flags
);
522 bitmap
->events_cleared
= bitmap
->mddev
->events
;
523 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
530 /* read the superblock from the bitmap file and initialize some bitmap fields */
531 static int bitmap_read_sb(struct bitmap
*bitmap
)
535 unsigned long chunksize
, daemon_sleep
, write_behind
;
536 unsigned long long events
;
538 struct page
*sb_page
;
540 if (!bitmap
->storage
.file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
541 chunksize
= 128 * 1024 * 1024;
542 daemon_sleep
= 5 * HZ
;
544 set_bit(BITMAP_STALE
, &bitmap
->flags
);
548 /* page 0 is the superblock, read it... */
549 sb_page
= alloc_page(GFP_KERNEL
);
552 bitmap
->storage
.sb_page
= sb_page
;
554 if (bitmap
->storage
.file
) {
555 loff_t isize
= i_size_read(bitmap
->storage
.file
->f_mapping
->host
);
556 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
558 err
= read_page(bitmap
->storage
.file
, 0,
559 bitmap
, bytes
, sb_page
);
561 err
= read_sb_page(bitmap
->mddev
,
562 bitmap
->mddev
->bitmap_info
.offset
,
564 0, sizeof(bitmap_super_t
));
569 sb
= kmap_atomic(sb_page
);
571 chunksize
= le32_to_cpu(sb
->chunksize
);
572 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
573 write_behind
= le32_to_cpu(sb
->write_behind
);
575 /* verify that the bitmap-specific fields are valid */
576 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
577 reason
= "bad magic";
578 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
579 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
580 reason
= "unrecognized superblock version";
581 else if (chunksize
< 512)
582 reason
= "bitmap chunksize too small";
583 else if (!is_power_of_2(chunksize
))
584 reason
= "bitmap chunksize not a power of 2";
585 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
586 reason
= "daemon sleep period out of range";
587 else if (write_behind
> COUNTER_MAX
)
588 reason
= "write-behind limit out of range (0 - 16383)";
590 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
591 bmname(bitmap
), reason
);
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 if (bitmap
->mddev
->persistent
) {
600 * We have a persistent array superblock, so compare the
601 * bitmap's UUID and event counter to the mddev's
603 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
605 "%s: bitmap superblock UUID mismatch\n",
609 events
= le64_to_cpu(sb
->events
);
610 if (events
< bitmap
->mddev
->events
) {
612 "%s: bitmap file is out of date (%llu < %llu) "
613 "-- forcing full recovery\n",
614 bmname(bitmap
), events
,
615 (unsigned long long) bitmap
->mddev
->events
);
616 set_bit(BITMAP_STALE
, &bitmap
->flags
);
620 /* assign fields using values from superblock */
621 bitmap
->flags
|= le32_to_cpu(sb
->state
);
622 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
623 set_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
);
624 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
629 if (test_bit(BITMAP_STALE
, &bitmap
->flags
))
630 bitmap
->events_cleared
= bitmap
->mddev
->events
;
631 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
632 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
633 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
635 bitmap_print_sb(bitmap
);
640 * general bitmap file operations
646 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
647 * file a page at a time. There's a superblock at the start of the file.
649 /* calculate the index of the page that contains this bit */
650 static inline unsigned long file_page_index(struct bitmap_storage
*store
,
654 chunk
+= sizeof(bitmap_super_t
) << 3;
655 return chunk
>> PAGE_BIT_SHIFT
;
658 /* calculate the (bit) offset of this bit within a page */
659 static inline unsigned long file_page_offset(struct bitmap_storage
*store
,
663 chunk
+= sizeof(bitmap_super_t
) << 3;
664 return chunk
& (PAGE_BITS
- 1);
668 * return a pointer to the page in the filemap that contains the given bit
670 * this lookup is complicated by the fact that the bitmap sb might be exactly
671 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
674 static inline struct page
*filemap_get_page(struct bitmap_storage
*store
,
677 if (file_page_index(store
, chunk
) >= store
->file_pages
)
679 return store
->filemap
[file_page_index(store
, chunk
)
680 - file_page_index(store
, 0)];
683 static int bitmap_storage_alloc(struct bitmap_storage
*store
,
684 unsigned long chunks
, int with_super
)
687 unsigned long num_pages
;
690 bytes
= DIV_ROUND_UP(chunks
, 8);
692 bytes
+= sizeof(bitmap_super_t
);
694 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
696 store
->filemap
= kmalloc(sizeof(struct page
*)
697 * num_pages
, GFP_KERNEL
);
701 if (with_super
&& !store
->sb_page
) {
702 store
->sb_page
= alloc_page(GFP_KERNEL
);
703 if (store
->sb_page
== NULL
)
705 store
->sb_page
->index
= 0;
708 if (store
->sb_page
) {
709 store
->filemap
[0] = store
->sb_page
;
712 for ( ; pnum
< num_pages
; pnum
++) {
713 store
->filemap
[pnum
] = alloc_page(GFP_KERNEL
);
714 if (!store
->filemap
[pnum
]) {
715 store
->file_pages
= pnum
;
718 store
->filemap
[pnum
]->index
= pnum
;
720 store
->file_pages
= pnum
;
722 /* We need 4 bits per page, rounded up to a multiple
723 * of sizeof(unsigned long) */
724 store
->filemap_attr
= kzalloc(
725 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
727 if (!store
->filemap_attr
)
730 store
->bytes
= bytes
;
735 static void bitmap_file_unmap(struct bitmap
*bitmap
)
737 struct page
**map
, *sb_page
;
740 struct bitmap_storage
*store
= &bitmap
->storage
;
742 spin_lock_irq(&bitmap
->lock
);
743 map
= store
->filemap
;
744 store
->filemap
= NULL
;
745 attr
= store
->filemap_attr
;
746 store
->filemap_attr
= NULL
;
747 pages
= store
->file_pages
;
748 store
->file_pages
= 0;
749 sb_page
= store
->sb_page
;
750 store
->sb_page
= NULL
;
751 spin_unlock_irq(&bitmap
->lock
);
754 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
755 free_buffers(map
[pages
]);
760 free_buffers(sb_page
);
763 static void bitmap_file_put(struct bitmap
*bitmap
)
767 spin_lock_irq(&bitmap
->lock
);
768 file
= bitmap
->storage
.file
;
769 bitmap
->storage
.file
= NULL
;
770 spin_unlock_irq(&bitmap
->lock
);
773 wait_event(bitmap
->write_wait
,
774 atomic_read(&bitmap
->pending_writes
)==0);
775 bitmap_file_unmap(bitmap
);
778 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
779 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
785 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
786 * then it is no longer reliable, so we stop using it and we mark the file
787 * as failed in the superblock
789 static void bitmap_file_kick(struct bitmap
*bitmap
)
791 char *path
, *ptr
= NULL
;
793 if (!test_and_set_bit(BITMAP_STALE
, &bitmap
->flags
)) {
794 bitmap_update_sb(bitmap
);
796 if (bitmap
->storage
.file
) {
797 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
799 ptr
= d_path(&bitmap
->storage
.file
->f_path
,
803 "%s: kicking failed bitmap file %s from array!\n",
804 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
809 "%s: disabling internal bitmap due to errors\n",
813 bitmap_file_put(bitmap
);
818 enum bitmap_page_attr
{
819 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
820 BITMAP_PAGE_PENDING
= 1, /* there are bits that are being cleaned.
821 * i.e. counter is 1 or 2. */
822 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
825 static inline void set_page_attr(struct bitmap
*bitmap
, int pnum
,
826 enum bitmap_page_attr attr
)
828 __set_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
831 static inline void clear_page_attr(struct bitmap
*bitmap
, int pnum
,
832 enum bitmap_page_attr attr
)
834 __clear_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
837 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, int pnum
,
838 enum bitmap_page_attr attr
)
840 return test_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
844 * bitmap_file_set_bit -- called before performing a write to the md device
845 * to set (and eventually sync) a particular bit in the bitmap file
847 * we set the bit immediately, then we record the page number so that
848 * when an unplug occurs, we can flush the dirty pages out to disk
850 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
855 unsigned long chunk
= block
>> bitmap
->chunkshift
;
857 page
= filemap_get_page(&bitmap
->storage
, chunk
);
860 bit
= file_page_offset(&bitmap
->storage
, chunk
);
863 kaddr
= kmap_atomic(page
);
864 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
867 __set_bit_le(bit
, kaddr
);
868 kunmap_atomic(kaddr
);
869 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
870 /* record page number so it gets flushed to disk when unplug occurs */
871 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_DIRTY
);
874 static void bitmap_file_clear_bit(struct bitmap
*bitmap
, sector_t block
)
879 unsigned long chunk
= block
>> bitmap
->chunkshift
;
881 page
= filemap_get_page(&bitmap
->storage
, chunk
);
884 bit
= file_page_offset(&bitmap
->storage
, chunk
);
885 paddr
= kmap_atomic(page
);
886 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
887 clear_bit(bit
, paddr
);
889 __clear_bit_le(bit
, paddr
);
890 kunmap_atomic(paddr
);
891 if (!test_page_attr(bitmap
, page
->index
, BITMAP_PAGE_NEEDWRITE
)) {
892 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_PENDING
);
893 bitmap
->allclean
= 0;
897 /* this gets called when the md device is ready to unplug its underlying
898 * (slave) device queues -- before we let any writes go down, we need to
899 * sync the dirty pages of the bitmap file to disk */
900 void bitmap_unplug(struct bitmap
*bitmap
)
903 int dirty
, need_write
;
906 if (!bitmap
|| !bitmap
->storage
.filemap
)
909 /* look at each page to see if there are any set bits that need to be
910 * flushed out to disk */
911 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++) {
912 spin_lock_irq(&bitmap
->lock
);
913 if (!bitmap
->storage
.filemap
) {
914 spin_unlock_irq(&bitmap
->lock
);
917 dirty
= test_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
918 need_write
= test_page_attr(bitmap
, i
, BITMAP_PAGE_NEEDWRITE
);
919 clear_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
920 clear_page_attr(bitmap
, i
, BITMAP_PAGE_NEEDWRITE
);
921 if (dirty
|| need_write
)
922 clear_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
);
925 spin_unlock_irq(&bitmap
->lock
);
927 if (dirty
|| need_write
)
928 write_page(bitmap
, bitmap
->storage
.filemap
[i
], 0);
930 if (wait
) { /* if any writes were performed, we need to wait on them */
931 if (bitmap
->storage
.file
)
932 wait_event(bitmap
->write_wait
,
933 atomic_read(&bitmap
->pending_writes
)==0);
935 md_super_wait(bitmap
->mddev
);
937 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
938 bitmap_file_kick(bitmap
);
940 EXPORT_SYMBOL(bitmap_unplug
);
942 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
943 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
944 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
945 * memory mapping of the bitmap file
947 * if there's no bitmap file, or if the bitmap file had been
948 * previously kicked from the array, we mark all the bits as
949 * 1's in order to cause a full resync.
951 * We ignore all bits for sectors that end earlier than 'start'.
952 * This is used when reading an out-of-date bitmap...
954 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
956 unsigned long i
, chunks
, index
, oldindex
, bit
;
957 struct page
*page
= NULL
;
958 unsigned long bit_cnt
= 0;
960 unsigned long offset
;
964 struct bitmap_storage
*store
= &bitmap
->storage
;
966 chunks
= bitmap
->chunks
;
969 if (!file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
970 /* No permanent bitmap - fill with '1s'. */
971 store
->filemap
= NULL
;
972 store
->file_pages
= 0;
973 for (i
= 0; i
< chunks
; i
++) {
974 /* if the disk bit is set, set the memory bit */
975 int needed
= ((sector_t
)(i
+1) << (bitmap
->chunkshift
)
977 bitmap_set_memory_bits(bitmap
,
978 (sector_t
)i
<< bitmap
->chunkshift
,
984 outofdate
= test_bit(BITMAP_STALE
, &bitmap
->flags
);
986 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
987 "recovery\n", bmname(bitmap
));
989 if (file
&& i_size_read(file
->f_mapping
->host
) < store
->bytes
) {
990 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
992 (unsigned long) i_size_read(file
->f_mapping
->host
),
999 if (!bitmap
->mddev
->bitmap_info
.external
)
1000 offset
= sizeof(bitmap_super_t
);
1002 for (i
= 0; i
< chunks
; i
++) {
1004 index
= file_page_index(&bitmap
->storage
, i
);
1005 bit
= file_page_offset(&bitmap
->storage
, i
);
1006 if (index
!= oldindex
) { /* this is a new page, read it in */
1008 /* unmap the old page, we're done with it */
1009 if (index
== store
->file_pages
-1)
1010 count
= store
->bytes
- index
* PAGE_SIZE
;
1013 page
= store
->filemap
[index
];
1015 ret
= read_page(file
, index
, bitmap
,
1020 bitmap
->mddev
->bitmap_info
.offset
,
1031 * if bitmap is out of date, dirty the
1032 * whole page and write it out
1034 paddr
= kmap_atomic(page
);
1035 memset(paddr
+ offset
, 0xff,
1036 PAGE_SIZE
- offset
);
1037 kunmap_atomic(paddr
);
1038 write_page(bitmap
, page
, 1);
1041 if (test_bit(BITMAP_WRITE_ERROR
,
1046 paddr
= kmap_atomic(page
);
1047 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
1048 b
= test_bit(bit
, paddr
);
1050 b
= test_bit_le(bit
, paddr
);
1051 kunmap_atomic(paddr
);
1053 /* if the disk bit is set, set the memory bit */
1054 int needed
= ((sector_t
)(i
+1) << bitmap
->chunkshift
1056 bitmap_set_memory_bits(bitmap
,
1057 (sector_t
)i
<< bitmap
->chunkshift
,
1064 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1065 "read %lu pages, set %lu of %lu bits\n",
1066 bmname(bitmap
), store
->file_pages
,
1072 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1073 bmname(bitmap
), ret
);
1077 void bitmap_write_all(struct bitmap
*bitmap
)
1079 /* We don't actually write all bitmap blocks here,
1080 * just flag them as needing to be written
1084 if (!bitmap
|| !bitmap
->storage
.filemap
)
1086 if (bitmap
->storage
.file
)
1087 /* Only one copy, so nothing needed */
1090 spin_lock_irq(&bitmap
->lock
);
1091 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1092 set_page_attr(bitmap
, i
,
1093 BITMAP_PAGE_NEEDWRITE
);
1094 bitmap
->allclean
= 0;
1095 spin_unlock_irq(&bitmap
->lock
);
1098 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
1100 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1101 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1102 bitmap
->bp
[page
].count
+= inc
;
1103 bitmap_checkfree(bitmap
, page
);
1106 static void bitmap_set_pending(struct bitmap
*bitmap
, sector_t offset
)
1108 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1109 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1110 struct bitmap_page
*bp
= &bitmap
->bp
[page
];
1116 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1117 sector_t offset
, sector_t
*blocks
,
1121 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1125 void bitmap_daemon_work(struct mddev
*mddev
)
1127 struct bitmap
*bitmap
;
1129 unsigned long nextpage
;
1132 /* Use a mutex to guard daemon_work against
1135 mutex_lock(&mddev
->bitmap_info
.mutex
);
1136 bitmap
= mddev
->bitmap
;
1137 if (bitmap
== NULL
) {
1138 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1141 if (time_before(jiffies
, bitmap
->daemon_lastrun
1142 + mddev
->bitmap_info
.daemon_sleep
))
1145 bitmap
->daemon_lastrun
= jiffies
;
1146 if (bitmap
->allclean
) {
1147 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1150 bitmap
->allclean
= 1;
1152 /* Any file-page which is PENDING now needs to be written.
1153 * So set NEEDWRITE now, then after we make any last-minute changes
1156 spin_lock_irq(&bitmap
->lock
);
1157 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++)
1158 if (test_page_attr(bitmap
, j
,
1159 BITMAP_PAGE_PENDING
)) {
1160 set_page_attr(bitmap
, j
,
1161 BITMAP_PAGE_NEEDWRITE
);
1162 clear_page_attr(bitmap
, j
,
1163 BITMAP_PAGE_PENDING
);
1166 if (bitmap
->need_sync
&&
1167 mddev
->bitmap_info
.external
== 0) {
1168 /* Arrange for superblock update as well as
1171 bitmap
->need_sync
= 0;
1172 if (bitmap
->storage
.filemap
) {
1173 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
1174 sb
->events_cleared
=
1175 cpu_to_le64(bitmap
->events_cleared
);
1177 set_page_attr(bitmap
, 0,
1178 BITMAP_PAGE_NEEDWRITE
);
1181 /* Now look at the bitmap counters and if any are '2' or '1',
1182 * decrement and handle accordingly.
1185 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1186 bitmap_counter_t
*bmc
;
1187 sector_t block
= (sector_t
)j
<< bitmap
->chunkshift
;
1189 if (j
== nextpage
) {
1190 nextpage
+= PAGE_COUNTER_RATIO
;
1191 if (!bitmap
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
) {
1192 j
|= PAGE_COUNTER_MASK
;
1195 bitmap
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
= 0;
1197 bmc
= bitmap_get_counter(bitmap
,
1202 j
|= PAGE_COUNTER_MASK
;
1205 if (*bmc
== 1 && !bitmap
->need_sync
) {
1206 /* We can clear the bit */
1208 bitmap_count_page(bitmap
, block
, -1);
1209 bitmap_file_clear_bit(bitmap
, block
);
1210 } else if (*bmc
&& *bmc
<= 2) {
1212 bitmap_set_pending(bitmap
, block
);
1213 bitmap
->allclean
= 0;
1217 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1218 * DIRTY pages need to be written by bitmap_unplug so it can wait
1220 * If we find any DIRTY page we stop there and let bitmap_unplug
1221 * handle all the rest. This is important in the case where
1222 * the first blocking holds the superblock and it has been updated.
1223 * We mustn't write any other blocks before the superblock.
1225 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++) {
1227 if (test_page_attr(bitmap
, j
,
1229 /* bitmap_unplug will handle the rest */
1231 if (test_page_attr(bitmap
, j
,
1232 BITMAP_PAGE_NEEDWRITE
)) {
1233 clear_page_attr(bitmap
, j
,
1234 BITMAP_PAGE_NEEDWRITE
);
1235 spin_unlock_irq(&bitmap
->lock
);
1236 write_page(bitmap
, bitmap
->storage
.filemap
[j
], 0);
1237 spin_lock_irq(&bitmap
->lock
);
1238 if (!bitmap
->storage
.filemap
)
1242 spin_unlock_irq(&bitmap
->lock
);
1245 if (bitmap
->allclean
== 0)
1246 mddev
->thread
->timeout
=
1247 mddev
->bitmap_info
.daemon_sleep
;
1248 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1251 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1252 sector_t offset
, sector_t
*blocks
,
1254 __releases(bitmap
->lock
)
1255 __acquires(bitmap
->lock
)
1257 /* If 'create', we might release the lock and reclaim it.
1258 * The lock must have been taken with interrupts enabled.
1259 * If !create, we don't release the lock.
1261 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1262 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1263 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1267 err
= bitmap_checkpage(bitmap
, page
, create
);
1269 if (bitmap
->bp
[page
].hijacked
||
1270 bitmap
->bp
[page
].map
== NULL
)
1271 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1272 PAGE_COUNTER_SHIFT
- 1);
1274 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1275 *blocks
= csize
- (offset
& (csize
- 1));
1280 /* now locked ... */
1282 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1283 /* should we use the first or second counter field
1284 * of the hijacked pointer? */
1285 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1286 return &((bitmap_counter_t
*)
1287 &bitmap
->bp
[page
].map
)[hi
];
1288 } else /* page is allocated */
1289 return (bitmap_counter_t
*)
1290 &(bitmap
->bp
[page
].map
[pageoff
]);
1293 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1300 atomic_inc(&bitmap
->behind_writes
);
1301 bw
= atomic_read(&bitmap
->behind_writes
);
1302 if (bw
> bitmap
->behind_writes_used
)
1303 bitmap
->behind_writes_used
= bw
;
1305 pr_debug("inc write-behind count %d/%lu\n",
1306 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1311 bitmap_counter_t
*bmc
;
1313 spin_lock_irq(&bitmap
->lock
);
1314 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1316 spin_unlock_irq(&bitmap
->lock
);
1320 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1321 DEFINE_WAIT(__wait
);
1322 /* note that it is safe to do the prepare_to_wait
1323 * after the test as long as we do it before dropping
1326 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1327 TASK_UNINTERRUPTIBLE
);
1328 spin_unlock_irq(&bitmap
->lock
);
1330 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1336 bitmap_file_set_bit(bitmap
, offset
);
1337 bitmap_count_page(bitmap
, offset
, 1);
1345 spin_unlock_irq(&bitmap
->lock
);
1348 if (sectors
> blocks
)
1355 EXPORT_SYMBOL(bitmap_startwrite
);
1357 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1358 int success
, int behind
)
1363 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1364 wake_up(&bitmap
->behind_wait
);
1365 pr_debug("dec write-behind count %d/%lu\n",
1366 atomic_read(&bitmap
->behind_writes
),
1367 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1372 unsigned long flags
;
1373 bitmap_counter_t
*bmc
;
1375 spin_lock_irqsave(&bitmap
->lock
, flags
);
1376 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1378 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1382 if (success
&& !bitmap
->mddev
->degraded
&&
1383 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1384 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1385 bitmap
->need_sync
= 1;
1386 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1389 if (!success
&& !NEEDED(*bmc
))
1390 *bmc
|= NEEDED_MASK
;
1392 if (COUNTER(*bmc
) == COUNTER_MAX
)
1393 wake_up(&bitmap
->overflow_wait
);
1397 bitmap_set_pending(bitmap
, offset
);
1398 bitmap
->allclean
= 0;
1400 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1402 if (sectors
> blocks
)
1408 EXPORT_SYMBOL(bitmap_endwrite
);
1410 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1413 bitmap_counter_t
*bmc
;
1415 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1417 return 1; /* always resync if no bitmap */
1419 spin_lock_irq(&bitmap
->lock
);
1420 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1426 else if (NEEDED(*bmc
)) {
1428 if (!degraded
) { /* don't set/clear bits if degraded */
1429 *bmc
|= RESYNC_MASK
;
1430 *bmc
&= ~NEEDED_MASK
;
1434 spin_unlock_irq(&bitmap
->lock
);
1438 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1441 /* bitmap_start_sync must always report on multiples of whole
1442 * pages, otherwise resync (which is very PAGE_SIZE based) will
1444 * So call __bitmap_start_sync repeatedly (if needed) until
1445 * At least PAGE_SIZE>>9 blocks are covered.
1446 * Return the 'or' of the result.
1452 while (*blocks
< (PAGE_SIZE
>>9)) {
1453 rv
|= __bitmap_start_sync(bitmap
, offset
,
1454 &blocks1
, degraded
);
1460 EXPORT_SYMBOL(bitmap_start_sync
);
1462 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1464 bitmap_counter_t
*bmc
;
1465 unsigned long flags
;
1467 if (bitmap
== NULL
) {
1471 spin_lock_irqsave(&bitmap
->lock
, flags
);
1472 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1477 *bmc
&= ~RESYNC_MASK
;
1479 if (!NEEDED(*bmc
) && aborted
)
1480 *bmc
|= NEEDED_MASK
;
1483 bitmap_set_pending(bitmap
, offset
);
1484 bitmap
->allclean
= 0;
1489 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1491 EXPORT_SYMBOL(bitmap_end_sync
);
1493 void bitmap_close_sync(struct bitmap
*bitmap
)
1495 /* Sync has finished, and any bitmap chunks that weren't synced
1496 * properly have been aborted. It remains to us to clear the
1497 * RESYNC bit wherever it is still on
1499 sector_t sector
= 0;
1503 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1504 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1508 EXPORT_SYMBOL(bitmap_close_sync
);
1510 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
)
1518 bitmap
->last_end_sync
= jiffies
;
1521 if (time_before(jiffies
, (bitmap
->last_end_sync
1522 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1524 wait_event(bitmap
->mddev
->recovery_wait
,
1525 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1527 bitmap
->mddev
->curr_resync_completed
= sector
;
1528 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1529 sector
&= ~((1ULL << bitmap
->chunkshift
) - 1);
1531 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1532 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1535 bitmap
->last_end_sync
= jiffies
;
1536 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1538 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1540 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1542 /* For each chunk covered by any of these sectors, set the
1543 * counter to 2 and possibly set resync_needed. They should all
1544 * be 0 at this point
1548 bitmap_counter_t
*bmc
;
1549 spin_lock_irq(&bitmap
->lock
);
1550 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1552 spin_unlock_irq(&bitmap
->lock
);
1556 *bmc
= 2 | (needed
? NEEDED_MASK
: 0);
1557 bitmap_count_page(bitmap
, offset
, 1);
1558 bitmap_set_pending(bitmap
, offset
);
1559 bitmap
->allclean
= 0;
1561 spin_unlock_irq(&bitmap
->lock
);
1564 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1565 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1567 unsigned long chunk
;
1569 for (chunk
= s
; chunk
<= e
; chunk
++) {
1570 sector_t sec
= (sector_t
)chunk
<< bitmap
->chunkshift
;
1571 bitmap_set_memory_bits(bitmap
, sec
, 1);
1572 spin_lock_irq(&bitmap
->lock
);
1573 bitmap_file_set_bit(bitmap
, sec
);
1574 spin_unlock_irq(&bitmap
->lock
);
1575 if (sec
< bitmap
->mddev
->recovery_cp
)
1576 /* We are asserting that the array is dirty,
1577 * so move the recovery_cp address back so
1578 * that it is obvious that it is dirty
1580 bitmap
->mddev
->recovery_cp
= sec
;
1585 * flush out any pending updates
1587 void bitmap_flush(struct mddev
*mddev
)
1589 struct bitmap
*bitmap
= mddev
->bitmap
;
1592 if (!bitmap
) /* there was no bitmap */
1595 /* run the daemon_work three time to ensure everything is flushed
1598 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1599 bitmap
->daemon_lastrun
-= sleep
;
1600 bitmap_daemon_work(mddev
);
1601 bitmap
->daemon_lastrun
-= sleep
;
1602 bitmap_daemon_work(mddev
);
1603 bitmap
->daemon_lastrun
-= sleep
;
1604 bitmap_daemon_work(mddev
);
1605 bitmap_update_sb(bitmap
);
1609 * free memory that was allocated
1611 static void bitmap_free(struct bitmap
*bitmap
)
1613 unsigned long k
, pages
;
1614 struct bitmap_page
*bp
;
1616 if (!bitmap
) /* there was no bitmap */
1619 /* release the bitmap file and kill the daemon */
1620 bitmap_file_put(bitmap
);
1623 pages
= bitmap
->pages
;
1625 /* free all allocated memory */
1627 if (bp
) /* deallocate the page memory */
1628 for (k
= 0; k
< pages
; k
++)
1629 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1635 void bitmap_destroy(struct mddev
*mddev
)
1637 struct bitmap
*bitmap
= mddev
->bitmap
;
1639 if (!bitmap
) /* there was no bitmap */
1642 mutex_lock(&mddev
->bitmap_info
.mutex
);
1643 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1644 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1646 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1648 if (bitmap
->sysfs_can_clear
)
1649 sysfs_put(bitmap
->sysfs_can_clear
);
1651 bitmap_free(bitmap
);
1655 * initialize the bitmap structure
1656 * if this returns an error, bitmap_destroy must be called to do clean up
1658 int bitmap_create(struct mddev
*mddev
)
1660 struct bitmap
*bitmap
;
1661 sector_t blocks
= mddev
->resync_max_sectors
;
1662 unsigned long chunks
;
1663 unsigned long pages
;
1664 struct file
*file
= mddev
->bitmap_info
.file
;
1666 struct sysfs_dirent
*bm
= NULL
;
1668 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1670 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1672 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1676 spin_lock_init(&bitmap
->lock
);
1677 atomic_set(&bitmap
->pending_writes
, 0);
1678 init_waitqueue_head(&bitmap
->write_wait
);
1679 init_waitqueue_head(&bitmap
->overflow_wait
);
1680 init_waitqueue_head(&bitmap
->behind_wait
);
1682 bitmap
->mddev
= mddev
;
1685 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "bitmap");
1687 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, NULL
, "can_clear");
1690 bitmap
->sysfs_can_clear
= NULL
;
1692 bitmap
->storage
.file
= file
;
1695 /* As future accesses to this file will use bmap,
1696 * and bypass the page cache, we must sync the file
1701 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1702 if (!mddev
->bitmap_info
.external
) {
1704 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1705 * instructing us to create a new on-disk bitmap instance.
1707 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1708 err
= bitmap_new_disk_sb(bitmap
);
1710 err
= bitmap_read_sb(bitmap
);
1713 if (mddev
->bitmap_info
.chunksize
== 0 ||
1714 mddev
->bitmap_info
.daemon_sleep
== 0)
1715 /* chunksize and time_base need to be
1722 bitmap
->daemon_lastrun
= jiffies
;
1723 bitmap
->chunkshift
= (ffz(~mddev
->bitmap_info
.chunksize
)
1724 - BITMAP_BLOCK_SHIFT
);
1726 chunks
= (blocks
+ (1 << bitmap
->chunkshift
) - 1) >>
1728 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1732 bitmap
->chunks
= chunks
;
1733 bitmap
->pages
= pages
;
1734 bitmap
->missing_pages
= pages
;
1736 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1742 if (file
|| mddev
->bitmap_info
.offset
) {
1743 err
= bitmap_storage_alloc(&bitmap
->storage
, bitmap
->chunks
,
1744 !mddev
->bitmap_info
.external
);
1748 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1749 pages
, bmname(bitmap
));
1751 mddev
->bitmap
= bitmap
;
1754 return test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
) ? -EIO
: 0;
1757 bitmap_free(bitmap
);
1761 int bitmap_load(struct mddev
*mddev
)
1765 sector_t sector
= 0;
1766 struct bitmap
*bitmap
= mddev
->bitmap
;
1771 /* Clear out old bitmap info first: Either there is none, or we
1772 * are resuming after someone else has possibly changed things,
1773 * so we should forget old cached info.
1774 * All chunks should be clean, but some might need_sync.
1776 while (sector
< mddev
->resync_max_sectors
) {
1778 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1781 bitmap_close_sync(bitmap
);
1783 if (mddev
->degraded
== 0
1784 || bitmap
->events_cleared
== mddev
->events
)
1785 /* no need to keep dirty bits to optimise a
1786 * re-add of a missing device */
1787 start
= mddev
->recovery_cp
;
1789 mutex_lock(&mddev
->bitmap_info
.mutex
);
1790 err
= bitmap_init_from_disk(bitmap
, start
);
1791 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1795 clear_bit(BITMAP_STALE
, &bitmap
->flags
);
1797 /* Kick recovery in case any bits were set */
1798 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1800 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1801 md_wakeup_thread(mddev
->thread
);
1803 bitmap_update_sb(bitmap
);
1805 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1810 EXPORT_SYMBOL_GPL(bitmap_load
);
1812 void bitmap_status(struct seq_file
*seq
, struct bitmap
*bitmap
)
1814 unsigned long chunk_kb
;
1819 spin_lock_irq(&bitmap
->lock
);
1820 chunk_kb
= bitmap
->mddev
->bitmap_info
.chunksize
>> 10;
1821 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
1823 bitmap
->pages
- bitmap
->missing_pages
,
1825 (bitmap
->pages
- bitmap
->missing_pages
)
1826 << (PAGE_SHIFT
- 10),
1827 chunk_kb
? chunk_kb
: bitmap
->mddev
->bitmap_info
.chunksize
,
1828 chunk_kb
? "KB" : "B");
1829 if (bitmap
->storage
.file
) {
1830 seq_printf(seq
, ", file: ");
1831 seq_path(seq
, &bitmap
->storage
.file
->f_path
, " \t\n");
1834 seq_printf(seq
, "\n");
1835 spin_unlock_irq(&bitmap
->lock
);
1839 location_show(struct mddev
*mddev
, char *page
)
1842 if (mddev
->bitmap_info
.file
)
1843 len
= sprintf(page
, "file");
1844 else if (mddev
->bitmap_info
.offset
)
1845 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
1847 len
= sprintf(page
, "none");
1848 len
+= sprintf(page
+len
, "\n");
1853 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1857 if (!mddev
->pers
->quiesce
)
1859 if (mddev
->recovery
|| mddev
->sync_thread
)
1863 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
1864 mddev
->bitmap_info
.offset
) {
1865 /* bitmap already configured. Only option is to clear it */
1866 if (strncmp(buf
, "none", 4) != 0)
1869 mddev
->pers
->quiesce(mddev
, 1);
1870 bitmap_destroy(mddev
);
1871 mddev
->pers
->quiesce(mddev
, 0);
1873 mddev
->bitmap_info
.offset
= 0;
1874 if (mddev
->bitmap_info
.file
) {
1875 struct file
*f
= mddev
->bitmap_info
.file
;
1876 mddev
->bitmap_info
.file
= NULL
;
1877 restore_bitmap_write_access(f
);
1881 /* No bitmap, OK to set a location */
1883 if (strncmp(buf
, "none", 4) == 0)
1884 /* nothing to be done */;
1885 else if (strncmp(buf
, "file:", 5) == 0) {
1886 /* Not supported yet */
1891 rv
= strict_strtoll(buf
+1, 10, &offset
);
1893 rv
= strict_strtoll(buf
, 10, &offset
);
1898 if (mddev
->bitmap_info
.external
== 0 &&
1899 mddev
->major_version
== 0 &&
1900 offset
!= mddev
->bitmap_info
.default_offset
)
1902 mddev
->bitmap_info
.offset
= offset
;
1904 mddev
->pers
->quiesce(mddev
, 1);
1905 rv
= bitmap_create(mddev
);
1907 rv
= bitmap_load(mddev
);
1909 bitmap_destroy(mddev
);
1910 mddev
->bitmap_info
.offset
= 0;
1912 mddev
->pers
->quiesce(mddev
, 0);
1918 if (!mddev
->external
) {
1919 /* Ensure new bitmap info is stored in
1920 * metadata promptly.
1922 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1923 md_wakeup_thread(mddev
->thread
);
1928 static struct md_sysfs_entry bitmap_location
=
1929 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
1931 /* 'bitmap/space' is the space available at 'location' for the
1932 * bitmap. This allows the kernel to know when it is safe to
1933 * resize the bitmap to match a resized array.
1936 space_show(struct mddev
*mddev
, char *page
)
1938 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.space
);
1942 space_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1944 unsigned long sectors
;
1947 rv
= kstrtoul(buf
, 10, §ors
);
1954 if (mddev
->bitmap
&&
1955 sectors
< (mddev
->bitmap
->storage
.bytes
+ 511) >> 9)
1956 return -EFBIG
; /* Bitmap is too big for this small space */
1958 /* could make sure it isn't too big, but that isn't really
1959 * needed - user-space should be careful.
1961 mddev
->bitmap_info
.space
= sectors
;
1965 static struct md_sysfs_entry bitmap_space
=
1966 __ATTR(space
, S_IRUGO
|S_IWUSR
, space_show
, space_store
);
1969 timeout_show(struct mddev
*mddev
, char *page
)
1972 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
1973 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
1975 len
= sprintf(page
, "%lu", secs
);
1977 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
1978 len
+= sprintf(page
+len
, "\n");
1983 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
1985 /* timeout can be set at any time */
1986 unsigned long timeout
;
1987 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
1991 /* just to make sure we don't overflow... */
1992 if (timeout
>= LONG_MAX
/ HZ
)
1995 timeout
= timeout
* HZ
/ 10000;
1997 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
1998 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2001 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2002 if (mddev
->thread
) {
2003 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2004 * the bitmap is all clean and we don't need to
2005 * adjust the timeout right now
2007 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2008 mddev
->thread
->timeout
= timeout
;
2009 md_wakeup_thread(mddev
->thread
);
2015 static struct md_sysfs_entry bitmap_timeout
=
2016 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2019 backlog_show(struct mddev
*mddev
, char *page
)
2021 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2025 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2027 unsigned long backlog
;
2028 int rv
= strict_strtoul(buf
, 10, &backlog
);
2031 if (backlog
> COUNTER_MAX
)
2033 mddev
->bitmap_info
.max_write_behind
= backlog
;
2037 static struct md_sysfs_entry bitmap_backlog
=
2038 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2041 chunksize_show(struct mddev
*mddev
, char *page
)
2043 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2047 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2049 /* Can only be changed when no bitmap is active */
2051 unsigned long csize
;
2054 rv
= strict_strtoul(buf
, 10, &csize
);
2058 !is_power_of_2(csize
))
2060 mddev
->bitmap_info
.chunksize
= csize
;
2064 static struct md_sysfs_entry bitmap_chunksize
=
2065 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2067 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2069 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2070 ? "external" : "internal"));
2073 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2075 if (mddev
->bitmap
||
2076 mddev
->bitmap_info
.file
||
2077 mddev
->bitmap_info
.offset
)
2079 if (strncmp(buf
, "external", 8) == 0)
2080 mddev
->bitmap_info
.external
= 1;
2081 else if (strncmp(buf
, "internal", 8) == 0)
2082 mddev
->bitmap_info
.external
= 0;
2088 static struct md_sysfs_entry bitmap_metadata
=
2089 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2091 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2095 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2098 len
= sprintf(page
, "\n");
2102 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2104 if (mddev
->bitmap
== NULL
)
2106 if (strncmp(buf
, "false", 5) == 0)
2107 mddev
->bitmap
->need_sync
= 1;
2108 else if (strncmp(buf
, "true", 4) == 0) {
2109 if (mddev
->degraded
)
2111 mddev
->bitmap
->need_sync
= 0;
2117 static struct md_sysfs_entry bitmap_can_clear
=
2118 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2121 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2123 if (mddev
->bitmap
== NULL
)
2124 return sprintf(page
, "0\n");
2125 return sprintf(page
, "%lu\n",
2126 mddev
->bitmap
->behind_writes_used
);
2130 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2133 mddev
->bitmap
->behind_writes_used
= 0;
2137 static struct md_sysfs_entry max_backlog_used
=
2138 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2139 behind_writes_used_show
, behind_writes_used_reset
);
2141 static struct attribute
*md_bitmap_attrs
[] = {
2142 &bitmap_location
.attr
,
2144 &bitmap_timeout
.attr
,
2145 &bitmap_backlog
.attr
,
2146 &bitmap_chunksize
.attr
,
2147 &bitmap_metadata
.attr
,
2148 &bitmap_can_clear
.attr
,
2149 &max_backlog_used
.attr
,
2152 struct attribute_group md_bitmap_group
= {
2154 .attrs
= md_bitmap_attrs
,