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).
16 * wait if count gets too high, wake when it drops to half.
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/config.h>
24 #include <linux/timer.h>
25 #include <linux/sched.h>
26 #include <linux/list.h>
27 #include <linux/file.h>
28 #include <linux/mount.h>
29 #include <linux/buffer_head.h>
30 #include <linux/raid/md.h>
31 #include <linux/raid/bitmap.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 */
54 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
55 #define DPRINTK(x...) do { } while(0)
59 # define PRINTK(x...) printk(KERN_DEBUG x)
65 static inline char * bmname(struct bitmap
*bitmap
)
67 return bitmap
->mddev
? mdname(bitmap
->mddev
) : "mdX";
72 * just a placeholder - calls kmalloc for bitmap pages
74 static unsigned char *bitmap_alloc_page(struct bitmap
*bitmap
)
78 #ifdef INJECT_FAULTS_1
81 page
= kmalloc(PAGE_SIZE
, GFP_NOIO
);
84 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap
));
86 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
87 bmname(bitmap
), page
);
92 * for now just a placeholder -- just calls kfree for bitmap pages
94 static void bitmap_free_page(struct bitmap
*bitmap
, unsigned char *page
)
96 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap
), page
);
101 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
103 * 1) check to see if this page is allocated, if it's not then try to alloc
104 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
105 * page pointer directly as a counter
107 * if we find our page, we increment the page's refcount so that it stays
108 * allocated while we're using it
110 static int bitmap_checkpage(struct bitmap
*bitmap
, unsigned long page
, int create
)
112 unsigned char *mappage
;
114 if (page
>= bitmap
->pages
) {
116 "%s: invalid bitmap page request: %lu (> %lu)\n",
117 bmname(bitmap
), page
, bitmap
->pages
-1);
122 if (bitmap
->bp
[page
].hijacked
) /* it's hijacked, don't try to alloc */
125 if (bitmap
->bp
[page
].map
) /* page is already allocated, just return */
131 spin_unlock_irq(&bitmap
->lock
);
133 /* this page has not been allocated yet */
135 if ((mappage
= bitmap_alloc_page(bitmap
)) == NULL
) {
136 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
138 /* failed - set the hijacked flag so that we can use the
139 * pointer as a counter */
140 spin_lock_irq(&bitmap
->lock
);
141 if (!bitmap
->bp
[page
].map
)
142 bitmap
->bp
[page
].hijacked
= 1;
148 spin_lock_irq(&bitmap
->lock
);
150 /* recheck the page */
152 if (bitmap
->bp
[page
].map
|| bitmap
->bp
[page
].hijacked
) {
153 /* somebody beat us to getting the page */
154 bitmap_free_page(bitmap
, mappage
);
158 /* no page was in place and we have one, so install it */
160 memset(mappage
, 0, PAGE_SIZE
);
161 bitmap
->bp
[page
].map
= mappage
;
162 bitmap
->missing_pages
--;
168 /* if page is completely empty, put it back on the free list, or dealloc it */
169 /* if page was hijacked, unmark the flag so it might get alloced next time */
170 /* Note: lock should be held when calling this */
171 static void bitmap_checkfree(struct bitmap
*bitmap
, unsigned long page
)
175 if (bitmap
->bp
[page
].count
) /* page is still busy */
178 /* page is no longer in use, it can be released */
180 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
181 bitmap
->bp
[page
].hijacked
= 0;
182 bitmap
->bp
[page
].map
= NULL
;
186 /* normal case, free the page */
189 /* actually ... let's not. We will probably need the page again exactly when
190 * memory is tight and we are flusing to disk
194 ptr
= bitmap
->bp
[page
].map
;
195 bitmap
->bp
[page
].map
= NULL
;
196 bitmap
->missing_pages
++;
197 bitmap_free_page(bitmap
, ptr
);
204 * bitmap file handling - read and write the bitmap file and its superblock
207 /* copy the pathname of a file to a buffer */
208 char *file_path(struct file
*file
, char *buf
, int count
)
219 buf
= d_path(d
, v
, buf
, count
);
221 return IS_ERR(buf
) ? NULL
: buf
;
225 * basic page I/O operations
228 /* IO operations when bitmap is stored near all superblocks */
229 static struct page
*read_sb_page(mddev_t
*mddev
, long offset
, unsigned long index
)
231 /* choose a good rdev and read the page from there */
234 struct list_head
*tmp
;
235 struct page
*page
= alloc_page(GFP_KERNEL
);
239 return ERR_PTR(-ENOMEM
);
241 ITERATE_RDEV(mddev
, rdev
, tmp
) {
242 if (! test_bit(In_sync
, &rdev
->flags
)
243 || test_bit(Faulty
, &rdev
->flags
))
246 target
= (rdev
->sb_offset
<< 1) + offset
+ index
* (PAGE_SIZE
/512);
248 if (sync_page_io(rdev
->bdev
, target
, PAGE_SIZE
, page
, READ
)) {
250 attach_page_buffers(page
, NULL
); /* so that free_buffer will
255 return ERR_PTR(-EIO
);
259 static int write_sb_page(mddev_t
*mddev
, long offset
, struct page
*page
, int wait
)
262 struct list_head
*tmp
;
264 ITERATE_RDEV(mddev
, rdev
, tmp
)
265 if (test_bit(In_sync
, &rdev
->flags
)
266 && !test_bit(Faulty
, &rdev
->flags
))
267 md_super_write(mddev
, rdev
,
268 (rdev
->sb_offset
<<1) + offset
269 + page
->index
* (PAGE_SIZE
/512),
274 md_super_wait(mddev
);
279 * write out a page to a file
281 static int write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
283 struct buffer_head
*bh
;
285 if (bitmap
->file
== NULL
)
286 return write_sb_page(bitmap
->mddev
, bitmap
->offset
, page
, wait
);
288 bh
= page_buffers(page
);
290 while (bh
&& bh
->b_blocknr
) {
291 atomic_inc(&bitmap
->pending_writes
);
292 set_buffer_locked(bh
);
293 set_buffer_mapped(bh
);
294 submit_bh(WRITE
, bh
);
295 bh
= bh
->b_this_page
;
299 wait_event(bitmap
->write_wait
,
300 atomic_read(&bitmap
->pending_writes
)==0);
301 return (bitmap
->flags
& BITMAP_WRITE_ERROR
) ? -EIO
: 0;
306 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
308 struct bitmap
*bitmap
= bh
->b_private
;
312 spin_lock_irqsave(&bitmap
->lock
, flags
);
313 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
314 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
316 if (atomic_dec_and_test(&bitmap
->pending_writes
))
317 wake_up(&bitmap
->write_wait
);
320 /* copied from buffer.c */
322 __clear_page_buffers(struct page
*page
)
324 ClearPagePrivate(page
);
325 set_page_private(page
, 0);
326 page_cache_release(page
);
328 static void free_buffers(struct page
*page
)
330 struct buffer_head
*bh
= page_buffers(page
);
333 struct buffer_head
*next
= bh
->b_this_page
;
334 free_buffer_head(bh
);
337 __clear_page_buffers(page
);
341 /* read a page from a file.
342 * We both read the page, and attach buffers to the page to record the
343 * address of each block (using bmap). These addresses will be used
344 * to write the block later, completely bypassing the filesystem.
345 * This usage is similar to how swap files are handled, and allows us
346 * to write to a file with no concerns of memory allocation failing.
348 static struct page
*read_page(struct file
*file
, unsigned long index
,
349 struct bitmap
*bitmap
,
352 struct page
*page
= NULL
;
353 struct inode
*inode
= file
->f_dentry
->d_inode
;
354 struct buffer_head
*bh
;
357 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE
,
358 (unsigned long long)index
<< PAGE_SHIFT
);
360 page
= alloc_page(GFP_KERNEL
);
362 page
= ERR_PTR(-ENOMEM
);
366 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
369 page
= ERR_PTR(-ENOMEM
);
372 attach_page_buffers(page
, bh
);
373 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
378 bh
->b_blocknr
= bmap(inode
, block
);
379 if (bh
->b_blocknr
== 0) {
380 /* Cannot use this file! */
382 page
= ERR_PTR(-EINVAL
);
385 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
386 if (count
< (1<<inode
->i_blkbits
))
389 count
-= (1<<inode
->i_blkbits
);
391 bh
->b_end_io
= end_bitmap_write
;
392 bh
->b_private
= bitmap
;
393 atomic_inc(&bitmap
->pending_writes
);
394 set_buffer_locked(bh
);
395 set_buffer_mapped(bh
);
399 bh
= bh
->b_this_page
;
403 wait_event(bitmap
->write_wait
,
404 atomic_read(&bitmap
->pending_writes
)==0);
405 if (bitmap
->flags
& BITMAP_WRITE_ERROR
) {
407 page
= ERR_PTR(-EIO
);
411 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %Lu): %ld\n",
413 (unsigned long long)index
<< PAGE_SHIFT
,
419 * bitmap file superblock operations
422 /* update the event counter and sync the superblock to disk */
423 int bitmap_update_sb(struct bitmap
*bitmap
)
428 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
430 spin_lock_irqsave(&bitmap
->lock
, flags
);
431 if (!bitmap
->sb_page
) { /* no superblock */
432 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
435 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
436 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
437 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
438 if (!bitmap
->mddev
->degraded
)
439 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
440 kunmap_atomic(sb
, KM_USER0
);
441 return write_page(bitmap
, bitmap
->sb_page
, 1);
444 /* print out the bitmap file superblock */
445 void bitmap_print_sb(struct bitmap
*bitmap
)
449 if (!bitmap
|| !bitmap
->sb_page
)
451 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
452 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
453 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
454 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
455 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
456 *(__u32
*)(sb
->uuid
+0),
457 *(__u32
*)(sb
->uuid
+4),
458 *(__u32
*)(sb
->uuid
+8),
459 *(__u32
*)(sb
->uuid
+12));
460 printk(KERN_DEBUG
" events: %llu\n",
461 (unsigned long long) le64_to_cpu(sb
->events
));
462 printk(KERN_DEBUG
"events cleared: %llu\n",
463 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
464 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
465 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
466 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
467 printk(KERN_DEBUG
" sync size: %llu KB\n",
468 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
469 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
470 kunmap_atomic(sb
, KM_USER0
);
473 /* read the superblock from the bitmap file and initialize some bitmap fields */
474 static int bitmap_read_sb(struct bitmap
*bitmap
)
478 unsigned long chunksize
, daemon_sleep
, write_behind
;
479 unsigned long long events
;
482 /* page 0 is the superblock, read it... */
484 bitmap
->sb_page
= read_page(bitmap
->file
, 0, bitmap
, PAGE_SIZE
);
486 bitmap
->sb_page
= read_sb_page(bitmap
->mddev
, bitmap
->offset
, 0);
488 if (IS_ERR(bitmap
->sb_page
)) {
489 err
= PTR_ERR(bitmap
->sb_page
);
490 bitmap
->sb_page
= NULL
;
494 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
496 chunksize
= le32_to_cpu(sb
->chunksize
);
497 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
);
498 write_behind
= le32_to_cpu(sb
->write_behind
);
500 /* verify that the bitmap-specific fields are valid */
501 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
502 reason
= "bad magic";
503 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
504 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
505 reason
= "unrecognized superblock version";
506 else if (chunksize
< PAGE_SIZE
)
507 reason
= "bitmap chunksize too small";
508 else if ((1 << ffz(~chunksize
)) != chunksize
)
509 reason
= "bitmap chunksize not a power of 2";
510 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
/ HZ
)
511 reason
= "daemon sleep period out of range";
512 else if (write_behind
> COUNTER_MAX
)
513 reason
= "write-behind limit out of range (0 - 16383)";
515 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
516 bmname(bitmap
), reason
);
520 /* keep the array size field of the bitmap superblock up to date */
521 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
523 if (!bitmap
->mddev
->persistent
)
527 * if we have a persistent array superblock, compare the
528 * bitmap's UUID and event counter to the mddev's
530 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
531 printk(KERN_INFO
"%s: bitmap superblock UUID mismatch\n",
535 events
= le64_to_cpu(sb
->events
);
536 if (events
< bitmap
->mddev
->events
) {
537 printk(KERN_INFO
"%s: bitmap file is out of date (%llu < %llu) "
538 "-- forcing full recovery\n", bmname(bitmap
), events
,
539 (unsigned long long) bitmap
->mddev
->events
);
540 sb
->state
|= BITMAP_STALE
;
543 /* assign fields using values from superblock */
544 bitmap
->chunksize
= chunksize
;
545 bitmap
->daemon_sleep
= daemon_sleep
;
546 bitmap
->daemon_lastrun
= jiffies
;
547 bitmap
->max_write_behind
= write_behind
;
548 bitmap
->flags
|= sb
->state
;
549 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
550 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
551 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
552 if (sb
->state
& BITMAP_STALE
)
553 bitmap
->events_cleared
= bitmap
->mddev
->events
;
556 kunmap_atomic(sb
, KM_USER0
);
558 bitmap_print_sb(bitmap
);
562 enum bitmap_mask_op
{
567 /* record the state of the bitmap in the superblock */
568 static void bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
569 enum bitmap_mask_op op
)
574 spin_lock_irqsave(&bitmap
->lock
, flags
);
575 if (!bitmap
->sb_page
) { /* can't set the state */
576 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
579 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
580 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
582 case MASK_SET
: sb
->state
|= bits
;
584 case MASK_UNSET
: sb
->state
&= ~bits
;
588 kunmap_atomic(sb
, KM_USER0
);
592 * general bitmap file operations
595 /* calculate the index of the page that contains this bit */
596 static inline unsigned long file_page_index(unsigned long chunk
)
598 return CHUNK_BIT_OFFSET(chunk
) >> PAGE_BIT_SHIFT
;
601 /* calculate the (bit) offset of this bit within a page */
602 static inline unsigned long file_page_offset(unsigned long chunk
)
604 return CHUNK_BIT_OFFSET(chunk
) & (PAGE_BITS
- 1);
608 * return a pointer to the page in the filemap that contains the given bit
610 * this lookup is complicated by the fact that the bitmap sb might be exactly
611 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
614 static inline struct page
*filemap_get_page(struct bitmap
*bitmap
,
617 return bitmap
->filemap
[file_page_index(chunk
) - file_page_index(0)];
621 static void bitmap_file_unmap(struct bitmap
*bitmap
)
623 struct page
**map
, *sb_page
;
628 spin_lock_irqsave(&bitmap
->lock
, flags
);
629 map
= bitmap
->filemap
;
630 bitmap
->filemap
= NULL
;
631 attr
= bitmap
->filemap_attr
;
632 bitmap
->filemap_attr
= NULL
;
633 pages
= bitmap
->file_pages
;
634 bitmap
->file_pages
= 0;
635 sb_page
= bitmap
->sb_page
;
636 bitmap
->sb_page
= NULL
;
637 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
640 if (map
[pages
]->index
!= 0) /* 0 is sb_page, release it below */
641 free_buffers(map
[pages
]);
646 free_buffers(sb_page
);
649 static void bitmap_file_put(struct bitmap
*bitmap
)
654 spin_lock_irqsave(&bitmap
->lock
, flags
);
657 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
660 wait_event(bitmap
->write_wait
,
661 atomic_read(&bitmap
->pending_writes
)==0);
662 bitmap_file_unmap(bitmap
);
665 struct inode
*inode
= file
->f_dentry
->d_inode
;
666 invalidate_inode_pages(inode
->i_mapping
);
673 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
674 * then it is no longer reliable, so we stop using it and we mark the file
675 * as failed in the superblock
677 static void bitmap_file_kick(struct bitmap
*bitmap
)
679 char *path
, *ptr
= NULL
;
681 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_SET
);
682 bitmap_update_sb(bitmap
);
685 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
687 ptr
= file_path(bitmap
->file
, path
, PAGE_SIZE
);
689 printk(KERN_ALERT
"%s: kicking failed bitmap file %s from array!\n",
690 bmname(bitmap
), ptr
? ptr
: "");
695 bitmap_file_put(bitmap
);
700 enum bitmap_page_attr
{
701 BITMAP_PAGE_DIRTY
= 0, // there are set bits that need to be synced
702 BITMAP_PAGE_CLEAN
= 1, // there are bits that might need to be cleared
703 BITMAP_PAGE_NEEDWRITE
=2, // there are cleared bits that need to be synced
706 static inline void set_page_attr(struct bitmap
*bitmap
, struct page
*page
,
707 enum bitmap_page_attr attr
)
709 __set_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
712 static inline void clear_page_attr(struct bitmap
*bitmap
, struct page
*page
,
713 enum bitmap_page_attr attr
)
715 __clear_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
718 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, struct page
*page
,
719 enum bitmap_page_attr attr
)
721 return test_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
725 * bitmap_file_set_bit -- called before performing a write to the md device
726 * to set (and eventually sync) a particular bit in the bitmap file
728 * we set the bit immediately, then we record the page number so that
729 * when an unplug occurs, we can flush the dirty pages out to disk
731 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
736 unsigned long chunk
= block
>> CHUNK_BLOCK_SHIFT(bitmap
);
738 if (!bitmap
->filemap
) {
742 page
= filemap_get_page(bitmap
, chunk
);
743 bit
= file_page_offset(chunk
);
746 kaddr
= kmap_atomic(page
, KM_USER0
);
747 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
750 ext2_set_bit(bit
, kaddr
);
751 kunmap_atomic(kaddr
, KM_USER0
);
752 PRINTK("set file bit %lu page %lu\n", bit
, page
->index
);
754 /* record page number so it gets flushed to disk when unplug occurs */
755 set_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
759 /* this gets called when the md device is ready to unplug its underlying
760 * (slave) device queues -- before we let any writes go down, we need to
761 * sync the dirty pages of the bitmap file to disk */
762 int bitmap_unplug(struct bitmap
*bitmap
)
764 unsigned long i
, flags
;
765 int dirty
, need_write
;
773 /* look at each page to see if there are any set bits that need to be
774 * flushed out to disk */
775 for (i
= 0; i
< bitmap
->file_pages
; i
++) {
776 spin_lock_irqsave(&bitmap
->lock
, flags
);
777 if (!bitmap
->filemap
) {
778 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
781 page
= bitmap
->filemap
[i
];
782 dirty
= test_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
783 need_write
= test_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
784 clear_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
785 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
788 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
790 if (dirty
| need_write
)
791 err
= write_page(bitmap
, page
, 0);
793 if (wait
) { /* if any writes were performed, we need to wait on them */
795 wait_event(bitmap
->write_wait
,
796 atomic_read(&bitmap
->pending_writes
)==0);
798 md_super_wait(bitmap
->mddev
);
800 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
801 bitmap_file_kick(bitmap
);
805 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
806 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
807 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
808 * memory mapping of the bitmap file
810 * if there's no bitmap file, or if the bitmap file had been
811 * previously kicked from the array, we mark all the bits as
812 * 1's in order to cause a full resync.
814 * We ignore all bits for sectors that end earlier than 'start'.
815 * This is used when reading an out-of-date bitmap...
817 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
819 unsigned long i
, chunks
, index
, oldindex
, bit
;
820 struct page
*page
= NULL
, *oldpage
= NULL
;
821 unsigned long num_pages
, bit_cnt
= 0;
823 unsigned long bytes
, offset
;
828 chunks
= bitmap
->chunks
;
831 BUG_ON(!file
&& !bitmap
->offset
);
833 #ifdef INJECT_FAULTS_3
836 outofdate
= bitmap
->flags
& BITMAP_STALE
;
839 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
840 "recovery\n", bmname(bitmap
));
842 bytes
= (chunks
+ 7) / 8;
844 num_pages
= (bytes
+ sizeof(bitmap_super_t
) + PAGE_SIZE
- 1) / PAGE_SIZE
;
846 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
+ sizeof(bitmap_super_t
)) {
847 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
849 (unsigned long) i_size_read(file
->f_mapping
->host
),
850 bytes
+ sizeof(bitmap_super_t
));
856 bitmap
->filemap
= kmalloc(sizeof(struct page
*) * num_pages
, GFP_KERNEL
);
857 if (!bitmap
->filemap
)
860 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
861 bitmap
->filemap_attr
= kzalloc(
862 (((num_pages
*4/8)+sizeof(unsigned long)-1)
863 /sizeof(unsigned long))
864 *sizeof(unsigned long),
866 if (!bitmap
->filemap_attr
)
871 for (i
= 0; i
< chunks
; i
++) {
873 index
= file_page_index(i
);
874 bit
= file_page_offset(i
);
875 if (index
!= oldindex
) { /* this is a new page, read it in */
877 /* unmap the old page, we're done with it */
878 if (index
== num_pages
-1)
879 count
= bytes
- index
* PAGE_SIZE
;
884 * if we're here then the superblock page
885 * contains some bits (PAGE_SIZE != sizeof sb)
886 * we've already read it in, so just use it
888 page
= bitmap
->sb_page
;
889 offset
= sizeof(bitmap_super_t
);
891 page
= read_page(file
, index
, bitmap
, count
);
894 page
= read_sb_page(bitmap
->mddev
, bitmap
->offset
, index
);
897 if (IS_ERR(page
)) { /* read error */
907 * if bitmap is out of date, dirty the
908 * whole page and write it out
910 paddr
= kmap_atomic(page
, KM_USER0
);
911 memset(paddr
+ offset
, 0xff,
913 kunmap_atomic(paddr
, KM_USER0
);
914 ret
= write_page(bitmap
, page
, 1);
916 /* release, page not in filemap yet */
922 bitmap
->filemap
[bitmap
->file_pages
++] = page
;
924 paddr
= kmap_atomic(page
, KM_USER0
);
925 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
926 b
= test_bit(bit
, paddr
);
928 b
= ext2_test_bit(bit
, paddr
);
929 kunmap_atomic(paddr
, KM_USER0
);
931 /* if the disk bit is set, set the memory bit */
932 bitmap_set_memory_bits(bitmap
, i
<< CHUNK_BLOCK_SHIFT(bitmap
),
933 ((i
+1) << (CHUNK_BLOCK_SHIFT(bitmap
)) >= start
)
936 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
940 /* everything went OK */
942 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
944 if (bit_cnt
) { /* Kick recovery if any bits were set */
945 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
946 md_wakeup_thread(bitmap
->mddev
->thread
);
950 printk(KERN_INFO
"%s: bitmap initialized from disk: "
951 "read %lu/%lu pages, set %lu bits, status: %d\n",
952 bmname(bitmap
), bitmap
->file_pages
, num_pages
, bit_cnt
, ret
);
957 void bitmap_write_all(struct bitmap
*bitmap
)
959 /* We don't actually write all bitmap blocks here,
960 * just flag them as needing to be written
964 for (i
=0; i
< bitmap
->file_pages
; i
++)
965 set_page_attr(bitmap
, bitmap
->filemap
[i
],
966 BITMAP_PAGE_NEEDWRITE
);
970 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
972 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
973 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
974 bitmap
->bp
[page
].count
+= inc
;
976 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
977 (unsigned long long)offset, inc, bitmap->bp[page].count);
979 bitmap_checkfree(bitmap
, page
);
981 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
982 sector_t offset
, int *blocks
,
986 * bitmap daemon -- periodically wakes up to clean bits and flush pages
990 int bitmap_daemon_work(struct bitmap
*bitmap
)
994 struct page
*page
= NULL
, *lastpage
= NULL
;
1001 if (time_before(jiffies
, bitmap
->daemon_lastrun
+ bitmap
->daemon_sleep
*HZ
))
1003 bitmap
->daemon_lastrun
= jiffies
;
1005 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1006 bitmap_counter_t
*bmc
;
1007 spin_lock_irqsave(&bitmap
->lock
, flags
);
1008 if (!bitmap
->filemap
) {
1009 /* error or shutdown */
1010 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1014 page
= filemap_get_page(bitmap
, j
);
1016 if (page
!= lastpage
) {
1017 /* skip this page unless it's marked as needing cleaning */
1018 if (!test_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
)) {
1019 int need_write
= test_page_attr(bitmap
, page
,
1020 BITMAP_PAGE_NEEDWRITE
);
1022 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
1024 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1026 switch (write_page(bitmap
, page
, 0)) {
1030 bitmap_file_kick(bitmap
);
1036 /* grab the new page, sync and release the old */
1037 if (lastpage
!= NULL
) {
1038 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1039 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1040 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1041 err
= write_page(bitmap
, lastpage
, 0);
1043 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1044 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1047 bitmap_file_kick(bitmap
);
1049 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1052 printk("bitmap clean at page %lu\n", j);
1054 spin_lock_irqsave(&bitmap
->lock
, flags
);
1055 clear_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1057 bmc
= bitmap_get_counter(bitmap
, j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1061 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1064 *bmc
=1; /* maybe clear the bit next time */
1065 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1066 } else if (*bmc
== 1) {
1067 /* we can clear the bit */
1069 bitmap_count_page(bitmap
, j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1073 paddr
= kmap_atomic(page
, KM_USER0
);
1074 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1075 clear_bit(file_page_offset(j
), paddr
);
1077 ext2_clear_bit(file_page_offset(j
), paddr
);
1078 kunmap_atomic(paddr
, KM_USER0
);
1081 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1084 /* now sync the final page */
1085 if (lastpage
!= NULL
) {
1086 spin_lock_irqsave(&bitmap
->lock
, flags
);
1087 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1088 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1089 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1090 err
= write_page(bitmap
, lastpage
, 0);
1092 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1093 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1100 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1101 sector_t offset
, int *blocks
,
1104 /* If 'create', we might release the lock and reclaim it.
1105 * The lock must have been taken with interrupts enabled.
1106 * If !create, we don't release the lock.
1108 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1109 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1110 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1113 if (bitmap_checkpage(bitmap
, page
, create
) < 0) {
1114 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1115 *blocks
= csize
- (offset
& (csize
- 1));
1118 /* now locked ... */
1120 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1121 /* should we use the first or second counter field
1122 * of the hijacked pointer? */
1123 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1124 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
) +
1125 PAGE_COUNTER_SHIFT
- 1);
1126 *blocks
= csize
- (offset
& (csize
- 1));
1127 return &((bitmap_counter_t
*)
1128 &bitmap
->bp
[page
].map
)[hi
];
1129 } else { /* page is allocated */
1130 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1131 *blocks
= csize
- (offset
& (csize
- 1));
1132 return (bitmap_counter_t
*)
1133 &(bitmap
->bp
[page
].map
[pageoff
]);
1137 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1139 if (!bitmap
) return 0;
1142 atomic_inc(&bitmap
->behind_writes
);
1143 PRINTK(KERN_DEBUG
"inc write-behind count %d/%d\n",
1144 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1149 bitmap_counter_t
*bmc
;
1151 spin_lock_irq(&bitmap
->lock
);
1152 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1154 spin_unlock_irq(&bitmap
->lock
);
1160 bitmap_file_set_bit(bitmap
, offset
);
1161 bitmap_count_page(bitmap
,offset
, 1);
1162 blk_plug_device(bitmap
->mddev
->queue
);
1167 BUG_ON((*bmc
& COUNTER_MAX
) == COUNTER_MAX
);
1170 spin_unlock_irq(&bitmap
->lock
);
1173 if (sectors
> blocks
)
1180 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1181 int success
, int behind
)
1183 if (!bitmap
) return;
1185 atomic_dec(&bitmap
->behind_writes
);
1186 PRINTK(KERN_DEBUG
"dec write-behind count %d/%d\n",
1187 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1192 unsigned long flags
;
1193 bitmap_counter_t
*bmc
;
1195 spin_lock_irqsave(&bitmap
->lock
, flags
);
1196 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1198 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1202 if (!success
&& ! (*bmc
& NEEDED_MASK
))
1203 *bmc
|= NEEDED_MASK
;
1207 set_page_attr(bitmap
,
1208 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1211 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1213 if (sectors
> blocks
)
1219 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
,
1222 bitmap_counter_t
*bmc
;
1224 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1226 return 1; /* always resync if no bitmap */
1228 spin_lock_irq(&bitmap
->lock
);
1229 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1235 else if (NEEDED(*bmc
)) {
1237 if (!degraded
) { /* don't set/clear bits if degraded */
1238 *bmc
|= RESYNC_MASK
;
1239 *bmc
&= ~NEEDED_MASK
;
1243 spin_unlock_irq(&bitmap
->lock
);
1247 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
, int aborted
)
1249 bitmap_counter_t
*bmc
;
1250 unsigned long flags
;
1252 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1253 */ if (bitmap
== NULL
) {
1257 spin_lock_irqsave(&bitmap
->lock
, flags
);
1258 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1263 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1266 *bmc
&= ~RESYNC_MASK
;
1268 if (!NEEDED(*bmc
) && aborted
)
1269 *bmc
|= NEEDED_MASK
;
1272 set_page_attr(bitmap
,
1273 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1279 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1282 void bitmap_close_sync(struct bitmap
*bitmap
)
1284 /* Sync has finished, and any bitmap chunks that weren't synced
1285 * properly have been aborted. It remains to us to clear the
1286 * RESYNC bit wherever it is still on
1288 sector_t sector
= 0;
1290 if (!bitmap
) return;
1291 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1292 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1294 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1295 (unsigned long long)sector, blocks);
1296 */ sector
+= blocks
;
1300 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1302 /* For each chunk covered by any of these sectors, set the
1303 * counter to 1 and set resync_needed. They should all
1304 * be 0 at this point
1308 bitmap_counter_t
*bmc
;
1309 spin_lock_irq(&bitmap
->lock
);
1310 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1312 spin_unlock_irq(&bitmap
->lock
);
1317 *bmc
= 1 | (needed
?NEEDED_MASK
:0);
1318 bitmap_count_page(bitmap
, offset
, 1);
1319 page
= filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
));
1320 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1322 spin_unlock_irq(&bitmap
->lock
);
1327 * flush out any pending updates
1329 void bitmap_flush(mddev_t
*mddev
)
1331 struct bitmap
*bitmap
= mddev
->bitmap
;
1334 if (!bitmap
) /* there was no bitmap */
1337 /* run the daemon_work three time to ensure everything is flushed
1340 sleep
= bitmap
->daemon_sleep
;
1341 bitmap
->daemon_sleep
= 0;
1342 bitmap_daemon_work(bitmap
);
1343 bitmap_daemon_work(bitmap
);
1344 bitmap_daemon_work(bitmap
);
1345 bitmap
->daemon_sleep
= sleep
;
1346 bitmap_update_sb(bitmap
);
1350 * free memory that was allocated
1352 static void bitmap_free(struct bitmap
*bitmap
)
1354 unsigned long k
, pages
;
1355 struct bitmap_page
*bp
;
1357 if (!bitmap
) /* there was no bitmap */
1360 /* release the bitmap file and kill the daemon */
1361 bitmap_file_put(bitmap
);
1364 pages
= bitmap
->pages
;
1366 /* free all allocated memory */
1368 if (bp
) /* deallocate the page memory */
1369 for (k
= 0; k
< pages
; k
++)
1370 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1375 void bitmap_destroy(mddev_t
*mddev
)
1377 struct bitmap
*bitmap
= mddev
->bitmap
;
1379 if (!bitmap
) /* there was no bitmap */
1382 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1384 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1386 bitmap_free(bitmap
);
1390 * initialize the bitmap structure
1391 * if this returns an error, bitmap_destroy must be called to do clean up
1393 int bitmap_create(mddev_t
*mddev
)
1395 struct bitmap
*bitmap
;
1396 unsigned long blocks
= mddev
->resync_max_sectors
;
1397 unsigned long chunks
;
1398 unsigned long pages
;
1399 struct file
*file
= mddev
->bitmap_file
;
1403 BUG_ON(sizeof(bitmap_super_t
) != 256);
1405 if (!file
&& !mddev
->bitmap_offset
) /* bitmap disabled, nothing to do */
1408 BUG_ON(file
&& mddev
->bitmap_offset
);
1410 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1414 spin_lock_init(&bitmap
->lock
);
1415 atomic_set(&bitmap
->pending_writes
, 0);
1416 init_waitqueue_head(&bitmap
->write_wait
);
1418 bitmap
->mddev
= mddev
;
1420 bitmap
->file
= file
;
1421 bitmap
->offset
= mddev
->bitmap_offset
;
1424 do_sync_file_range(file
, 0, LLONG_MAX
,
1425 SYNC_FILE_RANGE_WAIT_BEFORE
|
1426 SYNC_FILE_RANGE_WRITE
|
1427 SYNC_FILE_RANGE_WAIT_AFTER
);
1429 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1430 err
= bitmap_read_sb(bitmap
);
1434 bitmap
->chunkshift
= find_first_bit(&bitmap
->chunksize
,
1435 sizeof(bitmap
->chunksize
));
1437 /* now that chunksize and chunkshift are set, we can use these macros */
1438 chunks
= (blocks
+ CHUNK_BLOCK_RATIO(bitmap
) - 1) /
1439 CHUNK_BLOCK_RATIO(bitmap
);
1440 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1444 bitmap
->chunks
= chunks
;
1445 bitmap
->pages
= pages
;
1446 bitmap
->missing_pages
= pages
;
1447 bitmap
->counter_bits
= COUNTER_BITS
;
1449 bitmap
->syncchunk
= ~0UL;
1451 #ifdef INJECT_FATAL_FAULT_1
1454 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1460 /* now that we have some pages available, initialize the in-memory
1461 * bitmap from the on-disk bitmap */
1463 if (mddev
->degraded
== 0
1464 || bitmap
->events_cleared
== mddev
->events
)
1465 /* no need to keep dirty bits to optimise a re-add of a missing device */
1466 start
= mddev
->recovery_cp
;
1467 err
= bitmap_init_from_disk(bitmap
, start
);
1472 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1473 pages
, bmname(bitmap
));
1475 mddev
->bitmap
= bitmap
;
1477 mddev
->thread
->timeout
= bitmap
->daemon_sleep
* HZ
;
1479 return bitmap_update_sb(bitmap
);
1482 bitmap_free(bitmap
);
1486 /* the bitmap API -- for raid personalities */
1487 EXPORT_SYMBOL(bitmap_startwrite
);
1488 EXPORT_SYMBOL(bitmap_endwrite
);
1489 EXPORT_SYMBOL(bitmap_start_sync
);
1490 EXPORT_SYMBOL(bitmap_end_sync
);
1491 EXPORT_SYMBOL(bitmap_unplug
);
1492 EXPORT_SYMBOL(bitmap_close_sync
);