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/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/raid/md.h>
30 #include <linux/raid/bitmap.h>
37 /* these are for debugging purposes only! */
39 /* define one and only one of these */
40 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
41 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
42 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
43 #define INJECT_FAULTS_4 0 /* undef */
44 #define INJECT_FAULTS_5 0 /* undef */
45 #define INJECT_FAULTS_6 0
47 /* if these are defined, the driver will fail! debug only */
48 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
49 #define INJECT_FATAL_FAULT_2 0 /* undef */
50 #define INJECT_FATAL_FAULT_3 0 /* undef */
53 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
54 #define DPRINTK(x...) do { } while(0)
58 # define PRINTK(x...) printk(KERN_DEBUG x)
64 static inline char * bmname(struct bitmap
*bitmap
)
66 return bitmap
->mddev
? mdname(bitmap
->mddev
) : "mdX";
71 * just a placeholder - calls kmalloc for bitmap pages
73 static unsigned char *bitmap_alloc_page(struct bitmap
*bitmap
)
77 #ifdef INJECT_FAULTS_1
80 page
= kmalloc(PAGE_SIZE
, GFP_NOIO
);
83 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap
));
85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
86 bmname(bitmap
), page
);
91 * for now just a placeholder -- just calls kfree for bitmap pages
93 static void bitmap_free_page(struct bitmap
*bitmap
, unsigned char *page
)
95 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap
), page
);
100 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
102 * 1) check to see if this page is allocated, if it's not then try to alloc
103 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
104 * page pointer directly as a counter
106 * if we find our page, we increment the page's refcount so that it stays
107 * allocated while we're using it
109 static int bitmap_checkpage(struct bitmap
*bitmap
, unsigned long page
, int create
)
111 unsigned char *mappage
;
113 if (page
>= bitmap
->pages
) {
115 "%s: invalid bitmap page request: %lu (> %lu)\n",
116 bmname(bitmap
), page
, bitmap
->pages
-1);
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 spin_unlock_irq(&bitmap
->lock
);
132 /* this page has not been allocated yet */
134 if ((mappage
= bitmap_alloc_page(bitmap
)) == NULL
) {
135 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
137 /* failed - set the hijacked flag so that we can use the
138 * pointer as a counter */
139 spin_lock_irq(&bitmap
->lock
);
140 if (!bitmap
->bp
[page
].map
)
141 bitmap
->bp
[page
].hijacked
= 1;
147 spin_lock_irq(&bitmap
->lock
);
149 /* recheck the page */
151 if (bitmap
->bp
[page
].map
|| bitmap
->bp
[page
].hijacked
) {
152 /* somebody beat us to getting the page */
153 bitmap_free_page(bitmap
, mappage
);
157 /* no page was in place and we have one, so install it */
159 memset(mappage
, 0, PAGE_SIZE
);
160 bitmap
->bp
[page
].map
= mappage
;
161 bitmap
->missing_pages
--;
167 /* if page is completely empty, put it back on the free list, or dealloc it */
168 /* if page was hijacked, unmark the flag so it might get alloced next time */
169 /* Note: lock should be held when calling this */
170 static void bitmap_checkfree(struct bitmap
*bitmap
, unsigned long page
)
174 if (bitmap
->bp
[page
].count
) /* page is still busy */
177 /* page is no longer in use, it can be released */
179 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
180 bitmap
->bp
[page
].hijacked
= 0;
181 bitmap
->bp
[page
].map
= NULL
;
185 /* normal case, free the page */
188 /* actually ... let's not. We will probably need the page again exactly when
189 * memory is tight and we are flusing to disk
193 ptr
= bitmap
->bp
[page
].map
;
194 bitmap
->bp
[page
].map
= NULL
;
195 bitmap
->missing_pages
++;
196 bitmap_free_page(bitmap
, ptr
);
203 * bitmap file handling - read and write the bitmap file and its superblock
206 /* copy the pathname of a file to a buffer */
207 char *file_path(struct file
*file
, char *buf
, int count
)
218 buf
= d_path(d
, v
, buf
, count
);
220 return IS_ERR(buf
) ? NULL
: buf
;
224 * basic page I/O operations
227 /* IO operations when bitmap is stored near all superblocks */
228 static struct page
*read_sb_page(mddev_t
*mddev
, long offset
, unsigned long index
)
230 /* choose a good rdev and read the page from there */
233 struct list_head
*tmp
;
234 struct page
*page
= alloc_page(GFP_KERNEL
);
238 return ERR_PTR(-ENOMEM
);
240 ITERATE_RDEV(mddev
, rdev
, tmp
) {
241 if (! test_bit(In_sync
, &rdev
->flags
)
242 || test_bit(Faulty
, &rdev
->flags
))
245 target
= (rdev
->sb_offset
<< 1) + offset
+ index
* (PAGE_SIZE
/512);
247 if (sync_page_io(rdev
->bdev
, target
, PAGE_SIZE
, page
, READ
)) {
249 attach_page_buffers(page
, NULL
); /* so that free_buffer will
254 return ERR_PTR(-EIO
);
258 static int write_sb_page(mddev_t
*mddev
, long offset
, struct page
*page
, int wait
)
261 struct list_head
*tmp
;
263 ITERATE_RDEV(mddev
, rdev
, tmp
)
264 if (test_bit(In_sync
, &rdev
->flags
)
265 && !test_bit(Faulty
, &rdev
->flags
))
266 md_super_write(mddev
, rdev
,
267 (rdev
->sb_offset
<<1) + offset
268 + page
->index
* (PAGE_SIZE
/512),
273 md_super_wait(mddev
);
278 * write out a page to a file
280 static int write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
282 struct buffer_head
*bh
;
284 if (bitmap
->file
== NULL
)
285 return write_sb_page(bitmap
->mddev
, bitmap
->offset
, page
, wait
);
287 bh
= page_buffers(page
);
289 while (bh
&& bh
->b_blocknr
) {
290 atomic_inc(&bitmap
->pending_writes
);
291 set_buffer_locked(bh
);
292 set_buffer_mapped(bh
);
293 submit_bh(WRITE
, bh
);
294 bh
= bh
->b_this_page
;
298 wait_event(bitmap
->write_wait
,
299 atomic_read(&bitmap
->pending_writes
)==0);
300 return (bitmap
->flags
& BITMAP_WRITE_ERROR
) ? -EIO
: 0;
305 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
307 struct bitmap
*bitmap
= bh
->b_private
;
311 spin_lock_irqsave(&bitmap
->lock
, flags
);
312 bitmap
->flags
|= BITMAP_WRITE_ERROR
;
313 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
315 if (atomic_dec_and_test(&bitmap
->pending_writes
))
316 wake_up(&bitmap
->write_wait
);
319 /* copied from buffer.c */
321 __clear_page_buffers(struct page
*page
)
323 ClearPagePrivate(page
);
324 set_page_private(page
, 0);
325 page_cache_release(page
);
327 static void free_buffers(struct page
*page
)
329 struct buffer_head
*bh
= page_buffers(page
);
332 struct buffer_head
*next
= bh
->b_this_page
;
333 free_buffer_head(bh
);
336 __clear_page_buffers(page
);
340 /* read a page from a file.
341 * We both read the page, and attach buffers to the page to record the
342 * address of each block (using bmap). These addresses will be used
343 * to write the block later, completely bypassing the filesystem.
344 * This usage is similar to how swap files are handled, and allows us
345 * to write to a file with no concerns of memory allocation failing.
347 static struct page
*read_page(struct file
*file
, unsigned long index
,
348 struct bitmap
*bitmap
,
351 struct page
*page
= NULL
;
352 struct inode
*inode
= file
->f_dentry
->d_inode
;
353 struct buffer_head
*bh
;
356 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE
,
357 (unsigned long long)index
<< PAGE_SHIFT
);
359 page
= alloc_page(GFP_KERNEL
);
361 page
= ERR_PTR(-ENOMEM
);
365 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
368 page
= ERR_PTR(-ENOMEM
);
371 attach_page_buffers(page
, bh
);
372 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
377 bh
->b_blocknr
= bmap(inode
, block
);
378 if (bh
->b_blocknr
== 0) {
379 /* Cannot use this file! */
381 page
= ERR_PTR(-EINVAL
);
384 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
385 if (count
< (1<<inode
->i_blkbits
))
388 count
-= (1<<inode
->i_blkbits
);
390 bh
->b_end_io
= end_bitmap_write
;
391 bh
->b_private
= bitmap
;
392 atomic_inc(&bitmap
->pending_writes
);
393 set_buffer_locked(bh
);
394 set_buffer_mapped(bh
);
398 bh
= bh
->b_this_page
;
402 wait_event(bitmap
->write_wait
,
403 atomic_read(&bitmap
->pending_writes
)==0);
404 if (bitmap
->flags
& BITMAP_WRITE_ERROR
) {
406 page
= ERR_PTR(-EIO
);
410 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %Lu): %ld\n",
412 (unsigned long long)index
<< PAGE_SHIFT
,
418 * bitmap file superblock operations
421 /* update the event counter and sync the superblock to disk */
422 int bitmap_update_sb(struct bitmap
*bitmap
)
427 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
429 spin_lock_irqsave(&bitmap
->lock
, flags
);
430 if (!bitmap
->sb_page
) { /* no superblock */
431 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
434 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
435 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
436 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
437 if (!bitmap
->mddev
->degraded
)
438 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
439 kunmap_atomic(sb
, KM_USER0
);
440 return write_page(bitmap
, bitmap
->sb_page
, 1);
443 /* print out the bitmap file superblock */
444 void bitmap_print_sb(struct bitmap
*bitmap
)
448 if (!bitmap
|| !bitmap
->sb_page
)
450 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
451 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
452 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
453 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
454 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
455 *(__u32
*)(sb
->uuid
+0),
456 *(__u32
*)(sb
->uuid
+4),
457 *(__u32
*)(sb
->uuid
+8),
458 *(__u32
*)(sb
->uuid
+12));
459 printk(KERN_DEBUG
" events: %llu\n",
460 (unsigned long long) le64_to_cpu(sb
->events
));
461 printk(KERN_DEBUG
"events cleared: %llu\n",
462 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
463 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
464 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
465 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
466 printk(KERN_DEBUG
" sync size: %llu KB\n",
467 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
468 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
469 kunmap_atomic(sb
, KM_USER0
);
472 /* read the superblock from the bitmap file and initialize some bitmap fields */
473 static int bitmap_read_sb(struct bitmap
*bitmap
)
477 unsigned long chunksize
, daemon_sleep
, write_behind
;
478 unsigned long long events
;
481 /* page 0 is the superblock, read it... */
483 bitmap
->sb_page
= read_page(bitmap
->file
, 0, bitmap
, PAGE_SIZE
);
485 bitmap
->sb_page
= read_sb_page(bitmap
->mddev
, bitmap
->offset
, 0);
487 if (IS_ERR(bitmap
->sb_page
)) {
488 err
= PTR_ERR(bitmap
->sb_page
);
489 bitmap
->sb_page
= NULL
;
493 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
495 chunksize
= le32_to_cpu(sb
->chunksize
);
496 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
);
497 write_behind
= le32_to_cpu(sb
->write_behind
);
499 /* verify that the bitmap-specific fields are valid */
500 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
501 reason
= "bad magic";
502 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
503 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
504 reason
= "unrecognized superblock version";
505 else if (chunksize
< PAGE_SIZE
)
506 reason
= "bitmap chunksize too small";
507 else if ((1 << ffz(~chunksize
)) != chunksize
)
508 reason
= "bitmap chunksize not a power of 2";
509 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
/ HZ
)
510 reason
= "daemon sleep period out of range";
511 else if (write_behind
> COUNTER_MAX
)
512 reason
= "write-behind limit out of range (0 - 16383)";
514 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
515 bmname(bitmap
), reason
);
519 /* keep the array size field of the bitmap superblock up to date */
520 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
522 if (!bitmap
->mddev
->persistent
)
526 * if we have a persistent array superblock, compare the
527 * bitmap's UUID and event counter to the mddev's
529 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
530 printk(KERN_INFO
"%s: bitmap superblock UUID mismatch\n",
534 events
= le64_to_cpu(sb
->events
);
535 if (events
< bitmap
->mddev
->events
) {
536 printk(KERN_INFO
"%s: bitmap file is out of date (%llu < %llu) "
537 "-- forcing full recovery\n", bmname(bitmap
), events
,
538 (unsigned long long) bitmap
->mddev
->events
);
539 sb
->state
|= BITMAP_STALE
;
542 /* assign fields using values from superblock */
543 bitmap
->chunksize
= chunksize
;
544 bitmap
->daemon_sleep
= daemon_sleep
;
545 bitmap
->daemon_lastrun
= jiffies
;
546 bitmap
->max_write_behind
= write_behind
;
547 bitmap
->flags
|= sb
->state
;
548 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
549 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
550 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
551 if (sb
->state
& BITMAP_STALE
)
552 bitmap
->events_cleared
= bitmap
->mddev
->events
;
555 kunmap_atomic(sb
, KM_USER0
);
557 bitmap_print_sb(bitmap
);
561 enum bitmap_mask_op
{
566 /* record the state of the bitmap in the superblock */
567 static void bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
568 enum bitmap_mask_op op
)
573 spin_lock_irqsave(&bitmap
->lock
, flags
);
574 if (!bitmap
->sb_page
) { /* can't set the state */
575 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
578 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
579 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
581 case MASK_SET
: sb
->state
|= bits
;
583 case MASK_UNSET
: sb
->state
&= ~bits
;
587 kunmap_atomic(sb
, KM_USER0
);
591 * general bitmap file operations
594 /* calculate the index of the page that contains this bit */
595 static inline unsigned long file_page_index(unsigned long chunk
)
597 return CHUNK_BIT_OFFSET(chunk
) >> PAGE_BIT_SHIFT
;
600 /* calculate the (bit) offset of this bit within a page */
601 static inline unsigned long file_page_offset(unsigned long chunk
)
603 return CHUNK_BIT_OFFSET(chunk
) & (PAGE_BITS
- 1);
607 * return a pointer to the page in the filemap that contains the given bit
609 * this lookup is complicated by the fact that the bitmap sb might be exactly
610 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
613 static inline struct page
*filemap_get_page(struct bitmap
*bitmap
,
616 return bitmap
->filemap
[file_page_index(chunk
) - file_page_index(0)];
620 static void bitmap_file_unmap(struct bitmap
*bitmap
)
622 struct page
**map
, *sb_page
;
627 spin_lock_irqsave(&bitmap
->lock
, flags
);
628 map
= bitmap
->filemap
;
629 bitmap
->filemap
= NULL
;
630 attr
= bitmap
->filemap_attr
;
631 bitmap
->filemap_attr
= NULL
;
632 pages
= bitmap
->file_pages
;
633 bitmap
->file_pages
= 0;
634 sb_page
= bitmap
->sb_page
;
635 bitmap
->sb_page
= NULL
;
636 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
639 if (map
[pages
]->index
!= 0) /* 0 is sb_page, release it below */
640 free_buffers(map
[pages
]);
645 free_buffers(sb_page
);
648 static void bitmap_file_put(struct bitmap
*bitmap
)
653 spin_lock_irqsave(&bitmap
->lock
, flags
);
656 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
659 wait_event(bitmap
->write_wait
,
660 atomic_read(&bitmap
->pending_writes
)==0);
661 bitmap_file_unmap(bitmap
);
664 struct inode
*inode
= file
->f_dentry
->d_inode
;
665 invalidate_inode_pages(inode
->i_mapping
);
672 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
673 * then it is no longer reliable, so we stop using it and we mark the file
674 * as failed in the superblock
676 static void bitmap_file_kick(struct bitmap
*bitmap
)
678 char *path
, *ptr
= NULL
;
680 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_SET
);
681 bitmap_update_sb(bitmap
);
684 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
686 ptr
= file_path(bitmap
->file
, path
, PAGE_SIZE
);
688 printk(KERN_ALERT
"%s: kicking failed bitmap file %s from array!\n",
689 bmname(bitmap
), ptr
? ptr
: "");
694 bitmap_file_put(bitmap
);
699 enum bitmap_page_attr
{
700 BITMAP_PAGE_DIRTY
= 0, // there are set bits that need to be synced
701 BITMAP_PAGE_CLEAN
= 1, // there are bits that might need to be cleared
702 BITMAP_PAGE_NEEDWRITE
=2, // there are cleared bits that need to be synced
705 static inline void set_page_attr(struct bitmap
*bitmap
, struct page
*page
,
706 enum bitmap_page_attr attr
)
708 __set_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
711 static inline void clear_page_attr(struct bitmap
*bitmap
, struct page
*page
,
712 enum bitmap_page_attr attr
)
714 __clear_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
717 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, struct page
*page
,
718 enum bitmap_page_attr attr
)
720 return test_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
724 * bitmap_file_set_bit -- called before performing a write to the md device
725 * to set (and eventually sync) a particular bit in the bitmap file
727 * we set the bit immediately, then we record the page number so that
728 * when an unplug occurs, we can flush the dirty pages out to disk
730 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
735 unsigned long chunk
= block
>> CHUNK_BLOCK_SHIFT(bitmap
);
737 if (!bitmap
->filemap
) {
741 page
= filemap_get_page(bitmap
, chunk
);
742 bit
= file_page_offset(chunk
);
745 kaddr
= kmap_atomic(page
, KM_USER0
);
746 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
749 ext2_set_bit(bit
, kaddr
);
750 kunmap_atomic(kaddr
, KM_USER0
);
751 PRINTK("set file bit %lu page %lu\n", bit
, page
->index
);
753 /* record page number so it gets flushed to disk when unplug occurs */
754 set_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
758 /* this gets called when the md device is ready to unplug its underlying
759 * (slave) device queues -- before we let any writes go down, we need to
760 * sync the dirty pages of the bitmap file to disk */
761 int bitmap_unplug(struct bitmap
*bitmap
)
763 unsigned long i
, flags
;
764 int dirty
, need_write
;
772 /* look at each page to see if there are any set bits that need to be
773 * flushed out to disk */
774 for (i
= 0; i
< bitmap
->file_pages
; i
++) {
775 spin_lock_irqsave(&bitmap
->lock
, flags
);
776 if (!bitmap
->filemap
) {
777 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
780 page
= bitmap
->filemap
[i
];
781 dirty
= test_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
782 need_write
= test_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
783 clear_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
784 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
787 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
789 if (dirty
| need_write
)
790 err
= write_page(bitmap
, page
, 0);
792 if (wait
) { /* if any writes were performed, we need to wait on them */
794 wait_event(bitmap
->write_wait
,
795 atomic_read(&bitmap
->pending_writes
)==0);
797 md_super_wait(bitmap
->mddev
);
799 if (bitmap
->flags
& BITMAP_WRITE_ERROR
)
800 bitmap_file_kick(bitmap
);
804 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
805 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
806 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
807 * memory mapping of the bitmap file
809 * if there's no bitmap file, or if the bitmap file had been
810 * previously kicked from the array, we mark all the bits as
811 * 1's in order to cause a full resync.
813 * We ignore all bits for sectors that end earlier than 'start'.
814 * This is used when reading an out-of-date bitmap...
816 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
818 unsigned long i
, chunks
, index
, oldindex
, bit
;
819 struct page
*page
= NULL
, *oldpage
= NULL
;
820 unsigned long num_pages
, bit_cnt
= 0;
822 unsigned long bytes
, offset
;
827 chunks
= bitmap
->chunks
;
830 BUG_ON(!file
&& !bitmap
->offset
);
832 #ifdef INJECT_FAULTS_3
835 outofdate
= bitmap
->flags
& BITMAP_STALE
;
838 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
839 "recovery\n", bmname(bitmap
));
841 bytes
= (chunks
+ 7) / 8;
843 num_pages
= (bytes
+ sizeof(bitmap_super_t
) + PAGE_SIZE
- 1) / PAGE_SIZE
;
845 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
+ sizeof(bitmap_super_t
)) {
846 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
848 (unsigned long) i_size_read(file
->f_mapping
->host
),
849 bytes
+ sizeof(bitmap_super_t
));
855 bitmap
->filemap
= kmalloc(sizeof(struct page
*) * num_pages
, GFP_KERNEL
);
856 if (!bitmap
->filemap
)
859 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
860 bitmap
->filemap_attr
= kzalloc(
861 (((num_pages
*4/8)+sizeof(unsigned long)-1)
862 /sizeof(unsigned long))
863 *sizeof(unsigned long),
865 if (!bitmap
->filemap_attr
)
870 for (i
= 0; i
< chunks
; i
++) {
872 index
= file_page_index(i
);
873 bit
= file_page_offset(i
);
874 if (index
!= oldindex
) { /* this is a new page, read it in */
876 /* unmap the old page, we're done with it */
877 if (index
== num_pages
-1)
878 count
= bytes
- index
* PAGE_SIZE
;
883 * if we're here then the superblock page
884 * contains some bits (PAGE_SIZE != sizeof sb)
885 * we've already read it in, so just use it
887 page
= bitmap
->sb_page
;
888 offset
= sizeof(bitmap_super_t
);
890 page
= read_page(file
, index
, bitmap
, count
);
893 page
= read_sb_page(bitmap
->mddev
, bitmap
->offset
, index
);
896 if (IS_ERR(page
)) { /* read error */
906 * if bitmap is out of date, dirty the
907 * whole page and write it out
909 paddr
= kmap_atomic(page
, KM_USER0
);
910 memset(paddr
+ offset
, 0xff,
912 kunmap_atomic(paddr
, KM_USER0
);
913 ret
= write_page(bitmap
, page
, 1);
915 /* release, page not in filemap yet */
921 bitmap
->filemap
[bitmap
->file_pages
++] = page
;
923 paddr
= kmap_atomic(page
, KM_USER0
);
924 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
925 b
= test_bit(bit
, paddr
);
927 b
= ext2_test_bit(bit
, paddr
);
928 kunmap_atomic(paddr
, KM_USER0
);
930 /* if the disk bit is set, set the memory bit */
931 bitmap_set_memory_bits(bitmap
, i
<< CHUNK_BLOCK_SHIFT(bitmap
),
932 ((i
+1) << (CHUNK_BLOCK_SHIFT(bitmap
)) >= start
)
935 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
939 /* everything went OK */
941 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
943 if (bit_cnt
) { /* Kick recovery if any bits were set */
944 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
945 md_wakeup_thread(bitmap
->mddev
->thread
);
949 printk(KERN_INFO
"%s: bitmap initialized from disk: "
950 "read %lu/%lu pages, set %lu bits, status: %d\n",
951 bmname(bitmap
), bitmap
->file_pages
, num_pages
, bit_cnt
, ret
);
956 void bitmap_write_all(struct bitmap
*bitmap
)
958 /* We don't actually write all bitmap blocks here,
959 * just flag them as needing to be written
963 for (i
=0; i
< bitmap
->file_pages
; i
++)
964 set_page_attr(bitmap
, bitmap
->filemap
[i
],
965 BITMAP_PAGE_NEEDWRITE
);
969 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
971 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
972 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
973 bitmap
->bp
[page
].count
+= inc
;
975 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
976 (unsigned long long)offset, inc, bitmap->bp[page].count);
978 bitmap_checkfree(bitmap
, page
);
980 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
981 sector_t offset
, int *blocks
,
985 * bitmap daemon -- periodically wakes up to clean bits and flush pages
989 int bitmap_daemon_work(struct bitmap
*bitmap
)
993 struct page
*page
= NULL
, *lastpage
= NULL
;
1000 if (time_before(jiffies
, bitmap
->daemon_lastrun
+ bitmap
->daemon_sleep
*HZ
))
1002 bitmap
->daemon_lastrun
= jiffies
;
1004 for (j
= 0; j
< bitmap
->chunks
; j
++) {
1005 bitmap_counter_t
*bmc
;
1006 spin_lock_irqsave(&bitmap
->lock
, flags
);
1007 if (!bitmap
->filemap
) {
1008 /* error or shutdown */
1009 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1013 page
= filemap_get_page(bitmap
, j
);
1015 if (page
!= lastpage
) {
1016 /* skip this page unless it's marked as needing cleaning */
1017 if (!test_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
)) {
1018 int need_write
= test_page_attr(bitmap
, page
,
1019 BITMAP_PAGE_NEEDWRITE
);
1021 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
1023 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1025 switch (write_page(bitmap
, page
, 0)) {
1029 bitmap_file_kick(bitmap
);
1035 /* grab the new page, sync and release the old */
1036 if (lastpage
!= NULL
) {
1037 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1038 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1039 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1040 err
= write_page(bitmap
, lastpage
, 0);
1042 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1043 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1046 bitmap_file_kick(bitmap
);
1048 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1051 printk("bitmap clean at page %lu\n", j);
1053 spin_lock_irqsave(&bitmap
->lock
, flags
);
1054 clear_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1056 bmc
= bitmap_get_counter(bitmap
, j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1060 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1063 *bmc
=1; /* maybe clear the bit next time */
1064 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1065 } else if (*bmc
== 1) {
1066 /* we can clear the bit */
1068 bitmap_count_page(bitmap
, j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1072 paddr
= kmap_atomic(page
, KM_USER0
);
1073 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1074 clear_bit(file_page_offset(j
), paddr
);
1076 ext2_clear_bit(file_page_offset(j
), paddr
);
1077 kunmap_atomic(paddr
, KM_USER0
);
1080 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1083 /* now sync the final page */
1084 if (lastpage
!= NULL
) {
1085 spin_lock_irqsave(&bitmap
->lock
, flags
);
1086 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1087 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1088 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1089 err
= write_page(bitmap
, lastpage
, 0);
1091 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1092 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1099 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1100 sector_t offset
, int *blocks
,
1103 /* If 'create', we might release the lock and reclaim it.
1104 * The lock must have been taken with interrupts enabled.
1105 * If !create, we don't release the lock.
1107 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1108 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1109 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1112 if (bitmap_checkpage(bitmap
, page
, create
) < 0) {
1113 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1114 *blocks
= csize
- (offset
& (csize
- 1));
1117 /* now locked ... */
1119 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1120 /* should we use the first or second counter field
1121 * of the hijacked pointer? */
1122 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1123 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
) +
1124 PAGE_COUNTER_SHIFT
- 1);
1125 *blocks
= csize
- (offset
& (csize
- 1));
1126 return &((bitmap_counter_t
*)
1127 &bitmap
->bp
[page
].map
)[hi
];
1128 } else { /* page is allocated */
1129 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1130 *blocks
= csize
- (offset
& (csize
- 1));
1131 return (bitmap_counter_t
*)
1132 &(bitmap
->bp
[page
].map
[pageoff
]);
1136 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1138 if (!bitmap
) return 0;
1141 atomic_inc(&bitmap
->behind_writes
);
1142 PRINTK(KERN_DEBUG
"inc write-behind count %d/%d\n",
1143 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1148 bitmap_counter_t
*bmc
;
1150 spin_lock_irq(&bitmap
->lock
);
1151 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1153 spin_unlock_irq(&bitmap
->lock
);
1159 bitmap_file_set_bit(bitmap
, offset
);
1160 bitmap_count_page(bitmap
,offset
, 1);
1161 blk_plug_device(bitmap
->mddev
->queue
);
1166 BUG_ON((*bmc
& COUNTER_MAX
) == COUNTER_MAX
);
1169 spin_unlock_irq(&bitmap
->lock
);
1172 if (sectors
> blocks
)
1179 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1180 int success
, int behind
)
1182 if (!bitmap
) return;
1184 atomic_dec(&bitmap
->behind_writes
);
1185 PRINTK(KERN_DEBUG
"dec write-behind count %d/%d\n",
1186 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1191 unsigned long flags
;
1192 bitmap_counter_t
*bmc
;
1194 spin_lock_irqsave(&bitmap
->lock
, flags
);
1195 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1197 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1201 if (!success
&& ! (*bmc
& NEEDED_MASK
))
1202 *bmc
|= NEEDED_MASK
;
1206 set_page_attr(bitmap
,
1207 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1210 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1212 if (sectors
> blocks
)
1218 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
,
1221 bitmap_counter_t
*bmc
;
1223 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1225 return 1; /* always resync if no bitmap */
1227 spin_lock_irq(&bitmap
->lock
);
1228 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1234 else if (NEEDED(*bmc
)) {
1236 if (!degraded
) { /* don't set/clear bits if degraded */
1237 *bmc
|= RESYNC_MASK
;
1238 *bmc
&= ~NEEDED_MASK
;
1242 spin_unlock_irq(&bitmap
->lock
);
1246 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
, int aborted
)
1248 bitmap_counter_t
*bmc
;
1249 unsigned long flags
;
1251 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1252 */ if (bitmap
== NULL
) {
1256 spin_lock_irqsave(&bitmap
->lock
, flags
);
1257 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1262 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1265 *bmc
&= ~RESYNC_MASK
;
1267 if (!NEEDED(*bmc
) && aborted
)
1268 *bmc
|= NEEDED_MASK
;
1271 set_page_attr(bitmap
,
1272 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1278 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1281 void bitmap_close_sync(struct bitmap
*bitmap
)
1283 /* Sync has finished, and any bitmap chunks that weren't synced
1284 * properly have been aborted. It remains to us to clear the
1285 * RESYNC bit wherever it is still on
1287 sector_t sector
= 0;
1289 if (!bitmap
) return;
1290 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1291 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1293 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1294 (unsigned long long)sector, blocks);
1295 */ sector
+= blocks
;
1299 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1301 /* For each chunk covered by any of these sectors, set the
1302 * counter to 1 and set resync_needed. They should all
1303 * be 0 at this point
1307 bitmap_counter_t
*bmc
;
1308 spin_lock_irq(&bitmap
->lock
);
1309 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1311 spin_unlock_irq(&bitmap
->lock
);
1316 *bmc
= 1 | (needed
?NEEDED_MASK
:0);
1317 bitmap_count_page(bitmap
, offset
, 1);
1318 page
= filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
));
1319 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1321 spin_unlock_irq(&bitmap
->lock
);
1326 * flush out any pending updates
1328 void bitmap_flush(mddev_t
*mddev
)
1330 struct bitmap
*bitmap
= mddev
->bitmap
;
1333 if (!bitmap
) /* there was no bitmap */
1336 /* run the daemon_work three time to ensure everything is flushed
1339 sleep
= bitmap
->daemon_sleep
;
1340 bitmap
->daemon_sleep
= 0;
1341 bitmap_daemon_work(bitmap
);
1342 bitmap_daemon_work(bitmap
);
1343 bitmap_daemon_work(bitmap
);
1344 bitmap
->daemon_sleep
= sleep
;
1345 bitmap_update_sb(bitmap
);
1349 * free memory that was allocated
1351 static void bitmap_free(struct bitmap
*bitmap
)
1353 unsigned long k
, pages
;
1354 struct bitmap_page
*bp
;
1356 if (!bitmap
) /* there was no bitmap */
1359 /* release the bitmap file and kill the daemon */
1360 bitmap_file_put(bitmap
);
1363 pages
= bitmap
->pages
;
1365 /* free all allocated memory */
1367 if (bp
) /* deallocate the page memory */
1368 for (k
= 0; k
< pages
; k
++)
1369 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1374 void bitmap_destroy(mddev_t
*mddev
)
1376 struct bitmap
*bitmap
= mddev
->bitmap
;
1378 if (!bitmap
) /* there was no bitmap */
1381 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1383 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1385 bitmap_free(bitmap
);
1389 * initialize the bitmap structure
1390 * if this returns an error, bitmap_destroy must be called to do clean up
1392 int bitmap_create(mddev_t
*mddev
)
1394 struct bitmap
*bitmap
;
1395 unsigned long blocks
= mddev
->resync_max_sectors
;
1396 unsigned long chunks
;
1397 unsigned long pages
;
1398 struct file
*file
= mddev
->bitmap_file
;
1402 BUG_ON(sizeof(bitmap_super_t
) != 256);
1404 if (!file
&& !mddev
->bitmap_offset
) /* bitmap disabled, nothing to do */
1407 BUG_ON(file
&& mddev
->bitmap_offset
);
1409 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1413 spin_lock_init(&bitmap
->lock
);
1414 atomic_set(&bitmap
->pending_writes
, 0);
1415 init_waitqueue_head(&bitmap
->write_wait
);
1417 bitmap
->mddev
= mddev
;
1419 bitmap
->file
= file
;
1420 bitmap
->offset
= mddev
->bitmap_offset
;
1423 do_sync_file_range(file
, 0, LLONG_MAX
,
1424 SYNC_FILE_RANGE_WAIT_BEFORE
|
1425 SYNC_FILE_RANGE_WRITE
|
1426 SYNC_FILE_RANGE_WAIT_AFTER
);
1428 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1429 err
= bitmap_read_sb(bitmap
);
1433 bitmap
->chunkshift
= find_first_bit(&bitmap
->chunksize
,
1434 sizeof(bitmap
->chunksize
));
1436 /* now that chunksize and chunkshift are set, we can use these macros */
1437 chunks
= (blocks
+ CHUNK_BLOCK_RATIO(bitmap
) - 1) /
1438 CHUNK_BLOCK_RATIO(bitmap
);
1439 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1443 bitmap
->chunks
= chunks
;
1444 bitmap
->pages
= pages
;
1445 bitmap
->missing_pages
= pages
;
1446 bitmap
->counter_bits
= COUNTER_BITS
;
1448 bitmap
->syncchunk
= ~0UL;
1450 #ifdef INJECT_FATAL_FAULT_1
1453 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1459 /* now that we have some pages available, initialize the in-memory
1460 * bitmap from the on-disk bitmap */
1462 if (mddev
->degraded
== 0
1463 || bitmap
->events_cleared
== mddev
->events
)
1464 /* no need to keep dirty bits to optimise a re-add of a missing device */
1465 start
= mddev
->recovery_cp
;
1466 err
= bitmap_init_from_disk(bitmap
, start
);
1471 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1472 pages
, bmname(bitmap
));
1474 mddev
->bitmap
= bitmap
;
1476 mddev
->thread
->timeout
= bitmap
->daemon_sleep
* HZ
;
1478 return bitmap_update_sb(bitmap
);
1481 bitmap_free(bitmap
);
1485 /* the bitmap API -- for raid personalities */
1486 EXPORT_SYMBOL(bitmap_startwrite
);
1487 EXPORT_SYMBOL(bitmap_endwrite
);
1488 EXPORT_SYMBOL(bitmap_start_sync
);
1489 EXPORT_SYMBOL(bitmap_end_sync
);
1490 EXPORT_SYMBOL(bitmap_unplug
);
1491 EXPORT_SYMBOL(bitmap_close_sync
);