e40d3e4b8d27b4b29eba89ceddc4e63947ddba3f
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";
70 #define WRITE_POOL_SIZE 256
73 * just a placeholder - calls kmalloc for bitmap pages
75 static unsigned char *bitmap_alloc_page(struct bitmap
*bitmap
)
79 #ifdef INJECT_FAULTS_1
82 page
= kmalloc(PAGE_SIZE
, GFP_NOIO
);
85 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap
));
87 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
88 bmname(bitmap
), page
);
93 * for now just a placeholder -- just calls kfree for bitmap pages
95 static void bitmap_free_page(struct bitmap
*bitmap
, unsigned char *page
)
97 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap
), page
);
102 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
104 * 1) check to see if this page is allocated, if it's not then try to alloc
105 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
106 * page pointer directly as a counter
108 * if we find our page, we increment the page's refcount so that it stays
109 * allocated while we're using it
111 static int bitmap_checkpage(struct bitmap
*bitmap
, unsigned long page
, int create
)
113 unsigned char *mappage
;
115 if (page
>= bitmap
->pages
) {
117 "%s: invalid bitmap page request: %lu (> %lu)\n",
118 bmname(bitmap
), page
, bitmap
->pages
-1);
123 if (bitmap
->bp
[page
].hijacked
) /* it's hijacked, don't try to alloc */
126 if (bitmap
->bp
[page
].map
) /* page is already allocated, just return */
132 spin_unlock_irq(&bitmap
->lock
);
134 /* this page has not been allocated yet */
136 if ((mappage
= bitmap_alloc_page(bitmap
)) == NULL
) {
137 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
139 /* failed - set the hijacked flag so that we can use the
140 * pointer as a counter */
141 spin_lock_irq(&bitmap
->lock
);
142 if (!bitmap
->bp
[page
].map
)
143 bitmap
->bp
[page
].hijacked
= 1;
149 spin_lock_irq(&bitmap
->lock
);
151 /* recheck the page */
153 if (bitmap
->bp
[page
].map
|| bitmap
->bp
[page
].hijacked
) {
154 /* somebody beat us to getting the page */
155 bitmap_free_page(bitmap
, mappage
);
159 /* no page was in place and we have one, so install it */
161 memset(mappage
, 0, PAGE_SIZE
);
162 bitmap
->bp
[page
].map
= mappage
;
163 bitmap
->missing_pages
--;
169 /* if page is completely empty, put it back on the free list, or dealloc it */
170 /* if page was hijacked, unmark the flag so it might get alloced next time */
171 /* Note: lock should be held when calling this */
172 static void bitmap_checkfree(struct bitmap
*bitmap
, unsigned long page
)
176 if (bitmap
->bp
[page
].count
) /* page is still busy */
179 /* page is no longer in use, it can be released */
181 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
182 bitmap
->bp
[page
].hijacked
= 0;
183 bitmap
->bp
[page
].map
= NULL
;
187 /* normal case, free the page */
190 /* actually ... let's not. We will probably need the page again exactly when
191 * memory is tight and we are flusing to disk
195 ptr
= bitmap
->bp
[page
].map
;
196 bitmap
->bp
[page
].map
= NULL
;
197 bitmap
->missing_pages
++;
198 bitmap_free_page(bitmap
, ptr
);
205 * bitmap file handling - read and write the bitmap file and its superblock
208 /* copy the pathname of a file to a buffer */
209 char *file_path(struct file
*file
, char *buf
, int count
)
220 buf
= d_path(d
, v
, buf
, count
);
222 return IS_ERR(buf
) ? NULL
: buf
;
226 * basic page I/O operations
229 /* IO operations when bitmap is stored near all superblocks */
230 static struct page
*read_sb_page(mddev_t
*mddev
, long offset
, unsigned long index
)
232 /* choose a good rdev and read the page from there */
235 struct list_head
*tmp
;
236 struct page
*page
= alloc_page(GFP_KERNEL
);
240 return ERR_PTR(-ENOMEM
);
242 ITERATE_RDEV(mddev
, rdev
, tmp
) {
243 if (! test_bit(In_sync
, &rdev
->flags
)
244 || test_bit(Faulty
, &rdev
->flags
))
247 target
= (rdev
->sb_offset
<< 1) + offset
+ index
* (PAGE_SIZE
/512);
249 if (sync_page_io(rdev
->bdev
, target
, PAGE_SIZE
, page
, READ
)) {
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
)
284 if (bitmap
->file
== NULL
)
285 return write_sb_page(bitmap
->mddev
, bitmap
->offset
, page
, wait
);
287 flush_dcache_page(page
); /* make sure visible to anyone reading the file */
292 if (TestSetPageLocked(page
))
293 return -EAGAIN
; /* already locked */
294 if (PageWriteback(page
)) {
300 ret
= page
->mapping
->a_ops
->prepare_write(bitmap
->file
, page
, 0, PAGE_SIZE
);
302 ret
= page
->mapping
->a_ops
->commit_write(bitmap
->file
, page
, 0,
309 set_page_dirty(page
); /* force it to be written out */
312 /* add to list to be waited for */
313 struct page_list
*item
= mempool_alloc(bitmap
->write_pool
, GFP_NOIO
);
315 spin_lock(&bitmap
->write_lock
);
316 list_add(&item
->list
, &bitmap
->complete_pages
);
317 spin_unlock(&bitmap
->write_lock
);
319 return write_one_page(page
, wait
);
322 /* read a page from a file, pinning it into cache, and return bytes_read */
323 static struct page
*read_page(struct file
*file
, unsigned long index
,
324 unsigned long *bytes_read
)
326 struct inode
*inode
= file
->f_mapping
->host
;
327 struct page
*page
= NULL
;
328 loff_t isize
= i_size_read(inode
);
329 unsigned long end_index
= isize
>> PAGE_SHIFT
;
331 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE
,
332 (unsigned long long)index
<< PAGE_SHIFT
);
334 page
= read_cache_page(inode
->i_mapping
, index
,
335 (filler_t
*)inode
->i_mapping
->a_ops
->readpage
, file
);
338 wait_on_page_locked(page
);
339 if (!PageUptodate(page
) || PageError(page
)) {
341 page
= ERR_PTR(-EIO
);
345 if (index
> end_index
) /* we have read beyond EOF */
347 else if (index
== end_index
) /* possible short read */
348 *bytes_read
= isize
& ~PAGE_MASK
;
350 *bytes_read
= PAGE_SIZE
; /* got a full page */
353 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %Lu): %ld\n",
355 (unsigned long long)index
<< PAGE_SHIFT
,
361 * bitmap file superblock operations
364 /* update the event counter and sync the superblock to disk */
365 int bitmap_update_sb(struct bitmap
*bitmap
)
370 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
372 spin_lock_irqsave(&bitmap
->lock
, flags
);
373 if (!bitmap
->sb_page
) { /* no superblock */
374 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
377 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
378 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
379 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
380 if (!bitmap
->mddev
->degraded
)
381 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
382 kunmap_atomic(sb
, KM_USER0
);
383 return write_page(bitmap
, bitmap
->sb_page
, 1);
386 /* print out the bitmap file superblock */
387 void bitmap_print_sb(struct bitmap
*bitmap
)
391 if (!bitmap
|| !bitmap
->sb_page
)
393 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
394 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
395 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
396 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
397 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
398 *(__u32
*)(sb
->uuid
+0),
399 *(__u32
*)(sb
->uuid
+4),
400 *(__u32
*)(sb
->uuid
+8),
401 *(__u32
*)(sb
->uuid
+12));
402 printk(KERN_DEBUG
" events: %llu\n",
403 (unsigned long long) le64_to_cpu(sb
->events
));
404 printk(KERN_DEBUG
"events cleared: %llu\n",
405 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
406 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
407 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
408 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
409 printk(KERN_DEBUG
" sync size: %llu KB\n",
410 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
411 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
412 kunmap_atomic(sb
, KM_USER0
);
415 /* read the superblock from the bitmap file and initialize some bitmap fields */
416 static int bitmap_read_sb(struct bitmap
*bitmap
)
420 unsigned long chunksize
, daemon_sleep
, write_behind
;
421 unsigned long bytes_read
;
422 unsigned long long events
;
425 /* page 0 is the superblock, read it... */
427 bitmap
->sb_page
= read_page(bitmap
->file
, 0, &bytes_read
);
429 bitmap
->sb_page
= read_sb_page(bitmap
->mddev
, bitmap
->offset
, 0);
430 bytes_read
= PAGE_SIZE
;
432 if (IS_ERR(bitmap
->sb_page
)) {
433 err
= PTR_ERR(bitmap
->sb_page
);
434 bitmap
->sb_page
= NULL
;
438 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
440 if (bytes_read
< sizeof(*sb
)) { /* short read */
441 printk(KERN_INFO
"%s: bitmap file superblock truncated\n",
447 chunksize
= le32_to_cpu(sb
->chunksize
);
448 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
);
449 write_behind
= le32_to_cpu(sb
->write_behind
);
451 /* verify that the bitmap-specific fields are valid */
452 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
453 reason
= "bad magic";
454 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
455 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_HI
)
456 reason
= "unrecognized superblock version";
457 else if (chunksize
< PAGE_SIZE
)
458 reason
= "bitmap chunksize too small";
459 else if ((1 << ffz(~chunksize
)) != chunksize
)
460 reason
= "bitmap chunksize not a power of 2";
461 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
/ HZ
)
462 reason
= "daemon sleep period out of range";
463 else if (write_behind
> COUNTER_MAX
)
464 reason
= "write-behind limit out of range (0 - 16383)";
466 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
467 bmname(bitmap
), reason
);
471 /* keep the array size field of the bitmap superblock up to date */
472 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
474 if (!bitmap
->mddev
->persistent
)
478 * if we have a persistent array superblock, compare the
479 * bitmap's UUID and event counter to the mddev's
481 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
482 printk(KERN_INFO
"%s: bitmap superblock UUID mismatch\n",
486 events
= le64_to_cpu(sb
->events
);
487 if (events
< bitmap
->mddev
->events
) {
488 printk(KERN_INFO
"%s: bitmap file is out of date (%llu < %llu) "
489 "-- forcing full recovery\n", bmname(bitmap
), events
,
490 (unsigned long long) bitmap
->mddev
->events
);
491 sb
->state
|= BITMAP_STALE
;
494 /* assign fields using values from superblock */
495 bitmap
->chunksize
= chunksize
;
496 bitmap
->daemon_sleep
= daemon_sleep
;
497 bitmap
->daemon_lastrun
= jiffies
;
498 bitmap
->max_write_behind
= write_behind
;
499 bitmap
->flags
|= sb
->state
;
500 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
501 bitmap
->flags
|= BITMAP_HOSTENDIAN
;
502 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
503 if (sb
->state
& BITMAP_STALE
)
504 bitmap
->events_cleared
= bitmap
->mddev
->events
;
507 kunmap_atomic(sb
, KM_USER0
);
509 bitmap_print_sb(bitmap
);
513 enum bitmap_mask_op
{
518 /* record the state of the bitmap in the superblock */
519 static void bitmap_mask_state(struct bitmap
*bitmap
, enum bitmap_state bits
,
520 enum bitmap_mask_op op
)
525 spin_lock_irqsave(&bitmap
->lock
, flags
);
526 if (!bitmap
->sb_page
) { /* can't set the state */
527 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
530 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
531 sb
= (bitmap_super_t
*)kmap_atomic(bitmap
->sb_page
, KM_USER0
);
533 case MASK_SET
: sb
->state
|= bits
;
535 case MASK_UNSET
: sb
->state
&= ~bits
;
539 kunmap_atomic(sb
, KM_USER0
);
543 * general bitmap file operations
546 /* calculate the index of the page that contains this bit */
547 static inline unsigned long file_page_index(unsigned long chunk
)
549 return CHUNK_BIT_OFFSET(chunk
) >> PAGE_BIT_SHIFT
;
552 /* calculate the (bit) offset of this bit within a page */
553 static inline unsigned long file_page_offset(unsigned long chunk
)
555 return CHUNK_BIT_OFFSET(chunk
) & (PAGE_BITS
- 1);
559 * return a pointer to the page in the filemap that contains the given bit
561 * this lookup is complicated by the fact that the bitmap sb might be exactly
562 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
565 static inline struct page
*filemap_get_page(struct bitmap
*bitmap
,
568 return bitmap
->filemap
[file_page_index(chunk
) - file_page_index(0)];
572 static void bitmap_file_unmap(struct bitmap
*bitmap
)
574 struct page
**map
, *sb_page
;
579 spin_lock_irqsave(&bitmap
->lock
, flags
);
580 map
= bitmap
->filemap
;
581 bitmap
->filemap
= NULL
;
582 attr
= bitmap
->filemap_attr
;
583 bitmap
->filemap_attr
= NULL
;
584 pages
= bitmap
->file_pages
;
585 bitmap
->file_pages
= 0;
586 sb_page
= bitmap
->sb_page
;
587 bitmap
->sb_page
= NULL
;
588 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
591 if (map
[pages
]->index
!= 0) /* 0 is sb_page, release it below */
592 put_page(map
[pages
]);
596 safe_put_page(sb_page
);
599 /* dequeue the next item in a page list -- don't call from irq context */
600 static struct page_list
*dequeue_page(struct bitmap
*bitmap
)
602 struct page_list
*item
= NULL
;
603 struct list_head
*head
= &bitmap
->complete_pages
;
605 spin_lock(&bitmap
->write_lock
);
606 if (list_empty(head
))
608 item
= list_entry(head
->prev
, struct page_list
, list
);
609 list_del(head
->prev
);
611 spin_unlock(&bitmap
->write_lock
);
615 static void drain_write_queues(struct bitmap
*bitmap
)
617 struct page_list
*item
;
619 while ((item
= dequeue_page(bitmap
))) {
620 /* don't bother to wait */
621 mempool_free(item
, bitmap
->write_pool
);
625 static void bitmap_file_put(struct bitmap
*bitmap
)
630 spin_lock_irqsave(&bitmap
->lock
, flags
);
633 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
635 drain_write_queues(bitmap
);
637 bitmap_file_unmap(bitmap
);
645 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
646 * then it is no longer reliable, so we stop using it and we mark the file
647 * as failed in the superblock
649 static void bitmap_file_kick(struct bitmap
*bitmap
)
651 char *path
, *ptr
= NULL
;
653 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_SET
);
654 bitmap_update_sb(bitmap
);
657 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
659 ptr
= file_path(bitmap
->file
, path
, PAGE_SIZE
);
661 printk(KERN_ALERT
"%s: kicking failed bitmap file %s from array!\n",
662 bmname(bitmap
), ptr
? ptr
: "");
667 bitmap_file_put(bitmap
);
672 enum bitmap_page_attr
{
673 BITMAP_PAGE_DIRTY
= 0, // there are set bits that need to be synced
674 BITMAP_PAGE_CLEAN
= 1, // there are bits that might need to be cleared
675 BITMAP_PAGE_NEEDWRITE
=2, // there are cleared bits that need to be synced
678 static inline void set_page_attr(struct bitmap
*bitmap
, struct page
*page
,
679 enum bitmap_page_attr attr
)
681 __set_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
684 static inline void clear_page_attr(struct bitmap
*bitmap
, struct page
*page
,
685 enum bitmap_page_attr attr
)
687 __clear_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
690 static inline unsigned long test_page_attr(struct bitmap
*bitmap
, struct page
*page
,
691 enum bitmap_page_attr attr
)
693 return test_bit((page
->index
<<2) + attr
, bitmap
->filemap_attr
);
697 * bitmap_file_set_bit -- called before performing a write to the md device
698 * to set (and eventually sync) a particular bit in the bitmap file
700 * we set the bit immediately, then we record the page number so that
701 * when an unplug occurs, we can flush the dirty pages out to disk
703 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
708 unsigned long chunk
= block
>> CHUNK_BLOCK_SHIFT(bitmap
);
710 if (!bitmap
->filemap
) {
714 page
= filemap_get_page(bitmap
, chunk
);
715 bit
= file_page_offset(chunk
);
718 kaddr
= kmap_atomic(page
, KM_USER0
);
719 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
722 ext2_set_bit(bit
, kaddr
);
723 kunmap_atomic(kaddr
, KM_USER0
);
724 PRINTK("set file bit %lu page %lu\n", bit
, page
->index
);
726 /* record page number so it gets flushed to disk when unplug occurs */
727 set_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
731 static void bitmap_writeback(struct bitmap
*bitmap
);
733 /* this gets called when the md device is ready to unplug its underlying
734 * (slave) device queues -- before we let any writes go down, we need to
735 * sync the dirty pages of the bitmap file to disk */
736 int bitmap_unplug(struct bitmap
*bitmap
)
738 unsigned long i
, flags
;
739 int dirty
, need_write
;
747 /* look at each page to see if there are any set bits that need to be
748 * flushed out to disk */
749 for (i
= 0; i
< bitmap
->file_pages
; i
++) {
750 spin_lock_irqsave(&bitmap
->lock
, flags
);
751 if (!bitmap
->filemap
) {
752 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
755 page
= bitmap
->filemap
[i
];
756 dirty
= test_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
757 need_write
= test_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
758 clear_page_attr(bitmap
, page
, BITMAP_PAGE_DIRTY
);
759 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
762 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
764 if (dirty
| need_write
) {
765 err
= write_page(bitmap
, page
, 0);
766 if (err
== -EAGAIN
) {
768 err
= write_page(bitmap
, page
, 1);
776 if (wait
) { /* if any writes were performed, we need to wait on them */
778 bitmap_writeback(bitmap
);
780 md_super_wait(bitmap
->mddev
);
785 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
786 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
787 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
788 * memory mapping of the bitmap file
790 * if there's no bitmap file, or if the bitmap file had been
791 * previously kicked from the array, we mark all the bits as
792 * 1's in order to cause a full resync.
794 * We ignore all bits for sectors that end earlier than 'start'.
795 * This is used when reading an out-of-date bitmap...
797 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
799 unsigned long i
, chunks
, index
, oldindex
, bit
;
800 struct page
*page
= NULL
, *oldpage
= NULL
;
801 unsigned long num_pages
, bit_cnt
= 0;
803 unsigned long bytes
, offset
, dummy
;
808 chunks
= bitmap
->chunks
;
811 BUG_ON(!file
&& !bitmap
->offset
);
813 #ifdef INJECT_FAULTS_3
816 outofdate
= bitmap
->flags
& BITMAP_STALE
;
819 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
820 "recovery\n", bmname(bitmap
));
822 bytes
= (chunks
+ 7) / 8;
824 num_pages
= (bytes
+ sizeof(bitmap_super_t
) + PAGE_SIZE
- 1) / PAGE_SIZE
;
826 if (file
&& i_size_read(file
->f_mapping
->host
) < bytes
+ sizeof(bitmap_super_t
)) {
827 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
829 (unsigned long) i_size_read(file
->f_mapping
->host
),
830 bytes
+ sizeof(bitmap_super_t
));
836 bitmap
->filemap
= kmalloc(sizeof(struct page
*) * num_pages
, GFP_KERNEL
);
837 if (!bitmap
->filemap
)
840 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
841 bitmap
->filemap_attr
= kzalloc(
842 (((num_pages
*4/8)+sizeof(unsigned long)-1)
843 /sizeof(unsigned long))
844 *sizeof(unsigned long),
846 if (!bitmap
->filemap_attr
)
851 for (i
= 0; i
< chunks
; i
++) {
853 index
= file_page_index(i
);
854 bit
= file_page_offset(i
);
855 if (index
!= oldindex
) { /* this is a new page, read it in */
856 /* unmap the old page, we're done with it */
859 * if we're here then the superblock page
860 * contains some bits (PAGE_SIZE != sizeof sb)
861 * we've already read it in, so just use it
863 page
= bitmap
->sb_page
;
864 offset
= sizeof(bitmap_super_t
);
866 page
= read_page(file
, index
, &dummy
);
869 page
= read_sb_page(bitmap
->mddev
, bitmap
->offset
, index
);
872 if (IS_ERR(page
)) { /* read error */
882 * if bitmap is out of date, dirty the
883 * whole page and write it out
885 paddr
= kmap_atomic(page
, KM_USER0
);
886 memset(paddr
+ offset
, 0xff,
888 kunmap_atomic(paddr
, KM_USER0
);
889 ret
= write_page(bitmap
, page
, 1);
891 /* release, page not in filemap yet */
897 bitmap
->filemap
[bitmap
->file_pages
++] = page
;
899 paddr
= kmap_atomic(page
, KM_USER0
);
900 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
901 b
= test_bit(bit
, paddr
);
903 b
= ext2_test_bit(bit
, paddr
);
904 kunmap_atomic(paddr
, KM_USER0
);
906 /* if the disk bit is set, set the memory bit */
907 bitmap_set_memory_bits(bitmap
, i
<< CHUNK_BLOCK_SHIFT(bitmap
),
908 ((i
+1) << (CHUNK_BLOCK_SHIFT(bitmap
)) >= start
)
911 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
915 /* everything went OK */
917 bitmap_mask_state(bitmap
, BITMAP_STALE
, MASK_UNSET
);
919 if (bit_cnt
) { /* Kick recovery if any bits were set */
920 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
921 md_wakeup_thread(bitmap
->mddev
->thread
);
925 printk(KERN_INFO
"%s: bitmap initialized from disk: "
926 "read %lu/%lu pages, set %lu bits, status: %d\n",
927 bmname(bitmap
), bitmap
->file_pages
, num_pages
, bit_cnt
, ret
);
932 void bitmap_write_all(struct bitmap
*bitmap
)
934 /* We don't actually write all bitmap blocks here,
935 * just flag them as needing to be written
939 for (i
=0; i
< bitmap
->file_pages
; i
++)
940 set_page_attr(bitmap
, bitmap
->filemap
[i
],
941 BITMAP_PAGE_NEEDWRITE
);
945 static void bitmap_count_page(struct bitmap
*bitmap
, sector_t offset
, int inc
)
947 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
948 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
949 bitmap
->bp
[page
].count
+= inc
;
951 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
952 (unsigned long long)offset, inc, bitmap->bp[page].count);
954 bitmap_checkfree(bitmap
, page
);
956 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
957 sector_t offset
, int *blocks
,
961 * bitmap daemon -- periodically wakes up to clean bits and flush pages
965 int bitmap_daemon_work(struct bitmap
*bitmap
)
969 struct page
*page
= NULL
, *lastpage
= NULL
;
976 if (time_before(jiffies
, bitmap
->daemon_lastrun
+ bitmap
->daemon_sleep
*HZ
))
978 bitmap
->daemon_lastrun
= jiffies
;
980 for (j
= 0; j
< bitmap
->chunks
; j
++) {
981 bitmap_counter_t
*bmc
;
982 spin_lock_irqsave(&bitmap
->lock
, flags
);
983 if (!bitmap
->filemap
) {
984 /* error or shutdown */
985 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
989 page
= filemap_get_page(bitmap
, j
);
991 if (page
!= lastpage
) {
992 /* skip this page unless it's marked as needing cleaning */
993 if (!test_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
)) {
994 int need_write
= test_page_attr(bitmap
, page
,
995 BITMAP_PAGE_NEEDWRITE
);
997 clear_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
999 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1001 switch (write_page(bitmap
, page
, 0)) {
1003 set_page_attr(bitmap
, page
, BITMAP_PAGE_NEEDWRITE
);
1008 bitmap_file_kick(bitmap
);
1014 /* grab the new page, sync and release the old */
1015 if (lastpage
!= NULL
) {
1016 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1017 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1018 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1019 err
= write_page(bitmap
, lastpage
, 0);
1020 if (err
== -EAGAIN
) {
1022 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1025 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1026 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1029 bitmap_file_kick(bitmap
);
1031 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1034 printk("bitmap clean at page %lu\n", j);
1036 spin_lock_irqsave(&bitmap
->lock
, flags
);
1037 clear_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1039 bmc
= bitmap_get_counter(bitmap
, j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1043 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1046 *bmc
=1; /* maybe clear the bit next time */
1047 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1048 } else if (*bmc
== 1) {
1049 /* we can clear the bit */
1051 bitmap_count_page(bitmap
, j
<< CHUNK_BLOCK_SHIFT(bitmap
),
1055 paddr
= kmap_atomic(page
, KM_USER0
);
1056 if (bitmap
->flags
& BITMAP_HOSTENDIAN
)
1057 clear_bit(file_page_offset(j
), paddr
);
1059 ext2_clear_bit(file_page_offset(j
), paddr
);
1060 kunmap_atomic(paddr
, KM_USER0
);
1063 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1066 /* now sync the final page */
1067 if (lastpage
!= NULL
) {
1068 spin_lock_irqsave(&bitmap
->lock
, flags
);
1069 if (test_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
)) {
1070 clear_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1071 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1072 err
= write_page(bitmap
, lastpage
, 0);
1073 if (err
== -EAGAIN
) {
1074 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1078 set_page_attr(bitmap
, lastpage
, BITMAP_PAGE_NEEDWRITE
);
1079 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1086 static void bitmap_writeback(struct bitmap
*bitmap
)
1089 struct page_list
*item
;
1092 PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap
));
1093 /* wait on bitmap page writebacks */
1094 while ((item
= dequeue_page(bitmap
))) {
1096 mempool_free(item
, bitmap
->write_pool
);
1097 PRINTK("wait on page writeback: %p\n", page
);
1098 wait_on_page_writeback(page
);
1099 PRINTK("finished page writeback: %p\n", page
);
1101 err
= PageError(page
);
1103 printk(KERN_WARNING
"%s: bitmap file writeback "
1104 "failed (page %lu): %d\n",
1105 bmname(bitmap
), page
->index
, err
);
1106 bitmap_file_kick(bitmap
);
1112 static bitmap_counter_t
*bitmap_get_counter(struct bitmap
*bitmap
,
1113 sector_t offset
, int *blocks
,
1116 /* If 'create', we might release the lock and reclaim it.
1117 * The lock must have been taken with interrupts enabled.
1118 * If !create, we don't release the lock.
1120 sector_t chunk
= offset
>> CHUNK_BLOCK_SHIFT(bitmap
);
1121 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1122 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1125 if (bitmap_checkpage(bitmap
, page
, create
) < 0) {
1126 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1127 *blocks
= csize
- (offset
& (csize
- 1));
1130 /* now locked ... */
1132 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1133 /* should we use the first or second counter field
1134 * of the hijacked pointer? */
1135 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1136 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
) +
1137 PAGE_COUNTER_SHIFT
- 1);
1138 *blocks
= csize
- (offset
& (csize
- 1));
1139 return &((bitmap_counter_t
*)
1140 &bitmap
->bp
[page
].map
)[hi
];
1141 } else { /* page is allocated */
1142 csize
= ((sector_t
)1) << (CHUNK_BLOCK_SHIFT(bitmap
));
1143 *blocks
= csize
- (offset
& (csize
- 1));
1144 return (bitmap_counter_t
*)
1145 &(bitmap
->bp
[page
].map
[pageoff
]);
1149 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1151 if (!bitmap
) return 0;
1154 atomic_inc(&bitmap
->behind_writes
);
1155 PRINTK(KERN_DEBUG
"inc write-behind count %d/%d\n",
1156 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1161 bitmap_counter_t
*bmc
;
1163 spin_lock_irq(&bitmap
->lock
);
1164 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 1);
1166 spin_unlock_irq(&bitmap
->lock
);
1172 bitmap_file_set_bit(bitmap
, offset
);
1173 bitmap_count_page(bitmap
,offset
, 1);
1174 blk_plug_device(bitmap
->mddev
->queue
);
1179 BUG_ON((*bmc
& COUNTER_MAX
) == COUNTER_MAX
);
1182 spin_unlock_irq(&bitmap
->lock
);
1185 if (sectors
> blocks
)
1192 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1193 int success
, int behind
)
1195 if (!bitmap
) return;
1197 atomic_dec(&bitmap
->behind_writes
);
1198 PRINTK(KERN_DEBUG
"dec write-behind count %d/%d\n",
1199 atomic_read(&bitmap
->behind_writes
), bitmap
->max_write_behind
);
1204 unsigned long flags
;
1205 bitmap_counter_t
*bmc
;
1207 spin_lock_irqsave(&bitmap
->lock
, flags
);
1208 bmc
= bitmap_get_counter(bitmap
, offset
, &blocks
, 0);
1210 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1214 if (!success
&& ! (*bmc
& NEEDED_MASK
))
1215 *bmc
|= NEEDED_MASK
;
1219 set_page_attr(bitmap
,
1220 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1223 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1225 if (sectors
> blocks
)
1231 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
,
1234 bitmap_counter_t
*bmc
;
1236 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1238 return 1; /* always resync if no bitmap */
1240 spin_lock_irq(&bitmap
->lock
);
1241 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1247 else if (NEEDED(*bmc
)) {
1249 if (!degraded
) { /* don't set/clear bits if degraded */
1250 *bmc
|= RESYNC_MASK
;
1251 *bmc
&= ~NEEDED_MASK
;
1255 spin_unlock_irq(&bitmap
->lock
);
1259 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
, int aborted
)
1261 bitmap_counter_t
*bmc
;
1262 unsigned long flags
;
1264 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1265 */ if (bitmap
== NULL
) {
1269 spin_lock_irqsave(&bitmap
->lock
, flags
);
1270 bmc
= bitmap_get_counter(bitmap
, offset
, blocks
, 0);
1275 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1278 *bmc
&= ~RESYNC_MASK
;
1280 if (!NEEDED(*bmc
) && aborted
)
1281 *bmc
|= NEEDED_MASK
;
1284 set_page_attr(bitmap
,
1285 filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
)),
1291 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
1294 void bitmap_close_sync(struct bitmap
*bitmap
)
1296 /* Sync has finished, and any bitmap chunks that weren't synced
1297 * properly have been aborted. It remains to us to clear the
1298 * RESYNC bit wherever it is still on
1300 sector_t sector
= 0;
1302 if (!bitmap
) return;
1303 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1304 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1306 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1307 (unsigned long long)sector, blocks);
1308 */ sector
+= blocks
;
1312 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1314 /* For each chunk covered by any of these sectors, set the
1315 * counter to 1 and set resync_needed. They should all
1316 * be 0 at this point
1320 bitmap_counter_t
*bmc
;
1321 spin_lock_irq(&bitmap
->lock
);
1322 bmc
= bitmap_get_counter(bitmap
, offset
, &secs
, 1);
1324 spin_unlock_irq(&bitmap
->lock
);
1329 *bmc
= 1 | (needed
?NEEDED_MASK
:0);
1330 bitmap_count_page(bitmap
, offset
, 1);
1331 page
= filemap_get_page(bitmap
, offset
>> CHUNK_BLOCK_SHIFT(bitmap
));
1332 set_page_attr(bitmap
, page
, BITMAP_PAGE_CLEAN
);
1334 spin_unlock_irq(&bitmap
->lock
);
1339 * flush out any pending updates
1341 void bitmap_flush(mddev_t
*mddev
)
1343 struct bitmap
*bitmap
= mddev
->bitmap
;
1346 if (!bitmap
) /* there was no bitmap */
1349 /* run the daemon_work three time to ensure everything is flushed
1352 sleep
= bitmap
->daemon_sleep
;
1353 bitmap
->daemon_sleep
= 0;
1354 bitmap_daemon_work(bitmap
);
1355 bitmap_daemon_work(bitmap
);
1356 bitmap_daemon_work(bitmap
);
1357 bitmap
->daemon_sleep
= sleep
;
1358 bitmap_update_sb(bitmap
);
1362 * free memory that was allocated
1364 static void bitmap_free(struct bitmap
*bitmap
)
1366 unsigned long k
, pages
;
1367 struct bitmap_page
*bp
;
1369 if (!bitmap
) /* there was no bitmap */
1372 /* release the bitmap file and kill the daemon */
1373 bitmap_file_put(bitmap
);
1376 pages
= bitmap
->pages
;
1378 /* free all allocated memory */
1380 mempool_destroy(bitmap
->write_pool
);
1382 if (bp
) /* deallocate the page memory */
1383 for (k
= 0; k
< pages
; k
++)
1384 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1389 void bitmap_destroy(mddev_t
*mddev
)
1391 struct bitmap
*bitmap
= mddev
->bitmap
;
1393 if (!bitmap
) /* there was no bitmap */
1396 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1398 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1400 bitmap_free(bitmap
);
1404 * initialize the bitmap structure
1405 * if this returns an error, bitmap_destroy must be called to do clean up
1407 int bitmap_create(mddev_t
*mddev
)
1409 struct bitmap
*bitmap
;
1410 unsigned long blocks
= mddev
->resync_max_sectors
;
1411 unsigned long chunks
;
1412 unsigned long pages
;
1413 struct file
*file
= mddev
->bitmap_file
;
1417 BUG_ON(sizeof(bitmap_super_t
) != 256);
1419 if (!file
&& !mddev
->bitmap_offset
) /* bitmap disabled, nothing to do */
1422 BUG_ON(file
&& mddev
->bitmap_offset
);
1424 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1428 spin_lock_init(&bitmap
->lock
);
1429 bitmap
->mddev
= mddev
;
1431 spin_lock_init(&bitmap
->write_lock
);
1432 INIT_LIST_HEAD(&bitmap
->complete_pages
);
1433 bitmap
->write_pool
= mempool_create_kmalloc_pool(WRITE_POOL_SIZE
,
1434 sizeof(struct page_list
));
1436 if (!bitmap
->write_pool
)
1439 bitmap
->file
= file
;
1440 bitmap
->offset
= mddev
->bitmap_offset
;
1441 if (file
) get_file(file
);
1442 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1443 err
= bitmap_read_sb(bitmap
);
1447 bitmap
->chunkshift
= find_first_bit(&bitmap
->chunksize
,
1448 sizeof(bitmap
->chunksize
));
1450 /* now that chunksize and chunkshift are set, we can use these macros */
1451 chunks
= (blocks
+ CHUNK_BLOCK_RATIO(bitmap
) - 1) /
1452 CHUNK_BLOCK_RATIO(bitmap
);
1453 pages
= (chunks
+ PAGE_COUNTER_RATIO
- 1) / PAGE_COUNTER_RATIO
;
1457 bitmap
->chunks
= chunks
;
1458 bitmap
->pages
= pages
;
1459 bitmap
->missing_pages
= pages
;
1460 bitmap
->counter_bits
= COUNTER_BITS
;
1462 bitmap
->syncchunk
= ~0UL;
1464 #ifdef INJECT_FATAL_FAULT_1
1467 bitmap
->bp
= kzalloc(pages
* sizeof(*bitmap
->bp
), GFP_KERNEL
);
1473 /* now that we have some pages available, initialize the in-memory
1474 * bitmap from the on-disk bitmap */
1476 if (mddev
->degraded
== 0
1477 || bitmap
->events_cleared
== mddev
->events
)
1478 /* no need to keep dirty bits to optimise a re-add of a missing device */
1479 start
= mddev
->recovery_cp
;
1480 err
= bitmap_init_from_disk(bitmap
, start
);
1485 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1486 pages
, bmname(bitmap
));
1488 mddev
->bitmap
= bitmap
;
1490 mddev
->thread
->timeout
= bitmap
->daemon_sleep
* HZ
;
1492 return bitmap_update_sb(bitmap
);
1495 bitmap_free(bitmap
);
1499 /* the bitmap API -- for raid personalities */
1500 EXPORT_SYMBOL(bitmap_startwrite
);
1501 EXPORT_SYMBOL(bitmap_endwrite
);
1502 EXPORT_SYMBOL(bitmap_start_sync
);
1503 EXPORT_SYMBOL(bitmap_end_sync
);
1504 EXPORT_SYMBOL(bitmap_unplug
);
1505 EXPORT_SYMBOL(bitmap_close_sync
);
1506 EXPORT_SYMBOL(bitmap_daemon_work
);
This page took 0.08307 seconds and 4 git commands to generate.