2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
15 * flush after percent set rather than just time based. (maybe both).
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
33 static inline char *bmname(struct bitmap
*bitmap
)
35 return bitmap
->mddev
? mdname(bitmap
->mddev
) : "mdX";
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
48 static int bitmap_checkpage(struct bitmap_counts
*bitmap
,
49 unsigned long page
, int create
)
50 __releases(bitmap
->lock
)
51 __acquires(bitmap
->lock
)
53 unsigned char *mappage
;
55 if (page
>= bitmap
->pages
) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
63 if (bitmap
->bp
[page
].hijacked
) /* it's hijacked, don't try to alloc */
66 if (bitmap
->bp
[page
].map
) /* page is already allocated, just return */
72 /* this page has not been allocated yet */
74 spin_unlock_irq(&bitmap
->lock
);
75 /* It is possible that this is being called inside a
76 * prepare_to_wait/finish_wait loop from raid5c:make_request().
77 * In general it is not permitted to sleep in that context as it
78 * can cause the loop to spin freely.
79 * That doesn't apply here as we can only reach this point
81 * When this function completes, either bp[page].map or
82 * bp[page].hijacked. In either case, this function will
83 * abort before getting to this point again. So there is
84 * no risk of a free-spin, and so it is safe to assert
85 * that sleeping here is allowed.
87 sched_annotate_sleep();
88 mappage
= kzalloc(PAGE_SIZE
, GFP_NOIO
);
89 spin_lock_irq(&bitmap
->lock
);
91 if (mappage
== NULL
) {
92 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
93 /* failed - set the hijacked flag so that we can use the
94 * pointer as a counter */
95 if (!bitmap
->bp
[page
].map
)
96 bitmap
->bp
[page
].hijacked
= 1;
97 } else if (bitmap
->bp
[page
].map
||
98 bitmap
->bp
[page
].hijacked
) {
99 /* somebody beat us to getting the page */
104 /* no page was in place and we have one, so install it */
106 bitmap
->bp
[page
].map
= mappage
;
107 bitmap
->missing_pages
--;
112 /* if page is completely empty, put it back on the free list, or dealloc it */
113 /* if page was hijacked, unmark the flag so it might get alloced next time */
114 /* Note: lock should be held when calling this */
115 static void bitmap_checkfree(struct bitmap_counts
*bitmap
, unsigned long page
)
119 if (bitmap
->bp
[page
].count
) /* page is still busy */
122 /* page is no longer in use, it can be released */
124 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
125 bitmap
->bp
[page
].hijacked
= 0;
126 bitmap
->bp
[page
].map
= NULL
;
128 /* normal case, free the page */
129 ptr
= bitmap
->bp
[page
].map
;
130 bitmap
->bp
[page
].map
= NULL
;
131 bitmap
->missing_pages
++;
137 * bitmap file handling - read and write the bitmap file and its superblock
141 * basic page I/O operations
144 /* IO operations when bitmap is stored near all superblocks */
145 static int read_sb_page(struct mddev
*mddev
, loff_t offset
,
147 unsigned long index
, int size
)
149 /* choose a good rdev and read the page from there */
151 struct md_rdev
*rdev
;
154 rdev_for_each(rdev
, mddev
) {
155 if (! test_bit(In_sync
, &rdev
->flags
)
156 || test_bit(Faulty
, &rdev
->flags
))
159 target
= offset
+ index
* (PAGE_SIZE
/512);
161 if (sync_page_io(rdev
, target
,
162 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
171 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
173 /* Iterate the disks of an mddev, using rcu to protect access to the
174 * linked list, and raising the refcount of devices we return to ensure
175 * they don't disappear while in use.
176 * As devices are only added or removed when raid_disk is < 0 and
177 * nr_pending is 0 and In_sync is clear, the entries we return will
178 * still be in the same position on the list when we re-enter
179 * list_for_each_entry_continue_rcu.
181 * Note that if entered with 'rdev == NULL' to start at the
182 * beginning, we temporarily assign 'rdev' to an address which
183 * isn't really an rdev, but which can be used by
184 * list_for_each_entry_continue_rcu() to find the first entry.
188 /* start at the beginning */
189 rdev
= list_entry(&mddev
->disks
, struct md_rdev
, same_set
);
191 /* release the previous rdev and start from there. */
192 rdev_dec_pending(rdev
, mddev
);
194 list_for_each_entry_continue_rcu(rdev
, &mddev
->disks
, same_set
) {
195 if (rdev
->raid_disk
>= 0 &&
196 !test_bit(Faulty
, &rdev
->flags
)) {
197 /* this is a usable devices */
198 atomic_inc(&rdev
->nr_pending
);
207 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
209 struct md_rdev
*rdev
= NULL
;
210 struct block_device
*bdev
;
211 struct mddev
*mddev
= bitmap
->mddev
;
212 struct bitmap_storage
*store
= &bitmap
->storage
;
215 if (mddev_is_clustered(bitmap
->mddev
))
216 node_offset
= bitmap
->cluster_slot
* store
->file_pages
;
218 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
219 int size
= PAGE_SIZE
;
220 loff_t offset
= mddev
->bitmap_info
.offset
;
222 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
224 if (page
->index
== store
->file_pages
-1) {
225 int last_page_size
= store
->bytes
& (PAGE_SIZE
-1);
226 if (last_page_size
== 0)
227 last_page_size
= PAGE_SIZE
;
228 size
= roundup(last_page_size
,
229 bdev_logical_block_size(bdev
));
231 /* Just make sure we aren't corrupting data or
234 if (mddev
->external
) {
235 /* Bitmap could be anywhere. */
236 if (rdev
->sb_start
+ offset
+ (page
->index
240 rdev
->sb_start
+ offset
241 < (rdev
->data_offset
+ mddev
->dev_sectors
244 } else if (offset
< 0) {
245 /* DATA BITMAP METADATA */
247 + (long)(page
->index
* (PAGE_SIZE
/512))
249 /* bitmap runs in to metadata */
251 if (rdev
->data_offset
+ mddev
->dev_sectors
252 > rdev
->sb_start
+ offset
)
253 /* data runs in to bitmap */
255 } else if (rdev
->sb_start
< rdev
->data_offset
) {
256 /* METADATA BITMAP DATA */
259 + page
->index
*(PAGE_SIZE
/512) + size
/512
261 /* bitmap runs in to data */
264 /* DATA METADATA BITMAP - no problems */
266 md_super_write(mddev
, rdev
,
267 rdev
->sb_start
+ offset
268 + page
->index
* (PAGE_SIZE
/512),
274 md_super_wait(mddev
);
281 static void bitmap_file_kick(struct bitmap
*bitmap
);
283 * write out a page to a file
285 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
287 struct buffer_head
*bh
;
289 if (bitmap
->storage
.file
== NULL
) {
290 switch (write_sb_page(bitmap
, page
, wait
)) {
292 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
296 bh
= page_buffers(page
);
298 while (bh
&& bh
->b_blocknr
) {
299 atomic_inc(&bitmap
->pending_writes
);
300 set_buffer_locked(bh
);
301 set_buffer_mapped(bh
);
302 submit_bh(WRITE
| REQ_SYNC
, bh
);
303 bh
= bh
->b_this_page
;
307 wait_event(bitmap
->write_wait
,
308 atomic_read(&bitmap
->pending_writes
)==0);
310 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
311 bitmap_file_kick(bitmap
);
314 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
316 struct bitmap
*bitmap
= bh
->b_private
;
319 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
320 if (atomic_dec_and_test(&bitmap
->pending_writes
))
321 wake_up(&bitmap
->write_wait
);
324 /* copied from buffer.c */
326 __clear_page_buffers(struct page
*page
)
328 ClearPagePrivate(page
);
329 set_page_private(page
, 0);
330 page_cache_release(page
);
332 static void free_buffers(struct page
*page
)
334 struct buffer_head
*bh
;
336 if (!PagePrivate(page
))
339 bh
= page_buffers(page
);
341 struct buffer_head
*next
= bh
->b_this_page
;
342 free_buffer_head(bh
);
345 __clear_page_buffers(page
);
349 /* read a page from a file.
350 * We both read the page, and attach buffers to the page to record the
351 * address of each block (using bmap). These addresses will be used
352 * to write the block later, completely bypassing the filesystem.
353 * This usage is similar to how swap files are handled, and allows us
354 * to write to a file with no concerns of memory allocation failing.
356 static int read_page(struct file
*file
, unsigned long index
,
357 struct bitmap
*bitmap
,
362 struct inode
*inode
= file_inode(file
);
363 struct buffer_head
*bh
;
366 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
367 (unsigned long long)index
<< PAGE_SHIFT
);
369 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
374 attach_page_buffers(page
, bh
);
375 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
380 bh
->b_blocknr
= bmap(inode
, block
);
381 if (bh
->b_blocknr
== 0) {
382 /* Cannot use this file! */
386 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
387 if (count
< (1<<inode
->i_blkbits
))
390 count
-= (1<<inode
->i_blkbits
);
392 bh
->b_end_io
= end_bitmap_write
;
393 bh
->b_private
= bitmap
;
394 atomic_inc(&bitmap
->pending_writes
);
395 set_buffer_locked(bh
);
396 set_buffer_mapped(bh
);
400 bh
= bh
->b_this_page
;
404 wait_event(bitmap
->write_wait
,
405 atomic_read(&bitmap
->pending_writes
)==0);
406 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
410 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %d\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 void bitmap_update_sb(struct bitmap
*bitmap
)
426 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
428 if (bitmap
->mddev
->bitmap_info
.external
)
430 if (!bitmap
->storage
.sb_page
) /* no superblock */
432 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
433 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
434 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
435 /* rocking back to read-only */
436 bitmap
->events_cleared
= bitmap
->mddev
->events
;
437 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
438 sb
->state
= cpu_to_le32(bitmap
->flags
);
439 /* Just in case these have been changed via sysfs: */
440 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
441 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
442 /* This might have been changed by a reshape */
443 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
444 sb
->chunksize
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.chunksize
);
445 sb
->nodes
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.nodes
);
446 sb
->sectors_reserved
= cpu_to_le32(bitmap
->mddev
->
449 write_page(bitmap
, bitmap
->storage
.sb_page
, 1);
452 /* print out the bitmap file superblock */
453 void bitmap_print_sb(struct bitmap
*bitmap
)
457 if (!bitmap
|| !bitmap
->storage
.sb_page
)
459 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
460 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
461 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
462 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
463 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
464 *(__u32
*)(sb
->uuid
+0),
465 *(__u32
*)(sb
->uuid
+4),
466 *(__u32
*)(sb
->uuid
+8),
467 *(__u32
*)(sb
->uuid
+12));
468 printk(KERN_DEBUG
" events: %llu\n",
469 (unsigned long long) le64_to_cpu(sb
->events
));
470 printk(KERN_DEBUG
"events cleared: %llu\n",
471 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
472 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
473 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
474 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
475 printk(KERN_DEBUG
" sync size: %llu KB\n",
476 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
477 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
485 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
486 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
487 * This function verifies 'bitmap_info' and populates the on-disk bitmap
488 * structure, which is to be written to disk.
490 * Returns: 0 on success, -Exxx on error
492 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
495 unsigned long chunksize
, daemon_sleep
, write_behind
;
497 bitmap
->storage
.sb_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
498 if (bitmap
->storage
.sb_page
== NULL
)
500 bitmap
->storage
.sb_page
->index
= 0;
502 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
504 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
505 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
507 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
509 if (!is_power_of_2(chunksize
)) {
511 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
514 sb
->chunksize
= cpu_to_le32(chunksize
);
516 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
518 (daemon_sleep
< 1) || (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
519 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
520 daemon_sleep
= 5 * HZ
;
522 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
523 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
526 * FIXME: write_behind for RAID1. If not specified, what
527 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
529 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
530 if (write_behind
> COUNTER_MAX
)
531 write_behind
= COUNTER_MAX
/ 2;
532 sb
->write_behind
= cpu_to_le32(write_behind
);
533 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
535 /* keep the array size field of the bitmap superblock up to date */
536 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
538 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
540 set_bit(BITMAP_STALE
, &bitmap
->flags
);
541 sb
->state
= cpu_to_le32(bitmap
->flags
);
542 bitmap
->events_cleared
= bitmap
->mddev
->events
;
543 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
544 bitmap
->mddev
->bitmap_info
.nodes
= 0;
551 /* read the superblock from the bitmap file and initialize some bitmap fields */
552 static int bitmap_read_sb(struct bitmap
*bitmap
)
556 unsigned long chunksize
, daemon_sleep
, write_behind
;
557 unsigned long long events
;
559 unsigned long sectors_reserved
= 0;
561 struct page
*sb_page
;
562 loff_t offset
= bitmap
->mddev
->bitmap_info
.offset
;
564 if (!bitmap
->storage
.file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
565 chunksize
= 128 * 1024 * 1024;
566 daemon_sleep
= 5 * HZ
;
568 set_bit(BITMAP_STALE
, &bitmap
->flags
);
572 /* page 0 is the superblock, read it... */
573 sb_page
= alloc_page(GFP_KERNEL
);
576 bitmap
->storage
.sb_page
= sb_page
;
579 /* If cluster_slot is set, the cluster is setup */
580 if (bitmap
->cluster_slot
>= 0) {
581 sector_t bm_blocks
= bitmap
->mddev
->resync_max_sectors
;
583 sector_div(bm_blocks
,
584 bitmap
->mddev
->bitmap_info
.chunksize
>> 9);
586 bm_blocks
= ((bm_blocks
+7) >> 3) + sizeof(bitmap_super_t
);
588 bm_blocks
= DIV_ROUND_UP_SECTOR_T(bm_blocks
, 4096);
589 offset
= bitmap
->mddev
->bitmap_info
.offset
+ (bitmap
->cluster_slot
* (bm_blocks
<< 3));
590 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__
, __LINE__
,
591 bitmap
->cluster_slot
, offset
);
594 if (bitmap
->storage
.file
) {
595 loff_t isize
= i_size_read(bitmap
->storage
.file
->f_mapping
->host
);
596 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
598 err
= read_page(bitmap
->storage
.file
, 0,
599 bitmap
, bytes
, sb_page
);
601 err
= read_sb_page(bitmap
->mddev
,
604 0, sizeof(bitmap_super_t
));
610 sb
= kmap_atomic(sb_page
);
612 chunksize
= le32_to_cpu(sb
->chunksize
);
613 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
614 write_behind
= le32_to_cpu(sb
->write_behind
);
615 sectors_reserved
= le32_to_cpu(sb
->sectors_reserved
);
616 /* Setup nodes/clustername only if bitmap version is
619 if (sb
->version
== cpu_to_le32(BITMAP_MAJOR_CLUSTERED
)) {
620 nodes
= le32_to_cpu(sb
->nodes
);
621 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
,
622 sb
->cluster_name
, 64);
625 /* verify that the bitmap-specific fields are valid */
626 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
627 reason
= "bad magic";
628 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
629 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_CLUSTERED
)
630 reason
= "unrecognized superblock version";
631 else if (chunksize
< 512)
632 reason
= "bitmap chunksize too small";
633 else if (!is_power_of_2(chunksize
))
634 reason
= "bitmap chunksize not a power of 2";
635 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
636 reason
= "daemon sleep period out of range";
637 else if (write_behind
> COUNTER_MAX
)
638 reason
= "write-behind limit out of range (0 - 16383)";
640 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
641 bmname(bitmap
), reason
);
645 /* keep the array size field of the bitmap superblock up to date */
646 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
648 if (bitmap
->mddev
->persistent
) {
650 * We have a persistent array superblock, so compare the
651 * bitmap's UUID and event counter to the mddev's
653 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
655 "%s: bitmap superblock UUID mismatch\n",
659 events
= le64_to_cpu(sb
->events
);
660 if (!nodes
&& (events
< bitmap
->mddev
->events
)) {
662 "%s: bitmap file is out of date (%llu < %llu) "
663 "-- forcing full recovery\n",
664 bmname(bitmap
), events
,
665 (unsigned long long) bitmap
->mddev
->events
);
666 set_bit(BITMAP_STALE
, &bitmap
->flags
);
670 /* assign fields using values from superblock */
671 bitmap
->flags
|= le32_to_cpu(sb
->state
);
672 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
673 set_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
);
674 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
675 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
, sb
->cluster_name
, 64);
680 /* Assiging chunksize is required for "re_read" */
681 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
682 if (err
== 0 && nodes
&& (bitmap
->cluster_slot
< 0)) {
683 err
= md_setup_cluster(bitmap
->mddev
, nodes
);
685 pr_err("%s: Could not setup cluster service (%d)\n",
686 bmname(bitmap
), err
);
689 bitmap
->cluster_slot
= md_cluster_ops
->slot_number(bitmap
->mddev
);
695 if (test_bit(BITMAP_STALE
, &bitmap
->flags
))
696 bitmap
->events_cleared
= bitmap
->mddev
->events
;
697 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
698 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
699 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
700 bitmap
->mddev
->bitmap_info
.nodes
= nodes
;
701 if (bitmap
->mddev
->bitmap_info
.space
== 0 ||
702 bitmap
->mddev
->bitmap_info
.space
> sectors_reserved
)
703 bitmap
->mddev
->bitmap_info
.space
= sectors_reserved
;
705 bitmap_print_sb(bitmap
);
706 if (bitmap
->cluster_slot
< 0)
707 md_cluster_stop(bitmap
->mddev
);
713 * general bitmap file operations
719 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
720 * file a page at a time. There's a superblock at the start of the file.
722 /* calculate the index of the page that contains this bit */
723 static inline unsigned long file_page_index(struct bitmap_storage
*store
,
727 chunk
+= sizeof(bitmap_super_t
) << 3;
728 return chunk
>> PAGE_BIT_SHIFT
;
731 /* calculate the (bit) offset of this bit within a page */
732 static inline unsigned long file_page_offset(struct bitmap_storage
*store
,
736 chunk
+= sizeof(bitmap_super_t
) << 3;
737 return chunk
& (PAGE_BITS
- 1);
741 * return a pointer to the page in the filemap that contains the given bit
744 static inline struct page
*filemap_get_page(struct bitmap_storage
*store
,
747 if (file_page_index(store
, chunk
) >= store
->file_pages
)
749 return store
->filemap
[file_page_index(store
, chunk
)];
752 static int bitmap_storage_alloc(struct bitmap_storage
*store
,
753 unsigned long chunks
, int with_super
,
756 int pnum
, offset
= 0;
757 unsigned long num_pages
;
760 bytes
= DIV_ROUND_UP(chunks
, 8);
762 bytes
+= sizeof(bitmap_super_t
);
764 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
765 offset
= slot_number
* (num_pages
- 1);
767 store
->filemap
= kmalloc(sizeof(struct page
*)
768 * num_pages
, GFP_KERNEL
);
772 if (with_super
&& !store
->sb_page
) {
773 store
->sb_page
= alloc_page(GFP_KERNEL
|__GFP_ZERO
);
774 if (store
->sb_page
== NULL
)
779 if (store
->sb_page
) {
780 store
->filemap
[0] = store
->sb_page
;
782 store
->sb_page
->index
= offset
;
785 for ( ; pnum
< num_pages
; pnum
++) {
786 store
->filemap
[pnum
] = alloc_page(GFP_KERNEL
|__GFP_ZERO
);
787 if (!store
->filemap
[pnum
]) {
788 store
->file_pages
= pnum
;
791 store
->filemap
[pnum
]->index
= pnum
+ offset
;
793 store
->file_pages
= pnum
;
795 /* We need 4 bits per page, rounded up to a multiple
796 * of sizeof(unsigned long) */
797 store
->filemap_attr
= kzalloc(
798 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
800 if (!store
->filemap_attr
)
803 store
->bytes
= bytes
;
808 static void bitmap_file_unmap(struct bitmap_storage
*store
)
810 struct page
**map
, *sb_page
;
815 map
= store
->filemap
;
816 pages
= store
->file_pages
;
817 sb_page
= store
->sb_page
;
820 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
821 free_buffers(map
[pages
]);
823 kfree(store
->filemap_attr
);
826 free_buffers(sb_page
);
829 struct inode
*inode
= file_inode(file
);
830 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
836 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
837 * then it is no longer reliable, so we stop using it and we mark the file
838 * as failed in the superblock
840 static void bitmap_file_kick(struct bitmap
*bitmap
)
842 char *path
, *ptr
= NULL
;
844 if (!test_and_set_bit(BITMAP_STALE
, &bitmap
->flags
)) {
845 bitmap_update_sb(bitmap
);
847 if (bitmap
->storage
.file
) {
848 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
850 ptr
= file_path(bitmap
->storage
.file
,
854 "%s: kicking failed bitmap file %s from array!\n",
855 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
860 "%s: disabling internal bitmap due to errors\n",
865 enum bitmap_page_attr
{
866 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
867 BITMAP_PAGE_PENDING
= 1, /* there are bits that are being cleaned.
868 * i.e. counter is 1 or 2. */
869 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
872 static inline void set_page_attr(struct bitmap
*bitmap
, int pnum
,
873 enum bitmap_page_attr attr
)
875 set_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
878 static inline void clear_page_attr(struct bitmap
*bitmap
, int pnum
,
879 enum bitmap_page_attr attr
)
881 clear_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
884 static inline int test_page_attr(struct bitmap
*bitmap
, int pnum
,
885 enum bitmap_page_attr attr
)
887 return test_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
890 static inline int test_and_clear_page_attr(struct bitmap
*bitmap
, int pnum
,
891 enum bitmap_page_attr attr
)
893 return test_and_clear_bit((pnum
<<2) + attr
,
894 bitmap
->storage
.filemap_attr
);
897 * bitmap_file_set_bit -- called before performing a write to the md device
898 * to set (and eventually sync) a particular bit in the bitmap file
900 * we set the bit immediately, then we record the page number so that
901 * when an unplug occurs, we can flush the dirty pages out to disk
903 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
908 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
910 page
= filemap_get_page(&bitmap
->storage
, chunk
);
913 bit
= file_page_offset(&bitmap
->storage
, chunk
);
916 kaddr
= kmap_atomic(page
);
917 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
920 set_bit_le(bit
, kaddr
);
921 kunmap_atomic(kaddr
);
922 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
923 /* record page number so it gets flushed to disk when unplug occurs */
924 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_DIRTY
);
927 static void bitmap_file_clear_bit(struct bitmap
*bitmap
, sector_t block
)
932 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
934 page
= filemap_get_page(&bitmap
->storage
, chunk
);
937 bit
= file_page_offset(&bitmap
->storage
, chunk
);
938 paddr
= kmap_atomic(page
);
939 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
940 clear_bit(bit
, paddr
);
942 clear_bit_le(bit
, paddr
);
943 kunmap_atomic(paddr
);
944 if (!test_page_attr(bitmap
, page
->index
, BITMAP_PAGE_NEEDWRITE
)) {
945 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_PENDING
);
946 bitmap
->allclean
= 0;
950 static int bitmap_file_test_bit(struct bitmap
*bitmap
, sector_t block
)
955 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
958 page
= filemap_get_page(&bitmap
->storage
, chunk
);
961 bit
= file_page_offset(&bitmap
->storage
, chunk
);
962 paddr
= kmap_atomic(page
);
963 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
964 set
= test_bit(bit
, paddr
);
966 set
= test_bit_le(bit
, paddr
);
967 kunmap_atomic(paddr
);
972 /* this gets called when the md device is ready to unplug its underlying
973 * (slave) device queues -- before we let any writes go down, we need to
974 * sync the dirty pages of the bitmap file to disk */
975 void bitmap_unplug(struct bitmap
*bitmap
)
978 int dirty
, need_write
;
980 if (!bitmap
|| !bitmap
->storage
.filemap
||
981 test_bit(BITMAP_STALE
, &bitmap
->flags
))
984 /* look at each page to see if there are any set bits that need to be
985 * flushed out to disk */
986 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++) {
987 if (!bitmap
->storage
.filemap
)
989 dirty
= test_and_clear_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
990 need_write
= test_and_clear_page_attr(bitmap
, i
,
991 BITMAP_PAGE_NEEDWRITE
);
992 if (dirty
|| need_write
) {
993 clear_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
);
994 write_page(bitmap
, bitmap
->storage
.filemap
[i
], 0);
997 if (bitmap
->storage
.file
)
998 wait_event(bitmap
->write_wait
,
999 atomic_read(&bitmap
->pending_writes
)==0);
1001 md_super_wait(bitmap
->mddev
);
1003 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1004 bitmap_file_kick(bitmap
);
1006 EXPORT_SYMBOL(bitmap_unplug
);
1008 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
1009 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1010 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1011 * memory mapping of the bitmap file
1013 * if there's no bitmap file, or if the bitmap file had been
1014 * previously kicked from the array, we mark all the bits as
1015 * 1's in order to cause a full resync.
1017 * We ignore all bits for sectors that end earlier than 'start'.
1018 * This is used when reading an out-of-date bitmap...
1020 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
1022 unsigned long i
, chunks
, index
, oldindex
, bit
, node_offset
= 0;
1023 struct page
*page
= NULL
;
1024 unsigned long bit_cnt
= 0;
1026 unsigned long offset
;
1030 struct bitmap_storage
*store
= &bitmap
->storage
;
1032 chunks
= bitmap
->counts
.chunks
;
1035 if (!file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
1036 /* No permanent bitmap - fill with '1s'. */
1037 store
->filemap
= NULL
;
1038 store
->file_pages
= 0;
1039 for (i
= 0; i
< chunks
; i
++) {
1040 /* if the disk bit is set, set the memory bit */
1041 int needed
= ((sector_t
)(i
+1) << (bitmap
->counts
.chunkshift
)
1043 bitmap_set_memory_bits(bitmap
,
1044 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1050 outofdate
= test_bit(BITMAP_STALE
, &bitmap
->flags
);
1052 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
1053 "recovery\n", bmname(bitmap
));
1055 if (file
&& i_size_read(file
->f_mapping
->host
) < store
->bytes
) {
1056 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
1058 (unsigned long) i_size_read(file
->f_mapping
->host
),
1065 if (!bitmap
->mddev
->bitmap_info
.external
)
1066 offset
= sizeof(bitmap_super_t
);
1068 if (mddev_is_clustered(bitmap
->mddev
))
1069 node_offset
= bitmap
->cluster_slot
* (DIV_ROUND_UP(store
->bytes
, PAGE_SIZE
));
1071 for (i
= 0; i
< chunks
; i
++) {
1073 index
= file_page_index(&bitmap
->storage
, i
);
1074 bit
= file_page_offset(&bitmap
->storage
, i
);
1075 if (index
!= oldindex
) { /* this is a new page, read it in */
1077 /* unmap the old page, we're done with it */
1078 if (index
== store
->file_pages
-1)
1079 count
= store
->bytes
- index
* PAGE_SIZE
;
1082 page
= store
->filemap
[index
];
1084 ret
= read_page(file
, index
, bitmap
,
1089 bitmap
->mddev
->bitmap_info
.offset
,
1091 index
+ node_offset
, count
);
1100 * if bitmap is out of date, dirty the
1101 * whole page and write it out
1103 paddr
= kmap_atomic(page
);
1104 memset(paddr
+ offset
, 0xff,
1105 PAGE_SIZE
- offset
);
1106 kunmap_atomic(paddr
);
1107 write_page(bitmap
, page
, 1);
1110 if (test_bit(BITMAP_WRITE_ERROR
,
1115 paddr
= kmap_atomic(page
);
1116 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
1117 b
= test_bit(bit
, paddr
);
1119 b
= test_bit_le(bit
, paddr
);
1120 kunmap_atomic(paddr
);
1122 /* if the disk bit is set, set the memory bit */
1123 int needed
= ((sector_t
)(i
+1) << bitmap
->counts
.chunkshift
1125 bitmap_set_memory_bits(bitmap
,
1126 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1133 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1134 "read %lu pages, set %lu of %lu bits\n",
1135 bmname(bitmap
), store
->file_pages
,
1141 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1142 bmname(bitmap
), ret
);
1146 void bitmap_write_all(struct bitmap
*bitmap
)
1148 /* We don't actually write all bitmap blocks here,
1149 * just flag them as needing to be written
1153 if (!bitmap
|| !bitmap
->storage
.filemap
)
1155 if (bitmap
->storage
.file
)
1156 /* Only one copy, so nothing needed */
1159 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1160 set_page_attr(bitmap
, i
,
1161 BITMAP_PAGE_NEEDWRITE
);
1162 bitmap
->allclean
= 0;
1165 static void bitmap_count_page(struct bitmap_counts
*bitmap
,
1166 sector_t offset
, int inc
)
1168 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1169 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1170 bitmap
->bp
[page
].count
+= inc
;
1171 bitmap_checkfree(bitmap
, page
);
1174 static void bitmap_set_pending(struct bitmap_counts
*bitmap
, sector_t offset
)
1176 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1177 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1178 struct bitmap_page
*bp
= &bitmap
->bp
[page
];
1184 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1185 sector_t offset
, sector_t
*blocks
,
1189 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1193 void bitmap_daemon_work(struct mddev
*mddev
)
1195 struct bitmap
*bitmap
;
1197 unsigned long nextpage
;
1199 struct bitmap_counts
*counts
;
1201 /* Use a mutex to guard daemon_work against
1204 mutex_lock(&mddev
->bitmap_info
.mutex
);
1205 bitmap
= mddev
->bitmap
;
1206 if (bitmap
== NULL
) {
1207 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1210 if (time_before(jiffies
, bitmap
->daemon_lastrun
1211 + mddev
->bitmap_info
.daemon_sleep
))
1214 bitmap
->daemon_lastrun
= jiffies
;
1215 if (bitmap
->allclean
) {
1216 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1219 bitmap
->allclean
= 1;
1221 /* Any file-page which is PENDING now needs to be written.
1222 * So set NEEDWRITE now, then after we make any last-minute changes
1225 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++)
1226 if (test_and_clear_page_attr(bitmap
, j
,
1227 BITMAP_PAGE_PENDING
))
1228 set_page_attr(bitmap
, j
,
1229 BITMAP_PAGE_NEEDWRITE
);
1231 if (bitmap
->need_sync
&&
1232 mddev
->bitmap_info
.external
== 0) {
1233 /* Arrange for superblock update as well as
1236 bitmap
->need_sync
= 0;
1237 if (bitmap
->storage
.filemap
) {
1238 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
1239 sb
->events_cleared
=
1240 cpu_to_le64(bitmap
->events_cleared
);
1242 set_page_attr(bitmap
, 0,
1243 BITMAP_PAGE_NEEDWRITE
);
1246 /* Now look at the bitmap counters and if any are '2' or '1',
1247 * decrement and handle accordingly.
1249 counts
= &bitmap
->counts
;
1250 spin_lock_irq(&counts
->lock
);
1252 for (j
= 0; j
< counts
->chunks
; j
++) {
1253 bitmap_counter_t
*bmc
;
1254 sector_t block
= (sector_t
)j
<< counts
->chunkshift
;
1256 if (j
== nextpage
) {
1257 nextpage
+= PAGE_COUNTER_RATIO
;
1258 if (!counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
) {
1259 j
|= PAGE_COUNTER_MASK
;
1262 counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
= 0;
1264 bmc
= bitmap_get_counter(counts
,
1269 j
|= PAGE_COUNTER_MASK
;
1272 if (*bmc
== 1 && !bitmap
->need_sync
) {
1273 /* We can clear the bit */
1275 bitmap_count_page(counts
, block
, -1);
1276 bitmap_file_clear_bit(bitmap
, block
);
1277 } else if (*bmc
&& *bmc
<= 2) {
1279 bitmap_set_pending(counts
, block
);
1280 bitmap
->allclean
= 0;
1283 spin_unlock_irq(&counts
->lock
);
1285 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1286 * DIRTY pages need to be written by bitmap_unplug so it can wait
1288 * If we find any DIRTY page we stop there and let bitmap_unplug
1289 * handle all the rest. This is important in the case where
1290 * the first blocking holds the superblock and it has been updated.
1291 * We mustn't write any other blocks before the superblock.
1294 j
< bitmap
->storage
.file_pages
1295 && !test_bit(BITMAP_STALE
, &bitmap
->flags
);
1297 if (test_page_attr(bitmap
, j
,
1299 /* bitmap_unplug will handle the rest */
1301 if (test_and_clear_page_attr(bitmap
, j
,
1302 BITMAP_PAGE_NEEDWRITE
)) {
1303 write_page(bitmap
, bitmap
->storage
.filemap
[j
], 0);
1308 if (bitmap
->allclean
== 0)
1309 mddev
->thread
->timeout
=
1310 mddev
->bitmap_info
.daemon_sleep
;
1311 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1314 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1315 sector_t offset
, sector_t
*blocks
,
1317 __releases(bitmap
->lock
)
1318 __acquires(bitmap
->lock
)
1320 /* If 'create', we might release the lock and reclaim it.
1321 * The lock must have been taken with interrupts enabled.
1322 * If !create, we don't release the lock.
1324 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1325 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1326 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1330 err
= bitmap_checkpage(bitmap
, page
, create
);
1332 if (bitmap
->bp
[page
].hijacked
||
1333 bitmap
->bp
[page
].map
== NULL
)
1334 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1335 PAGE_COUNTER_SHIFT
- 1);
1337 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1338 *blocks
= csize
- (offset
& (csize
- 1));
1343 /* now locked ... */
1345 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1346 /* should we use the first or second counter field
1347 * of the hijacked pointer? */
1348 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1349 return &((bitmap_counter_t
*)
1350 &bitmap
->bp
[page
].map
)[hi
];
1351 } else /* page is allocated */
1352 return (bitmap_counter_t
*)
1353 &(bitmap
->bp
[page
].map
[pageoff
]);
1356 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1363 atomic_inc(&bitmap
->behind_writes
);
1364 bw
= atomic_read(&bitmap
->behind_writes
);
1365 if (bw
> bitmap
->behind_writes_used
)
1366 bitmap
->behind_writes_used
= bw
;
1368 pr_debug("inc write-behind count %d/%lu\n",
1369 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1374 bitmap_counter_t
*bmc
;
1376 spin_lock_irq(&bitmap
->counts
.lock
);
1377 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 1);
1379 spin_unlock_irq(&bitmap
->counts
.lock
);
1383 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1384 DEFINE_WAIT(__wait
);
1385 /* note that it is safe to do the prepare_to_wait
1386 * after the test as long as we do it before dropping
1389 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1390 TASK_UNINTERRUPTIBLE
);
1391 spin_unlock_irq(&bitmap
->counts
.lock
);
1393 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1399 bitmap_file_set_bit(bitmap
, offset
);
1400 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1408 spin_unlock_irq(&bitmap
->counts
.lock
);
1411 if (sectors
> blocks
)
1418 EXPORT_SYMBOL(bitmap_startwrite
);
1420 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1421 int success
, int behind
)
1426 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1427 wake_up(&bitmap
->behind_wait
);
1428 pr_debug("dec write-behind count %d/%lu\n",
1429 atomic_read(&bitmap
->behind_writes
),
1430 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1435 unsigned long flags
;
1436 bitmap_counter_t
*bmc
;
1438 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1439 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 0);
1441 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1445 if (success
&& !bitmap
->mddev
->degraded
&&
1446 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1447 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1448 bitmap
->need_sync
= 1;
1449 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1452 if (!success
&& !NEEDED(*bmc
))
1453 *bmc
|= NEEDED_MASK
;
1455 if (COUNTER(*bmc
) == COUNTER_MAX
)
1456 wake_up(&bitmap
->overflow_wait
);
1460 bitmap_set_pending(&bitmap
->counts
, offset
);
1461 bitmap
->allclean
= 0;
1463 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1465 if (sectors
> blocks
)
1471 EXPORT_SYMBOL(bitmap_endwrite
);
1473 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1476 bitmap_counter_t
*bmc
;
1478 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1480 return 1; /* always resync if no bitmap */
1482 spin_lock_irq(&bitmap
->counts
.lock
);
1483 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1489 else if (NEEDED(*bmc
)) {
1491 if (!degraded
) { /* don't set/clear bits if degraded */
1492 *bmc
|= RESYNC_MASK
;
1493 *bmc
&= ~NEEDED_MASK
;
1497 spin_unlock_irq(&bitmap
->counts
.lock
);
1501 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1504 /* bitmap_start_sync must always report on multiples of whole
1505 * pages, otherwise resync (which is very PAGE_SIZE based) will
1507 * So call __bitmap_start_sync repeatedly (if needed) until
1508 * At least PAGE_SIZE>>9 blocks are covered.
1509 * Return the 'or' of the result.
1515 while (*blocks
< (PAGE_SIZE
>>9)) {
1516 rv
|= __bitmap_start_sync(bitmap
, offset
,
1517 &blocks1
, degraded
);
1523 EXPORT_SYMBOL(bitmap_start_sync
);
1525 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1527 bitmap_counter_t
*bmc
;
1528 unsigned long flags
;
1530 if (bitmap
== NULL
) {
1534 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1535 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1540 *bmc
&= ~RESYNC_MASK
;
1542 if (!NEEDED(*bmc
) && aborted
)
1543 *bmc
|= NEEDED_MASK
;
1546 bitmap_set_pending(&bitmap
->counts
, offset
);
1547 bitmap
->allclean
= 0;
1552 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1554 EXPORT_SYMBOL(bitmap_end_sync
);
1556 void bitmap_close_sync(struct bitmap
*bitmap
)
1558 /* Sync has finished, and any bitmap chunks that weren't synced
1559 * properly have been aborted. It remains to us to clear the
1560 * RESYNC bit wherever it is still on
1562 sector_t sector
= 0;
1566 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1567 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1571 EXPORT_SYMBOL(bitmap_close_sync
);
1573 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
, bool force
)
1581 bitmap
->last_end_sync
= jiffies
;
1584 if (!force
&& time_before(jiffies
, (bitmap
->last_end_sync
1585 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1587 wait_event(bitmap
->mddev
->recovery_wait
,
1588 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1590 bitmap
->mddev
->curr_resync_completed
= sector
;
1591 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1592 sector
&= ~((1ULL << bitmap
->counts
.chunkshift
) - 1);
1594 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1595 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1598 bitmap
->last_end_sync
= jiffies
;
1599 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1601 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1603 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1605 /* For each chunk covered by any of these sectors, set the
1606 * counter to 2 and possibly set resync_needed. They should all
1607 * be 0 at this point
1611 bitmap_counter_t
*bmc
;
1612 spin_lock_irq(&bitmap
->counts
.lock
);
1613 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &secs
, 1);
1615 spin_unlock_irq(&bitmap
->counts
.lock
);
1620 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1621 bitmap_set_pending(&bitmap
->counts
, offset
);
1622 bitmap
->allclean
= 0;
1625 *bmc
|= NEEDED_MASK
;
1626 spin_unlock_irq(&bitmap
->counts
.lock
);
1629 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1630 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1632 unsigned long chunk
;
1634 for (chunk
= s
; chunk
<= e
; chunk
++) {
1635 sector_t sec
= (sector_t
)chunk
<< bitmap
->counts
.chunkshift
;
1636 bitmap_set_memory_bits(bitmap
, sec
, 1);
1637 bitmap_file_set_bit(bitmap
, sec
);
1638 if (sec
< bitmap
->mddev
->recovery_cp
)
1639 /* We are asserting that the array is dirty,
1640 * so move the recovery_cp address back so
1641 * that it is obvious that it is dirty
1643 bitmap
->mddev
->recovery_cp
= sec
;
1648 * flush out any pending updates
1650 void bitmap_flush(struct mddev
*mddev
)
1652 struct bitmap
*bitmap
= mddev
->bitmap
;
1655 if (!bitmap
) /* there was no bitmap */
1658 /* run the daemon_work three time to ensure everything is flushed
1661 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1662 bitmap
->daemon_lastrun
-= sleep
;
1663 bitmap_daemon_work(mddev
);
1664 bitmap
->daemon_lastrun
-= sleep
;
1665 bitmap_daemon_work(mddev
);
1666 bitmap
->daemon_lastrun
-= sleep
;
1667 bitmap_daemon_work(mddev
);
1668 bitmap_update_sb(bitmap
);
1672 * free memory that was allocated
1674 static void bitmap_free(struct bitmap
*bitmap
)
1676 unsigned long k
, pages
;
1677 struct bitmap_page
*bp
;
1679 if (!bitmap
) /* there was no bitmap */
1682 if (mddev_is_clustered(bitmap
->mddev
) && bitmap
->mddev
->cluster_info
&&
1683 bitmap
->cluster_slot
== md_cluster_ops
->slot_number(bitmap
->mddev
))
1684 md_cluster_stop(bitmap
->mddev
);
1686 /* Shouldn't be needed - but just in case.... */
1687 wait_event(bitmap
->write_wait
,
1688 atomic_read(&bitmap
->pending_writes
) == 0);
1690 /* release the bitmap file */
1691 bitmap_file_unmap(&bitmap
->storage
);
1693 bp
= bitmap
->counts
.bp
;
1694 pages
= bitmap
->counts
.pages
;
1696 /* free all allocated memory */
1698 if (bp
) /* deallocate the page memory */
1699 for (k
= 0; k
< pages
; k
++)
1700 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1706 void bitmap_destroy(struct mddev
*mddev
)
1708 struct bitmap
*bitmap
= mddev
->bitmap
;
1710 if (!bitmap
) /* there was no bitmap */
1713 mutex_lock(&mddev
->bitmap_info
.mutex
);
1714 spin_lock(&mddev
->lock
);
1715 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1716 spin_unlock(&mddev
->lock
);
1717 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1719 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1721 if (bitmap
->sysfs_can_clear
)
1722 sysfs_put(bitmap
->sysfs_can_clear
);
1724 bitmap_free(bitmap
);
1728 * initialize the bitmap structure
1729 * if this returns an error, bitmap_destroy must be called to do clean up
1731 struct bitmap
*bitmap_create(struct mddev
*mddev
, int slot
)
1733 struct bitmap
*bitmap
;
1734 sector_t blocks
= mddev
->resync_max_sectors
;
1735 struct file
*file
= mddev
->bitmap_info
.file
;
1737 struct kernfs_node
*bm
= NULL
;
1739 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1741 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1743 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1745 return ERR_PTR(-ENOMEM
);
1747 spin_lock_init(&bitmap
->counts
.lock
);
1748 atomic_set(&bitmap
->pending_writes
, 0);
1749 init_waitqueue_head(&bitmap
->write_wait
);
1750 init_waitqueue_head(&bitmap
->overflow_wait
);
1751 init_waitqueue_head(&bitmap
->behind_wait
);
1753 bitmap
->mddev
= mddev
;
1754 bitmap
->cluster_slot
= slot
;
1757 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, "bitmap");
1759 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, "can_clear");
1762 bitmap
->sysfs_can_clear
= NULL
;
1764 bitmap
->storage
.file
= file
;
1767 /* As future accesses to this file will use bmap,
1768 * and bypass the page cache, we must sync the file
1773 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1774 if (!mddev
->bitmap_info
.external
) {
1776 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1777 * instructing us to create a new on-disk bitmap instance.
1779 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1780 err
= bitmap_new_disk_sb(bitmap
);
1782 err
= bitmap_read_sb(bitmap
);
1785 if (mddev
->bitmap_info
.chunksize
== 0 ||
1786 mddev
->bitmap_info
.daemon_sleep
== 0)
1787 /* chunksize and time_base need to be
1794 bitmap
->daemon_lastrun
= jiffies
;
1795 err
= bitmap_resize(bitmap
, blocks
, mddev
->bitmap_info
.chunksize
, 1);
1799 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1800 bitmap
->counts
.pages
, bmname(bitmap
));
1802 err
= test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
) ? -EIO
: 0;
1808 bitmap_free(bitmap
);
1809 return ERR_PTR(err
);
1812 int bitmap_load(struct mddev
*mddev
)
1816 sector_t sector
= 0;
1817 struct bitmap
*bitmap
= mddev
->bitmap
;
1822 /* Clear out old bitmap info first: Either there is none, or we
1823 * are resuming after someone else has possibly changed things,
1824 * so we should forget old cached info.
1825 * All chunks should be clean, but some might need_sync.
1827 while (sector
< mddev
->resync_max_sectors
) {
1829 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1832 bitmap_close_sync(bitmap
);
1834 if (mddev
->degraded
== 0
1835 || bitmap
->events_cleared
== mddev
->events
)
1836 /* no need to keep dirty bits to optimise a
1837 * re-add of a missing device */
1838 start
= mddev
->recovery_cp
;
1840 mutex_lock(&mddev
->bitmap_info
.mutex
);
1841 err
= bitmap_init_from_disk(bitmap
, start
);
1842 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1846 clear_bit(BITMAP_STALE
, &bitmap
->flags
);
1848 /* Kick recovery in case any bits were set */
1849 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1851 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1852 md_wakeup_thread(mddev
->thread
);
1854 bitmap_update_sb(bitmap
);
1856 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1861 EXPORT_SYMBOL_GPL(bitmap_load
);
1863 /* Loads the bitmap associated with slot and copies the resync information
1866 int bitmap_copy_from_slot(struct mddev
*mddev
, int slot
,
1867 sector_t
*low
, sector_t
*high
, bool clear_bits
)
1870 sector_t block
, lo
= 0, hi
= 0;
1871 struct bitmap_counts
*counts
;
1872 struct bitmap
*bitmap
= bitmap_create(mddev
, slot
);
1875 return PTR_ERR(bitmap
);
1877 rv
= bitmap_init_from_disk(bitmap
, 0);
1881 counts
= &bitmap
->counts
;
1882 for (j
= 0; j
< counts
->chunks
; j
++) {
1883 block
= (sector_t
)j
<< counts
->chunkshift
;
1884 if (bitmap_file_test_bit(bitmap
, block
)) {
1888 bitmap_file_clear_bit(bitmap
, block
);
1889 bitmap_set_memory_bits(mddev
->bitmap
, block
, 1);
1890 bitmap_file_set_bit(mddev
->bitmap
, block
);
1895 bitmap_update_sb(bitmap
);
1896 /* Setting this for the ev_page should be enough.
1897 * And we do not require both write_all and PAGE_DIRT either
1899 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1900 set_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
1901 bitmap_write_all(bitmap
);
1902 bitmap_unplug(bitmap
);
1907 bitmap_free(bitmap
);
1910 EXPORT_SYMBOL_GPL(bitmap_copy_from_slot
);
1913 void bitmap_status(struct seq_file
*seq
, struct bitmap
*bitmap
)
1915 unsigned long chunk_kb
;
1916 struct bitmap_counts
*counts
;
1921 counts
= &bitmap
->counts
;
1923 chunk_kb
= bitmap
->mddev
->bitmap_info
.chunksize
>> 10;
1924 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
1926 counts
->pages
- counts
->missing_pages
,
1928 (counts
->pages
- counts
->missing_pages
)
1929 << (PAGE_SHIFT
- 10),
1930 chunk_kb
? chunk_kb
: bitmap
->mddev
->bitmap_info
.chunksize
,
1931 chunk_kb
? "KB" : "B");
1932 if (bitmap
->storage
.file
) {
1933 seq_printf(seq
, ", file: ");
1934 seq_file_path(seq
, bitmap
->storage
.file
, " \t\n");
1937 seq_printf(seq
, "\n");
1940 int bitmap_resize(struct bitmap
*bitmap
, sector_t blocks
,
1941 int chunksize
, int init
)
1943 /* If chunk_size is 0, choose an appropriate chunk size.
1944 * Then possibly allocate new storage space.
1945 * Then quiesce, copy bits, replace bitmap, and re-start
1947 * This function is called both to set up the initial bitmap
1948 * and to resize the bitmap while the array is active.
1949 * If this happens as a result of the array being resized,
1950 * chunksize will be zero, and we need to choose a suitable
1951 * chunksize, otherwise we use what we are given.
1953 struct bitmap_storage store
;
1954 struct bitmap_counts old_counts
;
1955 unsigned long chunks
;
1957 sector_t old_blocks
, new_blocks
;
1961 struct bitmap_page
*new_bp
;
1963 if (chunksize
== 0) {
1964 /* If there is enough space, leave the chunk size unchanged,
1965 * else increase by factor of two until there is enough space.
1968 long space
= bitmap
->mddev
->bitmap_info
.space
;
1971 /* We don't know how much space there is, so limit
1972 * to current size - in sectors.
1974 bytes
= DIV_ROUND_UP(bitmap
->counts
.chunks
, 8);
1975 if (!bitmap
->mddev
->bitmap_info
.external
)
1976 bytes
+= sizeof(bitmap_super_t
);
1977 space
= DIV_ROUND_UP(bytes
, 512);
1978 bitmap
->mddev
->bitmap_info
.space
= space
;
1980 chunkshift
= bitmap
->counts
.chunkshift
;
1983 /* 'chunkshift' is shift from block size to chunk size */
1985 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
1986 bytes
= DIV_ROUND_UP(chunks
, 8);
1987 if (!bitmap
->mddev
->bitmap_info
.external
)
1988 bytes
+= sizeof(bitmap_super_t
);
1989 } while (bytes
> (space
<< 9));
1991 chunkshift
= ffz(~chunksize
) - BITMAP_BLOCK_SHIFT
;
1993 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
1994 memset(&store
, 0, sizeof(store
));
1995 if (bitmap
->mddev
->bitmap_info
.offset
|| bitmap
->mddev
->bitmap_info
.file
)
1996 ret
= bitmap_storage_alloc(&store
, chunks
,
1997 !bitmap
->mddev
->bitmap_info
.external
,
1998 mddev_is_clustered(bitmap
->mddev
)
1999 ? bitmap
->cluster_slot
: 0);
2003 pages
= DIV_ROUND_UP(chunks
, PAGE_COUNTER_RATIO
);
2005 new_bp
= kzalloc(pages
* sizeof(*new_bp
), GFP_KERNEL
);
2008 bitmap_file_unmap(&store
);
2013 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 1);
2015 store
.file
= bitmap
->storage
.file
;
2016 bitmap
->storage
.file
= NULL
;
2018 if (store
.sb_page
&& bitmap
->storage
.sb_page
)
2019 memcpy(page_address(store
.sb_page
),
2020 page_address(bitmap
->storage
.sb_page
),
2021 sizeof(bitmap_super_t
));
2022 bitmap_file_unmap(&bitmap
->storage
);
2023 bitmap
->storage
= store
;
2025 old_counts
= bitmap
->counts
;
2026 bitmap
->counts
.bp
= new_bp
;
2027 bitmap
->counts
.pages
= pages
;
2028 bitmap
->counts
.missing_pages
= pages
;
2029 bitmap
->counts
.chunkshift
= chunkshift
;
2030 bitmap
->counts
.chunks
= chunks
;
2031 bitmap
->mddev
->bitmap_info
.chunksize
= 1 << (chunkshift
+
2032 BITMAP_BLOCK_SHIFT
);
2034 blocks
= min(old_counts
.chunks
<< old_counts
.chunkshift
,
2035 chunks
<< chunkshift
);
2037 spin_lock_irq(&bitmap
->counts
.lock
);
2038 for (block
= 0; block
< blocks
; ) {
2039 bitmap_counter_t
*bmc_old
, *bmc_new
;
2042 bmc_old
= bitmap_get_counter(&old_counts
, block
,
2044 set
= bmc_old
&& NEEDED(*bmc_old
);
2047 bmc_new
= bitmap_get_counter(&bitmap
->counts
, block
,
2049 if (*bmc_new
== 0) {
2050 /* need to set on-disk bits too. */
2051 sector_t end
= block
+ new_blocks
;
2052 sector_t start
= block
>> chunkshift
;
2053 start
<<= chunkshift
;
2054 while (start
< end
) {
2055 bitmap_file_set_bit(bitmap
, block
);
2056 start
+= 1 << chunkshift
;
2059 bitmap_count_page(&bitmap
->counts
,
2061 bitmap_set_pending(&bitmap
->counts
,
2064 *bmc_new
|= NEEDED_MASK
;
2065 if (new_blocks
< old_blocks
)
2066 old_blocks
= new_blocks
;
2068 block
+= old_blocks
;
2073 while (block
< (chunks
<< chunkshift
)) {
2074 bitmap_counter_t
*bmc
;
2075 bmc
= bitmap_get_counter(&bitmap
->counts
, block
,
2078 /* new space. It needs to be resynced, so
2079 * we set NEEDED_MASK.
2082 *bmc
= NEEDED_MASK
| 2;
2083 bitmap_count_page(&bitmap
->counts
,
2085 bitmap_set_pending(&bitmap
->counts
,
2089 block
+= new_blocks
;
2091 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
2092 set_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
2094 spin_unlock_irq(&bitmap
->counts
.lock
);
2097 bitmap_unplug(bitmap
);
2098 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 0);
2104 EXPORT_SYMBOL_GPL(bitmap_resize
);
2107 location_show(struct mddev
*mddev
, char *page
)
2110 if (mddev
->bitmap_info
.file
)
2111 len
= sprintf(page
, "file");
2112 else if (mddev
->bitmap_info
.offset
)
2113 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
2115 len
= sprintf(page
, "none");
2116 len
+= sprintf(page
+len
, "\n");
2121 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2125 if (!mddev
->pers
->quiesce
)
2127 if (mddev
->recovery
|| mddev
->sync_thread
)
2131 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
2132 mddev
->bitmap_info
.offset
) {
2133 /* bitmap already configured. Only option is to clear it */
2134 if (strncmp(buf
, "none", 4) != 0)
2137 mddev
->pers
->quiesce(mddev
, 1);
2138 bitmap_destroy(mddev
);
2139 mddev
->pers
->quiesce(mddev
, 0);
2141 mddev
->bitmap_info
.offset
= 0;
2142 if (mddev
->bitmap_info
.file
) {
2143 struct file
*f
= mddev
->bitmap_info
.file
;
2144 mddev
->bitmap_info
.file
= NULL
;
2148 /* No bitmap, OK to set a location */
2150 if (strncmp(buf
, "none", 4) == 0)
2151 /* nothing to be done */;
2152 else if (strncmp(buf
, "file:", 5) == 0) {
2153 /* Not supported yet */
2158 rv
= kstrtoll(buf
+1, 10, &offset
);
2160 rv
= kstrtoll(buf
, 10, &offset
);
2165 if (mddev
->bitmap_info
.external
== 0 &&
2166 mddev
->major_version
== 0 &&
2167 offset
!= mddev
->bitmap_info
.default_offset
)
2169 mddev
->bitmap_info
.offset
= offset
;
2171 struct bitmap
*bitmap
;
2172 mddev
->pers
->quiesce(mddev
, 1);
2173 bitmap
= bitmap_create(mddev
, -1);
2175 rv
= PTR_ERR(bitmap
);
2177 mddev
->bitmap
= bitmap
;
2178 rv
= bitmap_load(mddev
);
2180 bitmap_destroy(mddev
);
2181 mddev
->bitmap_info
.offset
= 0;
2184 mddev
->pers
->quiesce(mddev
, 0);
2190 if (!mddev
->external
) {
2191 /* Ensure new bitmap info is stored in
2192 * metadata promptly.
2194 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2195 md_wakeup_thread(mddev
->thread
);
2200 static struct md_sysfs_entry bitmap_location
=
2201 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
2203 /* 'bitmap/space' is the space available at 'location' for the
2204 * bitmap. This allows the kernel to know when it is safe to
2205 * resize the bitmap to match a resized array.
2208 space_show(struct mddev
*mddev
, char *page
)
2210 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.space
);
2214 space_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2216 unsigned long sectors
;
2219 rv
= kstrtoul(buf
, 10, §ors
);
2226 if (mddev
->bitmap
&&
2227 sectors
< (mddev
->bitmap
->storage
.bytes
+ 511) >> 9)
2228 return -EFBIG
; /* Bitmap is too big for this small space */
2230 /* could make sure it isn't too big, but that isn't really
2231 * needed - user-space should be careful.
2233 mddev
->bitmap_info
.space
= sectors
;
2237 static struct md_sysfs_entry bitmap_space
=
2238 __ATTR(space
, S_IRUGO
|S_IWUSR
, space_show
, space_store
);
2241 timeout_show(struct mddev
*mddev
, char *page
)
2244 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
2245 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
2247 len
= sprintf(page
, "%lu", secs
);
2249 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
2250 len
+= sprintf(page
+len
, "\n");
2255 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2257 /* timeout can be set at any time */
2258 unsigned long timeout
;
2259 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
2263 /* just to make sure we don't overflow... */
2264 if (timeout
>= LONG_MAX
/ HZ
)
2267 timeout
= timeout
* HZ
/ 10000;
2269 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
2270 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2273 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2274 if (mddev
->thread
) {
2275 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2276 * the bitmap is all clean and we don't need to
2277 * adjust the timeout right now
2279 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2280 mddev
->thread
->timeout
= timeout
;
2281 md_wakeup_thread(mddev
->thread
);
2287 static struct md_sysfs_entry bitmap_timeout
=
2288 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2291 backlog_show(struct mddev
*mddev
, char *page
)
2293 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2297 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2299 unsigned long backlog
;
2300 int rv
= kstrtoul(buf
, 10, &backlog
);
2303 if (backlog
> COUNTER_MAX
)
2305 mddev
->bitmap_info
.max_write_behind
= backlog
;
2309 static struct md_sysfs_entry bitmap_backlog
=
2310 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2313 chunksize_show(struct mddev
*mddev
, char *page
)
2315 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2319 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2321 /* Can only be changed when no bitmap is active */
2323 unsigned long csize
;
2326 rv
= kstrtoul(buf
, 10, &csize
);
2330 !is_power_of_2(csize
))
2332 mddev
->bitmap_info
.chunksize
= csize
;
2336 static struct md_sysfs_entry bitmap_chunksize
=
2337 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2339 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2341 if (mddev_is_clustered(mddev
))
2342 return sprintf(page
, "clustered\n");
2343 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2344 ? "external" : "internal"));
2347 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2349 if (mddev
->bitmap
||
2350 mddev
->bitmap_info
.file
||
2351 mddev
->bitmap_info
.offset
)
2353 if (strncmp(buf
, "external", 8) == 0)
2354 mddev
->bitmap_info
.external
= 1;
2355 else if ((strncmp(buf
, "internal", 8) == 0) ||
2356 (strncmp(buf
, "clustered", 9) == 0))
2357 mddev
->bitmap_info
.external
= 0;
2363 static struct md_sysfs_entry bitmap_metadata
=
2364 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2366 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2369 spin_lock(&mddev
->lock
);
2371 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2374 len
= sprintf(page
, "\n");
2375 spin_unlock(&mddev
->lock
);
2379 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2381 if (mddev
->bitmap
== NULL
)
2383 if (strncmp(buf
, "false", 5) == 0)
2384 mddev
->bitmap
->need_sync
= 1;
2385 else if (strncmp(buf
, "true", 4) == 0) {
2386 if (mddev
->degraded
)
2388 mddev
->bitmap
->need_sync
= 0;
2394 static struct md_sysfs_entry bitmap_can_clear
=
2395 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2398 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2401 spin_lock(&mddev
->lock
);
2402 if (mddev
->bitmap
== NULL
)
2403 ret
= sprintf(page
, "0\n");
2405 ret
= sprintf(page
, "%lu\n",
2406 mddev
->bitmap
->behind_writes_used
);
2407 spin_unlock(&mddev
->lock
);
2412 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2415 mddev
->bitmap
->behind_writes_used
= 0;
2419 static struct md_sysfs_entry max_backlog_used
=
2420 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2421 behind_writes_used_show
, behind_writes_used_reset
);
2423 static struct attribute
*md_bitmap_attrs
[] = {
2424 &bitmap_location
.attr
,
2426 &bitmap_timeout
.attr
,
2427 &bitmap_backlog
.attr
,
2428 &bitmap_chunksize
.attr
,
2429 &bitmap_metadata
.attr
,
2430 &bitmap_can_clear
.attr
,
2431 &max_backlog_used
.attr
,
2434 struct attribute_group md_bitmap_group
= {
2436 .attrs
= md_bitmap_attrs
,