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
, int no_hijack
)
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 /* We don't support hijack for cluster raid */
96 /* failed - set the hijacked flag so that we can use the
97 * pointer as a counter */
98 if (!bitmap
->bp
[page
].map
)
99 bitmap
->bp
[page
].hijacked
= 1;
100 } else if (bitmap
->bp
[page
].map
||
101 bitmap
->bp
[page
].hijacked
) {
102 /* somebody beat us to getting the page */
106 /* no page was in place and we have one, so install it */
108 bitmap
->bp
[page
].map
= mappage
;
109 bitmap
->missing_pages
--;
114 /* if page is completely empty, put it back on the free list, or dealloc it */
115 /* if page was hijacked, unmark the flag so it might get alloced next time */
116 /* Note: lock should be held when calling this */
117 static void bitmap_checkfree(struct bitmap_counts
*bitmap
, unsigned long page
)
121 if (bitmap
->bp
[page
].count
) /* page is still busy */
124 /* page is no longer in use, it can be released */
126 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
127 bitmap
->bp
[page
].hijacked
= 0;
128 bitmap
->bp
[page
].map
= NULL
;
130 /* normal case, free the page */
131 ptr
= bitmap
->bp
[page
].map
;
132 bitmap
->bp
[page
].map
= NULL
;
133 bitmap
->missing_pages
++;
139 * bitmap file handling - read and write the bitmap file and its superblock
143 * basic page I/O operations
146 /* IO operations when bitmap is stored near all superblocks */
147 static int read_sb_page(struct mddev
*mddev
, loff_t offset
,
149 unsigned long index
, int size
)
151 /* choose a good rdev and read the page from there */
153 struct md_rdev
*rdev
;
156 rdev_for_each(rdev
, mddev
) {
157 if (! test_bit(In_sync
, &rdev
->flags
)
158 || test_bit(Faulty
, &rdev
->flags
))
161 target
= offset
+ index
* (PAGE_SIZE
/512);
163 if (sync_page_io(rdev
, target
,
164 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
173 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
175 /* Iterate the disks of an mddev, using rcu to protect access to the
176 * linked list, and raising the refcount of devices we return to ensure
177 * they don't disappear while in use.
178 * As devices are only added or removed when raid_disk is < 0 and
179 * nr_pending is 0 and In_sync is clear, the entries we return will
180 * still be in the same position on the list when we re-enter
181 * list_for_each_entry_continue_rcu.
183 * Note that if entered with 'rdev == NULL' to start at the
184 * beginning, we temporarily assign 'rdev' to an address which
185 * isn't really an rdev, but which can be used by
186 * list_for_each_entry_continue_rcu() to find the first entry.
190 /* start at the beginning */
191 rdev
= list_entry(&mddev
->disks
, struct md_rdev
, same_set
);
193 /* release the previous rdev and start from there. */
194 rdev_dec_pending(rdev
, mddev
);
196 list_for_each_entry_continue_rcu(rdev
, &mddev
->disks
, same_set
) {
197 if (rdev
->raid_disk
>= 0 &&
198 !test_bit(Faulty
, &rdev
->flags
)) {
199 /* this is a usable devices */
200 atomic_inc(&rdev
->nr_pending
);
209 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
211 struct md_rdev
*rdev
= NULL
;
212 struct block_device
*bdev
;
213 struct mddev
*mddev
= bitmap
->mddev
;
214 struct bitmap_storage
*store
= &bitmap
->storage
;
216 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
217 int size
= PAGE_SIZE
;
218 loff_t offset
= mddev
->bitmap_info
.offset
;
220 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
222 if (page
->index
== store
->file_pages
-1) {
223 int last_page_size
= store
->bytes
& (PAGE_SIZE
-1);
224 if (last_page_size
== 0)
225 last_page_size
= PAGE_SIZE
;
226 size
= roundup(last_page_size
,
227 bdev_logical_block_size(bdev
));
229 /* Just make sure we aren't corrupting data or
232 if (mddev
->external
) {
233 /* Bitmap could be anywhere. */
234 if (rdev
->sb_start
+ offset
+ (page
->index
238 rdev
->sb_start
+ offset
239 < (rdev
->data_offset
+ mddev
->dev_sectors
242 } else if (offset
< 0) {
243 /* DATA BITMAP METADATA */
245 + (long)(page
->index
* (PAGE_SIZE
/512))
247 /* bitmap runs in to metadata */
249 if (rdev
->data_offset
+ mddev
->dev_sectors
250 > rdev
->sb_start
+ offset
)
251 /* data runs in to bitmap */
253 } else if (rdev
->sb_start
< rdev
->data_offset
) {
254 /* METADATA BITMAP DATA */
257 + page
->index
*(PAGE_SIZE
/512) + size
/512
259 /* bitmap runs in to data */
262 /* DATA METADATA BITMAP - no problems */
264 md_super_write(mddev
, rdev
,
265 rdev
->sb_start
+ offset
266 + page
->index
* (PAGE_SIZE
/512),
272 md_super_wait(mddev
);
279 static void bitmap_file_kick(struct bitmap
*bitmap
);
281 * write out a page to a file
283 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
285 struct buffer_head
*bh
;
287 if (bitmap
->storage
.file
== NULL
) {
288 switch (write_sb_page(bitmap
, page
, wait
)) {
290 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
294 bh
= page_buffers(page
);
296 while (bh
&& bh
->b_blocknr
) {
297 atomic_inc(&bitmap
->pending_writes
);
298 set_buffer_locked(bh
);
299 set_buffer_mapped(bh
);
300 submit_bh(WRITE
| REQ_SYNC
, bh
);
301 bh
= bh
->b_this_page
;
305 wait_event(bitmap
->write_wait
,
306 atomic_read(&bitmap
->pending_writes
)==0);
308 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
309 bitmap_file_kick(bitmap
);
312 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
314 struct bitmap
*bitmap
= bh
->b_private
;
317 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
318 if (atomic_dec_and_test(&bitmap
->pending_writes
))
319 wake_up(&bitmap
->write_wait
);
322 /* copied from buffer.c */
324 __clear_page_buffers(struct page
*page
)
326 ClearPagePrivate(page
);
327 set_page_private(page
, 0);
330 static void free_buffers(struct page
*page
)
332 struct buffer_head
*bh
;
334 if (!PagePrivate(page
))
337 bh
= page_buffers(page
);
339 struct buffer_head
*next
= bh
->b_this_page
;
340 free_buffer_head(bh
);
343 __clear_page_buffers(page
);
347 /* read a page from a file.
348 * We both read the page, and attach buffers to the page to record the
349 * address of each block (using bmap). These addresses will be used
350 * to write the block later, completely bypassing the filesystem.
351 * This usage is similar to how swap files are handled, and allows us
352 * to write to a file with no concerns of memory allocation failing.
354 static int read_page(struct file
*file
, unsigned long index
,
355 struct bitmap
*bitmap
,
360 struct inode
*inode
= file_inode(file
);
361 struct buffer_head
*bh
;
364 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
365 (unsigned long long)index
<< PAGE_SHIFT
);
367 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
372 attach_page_buffers(page
, bh
);
373 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
378 bh
->b_blocknr
= bmap(inode
, block
);
379 if (bh
->b_blocknr
== 0) {
380 /* Cannot use this file! */
384 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
385 if (count
< (1<<inode
->i_blkbits
))
388 count
-= (1<<inode
->i_blkbits
);
390 bh
->b_end_io
= end_bitmap_write
;
391 bh
->b_private
= bitmap
;
392 atomic_inc(&bitmap
->pending_writes
);
393 set_buffer_locked(bh
);
394 set_buffer_mapped(bh
);
398 bh
= bh
->b_this_page
;
402 wait_event(bitmap
->write_wait
,
403 atomic_read(&bitmap
->pending_writes
)==0);
404 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
408 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %d\n",
410 (unsigned long long)index
<< PAGE_SHIFT
,
416 * bitmap file superblock operations
419 /* update the event counter and sync the superblock to disk */
420 void bitmap_update_sb(struct bitmap
*bitmap
)
424 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
426 if (bitmap
->mddev
->bitmap_info
.external
)
428 if (!bitmap
->storage
.sb_page
) /* no superblock */
430 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
431 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
432 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
433 /* rocking back to read-only */
434 bitmap
->events_cleared
= bitmap
->mddev
->events
;
435 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
436 sb
->state
= cpu_to_le32(bitmap
->flags
);
437 /* Just in case these have been changed via sysfs: */
438 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
439 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
440 /* This might have been changed by a reshape */
441 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
442 sb
->chunksize
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.chunksize
);
443 sb
->nodes
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.nodes
);
444 sb
->sectors_reserved
= cpu_to_le32(bitmap
->mddev
->
447 write_page(bitmap
, bitmap
->storage
.sb_page
, 1);
450 /* print out the bitmap file superblock */
451 void bitmap_print_sb(struct bitmap
*bitmap
)
455 if (!bitmap
|| !bitmap
->storage
.sb_page
)
457 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
458 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
459 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
460 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
461 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
462 *(__u32
*)(sb
->uuid
+0),
463 *(__u32
*)(sb
->uuid
+4),
464 *(__u32
*)(sb
->uuid
+8),
465 *(__u32
*)(sb
->uuid
+12));
466 printk(KERN_DEBUG
" events: %llu\n",
467 (unsigned long long) le64_to_cpu(sb
->events
));
468 printk(KERN_DEBUG
"events cleared: %llu\n",
469 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
470 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
471 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
472 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
473 printk(KERN_DEBUG
" sync size: %llu KB\n",
474 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
475 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
483 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
484 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
485 * This function verifies 'bitmap_info' and populates the on-disk bitmap
486 * structure, which is to be written to disk.
488 * Returns: 0 on success, -Exxx on error
490 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
493 unsigned long chunksize
, daemon_sleep
, write_behind
;
495 bitmap
->storage
.sb_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
496 if (bitmap
->storage
.sb_page
== NULL
)
498 bitmap
->storage
.sb_page
->index
= 0;
500 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
502 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
503 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
505 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
507 if (!is_power_of_2(chunksize
)) {
509 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
512 sb
->chunksize
= cpu_to_le32(chunksize
);
514 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
515 if (!daemon_sleep
|| (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
516 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
517 daemon_sleep
= 5 * HZ
;
519 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
520 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
523 * FIXME: write_behind for RAID1. If not specified, what
524 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
526 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
527 if (write_behind
> COUNTER_MAX
)
528 write_behind
= COUNTER_MAX
/ 2;
529 sb
->write_behind
= cpu_to_le32(write_behind
);
530 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
532 /* keep the array size field of the bitmap superblock up to date */
533 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
535 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
537 set_bit(BITMAP_STALE
, &bitmap
->flags
);
538 sb
->state
= cpu_to_le32(bitmap
->flags
);
539 bitmap
->events_cleared
= bitmap
->mddev
->events
;
540 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
541 bitmap
->mddev
->bitmap_info
.nodes
= 0;
548 /* read the superblock from the bitmap file and initialize some bitmap fields */
549 static int bitmap_read_sb(struct bitmap
*bitmap
)
553 unsigned long chunksize
, daemon_sleep
, write_behind
;
554 unsigned long long events
;
556 unsigned long sectors_reserved
= 0;
558 struct page
*sb_page
;
559 loff_t offset
= bitmap
->mddev
->bitmap_info
.offset
;
561 if (!bitmap
->storage
.file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
562 chunksize
= 128 * 1024 * 1024;
563 daemon_sleep
= 5 * HZ
;
565 set_bit(BITMAP_STALE
, &bitmap
->flags
);
569 /* page 0 is the superblock, read it... */
570 sb_page
= alloc_page(GFP_KERNEL
);
573 bitmap
->storage
.sb_page
= sb_page
;
576 /* If cluster_slot is set, the cluster is setup */
577 if (bitmap
->cluster_slot
>= 0) {
578 sector_t bm_blocks
= bitmap
->mddev
->resync_max_sectors
;
580 sector_div(bm_blocks
,
581 bitmap
->mddev
->bitmap_info
.chunksize
>> 9);
583 bm_blocks
= ((bm_blocks
+7) >> 3) + sizeof(bitmap_super_t
);
585 bm_blocks
= DIV_ROUND_UP_SECTOR_T(bm_blocks
, 4096);
586 offset
= bitmap
->mddev
->bitmap_info
.offset
+ (bitmap
->cluster_slot
* (bm_blocks
<< 3));
587 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__
, __LINE__
,
588 bitmap
->cluster_slot
, offset
);
591 if (bitmap
->storage
.file
) {
592 loff_t isize
= i_size_read(bitmap
->storage
.file
->f_mapping
->host
);
593 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
595 err
= read_page(bitmap
->storage
.file
, 0,
596 bitmap
, bytes
, sb_page
);
598 err
= read_sb_page(bitmap
->mddev
,
601 0, sizeof(bitmap_super_t
));
607 sb
= kmap_atomic(sb_page
);
609 chunksize
= le32_to_cpu(sb
->chunksize
);
610 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
611 write_behind
= le32_to_cpu(sb
->write_behind
);
612 sectors_reserved
= le32_to_cpu(sb
->sectors_reserved
);
613 /* Setup nodes/clustername only if bitmap version is
616 if (sb
->version
== cpu_to_le32(BITMAP_MAJOR_CLUSTERED
)) {
617 nodes
= le32_to_cpu(sb
->nodes
);
618 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
,
619 sb
->cluster_name
, 64);
622 /* verify that the bitmap-specific fields are valid */
623 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
624 reason
= "bad magic";
625 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
626 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_CLUSTERED
)
627 reason
= "unrecognized superblock version";
628 else if (chunksize
< 512)
629 reason
= "bitmap chunksize too small";
630 else if (!is_power_of_2(chunksize
))
631 reason
= "bitmap chunksize not a power of 2";
632 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
633 reason
= "daemon sleep period out of range";
634 else if (write_behind
> COUNTER_MAX
)
635 reason
= "write-behind limit out of range (0 - 16383)";
637 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
638 bmname(bitmap
), reason
);
642 /* keep the array size field of the bitmap superblock up to date */
643 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
645 if (bitmap
->mddev
->persistent
) {
647 * We have a persistent array superblock, so compare the
648 * bitmap's UUID and event counter to the mddev's
650 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
652 "%s: bitmap superblock UUID mismatch\n",
656 events
= le64_to_cpu(sb
->events
);
657 if (!nodes
&& (events
< bitmap
->mddev
->events
)) {
659 "%s: bitmap file is out of date (%llu < %llu) "
660 "-- forcing full recovery\n",
661 bmname(bitmap
), events
,
662 (unsigned long long) bitmap
->mddev
->events
);
663 set_bit(BITMAP_STALE
, &bitmap
->flags
);
667 /* assign fields using values from superblock */
668 bitmap
->flags
|= le32_to_cpu(sb
->state
);
669 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
670 set_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
);
671 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
672 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
, sb
->cluster_name
, 64);
677 /* Assiging chunksize is required for "re_read" */
678 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
679 if (err
== 0 && nodes
&& (bitmap
->cluster_slot
< 0)) {
680 err
= md_setup_cluster(bitmap
->mddev
, nodes
);
682 pr_err("%s: Could not setup cluster service (%d)\n",
683 bmname(bitmap
), err
);
686 bitmap
->cluster_slot
= md_cluster_ops
->slot_number(bitmap
->mddev
);
692 if (test_bit(BITMAP_STALE
, &bitmap
->flags
))
693 bitmap
->events_cleared
= bitmap
->mddev
->events
;
694 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
695 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
696 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
697 bitmap
->mddev
->bitmap_info
.nodes
= nodes
;
698 if (bitmap
->mddev
->bitmap_info
.space
== 0 ||
699 bitmap
->mddev
->bitmap_info
.space
> sectors_reserved
)
700 bitmap
->mddev
->bitmap_info
.space
= sectors_reserved
;
702 bitmap_print_sb(bitmap
);
703 if (bitmap
->cluster_slot
< 0)
704 md_cluster_stop(bitmap
->mddev
);
710 * general bitmap file operations
716 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
717 * file a page at a time. There's a superblock at the start of the file.
719 /* calculate the index of the page that contains this bit */
720 static inline unsigned long file_page_index(struct bitmap_storage
*store
,
724 chunk
+= sizeof(bitmap_super_t
) << 3;
725 return chunk
>> PAGE_BIT_SHIFT
;
728 /* calculate the (bit) offset of this bit within a page */
729 static inline unsigned long file_page_offset(struct bitmap_storage
*store
,
733 chunk
+= sizeof(bitmap_super_t
) << 3;
734 return chunk
& (PAGE_BITS
- 1);
738 * return a pointer to the page in the filemap that contains the given bit
741 static inline struct page
*filemap_get_page(struct bitmap_storage
*store
,
744 if (file_page_index(store
, chunk
) >= store
->file_pages
)
746 return store
->filemap
[file_page_index(store
, chunk
)];
749 static int bitmap_storage_alloc(struct bitmap_storage
*store
,
750 unsigned long chunks
, int with_super
,
753 int pnum
, offset
= 0;
754 unsigned long num_pages
;
757 bytes
= DIV_ROUND_UP(chunks
, 8);
759 bytes
+= sizeof(bitmap_super_t
);
761 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
762 offset
= slot_number
* (num_pages
- 1);
764 store
->filemap
= kmalloc(sizeof(struct page
*)
765 * num_pages
, GFP_KERNEL
);
769 if (with_super
&& !store
->sb_page
) {
770 store
->sb_page
= alloc_page(GFP_KERNEL
|__GFP_ZERO
);
771 if (store
->sb_page
== NULL
)
776 if (store
->sb_page
) {
777 store
->filemap
[0] = store
->sb_page
;
779 store
->sb_page
->index
= offset
;
782 for ( ; pnum
< num_pages
; pnum
++) {
783 store
->filemap
[pnum
] = alloc_page(GFP_KERNEL
|__GFP_ZERO
);
784 if (!store
->filemap
[pnum
]) {
785 store
->file_pages
= pnum
;
788 store
->filemap
[pnum
]->index
= pnum
+ offset
;
790 store
->file_pages
= pnum
;
792 /* We need 4 bits per page, rounded up to a multiple
793 * of sizeof(unsigned long) */
794 store
->filemap_attr
= kzalloc(
795 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
797 if (!store
->filemap_attr
)
800 store
->bytes
= bytes
;
805 static void bitmap_file_unmap(struct bitmap_storage
*store
)
807 struct page
**map
, *sb_page
;
812 map
= store
->filemap
;
813 pages
= store
->file_pages
;
814 sb_page
= store
->sb_page
;
817 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
818 free_buffers(map
[pages
]);
820 kfree(store
->filemap_attr
);
823 free_buffers(sb_page
);
826 struct inode
*inode
= file_inode(file
);
827 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
833 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
834 * then it is no longer reliable, so we stop using it and we mark the file
835 * as failed in the superblock
837 static void bitmap_file_kick(struct bitmap
*bitmap
)
839 char *path
, *ptr
= NULL
;
841 if (!test_and_set_bit(BITMAP_STALE
, &bitmap
->flags
)) {
842 bitmap_update_sb(bitmap
);
844 if (bitmap
->storage
.file
) {
845 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
847 ptr
= file_path(bitmap
->storage
.file
,
851 "%s: kicking failed bitmap file %s from array!\n",
852 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
857 "%s: disabling internal bitmap due to errors\n",
862 enum bitmap_page_attr
{
863 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
864 BITMAP_PAGE_PENDING
= 1, /* there are bits that are being cleaned.
865 * i.e. counter is 1 or 2. */
866 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
869 static inline void set_page_attr(struct bitmap
*bitmap
, int pnum
,
870 enum bitmap_page_attr attr
)
872 set_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
875 static inline void clear_page_attr(struct bitmap
*bitmap
, int pnum
,
876 enum bitmap_page_attr attr
)
878 clear_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
881 static inline int test_page_attr(struct bitmap
*bitmap
, int pnum
,
882 enum bitmap_page_attr attr
)
884 return test_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
887 static inline int test_and_clear_page_attr(struct bitmap
*bitmap
, int pnum
,
888 enum bitmap_page_attr attr
)
890 return test_and_clear_bit((pnum
<<2) + attr
,
891 bitmap
->storage
.filemap_attr
);
894 * bitmap_file_set_bit -- called before performing a write to the md device
895 * to set (and eventually sync) a particular bit in the bitmap file
897 * we set the bit immediately, then we record the page number so that
898 * when an unplug occurs, we can flush the dirty pages out to disk
900 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
905 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
907 page
= filemap_get_page(&bitmap
->storage
, chunk
);
910 bit
= file_page_offset(&bitmap
->storage
, chunk
);
913 kaddr
= kmap_atomic(page
);
914 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
917 set_bit_le(bit
, kaddr
);
918 kunmap_atomic(kaddr
);
919 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
920 /* record page number so it gets flushed to disk when unplug occurs */
921 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_DIRTY
);
924 static void bitmap_file_clear_bit(struct bitmap
*bitmap
, sector_t block
)
929 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
931 page
= filemap_get_page(&bitmap
->storage
, chunk
);
934 bit
= file_page_offset(&bitmap
->storage
, chunk
);
935 paddr
= kmap_atomic(page
);
936 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
937 clear_bit(bit
, paddr
);
939 clear_bit_le(bit
, paddr
);
940 kunmap_atomic(paddr
);
941 if (!test_page_attr(bitmap
, page
->index
, BITMAP_PAGE_NEEDWRITE
)) {
942 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_PENDING
);
943 bitmap
->allclean
= 0;
947 static int bitmap_file_test_bit(struct bitmap
*bitmap
, sector_t block
)
952 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
955 page
= filemap_get_page(&bitmap
->storage
, chunk
);
958 bit
= file_page_offset(&bitmap
->storage
, chunk
);
959 paddr
= kmap_atomic(page
);
960 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
961 set
= test_bit(bit
, paddr
);
963 set
= test_bit_le(bit
, paddr
);
964 kunmap_atomic(paddr
);
969 /* this gets called when the md device is ready to unplug its underlying
970 * (slave) device queues -- before we let any writes go down, we need to
971 * sync the dirty pages of the bitmap file to disk */
972 void bitmap_unplug(struct bitmap
*bitmap
)
975 int dirty
, need_write
;
977 if (!bitmap
|| !bitmap
->storage
.filemap
||
978 test_bit(BITMAP_STALE
, &bitmap
->flags
))
981 /* look at each page to see if there are any set bits that need to be
982 * flushed out to disk */
983 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++) {
984 if (!bitmap
->storage
.filemap
)
986 dirty
= test_and_clear_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
987 need_write
= test_and_clear_page_attr(bitmap
, i
,
988 BITMAP_PAGE_NEEDWRITE
);
989 if (dirty
|| need_write
) {
990 clear_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
);
991 write_page(bitmap
, bitmap
->storage
.filemap
[i
], 0);
994 if (bitmap
->storage
.file
)
995 wait_event(bitmap
->write_wait
,
996 atomic_read(&bitmap
->pending_writes
)==0);
998 md_super_wait(bitmap
->mddev
);
1000 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1001 bitmap_file_kick(bitmap
);
1003 EXPORT_SYMBOL(bitmap_unplug
);
1005 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
1006 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1007 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1008 * memory mapping of the bitmap file
1010 * if there's no bitmap file, or if the bitmap file had been
1011 * previously kicked from the array, we mark all the bits as
1012 * 1's in order to cause a full resync.
1014 * We ignore all bits for sectors that end earlier than 'start'.
1015 * This is used when reading an out-of-date bitmap...
1017 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
1019 unsigned long i
, chunks
, index
, oldindex
, bit
, node_offset
= 0;
1020 struct page
*page
= NULL
;
1021 unsigned long bit_cnt
= 0;
1023 unsigned long offset
;
1027 struct bitmap_storage
*store
= &bitmap
->storage
;
1029 chunks
= bitmap
->counts
.chunks
;
1032 if (!file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
1033 /* No permanent bitmap - fill with '1s'. */
1034 store
->filemap
= NULL
;
1035 store
->file_pages
= 0;
1036 for (i
= 0; i
< chunks
; i
++) {
1037 /* if the disk bit is set, set the memory bit */
1038 int needed
= ((sector_t
)(i
+1) << (bitmap
->counts
.chunkshift
)
1040 bitmap_set_memory_bits(bitmap
,
1041 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1047 outofdate
= test_bit(BITMAP_STALE
, &bitmap
->flags
);
1049 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
1050 "recovery\n", bmname(bitmap
));
1052 if (file
&& i_size_read(file
->f_mapping
->host
) < store
->bytes
) {
1053 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
1055 (unsigned long) i_size_read(file
->f_mapping
->host
),
1062 if (!bitmap
->mddev
->bitmap_info
.external
)
1063 offset
= sizeof(bitmap_super_t
);
1065 if (mddev_is_clustered(bitmap
->mddev
))
1066 node_offset
= bitmap
->cluster_slot
* (DIV_ROUND_UP(store
->bytes
, PAGE_SIZE
));
1068 for (i
= 0; i
< chunks
; i
++) {
1070 index
= file_page_index(&bitmap
->storage
, i
);
1071 bit
= file_page_offset(&bitmap
->storage
, i
);
1072 if (index
!= oldindex
) { /* this is a new page, read it in */
1074 /* unmap the old page, we're done with it */
1075 if (index
== store
->file_pages
-1)
1076 count
= store
->bytes
- index
* PAGE_SIZE
;
1079 page
= store
->filemap
[index
];
1081 ret
= read_page(file
, index
, bitmap
,
1086 bitmap
->mddev
->bitmap_info
.offset
,
1088 index
+ node_offset
, count
);
1097 * if bitmap is out of date, dirty the
1098 * whole page and write it out
1100 paddr
= kmap_atomic(page
);
1101 memset(paddr
+ offset
, 0xff,
1102 PAGE_SIZE
- offset
);
1103 kunmap_atomic(paddr
);
1104 write_page(bitmap
, page
, 1);
1107 if (test_bit(BITMAP_WRITE_ERROR
,
1112 paddr
= kmap_atomic(page
);
1113 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
1114 b
= test_bit(bit
, paddr
);
1116 b
= test_bit_le(bit
, paddr
);
1117 kunmap_atomic(paddr
);
1119 /* if the disk bit is set, set the memory bit */
1120 int needed
= ((sector_t
)(i
+1) << bitmap
->counts
.chunkshift
1122 bitmap_set_memory_bits(bitmap
,
1123 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1130 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1131 "read %lu pages, set %lu of %lu bits\n",
1132 bmname(bitmap
), store
->file_pages
,
1138 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1139 bmname(bitmap
), ret
);
1143 void bitmap_write_all(struct bitmap
*bitmap
)
1145 /* We don't actually write all bitmap blocks here,
1146 * just flag them as needing to be written
1150 if (!bitmap
|| !bitmap
->storage
.filemap
)
1152 if (bitmap
->storage
.file
)
1153 /* Only one copy, so nothing needed */
1156 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1157 set_page_attr(bitmap
, i
,
1158 BITMAP_PAGE_NEEDWRITE
);
1159 bitmap
->allclean
= 0;
1162 static void bitmap_count_page(struct bitmap_counts
*bitmap
,
1163 sector_t offset
, int inc
)
1165 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1166 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1167 bitmap
->bp
[page
].count
+= inc
;
1168 bitmap_checkfree(bitmap
, page
);
1171 static void bitmap_set_pending(struct bitmap_counts
*bitmap
, sector_t offset
)
1173 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1174 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1175 struct bitmap_page
*bp
= &bitmap
->bp
[page
];
1181 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1182 sector_t offset
, sector_t
*blocks
,
1186 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1190 void bitmap_daemon_work(struct mddev
*mddev
)
1192 struct bitmap
*bitmap
;
1194 unsigned long nextpage
;
1196 struct bitmap_counts
*counts
;
1198 /* Use a mutex to guard daemon_work against
1201 mutex_lock(&mddev
->bitmap_info
.mutex
);
1202 bitmap
= mddev
->bitmap
;
1203 if (bitmap
== NULL
) {
1204 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1207 if (time_before(jiffies
, bitmap
->daemon_lastrun
1208 + mddev
->bitmap_info
.daemon_sleep
))
1211 bitmap
->daemon_lastrun
= jiffies
;
1212 if (bitmap
->allclean
) {
1213 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1216 bitmap
->allclean
= 1;
1218 /* Any file-page which is PENDING now needs to be written.
1219 * So set NEEDWRITE now, then after we make any last-minute changes
1222 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++)
1223 if (test_and_clear_page_attr(bitmap
, j
,
1224 BITMAP_PAGE_PENDING
))
1225 set_page_attr(bitmap
, j
,
1226 BITMAP_PAGE_NEEDWRITE
);
1228 if (bitmap
->need_sync
&&
1229 mddev
->bitmap_info
.external
== 0) {
1230 /* Arrange for superblock update as well as
1233 bitmap
->need_sync
= 0;
1234 if (bitmap
->storage
.filemap
) {
1235 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
1236 sb
->events_cleared
=
1237 cpu_to_le64(bitmap
->events_cleared
);
1239 set_page_attr(bitmap
, 0,
1240 BITMAP_PAGE_NEEDWRITE
);
1243 /* Now look at the bitmap counters and if any are '2' or '1',
1244 * decrement and handle accordingly.
1246 counts
= &bitmap
->counts
;
1247 spin_lock_irq(&counts
->lock
);
1249 for (j
= 0; j
< counts
->chunks
; j
++) {
1250 bitmap_counter_t
*bmc
;
1251 sector_t block
= (sector_t
)j
<< counts
->chunkshift
;
1253 if (j
== nextpage
) {
1254 nextpage
+= PAGE_COUNTER_RATIO
;
1255 if (!counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
) {
1256 j
|= PAGE_COUNTER_MASK
;
1259 counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
= 0;
1261 bmc
= bitmap_get_counter(counts
,
1266 j
|= PAGE_COUNTER_MASK
;
1269 if (*bmc
== 1 && !bitmap
->need_sync
) {
1270 /* We can clear the bit */
1272 bitmap_count_page(counts
, block
, -1);
1273 bitmap_file_clear_bit(bitmap
, block
);
1274 } else if (*bmc
&& *bmc
<= 2) {
1276 bitmap_set_pending(counts
, block
);
1277 bitmap
->allclean
= 0;
1280 spin_unlock_irq(&counts
->lock
);
1282 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1283 * DIRTY pages need to be written by bitmap_unplug so it can wait
1285 * If we find any DIRTY page we stop there and let bitmap_unplug
1286 * handle all the rest. This is important in the case where
1287 * the first blocking holds the superblock and it has been updated.
1288 * We mustn't write any other blocks before the superblock.
1291 j
< bitmap
->storage
.file_pages
1292 && !test_bit(BITMAP_STALE
, &bitmap
->flags
);
1294 if (test_page_attr(bitmap
, j
,
1296 /* bitmap_unplug will handle the rest */
1298 if (test_and_clear_page_attr(bitmap
, j
,
1299 BITMAP_PAGE_NEEDWRITE
)) {
1300 write_page(bitmap
, bitmap
->storage
.filemap
[j
], 0);
1305 if (bitmap
->allclean
== 0)
1306 mddev
->thread
->timeout
=
1307 mddev
->bitmap_info
.daemon_sleep
;
1308 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1311 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1312 sector_t offset
, sector_t
*blocks
,
1314 __releases(bitmap
->lock
)
1315 __acquires(bitmap
->lock
)
1317 /* If 'create', we might release the lock and reclaim it.
1318 * The lock must have been taken with interrupts enabled.
1319 * If !create, we don't release the lock.
1321 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1322 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1323 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1327 err
= bitmap_checkpage(bitmap
, page
, create
, 0);
1329 if (bitmap
->bp
[page
].hijacked
||
1330 bitmap
->bp
[page
].map
== NULL
)
1331 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1332 PAGE_COUNTER_SHIFT
- 1);
1334 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1335 *blocks
= csize
- (offset
& (csize
- 1));
1340 /* now locked ... */
1342 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1343 /* should we use the first or second counter field
1344 * of the hijacked pointer? */
1345 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1346 return &((bitmap_counter_t
*)
1347 &bitmap
->bp
[page
].map
)[hi
];
1348 } else /* page is allocated */
1349 return (bitmap_counter_t
*)
1350 &(bitmap
->bp
[page
].map
[pageoff
]);
1353 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1360 atomic_inc(&bitmap
->behind_writes
);
1361 bw
= atomic_read(&bitmap
->behind_writes
);
1362 if (bw
> bitmap
->behind_writes_used
)
1363 bitmap
->behind_writes_used
= bw
;
1365 pr_debug("inc write-behind count %d/%lu\n",
1366 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1371 bitmap_counter_t
*bmc
;
1373 spin_lock_irq(&bitmap
->counts
.lock
);
1374 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 1);
1376 spin_unlock_irq(&bitmap
->counts
.lock
);
1380 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1381 DEFINE_WAIT(__wait
);
1382 /* note that it is safe to do the prepare_to_wait
1383 * after the test as long as we do it before dropping
1386 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1387 TASK_UNINTERRUPTIBLE
);
1388 spin_unlock_irq(&bitmap
->counts
.lock
);
1390 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1396 bitmap_file_set_bit(bitmap
, offset
);
1397 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1405 spin_unlock_irq(&bitmap
->counts
.lock
);
1408 if (sectors
> blocks
)
1415 EXPORT_SYMBOL(bitmap_startwrite
);
1417 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1418 int success
, int behind
)
1423 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1424 wake_up(&bitmap
->behind_wait
);
1425 pr_debug("dec write-behind count %d/%lu\n",
1426 atomic_read(&bitmap
->behind_writes
),
1427 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1432 unsigned long flags
;
1433 bitmap_counter_t
*bmc
;
1435 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1436 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 0);
1438 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1442 if (success
&& !bitmap
->mddev
->degraded
&&
1443 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1444 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1445 bitmap
->need_sync
= 1;
1446 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1449 if (!success
&& !NEEDED(*bmc
))
1450 *bmc
|= NEEDED_MASK
;
1452 if (COUNTER(*bmc
) == COUNTER_MAX
)
1453 wake_up(&bitmap
->overflow_wait
);
1457 bitmap_set_pending(&bitmap
->counts
, offset
);
1458 bitmap
->allclean
= 0;
1460 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1462 if (sectors
> blocks
)
1468 EXPORT_SYMBOL(bitmap_endwrite
);
1470 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1473 bitmap_counter_t
*bmc
;
1475 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1477 return 1; /* always resync if no bitmap */
1479 spin_lock_irq(&bitmap
->counts
.lock
);
1480 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1486 else if (NEEDED(*bmc
)) {
1488 if (!degraded
) { /* don't set/clear bits if degraded */
1489 *bmc
|= RESYNC_MASK
;
1490 *bmc
&= ~NEEDED_MASK
;
1494 spin_unlock_irq(&bitmap
->counts
.lock
);
1498 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1501 /* bitmap_start_sync must always report on multiples of whole
1502 * pages, otherwise resync (which is very PAGE_SIZE based) will
1504 * So call __bitmap_start_sync repeatedly (if needed) until
1505 * At least PAGE_SIZE>>9 blocks are covered.
1506 * Return the 'or' of the result.
1512 while (*blocks
< (PAGE_SIZE
>>9)) {
1513 rv
|= __bitmap_start_sync(bitmap
, offset
,
1514 &blocks1
, degraded
);
1520 EXPORT_SYMBOL(bitmap_start_sync
);
1522 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1524 bitmap_counter_t
*bmc
;
1525 unsigned long flags
;
1527 if (bitmap
== NULL
) {
1531 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1532 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1537 *bmc
&= ~RESYNC_MASK
;
1539 if (!NEEDED(*bmc
) && aborted
)
1540 *bmc
|= NEEDED_MASK
;
1543 bitmap_set_pending(&bitmap
->counts
, offset
);
1544 bitmap
->allclean
= 0;
1549 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1551 EXPORT_SYMBOL(bitmap_end_sync
);
1553 void bitmap_close_sync(struct bitmap
*bitmap
)
1555 /* Sync has finished, and any bitmap chunks that weren't synced
1556 * properly have been aborted. It remains to us to clear the
1557 * RESYNC bit wherever it is still on
1559 sector_t sector
= 0;
1563 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1564 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1568 EXPORT_SYMBOL(bitmap_close_sync
);
1570 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
, bool force
)
1578 bitmap
->last_end_sync
= jiffies
;
1581 if (!force
&& time_before(jiffies
, (bitmap
->last_end_sync
1582 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1584 wait_event(bitmap
->mddev
->recovery_wait
,
1585 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1587 bitmap
->mddev
->curr_resync_completed
= sector
;
1588 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1589 sector
&= ~((1ULL << bitmap
->counts
.chunkshift
) - 1);
1591 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1592 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1595 bitmap
->last_end_sync
= jiffies
;
1596 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1598 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1600 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1602 /* For each chunk covered by any of these sectors, set the
1603 * counter to 2 and possibly set resync_needed. They should all
1604 * be 0 at this point
1608 bitmap_counter_t
*bmc
;
1609 spin_lock_irq(&bitmap
->counts
.lock
);
1610 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &secs
, 1);
1612 spin_unlock_irq(&bitmap
->counts
.lock
);
1617 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1618 bitmap_set_pending(&bitmap
->counts
, offset
);
1619 bitmap
->allclean
= 0;
1622 *bmc
|= NEEDED_MASK
;
1623 spin_unlock_irq(&bitmap
->counts
.lock
);
1626 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1627 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1629 unsigned long chunk
;
1631 for (chunk
= s
; chunk
<= e
; chunk
++) {
1632 sector_t sec
= (sector_t
)chunk
<< bitmap
->counts
.chunkshift
;
1633 bitmap_set_memory_bits(bitmap
, sec
, 1);
1634 bitmap_file_set_bit(bitmap
, sec
);
1635 if (sec
< bitmap
->mddev
->recovery_cp
)
1636 /* We are asserting that the array is dirty,
1637 * so move the recovery_cp address back so
1638 * that it is obvious that it is dirty
1640 bitmap
->mddev
->recovery_cp
= sec
;
1645 * flush out any pending updates
1647 void bitmap_flush(struct mddev
*mddev
)
1649 struct bitmap
*bitmap
= mddev
->bitmap
;
1652 if (!bitmap
) /* there was no bitmap */
1655 /* run the daemon_work three time to ensure everything is flushed
1658 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1659 bitmap
->daemon_lastrun
-= sleep
;
1660 bitmap_daemon_work(mddev
);
1661 bitmap
->daemon_lastrun
-= sleep
;
1662 bitmap_daemon_work(mddev
);
1663 bitmap
->daemon_lastrun
-= sleep
;
1664 bitmap_daemon_work(mddev
);
1665 bitmap_update_sb(bitmap
);
1669 * free memory that was allocated
1671 static void bitmap_free(struct bitmap
*bitmap
)
1673 unsigned long k
, pages
;
1674 struct bitmap_page
*bp
;
1676 if (!bitmap
) /* there was no bitmap */
1679 if (bitmap
->sysfs_can_clear
)
1680 sysfs_put(bitmap
->sysfs_can_clear
);
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 bitmap_free(bitmap
);
1725 * initialize the bitmap structure
1726 * if this returns an error, bitmap_destroy must be called to do clean up
1727 * once mddev->bitmap is set
1729 struct bitmap
*bitmap_create(struct mddev
*mddev
, int slot
)
1731 struct bitmap
*bitmap
;
1732 sector_t blocks
= mddev
->resync_max_sectors
;
1733 struct file
*file
= mddev
->bitmap_info
.file
;
1735 struct kernfs_node
*bm
= NULL
;
1737 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1739 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1741 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1743 return ERR_PTR(-ENOMEM
);
1745 spin_lock_init(&bitmap
->counts
.lock
);
1746 atomic_set(&bitmap
->pending_writes
, 0);
1747 init_waitqueue_head(&bitmap
->write_wait
);
1748 init_waitqueue_head(&bitmap
->overflow_wait
);
1749 init_waitqueue_head(&bitmap
->behind_wait
);
1751 bitmap
->mddev
= mddev
;
1752 bitmap
->cluster_slot
= slot
;
1755 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, "bitmap");
1757 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, "can_clear");
1760 bitmap
->sysfs_can_clear
= NULL
;
1762 bitmap
->storage
.file
= file
;
1765 /* As future accesses to this file will use bmap,
1766 * and bypass the page cache, we must sync the file
1771 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1772 if (!mddev
->bitmap_info
.external
) {
1774 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1775 * instructing us to create a new on-disk bitmap instance.
1777 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1778 err
= bitmap_new_disk_sb(bitmap
);
1780 err
= bitmap_read_sb(bitmap
);
1783 if (mddev
->bitmap_info
.chunksize
== 0 ||
1784 mddev
->bitmap_info
.daemon_sleep
== 0)
1785 /* chunksize and time_base need to be
1792 bitmap
->daemon_lastrun
= jiffies
;
1793 err
= bitmap_resize(bitmap
, blocks
, mddev
->bitmap_info
.chunksize
, 1);
1797 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1798 bitmap
->counts
.pages
, bmname(bitmap
));
1800 err
= test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
) ? -EIO
: 0;
1806 bitmap_free(bitmap
);
1807 return ERR_PTR(err
);
1810 int bitmap_load(struct mddev
*mddev
)
1814 sector_t sector
= 0;
1815 struct bitmap
*bitmap
= mddev
->bitmap
;
1820 /* Clear out old bitmap info first: Either there is none, or we
1821 * are resuming after someone else has possibly changed things,
1822 * so we should forget old cached info.
1823 * All chunks should be clean, but some might need_sync.
1825 while (sector
< mddev
->resync_max_sectors
) {
1827 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1830 bitmap_close_sync(bitmap
);
1832 if (mddev
->degraded
== 0
1833 || bitmap
->events_cleared
== mddev
->events
)
1834 /* no need to keep dirty bits to optimise a
1835 * re-add of a missing device */
1836 start
= mddev
->recovery_cp
;
1838 mutex_lock(&mddev
->bitmap_info
.mutex
);
1839 err
= bitmap_init_from_disk(bitmap
, start
);
1840 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1844 clear_bit(BITMAP_STALE
, &bitmap
->flags
);
1846 /* Kick recovery in case any bits were set */
1847 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1849 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1850 md_wakeup_thread(mddev
->thread
);
1852 bitmap_update_sb(bitmap
);
1854 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1859 EXPORT_SYMBOL_GPL(bitmap_load
);
1861 /* Loads the bitmap associated with slot and copies the resync information
1864 int bitmap_copy_from_slot(struct mddev
*mddev
, int slot
,
1865 sector_t
*low
, sector_t
*high
, bool clear_bits
)
1868 sector_t block
, lo
= 0, hi
= 0;
1869 struct bitmap_counts
*counts
;
1870 struct bitmap
*bitmap
= bitmap_create(mddev
, slot
);
1872 if (IS_ERR(bitmap
)) {
1873 bitmap_free(bitmap
);
1874 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 cluster raid, need to pre-allocate bitmap */
2039 if (mddev_is_clustered(bitmap
->mddev
)) {
2041 for (page
= 0; page
< pages
; page
++) {
2042 ret
= bitmap_checkpage(&bitmap
->counts
, page
, 1, 1);
2046 /* deallocate the page memory */
2047 for (k
= 0; k
< page
; k
++) {
2049 kfree(new_bp
[k
].map
);
2052 /* restore some fields from old_counts */
2053 bitmap
->counts
.bp
= old_counts
.bp
;
2054 bitmap
->counts
.pages
= old_counts
.pages
;
2055 bitmap
->counts
.missing_pages
= old_counts
.pages
;
2056 bitmap
->counts
.chunkshift
= old_counts
.chunkshift
;
2057 bitmap
->counts
.chunks
= old_counts
.chunks
;
2058 bitmap
->mddev
->bitmap_info
.chunksize
= 1 << (old_counts
.chunkshift
+
2059 BITMAP_BLOCK_SHIFT
);
2060 blocks
= old_counts
.chunks
<< old_counts
.chunkshift
;
2061 pr_err("Could not pre-allocate in-memory bitmap for cluster raid\n");
2064 bitmap
->counts
.bp
[page
].count
+= 1;
2068 for (block
= 0; block
< blocks
; ) {
2069 bitmap_counter_t
*bmc_old
, *bmc_new
;
2072 bmc_old
= bitmap_get_counter(&old_counts
, block
,
2074 set
= bmc_old
&& NEEDED(*bmc_old
);
2077 bmc_new
= bitmap_get_counter(&bitmap
->counts
, block
,
2079 if (*bmc_new
== 0) {
2080 /* need to set on-disk bits too. */
2081 sector_t end
= block
+ new_blocks
;
2082 sector_t start
= block
>> chunkshift
;
2083 start
<<= chunkshift
;
2084 while (start
< end
) {
2085 bitmap_file_set_bit(bitmap
, block
);
2086 start
+= 1 << chunkshift
;
2089 bitmap_count_page(&bitmap
->counts
,
2091 bitmap_set_pending(&bitmap
->counts
,
2094 *bmc_new
|= NEEDED_MASK
;
2095 if (new_blocks
< old_blocks
)
2096 old_blocks
= new_blocks
;
2098 block
+= old_blocks
;
2103 while (block
< (chunks
<< chunkshift
)) {
2104 bitmap_counter_t
*bmc
;
2105 bmc
= bitmap_get_counter(&bitmap
->counts
, block
,
2108 /* new space. It needs to be resynced, so
2109 * we set NEEDED_MASK.
2112 *bmc
= NEEDED_MASK
| 2;
2113 bitmap_count_page(&bitmap
->counts
,
2115 bitmap_set_pending(&bitmap
->counts
,
2119 block
+= new_blocks
;
2121 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
2122 set_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
2124 spin_unlock_irq(&bitmap
->counts
.lock
);
2127 bitmap_unplug(bitmap
);
2128 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 0);
2134 EXPORT_SYMBOL_GPL(bitmap_resize
);
2137 location_show(struct mddev
*mddev
, char *page
)
2140 if (mddev
->bitmap_info
.file
)
2141 len
= sprintf(page
, "file");
2142 else if (mddev
->bitmap_info
.offset
)
2143 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
2145 len
= sprintf(page
, "none");
2146 len
+= sprintf(page
+len
, "\n");
2151 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2155 if (!mddev
->pers
->quiesce
)
2157 if (mddev
->recovery
|| mddev
->sync_thread
)
2161 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
2162 mddev
->bitmap_info
.offset
) {
2163 /* bitmap already configured. Only option is to clear it */
2164 if (strncmp(buf
, "none", 4) != 0)
2167 mddev
->pers
->quiesce(mddev
, 1);
2168 bitmap_destroy(mddev
);
2169 mddev
->pers
->quiesce(mddev
, 0);
2171 mddev
->bitmap_info
.offset
= 0;
2172 if (mddev
->bitmap_info
.file
) {
2173 struct file
*f
= mddev
->bitmap_info
.file
;
2174 mddev
->bitmap_info
.file
= NULL
;
2178 /* No bitmap, OK to set a location */
2180 if (strncmp(buf
, "none", 4) == 0)
2181 /* nothing to be done */;
2182 else if (strncmp(buf
, "file:", 5) == 0) {
2183 /* Not supported yet */
2188 rv
= kstrtoll(buf
+1, 10, &offset
);
2190 rv
= kstrtoll(buf
, 10, &offset
);
2195 if (mddev
->bitmap_info
.external
== 0 &&
2196 mddev
->major_version
== 0 &&
2197 offset
!= mddev
->bitmap_info
.default_offset
)
2199 mddev
->bitmap_info
.offset
= offset
;
2201 struct bitmap
*bitmap
;
2202 mddev
->pers
->quiesce(mddev
, 1);
2203 bitmap
= bitmap_create(mddev
, -1);
2205 rv
= PTR_ERR(bitmap
);
2207 mddev
->bitmap
= bitmap
;
2208 rv
= bitmap_load(mddev
);
2210 mddev
->bitmap_info
.offset
= 0;
2212 mddev
->pers
->quiesce(mddev
, 0);
2214 bitmap_destroy(mddev
);
2220 if (!mddev
->external
) {
2221 /* Ensure new bitmap info is stored in
2222 * metadata promptly.
2224 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2225 md_wakeup_thread(mddev
->thread
);
2230 static struct md_sysfs_entry bitmap_location
=
2231 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
2233 /* 'bitmap/space' is the space available at 'location' for the
2234 * bitmap. This allows the kernel to know when it is safe to
2235 * resize the bitmap to match a resized array.
2238 space_show(struct mddev
*mddev
, char *page
)
2240 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.space
);
2244 space_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2246 unsigned long sectors
;
2249 rv
= kstrtoul(buf
, 10, §ors
);
2256 if (mddev
->bitmap
&&
2257 sectors
< (mddev
->bitmap
->storage
.bytes
+ 511) >> 9)
2258 return -EFBIG
; /* Bitmap is too big for this small space */
2260 /* could make sure it isn't too big, but that isn't really
2261 * needed - user-space should be careful.
2263 mddev
->bitmap_info
.space
= sectors
;
2267 static struct md_sysfs_entry bitmap_space
=
2268 __ATTR(space
, S_IRUGO
|S_IWUSR
, space_show
, space_store
);
2271 timeout_show(struct mddev
*mddev
, char *page
)
2274 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
2275 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
2277 len
= sprintf(page
, "%lu", secs
);
2279 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
2280 len
+= sprintf(page
+len
, "\n");
2285 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2287 /* timeout can be set at any time */
2288 unsigned long timeout
;
2289 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
2293 /* just to make sure we don't overflow... */
2294 if (timeout
>= LONG_MAX
/ HZ
)
2297 timeout
= timeout
* HZ
/ 10000;
2299 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
2300 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2303 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2304 if (mddev
->thread
) {
2305 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2306 * the bitmap is all clean and we don't need to
2307 * adjust the timeout right now
2309 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2310 mddev
->thread
->timeout
= timeout
;
2311 md_wakeup_thread(mddev
->thread
);
2317 static struct md_sysfs_entry bitmap_timeout
=
2318 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2321 backlog_show(struct mddev
*mddev
, char *page
)
2323 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2327 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2329 unsigned long backlog
;
2330 int rv
= kstrtoul(buf
, 10, &backlog
);
2333 if (backlog
> COUNTER_MAX
)
2335 mddev
->bitmap_info
.max_write_behind
= backlog
;
2339 static struct md_sysfs_entry bitmap_backlog
=
2340 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2343 chunksize_show(struct mddev
*mddev
, char *page
)
2345 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2349 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2351 /* Can only be changed when no bitmap is active */
2353 unsigned long csize
;
2356 rv
= kstrtoul(buf
, 10, &csize
);
2360 !is_power_of_2(csize
))
2362 mddev
->bitmap_info
.chunksize
= csize
;
2366 static struct md_sysfs_entry bitmap_chunksize
=
2367 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2369 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2371 if (mddev_is_clustered(mddev
))
2372 return sprintf(page
, "clustered\n");
2373 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2374 ? "external" : "internal"));
2377 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2379 if (mddev
->bitmap
||
2380 mddev
->bitmap_info
.file
||
2381 mddev
->bitmap_info
.offset
)
2383 if (strncmp(buf
, "external", 8) == 0)
2384 mddev
->bitmap_info
.external
= 1;
2385 else if ((strncmp(buf
, "internal", 8) == 0) ||
2386 (strncmp(buf
, "clustered", 9) == 0))
2387 mddev
->bitmap_info
.external
= 0;
2393 static struct md_sysfs_entry bitmap_metadata
=
2394 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2396 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2399 spin_lock(&mddev
->lock
);
2401 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2404 len
= sprintf(page
, "\n");
2405 spin_unlock(&mddev
->lock
);
2409 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2411 if (mddev
->bitmap
== NULL
)
2413 if (strncmp(buf
, "false", 5) == 0)
2414 mddev
->bitmap
->need_sync
= 1;
2415 else if (strncmp(buf
, "true", 4) == 0) {
2416 if (mddev
->degraded
)
2418 mddev
->bitmap
->need_sync
= 0;
2424 static struct md_sysfs_entry bitmap_can_clear
=
2425 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2428 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2431 spin_lock(&mddev
->lock
);
2432 if (mddev
->bitmap
== NULL
)
2433 ret
= sprintf(page
, "0\n");
2435 ret
= sprintf(page
, "%lu\n",
2436 mddev
->bitmap
->behind_writes_used
);
2437 spin_unlock(&mddev
->lock
);
2442 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2445 mddev
->bitmap
->behind_writes_used
= 0;
2449 static struct md_sysfs_entry max_backlog_used
=
2450 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2451 behind_writes_used_show
, behind_writes_used_reset
);
2453 static struct attribute
*md_bitmap_attrs
[] = {
2454 &bitmap_location
.attr
,
2456 &bitmap_timeout
.attr
,
2457 &bitmap_backlog
.attr
,
2458 &bitmap_chunksize
.attr
,
2459 &bitmap_metadata
.attr
,
2460 &bitmap_can_clear
.attr
,
2461 &max_backlog_used
.attr
,
2464 struct attribute_group md_bitmap_group
= {
2466 .attrs
= md_bitmap_attrs
,