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 */
103 /* no page was in place and we have one, so install it */
105 bitmap
->bp
[page
].map
= mappage
;
106 bitmap
->missing_pages
--;
111 /* if page is completely empty, put it back on the free list, or dealloc it */
112 /* if page was hijacked, unmark the flag so it might get alloced next time */
113 /* Note: lock should be held when calling this */
114 static void bitmap_checkfree(struct bitmap_counts
*bitmap
, unsigned long page
)
118 if (bitmap
->bp
[page
].count
) /* page is still busy */
121 /* page is no longer in use, it can be released */
123 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
124 bitmap
->bp
[page
].hijacked
= 0;
125 bitmap
->bp
[page
].map
= NULL
;
127 /* normal case, free the page */
128 ptr
= bitmap
->bp
[page
].map
;
129 bitmap
->bp
[page
].map
= NULL
;
130 bitmap
->missing_pages
++;
136 * bitmap file handling - read and write the bitmap file and its superblock
140 * basic page I/O operations
143 /* IO operations when bitmap is stored near all superblocks */
144 static int read_sb_page(struct mddev
*mddev
, loff_t offset
,
146 unsigned long index
, int size
)
148 /* choose a good rdev and read the page from there */
150 struct md_rdev
*rdev
;
153 rdev_for_each(rdev
, mddev
) {
154 if (! test_bit(In_sync
, &rdev
->flags
)
155 || test_bit(Faulty
, &rdev
->flags
))
158 target
= offset
+ index
* (PAGE_SIZE
/512);
160 if (sync_page_io(rdev
, target
,
161 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
170 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
172 /* Iterate the disks of an mddev, using rcu to protect access to the
173 * linked list, and raising the refcount of devices we return to ensure
174 * they don't disappear while in use.
175 * As devices are only added or removed when raid_disk is < 0 and
176 * nr_pending is 0 and In_sync is clear, the entries we return will
177 * still be in the same position on the list when we re-enter
178 * list_for_each_entry_continue_rcu.
180 * Note that if entered with 'rdev == NULL' to start at the
181 * beginning, we temporarily assign 'rdev' to an address which
182 * isn't really an rdev, but which can be used by
183 * list_for_each_entry_continue_rcu() to find the first entry.
187 /* start at the beginning */
188 rdev
= list_entry(&mddev
->disks
, struct md_rdev
, same_set
);
190 /* release the previous rdev and start from there. */
191 rdev_dec_pending(rdev
, mddev
);
193 list_for_each_entry_continue_rcu(rdev
, &mddev
->disks
, same_set
) {
194 if (rdev
->raid_disk
>= 0 &&
195 !test_bit(Faulty
, &rdev
->flags
)) {
196 /* this is a usable devices */
197 atomic_inc(&rdev
->nr_pending
);
206 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
208 struct md_rdev
*rdev
= NULL
;
209 struct block_device
*bdev
;
210 struct mddev
*mddev
= bitmap
->mddev
;
211 struct bitmap_storage
*store
= &bitmap
->storage
;
213 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
214 int size
= PAGE_SIZE
;
215 loff_t offset
= mddev
->bitmap_info
.offset
;
217 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
219 if (page
->index
== store
->file_pages
-1) {
220 int last_page_size
= store
->bytes
& (PAGE_SIZE
-1);
221 if (last_page_size
== 0)
222 last_page_size
= PAGE_SIZE
;
223 size
= roundup(last_page_size
,
224 bdev_logical_block_size(bdev
));
226 /* Just make sure we aren't corrupting data or
229 if (mddev
->external
) {
230 /* Bitmap could be anywhere. */
231 if (rdev
->sb_start
+ offset
+ (page
->index
235 rdev
->sb_start
+ offset
236 < (rdev
->data_offset
+ mddev
->dev_sectors
239 } else if (offset
< 0) {
240 /* DATA BITMAP METADATA */
242 + (long)(page
->index
* (PAGE_SIZE
/512))
244 /* bitmap runs in to metadata */
246 if (rdev
->data_offset
+ mddev
->dev_sectors
247 > rdev
->sb_start
+ offset
)
248 /* data runs in to bitmap */
250 } else if (rdev
->sb_start
< rdev
->data_offset
) {
251 /* METADATA BITMAP DATA */
254 + page
->index
*(PAGE_SIZE
/512) + size
/512
256 /* bitmap runs in to data */
259 /* DATA METADATA BITMAP - no problems */
261 md_super_write(mddev
, rdev
,
262 rdev
->sb_start
+ offset
263 + page
->index
* (PAGE_SIZE
/512),
269 md_super_wait(mddev
);
276 static void bitmap_file_kick(struct bitmap
*bitmap
);
278 * write out a page to a file
280 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
282 struct buffer_head
*bh
;
284 if (bitmap
->storage
.file
== NULL
) {
285 switch (write_sb_page(bitmap
, page
, wait
)) {
287 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
291 bh
= page_buffers(page
);
293 while (bh
&& bh
->b_blocknr
) {
294 atomic_inc(&bitmap
->pending_writes
);
295 set_buffer_locked(bh
);
296 set_buffer_mapped(bh
);
297 submit_bh(WRITE
| REQ_SYNC
, bh
);
298 bh
= bh
->b_this_page
;
302 wait_event(bitmap
->write_wait
,
303 atomic_read(&bitmap
->pending_writes
)==0);
305 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
306 bitmap_file_kick(bitmap
);
309 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
311 struct bitmap
*bitmap
= bh
->b_private
;
314 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
315 if (atomic_dec_and_test(&bitmap
->pending_writes
))
316 wake_up(&bitmap
->write_wait
);
319 /* copied from buffer.c */
321 __clear_page_buffers(struct page
*page
)
323 ClearPagePrivate(page
);
324 set_page_private(page
, 0);
325 page_cache_release(page
);
327 static void free_buffers(struct page
*page
)
329 struct buffer_head
*bh
;
331 if (!PagePrivate(page
))
334 bh
= page_buffers(page
);
336 struct buffer_head
*next
= bh
->b_this_page
;
337 free_buffer_head(bh
);
340 __clear_page_buffers(page
);
344 /* read a page from a file.
345 * We both read the page, and attach buffers to the page to record the
346 * address of each block (using bmap). These addresses will be used
347 * to write the block later, completely bypassing the filesystem.
348 * This usage is similar to how swap files are handled, and allows us
349 * to write to a file with no concerns of memory allocation failing.
351 static int read_page(struct file
*file
, unsigned long index
,
352 struct bitmap
*bitmap
,
357 struct inode
*inode
= file_inode(file
);
358 struct buffer_head
*bh
;
361 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
362 (unsigned long long)index
<< PAGE_SHIFT
);
364 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
369 attach_page_buffers(page
, bh
);
370 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
375 bh
->b_blocknr
= bmap(inode
, block
);
376 if (bh
->b_blocknr
== 0) {
377 /* Cannot use this file! */
381 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
382 if (count
< (1<<inode
->i_blkbits
))
385 count
-= (1<<inode
->i_blkbits
);
387 bh
->b_end_io
= end_bitmap_write
;
388 bh
->b_private
= bitmap
;
389 atomic_inc(&bitmap
->pending_writes
);
390 set_buffer_locked(bh
);
391 set_buffer_mapped(bh
);
395 bh
= bh
->b_this_page
;
399 wait_event(bitmap
->write_wait
,
400 atomic_read(&bitmap
->pending_writes
)==0);
401 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
405 printk(KERN_ALERT
"md: bitmap read error: (%dB @ %llu): %d\n",
407 (unsigned long long)index
<< PAGE_SHIFT
,
413 * bitmap file superblock operations
416 /* update the event counter and sync the superblock to disk */
417 void bitmap_update_sb(struct bitmap
*bitmap
)
421 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
423 if (bitmap
->mddev
->bitmap_info
.external
)
425 if (!bitmap
->storage
.sb_page
) /* no superblock */
427 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
428 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
429 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
430 /* rocking back to read-only */
431 bitmap
->events_cleared
= bitmap
->mddev
->events
;
432 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
433 sb
->state
= cpu_to_le32(bitmap
->flags
);
434 /* Just in case these have been changed via sysfs: */
435 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
436 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
437 /* This might have been changed by a reshape */
438 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
439 sb
->chunksize
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.chunksize
);
440 sb
->nodes
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.nodes
);
441 sb
->sectors_reserved
= cpu_to_le32(bitmap
->mddev
->
444 write_page(bitmap
, bitmap
->storage
.sb_page
, 1);
447 /* print out the bitmap file superblock */
448 void bitmap_print_sb(struct bitmap
*bitmap
)
452 if (!bitmap
|| !bitmap
->storage
.sb_page
)
454 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
455 printk(KERN_DEBUG
"%s: bitmap file superblock:\n", bmname(bitmap
));
456 printk(KERN_DEBUG
" magic: %08x\n", le32_to_cpu(sb
->magic
));
457 printk(KERN_DEBUG
" version: %d\n", le32_to_cpu(sb
->version
));
458 printk(KERN_DEBUG
" uuid: %08x.%08x.%08x.%08x\n",
459 *(__u32
*)(sb
->uuid
+0),
460 *(__u32
*)(sb
->uuid
+4),
461 *(__u32
*)(sb
->uuid
+8),
462 *(__u32
*)(sb
->uuid
+12));
463 printk(KERN_DEBUG
" events: %llu\n",
464 (unsigned long long) le64_to_cpu(sb
->events
));
465 printk(KERN_DEBUG
"events cleared: %llu\n",
466 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
467 printk(KERN_DEBUG
" state: %08x\n", le32_to_cpu(sb
->state
));
468 printk(KERN_DEBUG
" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
469 printk(KERN_DEBUG
" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
470 printk(KERN_DEBUG
" sync size: %llu KB\n",
471 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
472 printk(KERN_DEBUG
"max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
480 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
481 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
482 * This function verifies 'bitmap_info' and populates the on-disk bitmap
483 * structure, which is to be written to disk.
485 * Returns: 0 on success, -Exxx on error
487 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
490 unsigned long chunksize
, daemon_sleep
, write_behind
;
492 bitmap
->storage
.sb_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
493 if (bitmap
->storage
.sb_page
== NULL
)
495 bitmap
->storage
.sb_page
->index
= 0;
497 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
499 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
500 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
502 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
504 if (!is_power_of_2(chunksize
)) {
506 printk(KERN_ERR
"bitmap chunksize not a power of 2\n");
509 sb
->chunksize
= cpu_to_le32(chunksize
);
511 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
512 if (!daemon_sleep
|| (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
513 printk(KERN_INFO
"Choosing daemon_sleep default (5 sec)\n");
514 daemon_sleep
= 5 * HZ
;
516 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
517 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
520 * FIXME: write_behind for RAID1. If not specified, what
521 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
523 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
524 if (write_behind
> COUNTER_MAX
)
525 write_behind
= COUNTER_MAX
/ 2;
526 sb
->write_behind
= cpu_to_le32(write_behind
);
527 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
529 /* keep the array size field of the bitmap superblock up to date */
530 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
532 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
534 set_bit(BITMAP_STALE
, &bitmap
->flags
);
535 sb
->state
= cpu_to_le32(bitmap
->flags
);
536 bitmap
->events_cleared
= bitmap
->mddev
->events
;
537 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
538 bitmap
->mddev
->bitmap_info
.nodes
= 0;
545 /* read the superblock from the bitmap file and initialize some bitmap fields */
546 static int bitmap_read_sb(struct bitmap
*bitmap
)
550 unsigned long chunksize
, daemon_sleep
, write_behind
;
551 unsigned long long events
;
553 unsigned long sectors_reserved
= 0;
555 struct page
*sb_page
;
556 loff_t offset
= bitmap
->mddev
->bitmap_info
.offset
;
558 if (!bitmap
->storage
.file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
559 chunksize
= 128 * 1024 * 1024;
560 daemon_sleep
= 5 * HZ
;
562 set_bit(BITMAP_STALE
, &bitmap
->flags
);
566 /* page 0 is the superblock, read it... */
567 sb_page
= alloc_page(GFP_KERNEL
);
570 bitmap
->storage
.sb_page
= sb_page
;
573 /* If cluster_slot is set, the cluster is setup */
574 if (bitmap
->cluster_slot
>= 0) {
575 sector_t bm_blocks
= bitmap
->mddev
->resync_max_sectors
;
577 sector_div(bm_blocks
,
578 bitmap
->mddev
->bitmap_info
.chunksize
>> 9);
580 bm_blocks
= ((bm_blocks
+7) >> 3) + sizeof(bitmap_super_t
);
582 bm_blocks
= DIV_ROUND_UP_SECTOR_T(bm_blocks
, 4096);
583 offset
= bitmap
->mddev
->bitmap_info
.offset
+ (bitmap
->cluster_slot
* (bm_blocks
<< 3));
584 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__
, __LINE__
,
585 bitmap
->cluster_slot
, offset
);
588 if (bitmap
->storage
.file
) {
589 loff_t isize
= i_size_read(bitmap
->storage
.file
->f_mapping
->host
);
590 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
592 err
= read_page(bitmap
->storage
.file
, 0,
593 bitmap
, bytes
, sb_page
);
595 err
= read_sb_page(bitmap
->mddev
,
598 0, sizeof(bitmap_super_t
));
604 sb
= kmap_atomic(sb_page
);
606 chunksize
= le32_to_cpu(sb
->chunksize
);
607 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
608 write_behind
= le32_to_cpu(sb
->write_behind
);
609 sectors_reserved
= le32_to_cpu(sb
->sectors_reserved
);
610 /* Setup nodes/clustername only if bitmap version is
613 if (sb
->version
== cpu_to_le32(BITMAP_MAJOR_CLUSTERED
)) {
614 nodes
= le32_to_cpu(sb
->nodes
);
615 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
,
616 sb
->cluster_name
, 64);
619 /* verify that the bitmap-specific fields are valid */
620 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
621 reason
= "bad magic";
622 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
623 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_CLUSTERED
)
624 reason
= "unrecognized superblock version";
625 else if (chunksize
< 512)
626 reason
= "bitmap chunksize too small";
627 else if (!is_power_of_2(chunksize
))
628 reason
= "bitmap chunksize not a power of 2";
629 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
630 reason
= "daemon sleep period out of range";
631 else if (write_behind
> COUNTER_MAX
)
632 reason
= "write-behind limit out of range (0 - 16383)";
634 printk(KERN_INFO
"%s: invalid bitmap file superblock: %s\n",
635 bmname(bitmap
), reason
);
639 /* keep the array size field of the bitmap superblock up to date */
640 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
642 if (bitmap
->mddev
->persistent
) {
644 * We have a persistent array superblock, so compare the
645 * bitmap's UUID and event counter to the mddev's
647 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
649 "%s: bitmap superblock UUID mismatch\n",
653 events
= le64_to_cpu(sb
->events
);
654 if (!nodes
&& (events
< bitmap
->mddev
->events
)) {
656 "%s: bitmap file is out of date (%llu < %llu) "
657 "-- forcing full recovery\n",
658 bmname(bitmap
), events
,
659 (unsigned long long) bitmap
->mddev
->events
);
660 set_bit(BITMAP_STALE
, &bitmap
->flags
);
664 /* assign fields using values from superblock */
665 bitmap
->flags
|= le32_to_cpu(sb
->state
);
666 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
667 set_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
);
668 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
669 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
, sb
->cluster_name
, 64);
674 /* Assiging chunksize is required for "re_read" */
675 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
676 if (err
== 0 && nodes
&& (bitmap
->cluster_slot
< 0)) {
677 err
= md_setup_cluster(bitmap
->mddev
, nodes
);
679 pr_err("%s: Could not setup cluster service (%d)\n",
680 bmname(bitmap
), err
);
683 bitmap
->cluster_slot
= md_cluster_ops
->slot_number(bitmap
->mddev
);
689 if (test_bit(BITMAP_STALE
, &bitmap
->flags
))
690 bitmap
->events_cleared
= bitmap
->mddev
->events
;
691 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
692 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
693 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
694 bitmap
->mddev
->bitmap_info
.nodes
= nodes
;
695 if (bitmap
->mddev
->bitmap_info
.space
== 0 ||
696 bitmap
->mddev
->bitmap_info
.space
> sectors_reserved
)
697 bitmap
->mddev
->bitmap_info
.space
= sectors_reserved
;
699 bitmap_print_sb(bitmap
);
700 if (bitmap
->cluster_slot
< 0)
701 md_cluster_stop(bitmap
->mddev
);
707 * general bitmap file operations
713 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
714 * file a page at a time. There's a superblock at the start of the file.
716 /* calculate the index of the page that contains this bit */
717 static inline unsigned long file_page_index(struct bitmap_storage
*store
,
721 chunk
+= sizeof(bitmap_super_t
) << 3;
722 return chunk
>> PAGE_BIT_SHIFT
;
725 /* calculate the (bit) offset of this bit within a page */
726 static inline unsigned long file_page_offset(struct bitmap_storage
*store
,
730 chunk
+= sizeof(bitmap_super_t
) << 3;
731 return chunk
& (PAGE_BITS
- 1);
735 * return a pointer to the page in the filemap that contains the given bit
738 static inline struct page
*filemap_get_page(struct bitmap_storage
*store
,
741 if (file_page_index(store
, chunk
) >= store
->file_pages
)
743 return store
->filemap
[file_page_index(store
, chunk
)];
746 static int bitmap_storage_alloc(struct bitmap_storage
*store
,
747 unsigned long chunks
, int with_super
,
750 int pnum
, offset
= 0;
751 unsigned long num_pages
;
754 bytes
= DIV_ROUND_UP(chunks
, 8);
756 bytes
+= sizeof(bitmap_super_t
);
758 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
759 offset
= slot_number
* (num_pages
- 1);
761 store
->filemap
= kmalloc(sizeof(struct page
*)
762 * num_pages
, GFP_KERNEL
);
766 if (with_super
&& !store
->sb_page
) {
767 store
->sb_page
= alloc_page(GFP_KERNEL
|__GFP_ZERO
);
768 if (store
->sb_page
== NULL
)
773 if (store
->sb_page
) {
774 store
->filemap
[0] = store
->sb_page
;
776 store
->sb_page
->index
= offset
;
779 for ( ; pnum
< num_pages
; pnum
++) {
780 store
->filemap
[pnum
] = alloc_page(GFP_KERNEL
|__GFP_ZERO
);
781 if (!store
->filemap
[pnum
]) {
782 store
->file_pages
= pnum
;
785 store
->filemap
[pnum
]->index
= pnum
+ offset
;
787 store
->file_pages
= pnum
;
789 /* We need 4 bits per page, rounded up to a multiple
790 * of sizeof(unsigned long) */
791 store
->filemap_attr
= kzalloc(
792 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
794 if (!store
->filemap_attr
)
797 store
->bytes
= bytes
;
802 static void bitmap_file_unmap(struct bitmap_storage
*store
)
804 struct page
**map
, *sb_page
;
809 map
= store
->filemap
;
810 pages
= store
->file_pages
;
811 sb_page
= store
->sb_page
;
814 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
815 free_buffers(map
[pages
]);
817 kfree(store
->filemap_attr
);
820 free_buffers(sb_page
);
823 struct inode
*inode
= file_inode(file
);
824 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
830 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
831 * then it is no longer reliable, so we stop using it and we mark the file
832 * as failed in the superblock
834 static void bitmap_file_kick(struct bitmap
*bitmap
)
836 char *path
, *ptr
= NULL
;
838 if (!test_and_set_bit(BITMAP_STALE
, &bitmap
->flags
)) {
839 bitmap_update_sb(bitmap
);
841 if (bitmap
->storage
.file
) {
842 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
844 ptr
= file_path(bitmap
->storage
.file
,
848 "%s: kicking failed bitmap file %s from array!\n",
849 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
854 "%s: disabling internal bitmap due to errors\n",
859 enum bitmap_page_attr
{
860 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
861 BITMAP_PAGE_PENDING
= 1, /* there are bits that are being cleaned.
862 * i.e. counter is 1 or 2. */
863 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
866 static inline void set_page_attr(struct bitmap
*bitmap
, int pnum
,
867 enum bitmap_page_attr attr
)
869 set_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
872 static inline void clear_page_attr(struct bitmap
*bitmap
, int pnum
,
873 enum bitmap_page_attr attr
)
875 clear_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
878 static inline int test_page_attr(struct bitmap
*bitmap
, int pnum
,
879 enum bitmap_page_attr attr
)
881 return test_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
884 static inline int test_and_clear_page_attr(struct bitmap
*bitmap
, int pnum
,
885 enum bitmap_page_attr attr
)
887 return test_and_clear_bit((pnum
<<2) + attr
,
888 bitmap
->storage
.filemap_attr
);
891 * bitmap_file_set_bit -- called before performing a write to the md device
892 * to set (and eventually sync) a particular bit in the bitmap file
894 * we set the bit immediately, then we record the page number so that
895 * when an unplug occurs, we can flush the dirty pages out to disk
897 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
902 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
904 page
= filemap_get_page(&bitmap
->storage
, chunk
);
907 bit
= file_page_offset(&bitmap
->storage
, chunk
);
910 kaddr
= kmap_atomic(page
);
911 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
914 set_bit_le(bit
, kaddr
);
915 kunmap_atomic(kaddr
);
916 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
917 /* record page number so it gets flushed to disk when unplug occurs */
918 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_DIRTY
);
921 static void bitmap_file_clear_bit(struct bitmap
*bitmap
, sector_t block
)
926 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
928 page
= filemap_get_page(&bitmap
->storage
, chunk
);
931 bit
= file_page_offset(&bitmap
->storage
, chunk
);
932 paddr
= kmap_atomic(page
);
933 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
934 clear_bit(bit
, paddr
);
936 clear_bit_le(bit
, paddr
);
937 kunmap_atomic(paddr
);
938 if (!test_page_attr(bitmap
, page
->index
, BITMAP_PAGE_NEEDWRITE
)) {
939 set_page_attr(bitmap
, page
->index
, BITMAP_PAGE_PENDING
);
940 bitmap
->allclean
= 0;
944 static int bitmap_file_test_bit(struct bitmap
*bitmap
, sector_t block
)
949 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
952 page
= filemap_get_page(&bitmap
->storage
, chunk
);
955 bit
= file_page_offset(&bitmap
->storage
, chunk
);
956 paddr
= kmap_atomic(page
);
957 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
958 set
= test_bit(bit
, paddr
);
960 set
= test_bit_le(bit
, paddr
);
961 kunmap_atomic(paddr
);
966 /* this gets called when the md device is ready to unplug its underlying
967 * (slave) device queues -- before we let any writes go down, we need to
968 * sync the dirty pages of the bitmap file to disk */
969 void bitmap_unplug(struct bitmap
*bitmap
)
972 int dirty
, need_write
;
974 if (!bitmap
|| !bitmap
->storage
.filemap
||
975 test_bit(BITMAP_STALE
, &bitmap
->flags
))
978 /* look at each page to see if there are any set bits that need to be
979 * flushed out to disk */
980 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++) {
981 if (!bitmap
->storage
.filemap
)
983 dirty
= test_and_clear_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
984 need_write
= test_and_clear_page_attr(bitmap
, i
,
985 BITMAP_PAGE_NEEDWRITE
);
986 if (dirty
|| need_write
) {
987 clear_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
);
988 write_page(bitmap
, bitmap
->storage
.filemap
[i
], 0);
991 if (bitmap
->storage
.file
)
992 wait_event(bitmap
->write_wait
,
993 atomic_read(&bitmap
->pending_writes
)==0);
995 md_super_wait(bitmap
->mddev
);
997 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
998 bitmap_file_kick(bitmap
);
1000 EXPORT_SYMBOL(bitmap_unplug
);
1002 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
1003 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1004 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1005 * memory mapping of the bitmap file
1007 * if there's no bitmap file, or if the bitmap file had been
1008 * previously kicked from the array, we mark all the bits as
1009 * 1's in order to cause a full resync.
1011 * We ignore all bits for sectors that end earlier than 'start'.
1012 * This is used when reading an out-of-date bitmap...
1014 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
1016 unsigned long i
, chunks
, index
, oldindex
, bit
, node_offset
= 0;
1017 struct page
*page
= NULL
;
1018 unsigned long bit_cnt
= 0;
1020 unsigned long offset
;
1024 struct bitmap_storage
*store
= &bitmap
->storage
;
1026 chunks
= bitmap
->counts
.chunks
;
1029 if (!file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
1030 /* No permanent bitmap - fill with '1s'. */
1031 store
->filemap
= NULL
;
1032 store
->file_pages
= 0;
1033 for (i
= 0; i
< chunks
; i
++) {
1034 /* if the disk bit is set, set the memory bit */
1035 int needed
= ((sector_t
)(i
+1) << (bitmap
->counts
.chunkshift
)
1037 bitmap_set_memory_bits(bitmap
,
1038 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1044 outofdate
= test_bit(BITMAP_STALE
, &bitmap
->flags
);
1046 printk(KERN_INFO
"%s: bitmap file is out of date, doing full "
1047 "recovery\n", bmname(bitmap
));
1049 if (file
&& i_size_read(file
->f_mapping
->host
) < store
->bytes
) {
1050 printk(KERN_INFO
"%s: bitmap file too short %lu < %lu\n",
1052 (unsigned long) i_size_read(file
->f_mapping
->host
),
1059 if (!bitmap
->mddev
->bitmap_info
.external
)
1060 offset
= sizeof(bitmap_super_t
);
1062 if (mddev_is_clustered(bitmap
->mddev
))
1063 node_offset
= bitmap
->cluster_slot
* (DIV_ROUND_UP(store
->bytes
, PAGE_SIZE
));
1065 for (i
= 0; i
< chunks
; i
++) {
1067 index
= file_page_index(&bitmap
->storage
, i
);
1068 bit
= file_page_offset(&bitmap
->storage
, i
);
1069 if (index
!= oldindex
) { /* this is a new page, read it in */
1071 /* unmap the old page, we're done with it */
1072 if (index
== store
->file_pages
-1)
1073 count
= store
->bytes
- index
* PAGE_SIZE
;
1076 page
= store
->filemap
[index
];
1078 ret
= read_page(file
, index
, bitmap
,
1083 bitmap
->mddev
->bitmap_info
.offset
,
1085 index
+ node_offset
, count
);
1094 * if bitmap is out of date, dirty the
1095 * whole page and write it out
1097 paddr
= kmap_atomic(page
);
1098 memset(paddr
+ offset
, 0xff,
1099 PAGE_SIZE
- offset
);
1100 kunmap_atomic(paddr
);
1101 write_page(bitmap
, page
, 1);
1104 if (test_bit(BITMAP_WRITE_ERROR
,
1109 paddr
= kmap_atomic(page
);
1110 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
1111 b
= test_bit(bit
, paddr
);
1113 b
= test_bit_le(bit
, paddr
);
1114 kunmap_atomic(paddr
);
1116 /* if the disk bit is set, set the memory bit */
1117 int needed
= ((sector_t
)(i
+1) << bitmap
->counts
.chunkshift
1119 bitmap_set_memory_bits(bitmap
,
1120 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1127 printk(KERN_INFO
"%s: bitmap initialized from disk: "
1128 "read %lu pages, set %lu of %lu bits\n",
1129 bmname(bitmap
), store
->file_pages
,
1135 printk(KERN_INFO
"%s: bitmap initialisation failed: %d\n",
1136 bmname(bitmap
), ret
);
1140 void bitmap_write_all(struct bitmap
*bitmap
)
1142 /* We don't actually write all bitmap blocks here,
1143 * just flag them as needing to be written
1147 if (!bitmap
|| !bitmap
->storage
.filemap
)
1149 if (bitmap
->storage
.file
)
1150 /* Only one copy, so nothing needed */
1153 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1154 set_page_attr(bitmap
, i
,
1155 BITMAP_PAGE_NEEDWRITE
);
1156 bitmap
->allclean
= 0;
1159 static void bitmap_count_page(struct bitmap_counts
*bitmap
,
1160 sector_t offset
, int inc
)
1162 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1163 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1164 bitmap
->bp
[page
].count
+= inc
;
1165 bitmap_checkfree(bitmap
, page
);
1168 static void bitmap_set_pending(struct bitmap_counts
*bitmap
, sector_t offset
)
1170 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1171 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1172 struct bitmap_page
*bp
= &bitmap
->bp
[page
];
1178 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1179 sector_t offset
, sector_t
*blocks
,
1183 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1187 void bitmap_daemon_work(struct mddev
*mddev
)
1189 struct bitmap
*bitmap
;
1191 unsigned long nextpage
;
1193 struct bitmap_counts
*counts
;
1195 /* Use a mutex to guard daemon_work against
1198 mutex_lock(&mddev
->bitmap_info
.mutex
);
1199 bitmap
= mddev
->bitmap
;
1200 if (bitmap
== NULL
) {
1201 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1204 if (time_before(jiffies
, bitmap
->daemon_lastrun
1205 + mddev
->bitmap_info
.daemon_sleep
))
1208 bitmap
->daemon_lastrun
= jiffies
;
1209 if (bitmap
->allclean
) {
1210 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1213 bitmap
->allclean
= 1;
1215 /* Any file-page which is PENDING now needs to be written.
1216 * So set NEEDWRITE now, then after we make any last-minute changes
1219 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++)
1220 if (test_and_clear_page_attr(bitmap
, j
,
1221 BITMAP_PAGE_PENDING
))
1222 set_page_attr(bitmap
, j
,
1223 BITMAP_PAGE_NEEDWRITE
);
1225 if (bitmap
->need_sync
&&
1226 mddev
->bitmap_info
.external
== 0) {
1227 /* Arrange for superblock update as well as
1230 bitmap
->need_sync
= 0;
1231 if (bitmap
->storage
.filemap
) {
1232 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
1233 sb
->events_cleared
=
1234 cpu_to_le64(bitmap
->events_cleared
);
1236 set_page_attr(bitmap
, 0,
1237 BITMAP_PAGE_NEEDWRITE
);
1240 /* Now look at the bitmap counters and if any are '2' or '1',
1241 * decrement and handle accordingly.
1243 counts
= &bitmap
->counts
;
1244 spin_lock_irq(&counts
->lock
);
1246 for (j
= 0; j
< counts
->chunks
; j
++) {
1247 bitmap_counter_t
*bmc
;
1248 sector_t block
= (sector_t
)j
<< counts
->chunkshift
;
1250 if (j
== nextpage
) {
1251 nextpage
+= PAGE_COUNTER_RATIO
;
1252 if (!counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
) {
1253 j
|= PAGE_COUNTER_MASK
;
1256 counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
= 0;
1258 bmc
= bitmap_get_counter(counts
,
1263 j
|= PAGE_COUNTER_MASK
;
1266 if (*bmc
== 1 && !bitmap
->need_sync
) {
1267 /* We can clear the bit */
1269 bitmap_count_page(counts
, block
, -1);
1270 bitmap_file_clear_bit(bitmap
, block
);
1271 } else if (*bmc
&& *bmc
<= 2) {
1273 bitmap_set_pending(counts
, block
);
1274 bitmap
->allclean
= 0;
1277 spin_unlock_irq(&counts
->lock
);
1279 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1280 * DIRTY pages need to be written by bitmap_unplug so it can wait
1282 * If we find any DIRTY page we stop there and let bitmap_unplug
1283 * handle all the rest. This is important in the case where
1284 * the first blocking holds the superblock and it has been updated.
1285 * We mustn't write any other blocks before the superblock.
1288 j
< bitmap
->storage
.file_pages
1289 && !test_bit(BITMAP_STALE
, &bitmap
->flags
);
1291 if (test_page_attr(bitmap
, j
,
1293 /* bitmap_unplug will handle the rest */
1295 if (test_and_clear_page_attr(bitmap
, j
,
1296 BITMAP_PAGE_NEEDWRITE
)) {
1297 write_page(bitmap
, bitmap
->storage
.filemap
[j
], 0);
1302 if (bitmap
->allclean
== 0)
1303 mddev
->thread
->timeout
=
1304 mddev
->bitmap_info
.daemon_sleep
;
1305 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1308 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1309 sector_t offset
, sector_t
*blocks
,
1311 __releases(bitmap
->lock
)
1312 __acquires(bitmap
->lock
)
1314 /* If 'create', we might release the lock and reclaim it.
1315 * The lock must have been taken with interrupts enabled.
1316 * If !create, we don't release the lock.
1318 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1319 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1320 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1324 err
= bitmap_checkpage(bitmap
, page
, create
);
1326 if (bitmap
->bp
[page
].hijacked
||
1327 bitmap
->bp
[page
].map
== NULL
)
1328 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1329 PAGE_COUNTER_SHIFT
- 1);
1331 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1332 *blocks
= csize
- (offset
& (csize
- 1));
1337 /* now locked ... */
1339 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1340 /* should we use the first or second counter field
1341 * of the hijacked pointer? */
1342 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1343 return &((bitmap_counter_t
*)
1344 &bitmap
->bp
[page
].map
)[hi
];
1345 } else /* page is allocated */
1346 return (bitmap_counter_t
*)
1347 &(bitmap
->bp
[page
].map
[pageoff
]);
1350 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1357 atomic_inc(&bitmap
->behind_writes
);
1358 bw
= atomic_read(&bitmap
->behind_writes
);
1359 if (bw
> bitmap
->behind_writes_used
)
1360 bitmap
->behind_writes_used
= bw
;
1362 pr_debug("inc write-behind count %d/%lu\n",
1363 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1368 bitmap_counter_t
*bmc
;
1370 spin_lock_irq(&bitmap
->counts
.lock
);
1371 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 1);
1373 spin_unlock_irq(&bitmap
->counts
.lock
);
1377 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1378 DEFINE_WAIT(__wait
);
1379 /* note that it is safe to do the prepare_to_wait
1380 * after the test as long as we do it before dropping
1383 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1384 TASK_UNINTERRUPTIBLE
);
1385 spin_unlock_irq(&bitmap
->counts
.lock
);
1387 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1393 bitmap_file_set_bit(bitmap
, offset
);
1394 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1402 spin_unlock_irq(&bitmap
->counts
.lock
);
1405 if (sectors
> blocks
)
1412 EXPORT_SYMBOL(bitmap_startwrite
);
1414 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1415 int success
, int behind
)
1420 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1421 wake_up(&bitmap
->behind_wait
);
1422 pr_debug("dec write-behind count %d/%lu\n",
1423 atomic_read(&bitmap
->behind_writes
),
1424 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1429 unsigned long flags
;
1430 bitmap_counter_t
*bmc
;
1432 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1433 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 0);
1435 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1439 if (success
&& !bitmap
->mddev
->degraded
&&
1440 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1441 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1442 bitmap
->need_sync
= 1;
1443 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1446 if (!success
&& !NEEDED(*bmc
))
1447 *bmc
|= NEEDED_MASK
;
1449 if (COUNTER(*bmc
) == COUNTER_MAX
)
1450 wake_up(&bitmap
->overflow_wait
);
1454 bitmap_set_pending(&bitmap
->counts
, offset
);
1455 bitmap
->allclean
= 0;
1457 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1459 if (sectors
> blocks
)
1465 EXPORT_SYMBOL(bitmap_endwrite
);
1467 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1470 bitmap_counter_t
*bmc
;
1472 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1474 return 1; /* always resync if no bitmap */
1476 spin_lock_irq(&bitmap
->counts
.lock
);
1477 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1483 else if (NEEDED(*bmc
)) {
1485 if (!degraded
) { /* don't set/clear bits if degraded */
1486 *bmc
|= RESYNC_MASK
;
1487 *bmc
&= ~NEEDED_MASK
;
1491 spin_unlock_irq(&bitmap
->counts
.lock
);
1495 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1498 /* bitmap_start_sync must always report on multiples of whole
1499 * pages, otherwise resync (which is very PAGE_SIZE based) will
1501 * So call __bitmap_start_sync repeatedly (if needed) until
1502 * At least PAGE_SIZE>>9 blocks are covered.
1503 * Return the 'or' of the result.
1509 while (*blocks
< (PAGE_SIZE
>>9)) {
1510 rv
|= __bitmap_start_sync(bitmap
, offset
,
1511 &blocks1
, degraded
);
1517 EXPORT_SYMBOL(bitmap_start_sync
);
1519 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1521 bitmap_counter_t
*bmc
;
1522 unsigned long flags
;
1524 if (bitmap
== NULL
) {
1528 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1529 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1534 *bmc
&= ~RESYNC_MASK
;
1536 if (!NEEDED(*bmc
) && aborted
)
1537 *bmc
|= NEEDED_MASK
;
1540 bitmap_set_pending(&bitmap
->counts
, offset
);
1541 bitmap
->allclean
= 0;
1546 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1548 EXPORT_SYMBOL(bitmap_end_sync
);
1550 void bitmap_close_sync(struct bitmap
*bitmap
)
1552 /* Sync has finished, and any bitmap chunks that weren't synced
1553 * properly have been aborted. It remains to us to clear the
1554 * RESYNC bit wherever it is still on
1556 sector_t sector
= 0;
1560 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1561 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1565 EXPORT_SYMBOL(bitmap_close_sync
);
1567 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
, bool force
)
1575 bitmap
->last_end_sync
= jiffies
;
1578 if (!force
&& time_before(jiffies
, (bitmap
->last_end_sync
1579 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1581 wait_event(bitmap
->mddev
->recovery_wait
,
1582 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1584 bitmap
->mddev
->curr_resync_completed
= sector
;
1585 set_bit(MD_CHANGE_CLEAN
, &bitmap
->mddev
->flags
);
1586 sector
&= ~((1ULL << bitmap
->counts
.chunkshift
) - 1);
1588 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1589 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1592 bitmap
->last_end_sync
= jiffies
;
1593 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1595 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1597 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1599 /* For each chunk covered by any of these sectors, set the
1600 * counter to 2 and possibly set resync_needed. They should all
1601 * be 0 at this point
1605 bitmap_counter_t
*bmc
;
1606 spin_lock_irq(&bitmap
->counts
.lock
);
1607 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &secs
, 1);
1609 spin_unlock_irq(&bitmap
->counts
.lock
);
1614 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1615 bitmap_set_pending(&bitmap
->counts
, offset
);
1616 bitmap
->allclean
= 0;
1619 *bmc
|= NEEDED_MASK
;
1620 spin_unlock_irq(&bitmap
->counts
.lock
);
1623 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1624 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1626 unsigned long chunk
;
1628 for (chunk
= s
; chunk
<= e
; chunk
++) {
1629 sector_t sec
= (sector_t
)chunk
<< bitmap
->counts
.chunkshift
;
1630 bitmap_set_memory_bits(bitmap
, sec
, 1);
1631 bitmap_file_set_bit(bitmap
, sec
);
1632 if (sec
< bitmap
->mddev
->recovery_cp
)
1633 /* We are asserting that the array is dirty,
1634 * so move the recovery_cp address back so
1635 * that it is obvious that it is dirty
1637 bitmap
->mddev
->recovery_cp
= sec
;
1642 * flush out any pending updates
1644 void bitmap_flush(struct mddev
*mddev
)
1646 struct bitmap
*bitmap
= mddev
->bitmap
;
1649 if (!bitmap
) /* there was no bitmap */
1652 /* run the daemon_work three time to ensure everything is flushed
1655 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1656 bitmap
->daemon_lastrun
-= sleep
;
1657 bitmap_daemon_work(mddev
);
1658 bitmap
->daemon_lastrun
-= sleep
;
1659 bitmap_daemon_work(mddev
);
1660 bitmap
->daemon_lastrun
-= sleep
;
1661 bitmap_daemon_work(mddev
);
1662 bitmap_update_sb(bitmap
);
1666 * free memory that was allocated
1668 static void bitmap_free(struct bitmap
*bitmap
)
1670 unsigned long k
, pages
;
1671 struct bitmap_page
*bp
;
1673 if (!bitmap
) /* there was no bitmap */
1676 if (mddev_is_clustered(bitmap
->mddev
) && bitmap
->mddev
->cluster_info
&&
1677 bitmap
->cluster_slot
== md_cluster_ops
->slot_number(bitmap
->mddev
))
1678 md_cluster_stop(bitmap
->mddev
);
1680 /* Shouldn't be needed - but just in case.... */
1681 wait_event(bitmap
->write_wait
,
1682 atomic_read(&bitmap
->pending_writes
) == 0);
1684 /* release the bitmap file */
1685 bitmap_file_unmap(&bitmap
->storage
);
1687 bp
= bitmap
->counts
.bp
;
1688 pages
= bitmap
->counts
.pages
;
1690 /* free all allocated memory */
1692 if (bp
) /* deallocate the page memory */
1693 for (k
= 0; k
< pages
; k
++)
1694 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1700 void bitmap_destroy(struct mddev
*mddev
)
1702 struct bitmap
*bitmap
= mddev
->bitmap
;
1704 if (!bitmap
) /* there was no bitmap */
1707 mutex_lock(&mddev
->bitmap_info
.mutex
);
1708 spin_lock(&mddev
->lock
);
1709 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1710 spin_unlock(&mddev
->lock
);
1711 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1713 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1715 if (bitmap
->sysfs_can_clear
)
1716 sysfs_put(bitmap
->sysfs_can_clear
);
1718 bitmap_free(bitmap
);
1722 * initialize the bitmap structure
1723 * if this returns an error, bitmap_destroy must be called to do clean up
1725 struct bitmap
*bitmap_create(struct mddev
*mddev
, int slot
)
1727 struct bitmap
*bitmap
;
1728 sector_t blocks
= mddev
->resync_max_sectors
;
1729 struct file
*file
= mddev
->bitmap_info
.file
;
1731 struct kernfs_node
*bm
= NULL
;
1733 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1735 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1737 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1739 return ERR_PTR(-ENOMEM
);
1741 spin_lock_init(&bitmap
->counts
.lock
);
1742 atomic_set(&bitmap
->pending_writes
, 0);
1743 init_waitqueue_head(&bitmap
->write_wait
);
1744 init_waitqueue_head(&bitmap
->overflow_wait
);
1745 init_waitqueue_head(&bitmap
->behind_wait
);
1747 bitmap
->mddev
= mddev
;
1748 bitmap
->cluster_slot
= slot
;
1751 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, "bitmap");
1753 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, "can_clear");
1756 bitmap
->sysfs_can_clear
= NULL
;
1758 bitmap
->storage
.file
= file
;
1761 /* As future accesses to this file will use bmap,
1762 * and bypass the page cache, we must sync the file
1767 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1768 if (!mddev
->bitmap_info
.external
) {
1770 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1771 * instructing us to create a new on-disk bitmap instance.
1773 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1774 err
= bitmap_new_disk_sb(bitmap
);
1776 err
= bitmap_read_sb(bitmap
);
1779 if (mddev
->bitmap_info
.chunksize
== 0 ||
1780 mddev
->bitmap_info
.daemon_sleep
== 0)
1781 /* chunksize and time_base need to be
1788 bitmap
->daemon_lastrun
= jiffies
;
1789 err
= bitmap_resize(bitmap
, blocks
, mddev
->bitmap_info
.chunksize
, 1);
1793 printk(KERN_INFO
"created bitmap (%lu pages) for device %s\n",
1794 bitmap
->counts
.pages
, bmname(bitmap
));
1796 err
= test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
) ? -EIO
: 0;
1802 bitmap_free(bitmap
);
1803 return ERR_PTR(err
);
1806 int bitmap_load(struct mddev
*mddev
)
1810 sector_t sector
= 0;
1811 struct bitmap
*bitmap
= mddev
->bitmap
;
1816 /* Clear out old bitmap info first: Either there is none, or we
1817 * are resuming after someone else has possibly changed things,
1818 * so we should forget old cached info.
1819 * All chunks should be clean, but some might need_sync.
1821 while (sector
< mddev
->resync_max_sectors
) {
1823 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1826 bitmap_close_sync(bitmap
);
1828 if (mddev
->degraded
== 0
1829 || bitmap
->events_cleared
== mddev
->events
)
1830 /* no need to keep dirty bits to optimise a
1831 * re-add of a missing device */
1832 start
= mddev
->recovery_cp
;
1834 mutex_lock(&mddev
->bitmap_info
.mutex
);
1835 err
= bitmap_init_from_disk(bitmap
, start
);
1836 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1840 clear_bit(BITMAP_STALE
, &bitmap
->flags
);
1842 /* Kick recovery in case any bits were set */
1843 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1845 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1846 md_wakeup_thread(mddev
->thread
);
1848 bitmap_update_sb(bitmap
);
1850 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1855 EXPORT_SYMBOL_GPL(bitmap_load
);
1857 /* Loads the bitmap associated with slot and copies the resync information
1860 int bitmap_copy_from_slot(struct mddev
*mddev
, int slot
,
1861 sector_t
*low
, sector_t
*high
, bool clear_bits
)
1864 sector_t block
, lo
= 0, hi
= 0;
1865 struct bitmap_counts
*counts
;
1866 struct bitmap
*bitmap
= bitmap_create(mddev
, slot
);
1869 return PTR_ERR(bitmap
);
1871 rv
= bitmap_init_from_disk(bitmap
, 0);
1875 counts
= &bitmap
->counts
;
1876 for (j
= 0; j
< counts
->chunks
; j
++) {
1877 block
= (sector_t
)j
<< counts
->chunkshift
;
1878 if (bitmap_file_test_bit(bitmap
, block
)) {
1882 bitmap_file_clear_bit(bitmap
, block
);
1883 bitmap_set_memory_bits(mddev
->bitmap
, block
, 1);
1884 bitmap_file_set_bit(mddev
->bitmap
, block
);
1889 bitmap_update_sb(bitmap
);
1890 /* Setting this for the ev_page should be enough.
1891 * And we do not require both write_all and PAGE_DIRT either
1893 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1894 set_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
1895 bitmap_write_all(bitmap
);
1896 bitmap_unplug(bitmap
);
1901 bitmap_free(bitmap
);
1904 EXPORT_SYMBOL_GPL(bitmap_copy_from_slot
);
1907 void bitmap_status(struct seq_file
*seq
, struct bitmap
*bitmap
)
1909 unsigned long chunk_kb
;
1910 struct bitmap_counts
*counts
;
1915 counts
= &bitmap
->counts
;
1917 chunk_kb
= bitmap
->mddev
->bitmap_info
.chunksize
>> 10;
1918 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
1920 counts
->pages
- counts
->missing_pages
,
1922 (counts
->pages
- counts
->missing_pages
)
1923 << (PAGE_SHIFT
- 10),
1924 chunk_kb
? chunk_kb
: bitmap
->mddev
->bitmap_info
.chunksize
,
1925 chunk_kb
? "KB" : "B");
1926 if (bitmap
->storage
.file
) {
1927 seq_printf(seq
, ", file: ");
1928 seq_file_path(seq
, bitmap
->storage
.file
, " \t\n");
1931 seq_printf(seq
, "\n");
1934 int bitmap_resize(struct bitmap
*bitmap
, sector_t blocks
,
1935 int chunksize
, int init
)
1937 /* If chunk_size is 0, choose an appropriate chunk size.
1938 * Then possibly allocate new storage space.
1939 * Then quiesce, copy bits, replace bitmap, and re-start
1941 * This function is called both to set up the initial bitmap
1942 * and to resize the bitmap while the array is active.
1943 * If this happens as a result of the array being resized,
1944 * chunksize will be zero, and we need to choose a suitable
1945 * chunksize, otherwise we use what we are given.
1947 struct bitmap_storage store
;
1948 struct bitmap_counts old_counts
;
1949 unsigned long chunks
;
1951 sector_t old_blocks
, new_blocks
;
1955 struct bitmap_page
*new_bp
;
1957 if (chunksize
== 0) {
1958 /* If there is enough space, leave the chunk size unchanged,
1959 * else increase by factor of two until there is enough space.
1962 long space
= bitmap
->mddev
->bitmap_info
.space
;
1965 /* We don't know how much space there is, so limit
1966 * to current size - in sectors.
1968 bytes
= DIV_ROUND_UP(bitmap
->counts
.chunks
, 8);
1969 if (!bitmap
->mddev
->bitmap_info
.external
)
1970 bytes
+= sizeof(bitmap_super_t
);
1971 space
= DIV_ROUND_UP(bytes
, 512);
1972 bitmap
->mddev
->bitmap_info
.space
= space
;
1974 chunkshift
= bitmap
->counts
.chunkshift
;
1977 /* 'chunkshift' is shift from block size to chunk size */
1979 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
1980 bytes
= DIV_ROUND_UP(chunks
, 8);
1981 if (!bitmap
->mddev
->bitmap_info
.external
)
1982 bytes
+= sizeof(bitmap_super_t
);
1983 } while (bytes
> (space
<< 9));
1985 chunkshift
= ffz(~chunksize
) - BITMAP_BLOCK_SHIFT
;
1987 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
1988 memset(&store
, 0, sizeof(store
));
1989 if (bitmap
->mddev
->bitmap_info
.offset
|| bitmap
->mddev
->bitmap_info
.file
)
1990 ret
= bitmap_storage_alloc(&store
, chunks
,
1991 !bitmap
->mddev
->bitmap_info
.external
,
1992 mddev_is_clustered(bitmap
->mddev
)
1993 ? bitmap
->cluster_slot
: 0);
1997 pages
= DIV_ROUND_UP(chunks
, PAGE_COUNTER_RATIO
);
1999 new_bp
= kzalloc(pages
* sizeof(*new_bp
), GFP_KERNEL
);
2002 bitmap_file_unmap(&store
);
2007 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 1);
2009 store
.file
= bitmap
->storage
.file
;
2010 bitmap
->storage
.file
= NULL
;
2012 if (store
.sb_page
&& bitmap
->storage
.sb_page
)
2013 memcpy(page_address(store
.sb_page
),
2014 page_address(bitmap
->storage
.sb_page
),
2015 sizeof(bitmap_super_t
));
2016 bitmap_file_unmap(&bitmap
->storage
);
2017 bitmap
->storage
= store
;
2019 old_counts
= bitmap
->counts
;
2020 bitmap
->counts
.bp
= new_bp
;
2021 bitmap
->counts
.pages
= pages
;
2022 bitmap
->counts
.missing_pages
= pages
;
2023 bitmap
->counts
.chunkshift
= chunkshift
;
2024 bitmap
->counts
.chunks
= chunks
;
2025 bitmap
->mddev
->bitmap_info
.chunksize
= 1 << (chunkshift
+
2026 BITMAP_BLOCK_SHIFT
);
2028 blocks
= min(old_counts
.chunks
<< old_counts
.chunkshift
,
2029 chunks
<< chunkshift
);
2031 spin_lock_irq(&bitmap
->counts
.lock
);
2032 for (block
= 0; block
< blocks
; ) {
2033 bitmap_counter_t
*bmc_old
, *bmc_new
;
2036 bmc_old
= bitmap_get_counter(&old_counts
, block
,
2038 set
= bmc_old
&& NEEDED(*bmc_old
);
2041 bmc_new
= bitmap_get_counter(&bitmap
->counts
, block
,
2043 if (*bmc_new
== 0) {
2044 /* need to set on-disk bits too. */
2045 sector_t end
= block
+ new_blocks
;
2046 sector_t start
= block
>> chunkshift
;
2047 start
<<= chunkshift
;
2048 while (start
< end
) {
2049 bitmap_file_set_bit(bitmap
, block
);
2050 start
+= 1 << chunkshift
;
2053 bitmap_count_page(&bitmap
->counts
,
2055 bitmap_set_pending(&bitmap
->counts
,
2058 *bmc_new
|= NEEDED_MASK
;
2059 if (new_blocks
< old_blocks
)
2060 old_blocks
= new_blocks
;
2062 block
+= old_blocks
;
2067 while (block
< (chunks
<< chunkshift
)) {
2068 bitmap_counter_t
*bmc
;
2069 bmc
= bitmap_get_counter(&bitmap
->counts
, block
,
2072 /* new space. It needs to be resynced, so
2073 * we set NEEDED_MASK.
2076 *bmc
= NEEDED_MASK
| 2;
2077 bitmap_count_page(&bitmap
->counts
,
2079 bitmap_set_pending(&bitmap
->counts
,
2083 block
+= new_blocks
;
2085 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
2086 set_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
2088 spin_unlock_irq(&bitmap
->counts
.lock
);
2091 bitmap_unplug(bitmap
);
2092 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 0);
2098 EXPORT_SYMBOL_GPL(bitmap_resize
);
2101 location_show(struct mddev
*mddev
, char *page
)
2104 if (mddev
->bitmap_info
.file
)
2105 len
= sprintf(page
, "file");
2106 else if (mddev
->bitmap_info
.offset
)
2107 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
2109 len
= sprintf(page
, "none");
2110 len
+= sprintf(page
+len
, "\n");
2115 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2119 if (!mddev
->pers
->quiesce
)
2121 if (mddev
->recovery
|| mddev
->sync_thread
)
2125 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
2126 mddev
->bitmap_info
.offset
) {
2127 /* bitmap already configured. Only option is to clear it */
2128 if (strncmp(buf
, "none", 4) != 0)
2131 mddev
->pers
->quiesce(mddev
, 1);
2132 bitmap_destroy(mddev
);
2133 mddev
->pers
->quiesce(mddev
, 0);
2135 mddev
->bitmap_info
.offset
= 0;
2136 if (mddev
->bitmap_info
.file
) {
2137 struct file
*f
= mddev
->bitmap_info
.file
;
2138 mddev
->bitmap_info
.file
= NULL
;
2142 /* No bitmap, OK to set a location */
2144 if (strncmp(buf
, "none", 4) == 0)
2145 /* nothing to be done */;
2146 else if (strncmp(buf
, "file:", 5) == 0) {
2147 /* Not supported yet */
2152 rv
= kstrtoll(buf
+1, 10, &offset
);
2154 rv
= kstrtoll(buf
, 10, &offset
);
2159 if (mddev
->bitmap_info
.external
== 0 &&
2160 mddev
->major_version
== 0 &&
2161 offset
!= mddev
->bitmap_info
.default_offset
)
2163 mddev
->bitmap_info
.offset
= offset
;
2165 struct bitmap
*bitmap
;
2166 mddev
->pers
->quiesce(mddev
, 1);
2167 bitmap
= bitmap_create(mddev
, -1);
2169 rv
= PTR_ERR(bitmap
);
2171 mddev
->bitmap
= bitmap
;
2172 rv
= bitmap_load(mddev
);
2174 bitmap_destroy(mddev
);
2175 mddev
->bitmap_info
.offset
= 0;
2178 mddev
->pers
->quiesce(mddev
, 0);
2184 if (!mddev
->external
) {
2185 /* Ensure new bitmap info is stored in
2186 * metadata promptly.
2188 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2189 md_wakeup_thread(mddev
->thread
);
2194 static struct md_sysfs_entry bitmap_location
=
2195 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
2197 /* 'bitmap/space' is the space available at 'location' for the
2198 * bitmap. This allows the kernel to know when it is safe to
2199 * resize the bitmap to match a resized array.
2202 space_show(struct mddev
*mddev
, char *page
)
2204 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.space
);
2208 space_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2210 unsigned long sectors
;
2213 rv
= kstrtoul(buf
, 10, §ors
);
2220 if (mddev
->bitmap
&&
2221 sectors
< (mddev
->bitmap
->storage
.bytes
+ 511) >> 9)
2222 return -EFBIG
; /* Bitmap is too big for this small space */
2224 /* could make sure it isn't too big, but that isn't really
2225 * needed - user-space should be careful.
2227 mddev
->bitmap_info
.space
= sectors
;
2231 static struct md_sysfs_entry bitmap_space
=
2232 __ATTR(space
, S_IRUGO
|S_IWUSR
, space_show
, space_store
);
2235 timeout_show(struct mddev
*mddev
, char *page
)
2238 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
2239 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
2241 len
= sprintf(page
, "%lu", secs
);
2243 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
2244 len
+= sprintf(page
+len
, "\n");
2249 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2251 /* timeout can be set at any time */
2252 unsigned long timeout
;
2253 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
2257 /* just to make sure we don't overflow... */
2258 if (timeout
>= LONG_MAX
/ HZ
)
2261 timeout
= timeout
* HZ
/ 10000;
2263 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
2264 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2267 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2268 if (mddev
->thread
) {
2269 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2270 * the bitmap is all clean and we don't need to
2271 * adjust the timeout right now
2273 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2274 mddev
->thread
->timeout
= timeout
;
2275 md_wakeup_thread(mddev
->thread
);
2281 static struct md_sysfs_entry bitmap_timeout
=
2282 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2285 backlog_show(struct mddev
*mddev
, char *page
)
2287 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2291 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2293 unsigned long backlog
;
2294 int rv
= kstrtoul(buf
, 10, &backlog
);
2297 if (backlog
> COUNTER_MAX
)
2299 mddev
->bitmap_info
.max_write_behind
= backlog
;
2303 static struct md_sysfs_entry bitmap_backlog
=
2304 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2307 chunksize_show(struct mddev
*mddev
, char *page
)
2309 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2313 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2315 /* Can only be changed when no bitmap is active */
2317 unsigned long csize
;
2320 rv
= kstrtoul(buf
, 10, &csize
);
2324 !is_power_of_2(csize
))
2326 mddev
->bitmap_info
.chunksize
= csize
;
2330 static struct md_sysfs_entry bitmap_chunksize
=
2331 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2333 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2335 if (mddev_is_clustered(mddev
))
2336 return sprintf(page
, "clustered\n");
2337 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2338 ? "external" : "internal"));
2341 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2343 if (mddev
->bitmap
||
2344 mddev
->bitmap_info
.file
||
2345 mddev
->bitmap_info
.offset
)
2347 if (strncmp(buf
, "external", 8) == 0)
2348 mddev
->bitmap_info
.external
= 1;
2349 else if ((strncmp(buf
, "internal", 8) == 0) ||
2350 (strncmp(buf
, "clustered", 9) == 0))
2351 mddev
->bitmap_info
.external
= 0;
2357 static struct md_sysfs_entry bitmap_metadata
=
2358 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2360 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2363 spin_lock(&mddev
->lock
);
2365 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2368 len
= sprintf(page
, "\n");
2369 spin_unlock(&mddev
->lock
);
2373 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2375 if (mddev
->bitmap
== NULL
)
2377 if (strncmp(buf
, "false", 5) == 0)
2378 mddev
->bitmap
->need_sync
= 1;
2379 else if (strncmp(buf
, "true", 4) == 0) {
2380 if (mddev
->degraded
)
2382 mddev
->bitmap
->need_sync
= 0;
2388 static struct md_sysfs_entry bitmap_can_clear
=
2389 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2392 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2395 spin_lock(&mddev
->lock
);
2396 if (mddev
->bitmap
== NULL
)
2397 ret
= sprintf(page
, "0\n");
2399 ret
= sprintf(page
, "%lu\n",
2400 mddev
->bitmap
->behind_writes_used
);
2401 spin_unlock(&mddev
->lock
);
2406 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2409 mddev
->bitmap
->behind_writes_used
= 0;
2413 static struct md_sysfs_entry max_backlog_used
=
2414 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2415 behind_writes_used_show
, behind_writes_used_reset
);
2417 static struct attribute
*md_bitmap_attrs
[] = {
2418 &bitmap_location
.attr
,
2420 &bitmap_timeout
.attr
,
2421 &bitmap_backlog
.attr
,
2422 &bitmap_chunksize
.attr
,
2423 &bitmap_metadata
.attr
,
2424 &bitmap_can_clear
.attr
,
2425 &max_backlog_used
.attr
,
2428 struct attribute_group md_bitmap_group
= {
2430 .attrs
= md_bitmap_attrs
,