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>
30 #include <trace/events/block.h>
34 static inline char *bmname(struct bitmap
*bitmap
)
36 return bitmap
->mddev
? mdname(bitmap
->mddev
) : "mdX";
40 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
42 * 1) check to see if this page is allocated, if it's not then try to alloc
43 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
44 * page pointer directly as a counter
46 * if we find our page, we increment the page's refcount so that it stays
47 * allocated while we're using it
49 static int bitmap_checkpage(struct bitmap_counts
*bitmap
,
50 unsigned long page
, int create
, int no_hijack
)
51 __releases(bitmap
->lock
)
52 __acquires(bitmap
->lock
)
54 unsigned char *mappage
;
56 if (page
>= bitmap
->pages
) {
57 /* This can happen if bitmap_start_sync goes beyond
58 * End-of-device while looking for a whole page.
64 if (bitmap
->bp
[page
].hijacked
) /* it's hijacked, don't try to alloc */
67 if (bitmap
->bp
[page
].map
) /* page is already allocated, just return */
73 /* this page has not been allocated yet */
75 spin_unlock_irq(&bitmap
->lock
);
76 /* It is possible that this is being called inside a
77 * prepare_to_wait/finish_wait loop from raid5c:make_request().
78 * In general it is not permitted to sleep in that context as it
79 * can cause the loop to spin freely.
80 * That doesn't apply here as we can only reach this point
82 * When this function completes, either bp[page].map or
83 * bp[page].hijacked. In either case, this function will
84 * abort before getting to this point again. So there is
85 * no risk of a free-spin, and so it is safe to assert
86 * that sleeping here is allowed.
88 sched_annotate_sleep();
89 mappage
= kzalloc(PAGE_SIZE
, GFP_NOIO
);
90 spin_lock_irq(&bitmap
->lock
);
92 if (mappage
== NULL
) {
93 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
94 /* We don't support hijack for cluster raid */
97 /* failed - set the hijacked flag so that we can use the
98 * pointer as a counter */
99 if (!bitmap
->bp
[page
].map
)
100 bitmap
->bp
[page
].hijacked
= 1;
101 } else if (bitmap
->bp
[page
].map
||
102 bitmap
->bp
[page
].hijacked
) {
103 /* somebody beat us to getting the page */
107 /* no page was in place and we have one, so install it */
109 bitmap
->bp
[page
].map
= mappage
;
110 bitmap
->missing_pages
--;
115 /* if page is completely empty, put it back on the free list, or dealloc it */
116 /* if page was hijacked, unmark the flag so it might get alloced next time */
117 /* Note: lock should be held when calling this */
118 static void bitmap_checkfree(struct bitmap_counts
*bitmap
, unsigned long page
)
122 if (bitmap
->bp
[page
].count
) /* page is still busy */
125 /* page is no longer in use, it can be released */
127 if (bitmap
->bp
[page
].hijacked
) { /* page was hijacked, undo this now */
128 bitmap
->bp
[page
].hijacked
= 0;
129 bitmap
->bp
[page
].map
= NULL
;
131 /* normal case, free the page */
132 ptr
= bitmap
->bp
[page
].map
;
133 bitmap
->bp
[page
].map
= NULL
;
134 bitmap
->missing_pages
++;
140 * bitmap file handling - read and write the bitmap file and its superblock
144 * basic page I/O operations
147 /* IO operations when bitmap is stored near all superblocks */
148 static int read_sb_page(struct mddev
*mddev
, loff_t offset
,
150 unsigned long index
, int size
)
152 /* choose a good rdev and read the page from there */
154 struct md_rdev
*rdev
;
157 rdev_for_each(rdev
, mddev
) {
158 if (! test_bit(In_sync
, &rdev
->flags
)
159 || test_bit(Faulty
, &rdev
->flags
))
162 target
= offset
+ index
* (PAGE_SIZE
/512);
164 if (sync_page_io(rdev
, target
,
165 roundup(size
, bdev_logical_block_size(rdev
->bdev
)),
166 page
, REQ_OP_READ
, 0, true)) {
174 static struct md_rdev
*next_active_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
176 /* Iterate the disks of an mddev, using rcu to protect access to the
177 * linked list, and raising the refcount of devices we return to ensure
178 * they don't disappear while in use.
179 * As devices are only added or removed when raid_disk is < 0 and
180 * nr_pending is 0 and In_sync is clear, the entries we return will
181 * still be in the same position on the list when we re-enter
182 * list_for_each_entry_continue_rcu.
184 * Note that if entered with 'rdev == NULL' to start at the
185 * beginning, we temporarily assign 'rdev' to an address which
186 * isn't really an rdev, but which can be used by
187 * list_for_each_entry_continue_rcu() to find the first entry.
191 /* start at the beginning */
192 rdev
= list_entry(&mddev
->disks
, struct md_rdev
, same_set
);
194 /* release the previous rdev and start from there. */
195 rdev_dec_pending(rdev
, mddev
);
197 list_for_each_entry_continue_rcu(rdev
, &mddev
->disks
, same_set
) {
198 if (rdev
->raid_disk
>= 0 &&
199 !test_bit(Faulty
, &rdev
->flags
)) {
200 /* this is a usable devices */
201 atomic_inc(&rdev
->nr_pending
);
210 static int write_sb_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
212 struct md_rdev
*rdev
;
213 struct block_device
*bdev
;
214 struct mddev
*mddev
= bitmap
->mddev
;
215 struct bitmap_storage
*store
= &bitmap
->storage
;
219 while ((rdev
= next_active_rdev(rdev
, mddev
)) != NULL
) {
220 int size
= PAGE_SIZE
;
221 loff_t offset
= mddev
->bitmap_info
.offset
;
223 bdev
= (rdev
->meta_bdev
) ? rdev
->meta_bdev
: rdev
->bdev
;
225 if (page
->index
== store
->file_pages
-1) {
226 int last_page_size
= store
->bytes
& (PAGE_SIZE
-1);
227 if (last_page_size
== 0)
228 last_page_size
= PAGE_SIZE
;
229 size
= roundup(last_page_size
,
230 bdev_logical_block_size(bdev
));
232 /* Just make sure we aren't corrupting data or
235 if (mddev
->external
) {
236 /* Bitmap could be anywhere. */
237 if (rdev
->sb_start
+ offset
+ (page
->index
241 rdev
->sb_start
+ offset
242 < (rdev
->data_offset
+ mddev
->dev_sectors
245 } else if (offset
< 0) {
246 /* DATA BITMAP METADATA */
248 + (long)(page
->index
* (PAGE_SIZE
/512))
250 /* bitmap runs in to metadata */
252 if (rdev
->data_offset
+ mddev
->dev_sectors
253 > rdev
->sb_start
+ offset
)
254 /* data runs in to bitmap */
256 } else if (rdev
->sb_start
< rdev
->data_offset
) {
257 /* METADATA BITMAP DATA */
260 + page
->index
*(PAGE_SIZE
/512) + size
/512
262 /* bitmap runs in to data */
265 /* DATA METADATA BITMAP - no problems */
267 md_super_write(mddev
, rdev
,
268 rdev
->sb_start
+ offset
269 + page
->index
* (PAGE_SIZE
/512),
274 if (wait
&& md_super_wait(mddev
) < 0)
282 static void bitmap_file_kick(struct bitmap
*bitmap
);
284 * write out a page to a file
286 static void write_page(struct bitmap
*bitmap
, struct page
*page
, int wait
)
288 struct buffer_head
*bh
;
290 if (bitmap
->storage
.file
== NULL
) {
291 switch (write_sb_page(bitmap
, page
, wait
)) {
293 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
297 bh
= page_buffers(page
);
299 while (bh
&& bh
->b_blocknr
) {
300 atomic_inc(&bitmap
->pending_writes
);
301 set_buffer_locked(bh
);
302 set_buffer_mapped(bh
);
303 submit_bh(REQ_OP_WRITE
, REQ_SYNC
, bh
);
304 bh
= bh
->b_this_page
;
308 wait_event(bitmap
->write_wait
,
309 atomic_read(&bitmap
->pending_writes
)==0);
311 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
312 bitmap_file_kick(bitmap
);
315 static void end_bitmap_write(struct buffer_head
*bh
, int uptodate
)
317 struct bitmap
*bitmap
= bh
->b_private
;
320 set_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
);
321 if (atomic_dec_and_test(&bitmap
->pending_writes
))
322 wake_up(&bitmap
->write_wait
);
325 /* copied from buffer.c */
327 __clear_page_buffers(struct page
*page
)
329 ClearPagePrivate(page
);
330 set_page_private(page
, 0);
333 static void free_buffers(struct page
*page
)
335 struct buffer_head
*bh
;
337 if (!PagePrivate(page
))
340 bh
= page_buffers(page
);
342 struct buffer_head
*next
= bh
->b_this_page
;
343 free_buffer_head(bh
);
346 __clear_page_buffers(page
);
350 /* read a page from a file.
351 * We both read the page, and attach buffers to the page to record the
352 * address of each block (using bmap). These addresses will be used
353 * to write the block later, completely bypassing the filesystem.
354 * This usage is similar to how swap files are handled, and allows us
355 * to write to a file with no concerns of memory allocation failing.
357 static int read_page(struct file
*file
, unsigned long index
,
358 struct bitmap
*bitmap
,
363 struct inode
*inode
= file_inode(file
);
364 struct buffer_head
*bh
;
367 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE
,
368 (unsigned long long)index
<< PAGE_SHIFT
);
370 bh
= alloc_page_buffers(page
, 1<<inode
->i_blkbits
, 0);
375 attach_page_buffers(page
, bh
);
376 block
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
381 bh
->b_blocknr
= bmap(inode
, block
);
382 if (bh
->b_blocknr
== 0) {
383 /* Cannot use this file! */
387 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
388 if (count
< (1<<inode
->i_blkbits
))
391 count
-= (1<<inode
->i_blkbits
);
393 bh
->b_end_io
= end_bitmap_write
;
394 bh
->b_private
= bitmap
;
395 atomic_inc(&bitmap
->pending_writes
);
396 set_buffer_locked(bh
);
397 set_buffer_mapped(bh
);
398 submit_bh(REQ_OP_READ
, 0, bh
);
401 bh
= bh
->b_this_page
;
405 wait_event(bitmap
->write_wait
,
406 atomic_read(&bitmap
->pending_writes
)==0);
407 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
411 pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
413 (unsigned long long)index
<< PAGE_SHIFT
,
419 * bitmap file superblock operations
423 * bitmap_wait_writes() should be called before writing any bitmap
424 * blocks, to ensure previous writes, particularly from
425 * bitmap_daemon_work(), have completed.
427 static void bitmap_wait_writes(struct bitmap
*bitmap
)
429 if (bitmap
->storage
.file
)
430 wait_event(bitmap
->write_wait
,
431 atomic_read(&bitmap
->pending_writes
)==0);
433 /* Note that we ignore the return value. The writes
434 * might have failed, but that would just mean that
435 * some bits which should be cleared haven't been,
436 * which is safe. The relevant bitmap blocks will
437 * probably get written again, but there is no great
438 * loss if they aren't.
440 md_super_wait(bitmap
->mddev
);
444 /* update the event counter and sync the superblock to disk */
445 void bitmap_update_sb(struct bitmap
*bitmap
)
449 if (!bitmap
|| !bitmap
->mddev
) /* no bitmap for this array */
451 if (bitmap
->mddev
->bitmap_info
.external
)
453 if (!bitmap
->storage
.sb_page
) /* no superblock */
455 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
456 sb
->events
= cpu_to_le64(bitmap
->mddev
->events
);
457 if (bitmap
->mddev
->events
< bitmap
->events_cleared
)
458 /* rocking back to read-only */
459 bitmap
->events_cleared
= bitmap
->mddev
->events
;
460 sb
->events_cleared
= cpu_to_le64(bitmap
->events_cleared
);
461 sb
->state
= cpu_to_le32(bitmap
->flags
);
462 /* Just in case these have been changed via sysfs: */
463 sb
->daemon_sleep
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.daemon_sleep
/HZ
);
464 sb
->write_behind
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.max_write_behind
);
465 /* This might have been changed by a reshape */
466 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
467 sb
->chunksize
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.chunksize
);
468 sb
->nodes
= cpu_to_le32(bitmap
->mddev
->bitmap_info
.nodes
);
469 sb
->sectors_reserved
= cpu_to_le32(bitmap
->mddev
->
472 write_page(bitmap
, bitmap
->storage
.sb_page
, 1);
475 /* print out the bitmap file superblock */
476 void bitmap_print_sb(struct bitmap
*bitmap
)
480 if (!bitmap
|| !bitmap
->storage
.sb_page
)
482 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
483 pr_debug("%s: bitmap file superblock:\n", bmname(bitmap
));
484 pr_debug(" magic: %08x\n", le32_to_cpu(sb
->magic
));
485 pr_debug(" version: %d\n", le32_to_cpu(sb
->version
));
486 pr_debug(" uuid: %08x.%08x.%08x.%08x\n",
487 *(__u32
*)(sb
->uuid
+0),
488 *(__u32
*)(sb
->uuid
+4),
489 *(__u32
*)(sb
->uuid
+8),
490 *(__u32
*)(sb
->uuid
+12));
491 pr_debug(" events: %llu\n",
492 (unsigned long long) le64_to_cpu(sb
->events
));
493 pr_debug("events cleared: %llu\n",
494 (unsigned long long) le64_to_cpu(sb
->events_cleared
));
495 pr_debug(" state: %08x\n", le32_to_cpu(sb
->state
));
496 pr_debug(" chunksize: %d B\n", le32_to_cpu(sb
->chunksize
));
497 pr_debug(" daemon sleep: %ds\n", le32_to_cpu(sb
->daemon_sleep
));
498 pr_debug(" sync size: %llu KB\n",
499 (unsigned long long)le64_to_cpu(sb
->sync_size
)/2);
500 pr_debug("max write behind: %d\n", le32_to_cpu(sb
->write_behind
));
508 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
509 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
510 * This function verifies 'bitmap_info' and populates the on-disk bitmap
511 * structure, which is to be written to disk.
513 * Returns: 0 on success, -Exxx on error
515 static int bitmap_new_disk_sb(struct bitmap
*bitmap
)
518 unsigned long chunksize
, daemon_sleep
, write_behind
;
520 bitmap
->storage
.sb_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
521 if (bitmap
->storage
.sb_page
== NULL
)
523 bitmap
->storage
.sb_page
->index
= 0;
525 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
527 sb
->magic
= cpu_to_le32(BITMAP_MAGIC
);
528 sb
->version
= cpu_to_le32(BITMAP_MAJOR_HI
);
530 chunksize
= bitmap
->mddev
->bitmap_info
.chunksize
;
532 if (!is_power_of_2(chunksize
)) {
534 pr_warn("bitmap chunksize not a power of 2\n");
537 sb
->chunksize
= cpu_to_le32(chunksize
);
539 daemon_sleep
= bitmap
->mddev
->bitmap_info
.daemon_sleep
;
540 if (!daemon_sleep
|| (daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)) {
541 pr_debug("Choosing daemon_sleep default (5 sec)\n");
542 daemon_sleep
= 5 * HZ
;
544 sb
->daemon_sleep
= cpu_to_le32(daemon_sleep
);
545 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
548 * FIXME: write_behind for RAID1. If not specified, what
549 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
551 write_behind
= bitmap
->mddev
->bitmap_info
.max_write_behind
;
552 if (write_behind
> COUNTER_MAX
)
553 write_behind
= COUNTER_MAX
/ 2;
554 sb
->write_behind
= cpu_to_le32(write_behind
);
555 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
557 /* keep the array size field of the bitmap superblock up to date */
558 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
560 memcpy(sb
->uuid
, bitmap
->mddev
->uuid
, 16);
562 set_bit(BITMAP_STALE
, &bitmap
->flags
);
563 sb
->state
= cpu_to_le32(bitmap
->flags
);
564 bitmap
->events_cleared
= bitmap
->mddev
->events
;
565 sb
->events_cleared
= cpu_to_le64(bitmap
->mddev
->events
);
566 bitmap
->mddev
->bitmap_info
.nodes
= 0;
573 /* read the superblock from the bitmap file and initialize some bitmap fields */
574 static int bitmap_read_sb(struct bitmap
*bitmap
)
578 unsigned long chunksize
, daemon_sleep
, write_behind
;
579 unsigned long long events
;
581 unsigned long sectors_reserved
= 0;
583 struct page
*sb_page
;
584 loff_t offset
= bitmap
->mddev
->bitmap_info
.offset
;
586 if (!bitmap
->storage
.file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
587 chunksize
= 128 * 1024 * 1024;
588 daemon_sleep
= 5 * HZ
;
590 set_bit(BITMAP_STALE
, &bitmap
->flags
);
594 /* page 0 is the superblock, read it... */
595 sb_page
= alloc_page(GFP_KERNEL
);
598 bitmap
->storage
.sb_page
= sb_page
;
601 /* If cluster_slot is set, the cluster is setup */
602 if (bitmap
->cluster_slot
>= 0) {
603 sector_t bm_blocks
= bitmap
->mddev
->resync_max_sectors
;
605 sector_div(bm_blocks
,
606 bitmap
->mddev
->bitmap_info
.chunksize
>> 9);
608 bm_blocks
= ((bm_blocks
+7) >> 3) + sizeof(bitmap_super_t
);
610 bm_blocks
= DIV_ROUND_UP_SECTOR_T(bm_blocks
, 4096);
611 offset
= bitmap
->mddev
->bitmap_info
.offset
+ (bitmap
->cluster_slot
* (bm_blocks
<< 3));
612 pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__
, __LINE__
,
613 bitmap
->cluster_slot
, offset
);
616 if (bitmap
->storage
.file
) {
617 loff_t isize
= i_size_read(bitmap
->storage
.file
->f_mapping
->host
);
618 int bytes
= isize
> PAGE_SIZE
? PAGE_SIZE
: isize
;
620 err
= read_page(bitmap
->storage
.file
, 0,
621 bitmap
, bytes
, sb_page
);
623 err
= read_sb_page(bitmap
->mddev
,
626 0, sizeof(bitmap_super_t
));
632 sb
= kmap_atomic(sb_page
);
634 chunksize
= le32_to_cpu(sb
->chunksize
);
635 daemon_sleep
= le32_to_cpu(sb
->daemon_sleep
) * HZ
;
636 write_behind
= le32_to_cpu(sb
->write_behind
);
637 sectors_reserved
= le32_to_cpu(sb
->sectors_reserved
);
638 /* Setup nodes/clustername only if bitmap version is
641 if (sb
->version
== cpu_to_le32(BITMAP_MAJOR_CLUSTERED
)) {
642 nodes
= le32_to_cpu(sb
->nodes
);
643 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
,
644 sb
->cluster_name
, 64);
647 /* verify that the bitmap-specific fields are valid */
648 if (sb
->magic
!= cpu_to_le32(BITMAP_MAGIC
))
649 reason
= "bad magic";
650 else if (le32_to_cpu(sb
->version
) < BITMAP_MAJOR_LO
||
651 le32_to_cpu(sb
->version
) > BITMAP_MAJOR_CLUSTERED
)
652 reason
= "unrecognized superblock version";
653 else if (chunksize
< 512)
654 reason
= "bitmap chunksize too small";
655 else if (!is_power_of_2(chunksize
))
656 reason
= "bitmap chunksize not a power of 2";
657 else if (daemon_sleep
< 1 || daemon_sleep
> MAX_SCHEDULE_TIMEOUT
)
658 reason
= "daemon sleep period out of range";
659 else if (write_behind
> COUNTER_MAX
)
660 reason
= "write-behind limit out of range (0 - 16383)";
662 pr_warn("%s: invalid bitmap file superblock: %s\n",
663 bmname(bitmap
), reason
);
667 /* keep the array size field of the bitmap superblock up to date */
668 sb
->sync_size
= cpu_to_le64(bitmap
->mddev
->resync_max_sectors
);
670 if (bitmap
->mddev
->persistent
) {
672 * We have a persistent array superblock, so compare the
673 * bitmap's UUID and event counter to the mddev's
675 if (memcmp(sb
->uuid
, bitmap
->mddev
->uuid
, 16)) {
676 pr_warn("%s: bitmap superblock UUID mismatch\n",
680 events
= le64_to_cpu(sb
->events
);
681 if (!nodes
&& (events
< bitmap
->mddev
->events
)) {
682 pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
683 bmname(bitmap
), events
,
684 (unsigned long long) bitmap
->mddev
->events
);
685 set_bit(BITMAP_STALE
, &bitmap
->flags
);
689 /* assign fields using values from superblock */
690 bitmap
->flags
|= le32_to_cpu(sb
->state
);
691 if (le32_to_cpu(sb
->version
) == BITMAP_MAJOR_HOSTENDIAN
)
692 set_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
);
693 bitmap
->events_cleared
= le64_to_cpu(sb
->events_cleared
);
694 strlcpy(bitmap
->mddev
->bitmap_info
.cluster_name
, sb
->cluster_name
, 64);
699 /* Assiging chunksize is required for "re_read" */
700 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
701 if (err
== 0 && nodes
&& (bitmap
->cluster_slot
< 0)) {
702 err
= md_setup_cluster(bitmap
->mddev
, nodes
);
704 pr_warn("%s: Could not setup cluster service (%d)\n",
705 bmname(bitmap
), err
);
708 bitmap
->cluster_slot
= md_cluster_ops
->slot_number(bitmap
->mddev
);
714 if (test_bit(BITMAP_STALE
, &bitmap
->flags
))
715 bitmap
->events_cleared
= bitmap
->mddev
->events
;
716 bitmap
->mddev
->bitmap_info
.chunksize
= chunksize
;
717 bitmap
->mddev
->bitmap_info
.daemon_sleep
= daemon_sleep
;
718 bitmap
->mddev
->bitmap_info
.max_write_behind
= write_behind
;
719 bitmap
->mddev
->bitmap_info
.nodes
= nodes
;
720 if (bitmap
->mddev
->bitmap_info
.space
== 0 ||
721 bitmap
->mddev
->bitmap_info
.space
> sectors_reserved
)
722 bitmap
->mddev
->bitmap_info
.space
= sectors_reserved
;
724 bitmap_print_sb(bitmap
);
725 if (bitmap
->cluster_slot
< 0)
726 md_cluster_stop(bitmap
->mddev
);
732 * general bitmap file operations
738 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
739 * file a page at a time. There's a superblock at the start of the file.
741 /* calculate the index of the page that contains this bit */
742 static inline unsigned long file_page_index(struct bitmap_storage
*store
,
746 chunk
+= sizeof(bitmap_super_t
) << 3;
747 return chunk
>> PAGE_BIT_SHIFT
;
750 /* calculate the (bit) offset of this bit within a page */
751 static inline unsigned long file_page_offset(struct bitmap_storage
*store
,
755 chunk
+= sizeof(bitmap_super_t
) << 3;
756 return chunk
& (PAGE_BITS
- 1);
760 * return a pointer to the page in the filemap that contains the given bit
763 static inline struct page
*filemap_get_page(struct bitmap_storage
*store
,
766 if (file_page_index(store
, chunk
) >= store
->file_pages
)
768 return store
->filemap
[file_page_index(store
, chunk
)];
771 static int bitmap_storage_alloc(struct bitmap_storage
*store
,
772 unsigned long chunks
, int with_super
,
775 int pnum
, offset
= 0;
776 unsigned long num_pages
;
779 bytes
= DIV_ROUND_UP(chunks
, 8);
781 bytes
+= sizeof(bitmap_super_t
);
783 num_pages
= DIV_ROUND_UP(bytes
, PAGE_SIZE
);
784 offset
= slot_number
* num_pages
;
786 store
->filemap
= kmalloc(sizeof(struct page
*)
787 * num_pages
, GFP_KERNEL
);
791 if (with_super
&& !store
->sb_page
) {
792 store
->sb_page
= alloc_page(GFP_KERNEL
|__GFP_ZERO
);
793 if (store
->sb_page
== NULL
)
798 if (store
->sb_page
) {
799 store
->filemap
[0] = store
->sb_page
;
801 store
->sb_page
->index
= offset
;
804 for ( ; pnum
< num_pages
; pnum
++) {
805 store
->filemap
[pnum
] = alloc_page(GFP_KERNEL
|__GFP_ZERO
);
806 if (!store
->filemap
[pnum
]) {
807 store
->file_pages
= pnum
;
810 store
->filemap
[pnum
]->index
= pnum
+ offset
;
812 store
->file_pages
= pnum
;
814 /* We need 4 bits per page, rounded up to a multiple
815 * of sizeof(unsigned long) */
816 store
->filemap_attr
= kzalloc(
817 roundup(DIV_ROUND_UP(num_pages
*4, 8), sizeof(unsigned long)),
819 if (!store
->filemap_attr
)
822 store
->bytes
= bytes
;
827 static void bitmap_file_unmap(struct bitmap_storage
*store
)
829 struct page
**map
, *sb_page
;
834 map
= store
->filemap
;
835 pages
= store
->file_pages
;
836 sb_page
= store
->sb_page
;
839 if (map
[pages
] != sb_page
) /* 0 is sb_page, release it below */
840 free_buffers(map
[pages
]);
842 kfree(store
->filemap_attr
);
845 free_buffers(sb_page
);
848 struct inode
*inode
= file_inode(file
);
849 invalidate_mapping_pages(inode
->i_mapping
, 0, -1);
855 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
856 * then it is no longer reliable, so we stop using it and we mark the file
857 * as failed in the superblock
859 static void bitmap_file_kick(struct bitmap
*bitmap
)
861 char *path
, *ptr
= NULL
;
863 if (!test_and_set_bit(BITMAP_STALE
, &bitmap
->flags
)) {
864 bitmap_update_sb(bitmap
);
866 if (bitmap
->storage
.file
) {
867 path
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
869 ptr
= file_path(bitmap
->storage
.file
,
872 pr_warn("%s: kicking failed bitmap file %s from array!\n",
873 bmname(bitmap
), IS_ERR(ptr
) ? "" : ptr
);
877 pr_warn("%s: disabling internal bitmap due to errors\n",
882 enum bitmap_page_attr
{
883 BITMAP_PAGE_DIRTY
= 0, /* there are set bits that need to be synced */
884 BITMAP_PAGE_PENDING
= 1, /* there are bits that are being cleaned.
885 * i.e. counter is 1 or 2. */
886 BITMAP_PAGE_NEEDWRITE
= 2, /* there are cleared bits that need to be synced */
889 static inline void set_page_attr(struct bitmap
*bitmap
, int pnum
,
890 enum bitmap_page_attr attr
)
892 set_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
895 static inline void clear_page_attr(struct bitmap
*bitmap
, int pnum
,
896 enum bitmap_page_attr attr
)
898 clear_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
901 static inline int test_page_attr(struct bitmap
*bitmap
, int pnum
,
902 enum bitmap_page_attr attr
)
904 return test_bit((pnum
<<2) + attr
, bitmap
->storage
.filemap_attr
);
907 static inline int test_and_clear_page_attr(struct bitmap
*bitmap
, int pnum
,
908 enum bitmap_page_attr attr
)
910 return test_and_clear_bit((pnum
<<2) + attr
,
911 bitmap
->storage
.filemap_attr
);
914 * bitmap_file_set_bit -- called before performing a write to the md device
915 * to set (and eventually sync) a particular bit in the bitmap file
917 * we set the bit immediately, then we record the page number so that
918 * when an unplug occurs, we can flush the dirty pages out to disk
920 static void bitmap_file_set_bit(struct bitmap
*bitmap
, sector_t block
)
925 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
926 struct bitmap_storage
*store
= &bitmap
->storage
;
927 unsigned long node_offset
= 0;
929 if (mddev_is_clustered(bitmap
->mddev
))
930 node_offset
= bitmap
->cluster_slot
* store
->file_pages
;
932 page
= filemap_get_page(&bitmap
->storage
, chunk
);
935 bit
= file_page_offset(&bitmap
->storage
, chunk
);
938 kaddr
= kmap_atomic(page
);
939 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
942 set_bit_le(bit
, kaddr
);
943 kunmap_atomic(kaddr
);
944 pr_debug("set file bit %lu page %lu\n", bit
, page
->index
);
945 /* record page number so it gets flushed to disk when unplug occurs */
946 set_page_attr(bitmap
, page
->index
- node_offset
, BITMAP_PAGE_DIRTY
);
949 static void bitmap_file_clear_bit(struct bitmap
*bitmap
, sector_t block
)
954 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
955 struct bitmap_storage
*store
= &bitmap
->storage
;
956 unsigned long node_offset
= 0;
958 if (mddev_is_clustered(bitmap
->mddev
))
959 node_offset
= bitmap
->cluster_slot
* store
->file_pages
;
961 page
= filemap_get_page(&bitmap
->storage
, chunk
);
964 bit
= file_page_offset(&bitmap
->storage
, chunk
);
965 paddr
= kmap_atomic(page
);
966 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
967 clear_bit(bit
, paddr
);
969 clear_bit_le(bit
, paddr
);
970 kunmap_atomic(paddr
);
971 if (!test_page_attr(bitmap
, page
->index
- node_offset
, BITMAP_PAGE_NEEDWRITE
)) {
972 set_page_attr(bitmap
, page
->index
- node_offset
, BITMAP_PAGE_PENDING
);
973 bitmap
->allclean
= 0;
977 static int bitmap_file_test_bit(struct bitmap
*bitmap
, sector_t block
)
982 unsigned long chunk
= block
>> bitmap
->counts
.chunkshift
;
985 page
= filemap_get_page(&bitmap
->storage
, chunk
);
988 bit
= file_page_offset(&bitmap
->storage
, chunk
);
989 paddr
= kmap_atomic(page
);
990 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
991 set
= test_bit(bit
, paddr
);
993 set
= test_bit_le(bit
, paddr
);
994 kunmap_atomic(paddr
);
999 /* this gets called when the md device is ready to unplug its underlying
1000 * (slave) device queues -- before we let any writes go down, we need to
1001 * sync the dirty pages of the bitmap file to disk */
1002 void bitmap_unplug(struct bitmap
*bitmap
)
1005 int dirty
, need_write
;
1008 if (!bitmap
|| !bitmap
->storage
.filemap
||
1009 test_bit(BITMAP_STALE
, &bitmap
->flags
))
1012 /* look at each page to see if there are any set bits that need to be
1013 * flushed out to disk */
1014 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++) {
1015 if (!bitmap
->storage
.filemap
)
1017 dirty
= test_and_clear_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
1018 need_write
= test_and_clear_page_attr(bitmap
, i
,
1019 BITMAP_PAGE_NEEDWRITE
);
1020 if (dirty
|| need_write
) {
1022 bitmap_wait_writes(bitmap
);
1023 if (bitmap
->mddev
->queue
)
1024 blk_add_trace_msg(bitmap
->mddev
->queue
,
1025 "md bitmap_unplug");
1027 clear_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
);
1028 write_page(bitmap
, bitmap
->storage
.filemap
[i
], 0);
1033 bitmap_wait_writes(bitmap
);
1035 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1036 bitmap_file_kick(bitmap
);
1038 EXPORT_SYMBOL(bitmap_unplug
);
1040 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
);
1041 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1042 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1043 * memory mapping of the bitmap file
1045 * if there's no bitmap file, or if the bitmap file had been
1046 * previously kicked from the array, we mark all the bits as
1047 * 1's in order to cause a full resync.
1049 * We ignore all bits for sectors that end earlier than 'start'.
1050 * This is used when reading an out-of-date bitmap...
1052 static int bitmap_init_from_disk(struct bitmap
*bitmap
, sector_t start
)
1054 unsigned long i
, chunks
, index
, oldindex
, bit
, node_offset
= 0;
1055 struct page
*page
= NULL
;
1056 unsigned long bit_cnt
= 0;
1058 unsigned long offset
;
1062 struct bitmap_storage
*store
= &bitmap
->storage
;
1064 chunks
= bitmap
->counts
.chunks
;
1067 if (!file
&& !bitmap
->mddev
->bitmap_info
.offset
) {
1068 /* No permanent bitmap - fill with '1s'. */
1069 store
->filemap
= NULL
;
1070 store
->file_pages
= 0;
1071 for (i
= 0; i
< chunks
; i
++) {
1072 /* if the disk bit is set, set the memory bit */
1073 int needed
= ((sector_t
)(i
+1) << (bitmap
->counts
.chunkshift
)
1075 bitmap_set_memory_bits(bitmap
,
1076 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1082 outofdate
= test_bit(BITMAP_STALE
, &bitmap
->flags
);
1084 pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap
));
1086 if (file
&& i_size_read(file
->f_mapping
->host
) < store
->bytes
) {
1087 pr_warn("%s: bitmap file too short %lu < %lu\n",
1089 (unsigned long) i_size_read(file
->f_mapping
->host
),
1096 if (!bitmap
->mddev
->bitmap_info
.external
)
1097 offset
= sizeof(bitmap_super_t
);
1099 if (mddev_is_clustered(bitmap
->mddev
))
1100 node_offset
= bitmap
->cluster_slot
* (DIV_ROUND_UP(store
->bytes
, PAGE_SIZE
));
1102 for (i
= 0; i
< chunks
; i
++) {
1104 index
= file_page_index(&bitmap
->storage
, i
);
1105 bit
= file_page_offset(&bitmap
->storage
, i
);
1106 if (index
!= oldindex
) { /* this is a new page, read it in */
1108 /* unmap the old page, we're done with it */
1109 if (index
== store
->file_pages
-1)
1110 count
= store
->bytes
- index
* PAGE_SIZE
;
1113 page
= store
->filemap
[index
];
1115 ret
= read_page(file
, index
, bitmap
,
1120 bitmap
->mddev
->bitmap_info
.offset
,
1122 index
+ node_offset
, count
);
1131 * if bitmap is out of date, dirty the
1132 * whole page and write it out
1134 paddr
= kmap_atomic(page
);
1135 memset(paddr
+ offset
, 0xff,
1136 PAGE_SIZE
- offset
);
1137 kunmap_atomic(paddr
);
1138 write_page(bitmap
, page
, 1);
1141 if (test_bit(BITMAP_WRITE_ERROR
,
1146 paddr
= kmap_atomic(page
);
1147 if (test_bit(BITMAP_HOSTENDIAN
, &bitmap
->flags
))
1148 b
= test_bit(bit
, paddr
);
1150 b
= test_bit_le(bit
, paddr
);
1151 kunmap_atomic(paddr
);
1153 /* if the disk bit is set, set the memory bit */
1154 int needed
= ((sector_t
)(i
+1) << bitmap
->counts
.chunkshift
1156 bitmap_set_memory_bits(bitmap
,
1157 (sector_t
)i
<< bitmap
->counts
.chunkshift
,
1164 pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1165 bmname(bitmap
), store
->file_pages
,
1171 pr_warn("%s: bitmap initialisation failed: %d\n",
1172 bmname(bitmap
), ret
);
1176 void bitmap_write_all(struct bitmap
*bitmap
)
1178 /* We don't actually write all bitmap blocks here,
1179 * just flag them as needing to be written
1183 if (!bitmap
|| !bitmap
->storage
.filemap
)
1185 if (bitmap
->storage
.file
)
1186 /* Only one copy, so nothing needed */
1189 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1190 set_page_attr(bitmap
, i
,
1191 BITMAP_PAGE_NEEDWRITE
);
1192 bitmap
->allclean
= 0;
1195 static void bitmap_count_page(struct bitmap_counts
*bitmap
,
1196 sector_t offset
, int inc
)
1198 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1199 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1200 bitmap
->bp
[page
].count
+= inc
;
1201 bitmap_checkfree(bitmap
, page
);
1204 static void bitmap_set_pending(struct bitmap_counts
*bitmap
, sector_t offset
)
1206 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1207 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1208 struct bitmap_page
*bp
= &bitmap
->bp
[page
];
1214 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1215 sector_t offset
, sector_t
*blocks
,
1219 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1223 void bitmap_daemon_work(struct mddev
*mddev
)
1225 struct bitmap
*bitmap
;
1227 unsigned long nextpage
;
1229 struct bitmap_counts
*counts
;
1231 /* Use a mutex to guard daemon_work against
1234 mutex_lock(&mddev
->bitmap_info
.mutex
);
1235 bitmap
= mddev
->bitmap
;
1236 if (bitmap
== NULL
) {
1237 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1240 if (time_before(jiffies
, bitmap
->daemon_lastrun
1241 + mddev
->bitmap_info
.daemon_sleep
))
1244 bitmap
->daemon_lastrun
= jiffies
;
1245 if (bitmap
->allclean
) {
1246 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1249 bitmap
->allclean
= 1;
1251 if (bitmap
->mddev
->queue
)
1252 blk_add_trace_msg(bitmap
->mddev
->queue
,
1253 "md bitmap_daemon_work");
1255 /* Any file-page which is PENDING now needs to be written.
1256 * So set NEEDWRITE now, then after we make any last-minute changes
1259 for (j
= 0; j
< bitmap
->storage
.file_pages
; j
++)
1260 if (test_and_clear_page_attr(bitmap
, j
,
1261 BITMAP_PAGE_PENDING
))
1262 set_page_attr(bitmap
, j
,
1263 BITMAP_PAGE_NEEDWRITE
);
1265 if (bitmap
->need_sync
&&
1266 mddev
->bitmap_info
.external
== 0) {
1267 /* Arrange for superblock update as well as
1270 bitmap
->need_sync
= 0;
1271 if (bitmap
->storage
.filemap
) {
1272 sb
= kmap_atomic(bitmap
->storage
.sb_page
);
1273 sb
->events_cleared
=
1274 cpu_to_le64(bitmap
->events_cleared
);
1276 set_page_attr(bitmap
, 0,
1277 BITMAP_PAGE_NEEDWRITE
);
1280 /* Now look at the bitmap counters and if any are '2' or '1',
1281 * decrement and handle accordingly.
1283 counts
= &bitmap
->counts
;
1284 spin_lock_irq(&counts
->lock
);
1286 for (j
= 0; j
< counts
->chunks
; j
++) {
1287 bitmap_counter_t
*bmc
;
1288 sector_t block
= (sector_t
)j
<< counts
->chunkshift
;
1290 if (j
== nextpage
) {
1291 nextpage
+= PAGE_COUNTER_RATIO
;
1292 if (!counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
) {
1293 j
|= PAGE_COUNTER_MASK
;
1296 counts
->bp
[j
>> PAGE_COUNTER_SHIFT
].pending
= 0;
1298 bmc
= bitmap_get_counter(counts
,
1303 j
|= PAGE_COUNTER_MASK
;
1306 if (*bmc
== 1 && !bitmap
->need_sync
) {
1307 /* We can clear the bit */
1309 bitmap_count_page(counts
, block
, -1);
1310 bitmap_file_clear_bit(bitmap
, block
);
1311 } else if (*bmc
&& *bmc
<= 2) {
1313 bitmap_set_pending(counts
, block
);
1314 bitmap
->allclean
= 0;
1317 spin_unlock_irq(&counts
->lock
);
1319 bitmap_wait_writes(bitmap
);
1320 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1321 * DIRTY pages need to be written by bitmap_unplug so it can wait
1323 * If we find any DIRTY page we stop there and let bitmap_unplug
1324 * handle all the rest. This is important in the case where
1325 * the first blocking holds the superblock and it has been updated.
1326 * We mustn't write any other blocks before the superblock.
1329 j
< bitmap
->storage
.file_pages
1330 && !test_bit(BITMAP_STALE
, &bitmap
->flags
);
1332 if (test_page_attr(bitmap
, j
,
1334 /* bitmap_unplug will handle the rest */
1336 if (test_and_clear_page_attr(bitmap
, j
,
1337 BITMAP_PAGE_NEEDWRITE
)) {
1338 write_page(bitmap
, bitmap
->storage
.filemap
[j
], 0);
1343 if (bitmap
->allclean
== 0)
1344 mddev
->thread
->timeout
=
1345 mddev
->bitmap_info
.daemon_sleep
;
1346 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1349 static bitmap_counter_t
*bitmap_get_counter(struct bitmap_counts
*bitmap
,
1350 sector_t offset
, sector_t
*blocks
,
1352 __releases(bitmap
->lock
)
1353 __acquires(bitmap
->lock
)
1355 /* If 'create', we might release the lock and reclaim it.
1356 * The lock must have been taken with interrupts enabled.
1357 * If !create, we don't release the lock.
1359 sector_t chunk
= offset
>> bitmap
->chunkshift
;
1360 unsigned long page
= chunk
>> PAGE_COUNTER_SHIFT
;
1361 unsigned long pageoff
= (chunk
& PAGE_COUNTER_MASK
) << COUNTER_BYTE_SHIFT
;
1365 err
= bitmap_checkpage(bitmap
, page
, create
, 0);
1367 if (bitmap
->bp
[page
].hijacked
||
1368 bitmap
->bp
[page
].map
== NULL
)
1369 csize
= ((sector_t
)1) << (bitmap
->chunkshift
+
1370 PAGE_COUNTER_SHIFT
- 1);
1372 csize
= ((sector_t
)1) << bitmap
->chunkshift
;
1373 *blocks
= csize
- (offset
& (csize
- 1));
1378 /* now locked ... */
1380 if (bitmap
->bp
[page
].hijacked
) { /* hijacked pointer */
1381 /* should we use the first or second counter field
1382 * of the hijacked pointer? */
1383 int hi
= (pageoff
> PAGE_COUNTER_MASK
);
1384 return &((bitmap_counter_t
*)
1385 &bitmap
->bp
[page
].map
)[hi
];
1386 } else /* page is allocated */
1387 return (bitmap_counter_t
*)
1388 &(bitmap
->bp
[page
].map
[pageoff
]);
1391 int bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
, int behind
)
1398 atomic_inc(&bitmap
->behind_writes
);
1399 bw
= atomic_read(&bitmap
->behind_writes
);
1400 if (bw
> bitmap
->behind_writes_used
)
1401 bitmap
->behind_writes_used
= bw
;
1403 pr_debug("inc write-behind count %d/%lu\n",
1404 bw
, bitmap
->mddev
->bitmap_info
.max_write_behind
);
1409 bitmap_counter_t
*bmc
;
1411 spin_lock_irq(&bitmap
->counts
.lock
);
1412 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 1);
1414 spin_unlock_irq(&bitmap
->counts
.lock
);
1418 if (unlikely(COUNTER(*bmc
) == COUNTER_MAX
)) {
1419 DEFINE_WAIT(__wait
);
1420 /* note that it is safe to do the prepare_to_wait
1421 * after the test as long as we do it before dropping
1424 prepare_to_wait(&bitmap
->overflow_wait
, &__wait
,
1425 TASK_UNINTERRUPTIBLE
);
1426 spin_unlock_irq(&bitmap
->counts
.lock
);
1428 finish_wait(&bitmap
->overflow_wait
, &__wait
);
1434 bitmap_file_set_bit(bitmap
, offset
);
1435 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1443 spin_unlock_irq(&bitmap
->counts
.lock
);
1446 if (sectors
> blocks
)
1453 EXPORT_SYMBOL(bitmap_startwrite
);
1455 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
1456 int success
, int behind
)
1461 if (atomic_dec_and_test(&bitmap
->behind_writes
))
1462 wake_up(&bitmap
->behind_wait
);
1463 pr_debug("dec write-behind count %d/%lu\n",
1464 atomic_read(&bitmap
->behind_writes
),
1465 bitmap
->mddev
->bitmap_info
.max_write_behind
);
1470 unsigned long flags
;
1471 bitmap_counter_t
*bmc
;
1473 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1474 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &blocks
, 0);
1476 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1480 if (success
&& !bitmap
->mddev
->degraded
&&
1481 bitmap
->events_cleared
< bitmap
->mddev
->events
) {
1482 bitmap
->events_cleared
= bitmap
->mddev
->events
;
1483 bitmap
->need_sync
= 1;
1484 sysfs_notify_dirent_safe(bitmap
->sysfs_can_clear
);
1487 if (!success
&& !NEEDED(*bmc
))
1488 *bmc
|= NEEDED_MASK
;
1490 if (COUNTER(*bmc
) == COUNTER_MAX
)
1491 wake_up(&bitmap
->overflow_wait
);
1495 bitmap_set_pending(&bitmap
->counts
, offset
);
1496 bitmap
->allclean
= 0;
1498 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1500 if (sectors
> blocks
)
1506 EXPORT_SYMBOL(bitmap_endwrite
);
1508 static int __bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1511 bitmap_counter_t
*bmc
;
1513 if (bitmap
== NULL
) {/* FIXME or bitmap set as 'failed' */
1515 return 1; /* always resync if no bitmap */
1517 spin_lock_irq(&bitmap
->counts
.lock
);
1518 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1524 else if (NEEDED(*bmc
)) {
1526 if (!degraded
) { /* don't set/clear bits if degraded */
1527 *bmc
|= RESYNC_MASK
;
1528 *bmc
&= ~NEEDED_MASK
;
1532 spin_unlock_irq(&bitmap
->counts
.lock
);
1536 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
,
1539 /* bitmap_start_sync must always report on multiples of whole
1540 * pages, otherwise resync (which is very PAGE_SIZE based) will
1542 * So call __bitmap_start_sync repeatedly (if needed) until
1543 * At least PAGE_SIZE>>9 blocks are covered.
1544 * Return the 'or' of the result.
1550 while (*blocks
< (PAGE_SIZE
>>9)) {
1551 rv
|= __bitmap_start_sync(bitmap
, offset
,
1552 &blocks1
, degraded
);
1558 EXPORT_SYMBOL(bitmap_start_sync
);
1560 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, sector_t
*blocks
, int aborted
)
1562 bitmap_counter_t
*bmc
;
1563 unsigned long flags
;
1565 if (bitmap
== NULL
) {
1569 spin_lock_irqsave(&bitmap
->counts
.lock
, flags
);
1570 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, blocks
, 0);
1575 *bmc
&= ~RESYNC_MASK
;
1577 if (!NEEDED(*bmc
) && aborted
)
1578 *bmc
|= NEEDED_MASK
;
1581 bitmap_set_pending(&bitmap
->counts
, offset
);
1582 bitmap
->allclean
= 0;
1587 spin_unlock_irqrestore(&bitmap
->counts
.lock
, flags
);
1589 EXPORT_SYMBOL(bitmap_end_sync
);
1591 void bitmap_close_sync(struct bitmap
*bitmap
)
1593 /* Sync has finished, and any bitmap chunks that weren't synced
1594 * properly have been aborted. It remains to us to clear the
1595 * RESYNC bit wherever it is still on
1597 sector_t sector
= 0;
1601 while (sector
< bitmap
->mddev
->resync_max_sectors
) {
1602 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1606 EXPORT_SYMBOL(bitmap_close_sync
);
1608 void bitmap_cond_end_sync(struct bitmap
*bitmap
, sector_t sector
, bool force
)
1616 bitmap
->last_end_sync
= jiffies
;
1619 if (!force
&& time_before(jiffies
, (bitmap
->last_end_sync
1620 + bitmap
->mddev
->bitmap_info
.daemon_sleep
)))
1622 wait_event(bitmap
->mddev
->recovery_wait
,
1623 atomic_read(&bitmap
->mddev
->recovery_active
) == 0);
1625 bitmap
->mddev
->curr_resync_completed
= sector
;
1626 set_bit(MD_SB_CHANGE_CLEAN
, &bitmap
->mddev
->sb_flags
);
1627 sector
&= ~((1ULL << bitmap
->counts
.chunkshift
) - 1);
1629 while (s
< sector
&& s
< bitmap
->mddev
->resync_max_sectors
) {
1630 bitmap_end_sync(bitmap
, s
, &blocks
, 0);
1633 bitmap
->last_end_sync
= jiffies
;
1634 sysfs_notify(&bitmap
->mddev
->kobj
, NULL
, "sync_completed");
1636 EXPORT_SYMBOL(bitmap_cond_end_sync
);
1638 void bitmap_sync_with_cluster(struct mddev
*mddev
,
1639 sector_t old_lo
, sector_t old_hi
,
1640 sector_t new_lo
, sector_t new_hi
)
1642 struct bitmap
*bitmap
= mddev
->bitmap
;
1643 sector_t sector
, blocks
= 0;
1645 for (sector
= old_lo
; sector
< new_lo
; ) {
1646 bitmap_end_sync(bitmap
, sector
, &blocks
, 0);
1649 WARN((blocks
> new_lo
) && old_lo
, "alignment is not correct for lo\n");
1651 for (sector
= old_hi
; sector
< new_hi
; ) {
1652 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1655 WARN((blocks
> new_hi
) && old_hi
, "alignment is not correct for hi\n");
1657 EXPORT_SYMBOL(bitmap_sync_with_cluster
);
1659 static void bitmap_set_memory_bits(struct bitmap
*bitmap
, sector_t offset
, int needed
)
1661 /* For each chunk covered by any of these sectors, set the
1662 * counter to 2 and possibly set resync_needed. They should all
1663 * be 0 at this point
1667 bitmap_counter_t
*bmc
;
1668 spin_lock_irq(&bitmap
->counts
.lock
);
1669 bmc
= bitmap_get_counter(&bitmap
->counts
, offset
, &secs
, 1);
1671 spin_unlock_irq(&bitmap
->counts
.lock
);
1676 bitmap_count_page(&bitmap
->counts
, offset
, 1);
1677 bitmap_set_pending(&bitmap
->counts
, offset
);
1678 bitmap
->allclean
= 0;
1681 *bmc
|= NEEDED_MASK
;
1682 spin_unlock_irq(&bitmap
->counts
.lock
);
1685 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1686 void bitmap_dirty_bits(struct bitmap
*bitmap
, unsigned long s
, unsigned long e
)
1688 unsigned long chunk
;
1690 for (chunk
= s
; chunk
<= e
; chunk
++) {
1691 sector_t sec
= (sector_t
)chunk
<< bitmap
->counts
.chunkshift
;
1692 bitmap_set_memory_bits(bitmap
, sec
, 1);
1693 bitmap_file_set_bit(bitmap
, sec
);
1694 if (sec
< bitmap
->mddev
->recovery_cp
)
1695 /* We are asserting that the array is dirty,
1696 * so move the recovery_cp address back so
1697 * that it is obvious that it is dirty
1699 bitmap
->mddev
->recovery_cp
= sec
;
1704 * flush out any pending updates
1706 void bitmap_flush(struct mddev
*mddev
)
1708 struct bitmap
*bitmap
= mddev
->bitmap
;
1711 if (!bitmap
) /* there was no bitmap */
1714 /* run the daemon_work three time to ensure everything is flushed
1717 sleep
= mddev
->bitmap_info
.daemon_sleep
* 2;
1718 bitmap
->daemon_lastrun
-= sleep
;
1719 bitmap_daemon_work(mddev
);
1720 bitmap
->daemon_lastrun
-= sleep
;
1721 bitmap_daemon_work(mddev
);
1722 bitmap
->daemon_lastrun
-= sleep
;
1723 bitmap_daemon_work(mddev
);
1724 bitmap_update_sb(bitmap
);
1728 * free memory that was allocated
1730 static void bitmap_free(struct bitmap
*bitmap
)
1732 unsigned long k
, pages
;
1733 struct bitmap_page
*bp
;
1735 if (!bitmap
) /* there was no bitmap */
1738 if (bitmap
->sysfs_can_clear
)
1739 sysfs_put(bitmap
->sysfs_can_clear
);
1741 if (mddev_is_clustered(bitmap
->mddev
) && bitmap
->mddev
->cluster_info
&&
1742 bitmap
->cluster_slot
== md_cluster_ops
->slot_number(bitmap
->mddev
))
1743 md_cluster_stop(bitmap
->mddev
);
1745 /* Shouldn't be needed - but just in case.... */
1746 wait_event(bitmap
->write_wait
,
1747 atomic_read(&bitmap
->pending_writes
) == 0);
1749 /* release the bitmap file */
1750 bitmap_file_unmap(&bitmap
->storage
);
1752 bp
= bitmap
->counts
.bp
;
1753 pages
= bitmap
->counts
.pages
;
1755 /* free all allocated memory */
1757 if (bp
) /* deallocate the page memory */
1758 for (k
= 0; k
< pages
; k
++)
1759 if (bp
[k
].map
&& !bp
[k
].hijacked
)
1765 void bitmap_destroy(struct mddev
*mddev
)
1767 struct bitmap
*bitmap
= mddev
->bitmap
;
1769 if (!bitmap
) /* there was no bitmap */
1772 mutex_lock(&mddev
->bitmap_info
.mutex
);
1773 spin_lock(&mddev
->lock
);
1774 mddev
->bitmap
= NULL
; /* disconnect from the md device */
1775 spin_unlock(&mddev
->lock
);
1776 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1778 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
1780 bitmap_free(bitmap
);
1784 * initialize the bitmap structure
1785 * if this returns an error, bitmap_destroy must be called to do clean up
1786 * once mddev->bitmap is set
1788 struct bitmap
*bitmap_create(struct mddev
*mddev
, int slot
)
1790 struct bitmap
*bitmap
;
1791 sector_t blocks
= mddev
->resync_max_sectors
;
1792 struct file
*file
= mddev
->bitmap_info
.file
;
1794 struct kernfs_node
*bm
= NULL
;
1796 BUILD_BUG_ON(sizeof(bitmap_super_t
) != 256);
1798 BUG_ON(file
&& mddev
->bitmap_info
.offset
);
1800 bitmap
= kzalloc(sizeof(*bitmap
), GFP_KERNEL
);
1802 return ERR_PTR(-ENOMEM
);
1804 spin_lock_init(&bitmap
->counts
.lock
);
1805 atomic_set(&bitmap
->pending_writes
, 0);
1806 init_waitqueue_head(&bitmap
->write_wait
);
1807 init_waitqueue_head(&bitmap
->overflow_wait
);
1808 init_waitqueue_head(&bitmap
->behind_wait
);
1810 bitmap
->mddev
= mddev
;
1811 bitmap
->cluster_slot
= slot
;
1814 bm
= sysfs_get_dirent(mddev
->kobj
.sd
, "bitmap");
1816 bitmap
->sysfs_can_clear
= sysfs_get_dirent(bm
, "can_clear");
1819 bitmap
->sysfs_can_clear
= NULL
;
1821 bitmap
->storage
.file
= file
;
1824 /* As future accesses to this file will use bmap,
1825 * and bypass the page cache, we must sync the file
1830 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1831 if (!mddev
->bitmap_info
.external
) {
1833 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1834 * instructing us to create a new on-disk bitmap instance.
1836 if (test_and_clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
))
1837 err
= bitmap_new_disk_sb(bitmap
);
1839 err
= bitmap_read_sb(bitmap
);
1842 if (mddev
->bitmap_info
.chunksize
== 0 ||
1843 mddev
->bitmap_info
.daemon_sleep
== 0)
1844 /* chunksize and time_base need to be
1851 bitmap
->daemon_lastrun
= jiffies
;
1852 err
= bitmap_resize(bitmap
, blocks
, mddev
->bitmap_info
.chunksize
, 1);
1856 pr_debug("created bitmap (%lu pages) for device %s\n",
1857 bitmap
->counts
.pages
, bmname(bitmap
));
1859 err
= test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
) ? -EIO
: 0;
1865 bitmap_free(bitmap
);
1866 return ERR_PTR(err
);
1869 int bitmap_load(struct mddev
*mddev
)
1873 sector_t sector
= 0;
1874 struct bitmap
*bitmap
= mddev
->bitmap
;
1879 if (mddev_is_clustered(mddev
))
1880 md_cluster_ops
->load_bitmaps(mddev
, mddev
->bitmap_info
.nodes
);
1882 /* Clear out old bitmap info first: Either there is none, or we
1883 * are resuming after someone else has possibly changed things,
1884 * so we should forget old cached info.
1885 * All chunks should be clean, but some might need_sync.
1887 while (sector
< mddev
->resync_max_sectors
) {
1889 bitmap_start_sync(bitmap
, sector
, &blocks
, 0);
1892 bitmap_close_sync(bitmap
);
1894 if (mddev
->degraded
== 0
1895 || bitmap
->events_cleared
== mddev
->events
)
1896 /* no need to keep dirty bits to optimise a
1897 * re-add of a missing device */
1898 start
= mddev
->recovery_cp
;
1900 mutex_lock(&mddev
->bitmap_info
.mutex
);
1901 err
= bitmap_init_from_disk(bitmap
, start
);
1902 mutex_unlock(&mddev
->bitmap_info
.mutex
);
1906 clear_bit(BITMAP_STALE
, &bitmap
->flags
);
1908 /* Kick recovery in case any bits were set */
1909 set_bit(MD_RECOVERY_NEEDED
, &bitmap
->mddev
->recovery
);
1911 mddev
->thread
->timeout
= mddev
->bitmap_info
.daemon_sleep
;
1912 md_wakeup_thread(mddev
->thread
);
1914 bitmap_update_sb(bitmap
);
1916 if (test_bit(BITMAP_WRITE_ERROR
, &bitmap
->flags
))
1921 EXPORT_SYMBOL_GPL(bitmap_load
);
1923 /* Loads the bitmap associated with slot and copies the resync information
1926 int bitmap_copy_from_slot(struct mddev
*mddev
, int slot
,
1927 sector_t
*low
, sector_t
*high
, bool clear_bits
)
1930 sector_t block
, lo
= 0, hi
= 0;
1931 struct bitmap_counts
*counts
;
1932 struct bitmap
*bitmap
= bitmap_create(mddev
, slot
);
1935 return PTR_ERR(bitmap
);
1937 rv
= bitmap_init_from_disk(bitmap
, 0);
1941 counts
= &bitmap
->counts
;
1942 for (j
= 0; j
< counts
->chunks
; j
++) {
1943 block
= (sector_t
)j
<< counts
->chunkshift
;
1944 if (bitmap_file_test_bit(bitmap
, block
)) {
1948 bitmap_file_clear_bit(bitmap
, block
);
1949 bitmap_set_memory_bits(mddev
->bitmap
, block
, 1);
1950 bitmap_file_set_bit(mddev
->bitmap
, block
);
1955 bitmap_update_sb(bitmap
);
1956 /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
1957 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
1958 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
1959 if (test_page_attr(bitmap
, i
, BITMAP_PAGE_PENDING
))
1960 set_page_attr(bitmap
, i
, BITMAP_PAGE_NEEDWRITE
);
1961 bitmap_unplug(bitmap
);
1963 bitmap_unplug(mddev
->bitmap
);
1967 bitmap_free(bitmap
);
1970 EXPORT_SYMBOL_GPL(bitmap_copy_from_slot
);
1973 void bitmap_status(struct seq_file
*seq
, struct bitmap
*bitmap
)
1975 unsigned long chunk_kb
;
1976 struct bitmap_counts
*counts
;
1981 counts
= &bitmap
->counts
;
1983 chunk_kb
= bitmap
->mddev
->bitmap_info
.chunksize
>> 10;
1984 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
1986 counts
->pages
- counts
->missing_pages
,
1988 (counts
->pages
- counts
->missing_pages
)
1989 << (PAGE_SHIFT
- 10),
1990 chunk_kb
? chunk_kb
: bitmap
->mddev
->bitmap_info
.chunksize
,
1991 chunk_kb
? "KB" : "B");
1992 if (bitmap
->storage
.file
) {
1993 seq_printf(seq
, ", file: ");
1994 seq_file_path(seq
, bitmap
->storage
.file
, " \t\n");
1997 seq_printf(seq
, "\n");
2000 int bitmap_resize(struct bitmap
*bitmap
, sector_t blocks
,
2001 int chunksize
, int init
)
2003 /* If chunk_size is 0, choose an appropriate chunk size.
2004 * Then possibly allocate new storage space.
2005 * Then quiesce, copy bits, replace bitmap, and re-start
2007 * This function is called both to set up the initial bitmap
2008 * and to resize the bitmap while the array is active.
2009 * If this happens as a result of the array being resized,
2010 * chunksize will be zero, and we need to choose a suitable
2011 * chunksize, otherwise we use what we are given.
2013 struct bitmap_storage store
;
2014 struct bitmap_counts old_counts
;
2015 unsigned long chunks
;
2017 sector_t old_blocks
, new_blocks
;
2021 struct bitmap_page
*new_bp
;
2023 if (chunksize
== 0) {
2024 /* If there is enough space, leave the chunk size unchanged,
2025 * else increase by factor of two until there is enough space.
2028 long space
= bitmap
->mddev
->bitmap_info
.space
;
2031 /* We don't know how much space there is, so limit
2032 * to current size - in sectors.
2034 bytes
= DIV_ROUND_UP(bitmap
->counts
.chunks
, 8);
2035 if (!bitmap
->mddev
->bitmap_info
.external
)
2036 bytes
+= sizeof(bitmap_super_t
);
2037 space
= DIV_ROUND_UP(bytes
, 512);
2038 bitmap
->mddev
->bitmap_info
.space
= space
;
2040 chunkshift
= bitmap
->counts
.chunkshift
;
2043 /* 'chunkshift' is shift from block size to chunk size */
2045 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
2046 bytes
= DIV_ROUND_UP(chunks
, 8);
2047 if (!bitmap
->mddev
->bitmap_info
.external
)
2048 bytes
+= sizeof(bitmap_super_t
);
2049 } while (bytes
> (space
<< 9));
2051 chunkshift
= ffz(~chunksize
) - BITMAP_BLOCK_SHIFT
;
2053 chunks
= DIV_ROUND_UP_SECTOR_T(blocks
, 1 << chunkshift
);
2054 memset(&store
, 0, sizeof(store
));
2055 if (bitmap
->mddev
->bitmap_info
.offset
|| bitmap
->mddev
->bitmap_info
.file
)
2056 ret
= bitmap_storage_alloc(&store
, chunks
,
2057 !bitmap
->mddev
->bitmap_info
.external
,
2058 mddev_is_clustered(bitmap
->mddev
)
2059 ? bitmap
->cluster_slot
: 0);
2061 bitmap_file_unmap(&store
);
2065 pages
= DIV_ROUND_UP(chunks
, PAGE_COUNTER_RATIO
);
2067 new_bp
= kzalloc(pages
* sizeof(*new_bp
), GFP_KERNEL
);
2070 bitmap_file_unmap(&store
);
2075 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 1);
2077 store
.file
= bitmap
->storage
.file
;
2078 bitmap
->storage
.file
= NULL
;
2080 if (store
.sb_page
&& bitmap
->storage
.sb_page
)
2081 memcpy(page_address(store
.sb_page
),
2082 page_address(bitmap
->storage
.sb_page
),
2083 sizeof(bitmap_super_t
));
2084 bitmap_file_unmap(&bitmap
->storage
);
2085 bitmap
->storage
= store
;
2087 old_counts
= bitmap
->counts
;
2088 bitmap
->counts
.bp
= new_bp
;
2089 bitmap
->counts
.pages
= pages
;
2090 bitmap
->counts
.missing_pages
= pages
;
2091 bitmap
->counts
.chunkshift
= chunkshift
;
2092 bitmap
->counts
.chunks
= chunks
;
2093 bitmap
->mddev
->bitmap_info
.chunksize
= 1 << (chunkshift
+
2094 BITMAP_BLOCK_SHIFT
);
2096 blocks
= min(old_counts
.chunks
<< old_counts
.chunkshift
,
2097 chunks
<< chunkshift
);
2099 spin_lock_irq(&bitmap
->counts
.lock
);
2100 /* For cluster raid, need to pre-allocate bitmap */
2101 if (mddev_is_clustered(bitmap
->mddev
)) {
2103 for (page
= 0; page
< pages
; page
++) {
2104 ret
= bitmap_checkpage(&bitmap
->counts
, page
, 1, 1);
2108 /* deallocate the page memory */
2109 for (k
= 0; k
< page
; k
++) {
2110 kfree(new_bp
[k
].map
);
2113 /* restore some fields from old_counts */
2114 bitmap
->counts
.bp
= old_counts
.bp
;
2115 bitmap
->counts
.pages
= old_counts
.pages
;
2116 bitmap
->counts
.missing_pages
= old_counts
.pages
;
2117 bitmap
->counts
.chunkshift
= old_counts
.chunkshift
;
2118 bitmap
->counts
.chunks
= old_counts
.chunks
;
2119 bitmap
->mddev
->bitmap_info
.chunksize
= 1 << (old_counts
.chunkshift
+
2120 BITMAP_BLOCK_SHIFT
);
2121 blocks
= old_counts
.chunks
<< old_counts
.chunkshift
;
2122 pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2125 bitmap
->counts
.bp
[page
].count
+= 1;
2129 for (block
= 0; block
< blocks
; ) {
2130 bitmap_counter_t
*bmc_old
, *bmc_new
;
2133 bmc_old
= bitmap_get_counter(&old_counts
, block
,
2135 set
= bmc_old
&& NEEDED(*bmc_old
);
2138 bmc_new
= bitmap_get_counter(&bitmap
->counts
, block
,
2140 if (*bmc_new
== 0) {
2141 /* need to set on-disk bits too. */
2142 sector_t end
= block
+ new_blocks
;
2143 sector_t start
= block
>> chunkshift
;
2144 start
<<= chunkshift
;
2145 while (start
< end
) {
2146 bitmap_file_set_bit(bitmap
, block
);
2147 start
+= 1 << chunkshift
;
2150 bitmap_count_page(&bitmap
->counts
,
2152 bitmap_set_pending(&bitmap
->counts
,
2155 *bmc_new
|= NEEDED_MASK
;
2156 if (new_blocks
< old_blocks
)
2157 old_blocks
= new_blocks
;
2159 block
+= old_blocks
;
2164 while (block
< (chunks
<< chunkshift
)) {
2165 bitmap_counter_t
*bmc
;
2166 bmc
= bitmap_get_counter(&bitmap
->counts
, block
,
2169 /* new space. It needs to be resynced, so
2170 * we set NEEDED_MASK.
2173 *bmc
= NEEDED_MASK
| 2;
2174 bitmap_count_page(&bitmap
->counts
,
2176 bitmap_set_pending(&bitmap
->counts
,
2180 block
+= new_blocks
;
2182 for (i
= 0; i
< bitmap
->storage
.file_pages
; i
++)
2183 set_page_attr(bitmap
, i
, BITMAP_PAGE_DIRTY
);
2185 spin_unlock_irq(&bitmap
->counts
.lock
);
2188 bitmap_unplug(bitmap
);
2189 bitmap
->mddev
->pers
->quiesce(bitmap
->mddev
, 0);
2195 EXPORT_SYMBOL_GPL(bitmap_resize
);
2198 location_show(struct mddev
*mddev
, char *page
)
2201 if (mddev
->bitmap_info
.file
)
2202 len
= sprintf(page
, "file");
2203 else if (mddev
->bitmap_info
.offset
)
2204 len
= sprintf(page
, "%+lld", (long long)mddev
->bitmap_info
.offset
);
2206 len
= sprintf(page
, "none");
2207 len
+= sprintf(page
+len
, "\n");
2212 location_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2216 rv
= mddev_lock(mddev
);
2220 if (!mddev
->pers
->quiesce
) {
2224 if (mddev
->recovery
|| mddev
->sync_thread
) {
2230 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
||
2231 mddev
->bitmap_info
.offset
) {
2232 /* bitmap already configured. Only option is to clear it */
2233 if (strncmp(buf
, "none", 4) != 0) {
2238 mddev
->pers
->quiesce(mddev
, 1);
2239 bitmap_destroy(mddev
);
2240 mddev
->pers
->quiesce(mddev
, 0);
2242 mddev
->bitmap_info
.offset
= 0;
2243 if (mddev
->bitmap_info
.file
) {
2244 struct file
*f
= mddev
->bitmap_info
.file
;
2245 mddev
->bitmap_info
.file
= NULL
;
2249 /* No bitmap, OK to set a location */
2251 if (strncmp(buf
, "none", 4) == 0)
2252 /* nothing to be done */;
2253 else if (strncmp(buf
, "file:", 5) == 0) {
2254 /* Not supported yet */
2259 rv
= kstrtoll(buf
+1, 10, &offset
);
2261 rv
= kstrtoll(buf
, 10, &offset
);
2268 if (mddev
->bitmap_info
.external
== 0 &&
2269 mddev
->major_version
== 0 &&
2270 offset
!= mddev
->bitmap_info
.default_offset
) {
2274 mddev
->bitmap_info
.offset
= offset
;
2276 struct bitmap
*bitmap
;
2277 mddev
->pers
->quiesce(mddev
, 1);
2278 bitmap
= bitmap_create(mddev
, -1);
2280 rv
= PTR_ERR(bitmap
);
2282 mddev
->bitmap
= bitmap
;
2283 rv
= bitmap_load(mddev
);
2285 mddev
->bitmap_info
.offset
= 0;
2287 mddev
->pers
->quiesce(mddev
, 0);
2289 bitmap_destroy(mddev
);
2295 if (!mddev
->external
) {
2296 /* Ensure new bitmap info is stored in
2297 * metadata promptly.
2299 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2300 md_wakeup_thread(mddev
->thread
);
2304 mddev_unlock(mddev
);
2310 static struct md_sysfs_entry bitmap_location
=
2311 __ATTR(location
, S_IRUGO
|S_IWUSR
, location_show
, location_store
);
2313 /* 'bitmap/space' is the space available at 'location' for the
2314 * bitmap. This allows the kernel to know when it is safe to
2315 * resize the bitmap to match a resized array.
2318 space_show(struct mddev
*mddev
, char *page
)
2320 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.space
);
2324 space_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2326 unsigned long sectors
;
2329 rv
= kstrtoul(buf
, 10, §ors
);
2336 if (mddev
->bitmap
&&
2337 sectors
< (mddev
->bitmap
->storage
.bytes
+ 511) >> 9)
2338 return -EFBIG
; /* Bitmap is too big for this small space */
2340 /* could make sure it isn't too big, but that isn't really
2341 * needed - user-space should be careful.
2343 mddev
->bitmap_info
.space
= sectors
;
2347 static struct md_sysfs_entry bitmap_space
=
2348 __ATTR(space
, S_IRUGO
|S_IWUSR
, space_show
, space_store
);
2351 timeout_show(struct mddev
*mddev
, char *page
)
2354 unsigned long secs
= mddev
->bitmap_info
.daemon_sleep
/ HZ
;
2355 unsigned long jifs
= mddev
->bitmap_info
.daemon_sleep
% HZ
;
2357 len
= sprintf(page
, "%lu", secs
);
2359 len
+= sprintf(page
+len
, ".%03u", jiffies_to_msecs(jifs
));
2360 len
+= sprintf(page
+len
, "\n");
2365 timeout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2367 /* timeout can be set at any time */
2368 unsigned long timeout
;
2369 int rv
= strict_strtoul_scaled(buf
, &timeout
, 4);
2373 /* just to make sure we don't overflow... */
2374 if (timeout
>= LONG_MAX
/ HZ
)
2377 timeout
= timeout
* HZ
/ 10000;
2379 if (timeout
>= MAX_SCHEDULE_TIMEOUT
)
2380 timeout
= MAX_SCHEDULE_TIMEOUT
-1;
2383 mddev
->bitmap_info
.daemon_sleep
= timeout
;
2384 if (mddev
->thread
) {
2385 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2386 * the bitmap is all clean and we don't need to
2387 * adjust the timeout right now
2389 if (mddev
->thread
->timeout
< MAX_SCHEDULE_TIMEOUT
) {
2390 mddev
->thread
->timeout
= timeout
;
2391 md_wakeup_thread(mddev
->thread
);
2397 static struct md_sysfs_entry bitmap_timeout
=
2398 __ATTR(time_base
, S_IRUGO
|S_IWUSR
, timeout_show
, timeout_store
);
2401 backlog_show(struct mddev
*mddev
, char *page
)
2403 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.max_write_behind
);
2407 backlog_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2409 unsigned long backlog
;
2410 int rv
= kstrtoul(buf
, 10, &backlog
);
2413 if (backlog
> COUNTER_MAX
)
2415 mddev
->bitmap_info
.max_write_behind
= backlog
;
2419 static struct md_sysfs_entry bitmap_backlog
=
2420 __ATTR(backlog
, S_IRUGO
|S_IWUSR
, backlog_show
, backlog_store
);
2423 chunksize_show(struct mddev
*mddev
, char *page
)
2425 return sprintf(page
, "%lu\n", mddev
->bitmap_info
.chunksize
);
2429 chunksize_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2431 /* Can only be changed when no bitmap is active */
2433 unsigned long csize
;
2436 rv
= kstrtoul(buf
, 10, &csize
);
2440 !is_power_of_2(csize
))
2442 mddev
->bitmap_info
.chunksize
= csize
;
2446 static struct md_sysfs_entry bitmap_chunksize
=
2447 __ATTR(chunksize
, S_IRUGO
|S_IWUSR
, chunksize_show
, chunksize_store
);
2449 static ssize_t
metadata_show(struct mddev
*mddev
, char *page
)
2451 if (mddev_is_clustered(mddev
))
2452 return sprintf(page
, "clustered\n");
2453 return sprintf(page
, "%s\n", (mddev
->bitmap_info
.external
2454 ? "external" : "internal"));
2457 static ssize_t
metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2459 if (mddev
->bitmap
||
2460 mddev
->bitmap_info
.file
||
2461 mddev
->bitmap_info
.offset
)
2463 if (strncmp(buf
, "external", 8) == 0)
2464 mddev
->bitmap_info
.external
= 1;
2465 else if ((strncmp(buf
, "internal", 8) == 0) ||
2466 (strncmp(buf
, "clustered", 9) == 0))
2467 mddev
->bitmap_info
.external
= 0;
2473 static struct md_sysfs_entry bitmap_metadata
=
2474 __ATTR(metadata
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2476 static ssize_t
can_clear_show(struct mddev
*mddev
, char *page
)
2479 spin_lock(&mddev
->lock
);
2481 len
= sprintf(page
, "%s\n", (mddev
->bitmap
->need_sync
?
2484 len
= sprintf(page
, "\n");
2485 spin_unlock(&mddev
->lock
);
2489 static ssize_t
can_clear_store(struct mddev
*mddev
, const char *buf
, size_t len
)
2491 if (mddev
->bitmap
== NULL
)
2493 if (strncmp(buf
, "false", 5) == 0)
2494 mddev
->bitmap
->need_sync
= 1;
2495 else if (strncmp(buf
, "true", 4) == 0) {
2496 if (mddev
->degraded
)
2498 mddev
->bitmap
->need_sync
= 0;
2504 static struct md_sysfs_entry bitmap_can_clear
=
2505 __ATTR(can_clear
, S_IRUGO
|S_IWUSR
, can_clear_show
, can_clear_store
);
2508 behind_writes_used_show(struct mddev
*mddev
, char *page
)
2511 spin_lock(&mddev
->lock
);
2512 if (mddev
->bitmap
== NULL
)
2513 ret
= sprintf(page
, "0\n");
2515 ret
= sprintf(page
, "%lu\n",
2516 mddev
->bitmap
->behind_writes_used
);
2517 spin_unlock(&mddev
->lock
);
2522 behind_writes_used_reset(struct mddev
*mddev
, const char *buf
, size_t len
)
2525 mddev
->bitmap
->behind_writes_used
= 0;
2529 static struct md_sysfs_entry max_backlog_used
=
2530 __ATTR(max_backlog_used
, S_IRUGO
| S_IWUSR
,
2531 behind_writes_used_show
, behind_writes_used_reset
);
2533 static struct attribute
*md_bitmap_attrs
[] = {
2534 &bitmap_location
.attr
,
2536 &bitmap_timeout
.attr
,
2537 &bitmap_backlog
.attr
,
2538 &bitmap_chunksize
.attr
,
2539 &bitmap_metadata
.attr
,
2540 &bitmap_can_clear
.attr
,
2541 &max_backlog_used
.attr
,
2544 struct attribute_group md_bitmap_group
= {
2546 .attrs
= md_bitmap_attrs
,