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 *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 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
76 spin_lock_irq(&bitmap->lock);
78 if (mappage == NULL) {
79 pr_debug("%s: bitmap map page allocation failed, hijacking\n",
81 /* failed - set the hijacked flag so that we can use the
82 * pointer as a counter */
83 if (!bitmap->bp[page].map)
84 bitmap->bp[page].hijacked = 1;
85 } else if (bitmap->bp[page].map ||
86 bitmap->bp[page].hijacked) {
87 /* somebody beat us to getting the page */
92 /* no page was in place and we have one, so install it */
94 bitmap->bp[page].map = mappage;
95 bitmap->missing_pages--;
100 /* if page is completely empty, put it back on the free list, or dealloc it */
101 /* if page was hijacked, unmark the flag so it might get alloced next time */
102 /* Note: lock should be held when calling this */
103 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
107 if (bitmap->bp[page].count) /* page is still busy */
110 /* page is no longer in use, it can be released */
112 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
113 bitmap->bp[page].hijacked = 0;
114 bitmap->bp[page].map = NULL;
116 /* normal case, free the page */
117 ptr = bitmap->bp[page].map;
118 bitmap->bp[page].map = NULL;
119 bitmap->missing_pages++;
125 * bitmap file handling - read and write the bitmap file and its superblock
129 * basic page I/O operations
132 /* IO operations when bitmap is stored near all superblocks */
133 static int read_sb_page(struct mddev *mddev, loff_t offset,
135 unsigned long index, int size)
137 /* choose a good rdev and read the page from there */
139 struct md_rdev *rdev;
142 rdev_for_each(rdev, mddev) {
143 if (! test_bit(In_sync, &rdev->flags)
144 || test_bit(Faulty, &rdev->flags))
147 target = offset + index * (PAGE_SIZE/512);
149 if (sync_page_io(rdev, target,
150 roundup(size, bdev_logical_block_size(rdev->bdev)),
159 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
161 /* Iterate the disks of an mddev, using rcu to protect access to the
162 * linked list, and raising the refcount of devices we return to ensure
163 * they don't disappear while in use.
164 * As devices are only added or removed when raid_disk is < 0 and
165 * nr_pending is 0 and In_sync is clear, the entries we return will
166 * still be in the same position on the list when we re-enter
167 * list_for_each_continue_rcu.
169 struct list_head *pos;
172 /* start at the beginning */
175 /* release the previous rdev and start from there. */
176 rdev_dec_pending(rdev, mddev);
177 pos = &rdev->same_set;
179 list_for_each_continue_rcu(pos, &mddev->disks) {
180 rdev = list_entry(pos, struct md_rdev, same_set);
181 if (rdev->raid_disk >= 0 &&
182 !test_bit(Faulty, &rdev->flags)) {
183 /* this is a usable devices */
184 atomic_inc(&rdev->nr_pending);
193 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
195 struct md_rdev *rdev = NULL;
196 struct block_device *bdev;
197 struct mddev *mddev = bitmap->mddev;
199 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
200 int size = PAGE_SIZE;
201 loff_t offset = mddev->bitmap_info.offset;
203 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
205 if (page->index == bitmap->file_pages-1)
206 size = roundup(bitmap->last_page_size,
207 bdev_logical_block_size(bdev));
208 /* Just make sure we aren't corrupting data or
211 if (mddev->external) {
212 /* Bitmap could be anywhere. */
213 if (rdev->sb_start + offset + (page->index
217 rdev->sb_start + offset
218 < (rdev->data_offset + mddev->dev_sectors
221 } else if (offset < 0) {
222 /* DATA BITMAP METADATA */
224 + (long)(page->index * (PAGE_SIZE/512))
226 /* bitmap runs in to metadata */
228 if (rdev->data_offset + mddev->dev_sectors
229 > rdev->sb_start + offset)
230 /* data runs in to bitmap */
232 } else if (rdev->sb_start < rdev->data_offset) {
233 /* METADATA BITMAP DATA */
236 + page->index*(PAGE_SIZE/512) + size/512
238 /* bitmap runs in to data */
241 /* DATA METADATA BITMAP - no problems */
243 md_super_write(mddev, rdev,
244 rdev->sb_start + offset
245 + page->index * (PAGE_SIZE/512),
251 md_super_wait(mddev);
258 static void bitmap_file_kick(struct bitmap *bitmap);
260 * write out a page to a file
262 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
264 struct buffer_head *bh;
266 if (bitmap->file == NULL) {
267 switch (write_sb_page(bitmap, page, wait)) {
269 bitmap->flags |= BITMAP_WRITE_ERROR;
273 bh = page_buffers(page);
275 while (bh && bh->b_blocknr) {
276 atomic_inc(&bitmap->pending_writes);
277 set_buffer_locked(bh);
278 set_buffer_mapped(bh);
279 submit_bh(WRITE | REQ_SYNC, bh);
280 bh = bh->b_this_page;
284 wait_event(bitmap->write_wait,
285 atomic_read(&bitmap->pending_writes)==0);
287 if (bitmap->flags & BITMAP_WRITE_ERROR)
288 bitmap_file_kick(bitmap);
291 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
293 struct bitmap *bitmap = bh->b_private;
297 spin_lock_irqsave(&bitmap->lock, flags);
298 bitmap->flags |= BITMAP_WRITE_ERROR;
299 spin_unlock_irqrestore(&bitmap->lock, flags);
301 if (atomic_dec_and_test(&bitmap->pending_writes))
302 wake_up(&bitmap->write_wait);
305 /* copied from buffer.c */
307 __clear_page_buffers(struct page *page)
309 ClearPagePrivate(page);
310 set_page_private(page, 0);
311 page_cache_release(page);
313 static void free_buffers(struct page *page)
315 struct buffer_head *bh;
317 if (!PagePrivate(page))
320 bh = page_buffers(page);
322 struct buffer_head *next = bh->b_this_page;
323 free_buffer_head(bh);
326 __clear_page_buffers(page);
330 /* read a page from a file.
331 * We both read the page, and attach buffers to the page to record the
332 * address of each block (using bmap). These addresses will be used
333 * to write the block later, completely bypassing the filesystem.
334 * This usage is similar to how swap files are handled, and allows us
335 * to write to a file with no concerns of memory allocation failing.
337 static int read_page(struct file *file, unsigned long index,
338 struct bitmap *bitmap,
343 struct inode *inode = file->f_path.dentry->d_inode;
344 struct buffer_head *bh;
347 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
348 (unsigned long long)index << PAGE_SHIFT);
350 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
355 attach_page_buffers(page, bh);
356 block = index << (PAGE_SHIFT - inode->i_blkbits);
361 bh->b_blocknr = bmap(inode, block);
362 if (bh->b_blocknr == 0) {
363 /* Cannot use this file! */
367 bh->b_bdev = inode->i_sb->s_bdev;
368 if (count < (1<<inode->i_blkbits))
371 count -= (1<<inode->i_blkbits);
373 bh->b_end_io = end_bitmap_write;
374 bh->b_private = bitmap;
375 atomic_inc(&bitmap->pending_writes);
376 set_buffer_locked(bh);
377 set_buffer_mapped(bh);
381 bh = bh->b_this_page;
385 wait_event(bitmap->write_wait,
386 atomic_read(&bitmap->pending_writes)==0);
387 if (bitmap->flags & BITMAP_WRITE_ERROR)
391 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
393 (unsigned long long)index << PAGE_SHIFT,
399 * bitmap file superblock operations
402 /* update the event counter and sync the superblock to disk */
403 void bitmap_update_sb(struct bitmap *bitmap)
407 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
409 if (bitmap->mddev->bitmap_info.external)
411 if (!bitmap->sb_page) /* no superblock */
413 sb = kmap_atomic(bitmap->sb_page);
414 sb->events = cpu_to_le64(bitmap->mddev->events);
415 if (bitmap->mddev->events < bitmap->events_cleared)
416 /* rocking back to read-only */
417 bitmap->events_cleared = bitmap->mddev->events;
418 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
419 sb->state = cpu_to_le32(bitmap->flags);
420 /* Just in case these have been changed via sysfs: */
421 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
422 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
424 write_page(bitmap, bitmap->sb_page, 1);
427 /* print out the bitmap file superblock */
428 void bitmap_print_sb(struct bitmap *bitmap)
432 if (!bitmap || !bitmap->sb_page)
434 sb = kmap_atomic(bitmap->sb_page);
435 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
436 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
437 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
438 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
439 *(__u32 *)(sb->uuid+0),
440 *(__u32 *)(sb->uuid+4),
441 *(__u32 *)(sb->uuid+8),
442 *(__u32 *)(sb->uuid+12));
443 printk(KERN_DEBUG " events: %llu\n",
444 (unsigned long long) le64_to_cpu(sb->events));
445 printk(KERN_DEBUG "events cleared: %llu\n",
446 (unsigned long long) le64_to_cpu(sb->events_cleared));
447 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
448 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
449 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
450 printk(KERN_DEBUG " sync size: %llu KB\n",
451 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
452 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
460 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
461 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
462 * This function verifies 'bitmap_info' and populates the on-disk bitmap
463 * structure, which is to be written to disk.
465 * Returns: 0 on success, -Exxx on error
467 static int bitmap_new_disk_sb(struct bitmap *bitmap)
470 unsigned long chunksize, daemon_sleep, write_behind;
473 bitmap->sb_page = alloc_page(GFP_KERNEL);
474 if (IS_ERR(bitmap->sb_page)) {
475 err = PTR_ERR(bitmap->sb_page);
476 bitmap->sb_page = NULL;
479 bitmap->sb_page->index = 0;
481 sb = kmap_atomic(bitmap->sb_page);
483 sb->magic = cpu_to_le32(BITMAP_MAGIC);
484 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
486 chunksize = bitmap->mddev->bitmap_info.chunksize;
488 if (!is_power_of_2(chunksize)) {
490 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
493 sb->chunksize = cpu_to_le32(chunksize);
495 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
497 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
498 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
499 daemon_sleep = 5 * HZ;
501 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
502 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
505 * FIXME: write_behind for RAID1. If not specified, what
506 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
508 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
509 if (write_behind > COUNTER_MAX)
510 write_behind = COUNTER_MAX / 2;
511 sb->write_behind = cpu_to_le32(write_behind);
512 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
514 /* keep the array size field of the bitmap superblock up to date */
515 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
517 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
519 bitmap->flags |= BITMAP_STALE;
520 sb->state |= cpu_to_le32(BITMAP_STALE);
521 bitmap->events_cleared = bitmap->mddev->events;
522 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
529 /* read the superblock from the bitmap file and initialize some bitmap fields */
530 static int bitmap_read_sb(struct bitmap *bitmap)
534 unsigned long chunksize, daemon_sleep, write_behind;
535 unsigned long long events;
537 struct page *sb_page;
539 if (!bitmap->file && !bitmap->mddev->bitmap_info.offset) {
540 chunksize = 128 * 1024 * 1024;
541 daemon_sleep = 5 * HZ;
543 bitmap->flags = BITMAP_STALE;
547 /* page 0 is the superblock, read it... */
548 sb_page = alloc_page(GFP_KERNEL);
551 bitmap->sb_page = sb_page;
554 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
555 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
557 err = read_page(bitmap->file, 0,
558 bitmap, bytes, sb_page);
560 err = read_sb_page(bitmap->mddev,
561 bitmap->mddev->bitmap_info.offset,
563 0, sizeof(bitmap_super_t));
568 sb = kmap_atomic(sb_page);
570 chunksize = le32_to_cpu(sb->chunksize);
571 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
572 write_behind = le32_to_cpu(sb->write_behind);
574 /* verify that the bitmap-specific fields are valid */
575 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
576 reason = "bad magic";
577 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
578 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
579 reason = "unrecognized superblock version";
580 else if (chunksize < 512)
581 reason = "bitmap chunksize too small";
582 else if (!is_power_of_2(chunksize))
583 reason = "bitmap chunksize not a power of 2";
584 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
585 reason = "daemon sleep period out of range";
586 else if (write_behind > COUNTER_MAX)
587 reason = "write-behind limit out of range (0 - 16383)";
589 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
590 bmname(bitmap), reason);
594 /* keep the array size field of the bitmap superblock up to date */
595 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
597 if (bitmap->mddev->persistent) {
599 * We have a persistent array superblock, so compare the
600 * bitmap's UUID and event counter to the mddev's
602 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
604 "%s: bitmap superblock UUID mismatch\n",
608 events = le64_to_cpu(sb->events);
609 if (events < bitmap->mddev->events) {
611 "%s: bitmap file is out of date (%llu < %llu) "
612 "-- forcing full recovery\n",
613 bmname(bitmap), events,
614 (unsigned long long) bitmap->mddev->events);
615 sb->state |= cpu_to_le32(BITMAP_STALE);
619 /* assign fields using values from superblock */
620 bitmap->flags |= le32_to_cpu(sb->state);
621 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
622 bitmap->flags |= BITMAP_HOSTENDIAN;
623 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
628 if (bitmap->flags & BITMAP_STALE)
629 bitmap->events_cleared = bitmap->mddev->events;
630 bitmap->mddev->bitmap_info.chunksize = chunksize;
631 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
632 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
634 bitmap_print_sb(bitmap);
638 enum bitmap_mask_op {
643 /* record the state of the bitmap in the superblock. Return the old value */
644 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
645 enum bitmap_mask_op op)
650 if (!bitmap->sb_page) /* can't set the state */
652 sb = kmap_atomic(bitmap->sb_page);
653 old = le32_to_cpu(sb->state) & bits;
656 sb->state |= cpu_to_le32(bits);
657 bitmap->flags |= bits;
660 sb->state &= cpu_to_le32(~bits);
661 bitmap->flags &= ~bits;
671 * general bitmap file operations
677 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
678 * file a page at a time. There's a superblock at the start of the file.
680 /* calculate the index of the page that contains this bit */
681 static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
683 if (!bitmap->mddev->bitmap_info.external)
684 chunk += sizeof(bitmap_super_t) << 3;
685 return chunk >> PAGE_BIT_SHIFT;
688 /* calculate the (bit) offset of this bit within a page */
689 static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
691 if (!bitmap->mddev->bitmap_info.external)
692 chunk += sizeof(bitmap_super_t) << 3;
693 return chunk & (PAGE_BITS - 1);
697 * return a pointer to the page in the filemap that contains the given bit
699 * this lookup is complicated by the fact that the bitmap sb might be exactly
700 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
703 static inline struct page *filemap_get_page(struct bitmap *bitmap,
706 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
708 return bitmap->filemap[file_page_index(bitmap, chunk)
709 - file_page_index(bitmap, 0)];
712 static void bitmap_file_unmap(struct bitmap *bitmap)
714 struct page **map, *sb_page;
719 spin_lock_irqsave(&bitmap->lock, flags);
720 map = bitmap->filemap;
721 bitmap->filemap = NULL;
722 attr = bitmap->filemap_attr;
723 bitmap->filemap_attr = NULL;
724 pages = bitmap->file_pages;
725 bitmap->file_pages = 0;
726 sb_page = bitmap->sb_page;
727 bitmap->sb_page = NULL;
728 spin_unlock_irqrestore(&bitmap->lock, flags);
731 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
732 free_buffers(map[pages]);
737 free_buffers(sb_page);
740 static void bitmap_file_put(struct bitmap *bitmap)
745 spin_lock_irqsave(&bitmap->lock, flags);
748 spin_unlock_irqrestore(&bitmap->lock, flags);
751 wait_event(bitmap->write_wait,
752 atomic_read(&bitmap->pending_writes)==0);
753 bitmap_file_unmap(bitmap);
756 struct inode *inode = file->f_path.dentry->d_inode;
757 invalidate_mapping_pages(inode->i_mapping, 0, -1);
763 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
764 * then it is no longer reliable, so we stop using it and we mark the file
765 * as failed in the superblock
767 static void bitmap_file_kick(struct bitmap *bitmap)
769 char *path, *ptr = NULL;
771 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
772 bitmap_update_sb(bitmap);
775 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
777 ptr = d_path(&bitmap->file->f_path, path,
781 "%s: kicking failed bitmap file %s from array!\n",
782 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
787 "%s: disabling internal bitmap due to errors\n",
791 bitmap_file_put(bitmap);
796 enum bitmap_page_attr {
797 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
798 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
799 * i.e. counter is 1 or 2. */
800 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
803 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
804 enum bitmap_page_attr attr)
806 __set_bit((pnum<<2) + attr, bitmap->filemap_attr);
809 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
810 enum bitmap_page_attr attr)
812 __clear_bit((pnum<<2) + attr, bitmap->filemap_attr);
815 static inline unsigned long test_page_attr(struct bitmap *bitmap, int pnum,
816 enum bitmap_page_attr attr)
818 return test_bit((pnum<<2) + attr, bitmap->filemap_attr);
822 * bitmap_file_set_bit -- called before performing a write to the md device
823 * to set (and eventually sync) a particular bit in the bitmap file
825 * we set the bit immediately, then we record the page number so that
826 * when an unplug occurs, we can flush the dirty pages out to disk
828 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
833 unsigned long chunk = block >> bitmap->chunkshift;
835 page = filemap_get_page(bitmap, chunk);
838 bit = file_page_offset(bitmap, chunk);
841 kaddr = kmap_atomic(page);
842 if (bitmap->flags & BITMAP_HOSTENDIAN)
845 __set_bit_le(bit, kaddr);
846 kunmap_atomic(kaddr);
847 pr_debug("set file bit %lu page %lu\n", bit, page->index);
848 /* record page number so it gets flushed to disk when unplug occurs */
849 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
852 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
857 unsigned long chunk = block >> bitmap->chunkshift;
859 page = filemap_get_page(bitmap, chunk);
862 bit = file_page_offset(bitmap, chunk);
863 paddr = kmap_atomic(page);
864 if (bitmap->flags & BITMAP_HOSTENDIAN)
865 clear_bit(bit, paddr);
867 __clear_bit_le(bit, paddr);
868 kunmap_atomic(paddr);
869 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
870 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
871 bitmap->allclean = 0;
875 /* this gets called when the md device is ready to unplug its underlying
876 * (slave) device queues -- before we let any writes go down, we need to
877 * sync the dirty pages of the bitmap file to disk */
878 void bitmap_unplug(struct bitmap *bitmap)
880 unsigned long i, flags;
881 int dirty, need_write;
884 if (!bitmap || !bitmap->filemap)
887 /* look at each page to see if there are any set bits that need to be
888 * flushed out to disk */
889 for (i = 0; i < bitmap->file_pages; i++) {
890 spin_lock_irqsave(&bitmap->lock, flags);
891 if (!bitmap->filemap) {
892 spin_unlock_irqrestore(&bitmap->lock, flags);
895 dirty = test_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
896 need_write = test_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
897 clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
898 clear_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
899 if (dirty || need_write)
900 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
903 spin_unlock_irqrestore(&bitmap->lock, flags);
905 if (dirty || need_write)
906 write_page(bitmap, bitmap->filemap[i], 0);
908 if (wait) { /* if any writes were performed, we need to wait on them */
910 wait_event(bitmap->write_wait,
911 atomic_read(&bitmap->pending_writes)==0);
913 md_super_wait(bitmap->mddev);
915 if (bitmap->flags & BITMAP_WRITE_ERROR)
916 bitmap_file_kick(bitmap);
918 EXPORT_SYMBOL(bitmap_unplug);
920 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
921 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
922 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
923 * memory mapping of the bitmap file
925 * if there's no bitmap file, or if the bitmap file had been
926 * previously kicked from the array, we mark all the bits as
927 * 1's in order to cause a full resync.
929 * We ignore all bits for sectors that end earlier than 'start'.
930 * This is used when reading an out-of-date bitmap...
932 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
934 unsigned long i, chunks, index, oldindex, bit;
936 struct page *page = NULL;
937 unsigned long num_pages, bit_cnt = 0;
939 unsigned long bytes, offset;
944 chunks = bitmap->chunks;
947 if (!file && !bitmap->mddev->bitmap_info.offset) {
948 /* No permanent bitmap - fill with '1s'. */
949 bitmap->filemap = NULL;
950 bitmap->file_pages = 0;
951 for (i = 0; i < chunks ; i++) {
952 /* if the disk bit is set, set the memory bit */
953 int needed = ((sector_t)(i+1) << (bitmap->chunkshift)
955 bitmap_set_memory_bits(bitmap,
956 (sector_t)i << bitmap->chunkshift,
962 outofdate = bitmap->flags & BITMAP_STALE;
964 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
965 "recovery\n", bmname(bitmap));
967 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
968 if (!bitmap->mddev->bitmap_info.external)
969 bytes += sizeof(bitmap_super_t);
971 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
973 if (file && i_size_read(file->f_mapping->host) < bytes) {
974 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
976 (unsigned long) i_size_read(file->f_mapping->host),
983 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
984 if (!bitmap->filemap)
989 if (bitmap->sb_page) {
990 bitmap->filemap[0] = bitmap->sb_page;
992 offset = sizeof(bitmap_super_t);
994 for ( ; pnum < num_pages; pnum++) {
995 bitmap->filemap[pnum] = alloc_page(GFP_KERNEL);
996 if (!bitmap->filemap[pnum]) {
997 bitmap->file_pages = pnum;
1001 bitmap->file_pages = pnum;
1003 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
1004 bitmap->filemap_attr = kzalloc(
1005 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
1007 if (!bitmap->filemap_attr)
1012 for (i = 0; i < chunks; i++) {
1014 index = file_page_index(bitmap, i);
1015 bit = file_page_offset(bitmap, i);
1016 if (index != oldindex) { /* this is a new page, read it in */
1018 /* unmap the old page, we're done with it */
1019 if (index == num_pages-1)
1020 count = bytes - index * PAGE_SIZE;
1023 page = bitmap->filemap[index];
1025 ret = read_page(file, index, bitmap,
1030 bitmap->mddev->bitmap_info.offset,
1039 bitmap->last_page_size = count;
1043 * if bitmap is out of date, dirty the
1044 * whole page and write it out
1046 paddr = kmap_atomic(page);
1047 memset(paddr + offset, 0xff,
1048 PAGE_SIZE - offset);
1049 kunmap_atomic(paddr);
1050 write_page(bitmap, page, 1);
1053 if (bitmap->flags & BITMAP_WRITE_ERROR)
1057 paddr = kmap_atomic(page);
1058 if (bitmap->flags & BITMAP_HOSTENDIAN)
1059 b = test_bit(bit, paddr);
1061 b = test_bit_le(bit, paddr);
1062 kunmap_atomic(paddr);
1064 /* if the disk bit is set, set the memory bit */
1065 int needed = ((sector_t)(i+1) << bitmap->chunkshift
1067 bitmap_set_memory_bits(bitmap,
1068 (sector_t)i << bitmap->chunkshift,
1075 printk(KERN_INFO "%s: bitmap initialized from disk: "
1076 "read %lu/%lu pages, set %lu of %lu bits\n",
1077 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, chunks);
1082 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1083 bmname(bitmap), ret);
1087 void bitmap_write_all(struct bitmap *bitmap)
1089 /* We don't actually write all bitmap blocks here,
1090 * just flag them as needing to be written
1094 if (!bitmap || !bitmap->filemap)
1097 /* Only one copy, so nothing needed */
1100 spin_lock_irq(&bitmap->lock);
1101 for (i = 0; i < bitmap->file_pages; i++)
1102 set_page_attr(bitmap, i,
1103 BITMAP_PAGE_NEEDWRITE);
1104 bitmap->allclean = 0;
1105 spin_unlock_irq(&bitmap->lock);
1108 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1110 sector_t chunk = offset >> bitmap->chunkshift;
1111 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1112 bitmap->bp[page].count += inc;
1113 bitmap_checkfree(bitmap, page);
1116 static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset)
1118 sector_t chunk = offset >> bitmap->chunkshift;
1119 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1120 struct bitmap_page *bp = &bitmap->bp[page];
1126 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1127 sector_t offset, sector_t *blocks,
1131 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1135 void bitmap_daemon_work(struct mddev *mddev)
1137 struct bitmap *bitmap;
1139 unsigned long nextpage;
1140 unsigned long flags;
1143 /* Use a mutex to guard daemon_work against
1146 mutex_lock(&mddev->bitmap_info.mutex);
1147 bitmap = mddev->bitmap;
1148 if (bitmap == NULL) {
1149 mutex_unlock(&mddev->bitmap_info.mutex);
1152 if (time_before(jiffies, bitmap->daemon_lastrun
1153 + mddev->bitmap_info.daemon_sleep))
1156 bitmap->daemon_lastrun = jiffies;
1157 if (bitmap->allclean) {
1158 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1161 bitmap->allclean = 1;
1163 /* Any file-page which is PENDING now needs to be written.
1164 * So set NEEDWRITE now, then after we make any last-minute changes
1167 spin_lock_irqsave(&bitmap->lock, flags);
1168 for (j = 0; j < bitmap->file_pages; j++)
1169 if (test_page_attr(bitmap, j,
1170 BITMAP_PAGE_PENDING)) {
1171 set_page_attr(bitmap, j,
1172 BITMAP_PAGE_NEEDWRITE);
1173 clear_page_attr(bitmap, j,
1174 BITMAP_PAGE_PENDING);
1177 if (bitmap->need_sync &&
1178 mddev->bitmap_info.external == 0) {
1179 /* Arrange for superblock update as well as
1182 bitmap->need_sync = 0;
1183 if (bitmap->filemap) {
1184 sb = kmap_atomic(bitmap->sb_page);
1185 sb->events_cleared =
1186 cpu_to_le64(bitmap->events_cleared);
1188 set_page_attr(bitmap, 0,
1189 BITMAP_PAGE_NEEDWRITE);
1192 /* Now look at the bitmap counters and if any are '2' or '1',
1193 * decrement and handle accordingly.
1196 for (j = 0; j < bitmap->chunks; j++) {
1197 bitmap_counter_t *bmc;
1198 sector_t block = (sector_t)j << bitmap->chunkshift;
1200 if (j == nextpage) {
1201 nextpage += PAGE_COUNTER_RATIO;
1202 if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1203 j |= PAGE_COUNTER_MASK;
1206 bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1208 bmc = bitmap_get_counter(bitmap,
1213 j |= PAGE_COUNTER_MASK;
1216 if (*bmc == 1 && !bitmap->need_sync) {
1217 /* We can clear the bit */
1219 bitmap_count_page(bitmap, block, -1);
1220 bitmap_file_clear_bit(bitmap, block);
1221 } else if (*bmc && *bmc <= 2) {
1223 bitmap_set_pending(bitmap, block);
1224 bitmap->allclean = 0;
1228 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1229 * DIRTY pages need to be written by bitmap_unplug so it can wait
1231 * If we find any DIRTY page we stop there and let bitmap_unplug
1232 * handle all the rest. This is important in the case where
1233 * the first blocking holds the superblock and it has been updated.
1234 * We mustn't write any other blocks before the superblock.
1236 for (j = 0; j < bitmap->file_pages; j++) {
1238 if (test_page_attr(bitmap, j,
1240 /* bitmap_unplug will handle the rest */
1242 if (test_page_attr(bitmap, j,
1243 BITMAP_PAGE_NEEDWRITE)) {
1244 clear_page_attr(bitmap, j,
1245 BITMAP_PAGE_NEEDWRITE);
1246 spin_unlock_irqrestore(&bitmap->lock, flags);
1247 write_page(bitmap, bitmap->filemap[j], 0);
1248 spin_lock_irqsave(&bitmap->lock, flags);
1249 if (!bitmap->filemap)
1253 spin_unlock_irqrestore(&bitmap->lock, flags);
1256 if (bitmap->allclean == 0)
1257 mddev->thread->timeout =
1258 mddev->bitmap_info.daemon_sleep;
1259 mutex_unlock(&mddev->bitmap_info.mutex);
1262 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1263 sector_t offset, sector_t *blocks,
1265 __releases(bitmap->lock)
1266 __acquires(bitmap->lock)
1268 /* If 'create', we might release the lock and reclaim it.
1269 * The lock must have been taken with interrupts enabled.
1270 * If !create, we don't release the lock.
1272 sector_t chunk = offset >> bitmap->chunkshift;
1273 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1274 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1278 err = bitmap_checkpage(bitmap, page, create);
1280 if (bitmap->bp[page].hijacked ||
1281 bitmap->bp[page].map == NULL)
1282 csize = ((sector_t)1) << (bitmap->chunkshift +
1283 PAGE_COUNTER_SHIFT - 1);
1285 csize = ((sector_t)1) << bitmap->chunkshift;
1286 *blocks = csize - (offset & (csize - 1));
1291 /* now locked ... */
1293 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1294 /* should we use the first or second counter field
1295 * of the hijacked pointer? */
1296 int hi = (pageoff > PAGE_COUNTER_MASK);
1297 return &((bitmap_counter_t *)
1298 &bitmap->bp[page].map)[hi];
1299 } else /* page is allocated */
1300 return (bitmap_counter_t *)
1301 &(bitmap->bp[page].map[pageoff]);
1304 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1311 atomic_inc(&bitmap->behind_writes);
1312 bw = atomic_read(&bitmap->behind_writes);
1313 if (bw > bitmap->behind_writes_used)
1314 bitmap->behind_writes_used = bw;
1316 pr_debug("inc write-behind count %d/%lu\n",
1317 bw, bitmap->mddev->bitmap_info.max_write_behind);
1322 bitmap_counter_t *bmc;
1324 spin_lock_irq(&bitmap->lock);
1325 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1327 spin_unlock_irq(&bitmap->lock);
1331 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1332 DEFINE_WAIT(__wait);
1333 /* note that it is safe to do the prepare_to_wait
1334 * after the test as long as we do it before dropping
1337 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1338 TASK_UNINTERRUPTIBLE);
1339 spin_unlock_irq(&bitmap->lock);
1341 finish_wait(&bitmap->overflow_wait, &__wait);
1347 bitmap_file_set_bit(bitmap, offset);
1348 bitmap_count_page(bitmap, offset, 1);
1356 spin_unlock_irq(&bitmap->lock);
1359 if (sectors > blocks)
1366 EXPORT_SYMBOL(bitmap_startwrite);
1368 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1369 int success, int behind)
1374 if (atomic_dec_and_test(&bitmap->behind_writes))
1375 wake_up(&bitmap->behind_wait);
1376 pr_debug("dec write-behind count %d/%lu\n",
1377 atomic_read(&bitmap->behind_writes),
1378 bitmap->mddev->bitmap_info.max_write_behind);
1383 unsigned long flags;
1384 bitmap_counter_t *bmc;
1386 spin_lock_irqsave(&bitmap->lock, flags);
1387 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1389 spin_unlock_irqrestore(&bitmap->lock, flags);
1393 if (success && !bitmap->mddev->degraded &&
1394 bitmap->events_cleared < bitmap->mddev->events) {
1395 bitmap->events_cleared = bitmap->mddev->events;
1396 bitmap->need_sync = 1;
1397 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1400 if (!success && !NEEDED(*bmc))
1401 *bmc |= NEEDED_MASK;
1403 if (COUNTER(*bmc) == COUNTER_MAX)
1404 wake_up(&bitmap->overflow_wait);
1408 bitmap_set_pending(bitmap, offset);
1409 bitmap->allclean = 0;
1411 spin_unlock_irqrestore(&bitmap->lock, flags);
1413 if (sectors > blocks)
1419 EXPORT_SYMBOL(bitmap_endwrite);
1421 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1424 bitmap_counter_t *bmc;
1426 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1428 return 1; /* always resync if no bitmap */
1430 spin_lock_irq(&bitmap->lock);
1431 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1437 else if (NEEDED(*bmc)) {
1439 if (!degraded) { /* don't set/clear bits if degraded */
1440 *bmc |= RESYNC_MASK;
1441 *bmc &= ~NEEDED_MASK;
1445 spin_unlock_irq(&bitmap->lock);
1449 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1452 /* bitmap_start_sync must always report on multiples of whole
1453 * pages, otherwise resync (which is very PAGE_SIZE based) will
1455 * So call __bitmap_start_sync repeatedly (if needed) until
1456 * At least PAGE_SIZE>>9 blocks are covered.
1457 * Return the 'or' of the result.
1463 while (*blocks < (PAGE_SIZE>>9)) {
1464 rv |= __bitmap_start_sync(bitmap, offset,
1465 &blocks1, degraded);
1471 EXPORT_SYMBOL(bitmap_start_sync);
1473 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1475 bitmap_counter_t *bmc;
1476 unsigned long flags;
1478 if (bitmap == NULL) {
1482 spin_lock_irqsave(&bitmap->lock, flags);
1483 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1488 *bmc &= ~RESYNC_MASK;
1490 if (!NEEDED(*bmc) && aborted)
1491 *bmc |= NEEDED_MASK;
1494 bitmap_set_pending(bitmap, offset);
1495 bitmap->allclean = 0;
1500 spin_unlock_irqrestore(&bitmap->lock, flags);
1502 EXPORT_SYMBOL(bitmap_end_sync);
1504 void bitmap_close_sync(struct bitmap *bitmap)
1506 /* Sync has finished, and any bitmap chunks that weren't synced
1507 * properly have been aborted. It remains to us to clear the
1508 * RESYNC bit wherever it is still on
1510 sector_t sector = 0;
1514 while (sector < bitmap->mddev->resync_max_sectors) {
1515 bitmap_end_sync(bitmap, sector, &blocks, 0);
1519 EXPORT_SYMBOL(bitmap_close_sync);
1521 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1529 bitmap->last_end_sync = jiffies;
1532 if (time_before(jiffies, (bitmap->last_end_sync
1533 + bitmap->mddev->bitmap_info.daemon_sleep)))
1535 wait_event(bitmap->mddev->recovery_wait,
1536 atomic_read(&bitmap->mddev->recovery_active) == 0);
1538 bitmap->mddev->curr_resync_completed = sector;
1539 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1540 sector &= ~((1ULL << bitmap->chunkshift) - 1);
1542 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1543 bitmap_end_sync(bitmap, s, &blocks, 0);
1546 bitmap->last_end_sync = jiffies;
1547 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1549 EXPORT_SYMBOL(bitmap_cond_end_sync);
1551 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1553 /* For each chunk covered by any of these sectors, set the
1554 * counter to 2 and possibly set resync_needed. They should all
1555 * be 0 at this point
1559 bitmap_counter_t *bmc;
1560 spin_lock_irq(&bitmap->lock);
1561 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1563 spin_unlock_irq(&bitmap->lock);
1567 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1568 bitmap_count_page(bitmap, offset, 1);
1569 bitmap_set_pending(bitmap, offset);
1570 bitmap->allclean = 0;
1572 spin_unlock_irq(&bitmap->lock);
1575 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1576 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1578 unsigned long chunk;
1580 for (chunk = s; chunk <= e; chunk++) {
1581 sector_t sec = (sector_t)chunk << bitmap->chunkshift;
1582 bitmap_set_memory_bits(bitmap, sec, 1);
1583 spin_lock_irq(&bitmap->lock);
1584 bitmap_file_set_bit(bitmap, sec);
1585 spin_unlock_irq(&bitmap->lock);
1586 if (sec < bitmap->mddev->recovery_cp)
1587 /* We are asserting that the array is dirty,
1588 * so move the recovery_cp address back so
1589 * that it is obvious that it is dirty
1591 bitmap->mddev->recovery_cp = sec;
1596 * flush out any pending updates
1598 void bitmap_flush(struct mddev *mddev)
1600 struct bitmap *bitmap = mddev->bitmap;
1603 if (!bitmap) /* there was no bitmap */
1606 /* run the daemon_work three time to ensure everything is flushed
1609 sleep = mddev->bitmap_info.daemon_sleep * 2;
1610 bitmap->daemon_lastrun -= sleep;
1611 bitmap_daemon_work(mddev);
1612 bitmap->daemon_lastrun -= sleep;
1613 bitmap_daemon_work(mddev);
1614 bitmap->daemon_lastrun -= sleep;
1615 bitmap_daemon_work(mddev);
1616 bitmap_update_sb(bitmap);
1620 * free memory that was allocated
1622 static void bitmap_free(struct bitmap *bitmap)
1624 unsigned long k, pages;
1625 struct bitmap_page *bp;
1627 if (!bitmap) /* there was no bitmap */
1630 /* release the bitmap file and kill the daemon */
1631 bitmap_file_put(bitmap);
1634 pages = bitmap->pages;
1636 /* free all allocated memory */
1638 if (bp) /* deallocate the page memory */
1639 for (k = 0; k < pages; k++)
1640 if (bp[k].map && !bp[k].hijacked)
1646 void bitmap_destroy(struct mddev *mddev)
1648 struct bitmap *bitmap = mddev->bitmap;
1650 if (!bitmap) /* there was no bitmap */
1653 mutex_lock(&mddev->bitmap_info.mutex);
1654 mddev->bitmap = NULL; /* disconnect from the md device */
1655 mutex_unlock(&mddev->bitmap_info.mutex);
1657 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1659 if (bitmap->sysfs_can_clear)
1660 sysfs_put(bitmap->sysfs_can_clear);
1662 bitmap_free(bitmap);
1666 * initialize the bitmap structure
1667 * if this returns an error, bitmap_destroy must be called to do clean up
1669 int bitmap_create(struct mddev *mddev)
1671 struct bitmap *bitmap;
1672 sector_t blocks = mddev->resync_max_sectors;
1673 unsigned long chunks;
1674 unsigned long pages;
1675 struct file *file = mddev->bitmap_info.file;
1677 struct sysfs_dirent *bm = NULL;
1679 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1681 BUG_ON(file && mddev->bitmap_info.offset);
1683 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1687 spin_lock_init(&bitmap->lock);
1688 atomic_set(&bitmap->pending_writes, 0);
1689 init_waitqueue_head(&bitmap->write_wait);
1690 init_waitqueue_head(&bitmap->overflow_wait);
1691 init_waitqueue_head(&bitmap->behind_wait);
1693 bitmap->mddev = mddev;
1696 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1698 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1701 bitmap->sysfs_can_clear = NULL;
1703 bitmap->file = file;
1706 /* As future accesses to this file will use bmap,
1707 * and bypass the page cache, we must sync the file
1712 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1713 if (!mddev->bitmap_info.external) {
1715 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1716 * instructing us to create a new on-disk bitmap instance.
1718 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1719 err = bitmap_new_disk_sb(bitmap);
1721 err = bitmap_read_sb(bitmap);
1724 if (mddev->bitmap_info.chunksize == 0 ||
1725 mddev->bitmap_info.daemon_sleep == 0)
1726 /* chunksize and time_base need to be
1733 bitmap->daemon_lastrun = jiffies;
1734 bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize)
1735 - BITMAP_BLOCK_SHIFT);
1737 chunks = (blocks + (1 << bitmap->chunkshift) - 1) >>
1739 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1743 bitmap->chunks = chunks;
1744 bitmap->pages = pages;
1745 bitmap->missing_pages = pages;
1747 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1753 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1754 pages, bmname(bitmap));
1756 mddev->bitmap = bitmap;
1759 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1762 bitmap_free(bitmap);
1766 int bitmap_load(struct mddev *mddev)
1770 sector_t sector = 0;
1771 struct bitmap *bitmap = mddev->bitmap;
1776 /* Clear out old bitmap info first: Either there is none, or we
1777 * are resuming after someone else has possibly changed things,
1778 * so we should forget old cached info.
1779 * All chunks should be clean, but some might need_sync.
1781 while (sector < mddev->resync_max_sectors) {
1783 bitmap_start_sync(bitmap, sector, &blocks, 0);
1786 bitmap_close_sync(bitmap);
1788 if (mddev->degraded == 0
1789 || bitmap->events_cleared == mddev->events)
1790 /* no need to keep dirty bits to optimise a
1791 * re-add of a missing device */
1792 start = mddev->recovery_cp;
1794 mutex_lock(&mddev->bitmap_info.mutex);
1795 err = bitmap_init_from_disk(bitmap, start);
1796 mutex_unlock(&mddev->bitmap_info.mutex);
1800 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1802 /* Kick recovery in case any bits were set */
1803 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1805 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1806 md_wakeup_thread(mddev->thread);
1808 bitmap_update_sb(bitmap);
1810 if (bitmap->flags & BITMAP_WRITE_ERROR)
1815 EXPORT_SYMBOL_GPL(bitmap_load);
1817 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1819 unsigned long chunk_kb;
1820 unsigned long flags;
1825 spin_lock_irqsave(&bitmap->lock, flags);
1826 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1827 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1829 bitmap->pages - bitmap->missing_pages,
1831 (bitmap->pages - bitmap->missing_pages)
1832 << (PAGE_SHIFT - 10),
1833 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1834 chunk_kb ? "KB" : "B");
1836 seq_printf(seq, ", file: ");
1837 seq_path(seq, &bitmap->file->f_path, " \t\n");
1840 seq_printf(seq, "\n");
1841 spin_unlock_irqrestore(&bitmap->lock, flags);
1845 location_show(struct mddev *mddev, char *page)
1848 if (mddev->bitmap_info.file)
1849 len = sprintf(page, "file");
1850 else if (mddev->bitmap_info.offset)
1851 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1853 len = sprintf(page, "none");
1854 len += sprintf(page+len, "\n");
1859 location_store(struct mddev *mddev, const char *buf, size_t len)
1863 if (!mddev->pers->quiesce)
1865 if (mddev->recovery || mddev->sync_thread)
1869 if (mddev->bitmap || mddev->bitmap_info.file ||
1870 mddev->bitmap_info.offset) {
1871 /* bitmap already configured. Only option is to clear it */
1872 if (strncmp(buf, "none", 4) != 0)
1875 mddev->pers->quiesce(mddev, 1);
1876 bitmap_destroy(mddev);
1877 mddev->pers->quiesce(mddev, 0);
1879 mddev->bitmap_info.offset = 0;
1880 if (mddev->bitmap_info.file) {
1881 struct file *f = mddev->bitmap_info.file;
1882 mddev->bitmap_info.file = NULL;
1883 restore_bitmap_write_access(f);
1887 /* No bitmap, OK to set a location */
1889 if (strncmp(buf, "none", 4) == 0)
1890 /* nothing to be done */;
1891 else if (strncmp(buf, "file:", 5) == 0) {
1892 /* Not supported yet */
1897 rv = strict_strtoll(buf+1, 10, &offset);
1899 rv = strict_strtoll(buf, 10, &offset);
1904 if (mddev->bitmap_info.external == 0 &&
1905 mddev->major_version == 0 &&
1906 offset != mddev->bitmap_info.default_offset)
1908 mddev->bitmap_info.offset = offset;
1910 mddev->pers->quiesce(mddev, 1);
1911 rv = bitmap_create(mddev);
1913 rv = bitmap_load(mddev);
1915 bitmap_destroy(mddev);
1916 mddev->bitmap_info.offset = 0;
1918 mddev->pers->quiesce(mddev, 0);
1924 if (!mddev->external) {
1925 /* Ensure new bitmap info is stored in
1926 * metadata promptly.
1928 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1929 md_wakeup_thread(mddev->thread);
1934 static struct md_sysfs_entry bitmap_location =
1935 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1937 /* 'bitmap/space' is the space available at 'location' for the
1938 * bitmap. This allows the kernel to know when it is safe to
1939 * resize the bitmap to match a resized array.
1942 space_show(struct mddev *mddev, char *page)
1944 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
1948 space_store(struct mddev *mddev, const char *buf, size_t len)
1950 unsigned long sectors;
1953 rv = kstrtoul(buf, 10, §ors);
1960 if (mddev->bitmap &&
1961 sectors < ((mddev->bitmap->file_pages - 1) * PAGE_SIZE
1962 + mddev->bitmap->last_page_size + 511) >> 9)
1963 return -EFBIG; /* Bitmap is too big for this small space */
1965 /* could make sure it isn't too big, but that isn't really
1966 * needed - user-space should be careful.
1968 mddev->bitmap_info.space = sectors;
1972 static struct md_sysfs_entry bitmap_space =
1973 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
1976 timeout_show(struct mddev *mddev, char *page)
1979 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1980 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1982 len = sprintf(page, "%lu", secs);
1984 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1985 len += sprintf(page+len, "\n");
1990 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1992 /* timeout can be set at any time */
1993 unsigned long timeout;
1994 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1998 /* just to make sure we don't overflow... */
1999 if (timeout >= LONG_MAX / HZ)
2002 timeout = timeout * HZ / 10000;
2004 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2005 timeout = MAX_SCHEDULE_TIMEOUT-1;
2008 mddev->bitmap_info.daemon_sleep = timeout;
2009 if (mddev->thread) {
2010 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2011 * the bitmap is all clean and we don't need to
2012 * adjust the timeout right now
2014 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2015 mddev->thread->timeout = timeout;
2016 md_wakeup_thread(mddev->thread);
2022 static struct md_sysfs_entry bitmap_timeout =
2023 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2026 backlog_show(struct mddev *mddev, char *page)
2028 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2032 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2034 unsigned long backlog;
2035 int rv = strict_strtoul(buf, 10, &backlog);
2038 if (backlog > COUNTER_MAX)
2040 mddev->bitmap_info.max_write_behind = backlog;
2044 static struct md_sysfs_entry bitmap_backlog =
2045 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2048 chunksize_show(struct mddev *mddev, char *page)
2050 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2054 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2056 /* Can only be changed when no bitmap is active */
2058 unsigned long csize;
2061 rv = strict_strtoul(buf, 10, &csize);
2065 !is_power_of_2(csize))
2067 mddev->bitmap_info.chunksize = csize;
2071 static struct md_sysfs_entry bitmap_chunksize =
2072 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2074 static ssize_t metadata_show(struct mddev *mddev, char *page)
2076 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2077 ? "external" : "internal"));
2080 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2082 if (mddev->bitmap ||
2083 mddev->bitmap_info.file ||
2084 mddev->bitmap_info.offset)
2086 if (strncmp(buf, "external", 8) == 0)
2087 mddev->bitmap_info.external = 1;
2088 else if (strncmp(buf, "internal", 8) == 0)
2089 mddev->bitmap_info.external = 0;
2095 static struct md_sysfs_entry bitmap_metadata =
2096 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2098 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2102 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2105 len = sprintf(page, "\n");
2109 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2111 if (mddev->bitmap == NULL)
2113 if (strncmp(buf, "false", 5) == 0)
2114 mddev->bitmap->need_sync = 1;
2115 else if (strncmp(buf, "true", 4) == 0) {
2116 if (mddev->degraded)
2118 mddev->bitmap->need_sync = 0;
2124 static struct md_sysfs_entry bitmap_can_clear =
2125 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2128 behind_writes_used_show(struct mddev *mddev, char *page)
2130 if (mddev->bitmap == NULL)
2131 return sprintf(page, "0\n");
2132 return sprintf(page, "%lu\n",
2133 mddev->bitmap->behind_writes_used);
2137 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2140 mddev->bitmap->behind_writes_used = 0;
2144 static struct md_sysfs_entry max_backlog_used =
2145 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2146 behind_writes_used_show, behind_writes_used_reset);
2148 static struct attribute *md_bitmap_attrs[] = {
2149 &bitmap_location.attr,
2151 &bitmap_timeout.attr,
2152 &bitmap_backlog.attr,
2153 &bitmap_chunksize.attr,
2154 &bitmap_metadata.attr,
2155 &bitmap_can_clear.attr,
2156 &max_backlog_used.attr,
2159 struct attribute_group md_bitmap_group = {
2161 .attrs = md_bitmap_attrs,