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>
32 static inline char *bmname(struct bitmap *bitmap)
34 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
38 * just a placeholder - calls kmalloc for bitmap pages
40 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
44 page = kzalloc(PAGE_SIZE, GFP_NOIO);
46 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
48 pr_debug("%s: bitmap_alloc_page: allocated page at %p\n",
49 bmname(bitmap), page);
54 * for now just a placeholder -- just calls kfree for bitmap pages
56 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
58 pr_debug("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
63 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
65 * 1) check to see if this page is allocated, if it's not then try to alloc
66 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
67 * page pointer directly as a counter
69 * if we find our page, we increment the page's refcount so that it stays
70 * allocated while we're using it
72 static int bitmap_checkpage(struct bitmap *bitmap,
73 unsigned long page, int create)
74 __releases(bitmap->lock)
75 __acquires(bitmap->lock)
77 unsigned char *mappage;
79 if (page >= bitmap->pages) {
80 /* This can happen if bitmap_start_sync goes beyond
81 * End-of-device while looking for a whole page.
87 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
90 if (bitmap->bp[page].map) /* page is already allocated, just return */
96 /* this page has not been allocated yet */
98 spin_unlock_irq(&bitmap->lock);
99 mappage = bitmap_alloc_page(bitmap);
100 spin_lock_irq(&bitmap->lock);
102 if (mappage == NULL) {
103 pr_debug("%s: bitmap map page allocation failed, hijacking\n",
105 /* failed - set the hijacked flag so that we can use the
106 * pointer as a counter */
107 if (!bitmap->bp[page].map)
108 bitmap->bp[page].hijacked = 1;
109 } else if (bitmap->bp[page].map ||
110 bitmap->bp[page].hijacked) {
111 /* somebody beat us to getting the page */
112 bitmap_free_page(bitmap, mappage);
116 /* no page was in place and we have one, so install it */
118 bitmap->bp[page].map = mappage;
119 bitmap->missing_pages--;
124 /* if page is completely empty, put it back on the free list, or dealloc it */
125 /* if page was hijacked, unmark the flag so it might get alloced next time */
126 /* Note: lock should be held when calling this */
127 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
131 if (bitmap->bp[page].count) /* page is still busy */
134 /* page is no longer in use, it can be released */
136 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
137 bitmap->bp[page].hijacked = 0;
138 bitmap->bp[page].map = NULL;
140 /* normal case, free the page */
141 ptr = bitmap->bp[page].map;
142 bitmap->bp[page].map = NULL;
143 bitmap->missing_pages++;
144 bitmap_free_page(bitmap, ptr);
149 * bitmap file handling - read and write the bitmap file and its superblock
153 * basic page I/O operations
156 /* IO operations when bitmap is stored near all superblocks */
157 static struct page *read_sb_page(struct mddev *mddev, loff_t offset,
159 unsigned long index, int size)
161 /* choose a good rdev and read the page from there */
163 struct md_rdev *rdev;
168 page = alloc_page(GFP_KERNEL);
170 return ERR_PTR(-ENOMEM);
174 list_for_each_entry(rdev, &mddev->disks, same_set) {
175 if (! test_bit(In_sync, &rdev->flags)
176 || test_bit(Faulty, &rdev->flags))
179 target = offset + index * (PAGE_SIZE/512);
181 if (sync_page_io(rdev, target,
182 roundup(size, bdev_logical_block_size(rdev->bdev)),
185 attach_page_buffers(page, NULL); /* so that free_buffer will
192 return ERR_PTR(-EIO);
196 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
198 /* Iterate the disks of an mddev, using rcu to protect access to the
199 * linked list, and raising the refcount of devices we return to ensure
200 * they don't disappear while in use.
201 * As devices are only added or removed when raid_disk is < 0 and
202 * nr_pending is 0 and In_sync is clear, the entries we return will
203 * still be in the same position on the list when we re-enter
204 * list_for_each_continue_rcu.
206 struct list_head *pos;
209 /* start at the beginning */
212 /* release the previous rdev and start from there. */
213 rdev_dec_pending(rdev, mddev);
214 pos = &rdev->same_set;
216 list_for_each_continue_rcu(pos, &mddev->disks) {
217 rdev = list_entry(pos, struct md_rdev, same_set);
218 if (rdev->raid_disk >= 0 &&
219 !test_bit(Faulty, &rdev->flags)) {
220 /* this is a usable devices */
221 atomic_inc(&rdev->nr_pending);
230 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
232 struct md_rdev *rdev = NULL;
233 struct block_device *bdev;
234 struct mddev *mddev = bitmap->mddev;
236 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
237 int size = PAGE_SIZE;
238 loff_t offset = mddev->bitmap_info.offset;
240 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
242 if (page->index == bitmap->file_pages-1)
243 size = roundup(bitmap->last_page_size,
244 bdev_logical_block_size(bdev));
245 /* Just make sure we aren't corrupting data or
248 if (mddev->external) {
249 /* Bitmap could be anywhere. */
250 if (rdev->sb_start + offset + (page->index
254 rdev->sb_start + offset
255 < (rdev->data_offset + mddev->dev_sectors
258 } else if (offset < 0) {
259 /* DATA BITMAP METADATA */
261 + (long)(page->index * (PAGE_SIZE/512))
263 /* bitmap runs in to metadata */
265 if (rdev->data_offset + mddev->dev_sectors
266 > rdev->sb_start + offset)
267 /* data runs in to bitmap */
269 } else if (rdev->sb_start < rdev->data_offset) {
270 /* METADATA BITMAP DATA */
273 + page->index*(PAGE_SIZE/512) + size/512
275 /* bitmap runs in to data */
278 /* DATA METADATA BITMAP - no problems */
280 md_super_write(mddev, rdev,
281 rdev->sb_start + offset
282 + page->index * (PAGE_SIZE/512),
288 md_super_wait(mddev);
295 static void bitmap_file_kick(struct bitmap *bitmap);
297 * write out a page to a file
299 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
301 struct buffer_head *bh;
303 if (bitmap->file == NULL) {
304 switch (write_sb_page(bitmap, page, wait)) {
306 bitmap->flags |= BITMAP_WRITE_ERROR;
310 bh = page_buffers(page);
312 while (bh && bh->b_blocknr) {
313 atomic_inc(&bitmap->pending_writes);
314 set_buffer_locked(bh);
315 set_buffer_mapped(bh);
316 submit_bh(WRITE | REQ_SYNC, bh);
317 bh = bh->b_this_page;
321 wait_event(bitmap->write_wait,
322 atomic_read(&bitmap->pending_writes)==0);
324 if (bitmap->flags & BITMAP_WRITE_ERROR)
325 bitmap_file_kick(bitmap);
328 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
330 struct bitmap *bitmap = bh->b_private;
334 spin_lock_irqsave(&bitmap->lock, flags);
335 bitmap->flags |= BITMAP_WRITE_ERROR;
336 spin_unlock_irqrestore(&bitmap->lock, flags);
338 if (atomic_dec_and_test(&bitmap->pending_writes))
339 wake_up(&bitmap->write_wait);
342 /* copied from buffer.c */
344 __clear_page_buffers(struct page *page)
346 ClearPagePrivate(page);
347 set_page_private(page, 0);
348 page_cache_release(page);
350 static void free_buffers(struct page *page)
352 struct buffer_head *bh = page_buffers(page);
355 struct buffer_head *next = bh->b_this_page;
356 free_buffer_head(bh);
359 __clear_page_buffers(page);
363 /* read a page from a file.
364 * We both read the page, and attach buffers to the page to record the
365 * address of each block (using bmap). These addresses will be used
366 * to write the block later, completely bypassing the filesystem.
367 * This usage is similar to how swap files are handled, and allows us
368 * to write to a file with no concerns of memory allocation failing.
370 static struct page *read_page(struct file *file, unsigned long index,
371 struct bitmap *bitmap,
374 struct page *page = NULL;
375 struct inode *inode = file->f_path.dentry->d_inode;
376 struct buffer_head *bh;
379 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
380 (unsigned long long)index << PAGE_SHIFT);
382 page = alloc_page(GFP_KERNEL);
384 page = ERR_PTR(-ENOMEM);
388 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
391 page = ERR_PTR(-ENOMEM);
394 attach_page_buffers(page, bh);
395 block = index << (PAGE_SHIFT - inode->i_blkbits);
400 bh->b_blocknr = bmap(inode, block);
401 if (bh->b_blocknr == 0) {
402 /* Cannot use this file! */
404 page = ERR_PTR(-EINVAL);
407 bh->b_bdev = inode->i_sb->s_bdev;
408 if (count < (1<<inode->i_blkbits))
411 count -= (1<<inode->i_blkbits);
413 bh->b_end_io = end_bitmap_write;
414 bh->b_private = bitmap;
415 atomic_inc(&bitmap->pending_writes);
416 set_buffer_locked(bh);
417 set_buffer_mapped(bh);
421 bh = bh->b_this_page;
425 wait_event(bitmap->write_wait,
426 atomic_read(&bitmap->pending_writes)==0);
427 if (bitmap->flags & BITMAP_WRITE_ERROR) {
429 page = ERR_PTR(-EIO);
433 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n",
435 (unsigned long long)index << PAGE_SHIFT,
441 * bitmap file superblock operations
444 /* update the event counter and sync the superblock to disk */
445 void bitmap_update_sb(struct bitmap *bitmap)
450 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
452 if (bitmap->mddev->bitmap_info.external)
454 spin_lock_irqsave(&bitmap->lock, flags);
455 if (!bitmap->sb_page) { /* no superblock */
456 spin_unlock_irqrestore(&bitmap->lock, flags);
459 spin_unlock_irqrestore(&bitmap->lock, flags);
460 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
461 sb->events = cpu_to_le64(bitmap->mddev->events);
462 if (bitmap->mddev->events < bitmap->events_cleared)
463 /* rocking back to read-only */
464 bitmap->events_cleared = bitmap->mddev->events;
465 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
466 sb->state = cpu_to_le32(bitmap->flags);
467 /* Just in case these have been changed via sysfs: */
468 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
469 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
470 kunmap_atomic(sb, KM_USER0);
471 write_page(bitmap, bitmap->sb_page, 1);
474 /* print out the bitmap file superblock */
475 void bitmap_print_sb(struct bitmap *bitmap)
479 if (!bitmap || !bitmap->sb_page)
481 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
482 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
483 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
484 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
485 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
486 *(__u32 *)(sb->uuid+0),
487 *(__u32 *)(sb->uuid+4),
488 *(__u32 *)(sb->uuid+8),
489 *(__u32 *)(sb->uuid+12));
490 printk(KERN_DEBUG " events: %llu\n",
491 (unsigned long long) le64_to_cpu(sb->events));
492 printk(KERN_DEBUG "events cleared: %llu\n",
493 (unsigned long long) le64_to_cpu(sb->events_cleared));
494 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
495 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
496 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
497 printk(KERN_DEBUG " sync size: %llu KB\n",
498 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
499 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
500 kunmap_atomic(sb, KM_USER0);
507 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
508 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
509 * This function verifies 'bitmap_info' and populates the on-disk bitmap
510 * structure, which is to be written to disk.
512 * Returns: 0 on success, -Exxx on error
514 static int bitmap_new_disk_sb(struct bitmap *bitmap)
517 unsigned long chunksize, daemon_sleep, write_behind;
520 bitmap->sb_page = alloc_page(GFP_KERNEL);
521 if (IS_ERR(bitmap->sb_page)) {
522 err = PTR_ERR(bitmap->sb_page);
523 bitmap->sb_page = NULL;
526 bitmap->sb_page->index = 0;
528 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
530 sb->magic = cpu_to_le32(BITMAP_MAGIC);
531 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
533 chunksize = bitmap->mddev->bitmap_info.chunksize;
535 if (!is_power_of_2(chunksize)) {
536 kunmap_atomic(sb, KM_USER0);
537 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
540 sb->chunksize = cpu_to_le32(chunksize);
542 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
544 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
545 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
546 daemon_sleep = 5 * HZ;
548 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
549 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
552 * FIXME: write_behind for RAID1. If not specified, what
553 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
555 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
556 if (write_behind > COUNTER_MAX)
557 write_behind = COUNTER_MAX / 2;
558 sb->write_behind = cpu_to_le32(write_behind);
559 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
561 /* keep the array size field of the bitmap superblock up to date */
562 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
564 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
566 bitmap->flags |= BITMAP_STALE;
567 sb->state |= cpu_to_le32(BITMAP_STALE);
568 bitmap->events_cleared = bitmap->mddev->events;
569 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
571 bitmap->flags |= BITMAP_HOSTENDIAN;
572 sb->version = cpu_to_le32(BITMAP_MAJOR_HOSTENDIAN);
574 kunmap_atomic(sb, KM_USER0);
579 /* read the superblock from the bitmap file and initialize some bitmap fields */
580 static int bitmap_read_sb(struct bitmap *bitmap)
584 unsigned long chunksize, daemon_sleep, write_behind;
585 unsigned long long events;
588 /* page 0 is the superblock, read it... */
590 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
591 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
593 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
595 bitmap->sb_page = read_sb_page(bitmap->mddev,
596 bitmap->mddev->bitmap_info.offset,
598 0, sizeof(bitmap_super_t));
600 if (IS_ERR(bitmap->sb_page)) {
601 err = PTR_ERR(bitmap->sb_page);
602 bitmap->sb_page = NULL;
606 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
608 chunksize = le32_to_cpu(sb->chunksize);
609 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
610 write_behind = le32_to_cpu(sb->write_behind);
612 /* verify that the bitmap-specific fields are valid */
613 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
614 reason = "bad magic";
615 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
616 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
617 reason = "unrecognized superblock version";
618 else if (chunksize < 512)
619 reason = "bitmap chunksize too small";
620 else if (!is_power_of_2(chunksize))
621 reason = "bitmap chunksize not a power of 2";
622 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
623 reason = "daemon sleep period out of range";
624 else if (write_behind > COUNTER_MAX)
625 reason = "write-behind limit out of range (0 - 16383)";
627 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
628 bmname(bitmap), reason);
632 /* keep the array size field of the bitmap superblock up to date */
633 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
635 if (!bitmap->mddev->persistent)
639 * if we have a persistent array superblock, compare the
640 * bitmap's UUID and event counter to the mddev's
642 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
643 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
647 events = le64_to_cpu(sb->events);
648 if (events < bitmap->mddev->events) {
649 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
650 "-- forcing full recovery\n", bmname(bitmap), events,
651 (unsigned long long) bitmap->mddev->events);
652 sb->state |= cpu_to_le32(BITMAP_STALE);
655 /* assign fields using values from superblock */
656 bitmap->mddev->bitmap_info.chunksize = chunksize;
657 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
658 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
659 bitmap->flags |= le32_to_cpu(sb->state);
660 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
661 bitmap->flags |= BITMAP_HOSTENDIAN;
662 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
663 if (bitmap->flags & BITMAP_STALE)
664 bitmap->events_cleared = bitmap->mddev->events;
667 kunmap_atomic(sb, KM_USER0);
669 bitmap_print_sb(bitmap);
673 enum bitmap_mask_op {
678 /* record the state of the bitmap in the superblock. Return the old value */
679 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
680 enum bitmap_mask_op op)
686 spin_lock_irqsave(&bitmap->lock, flags);
687 if (!bitmap->sb_page) { /* can't set the state */
688 spin_unlock_irqrestore(&bitmap->lock, flags);
691 spin_unlock_irqrestore(&bitmap->lock, flags);
692 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
693 old = le32_to_cpu(sb->state) & bits;
696 sb->state |= cpu_to_le32(bits);
697 bitmap->flags |= bits;
700 sb->state &= cpu_to_le32(~bits);
701 bitmap->flags &= ~bits;
706 kunmap_atomic(sb, KM_USER0);
711 * general bitmap file operations
717 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
718 * file a page at a time. There's a superblock at the start of the file.
720 /* calculate the index of the page that contains this bit */
721 static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
723 if (!bitmap->mddev->bitmap_info.external)
724 chunk += sizeof(bitmap_super_t) << 3;
725 return chunk >> PAGE_BIT_SHIFT;
728 /* calculate the (bit) offset of this bit within a page */
729 static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
731 if (!bitmap->mddev->bitmap_info.external)
732 chunk += sizeof(bitmap_super_t) << 3;
733 return chunk & (PAGE_BITS - 1);
737 * return a pointer to the page in the filemap that contains the given bit
739 * this lookup is complicated by the fact that the bitmap sb might be exactly
740 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
743 static inline struct page *filemap_get_page(struct bitmap *bitmap,
746 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
748 return bitmap->filemap[file_page_index(bitmap, chunk)
749 - file_page_index(bitmap, 0)];
752 static void bitmap_file_unmap(struct bitmap *bitmap)
754 struct page **map, *sb_page;
759 spin_lock_irqsave(&bitmap->lock, flags);
760 map = bitmap->filemap;
761 bitmap->filemap = NULL;
762 attr = bitmap->filemap_attr;
763 bitmap->filemap_attr = NULL;
764 pages = bitmap->file_pages;
765 bitmap->file_pages = 0;
766 sb_page = bitmap->sb_page;
767 bitmap->sb_page = NULL;
768 spin_unlock_irqrestore(&bitmap->lock, flags);
771 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
772 free_buffers(map[pages]);
777 free_buffers(sb_page);
780 static void bitmap_file_put(struct bitmap *bitmap)
785 spin_lock_irqsave(&bitmap->lock, flags);
788 spin_unlock_irqrestore(&bitmap->lock, flags);
791 wait_event(bitmap->write_wait,
792 atomic_read(&bitmap->pending_writes)==0);
793 bitmap_file_unmap(bitmap);
796 struct inode *inode = file->f_path.dentry->d_inode;
797 invalidate_mapping_pages(inode->i_mapping, 0, -1);
803 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
804 * then it is no longer reliable, so we stop using it and we mark the file
805 * as failed in the superblock
807 static void bitmap_file_kick(struct bitmap *bitmap)
809 char *path, *ptr = NULL;
811 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
812 bitmap_update_sb(bitmap);
815 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
817 ptr = d_path(&bitmap->file->f_path, path,
821 "%s: kicking failed bitmap file %s from array!\n",
822 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
827 "%s: disabling internal bitmap due to errors\n",
831 bitmap_file_put(bitmap);
836 enum bitmap_page_attr {
837 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
838 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
839 * i.e. counter is 1 or 2. */
840 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
843 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
844 enum bitmap_page_attr attr)
846 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
849 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
850 enum bitmap_page_attr attr)
852 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
855 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
856 enum bitmap_page_attr attr)
858 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
862 * bitmap_file_set_bit -- called before performing a write to the md device
863 * to set (and eventually sync) a particular bit in the bitmap file
865 * we set the bit immediately, then we record the page number so that
866 * when an unplug occurs, we can flush the dirty pages out to disk
868 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
873 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
875 if (!bitmap->filemap)
878 page = filemap_get_page(bitmap, chunk);
881 bit = file_page_offset(bitmap, chunk);
884 kaddr = kmap_atomic(page, KM_USER0);
885 if (bitmap->flags & BITMAP_HOSTENDIAN)
888 __set_bit_le(bit, kaddr);
889 kunmap_atomic(kaddr, KM_USER0);
890 pr_debug("set file bit %lu page %lu\n", bit, page->index);
891 /* record page number so it gets flushed to disk when unplug occurs */
892 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
895 /* this gets called when the md device is ready to unplug its underlying
896 * (slave) device queues -- before we let any writes go down, we need to
897 * sync the dirty pages of the bitmap file to disk */
898 void bitmap_unplug(struct bitmap *bitmap)
900 unsigned long i, flags;
901 int dirty, need_write;
908 /* look at each page to see if there are any set bits that need to be
909 * flushed out to disk */
910 for (i = 0; i < bitmap->file_pages; i++) {
911 spin_lock_irqsave(&bitmap->lock, flags);
912 if (!bitmap->filemap) {
913 spin_unlock_irqrestore(&bitmap->lock, flags);
916 page = bitmap->filemap[i];
917 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
918 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
919 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
920 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
923 spin_unlock_irqrestore(&bitmap->lock, flags);
925 if (dirty || need_write)
926 write_page(bitmap, page, 0);
928 if (wait) { /* if any writes were performed, we need to wait on them */
930 wait_event(bitmap->write_wait,
931 atomic_read(&bitmap->pending_writes)==0);
933 md_super_wait(bitmap->mddev);
935 if (bitmap->flags & BITMAP_WRITE_ERROR)
936 bitmap_file_kick(bitmap);
938 EXPORT_SYMBOL(bitmap_unplug);
940 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
941 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
942 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
943 * memory mapping of the bitmap file
945 * if there's no bitmap file, or if the bitmap file had been
946 * previously kicked from the array, we mark all the bits as
947 * 1's in order to cause a full resync.
949 * We ignore all bits for sectors that end earlier than 'start'.
950 * This is used when reading an out-of-date bitmap...
952 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
954 unsigned long i, chunks, index, oldindex, bit;
955 struct page *page = NULL, *oldpage = NULL;
956 unsigned long num_pages, bit_cnt = 0;
958 unsigned long bytes, offset;
963 chunks = bitmap->chunks;
966 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
968 outofdate = bitmap->flags & BITMAP_STALE;
970 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
971 "recovery\n", bmname(bitmap));
973 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
974 if (!bitmap->mddev->bitmap_info.external)
975 bytes += sizeof(bitmap_super_t);
977 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
979 if (file && i_size_read(file->f_mapping->host) < bytes) {
980 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
982 (unsigned long) i_size_read(file->f_mapping->host),
989 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
990 if (!bitmap->filemap)
993 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
994 bitmap->filemap_attr = kzalloc(
995 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
997 if (!bitmap->filemap_attr)
1002 for (i = 0; i < chunks; i++) {
1004 index = file_page_index(bitmap, i);
1005 bit = file_page_offset(bitmap, i);
1006 if (index != oldindex) { /* this is a new page, read it in */
1008 /* unmap the old page, we're done with it */
1009 if (index == num_pages-1)
1010 count = bytes - index * PAGE_SIZE;
1013 if (index == 0 && bitmap->sb_page) {
1015 * if we're here then the superblock page
1016 * contains some bits (PAGE_SIZE != sizeof sb)
1017 * we've already read it in, so just use it
1019 page = bitmap->sb_page;
1020 offset = sizeof(bitmap_super_t);
1022 page = read_sb_page(
1024 bitmap->mddev->bitmap_info.offset,
1028 page = read_page(file, index, bitmap, count);
1031 page = read_sb_page(bitmap->mddev,
1032 bitmap->mddev->bitmap_info.offset,
1037 if (IS_ERR(page)) { /* read error */
1038 ret = PTR_ERR(page);
1045 bitmap->filemap[bitmap->file_pages++] = page;
1046 bitmap->last_page_size = count;
1050 * if bitmap is out of date, dirty the
1051 * whole page and write it out
1053 paddr = kmap_atomic(page, KM_USER0);
1054 memset(paddr + offset, 0xff,
1055 PAGE_SIZE - offset);
1056 kunmap_atomic(paddr, KM_USER0);
1057 write_page(bitmap, page, 1);
1060 if (bitmap->flags & BITMAP_WRITE_ERROR)
1064 paddr = kmap_atomic(page, KM_USER0);
1065 if (bitmap->flags & BITMAP_HOSTENDIAN)
1066 b = test_bit(bit, paddr);
1068 b = test_bit_le(bit, paddr);
1069 kunmap_atomic(paddr, KM_USER0);
1071 /* if the disk bit is set, set the memory bit */
1072 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1074 bitmap_set_memory_bits(bitmap,
1075 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1081 /* everything went OK */
1083 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1085 if (bit_cnt) { /* Kick recovery if any bits were set */
1086 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1087 md_wakeup_thread(bitmap->mddev->thread);
1090 printk(KERN_INFO "%s: bitmap initialized from disk: "
1091 "read %lu/%lu pages, set %lu of %lu bits\n",
1092 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, chunks);
1097 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1098 bmname(bitmap), ret);
1102 void bitmap_write_all(struct bitmap *bitmap)
1104 /* We don't actually write all bitmap blocks here,
1105 * just flag them as needing to be written
1109 for (i = 0; i < bitmap->file_pages; i++)
1110 set_page_attr(bitmap, bitmap->filemap[i],
1111 BITMAP_PAGE_NEEDWRITE);
1112 bitmap->allclean = 0;
1115 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1117 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1118 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1119 bitmap->bp[page].count += inc;
1120 bitmap_checkfree(bitmap, page);
1122 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1123 sector_t offset, sector_t *blocks,
1127 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1131 void bitmap_daemon_work(struct mddev *mddev)
1133 struct bitmap *bitmap;
1135 unsigned long flags;
1136 struct page *page = NULL, *lastpage = NULL;
1140 /* Use a mutex to guard daemon_work against
1143 mutex_lock(&mddev->bitmap_info.mutex);
1144 bitmap = mddev->bitmap;
1145 if (bitmap == NULL) {
1146 mutex_unlock(&mddev->bitmap_info.mutex);
1149 if (time_before(jiffies, bitmap->daemon_lastrun
1150 + bitmap->mddev->bitmap_info.daemon_sleep))
1153 bitmap->daemon_lastrun = jiffies;
1154 if (bitmap->allclean) {
1155 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1158 bitmap->allclean = 1;
1160 spin_lock_irqsave(&bitmap->lock, flags);
1161 for (j = 0; j < bitmap->chunks; j++) {
1162 bitmap_counter_t *bmc;
1163 if (!bitmap->filemap)
1164 /* error or shutdown */
1167 page = filemap_get_page(bitmap, j);
1169 if (page != lastpage) {
1170 /* skip this page unless it's marked as needing cleaning */
1171 if (!test_page_attr(bitmap, page, BITMAP_PAGE_PENDING)) {
1172 int need_write = test_page_attr(bitmap, page,
1173 BITMAP_PAGE_NEEDWRITE);
1175 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1177 spin_unlock_irqrestore(&bitmap->lock, flags);
1179 write_page(bitmap, page, 0);
1180 spin_lock_irqsave(&bitmap->lock, flags);
1181 j |= (PAGE_BITS - 1);
1185 /* grab the new page, sync and release the old */
1186 if (lastpage != NULL) {
1187 if (test_page_attr(bitmap, lastpage,
1188 BITMAP_PAGE_NEEDWRITE)) {
1189 clear_page_attr(bitmap, lastpage,
1190 BITMAP_PAGE_NEEDWRITE);
1191 spin_unlock_irqrestore(&bitmap->lock, flags);
1192 write_page(bitmap, lastpage, 0);
1194 set_page_attr(bitmap, lastpage,
1195 BITMAP_PAGE_NEEDWRITE);
1196 bitmap->allclean = 0;
1197 spin_unlock_irqrestore(&bitmap->lock, flags);
1200 spin_unlock_irqrestore(&bitmap->lock, flags);
1203 /* We are possibly going to clear some bits, so make
1204 * sure that events_cleared is up-to-date.
1206 if (bitmap->need_sync &&
1207 bitmap->mddev->bitmap_info.external == 0) {
1209 bitmap->need_sync = 0;
1210 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1211 sb->events_cleared =
1212 cpu_to_le64(bitmap->events_cleared);
1213 kunmap_atomic(sb, KM_USER0);
1214 write_page(bitmap, bitmap->sb_page, 1);
1216 spin_lock_irqsave(&bitmap->lock, flags);
1217 if (!bitmap->need_sync)
1218 clear_page_attr(bitmap, page, BITMAP_PAGE_PENDING);
1220 bitmap->allclean = 0;
1222 bmc = bitmap_get_counter(bitmap,
1223 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1226 j |= PAGE_COUNTER_MASK;
1228 if (*bmc == 1 && !bitmap->need_sync) {
1229 /* we can clear the bit */
1231 bitmap_count_page(bitmap,
1232 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1236 paddr = kmap_atomic(page, KM_USER0);
1237 if (bitmap->flags & BITMAP_HOSTENDIAN)
1238 clear_bit(file_page_offset(bitmap, j),
1242 file_page_offset(bitmap,
1245 kunmap_atomic(paddr, KM_USER0);
1246 } else if (*bmc <= 2) {
1247 *bmc = 1; /* maybe clear the bit next time */
1248 set_page_attr(bitmap, page, BITMAP_PAGE_PENDING);
1249 bitmap->allclean = 0;
1253 spin_unlock_irqrestore(&bitmap->lock, flags);
1255 /* now sync the final page */
1256 if (lastpage != NULL) {
1257 spin_lock_irqsave(&bitmap->lock, flags);
1258 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1259 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1260 spin_unlock_irqrestore(&bitmap->lock, flags);
1261 write_page(bitmap, lastpage, 0);
1263 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1264 bitmap->allclean = 0;
1265 spin_unlock_irqrestore(&bitmap->lock, flags);
1270 if (bitmap->allclean == 0)
1271 bitmap->mddev->thread->timeout =
1272 bitmap->mddev->bitmap_info.daemon_sleep;
1273 mutex_unlock(&mddev->bitmap_info.mutex);
1276 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1277 sector_t offset, sector_t *blocks,
1279 __releases(bitmap->lock)
1280 __acquires(bitmap->lock)
1282 /* If 'create', we might release the lock and reclaim it.
1283 * The lock must have been taken with interrupts enabled.
1284 * If !create, we don't release the lock.
1286 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1287 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1288 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1292 err = bitmap_checkpage(bitmap, page, create);
1294 if (bitmap->bp[page].hijacked ||
1295 bitmap->bp[page].map == NULL)
1296 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1297 PAGE_COUNTER_SHIFT - 1);
1299 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1300 *blocks = csize - (offset & (csize - 1));
1305 /* now locked ... */
1307 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1308 /* should we use the first or second counter field
1309 * of the hijacked pointer? */
1310 int hi = (pageoff > PAGE_COUNTER_MASK);
1311 return &((bitmap_counter_t *)
1312 &bitmap->bp[page].map)[hi];
1313 } else /* page is allocated */
1314 return (bitmap_counter_t *)
1315 &(bitmap->bp[page].map[pageoff]);
1318 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1325 atomic_inc(&bitmap->behind_writes);
1326 bw = atomic_read(&bitmap->behind_writes);
1327 if (bw > bitmap->behind_writes_used)
1328 bitmap->behind_writes_used = bw;
1330 pr_debug("inc write-behind count %d/%lu\n",
1331 bw, bitmap->mddev->bitmap_info.max_write_behind);
1336 bitmap_counter_t *bmc;
1338 spin_lock_irq(&bitmap->lock);
1339 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1341 spin_unlock_irq(&bitmap->lock);
1345 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1346 DEFINE_WAIT(__wait);
1347 /* note that it is safe to do the prepare_to_wait
1348 * after the test as long as we do it before dropping
1351 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1352 TASK_UNINTERRUPTIBLE);
1353 spin_unlock_irq(&bitmap->lock);
1355 finish_wait(&bitmap->overflow_wait, &__wait);
1361 bitmap_file_set_bit(bitmap, offset);
1362 bitmap_count_page(bitmap, offset, 1);
1370 spin_unlock_irq(&bitmap->lock);
1373 if (sectors > blocks)
1380 EXPORT_SYMBOL(bitmap_startwrite);
1382 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1383 int success, int behind)
1388 if (atomic_dec_and_test(&bitmap->behind_writes))
1389 wake_up(&bitmap->behind_wait);
1390 pr_debug("dec write-behind count %d/%lu\n",
1391 atomic_read(&bitmap->behind_writes),
1392 bitmap->mddev->bitmap_info.max_write_behind);
1394 if (bitmap->mddev->degraded)
1395 /* Never clear bits or update events_cleared when degraded */
1400 unsigned long flags;
1401 bitmap_counter_t *bmc;
1403 spin_lock_irqsave(&bitmap->lock, flags);
1404 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1406 spin_unlock_irqrestore(&bitmap->lock, flags);
1411 bitmap->events_cleared < bitmap->mddev->events) {
1412 bitmap->events_cleared = bitmap->mddev->events;
1413 bitmap->need_sync = 1;
1414 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1417 if (!success && !NEEDED(*bmc))
1418 *bmc |= NEEDED_MASK;
1420 if (COUNTER(*bmc) == COUNTER_MAX)
1421 wake_up(&bitmap->overflow_wait);
1425 set_page_attr(bitmap,
1428 offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1429 BITMAP_PAGE_PENDING);
1430 bitmap->allclean = 0;
1432 spin_unlock_irqrestore(&bitmap->lock, flags);
1434 if (sectors > blocks)
1440 EXPORT_SYMBOL(bitmap_endwrite);
1442 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1445 bitmap_counter_t *bmc;
1447 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1449 return 1; /* always resync if no bitmap */
1451 spin_lock_irq(&bitmap->lock);
1452 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1458 else if (NEEDED(*bmc)) {
1460 if (!degraded) { /* don't set/clear bits if degraded */
1461 *bmc |= RESYNC_MASK;
1462 *bmc &= ~NEEDED_MASK;
1466 spin_unlock_irq(&bitmap->lock);
1470 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1473 /* bitmap_start_sync must always report on multiples of whole
1474 * pages, otherwise resync (which is very PAGE_SIZE based) will
1476 * So call __bitmap_start_sync repeatedly (if needed) until
1477 * At least PAGE_SIZE>>9 blocks are covered.
1478 * Return the 'or' of the result.
1484 while (*blocks < (PAGE_SIZE>>9)) {
1485 rv |= __bitmap_start_sync(bitmap, offset,
1486 &blocks1, degraded);
1492 EXPORT_SYMBOL(bitmap_start_sync);
1494 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1496 bitmap_counter_t *bmc;
1497 unsigned long flags;
1499 if (bitmap == NULL) {
1503 spin_lock_irqsave(&bitmap->lock, flags);
1504 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1509 *bmc &= ~RESYNC_MASK;
1511 if (!NEEDED(*bmc) && aborted)
1512 *bmc |= NEEDED_MASK;
1515 set_page_attr(bitmap,
1516 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1517 BITMAP_PAGE_PENDING);
1518 bitmap->allclean = 0;
1523 spin_unlock_irqrestore(&bitmap->lock, flags);
1525 EXPORT_SYMBOL(bitmap_end_sync);
1527 void bitmap_close_sync(struct bitmap *bitmap)
1529 /* Sync has finished, and any bitmap chunks that weren't synced
1530 * properly have been aborted. It remains to us to clear the
1531 * RESYNC bit wherever it is still on
1533 sector_t sector = 0;
1537 while (sector < bitmap->mddev->resync_max_sectors) {
1538 bitmap_end_sync(bitmap, sector, &blocks, 0);
1542 EXPORT_SYMBOL(bitmap_close_sync);
1544 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1552 bitmap->last_end_sync = jiffies;
1555 if (time_before(jiffies, (bitmap->last_end_sync
1556 + bitmap->mddev->bitmap_info.daemon_sleep)))
1558 wait_event(bitmap->mddev->recovery_wait,
1559 atomic_read(&bitmap->mddev->recovery_active) == 0);
1561 bitmap->mddev->curr_resync_completed = sector;
1562 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1563 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1565 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1566 bitmap_end_sync(bitmap, s, &blocks, 0);
1569 bitmap->last_end_sync = jiffies;
1570 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1572 EXPORT_SYMBOL(bitmap_cond_end_sync);
1574 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1576 /* For each chunk covered by any of these sectors, set the
1577 * counter to 1 and set resync_needed. They should all
1578 * be 0 at this point
1582 bitmap_counter_t *bmc;
1583 spin_lock_irq(&bitmap->lock);
1584 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1586 spin_unlock_irq(&bitmap->lock);
1591 *bmc = 1 | (needed ? NEEDED_MASK : 0);
1592 bitmap_count_page(bitmap, offset, 1);
1593 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1594 set_page_attr(bitmap, page, BITMAP_PAGE_PENDING);
1595 bitmap->allclean = 0;
1597 spin_unlock_irq(&bitmap->lock);
1600 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1601 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1603 unsigned long chunk;
1605 for (chunk = s; chunk <= e; chunk++) {
1606 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
1607 bitmap_set_memory_bits(bitmap, sec, 1);
1608 bitmap_file_set_bit(bitmap, sec);
1609 if (sec < bitmap->mddev->recovery_cp)
1610 /* We are asserting that the array is dirty,
1611 * so move the recovery_cp address back so
1612 * that it is obvious that it is dirty
1614 bitmap->mddev->recovery_cp = sec;
1619 * flush out any pending updates
1621 void bitmap_flush(struct mddev *mddev)
1623 struct bitmap *bitmap = mddev->bitmap;
1626 if (!bitmap) /* there was no bitmap */
1629 /* run the daemon_work three time to ensure everything is flushed
1632 sleep = mddev->bitmap_info.daemon_sleep * 2;
1633 bitmap->daemon_lastrun -= sleep;
1634 bitmap_daemon_work(mddev);
1635 bitmap->daemon_lastrun -= sleep;
1636 bitmap_daemon_work(mddev);
1637 bitmap->daemon_lastrun -= sleep;
1638 bitmap_daemon_work(mddev);
1639 bitmap_update_sb(bitmap);
1643 * free memory that was allocated
1645 static void bitmap_free(struct bitmap *bitmap)
1647 unsigned long k, pages;
1648 struct bitmap_page *bp;
1650 if (!bitmap) /* there was no bitmap */
1653 /* release the bitmap file and kill the daemon */
1654 bitmap_file_put(bitmap);
1657 pages = bitmap->pages;
1659 /* free all allocated memory */
1661 if (bp) /* deallocate the page memory */
1662 for (k = 0; k < pages; k++)
1663 if (bp[k].map && !bp[k].hijacked)
1669 void bitmap_destroy(struct mddev *mddev)
1671 struct bitmap *bitmap = mddev->bitmap;
1673 if (!bitmap) /* there was no bitmap */
1676 mutex_lock(&mddev->bitmap_info.mutex);
1677 mddev->bitmap = NULL; /* disconnect from the md device */
1678 mutex_unlock(&mddev->bitmap_info.mutex);
1680 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1682 if (bitmap->sysfs_can_clear)
1683 sysfs_put(bitmap->sysfs_can_clear);
1685 bitmap_free(bitmap);
1689 * initialize the bitmap structure
1690 * if this returns an error, bitmap_destroy must be called to do clean up
1692 int bitmap_create(struct mddev *mddev)
1694 struct bitmap *bitmap;
1695 sector_t blocks = mddev->resync_max_sectors;
1696 unsigned long chunks;
1697 unsigned long pages;
1698 struct file *file = mddev->bitmap_info.file;
1700 struct sysfs_dirent *bm = NULL;
1702 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1705 && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */
1708 BUG_ON(file && mddev->bitmap_info.offset);
1710 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1714 spin_lock_init(&bitmap->lock);
1715 atomic_set(&bitmap->pending_writes, 0);
1716 init_waitqueue_head(&bitmap->write_wait);
1717 init_waitqueue_head(&bitmap->overflow_wait);
1718 init_waitqueue_head(&bitmap->behind_wait);
1720 bitmap->mddev = mddev;
1723 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1725 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1728 bitmap->sysfs_can_clear = NULL;
1730 bitmap->file = file;
1733 /* As future accesses to this file will use bmap,
1734 * and bypass the page cache, we must sync the file
1739 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1740 if (!mddev->bitmap_info.external) {
1742 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1743 * instructing us to create a new on-disk bitmap instance.
1745 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1746 err = bitmap_new_disk_sb(bitmap);
1748 err = bitmap_read_sb(bitmap);
1751 if (mddev->bitmap_info.chunksize == 0 ||
1752 mddev->bitmap_info.daemon_sleep == 0)
1753 /* chunksize and time_base need to be
1760 bitmap->daemon_lastrun = jiffies;
1761 bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize);
1763 /* now that chunksize and chunkshift are set, we can use these macros */
1764 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1765 CHUNK_BLOCK_SHIFT(bitmap);
1766 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1770 bitmap->chunks = chunks;
1771 bitmap->pages = pages;
1772 bitmap->missing_pages = pages;
1774 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1780 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1781 pages, bmname(bitmap));
1783 mddev->bitmap = bitmap;
1786 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1789 bitmap_free(bitmap);
1793 int bitmap_load(struct mddev *mddev)
1797 sector_t sector = 0;
1798 struct bitmap *bitmap = mddev->bitmap;
1803 /* Clear out old bitmap info first: Either there is none, or we
1804 * are resuming after someone else has possibly changed things,
1805 * so we should forget old cached info.
1806 * All chunks should be clean, but some might need_sync.
1808 while (sector < mddev->resync_max_sectors) {
1810 bitmap_start_sync(bitmap, sector, &blocks, 0);
1813 bitmap_close_sync(bitmap);
1815 if (mddev->degraded == 0
1816 || bitmap->events_cleared == mddev->events)
1817 /* no need to keep dirty bits to optimise a
1818 * re-add of a missing device */
1819 start = mddev->recovery_cp;
1821 err = bitmap_init_from_disk(bitmap, start);
1826 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1827 md_wakeup_thread(mddev->thread);
1829 bitmap_update_sb(bitmap);
1831 if (bitmap->flags & BITMAP_WRITE_ERROR)
1836 EXPORT_SYMBOL_GPL(bitmap_load);
1839 location_show(struct mddev *mddev, char *page)
1842 if (mddev->bitmap_info.file)
1843 len = sprintf(page, "file");
1844 else if (mddev->bitmap_info.offset)
1845 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1847 len = sprintf(page, "none");
1848 len += sprintf(page+len, "\n");
1853 location_store(struct mddev *mddev, const char *buf, size_t len)
1857 if (!mddev->pers->quiesce)
1859 if (mddev->recovery || mddev->sync_thread)
1863 if (mddev->bitmap || mddev->bitmap_info.file ||
1864 mddev->bitmap_info.offset) {
1865 /* bitmap already configured. Only option is to clear it */
1866 if (strncmp(buf, "none", 4) != 0)
1869 mddev->pers->quiesce(mddev, 1);
1870 bitmap_destroy(mddev);
1871 mddev->pers->quiesce(mddev, 0);
1873 mddev->bitmap_info.offset = 0;
1874 if (mddev->bitmap_info.file) {
1875 struct file *f = mddev->bitmap_info.file;
1876 mddev->bitmap_info.file = NULL;
1877 restore_bitmap_write_access(f);
1881 /* No bitmap, OK to set a location */
1883 if (strncmp(buf, "none", 4) == 0)
1884 /* nothing to be done */;
1885 else if (strncmp(buf, "file:", 5) == 0) {
1886 /* Not supported yet */
1891 rv = strict_strtoll(buf+1, 10, &offset);
1893 rv = strict_strtoll(buf, 10, &offset);
1898 if (mddev->bitmap_info.external == 0 &&
1899 mddev->major_version == 0 &&
1900 offset != mddev->bitmap_info.default_offset)
1902 mddev->bitmap_info.offset = offset;
1904 mddev->pers->quiesce(mddev, 1);
1905 rv = bitmap_create(mddev);
1907 bitmap_destroy(mddev);
1908 mddev->bitmap_info.offset = 0;
1910 mddev->pers->quiesce(mddev, 0);
1916 if (!mddev->external) {
1917 /* Ensure new bitmap info is stored in
1918 * metadata promptly.
1920 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1921 md_wakeup_thread(mddev->thread);
1926 static struct md_sysfs_entry bitmap_location =
1927 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1930 timeout_show(struct mddev *mddev, char *page)
1933 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1934 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1936 len = sprintf(page, "%lu", secs);
1938 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1939 len += sprintf(page+len, "\n");
1944 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1946 /* timeout can be set at any time */
1947 unsigned long timeout;
1948 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1952 /* just to make sure we don't overflow... */
1953 if (timeout >= LONG_MAX / HZ)
1956 timeout = timeout * HZ / 10000;
1958 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1959 timeout = MAX_SCHEDULE_TIMEOUT-1;
1962 mddev->bitmap_info.daemon_sleep = timeout;
1963 if (mddev->thread) {
1964 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1965 * the bitmap is all clean and we don't need to
1966 * adjust the timeout right now
1968 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1969 mddev->thread->timeout = timeout;
1970 md_wakeup_thread(mddev->thread);
1976 static struct md_sysfs_entry bitmap_timeout =
1977 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1980 backlog_show(struct mddev *mddev, char *page)
1982 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1986 backlog_store(struct mddev *mddev, const char *buf, size_t len)
1988 unsigned long backlog;
1989 int rv = strict_strtoul(buf, 10, &backlog);
1992 if (backlog > COUNTER_MAX)
1994 mddev->bitmap_info.max_write_behind = backlog;
1998 static struct md_sysfs_entry bitmap_backlog =
1999 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2002 chunksize_show(struct mddev *mddev, char *page)
2004 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2008 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2010 /* Can only be changed when no bitmap is active */
2012 unsigned long csize;
2015 rv = strict_strtoul(buf, 10, &csize);
2019 !is_power_of_2(csize))
2021 mddev->bitmap_info.chunksize = csize;
2025 static struct md_sysfs_entry bitmap_chunksize =
2026 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2028 static ssize_t metadata_show(struct mddev *mddev, char *page)
2030 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2031 ? "external" : "internal"));
2034 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2036 if (mddev->bitmap ||
2037 mddev->bitmap_info.file ||
2038 mddev->bitmap_info.offset)
2040 if (strncmp(buf, "external", 8) == 0)
2041 mddev->bitmap_info.external = 1;
2042 else if (strncmp(buf, "internal", 8) == 0)
2043 mddev->bitmap_info.external = 0;
2049 static struct md_sysfs_entry bitmap_metadata =
2050 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2052 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2056 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2059 len = sprintf(page, "\n");
2063 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2065 if (mddev->bitmap == NULL)
2067 if (strncmp(buf, "false", 5) == 0)
2068 mddev->bitmap->need_sync = 1;
2069 else if (strncmp(buf, "true", 4) == 0) {
2070 if (mddev->degraded)
2072 mddev->bitmap->need_sync = 0;
2078 static struct md_sysfs_entry bitmap_can_clear =
2079 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2082 behind_writes_used_show(struct mddev *mddev, char *page)
2084 if (mddev->bitmap == NULL)
2085 return sprintf(page, "0\n");
2086 return sprintf(page, "%lu\n",
2087 mddev->bitmap->behind_writes_used);
2091 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2094 mddev->bitmap->behind_writes_used = 0;
2098 static struct md_sysfs_entry max_backlog_used =
2099 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2100 behind_writes_used_show, behind_writes_used_reset);
2102 static struct attribute *md_bitmap_attrs[] = {
2103 &bitmap_location.attr,
2104 &bitmap_timeout.attr,
2105 &bitmap_backlog.attr,
2106 &bitmap_chunksize.attr,
2107 &bitmap_metadata.attr,
2108 &bitmap_can_clear.attr,
2109 &max_backlog_used.attr,
2112 struct attribute_group md_bitmap_group = {
2114 .attrs = md_bitmap_attrs,