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).
16 * wait if count gets too high, wake when it drops to half.
19 #include <linux/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/timer.h>
25 #include <linux/sched.h>
26 #include <linux/list.h>
27 #include <linux/file.h>
28 #include <linux/mount.h>
29 #include <linux/buffer_head.h>
38 /* these are for debugging purposes only! */
40 /* define one and only one of these */
41 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
42 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
43 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
44 #define INJECT_FAULTS_4 0 /* undef */
45 #define INJECT_FAULTS_5 0 /* undef */
46 #define INJECT_FAULTS_6 0
48 /* if these are defined, the driver will fail! debug only */
49 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
50 #define INJECT_FATAL_FAULT_2 0 /* undef */
51 #define INJECT_FATAL_FAULT_3 0 /* undef */
54 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
55 #define DPRINTK(x...) do { } while(0)
59 # define PRINTK(x...) printk(KERN_DEBUG x)
65 static inline char * bmname(struct bitmap *bitmap)
67 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
72 * just a placeholder - calls kmalloc for bitmap pages
74 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
78 #ifdef INJECT_FAULTS_1
81 page = kmalloc(PAGE_SIZE, GFP_NOIO);
84 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
86 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
87 bmname(bitmap), page);
92 * for now just a placeholder -- just calls kfree for bitmap pages
94 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
96 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
101 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
103 * 1) check to see if this page is allocated, if it's not then try to alloc
104 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
105 * page pointer directly as a counter
107 * if we find our page, we increment the page's refcount so that it stays
108 * allocated while we're using it
110 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
111 __releases(bitmap->lock)
112 __acquires(bitmap->lock)
114 unsigned char *mappage;
116 if (page >= bitmap->pages) {
117 /* This can happen if bitmap_start_sync goes beyond
118 * End-of-device while looking for a whole page.
125 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
128 if (bitmap->bp[page].map) /* page is already allocated, just return */
134 spin_unlock_irq(&bitmap->lock);
136 /* this page has not been allocated yet */
138 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
139 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
141 /* failed - set the hijacked flag so that we can use the
142 * pointer as a counter */
143 spin_lock_irq(&bitmap->lock);
144 if (!bitmap->bp[page].map)
145 bitmap->bp[page].hijacked = 1;
151 spin_lock_irq(&bitmap->lock);
153 /* recheck the page */
155 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
156 /* somebody beat us to getting the page */
157 bitmap_free_page(bitmap, mappage);
161 /* no page was in place and we have one, so install it */
163 memset(mappage, 0, PAGE_SIZE);
164 bitmap->bp[page].map = mappage;
165 bitmap->missing_pages--;
171 /* if page is completely empty, put it back on the free list, or dealloc it */
172 /* if page was hijacked, unmark the flag so it might get alloced next time */
173 /* Note: lock should be held when calling this */
174 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
178 if (bitmap->bp[page].count) /* page is still busy */
181 /* page is no longer in use, it can be released */
183 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
184 bitmap->bp[page].hijacked = 0;
185 bitmap->bp[page].map = NULL;
189 /* normal case, free the page */
192 /* actually ... let's not. We will probably need the page again exactly when
193 * memory is tight and we are flusing to disk
197 ptr = bitmap->bp[page].map;
198 bitmap->bp[page].map = NULL;
199 bitmap->missing_pages++;
200 bitmap_free_page(bitmap, ptr);
207 * bitmap file handling - read and write the bitmap file and its superblock
211 * basic page I/O operations
214 /* IO operations when bitmap is stored near all superblocks */
215 static struct page *read_sb_page(mddev_t *mddev, loff_t offset,
217 unsigned long index, int size)
219 /* choose a good rdev and read the page from there */
225 page = alloc_page(GFP_KERNEL);
227 return ERR_PTR(-ENOMEM);
229 list_for_each_entry(rdev, &mddev->disks, same_set) {
230 if (! test_bit(In_sync, &rdev->flags)
231 || test_bit(Faulty, &rdev->flags))
234 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
236 if (sync_page_io(rdev->bdev, target,
237 roundup(size, bdev_logical_block_size(rdev->bdev)),
240 attach_page_buffers(page, NULL); /* so that free_buffer will
245 return ERR_PTR(-EIO);
249 static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
251 /* Iterate the disks of an mddev, using rcu to protect access to the
252 * linked list, and raising the refcount of devices we return to ensure
253 * they don't disappear while in use.
254 * As devices are only added or removed when raid_disk is < 0 and
255 * nr_pending is 0 and In_sync is clear, the entries we return will
256 * still be in the same position on the list when we re-enter
257 * list_for_each_continue_rcu.
259 struct list_head *pos;
262 /* start at the beginning */
265 /* release the previous rdev and start from there. */
266 rdev_dec_pending(rdev, mddev);
267 pos = &rdev->same_set;
269 list_for_each_continue_rcu(pos, &mddev->disks) {
270 rdev = list_entry(pos, mdk_rdev_t, same_set);
271 if (rdev->raid_disk >= 0 &&
272 !test_bit(Faulty, &rdev->flags)) {
273 /* this is a usable devices */
274 atomic_inc(&rdev->nr_pending);
283 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
285 mdk_rdev_t *rdev = NULL;
286 mddev_t *mddev = bitmap->mddev;
288 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
289 int size = PAGE_SIZE;
290 loff_t offset = mddev->bitmap_info.offset;
291 if (page->index == bitmap->file_pages-1)
292 size = roundup(bitmap->last_page_size,
293 bdev_logical_block_size(rdev->bdev));
294 /* Just make sure we aren't corrupting data or
297 if (mddev->external) {
298 /* Bitmap could be anywhere. */
299 if (rdev->sb_start + offset + (page->index *(PAGE_SIZE/512)) >
301 rdev->sb_start + offset <
302 rdev->data_offset + mddev->dev_sectors +
305 } else if (offset < 0) {
306 /* DATA BITMAP METADATA */
308 + (long)(page->index * (PAGE_SIZE/512))
310 /* bitmap runs in to metadata */
312 if (rdev->data_offset + mddev->dev_sectors
313 > rdev->sb_start + offset)
314 /* data runs in to bitmap */
316 } else if (rdev->sb_start < rdev->data_offset) {
317 /* METADATA BITMAP DATA */
320 + page->index*(PAGE_SIZE/512) + size/512
322 /* bitmap runs in to data */
325 /* DATA METADATA BITMAP - no problems */
327 md_super_write(mddev, rdev,
328 rdev->sb_start + offset
329 + page->index * (PAGE_SIZE/512),
335 md_super_wait(mddev);
342 static void bitmap_file_kick(struct bitmap *bitmap);
344 * write out a page to a file
346 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
348 struct buffer_head *bh;
350 if (bitmap->file == NULL) {
351 switch (write_sb_page(bitmap, page, wait)) {
353 bitmap->flags |= BITMAP_WRITE_ERROR;
357 bh = page_buffers(page);
359 while (bh && bh->b_blocknr) {
360 atomic_inc(&bitmap->pending_writes);
361 set_buffer_locked(bh);
362 set_buffer_mapped(bh);
363 submit_bh(WRITE, bh);
364 bh = bh->b_this_page;
368 wait_event(bitmap->write_wait,
369 atomic_read(&bitmap->pending_writes)==0);
372 if (bitmap->flags & BITMAP_WRITE_ERROR)
373 bitmap_file_kick(bitmap);
376 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
378 struct bitmap *bitmap = bh->b_private;
382 spin_lock_irqsave(&bitmap->lock, flags);
383 bitmap->flags |= BITMAP_WRITE_ERROR;
384 spin_unlock_irqrestore(&bitmap->lock, flags);
386 if (atomic_dec_and_test(&bitmap->pending_writes))
387 wake_up(&bitmap->write_wait);
390 /* copied from buffer.c */
392 __clear_page_buffers(struct page *page)
394 ClearPagePrivate(page);
395 set_page_private(page, 0);
396 page_cache_release(page);
398 static void free_buffers(struct page *page)
400 struct buffer_head *bh = page_buffers(page);
403 struct buffer_head *next = bh->b_this_page;
404 free_buffer_head(bh);
407 __clear_page_buffers(page);
411 /* read a page from a file.
412 * We both read the page, and attach buffers to the page to record the
413 * address of each block (using bmap). These addresses will be used
414 * to write the block later, completely bypassing the filesystem.
415 * This usage is similar to how swap files are handled, and allows us
416 * to write to a file with no concerns of memory allocation failing.
418 static struct page *read_page(struct file *file, unsigned long index,
419 struct bitmap *bitmap,
422 struct page *page = NULL;
423 struct inode *inode = file->f_path.dentry->d_inode;
424 struct buffer_head *bh;
427 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
428 (unsigned long long)index << PAGE_SHIFT);
430 page = alloc_page(GFP_KERNEL);
432 page = ERR_PTR(-ENOMEM);
436 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
439 page = ERR_PTR(-ENOMEM);
442 attach_page_buffers(page, bh);
443 block = index << (PAGE_SHIFT - inode->i_blkbits);
448 bh->b_blocknr = bmap(inode, block);
449 if (bh->b_blocknr == 0) {
450 /* Cannot use this file! */
452 page = ERR_PTR(-EINVAL);
455 bh->b_bdev = inode->i_sb->s_bdev;
456 if (count < (1<<inode->i_blkbits))
459 count -= (1<<inode->i_blkbits);
461 bh->b_end_io = end_bitmap_write;
462 bh->b_private = bitmap;
463 atomic_inc(&bitmap->pending_writes);
464 set_buffer_locked(bh);
465 set_buffer_mapped(bh);
469 bh = bh->b_this_page;
473 wait_event(bitmap->write_wait,
474 atomic_read(&bitmap->pending_writes)==0);
475 if (bitmap->flags & BITMAP_WRITE_ERROR) {
477 page = ERR_PTR(-EIO);
481 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
483 (unsigned long long)index << PAGE_SHIFT,
489 * bitmap file superblock operations
492 /* update the event counter and sync the superblock to disk */
493 void bitmap_update_sb(struct bitmap *bitmap)
498 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
500 spin_lock_irqsave(&bitmap->lock, flags);
501 if (!bitmap->sb_page) { /* no superblock */
502 spin_unlock_irqrestore(&bitmap->lock, flags);
505 spin_unlock_irqrestore(&bitmap->lock, flags);
506 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
507 sb->events = cpu_to_le64(bitmap->mddev->events);
508 if (bitmap->mddev->events < bitmap->events_cleared) {
509 /* rocking back to read-only */
510 bitmap->events_cleared = bitmap->mddev->events;
511 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
513 /* Just in case these have been changed via sysfs: */
514 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
515 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
516 kunmap_atomic(sb, KM_USER0);
517 write_page(bitmap, bitmap->sb_page, 1);
520 /* print out the bitmap file superblock */
521 void bitmap_print_sb(struct bitmap *bitmap)
525 if (!bitmap || !bitmap->sb_page)
527 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
528 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
529 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
530 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
531 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
532 *(__u32 *)(sb->uuid+0),
533 *(__u32 *)(sb->uuid+4),
534 *(__u32 *)(sb->uuid+8),
535 *(__u32 *)(sb->uuid+12));
536 printk(KERN_DEBUG " events: %llu\n",
537 (unsigned long long) le64_to_cpu(sb->events));
538 printk(KERN_DEBUG "events cleared: %llu\n",
539 (unsigned long long) le64_to_cpu(sb->events_cleared));
540 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
541 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
542 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
543 printk(KERN_DEBUG " sync size: %llu KB\n",
544 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
545 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
546 kunmap_atomic(sb, KM_USER0);
549 /* read the superblock from the bitmap file and initialize some bitmap fields */
550 static int bitmap_read_sb(struct bitmap *bitmap)
554 unsigned long chunksize, daemon_sleep, write_behind;
555 unsigned long long events;
558 /* page 0 is the superblock, read it... */
560 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
561 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
563 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
565 bitmap->sb_page = read_sb_page(bitmap->mddev,
566 bitmap->mddev->bitmap_info.offset,
568 0, sizeof(bitmap_super_t));
570 if (IS_ERR(bitmap->sb_page)) {
571 err = PTR_ERR(bitmap->sb_page);
572 bitmap->sb_page = NULL;
576 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
578 chunksize = le32_to_cpu(sb->chunksize);
579 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
580 write_behind = le32_to_cpu(sb->write_behind);
582 /* verify that the bitmap-specific fields are valid */
583 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
584 reason = "bad magic";
585 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
586 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
587 reason = "unrecognized superblock version";
588 else if (chunksize < 512)
589 reason = "bitmap chunksize too small";
590 else if ((1 << ffz(~chunksize)) != chunksize)
591 reason = "bitmap chunksize not a power of 2";
592 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
593 reason = "daemon sleep period out of range";
594 else if (write_behind > COUNTER_MAX)
595 reason = "write-behind limit out of range (0 - 16383)";
597 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
598 bmname(bitmap), reason);
602 /* keep the array size field of the bitmap superblock up to date */
603 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
605 if (!bitmap->mddev->persistent)
609 * if we have a persistent array superblock, compare the
610 * bitmap's UUID and event counter to the mddev's
612 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
613 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
617 events = le64_to_cpu(sb->events);
618 if (events < bitmap->mddev->events) {
619 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
620 "-- forcing full recovery\n", bmname(bitmap), events,
621 (unsigned long long) bitmap->mddev->events);
622 sb->state |= cpu_to_le32(BITMAP_STALE);
625 /* assign fields using values from superblock */
626 bitmap->mddev->bitmap_info.chunksize = chunksize;
627 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
628 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
629 bitmap->flags |= le32_to_cpu(sb->state);
630 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
631 bitmap->flags |= BITMAP_HOSTENDIAN;
632 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
633 if (sb->state & cpu_to_le32(BITMAP_STALE))
634 bitmap->events_cleared = bitmap->mddev->events;
637 kunmap_atomic(sb, KM_USER0);
639 bitmap_print_sb(bitmap);
643 enum bitmap_mask_op {
648 /* record the state of the bitmap in the superblock. Return the old value */
649 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
650 enum bitmap_mask_op op)
656 spin_lock_irqsave(&bitmap->lock, flags);
657 if (!bitmap->sb_page) { /* can't set the state */
658 spin_unlock_irqrestore(&bitmap->lock, flags);
661 spin_unlock_irqrestore(&bitmap->lock, flags);
662 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
663 old = le32_to_cpu(sb->state) & bits;
665 case MASK_SET: sb->state |= cpu_to_le32(bits);
667 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
671 kunmap_atomic(sb, KM_USER0);
676 * general bitmap file operations
679 /* calculate the index of the page that contains this bit */
680 static inline unsigned long file_page_index(unsigned long chunk)
682 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
685 /* calculate the (bit) offset of this bit within a page */
686 static inline unsigned long file_page_offset(unsigned long chunk)
688 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
692 * return a pointer to the page in the filemap that contains the given bit
694 * this lookup is complicated by the fact that the bitmap sb might be exactly
695 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
698 static inline struct page *filemap_get_page(struct bitmap *bitmap,
701 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
702 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
706 static void bitmap_file_unmap(struct bitmap *bitmap)
708 struct page **map, *sb_page;
713 spin_lock_irqsave(&bitmap->lock, flags);
714 map = bitmap->filemap;
715 bitmap->filemap = NULL;
716 attr = bitmap->filemap_attr;
717 bitmap->filemap_attr = NULL;
718 pages = bitmap->file_pages;
719 bitmap->file_pages = 0;
720 sb_page = bitmap->sb_page;
721 bitmap->sb_page = NULL;
722 spin_unlock_irqrestore(&bitmap->lock, flags);
725 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
726 free_buffers(map[pages]);
731 free_buffers(sb_page);
734 static void bitmap_file_put(struct bitmap *bitmap)
739 spin_lock_irqsave(&bitmap->lock, flags);
742 spin_unlock_irqrestore(&bitmap->lock, flags);
745 wait_event(bitmap->write_wait,
746 atomic_read(&bitmap->pending_writes)==0);
747 bitmap_file_unmap(bitmap);
750 struct inode *inode = file->f_path.dentry->d_inode;
751 invalidate_mapping_pages(inode->i_mapping, 0, -1);
758 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
759 * then it is no longer reliable, so we stop using it and we mark the file
760 * as failed in the superblock
762 static void bitmap_file_kick(struct bitmap *bitmap)
764 char *path, *ptr = NULL;
766 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
767 bitmap_update_sb(bitmap);
770 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
772 ptr = d_path(&bitmap->file->f_path, path,
777 "%s: kicking failed bitmap file %s from array!\n",
778 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
783 "%s: disabling internal bitmap due to errors\n",
787 bitmap_file_put(bitmap);
792 enum bitmap_page_attr {
793 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
794 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
795 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
798 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
799 enum bitmap_page_attr attr)
801 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
804 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
805 enum bitmap_page_attr attr)
807 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
810 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
811 enum bitmap_page_attr attr)
813 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
817 * bitmap_file_set_bit -- called before performing a write to the md device
818 * to set (and eventually sync) a particular bit in the bitmap file
820 * we set the bit immediately, then we record the page number so that
821 * when an unplug occurs, we can flush the dirty pages out to disk
823 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
828 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
830 if (!bitmap->filemap) {
834 page = filemap_get_page(bitmap, chunk);
836 bit = file_page_offset(chunk);
839 kaddr = kmap_atomic(page, KM_USER0);
840 if (bitmap->flags & BITMAP_HOSTENDIAN)
843 ext2_set_bit(bit, kaddr);
844 kunmap_atomic(kaddr, KM_USER0);
845 PRINTK("set file bit %lu page %lu\n", bit, page->index);
847 /* record page number so it gets flushed to disk when unplug occurs */
848 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
852 /* this gets called when the md device is ready to unplug its underlying
853 * (slave) device queues -- before we let any writes go down, we need to
854 * sync the dirty pages of the bitmap file to disk */
855 void bitmap_unplug(struct bitmap *bitmap)
857 unsigned long i, flags;
858 int dirty, need_write;
865 /* look at each page to see if there are any set bits that need to be
866 * flushed out to disk */
867 for (i = 0; i < bitmap->file_pages; i++) {
868 spin_lock_irqsave(&bitmap->lock, flags);
869 if (!bitmap->filemap) {
870 spin_unlock_irqrestore(&bitmap->lock, flags);
873 page = bitmap->filemap[i];
874 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
875 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
876 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
877 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
880 spin_unlock_irqrestore(&bitmap->lock, flags);
882 if (dirty | need_write)
883 write_page(bitmap, page, 0);
885 if (wait) { /* if any writes were performed, we need to wait on them */
887 wait_event(bitmap->write_wait,
888 atomic_read(&bitmap->pending_writes)==0);
890 md_super_wait(bitmap->mddev);
892 if (bitmap->flags & BITMAP_WRITE_ERROR)
893 bitmap_file_kick(bitmap);
896 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
897 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
898 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
899 * memory mapping of the bitmap file
901 * if there's no bitmap file, or if the bitmap file had been
902 * previously kicked from the array, we mark all the bits as
903 * 1's in order to cause a full resync.
905 * We ignore all bits for sectors that end earlier than 'start'.
906 * This is used when reading an out-of-date bitmap...
908 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
910 unsigned long i, chunks, index, oldindex, bit;
911 struct page *page = NULL, *oldpage = NULL;
912 unsigned long num_pages, bit_cnt = 0;
914 unsigned long bytes, offset;
919 chunks = bitmap->chunks;
922 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
924 #ifdef INJECT_FAULTS_3
927 outofdate = bitmap->flags & BITMAP_STALE;
930 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
931 "recovery\n", bmname(bitmap));
933 bytes = (chunks + 7) / 8;
935 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
937 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
938 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
940 (unsigned long) i_size_read(file->f_mapping->host),
941 bytes + sizeof(bitmap_super_t));
947 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
948 if (!bitmap->filemap)
951 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
952 bitmap->filemap_attr = kzalloc(
953 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
955 if (!bitmap->filemap_attr)
960 for (i = 0; i < chunks; i++) {
962 index = file_page_index(i);
963 bit = file_page_offset(i);
964 if (index != oldindex) { /* this is a new page, read it in */
966 /* unmap the old page, we're done with it */
967 if (index == num_pages-1)
968 count = bytes + sizeof(bitmap_super_t)
974 * if we're here then the superblock page
975 * contains some bits (PAGE_SIZE != sizeof sb)
976 * we've already read it in, so just use it
978 page = bitmap->sb_page;
979 offset = sizeof(bitmap_super_t);
981 read_sb_page(bitmap->mddev,
982 bitmap->mddev->bitmap_info.offset,
986 page = read_page(file, index, bitmap, count);
989 page = read_sb_page(bitmap->mddev,
990 bitmap->mddev->bitmap_info.offset,
995 if (IS_ERR(page)) { /* read error */
1003 bitmap->filemap[bitmap->file_pages++] = page;
1004 bitmap->last_page_size = count;
1008 * if bitmap is out of date, dirty the
1009 * whole page and write it out
1011 paddr = kmap_atomic(page, KM_USER0);
1012 memset(paddr + offset, 0xff,
1013 PAGE_SIZE - offset);
1014 kunmap_atomic(paddr, KM_USER0);
1015 write_page(bitmap, page, 1);
1018 if (bitmap->flags & BITMAP_WRITE_ERROR)
1022 paddr = kmap_atomic(page, KM_USER0);
1023 if (bitmap->flags & BITMAP_HOSTENDIAN)
1024 b = test_bit(bit, paddr);
1026 b = ext2_test_bit(bit, paddr);
1027 kunmap_atomic(paddr, KM_USER0);
1029 /* if the disk bit is set, set the memory bit */
1030 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1032 bitmap_set_memory_bits(bitmap,
1033 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1036 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1040 /* everything went OK */
1042 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1044 if (bit_cnt) { /* Kick recovery if any bits were set */
1045 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1046 md_wakeup_thread(bitmap->mddev->thread);
1049 printk(KERN_INFO "%s: bitmap initialized from disk: "
1050 "read %lu/%lu pages, set %lu bits\n",
1051 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1056 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1057 bmname(bitmap), ret);
1061 void bitmap_write_all(struct bitmap *bitmap)
1063 /* We don't actually write all bitmap blocks here,
1064 * just flag them as needing to be written
1068 for (i=0; i < bitmap->file_pages; i++)
1069 set_page_attr(bitmap, bitmap->filemap[i],
1070 BITMAP_PAGE_NEEDWRITE);
1074 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1076 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1077 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1078 bitmap->bp[page].count += inc;
1080 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1081 (unsigned long long)offset, inc, bitmap->bp[page].count);
1083 bitmap_checkfree(bitmap, page);
1085 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1086 sector_t offset, int *blocks,
1090 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1094 void bitmap_daemon_work(mddev_t *mddev)
1096 struct bitmap *bitmap;
1098 unsigned long flags;
1099 struct page *page = NULL, *lastpage = NULL;
1103 /* Use a mutex to guard daemon_work against
1106 mutex_lock(&mddev->bitmap_info.mutex);
1107 bitmap = mddev->bitmap;
1108 if (bitmap == NULL) {
1109 mutex_unlock(&mddev->bitmap_info.mutex);
1112 if (time_before(jiffies, bitmap->daemon_lastrun
1113 + bitmap->mddev->bitmap_info.daemon_sleep))
1116 bitmap->daemon_lastrun = jiffies;
1117 if (bitmap->allclean) {
1118 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1121 bitmap->allclean = 1;
1123 spin_lock_irqsave(&bitmap->lock, flags);
1124 for (j = 0; j < bitmap->chunks; j++) {
1125 bitmap_counter_t *bmc;
1126 if (!bitmap->filemap)
1127 /* error or shutdown */
1130 page = filemap_get_page(bitmap, j);
1132 if (page != lastpage) {
1133 /* skip this page unless it's marked as needing cleaning */
1134 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1135 int need_write = test_page_attr(bitmap, page,
1136 BITMAP_PAGE_NEEDWRITE);
1138 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1140 spin_unlock_irqrestore(&bitmap->lock, flags);
1142 write_page(bitmap, page, 0);
1143 bitmap->allclean = 0;
1145 spin_lock_irqsave(&bitmap->lock, flags);
1146 j |= (PAGE_BITS - 1);
1150 /* grab the new page, sync and release the old */
1151 if (lastpage != NULL) {
1152 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1153 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1154 spin_unlock_irqrestore(&bitmap->lock, flags);
1155 write_page(bitmap, lastpage, 0);
1157 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1158 spin_unlock_irqrestore(&bitmap->lock, flags);
1161 spin_unlock_irqrestore(&bitmap->lock, flags);
1164 /* We are possibly going to clear some bits, so make
1165 * sure that events_cleared is up-to-date.
1167 if (bitmap->need_sync) {
1169 bitmap->need_sync = 0;
1170 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1171 sb->events_cleared =
1172 cpu_to_le64(bitmap->events_cleared);
1173 kunmap_atomic(sb, KM_USER0);
1174 write_page(bitmap, bitmap->sb_page, 1);
1176 spin_lock_irqsave(&bitmap->lock, flags);
1177 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1179 bmc = bitmap_get_counter(bitmap,
1180 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1184 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1187 bitmap->allclean = 0;
1190 *bmc=1; /* maybe clear the bit next time */
1191 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1192 } else if (*bmc == 1) {
1193 /* we can clear the bit */
1195 bitmap_count_page(bitmap,
1196 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1200 paddr = kmap_atomic(page, KM_USER0);
1201 if (bitmap->flags & BITMAP_HOSTENDIAN)
1202 clear_bit(file_page_offset(j), paddr);
1204 ext2_clear_bit(file_page_offset(j), paddr);
1205 kunmap_atomic(paddr, KM_USER0);
1208 j |= PAGE_COUNTER_MASK;
1210 spin_unlock_irqrestore(&bitmap->lock, flags);
1212 /* now sync the final page */
1213 if (lastpage != NULL) {
1214 spin_lock_irqsave(&bitmap->lock, flags);
1215 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1216 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1217 spin_unlock_irqrestore(&bitmap->lock, flags);
1218 write_page(bitmap, lastpage, 0);
1220 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1221 spin_unlock_irqrestore(&bitmap->lock, flags);
1226 if (bitmap->allclean == 0)
1227 bitmap->mddev->thread->timeout =
1228 bitmap->mddev->bitmap_info.daemon_sleep;
1229 mutex_unlock(&mddev->bitmap_info.mutex);
1232 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1233 sector_t offset, int *blocks,
1235 __releases(bitmap->lock)
1236 __acquires(bitmap->lock)
1238 /* If 'create', we might release the lock and reclaim it.
1239 * The lock must have been taken with interrupts enabled.
1240 * If !create, we don't release the lock.
1242 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1243 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1244 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1247 if (bitmap_checkpage(bitmap, page, create) < 0) {
1248 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1249 *blocks = csize - (offset & (csize- 1));
1252 /* now locked ... */
1254 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1255 /* should we use the first or second counter field
1256 * of the hijacked pointer? */
1257 int hi = (pageoff > PAGE_COUNTER_MASK);
1258 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1259 PAGE_COUNTER_SHIFT - 1);
1260 *blocks = csize - (offset & (csize- 1));
1261 return &((bitmap_counter_t *)
1262 &bitmap->bp[page].map)[hi];
1263 } else { /* page is allocated */
1264 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1265 *blocks = csize - (offset & (csize- 1));
1266 return (bitmap_counter_t *)
1267 &(bitmap->bp[page].map[pageoff]);
1271 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1273 if (!bitmap) return 0;
1276 atomic_inc(&bitmap->behind_writes);
1277 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1278 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1283 bitmap_counter_t *bmc;
1285 spin_lock_irq(&bitmap->lock);
1286 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1288 spin_unlock_irq(&bitmap->lock);
1292 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1293 DEFINE_WAIT(__wait);
1294 /* note that it is safe to do the prepare_to_wait
1295 * after the test as long as we do it before dropping
1298 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1299 TASK_UNINTERRUPTIBLE);
1300 spin_unlock_irq(&bitmap->lock);
1301 blk_unplug(bitmap->mddev->queue);
1303 finish_wait(&bitmap->overflow_wait, &__wait);
1309 bitmap_file_set_bit(bitmap, offset);
1310 bitmap_count_page(bitmap,offset, 1);
1311 blk_plug_device_unlocked(bitmap->mddev->queue);
1319 spin_unlock_irq(&bitmap->lock);
1322 if (sectors > blocks)
1326 bitmap->allclean = 0;
1330 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1331 int success, int behind)
1333 if (!bitmap) return;
1335 atomic_dec(&bitmap->behind_writes);
1336 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1337 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1339 if (bitmap->mddev->degraded)
1340 /* Never clear bits or update events_cleared when degraded */
1345 unsigned long flags;
1346 bitmap_counter_t *bmc;
1348 spin_lock_irqsave(&bitmap->lock, flags);
1349 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1351 spin_unlock_irqrestore(&bitmap->lock, flags);
1356 bitmap->events_cleared < bitmap->mddev->events) {
1357 bitmap->events_cleared = bitmap->mddev->events;
1358 bitmap->need_sync = 1;
1361 if (!success && ! (*bmc & NEEDED_MASK))
1362 *bmc |= NEEDED_MASK;
1364 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1365 wake_up(&bitmap->overflow_wait);
1369 set_page_attr(bitmap,
1370 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1373 spin_unlock_irqrestore(&bitmap->lock, flags);
1375 if (sectors > blocks)
1381 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1384 bitmap_counter_t *bmc;
1386 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1388 return 1; /* always resync if no bitmap */
1390 spin_lock_irq(&bitmap->lock);
1391 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1397 else if (NEEDED(*bmc)) {
1399 if (!degraded) { /* don't set/clear bits if degraded */
1400 *bmc |= RESYNC_MASK;
1401 *bmc &= ~NEEDED_MASK;
1405 spin_unlock_irq(&bitmap->lock);
1406 bitmap->allclean = 0;
1410 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1413 /* bitmap_start_sync must always report on multiples of whole
1414 * pages, otherwise resync (which is very PAGE_SIZE based) will
1416 * So call __bitmap_start_sync repeatedly (if needed) until
1417 * At least PAGE_SIZE>>9 blocks are covered.
1418 * Return the 'or' of the result.
1424 while (*blocks < (PAGE_SIZE>>9)) {
1425 rv |= __bitmap_start_sync(bitmap, offset,
1426 &blocks1, degraded);
1433 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1435 bitmap_counter_t *bmc;
1436 unsigned long flags;
1438 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1439 */ if (bitmap == NULL) {
1443 spin_lock_irqsave(&bitmap->lock, flags);
1444 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1449 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1452 *bmc &= ~RESYNC_MASK;
1454 if (!NEEDED(*bmc) && aborted)
1455 *bmc |= NEEDED_MASK;
1458 set_page_attr(bitmap,
1459 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1465 spin_unlock_irqrestore(&bitmap->lock, flags);
1466 bitmap->allclean = 0;
1469 void bitmap_close_sync(struct bitmap *bitmap)
1471 /* Sync has finished, and any bitmap chunks that weren't synced
1472 * properly have been aborted. It remains to us to clear the
1473 * RESYNC bit wherever it is still on
1475 sector_t sector = 0;
1479 while (sector < bitmap->mddev->resync_max_sectors) {
1480 bitmap_end_sync(bitmap, sector, &blocks, 0);
1485 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1493 bitmap->last_end_sync = jiffies;
1496 if (time_before(jiffies, (bitmap->last_end_sync
1497 + bitmap->mddev->bitmap_info.daemon_sleep)))
1499 wait_event(bitmap->mddev->recovery_wait,
1500 atomic_read(&bitmap->mddev->recovery_active) == 0);
1502 bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync;
1503 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1504 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1506 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1507 bitmap_end_sync(bitmap, s, &blocks, 0);
1510 bitmap->last_end_sync = jiffies;
1511 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1514 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1516 /* For each chunk covered by any of these sectors, set the
1517 * counter to 1 and set resync_needed. They should all
1518 * be 0 at this point
1522 bitmap_counter_t *bmc;
1523 spin_lock_irq(&bitmap->lock);
1524 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1526 spin_unlock_irq(&bitmap->lock);
1531 *bmc = 1 | (needed?NEEDED_MASK:0);
1532 bitmap_count_page(bitmap, offset, 1);
1533 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1534 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1536 spin_unlock_irq(&bitmap->lock);
1537 bitmap->allclean = 0;
1540 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1541 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1543 unsigned long chunk;
1545 for (chunk = s; chunk <= e; chunk++) {
1546 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
1547 bitmap_set_memory_bits(bitmap, sec, 1);
1548 bitmap_file_set_bit(bitmap, sec);
1553 * flush out any pending updates
1555 void bitmap_flush(mddev_t *mddev)
1557 struct bitmap *bitmap = mddev->bitmap;
1560 if (!bitmap) /* there was no bitmap */
1563 /* run the daemon_work three time to ensure everything is flushed
1566 sleep = mddev->bitmap_info.daemon_sleep * 2;
1567 bitmap->daemon_lastrun -= sleep;
1568 bitmap_daemon_work(mddev);
1569 bitmap->daemon_lastrun -= sleep;
1570 bitmap_daemon_work(mddev);
1571 bitmap->daemon_lastrun -= sleep;
1572 bitmap_daemon_work(mddev);
1573 bitmap_update_sb(bitmap);
1577 * free memory that was allocated
1579 static void bitmap_free(struct bitmap *bitmap)
1581 unsigned long k, pages;
1582 struct bitmap_page *bp;
1584 if (!bitmap) /* there was no bitmap */
1587 /* release the bitmap file and kill the daemon */
1588 bitmap_file_put(bitmap);
1591 pages = bitmap->pages;
1593 /* free all allocated memory */
1595 if (bp) /* deallocate the page memory */
1596 for (k = 0; k < pages; k++)
1597 if (bp[k].map && !bp[k].hijacked)
1603 void bitmap_destroy(mddev_t *mddev)
1605 struct bitmap *bitmap = mddev->bitmap;
1607 if (!bitmap) /* there was no bitmap */
1610 mutex_lock(&mddev->bitmap_info.mutex);
1611 mddev->bitmap = NULL; /* disconnect from the md device */
1612 mutex_unlock(&mddev->bitmap_info.mutex);
1614 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1616 bitmap_free(bitmap);
1620 * initialize the bitmap structure
1621 * if this returns an error, bitmap_destroy must be called to do clean up
1623 int bitmap_create(mddev_t *mddev)
1625 struct bitmap *bitmap;
1626 sector_t blocks = mddev->resync_max_sectors;
1627 unsigned long chunks;
1628 unsigned long pages;
1629 struct file *file = mddev->bitmap_info.file;
1633 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1635 if (!file && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */
1638 BUG_ON(file && mddev->bitmap_info.offset);
1640 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1644 spin_lock_init(&bitmap->lock);
1645 atomic_set(&bitmap->pending_writes, 0);
1646 init_waitqueue_head(&bitmap->write_wait);
1647 init_waitqueue_head(&bitmap->overflow_wait);
1649 bitmap->mddev = mddev;
1651 bitmap->file = file;
1654 /* As future accesses to this file will use bmap,
1655 * and bypass the page cache, we must sync the file
1658 vfs_fsync(file, file->f_dentry, 1);
1660 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1661 err = bitmap_read_sb(bitmap);
1665 bitmap->daemon_lastrun = jiffies;
1666 bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize);
1668 /* now that chunksize and chunkshift are set, we can use these macros */
1669 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1670 CHUNK_BLOCK_SHIFT(bitmap);
1671 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1675 bitmap->chunks = chunks;
1676 bitmap->pages = pages;
1677 bitmap->missing_pages = pages;
1678 bitmap->counter_bits = COUNTER_BITS;
1680 bitmap->syncchunk = ~0UL;
1682 #ifdef INJECT_FATAL_FAULT_1
1685 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1691 /* now that we have some pages available, initialize the in-memory
1692 * bitmap from the on-disk bitmap */
1694 if (mddev->degraded == 0
1695 || bitmap->events_cleared == mddev->events)
1696 /* no need to keep dirty bits to optimise a re-add of a missing device */
1697 start = mddev->recovery_cp;
1698 err = bitmap_init_from_disk(bitmap, start);
1703 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1704 pages, bmname(bitmap));
1706 mddev->bitmap = bitmap;
1708 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1709 md_wakeup_thread(mddev->thread);
1711 bitmap_update_sb(bitmap);
1713 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1716 bitmap_free(bitmap);
1721 location_show(mddev_t *mddev, char *page)
1724 if (mddev->bitmap_info.file) {
1725 len = sprintf(page, "file");
1726 } else if (mddev->bitmap_info.offset) {
1727 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1729 len = sprintf(page, "none");
1730 len += sprintf(page+len, "\n");
1735 location_store(mddev_t *mddev, const char *buf, size_t len)
1739 if (!mddev->pers->quiesce)
1741 if (mddev->recovery || mddev->sync_thread)
1745 if (mddev->bitmap || mddev->bitmap_info.file ||
1746 mddev->bitmap_info.offset) {
1747 /* bitmap already configured. Only option is to clear it */
1748 if (strncmp(buf, "none", 4) != 0)
1751 mddev->pers->quiesce(mddev, 1);
1752 bitmap_destroy(mddev);
1753 mddev->pers->quiesce(mddev, 0);
1755 mddev->bitmap_info.offset = 0;
1756 if (mddev->bitmap_info.file) {
1757 struct file *f = mddev->bitmap_info.file;
1758 mddev->bitmap_info.file = NULL;
1759 restore_bitmap_write_access(f);
1763 /* No bitmap, OK to set a location */
1765 if (strncmp(buf, "none", 4) == 0)
1766 /* nothing to be done */;
1767 else if (strncmp(buf, "file:", 5) == 0) {
1768 /* Not supported yet */
1773 rv = strict_strtoll(buf+1, 10, &offset);
1775 rv = strict_strtoll(buf, 10, &offset);
1780 if (mddev->major_version == 0 &&
1781 offset != mddev->bitmap_info.default_offset)
1783 mddev->bitmap_info.offset = offset;
1785 mddev->pers->quiesce(mddev, 1);
1786 rv = bitmap_create(mddev);
1788 bitmap_destroy(mddev);
1789 mddev->bitmap_info.offset = 0;
1791 mddev->pers->quiesce(mddev, 0);
1797 if (!mddev->external) {
1798 /* Ensure new bitmap info is stored in
1799 * metadata promptly.
1801 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1802 md_wakeup_thread(mddev->thread);
1807 static struct md_sysfs_entry bitmap_location =
1808 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1811 timeout_show(mddev_t *mddev, char *page)
1814 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1815 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1817 len = sprintf(page, "%lu", secs);
1819 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1820 len += sprintf(page+len, "\n");
1825 timeout_store(mddev_t *mddev, const char *buf, size_t len)
1827 /* timeout can be set at any time */
1828 unsigned long timeout;
1829 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1833 /* just to make sure we don't overflow... */
1834 if (timeout >= LONG_MAX / HZ)
1837 timeout = timeout * HZ / 10000;
1839 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1840 timeout = MAX_SCHEDULE_TIMEOUT-1;
1843 mddev->bitmap_info.daemon_sleep = timeout;
1844 if (mddev->thread) {
1845 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1846 * the bitmap is all clean and we don't need to
1847 * adjust the timeout right now
1849 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1850 mddev->thread->timeout = timeout;
1851 md_wakeup_thread(mddev->thread);
1857 static struct md_sysfs_entry bitmap_timeout =
1858 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1861 backlog_show(mddev_t *mddev, char *page)
1863 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1867 backlog_store(mddev_t *mddev, const char *buf, size_t len)
1869 unsigned long backlog;
1870 int rv = strict_strtoul(buf, 10, &backlog);
1873 if (backlog > COUNTER_MAX)
1875 mddev->bitmap_info.max_write_behind = backlog;
1879 static struct md_sysfs_entry bitmap_backlog =
1880 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
1883 chunksize_show(mddev_t *mddev, char *page)
1885 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
1889 chunksize_store(mddev_t *mddev, const char *buf, size_t len)
1891 /* Can only be changed when no bitmap is active */
1893 unsigned long csize;
1896 rv = strict_strtoul(buf, 10, &csize);
1900 !is_power_of_2(csize))
1902 mddev->bitmap_info.chunksize = csize;
1906 static struct md_sysfs_entry bitmap_chunksize =
1907 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
1909 static struct attribute *md_bitmap_attrs[] = {
1910 &bitmap_location.attr,
1911 &bitmap_timeout.attr,
1912 &bitmap_backlog.attr,
1913 &bitmap_chunksize.attr,
1916 struct attribute_group md_bitmap_group = {
1918 .attrs = md_bitmap_attrs,
1922 /* the bitmap API -- for raid personalities */
1923 EXPORT_SYMBOL(bitmap_startwrite);
1924 EXPORT_SYMBOL(bitmap_endwrite);
1925 EXPORT_SYMBOL(bitmap_start_sync);
1926 EXPORT_SYMBOL(bitmap_end_sync);
1927 EXPORT_SYMBOL(bitmap_unplug);
1928 EXPORT_SYMBOL(bitmap_close_sync);
1929 EXPORT_SYMBOL(bitmap_cond_end_sync);