2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
15 * flush after percent set rather than just time based. (maybe both).
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
33 static inline char *bmname(struct bitmap *bitmap)
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
48 static int bitmap_checkpage(struct bitmap_counts *bitmap,
49 unsigned long page, int create)
50 __releases(bitmap->lock)
51 __acquires(bitmap->lock)
53 unsigned char *mappage;
55 if (page >= bitmap->pages) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
63 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
72 /* this page has not been allocated yet */
74 spin_unlock_irq(&bitmap->lock);
75 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
76 spin_lock_irq(&bitmap->lock);
78 if (mappage == NULL) {
79 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
80 /* failed - set the hijacked flag so that we can use the
81 * pointer as a counter */
82 if (!bitmap->bp[page].map)
83 bitmap->bp[page].hijacked = 1;
84 } else if (bitmap->bp[page].map ||
85 bitmap->bp[page].hijacked) {
86 /* somebody beat us to getting the page */
91 /* no page was in place and we have one, so install it */
93 bitmap->bp[page].map = mappage;
94 bitmap->missing_pages--;
99 /* if page is completely empty, put it back on the free list, or dealloc it */
100 /* if page was hijacked, unmark the flag so it might get alloced next time */
101 /* Note: lock should be held when calling this */
102 static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
106 if (bitmap->bp[page].count) /* page is still busy */
109 /* page is no longer in use, it can be released */
111 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
112 bitmap->bp[page].hijacked = 0;
113 bitmap->bp[page].map = NULL;
115 /* normal case, free the page */
116 ptr = bitmap->bp[page].map;
117 bitmap->bp[page].map = NULL;
118 bitmap->missing_pages++;
124 * bitmap file handling - read and write the bitmap file and its superblock
128 * basic page I/O operations
131 /* IO operations when bitmap is stored near all superblocks */
132 static int read_sb_page(struct mddev *mddev, loff_t offset,
134 unsigned long index, int size)
136 /* choose a good rdev and read the page from there */
138 struct md_rdev *rdev;
141 rdev_for_each(rdev, mddev) {
142 if (! test_bit(In_sync, &rdev->flags)
143 || test_bit(Faulty, &rdev->flags))
146 target = offset + index * (PAGE_SIZE/512);
148 if (sync_page_io(rdev, target,
149 roundup(size, bdev_logical_block_size(rdev->bdev)),
158 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
160 /* Iterate the disks of an mddev, using rcu to protect access to the
161 * linked list, and raising the refcount of devices we return to ensure
162 * they don't disappear while in use.
163 * As devices are only added or removed when raid_disk is < 0 and
164 * nr_pending is 0 and In_sync is clear, the entries we return will
165 * still be in the same position on the list when we re-enter
166 * list_for_each_entry_continue_rcu.
170 /* start at the beginning */
171 rdev = list_entry_rcu(&mddev->disks, struct md_rdev, same_set);
173 /* release the previous rdev and start from there. */
174 rdev_dec_pending(rdev, mddev);
176 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
177 if (rdev->raid_disk >= 0 &&
178 !test_bit(Faulty, &rdev->flags)) {
179 /* this is a usable devices */
180 atomic_inc(&rdev->nr_pending);
189 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
191 struct md_rdev *rdev = NULL;
192 struct block_device *bdev;
193 struct mddev *mddev = bitmap->mddev;
194 struct bitmap_storage *store = &bitmap->storage;
196 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
197 int size = PAGE_SIZE;
198 loff_t offset = mddev->bitmap_info.offset;
200 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
202 if (page->index == store->file_pages-1) {
203 int last_page_size = store->bytes & (PAGE_SIZE-1);
204 if (last_page_size == 0)
205 last_page_size = PAGE_SIZE;
206 size = roundup(last_page_size,
207 bdev_logical_block_size(bdev));
209 /* Just make sure we aren't corrupting data or
212 if (mddev->external) {
213 /* Bitmap could be anywhere. */
214 if (rdev->sb_start + offset + (page->index
218 rdev->sb_start + offset
219 < (rdev->data_offset + mddev->dev_sectors
222 } else if (offset < 0) {
223 /* DATA BITMAP METADATA */
225 + (long)(page->index * (PAGE_SIZE/512))
227 /* bitmap runs in to metadata */
229 if (rdev->data_offset + mddev->dev_sectors
230 > rdev->sb_start + offset)
231 /* data runs in to bitmap */
233 } else if (rdev->sb_start < rdev->data_offset) {
234 /* METADATA BITMAP DATA */
237 + page->index*(PAGE_SIZE/512) + size/512
239 /* bitmap runs in to data */
242 /* DATA METADATA BITMAP - no problems */
244 md_super_write(mddev, rdev,
245 rdev->sb_start + offset
246 + page->index * (PAGE_SIZE/512),
252 md_super_wait(mddev);
259 static void bitmap_file_kick(struct bitmap *bitmap);
261 * write out a page to a file
263 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
265 struct buffer_head *bh;
267 if (bitmap->storage.file == NULL) {
268 switch (write_sb_page(bitmap, page, wait)) {
270 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
274 bh = page_buffers(page);
276 while (bh && bh->b_blocknr) {
277 atomic_inc(&bitmap->pending_writes);
278 set_buffer_locked(bh);
279 set_buffer_mapped(bh);
280 submit_bh(WRITE | REQ_SYNC, bh);
281 bh = bh->b_this_page;
285 wait_event(bitmap->write_wait,
286 atomic_read(&bitmap->pending_writes)==0);
288 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
289 bitmap_file_kick(bitmap);
292 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
294 struct bitmap *bitmap = bh->b_private;
297 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
298 if (atomic_dec_and_test(&bitmap->pending_writes))
299 wake_up(&bitmap->write_wait);
302 /* copied from buffer.c */
304 __clear_page_buffers(struct page *page)
306 ClearPagePrivate(page);
307 set_page_private(page, 0);
308 page_cache_release(page);
310 static void free_buffers(struct page *page)
312 struct buffer_head *bh;
314 if (!PagePrivate(page))
317 bh = page_buffers(page);
319 struct buffer_head *next = bh->b_this_page;
320 free_buffer_head(bh);
323 __clear_page_buffers(page);
327 /* read a page from a file.
328 * We both read the page, and attach buffers to the page to record the
329 * address of each block (using bmap). These addresses will be used
330 * to write the block later, completely bypassing the filesystem.
331 * This usage is similar to how swap files are handled, and allows us
332 * to write to a file with no concerns of memory allocation failing.
334 static int read_page(struct file *file, unsigned long index,
335 struct bitmap *bitmap,
340 struct inode *inode = file_inode(file);
341 struct buffer_head *bh;
344 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
345 (unsigned long long)index << PAGE_SHIFT);
347 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
352 attach_page_buffers(page, bh);
353 block = index << (PAGE_SHIFT - inode->i_blkbits);
358 bh->b_blocknr = bmap(inode, block);
359 if (bh->b_blocknr == 0) {
360 /* Cannot use this file! */
364 bh->b_bdev = inode->i_sb->s_bdev;
365 if (count < (1<<inode->i_blkbits))
368 count -= (1<<inode->i_blkbits);
370 bh->b_end_io = end_bitmap_write;
371 bh->b_private = bitmap;
372 atomic_inc(&bitmap->pending_writes);
373 set_buffer_locked(bh);
374 set_buffer_mapped(bh);
378 bh = bh->b_this_page;
382 wait_event(bitmap->write_wait,
383 atomic_read(&bitmap->pending_writes)==0);
384 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
388 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
390 (unsigned long long)index << PAGE_SHIFT,
396 * bitmap file superblock operations
399 /* update the event counter and sync the superblock to disk */
400 void bitmap_update_sb(struct bitmap *bitmap)
404 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
406 if (bitmap->mddev->bitmap_info.external)
408 if (!bitmap->storage.sb_page) /* no superblock */
410 sb = kmap_atomic(bitmap->storage.sb_page);
411 sb->events = cpu_to_le64(bitmap->mddev->events);
412 if (bitmap->mddev->events < bitmap->events_cleared)
413 /* rocking back to read-only */
414 bitmap->events_cleared = bitmap->mddev->events;
415 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
416 sb->state = cpu_to_le32(bitmap->flags);
417 /* Just in case these have been changed via sysfs: */
418 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
419 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
420 /* This might have been changed by a reshape */
421 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
422 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
423 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
426 write_page(bitmap, bitmap->storage.sb_page, 1);
429 /* print out the bitmap file superblock */
430 void bitmap_print_sb(struct bitmap *bitmap)
434 if (!bitmap || !bitmap->storage.sb_page)
436 sb = kmap_atomic(bitmap->storage.sb_page);
437 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
438 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
439 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
440 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
441 *(__u32 *)(sb->uuid+0),
442 *(__u32 *)(sb->uuid+4),
443 *(__u32 *)(sb->uuid+8),
444 *(__u32 *)(sb->uuid+12));
445 printk(KERN_DEBUG " events: %llu\n",
446 (unsigned long long) le64_to_cpu(sb->events));
447 printk(KERN_DEBUG "events cleared: %llu\n",
448 (unsigned long long) le64_to_cpu(sb->events_cleared));
449 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
450 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
451 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
452 printk(KERN_DEBUG " sync size: %llu KB\n",
453 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
454 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
462 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
463 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
464 * This function verifies 'bitmap_info' and populates the on-disk bitmap
465 * structure, which is to be written to disk.
467 * Returns: 0 on success, -Exxx on error
469 static int bitmap_new_disk_sb(struct bitmap *bitmap)
472 unsigned long chunksize, daemon_sleep, write_behind;
474 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
475 if (bitmap->storage.sb_page == NULL)
477 bitmap->storage.sb_page->index = 0;
479 sb = kmap_atomic(bitmap->storage.sb_page);
481 sb->magic = cpu_to_le32(BITMAP_MAGIC);
482 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
484 chunksize = bitmap->mddev->bitmap_info.chunksize;
486 if (!is_power_of_2(chunksize)) {
488 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
491 sb->chunksize = cpu_to_le32(chunksize);
493 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
495 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
496 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
497 daemon_sleep = 5 * HZ;
499 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
500 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
503 * FIXME: write_behind for RAID1. If not specified, what
504 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
506 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
507 if (write_behind > COUNTER_MAX)
508 write_behind = COUNTER_MAX / 2;
509 sb->write_behind = cpu_to_le32(write_behind);
510 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
512 /* keep the array size field of the bitmap superblock up to date */
513 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
515 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
517 set_bit(BITMAP_STALE, &bitmap->flags);
518 sb->state = cpu_to_le32(bitmap->flags);
519 bitmap->events_cleared = bitmap->mddev->events;
520 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
527 /* read the superblock from the bitmap file and initialize some bitmap fields */
528 static int bitmap_read_sb(struct bitmap *bitmap)
532 unsigned long chunksize, daemon_sleep, write_behind;
533 unsigned long long events;
534 unsigned long sectors_reserved = 0;
536 struct page *sb_page;
538 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
539 chunksize = 128 * 1024 * 1024;
540 daemon_sleep = 5 * HZ;
542 set_bit(BITMAP_STALE, &bitmap->flags);
546 /* page 0 is the superblock, read it... */
547 sb_page = alloc_page(GFP_KERNEL);
550 bitmap->storage.sb_page = sb_page;
552 if (bitmap->storage.file) {
553 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
554 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
556 err = read_page(bitmap->storage.file, 0,
557 bitmap, bytes, sb_page);
559 err = read_sb_page(bitmap->mddev,
560 bitmap->mddev->bitmap_info.offset,
562 0, sizeof(bitmap_super_t));
567 sb = kmap_atomic(sb_page);
569 chunksize = le32_to_cpu(sb->chunksize);
570 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
571 write_behind = le32_to_cpu(sb->write_behind);
572 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
574 /* verify that the bitmap-specific fields are valid */
575 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
576 reason = "bad magic";
577 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
578 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
579 reason = "unrecognized superblock version";
580 else if (chunksize < 512)
581 reason = "bitmap chunksize too small";
582 else if (!is_power_of_2(chunksize))
583 reason = "bitmap chunksize not a power of 2";
584 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
585 reason = "daemon sleep period out of range";
586 else if (write_behind > COUNTER_MAX)
587 reason = "write-behind limit out of range (0 - 16383)";
589 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
590 bmname(bitmap), reason);
594 /* keep the array size field of the bitmap superblock up to date */
595 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
597 if (bitmap->mddev->persistent) {
599 * We have a persistent array superblock, so compare the
600 * bitmap's UUID and event counter to the mddev's
602 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
604 "%s: bitmap superblock UUID mismatch\n",
608 events = le64_to_cpu(sb->events);
609 if (events < bitmap->mddev->events) {
611 "%s: bitmap file is out of date (%llu < %llu) "
612 "-- forcing full recovery\n",
613 bmname(bitmap), events,
614 (unsigned long long) bitmap->mddev->events);
615 set_bit(BITMAP_STALE, &bitmap->flags);
619 /* assign fields using values from superblock */
620 bitmap->flags |= le32_to_cpu(sb->state);
621 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
622 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
623 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
628 if (test_bit(BITMAP_STALE, &bitmap->flags))
629 bitmap->events_cleared = bitmap->mddev->events;
630 bitmap->mddev->bitmap_info.chunksize = chunksize;
631 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
632 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
633 if (bitmap->mddev->bitmap_info.space == 0 ||
634 bitmap->mddev->bitmap_info.space > sectors_reserved)
635 bitmap->mddev->bitmap_info.space = sectors_reserved;
637 bitmap_print_sb(bitmap);
642 * general bitmap file operations
648 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
649 * file a page at a time. There's a superblock at the start of the file.
651 /* calculate the index of the page that contains this bit */
652 static inline unsigned long file_page_index(struct bitmap_storage *store,
656 chunk += sizeof(bitmap_super_t) << 3;
657 return chunk >> PAGE_BIT_SHIFT;
660 /* calculate the (bit) offset of this bit within a page */
661 static inline unsigned long file_page_offset(struct bitmap_storage *store,
665 chunk += sizeof(bitmap_super_t) << 3;
666 return chunk & (PAGE_BITS - 1);
670 * return a pointer to the page in the filemap that contains the given bit
672 * this lookup is complicated by the fact that the bitmap sb might be exactly
673 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
676 static inline struct page *filemap_get_page(struct bitmap_storage *store,
679 if (file_page_index(store, chunk) >= store->file_pages)
681 return store->filemap[file_page_index(store, chunk)
682 - file_page_index(store, 0)];
685 static int bitmap_storage_alloc(struct bitmap_storage *store,
686 unsigned long chunks, int with_super)
689 unsigned long num_pages;
692 bytes = DIV_ROUND_UP(chunks, 8);
694 bytes += sizeof(bitmap_super_t);
696 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
698 store->filemap = kmalloc(sizeof(struct page *)
699 * num_pages, GFP_KERNEL);
703 if (with_super && !store->sb_page) {
704 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
705 if (store->sb_page == NULL)
707 store->sb_page->index = 0;
710 if (store->sb_page) {
711 store->filemap[0] = store->sb_page;
714 for ( ; pnum < num_pages; pnum++) {
715 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
716 if (!store->filemap[pnum]) {
717 store->file_pages = pnum;
720 store->filemap[pnum]->index = pnum;
722 store->file_pages = pnum;
724 /* We need 4 bits per page, rounded up to a multiple
725 * of sizeof(unsigned long) */
726 store->filemap_attr = kzalloc(
727 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
729 if (!store->filemap_attr)
732 store->bytes = bytes;
737 static void bitmap_file_unmap(struct bitmap_storage *store)
739 struct page **map, *sb_page;
744 map = store->filemap;
745 pages = store->file_pages;
746 sb_page = store->sb_page;
749 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
750 free_buffers(map[pages]);
752 kfree(store->filemap_attr);
755 free_buffers(sb_page);
758 struct inode *inode = file_inode(file);
759 invalidate_mapping_pages(inode->i_mapping, 0, -1);
765 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
766 * then it is no longer reliable, so we stop using it and we mark the file
767 * as failed in the superblock
769 static void bitmap_file_kick(struct bitmap *bitmap)
771 char *path, *ptr = NULL;
773 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
774 bitmap_update_sb(bitmap);
776 if (bitmap->storage.file) {
777 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
779 ptr = d_path(&bitmap->storage.file->f_path,
783 "%s: kicking failed bitmap file %s from array!\n",
784 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
789 "%s: disabling internal bitmap due to errors\n",
794 enum bitmap_page_attr {
795 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
796 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
797 * i.e. counter is 1 or 2. */
798 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
801 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
802 enum bitmap_page_attr attr)
804 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
807 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
808 enum bitmap_page_attr attr)
810 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
813 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
814 enum bitmap_page_attr attr)
816 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
819 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
820 enum bitmap_page_attr attr)
822 return test_and_clear_bit((pnum<<2) + attr,
823 bitmap->storage.filemap_attr);
826 * bitmap_file_set_bit -- called before performing a write to the md device
827 * to set (and eventually sync) a particular bit in the bitmap file
829 * we set the bit immediately, then we record the page number so that
830 * when an unplug occurs, we can flush the dirty pages out to disk
832 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
837 unsigned long chunk = block >> bitmap->counts.chunkshift;
839 page = filemap_get_page(&bitmap->storage, chunk);
842 bit = file_page_offset(&bitmap->storage, chunk);
845 kaddr = kmap_atomic(page);
846 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
849 set_bit_le(bit, kaddr);
850 kunmap_atomic(kaddr);
851 pr_debug("set file bit %lu page %lu\n", bit, page->index);
852 /* record page number so it gets flushed to disk when unplug occurs */
853 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
856 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
861 unsigned long chunk = block >> bitmap->counts.chunkshift;
863 page = filemap_get_page(&bitmap->storage, chunk);
866 bit = file_page_offset(&bitmap->storage, chunk);
867 paddr = kmap_atomic(page);
868 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
869 clear_bit(bit, paddr);
871 clear_bit_le(bit, paddr);
872 kunmap_atomic(paddr);
873 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
874 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
875 bitmap->allclean = 0;
879 /* this gets called when the md device is ready to unplug its underlying
880 * (slave) device queues -- before we let any writes go down, we need to
881 * sync the dirty pages of the bitmap file to disk */
882 void bitmap_unplug(struct bitmap *bitmap)
885 int dirty, need_write;
887 if (!bitmap || !bitmap->storage.filemap ||
888 test_bit(BITMAP_STALE, &bitmap->flags))
891 /* look at each page to see if there are any set bits that need to be
892 * flushed out to disk */
893 for (i = 0; i < bitmap->storage.file_pages; i++) {
894 if (!bitmap->storage.filemap)
896 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
897 need_write = test_and_clear_page_attr(bitmap, i,
898 BITMAP_PAGE_NEEDWRITE);
899 if (dirty || need_write) {
900 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
901 write_page(bitmap, bitmap->storage.filemap[i], 0);
904 if (bitmap->storage.file)
905 wait_event(bitmap->write_wait,
906 atomic_read(&bitmap->pending_writes)==0);
908 md_super_wait(bitmap->mddev);
910 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
911 bitmap_file_kick(bitmap);
913 EXPORT_SYMBOL(bitmap_unplug);
915 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
916 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
917 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
918 * memory mapping of the bitmap file
920 * if there's no bitmap file, or if the bitmap file had been
921 * previously kicked from the array, we mark all the bits as
922 * 1's in order to cause a full resync.
924 * We ignore all bits for sectors that end earlier than 'start'.
925 * This is used when reading an out-of-date bitmap...
927 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
929 unsigned long i, chunks, index, oldindex, bit;
930 struct page *page = NULL;
931 unsigned long bit_cnt = 0;
933 unsigned long offset;
937 struct bitmap_storage *store = &bitmap->storage;
939 chunks = bitmap->counts.chunks;
942 if (!file && !bitmap->mddev->bitmap_info.offset) {
943 /* No permanent bitmap - fill with '1s'. */
944 store->filemap = NULL;
945 store->file_pages = 0;
946 for (i = 0; i < chunks ; i++) {
947 /* if the disk bit is set, set the memory bit */
948 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
950 bitmap_set_memory_bits(bitmap,
951 (sector_t)i << bitmap->counts.chunkshift,
957 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
959 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
960 "recovery\n", bmname(bitmap));
962 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
963 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
965 (unsigned long) i_size_read(file->f_mapping->host),
972 if (!bitmap->mddev->bitmap_info.external)
973 offset = sizeof(bitmap_super_t);
975 for (i = 0; i < chunks; i++) {
977 index = file_page_index(&bitmap->storage, i);
978 bit = file_page_offset(&bitmap->storage, i);
979 if (index != oldindex) { /* this is a new page, read it in */
981 /* unmap the old page, we're done with it */
982 if (index == store->file_pages-1)
983 count = store->bytes - index * PAGE_SIZE;
986 page = store->filemap[index];
988 ret = read_page(file, index, bitmap,
993 bitmap->mddev->bitmap_info.offset,
1004 * if bitmap is out of date, dirty the
1005 * whole page and write it out
1007 paddr = kmap_atomic(page);
1008 memset(paddr + offset, 0xff,
1009 PAGE_SIZE - offset);
1010 kunmap_atomic(paddr);
1011 write_page(bitmap, page, 1);
1014 if (test_bit(BITMAP_WRITE_ERROR,
1019 paddr = kmap_atomic(page);
1020 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1021 b = test_bit(bit, paddr);
1023 b = test_bit_le(bit, paddr);
1024 kunmap_atomic(paddr);
1026 /* if the disk bit is set, set the memory bit */
1027 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1029 bitmap_set_memory_bits(bitmap,
1030 (sector_t)i << bitmap->counts.chunkshift,
1037 printk(KERN_INFO "%s: bitmap initialized from disk: "
1038 "read %lu pages, set %lu of %lu bits\n",
1039 bmname(bitmap), store->file_pages,
1045 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1046 bmname(bitmap), ret);
1050 void bitmap_write_all(struct bitmap *bitmap)
1052 /* We don't actually write all bitmap blocks here,
1053 * just flag them as needing to be written
1057 if (!bitmap || !bitmap->storage.filemap)
1059 if (bitmap->storage.file)
1060 /* Only one copy, so nothing needed */
1063 for (i = 0; i < bitmap->storage.file_pages; i++)
1064 set_page_attr(bitmap, i,
1065 BITMAP_PAGE_NEEDWRITE);
1066 bitmap->allclean = 0;
1069 static void bitmap_count_page(struct bitmap_counts *bitmap,
1070 sector_t offset, int inc)
1072 sector_t chunk = offset >> bitmap->chunkshift;
1073 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1074 bitmap->bp[page].count += inc;
1075 bitmap_checkfree(bitmap, page);
1078 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1080 sector_t chunk = offset >> bitmap->chunkshift;
1081 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1082 struct bitmap_page *bp = &bitmap->bp[page];
1088 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1089 sector_t offset, sector_t *blocks,
1093 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1097 void bitmap_daemon_work(struct mddev *mddev)
1099 struct bitmap *bitmap;
1101 unsigned long nextpage;
1103 struct bitmap_counts *counts;
1105 /* Use a mutex to guard daemon_work against
1108 mutex_lock(&mddev->bitmap_info.mutex);
1109 bitmap = mddev->bitmap;
1110 if (bitmap == NULL) {
1111 mutex_unlock(&mddev->bitmap_info.mutex);
1114 if (time_before(jiffies, bitmap->daemon_lastrun
1115 + mddev->bitmap_info.daemon_sleep))
1118 bitmap->daemon_lastrun = jiffies;
1119 if (bitmap->allclean) {
1120 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1123 bitmap->allclean = 1;
1125 /* Any file-page which is PENDING now needs to be written.
1126 * So set NEEDWRITE now, then after we make any last-minute changes
1129 for (j = 0; j < bitmap->storage.file_pages; j++)
1130 if (test_and_clear_page_attr(bitmap, j,
1131 BITMAP_PAGE_PENDING))
1132 set_page_attr(bitmap, j,
1133 BITMAP_PAGE_NEEDWRITE);
1135 if (bitmap->need_sync &&
1136 mddev->bitmap_info.external == 0) {
1137 /* Arrange for superblock update as well as
1140 bitmap->need_sync = 0;
1141 if (bitmap->storage.filemap) {
1142 sb = kmap_atomic(bitmap->storage.sb_page);
1143 sb->events_cleared =
1144 cpu_to_le64(bitmap->events_cleared);
1146 set_page_attr(bitmap, 0,
1147 BITMAP_PAGE_NEEDWRITE);
1150 /* Now look at the bitmap counters and if any are '2' or '1',
1151 * decrement and handle accordingly.
1153 counts = &bitmap->counts;
1154 spin_lock_irq(&counts->lock);
1156 for (j = 0; j < counts->chunks; j++) {
1157 bitmap_counter_t *bmc;
1158 sector_t block = (sector_t)j << counts->chunkshift;
1160 if (j == nextpage) {
1161 nextpage += PAGE_COUNTER_RATIO;
1162 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1163 j |= PAGE_COUNTER_MASK;
1166 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1168 bmc = bitmap_get_counter(counts,
1173 j |= PAGE_COUNTER_MASK;
1176 if (*bmc == 1 && !bitmap->need_sync) {
1177 /* We can clear the bit */
1179 bitmap_count_page(counts, block, -1);
1180 bitmap_file_clear_bit(bitmap, block);
1181 } else if (*bmc && *bmc <= 2) {
1183 bitmap_set_pending(counts, block);
1184 bitmap->allclean = 0;
1187 spin_unlock_irq(&counts->lock);
1189 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1190 * DIRTY pages need to be written by bitmap_unplug so it can wait
1192 * If we find any DIRTY page we stop there and let bitmap_unplug
1193 * handle all the rest. This is important in the case where
1194 * the first blocking holds the superblock and it has been updated.
1195 * We mustn't write any other blocks before the superblock.
1198 j < bitmap->storage.file_pages
1199 && !test_bit(BITMAP_STALE, &bitmap->flags);
1202 if (test_page_attr(bitmap, j,
1204 /* bitmap_unplug will handle the rest */
1206 if (test_and_clear_page_attr(bitmap, j,
1207 BITMAP_PAGE_NEEDWRITE)) {
1208 write_page(bitmap, bitmap->storage.filemap[j], 0);
1213 if (bitmap->allclean == 0)
1214 mddev->thread->timeout =
1215 mddev->bitmap_info.daemon_sleep;
1216 mutex_unlock(&mddev->bitmap_info.mutex);
1219 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1220 sector_t offset, sector_t *blocks,
1222 __releases(bitmap->lock)
1223 __acquires(bitmap->lock)
1225 /* If 'create', we might release the lock and reclaim it.
1226 * The lock must have been taken with interrupts enabled.
1227 * If !create, we don't release the lock.
1229 sector_t chunk = offset >> bitmap->chunkshift;
1230 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1231 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1235 err = bitmap_checkpage(bitmap, page, create);
1237 if (bitmap->bp[page].hijacked ||
1238 bitmap->bp[page].map == NULL)
1239 csize = ((sector_t)1) << (bitmap->chunkshift +
1240 PAGE_COUNTER_SHIFT - 1);
1242 csize = ((sector_t)1) << bitmap->chunkshift;
1243 *blocks = csize - (offset & (csize - 1));
1248 /* now locked ... */
1250 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1251 /* should we use the first or second counter field
1252 * of the hijacked pointer? */
1253 int hi = (pageoff > PAGE_COUNTER_MASK);
1254 return &((bitmap_counter_t *)
1255 &bitmap->bp[page].map)[hi];
1256 } else /* page is allocated */
1257 return (bitmap_counter_t *)
1258 &(bitmap->bp[page].map[pageoff]);
1261 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1268 atomic_inc(&bitmap->behind_writes);
1269 bw = atomic_read(&bitmap->behind_writes);
1270 if (bw > bitmap->behind_writes_used)
1271 bitmap->behind_writes_used = bw;
1273 pr_debug("inc write-behind count %d/%lu\n",
1274 bw, bitmap->mddev->bitmap_info.max_write_behind);
1279 bitmap_counter_t *bmc;
1281 spin_lock_irq(&bitmap->counts.lock);
1282 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1284 spin_unlock_irq(&bitmap->counts.lock);
1288 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1289 DEFINE_WAIT(__wait);
1290 /* note that it is safe to do the prepare_to_wait
1291 * after the test as long as we do it before dropping
1294 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1295 TASK_UNINTERRUPTIBLE);
1296 spin_unlock_irq(&bitmap->counts.lock);
1298 finish_wait(&bitmap->overflow_wait, &__wait);
1304 bitmap_file_set_bit(bitmap, offset);
1305 bitmap_count_page(&bitmap->counts, offset, 1);
1313 spin_unlock_irq(&bitmap->counts.lock);
1316 if (sectors > blocks)
1323 EXPORT_SYMBOL(bitmap_startwrite);
1325 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1326 int success, int behind)
1331 if (atomic_dec_and_test(&bitmap->behind_writes))
1332 wake_up(&bitmap->behind_wait);
1333 pr_debug("dec write-behind count %d/%lu\n",
1334 atomic_read(&bitmap->behind_writes),
1335 bitmap->mddev->bitmap_info.max_write_behind);
1340 unsigned long flags;
1341 bitmap_counter_t *bmc;
1343 spin_lock_irqsave(&bitmap->counts.lock, flags);
1344 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1346 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1350 if (success && !bitmap->mddev->degraded &&
1351 bitmap->events_cleared < bitmap->mddev->events) {
1352 bitmap->events_cleared = bitmap->mddev->events;
1353 bitmap->need_sync = 1;
1354 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1357 if (!success && !NEEDED(*bmc))
1358 *bmc |= NEEDED_MASK;
1360 if (COUNTER(*bmc) == COUNTER_MAX)
1361 wake_up(&bitmap->overflow_wait);
1365 bitmap_set_pending(&bitmap->counts, offset);
1366 bitmap->allclean = 0;
1368 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1370 if (sectors > blocks)
1376 EXPORT_SYMBOL(bitmap_endwrite);
1378 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1381 bitmap_counter_t *bmc;
1383 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1385 return 1; /* always resync if no bitmap */
1387 spin_lock_irq(&bitmap->counts.lock);
1388 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1394 else if (NEEDED(*bmc)) {
1396 if (!degraded) { /* don't set/clear bits if degraded */
1397 *bmc |= RESYNC_MASK;
1398 *bmc &= ~NEEDED_MASK;
1402 spin_unlock_irq(&bitmap->counts.lock);
1406 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1409 /* bitmap_start_sync must always report on multiples of whole
1410 * pages, otherwise resync (which is very PAGE_SIZE based) will
1412 * So call __bitmap_start_sync repeatedly (if needed) until
1413 * At least PAGE_SIZE>>9 blocks are covered.
1414 * Return the 'or' of the result.
1420 while (*blocks < (PAGE_SIZE>>9)) {
1421 rv |= __bitmap_start_sync(bitmap, offset,
1422 &blocks1, degraded);
1428 EXPORT_SYMBOL(bitmap_start_sync);
1430 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1432 bitmap_counter_t *bmc;
1433 unsigned long flags;
1435 if (bitmap == NULL) {
1439 spin_lock_irqsave(&bitmap->counts.lock, flags);
1440 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1445 *bmc &= ~RESYNC_MASK;
1447 if (!NEEDED(*bmc) && aborted)
1448 *bmc |= NEEDED_MASK;
1451 bitmap_set_pending(&bitmap->counts, offset);
1452 bitmap->allclean = 0;
1457 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1459 EXPORT_SYMBOL(bitmap_end_sync);
1461 void bitmap_close_sync(struct bitmap *bitmap)
1463 /* Sync has finished, and any bitmap chunks that weren't synced
1464 * properly have been aborted. It remains to us to clear the
1465 * RESYNC bit wherever it is still on
1467 sector_t sector = 0;
1471 while (sector < bitmap->mddev->resync_max_sectors) {
1472 bitmap_end_sync(bitmap, sector, &blocks, 0);
1476 EXPORT_SYMBOL(bitmap_close_sync);
1478 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1486 bitmap->last_end_sync = jiffies;
1489 if (time_before(jiffies, (bitmap->last_end_sync
1490 + bitmap->mddev->bitmap_info.daemon_sleep)))
1492 wait_event(bitmap->mddev->recovery_wait,
1493 atomic_read(&bitmap->mddev->recovery_active) == 0);
1495 bitmap->mddev->curr_resync_completed = sector;
1496 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1497 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1499 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1500 bitmap_end_sync(bitmap, s, &blocks, 0);
1503 bitmap->last_end_sync = jiffies;
1504 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1506 EXPORT_SYMBOL(bitmap_cond_end_sync);
1508 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1510 /* For each chunk covered by any of these sectors, set the
1511 * counter to 2 and possibly set resync_needed. They should all
1512 * be 0 at this point
1516 bitmap_counter_t *bmc;
1517 spin_lock_irq(&bitmap->counts.lock);
1518 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1520 spin_unlock_irq(&bitmap->counts.lock);
1524 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1525 bitmap_count_page(&bitmap->counts, offset, 1);
1526 bitmap_set_pending(&bitmap->counts, offset);
1527 bitmap->allclean = 0;
1529 spin_unlock_irq(&bitmap->counts.lock);
1532 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1533 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1535 unsigned long chunk;
1537 for (chunk = s; chunk <= e; chunk++) {
1538 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1539 bitmap_set_memory_bits(bitmap, sec, 1);
1540 bitmap_file_set_bit(bitmap, sec);
1541 if (sec < bitmap->mddev->recovery_cp)
1542 /* We are asserting that the array is dirty,
1543 * so move the recovery_cp address back so
1544 * that it is obvious that it is dirty
1546 bitmap->mddev->recovery_cp = sec;
1551 * flush out any pending updates
1553 void bitmap_flush(struct mddev *mddev)
1555 struct bitmap *bitmap = mddev->bitmap;
1558 if (!bitmap) /* there was no bitmap */
1561 /* run the daemon_work three time to ensure everything is flushed
1564 sleep = mddev->bitmap_info.daemon_sleep * 2;
1565 bitmap->daemon_lastrun -= sleep;
1566 bitmap_daemon_work(mddev);
1567 bitmap->daemon_lastrun -= sleep;
1568 bitmap_daemon_work(mddev);
1569 bitmap->daemon_lastrun -= sleep;
1570 bitmap_daemon_work(mddev);
1571 bitmap_update_sb(bitmap);
1575 * free memory that was allocated
1577 static void bitmap_free(struct bitmap *bitmap)
1579 unsigned long k, pages;
1580 struct bitmap_page *bp;
1582 if (!bitmap) /* there was no bitmap */
1585 /* Shouldn't be needed - but just in case.... */
1586 wait_event(bitmap->write_wait,
1587 atomic_read(&bitmap->pending_writes) == 0);
1589 /* release the bitmap file */
1590 bitmap_file_unmap(&bitmap->storage);
1592 bp = bitmap->counts.bp;
1593 pages = bitmap->counts.pages;
1595 /* free all allocated memory */
1597 if (bp) /* deallocate the page memory */
1598 for (k = 0; k < pages; k++)
1599 if (bp[k].map && !bp[k].hijacked)
1605 void bitmap_destroy(struct mddev *mddev)
1607 struct bitmap *bitmap = mddev->bitmap;
1609 if (!bitmap) /* there was no bitmap */
1612 mutex_lock(&mddev->bitmap_info.mutex);
1613 mddev->bitmap = NULL; /* disconnect from the md device */
1614 mutex_unlock(&mddev->bitmap_info.mutex);
1616 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1618 if (bitmap->sysfs_can_clear)
1619 sysfs_put(bitmap->sysfs_can_clear);
1621 bitmap_free(bitmap);
1625 * initialize the bitmap structure
1626 * if this returns an error, bitmap_destroy must be called to do clean up
1628 int bitmap_create(struct mddev *mddev)
1630 struct bitmap *bitmap;
1631 sector_t blocks = mddev->resync_max_sectors;
1632 struct file *file = mddev->bitmap_info.file;
1634 struct sysfs_dirent *bm = NULL;
1636 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1638 BUG_ON(file && mddev->bitmap_info.offset);
1640 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1644 spin_lock_init(&bitmap->counts.lock);
1645 atomic_set(&bitmap->pending_writes, 0);
1646 init_waitqueue_head(&bitmap->write_wait);
1647 init_waitqueue_head(&bitmap->overflow_wait);
1648 init_waitqueue_head(&bitmap->behind_wait);
1650 bitmap->mddev = mddev;
1653 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1655 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1658 bitmap->sysfs_can_clear = NULL;
1660 bitmap->storage.file = file;
1663 /* As future accesses to this file will use bmap,
1664 * and bypass the page cache, we must sync the file
1669 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1670 if (!mddev->bitmap_info.external) {
1672 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1673 * instructing us to create a new on-disk bitmap instance.
1675 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1676 err = bitmap_new_disk_sb(bitmap);
1678 err = bitmap_read_sb(bitmap);
1681 if (mddev->bitmap_info.chunksize == 0 ||
1682 mddev->bitmap_info.daemon_sleep == 0)
1683 /* chunksize and time_base need to be
1690 bitmap->daemon_lastrun = jiffies;
1691 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1695 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1696 bitmap->counts.pages, bmname(bitmap));
1698 mddev->bitmap = bitmap;
1699 return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1702 bitmap_free(bitmap);
1706 int bitmap_load(struct mddev *mddev)
1710 sector_t sector = 0;
1711 struct bitmap *bitmap = mddev->bitmap;
1716 /* Clear out old bitmap info first: Either there is none, or we
1717 * are resuming after someone else has possibly changed things,
1718 * so we should forget old cached info.
1719 * All chunks should be clean, but some might need_sync.
1721 while (sector < mddev->resync_max_sectors) {
1723 bitmap_start_sync(bitmap, sector, &blocks, 0);
1726 bitmap_close_sync(bitmap);
1728 if (mddev->degraded == 0
1729 || bitmap->events_cleared == mddev->events)
1730 /* no need to keep dirty bits to optimise a
1731 * re-add of a missing device */
1732 start = mddev->recovery_cp;
1734 mutex_lock(&mddev->bitmap_info.mutex);
1735 err = bitmap_init_from_disk(bitmap, start);
1736 mutex_unlock(&mddev->bitmap_info.mutex);
1740 clear_bit(BITMAP_STALE, &bitmap->flags);
1742 /* Kick recovery in case any bits were set */
1743 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1745 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1746 md_wakeup_thread(mddev->thread);
1748 bitmap_update_sb(bitmap);
1750 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1755 EXPORT_SYMBOL_GPL(bitmap_load);
1757 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1759 unsigned long chunk_kb;
1760 struct bitmap_counts *counts;
1765 counts = &bitmap->counts;
1767 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1768 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1770 counts->pages - counts->missing_pages,
1772 (counts->pages - counts->missing_pages)
1773 << (PAGE_SHIFT - 10),
1774 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1775 chunk_kb ? "KB" : "B");
1776 if (bitmap->storage.file) {
1777 seq_printf(seq, ", file: ");
1778 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1781 seq_printf(seq, "\n");
1784 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1785 int chunksize, int init)
1787 /* If chunk_size is 0, choose an appropriate chunk size.
1788 * Then possibly allocate new storage space.
1789 * Then quiesce, copy bits, replace bitmap, and re-start
1791 * This function is called both to set up the initial bitmap
1792 * and to resize the bitmap while the array is active.
1793 * If this happens as a result of the array being resized,
1794 * chunksize will be zero, and we need to choose a suitable
1795 * chunksize, otherwise we use what we are given.
1797 struct bitmap_storage store;
1798 struct bitmap_counts old_counts;
1799 unsigned long chunks;
1801 sector_t old_blocks, new_blocks;
1805 struct bitmap_page *new_bp;
1807 if (chunksize == 0) {
1808 /* If there is enough space, leave the chunk size unchanged,
1809 * else increase by factor of two until there is enough space.
1812 long space = bitmap->mddev->bitmap_info.space;
1815 /* We don't know how much space there is, so limit
1816 * to current size - in sectors.
1818 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1819 if (!bitmap->mddev->bitmap_info.external)
1820 bytes += sizeof(bitmap_super_t);
1821 space = DIV_ROUND_UP(bytes, 512);
1822 bitmap->mddev->bitmap_info.space = space;
1824 chunkshift = bitmap->counts.chunkshift;
1827 /* 'chunkshift' is shift from block size to chunk size */
1829 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1830 bytes = DIV_ROUND_UP(chunks, 8);
1831 if (!bitmap->mddev->bitmap_info.external)
1832 bytes += sizeof(bitmap_super_t);
1833 } while (bytes > (space << 9));
1835 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1837 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1838 memset(&store, 0, sizeof(store));
1839 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1840 ret = bitmap_storage_alloc(&store, chunks,
1841 !bitmap->mddev->bitmap_info.external);
1845 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
1847 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
1850 bitmap_file_unmap(&store);
1855 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
1857 store.file = bitmap->storage.file;
1858 bitmap->storage.file = NULL;
1860 if (store.sb_page && bitmap->storage.sb_page)
1861 memcpy(page_address(store.sb_page),
1862 page_address(bitmap->storage.sb_page),
1863 sizeof(bitmap_super_t));
1864 bitmap_file_unmap(&bitmap->storage);
1865 bitmap->storage = store;
1867 old_counts = bitmap->counts;
1868 bitmap->counts.bp = new_bp;
1869 bitmap->counts.pages = pages;
1870 bitmap->counts.missing_pages = pages;
1871 bitmap->counts.chunkshift = chunkshift;
1872 bitmap->counts.chunks = chunks;
1873 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
1874 BITMAP_BLOCK_SHIFT);
1876 blocks = min(old_counts.chunks << old_counts.chunkshift,
1877 chunks << chunkshift);
1879 spin_lock_irq(&bitmap->counts.lock);
1880 for (block = 0; block < blocks; ) {
1881 bitmap_counter_t *bmc_old, *bmc_new;
1884 bmc_old = bitmap_get_counter(&old_counts, block,
1886 set = bmc_old && NEEDED(*bmc_old);
1889 bmc_new = bitmap_get_counter(&bitmap->counts, block,
1891 if (*bmc_new == 0) {
1892 /* need to set on-disk bits too. */
1893 sector_t end = block + new_blocks;
1894 sector_t start = block >> chunkshift;
1895 start <<= chunkshift;
1896 while (start < end) {
1897 bitmap_file_set_bit(bitmap, block);
1898 start += 1 << chunkshift;
1901 bitmap_count_page(&bitmap->counts,
1903 bitmap_set_pending(&bitmap->counts,
1906 *bmc_new |= NEEDED_MASK;
1907 if (new_blocks < old_blocks)
1908 old_blocks = new_blocks;
1910 block += old_blocks;
1915 while (block < (chunks << chunkshift)) {
1916 bitmap_counter_t *bmc;
1917 bmc = bitmap_get_counter(&bitmap->counts, block,
1920 /* new space. It needs to be resynced, so
1921 * we set NEEDED_MASK.
1924 *bmc = NEEDED_MASK | 2;
1925 bitmap_count_page(&bitmap->counts,
1927 bitmap_set_pending(&bitmap->counts,
1931 block += new_blocks;
1933 for (i = 0; i < bitmap->storage.file_pages; i++)
1934 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1936 spin_unlock_irq(&bitmap->counts.lock);
1939 bitmap_unplug(bitmap);
1940 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
1946 EXPORT_SYMBOL_GPL(bitmap_resize);
1949 location_show(struct mddev *mddev, char *page)
1952 if (mddev->bitmap_info.file)
1953 len = sprintf(page, "file");
1954 else if (mddev->bitmap_info.offset)
1955 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1957 len = sprintf(page, "none");
1958 len += sprintf(page+len, "\n");
1963 location_store(struct mddev *mddev, const char *buf, size_t len)
1967 if (!mddev->pers->quiesce)
1969 if (mddev->recovery || mddev->sync_thread)
1973 if (mddev->bitmap || mddev->bitmap_info.file ||
1974 mddev->bitmap_info.offset) {
1975 /* bitmap already configured. Only option is to clear it */
1976 if (strncmp(buf, "none", 4) != 0)
1979 mddev->pers->quiesce(mddev, 1);
1980 bitmap_destroy(mddev);
1981 mddev->pers->quiesce(mddev, 0);
1983 mddev->bitmap_info.offset = 0;
1984 if (mddev->bitmap_info.file) {
1985 struct file *f = mddev->bitmap_info.file;
1986 mddev->bitmap_info.file = NULL;
1987 restore_bitmap_write_access(f);
1991 /* No bitmap, OK to set a location */
1993 if (strncmp(buf, "none", 4) == 0)
1994 /* nothing to be done */;
1995 else if (strncmp(buf, "file:", 5) == 0) {
1996 /* Not supported yet */
2001 rv = strict_strtoll(buf+1, 10, &offset);
2003 rv = strict_strtoll(buf, 10, &offset);
2008 if (mddev->bitmap_info.external == 0 &&
2009 mddev->major_version == 0 &&
2010 offset != mddev->bitmap_info.default_offset)
2012 mddev->bitmap_info.offset = offset;
2014 mddev->pers->quiesce(mddev, 1);
2015 rv = bitmap_create(mddev);
2017 rv = bitmap_load(mddev);
2019 bitmap_destroy(mddev);
2020 mddev->bitmap_info.offset = 0;
2022 mddev->pers->quiesce(mddev, 0);
2028 if (!mddev->external) {
2029 /* Ensure new bitmap info is stored in
2030 * metadata promptly.
2032 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2033 md_wakeup_thread(mddev->thread);
2038 static struct md_sysfs_entry bitmap_location =
2039 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2041 /* 'bitmap/space' is the space available at 'location' for the
2042 * bitmap. This allows the kernel to know when it is safe to
2043 * resize the bitmap to match a resized array.
2046 space_show(struct mddev *mddev, char *page)
2048 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2052 space_store(struct mddev *mddev, const char *buf, size_t len)
2054 unsigned long sectors;
2057 rv = kstrtoul(buf, 10, §ors);
2064 if (mddev->bitmap &&
2065 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2066 return -EFBIG; /* Bitmap is too big for this small space */
2068 /* could make sure it isn't too big, but that isn't really
2069 * needed - user-space should be careful.
2071 mddev->bitmap_info.space = sectors;
2075 static struct md_sysfs_entry bitmap_space =
2076 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2079 timeout_show(struct mddev *mddev, char *page)
2082 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2083 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2085 len = sprintf(page, "%lu", secs);
2087 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2088 len += sprintf(page+len, "\n");
2093 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2095 /* timeout can be set at any time */
2096 unsigned long timeout;
2097 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2101 /* just to make sure we don't overflow... */
2102 if (timeout >= LONG_MAX / HZ)
2105 timeout = timeout * HZ / 10000;
2107 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2108 timeout = MAX_SCHEDULE_TIMEOUT-1;
2111 mddev->bitmap_info.daemon_sleep = timeout;
2112 if (mddev->thread) {
2113 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2114 * the bitmap is all clean and we don't need to
2115 * adjust the timeout right now
2117 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2118 mddev->thread->timeout = timeout;
2119 md_wakeup_thread(mddev->thread);
2125 static struct md_sysfs_entry bitmap_timeout =
2126 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2129 backlog_show(struct mddev *mddev, char *page)
2131 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2135 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2137 unsigned long backlog;
2138 int rv = strict_strtoul(buf, 10, &backlog);
2141 if (backlog > COUNTER_MAX)
2143 mddev->bitmap_info.max_write_behind = backlog;
2147 static struct md_sysfs_entry bitmap_backlog =
2148 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2151 chunksize_show(struct mddev *mddev, char *page)
2153 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2157 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2159 /* Can only be changed when no bitmap is active */
2161 unsigned long csize;
2164 rv = strict_strtoul(buf, 10, &csize);
2168 !is_power_of_2(csize))
2170 mddev->bitmap_info.chunksize = csize;
2174 static struct md_sysfs_entry bitmap_chunksize =
2175 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2177 static ssize_t metadata_show(struct mddev *mddev, char *page)
2179 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2180 ? "external" : "internal"));
2183 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2185 if (mddev->bitmap ||
2186 mddev->bitmap_info.file ||
2187 mddev->bitmap_info.offset)
2189 if (strncmp(buf, "external", 8) == 0)
2190 mddev->bitmap_info.external = 1;
2191 else if (strncmp(buf, "internal", 8) == 0)
2192 mddev->bitmap_info.external = 0;
2198 static struct md_sysfs_entry bitmap_metadata =
2199 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2201 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2205 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2208 len = sprintf(page, "\n");
2212 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2214 if (mddev->bitmap == NULL)
2216 if (strncmp(buf, "false", 5) == 0)
2217 mddev->bitmap->need_sync = 1;
2218 else if (strncmp(buf, "true", 4) == 0) {
2219 if (mddev->degraded)
2221 mddev->bitmap->need_sync = 0;
2227 static struct md_sysfs_entry bitmap_can_clear =
2228 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2231 behind_writes_used_show(struct mddev *mddev, char *page)
2233 if (mddev->bitmap == NULL)
2234 return sprintf(page, "0\n");
2235 return sprintf(page, "%lu\n",
2236 mddev->bitmap->behind_writes_used);
2240 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2243 mddev->bitmap->behind_writes_used = 0;
2247 static struct md_sysfs_entry max_backlog_used =
2248 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2249 behind_writes_used_show, behind_writes_used_reset);
2251 static struct attribute *md_bitmap_attrs[] = {
2252 &bitmap_location.attr,
2254 &bitmap_timeout.attr,
2255 &bitmap_backlog.attr,
2256 &bitmap_chunksize.attr,
2257 &bitmap_metadata.attr,
2258 &bitmap_can_clear.attr,
2259 &max_backlog_used.attr,
2262 struct attribute_group md_bitmap_group = {
2264 .attrs = md_bitmap_attrs,