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/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/config.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>
30 #include <linux/raid/md.h>
31 #include <linux/raid/bitmap.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)
112 unsigned char *mappage;
114 if (page >= bitmap->pages) {
116 "%s: invalid bitmap page request: %lu (> %lu)\n",
117 bmname(bitmap), page, bitmap->pages-1);
122 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
125 if (bitmap->bp[page].map) /* page is already allocated, just return */
131 spin_unlock_irq(&bitmap->lock);
133 /* this page has not been allocated yet */
135 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
136 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
138 /* failed - set the hijacked flag so that we can use the
139 * pointer as a counter */
140 spin_lock_irq(&bitmap->lock);
141 if (!bitmap->bp[page].map)
142 bitmap->bp[page].hijacked = 1;
148 spin_lock_irq(&bitmap->lock);
150 /* recheck the page */
152 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
153 /* somebody beat us to getting the page */
154 bitmap_free_page(bitmap, mappage);
158 /* no page was in place and we have one, so install it */
160 memset(mappage, 0, PAGE_SIZE);
161 bitmap->bp[page].map = mappage;
162 bitmap->missing_pages--;
168 /* if page is completely empty, put it back on the free list, or dealloc it */
169 /* if page was hijacked, unmark the flag so it might get alloced next time */
170 /* Note: lock should be held when calling this */
171 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
175 if (bitmap->bp[page].count) /* page is still busy */
178 /* page is no longer in use, it can be released */
180 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
181 bitmap->bp[page].hijacked = 0;
182 bitmap->bp[page].map = NULL;
186 /* normal case, free the page */
189 /* actually ... let's not. We will probably need the page again exactly when
190 * memory is tight and we are flusing to disk
194 ptr = bitmap->bp[page].map;
195 bitmap->bp[page].map = NULL;
196 bitmap->missing_pages++;
197 bitmap_free_page(bitmap, ptr);
204 * bitmap file handling - read and write the bitmap file and its superblock
207 /* copy the pathname of a file to a buffer */
208 char *file_path(struct file *file, char *buf, int count)
219 buf = d_path(d, v, buf, count);
221 return IS_ERR(buf) ? NULL : buf;
225 * basic page I/O operations
228 /* IO operations when bitmap is stored near all superblocks */
229 static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
231 /* choose a good rdev and read the page from there */
234 struct list_head *tmp;
235 struct page *page = alloc_page(GFP_KERNEL);
239 return ERR_PTR(-ENOMEM);
241 ITERATE_RDEV(mddev, rdev, tmp) {
242 if (! test_bit(In_sync, &rdev->flags)
243 || test_bit(Faulty, &rdev->flags))
246 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
248 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
253 return ERR_PTR(-EIO);
257 static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
260 struct list_head *tmp;
262 ITERATE_RDEV(mddev, rdev, tmp)
263 if (test_bit(In_sync, &rdev->flags)
264 && !test_bit(Faulty, &rdev->flags))
265 md_super_write(mddev, rdev,
266 (rdev->sb_offset<<1) + offset
267 + page->index * (PAGE_SIZE/512),
272 md_super_wait(mddev);
277 * write out a page to a file
279 static int write_page(struct bitmap *bitmap, struct page *page, int wait)
281 struct buffer_head *bh;
283 if (bitmap->file == NULL)
284 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
286 bh = page_buffers(page);
288 while (bh && bh->b_blocknr) {
289 atomic_inc(&bitmap->pending_writes);
290 set_buffer_locked(bh);
291 set_buffer_mapped(bh);
292 submit_bh(WRITE, bh);
293 bh = bh->b_this_page;
297 wait_event(bitmap->write_wait,
298 atomic_read(&bitmap->pending_writes)==0);
299 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
304 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
306 struct bitmap *bitmap = bh->b_private;
310 spin_lock_irqsave(&bitmap->lock, flags);
311 bitmap->flags |= BITMAP_WRITE_ERROR;
312 spin_unlock_irqrestore(&bitmap->lock, flags);
314 if (atomic_dec_and_test(&bitmap->pending_writes))
315 wake_up(&bitmap->write_wait);
318 /* copied from buffer.c */
320 __clear_page_buffers(struct page *page)
322 ClearPagePrivate(page);
323 set_page_private(page, 0);
324 page_cache_release(page);
326 static void free_buffers(struct page *page)
328 struct buffer_head *bh = page_buffers(page);
331 struct buffer_head *next = bh->b_this_page;
332 free_buffer_head(bh);
335 __clear_page_buffers(page);
339 /* read a page from a file.
340 * We both read the page, and attach buffers to the page to record the
341 * address of each block (using bmap). These addresses will be used
342 * to write the block later, completely bypassing the filesystem.
343 * This usage is similar to how swap files are handled, and allows us
344 * to write to a file with no concerns of memory allocation failing.
346 static struct page *read_page(struct file *file, unsigned long index,
347 struct bitmap *bitmap,
350 struct page *page = NULL;
351 struct inode *inode = file->f_dentry->d_inode;
352 loff_t pos = index << PAGE_SHIFT;
354 struct buffer_head *bh;
358 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
359 (unsigned long long)index << PAGE_SHIFT);
361 page = alloc_page(GFP_KERNEL);
363 page = ERR_PTR(-ENOMEM);
369 ret = vfs_read(file, (char __user*) page_address(page), count, &pos);
372 if (ret >= 0 && ret != count)
379 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
382 page = ERR_PTR(-ENOMEM);
385 attach_page_buffers(page, bh);
386 block = index << (PAGE_SHIFT - inode->i_blkbits);
391 bh->b_blocknr = bmap(inode, block);
392 if (bh->b_blocknr == 0) {
393 /* Cannot use this file! */
395 page = ERR_PTR(-EINVAL);
398 bh->b_bdev = inode->i_sb->s_bdev;
399 if (count < (1<<inode->i_blkbits))
402 count -= (1<<inode->i_blkbits);
404 bh->b_end_io = end_bitmap_write;
405 bh->b_private = bitmap;
408 bh = bh->b_this_page;
414 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
416 (unsigned long long)index << PAGE_SHIFT,
422 * bitmap file superblock operations
425 /* update the event counter and sync the superblock to disk */
426 int bitmap_update_sb(struct bitmap *bitmap)
431 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
433 spin_lock_irqsave(&bitmap->lock, flags);
434 if (!bitmap->sb_page) { /* no superblock */
435 spin_unlock_irqrestore(&bitmap->lock, flags);
438 spin_unlock_irqrestore(&bitmap->lock, flags);
439 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
440 sb->events = cpu_to_le64(bitmap->mddev->events);
441 if (!bitmap->mddev->degraded)
442 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
443 kunmap_atomic(sb, KM_USER0);
444 return write_page(bitmap, bitmap->sb_page, 1);
447 /* print out the bitmap file superblock */
448 void bitmap_print_sb(struct bitmap *bitmap)
452 if (!bitmap || !bitmap->sb_page)
454 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
455 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
456 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
457 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
458 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
459 *(__u32 *)(sb->uuid+0),
460 *(__u32 *)(sb->uuid+4),
461 *(__u32 *)(sb->uuid+8),
462 *(__u32 *)(sb->uuid+12));
463 printk(KERN_DEBUG " events: %llu\n",
464 (unsigned long long) le64_to_cpu(sb->events));
465 printk(KERN_DEBUG "events cleared: %llu\n",
466 (unsigned long long) le64_to_cpu(sb->events_cleared));
467 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
468 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
469 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
470 printk(KERN_DEBUG " sync size: %llu KB\n",
471 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
472 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
473 kunmap_atomic(sb, KM_USER0);
476 /* read the superblock from the bitmap file and initialize some bitmap fields */
477 static int bitmap_read_sb(struct bitmap *bitmap)
481 unsigned long chunksize, daemon_sleep, write_behind;
482 unsigned long long events;
485 /* page 0 is the superblock, read it... */
487 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, PAGE_SIZE);
489 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
491 if (IS_ERR(bitmap->sb_page)) {
492 err = PTR_ERR(bitmap->sb_page);
493 bitmap->sb_page = NULL;
497 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
499 chunksize = le32_to_cpu(sb->chunksize);
500 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
501 write_behind = le32_to_cpu(sb->write_behind);
503 /* verify that the bitmap-specific fields are valid */
504 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
505 reason = "bad magic";
506 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
507 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
508 reason = "unrecognized superblock version";
509 else if (chunksize < PAGE_SIZE)
510 reason = "bitmap chunksize too small";
511 else if ((1 << ffz(~chunksize)) != chunksize)
512 reason = "bitmap chunksize not a power of 2";
513 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
514 reason = "daemon sleep period out of range";
515 else if (write_behind > COUNTER_MAX)
516 reason = "write-behind limit out of range (0 - 16383)";
518 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
519 bmname(bitmap), reason);
523 /* keep the array size field of the bitmap superblock up to date */
524 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
526 if (!bitmap->mddev->persistent)
530 * if we have a persistent array superblock, compare the
531 * bitmap's UUID and event counter to the mddev's
533 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
534 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
538 events = le64_to_cpu(sb->events);
539 if (events < bitmap->mddev->events) {
540 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
541 "-- forcing full recovery\n", bmname(bitmap), events,
542 (unsigned long long) bitmap->mddev->events);
543 sb->state |= BITMAP_STALE;
546 /* assign fields using values from superblock */
547 bitmap->chunksize = chunksize;
548 bitmap->daemon_sleep = daemon_sleep;
549 bitmap->daemon_lastrun = jiffies;
550 bitmap->max_write_behind = write_behind;
551 bitmap->flags |= sb->state;
552 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
553 bitmap->flags |= BITMAP_HOSTENDIAN;
554 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
555 if (sb->state & BITMAP_STALE)
556 bitmap->events_cleared = bitmap->mddev->events;
559 kunmap_atomic(sb, KM_USER0);
561 bitmap_print_sb(bitmap);
565 enum bitmap_mask_op {
570 /* record the state of the bitmap in the superblock */
571 static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
572 enum bitmap_mask_op op)
577 spin_lock_irqsave(&bitmap->lock, flags);
578 if (!bitmap->sb_page) { /* can't set the state */
579 spin_unlock_irqrestore(&bitmap->lock, flags);
582 spin_unlock_irqrestore(&bitmap->lock, flags);
583 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
585 case MASK_SET: sb->state |= bits;
587 case MASK_UNSET: sb->state &= ~bits;
591 kunmap_atomic(sb, KM_USER0);
595 * general bitmap file operations
598 /* calculate the index of the page that contains this bit */
599 static inline unsigned long file_page_index(unsigned long chunk)
601 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
604 /* calculate the (bit) offset of this bit within a page */
605 static inline unsigned long file_page_offset(unsigned long chunk)
607 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
611 * return a pointer to the page in the filemap that contains the given bit
613 * this lookup is complicated by the fact that the bitmap sb might be exactly
614 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
617 static inline struct page *filemap_get_page(struct bitmap *bitmap,
620 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
624 static void bitmap_file_unmap(struct bitmap *bitmap)
626 struct page **map, *sb_page;
631 spin_lock_irqsave(&bitmap->lock, flags);
632 map = bitmap->filemap;
633 bitmap->filemap = NULL;
634 attr = bitmap->filemap_attr;
635 bitmap->filemap_attr = NULL;
636 pages = bitmap->file_pages;
637 bitmap->file_pages = 0;
638 sb_page = bitmap->sb_page;
639 bitmap->sb_page = NULL;
640 spin_unlock_irqrestore(&bitmap->lock, flags);
643 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
644 free_buffers(map[pages]);
649 free_buffers(sb_page);
652 static void bitmap_file_put(struct bitmap *bitmap)
657 spin_lock_irqsave(&bitmap->lock, flags);
660 spin_unlock_irqrestore(&bitmap->lock, flags);
663 wait_event(bitmap->write_wait,
664 atomic_read(&bitmap->pending_writes)==0);
665 bitmap_file_unmap(bitmap);
668 struct inode *inode = file->f_dentry->d_inode;
669 invalidate_inode_pages(inode->i_mapping);
676 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
677 * then it is no longer reliable, so we stop using it and we mark the file
678 * as failed in the superblock
680 static void bitmap_file_kick(struct bitmap *bitmap)
682 char *path, *ptr = NULL;
684 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
685 bitmap_update_sb(bitmap);
688 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
690 ptr = file_path(bitmap->file, path, PAGE_SIZE);
692 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
693 bmname(bitmap), ptr ? ptr : "");
698 bitmap_file_put(bitmap);
703 enum bitmap_page_attr {
704 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
705 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
706 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
709 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
710 enum bitmap_page_attr attr)
712 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
715 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
716 enum bitmap_page_attr attr)
718 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
721 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
722 enum bitmap_page_attr attr)
724 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
728 * bitmap_file_set_bit -- called before performing a write to the md device
729 * to set (and eventually sync) a particular bit in the bitmap file
731 * we set the bit immediately, then we record the page number so that
732 * when an unplug occurs, we can flush the dirty pages out to disk
734 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
739 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
741 if (!bitmap->filemap) {
745 page = filemap_get_page(bitmap, chunk);
746 bit = file_page_offset(chunk);
749 kaddr = kmap_atomic(page, KM_USER0);
750 if (bitmap->flags & BITMAP_HOSTENDIAN)
753 ext2_set_bit(bit, kaddr);
754 kunmap_atomic(kaddr, KM_USER0);
755 PRINTK("set file bit %lu page %lu\n", bit, page->index);
757 /* record page number so it gets flushed to disk when unplug occurs */
758 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
762 /* this gets called when the md device is ready to unplug its underlying
763 * (slave) device queues -- before we let any writes go down, we need to
764 * sync the dirty pages of the bitmap file to disk */
765 int bitmap_unplug(struct bitmap *bitmap)
767 unsigned long i, flags;
768 int dirty, need_write;
776 /* look at each page to see if there are any set bits that need to be
777 * flushed out to disk */
778 for (i = 0; i < bitmap->file_pages; i++) {
779 spin_lock_irqsave(&bitmap->lock, flags);
780 if (!bitmap->filemap) {
781 spin_unlock_irqrestore(&bitmap->lock, flags);
784 page = bitmap->filemap[i];
785 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
786 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
787 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
788 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
791 spin_unlock_irqrestore(&bitmap->lock, flags);
793 if (dirty | need_write)
794 err = write_page(bitmap, page, 0);
796 if (wait) { /* if any writes were performed, we need to wait on them */
798 wait_event(bitmap->write_wait,
799 atomic_read(&bitmap->pending_writes)==0);
801 md_super_wait(bitmap->mddev);
803 if (bitmap->flags & BITMAP_WRITE_ERROR)
804 bitmap_file_kick(bitmap);
808 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
809 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
810 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
811 * memory mapping of the bitmap file
813 * if there's no bitmap file, or if the bitmap file had been
814 * previously kicked from the array, we mark all the bits as
815 * 1's in order to cause a full resync.
817 * We ignore all bits for sectors that end earlier than 'start'.
818 * This is used when reading an out-of-date bitmap...
820 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
822 unsigned long i, chunks, index, oldindex, bit;
823 struct page *page = NULL, *oldpage = NULL;
824 unsigned long num_pages, bit_cnt = 0;
826 unsigned long bytes, offset;
831 chunks = bitmap->chunks;
834 BUG_ON(!file && !bitmap->offset);
836 #ifdef INJECT_FAULTS_3
839 outofdate = bitmap->flags & BITMAP_STALE;
842 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
843 "recovery\n", bmname(bitmap));
845 bytes = (chunks + 7) / 8;
847 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
849 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
850 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
852 (unsigned long) i_size_read(file->f_mapping->host),
853 bytes + sizeof(bitmap_super_t));
859 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
860 if (!bitmap->filemap)
863 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
864 bitmap->filemap_attr = kzalloc(
865 (((num_pages*4/8)+sizeof(unsigned long)-1)
866 /sizeof(unsigned long))
867 *sizeof(unsigned long),
869 if (!bitmap->filemap_attr)
874 for (i = 0; i < chunks; i++) {
876 index = file_page_index(i);
877 bit = file_page_offset(i);
878 if (index != oldindex) { /* this is a new page, read it in */
880 /* unmap the old page, we're done with it */
881 if (index == num_pages-1)
882 count = bytes - index * PAGE_SIZE;
887 * if we're here then the superblock page
888 * contains some bits (PAGE_SIZE != sizeof sb)
889 * we've already read it in, so just use it
891 page = bitmap->sb_page;
892 offset = sizeof(bitmap_super_t);
894 page = read_page(file, index, bitmap, count);
897 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
900 if (IS_ERR(page)) { /* read error */
910 * if bitmap is out of date, dirty the
911 * whole page and write it out
913 paddr = kmap_atomic(page, KM_USER0);
914 memset(paddr + offset, 0xff,
916 kunmap_atomic(paddr, KM_USER0);
917 ret = write_page(bitmap, page, 1);
919 /* release, page not in filemap yet */
925 bitmap->filemap[bitmap->file_pages++] = page;
927 paddr = kmap_atomic(page, KM_USER0);
928 if (bitmap->flags & BITMAP_HOSTENDIAN)
929 b = test_bit(bit, paddr);
931 b = ext2_test_bit(bit, paddr);
932 kunmap_atomic(paddr, KM_USER0);
934 /* if the disk bit is set, set the memory bit */
935 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
936 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
939 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
943 /* everything went OK */
945 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
947 if (bit_cnt) { /* Kick recovery if any bits were set */
948 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
949 md_wakeup_thread(bitmap->mddev->thread);
953 printk(KERN_INFO "%s: bitmap initialized from disk: "
954 "read %lu/%lu pages, set %lu bits, status: %d\n",
955 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
960 void bitmap_write_all(struct bitmap *bitmap)
962 /* We don't actually write all bitmap blocks here,
963 * just flag them as needing to be written
967 for (i=0; i < bitmap->file_pages; i++)
968 set_page_attr(bitmap, bitmap->filemap[i],
969 BITMAP_PAGE_NEEDWRITE);
973 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
975 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
976 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
977 bitmap->bp[page].count += inc;
979 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
980 (unsigned long long)offset, inc, bitmap->bp[page].count);
982 bitmap_checkfree(bitmap, page);
984 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
985 sector_t offset, int *blocks,
989 * bitmap daemon -- periodically wakes up to clean bits and flush pages
993 int bitmap_daemon_work(struct bitmap *bitmap)
997 struct page *page = NULL, *lastpage = NULL;
1004 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1006 bitmap->daemon_lastrun = jiffies;
1008 for (j = 0; j < bitmap->chunks; j++) {
1009 bitmap_counter_t *bmc;
1010 spin_lock_irqsave(&bitmap->lock, flags);
1011 if (!bitmap->filemap) {
1012 /* error or shutdown */
1013 spin_unlock_irqrestore(&bitmap->lock, flags);
1017 page = filemap_get_page(bitmap, j);
1019 if (page != lastpage) {
1020 /* skip this page unless it's marked as needing cleaning */
1021 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1022 int need_write = test_page_attr(bitmap, page,
1023 BITMAP_PAGE_NEEDWRITE);
1025 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1027 spin_unlock_irqrestore(&bitmap->lock, flags);
1029 switch (write_page(bitmap, page, 0)) {
1033 bitmap_file_kick(bitmap);
1039 /* grab the new page, sync and release the old */
1040 if (lastpage != NULL) {
1041 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1042 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1043 spin_unlock_irqrestore(&bitmap->lock, flags);
1044 err = write_page(bitmap, lastpage, 0);
1046 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1047 spin_unlock_irqrestore(&bitmap->lock, flags);
1050 bitmap_file_kick(bitmap);
1052 spin_unlock_irqrestore(&bitmap->lock, flags);
1055 printk("bitmap clean at page %lu\n", j);
1057 spin_lock_irqsave(&bitmap->lock, flags);
1058 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1060 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1064 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1067 *bmc=1; /* maybe clear the bit next time */
1068 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1069 } else if (*bmc == 1) {
1070 /* we can clear the bit */
1072 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1076 paddr = kmap_atomic(page, KM_USER0);
1077 if (bitmap->flags & BITMAP_HOSTENDIAN)
1078 clear_bit(file_page_offset(j), paddr);
1080 ext2_clear_bit(file_page_offset(j), paddr);
1081 kunmap_atomic(paddr, KM_USER0);
1084 spin_unlock_irqrestore(&bitmap->lock, flags);
1087 /* now sync the final page */
1088 if (lastpage != NULL) {
1089 spin_lock_irqsave(&bitmap->lock, flags);
1090 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1091 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1092 spin_unlock_irqrestore(&bitmap->lock, flags);
1093 err = write_page(bitmap, lastpage, 0);
1095 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1096 spin_unlock_irqrestore(&bitmap->lock, flags);
1103 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1104 sector_t offset, int *blocks,
1107 /* If 'create', we might release the lock and reclaim it.
1108 * The lock must have been taken with interrupts enabled.
1109 * If !create, we don't release the lock.
1111 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1112 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1113 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1116 if (bitmap_checkpage(bitmap, page, create) < 0) {
1117 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1118 *blocks = csize - (offset & (csize- 1));
1121 /* now locked ... */
1123 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1124 /* should we use the first or second counter field
1125 * of the hijacked pointer? */
1126 int hi = (pageoff > PAGE_COUNTER_MASK);
1127 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1128 PAGE_COUNTER_SHIFT - 1);
1129 *blocks = csize - (offset & (csize- 1));
1130 return &((bitmap_counter_t *)
1131 &bitmap->bp[page].map)[hi];
1132 } else { /* page is allocated */
1133 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1134 *blocks = csize - (offset & (csize- 1));
1135 return (bitmap_counter_t *)
1136 &(bitmap->bp[page].map[pageoff]);
1140 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1142 if (!bitmap) return 0;
1145 atomic_inc(&bitmap->behind_writes);
1146 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1147 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1152 bitmap_counter_t *bmc;
1154 spin_lock_irq(&bitmap->lock);
1155 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1157 spin_unlock_irq(&bitmap->lock);
1163 bitmap_file_set_bit(bitmap, offset);
1164 bitmap_count_page(bitmap,offset, 1);
1165 blk_plug_device(bitmap->mddev->queue);
1170 BUG_ON((*bmc & COUNTER_MAX) == COUNTER_MAX);
1173 spin_unlock_irq(&bitmap->lock);
1176 if (sectors > blocks)
1183 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1184 int success, int behind)
1186 if (!bitmap) return;
1188 atomic_dec(&bitmap->behind_writes);
1189 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1190 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1195 unsigned long flags;
1196 bitmap_counter_t *bmc;
1198 spin_lock_irqsave(&bitmap->lock, flags);
1199 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1201 spin_unlock_irqrestore(&bitmap->lock, flags);
1205 if (!success && ! (*bmc & NEEDED_MASK))
1206 *bmc |= NEEDED_MASK;
1210 set_page_attr(bitmap,
1211 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1214 spin_unlock_irqrestore(&bitmap->lock, flags);
1216 if (sectors > blocks)
1222 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1225 bitmap_counter_t *bmc;
1227 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1229 return 1; /* always resync if no bitmap */
1231 spin_lock_irq(&bitmap->lock);
1232 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1238 else if (NEEDED(*bmc)) {
1240 if (!degraded) { /* don't set/clear bits if degraded */
1241 *bmc |= RESYNC_MASK;
1242 *bmc &= ~NEEDED_MASK;
1246 spin_unlock_irq(&bitmap->lock);
1250 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1252 bitmap_counter_t *bmc;
1253 unsigned long flags;
1255 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1256 */ if (bitmap == NULL) {
1260 spin_lock_irqsave(&bitmap->lock, flags);
1261 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1266 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1269 *bmc &= ~RESYNC_MASK;
1271 if (!NEEDED(*bmc) && aborted)
1272 *bmc |= NEEDED_MASK;
1275 set_page_attr(bitmap,
1276 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1282 spin_unlock_irqrestore(&bitmap->lock, flags);
1285 void bitmap_close_sync(struct bitmap *bitmap)
1287 /* Sync has finished, and any bitmap chunks that weren't synced
1288 * properly have been aborted. It remains to us to clear the
1289 * RESYNC bit wherever it is still on
1291 sector_t sector = 0;
1293 if (!bitmap) return;
1294 while (sector < bitmap->mddev->resync_max_sectors) {
1295 bitmap_end_sync(bitmap, sector, &blocks, 0);
1297 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1298 (unsigned long long)sector, blocks);
1299 */ sector += blocks;
1303 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1305 /* For each chunk covered by any of these sectors, set the
1306 * counter to 1 and set resync_needed. They should all
1307 * be 0 at this point
1311 bitmap_counter_t *bmc;
1312 spin_lock_irq(&bitmap->lock);
1313 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1315 spin_unlock_irq(&bitmap->lock);
1320 *bmc = 1 | (needed?NEEDED_MASK:0);
1321 bitmap_count_page(bitmap, offset, 1);
1322 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1323 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1325 spin_unlock_irq(&bitmap->lock);
1330 * flush out any pending updates
1332 void bitmap_flush(mddev_t *mddev)
1334 struct bitmap *bitmap = mddev->bitmap;
1337 if (!bitmap) /* there was no bitmap */
1340 /* run the daemon_work three time to ensure everything is flushed
1343 sleep = bitmap->daemon_sleep;
1344 bitmap->daemon_sleep = 0;
1345 bitmap_daemon_work(bitmap);
1346 bitmap_daemon_work(bitmap);
1347 bitmap_daemon_work(bitmap);
1348 bitmap->daemon_sleep = sleep;
1349 bitmap_update_sb(bitmap);
1353 * free memory that was allocated
1355 static void bitmap_free(struct bitmap *bitmap)
1357 unsigned long k, pages;
1358 struct bitmap_page *bp;
1360 if (!bitmap) /* there was no bitmap */
1363 /* release the bitmap file and kill the daemon */
1364 bitmap_file_put(bitmap);
1367 pages = bitmap->pages;
1369 /* free all allocated memory */
1371 if (bp) /* deallocate the page memory */
1372 for (k = 0; k < pages; k++)
1373 if (bp[k].map && !bp[k].hijacked)
1378 void bitmap_destroy(mddev_t *mddev)
1380 struct bitmap *bitmap = mddev->bitmap;
1382 if (!bitmap) /* there was no bitmap */
1385 mddev->bitmap = NULL; /* disconnect from the md device */
1387 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1389 bitmap_free(bitmap);
1393 * initialize the bitmap structure
1394 * if this returns an error, bitmap_destroy must be called to do clean up
1396 int bitmap_create(mddev_t *mddev)
1398 struct bitmap *bitmap;
1399 unsigned long blocks = mddev->resync_max_sectors;
1400 unsigned long chunks;
1401 unsigned long pages;
1402 struct file *file = mddev->bitmap_file;
1406 BUG_ON(sizeof(bitmap_super_t) != 256);
1408 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1411 BUG_ON(file && mddev->bitmap_offset);
1413 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1417 spin_lock_init(&bitmap->lock);
1418 bitmap->mddev = mddev;
1420 bitmap->file = file;
1421 bitmap->offset = mddev->bitmap_offset;
1422 if (file) get_file(file);
1424 /* Ensure we read fresh data */
1425 invalidate_inode_pages(file->f_dentry->d_inode->i_mapping);
1427 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1428 err = bitmap_read_sb(bitmap);
1432 bitmap->chunkshift = find_first_bit(&bitmap->chunksize,
1433 sizeof(bitmap->chunksize));
1435 /* now that chunksize and chunkshift are set, we can use these macros */
1436 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1437 CHUNK_BLOCK_RATIO(bitmap);
1438 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1442 bitmap->chunks = chunks;
1443 bitmap->pages = pages;
1444 bitmap->missing_pages = pages;
1445 bitmap->counter_bits = COUNTER_BITS;
1447 bitmap->syncchunk = ~0UL;
1449 atomic_set(&bitmap->pending_writes, 0);
1450 init_waitqueue_head(&bitmap->write_wait);
1452 #ifdef INJECT_FATAL_FAULT_1
1455 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1461 /* now that we have some pages available, initialize the in-memory
1462 * bitmap from the on-disk bitmap */
1464 if (mddev->degraded == 0
1465 || bitmap->events_cleared == mddev->events)
1466 /* no need to keep dirty bits to optimise a re-add of a missing device */
1467 start = mddev->recovery_cp;
1468 err = bitmap_init_from_disk(bitmap, start);
1473 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1474 pages, bmname(bitmap));
1476 mddev->bitmap = bitmap;
1478 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1480 return bitmap_update_sb(bitmap);
1483 bitmap_free(bitmap);
1487 /* the bitmap API -- for raid personalities */
1488 EXPORT_SYMBOL(bitmap_startwrite);
1489 EXPORT_SYMBOL(bitmap_endwrite);
1490 EXPORT_SYMBOL(bitmap_start_sync);
1491 EXPORT_SYMBOL(bitmap_end_sync);
1492 EXPORT_SYMBOL(bitmap_unplug);
1493 EXPORT_SYMBOL(bitmap_close_sync);