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/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/raid/md.h>
30 #include <linux/raid/bitmap.h>
37 /* these are for debugging purposes only! */
39 /* define one and only one of these */
40 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
41 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
42 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
43 #define INJECT_FAULTS_4 0 /* undef */
44 #define INJECT_FAULTS_5 0 /* undef */
45 #define INJECT_FAULTS_6 0
47 /* if these are defined, the driver will fail! debug only */
48 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
49 #define INJECT_FATAL_FAULT_2 0 /* undef */
50 #define INJECT_FATAL_FAULT_3 0 /* undef */
53 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
54 #define DPRINTK(x...) do { } while(0)
58 # define PRINTK(x...) printk(KERN_DEBUG x)
64 static inline char * bmname(struct bitmap *bitmap)
66 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
71 * just a placeholder - calls kmalloc for bitmap pages
73 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
77 #ifdef INJECT_FAULTS_1
80 page = kmalloc(PAGE_SIZE, GFP_NOIO);
83 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
86 bmname(bitmap), page);
91 * for now just a placeholder -- just calls kfree for bitmap pages
93 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
95 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
100 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
102 * 1) check to see if this page is allocated, if it's not then try to alloc
103 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
104 * page pointer directly as a counter
106 * if we find our page, we increment the page's refcount so that it stays
107 * allocated while we're using it
109 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
111 unsigned char *mappage;
113 if (page >= bitmap->pages) {
115 "%s: invalid bitmap page request: %lu (> %lu)\n",
116 bmname(bitmap), page, bitmap->pages-1);
121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
130 spin_unlock_irq(&bitmap->lock);
132 /* this page has not been allocated yet */
134 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
135 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
137 /* failed - set the hijacked flag so that we can use the
138 * pointer as a counter */
139 spin_lock_irq(&bitmap->lock);
140 if (!bitmap->bp[page].map)
141 bitmap->bp[page].hijacked = 1;
147 spin_lock_irq(&bitmap->lock);
149 /* recheck the page */
151 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
152 /* somebody beat us to getting the page */
153 bitmap_free_page(bitmap, mappage);
157 /* no page was in place and we have one, so install it */
159 memset(mappage, 0, PAGE_SIZE);
160 bitmap->bp[page].map = mappage;
161 bitmap->missing_pages--;
167 /* if page is completely empty, put it back on the free list, or dealloc it */
168 /* if page was hijacked, unmark the flag so it might get alloced next time */
169 /* Note: lock should be held when calling this */
170 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
174 if (bitmap->bp[page].count) /* page is still busy */
177 /* page is no longer in use, it can be released */
179 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
180 bitmap->bp[page].hijacked = 0;
181 bitmap->bp[page].map = NULL;
185 /* normal case, free the page */
188 /* actually ... let's not. We will probably need the page again exactly when
189 * memory is tight and we are flusing to disk
193 ptr = bitmap->bp[page].map;
194 bitmap->bp[page].map = NULL;
195 bitmap->missing_pages++;
196 bitmap_free_page(bitmap, ptr);
203 * bitmap file handling - read and write the bitmap file and its superblock
206 /* copy the pathname of a file to a buffer */
207 char *file_path(struct file *file, char *buf, int count)
215 d = file->f_path.dentry;
216 v = file->f_path.mnt;
218 buf = d_path(d, v, buf, count);
220 return IS_ERR(buf) ? NULL : buf;
224 * basic page I/O operations
227 /* IO operations when bitmap is stored near all superblocks */
228 static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
230 /* choose a good rdev and read the page from there */
233 struct list_head *tmp;
234 struct page *page = alloc_page(GFP_KERNEL);
238 return ERR_PTR(-ENOMEM);
240 ITERATE_RDEV(mddev, rdev, tmp) {
241 if (! test_bit(In_sync, &rdev->flags)
242 || test_bit(Faulty, &rdev->flags))
245 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
247 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
249 attach_page_buffers(page, NULL); /* so that free_buffer will
254 return ERR_PTR(-EIO);
258 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
261 struct list_head *tmp;
262 mddev_t *mddev = bitmap->mddev;
264 ITERATE_RDEV(mddev, rdev, tmp)
265 if (test_bit(In_sync, &rdev->flags)
266 && !test_bit(Faulty, &rdev->flags)) {
267 int size = PAGE_SIZE;
268 if (page->index == bitmap->file_pages-1)
269 size = roundup(bitmap->last_page_size,
270 bdev_hardsect_size(rdev->bdev));
271 /* Just make sure we aren't corrupting data or
274 if (bitmap->offset < 0) {
275 /* DATA BITMAP METADATA */
277 + page->index * (PAGE_SIZE/512)
279 /* bitmap runs in to metadata */
281 if (rdev->data_offset + mddev->size*2
282 > rdev->sb_offset*2 + bitmap->offset)
283 /* data runs in to bitmap */
285 } else if (rdev->sb_offset*2 < rdev->data_offset) {
286 /* METADATA BITMAP DATA */
287 if (rdev->sb_offset*2
289 + page->index*(PAGE_SIZE/512) + size/512
291 /* bitmap runs in to data */
294 /* DATA METADATA BITMAP - no problems */
296 md_super_write(mddev, rdev,
297 (rdev->sb_offset<<1) + bitmap->offset
298 + page->index * (PAGE_SIZE/512),
304 md_super_wait(mddev);
309 * write out a page to a file
311 static int write_page(struct bitmap *bitmap, struct page *page, int wait)
313 struct buffer_head *bh;
315 if (bitmap->file == NULL) {
316 switch (write_sb_page(bitmap, page, wait)) {
318 bitmap->flags |= BITMAP_WRITE_ERROR;
324 bh = page_buffers(page);
326 while (bh && bh->b_blocknr) {
327 atomic_inc(&bitmap->pending_writes);
328 set_buffer_locked(bh);
329 set_buffer_mapped(bh);
330 submit_bh(WRITE, bh);
331 bh = bh->b_this_page;
335 wait_event(bitmap->write_wait,
336 atomic_read(&bitmap->pending_writes)==0);
337 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
342 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
344 struct bitmap *bitmap = bh->b_private;
348 spin_lock_irqsave(&bitmap->lock, flags);
349 bitmap->flags |= BITMAP_WRITE_ERROR;
350 spin_unlock_irqrestore(&bitmap->lock, flags);
352 if (atomic_dec_and_test(&bitmap->pending_writes))
353 wake_up(&bitmap->write_wait);
356 /* copied from buffer.c */
358 __clear_page_buffers(struct page *page)
360 ClearPagePrivate(page);
361 set_page_private(page, 0);
362 page_cache_release(page);
364 static void free_buffers(struct page *page)
366 struct buffer_head *bh = page_buffers(page);
369 struct buffer_head *next = bh->b_this_page;
370 free_buffer_head(bh);
373 __clear_page_buffers(page);
377 /* read a page from a file.
378 * We both read the page, and attach buffers to the page to record the
379 * address of each block (using bmap). These addresses will be used
380 * to write the block later, completely bypassing the filesystem.
381 * This usage is similar to how swap files are handled, and allows us
382 * to write to a file with no concerns of memory allocation failing.
384 static struct page *read_page(struct file *file, unsigned long index,
385 struct bitmap *bitmap,
388 struct page *page = NULL;
389 struct inode *inode = file->f_path.dentry->d_inode;
390 struct buffer_head *bh;
393 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
394 (unsigned long long)index << PAGE_SHIFT);
396 page = alloc_page(GFP_KERNEL);
398 page = ERR_PTR(-ENOMEM);
402 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
405 page = ERR_PTR(-ENOMEM);
408 attach_page_buffers(page, bh);
409 block = index << (PAGE_SHIFT - inode->i_blkbits);
414 bh->b_blocknr = bmap(inode, block);
415 if (bh->b_blocknr == 0) {
416 /* Cannot use this file! */
418 page = ERR_PTR(-EINVAL);
421 bh->b_bdev = inode->i_sb->s_bdev;
422 if (count < (1<<inode->i_blkbits))
425 count -= (1<<inode->i_blkbits);
427 bh->b_end_io = end_bitmap_write;
428 bh->b_private = bitmap;
429 atomic_inc(&bitmap->pending_writes);
430 set_buffer_locked(bh);
431 set_buffer_mapped(bh);
435 bh = bh->b_this_page;
439 wait_event(bitmap->write_wait,
440 atomic_read(&bitmap->pending_writes)==0);
441 if (bitmap->flags & BITMAP_WRITE_ERROR) {
443 page = ERR_PTR(-EIO);
447 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
449 (unsigned long long)index << PAGE_SHIFT,
455 * bitmap file superblock operations
458 /* update the event counter and sync the superblock to disk */
459 int bitmap_update_sb(struct bitmap *bitmap)
464 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
466 spin_lock_irqsave(&bitmap->lock, flags);
467 if (!bitmap->sb_page) { /* no superblock */
468 spin_unlock_irqrestore(&bitmap->lock, flags);
471 spin_unlock_irqrestore(&bitmap->lock, flags);
472 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
473 sb->events = cpu_to_le64(bitmap->mddev->events);
474 if (!bitmap->mddev->degraded)
475 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
476 kunmap_atomic(sb, KM_USER0);
477 return write_page(bitmap, bitmap->sb_page, 1);
480 /* print out the bitmap file superblock */
481 void bitmap_print_sb(struct bitmap *bitmap)
485 if (!bitmap || !bitmap->sb_page)
487 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
488 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
489 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
490 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
491 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
492 *(__u32 *)(sb->uuid+0),
493 *(__u32 *)(sb->uuid+4),
494 *(__u32 *)(sb->uuid+8),
495 *(__u32 *)(sb->uuid+12));
496 printk(KERN_DEBUG " events: %llu\n",
497 (unsigned long long) le64_to_cpu(sb->events));
498 printk(KERN_DEBUG "events cleared: %llu\n",
499 (unsigned long long) le64_to_cpu(sb->events_cleared));
500 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
501 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
502 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
503 printk(KERN_DEBUG " sync size: %llu KB\n",
504 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
505 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
506 kunmap_atomic(sb, KM_USER0);
509 /* read the superblock from the bitmap file and initialize some bitmap fields */
510 static int bitmap_read_sb(struct bitmap *bitmap)
514 unsigned long chunksize, daemon_sleep, write_behind;
515 unsigned long long events;
518 /* page 0 is the superblock, read it... */
520 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
521 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
523 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
525 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
527 if (IS_ERR(bitmap->sb_page)) {
528 err = PTR_ERR(bitmap->sb_page);
529 bitmap->sb_page = NULL;
533 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
535 chunksize = le32_to_cpu(sb->chunksize);
536 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
537 write_behind = le32_to_cpu(sb->write_behind);
539 /* verify that the bitmap-specific fields are valid */
540 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
541 reason = "bad magic";
542 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
543 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
544 reason = "unrecognized superblock version";
545 else if (chunksize < PAGE_SIZE)
546 reason = "bitmap chunksize too small";
547 else if ((1 << ffz(~chunksize)) != chunksize)
548 reason = "bitmap chunksize not a power of 2";
549 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
550 reason = "daemon sleep period out of range";
551 else if (write_behind > COUNTER_MAX)
552 reason = "write-behind limit out of range (0 - 16383)";
554 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
555 bmname(bitmap), reason);
559 /* keep the array size field of the bitmap superblock up to date */
560 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
562 if (!bitmap->mddev->persistent)
566 * if we have a persistent array superblock, compare the
567 * bitmap's UUID and event counter to the mddev's
569 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
570 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
574 events = le64_to_cpu(sb->events);
575 if (events < bitmap->mddev->events) {
576 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
577 "-- forcing full recovery\n", bmname(bitmap), events,
578 (unsigned long long) bitmap->mddev->events);
579 sb->state |= cpu_to_le32(BITMAP_STALE);
582 /* assign fields using values from superblock */
583 bitmap->chunksize = chunksize;
584 bitmap->daemon_sleep = daemon_sleep;
585 bitmap->daemon_lastrun = jiffies;
586 bitmap->max_write_behind = write_behind;
587 bitmap->flags |= le32_to_cpu(sb->state);
588 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
589 bitmap->flags |= BITMAP_HOSTENDIAN;
590 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
591 if (sb->state & cpu_to_le32(BITMAP_STALE))
592 bitmap->events_cleared = bitmap->mddev->events;
595 kunmap_atomic(sb, KM_USER0);
597 bitmap_print_sb(bitmap);
601 enum bitmap_mask_op {
606 /* record the state of the bitmap in the superblock */
607 static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
608 enum bitmap_mask_op op)
613 spin_lock_irqsave(&bitmap->lock, flags);
614 if (!bitmap->sb_page) { /* can't set the state */
615 spin_unlock_irqrestore(&bitmap->lock, flags);
618 spin_unlock_irqrestore(&bitmap->lock, flags);
619 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
621 case MASK_SET: sb->state |= cpu_to_le32(bits);
623 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
627 kunmap_atomic(sb, KM_USER0);
631 * general bitmap file operations
634 /* calculate the index of the page that contains this bit */
635 static inline unsigned long file_page_index(unsigned long chunk)
637 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
640 /* calculate the (bit) offset of this bit within a page */
641 static inline unsigned long file_page_offset(unsigned long chunk)
643 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
647 * return a pointer to the page in the filemap that contains the given bit
649 * this lookup is complicated by the fact that the bitmap sb might be exactly
650 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
653 static inline struct page *filemap_get_page(struct bitmap *bitmap,
656 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
657 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
661 static void bitmap_file_unmap(struct bitmap *bitmap)
663 struct page **map, *sb_page;
668 spin_lock_irqsave(&bitmap->lock, flags);
669 map = bitmap->filemap;
670 bitmap->filemap = NULL;
671 attr = bitmap->filemap_attr;
672 bitmap->filemap_attr = NULL;
673 pages = bitmap->file_pages;
674 bitmap->file_pages = 0;
675 sb_page = bitmap->sb_page;
676 bitmap->sb_page = NULL;
677 spin_unlock_irqrestore(&bitmap->lock, flags);
680 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
681 free_buffers(map[pages]);
686 free_buffers(sb_page);
689 static void bitmap_file_put(struct bitmap *bitmap)
694 spin_lock_irqsave(&bitmap->lock, flags);
697 spin_unlock_irqrestore(&bitmap->lock, flags);
700 wait_event(bitmap->write_wait,
701 atomic_read(&bitmap->pending_writes)==0);
702 bitmap_file_unmap(bitmap);
705 struct inode *inode = file->f_path.dentry->d_inode;
706 invalidate_mapping_pages(inode->i_mapping, 0, -1);
713 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
714 * then it is no longer reliable, so we stop using it and we mark the file
715 * as failed in the superblock
717 static void bitmap_file_kick(struct bitmap *bitmap)
719 char *path, *ptr = NULL;
721 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
722 bitmap_update_sb(bitmap);
725 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
727 ptr = file_path(bitmap->file, path, PAGE_SIZE);
729 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
730 bmname(bitmap), ptr ? ptr : "");
735 bitmap_file_put(bitmap);
740 enum bitmap_page_attr {
741 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
742 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
743 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
746 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
747 enum bitmap_page_attr attr)
749 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
752 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
753 enum bitmap_page_attr attr)
755 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
758 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
759 enum bitmap_page_attr attr)
761 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
765 * bitmap_file_set_bit -- called before performing a write to the md device
766 * to set (and eventually sync) a particular bit in the bitmap file
768 * we set the bit immediately, then we record the page number so that
769 * when an unplug occurs, we can flush the dirty pages out to disk
771 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
776 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
778 if (!bitmap->filemap) {
782 page = filemap_get_page(bitmap, chunk);
784 bit = file_page_offset(chunk);
787 kaddr = kmap_atomic(page, KM_USER0);
788 if (bitmap->flags & BITMAP_HOSTENDIAN)
791 ext2_set_bit(bit, kaddr);
792 kunmap_atomic(kaddr, KM_USER0);
793 PRINTK("set file bit %lu page %lu\n", bit, page->index);
795 /* record page number so it gets flushed to disk when unplug occurs */
796 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
800 /* this gets called when the md device is ready to unplug its underlying
801 * (slave) device queues -- before we let any writes go down, we need to
802 * sync the dirty pages of the bitmap file to disk */
803 int bitmap_unplug(struct bitmap *bitmap)
805 unsigned long i, flags;
806 int dirty, need_write;
814 /* look at each page to see if there are any set bits that need to be
815 * flushed out to disk */
816 for (i = 0; i < bitmap->file_pages; i++) {
817 spin_lock_irqsave(&bitmap->lock, flags);
818 if (!bitmap->filemap) {
819 spin_unlock_irqrestore(&bitmap->lock, flags);
822 page = bitmap->filemap[i];
823 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
824 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
825 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
826 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
829 spin_unlock_irqrestore(&bitmap->lock, flags);
831 if (dirty | need_write)
832 err = write_page(bitmap, page, 0);
834 if (wait) { /* if any writes were performed, we need to wait on them */
836 wait_event(bitmap->write_wait,
837 atomic_read(&bitmap->pending_writes)==0);
839 md_super_wait(bitmap->mddev);
841 if (bitmap->flags & BITMAP_WRITE_ERROR)
842 bitmap_file_kick(bitmap);
846 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
847 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
848 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
849 * memory mapping of the bitmap file
851 * if there's no bitmap file, or if the bitmap file had been
852 * previously kicked from the array, we mark all the bits as
853 * 1's in order to cause a full resync.
855 * We ignore all bits for sectors that end earlier than 'start'.
856 * This is used when reading an out-of-date bitmap...
858 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
860 unsigned long i, chunks, index, oldindex, bit;
861 struct page *page = NULL, *oldpage = NULL;
862 unsigned long num_pages, bit_cnt = 0;
864 unsigned long bytes, offset;
869 chunks = bitmap->chunks;
872 BUG_ON(!file && !bitmap->offset);
874 #ifdef INJECT_FAULTS_3
877 outofdate = bitmap->flags & BITMAP_STALE;
880 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
881 "recovery\n", bmname(bitmap));
883 bytes = (chunks + 7) / 8;
885 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
887 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
888 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
890 (unsigned long) i_size_read(file->f_mapping->host),
891 bytes + sizeof(bitmap_super_t));
897 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
898 if (!bitmap->filemap)
901 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
902 bitmap->filemap_attr = kzalloc(
903 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
905 if (!bitmap->filemap_attr)
910 for (i = 0; i < chunks; i++) {
912 index = file_page_index(i);
913 bit = file_page_offset(i);
914 if (index != oldindex) { /* this is a new page, read it in */
916 /* unmap the old page, we're done with it */
917 if (index == num_pages-1)
918 count = bytes + sizeof(bitmap_super_t)
924 * if we're here then the superblock page
925 * contains some bits (PAGE_SIZE != sizeof sb)
926 * we've already read it in, so just use it
928 page = bitmap->sb_page;
929 offset = sizeof(bitmap_super_t);
931 page = read_page(file, index, bitmap, count);
934 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
937 if (IS_ERR(page)) { /* read error */
947 * if bitmap is out of date, dirty the
948 * whole page and write it out
950 paddr = kmap_atomic(page, KM_USER0);
951 memset(paddr + offset, 0xff,
953 kunmap_atomic(paddr, KM_USER0);
954 ret = write_page(bitmap, page, 1);
956 /* release, page not in filemap yet */
962 bitmap->filemap[bitmap->file_pages++] = page;
963 bitmap->last_page_size = count;
965 paddr = kmap_atomic(page, KM_USER0);
966 if (bitmap->flags & BITMAP_HOSTENDIAN)
967 b = test_bit(bit, paddr);
969 b = ext2_test_bit(bit, paddr);
970 kunmap_atomic(paddr, KM_USER0);
972 /* if the disk bit is set, set the memory bit */
973 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
974 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
977 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
981 /* everything went OK */
983 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
985 if (bit_cnt) { /* Kick recovery if any bits were set */
986 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
987 md_wakeup_thread(bitmap->mddev->thread);
991 printk(KERN_INFO "%s: bitmap initialized from disk: "
992 "read %lu/%lu pages, set %lu bits, status: %d\n",
993 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
998 void bitmap_write_all(struct bitmap *bitmap)
1000 /* We don't actually write all bitmap blocks here,
1001 * just flag them as needing to be written
1005 for (i=0; i < bitmap->file_pages; i++)
1006 set_page_attr(bitmap, bitmap->filemap[i],
1007 BITMAP_PAGE_NEEDWRITE);
1011 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1013 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1014 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1015 bitmap->bp[page].count += inc;
1017 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1018 (unsigned long long)offset, inc, bitmap->bp[page].count);
1020 bitmap_checkfree(bitmap, page);
1022 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1023 sector_t offset, int *blocks,
1027 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1031 int bitmap_daemon_work(struct bitmap *bitmap)
1034 unsigned long flags;
1035 struct page *page = NULL, *lastpage = NULL;
1042 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1044 bitmap->daemon_lastrun = jiffies;
1046 for (j = 0; j < bitmap->chunks; j++) {
1047 bitmap_counter_t *bmc;
1048 spin_lock_irqsave(&bitmap->lock, flags);
1049 if (!bitmap->filemap) {
1050 /* error or shutdown */
1051 spin_unlock_irqrestore(&bitmap->lock, flags);
1055 page = filemap_get_page(bitmap, j);
1057 if (page != lastpage) {
1058 /* skip this page unless it's marked as needing cleaning */
1059 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1060 int need_write = test_page_attr(bitmap, page,
1061 BITMAP_PAGE_NEEDWRITE);
1063 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1065 spin_unlock_irqrestore(&bitmap->lock, flags);
1067 switch (write_page(bitmap, page, 0)) {
1071 bitmap_file_kick(bitmap);
1077 /* grab the new page, sync and release the old */
1078 if (lastpage != NULL) {
1079 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1080 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1081 spin_unlock_irqrestore(&bitmap->lock, flags);
1082 err = write_page(bitmap, lastpage, 0);
1084 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1085 spin_unlock_irqrestore(&bitmap->lock, flags);
1088 bitmap_file_kick(bitmap);
1090 spin_unlock_irqrestore(&bitmap->lock, flags);
1093 printk("bitmap clean at page %lu\n", j);
1095 spin_lock_irqsave(&bitmap->lock, flags);
1096 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1098 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1102 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1105 *bmc=1; /* maybe clear the bit next time */
1106 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1107 } else if (*bmc == 1) {
1108 /* we can clear the bit */
1110 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1114 paddr = kmap_atomic(page, KM_USER0);
1115 if (bitmap->flags & BITMAP_HOSTENDIAN)
1116 clear_bit(file_page_offset(j), paddr);
1118 ext2_clear_bit(file_page_offset(j), paddr);
1119 kunmap_atomic(paddr, KM_USER0);
1122 spin_unlock_irqrestore(&bitmap->lock, flags);
1125 /* now sync the final page */
1126 if (lastpage != NULL) {
1127 spin_lock_irqsave(&bitmap->lock, flags);
1128 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1129 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1130 spin_unlock_irqrestore(&bitmap->lock, flags);
1131 err = write_page(bitmap, lastpage, 0);
1133 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1134 spin_unlock_irqrestore(&bitmap->lock, flags);
1141 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1142 sector_t offset, int *blocks,
1145 /* If 'create', we might release the lock and reclaim it.
1146 * The lock must have been taken with interrupts enabled.
1147 * If !create, we don't release the lock.
1149 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1150 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1151 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1154 if (bitmap_checkpage(bitmap, page, create) < 0) {
1155 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1156 *blocks = csize - (offset & (csize- 1));
1159 /* now locked ... */
1161 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1162 /* should we use the first or second counter field
1163 * of the hijacked pointer? */
1164 int hi = (pageoff > PAGE_COUNTER_MASK);
1165 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1166 PAGE_COUNTER_SHIFT - 1);
1167 *blocks = csize - (offset & (csize- 1));
1168 return &((bitmap_counter_t *)
1169 &bitmap->bp[page].map)[hi];
1170 } else { /* page is allocated */
1171 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1172 *blocks = csize - (offset & (csize- 1));
1173 return (bitmap_counter_t *)
1174 &(bitmap->bp[page].map[pageoff]);
1178 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1180 if (!bitmap) return 0;
1183 atomic_inc(&bitmap->behind_writes);
1184 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1185 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1190 bitmap_counter_t *bmc;
1192 spin_lock_irq(&bitmap->lock);
1193 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1195 spin_unlock_irq(&bitmap->lock);
1199 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1200 DEFINE_WAIT(__wait);
1201 /* note that it is safe to do the prepare_to_wait
1202 * after the test as long as we do it before dropping
1205 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1206 TASK_UNINTERRUPTIBLE);
1207 spin_unlock_irq(&bitmap->lock);
1208 bitmap->mddev->queue
1209 ->unplug_fn(bitmap->mddev->queue);
1211 finish_wait(&bitmap->overflow_wait, &__wait);
1217 bitmap_file_set_bit(bitmap, offset);
1218 bitmap_count_page(bitmap,offset, 1);
1219 blk_plug_device(bitmap->mddev->queue);
1227 spin_unlock_irq(&bitmap->lock);
1230 if (sectors > blocks)
1237 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1238 int success, int behind)
1240 if (!bitmap) return;
1242 atomic_dec(&bitmap->behind_writes);
1243 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1244 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1249 unsigned long flags;
1250 bitmap_counter_t *bmc;
1252 spin_lock_irqsave(&bitmap->lock, flags);
1253 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1255 spin_unlock_irqrestore(&bitmap->lock, flags);
1259 if (!success && ! (*bmc & NEEDED_MASK))
1260 *bmc |= NEEDED_MASK;
1262 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1263 wake_up(&bitmap->overflow_wait);
1267 set_page_attr(bitmap,
1268 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1271 spin_unlock_irqrestore(&bitmap->lock, flags);
1273 if (sectors > blocks)
1279 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1282 bitmap_counter_t *bmc;
1284 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1286 return 1; /* always resync if no bitmap */
1288 spin_lock_irq(&bitmap->lock);
1289 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1295 else if (NEEDED(*bmc)) {
1297 if (!degraded) { /* don't set/clear bits if degraded */
1298 *bmc |= RESYNC_MASK;
1299 *bmc &= ~NEEDED_MASK;
1303 spin_unlock_irq(&bitmap->lock);
1307 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1309 bitmap_counter_t *bmc;
1310 unsigned long flags;
1312 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1313 */ if (bitmap == NULL) {
1317 spin_lock_irqsave(&bitmap->lock, flags);
1318 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1323 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1326 *bmc &= ~RESYNC_MASK;
1328 if (!NEEDED(*bmc) && aborted)
1329 *bmc |= NEEDED_MASK;
1332 set_page_attr(bitmap,
1333 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1339 spin_unlock_irqrestore(&bitmap->lock, flags);
1342 void bitmap_close_sync(struct bitmap *bitmap)
1344 /* Sync has finished, and any bitmap chunks that weren't synced
1345 * properly have been aborted. It remains to us to clear the
1346 * RESYNC bit wherever it is still on
1348 sector_t sector = 0;
1350 if (!bitmap) return;
1351 while (sector < bitmap->mddev->resync_max_sectors) {
1352 bitmap_end_sync(bitmap, sector, &blocks, 0);
1354 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1355 (unsigned long long)sector, blocks);
1356 */ sector += blocks;
1360 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1362 /* For each chunk covered by any of these sectors, set the
1363 * counter to 1 and set resync_needed. They should all
1364 * be 0 at this point
1368 bitmap_counter_t *bmc;
1369 spin_lock_irq(&bitmap->lock);
1370 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1372 spin_unlock_irq(&bitmap->lock);
1377 *bmc = 1 | (needed?NEEDED_MASK:0);
1378 bitmap_count_page(bitmap, offset, 1);
1379 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1380 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1382 spin_unlock_irq(&bitmap->lock);
1386 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1387 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1389 unsigned long chunk;
1391 for (chunk = s; chunk <= e; chunk++) {
1392 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1393 bitmap_set_memory_bits(bitmap, sec, 1);
1394 bitmap_file_set_bit(bitmap, sec);
1399 * flush out any pending updates
1401 void bitmap_flush(mddev_t *mddev)
1403 struct bitmap *bitmap = mddev->bitmap;
1406 if (!bitmap) /* there was no bitmap */
1409 /* run the daemon_work three time to ensure everything is flushed
1412 sleep = bitmap->daemon_sleep;
1413 bitmap->daemon_sleep = 0;
1414 bitmap_daemon_work(bitmap);
1415 bitmap_daemon_work(bitmap);
1416 bitmap_daemon_work(bitmap);
1417 bitmap->daemon_sleep = sleep;
1418 bitmap_update_sb(bitmap);
1422 * free memory that was allocated
1424 static void bitmap_free(struct bitmap *bitmap)
1426 unsigned long k, pages;
1427 struct bitmap_page *bp;
1429 if (!bitmap) /* there was no bitmap */
1432 /* release the bitmap file and kill the daemon */
1433 bitmap_file_put(bitmap);
1436 pages = bitmap->pages;
1438 /* free all allocated memory */
1440 if (bp) /* deallocate the page memory */
1441 for (k = 0; k < pages; k++)
1442 if (bp[k].map && !bp[k].hijacked)
1447 void bitmap_destroy(mddev_t *mddev)
1449 struct bitmap *bitmap = mddev->bitmap;
1451 if (!bitmap) /* there was no bitmap */
1454 mddev->bitmap = NULL; /* disconnect from the md device */
1456 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1458 bitmap_free(bitmap);
1462 * initialize the bitmap structure
1463 * if this returns an error, bitmap_destroy must be called to do clean up
1465 int bitmap_create(mddev_t *mddev)
1467 struct bitmap *bitmap;
1468 unsigned long blocks = mddev->resync_max_sectors;
1469 unsigned long chunks;
1470 unsigned long pages;
1471 struct file *file = mddev->bitmap_file;
1475 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1477 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1480 BUG_ON(file && mddev->bitmap_offset);
1482 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1486 spin_lock_init(&bitmap->lock);
1487 atomic_set(&bitmap->pending_writes, 0);
1488 init_waitqueue_head(&bitmap->write_wait);
1489 init_waitqueue_head(&bitmap->overflow_wait);
1491 bitmap->mddev = mddev;
1493 bitmap->file = file;
1494 bitmap->offset = mddev->bitmap_offset;
1497 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1498 SYNC_FILE_RANGE_WAIT_BEFORE |
1499 SYNC_FILE_RANGE_WRITE |
1500 SYNC_FILE_RANGE_WAIT_AFTER);
1502 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1503 err = bitmap_read_sb(bitmap);
1507 bitmap->chunkshift = ffz(~bitmap->chunksize);
1509 /* now that chunksize and chunkshift are set, we can use these macros */
1510 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1511 CHUNK_BLOCK_RATIO(bitmap);
1512 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1516 bitmap->chunks = chunks;
1517 bitmap->pages = pages;
1518 bitmap->missing_pages = pages;
1519 bitmap->counter_bits = COUNTER_BITS;
1521 bitmap->syncchunk = ~0UL;
1523 #ifdef INJECT_FATAL_FAULT_1
1526 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1532 /* now that we have some pages available, initialize the in-memory
1533 * bitmap from the on-disk bitmap */
1535 if (mddev->degraded == 0
1536 || bitmap->events_cleared == mddev->events)
1537 /* no need to keep dirty bits to optimise a re-add of a missing device */
1538 start = mddev->recovery_cp;
1539 err = bitmap_init_from_disk(bitmap, start);
1544 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1545 pages, bmname(bitmap));
1547 mddev->bitmap = bitmap;
1549 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1551 return bitmap_update_sb(bitmap);
1554 bitmap_free(bitmap);
1558 /* the bitmap API -- for raid personalities */
1559 EXPORT_SYMBOL(bitmap_startwrite);
1560 EXPORT_SYMBOL(bitmap_endwrite);
1561 EXPORT_SYMBOL(bitmap_start_sync);
1562 EXPORT_SYMBOL(bitmap_end_sync);
1563 EXPORT_SYMBOL(bitmap_unplug);
1564 EXPORT_SYMBOL(bitmap_close_sync);