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.
17 * allow bitmap to be mirrored with superblock (before or after...)
18 * allow hot-add to re-instate a current device.
19 * allow hot-add of bitmap after quiessing device
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/config.h>
27 #include <linux/timer.h>
28 #include <linux/sched.h>
29 #include <linux/list.h>
30 #include <linux/file.h>
31 #include <linux/mount.h>
32 #include <linux/buffer_head.h>
33 #include <linux/raid/md.h>
34 #include <linux/raid/bitmap.h>
41 /* these are for debugging purposes only! */
43 /* define one and only one of these */
44 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
45 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
46 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
47 #define INJECT_FAULTS_4 0 /* undef */
48 #define INJECT_FAULTS_5 0 /* undef */
49 #define INJECT_FAULTS_6 0
51 /* if these are defined, the driver will fail! debug only */
52 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
53 #define INJECT_FATAL_FAULT_2 0 /* undef */
54 #define INJECT_FATAL_FAULT_3 0 /* undef */
57 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
58 #define DPRINTK(x...) do { } while(0)
62 # define PRINTK(x...) printk(KERN_DEBUG x)
68 static inline char * bmname(struct bitmap *bitmap)
70 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
75 * test if the bitmap is active
77 int bitmap_active(struct bitmap *bitmap)
84 spin_lock_irqsave(&bitmap->lock, flags);
85 res = bitmap->flags & BITMAP_ACTIVE;
86 spin_unlock_irqrestore(&bitmap->lock, flags);
90 #define WRITE_POOL_SIZE 256
93 * just a placeholder - calls kmalloc for bitmap pages
95 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
99 #ifdef INJECT_FAULTS_1
102 page = kmalloc(PAGE_SIZE, GFP_NOIO);
105 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
107 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
108 bmname(bitmap), page);
113 * for now just a placeholder -- just calls kfree for bitmap pages
115 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
117 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
122 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
124 * 1) check to see if this page is allocated, if it's not then try to alloc
125 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
126 * page pointer directly as a counter
128 * if we find our page, we increment the page's refcount so that it stays
129 * allocated while we're using it
131 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
133 unsigned char *mappage;
135 if (page >= bitmap->pages) {
137 "%s: invalid bitmap page request: %lu (> %lu)\n",
138 bmname(bitmap), page, bitmap->pages-1);
143 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
146 if (bitmap->bp[page].map) /* page is already allocated, just return */
152 spin_unlock_irq(&bitmap->lock);
154 /* this page has not been allocated yet */
156 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
157 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
159 /* failed - set the hijacked flag so that we can use the
160 * pointer as a counter */
161 spin_lock_irq(&bitmap->lock);
162 if (!bitmap->bp[page].map)
163 bitmap->bp[page].hijacked = 1;
169 spin_lock_irq(&bitmap->lock);
171 /* recheck the page */
173 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
174 /* somebody beat us to getting the page */
175 bitmap_free_page(bitmap, mappage);
179 /* no page was in place and we have one, so install it */
181 memset(mappage, 0, PAGE_SIZE);
182 bitmap->bp[page].map = mappage;
183 bitmap->missing_pages--;
189 /* if page is completely empty, put it back on the free list, or dealloc it */
190 /* if page was hijacked, unmark the flag so it might get alloced next time */
191 /* Note: lock should be held when calling this */
192 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
196 if (bitmap->bp[page].count) /* page is still busy */
199 /* page is no longer in use, it can be released */
201 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
202 bitmap->bp[page].hijacked = 0;
203 bitmap->bp[page].map = NULL;
207 /* normal case, free the page */
210 /* actually ... let's not. We will probably need the page again exactly when
211 * memory is tight and we are flusing to disk
215 ptr = bitmap->bp[page].map;
216 bitmap->bp[page].map = NULL;
217 bitmap->missing_pages++;
218 bitmap_free_page(bitmap, ptr);
225 * bitmap file handling - read and write the bitmap file and its superblock
228 /* copy the pathname of a file to a buffer */
229 char *file_path(struct file *file, char *buf, int count)
240 buf = d_path(d, v, buf, count);
242 return IS_ERR(buf) ? NULL : buf;
246 * basic page I/O operations
249 /* IO operations when bitmap is stored near all superblocks */
250 static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
252 /* choose a good rdev and read the page from there */
255 struct list_head *tmp;
256 struct page *page = alloc_page(GFP_KERNEL);
260 return ERR_PTR(-ENOMEM);
262 ITERATE_RDEV(mddev, rdev, tmp) {
263 if (! test_bit(In_sync, &rdev->flags)
264 || test_bit(Faulty, &rdev->flags))
267 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
269 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
274 return ERR_PTR(-EIO);
278 static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
281 struct list_head *tmp;
283 ITERATE_RDEV(mddev, rdev, tmp)
284 if (test_bit(In_sync, &rdev->flags)
285 && !test_bit(Faulty, &rdev->flags))
286 md_super_write(mddev, rdev,
287 (rdev->sb_offset<<1) + offset
288 + page->index * (PAGE_SIZE/512),
293 md_super_wait(mddev);
298 * write out a page to a file
300 static int write_page(struct bitmap *bitmap, struct page *page, int wait)
304 if (bitmap->file == NULL)
305 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
307 flush_dcache_page(page); /* make sure visible to anyone reading the file */
312 if (TestSetPageLocked(page))
313 return -EAGAIN; /* already locked */
314 if (PageWriteback(page)) {
320 ret = page->mapping->a_ops->prepare_write(bitmap->file, page, 0, PAGE_SIZE);
322 ret = page->mapping->a_ops->commit_write(bitmap->file, page, 0,
329 set_page_dirty(page); /* force it to be written out */
332 /* add to list to be waited for */
333 struct page_list *item = mempool_alloc(bitmap->write_pool, GFP_NOIO);
336 spin_lock(&bitmap->write_lock);
337 list_add(&item->list, &bitmap->complete_pages);
338 spin_unlock(&bitmap->write_lock);
340 return write_one_page(page, wait);
343 /* read a page from a file, pinning it into cache, and return bytes_read */
344 static struct page *read_page(struct file *file, unsigned long index,
345 unsigned long *bytes_read)
347 struct inode *inode = file->f_mapping->host;
348 struct page *page = NULL;
349 loff_t isize = i_size_read(inode);
350 unsigned long end_index = isize >> PAGE_SHIFT;
352 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
353 (unsigned long long)index << PAGE_SHIFT);
355 page = read_cache_page(inode->i_mapping, index,
356 (filler_t *)inode->i_mapping->a_ops->readpage, file);
359 wait_on_page_locked(page);
360 if (!PageUptodate(page) || PageError(page)) {
362 page = ERR_PTR(-EIO);
366 if (index > end_index) /* we have read beyond EOF */
368 else if (index == end_index) /* possible short read */
369 *bytes_read = isize & ~PAGE_MASK;
371 *bytes_read = PAGE_SIZE; /* got a full page */
374 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
376 (unsigned long long)index << PAGE_SHIFT,
382 * bitmap file superblock operations
385 /* update the event counter and sync the superblock to disk */
386 int bitmap_update_sb(struct bitmap *bitmap)
391 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
393 spin_lock_irqsave(&bitmap->lock, flags);
394 if (!bitmap->sb_page) { /* no superblock */
395 spin_unlock_irqrestore(&bitmap->lock, flags);
398 spin_unlock_irqrestore(&bitmap->lock, flags);
399 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
400 sb->events = cpu_to_le64(bitmap->mddev->events);
401 if (!bitmap->mddev->degraded)
402 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
403 kunmap_atomic(sb, KM_USER0);
404 return write_page(bitmap, bitmap->sb_page, 1);
407 /* print out the bitmap file superblock */
408 void bitmap_print_sb(struct bitmap *bitmap)
412 if (!bitmap || !bitmap->sb_page)
414 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
415 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
416 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
417 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
418 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
419 *(__u32 *)(sb->uuid+0),
420 *(__u32 *)(sb->uuid+4),
421 *(__u32 *)(sb->uuid+8),
422 *(__u32 *)(sb->uuid+12));
423 printk(KERN_DEBUG " events: %llu\n",
424 (unsigned long long) le64_to_cpu(sb->events));
425 printk(KERN_DEBUG "events cleared: %llu\n",
426 (unsigned long long) le64_to_cpu(sb->events_cleared));
427 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
428 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
429 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
430 printk(KERN_DEBUG " sync size: %llu KB\n",
431 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
432 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
433 kunmap_atomic(sb, KM_USER0);
436 /* read the superblock from the bitmap file and initialize some bitmap fields */
437 static int bitmap_read_sb(struct bitmap *bitmap)
441 unsigned long chunksize, daemon_sleep, write_behind;
442 unsigned long bytes_read;
443 unsigned long long events;
446 /* page 0 is the superblock, read it... */
448 bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read);
450 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
451 bytes_read = PAGE_SIZE;
453 if (IS_ERR(bitmap->sb_page)) {
454 err = PTR_ERR(bitmap->sb_page);
455 bitmap->sb_page = NULL;
459 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
461 if (bytes_read < sizeof(*sb)) { /* short read */
462 printk(KERN_INFO "%s: bitmap file superblock truncated\n",
468 chunksize = le32_to_cpu(sb->chunksize);
469 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
470 write_behind = le32_to_cpu(sb->write_behind);
472 /* verify that the bitmap-specific fields are valid */
473 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
474 reason = "bad magic";
475 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
476 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
477 reason = "unrecognized superblock version";
478 else if (chunksize < PAGE_SIZE)
479 reason = "bitmap chunksize too small";
480 else if ((1 << ffz(~chunksize)) != chunksize)
481 reason = "bitmap chunksize not a power of 2";
482 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
483 reason = "daemon sleep period out of range";
484 else if (write_behind > COUNTER_MAX)
485 reason = "write-behind limit out of range (0 - 16383)";
487 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
488 bmname(bitmap), reason);
492 /* keep the array size field of the bitmap superblock up to date */
493 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
495 if (!bitmap->mddev->persistent)
499 * if we have a persistent array superblock, compare the
500 * bitmap's UUID and event counter to the mddev's
502 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
503 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
507 events = le64_to_cpu(sb->events);
508 if (events < bitmap->mddev->events) {
509 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
510 "-- forcing full recovery\n", bmname(bitmap), events,
511 (unsigned long long) bitmap->mddev->events);
512 sb->state |= BITMAP_STALE;
515 /* assign fields using values from superblock */
516 bitmap->chunksize = chunksize;
517 bitmap->daemon_sleep = daemon_sleep;
518 bitmap->daemon_lastrun = jiffies;
519 bitmap->max_write_behind = write_behind;
520 bitmap->flags |= sb->state;
521 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
522 bitmap->flags |= BITMAP_HOSTENDIAN;
523 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
524 if (sb->state & BITMAP_STALE)
525 bitmap->events_cleared = bitmap->mddev->events;
528 kunmap_atomic(sb, KM_USER0);
530 bitmap_print_sb(bitmap);
534 enum bitmap_mask_op {
539 /* record the state of the bitmap in the superblock */
540 static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
541 enum bitmap_mask_op op)
546 spin_lock_irqsave(&bitmap->lock, flags);
547 if (!bitmap->sb_page) { /* can't set the state */
548 spin_unlock_irqrestore(&bitmap->lock, flags);
551 get_page(bitmap->sb_page);
552 spin_unlock_irqrestore(&bitmap->lock, flags);
553 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
555 case MASK_SET: sb->state |= bits;
557 case MASK_UNSET: sb->state &= ~bits;
561 kunmap_atomic(sb, KM_USER0);
562 put_page(bitmap->sb_page);
566 * general bitmap file operations
569 /* calculate the index of the page that contains this bit */
570 static inline unsigned long file_page_index(unsigned long chunk)
572 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
575 /* calculate the (bit) offset of this bit within a page */
576 static inline unsigned long file_page_offset(unsigned long chunk)
578 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
582 * return a pointer to the page in the filemap that contains the given bit
584 * this lookup is complicated by the fact that the bitmap sb might be exactly
585 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
588 static inline struct page *filemap_get_page(struct bitmap *bitmap,
591 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
595 static void bitmap_file_unmap(struct bitmap *bitmap)
597 struct page **map, *sb_page;
602 spin_lock_irqsave(&bitmap->lock, flags);
603 map = bitmap->filemap;
604 bitmap->filemap = NULL;
605 attr = bitmap->filemap_attr;
606 bitmap->filemap_attr = NULL;
607 pages = bitmap->file_pages;
608 bitmap->file_pages = 0;
609 sb_page = bitmap->sb_page;
610 bitmap->sb_page = NULL;
611 spin_unlock_irqrestore(&bitmap->lock, flags);
614 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
615 put_page(map[pages]);
619 safe_put_page(sb_page);
622 /* dequeue the next item in a page list -- don't call from irq context */
623 static struct page_list *dequeue_page(struct bitmap *bitmap)
625 struct page_list *item = NULL;
626 struct list_head *head = &bitmap->complete_pages;
628 spin_lock(&bitmap->write_lock);
629 if (list_empty(head))
631 item = list_entry(head->prev, struct page_list, list);
632 list_del(head->prev);
634 spin_unlock(&bitmap->write_lock);
638 static void drain_write_queues(struct bitmap *bitmap)
640 struct page_list *item;
642 while ((item = dequeue_page(bitmap))) {
643 /* don't bother to wait */
644 put_page(item->page);
645 mempool_free(item, bitmap->write_pool);
649 static void bitmap_file_put(struct bitmap *bitmap)
655 spin_lock_irqsave(&bitmap->lock, flags);
658 spin_unlock_irqrestore(&bitmap->lock, flags);
660 drain_write_queues(bitmap);
662 bitmap_file_unmap(bitmap);
665 inode = file->f_mapping->host;
666 spin_lock(&inode->i_lock);
667 atomic_set(&inode->i_writecount, 1); /* allow writes again */
668 spin_unlock(&inode->i_lock);
675 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
676 * then it is no longer reliable, so we stop using it and we mark the file
677 * as failed in the superblock
679 static void bitmap_file_kick(struct bitmap *bitmap)
681 char *path, *ptr = NULL;
683 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
684 bitmap_update_sb(bitmap);
687 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
689 ptr = file_path(bitmap->file, path, PAGE_SIZE);
691 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
692 bmname(bitmap), ptr ? ptr : "");
697 bitmap_file_put(bitmap);
702 enum bitmap_page_attr {
703 BITMAP_PAGE_DIRTY = 1, // there are set bits that need to be synced
704 BITMAP_PAGE_CLEAN = 2, // there are bits that might need to be cleared
705 BITMAP_PAGE_NEEDWRITE=4, // there are cleared bits that need to be synced
708 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
709 enum bitmap_page_attr attr)
711 bitmap->filemap_attr[page->index] |= attr;
714 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
715 enum bitmap_page_attr attr)
717 bitmap->filemap_attr[page->index] &= ~attr;
720 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
721 enum bitmap_page_attr attr)
723 return bitmap->filemap_attr[page->index] & attr;
727 * bitmap_file_set_bit -- called before performing a write to the md device
728 * to set (and eventually sync) a particular bit in the bitmap file
730 * we set the bit immediately, then we record the page number so that
731 * when an unplug occurs, we can flush the dirty pages out to disk
733 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
738 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
740 if (!bitmap->filemap) {
744 page = filemap_get_page(bitmap, chunk);
745 bit = file_page_offset(chunk);
748 /* make sure the page stays cached until it gets written out */
749 if (! test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY))
753 kaddr = kmap_atomic(page, KM_USER0);
754 if (bitmap->flags & BITMAP_HOSTENDIAN)
757 ext2_set_bit(bit, kaddr);
758 kunmap_atomic(kaddr, KM_USER0);
759 PRINTK("set file bit %lu page %lu\n", bit, page->index);
761 /* record page number so it gets flushed to disk when unplug occurs */
762 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
766 static void bitmap_writeback(struct bitmap *bitmap);
768 /* this gets called when the md device is ready to unplug its underlying
769 * (slave) device queues -- before we let any writes go down, we need to
770 * sync the dirty pages of the bitmap file to disk */
771 int bitmap_unplug(struct bitmap *bitmap)
773 unsigned long i, flags;
774 int dirty, need_write;
782 /* look at each page to see if there are any set bits that need to be
783 * flushed out to disk */
784 for (i = 0; i < bitmap->file_pages; i++) {
785 spin_lock_irqsave(&bitmap->lock, flags);
786 if (!bitmap->filemap) {
787 spin_unlock_irqrestore(&bitmap->lock, flags);
790 page = bitmap->filemap[i];
791 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
792 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
793 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
794 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
797 spin_unlock_irqrestore(&bitmap->lock, flags);
799 if (dirty | need_write) {
800 err = write_page(bitmap, page, 0);
801 if (err == -EAGAIN) {
803 err = write_page(bitmap, page, 1);
811 if (wait) { /* if any writes were performed, we need to wait on them */
813 bitmap_writeback(bitmap);
815 md_super_wait(bitmap->mddev);
820 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
821 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
822 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
823 * memory mapping of the bitmap file
825 * if there's no bitmap file, or if the bitmap file had been
826 * previously kicked from the array, we mark all the bits as
827 * 1's in order to cause a full resync.
829 * We ignore all bits for sectors that end earlier than 'start'.
830 * This is used when reading an out-of-date bitmap...
832 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
834 unsigned long i, chunks, index, oldindex, bit;
835 struct page *page = NULL, *oldpage = NULL;
836 unsigned long num_pages, bit_cnt = 0;
838 unsigned long bytes, offset, dummy;
843 chunks = bitmap->chunks;
846 BUG_ON(!file && !bitmap->offset);
848 #ifdef INJECT_FAULTS_3
851 outofdate = bitmap->flags & BITMAP_STALE;
854 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
855 "recovery\n", bmname(bitmap));
857 bytes = (chunks + 7) / 8;
859 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
861 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
862 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
864 (unsigned long) i_size_read(file->f_mapping->host),
865 bytes + sizeof(bitmap_super_t));
871 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
872 if (!bitmap->filemap)
875 bitmap->filemap_attr = kzalloc(sizeof(long) * num_pages, GFP_KERNEL);
876 if (!bitmap->filemap_attr)
881 for (i = 0; i < chunks; i++) {
883 index = file_page_index(i);
884 bit = file_page_offset(i);
885 if (index != oldindex) { /* this is a new page, read it in */
886 /* unmap the old page, we're done with it */
889 * if we're here then the superblock page
890 * contains some bits (PAGE_SIZE != sizeof sb)
891 * we've already read it in, so just use it
893 page = bitmap->sb_page;
894 offset = sizeof(bitmap_super_t);
896 page = read_page(file, index, &dummy);
899 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
902 if (IS_ERR(page)) { /* read error */
912 * if bitmap is out of date, dirty the
913 * whole page and write it out
915 paddr = kmap_atomic(page, KM_USER0);
916 memset(paddr + offset, 0xff,
918 kunmap_atomic(paddr, KM_USER0);
919 ret = write_page(bitmap, page, 1);
921 /* release, page not in filemap yet */
927 bitmap->filemap[bitmap->file_pages++] = page;
929 paddr = kmap_atomic(page, KM_USER0);
930 if (bitmap->flags & BITMAP_HOSTENDIAN)
931 b = test_bit(bit, paddr);
933 b = ext2_test_bit(bit, paddr);
934 kunmap_atomic(paddr, KM_USER0);
936 /* if the disk bit is set, set the memory bit */
937 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
938 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
941 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
945 /* everything went OK */
947 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
949 if (bit_cnt) { /* Kick recovery if any bits were set */
950 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
951 md_wakeup_thread(bitmap->mddev->thread);
955 printk(KERN_INFO "%s: bitmap initialized from disk: "
956 "read %lu/%lu pages, set %lu bits, status: %d\n",
957 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
962 void bitmap_write_all(struct bitmap *bitmap)
964 /* We don't actually write all bitmap blocks here,
965 * just flag them as needing to be written
969 for (i=0; i < bitmap->file_pages; i++)
970 set_page_attr(bitmap, bitmap->filemap[i],
971 BITMAP_PAGE_NEEDWRITE);
975 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
977 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
978 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
979 bitmap->bp[page].count += inc;
981 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
982 (unsigned long long)offset, inc, bitmap->bp[page].count);
984 bitmap_checkfree(bitmap, page);
986 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
987 sector_t offset, int *blocks,
991 * bitmap daemon -- periodically wakes up to clean bits and flush pages
995 int bitmap_daemon_work(struct bitmap *bitmap)
999 struct page *page = NULL, *lastpage = NULL;
1006 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1008 bitmap->daemon_lastrun = jiffies;
1010 for (j = 0; j < bitmap->chunks; j++) {
1011 bitmap_counter_t *bmc;
1012 spin_lock_irqsave(&bitmap->lock, flags);
1013 if (!bitmap->filemap) {
1014 /* error or shutdown */
1015 spin_unlock_irqrestore(&bitmap->lock, flags);
1019 page = filemap_get_page(bitmap, j);
1021 if (page != lastpage) {
1022 /* skip this page unless it's marked as needing cleaning */
1023 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1024 int need_write = test_page_attr(bitmap, page,
1025 BITMAP_PAGE_NEEDWRITE);
1028 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1030 spin_unlock_irqrestore(&bitmap->lock, flags);
1032 switch (write_page(bitmap, page, 0)) {
1034 set_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1039 bitmap_file_kick(bitmap);
1046 /* grab the new page, sync and release the old */
1048 if (lastpage != NULL) {
1049 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1050 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1051 spin_unlock_irqrestore(&bitmap->lock, flags);
1052 err = write_page(bitmap, lastpage, 0);
1053 if (err == -EAGAIN) {
1055 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1058 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1059 spin_unlock_irqrestore(&bitmap->lock, flags);
1063 bitmap_file_kick(bitmap);
1065 spin_unlock_irqrestore(&bitmap->lock, flags);
1068 printk("bitmap clean at page %lu\n", j);
1070 spin_lock_irqsave(&bitmap->lock, flags);
1071 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1073 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1077 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1080 *bmc=1; /* maybe clear the bit next time */
1081 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1082 } else if (*bmc == 1) {
1083 /* we can clear the bit */
1085 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1089 paddr = kmap_atomic(page, KM_USER0);
1090 if (bitmap->flags & BITMAP_HOSTENDIAN)
1091 clear_bit(file_page_offset(j), paddr);
1093 ext2_clear_bit(file_page_offset(j), paddr);
1094 kunmap_atomic(paddr, KM_USER0);
1097 spin_unlock_irqrestore(&bitmap->lock, flags);
1100 /* now sync the final page */
1101 if (lastpage != NULL) {
1102 spin_lock_irqsave(&bitmap->lock, flags);
1103 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1104 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1105 spin_unlock_irqrestore(&bitmap->lock, flags);
1106 err = write_page(bitmap, lastpage, 0);
1107 if (err == -EAGAIN) {
1108 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1112 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1113 spin_unlock_irqrestore(&bitmap->lock, flags);
1122 static void bitmap_writeback(struct bitmap *bitmap)
1125 struct page_list *item;
1128 PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap));
1129 /* wait on bitmap page writebacks */
1130 while ((item = dequeue_page(bitmap))) {
1132 mempool_free(item, bitmap->write_pool);
1133 PRINTK("wait on page writeback: %p\n", page);
1134 wait_on_page_writeback(page);
1135 PRINTK("finished page writeback: %p\n", page);
1137 err = PageError(page);
1140 printk(KERN_WARNING "%s: bitmap file writeback "
1141 "failed (page %lu): %d\n",
1142 bmname(bitmap), page->index, err);
1143 bitmap_file_kick(bitmap);
1149 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1150 sector_t offset, int *blocks,
1153 /* If 'create', we might release the lock and reclaim it.
1154 * The lock must have been taken with interrupts enabled.
1155 * If !create, we don't release the lock.
1157 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1158 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1159 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1162 if (bitmap_checkpage(bitmap, page, create) < 0) {
1163 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1164 *blocks = csize - (offset & (csize- 1));
1167 /* now locked ... */
1169 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1170 /* should we use the first or second counter field
1171 * of the hijacked pointer? */
1172 int hi = (pageoff > PAGE_COUNTER_MASK);
1173 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1174 PAGE_COUNTER_SHIFT - 1);
1175 *blocks = csize - (offset & (csize- 1));
1176 return &((bitmap_counter_t *)
1177 &bitmap->bp[page].map)[hi];
1178 } else { /* page is allocated */
1179 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1180 *blocks = csize - (offset & (csize- 1));
1181 return (bitmap_counter_t *)
1182 &(bitmap->bp[page].map[pageoff]);
1186 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1188 if (!bitmap) return 0;
1191 atomic_inc(&bitmap->behind_writes);
1192 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1193 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1198 bitmap_counter_t *bmc;
1200 spin_lock_irq(&bitmap->lock);
1201 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1203 spin_unlock_irq(&bitmap->lock);
1209 bitmap_file_set_bit(bitmap, offset);
1210 bitmap_count_page(bitmap,offset, 1);
1211 blk_plug_device(bitmap->mddev->queue);
1216 BUG_ON((*bmc & COUNTER_MAX) == COUNTER_MAX);
1219 spin_unlock_irq(&bitmap->lock);
1222 if (sectors > blocks)
1229 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1230 int success, int behind)
1232 if (!bitmap) return;
1234 atomic_dec(&bitmap->behind_writes);
1235 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1236 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1241 unsigned long flags;
1242 bitmap_counter_t *bmc;
1244 spin_lock_irqsave(&bitmap->lock, flags);
1245 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1247 spin_unlock_irqrestore(&bitmap->lock, flags);
1251 if (!success && ! (*bmc & NEEDED_MASK))
1252 *bmc |= NEEDED_MASK;
1256 set_page_attr(bitmap,
1257 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1260 spin_unlock_irqrestore(&bitmap->lock, flags);
1262 if (sectors > blocks)
1268 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1271 bitmap_counter_t *bmc;
1273 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1275 return 1; /* always resync if no bitmap */
1277 spin_lock_irq(&bitmap->lock);
1278 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1284 else if (NEEDED(*bmc)) {
1286 if (!degraded) { /* don't set/clear bits if degraded */
1287 *bmc |= RESYNC_MASK;
1288 *bmc &= ~NEEDED_MASK;
1292 spin_unlock_irq(&bitmap->lock);
1296 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1298 bitmap_counter_t *bmc;
1299 unsigned long flags;
1301 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1302 */ if (bitmap == NULL) {
1306 spin_lock_irqsave(&bitmap->lock, flags);
1307 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1312 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1315 *bmc &= ~RESYNC_MASK;
1317 if (!NEEDED(*bmc) && aborted)
1318 *bmc |= NEEDED_MASK;
1321 set_page_attr(bitmap,
1322 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1328 spin_unlock_irqrestore(&bitmap->lock, flags);
1331 void bitmap_close_sync(struct bitmap *bitmap)
1333 /* Sync has finished, and any bitmap chunks that weren't synced
1334 * properly have been aborted. It remains to us to clear the
1335 * RESYNC bit wherever it is still on
1337 sector_t sector = 0;
1339 if (!bitmap) return;
1340 while (sector < bitmap->mddev->resync_max_sectors) {
1341 bitmap_end_sync(bitmap, sector, &blocks, 0);
1343 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1344 (unsigned long long)sector, blocks);
1345 */ sector += blocks;
1349 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1351 /* For each chunk covered by any of these sectors, set the
1352 * counter to 1 and set resync_needed. They should all
1353 * be 0 at this point
1357 bitmap_counter_t *bmc;
1358 spin_lock_irq(&bitmap->lock);
1359 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1361 spin_unlock_irq(&bitmap->lock);
1366 *bmc = 1 | (needed?NEEDED_MASK:0);
1367 bitmap_count_page(bitmap, offset, 1);
1368 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1369 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1371 spin_unlock_irq(&bitmap->lock);
1376 * flush out any pending updates
1378 void bitmap_flush(mddev_t *mddev)
1380 struct bitmap *bitmap = mddev->bitmap;
1383 if (!bitmap) /* there was no bitmap */
1386 /* run the daemon_work three time to ensure everything is flushed
1389 sleep = bitmap->daemon_sleep;
1390 bitmap->daemon_sleep = 0;
1391 bitmap_daemon_work(bitmap);
1392 bitmap_daemon_work(bitmap);
1393 bitmap_daemon_work(bitmap);
1394 bitmap->daemon_sleep = sleep;
1395 bitmap_update_sb(bitmap);
1399 * free memory that was allocated
1401 static void bitmap_free(struct bitmap *bitmap)
1403 unsigned long k, pages;
1404 struct bitmap_page *bp;
1406 if (!bitmap) /* there was no bitmap */
1409 /* release the bitmap file and kill the daemon */
1410 bitmap_file_put(bitmap);
1413 pages = bitmap->pages;
1415 /* free all allocated memory */
1417 mempool_destroy(bitmap->write_pool);
1419 if (bp) /* deallocate the page memory */
1420 for (k = 0; k < pages; k++)
1421 if (bp[k].map && !bp[k].hijacked)
1426 void bitmap_destroy(mddev_t *mddev)
1428 struct bitmap *bitmap = mddev->bitmap;
1430 if (!bitmap) /* there was no bitmap */
1433 mddev->bitmap = NULL; /* disconnect from the md device */
1435 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1437 bitmap_free(bitmap);
1441 * initialize the bitmap structure
1442 * if this returns an error, bitmap_destroy must be called to do clean up
1444 int bitmap_create(mddev_t *mddev)
1446 struct bitmap *bitmap;
1447 unsigned long blocks = mddev->resync_max_sectors;
1448 unsigned long chunks;
1449 unsigned long pages;
1450 struct file *file = mddev->bitmap_file;
1454 BUG_ON(sizeof(bitmap_super_t) != 256);
1456 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1459 BUG_ON(file && mddev->bitmap_offset);
1461 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1465 spin_lock_init(&bitmap->lock);
1466 bitmap->mddev = mddev;
1468 spin_lock_init(&bitmap->write_lock);
1469 INIT_LIST_HEAD(&bitmap->complete_pages);
1470 bitmap->write_pool = mempool_create_kmalloc_pool(WRITE_POOL_SIZE,
1471 sizeof(struct page_list));
1473 if (!bitmap->write_pool)
1476 bitmap->file = file;
1477 bitmap->offset = mddev->bitmap_offset;
1478 if (file) get_file(file);
1479 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1480 err = bitmap_read_sb(bitmap);
1484 bitmap->chunkshift = find_first_bit(&bitmap->chunksize,
1485 sizeof(bitmap->chunksize));
1487 /* now that chunksize and chunkshift are set, we can use these macros */
1488 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1489 CHUNK_BLOCK_RATIO(bitmap);
1490 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1494 bitmap->chunks = chunks;
1495 bitmap->pages = pages;
1496 bitmap->missing_pages = pages;
1497 bitmap->counter_bits = COUNTER_BITS;
1499 bitmap->syncchunk = ~0UL;
1501 #ifdef INJECT_FATAL_FAULT_1
1504 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1510 bitmap->flags |= BITMAP_ACTIVE;
1512 /* now that we have some pages available, initialize the in-memory
1513 * bitmap from the on-disk bitmap */
1515 if (mddev->degraded == 0
1516 || bitmap->events_cleared == mddev->events)
1517 /* no need to keep dirty bits to optimise a re-add of a missing device */
1518 start = mddev->recovery_cp;
1519 err = bitmap_init_from_disk(bitmap, start);
1524 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1525 pages, bmname(bitmap));
1527 mddev->bitmap = bitmap;
1529 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1531 return bitmap_update_sb(bitmap);
1534 bitmap_free(bitmap);
1538 /* the bitmap API -- for raid personalities */
1539 EXPORT_SYMBOL(bitmap_startwrite);
1540 EXPORT_SYMBOL(bitmap_endwrite);
1541 EXPORT_SYMBOL(bitmap_start_sync);
1542 EXPORT_SYMBOL(bitmap_end_sync);
1543 EXPORT_SYMBOL(bitmap_unplug);
1544 EXPORT_SYMBOL(bitmap_close_sync);
1545 EXPORT_SYMBOL(bitmap_daemon_work);