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
10 * - added bitmap daemon (to asynchronously clear bitmap bits from disk)
16 * flush after percent set rather than just time based. (maybe both).
17 * wait if count gets too high, wake when it drops to half.
18 * allow bitmap to be mirrored with superblock (before or after...)
19 * allow hot-add to re-instate a current device.
20 * allow hot-add of bitmap after quiessing device
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/slab.h>
26 #include <linux/init.h>
27 #include <linux/config.h>
28 #include <linux/timer.h>
29 #include <linux/sched.h>
30 #include <linux/list.h>
31 #include <linux/file.h>
32 #include <linux/mount.h>
33 #include <linux/buffer_head.h>
34 #include <linux/raid/md.h>
35 #include <linux/raid/bitmap.h>
42 /* these are for debugging purposes only! */
44 /* define one and only one of these */
45 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
46 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
47 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
48 #define INJECT_FAULTS_4 0 /* undef */
49 #define INJECT_FAULTS_5 0 /* undef */
50 #define INJECT_FAULTS_6 0
52 /* if these are defined, the driver will fail! debug only */
53 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
54 #define INJECT_FATAL_FAULT_2 0 /* undef */
55 #define INJECT_FATAL_FAULT_3 0 /* undef */
58 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
59 #define DPRINTK(x...) do { } while(0)
63 # define PRINTK(x...) printk(KERN_DEBUG x)
69 static inline char * bmname(struct bitmap *bitmap)
71 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
76 * test if the bitmap is active
78 int bitmap_active(struct bitmap *bitmap)
85 spin_lock_irqsave(&bitmap->lock, flags);
86 res = bitmap->flags & BITMAP_ACTIVE;
87 spin_unlock_irqrestore(&bitmap->lock, flags);
91 #define WRITE_POOL_SIZE 256
92 /* mempool for queueing pending writes on the bitmap file */
93 static void *write_pool_alloc(gfp_t gfp_flags, void *data)
95 return kmalloc(sizeof(struct page_list), gfp_flags);
98 static void write_pool_free(void *ptr, void *data)
104 * just a placeholder - calls kmalloc for bitmap pages
106 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
110 #ifdef INJECT_FAULTS_1
113 page = kmalloc(PAGE_SIZE, GFP_NOIO);
116 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
118 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
119 bmname(bitmap), page);
124 * for now just a placeholder -- just calls kfree for bitmap pages
126 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
128 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
133 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
135 * 1) check to see if this page is allocated, if it's not then try to alloc
136 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
137 * page pointer directly as a counter
139 * if we find our page, we increment the page's refcount so that it stays
140 * allocated while we're using it
142 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
144 unsigned char *mappage;
146 if (page >= bitmap->pages) {
148 "%s: invalid bitmap page request: %lu (> %lu)\n",
149 bmname(bitmap), page, bitmap->pages-1);
154 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
157 if (bitmap->bp[page].map) /* page is already allocated, just return */
163 spin_unlock_irq(&bitmap->lock);
165 /* this page has not been allocated yet */
167 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
168 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
170 /* failed - set the hijacked flag so that we can use the
171 * pointer as a counter */
172 spin_lock_irq(&bitmap->lock);
173 if (!bitmap->bp[page].map)
174 bitmap->bp[page].hijacked = 1;
180 spin_lock_irq(&bitmap->lock);
182 /* recheck the page */
184 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
185 /* somebody beat us to getting the page */
186 bitmap_free_page(bitmap, mappage);
190 /* no page was in place and we have one, so install it */
192 memset(mappage, 0, PAGE_SIZE);
193 bitmap->bp[page].map = mappage;
194 bitmap->missing_pages--;
200 /* if page is completely empty, put it back on the free list, or dealloc it */
201 /* if page was hijacked, unmark the flag so it might get alloced next time */
202 /* Note: lock should be held when calling this */
203 static inline void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
207 if (bitmap->bp[page].count) /* page is still busy */
210 /* page is no longer in use, it can be released */
212 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
213 bitmap->bp[page].hijacked = 0;
214 bitmap->bp[page].map = NULL;
218 /* normal case, free the page */
221 /* actually ... let's not. We will probably need the page again exactly when
222 * memory is tight and we are flusing to disk
226 ptr = bitmap->bp[page].map;
227 bitmap->bp[page].map = NULL;
228 bitmap->missing_pages++;
229 bitmap_free_page(bitmap, ptr);
236 * bitmap file handling - read and write the bitmap file and its superblock
239 /* copy the pathname of a file to a buffer */
240 char *file_path(struct file *file, char *buf, int count)
251 buf = d_path(d, v, buf, count);
253 return IS_ERR(buf) ? NULL : buf;
257 * basic page I/O operations
260 /* IO operations when bitmap is stored near all superblocks */
261 static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
263 /* choose a good rdev and read the page from there */
266 struct list_head *tmp;
267 struct page *page = alloc_page(GFP_KERNEL);
271 return ERR_PTR(-ENOMEM);
273 ITERATE_RDEV(mddev, rdev, tmp) {
274 if (! test_bit(In_sync, &rdev->flags)
275 || test_bit(Faulty, &rdev->flags))
278 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
280 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
285 return ERR_PTR(-EIO);
289 static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
292 struct list_head *tmp;
294 ITERATE_RDEV(mddev, rdev, tmp)
295 if (test_bit(In_sync, &rdev->flags)
296 && !test_bit(Faulty, &rdev->flags))
297 md_super_write(mddev, rdev,
298 (rdev->sb_offset<<1) + offset
299 + page->index * (PAGE_SIZE/512),
304 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
309 * write out a page to a file
311 static int write_page(struct bitmap *bitmap, struct page *page, int wait)
315 if (bitmap->file == NULL)
316 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
321 if (TestSetPageLocked(page))
322 return -EAGAIN; /* already locked */
323 if (PageWriteback(page)) {
329 ret = page->mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE);
331 ret = page->mapping->a_ops->commit_write(NULL, page, 0,
338 set_page_dirty(page); /* force it to be written out */
341 /* add to list to be waited for by daemon */
342 struct page_list *item = mempool_alloc(bitmap->write_pool, GFP_NOIO);
344 page_cache_get(page);
345 spin_lock(&bitmap->write_lock);
346 list_add(&item->list, &bitmap->complete_pages);
347 spin_unlock(&bitmap->write_lock);
348 md_wakeup_thread(bitmap->writeback_daemon);
350 return write_one_page(page, wait);
353 /* read a page from a file, pinning it into cache, and return bytes_read */
354 static struct page *read_page(struct file *file, unsigned long index,
355 unsigned long *bytes_read)
357 struct inode *inode = file->f_mapping->host;
358 struct page *page = NULL;
359 loff_t isize = i_size_read(inode);
360 unsigned long end_index = isize >> PAGE_CACHE_SHIFT;
362 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_CACHE_SIZE,
363 (unsigned long long)index << PAGE_CACHE_SHIFT);
365 page = read_cache_page(inode->i_mapping, index,
366 (filler_t *)inode->i_mapping->a_ops->readpage, file);
369 wait_on_page_locked(page);
370 if (!PageUptodate(page) || PageError(page)) {
371 page_cache_release(page);
372 page = ERR_PTR(-EIO);
376 if (index > end_index) /* we have read beyond EOF */
378 else if (index == end_index) /* possible short read */
379 *bytes_read = isize & ~PAGE_CACHE_MASK;
381 *bytes_read = PAGE_CACHE_SIZE; /* got a full page */
384 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
385 (int)PAGE_CACHE_SIZE,
386 (unsigned long long)index << PAGE_CACHE_SHIFT,
392 * bitmap file superblock operations
395 /* update the event counter and sync the superblock to disk */
396 int bitmap_update_sb(struct bitmap *bitmap)
401 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
403 spin_lock_irqsave(&bitmap->lock, flags);
404 if (!bitmap->sb_page) { /* no superblock */
405 spin_unlock_irqrestore(&bitmap->lock, flags);
408 spin_unlock_irqrestore(&bitmap->lock, flags);
409 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
410 sb->events = cpu_to_le64(bitmap->mddev->events);
411 if (!bitmap->mddev->degraded)
412 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
413 kunmap(bitmap->sb_page);
414 return write_page(bitmap, bitmap->sb_page, 1);
417 /* print out the bitmap file superblock */
418 void bitmap_print_sb(struct bitmap *bitmap)
422 if (!bitmap || !bitmap->sb_page)
424 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
425 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
426 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
427 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
428 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
429 *(__u32 *)(sb->uuid+0),
430 *(__u32 *)(sb->uuid+4),
431 *(__u32 *)(sb->uuid+8),
432 *(__u32 *)(sb->uuid+12));
433 printk(KERN_DEBUG " events: %llu\n",
434 (unsigned long long) le64_to_cpu(sb->events));
435 printk(KERN_DEBUG "events cleared: %llu\n",
436 (unsigned long long) le64_to_cpu(sb->events_cleared));
437 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
438 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
439 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
440 printk(KERN_DEBUG " sync size: %llu KB\n",
441 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
442 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
443 kunmap(bitmap->sb_page);
446 /* read the superblock from the bitmap file and initialize some bitmap fields */
447 static int bitmap_read_sb(struct bitmap *bitmap)
451 unsigned long chunksize, daemon_sleep, write_behind;
452 unsigned long bytes_read;
453 unsigned long long events;
456 /* page 0 is the superblock, read it... */
458 bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read);
460 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
461 bytes_read = PAGE_SIZE;
463 if (IS_ERR(bitmap->sb_page)) {
464 err = PTR_ERR(bitmap->sb_page);
465 bitmap->sb_page = NULL;
469 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
471 if (bytes_read < sizeof(*sb)) { /* short read */
472 printk(KERN_INFO "%s: bitmap file superblock truncated\n",
478 chunksize = le32_to_cpu(sb->chunksize);
479 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
480 write_behind = le32_to_cpu(sb->write_behind);
482 /* verify that the bitmap-specific fields are valid */
483 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
484 reason = "bad magic";
485 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
486 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
487 reason = "unrecognized superblock version";
488 else if (chunksize < 512 || chunksize > (1024 * 1024 * 4))
489 reason = "bitmap chunksize out of range (512B - 4MB)";
490 else if ((1 << ffz(~chunksize)) != chunksize)
491 reason = "bitmap chunksize not a power of 2";
492 else if (daemon_sleep < 1 || daemon_sleep > 15)
493 reason = "daemon sleep period out of range (1-15s)";
494 else if (write_behind > COUNTER_MAX)
495 reason = "write-behind limit out of range (0 - 16383)";
497 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
498 bmname(bitmap), reason);
502 /* keep the array size field of the bitmap superblock up to date */
503 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
505 if (!bitmap->mddev->persistent)
509 * if we have a persistent array superblock, compare the
510 * bitmap's UUID and event counter to the mddev's
512 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
513 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
517 events = le64_to_cpu(sb->events);
518 if (events < bitmap->mddev->events) {
519 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
520 "-- forcing full recovery\n", bmname(bitmap), events,
521 (unsigned long long) bitmap->mddev->events);
522 sb->state |= BITMAP_STALE;
525 /* assign fields using values from superblock */
526 bitmap->chunksize = chunksize;
527 bitmap->daemon_sleep = daemon_sleep;
528 bitmap->daemon_lastrun = jiffies;
529 bitmap->max_write_behind = write_behind;
530 bitmap->flags |= sb->state;
531 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
532 bitmap->flags |= BITMAP_HOSTENDIAN;
533 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
534 if (sb->state & BITMAP_STALE)
535 bitmap->events_cleared = bitmap->mddev->events;
538 kunmap(bitmap->sb_page);
540 bitmap_print_sb(bitmap);
544 enum bitmap_mask_op {
549 /* record the state of the bitmap in the superblock */
550 static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
551 enum bitmap_mask_op op)
556 spin_lock_irqsave(&bitmap->lock, flags);
557 if (!bitmap || !bitmap->sb_page) { /* can't set the state */
558 spin_unlock_irqrestore(&bitmap->lock, flags);
561 page_cache_get(bitmap->sb_page);
562 spin_unlock_irqrestore(&bitmap->lock, flags);
563 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
565 case MASK_SET: sb->state |= bits;
567 case MASK_UNSET: sb->state &= ~bits;
571 kunmap(bitmap->sb_page);
572 page_cache_release(bitmap->sb_page);
576 * general bitmap file operations
579 /* calculate the index of the page that contains this bit */
580 static inline unsigned long file_page_index(unsigned long chunk)
582 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
585 /* calculate the (bit) offset of this bit within a page */
586 static inline unsigned long file_page_offset(unsigned long chunk)
588 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
592 * return a pointer to the page in the filemap that contains the given bit
594 * this lookup is complicated by the fact that the bitmap sb might be exactly
595 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
598 static inline struct page *filemap_get_page(struct bitmap *bitmap,
601 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
605 static void bitmap_file_unmap(struct bitmap *bitmap)
607 struct page **map, *sb_page;
612 spin_lock_irqsave(&bitmap->lock, flags);
613 map = bitmap->filemap;
614 bitmap->filemap = NULL;
615 attr = bitmap->filemap_attr;
616 bitmap->filemap_attr = NULL;
617 pages = bitmap->file_pages;
618 bitmap->file_pages = 0;
619 sb_page = bitmap->sb_page;
620 bitmap->sb_page = NULL;
621 spin_unlock_irqrestore(&bitmap->lock, flags);
624 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
625 page_cache_release(map[pages]);
630 page_cache_release(sb_page);
633 static void bitmap_stop_daemon(struct bitmap *bitmap);
635 /* dequeue the next item in a page list -- don't call from irq context */
636 static struct page_list *dequeue_page(struct bitmap *bitmap)
638 struct page_list *item = NULL;
639 struct list_head *head = &bitmap->complete_pages;
641 spin_lock(&bitmap->write_lock);
642 if (list_empty(head))
644 item = list_entry(head->prev, struct page_list, list);
645 list_del(head->prev);
647 spin_unlock(&bitmap->write_lock);
651 static void drain_write_queues(struct bitmap *bitmap)
653 struct page_list *item;
655 while ((item = dequeue_page(bitmap))) {
656 /* don't bother to wait */
657 page_cache_release(item->page);
658 mempool_free(item, bitmap->write_pool);
661 wake_up(&bitmap->write_wait);
664 static void bitmap_file_put(struct bitmap *bitmap)
670 spin_lock_irqsave(&bitmap->lock, flags);
673 spin_unlock_irqrestore(&bitmap->lock, flags);
675 bitmap_stop_daemon(bitmap);
677 drain_write_queues(bitmap);
679 bitmap_file_unmap(bitmap);
682 inode = file->f_mapping->host;
683 spin_lock(&inode->i_lock);
684 atomic_set(&inode->i_writecount, 1); /* allow writes again */
685 spin_unlock(&inode->i_lock);
692 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
693 * then it is no longer reliable, so we stop using it and we mark the file
694 * as failed in the superblock
696 static void bitmap_file_kick(struct bitmap *bitmap)
698 char *path, *ptr = NULL;
700 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
701 bitmap_update_sb(bitmap);
704 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
706 ptr = file_path(bitmap->file, path, PAGE_SIZE);
708 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
709 bmname(bitmap), ptr ? ptr : "");
714 bitmap_file_put(bitmap);
719 enum bitmap_page_attr {
720 BITMAP_PAGE_DIRTY = 1, // there are set bits that need to be synced
721 BITMAP_PAGE_CLEAN = 2, // there are bits that might need to be cleared
722 BITMAP_PAGE_NEEDWRITE=4, // there are cleared bits that need to be synced
725 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
726 enum bitmap_page_attr attr)
728 bitmap->filemap_attr[page->index] |= attr;
731 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
732 enum bitmap_page_attr attr)
734 bitmap->filemap_attr[page->index] &= ~attr;
737 static inline unsigned long get_page_attr(struct bitmap *bitmap, struct page *page)
739 return bitmap->filemap_attr[page->index];
743 * bitmap_file_set_bit -- called before performing a write to the md device
744 * to set (and eventually sync) a particular bit in the bitmap file
746 * we set the bit immediately, then we record the page number so that
747 * when an unplug occurs, we can flush the dirty pages out to disk
749 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
754 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
756 if (!bitmap->filemap) {
760 page = filemap_get_page(bitmap, chunk);
761 bit = file_page_offset(chunk);
764 /* make sure the page stays cached until it gets written out */
765 if (! (get_page_attr(bitmap, page) & BITMAP_PAGE_DIRTY))
766 page_cache_get(page);
769 kaddr = kmap_atomic(page, KM_USER0);
770 if (bitmap->flags & BITMAP_HOSTENDIAN)
773 ext2_set_bit(bit, kaddr);
774 kunmap_atomic(kaddr, KM_USER0);
775 PRINTK("set file bit %lu page %lu\n", bit, page->index);
777 /* record page number so it gets flushed to disk when unplug occurs */
778 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
782 /* this gets called when the md device is ready to unplug its underlying
783 * (slave) device queues -- before we let any writes go down, we need to
784 * sync the dirty pages of the bitmap file to disk */
785 int bitmap_unplug(struct bitmap *bitmap)
787 unsigned long i, attr, flags;
795 /* look at each page to see if there are any set bits that need to be
796 * flushed out to disk */
797 for (i = 0; i < bitmap->file_pages; i++) {
798 spin_lock_irqsave(&bitmap->lock, flags);
799 if (!bitmap->filemap) {
800 spin_unlock_irqrestore(&bitmap->lock, flags);
803 page = bitmap->filemap[i];
804 attr = get_page_attr(bitmap, page);
805 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
806 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
807 if ((attr & BITMAP_PAGE_DIRTY))
809 spin_unlock_irqrestore(&bitmap->lock, flags);
811 if (attr & (BITMAP_PAGE_DIRTY | BITMAP_PAGE_NEEDWRITE)) {
812 err = write_page(bitmap, page, 0);
813 if (err == -EAGAIN) {
814 if (attr & BITMAP_PAGE_DIRTY)
815 err = write_page(bitmap, page, 1);
823 if (wait) { /* if any writes were performed, we need to wait on them */
825 spin_lock_irq(&bitmap->write_lock);
826 wait_event_lock_irq(bitmap->write_wait,
827 list_empty(&bitmap->complete_pages), bitmap->write_lock,
828 wake_up_process(bitmap->writeback_daemon->tsk));
829 spin_unlock_irq(&bitmap->write_lock);
831 wait_event(bitmap->mddev->sb_wait,
832 atomic_read(&bitmap->mddev->pending_writes)==0);
837 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
838 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
839 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
840 * memory mapping of the bitmap file
842 * if there's no bitmap file, or if the bitmap file had been
843 * previously kicked from the array, we mark all the bits as
844 * 1's in order to cause a full resync.
846 * We ignore all bits for sectors that end earlier than 'start'.
847 * This is used when reading an out-of-date bitmap...
849 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
851 unsigned long i, chunks, index, oldindex, bit;
852 struct page *page = NULL, *oldpage = NULL;
853 unsigned long num_pages, bit_cnt = 0;
855 unsigned long bytes, offset, dummy;
859 chunks = bitmap->chunks;
862 BUG_ON(!file && !bitmap->offset);
864 #ifdef INJECT_FAULTS_3
867 outofdate = bitmap->flags & BITMAP_STALE;
870 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
871 "recovery\n", bmname(bitmap));
873 bytes = (chunks + 7) / 8;
875 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
877 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
878 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
880 (unsigned long) i_size_read(file->f_mapping->host),
881 bytes + sizeof(bitmap_super_t));
887 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
888 if (!bitmap->filemap)
891 bitmap->filemap_attr = kmalloc(sizeof(long) * num_pages, GFP_KERNEL);
892 if (!bitmap->filemap_attr)
895 memset(bitmap->filemap_attr, 0, sizeof(long) * num_pages);
899 for (i = 0; i < chunks; i++) {
901 index = file_page_index(i);
902 bit = file_page_offset(i);
903 if (index != oldindex) { /* this is a new page, read it in */
904 /* unmap the old page, we're done with it */
909 * if we're here then the superblock page
910 * contains some bits (PAGE_SIZE != sizeof sb)
911 * we've already read it in, so just use it
913 page = bitmap->sb_page;
914 offset = sizeof(bitmap_super_t);
916 page = read_page(file, index, &dummy);
919 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
922 if (IS_ERR(page)) { /* read error */
933 * if bitmap is out of date, dirty the
934 * whole page and write it out
936 memset(page_address(page) + offset, 0xff,
938 ret = write_page(bitmap, page, 1);
941 /* release, page not in filemap yet */
942 page_cache_release(page);
947 bitmap->filemap[bitmap->file_pages++] = page;
949 if (bitmap->flags & BITMAP_HOSTENDIAN)
950 b = test_bit(bit, page_address(page));
952 b = ext2_test_bit(bit, page_address(page));
954 /* if the disk bit is set, set the memory bit */
955 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
956 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
959 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
963 /* everything went OK */
965 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
967 if (page) /* unmap the last page */
970 if (bit_cnt) { /* Kick recovery if any bits were set */
971 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
972 md_wakeup_thread(bitmap->mddev->thread);
976 printk(KERN_INFO "%s: bitmap initialized from disk: "
977 "read %lu/%lu pages, set %lu bits, status: %d\n",
978 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
983 void bitmap_write_all(struct bitmap *bitmap)
985 /* We don't actually write all bitmap blocks here,
986 * just flag them as needing to be written
989 unsigned long chunks = bitmap->chunks;
990 unsigned long bytes = (chunks+7)/8 + sizeof(bitmap_super_t);
991 unsigned long num_pages = (bytes + PAGE_SIZE-1) / PAGE_SIZE;
993 bitmap->filemap_attr[num_pages] |= BITMAP_PAGE_NEEDWRITE;
997 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
999 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1000 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1001 bitmap->bp[page].count += inc;
1003 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1004 (unsigned long long)offset, inc, bitmap->bp[page].count);
1006 bitmap_checkfree(bitmap, page);
1008 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1009 sector_t offset, int *blocks,
1013 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1017 int bitmap_daemon_work(struct bitmap *bitmap)
1020 unsigned long flags;
1021 struct page *page = NULL, *lastpage = NULL;
1028 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1030 bitmap->daemon_lastrun = jiffies;
1032 for (j = 0; j < bitmap->chunks; j++) {
1033 bitmap_counter_t *bmc;
1034 spin_lock_irqsave(&bitmap->lock, flags);
1035 if (!bitmap->filemap) {
1036 /* error or shutdown */
1037 spin_unlock_irqrestore(&bitmap->lock, flags);
1041 page = filemap_get_page(bitmap, j);
1043 if (page != lastpage) {
1044 /* skip this page unless it's marked as needing cleaning */
1045 if (!((attr=get_page_attr(bitmap, page)) & BITMAP_PAGE_CLEAN)) {
1046 if (attr & BITMAP_PAGE_NEEDWRITE) {
1047 page_cache_get(page);
1048 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1050 spin_unlock_irqrestore(&bitmap->lock, flags);
1051 if (attr & BITMAP_PAGE_NEEDWRITE) {
1052 switch (write_page(bitmap, page, 0)) {
1054 set_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1059 bitmap_file_kick(bitmap);
1061 page_cache_release(page);
1066 /* grab the new page, sync and release the old */
1067 page_cache_get(page);
1068 if (lastpage != NULL) {
1069 if (get_page_attr(bitmap, lastpage) & BITMAP_PAGE_NEEDWRITE) {
1070 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1071 spin_unlock_irqrestore(&bitmap->lock, flags);
1072 err = write_page(bitmap, lastpage, 0);
1073 if (err == -EAGAIN) {
1075 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1078 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1079 spin_unlock_irqrestore(&bitmap->lock, flags);
1082 page_cache_release(lastpage);
1084 bitmap_file_kick(bitmap);
1086 spin_unlock_irqrestore(&bitmap->lock, flags);
1090 printk("bitmap clean at page %lu\n", j);
1092 spin_lock_irqsave(&bitmap->lock, flags);
1093 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1095 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1099 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1102 *bmc=1; /* maybe clear the bit next time */
1103 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1104 } else if (*bmc == 1) {
1105 /* we can clear the bit */
1107 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1111 if (bitmap->flags & BITMAP_HOSTENDIAN)
1112 clear_bit(file_page_offset(j), page_address(page));
1114 ext2_clear_bit(file_page_offset(j), page_address(page));
1117 spin_unlock_irqrestore(&bitmap->lock, flags);
1120 /* now sync the final page */
1121 if (lastpage != NULL) {
1123 spin_lock_irqsave(&bitmap->lock, flags);
1124 if (get_page_attr(bitmap, lastpage) &BITMAP_PAGE_NEEDWRITE) {
1125 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1126 spin_unlock_irqrestore(&bitmap->lock, flags);
1127 err = write_page(bitmap, lastpage, 0);
1128 if (err == -EAGAIN) {
1129 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1133 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1134 spin_unlock_irqrestore(&bitmap->lock, flags);
1137 page_cache_release(lastpage);
1143 static void daemon_exit(struct bitmap *bitmap, mdk_thread_t **daemon)
1146 unsigned long flags;
1148 /* if no one is waiting on us, we'll free the md thread struct
1149 * and exit, otherwise we let the waiter clean things up */
1150 spin_lock_irqsave(&bitmap->lock, flags);
1151 if ((dmn = *daemon)) { /* no one is waiting, cleanup and exit */
1153 spin_unlock_irqrestore(&bitmap->lock, flags);
1155 complete_and_exit(NULL, 0); /* do_exit not exported */
1157 spin_unlock_irqrestore(&bitmap->lock, flags);
1160 static void bitmap_writeback_daemon(mddev_t *mddev)
1162 struct bitmap *bitmap = mddev->bitmap;
1164 struct page_list *item;
1167 if (signal_pending(current)) {
1169 "%s: bitmap writeback daemon got signal, exiting...\n",
1175 /* about to be stopped. */
1178 PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap));
1179 /* wait on bitmap page writebacks */
1180 while ((item = dequeue_page(bitmap))) {
1182 mempool_free(item, bitmap->write_pool);
1183 PRINTK("wait on page writeback: %p\n", page);
1184 wait_on_page_writeback(page);
1185 PRINTK("finished page writeback: %p\n", page);
1187 err = PageError(page);
1188 page_cache_release(page);
1190 printk(KERN_WARNING "%s: bitmap file writeback "
1191 "failed (page %lu): %d\n",
1192 bmname(bitmap), page->index, err);
1193 bitmap_file_kick(bitmap);
1198 wake_up(&bitmap->write_wait);
1200 printk(KERN_INFO "%s: bitmap writeback daemon exiting (%d)\n",
1201 bmname(bitmap), err);
1202 daemon_exit(bitmap, &bitmap->writeback_daemon);
1206 static mdk_thread_t *bitmap_start_daemon(struct bitmap *bitmap,
1207 void (*func)(mddev_t *), char *name)
1209 mdk_thread_t *daemon;
1212 #ifdef INJECT_FATAL_FAULT_2
1215 sprintf(namebuf, "%%s_%s", name);
1216 daemon = md_register_thread(func, bitmap->mddev, namebuf);
1219 printk(KERN_ERR "%s: failed to start bitmap daemon\n",
1221 return ERR_PTR(-ECHILD);
1224 md_wakeup_thread(daemon); /* start it running */
1226 PRINTK("%s: %s daemon (pid %d) started...\n",
1227 bmname(bitmap), name, daemon->tsk->pid);
1232 static void bitmap_stop_daemon(struct bitmap *bitmap)
1234 /* the daemon can't stop itself... it'll just exit instead... */
1235 if (bitmap->writeback_daemon && ! IS_ERR(bitmap->writeback_daemon) &&
1236 current->pid != bitmap->writeback_daemon->tsk->pid) {
1237 mdk_thread_t *daemon;
1238 unsigned long flags;
1240 spin_lock_irqsave(&bitmap->lock, flags);
1241 daemon = bitmap->writeback_daemon;
1242 bitmap->writeback_daemon = NULL;
1243 spin_unlock_irqrestore(&bitmap->lock, flags);
1244 if (daemon && ! IS_ERR(daemon))
1245 md_unregister_thread(daemon); /* destroy the thread */
1249 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1250 sector_t offset, int *blocks,
1253 /* If 'create', we might release the lock and reclaim it.
1254 * The lock must have been taken with interrupts enabled.
1255 * If !create, we don't release the lock.
1257 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1258 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1259 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1262 if (bitmap_checkpage(bitmap, page, create) < 0) {
1263 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1264 *blocks = csize - (offset & (csize- 1));
1267 /* now locked ... */
1269 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1270 /* should we use the first or second counter field
1271 * of the hijacked pointer? */
1272 int hi = (pageoff > PAGE_COUNTER_MASK);
1273 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1274 PAGE_COUNTER_SHIFT - 1);
1275 *blocks = csize - (offset & (csize- 1));
1276 return &((bitmap_counter_t *)
1277 &bitmap->bp[page].map)[hi];
1278 } else { /* page is allocated */
1279 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1280 *blocks = csize - (offset & (csize- 1));
1281 return (bitmap_counter_t *)
1282 &(bitmap->bp[page].map[pageoff]);
1286 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1288 if (!bitmap) return 0;
1291 atomic_inc(&bitmap->behind_writes);
1292 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1293 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1298 bitmap_counter_t *bmc;
1300 spin_lock_irq(&bitmap->lock);
1301 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1303 spin_unlock_irq(&bitmap->lock);
1309 bitmap_file_set_bit(bitmap, offset);
1310 bitmap_count_page(bitmap,offset, 1);
1311 blk_plug_device(bitmap->mddev->queue);
1316 if ((*bmc & COUNTER_MAX) == COUNTER_MAX) BUG();
1319 spin_unlock_irq(&bitmap->lock);
1322 if (sectors > blocks)
1329 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1330 int success, int behind)
1332 if (!bitmap) return;
1334 atomic_dec(&bitmap->behind_writes);
1335 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1336 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1341 unsigned long flags;
1342 bitmap_counter_t *bmc;
1344 spin_lock_irqsave(&bitmap->lock, flags);
1345 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1347 spin_unlock_irqrestore(&bitmap->lock, flags);
1351 if (!success && ! (*bmc & NEEDED_MASK))
1352 *bmc |= NEEDED_MASK;
1356 set_page_attr(bitmap,
1357 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1360 spin_unlock_irqrestore(&bitmap->lock, flags);
1362 if (sectors > blocks)
1368 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1371 bitmap_counter_t *bmc;
1373 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1375 return 1; /* always resync if no bitmap */
1377 spin_lock_irq(&bitmap->lock);
1378 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1384 else if (NEEDED(*bmc)) {
1386 if (!degraded) { /* don't set/clear bits if degraded */
1387 *bmc |= RESYNC_MASK;
1388 *bmc &= ~NEEDED_MASK;
1392 spin_unlock_irq(&bitmap->lock);
1396 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1398 bitmap_counter_t *bmc;
1399 unsigned long flags;
1401 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1402 */ if (bitmap == NULL) {
1406 spin_lock_irqsave(&bitmap->lock, flags);
1407 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1412 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1415 *bmc &= ~RESYNC_MASK;
1417 if (!NEEDED(*bmc) && aborted)
1418 *bmc |= NEEDED_MASK;
1421 set_page_attr(bitmap,
1422 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1428 spin_unlock_irqrestore(&bitmap->lock, flags);
1431 void bitmap_close_sync(struct bitmap *bitmap)
1433 /* Sync has finished, and any bitmap chunks that weren't synced
1434 * properly have been aborted. It remains to us to clear the
1435 * RESYNC bit wherever it is still on
1437 sector_t sector = 0;
1439 if (!bitmap) return;
1440 while (sector < bitmap->mddev->resync_max_sectors) {
1441 bitmap_end_sync(bitmap, sector, &blocks, 0);
1443 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1444 (unsigned long long)sector, blocks);
1445 */ sector += blocks;
1449 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1451 /* For each chunk covered by any of these sectors, set the
1452 * counter to 1 and set resync_needed. They should all
1453 * be 0 at this point
1457 bitmap_counter_t *bmc;
1458 spin_lock_irq(&bitmap->lock);
1459 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1461 spin_unlock_irq(&bitmap->lock);
1466 *bmc = 1 | (needed?NEEDED_MASK:0);
1467 bitmap_count_page(bitmap, offset, 1);
1468 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1469 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1471 spin_unlock_irq(&bitmap->lock);
1476 * flush out any pending updates
1478 void bitmap_flush(mddev_t *mddev)
1480 struct bitmap *bitmap = mddev->bitmap;
1483 if (!bitmap) /* there was no bitmap */
1486 /* run the daemon_work three time to ensure everything is flushed
1489 sleep = bitmap->daemon_sleep;
1490 bitmap->daemon_sleep = 0;
1491 bitmap_daemon_work(bitmap);
1492 bitmap_daemon_work(bitmap);
1493 bitmap_daemon_work(bitmap);
1494 bitmap->daemon_sleep = sleep;
1495 bitmap_update_sb(bitmap);
1499 * free memory that was allocated
1501 static void bitmap_free(struct bitmap *bitmap)
1503 unsigned long k, pages;
1504 struct bitmap_page *bp;
1506 if (!bitmap) /* there was no bitmap */
1509 /* release the bitmap file and kill the daemon */
1510 bitmap_file_put(bitmap);
1513 pages = bitmap->pages;
1515 /* free all allocated memory */
1517 mempool_destroy(bitmap->write_pool);
1519 if (bp) /* deallocate the page memory */
1520 for (k = 0; k < pages; k++)
1521 if (bp[k].map && !bp[k].hijacked)
1526 void bitmap_destroy(mddev_t *mddev)
1528 struct bitmap *bitmap = mddev->bitmap;
1530 if (!bitmap) /* there was no bitmap */
1533 mddev->bitmap = NULL; /* disconnect from the md device */
1535 bitmap_free(bitmap);
1539 * initialize the bitmap structure
1540 * if this returns an error, bitmap_destroy must be called to do clean up
1542 int bitmap_create(mddev_t *mddev)
1544 struct bitmap *bitmap;
1545 unsigned long blocks = mddev->resync_max_sectors;
1546 unsigned long chunks;
1547 unsigned long pages;
1548 struct file *file = mddev->bitmap_file;
1552 BUG_ON(sizeof(bitmap_super_t) != 256);
1554 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1557 BUG_ON(file && mddev->bitmap_offset);
1559 bitmap = kmalloc(sizeof(*bitmap), GFP_KERNEL);
1563 memset(bitmap, 0, sizeof(*bitmap));
1565 spin_lock_init(&bitmap->lock);
1566 bitmap->mddev = mddev;
1568 spin_lock_init(&bitmap->write_lock);
1569 INIT_LIST_HEAD(&bitmap->complete_pages);
1570 init_waitqueue_head(&bitmap->write_wait);
1571 bitmap->write_pool = mempool_create(WRITE_POOL_SIZE, write_pool_alloc,
1572 write_pool_free, NULL);
1574 if (!bitmap->write_pool)
1577 bitmap->file = file;
1578 bitmap->offset = mddev->bitmap_offset;
1579 if (file) get_file(file);
1580 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1581 err = bitmap_read_sb(bitmap);
1585 bitmap->chunkshift = find_first_bit(&bitmap->chunksize,
1586 sizeof(bitmap->chunksize));
1588 /* now that chunksize and chunkshift are set, we can use these macros */
1589 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1590 CHUNK_BLOCK_RATIO(bitmap);
1591 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1595 bitmap->chunks = chunks;
1596 bitmap->pages = pages;
1597 bitmap->missing_pages = pages;
1598 bitmap->counter_bits = COUNTER_BITS;
1600 bitmap->syncchunk = ~0UL;
1602 #ifdef INJECT_FATAL_FAULT_1
1605 bitmap->bp = kmalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1610 memset(bitmap->bp, 0, pages * sizeof(*bitmap->bp));
1612 bitmap->flags |= BITMAP_ACTIVE;
1614 /* now that we have some pages available, initialize the in-memory
1615 * bitmap from the on-disk bitmap */
1617 if (mddev->degraded == 0
1618 || bitmap->events_cleared == mddev->events)
1619 /* no need to keep dirty bits to optimise a re-add of a missing device */
1620 start = mddev->recovery_cp;
1621 err = bitmap_init_from_disk(bitmap, start);
1626 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1627 pages, bmname(bitmap));
1629 mddev->bitmap = bitmap;
1632 /* kick off the bitmap writeback daemon */
1633 bitmap->writeback_daemon =
1634 bitmap_start_daemon(bitmap,
1635 bitmap_writeback_daemon,
1638 if (IS_ERR(bitmap->writeback_daemon))
1639 return PTR_ERR(bitmap->writeback_daemon);
1640 return bitmap_update_sb(bitmap);
1643 bitmap_free(bitmap);
1647 /* the bitmap API -- for raid personalities */
1648 EXPORT_SYMBOL(bitmap_startwrite);
1649 EXPORT_SYMBOL(bitmap_endwrite);
1650 EXPORT_SYMBOL(bitmap_start_sync);
1651 EXPORT_SYMBOL(bitmap_end_sync);
1652 EXPORT_SYMBOL(bitmap_unplug);
1653 EXPORT_SYMBOL(bitmap_close_sync);
1654 EXPORT_SYMBOL(bitmap_daemon_work);