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/version.h>
25 #include <linux/errno.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/config.h>
29 #include <linux/timer.h>
30 #include <linux/sched.h>
31 #include <linux/list.h>
32 #include <linux/file.h>
33 #include <linux/mount.h>
34 #include <linux/buffer_head.h>
35 #include <linux/raid/md.h>
36 #include <linux/raid/bitmap.h>
43 /* these are for debugging purposes only! */
45 /* define one and only one of these */
46 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
47 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
48 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
49 #define INJECT_FAULTS_4 0 /* undef */
50 #define INJECT_FAULTS_5 0 /* undef */
51 #define INJECT_FAULTS_6 0
53 /* if these are defined, the driver will fail! debug only */
54 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
55 #define INJECT_FATAL_FAULT_2 0 /* undef */
56 #define INJECT_FATAL_FAULT_3 0 /* undef */
59 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
60 #define DPRINTK(x...) do { } while(0)
64 # define PRINTK(x...) printk(KERN_DEBUG x)
70 static inline char * bmname(struct bitmap *bitmap)
72 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
77 * test if the bitmap is active
79 int bitmap_active(struct bitmap *bitmap)
86 spin_lock_irqsave(&bitmap->lock, flags);
87 res = bitmap->flags & BITMAP_ACTIVE;
88 spin_unlock_irqrestore(&bitmap->lock, flags);
92 #define WRITE_POOL_SIZE 256
93 /* mempool for queueing pending writes on the bitmap file */
94 static void *write_pool_alloc(unsigned int gfp_flags, void *data)
96 return kmalloc(sizeof(struct page_list), gfp_flags);
99 static void write_pool_free(void *ptr, void *data)
105 * just a placeholder - calls kmalloc for bitmap pages
107 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
111 #ifdef INJECT_FAULTS_1
114 page = kmalloc(PAGE_SIZE, GFP_NOIO);
117 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
119 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
120 bmname(bitmap), page);
125 * for now just a placeholder -- just calls kfree for bitmap pages
127 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
129 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
134 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
136 * 1) check to see if this page is allocated, if it's not then try to alloc
137 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
138 * page pointer directly as a counter
140 * if we find our page, we increment the page's refcount so that it stays
141 * allocated while we're using it
143 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
145 unsigned char *mappage;
147 if (page >= bitmap->pages) {
149 "%s: invalid bitmap page request: %lu (> %lu)\n",
150 bmname(bitmap), page, bitmap->pages-1);
155 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
158 if (bitmap->bp[page].map) /* page is already allocated, just return */
164 spin_unlock_irq(&bitmap->lock);
166 /* this page has not been allocated yet */
168 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
169 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
171 /* failed - set the hijacked flag so that we can use the
172 * pointer as a counter */
173 spin_lock_irq(&bitmap->lock);
174 if (!bitmap->bp[page].map)
175 bitmap->bp[page].hijacked = 1;
181 spin_lock_irq(&bitmap->lock);
183 /* recheck the page */
185 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
186 /* somebody beat us to getting the page */
187 bitmap_free_page(bitmap, mappage);
191 /* no page was in place and we have one, so install it */
193 memset(mappage, 0, PAGE_SIZE);
194 bitmap->bp[page].map = mappage;
195 bitmap->missing_pages--;
201 /* if page is completely empty, put it back on the free list, or dealloc it */
202 /* if page was hijacked, unmark the flag so it might get alloced next time */
203 /* Note: lock should be held when calling this */
204 static inline void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
208 if (bitmap->bp[page].count) /* page is still busy */
211 /* page is no longer in use, it can be released */
213 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
214 bitmap->bp[page].hijacked = 0;
215 bitmap->bp[page].map = NULL;
219 /* normal case, free the page */
222 /* actually ... let's not. We will probably need the page again exactly when
223 * memory is tight and we are flusing to disk
227 ptr = bitmap->bp[page].map;
228 bitmap->bp[page].map = NULL;
229 bitmap->missing_pages++;
230 bitmap_free_page(bitmap, ptr);
237 * bitmap file handling - read and write the bitmap file and its superblock
240 /* copy the pathname of a file to a buffer */
241 char *file_path(struct file *file, char *buf, int count)
252 buf = d_path(d, v, buf, count);
254 return IS_ERR(buf) ? NULL : buf;
258 * basic page I/O operations
261 /* IO operations when bitmap is stored near all superblocks */
262 static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
264 /* choose a good rdev and read the page from there */
267 struct list_head *tmp;
268 struct page *page = alloc_page(GFP_KERNEL);
272 return ERR_PTR(-ENOMEM);
274 ITERATE_RDEV(mddev, rdev, tmp)
275 if (rdev->in_sync && !rdev->faulty)
277 return ERR_PTR(-EIO);
280 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
282 } while (!sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ));
288 static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
291 struct list_head *tmp;
293 ITERATE_RDEV(mddev, rdev, tmp)
294 if (rdev->in_sync && !rdev->faulty)
295 md_super_write(mddev, rdev,
296 (rdev->sb_offset<<1) + offset
297 + page->index * (PAGE_SIZE/512),
302 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
307 * write out a page to a file
309 static int write_page(struct bitmap *bitmap, struct page *page, int wait)
313 if (bitmap->file == NULL)
314 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
319 if (TestSetPageLocked(page))
320 return -EAGAIN; /* already locked */
321 if (PageWriteback(page)) {
327 ret = page->mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE);
329 ret = page->mapping->a_ops->commit_write(NULL, page, 0,
336 set_page_dirty(page); /* force it to be written out */
339 /* add to list to be waited for by daemon */
340 struct page_list *item = mempool_alloc(bitmap->write_pool, GFP_NOIO);
342 page_cache_get(page);
343 spin_lock(&bitmap->write_lock);
344 list_add(&item->list, &bitmap->complete_pages);
345 spin_unlock(&bitmap->write_lock);
346 md_wakeup_thread(bitmap->writeback_daemon);
348 return write_one_page(page, wait);
351 /* read a page from a file, pinning it into cache, and return bytes_read */
352 static struct page *read_page(struct file *file, unsigned long index,
353 unsigned long *bytes_read)
355 struct inode *inode = file->f_mapping->host;
356 struct page *page = NULL;
357 loff_t isize = i_size_read(inode);
358 unsigned long end_index = isize >> PAGE_CACHE_SHIFT;
360 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_CACHE_SIZE,
361 (unsigned long long)index << PAGE_CACHE_SHIFT);
363 page = read_cache_page(inode->i_mapping, index,
364 (filler_t *)inode->i_mapping->a_ops->readpage, file);
367 wait_on_page_locked(page);
368 if (!PageUptodate(page) || PageError(page)) {
369 page_cache_release(page);
370 page = ERR_PTR(-EIO);
374 if (index > end_index) /* we have read beyond EOF */
376 else if (index == end_index) /* possible short read */
377 *bytes_read = isize & ~PAGE_CACHE_MASK;
379 *bytes_read = PAGE_CACHE_SIZE; /* got a full page */
382 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
383 (int)PAGE_CACHE_SIZE,
384 (unsigned long long)index << PAGE_CACHE_SHIFT,
390 * bitmap file superblock operations
393 /* update the event counter and sync the superblock to disk */
394 int bitmap_update_sb(struct bitmap *bitmap)
399 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
401 spin_lock_irqsave(&bitmap->lock, flags);
402 if (!bitmap->sb_page) { /* no superblock */
403 spin_unlock_irqrestore(&bitmap->lock, flags);
406 spin_unlock_irqrestore(&bitmap->lock, flags);
407 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
408 sb->events = cpu_to_le64(bitmap->mddev->events);
409 if (!bitmap->mddev->degraded)
410 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
411 kunmap(bitmap->sb_page);
412 return write_page(bitmap, bitmap->sb_page, 1);
415 /* print out the bitmap file superblock */
416 void bitmap_print_sb(struct bitmap *bitmap)
420 if (!bitmap || !bitmap->sb_page)
422 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
423 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
424 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
425 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
426 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
427 *(__u32 *)(sb->uuid+0),
428 *(__u32 *)(sb->uuid+4),
429 *(__u32 *)(sb->uuid+8),
430 *(__u32 *)(sb->uuid+12));
431 printk(KERN_DEBUG " events: %llu\n",
432 (unsigned long long) le64_to_cpu(sb->events));
433 printk(KERN_DEBUG "events cleared: %llu\n",
434 (unsigned long long) le64_to_cpu(sb->events_cleared));
435 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
436 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
437 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
438 printk(KERN_DEBUG " sync size: %llu KB\n",
439 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
440 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
441 kunmap(bitmap->sb_page);
444 /* read the superblock from the bitmap file and initialize some bitmap fields */
445 static int bitmap_read_sb(struct bitmap *bitmap)
449 unsigned long chunksize, daemon_sleep, write_behind;
450 unsigned long bytes_read;
451 unsigned long long events;
454 /* page 0 is the superblock, read it... */
456 bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read);
458 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
459 bytes_read = PAGE_SIZE;
461 if (IS_ERR(bitmap->sb_page)) {
462 err = PTR_ERR(bitmap->sb_page);
463 bitmap->sb_page = NULL;
467 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
469 if (bytes_read < sizeof(*sb)) { /* short read */
470 printk(KERN_INFO "%s: bitmap file superblock truncated\n",
476 chunksize = le32_to_cpu(sb->chunksize);
477 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
478 write_behind = le32_to_cpu(sb->write_behind);
480 /* verify that the bitmap-specific fields are valid */
481 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
482 reason = "bad magic";
483 else if (sb->version != cpu_to_le32(BITMAP_MAJOR))
484 reason = "unrecognized superblock version";
485 else if (chunksize < 512 || chunksize > (1024 * 1024 * 4))
486 reason = "bitmap chunksize out of range (512B - 4MB)";
487 else if ((1 << ffz(~chunksize)) != chunksize)
488 reason = "bitmap chunksize not a power of 2";
489 else if (daemon_sleep < 1 || daemon_sleep > 15)
490 reason = "daemon sleep period out of range (1-15s)";
491 else if (write_behind > COUNTER_MAX)
492 reason = "write-behind limit out of range (0 - 16383)";
494 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
495 bmname(bitmap), reason);
499 /* keep the array size field of the bitmap superblock up to date */
500 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
502 if (!bitmap->mddev->persistent)
506 * if we have a persistent array superblock, compare the
507 * bitmap's UUID and event counter to the mddev's
509 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
510 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
514 events = le64_to_cpu(sb->events);
515 if (events < bitmap->mddev->events) {
516 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
517 "-- forcing full recovery\n", bmname(bitmap), events,
518 (unsigned long long) bitmap->mddev->events);
519 sb->state |= BITMAP_STALE;
522 /* assign fields using values from superblock */
523 bitmap->chunksize = chunksize;
524 bitmap->daemon_sleep = daemon_sleep;
525 bitmap->max_write_behind = write_behind;
526 bitmap->flags |= sb->state;
527 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
528 if (sb->state & BITMAP_STALE)
529 bitmap->events_cleared = bitmap->mddev->events;
532 kunmap(bitmap->sb_page);
534 bitmap_print_sb(bitmap);
538 enum bitmap_mask_op {
543 /* record the state of the bitmap in the superblock */
544 static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
545 enum bitmap_mask_op op)
550 spin_lock_irqsave(&bitmap->lock, flags);
551 if (!bitmap || !bitmap->sb_page) { /* can't set the state */
552 spin_unlock_irqrestore(&bitmap->lock, flags);
555 page_cache_get(bitmap->sb_page);
556 spin_unlock_irqrestore(&bitmap->lock, flags);
557 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
559 case MASK_SET: sb->state |= bits;
561 case MASK_UNSET: sb->state &= ~bits;
565 kunmap(bitmap->sb_page);
566 page_cache_release(bitmap->sb_page);
570 * general bitmap file operations
573 /* calculate the index of the page that contains this bit */
574 static inline unsigned long file_page_index(unsigned long chunk)
576 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
579 /* calculate the (bit) offset of this bit within a page */
580 static inline unsigned long file_page_offset(unsigned long chunk)
582 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
586 * return a pointer to the page in the filemap that contains the given bit
588 * this lookup is complicated by the fact that the bitmap sb might be exactly
589 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
592 static inline struct page *filemap_get_page(struct bitmap *bitmap,
595 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
599 static void bitmap_file_unmap(struct bitmap *bitmap)
601 struct page **map, *sb_page;
606 spin_lock_irqsave(&bitmap->lock, flags);
607 map = bitmap->filemap;
608 bitmap->filemap = NULL;
609 attr = bitmap->filemap_attr;
610 bitmap->filemap_attr = NULL;
611 pages = bitmap->file_pages;
612 bitmap->file_pages = 0;
613 sb_page = bitmap->sb_page;
614 bitmap->sb_page = NULL;
615 spin_unlock_irqrestore(&bitmap->lock, flags);
618 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
619 page_cache_release(map[pages]);
624 page_cache_release(sb_page);
627 static void bitmap_stop_daemons(struct bitmap *bitmap);
629 /* dequeue the next item in a page list -- don't call from irq context */
630 static struct page_list *dequeue_page(struct bitmap *bitmap)
632 struct page_list *item = NULL;
633 struct list_head *head = &bitmap->complete_pages;
635 spin_lock(&bitmap->write_lock);
636 if (list_empty(head))
638 item = list_entry(head->prev, struct page_list, list);
639 list_del(head->prev);
641 spin_unlock(&bitmap->write_lock);
645 static void drain_write_queues(struct bitmap *bitmap)
647 struct page_list *item;
649 while ((item = dequeue_page(bitmap))) {
650 /* don't bother to wait */
651 page_cache_release(item->page);
652 mempool_free(item, bitmap->write_pool);
655 wake_up(&bitmap->write_wait);
658 static void bitmap_file_put(struct bitmap *bitmap)
664 spin_lock_irqsave(&bitmap->lock, flags);
667 spin_unlock_irqrestore(&bitmap->lock, flags);
669 bitmap_stop_daemons(bitmap);
671 drain_write_queues(bitmap);
673 bitmap_file_unmap(bitmap);
676 inode = file->f_mapping->host;
677 spin_lock(&inode->i_lock);
678 atomic_set(&inode->i_writecount, 1); /* allow writes again */
679 spin_unlock(&inode->i_lock);
686 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
687 * then it is no longer reliable, so we stop using it and we mark the file
688 * as failed in the superblock
690 static void bitmap_file_kick(struct bitmap *bitmap)
692 char *path, *ptr = NULL;
694 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
695 bitmap_update_sb(bitmap);
698 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
700 ptr = file_path(bitmap->file, path, PAGE_SIZE);
702 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
703 bmname(bitmap), ptr ? ptr : "");
708 bitmap_file_put(bitmap);
713 enum bitmap_page_attr {
714 BITMAP_PAGE_DIRTY = 1, // there are set bits that need to be synced
715 BITMAP_PAGE_CLEAN = 2, // there are bits that might need to be cleared
716 BITMAP_PAGE_NEEDWRITE=4, // there are cleared bits that need to be synced
719 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
720 enum bitmap_page_attr attr)
722 bitmap->filemap_attr[page->index] |= attr;
725 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
726 enum bitmap_page_attr attr)
728 bitmap->filemap_attr[page->index] &= ~attr;
731 static inline unsigned long get_page_attr(struct bitmap *bitmap, struct page *page)
733 return bitmap->filemap_attr[page->index];
737 * bitmap_file_set_bit -- called before performing a write to the md device
738 * to set (and eventually sync) a particular bit in the bitmap file
740 * we set the bit immediately, then we record the page number so that
741 * when an unplug occurs, we can flush the dirty pages out to disk
743 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
748 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
750 if (!bitmap->filemap) {
754 page = filemap_get_page(bitmap, chunk);
755 bit = file_page_offset(chunk);
758 /* make sure the page stays cached until it gets written out */
759 if (! (get_page_attr(bitmap, page) & BITMAP_PAGE_DIRTY))
760 page_cache_get(page);
763 kaddr = kmap_atomic(page, KM_USER0);
765 kunmap_atomic(kaddr, KM_USER0);
766 PRINTK("set file bit %lu page %lu\n", bit, page->index);
768 /* record page number so it gets flushed to disk when unplug occurs */
769 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
773 /* this gets called when the md device is ready to unplug its underlying
774 * (slave) device queues -- before we let any writes go down, we need to
775 * sync the dirty pages of the bitmap file to disk */
776 int bitmap_unplug(struct bitmap *bitmap)
778 unsigned long i, attr, flags;
786 /* look at each page to see if there are any set bits that need to be
787 * flushed out to disk */
788 for (i = 0; i < bitmap->file_pages; i++) {
789 spin_lock_irqsave(&bitmap->lock, flags);
790 if (!bitmap->filemap) {
791 spin_unlock_irqrestore(&bitmap->lock, flags);
794 page = bitmap->filemap[i];
795 attr = get_page_attr(bitmap, page);
796 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
797 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
798 if ((attr & BITMAP_PAGE_DIRTY))
800 spin_unlock_irqrestore(&bitmap->lock, flags);
802 if (attr & (BITMAP_PAGE_DIRTY | BITMAP_PAGE_NEEDWRITE)) {
803 err = write_page(bitmap, page, 0);
804 if (err == -EAGAIN) {
805 if (attr & BITMAP_PAGE_DIRTY)
806 err = write_page(bitmap, page, 1);
814 if (wait) { /* if any writes were performed, we need to wait on them */
816 spin_lock_irq(&bitmap->write_lock);
817 wait_event_lock_irq(bitmap->write_wait,
818 list_empty(&bitmap->complete_pages), bitmap->write_lock,
819 wake_up_process(bitmap->writeback_daemon->tsk));
820 spin_unlock_irq(&bitmap->write_lock);
822 wait_event(bitmap->mddev->sb_wait,
823 atomic_read(&bitmap->mddev->pending_writes)==0);
828 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
829 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
830 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
831 * memory mapping of the bitmap file
833 * if there's no bitmap file, or if the bitmap file had been
834 * previously kicked from the array, we mark all the bits as
835 * 1's in order to cause a full resync.
837 * We ignore all bits for sectors that end earlier than 'start'.
838 * This is used when reading an out-of-date bitmap...
840 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
842 unsigned long i, chunks, index, oldindex, bit;
843 struct page *page = NULL, *oldpage = NULL;
844 unsigned long num_pages, bit_cnt = 0;
846 unsigned long bytes, offset, dummy;
850 chunks = bitmap->chunks;
853 BUG_ON(!file && !bitmap->offset);
855 #ifdef INJECT_FAULTS_3
858 outofdate = bitmap->flags & BITMAP_STALE;
861 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
862 "recovery\n", bmname(bitmap));
864 bytes = (chunks + 7) / 8;
866 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
868 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
869 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
871 (unsigned long) i_size_read(file->f_mapping->host),
872 bytes + sizeof(bitmap_super_t));
878 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
879 if (!bitmap->filemap)
882 bitmap->filemap_attr = kmalloc(sizeof(long) * num_pages, GFP_KERNEL);
883 if (!bitmap->filemap_attr)
886 memset(bitmap->filemap_attr, 0, sizeof(long) * num_pages);
890 for (i = 0; i < chunks; i++) {
891 index = file_page_index(i);
892 bit = file_page_offset(i);
893 if (index != oldindex) { /* this is a new page, read it in */
894 /* unmap the old page, we're done with it */
899 * if we're here then the superblock page
900 * contains some bits (PAGE_SIZE != sizeof sb)
901 * we've already read it in, so just use it
903 page = bitmap->sb_page;
904 offset = sizeof(bitmap_super_t);
906 page = read_page(file, index, &dummy);
909 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
912 if (IS_ERR(page)) { /* read error */
923 * if bitmap is out of date, dirty the
924 * whole page and write it out
926 memset(page_address(page) + offset, 0xff,
928 ret = write_page(bitmap, page, 1);
931 /* release, page not in filemap yet */
932 page_cache_release(page);
937 bitmap->filemap[bitmap->file_pages++] = page;
939 if (test_bit(bit, page_address(page))) {
940 /* if the disk bit is set, set the memory bit */
941 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
942 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
945 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
949 /* everything went OK */
951 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
953 if (page) /* unmap the last page */
956 if (bit_cnt) { /* Kick recovery if any bits were set */
957 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
958 md_wakeup_thread(bitmap->mddev->thread);
962 printk(KERN_INFO "%s: bitmap initialized from disk: "
963 "read %lu/%lu pages, set %lu bits, status: %d\n",
964 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
969 void bitmap_write_all(struct bitmap *bitmap)
971 /* We don't actually write all bitmap blocks here,
972 * just flag them as needing to be written
975 unsigned long chunks = bitmap->chunks;
976 unsigned long bytes = (chunks+7)/8 + sizeof(bitmap_super_t);
977 unsigned long num_pages = (bytes + PAGE_SIZE-1) / PAGE_SIZE;
979 bitmap->filemap_attr[num_pages] |= BITMAP_PAGE_NEEDWRITE;
983 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
985 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
986 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
987 bitmap->bp[page].count += inc;
989 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
990 (unsigned long long)offset, inc, bitmap->bp[page].count);
992 bitmap_checkfree(bitmap, page);
994 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
995 sector_t offset, int *blocks,
999 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1003 int bitmap_daemon_work(struct bitmap *bitmap)
1006 unsigned long flags;
1007 struct page *page = NULL, *lastpage = NULL;
1014 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1016 bitmap->daemon_lastrun = jiffies;
1018 for (j = 0; j < bitmap->chunks; j++) {
1019 bitmap_counter_t *bmc;
1020 spin_lock_irqsave(&bitmap->lock, flags);
1021 if (!bitmap->filemap) {
1022 /* error or shutdown */
1023 spin_unlock_irqrestore(&bitmap->lock, flags);
1027 page = filemap_get_page(bitmap, j);
1029 if (page != lastpage) {
1030 /* skip this page unless it's marked as needing cleaning */
1031 if (!((attr=get_page_attr(bitmap, page)) & BITMAP_PAGE_CLEAN)) {
1032 if (attr & BITMAP_PAGE_NEEDWRITE) {
1033 page_cache_get(page);
1034 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1036 spin_unlock_irqrestore(&bitmap->lock, flags);
1037 if (attr & BITMAP_PAGE_NEEDWRITE) {
1038 switch (write_page(bitmap, page, 0)) {
1040 set_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1045 bitmap_file_kick(bitmap);
1047 page_cache_release(page);
1052 /* grab the new page, sync and release the old */
1053 page_cache_get(page);
1054 if (lastpage != NULL) {
1055 if (get_page_attr(bitmap, lastpage) & BITMAP_PAGE_NEEDWRITE) {
1056 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1057 spin_unlock_irqrestore(&bitmap->lock, flags);
1058 err = write_page(bitmap, lastpage, 0);
1059 if (err == -EAGAIN) {
1061 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1064 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1065 spin_unlock_irqrestore(&bitmap->lock, flags);
1068 page_cache_release(lastpage);
1070 bitmap_file_kick(bitmap);
1072 spin_unlock_irqrestore(&bitmap->lock, flags);
1076 printk("bitmap clean at page %lu\n", j);
1078 spin_lock_irqsave(&bitmap->lock, flags);
1079 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1081 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1085 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1088 *bmc=1; /* maybe clear the bit next time */
1089 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1090 } else if (*bmc == 1) {
1091 /* we can clear the bit */
1093 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1097 clear_bit(file_page_offset(j), page_address(page));
1100 spin_unlock_irqrestore(&bitmap->lock, flags);
1103 /* now sync the final page */
1104 if (lastpage != NULL) {
1106 spin_lock_irqsave(&bitmap->lock, flags);
1107 if (get_page_attr(bitmap, lastpage) &BITMAP_PAGE_NEEDWRITE) {
1108 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1109 spin_unlock_irqrestore(&bitmap->lock, flags);
1110 err = write_page(bitmap, lastpage, 0);
1111 if (err == -EAGAIN) {
1112 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1116 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1117 spin_unlock_irqrestore(&bitmap->lock, flags);
1120 page_cache_release(lastpage);
1126 static void daemon_exit(struct bitmap *bitmap, mdk_thread_t **daemon)
1129 unsigned long flags;
1131 /* if no one is waiting on us, we'll free the md thread struct
1132 * and exit, otherwise we let the waiter clean things up */
1133 spin_lock_irqsave(&bitmap->lock, flags);
1134 if ((dmn = *daemon)) { /* no one is waiting, cleanup and exit */
1136 spin_unlock_irqrestore(&bitmap->lock, flags);
1138 complete_and_exit(NULL, 0); /* do_exit not exported */
1140 spin_unlock_irqrestore(&bitmap->lock, flags);
1143 static void bitmap_writeback_daemon(mddev_t *mddev)
1145 struct bitmap *bitmap = mddev->bitmap;
1147 struct page_list *item;
1150 if (signal_pending(current)) {
1152 "%s: bitmap writeback daemon got signal, exiting...\n",
1158 PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap));
1159 /* wait on bitmap page writebacks */
1160 while ((item = dequeue_page(bitmap))) {
1162 mempool_free(item, bitmap->write_pool);
1163 PRINTK("wait on page writeback: %p\n", page);
1164 wait_on_page_writeback(page);
1165 PRINTK("finished page writeback: %p\n", page);
1167 err = PageError(page);
1168 page_cache_release(page);
1170 printk(KERN_WARNING "%s: bitmap file writeback "
1171 "failed (page %lu): %d\n",
1172 bmname(bitmap), page->index, err);
1173 bitmap_file_kick(bitmap);
1178 wake_up(&bitmap->write_wait);
1180 printk(KERN_INFO "%s: bitmap writeback daemon exiting (%d)\n",
1181 bmname(bitmap), err);
1182 daemon_exit(bitmap, &bitmap->writeback_daemon);
1186 static int bitmap_start_daemon(struct bitmap *bitmap, mdk_thread_t **ptr,
1187 void (*func)(mddev_t *), char *name)
1189 mdk_thread_t *daemon;
1190 unsigned long flags;
1193 spin_lock_irqsave(&bitmap->lock, flags);
1196 if (!bitmap->file) /* no need for daemon if there's no backing file */
1199 spin_unlock_irqrestore(&bitmap->lock, flags);
1201 #ifdef INJECT_FATAL_FAULT_2
1204 sprintf(namebuf, "%%s_%s", name);
1205 daemon = md_register_thread(func, bitmap->mddev, namebuf);
1208 printk(KERN_ERR "%s: failed to start bitmap daemon\n",
1213 spin_lock_irqsave(&bitmap->lock, flags);
1216 md_wakeup_thread(daemon); /* start it running */
1218 PRINTK("%s: %s daemon (pid %d) started...\n",
1219 bmname(bitmap), name, daemon->tsk->pid);
1221 spin_unlock_irqrestore(&bitmap->lock, flags);
1225 static int bitmap_start_daemons(struct bitmap *bitmap)
1227 int err = bitmap_start_daemon(bitmap, &bitmap->writeback_daemon,
1228 bitmap_writeback_daemon, "bitmap_wb");
1232 static void bitmap_stop_daemon(struct bitmap *bitmap, mdk_thread_t **ptr)
1234 mdk_thread_t *daemon;
1235 unsigned long flags;
1237 spin_lock_irqsave(&bitmap->lock, flags);
1240 spin_unlock_irqrestore(&bitmap->lock, flags);
1242 md_unregister_thread(daemon); /* destroy the thread */
1245 static void bitmap_stop_daemons(struct bitmap *bitmap)
1247 /* the daemons can't stop themselves... they'll just exit instead... */
1248 if (bitmap->writeback_daemon &&
1249 current->pid != bitmap->writeback_daemon->tsk->pid)
1250 bitmap_stop_daemon(bitmap, &bitmap->writeback_daemon);
1253 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1254 sector_t offset, int *blocks,
1257 /* If 'create', we might release the lock and reclaim it.
1258 * The lock must have been taken with interrupts enabled.
1259 * If !create, we don't release the lock.
1261 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1262 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1263 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1266 if (bitmap_checkpage(bitmap, page, create) < 0) {
1267 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1268 *blocks = csize - (offset & (csize- 1));
1271 /* now locked ... */
1273 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1274 /* should we use the first or second counter field
1275 * of the hijacked pointer? */
1276 int hi = (pageoff > PAGE_COUNTER_MASK);
1277 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1278 PAGE_COUNTER_SHIFT - 1);
1279 *blocks = csize - (offset & (csize- 1));
1280 return &((bitmap_counter_t *)
1281 &bitmap->bp[page].map)[hi];
1282 } else { /* page is allocated */
1283 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1284 *blocks = csize - (offset & (csize- 1));
1285 return (bitmap_counter_t *)
1286 &(bitmap->bp[page].map[pageoff]);
1290 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1292 if (!bitmap) return 0;
1295 atomic_inc(&bitmap->behind_writes);
1296 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1297 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1302 bitmap_counter_t *bmc;
1304 spin_lock_irq(&bitmap->lock);
1305 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1307 spin_unlock_irq(&bitmap->lock);
1313 bitmap_file_set_bit(bitmap, offset);
1314 bitmap_count_page(bitmap,offset, 1);
1315 blk_plug_device(bitmap->mddev->queue);
1320 if ((*bmc & COUNTER_MAX) == COUNTER_MAX) BUG();
1323 spin_unlock_irq(&bitmap->lock);
1326 if (sectors > blocks)
1333 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1334 int success, int behind)
1336 if (!bitmap) return;
1338 atomic_dec(&bitmap->behind_writes);
1339 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1340 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1345 unsigned long flags;
1346 bitmap_counter_t *bmc;
1348 spin_lock_irqsave(&bitmap->lock, flags);
1349 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1351 spin_unlock_irqrestore(&bitmap->lock, flags);
1355 if (!success && ! (*bmc & NEEDED_MASK))
1356 *bmc |= NEEDED_MASK;
1360 set_page_attr(bitmap,
1361 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1364 spin_unlock_irqrestore(&bitmap->lock, flags);
1366 if (sectors > blocks)
1372 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1375 bitmap_counter_t *bmc;
1377 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1379 return 1; /* always resync if no bitmap */
1381 spin_lock_irq(&bitmap->lock);
1382 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1388 else if (NEEDED(*bmc)) {
1390 if (!degraded) { /* don't set/clear bits if degraded */
1391 *bmc |= RESYNC_MASK;
1392 *bmc &= ~NEEDED_MASK;
1396 spin_unlock_irq(&bitmap->lock);
1400 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1402 bitmap_counter_t *bmc;
1403 unsigned long flags;
1405 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1406 */ if (bitmap == NULL) {
1410 spin_lock_irqsave(&bitmap->lock, flags);
1411 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1416 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1419 *bmc &= ~RESYNC_MASK;
1421 if (!NEEDED(*bmc) && aborted)
1422 *bmc |= NEEDED_MASK;
1425 set_page_attr(bitmap,
1426 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1432 spin_unlock_irqrestore(&bitmap->lock, flags);
1435 void bitmap_close_sync(struct bitmap *bitmap)
1437 /* Sync has finished, and any bitmap chunks that weren't synced
1438 * properly have been aborted. It remains to us to clear the
1439 * RESYNC bit wherever it is still on
1441 sector_t sector = 0;
1443 if (!bitmap) return;
1444 while (sector < bitmap->mddev->resync_max_sectors) {
1445 bitmap_end_sync(bitmap, sector, &blocks, 0);
1447 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1448 (unsigned long long)sector, blocks);
1449 */ sector += blocks;
1453 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1455 /* For each chunk covered by any of these sectors, set the
1456 * counter to 1 and set resync_needed. They should all
1457 * be 0 at this point
1461 bitmap_counter_t *bmc;
1462 spin_lock_irq(&bitmap->lock);
1463 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1465 spin_unlock_irq(&bitmap->lock);
1470 *bmc = 1 | (needed?NEEDED_MASK:0);
1471 bitmap_count_page(bitmap, offset, 1);
1472 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1473 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1475 spin_unlock_irq(&bitmap->lock);
1480 * flush out any pending updates
1482 void bitmap_flush(mddev_t *mddev)
1484 struct bitmap *bitmap = mddev->bitmap;
1487 if (!bitmap) /* there was no bitmap */
1490 /* run the daemon_work three time to ensure everything is flushed
1493 sleep = bitmap->daemon_sleep;
1494 bitmap->daemon_sleep = 0;
1495 bitmap_daemon_work(bitmap);
1496 bitmap_daemon_work(bitmap);
1497 bitmap_daemon_work(bitmap);
1498 bitmap->daemon_sleep = sleep;
1499 bitmap_update_sb(bitmap);
1503 * free memory that was allocated
1505 void bitmap_destroy(mddev_t *mddev)
1507 unsigned long k, pages;
1508 struct bitmap_page *bp;
1509 struct bitmap *bitmap = mddev->bitmap;
1511 if (!bitmap) /* there was no bitmap */
1514 mddev->bitmap = NULL; /* disconnect from the md device */
1516 /* release the bitmap file and kill the daemon */
1517 bitmap_file_put(bitmap);
1520 pages = bitmap->pages;
1522 /* free all allocated memory */
1524 mempool_destroy(bitmap->write_pool);
1526 if (bp) /* deallocate the page memory */
1527 for (k = 0; k < pages; k++)
1528 if (bp[k].map && !bp[k].hijacked)
1535 * initialize the bitmap structure
1536 * if this returns an error, bitmap_destroy must be called to do clean up
1538 int bitmap_create(mddev_t *mddev)
1540 struct bitmap *bitmap;
1541 unsigned long blocks = mddev->resync_max_sectors;
1542 unsigned long chunks;
1543 unsigned long pages;
1544 struct file *file = mddev->bitmap_file;
1548 BUG_ON(sizeof(bitmap_super_t) != 256);
1550 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1553 BUG_ON(file && mddev->bitmap_offset);
1555 bitmap = kmalloc(sizeof(*bitmap), GFP_KERNEL);
1559 memset(bitmap, 0, sizeof(*bitmap));
1561 spin_lock_init(&bitmap->lock);
1562 bitmap->mddev = mddev;
1563 mddev->bitmap = bitmap;
1565 spin_lock_init(&bitmap->write_lock);
1566 INIT_LIST_HEAD(&bitmap->complete_pages);
1567 init_waitqueue_head(&bitmap->write_wait);
1568 bitmap->write_pool = mempool_create(WRITE_POOL_SIZE, write_pool_alloc,
1569 write_pool_free, NULL);
1570 if (!bitmap->write_pool)
1573 bitmap->file = file;
1574 bitmap->offset = mddev->bitmap_offset;
1575 if (file) get_file(file);
1576 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1577 err = bitmap_read_sb(bitmap);
1581 bitmap->chunkshift = find_first_bit(&bitmap->chunksize,
1582 sizeof(bitmap->chunksize));
1584 /* now that chunksize and chunkshift are set, we can use these macros */
1585 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1586 CHUNK_BLOCK_RATIO(bitmap);
1587 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1591 bitmap->chunks = chunks;
1592 bitmap->pages = pages;
1593 bitmap->missing_pages = pages;
1594 bitmap->counter_bits = COUNTER_BITS;
1596 bitmap->syncchunk = ~0UL;
1598 #ifdef INJECT_FATAL_FAULT_1
1601 bitmap->bp = kmalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1605 memset(bitmap->bp, 0, pages * sizeof(*bitmap->bp));
1607 bitmap->flags |= BITMAP_ACTIVE;
1609 /* now that we have some pages available, initialize the in-memory
1610 * bitmap from the on-disk bitmap */
1612 if (mddev->degraded == 0
1613 || bitmap->events_cleared == mddev->events)
1614 /* no need to keep dirty bits to optimise a re-add of a missing device */
1615 start = mddev->recovery_cp;
1616 err = bitmap_init_from_disk(bitmap, start);
1621 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1622 pages, bmname(bitmap));
1624 /* kick off the bitmap daemons */
1625 err = bitmap_start_daemons(bitmap);
1628 return bitmap_update_sb(bitmap);
1631 /* the bitmap API -- for raid personalities */
1632 EXPORT_SYMBOL(bitmap_startwrite);
1633 EXPORT_SYMBOL(bitmap_endwrite);
1634 EXPORT_SYMBOL(bitmap_start_sync);
1635 EXPORT_SYMBOL(bitmap_end_sync);
1636 EXPORT_SYMBOL(bitmap_unplug);
1637 EXPORT_SYMBOL(bitmap_close_sync);
1638 EXPORT_SYMBOL(bitmap_daemon_work);