jbd2: enable journal clients to enable v2 checksumming
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / bitmap.c
blob97e73e555d112c8762b7934bd9892d39f5c6e4ce
1 /*
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
13 * Still to do:
15 * flush after percent set rather than just time based. (maybe both).
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
30 #include "md.h"
31 #include "bitmap.h"
33 static inline char *bmname(struct bitmap *bitmap)
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
48 static int bitmap_checkpage(struct bitmap *bitmap,
49 unsigned long page, int create)
50 __releases(bitmap->lock)
51 __acquires(bitmap->lock)
53 unsigned char *mappage;
55 if (page >= bitmap->pages) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
58 * It is harmless.
60 return -EINVAL;
63 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
64 return 0;
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
67 return 0;
69 if (!create)
70 return -ENOENT;
72 /* this page has not been allocated yet */
74 spin_unlock_irq(&bitmap->lock);
75 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
76 spin_lock_irq(&bitmap->lock);
78 if (mappage == NULL) {
79 pr_debug("%s: bitmap map page allocation failed, hijacking\n",
80 bmname(bitmap));
81 /* failed - set the hijacked flag so that we can use the
82 * pointer as a counter */
83 if (!bitmap->bp[page].map)
84 bitmap->bp[page].hijacked = 1;
85 } else if (bitmap->bp[page].map ||
86 bitmap->bp[page].hijacked) {
87 /* somebody beat us to getting the page */
88 kfree(mappage);
89 return 0;
90 } else {
92 /* no page was in place and we have one, so install it */
94 bitmap->bp[page].map = mappage;
95 bitmap->missing_pages--;
97 return 0;
100 /* if page is completely empty, put it back on the free list, or dealloc it */
101 /* if page was hijacked, unmark the flag so it might get alloced next time */
102 /* Note: lock should be held when calling this */
103 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
105 char *ptr;
107 if (bitmap->bp[page].count) /* page is still busy */
108 return;
110 /* page is no longer in use, it can be released */
112 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
113 bitmap->bp[page].hijacked = 0;
114 bitmap->bp[page].map = NULL;
115 } else {
116 /* normal case, free the page */
117 ptr = bitmap->bp[page].map;
118 bitmap->bp[page].map = NULL;
119 bitmap->missing_pages++;
120 kfree(ptr);
125 * bitmap file handling - read and write the bitmap file and its superblock
129 * basic page I/O operations
132 /* IO operations when bitmap is stored near all superblocks */
133 static struct page *read_sb_page(struct mddev *mddev, loff_t offset,
134 struct page *page,
135 unsigned long index, int size)
137 /* choose a good rdev and read the page from there */
139 struct md_rdev *rdev;
140 sector_t target;
141 int did_alloc = 0;
143 if (!page) {
144 page = alloc_page(GFP_KERNEL);
145 if (!page)
146 return ERR_PTR(-ENOMEM);
147 did_alloc = 1;
150 rdev_for_each(rdev, mddev) {
151 if (! test_bit(In_sync, &rdev->flags)
152 || test_bit(Faulty, &rdev->flags))
153 continue;
155 target = offset + index * (PAGE_SIZE/512);
157 if (sync_page_io(rdev, target,
158 roundup(size, bdev_logical_block_size(rdev->bdev)),
159 page, READ, true)) {
160 page->index = index;
161 attach_page_buffers(page, NULL); /* so that free_buffer will
162 * quietly no-op */
163 return page;
166 if (did_alloc)
167 put_page(page);
168 return ERR_PTR(-EIO);
172 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
174 /* Iterate the disks of an mddev, using rcu to protect access to the
175 * linked list, and raising the refcount of devices we return to ensure
176 * they don't disappear while in use.
177 * As devices are only added or removed when raid_disk is < 0 and
178 * nr_pending is 0 and In_sync is clear, the entries we return will
179 * still be in the same position on the list when we re-enter
180 * list_for_each_continue_rcu.
182 struct list_head *pos;
183 rcu_read_lock();
184 if (rdev == NULL)
185 /* start at the beginning */
186 pos = &mddev->disks;
187 else {
188 /* release the previous rdev and start from there. */
189 rdev_dec_pending(rdev, mddev);
190 pos = &rdev->same_set;
192 list_for_each_continue_rcu(pos, &mddev->disks) {
193 rdev = list_entry(pos, struct md_rdev, same_set);
194 if (rdev->raid_disk >= 0 &&
195 !test_bit(Faulty, &rdev->flags)) {
196 /* this is a usable devices */
197 atomic_inc(&rdev->nr_pending);
198 rcu_read_unlock();
199 return rdev;
202 rcu_read_unlock();
203 return NULL;
206 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
208 struct md_rdev *rdev = NULL;
209 struct block_device *bdev;
210 struct mddev *mddev = bitmap->mddev;
212 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
213 int size = PAGE_SIZE;
214 loff_t offset = mddev->bitmap_info.offset;
216 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
218 if (page->index == bitmap->file_pages-1)
219 size = roundup(bitmap->last_page_size,
220 bdev_logical_block_size(bdev));
221 /* Just make sure we aren't corrupting data or
222 * metadata
224 if (mddev->external) {
225 /* Bitmap could be anywhere. */
226 if (rdev->sb_start + offset + (page->index
227 * (PAGE_SIZE/512))
228 > rdev->data_offset
230 rdev->sb_start + offset
231 < (rdev->data_offset + mddev->dev_sectors
232 + (PAGE_SIZE/512)))
233 goto bad_alignment;
234 } else if (offset < 0) {
235 /* DATA BITMAP METADATA */
236 if (offset
237 + (long)(page->index * (PAGE_SIZE/512))
238 + size/512 > 0)
239 /* bitmap runs in to metadata */
240 goto bad_alignment;
241 if (rdev->data_offset + mddev->dev_sectors
242 > rdev->sb_start + offset)
243 /* data runs in to bitmap */
244 goto bad_alignment;
245 } else if (rdev->sb_start < rdev->data_offset) {
246 /* METADATA BITMAP DATA */
247 if (rdev->sb_start
248 + offset
249 + page->index*(PAGE_SIZE/512) + size/512
250 > rdev->data_offset)
251 /* bitmap runs in to data */
252 goto bad_alignment;
253 } else {
254 /* DATA METADATA BITMAP - no problems */
256 md_super_write(mddev, rdev,
257 rdev->sb_start + offset
258 + page->index * (PAGE_SIZE/512),
259 size,
260 page);
263 if (wait)
264 md_super_wait(mddev);
265 return 0;
267 bad_alignment:
268 return -EINVAL;
271 static void bitmap_file_kick(struct bitmap *bitmap);
273 * write out a page to a file
275 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
277 struct buffer_head *bh;
279 if (bitmap->file == NULL) {
280 switch (write_sb_page(bitmap, page, wait)) {
281 case -EINVAL:
282 bitmap->flags |= BITMAP_WRITE_ERROR;
284 } else {
286 bh = page_buffers(page);
288 while (bh && bh->b_blocknr) {
289 atomic_inc(&bitmap->pending_writes);
290 set_buffer_locked(bh);
291 set_buffer_mapped(bh);
292 submit_bh(WRITE | REQ_SYNC, bh);
293 bh = bh->b_this_page;
296 if (wait)
297 wait_event(bitmap->write_wait,
298 atomic_read(&bitmap->pending_writes)==0);
300 if (bitmap->flags & BITMAP_WRITE_ERROR)
301 bitmap_file_kick(bitmap);
304 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
306 struct bitmap *bitmap = bh->b_private;
307 unsigned long flags;
309 if (!uptodate) {
310 spin_lock_irqsave(&bitmap->lock, flags);
311 bitmap->flags |= BITMAP_WRITE_ERROR;
312 spin_unlock_irqrestore(&bitmap->lock, flags);
314 if (atomic_dec_and_test(&bitmap->pending_writes))
315 wake_up(&bitmap->write_wait);
318 /* copied from buffer.c */
319 static void
320 __clear_page_buffers(struct page *page)
322 ClearPagePrivate(page);
323 set_page_private(page, 0);
324 page_cache_release(page);
326 static void free_buffers(struct page *page)
328 struct buffer_head *bh = page_buffers(page);
330 while (bh) {
331 struct buffer_head *next = bh->b_this_page;
332 free_buffer_head(bh);
333 bh = next;
335 __clear_page_buffers(page);
336 put_page(page);
339 /* read a page from a file.
340 * We both read the page, and attach buffers to the page to record the
341 * address of each block (using bmap). These addresses will be used
342 * to write the block later, completely bypassing the filesystem.
343 * This usage is similar to how swap files are handled, and allows us
344 * to write to a file with no concerns of memory allocation failing.
346 static struct page *read_page(struct file *file, unsigned long index,
347 struct bitmap *bitmap,
348 unsigned long count)
350 struct page *page = NULL;
351 struct inode *inode = file->f_path.dentry->d_inode;
352 struct buffer_head *bh;
353 sector_t block;
355 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
356 (unsigned long long)index << PAGE_SHIFT);
358 page = alloc_page(GFP_KERNEL);
359 if (!page)
360 page = ERR_PTR(-ENOMEM);
361 if (IS_ERR(page))
362 goto out;
364 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
365 if (!bh) {
366 put_page(page);
367 page = ERR_PTR(-ENOMEM);
368 goto out;
370 attach_page_buffers(page, bh);
371 block = index << (PAGE_SHIFT - inode->i_blkbits);
372 while (bh) {
373 if (count == 0)
374 bh->b_blocknr = 0;
375 else {
376 bh->b_blocknr = bmap(inode, block);
377 if (bh->b_blocknr == 0) {
378 /* Cannot use this file! */
379 free_buffers(page);
380 page = ERR_PTR(-EINVAL);
381 goto out;
383 bh->b_bdev = inode->i_sb->s_bdev;
384 if (count < (1<<inode->i_blkbits))
385 count = 0;
386 else
387 count -= (1<<inode->i_blkbits);
389 bh->b_end_io = end_bitmap_write;
390 bh->b_private = bitmap;
391 atomic_inc(&bitmap->pending_writes);
392 set_buffer_locked(bh);
393 set_buffer_mapped(bh);
394 submit_bh(READ, bh);
396 block++;
397 bh = bh->b_this_page;
399 page->index = index;
401 wait_event(bitmap->write_wait,
402 atomic_read(&bitmap->pending_writes)==0);
403 if (bitmap->flags & BITMAP_WRITE_ERROR) {
404 free_buffers(page);
405 page = ERR_PTR(-EIO);
407 out:
408 if (IS_ERR(page))
409 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n",
410 (int)PAGE_SIZE,
411 (unsigned long long)index << PAGE_SHIFT,
412 PTR_ERR(page));
413 return page;
417 * bitmap file superblock operations
420 /* update the event counter and sync the superblock to disk */
421 void bitmap_update_sb(struct bitmap *bitmap)
423 bitmap_super_t *sb;
425 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
426 return;
427 if (bitmap->mddev->bitmap_info.external)
428 return;
429 if (!bitmap->sb_page) /* no superblock */
430 return;
431 sb = kmap_atomic(bitmap->sb_page);
432 sb->events = cpu_to_le64(bitmap->mddev->events);
433 if (bitmap->mddev->events < bitmap->events_cleared)
434 /* rocking back to read-only */
435 bitmap->events_cleared = bitmap->mddev->events;
436 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
437 sb->state = cpu_to_le32(bitmap->flags);
438 /* Just in case these have been changed via sysfs: */
439 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
440 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
441 kunmap_atomic(sb);
442 write_page(bitmap, bitmap->sb_page, 1);
445 /* print out the bitmap file superblock */
446 void bitmap_print_sb(struct bitmap *bitmap)
448 bitmap_super_t *sb;
450 if (!bitmap || !bitmap->sb_page)
451 return;
452 sb = kmap_atomic(bitmap->sb_page);
453 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
454 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
455 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
456 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
457 *(__u32 *)(sb->uuid+0),
458 *(__u32 *)(sb->uuid+4),
459 *(__u32 *)(sb->uuid+8),
460 *(__u32 *)(sb->uuid+12));
461 printk(KERN_DEBUG " events: %llu\n",
462 (unsigned long long) le64_to_cpu(sb->events));
463 printk(KERN_DEBUG "events cleared: %llu\n",
464 (unsigned long long) le64_to_cpu(sb->events_cleared));
465 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
466 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
467 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
468 printk(KERN_DEBUG " sync size: %llu KB\n",
469 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
470 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
471 kunmap_atomic(sb);
475 * bitmap_new_disk_sb
476 * @bitmap
478 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
479 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
480 * This function verifies 'bitmap_info' and populates the on-disk bitmap
481 * structure, which is to be written to disk.
483 * Returns: 0 on success, -Exxx on error
485 static int bitmap_new_disk_sb(struct bitmap *bitmap)
487 bitmap_super_t *sb;
488 unsigned long chunksize, daemon_sleep, write_behind;
489 int err = -EINVAL;
491 bitmap->sb_page = alloc_page(GFP_KERNEL);
492 if (IS_ERR(bitmap->sb_page)) {
493 err = PTR_ERR(bitmap->sb_page);
494 bitmap->sb_page = NULL;
495 return err;
497 bitmap->sb_page->index = 0;
499 sb = kmap_atomic(bitmap->sb_page);
501 sb->magic = cpu_to_le32(BITMAP_MAGIC);
502 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
504 chunksize = bitmap->mddev->bitmap_info.chunksize;
505 BUG_ON(!chunksize);
506 if (!is_power_of_2(chunksize)) {
507 kunmap_atomic(sb);
508 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
509 return -EINVAL;
511 sb->chunksize = cpu_to_le32(chunksize);
513 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
514 if (!daemon_sleep ||
515 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
516 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
517 daemon_sleep = 5 * HZ;
519 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
520 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
523 * FIXME: write_behind for RAID1. If not specified, what
524 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
526 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
527 if (write_behind > COUNTER_MAX)
528 write_behind = COUNTER_MAX / 2;
529 sb->write_behind = cpu_to_le32(write_behind);
530 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
532 /* keep the array size field of the bitmap superblock up to date */
533 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
535 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
537 bitmap->flags |= BITMAP_STALE;
538 sb->state |= cpu_to_le32(BITMAP_STALE);
539 bitmap->events_cleared = bitmap->mddev->events;
540 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
542 kunmap_atomic(sb);
544 return 0;
547 /* read the superblock from the bitmap file and initialize some bitmap fields */
548 static int bitmap_read_sb(struct bitmap *bitmap)
550 char *reason = NULL;
551 bitmap_super_t *sb;
552 unsigned long chunksize, daemon_sleep, write_behind;
553 unsigned long long events;
554 int err = -EINVAL;
556 /* page 0 is the superblock, read it... */
557 if (bitmap->file) {
558 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
559 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
561 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
562 } else {
563 bitmap->sb_page = read_sb_page(bitmap->mddev,
564 bitmap->mddev->bitmap_info.offset,
565 NULL,
566 0, sizeof(bitmap_super_t));
568 if (IS_ERR(bitmap->sb_page)) {
569 err = PTR_ERR(bitmap->sb_page);
570 bitmap->sb_page = NULL;
571 return err;
574 sb = kmap_atomic(bitmap->sb_page);
576 chunksize = le32_to_cpu(sb->chunksize);
577 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
578 write_behind = le32_to_cpu(sb->write_behind);
580 /* verify that the bitmap-specific fields are valid */
581 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
582 reason = "bad magic";
583 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
584 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
585 reason = "unrecognized superblock version";
586 else if (chunksize < 512)
587 reason = "bitmap chunksize too small";
588 else if (!is_power_of_2(chunksize))
589 reason = "bitmap chunksize not a power of 2";
590 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
591 reason = "daemon sleep period out of range";
592 else if (write_behind > COUNTER_MAX)
593 reason = "write-behind limit out of range (0 - 16383)";
594 if (reason) {
595 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
596 bmname(bitmap), reason);
597 goto out;
600 /* keep the array size field of the bitmap superblock up to date */
601 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
603 if (bitmap->mddev->persistent) {
605 * We have a persistent array superblock, so compare the
606 * bitmap's UUID and event counter to the mddev's
608 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
609 printk(KERN_INFO
610 "%s: bitmap superblock UUID mismatch\n",
611 bmname(bitmap));
612 goto out;
614 events = le64_to_cpu(sb->events);
615 if (events < bitmap->mddev->events) {
616 printk(KERN_INFO
617 "%s: bitmap file is out of date (%llu < %llu) "
618 "-- forcing full recovery\n",
619 bmname(bitmap), events,
620 (unsigned long long) bitmap->mddev->events);
621 sb->state |= cpu_to_le32(BITMAP_STALE);
625 /* assign fields using values from superblock */
626 bitmap->mddev->bitmap_info.chunksize = chunksize;
627 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
628 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
629 bitmap->flags |= le32_to_cpu(sb->state);
630 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
631 bitmap->flags |= BITMAP_HOSTENDIAN;
632 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
633 if (bitmap->flags & BITMAP_STALE)
634 bitmap->events_cleared = bitmap->mddev->events;
635 err = 0;
636 out:
637 kunmap_atomic(sb);
638 if (err)
639 bitmap_print_sb(bitmap);
640 return err;
643 enum bitmap_mask_op {
644 MASK_SET,
645 MASK_UNSET
648 /* record the state of the bitmap in the superblock. Return the old value */
649 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
650 enum bitmap_mask_op op)
652 bitmap_super_t *sb;
653 int old;
655 if (!bitmap->sb_page) /* can't set the state */
656 return 0;
657 sb = kmap_atomic(bitmap->sb_page);
658 old = le32_to_cpu(sb->state) & bits;
659 switch (op) {
660 case MASK_SET:
661 sb->state |= cpu_to_le32(bits);
662 bitmap->flags |= bits;
663 break;
664 case MASK_UNSET:
665 sb->state &= cpu_to_le32(~bits);
666 bitmap->flags &= ~bits;
667 break;
668 default:
669 BUG();
671 kunmap_atomic(sb);
672 return old;
676 * general bitmap file operations
680 * on-disk bitmap:
682 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
683 * file a page at a time. There's a superblock at the start of the file.
685 /* calculate the index of the page that contains this bit */
686 static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
688 if (!bitmap->mddev->bitmap_info.external)
689 chunk += sizeof(bitmap_super_t) << 3;
690 return chunk >> PAGE_BIT_SHIFT;
693 /* calculate the (bit) offset of this bit within a page */
694 static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
696 if (!bitmap->mddev->bitmap_info.external)
697 chunk += sizeof(bitmap_super_t) << 3;
698 return chunk & (PAGE_BITS - 1);
702 * return a pointer to the page in the filemap that contains the given bit
704 * this lookup is complicated by the fact that the bitmap sb might be exactly
705 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
706 * 0 or page 1
708 static inline struct page *filemap_get_page(struct bitmap *bitmap,
709 unsigned long chunk)
711 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
712 return NULL;
713 return bitmap->filemap[file_page_index(bitmap, chunk)
714 - file_page_index(bitmap, 0)];
717 static void bitmap_file_unmap(struct bitmap *bitmap)
719 struct page **map, *sb_page;
720 unsigned long *attr;
721 int pages;
722 unsigned long flags;
724 spin_lock_irqsave(&bitmap->lock, flags);
725 map = bitmap->filemap;
726 bitmap->filemap = NULL;
727 attr = bitmap->filemap_attr;
728 bitmap->filemap_attr = NULL;
729 pages = bitmap->file_pages;
730 bitmap->file_pages = 0;
731 sb_page = bitmap->sb_page;
732 bitmap->sb_page = NULL;
733 spin_unlock_irqrestore(&bitmap->lock, flags);
735 while (pages--)
736 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
737 free_buffers(map[pages]);
738 kfree(map);
739 kfree(attr);
741 if (sb_page)
742 free_buffers(sb_page);
745 static void bitmap_file_put(struct bitmap *bitmap)
747 struct file *file;
748 unsigned long flags;
750 spin_lock_irqsave(&bitmap->lock, flags);
751 file = bitmap->file;
752 bitmap->file = NULL;
753 spin_unlock_irqrestore(&bitmap->lock, flags);
755 if (file)
756 wait_event(bitmap->write_wait,
757 atomic_read(&bitmap->pending_writes)==0);
758 bitmap_file_unmap(bitmap);
760 if (file) {
761 struct inode *inode = file->f_path.dentry->d_inode;
762 invalidate_mapping_pages(inode->i_mapping, 0, -1);
763 fput(file);
768 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
769 * then it is no longer reliable, so we stop using it and we mark the file
770 * as failed in the superblock
772 static void bitmap_file_kick(struct bitmap *bitmap)
774 char *path, *ptr = NULL;
776 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
777 bitmap_update_sb(bitmap);
779 if (bitmap->file) {
780 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
781 if (path)
782 ptr = d_path(&bitmap->file->f_path, path,
783 PAGE_SIZE);
785 printk(KERN_ALERT
786 "%s: kicking failed bitmap file %s from array!\n",
787 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
789 kfree(path);
790 } else
791 printk(KERN_ALERT
792 "%s: disabling internal bitmap due to errors\n",
793 bmname(bitmap));
796 bitmap_file_put(bitmap);
798 return;
801 enum bitmap_page_attr {
802 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
803 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
804 * i.e. counter is 1 or 2. */
805 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
808 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
809 enum bitmap_page_attr attr)
811 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
814 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
815 enum bitmap_page_attr attr)
817 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
820 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
821 enum bitmap_page_attr attr)
823 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
827 * bitmap_file_set_bit -- called before performing a write to the md device
828 * to set (and eventually sync) a particular bit in the bitmap file
830 * we set the bit immediately, then we record the page number so that
831 * when an unplug occurs, we can flush the dirty pages out to disk
833 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
835 unsigned long bit;
836 struct page *page;
837 void *kaddr;
838 unsigned long chunk = block >> bitmap->chunkshift;
840 if (!bitmap->filemap)
841 return;
843 page = filemap_get_page(bitmap, chunk);
844 if (!page)
845 return;
846 bit = file_page_offset(bitmap, chunk);
848 /* set the bit */
849 kaddr = kmap_atomic(page);
850 if (bitmap->flags & BITMAP_HOSTENDIAN)
851 set_bit(bit, kaddr);
852 else
853 __set_bit_le(bit, kaddr);
854 kunmap_atomic(kaddr);
855 pr_debug("set file bit %lu page %lu\n", bit, page->index);
856 /* record page number so it gets flushed to disk when unplug occurs */
857 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
860 /* this gets called when the md device is ready to unplug its underlying
861 * (slave) device queues -- before we let any writes go down, we need to
862 * sync the dirty pages of the bitmap file to disk */
863 void bitmap_unplug(struct bitmap *bitmap)
865 unsigned long i, flags;
866 int dirty, need_write;
867 struct page *page;
868 int wait = 0;
870 if (!bitmap)
871 return;
873 /* look at each page to see if there are any set bits that need to be
874 * flushed out to disk */
875 for (i = 0; i < bitmap->file_pages; i++) {
876 spin_lock_irqsave(&bitmap->lock, flags);
877 if (!bitmap->filemap) {
878 spin_unlock_irqrestore(&bitmap->lock, flags);
879 return;
881 page = bitmap->filemap[i];
882 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
883 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
884 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
885 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
886 if (dirty)
887 wait = 1;
888 spin_unlock_irqrestore(&bitmap->lock, flags);
890 if (dirty || need_write)
891 write_page(bitmap, page, 0);
893 if (wait) { /* if any writes were performed, we need to wait on them */
894 if (bitmap->file)
895 wait_event(bitmap->write_wait,
896 atomic_read(&bitmap->pending_writes)==0);
897 else
898 md_super_wait(bitmap->mddev);
900 if (bitmap->flags & BITMAP_WRITE_ERROR)
901 bitmap_file_kick(bitmap);
903 EXPORT_SYMBOL(bitmap_unplug);
905 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
906 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
907 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
908 * memory mapping of the bitmap file
909 * Special cases:
910 * if there's no bitmap file, or if the bitmap file had been
911 * previously kicked from the array, we mark all the bits as
912 * 1's in order to cause a full resync.
914 * We ignore all bits for sectors that end earlier than 'start'.
915 * This is used when reading an out-of-date bitmap...
917 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
919 unsigned long i, chunks, index, oldindex, bit;
920 struct page *page = NULL, *oldpage = NULL;
921 unsigned long num_pages, bit_cnt = 0;
922 struct file *file;
923 unsigned long bytes, offset;
924 int outofdate;
925 int ret = -ENOSPC;
926 void *paddr;
928 chunks = bitmap->chunks;
929 file = bitmap->file;
931 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
933 outofdate = bitmap->flags & BITMAP_STALE;
934 if (outofdate)
935 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
936 "recovery\n", bmname(bitmap));
938 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
939 if (!bitmap->mddev->bitmap_info.external)
940 bytes += sizeof(bitmap_super_t);
942 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
944 if (file && i_size_read(file->f_mapping->host) < bytes) {
945 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
946 bmname(bitmap),
947 (unsigned long) i_size_read(file->f_mapping->host),
948 bytes);
949 goto err;
952 ret = -ENOMEM;
954 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
955 if (!bitmap->filemap)
956 goto err;
958 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
959 bitmap->filemap_attr = kzalloc(
960 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
961 GFP_KERNEL);
962 if (!bitmap->filemap_attr)
963 goto err;
965 oldindex = ~0L;
967 for (i = 0; i < chunks; i++) {
968 int b;
969 index = file_page_index(bitmap, i);
970 bit = file_page_offset(bitmap, i);
971 if (index != oldindex) { /* this is a new page, read it in */
972 int count;
973 /* unmap the old page, we're done with it */
974 if (index == num_pages-1)
975 count = bytes - index * PAGE_SIZE;
976 else
977 count = PAGE_SIZE;
978 if (index == 0 && bitmap->sb_page) {
980 * if we're here then the superblock page
981 * contains some bits (PAGE_SIZE != sizeof sb)
982 * we've already read it in, so just use it
984 page = bitmap->sb_page;
985 offset = sizeof(bitmap_super_t);
986 if (!file)
987 page = read_sb_page(
988 bitmap->mddev,
989 bitmap->mddev->bitmap_info.offset,
990 page,
991 index, count);
992 } else if (file) {
993 page = read_page(file, index, bitmap, count);
994 offset = 0;
995 } else {
996 page = read_sb_page(bitmap->mddev,
997 bitmap->mddev->bitmap_info.offset,
998 NULL,
999 index, count);
1000 offset = 0;
1002 if (IS_ERR(page)) { /* read error */
1003 ret = PTR_ERR(page);
1004 goto err;
1007 oldindex = index;
1008 oldpage = page;
1010 bitmap->filemap[bitmap->file_pages++] = page;
1011 bitmap->last_page_size = count;
1013 if (outofdate) {
1015 * if bitmap is out of date, dirty the
1016 * whole page and write it out
1018 paddr = kmap_atomic(page);
1019 memset(paddr + offset, 0xff,
1020 PAGE_SIZE - offset);
1021 kunmap_atomic(paddr);
1022 write_page(bitmap, page, 1);
1024 ret = -EIO;
1025 if (bitmap->flags & BITMAP_WRITE_ERROR)
1026 goto err;
1029 paddr = kmap_atomic(page);
1030 if (bitmap->flags & BITMAP_HOSTENDIAN)
1031 b = test_bit(bit, paddr);
1032 else
1033 b = test_bit_le(bit, paddr);
1034 kunmap_atomic(paddr);
1035 if (b) {
1036 /* if the disk bit is set, set the memory bit */
1037 int needed = ((sector_t)(i+1) << bitmap->chunkshift
1038 >= start);
1039 bitmap_set_memory_bits(bitmap,
1040 (sector_t)i << bitmap->chunkshift,
1041 needed);
1042 bit_cnt++;
1046 /* everything went OK */
1047 ret = 0;
1048 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1050 if (bit_cnt) { /* Kick recovery if any bits were set */
1051 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1052 md_wakeup_thread(bitmap->mddev->thread);
1055 printk(KERN_INFO "%s: bitmap initialized from disk: "
1056 "read %lu/%lu pages, set %lu of %lu bits\n",
1057 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, chunks);
1059 return 0;
1061 err:
1062 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1063 bmname(bitmap), ret);
1064 return ret;
1067 void bitmap_write_all(struct bitmap *bitmap)
1069 /* We don't actually write all bitmap blocks here,
1070 * just flag them as needing to be written
1072 int i;
1074 spin_lock_irq(&bitmap->lock);
1075 for (i = 0; i < bitmap->file_pages; i++)
1076 set_page_attr(bitmap, bitmap->filemap[i],
1077 BITMAP_PAGE_NEEDWRITE);
1078 bitmap->allclean = 0;
1079 spin_unlock_irq(&bitmap->lock);
1082 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1084 sector_t chunk = offset >> bitmap->chunkshift;
1085 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1086 bitmap->bp[page].count += inc;
1087 bitmap_checkfree(bitmap, page);
1089 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1090 sector_t offset, sector_t *blocks,
1091 int create);
1094 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1095 * out to disk
1098 void bitmap_daemon_work(struct mddev *mddev)
1100 struct bitmap *bitmap;
1101 unsigned long j;
1102 unsigned long flags;
1103 struct page *page = NULL, *lastpage = NULL;
1104 sector_t blocks;
1105 void *paddr;
1107 /* Use a mutex to guard daemon_work against
1108 * bitmap_destroy.
1110 mutex_lock(&mddev->bitmap_info.mutex);
1111 bitmap = mddev->bitmap;
1112 if (bitmap == NULL) {
1113 mutex_unlock(&mddev->bitmap_info.mutex);
1114 return;
1116 if (time_before(jiffies, bitmap->daemon_lastrun
1117 + mddev->bitmap_info.daemon_sleep))
1118 goto done;
1120 bitmap->daemon_lastrun = jiffies;
1121 if (bitmap->allclean) {
1122 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1123 goto done;
1125 bitmap->allclean = 1;
1127 spin_lock_irqsave(&bitmap->lock, flags);
1128 for (j = 0; j < bitmap->chunks; j++) {
1129 bitmap_counter_t *bmc;
1130 if (!bitmap->filemap)
1131 /* error or shutdown */
1132 break;
1134 page = filemap_get_page(bitmap, j);
1136 if (page != lastpage) {
1137 /* skip this page unless it's marked as needing cleaning */
1138 if (!test_page_attr(bitmap, page, BITMAP_PAGE_PENDING)) {
1139 int need_write = test_page_attr(bitmap, page,
1140 BITMAP_PAGE_NEEDWRITE);
1141 if (need_write)
1142 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1144 spin_unlock_irqrestore(&bitmap->lock, flags);
1145 if (need_write)
1146 write_page(bitmap, page, 0);
1147 spin_lock_irqsave(&bitmap->lock, flags);
1148 j |= (PAGE_BITS - 1);
1149 continue;
1152 /* grab the new page, sync and release the old */
1153 if (lastpage != NULL) {
1154 if (test_page_attr(bitmap, lastpage,
1155 BITMAP_PAGE_NEEDWRITE)) {
1156 clear_page_attr(bitmap, lastpage,
1157 BITMAP_PAGE_NEEDWRITE);
1158 spin_unlock_irqrestore(&bitmap->lock, flags);
1159 write_page(bitmap, lastpage, 0);
1160 } else {
1161 set_page_attr(bitmap, lastpage,
1162 BITMAP_PAGE_NEEDWRITE);
1163 bitmap->allclean = 0;
1164 spin_unlock_irqrestore(&bitmap->lock, flags);
1166 } else
1167 spin_unlock_irqrestore(&bitmap->lock, flags);
1168 lastpage = page;
1170 /* We are possibly going to clear some bits, so make
1171 * sure that events_cleared is up-to-date.
1173 if (bitmap->need_sync &&
1174 mddev->bitmap_info.external == 0) {
1175 bitmap_super_t *sb;
1176 bitmap->need_sync = 0;
1177 sb = kmap_atomic(bitmap->sb_page);
1178 sb->events_cleared =
1179 cpu_to_le64(bitmap->events_cleared);
1180 kunmap_atomic(sb);
1181 write_page(bitmap, bitmap->sb_page, 1);
1183 spin_lock_irqsave(&bitmap->lock, flags);
1184 if (!bitmap->need_sync)
1185 clear_page_attr(bitmap, page, BITMAP_PAGE_PENDING);
1186 else
1187 bitmap->allclean = 0;
1189 bmc = bitmap_get_counter(bitmap,
1190 (sector_t)j << bitmap->chunkshift,
1191 &blocks, 0);
1192 if (!bmc)
1193 j |= PAGE_COUNTER_MASK;
1194 else if (*bmc) {
1195 if (*bmc == 1 && !bitmap->need_sync) {
1196 /* we can clear the bit */
1197 *bmc = 0;
1198 bitmap_count_page(bitmap,
1199 (sector_t)j << bitmap->chunkshift,
1200 -1);
1202 /* clear the bit */
1203 paddr = kmap_atomic(page);
1204 if (bitmap->flags & BITMAP_HOSTENDIAN)
1205 clear_bit(file_page_offset(bitmap, j),
1206 paddr);
1207 else
1208 __clear_bit_le(
1209 file_page_offset(bitmap,
1211 paddr);
1212 kunmap_atomic(paddr);
1213 } else if (*bmc <= 2) {
1214 *bmc = 1; /* maybe clear the bit next time */
1215 set_page_attr(bitmap, page, BITMAP_PAGE_PENDING);
1216 bitmap->allclean = 0;
1220 spin_unlock_irqrestore(&bitmap->lock, flags);
1222 /* now sync the final page */
1223 if (lastpage != NULL) {
1224 spin_lock_irqsave(&bitmap->lock, flags);
1225 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1226 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1227 spin_unlock_irqrestore(&bitmap->lock, flags);
1228 write_page(bitmap, lastpage, 0);
1229 } else {
1230 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1231 bitmap->allclean = 0;
1232 spin_unlock_irqrestore(&bitmap->lock, flags);
1236 done:
1237 if (bitmap->allclean == 0)
1238 mddev->thread->timeout =
1239 mddev->bitmap_info.daemon_sleep;
1240 mutex_unlock(&mddev->bitmap_info.mutex);
1243 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1244 sector_t offset, sector_t *blocks,
1245 int create)
1246 __releases(bitmap->lock)
1247 __acquires(bitmap->lock)
1249 /* If 'create', we might release the lock and reclaim it.
1250 * The lock must have been taken with interrupts enabled.
1251 * If !create, we don't release the lock.
1253 sector_t chunk = offset >> bitmap->chunkshift;
1254 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1255 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1256 sector_t csize;
1257 int err;
1259 err = bitmap_checkpage(bitmap, page, create);
1261 if (bitmap->bp[page].hijacked ||
1262 bitmap->bp[page].map == NULL)
1263 csize = ((sector_t)1) << (bitmap->chunkshift +
1264 PAGE_COUNTER_SHIFT - 1);
1265 else
1266 csize = ((sector_t)1) << bitmap->chunkshift;
1267 *blocks = csize - (offset & (csize - 1));
1269 if (err < 0)
1270 return NULL;
1272 /* now locked ... */
1274 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1275 /* should we use the first or second counter field
1276 * of the hijacked pointer? */
1277 int hi = (pageoff > PAGE_COUNTER_MASK);
1278 return &((bitmap_counter_t *)
1279 &bitmap->bp[page].map)[hi];
1280 } else /* page is allocated */
1281 return (bitmap_counter_t *)
1282 &(bitmap->bp[page].map[pageoff]);
1285 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1287 if (!bitmap)
1288 return 0;
1290 if (behind) {
1291 int bw;
1292 atomic_inc(&bitmap->behind_writes);
1293 bw = atomic_read(&bitmap->behind_writes);
1294 if (bw > bitmap->behind_writes_used)
1295 bitmap->behind_writes_used = bw;
1297 pr_debug("inc write-behind count %d/%lu\n",
1298 bw, bitmap->mddev->bitmap_info.max_write_behind);
1301 while (sectors) {
1302 sector_t blocks;
1303 bitmap_counter_t *bmc;
1305 spin_lock_irq(&bitmap->lock);
1306 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1307 if (!bmc) {
1308 spin_unlock_irq(&bitmap->lock);
1309 return 0;
1312 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1313 DEFINE_WAIT(__wait);
1314 /* note that it is safe to do the prepare_to_wait
1315 * after the test as long as we do it before dropping
1316 * the spinlock.
1318 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1319 TASK_UNINTERRUPTIBLE);
1320 spin_unlock_irq(&bitmap->lock);
1321 io_schedule();
1322 finish_wait(&bitmap->overflow_wait, &__wait);
1323 continue;
1326 switch (*bmc) {
1327 case 0:
1328 bitmap_file_set_bit(bitmap, offset);
1329 bitmap_count_page(bitmap, offset, 1);
1330 /* fall through */
1331 case 1:
1332 *bmc = 2;
1335 (*bmc)++;
1337 spin_unlock_irq(&bitmap->lock);
1339 offset += blocks;
1340 if (sectors > blocks)
1341 sectors -= blocks;
1342 else
1343 sectors = 0;
1345 return 0;
1347 EXPORT_SYMBOL(bitmap_startwrite);
1349 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1350 int success, int behind)
1352 if (!bitmap)
1353 return;
1354 if (behind) {
1355 if (atomic_dec_and_test(&bitmap->behind_writes))
1356 wake_up(&bitmap->behind_wait);
1357 pr_debug("dec write-behind count %d/%lu\n",
1358 atomic_read(&bitmap->behind_writes),
1359 bitmap->mddev->bitmap_info.max_write_behind);
1362 while (sectors) {
1363 sector_t blocks;
1364 unsigned long flags;
1365 bitmap_counter_t *bmc;
1367 spin_lock_irqsave(&bitmap->lock, flags);
1368 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1369 if (!bmc) {
1370 spin_unlock_irqrestore(&bitmap->lock, flags);
1371 return;
1374 if (success && !bitmap->mddev->degraded &&
1375 bitmap->events_cleared < bitmap->mddev->events) {
1376 bitmap->events_cleared = bitmap->mddev->events;
1377 bitmap->need_sync = 1;
1378 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1381 if (!success && !NEEDED(*bmc))
1382 *bmc |= NEEDED_MASK;
1384 if (COUNTER(*bmc) == COUNTER_MAX)
1385 wake_up(&bitmap->overflow_wait);
1387 (*bmc)--;
1388 if (*bmc <= 2) {
1389 set_page_attr(bitmap,
1390 filemap_get_page(
1391 bitmap,
1392 offset >> bitmap->chunkshift),
1393 BITMAP_PAGE_PENDING);
1394 bitmap->allclean = 0;
1396 spin_unlock_irqrestore(&bitmap->lock, flags);
1397 offset += blocks;
1398 if (sectors > blocks)
1399 sectors -= blocks;
1400 else
1401 sectors = 0;
1404 EXPORT_SYMBOL(bitmap_endwrite);
1406 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1407 int degraded)
1409 bitmap_counter_t *bmc;
1410 int rv;
1411 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1412 *blocks = 1024;
1413 return 1; /* always resync if no bitmap */
1415 spin_lock_irq(&bitmap->lock);
1416 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1417 rv = 0;
1418 if (bmc) {
1419 /* locked */
1420 if (RESYNC(*bmc))
1421 rv = 1;
1422 else if (NEEDED(*bmc)) {
1423 rv = 1;
1424 if (!degraded) { /* don't set/clear bits if degraded */
1425 *bmc |= RESYNC_MASK;
1426 *bmc &= ~NEEDED_MASK;
1430 spin_unlock_irq(&bitmap->lock);
1431 return rv;
1434 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1435 int degraded)
1437 /* bitmap_start_sync must always report on multiples of whole
1438 * pages, otherwise resync (which is very PAGE_SIZE based) will
1439 * get confused.
1440 * So call __bitmap_start_sync repeatedly (if needed) until
1441 * At least PAGE_SIZE>>9 blocks are covered.
1442 * Return the 'or' of the result.
1444 int rv = 0;
1445 sector_t blocks1;
1447 *blocks = 0;
1448 while (*blocks < (PAGE_SIZE>>9)) {
1449 rv |= __bitmap_start_sync(bitmap, offset,
1450 &blocks1, degraded);
1451 offset += blocks1;
1452 *blocks += blocks1;
1454 return rv;
1456 EXPORT_SYMBOL(bitmap_start_sync);
1458 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1460 bitmap_counter_t *bmc;
1461 unsigned long flags;
1463 if (bitmap == NULL) {
1464 *blocks = 1024;
1465 return;
1467 spin_lock_irqsave(&bitmap->lock, flags);
1468 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1469 if (bmc == NULL)
1470 goto unlock;
1471 /* locked */
1472 if (RESYNC(*bmc)) {
1473 *bmc &= ~RESYNC_MASK;
1475 if (!NEEDED(*bmc) && aborted)
1476 *bmc |= NEEDED_MASK;
1477 else {
1478 if (*bmc <= 2) {
1479 set_page_attr(bitmap,
1480 filemap_get_page(bitmap, offset >> bitmap->chunkshift),
1481 BITMAP_PAGE_PENDING);
1482 bitmap->allclean = 0;
1486 unlock:
1487 spin_unlock_irqrestore(&bitmap->lock, flags);
1489 EXPORT_SYMBOL(bitmap_end_sync);
1491 void bitmap_close_sync(struct bitmap *bitmap)
1493 /* Sync has finished, and any bitmap chunks that weren't synced
1494 * properly have been aborted. It remains to us to clear the
1495 * RESYNC bit wherever it is still on
1497 sector_t sector = 0;
1498 sector_t blocks;
1499 if (!bitmap)
1500 return;
1501 while (sector < bitmap->mddev->resync_max_sectors) {
1502 bitmap_end_sync(bitmap, sector, &blocks, 0);
1503 sector += blocks;
1506 EXPORT_SYMBOL(bitmap_close_sync);
1508 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1510 sector_t s = 0;
1511 sector_t blocks;
1513 if (!bitmap)
1514 return;
1515 if (sector == 0) {
1516 bitmap->last_end_sync = jiffies;
1517 return;
1519 if (time_before(jiffies, (bitmap->last_end_sync
1520 + bitmap->mddev->bitmap_info.daemon_sleep)))
1521 return;
1522 wait_event(bitmap->mddev->recovery_wait,
1523 atomic_read(&bitmap->mddev->recovery_active) == 0);
1525 bitmap->mddev->curr_resync_completed = sector;
1526 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1527 sector &= ~((1ULL << bitmap->chunkshift) - 1);
1528 s = 0;
1529 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1530 bitmap_end_sync(bitmap, s, &blocks, 0);
1531 s += blocks;
1533 bitmap->last_end_sync = jiffies;
1534 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1536 EXPORT_SYMBOL(bitmap_cond_end_sync);
1538 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1540 /* For each chunk covered by any of these sectors, set the
1541 * counter to 1 and set resync_needed. They should all
1542 * be 0 at this point
1545 sector_t secs;
1546 bitmap_counter_t *bmc;
1547 spin_lock_irq(&bitmap->lock);
1548 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1549 if (!bmc) {
1550 spin_unlock_irq(&bitmap->lock);
1551 return;
1553 if (!*bmc) {
1554 struct page *page;
1555 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1556 bitmap_count_page(bitmap, offset, 1);
1557 page = filemap_get_page(bitmap, offset >> bitmap->chunkshift);
1558 set_page_attr(bitmap, page, BITMAP_PAGE_PENDING);
1559 bitmap->allclean = 0;
1561 spin_unlock_irq(&bitmap->lock);
1564 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1565 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1567 unsigned long chunk;
1569 for (chunk = s; chunk <= e; chunk++) {
1570 sector_t sec = (sector_t)chunk << bitmap->chunkshift;
1571 bitmap_set_memory_bits(bitmap, sec, 1);
1572 spin_lock_irq(&bitmap->lock);
1573 bitmap_file_set_bit(bitmap, sec);
1574 spin_unlock_irq(&bitmap->lock);
1575 if (sec < bitmap->mddev->recovery_cp)
1576 /* We are asserting that the array is dirty,
1577 * so move the recovery_cp address back so
1578 * that it is obvious that it is dirty
1580 bitmap->mddev->recovery_cp = sec;
1585 * flush out any pending updates
1587 void bitmap_flush(struct mddev *mddev)
1589 struct bitmap *bitmap = mddev->bitmap;
1590 long sleep;
1592 if (!bitmap) /* there was no bitmap */
1593 return;
1595 /* run the daemon_work three time to ensure everything is flushed
1596 * that can be
1598 sleep = mddev->bitmap_info.daemon_sleep * 2;
1599 bitmap->daemon_lastrun -= sleep;
1600 bitmap_daemon_work(mddev);
1601 bitmap->daemon_lastrun -= sleep;
1602 bitmap_daemon_work(mddev);
1603 bitmap->daemon_lastrun -= sleep;
1604 bitmap_daemon_work(mddev);
1605 bitmap_update_sb(bitmap);
1609 * free memory that was allocated
1611 static void bitmap_free(struct bitmap *bitmap)
1613 unsigned long k, pages;
1614 struct bitmap_page *bp;
1616 if (!bitmap) /* there was no bitmap */
1617 return;
1619 /* release the bitmap file and kill the daemon */
1620 bitmap_file_put(bitmap);
1622 bp = bitmap->bp;
1623 pages = bitmap->pages;
1625 /* free all allocated memory */
1627 if (bp) /* deallocate the page memory */
1628 for (k = 0; k < pages; k++)
1629 if (bp[k].map && !bp[k].hijacked)
1630 kfree(bp[k].map);
1631 kfree(bp);
1632 kfree(bitmap);
1635 void bitmap_destroy(struct mddev *mddev)
1637 struct bitmap *bitmap = mddev->bitmap;
1639 if (!bitmap) /* there was no bitmap */
1640 return;
1642 mutex_lock(&mddev->bitmap_info.mutex);
1643 mddev->bitmap = NULL; /* disconnect from the md device */
1644 mutex_unlock(&mddev->bitmap_info.mutex);
1645 if (mddev->thread)
1646 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1648 if (bitmap->sysfs_can_clear)
1649 sysfs_put(bitmap->sysfs_can_clear);
1651 bitmap_free(bitmap);
1655 * initialize the bitmap structure
1656 * if this returns an error, bitmap_destroy must be called to do clean up
1658 int bitmap_create(struct mddev *mddev)
1660 struct bitmap *bitmap;
1661 sector_t blocks = mddev->resync_max_sectors;
1662 unsigned long chunks;
1663 unsigned long pages;
1664 struct file *file = mddev->bitmap_info.file;
1665 int err;
1666 struct sysfs_dirent *bm = NULL;
1668 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1670 if (!file
1671 && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */
1672 return 0;
1674 BUG_ON(file && mddev->bitmap_info.offset);
1676 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1677 if (!bitmap)
1678 return -ENOMEM;
1680 spin_lock_init(&bitmap->lock);
1681 atomic_set(&bitmap->pending_writes, 0);
1682 init_waitqueue_head(&bitmap->write_wait);
1683 init_waitqueue_head(&bitmap->overflow_wait);
1684 init_waitqueue_head(&bitmap->behind_wait);
1686 bitmap->mddev = mddev;
1688 if (mddev->kobj.sd)
1689 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1690 if (bm) {
1691 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1692 sysfs_put(bm);
1693 } else
1694 bitmap->sysfs_can_clear = NULL;
1696 bitmap->file = file;
1697 if (file) {
1698 get_file(file);
1699 /* As future accesses to this file will use bmap,
1700 * and bypass the page cache, we must sync the file
1701 * first.
1703 vfs_fsync(file, 1);
1705 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1706 if (!mddev->bitmap_info.external) {
1708 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1709 * instructing us to create a new on-disk bitmap instance.
1711 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1712 err = bitmap_new_disk_sb(bitmap);
1713 else
1714 err = bitmap_read_sb(bitmap);
1715 } else {
1716 err = 0;
1717 if (mddev->bitmap_info.chunksize == 0 ||
1718 mddev->bitmap_info.daemon_sleep == 0)
1719 /* chunksize and time_base need to be
1720 * set first. */
1721 err = -EINVAL;
1723 if (err)
1724 goto error;
1726 bitmap->daemon_lastrun = jiffies;
1727 bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize)
1728 - BITMAP_BLOCK_SHIFT);
1730 /* now that chunksize and chunkshift are set, we can use these macros */
1731 chunks = (blocks + bitmap->chunkshift - 1) >>
1732 bitmap->chunkshift;
1733 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1735 BUG_ON(!pages);
1737 bitmap->chunks = chunks;
1738 bitmap->pages = pages;
1739 bitmap->missing_pages = pages;
1741 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1743 err = -ENOMEM;
1744 if (!bitmap->bp)
1745 goto error;
1747 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1748 pages, bmname(bitmap));
1750 mddev->bitmap = bitmap;
1753 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1755 error:
1756 bitmap_free(bitmap);
1757 return err;
1760 int bitmap_load(struct mddev *mddev)
1762 int err = 0;
1763 sector_t start = 0;
1764 sector_t sector = 0;
1765 struct bitmap *bitmap = mddev->bitmap;
1767 if (!bitmap)
1768 goto out;
1770 /* Clear out old bitmap info first: Either there is none, or we
1771 * are resuming after someone else has possibly changed things,
1772 * so we should forget old cached info.
1773 * All chunks should be clean, but some might need_sync.
1775 while (sector < mddev->resync_max_sectors) {
1776 sector_t blocks;
1777 bitmap_start_sync(bitmap, sector, &blocks, 0);
1778 sector += blocks;
1780 bitmap_close_sync(bitmap);
1782 if (mddev->degraded == 0
1783 || bitmap->events_cleared == mddev->events)
1784 /* no need to keep dirty bits to optimise a
1785 * re-add of a missing device */
1786 start = mddev->recovery_cp;
1788 mutex_lock(&mddev->bitmap_info.mutex);
1789 err = bitmap_init_from_disk(bitmap, start);
1790 mutex_unlock(&mddev->bitmap_info.mutex);
1792 if (err)
1793 goto out;
1795 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1796 md_wakeup_thread(mddev->thread);
1798 bitmap_update_sb(bitmap);
1800 if (bitmap->flags & BITMAP_WRITE_ERROR)
1801 err = -EIO;
1802 out:
1803 return err;
1805 EXPORT_SYMBOL_GPL(bitmap_load);
1807 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1809 unsigned long chunk_kb;
1810 unsigned long flags;
1812 if (!bitmap)
1813 return;
1815 spin_lock_irqsave(&bitmap->lock, flags);
1816 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1817 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1818 "%lu%s chunk",
1819 bitmap->pages - bitmap->missing_pages,
1820 bitmap->pages,
1821 (bitmap->pages - bitmap->missing_pages)
1822 << (PAGE_SHIFT - 10),
1823 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1824 chunk_kb ? "KB" : "B");
1825 if (bitmap->file) {
1826 seq_printf(seq, ", file: ");
1827 seq_path(seq, &bitmap->file->f_path, " \t\n");
1830 seq_printf(seq, "\n");
1831 spin_unlock_irqrestore(&bitmap->lock, flags);
1834 static ssize_t
1835 location_show(struct mddev *mddev, char *page)
1837 ssize_t len;
1838 if (mddev->bitmap_info.file)
1839 len = sprintf(page, "file");
1840 else if (mddev->bitmap_info.offset)
1841 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1842 else
1843 len = sprintf(page, "none");
1844 len += sprintf(page+len, "\n");
1845 return len;
1848 static ssize_t
1849 location_store(struct mddev *mddev, const char *buf, size_t len)
1852 if (mddev->pers) {
1853 if (!mddev->pers->quiesce)
1854 return -EBUSY;
1855 if (mddev->recovery || mddev->sync_thread)
1856 return -EBUSY;
1859 if (mddev->bitmap || mddev->bitmap_info.file ||
1860 mddev->bitmap_info.offset) {
1861 /* bitmap already configured. Only option is to clear it */
1862 if (strncmp(buf, "none", 4) != 0)
1863 return -EBUSY;
1864 if (mddev->pers) {
1865 mddev->pers->quiesce(mddev, 1);
1866 bitmap_destroy(mddev);
1867 mddev->pers->quiesce(mddev, 0);
1869 mddev->bitmap_info.offset = 0;
1870 if (mddev->bitmap_info.file) {
1871 struct file *f = mddev->bitmap_info.file;
1872 mddev->bitmap_info.file = NULL;
1873 restore_bitmap_write_access(f);
1874 fput(f);
1876 } else {
1877 /* No bitmap, OK to set a location */
1878 long long offset;
1879 if (strncmp(buf, "none", 4) == 0)
1880 /* nothing to be done */;
1881 else if (strncmp(buf, "file:", 5) == 0) {
1882 /* Not supported yet */
1883 return -EINVAL;
1884 } else {
1885 int rv;
1886 if (buf[0] == '+')
1887 rv = strict_strtoll(buf+1, 10, &offset);
1888 else
1889 rv = strict_strtoll(buf, 10, &offset);
1890 if (rv)
1891 return rv;
1892 if (offset == 0)
1893 return -EINVAL;
1894 if (mddev->bitmap_info.external == 0 &&
1895 mddev->major_version == 0 &&
1896 offset != mddev->bitmap_info.default_offset)
1897 return -EINVAL;
1898 mddev->bitmap_info.offset = offset;
1899 if (mddev->pers) {
1900 mddev->pers->quiesce(mddev, 1);
1901 rv = bitmap_create(mddev);
1902 if (!rv)
1903 rv = bitmap_load(mddev);
1904 if (rv) {
1905 bitmap_destroy(mddev);
1906 mddev->bitmap_info.offset = 0;
1908 mddev->pers->quiesce(mddev, 0);
1909 if (rv)
1910 return rv;
1914 if (!mddev->external) {
1915 /* Ensure new bitmap info is stored in
1916 * metadata promptly.
1918 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1919 md_wakeup_thread(mddev->thread);
1921 return len;
1924 static struct md_sysfs_entry bitmap_location =
1925 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1927 static ssize_t
1928 timeout_show(struct mddev *mddev, char *page)
1930 ssize_t len;
1931 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1932 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1934 len = sprintf(page, "%lu", secs);
1935 if (jifs)
1936 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1937 len += sprintf(page+len, "\n");
1938 return len;
1941 static ssize_t
1942 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1944 /* timeout can be set at any time */
1945 unsigned long timeout;
1946 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1947 if (rv)
1948 return rv;
1950 /* just to make sure we don't overflow... */
1951 if (timeout >= LONG_MAX / HZ)
1952 return -EINVAL;
1954 timeout = timeout * HZ / 10000;
1956 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1957 timeout = MAX_SCHEDULE_TIMEOUT-1;
1958 if (timeout < 1)
1959 timeout = 1;
1960 mddev->bitmap_info.daemon_sleep = timeout;
1961 if (mddev->thread) {
1962 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1963 * the bitmap is all clean and we don't need to
1964 * adjust the timeout right now
1966 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1967 mddev->thread->timeout = timeout;
1968 md_wakeup_thread(mddev->thread);
1971 return len;
1974 static struct md_sysfs_entry bitmap_timeout =
1975 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1977 static ssize_t
1978 backlog_show(struct mddev *mddev, char *page)
1980 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1983 static ssize_t
1984 backlog_store(struct mddev *mddev, const char *buf, size_t len)
1986 unsigned long backlog;
1987 int rv = strict_strtoul(buf, 10, &backlog);
1988 if (rv)
1989 return rv;
1990 if (backlog > COUNTER_MAX)
1991 return -EINVAL;
1992 mddev->bitmap_info.max_write_behind = backlog;
1993 return len;
1996 static struct md_sysfs_entry bitmap_backlog =
1997 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
1999 static ssize_t
2000 chunksize_show(struct mddev *mddev, char *page)
2002 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2005 static ssize_t
2006 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2008 /* Can only be changed when no bitmap is active */
2009 int rv;
2010 unsigned long csize;
2011 if (mddev->bitmap)
2012 return -EBUSY;
2013 rv = strict_strtoul(buf, 10, &csize);
2014 if (rv)
2015 return rv;
2016 if (csize < 512 ||
2017 !is_power_of_2(csize))
2018 return -EINVAL;
2019 mddev->bitmap_info.chunksize = csize;
2020 return len;
2023 static struct md_sysfs_entry bitmap_chunksize =
2024 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2026 static ssize_t metadata_show(struct mddev *mddev, char *page)
2028 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2029 ? "external" : "internal"));
2032 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2034 if (mddev->bitmap ||
2035 mddev->bitmap_info.file ||
2036 mddev->bitmap_info.offset)
2037 return -EBUSY;
2038 if (strncmp(buf, "external", 8) == 0)
2039 mddev->bitmap_info.external = 1;
2040 else if (strncmp(buf, "internal", 8) == 0)
2041 mddev->bitmap_info.external = 0;
2042 else
2043 return -EINVAL;
2044 return len;
2047 static struct md_sysfs_entry bitmap_metadata =
2048 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2050 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2052 int len;
2053 if (mddev->bitmap)
2054 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2055 "false" : "true"));
2056 else
2057 len = sprintf(page, "\n");
2058 return len;
2061 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2063 if (mddev->bitmap == NULL)
2064 return -ENOENT;
2065 if (strncmp(buf, "false", 5) == 0)
2066 mddev->bitmap->need_sync = 1;
2067 else if (strncmp(buf, "true", 4) == 0) {
2068 if (mddev->degraded)
2069 return -EBUSY;
2070 mddev->bitmap->need_sync = 0;
2071 } else
2072 return -EINVAL;
2073 return len;
2076 static struct md_sysfs_entry bitmap_can_clear =
2077 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2079 static ssize_t
2080 behind_writes_used_show(struct mddev *mddev, char *page)
2082 if (mddev->bitmap == NULL)
2083 return sprintf(page, "0\n");
2084 return sprintf(page, "%lu\n",
2085 mddev->bitmap->behind_writes_used);
2088 static ssize_t
2089 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2091 if (mddev->bitmap)
2092 mddev->bitmap->behind_writes_used = 0;
2093 return len;
2096 static struct md_sysfs_entry max_backlog_used =
2097 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2098 behind_writes_used_show, behind_writes_used_reset);
2100 static struct attribute *md_bitmap_attrs[] = {
2101 &bitmap_location.attr,
2102 &bitmap_timeout.attr,
2103 &bitmap_backlog.attr,
2104 &bitmap_chunksize.attr,
2105 &bitmap_metadata.attr,
2106 &bitmap_can_clear.attr,
2107 &max_backlog_used.attr,
2108 NULL
2110 struct attribute_group md_bitmap_group = {
2111 .name = "bitmap",
2112 .attrs = md_bitmap_attrs,