Bug fixes
[cbs-scheduler.git] / drivers / md / bitmap.c
blobb411249b1922ab319cdb0d8d58ea94acc0534b5d
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).
16 * wait if count gets too high, wake when it drops to half.
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/raid/md.h>
30 #include <linux/raid/bitmap.h>
32 /* debug macros */
34 #define DEBUG 0
36 #if DEBUG
37 /* these are for debugging purposes only! */
39 /* define one and only one of these */
40 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
41 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
42 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
43 #define INJECT_FAULTS_4 0 /* undef */
44 #define INJECT_FAULTS_5 0 /* undef */
45 #define INJECT_FAULTS_6 0
47 /* if these are defined, the driver will fail! debug only */
48 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
49 #define INJECT_FATAL_FAULT_2 0 /* undef */
50 #define INJECT_FATAL_FAULT_3 0 /* undef */
51 #endif
53 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
54 #define DPRINTK(x...) do { } while(0)
56 #ifndef PRINTK
57 # if DEBUG > 0
58 # define PRINTK(x...) printk(KERN_DEBUG x)
59 # else
60 # define PRINTK(x...)
61 # endif
62 #endif
64 static inline char * bmname(struct bitmap *bitmap)
66 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
71 * just a placeholder - calls kmalloc for bitmap pages
73 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
75 unsigned char *page;
77 #ifdef INJECT_FAULTS_1
78 page = NULL;
79 #else
80 page = kmalloc(PAGE_SIZE, GFP_NOIO);
81 #endif
82 if (!page)
83 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
84 else
85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
86 bmname(bitmap), page);
87 return page;
91 * for now just a placeholder -- just calls kfree for bitmap pages
93 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
95 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
96 kfree(page);
100 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
102 * 1) check to see if this page is allocated, if it's not then try to alloc
103 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
104 * page pointer directly as a counter
106 * if we find our page, we increment the page's refcount so that it stays
107 * allocated while we're using it
109 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
111 unsigned char *mappage;
113 if (page >= bitmap->pages) {
114 printk(KERN_ALERT
115 "%s: invalid bitmap page request: %lu (> %lu)\n",
116 bmname(bitmap), page, bitmap->pages-1);
117 return -EINVAL;
121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
122 return 0;
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
125 return 0;
127 if (!create)
128 return -ENOENT;
130 spin_unlock_irq(&bitmap->lock);
132 /* this page has not been allocated yet */
134 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
135 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
136 bmname(bitmap));
137 /* failed - set the hijacked flag so that we can use the
138 * pointer as a counter */
139 spin_lock_irq(&bitmap->lock);
140 if (!bitmap->bp[page].map)
141 bitmap->bp[page].hijacked = 1;
142 goto out;
145 /* got a page */
147 spin_lock_irq(&bitmap->lock);
149 /* recheck the page */
151 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
152 /* somebody beat us to getting the page */
153 bitmap_free_page(bitmap, mappage);
154 return 0;
157 /* no page was in place and we have one, so install it */
159 memset(mappage, 0, PAGE_SIZE);
160 bitmap->bp[page].map = mappage;
161 bitmap->missing_pages--;
162 out:
163 return 0;
167 /* if page is completely empty, put it back on the free list, or dealloc it */
168 /* if page was hijacked, unmark the flag so it might get alloced next time */
169 /* Note: lock should be held when calling this */
170 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
172 char *ptr;
174 if (bitmap->bp[page].count) /* page is still busy */
175 return;
177 /* page is no longer in use, it can be released */
179 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
180 bitmap->bp[page].hijacked = 0;
181 bitmap->bp[page].map = NULL;
182 return;
185 /* normal case, free the page */
187 #if 0
188 /* actually ... let's not. We will probably need the page again exactly when
189 * memory is tight and we are flusing to disk
191 return;
192 #else
193 ptr = bitmap->bp[page].map;
194 bitmap->bp[page].map = NULL;
195 bitmap->missing_pages++;
196 bitmap_free_page(bitmap, ptr);
197 return;
198 #endif
203 * bitmap file handling - read and write the bitmap file and its superblock
207 * basic page I/O operations
210 /* IO operations when bitmap is stored near all superblocks */
211 static struct page *read_sb_page(mddev_t *mddev, long offset,
212 struct page *page,
213 unsigned long index, int size)
215 /* choose a good rdev and read the page from there */
217 mdk_rdev_t *rdev;
218 sector_t target;
220 if (!page)
221 page = alloc_page(GFP_KERNEL);
222 if (!page)
223 return ERR_PTR(-ENOMEM);
225 list_for_each_entry(rdev, &mddev->disks, same_set) {
226 if (! test_bit(In_sync, &rdev->flags)
227 || test_bit(Faulty, &rdev->flags))
228 continue;
230 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
232 if (sync_page_io(rdev->bdev, target,
233 roundup(size, bdev_hardsect_size(rdev->bdev)),
234 page, READ)) {
235 page->index = index;
236 attach_page_buffers(page, NULL); /* so that free_buffer will
237 * quietly no-op */
238 return page;
241 return ERR_PTR(-EIO);
245 static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
247 /* Iterate the disks of an mddev, using rcu to protect access to the
248 * linked list, and raising the refcount of devices we return to ensure
249 * they don't disappear while in use.
250 * As devices are only added or removed when raid_disk is < 0 and
251 * nr_pending is 0 and In_sync is clear, the entries we return will
252 * still be in the same position on the list when we re-enter
253 * list_for_each_continue_rcu.
255 struct list_head *pos;
256 rcu_read_lock();
257 if (rdev == NULL)
258 /* start at the beginning */
259 pos = &mddev->disks;
260 else {
261 /* release the previous rdev and start from there. */
262 rdev_dec_pending(rdev, mddev);
263 pos = &rdev->same_set;
265 list_for_each_continue_rcu(pos, &mddev->disks) {
266 rdev = list_entry(pos, mdk_rdev_t, same_set);
267 if (rdev->raid_disk >= 0 &&
268 test_bit(In_sync, &rdev->flags) &&
269 !test_bit(Faulty, &rdev->flags)) {
270 /* this is a usable devices */
271 atomic_inc(&rdev->nr_pending);
272 rcu_read_unlock();
273 return rdev;
276 rcu_read_unlock();
277 return NULL;
280 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
282 mdk_rdev_t *rdev = NULL;
283 mddev_t *mddev = bitmap->mddev;
285 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
286 int size = PAGE_SIZE;
287 if (page->index == bitmap->file_pages-1)
288 size = roundup(bitmap->last_page_size,
289 bdev_hardsect_size(rdev->bdev));
290 /* Just make sure we aren't corrupting data or
291 * metadata
293 if (bitmap->offset < 0) {
294 /* DATA BITMAP METADATA */
295 if (bitmap->offset
296 + (long)(page->index * (PAGE_SIZE/512))
297 + size/512 > 0)
298 /* bitmap runs in to metadata */
299 goto bad_alignment;
300 if (rdev->data_offset + mddev->size*2
301 > rdev->sb_start + bitmap->offset)
302 /* data runs in to bitmap */
303 goto bad_alignment;
304 } else if (rdev->sb_start < rdev->data_offset) {
305 /* METADATA BITMAP DATA */
306 if (rdev->sb_start
307 + bitmap->offset
308 + page->index*(PAGE_SIZE/512) + size/512
309 > rdev->data_offset)
310 /* bitmap runs in to data */
311 goto bad_alignment;
312 } else {
313 /* DATA METADATA BITMAP - no problems */
315 md_super_write(mddev, rdev,
316 rdev->sb_start + bitmap->offset
317 + page->index * (PAGE_SIZE/512),
318 size,
319 page);
322 if (wait)
323 md_super_wait(mddev);
324 return 0;
326 bad_alignment:
327 rcu_read_unlock();
328 return -EINVAL;
331 static void bitmap_file_kick(struct bitmap *bitmap);
333 * write out a page to a file
335 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
337 struct buffer_head *bh;
339 if (bitmap->file == NULL) {
340 switch (write_sb_page(bitmap, page, wait)) {
341 case -EINVAL:
342 bitmap->flags |= BITMAP_WRITE_ERROR;
344 } else {
346 bh = page_buffers(page);
348 while (bh && bh->b_blocknr) {
349 atomic_inc(&bitmap->pending_writes);
350 set_buffer_locked(bh);
351 set_buffer_mapped(bh);
352 submit_bh(WRITE, bh);
353 bh = bh->b_this_page;
356 if (wait) {
357 wait_event(bitmap->write_wait,
358 atomic_read(&bitmap->pending_writes)==0);
361 if (bitmap->flags & BITMAP_WRITE_ERROR)
362 bitmap_file_kick(bitmap);
365 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
367 struct bitmap *bitmap = bh->b_private;
368 unsigned long flags;
370 if (!uptodate) {
371 spin_lock_irqsave(&bitmap->lock, flags);
372 bitmap->flags |= BITMAP_WRITE_ERROR;
373 spin_unlock_irqrestore(&bitmap->lock, flags);
375 if (atomic_dec_and_test(&bitmap->pending_writes))
376 wake_up(&bitmap->write_wait);
379 /* copied from buffer.c */
380 static void
381 __clear_page_buffers(struct page *page)
383 ClearPagePrivate(page);
384 set_page_private(page, 0);
385 page_cache_release(page);
387 static void free_buffers(struct page *page)
389 struct buffer_head *bh = page_buffers(page);
391 while (bh) {
392 struct buffer_head *next = bh->b_this_page;
393 free_buffer_head(bh);
394 bh = next;
396 __clear_page_buffers(page);
397 put_page(page);
400 /* read a page from a file.
401 * We both read the page, and attach buffers to the page to record the
402 * address of each block (using bmap). These addresses will be used
403 * to write the block later, completely bypassing the filesystem.
404 * This usage is similar to how swap files are handled, and allows us
405 * to write to a file with no concerns of memory allocation failing.
407 static struct page *read_page(struct file *file, unsigned long index,
408 struct bitmap *bitmap,
409 unsigned long count)
411 struct page *page = NULL;
412 struct inode *inode = file->f_path.dentry->d_inode;
413 struct buffer_head *bh;
414 sector_t block;
416 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
417 (unsigned long long)index << PAGE_SHIFT);
419 page = alloc_page(GFP_KERNEL);
420 if (!page)
421 page = ERR_PTR(-ENOMEM);
422 if (IS_ERR(page))
423 goto out;
425 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
426 if (!bh) {
427 put_page(page);
428 page = ERR_PTR(-ENOMEM);
429 goto out;
431 attach_page_buffers(page, bh);
432 block = index << (PAGE_SHIFT - inode->i_blkbits);
433 while (bh) {
434 if (count == 0)
435 bh->b_blocknr = 0;
436 else {
437 bh->b_blocknr = bmap(inode, block);
438 if (bh->b_blocknr == 0) {
439 /* Cannot use this file! */
440 free_buffers(page);
441 page = ERR_PTR(-EINVAL);
442 goto out;
444 bh->b_bdev = inode->i_sb->s_bdev;
445 if (count < (1<<inode->i_blkbits))
446 count = 0;
447 else
448 count -= (1<<inode->i_blkbits);
450 bh->b_end_io = end_bitmap_write;
451 bh->b_private = bitmap;
452 atomic_inc(&bitmap->pending_writes);
453 set_buffer_locked(bh);
454 set_buffer_mapped(bh);
455 submit_bh(READ, bh);
457 block++;
458 bh = bh->b_this_page;
460 page->index = index;
462 wait_event(bitmap->write_wait,
463 atomic_read(&bitmap->pending_writes)==0);
464 if (bitmap->flags & BITMAP_WRITE_ERROR) {
465 free_buffers(page);
466 page = ERR_PTR(-EIO);
468 out:
469 if (IS_ERR(page))
470 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
471 (int)PAGE_SIZE,
472 (unsigned long long)index << PAGE_SHIFT,
473 PTR_ERR(page));
474 return page;
478 * bitmap file superblock operations
481 /* update the event counter and sync the superblock to disk */
482 void bitmap_update_sb(struct bitmap *bitmap)
484 bitmap_super_t *sb;
485 unsigned long flags;
487 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
488 return;
489 spin_lock_irqsave(&bitmap->lock, flags);
490 if (!bitmap->sb_page) { /* no superblock */
491 spin_unlock_irqrestore(&bitmap->lock, flags);
492 return;
494 spin_unlock_irqrestore(&bitmap->lock, flags);
495 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
496 sb->events = cpu_to_le64(bitmap->mddev->events);
497 if (bitmap->mddev->events < bitmap->events_cleared) {
498 /* rocking back to read-only */
499 bitmap->events_cleared = bitmap->mddev->events;
500 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
502 kunmap_atomic(sb, KM_USER0);
503 write_page(bitmap, bitmap->sb_page, 1);
506 /* print out the bitmap file superblock */
507 void bitmap_print_sb(struct bitmap *bitmap)
509 bitmap_super_t *sb;
511 if (!bitmap || !bitmap->sb_page)
512 return;
513 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
514 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
515 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
516 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
517 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
518 *(__u32 *)(sb->uuid+0),
519 *(__u32 *)(sb->uuid+4),
520 *(__u32 *)(sb->uuid+8),
521 *(__u32 *)(sb->uuid+12));
522 printk(KERN_DEBUG " events: %llu\n",
523 (unsigned long long) le64_to_cpu(sb->events));
524 printk(KERN_DEBUG "events cleared: %llu\n",
525 (unsigned long long) le64_to_cpu(sb->events_cleared));
526 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
527 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
528 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
529 printk(KERN_DEBUG " sync size: %llu KB\n",
530 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
531 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
532 kunmap_atomic(sb, KM_USER0);
535 /* read the superblock from the bitmap file and initialize some bitmap fields */
536 static int bitmap_read_sb(struct bitmap *bitmap)
538 char *reason = NULL;
539 bitmap_super_t *sb;
540 unsigned long chunksize, daemon_sleep, write_behind;
541 unsigned long long events;
542 int err = -EINVAL;
544 /* page 0 is the superblock, read it... */
545 if (bitmap->file) {
546 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
547 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
549 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
550 } else {
551 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset,
552 NULL,
553 0, sizeof(bitmap_super_t));
555 if (IS_ERR(bitmap->sb_page)) {
556 err = PTR_ERR(bitmap->sb_page);
557 bitmap->sb_page = NULL;
558 return err;
561 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
563 chunksize = le32_to_cpu(sb->chunksize);
564 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
565 write_behind = le32_to_cpu(sb->write_behind);
567 /* verify that the bitmap-specific fields are valid */
568 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
569 reason = "bad magic";
570 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
571 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
572 reason = "unrecognized superblock version";
573 else if (chunksize < PAGE_SIZE)
574 reason = "bitmap chunksize too small";
575 else if ((1 << ffz(~chunksize)) != chunksize)
576 reason = "bitmap chunksize not a power of 2";
577 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
578 reason = "daemon sleep period out of range";
579 else if (write_behind > COUNTER_MAX)
580 reason = "write-behind limit out of range (0 - 16383)";
581 if (reason) {
582 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
583 bmname(bitmap), reason);
584 goto out;
587 /* keep the array size field of the bitmap superblock up to date */
588 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
590 if (!bitmap->mddev->persistent)
591 goto success;
594 * if we have a persistent array superblock, compare the
595 * bitmap's UUID and event counter to the mddev's
597 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
598 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
599 bmname(bitmap));
600 goto out;
602 events = le64_to_cpu(sb->events);
603 if (events < bitmap->mddev->events) {
604 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
605 "-- forcing full recovery\n", bmname(bitmap), events,
606 (unsigned long long) bitmap->mddev->events);
607 sb->state |= cpu_to_le32(BITMAP_STALE);
609 success:
610 /* assign fields using values from superblock */
611 bitmap->chunksize = chunksize;
612 bitmap->daemon_sleep = daemon_sleep;
613 bitmap->daemon_lastrun = jiffies;
614 bitmap->max_write_behind = write_behind;
615 bitmap->flags |= le32_to_cpu(sb->state);
616 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
617 bitmap->flags |= BITMAP_HOSTENDIAN;
618 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
619 if (sb->state & cpu_to_le32(BITMAP_STALE))
620 bitmap->events_cleared = bitmap->mddev->events;
621 err = 0;
622 out:
623 kunmap_atomic(sb, KM_USER0);
624 if (err)
625 bitmap_print_sb(bitmap);
626 return err;
629 enum bitmap_mask_op {
630 MASK_SET,
631 MASK_UNSET
634 /* record the state of the bitmap in the superblock. Return the old value */
635 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
636 enum bitmap_mask_op op)
638 bitmap_super_t *sb;
639 unsigned long flags;
640 int old;
642 spin_lock_irqsave(&bitmap->lock, flags);
643 if (!bitmap->sb_page) { /* can't set the state */
644 spin_unlock_irqrestore(&bitmap->lock, flags);
645 return 0;
647 spin_unlock_irqrestore(&bitmap->lock, flags);
648 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
649 old = le32_to_cpu(sb->state) & bits;
650 switch (op) {
651 case MASK_SET: sb->state |= cpu_to_le32(bits);
652 break;
653 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
654 break;
655 default: BUG();
657 kunmap_atomic(sb, KM_USER0);
658 return old;
662 * general bitmap file operations
665 /* calculate the index of the page that contains this bit */
666 static inline unsigned long file_page_index(unsigned long chunk)
668 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
671 /* calculate the (bit) offset of this bit within a page */
672 static inline unsigned long file_page_offset(unsigned long chunk)
674 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
678 * return a pointer to the page in the filemap that contains the given bit
680 * this lookup is complicated by the fact that the bitmap sb might be exactly
681 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
682 * 0 or page 1
684 static inline struct page *filemap_get_page(struct bitmap *bitmap,
685 unsigned long chunk)
687 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
688 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
692 static void bitmap_file_unmap(struct bitmap *bitmap)
694 struct page **map, *sb_page;
695 unsigned long *attr;
696 int pages;
697 unsigned long flags;
699 spin_lock_irqsave(&bitmap->lock, flags);
700 map = bitmap->filemap;
701 bitmap->filemap = NULL;
702 attr = bitmap->filemap_attr;
703 bitmap->filemap_attr = NULL;
704 pages = bitmap->file_pages;
705 bitmap->file_pages = 0;
706 sb_page = bitmap->sb_page;
707 bitmap->sb_page = NULL;
708 spin_unlock_irqrestore(&bitmap->lock, flags);
710 while (pages--)
711 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
712 free_buffers(map[pages]);
713 kfree(map);
714 kfree(attr);
716 if (sb_page)
717 free_buffers(sb_page);
720 static void bitmap_file_put(struct bitmap *bitmap)
722 struct file *file;
723 unsigned long flags;
725 spin_lock_irqsave(&bitmap->lock, flags);
726 file = bitmap->file;
727 bitmap->file = NULL;
728 spin_unlock_irqrestore(&bitmap->lock, flags);
730 if (file)
731 wait_event(bitmap->write_wait,
732 atomic_read(&bitmap->pending_writes)==0);
733 bitmap_file_unmap(bitmap);
735 if (file) {
736 struct inode *inode = file->f_path.dentry->d_inode;
737 invalidate_mapping_pages(inode->i_mapping, 0, -1);
738 fput(file);
744 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
745 * then it is no longer reliable, so we stop using it and we mark the file
746 * as failed in the superblock
748 static void bitmap_file_kick(struct bitmap *bitmap)
750 char *path, *ptr = NULL;
752 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
753 bitmap_update_sb(bitmap);
755 if (bitmap->file) {
756 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
757 if (path)
758 ptr = d_path(&bitmap->file->f_path, path,
759 PAGE_SIZE);
762 printk(KERN_ALERT
763 "%s: kicking failed bitmap file %s from array!\n",
764 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
766 kfree(path);
767 } else
768 printk(KERN_ALERT
769 "%s: disabling internal bitmap due to errors\n",
770 bmname(bitmap));
773 bitmap_file_put(bitmap);
775 return;
778 enum bitmap_page_attr {
779 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
780 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
781 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
784 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
785 enum bitmap_page_attr attr)
787 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
790 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
791 enum bitmap_page_attr attr)
793 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
796 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
797 enum bitmap_page_attr attr)
799 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
803 * bitmap_file_set_bit -- called before performing a write to the md device
804 * to set (and eventually sync) a particular bit in the bitmap file
806 * we set the bit immediately, then we record the page number so that
807 * when an unplug occurs, we can flush the dirty pages out to disk
809 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
811 unsigned long bit;
812 struct page *page;
813 void *kaddr;
814 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
816 if (!bitmap->filemap) {
817 return;
820 page = filemap_get_page(bitmap, chunk);
821 if (!page) return;
822 bit = file_page_offset(chunk);
824 /* set the bit */
825 kaddr = kmap_atomic(page, KM_USER0);
826 if (bitmap->flags & BITMAP_HOSTENDIAN)
827 set_bit(bit, kaddr);
828 else
829 ext2_set_bit(bit, kaddr);
830 kunmap_atomic(kaddr, KM_USER0);
831 PRINTK("set file bit %lu page %lu\n", bit, page->index);
833 /* record page number so it gets flushed to disk when unplug occurs */
834 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
838 /* this gets called when the md device is ready to unplug its underlying
839 * (slave) device queues -- before we let any writes go down, we need to
840 * sync the dirty pages of the bitmap file to disk */
841 void bitmap_unplug(struct bitmap *bitmap)
843 unsigned long i, flags;
844 int dirty, need_write;
845 struct page *page;
846 int wait = 0;
848 if (!bitmap)
849 return;
851 /* look at each page to see if there are any set bits that need to be
852 * flushed out to disk */
853 for (i = 0; i < bitmap->file_pages; i++) {
854 spin_lock_irqsave(&bitmap->lock, flags);
855 if (!bitmap->filemap) {
856 spin_unlock_irqrestore(&bitmap->lock, flags);
857 return;
859 page = bitmap->filemap[i];
860 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
861 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
862 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
863 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
864 if (dirty)
865 wait = 1;
866 spin_unlock_irqrestore(&bitmap->lock, flags);
868 if (dirty | need_write)
869 write_page(bitmap, page, 0);
871 if (wait) { /* if any writes were performed, we need to wait on them */
872 if (bitmap->file)
873 wait_event(bitmap->write_wait,
874 atomic_read(&bitmap->pending_writes)==0);
875 else
876 md_super_wait(bitmap->mddev);
878 if (bitmap->flags & BITMAP_WRITE_ERROR)
879 bitmap_file_kick(bitmap);
882 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
883 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
884 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
885 * memory mapping of the bitmap file
886 * Special cases:
887 * if there's no bitmap file, or if the bitmap file had been
888 * previously kicked from the array, we mark all the bits as
889 * 1's in order to cause a full resync.
891 * We ignore all bits for sectors that end earlier than 'start'.
892 * This is used when reading an out-of-date bitmap...
894 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
896 unsigned long i, chunks, index, oldindex, bit;
897 struct page *page = NULL, *oldpage = NULL;
898 unsigned long num_pages, bit_cnt = 0;
899 struct file *file;
900 unsigned long bytes, offset;
901 int outofdate;
902 int ret = -ENOSPC;
903 void *paddr;
905 chunks = bitmap->chunks;
906 file = bitmap->file;
908 BUG_ON(!file && !bitmap->offset);
910 #ifdef INJECT_FAULTS_3
911 outofdate = 1;
912 #else
913 outofdate = bitmap->flags & BITMAP_STALE;
914 #endif
915 if (outofdate)
916 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
917 "recovery\n", bmname(bitmap));
919 bytes = (chunks + 7) / 8;
921 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
923 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
924 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
925 bmname(bitmap),
926 (unsigned long) i_size_read(file->f_mapping->host),
927 bytes + sizeof(bitmap_super_t));
928 goto err;
931 ret = -ENOMEM;
933 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
934 if (!bitmap->filemap)
935 goto err;
937 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
938 bitmap->filemap_attr = kzalloc(
939 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
940 GFP_KERNEL);
941 if (!bitmap->filemap_attr)
942 goto err;
944 oldindex = ~0L;
946 for (i = 0; i < chunks; i++) {
947 int b;
948 index = file_page_index(i);
949 bit = file_page_offset(i);
950 if (index != oldindex) { /* this is a new page, read it in */
951 int count;
952 /* unmap the old page, we're done with it */
953 if (index == num_pages-1)
954 count = bytes + sizeof(bitmap_super_t)
955 - index * PAGE_SIZE;
956 else
957 count = PAGE_SIZE;
958 if (index == 0) {
960 * if we're here then the superblock page
961 * contains some bits (PAGE_SIZE != sizeof sb)
962 * we've already read it in, so just use it
964 page = bitmap->sb_page;
965 offset = sizeof(bitmap_super_t);
966 if (!file)
967 read_sb_page(bitmap->mddev,
968 bitmap->offset,
969 page,
970 index, count);
971 } else if (file) {
972 page = read_page(file, index, bitmap, count);
973 offset = 0;
974 } else {
975 page = read_sb_page(bitmap->mddev, bitmap->offset,
976 NULL,
977 index, count);
978 offset = 0;
980 if (IS_ERR(page)) { /* read error */
981 ret = PTR_ERR(page);
982 goto err;
985 oldindex = index;
986 oldpage = page;
988 bitmap->filemap[bitmap->file_pages++] = page;
989 bitmap->last_page_size = count;
991 if (outofdate) {
993 * if bitmap is out of date, dirty the
994 * whole page and write it out
996 paddr = kmap_atomic(page, KM_USER0);
997 memset(paddr + offset, 0xff,
998 PAGE_SIZE - offset);
999 kunmap_atomic(paddr, KM_USER0);
1000 write_page(bitmap, page, 1);
1002 ret = -EIO;
1003 if (bitmap->flags & BITMAP_WRITE_ERROR)
1004 goto err;
1007 paddr = kmap_atomic(page, KM_USER0);
1008 if (bitmap->flags & BITMAP_HOSTENDIAN)
1009 b = test_bit(bit, paddr);
1010 else
1011 b = ext2_test_bit(bit, paddr);
1012 kunmap_atomic(paddr, KM_USER0);
1013 if (b) {
1014 /* if the disk bit is set, set the memory bit */
1015 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1016 >= start);
1017 bitmap_set_memory_bits(bitmap,
1018 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1019 needed);
1020 bit_cnt++;
1021 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1025 /* everything went OK */
1026 ret = 0;
1027 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1029 if (bit_cnt) { /* Kick recovery if any bits were set */
1030 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1031 md_wakeup_thread(bitmap->mddev->thread);
1034 printk(KERN_INFO "%s: bitmap initialized from disk: "
1035 "read %lu/%lu pages, set %lu bits\n",
1036 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1038 return 0;
1040 err:
1041 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1042 bmname(bitmap), ret);
1043 return ret;
1046 void bitmap_write_all(struct bitmap *bitmap)
1048 /* We don't actually write all bitmap blocks here,
1049 * just flag them as needing to be written
1051 int i;
1053 for (i=0; i < bitmap->file_pages; i++)
1054 set_page_attr(bitmap, bitmap->filemap[i],
1055 BITMAP_PAGE_NEEDWRITE);
1059 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1061 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1062 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1063 bitmap->bp[page].count += inc;
1065 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1066 (unsigned long long)offset, inc, bitmap->bp[page].count);
1068 bitmap_checkfree(bitmap, page);
1070 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1071 sector_t offset, int *blocks,
1072 int create);
1075 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1076 * out to disk
1079 void bitmap_daemon_work(struct bitmap *bitmap)
1081 unsigned long j;
1082 unsigned long flags;
1083 struct page *page = NULL, *lastpage = NULL;
1084 int blocks;
1085 void *paddr;
1087 if (bitmap == NULL)
1088 return;
1089 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1090 goto done;
1092 bitmap->daemon_lastrun = jiffies;
1093 if (bitmap->allclean) {
1094 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1095 return;
1097 bitmap->allclean = 1;
1099 for (j = 0; j < bitmap->chunks; j++) {
1100 bitmap_counter_t *bmc;
1101 spin_lock_irqsave(&bitmap->lock, flags);
1102 if (!bitmap->filemap) {
1103 /* error or shutdown */
1104 spin_unlock_irqrestore(&bitmap->lock, flags);
1105 break;
1108 page = filemap_get_page(bitmap, j);
1110 if (page != lastpage) {
1111 /* skip this page unless it's marked as needing cleaning */
1112 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1113 int need_write = test_page_attr(bitmap, page,
1114 BITMAP_PAGE_NEEDWRITE);
1115 if (need_write)
1116 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1118 spin_unlock_irqrestore(&bitmap->lock, flags);
1119 if (need_write) {
1120 write_page(bitmap, page, 0);
1121 bitmap->allclean = 0;
1123 continue;
1126 /* grab the new page, sync and release the old */
1127 if (lastpage != NULL) {
1128 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1129 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1130 spin_unlock_irqrestore(&bitmap->lock, flags);
1131 write_page(bitmap, lastpage, 0);
1132 } else {
1133 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1134 spin_unlock_irqrestore(&bitmap->lock, flags);
1136 } else
1137 spin_unlock_irqrestore(&bitmap->lock, flags);
1138 lastpage = page;
1140 /* We are possibly going to clear some bits, so make
1141 * sure that events_cleared is up-to-date.
1143 if (bitmap->need_sync) {
1144 bitmap_super_t *sb;
1145 bitmap->need_sync = 0;
1146 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1147 sb->events_cleared =
1148 cpu_to_le64(bitmap->events_cleared);
1149 kunmap_atomic(sb, KM_USER0);
1150 write_page(bitmap, bitmap->sb_page, 1);
1152 spin_lock_irqsave(&bitmap->lock, flags);
1153 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1155 bmc = bitmap_get_counter(bitmap,
1156 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1157 &blocks, 0);
1158 if (bmc) {
1160 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1162 if (*bmc)
1163 bitmap->allclean = 0;
1165 if (*bmc == 2) {
1166 *bmc=1; /* maybe clear the bit next time */
1167 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1168 } else if (*bmc == 1) {
1169 /* we can clear the bit */
1170 *bmc = 0;
1171 bitmap_count_page(bitmap,
1172 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1173 -1);
1175 /* clear the bit */
1176 paddr = kmap_atomic(page, KM_USER0);
1177 if (bitmap->flags & BITMAP_HOSTENDIAN)
1178 clear_bit(file_page_offset(j), paddr);
1179 else
1180 ext2_clear_bit(file_page_offset(j), paddr);
1181 kunmap_atomic(paddr, KM_USER0);
1184 spin_unlock_irqrestore(&bitmap->lock, flags);
1187 /* now sync the final page */
1188 if (lastpage != NULL) {
1189 spin_lock_irqsave(&bitmap->lock, flags);
1190 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1191 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1192 spin_unlock_irqrestore(&bitmap->lock, flags);
1193 write_page(bitmap, lastpage, 0);
1194 } else {
1195 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1196 spin_unlock_irqrestore(&bitmap->lock, flags);
1200 done:
1201 if (bitmap->allclean == 0)
1202 bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1205 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1206 sector_t offset, int *blocks,
1207 int create)
1209 /* If 'create', we might release the lock and reclaim it.
1210 * The lock must have been taken with interrupts enabled.
1211 * If !create, we don't release the lock.
1213 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1214 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1215 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1216 sector_t csize;
1218 if (bitmap_checkpage(bitmap, page, create) < 0) {
1219 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1220 *blocks = csize - (offset & (csize- 1));
1221 return NULL;
1223 /* now locked ... */
1225 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1226 /* should we use the first or second counter field
1227 * of the hijacked pointer? */
1228 int hi = (pageoff > PAGE_COUNTER_MASK);
1229 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1230 PAGE_COUNTER_SHIFT - 1);
1231 *blocks = csize - (offset & (csize- 1));
1232 return &((bitmap_counter_t *)
1233 &bitmap->bp[page].map)[hi];
1234 } else { /* page is allocated */
1235 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1236 *blocks = csize - (offset & (csize- 1));
1237 return (bitmap_counter_t *)
1238 &(bitmap->bp[page].map[pageoff]);
1242 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1244 if (!bitmap) return 0;
1246 if (behind) {
1247 atomic_inc(&bitmap->behind_writes);
1248 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1249 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1252 while (sectors) {
1253 int blocks;
1254 bitmap_counter_t *bmc;
1256 spin_lock_irq(&bitmap->lock);
1257 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1258 if (!bmc) {
1259 spin_unlock_irq(&bitmap->lock);
1260 return 0;
1263 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1264 DEFINE_WAIT(__wait);
1265 /* note that it is safe to do the prepare_to_wait
1266 * after the test as long as we do it before dropping
1267 * the spinlock.
1269 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1270 TASK_UNINTERRUPTIBLE);
1271 spin_unlock_irq(&bitmap->lock);
1272 blk_unplug(bitmap->mddev->queue);
1273 schedule();
1274 finish_wait(&bitmap->overflow_wait, &__wait);
1275 continue;
1278 switch(*bmc) {
1279 case 0:
1280 bitmap_file_set_bit(bitmap, offset);
1281 bitmap_count_page(bitmap,offset, 1);
1282 blk_plug_device_unlocked(bitmap->mddev->queue);
1283 /* fall through */
1284 case 1:
1285 *bmc = 2;
1288 (*bmc)++;
1290 spin_unlock_irq(&bitmap->lock);
1292 offset += blocks;
1293 if (sectors > blocks)
1294 sectors -= blocks;
1295 else sectors = 0;
1297 bitmap->allclean = 0;
1298 return 0;
1301 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1302 int success, int behind)
1304 if (!bitmap) return;
1305 if (behind) {
1306 atomic_dec(&bitmap->behind_writes);
1307 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1308 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1311 while (sectors) {
1312 int blocks;
1313 unsigned long flags;
1314 bitmap_counter_t *bmc;
1316 spin_lock_irqsave(&bitmap->lock, flags);
1317 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1318 if (!bmc) {
1319 spin_unlock_irqrestore(&bitmap->lock, flags);
1320 return;
1323 if (success &&
1324 bitmap->events_cleared < bitmap->mddev->events) {
1325 bitmap->events_cleared = bitmap->mddev->events;
1326 bitmap->need_sync = 1;
1329 if (!success && ! (*bmc & NEEDED_MASK))
1330 *bmc |= NEEDED_MASK;
1332 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1333 wake_up(&bitmap->overflow_wait);
1335 (*bmc)--;
1336 if (*bmc <= 2) {
1337 set_page_attr(bitmap,
1338 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1339 BITMAP_PAGE_CLEAN);
1341 spin_unlock_irqrestore(&bitmap->lock, flags);
1342 offset += blocks;
1343 if (sectors > blocks)
1344 sectors -= blocks;
1345 else sectors = 0;
1349 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1350 int degraded)
1352 bitmap_counter_t *bmc;
1353 int rv;
1354 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1355 *blocks = 1024;
1356 return 1; /* always resync if no bitmap */
1358 spin_lock_irq(&bitmap->lock);
1359 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1360 rv = 0;
1361 if (bmc) {
1362 /* locked */
1363 if (RESYNC(*bmc))
1364 rv = 1;
1365 else if (NEEDED(*bmc)) {
1366 rv = 1;
1367 if (!degraded) { /* don't set/clear bits if degraded */
1368 *bmc |= RESYNC_MASK;
1369 *bmc &= ~NEEDED_MASK;
1373 spin_unlock_irq(&bitmap->lock);
1374 bitmap->allclean = 0;
1375 return rv;
1378 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1380 bitmap_counter_t *bmc;
1381 unsigned long flags;
1383 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1384 */ if (bitmap == NULL) {
1385 *blocks = 1024;
1386 return;
1388 spin_lock_irqsave(&bitmap->lock, flags);
1389 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1390 if (bmc == NULL)
1391 goto unlock;
1392 /* locked */
1394 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1396 if (RESYNC(*bmc)) {
1397 *bmc &= ~RESYNC_MASK;
1399 if (!NEEDED(*bmc) && aborted)
1400 *bmc |= NEEDED_MASK;
1401 else {
1402 if (*bmc <= 2) {
1403 set_page_attr(bitmap,
1404 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1405 BITMAP_PAGE_CLEAN);
1409 unlock:
1410 spin_unlock_irqrestore(&bitmap->lock, flags);
1411 bitmap->allclean = 0;
1414 void bitmap_close_sync(struct bitmap *bitmap)
1416 /* Sync has finished, and any bitmap chunks that weren't synced
1417 * properly have been aborted. It remains to us to clear the
1418 * RESYNC bit wherever it is still on
1420 sector_t sector = 0;
1421 int blocks;
1422 if (!bitmap)
1423 return;
1424 while (sector < bitmap->mddev->resync_max_sectors) {
1425 bitmap_end_sync(bitmap, sector, &blocks, 0);
1426 sector += blocks;
1430 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1432 sector_t s = 0;
1433 int blocks;
1435 if (!bitmap)
1436 return;
1437 if (sector == 0) {
1438 bitmap->last_end_sync = jiffies;
1439 return;
1441 if (time_before(jiffies, (bitmap->last_end_sync
1442 + bitmap->daemon_sleep * HZ)))
1443 return;
1444 wait_event(bitmap->mddev->recovery_wait,
1445 atomic_read(&bitmap->mddev->recovery_active) == 0);
1447 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1448 s = 0;
1449 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1450 bitmap_end_sync(bitmap, s, &blocks, 0);
1451 s += blocks;
1453 bitmap->last_end_sync = jiffies;
1456 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1458 /* For each chunk covered by any of these sectors, set the
1459 * counter to 1 and set resync_needed. They should all
1460 * be 0 at this point
1463 int secs;
1464 bitmap_counter_t *bmc;
1465 spin_lock_irq(&bitmap->lock);
1466 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1467 if (!bmc) {
1468 spin_unlock_irq(&bitmap->lock);
1469 return;
1471 if (! *bmc) {
1472 struct page *page;
1473 *bmc = 1 | (needed?NEEDED_MASK:0);
1474 bitmap_count_page(bitmap, offset, 1);
1475 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1476 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1478 spin_unlock_irq(&bitmap->lock);
1479 bitmap->allclean = 0;
1482 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1483 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1485 unsigned long chunk;
1487 for (chunk = s; chunk <= e; chunk++) {
1488 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
1489 bitmap_set_memory_bits(bitmap, sec, 1);
1490 bitmap_file_set_bit(bitmap, sec);
1495 * flush out any pending updates
1497 void bitmap_flush(mddev_t *mddev)
1499 struct bitmap *bitmap = mddev->bitmap;
1500 int sleep;
1502 if (!bitmap) /* there was no bitmap */
1503 return;
1505 /* run the daemon_work three time to ensure everything is flushed
1506 * that can be
1508 sleep = bitmap->daemon_sleep;
1509 bitmap->daemon_sleep = 0;
1510 bitmap_daemon_work(bitmap);
1511 bitmap_daemon_work(bitmap);
1512 bitmap_daemon_work(bitmap);
1513 bitmap->daemon_sleep = sleep;
1514 bitmap_update_sb(bitmap);
1518 * free memory that was allocated
1520 static void bitmap_free(struct bitmap *bitmap)
1522 unsigned long k, pages;
1523 struct bitmap_page *bp;
1525 if (!bitmap) /* there was no bitmap */
1526 return;
1528 /* release the bitmap file and kill the daemon */
1529 bitmap_file_put(bitmap);
1531 bp = bitmap->bp;
1532 pages = bitmap->pages;
1534 /* free all allocated memory */
1536 if (bp) /* deallocate the page memory */
1537 for (k = 0; k < pages; k++)
1538 if (bp[k].map && !bp[k].hijacked)
1539 kfree(bp[k].map);
1540 kfree(bp);
1541 kfree(bitmap);
1543 void bitmap_destroy(mddev_t *mddev)
1545 struct bitmap *bitmap = mddev->bitmap;
1547 if (!bitmap) /* there was no bitmap */
1548 return;
1550 mddev->bitmap = NULL; /* disconnect from the md device */
1551 if (mddev->thread)
1552 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1554 bitmap_free(bitmap);
1558 * initialize the bitmap structure
1559 * if this returns an error, bitmap_destroy must be called to do clean up
1561 int bitmap_create(mddev_t *mddev)
1563 struct bitmap *bitmap;
1564 unsigned long blocks = mddev->resync_max_sectors;
1565 unsigned long chunks;
1566 unsigned long pages;
1567 struct file *file = mddev->bitmap_file;
1568 int err;
1569 sector_t start;
1571 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1573 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1574 return 0;
1576 BUG_ON(file && mddev->bitmap_offset);
1578 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1579 if (!bitmap)
1580 return -ENOMEM;
1582 spin_lock_init(&bitmap->lock);
1583 atomic_set(&bitmap->pending_writes, 0);
1584 init_waitqueue_head(&bitmap->write_wait);
1585 init_waitqueue_head(&bitmap->overflow_wait);
1587 bitmap->mddev = mddev;
1589 bitmap->file = file;
1590 bitmap->offset = mddev->bitmap_offset;
1591 if (file) {
1592 get_file(file);
1593 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1594 SYNC_FILE_RANGE_WAIT_BEFORE |
1595 SYNC_FILE_RANGE_WRITE |
1596 SYNC_FILE_RANGE_WAIT_AFTER);
1598 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1599 err = bitmap_read_sb(bitmap);
1600 if (err)
1601 goto error;
1603 bitmap->chunkshift = ffz(~bitmap->chunksize);
1605 /* now that chunksize and chunkshift are set, we can use these macros */
1606 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1607 CHUNK_BLOCK_RATIO(bitmap);
1608 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1610 BUG_ON(!pages);
1612 bitmap->chunks = chunks;
1613 bitmap->pages = pages;
1614 bitmap->missing_pages = pages;
1615 bitmap->counter_bits = COUNTER_BITS;
1617 bitmap->syncchunk = ~0UL;
1619 #ifdef INJECT_FATAL_FAULT_1
1620 bitmap->bp = NULL;
1621 #else
1622 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1623 #endif
1624 err = -ENOMEM;
1625 if (!bitmap->bp)
1626 goto error;
1628 /* now that we have some pages available, initialize the in-memory
1629 * bitmap from the on-disk bitmap */
1630 start = 0;
1631 if (mddev->degraded == 0
1632 || bitmap->events_cleared == mddev->events)
1633 /* no need to keep dirty bits to optimise a re-add of a missing device */
1634 start = mddev->recovery_cp;
1635 err = bitmap_init_from_disk(bitmap, start);
1637 if (err)
1638 goto error;
1640 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1641 pages, bmname(bitmap));
1643 mddev->bitmap = bitmap;
1645 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1647 bitmap_update_sb(bitmap);
1649 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1651 error:
1652 bitmap_free(bitmap);
1653 return err;
1656 /* the bitmap API -- for raid personalities */
1657 EXPORT_SYMBOL(bitmap_startwrite);
1658 EXPORT_SYMBOL(bitmap_endwrite);
1659 EXPORT_SYMBOL(bitmap_start_sync);
1660 EXPORT_SYMBOL(bitmap_end_sync);
1661 EXPORT_SYMBOL(bitmap_unplug);
1662 EXPORT_SYMBOL(bitmap_close_sync);
1663 EXPORT_SYMBOL(bitmap_cond_end_sync);