Merge master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6/mini2440.git] / drivers / md / bitmap.c
blob5554adaa58f9be01d37bc5623f63384c084e69de
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
206 /* copy the pathname of a file to a buffer */
207 char *file_path(struct file *file, char *buf, int count)
209 struct dentry *d;
210 struct vfsmount *v;
212 if (!buf)
213 return NULL;
215 d = file->f_path.dentry;
216 v = file->f_path.mnt;
218 buf = d_path(d, v, buf, count);
220 return IS_ERR(buf) ? NULL : buf;
224 * basic page I/O operations
227 /* IO operations when bitmap is stored near all superblocks */
228 static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
230 /* choose a good rdev and read the page from there */
232 mdk_rdev_t *rdev;
233 struct list_head *tmp;
234 struct page *page = alloc_page(GFP_KERNEL);
235 sector_t target;
237 if (!page)
238 return ERR_PTR(-ENOMEM);
240 ITERATE_RDEV(mddev, rdev, tmp) {
241 if (! test_bit(In_sync, &rdev->flags)
242 || test_bit(Faulty, &rdev->flags))
243 continue;
245 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
247 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
248 page->index = index;
249 attach_page_buffers(page, NULL); /* so that free_buffer will
250 * quietly no-op */
251 return page;
254 return ERR_PTR(-EIO);
258 static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
260 mdk_rdev_t *rdev;
261 struct list_head *tmp;
263 ITERATE_RDEV(mddev, rdev, tmp)
264 if (test_bit(In_sync, &rdev->flags)
265 && !test_bit(Faulty, &rdev->flags))
266 md_super_write(mddev, rdev,
267 (rdev->sb_offset<<1) + offset
268 + page->index * (PAGE_SIZE/512),
269 PAGE_SIZE,
270 page);
272 if (wait)
273 md_super_wait(mddev);
274 return 0;
278 * write out a page to a file
280 static int write_page(struct bitmap *bitmap, struct page *page, int wait)
282 struct buffer_head *bh;
284 if (bitmap->file == NULL)
285 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
287 bh = page_buffers(page);
289 while (bh && bh->b_blocknr) {
290 atomic_inc(&bitmap->pending_writes);
291 set_buffer_locked(bh);
292 set_buffer_mapped(bh);
293 submit_bh(WRITE, bh);
294 bh = bh->b_this_page;
297 if (wait) {
298 wait_event(bitmap->write_wait,
299 atomic_read(&bitmap->pending_writes)==0);
300 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
302 return 0;
305 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
307 struct bitmap *bitmap = bh->b_private;
308 unsigned long flags;
310 if (!uptodate) {
311 spin_lock_irqsave(&bitmap->lock, flags);
312 bitmap->flags |= BITMAP_WRITE_ERROR;
313 spin_unlock_irqrestore(&bitmap->lock, flags);
315 if (atomic_dec_and_test(&bitmap->pending_writes))
316 wake_up(&bitmap->write_wait);
319 /* copied from buffer.c */
320 static void
321 __clear_page_buffers(struct page *page)
323 ClearPagePrivate(page);
324 set_page_private(page, 0);
325 page_cache_release(page);
327 static void free_buffers(struct page *page)
329 struct buffer_head *bh = page_buffers(page);
331 while (bh) {
332 struct buffer_head *next = bh->b_this_page;
333 free_buffer_head(bh);
334 bh = next;
336 __clear_page_buffers(page);
337 put_page(page);
340 /* read a page from a file.
341 * We both read the page, and attach buffers to the page to record the
342 * address of each block (using bmap). These addresses will be used
343 * to write the block later, completely bypassing the filesystem.
344 * This usage is similar to how swap files are handled, and allows us
345 * to write to a file with no concerns of memory allocation failing.
347 static struct page *read_page(struct file *file, unsigned long index,
348 struct bitmap *bitmap,
349 unsigned long count)
351 struct page *page = NULL;
352 struct inode *inode = file->f_path.dentry->d_inode;
353 struct buffer_head *bh;
354 sector_t block;
356 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
357 (unsigned long long)index << PAGE_SHIFT);
359 page = alloc_page(GFP_KERNEL);
360 if (!page)
361 page = ERR_PTR(-ENOMEM);
362 if (IS_ERR(page))
363 goto out;
365 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
366 if (!bh) {
367 put_page(page);
368 page = ERR_PTR(-ENOMEM);
369 goto out;
371 attach_page_buffers(page, bh);
372 block = index << (PAGE_SHIFT - inode->i_blkbits);
373 while (bh) {
374 if (count == 0)
375 bh->b_blocknr = 0;
376 else {
377 bh->b_blocknr = bmap(inode, block);
378 if (bh->b_blocknr == 0) {
379 /* Cannot use this file! */
380 free_buffers(page);
381 page = ERR_PTR(-EINVAL);
382 goto out;
384 bh->b_bdev = inode->i_sb->s_bdev;
385 if (count < (1<<inode->i_blkbits))
386 count = 0;
387 else
388 count -= (1<<inode->i_blkbits);
390 bh->b_end_io = end_bitmap_write;
391 bh->b_private = bitmap;
392 atomic_inc(&bitmap->pending_writes);
393 set_buffer_locked(bh);
394 set_buffer_mapped(bh);
395 submit_bh(READ, bh);
397 block++;
398 bh = bh->b_this_page;
400 page->index = index;
402 wait_event(bitmap->write_wait,
403 atomic_read(&bitmap->pending_writes)==0);
404 if (bitmap->flags & BITMAP_WRITE_ERROR) {
405 free_buffers(page);
406 page = ERR_PTR(-EIO);
408 out:
409 if (IS_ERR(page))
410 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
411 (int)PAGE_SIZE,
412 (unsigned long long)index << PAGE_SHIFT,
413 PTR_ERR(page));
414 return page;
418 * bitmap file superblock operations
421 /* update the event counter and sync the superblock to disk */
422 int bitmap_update_sb(struct bitmap *bitmap)
424 bitmap_super_t *sb;
425 unsigned long flags;
427 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
428 return 0;
429 spin_lock_irqsave(&bitmap->lock, flags);
430 if (!bitmap->sb_page) { /* no superblock */
431 spin_unlock_irqrestore(&bitmap->lock, flags);
432 return 0;
434 spin_unlock_irqrestore(&bitmap->lock, flags);
435 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
436 sb->events = cpu_to_le64(bitmap->mddev->events);
437 if (!bitmap->mddev->degraded)
438 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
439 kunmap_atomic(sb, KM_USER0);
440 return write_page(bitmap, bitmap->sb_page, 1);
443 /* print out the bitmap file superblock */
444 void bitmap_print_sb(struct bitmap *bitmap)
446 bitmap_super_t *sb;
448 if (!bitmap || !bitmap->sb_page)
449 return;
450 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
451 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
452 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
453 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
454 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
455 *(__u32 *)(sb->uuid+0),
456 *(__u32 *)(sb->uuid+4),
457 *(__u32 *)(sb->uuid+8),
458 *(__u32 *)(sb->uuid+12));
459 printk(KERN_DEBUG " events: %llu\n",
460 (unsigned long long) le64_to_cpu(sb->events));
461 printk(KERN_DEBUG "events cleared: %llu\n",
462 (unsigned long long) le64_to_cpu(sb->events_cleared));
463 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
464 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
465 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
466 printk(KERN_DEBUG " sync size: %llu KB\n",
467 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
468 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
469 kunmap_atomic(sb, KM_USER0);
472 /* read the superblock from the bitmap file and initialize some bitmap fields */
473 static int bitmap_read_sb(struct bitmap *bitmap)
475 char *reason = NULL;
476 bitmap_super_t *sb;
477 unsigned long chunksize, daemon_sleep, write_behind;
478 unsigned long long events;
479 int err = -EINVAL;
481 /* page 0 is the superblock, read it... */
482 if (bitmap->file) {
483 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
484 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
486 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
487 } else {
488 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
490 if (IS_ERR(bitmap->sb_page)) {
491 err = PTR_ERR(bitmap->sb_page);
492 bitmap->sb_page = NULL;
493 return err;
496 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
498 chunksize = le32_to_cpu(sb->chunksize);
499 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
500 write_behind = le32_to_cpu(sb->write_behind);
502 /* verify that the bitmap-specific fields are valid */
503 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
504 reason = "bad magic";
505 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
506 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
507 reason = "unrecognized superblock version";
508 else if (chunksize < PAGE_SIZE)
509 reason = "bitmap chunksize too small";
510 else if ((1 << ffz(~chunksize)) != chunksize)
511 reason = "bitmap chunksize not a power of 2";
512 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
513 reason = "daemon sleep period out of range";
514 else if (write_behind > COUNTER_MAX)
515 reason = "write-behind limit out of range (0 - 16383)";
516 if (reason) {
517 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
518 bmname(bitmap), reason);
519 goto out;
522 /* keep the array size field of the bitmap superblock up to date */
523 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
525 if (!bitmap->mddev->persistent)
526 goto success;
529 * if we have a persistent array superblock, compare the
530 * bitmap's UUID and event counter to the mddev's
532 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
533 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
534 bmname(bitmap));
535 goto out;
537 events = le64_to_cpu(sb->events);
538 if (events < bitmap->mddev->events) {
539 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
540 "-- forcing full recovery\n", bmname(bitmap), events,
541 (unsigned long long) bitmap->mddev->events);
542 sb->state |= cpu_to_le32(BITMAP_STALE);
544 success:
545 /* assign fields using values from superblock */
546 bitmap->chunksize = chunksize;
547 bitmap->daemon_sleep = daemon_sleep;
548 bitmap->daemon_lastrun = jiffies;
549 bitmap->max_write_behind = write_behind;
550 bitmap->flags |= le32_to_cpu(sb->state);
551 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
552 bitmap->flags |= BITMAP_HOSTENDIAN;
553 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
554 if (sb->state & cpu_to_le32(BITMAP_STALE))
555 bitmap->events_cleared = bitmap->mddev->events;
556 err = 0;
557 out:
558 kunmap_atomic(sb, KM_USER0);
559 if (err)
560 bitmap_print_sb(bitmap);
561 return err;
564 enum bitmap_mask_op {
565 MASK_SET,
566 MASK_UNSET
569 /* record the state of the bitmap in the superblock */
570 static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
571 enum bitmap_mask_op op)
573 bitmap_super_t *sb;
574 unsigned long flags;
576 spin_lock_irqsave(&bitmap->lock, flags);
577 if (!bitmap->sb_page) { /* can't set the state */
578 spin_unlock_irqrestore(&bitmap->lock, flags);
579 return;
581 spin_unlock_irqrestore(&bitmap->lock, flags);
582 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
583 switch (op) {
584 case MASK_SET: sb->state |= cpu_to_le32(bits);
585 break;
586 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
587 break;
588 default: BUG();
590 kunmap_atomic(sb, KM_USER0);
594 * general bitmap file operations
597 /* calculate the index of the page that contains this bit */
598 static inline unsigned long file_page_index(unsigned long chunk)
600 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
603 /* calculate the (bit) offset of this bit within a page */
604 static inline unsigned long file_page_offset(unsigned long chunk)
606 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
610 * return a pointer to the page in the filemap that contains the given bit
612 * this lookup is complicated by the fact that the bitmap sb might be exactly
613 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
614 * 0 or page 1
616 static inline struct page *filemap_get_page(struct bitmap *bitmap,
617 unsigned long chunk)
619 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
620 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
624 static void bitmap_file_unmap(struct bitmap *bitmap)
626 struct page **map, *sb_page;
627 unsigned long *attr;
628 int pages;
629 unsigned long flags;
631 spin_lock_irqsave(&bitmap->lock, flags);
632 map = bitmap->filemap;
633 bitmap->filemap = NULL;
634 attr = bitmap->filemap_attr;
635 bitmap->filemap_attr = NULL;
636 pages = bitmap->file_pages;
637 bitmap->file_pages = 0;
638 sb_page = bitmap->sb_page;
639 bitmap->sb_page = NULL;
640 spin_unlock_irqrestore(&bitmap->lock, flags);
642 while (pages--)
643 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
644 free_buffers(map[pages]);
645 kfree(map);
646 kfree(attr);
648 if (sb_page)
649 free_buffers(sb_page);
652 static void bitmap_file_put(struct bitmap *bitmap)
654 struct file *file;
655 unsigned long flags;
657 spin_lock_irqsave(&bitmap->lock, flags);
658 file = bitmap->file;
659 bitmap->file = NULL;
660 spin_unlock_irqrestore(&bitmap->lock, flags);
662 if (file)
663 wait_event(bitmap->write_wait,
664 atomic_read(&bitmap->pending_writes)==0);
665 bitmap_file_unmap(bitmap);
667 if (file) {
668 struct inode *inode = file->f_path.dentry->d_inode;
669 invalidate_mapping_pages(inode->i_mapping, 0, -1);
670 fput(file);
676 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
677 * then it is no longer reliable, so we stop using it and we mark the file
678 * as failed in the superblock
680 static void bitmap_file_kick(struct bitmap *bitmap)
682 char *path, *ptr = NULL;
684 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
685 bitmap_update_sb(bitmap);
687 if (bitmap->file) {
688 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
689 if (path)
690 ptr = file_path(bitmap->file, path, PAGE_SIZE);
692 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
693 bmname(bitmap), ptr ? ptr : "");
695 kfree(path);
698 bitmap_file_put(bitmap);
700 return;
703 enum bitmap_page_attr {
704 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
705 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
706 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
709 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
710 enum bitmap_page_attr attr)
712 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
715 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
716 enum bitmap_page_attr attr)
718 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
721 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
722 enum bitmap_page_attr attr)
724 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
728 * bitmap_file_set_bit -- called before performing a write to the md device
729 * to set (and eventually sync) a particular bit in the bitmap file
731 * we set the bit immediately, then we record the page number so that
732 * when an unplug occurs, we can flush the dirty pages out to disk
734 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
736 unsigned long bit;
737 struct page *page;
738 void *kaddr;
739 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
741 if (!bitmap->filemap) {
742 return;
745 page = filemap_get_page(bitmap, chunk);
746 if (!page) return;
747 bit = file_page_offset(chunk);
749 /* set the bit */
750 kaddr = kmap_atomic(page, KM_USER0);
751 if (bitmap->flags & BITMAP_HOSTENDIAN)
752 set_bit(bit, kaddr);
753 else
754 ext2_set_bit(bit, kaddr);
755 kunmap_atomic(kaddr, KM_USER0);
756 PRINTK("set file bit %lu page %lu\n", bit, page->index);
758 /* record page number so it gets flushed to disk when unplug occurs */
759 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
763 /* this gets called when the md device is ready to unplug its underlying
764 * (slave) device queues -- before we let any writes go down, we need to
765 * sync the dirty pages of the bitmap file to disk */
766 int bitmap_unplug(struct bitmap *bitmap)
768 unsigned long i, flags;
769 int dirty, need_write;
770 struct page *page;
771 int wait = 0;
772 int err;
774 if (!bitmap)
775 return 0;
777 /* look at each page to see if there are any set bits that need to be
778 * flushed out to disk */
779 for (i = 0; i < bitmap->file_pages; i++) {
780 spin_lock_irqsave(&bitmap->lock, flags);
781 if (!bitmap->filemap) {
782 spin_unlock_irqrestore(&bitmap->lock, flags);
783 return 0;
785 page = bitmap->filemap[i];
786 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
787 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
788 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
789 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
790 if (dirty)
791 wait = 1;
792 spin_unlock_irqrestore(&bitmap->lock, flags);
794 if (dirty | need_write)
795 err = write_page(bitmap, page, 0);
797 if (wait) { /* if any writes were performed, we need to wait on them */
798 if (bitmap->file)
799 wait_event(bitmap->write_wait,
800 atomic_read(&bitmap->pending_writes)==0);
801 else
802 md_super_wait(bitmap->mddev);
804 if (bitmap->flags & BITMAP_WRITE_ERROR)
805 bitmap_file_kick(bitmap);
806 return 0;
809 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
810 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
811 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
812 * memory mapping of the bitmap file
813 * Special cases:
814 * if there's no bitmap file, or if the bitmap file had been
815 * previously kicked from the array, we mark all the bits as
816 * 1's in order to cause a full resync.
818 * We ignore all bits for sectors that end earlier than 'start'.
819 * This is used when reading an out-of-date bitmap...
821 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
823 unsigned long i, chunks, index, oldindex, bit;
824 struct page *page = NULL, *oldpage = NULL;
825 unsigned long num_pages, bit_cnt = 0;
826 struct file *file;
827 unsigned long bytes, offset;
828 int outofdate;
829 int ret = -ENOSPC;
830 void *paddr;
832 chunks = bitmap->chunks;
833 file = bitmap->file;
835 BUG_ON(!file && !bitmap->offset);
837 #ifdef INJECT_FAULTS_3
838 outofdate = 1;
839 #else
840 outofdate = bitmap->flags & BITMAP_STALE;
841 #endif
842 if (outofdate)
843 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
844 "recovery\n", bmname(bitmap));
846 bytes = (chunks + 7) / 8;
848 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
850 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
851 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
852 bmname(bitmap),
853 (unsigned long) i_size_read(file->f_mapping->host),
854 bytes + sizeof(bitmap_super_t));
855 goto out;
858 ret = -ENOMEM;
860 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
861 if (!bitmap->filemap)
862 goto out;
864 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
865 bitmap->filemap_attr = kzalloc(
866 (((num_pages*4/8)+sizeof(unsigned long)-1)
867 /sizeof(unsigned long))
868 *sizeof(unsigned long),
869 GFP_KERNEL);
870 if (!bitmap->filemap_attr)
871 goto out;
873 oldindex = ~0L;
875 for (i = 0; i < chunks; i++) {
876 int b;
877 index = file_page_index(i);
878 bit = file_page_offset(i);
879 if (index != oldindex) { /* this is a new page, read it in */
880 int count;
881 /* unmap the old page, we're done with it */
882 if (index == num_pages-1)
883 count = bytes + sizeof(bitmap_super_t)
884 - index * PAGE_SIZE;
885 else
886 count = PAGE_SIZE;
887 if (index == 0) {
889 * if we're here then the superblock page
890 * contains some bits (PAGE_SIZE != sizeof sb)
891 * we've already read it in, so just use it
893 page = bitmap->sb_page;
894 offset = sizeof(bitmap_super_t);
895 } else if (file) {
896 page = read_page(file, index, bitmap, count);
897 offset = 0;
898 } else {
899 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
900 offset = 0;
902 if (IS_ERR(page)) { /* read error */
903 ret = PTR_ERR(page);
904 goto out;
907 oldindex = index;
908 oldpage = page;
910 if (outofdate) {
912 * if bitmap is out of date, dirty the
913 * whole page and write it out
915 paddr = kmap_atomic(page, KM_USER0);
916 memset(paddr + offset, 0xff,
917 PAGE_SIZE - offset);
918 kunmap_atomic(paddr, KM_USER0);
919 ret = write_page(bitmap, page, 1);
920 if (ret) {
921 /* release, page not in filemap yet */
922 put_page(page);
923 goto out;
927 bitmap->filemap[bitmap->file_pages++] = page;
929 paddr = kmap_atomic(page, KM_USER0);
930 if (bitmap->flags & BITMAP_HOSTENDIAN)
931 b = test_bit(bit, paddr);
932 else
933 b = ext2_test_bit(bit, paddr);
934 kunmap_atomic(paddr, KM_USER0);
935 if (b) {
936 /* if the disk bit is set, set the memory bit */
937 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
938 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
940 bit_cnt++;
941 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
945 /* everything went OK */
946 ret = 0;
947 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
949 if (bit_cnt) { /* Kick recovery if any bits were set */
950 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
951 md_wakeup_thread(bitmap->mddev->thread);
954 out:
955 printk(KERN_INFO "%s: bitmap initialized from disk: "
956 "read %lu/%lu pages, set %lu bits, status: %d\n",
957 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
959 return ret;
962 void bitmap_write_all(struct bitmap *bitmap)
964 /* We don't actually write all bitmap blocks here,
965 * just flag them as needing to be written
967 int i;
969 for (i=0; i < bitmap->file_pages; i++)
970 set_page_attr(bitmap, bitmap->filemap[i],
971 BITMAP_PAGE_NEEDWRITE);
975 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
977 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
978 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
979 bitmap->bp[page].count += inc;
981 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
982 (unsigned long long)offset, inc, bitmap->bp[page].count);
984 bitmap_checkfree(bitmap, page);
986 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
987 sector_t offset, int *blocks,
988 int create);
991 * bitmap daemon -- periodically wakes up to clean bits and flush pages
992 * out to disk
995 int bitmap_daemon_work(struct bitmap *bitmap)
997 unsigned long j;
998 unsigned long flags;
999 struct page *page = NULL, *lastpage = NULL;
1000 int err = 0;
1001 int blocks;
1002 void *paddr;
1004 if (bitmap == NULL)
1005 return 0;
1006 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1007 return 0;
1008 bitmap->daemon_lastrun = jiffies;
1010 for (j = 0; j < bitmap->chunks; j++) {
1011 bitmap_counter_t *bmc;
1012 spin_lock_irqsave(&bitmap->lock, flags);
1013 if (!bitmap->filemap) {
1014 /* error or shutdown */
1015 spin_unlock_irqrestore(&bitmap->lock, flags);
1016 break;
1019 page = filemap_get_page(bitmap, j);
1021 if (page != lastpage) {
1022 /* skip this page unless it's marked as needing cleaning */
1023 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1024 int need_write = test_page_attr(bitmap, page,
1025 BITMAP_PAGE_NEEDWRITE);
1026 if (need_write)
1027 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1029 spin_unlock_irqrestore(&bitmap->lock, flags);
1030 if (need_write) {
1031 switch (write_page(bitmap, page, 0)) {
1032 case 0:
1033 break;
1034 default:
1035 bitmap_file_kick(bitmap);
1038 continue;
1041 /* grab the new page, sync and release the old */
1042 if (lastpage != NULL) {
1043 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1044 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1045 spin_unlock_irqrestore(&bitmap->lock, flags);
1046 err = write_page(bitmap, lastpage, 0);
1047 } else {
1048 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1049 spin_unlock_irqrestore(&bitmap->lock, flags);
1051 if (err)
1052 bitmap_file_kick(bitmap);
1053 } else
1054 spin_unlock_irqrestore(&bitmap->lock, flags);
1055 lastpage = page;
1057 printk("bitmap clean at page %lu\n", j);
1059 spin_lock_irqsave(&bitmap->lock, flags);
1060 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1062 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1063 &blocks, 0);
1064 if (bmc) {
1066 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1068 if (*bmc == 2) {
1069 *bmc=1; /* maybe clear the bit next time */
1070 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1071 } else if (*bmc == 1) {
1072 /* we can clear the bit */
1073 *bmc = 0;
1074 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1075 -1);
1077 /* clear the bit */
1078 paddr = kmap_atomic(page, KM_USER0);
1079 if (bitmap->flags & BITMAP_HOSTENDIAN)
1080 clear_bit(file_page_offset(j), paddr);
1081 else
1082 ext2_clear_bit(file_page_offset(j), paddr);
1083 kunmap_atomic(paddr, KM_USER0);
1086 spin_unlock_irqrestore(&bitmap->lock, flags);
1089 /* now sync the final page */
1090 if (lastpage != NULL) {
1091 spin_lock_irqsave(&bitmap->lock, flags);
1092 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1093 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1094 spin_unlock_irqrestore(&bitmap->lock, flags);
1095 err = write_page(bitmap, lastpage, 0);
1096 } else {
1097 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1098 spin_unlock_irqrestore(&bitmap->lock, flags);
1102 return err;
1105 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1106 sector_t offset, int *blocks,
1107 int create)
1109 /* If 'create', we might release the lock and reclaim it.
1110 * The lock must have been taken with interrupts enabled.
1111 * If !create, we don't release the lock.
1113 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1114 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1115 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1116 sector_t csize;
1118 if (bitmap_checkpage(bitmap, page, create) < 0) {
1119 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1120 *blocks = csize - (offset & (csize- 1));
1121 return NULL;
1123 /* now locked ... */
1125 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1126 /* should we use the first or second counter field
1127 * of the hijacked pointer? */
1128 int hi = (pageoff > PAGE_COUNTER_MASK);
1129 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1130 PAGE_COUNTER_SHIFT - 1);
1131 *blocks = csize - (offset & (csize- 1));
1132 return &((bitmap_counter_t *)
1133 &bitmap->bp[page].map)[hi];
1134 } else { /* page is allocated */
1135 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1136 *blocks = csize - (offset & (csize- 1));
1137 return (bitmap_counter_t *)
1138 &(bitmap->bp[page].map[pageoff]);
1142 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1144 if (!bitmap) return 0;
1146 if (behind) {
1147 atomic_inc(&bitmap->behind_writes);
1148 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1149 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1152 while (sectors) {
1153 int blocks;
1154 bitmap_counter_t *bmc;
1156 spin_lock_irq(&bitmap->lock);
1157 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1158 if (!bmc) {
1159 spin_unlock_irq(&bitmap->lock);
1160 return 0;
1163 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1164 DEFINE_WAIT(__wait);
1165 /* note that it is safe to do the prepare_to_wait
1166 * after the test as long as we do it before dropping
1167 * the spinlock.
1169 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1170 TASK_UNINTERRUPTIBLE);
1171 spin_unlock_irq(&bitmap->lock);
1172 bitmap->mddev->queue
1173 ->unplug_fn(bitmap->mddev->queue);
1174 schedule();
1175 finish_wait(&bitmap->overflow_wait, &__wait);
1176 continue;
1179 switch(*bmc) {
1180 case 0:
1181 bitmap_file_set_bit(bitmap, offset);
1182 bitmap_count_page(bitmap,offset, 1);
1183 blk_plug_device(bitmap->mddev->queue);
1184 /* fall through */
1185 case 1:
1186 *bmc = 2;
1189 (*bmc)++;
1191 spin_unlock_irq(&bitmap->lock);
1193 offset += blocks;
1194 if (sectors > blocks)
1195 sectors -= blocks;
1196 else sectors = 0;
1198 return 0;
1201 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1202 int success, int behind)
1204 if (!bitmap) return;
1205 if (behind) {
1206 atomic_dec(&bitmap->behind_writes);
1207 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1208 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1211 while (sectors) {
1212 int blocks;
1213 unsigned long flags;
1214 bitmap_counter_t *bmc;
1216 spin_lock_irqsave(&bitmap->lock, flags);
1217 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1218 if (!bmc) {
1219 spin_unlock_irqrestore(&bitmap->lock, flags);
1220 return;
1223 if (!success && ! (*bmc & NEEDED_MASK))
1224 *bmc |= NEEDED_MASK;
1226 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1227 wake_up(&bitmap->overflow_wait);
1229 (*bmc)--;
1230 if (*bmc <= 2) {
1231 set_page_attr(bitmap,
1232 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1233 BITMAP_PAGE_CLEAN);
1235 spin_unlock_irqrestore(&bitmap->lock, flags);
1236 offset += blocks;
1237 if (sectors > blocks)
1238 sectors -= blocks;
1239 else sectors = 0;
1243 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1244 int degraded)
1246 bitmap_counter_t *bmc;
1247 int rv;
1248 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1249 *blocks = 1024;
1250 return 1; /* always resync if no bitmap */
1252 spin_lock_irq(&bitmap->lock);
1253 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1254 rv = 0;
1255 if (bmc) {
1256 /* locked */
1257 if (RESYNC(*bmc))
1258 rv = 1;
1259 else if (NEEDED(*bmc)) {
1260 rv = 1;
1261 if (!degraded) { /* don't set/clear bits if degraded */
1262 *bmc |= RESYNC_MASK;
1263 *bmc &= ~NEEDED_MASK;
1267 spin_unlock_irq(&bitmap->lock);
1268 return rv;
1271 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1273 bitmap_counter_t *bmc;
1274 unsigned long flags;
1276 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1277 */ if (bitmap == NULL) {
1278 *blocks = 1024;
1279 return;
1281 spin_lock_irqsave(&bitmap->lock, flags);
1282 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1283 if (bmc == NULL)
1284 goto unlock;
1285 /* locked */
1287 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1289 if (RESYNC(*bmc)) {
1290 *bmc &= ~RESYNC_MASK;
1292 if (!NEEDED(*bmc) && aborted)
1293 *bmc |= NEEDED_MASK;
1294 else {
1295 if (*bmc <= 2) {
1296 set_page_attr(bitmap,
1297 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1298 BITMAP_PAGE_CLEAN);
1302 unlock:
1303 spin_unlock_irqrestore(&bitmap->lock, flags);
1306 void bitmap_close_sync(struct bitmap *bitmap)
1308 /* Sync has finished, and any bitmap chunks that weren't synced
1309 * properly have been aborted. It remains to us to clear the
1310 * RESYNC bit wherever it is still on
1312 sector_t sector = 0;
1313 int blocks;
1314 if (!bitmap) return;
1315 while (sector < bitmap->mddev->resync_max_sectors) {
1316 bitmap_end_sync(bitmap, sector, &blocks, 0);
1318 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1319 (unsigned long long)sector, blocks);
1320 */ sector += blocks;
1324 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1326 /* For each chunk covered by any of these sectors, set the
1327 * counter to 1 and set resync_needed. They should all
1328 * be 0 at this point
1331 int secs;
1332 bitmap_counter_t *bmc;
1333 spin_lock_irq(&bitmap->lock);
1334 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1335 if (!bmc) {
1336 spin_unlock_irq(&bitmap->lock);
1337 return;
1339 if (! *bmc) {
1340 struct page *page;
1341 *bmc = 1 | (needed?NEEDED_MASK:0);
1342 bitmap_count_page(bitmap, offset, 1);
1343 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1344 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1346 spin_unlock_irq(&bitmap->lock);
1350 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1351 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1353 unsigned long chunk;
1355 for (chunk = s; chunk <= e; chunk++) {
1356 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1357 bitmap_set_memory_bits(bitmap, sec, 1);
1358 bitmap_file_set_bit(bitmap, sec);
1363 * flush out any pending updates
1365 void bitmap_flush(mddev_t *mddev)
1367 struct bitmap *bitmap = mddev->bitmap;
1368 int sleep;
1370 if (!bitmap) /* there was no bitmap */
1371 return;
1373 /* run the daemon_work three time to ensure everything is flushed
1374 * that can be
1376 sleep = bitmap->daemon_sleep;
1377 bitmap->daemon_sleep = 0;
1378 bitmap_daemon_work(bitmap);
1379 bitmap_daemon_work(bitmap);
1380 bitmap_daemon_work(bitmap);
1381 bitmap->daemon_sleep = sleep;
1382 bitmap_update_sb(bitmap);
1386 * free memory that was allocated
1388 static void bitmap_free(struct bitmap *bitmap)
1390 unsigned long k, pages;
1391 struct bitmap_page *bp;
1393 if (!bitmap) /* there was no bitmap */
1394 return;
1396 /* release the bitmap file and kill the daemon */
1397 bitmap_file_put(bitmap);
1399 bp = bitmap->bp;
1400 pages = bitmap->pages;
1402 /* free all allocated memory */
1404 if (bp) /* deallocate the page memory */
1405 for (k = 0; k < pages; k++)
1406 if (bp[k].map && !bp[k].hijacked)
1407 kfree(bp[k].map);
1408 kfree(bp);
1409 kfree(bitmap);
1411 void bitmap_destroy(mddev_t *mddev)
1413 struct bitmap *bitmap = mddev->bitmap;
1415 if (!bitmap) /* there was no bitmap */
1416 return;
1418 mddev->bitmap = NULL; /* disconnect from the md device */
1419 if (mddev->thread)
1420 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1422 bitmap_free(bitmap);
1426 * initialize the bitmap structure
1427 * if this returns an error, bitmap_destroy must be called to do clean up
1429 int bitmap_create(mddev_t *mddev)
1431 struct bitmap *bitmap;
1432 unsigned long blocks = mddev->resync_max_sectors;
1433 unsigned long chunks;
1434 unsigned long pages;
1435 struct file *file = mddev->bitmap_file;
1436 int err;
1437 sector_t start;
1439 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1441 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1442 return 0;
1444 BUG_ON(file && mddev->bitmap_offset);
1446 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1447 if (!bitmap)
1448 return -ENOMEM;
1450 spin_lock_init(&bitmap->lock);
1451 atomic_set(&bitmap->pending_writes, 0);
1452 init_waitqueue_head(&bitmap->write_wait);
1453 init_waitqueue_head(&bitmap->overflow_wait);
1455 bitmap->mddev = mddev;
1457 bitmap->file = file;
1458 bitmap->offset = mddev->bitmap_offset;
1459 if (file) {
1460 get_file(file);
1461 do_sync_file_range(file, 0, LLONG_MAX,
1462 SYNC_FILE_RANGE_WAIT_BEFORE |
1463 SYNC_FILE_RANGE_WRITE |
1464 SYNC_FILE_RANGE_WAIT_AFTER);
1466 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1467 err = bitmap_read_sb(bitmap);
1468 if (err)
1469 goto error;
1471 bitmap->chunkshift = ffz(~bitmap->chunksize);
1473 /* now that chunksize and chunkshift are set, we can use these macros */
1474 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1475 CHUNK_BLOCK_RATIO(bitmap);
1476 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1478 BUG_ON(!pages);
1480 bitmap->chunks = chunks;
1481 bitmap->pages = pages;
1482 bitmap->missing_pages = pages;
1483 bitmap->counter_bits = COUNTER_BITS;
1485 bitmap->syncchunk = ~0UL;
1487 #ifdef INJECT_FATAL_FAULT_1
1488 bitmap->bp = NULL;
1489 #else
1490 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1491 #endif
1492 err = -ENOMEM;
1493 if (!bitmap->bp)
1494 goto error;
1496 /* now that we have some pages available, initialize the in-memory
1497 * bitmap from the on-disk bitmap */
1498 start = 0;
1499 if (mddev->degraded == 0
1500 || bitmap->events_cleared == mddev->events)
1501 /* no need to keep dirty bits to optimise a re-add of a missing device */
1502 start = mddev->recovery_cp;
1503 err = bitmap_init_from_disk(bitmap, start);
1505 if (err)
1506 goto error;
1508 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1509 pages, bmname(bitmap));
1511 mddev->bitmap = bitmap;
1513 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1515 return bitmap_update_sb(bitmap);
1517 error:
1518 bitmap_free(bitmap);
1519 return err;
1522 /* the bitmap API -- for raid personalities */
1523 EXPORT_SYMBOL(bitmap_startwrite);
1524 EXPORT_SYMBOL(bitmap_endwrite);
1525 EXPORT_SYMBOL(bitmap_start_sync);
1526 EXPORT_SYMBOL(bitmap_end_sync);
1527 EXPORT_SYMBOL(bitmap_unplug);
1528 EXPORT_SYMBOL(bitmap_close_sync);