kconfig: set all new symbols automatically
[linux-2.6/x86.git] / drivers / md / bitmap.c
blob621a272a2c7495b9cd8866ec6bb6d0454d5d1396
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, unsigned long index)
213 /* choose a good rdev and read the page from there */
215 mdk_rdev_t *rdev;
216 struct list_head *tmp;
217 struct page *page = alloc_page(GFP_KERNEL);
218 sector_t target;
220 if (!page)
221 return ERR_PTR(-ENOMEM);
223 rdev_for_each(rdev, tmp, mddev) {
224 if (! test_bit(In_sync, &rdev->flags)
225 || test_bit(Faulty, &rdev->flags))
226 continue;
228 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
230 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
231 page->index = index;
232 attach_page_buffers(page, NULL); /* so that free_buffer will
233 * quietly no-op */
234 return page;
237 return ERR_PTR(-EIO);
241 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
243 mdk_rdev_t *rdev;
244 mddev_t *mddev = bitmap->mddev;
246 rcu_read_lock();
247 rdev_for_each_rcu(rdev, mddev)
248 if (test_bit(In_sync, &rdev->flags)
249 && !test_bit(Faulty, &rdev->flags)) {
250 int size = PAGE_SIZE;
251 if (page->index == bitmap->file_pages-1)
252 size = roundup(bitmap->last_page_size,
253 bdev_hardsect_size(rdev->bdev));
254 /* Just make sure we aren't corrupting data or
255 * metadata
257 if (bitmap->offset < 0) {
258 /* DATA BITMAP METADATA */
259 if (bitmap->offset
260 + (long)(page->index * (PAGE_SIZE/512))
261 + size/512 > 0)
262 /* bitmap runs in to metadata */
263 goto bad_alignment;
264 if (rdev->data_offset + mddev->size*2
265 > rdev->sb_start + bitmap->offset)
266 /* data runs in to bitmap */
267 goto bad_alignment;
268 } else if (rdev->sb_start < rdev->data_offset) {
269 /* METADATA BITMAP DATA */
270 if (rdev->sb_start
271 + bitmap->offset
272 + page->index*(PAGE_SIZE/512) + size/512
273 > rdev->data_offset)
274 /* bitmap runs in to data */
275 goto bad_alignment;
276 } else {
277 /* DATA METADATA BITMAP - no problems */
279 md_super_write(mddev, rdev,
280 rdev->sb_start + bitmap->offset
281 + page->index * (PAGE_SIZE/512),
282 size,
283 page);
285 rcu_read_unlock();
287 if (wait)
288 md_super_wait(mddev);
289 return 0;
291 bad_alignment:
292 rcu_read_unlock();
293 return -EINVAL;
296 static void bitmap_file_kick(struct bitmap *bitmap);
298 * write out a page to a file
300 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
302 struct buffer_head *bh;
304 if (bitmap->file == NULL) {
305 switch (write_sb_page(bitmap, page, wait)) {
306 case -EINVAL:
307 bitmap->flags |= BITMAP_WRITE_ERROR;
309 } else {
311 bh = page_buffers(page);
313 while (bh && bh->b_blocknr) {
314 atomic_inc(&bitmap->pending_writes);
315 set_buffer_locked(bh);
316 set_buffer_mapped(bh);
317 submit_bh(WRITE, bh);
318 bh = bh->b_this_page;
321 if (wait) {
322 wait_event(bitmap->write_wait,
323 atomic_read(&bitmap->pending_writes)==0);
326 if (bitmap->flags & BITMAP_WRITE_ERROR)
327 bitmap_file_kick(bitmap);
330 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
332 struct bitmap *bitmap = bh->b_private;
333 unsigned long flags;
335 if (!uptodate) {
336 spin_lock_irqsave(&bitmap->lock, flags);
337 bitmap->flags |= BITMAP_WRITE_ERROR;
338 spin_unlock_irqrestore(&bitmap->lock, flags);
340 if (atomic_dec_and_test(&bitmap->pending_writes))
341 wake_up(&bitmap->write_wait);
344 /* copied from buffer.c */
345 static void
346 __clear_page_buffers(struct page *page)
348 ClearPagePrivate(page);
349 set_page_private(page, 0);
350 page_cache_release(page);
352 static void free_buffers(struct page *page)
354 struct buffer_head *bh = page_buffers(page);
356 while (bh) {
357 struct buffer_head *next = bh->b_this_page;
358 free_buffer_head(bh);
359 bh = next;
361 __clear_page_buffers(page);
362 put_page(page);
365 /* read a page from a file.
366 * We both read the page, and attach buffers to the page to record the
367 * address of each block (using bmap). These addresses will be used
368 * to write the block later, completely bypassing the filesystem.
369 * This usage is similar to how swap files are handled, and allows us
370 * to write to a file with no concerns of memory allocation failing.
372 static struct page *read_page(struct file *file, unsigned long index,
373 struct bitmap *bitmap,
374 unsigned long count)
376 struct page *page = NULL;
377 struct inode *inode = file->f_path.dentry->d_inode;
378 struct buffer_head *bh;
379 sector_t block;
381 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
382 (unsigned long long)index << PAGE_SHIFT);
384 page = alloc_page(GFP_KERNEL);
385 if (!page)
386 page = ERR_PTR(-ENOMEM);
387 if (IS_ERR(page))
388 goto out;
390 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
391 if (!bh) {
392 put_page(page);
393 page = ERR_PTR(-ENOMEM);
394 goto out;
396 attach_page_buffers(page, bh);
397 block = index << (PAGE_SHIFT - inode->i_blkbits);
398 while (bh) {
399 if (count == 0)
400 bh->b_blocknr = 0;
401 else {
402 bh->b_blocknr = bmap(inode, block);
403 if (bh->b_blocknr == 0) {
404 /* Cannot use this file! */
405 free_buffers(page);
406 page = ERR_PTR(-EINVAL);
407 goto out;
409 bh->b_bdev = inode->i_sb->s_bdev;
410 if (count < (1<<inode->i_blkbits))
411 count = 0;
412 else
413 count -= (1<<inode->i_blkbits);
415 bh->b_end_io = end_bitmap_write;
416 bh->b_private = bitmap;
417 atomic_inc(&bitmap->pending_writes);
418 set_buffer_locked(bh);
419 set_buffer_mapped(bh);
420 submit_bh(READ, bh);
422 block++;
423 bh = bh->b_this_page;
425 page->index = index;
427 wait_event(bitmap->write_wait,
428 atomic_read(&bitmap->pending_writes)==0);
429 if (bitmap->flags & BITMAP_WRITE_ERROR) {
430 free_buffers(page);
431 page = ERR_PTR(-EIO);
433 out:
434 if (IS_ERR(page))
435 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
436 (int)PAGE_SIZE,
437 (unsigned long long)index << PAGE_SHIFT,
438 PTR_ERR(page));
439 return page;
443 * bitmap file superblock operations
446 /* update the event counter and sync the superblock to disk */
447 void bitmap_update_sb(struct bitmap *bitmap)
449 bitmap_super_t *sb;
450 unsigned long flags;
452 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
453 return;
454 spin_lock_irqsave(&bitmap->lock, flags);
455 if (!bitmap->sb_page) { /* no superblock */
456 spin_unlock_irqrestore(&bitmap->lock, flags);
457 return;
459 spin_unlock_irqrestore(&bitmap->lock, flags);
460 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
461 sb->events = cpu_to_le64(bitmap->mddev->events);
462 if (bitmap->mddev->events < bitmap->events_cleared) {
463 /* rocking back to read-only */
464 bitmap->events_cleared = bitmap->mddev->events;
465 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
467 kunmap_atomic(sb, KM_USER0);
468 write_page(bitmap, bitmap->sb_page, 1);
471 /* print out the bitmap file superblock */
472 void bitmap_print_sb(struct bitmap *bitmap)
474 bitmap_super_t *sb;
476 if (!bitmap || !bitmap->sb_page)
477 return;
478 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
479 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
480 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
481 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
482 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
483 *(__u32 *)(sb->uuid+0),
484 *(__u32 *)(sb->uuid+4),
485 *(__u32 *)(sb->uuid+8),
486 *(__u32 *)(sb->uuid+12));
487 printk(KERN_DEBUG " events: %llu\n",
488 (unsigned long long) le64_to_cpu(sb->events));
489 printk(KERN_DEBUG "events cleared: %llu\n",
490 (unsigned long long) le64_to_cpu(sb->events_cleared));
491 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
492 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
493 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
494 printk(KERN_DEBUG " sync size: %llu KB\n",
495 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
496 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
497 kunmap_atomic(sb, KM_USER0);
500 /* read the superblock from the bitmap file and initialize some bitmap fields */
501 static int bitmap_read_sb(struct bitmap *bitmap)
503 char *reason = NULL;
504 bitmap_super_t *sb;
505 unsigned long chunksize, daemon_sleep, write_behind;
506 unsigned long long events;
507 int err = -EINVAL;
509 /* page 0 is the superblock, read it... */
510 if (bitmap->file) {
511 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
512 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
514 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
515 } else {
516 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
518 if (IS_ERR(bitmap->sb_page)) {
519 err = PTR_ERR(bitmap->sb_page);
520 bitmap->sb_page = NULL;
521 return err;
524 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
526 chunksize = le32_to_cpu(sb->chunksize);
527 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
528 write_behind = le32_to_cpu(sb->write_behind);
530 /* verify that the bitmap-specific fields are valid */
531 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
532 reason = "bad magic";
533 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
534 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
535 reason = "unrecognized superblock version";
536 else if (chunksize < PAGE_SIZE)
537 reason = "bitmap chunksize too small";
538 else if ((1 << ffz(~chunksize)) != chunksize)
539 reason = "bitmap chunksize not a power of 2";
540 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
541 reason = "daemon sleep period out of range";
542 else if (write_behind > COUNTER_MAX)
543 reason = "write-behind limit out of range (0 - 16383)";
544 if (reason) {
545 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
546 bmname(bitmap), reason);
547 goto out;
550 /* keep the array size field of the bitmap superblock up to date */
551 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
553 if (!bitmap->mddev->persistent)
554 goto success;
557 * if we have a persistent array superblock, compare the
558 * bitmap's UUID and event counter to the mddev's
560 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
561 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
562 bmname(bitmap));
563 goto out;
565 events = le64_to_cpu(sb->events);
566 if (events < bitmap->mddev->events) {
567 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
568 "-- forcing full recovery\n", bmname(bitmap), events,
569 (unsigned long long) bitmap->mddev->events);
570 sb->state |= cpu_to_le32(BITMAP_STALE);
572 success:
573 /* assign fields using values from superblock */
574 bitmap->chunksize = chunksize;
575 bitmap->daemon_sleep = daemon_sleep;
576 bitmap->daemon_lastrun = jiffies;
577 bitmap->max_write_behind = write_behind;
578 bitmap->flags |= le32_to_cpu(sb->state);
579 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
580 bitmap->flags |= BITMAP_HOSTENDIAN;
581 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
582 if (sb->state & cpu_to_le32(BITMAP_STALE))
583 bitmap->events_cleared = bitmap->mddev->events;
584 err = 0;
585 out:
586 kunmap_atomic(sb, KM_USER0);
587 if (err)
588 bitmap_print_sb(bitmap);
589 return err;
592 enum bitmap_mask_op {
593 MASK_SET,
594 MASK_UNSET
597 /* record the state of the bitmap in the superblock. Return the old value */
598 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
599 enum bitmap_mask_op op)
601 bitmap_super_t *sb;
602 unsigned long flags;
603 int old;
605 spin_lock_irqsave(&bitmap->lock, flags);
606 if (!bitmap->sb_page) { /* can't set the state */
607 spin_unlock_irqrestore(&bitmap->lock, flags);
608 return 0;
610 spin_unlock_irqrestore(&bitmap->lock, flags);
611 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
612 old = le32_to_cpu(sb->state) & bits;
613 switch (op) {
614 case MASK_SET: sb->state |= cpu_to_le32(bits);
615 break;
616 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
617 break;
618 default: BUG();
620 kunmap_atomic(sb, KM_USER0);
621 return old;
625 * general bitmap file operations
628 /* calculate the index of the page that contains this bit */
629 static inline unsigned long file_page_index(unsigned long chunk)
631 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
634 /* calculate the (bit) offset of this bit within a page */
635 static inline unsigned long file_page_offset(unsigned long chunk)
637 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
641 * return a pointer to the page in the filemap that contains the given bit
643 * this lookup is complicated by the fact that the bitmap sb might be exactly
644 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
645 * 0 or page 1
647 static inline struct page *filemap_get_page(struct bitmap *bitmap,
648 unsigned long chunk)
650 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
651 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
655 static void bitmap_file_unmap(struct bitmap *bitmap)
657 struct page **map, *sb_page;
658 unsigned long *attr;
659 int pages;
660 unsigned long flags;
662 spin_lock_irqsave(&bitmap->lock, flags);
663 map = bitmap->filemap;
664 bitmap->filemap = NULL;
665 attr = bitmap->filemap_attr;
666 bitmap->filemap_attr = NULL;
667 pages = bitmap->file_pages;
668 bitmap->file_pages = 0;
669 sb_page = bitmap->sb_page;
670 bitmap->sb_page = NULL;
671 spin_unlock_irqrestore(&bitmap->lock, flags);
673 while (pages--)
674 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
675 free_buffers(map[pages]);
676 kfree(map);
677 kfree(attr);
679 if (sb_page)
680 free_buffers(sb_page);
683 static void bitmap_file_put(struct bitmap *bitmap)
685 struct file *file;
686 unsigned long flags;
688 spin_lock_irqsave(&bitmap->lock, flags);
689 file = bitmap->file;
690 bitmap->file = NULL;
691 spin_unlock_irqrestore(&bitmap->lock, flags);
693 if (file)
694 wait_event(bitmap->write_wait,
695 atomic_read(&bitmap->pending_writes)==0);
696 bitmap_file_unmap(bitmap);
698 if (file) {
699 struct inode *inode = file->f_path.dentry->d_inode;
700 invalidate_mapping_pages(inode->i_mapping, 0, -1);
701 fput(file);
707 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
708 * then it is no longer reliable, so we stop using it and we mark the file
709 * as failed in the superblock
711 static void bitmap_file_kick(struct bitmap *bitmap)
713 char *path, *ptr = NULL;
715 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
716 bitmap_update_sb(bitmap);
718 if (bitmap->file) {
719 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
720 if (path)
721 ptr = d_path(&bitmap->file->f_path, path,
722 PAGE_SIZE);
725 printk(KERN_ALERT
726 "%s: kicking failed bitmap file %s from array!\n",
727 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
729 kfree(path);
730 } else
731 printk(KERN_ALERT
732 "%s: disabling internal bitmap due to errors\n",
733 bmname(bitmap));
736 bitmap_file_put(bitmap);
738 return;
741 enum bitmap_page_attr {
742 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
743 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
744 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
747 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
748 enum bitmap_page_attr attr)
750 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
753 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
754 enum bitmap_page_attr attr)
756 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
759 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
760 enum bitmap_page_attr attr)
762 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
766 * bitmap_file_set_bit -- called before performing a write to the md device
767 * to set (and eventually sync) a particular bit in the bitmap file
769 * we set the bit immediately, then we record the page number so that
770 * when an unplug occurs, we can flush the dirty pages out to disk
772 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
774 unsigned long bit;
775 struct page *page;
776 void *kaddr;
777 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
779 if (!bitmap->filemap) {
780 return;
783 page = filemap_get_page(bitmap, chunk);
784 if (!page) return;
785 bit = file_page_offset(chunk);
787 /* set the bit */
788 kaddr = kmap_atomic(page, KM_USER0);
789 if (bitmap->flags & BITMAP_HOSTENDIAN)
790 set_bit(bit, kaddr);
791 else
792 ext2_set_bit(bit, kaddr);
793 kunmap_atomic(kaddr, KM_USER0);
794 PRINTK("set file bit %lu page %lu\n", bit, page->index);
796 /* record page number so it gets flushed to disk when unplug occurs */
797 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
801 /* this gets called when the md device is ready to unplug its underlying
802 * (slave) device queues -- before we let any writes go down, we need to
803 * sync the dirty pages of the bitmap file to disk */
804 void bitmap_unplug(struct bitmap *bitmap)
806 unsigned long i, flags;
807 int dirty, need_write;
808 struct page *page;
809 int wait = 0;
811 if (!bitmap)
812 return;
814 /* look at each page to see if there are any set bits that need to be
815 * flushed out to disk */
816 for (i = 0; i < bitmap->file_pages; i++) {
817 spin_lock_irqsave(&bitmap->lock, flags);
818 if (!bitmap->filemap) {
819 spin_unlock_irqrestore(&bitmap->lock, flags);
820 return;
822 page = bitmap->filemap[i];
823 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
824 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
825 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
826 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
827 if (dirty)
828 wait = 1;
829 spin_unlock_irqrestore(&bitmap->lock, flags);
831 if (dirty | need_write)
832 write_page(bitmap, page, 0);
834 if (wait) { /* if any writes were performed, we need to wait on them */
835 if (bitmap->file)
836 wait_event(bitmap->write_wait,
837 atomic_read(&bitmap->pending_writes)==0);
838 else
839 md_super_wait(bitmap->mddev);
841 if (bitmap->flags & BITMAP_WRITE_ERROR)
842 bitmap_file_kick(bitmap);
845 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
846 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
847 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
848 * memory mapping of the bitmap file
849 * Special cases:
850 * if there's no bitmap file, or if the bitmap file had been
851 * previously kicked from the array, we mark all the bits as
852 * 1's in order to cause a full resync.
854 * We ignore all bits for sectors that end earlier than 'start'.
855 * This is used when reading an out-of-date bitmap...
857 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
859 unsigned long i, chunks, index, oldindex, bit;
860 struct page *page = NULL, *oldpage = NULL;
861 unsigned long num_pages, bit_cnt = 0;
862 struct file *file;
863 unsigned long bytes, offset;
864 int outofdate;
865 int ret = -ENOSPC;
866 void *paddr;
868 chunks = bitmap->chunks;
869 file = bitmap->file;
871 BUG_ON(!file && !bitmap->offset);
873 #ifdef INJECT_FAULTS_3
874 outofdate = 1;
875 #else
876 outofdate = bitmap->flags & BITMAP_STALE;
877 #endif
878 if (outofdate)
879 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
880 "recovery\n", bmname(bitmap));
882 bytes = (chunks + 7) / 8;
884 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
886 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
887 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
888 bmname(bitmap),
889 (unsigned long) i_size_read(file->f_mapping->host),
890 bytes + sizeof(bitmap_super_t));
891 goto err;
894 ret = -ENOMEM;
896 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
897 if (!bitmap->filemap)
898 goto err;
900 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
901 bitmap->filemap_attr = kzalloc(
902 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
903 GFP_KERNEL);
904 if (!bitmap->filemap_attr)
905 goto err;
907 oldindex = ~0L;
909 for (i = 0; i < chunks; i++) {
910 int b;
911 index = file_page_index(i);
912 bit = file_page_offset(i);
913 if (index != oldindex) { /* this is a new page, read it in */
914 int count;
915 /* unmap the old page, we're done with it */
916 if (index == num_pages-1)
917 count = bytes + sizeof(bitmap_super_t)
918 - index * PAGE_SIZE;
919 else
920 count = PAGE_SIZE;
921 if (index == 0) {
923 * if we're here then the superblock page
924 * contains some bits (PAGE_SIZE != sizeof sb)
925 * we've already read it in, so just use it
927 page = bitmap->sb_page;
928 offset = sizeof(bitmap_super_t);
929 } else if (file) {
930 page = read_page(file, index, bitmap, count);
931 offset = 0;
932 } else {
933 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
934 offset = 0;
936 if (IS_ERR(page)) { /* read error */
937 ret = PTR_ERR(page);
938 goto err;
941 oldindex = index;
942 oldpage = page;
944 if (outofdate) {
946 * if bitmap is out of date, dirty the
947 * whole page and write it out
949 paddr = kmap_atomic(page, KM_USER0);
950 memset(paddr + offset, 0xff,
951 PAGE_SIZE - offset);
952 kunmap_atomic(paddr, KM_USER0);
953 write_page(bitmap, page, 1);
955 ret = -EIO;
956 if (bitmap->flags & BITMAP_WRITE_ERROR) {
957 /* release, page not in filemap yet */
958 put_page(page);
959 goto err;
963 bitmap->filemap[bitmap->file_pages++] = page;
964 bitmap->last_page_size = count;
966 paddr = kmap_atomic(page, KM_USER0);
967 if (bitmap->flags & BITMAP_HOSTENDIAN)
968 b = test_bit(bit, paddr);
969 else
970 b = ext2_test_bit(bit, paddr);
971 kunmap_atomic(paddr, KM_USER0);
972 if (b) {
973 /* if the disk bit is set, set the memory bit */
974 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
975 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
977 bit_cnt++;
978 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
982 /* everything went OK */
983 ret = 0;
984 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
986 if (bit_cnt) { /* Kick recovery if any bits were set */
987 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
988 md_wakeup_thread(bitmap->mddev->thread);
991 printk(KERN_INFO "%s: bitmap initialized from disk: "
992 "read %lu/%lu pages, set %lu bits\n",
993 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
995 return 0;
997 err:
998 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
999 bmname(bitmap), ret);
1000 return ret;
1003 void bitmap_write_all(struct bitmap *bitmap)
1005 /* We don't actually write all bitmap blocks here,
1006 * just flag them as needing to be written
1008 int i;
1010 for (i=0; i < bitmap->file_pages; i++)
1011 set_page_attr(bitmap, bitmap->filemap[i],
1012 BITMAP_PAGE_NEEDWRITE);
1016 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1018 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1019 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1020 bitmap->bp[page].count += inc;
1022 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1023 (unsigned long long)offset, inc, bitmap->bp[page].count);
1025 bitmap_checkfree(bitmap, page);
1027 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1028 sector_t offset, int *blocks,
1029 int create);
1032 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1033 * out to disk
1036 void bitmap_daemon_work(struct bitmap *bitmap)
1038 unsigned long j;
1039 unsigned long flags;
1040 struct page *page = NULL, *lastpage = NULL;
1041 int blocks;
1042 void *paddr;
1044 if (bitmap == NULL)
1045 return;
1046 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1047 goto done;
1049 bitmap->daemon_lastrun = jiffies;
1050 if (bitmap->allclean) {
1051 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1052 return;
1054 bitmap->allclean = 1;
1056 for (j = 0; j < bitmap->chunks; j++) {
1057 bitmap_counter_t *bmc;
1058 spin_lock_irqsave(&bitmap->lock, flags);
1059 if (!bitmap->filemap) {
1060 /* error or shutdown */
1061 spin_unlock_irqrestore(&bitmap->lock, flags);
1062 break;
1065 page = filemap_get_page(bitmap, j);
1067 if (page != lastpage) {
1068 /* skip this page unless it's marked as needing cleaning */
1069 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1070 int need_write = test_page_attr(bitmap, page,
1071 BITMAP_PAGE_NEEDWRITE);
1072 if (need_write)
1073 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1075 spin_unlock_irqrestore(&bitmap->lock, flags);
1076 if (need_write) {
1077 write_page(bitmap, page, 0);
1078 bitmap->allclean = 0;
1080 continue;
1083 /* grab the new page, sync and release the old */
1084 if (lastpage != NULL) {
1085 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1086 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1087 spin_unlock_irqrestore(&bitmap->lock, flags);
1088 write_page(bitmap, lastpage, 0);
1089 } else {
1090 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1091 spin_unlock_irqrestore(&bitmap->lock, flags);
1093 } else
1094 spin_unlock_irqrestore(&bitmap->lock, flags);
1095 lastpage = page;
1097 /* We are possibly going to clear some bits, so make
1098 * sure that events_cleared is up-to-date.
1100 if (bitmap->need_sync) {
1101 bitmap_super_t *sb;
1102 bitmap->need_sync = 0;
1103 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1104 sb->events_cleared =
1105 cpu_to_le64(bitmap->events_cleared);
1106 kunmap_atomic(sb, KM_USER0);
1107 write_page(bitmap, bitmap->sb_page, 1);
1109 spin_lock_irqsave(&bitmap->lock, flags);
1110 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1112 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1113 &blocks, 0);
1114 if (bmc) {
1116 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1118 if (*bmc)
1119 bitmap->allclean = 0;
1121 if (*bmc == 2) {
1122 *bmc=1; /* maybe clear the bit next time */
1123 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1124 } else if (*bmc == 1) {
1125 /* we can clear the bit */
1126 *bmc = 0;
1127 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1128 -1);
1130 /* clear the bit */
1131 paddr = kmap_atomic(page, KM_USER0);
1132 if (bitmap->flags & BITMAP_HOSTENDIAN)
1133 clear_bit(file_page_offset(j), paddr);
1134 else
1135 ext2_clear_bit(file_page_offset(j), paddr);
1136 kunmap_atomic(paddr, KM_USER0);
1139 spin_unlock_irqrestore(&bitmap->lock, flags);
1142 /* now sync the final page */
1143 if (lastpage != NULL) {
1144 spin_lock_irqsave(&bitmap->lock, flags);
1145 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1146 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1147 spin_unlock_irqrestore(&bitmap->lock, flags);
1148 write_page(bitmap, lastpage, 0);
1149 } else {
1150 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1151 spin_unlock_irqrestore(&bitmap->lock, flags);
1155 done:
1156 if (bitmap->allclean == 0)
1157 bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1160 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1161 sector_t offset, int *blocks,
1162 int create)
1164 /* If 'create', we might release the lock and reclaim it.
1165 * The lock must have been taken with interrupts enabled.
1166 * If !create, we don't release the lock.
1168 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1169 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1170 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1171 sector_t csize;
1173 if (bitmap_checkpage(bitmap, page, create) < 0) {
1174 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1175 *blocks = csize - (offset & (csize- 1));
1176 return NULL;
1178 /* now locked ... */
1180 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1181 /* should we use the first or second counter field
1182 * of the hijacked pointer? */
1183 int hi = (pageoff > PAGE_COUNTER_MASK);
1184 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1185 PAGE_COUNTER_SHIFT - 1);
1186 *blocks = csize - (offset & (csize- 1));
1187 return &((bitmap_counter_t *)
1188 &bitmap->bp[page].map)[hi];
1189 } else { /* page is allocated */
1190 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1191 *blocks = csize - (offset & (csize- 1));
1192 return (bitmap_counter_t *)
1193 &(bitmap->bp[page].map[pageoff]);
1197 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1199 if (!bitmap) return 0;
1201 if (behind) {
1202 atomic_inc(&bitmap->behind_writes);
1203 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1204 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1207 while (sectors) {
1208 int blocks;
1209 bitmap_counter_t *bmc;
1211 spin_lock_irq(&bitmap->lock);
1212 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1213 if (!bmc) {
1214 spin_unlock_irq(&bitmap->lock);
1215 return 0;
1218 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1219 DEFINE_WAIT(__wait);
1220 /* note that it is safe to do the prepare_to_wait
1221 * after the test as long as we do it before dropping
1222 * the spinlock.
1224 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1225 TASK_UNINTERRUPTIBLE);
1226 spin_unlock_irq(&bitmap->lock);
1227 blk_unplug(bitmap->mddev->queue);
1228 schedule();
1229 finish_wait(&bitmap->overflow_wait, &__wait);
1230 continue;
1233 switch(*bmc) {
1234 case 0:
1235 bitmap_file_set_bit(bitmap, offset);
1236 bitmap_count_page(bitmap,offset, 1);
1237 blk_plug_device(bitmap->mddev->queue);
1238 /* fall through */
1239 case 1:
1240 *bmc = 2;
1243 (*bmc)++;
1245 spin_unlock_irq(&bitmap->lock);
1247 offset += blocks;
1248 if (sectors > blocks)
1249 sectors -= blocks;
1250 else sectors = 0;
1252 bitmap->allclean = 0;
1253 return 0;
1256 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1257 int success, int behind)
1259 if (!bitmap) return;
1260 if (behind) {
1261 atomic_dec(&bitmap->behind_writes);
1262 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1263 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1266 while (sectors) {
1267 int blocks;
1268 unsigned long flags;
1269 bitmap_counter_t *bmc;
1271 spin_lock_irqsave(&bitmap->lock, flags);
1272 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1273 if (!bmc) {
1274 spin_unlock_irqrestore(&bitmap->lock, flags);
1275 return;
1278 if (success &&
1279 bitmap->events_cleared < bitmap->mddev->events) {
1280 bitmap->events_cleared = bitmap->mddev->events;
1281 bitmap->need_sync = 1;
1284 if (!success && ! (*bmc & NEEDED_MASK))
1285 *bmc |= NEEDED_MASK;
1287 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1288 wake_up(&bitmap->overflow_wait);
1290 (*bmc)--;
1291 if (*bmc <= 2) {
1292 set_page_attr(bitmap,
1293 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1294 BITMAP_PAGE_CLEAN);
1296 spin_unlock_irqrestore(&bitmap->lock, flags);
1297 offset += blocks;
1298 if (sectors > blocks)
1299 sectors -= blocks;
1300 else sectors = 0;
1304 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1305 int degraded)
1307 bitmap_counter_t *bmc;
1308 int rv;
1309 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1310 *blocks = 1024;
1311 return 1; /* always resync if no bitmap */
1313 spin_lock_irq(&bitmap->lock);
1314 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1315 rv = 0;
1316 if (bmc) {
1317 /* locked */
1318 if (RESYNC(*bmc))
1319 rv = 1;
1320 else if (NEEDED(*bmc)) {
1321 rv = 1;
1322 if (!degraded) { /* don't set/clear bits if degraded */
1323 *bmc |= RESYNC_MASK;
1324 *bmc &= ~NEEDED_MASK;
1328 spin_unlock_irq(&bitmap->lock);
1329 bitmap->allclean = 0;
1330 return rv;
1333 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1335 bitmap_counter_t *bmc;
1336 unsigned long flags;
1338 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1339 */ if (bitmap == NULL) {
1340 *blocks = 1024;
1341 return;
1343 spin_lock_irqsave(&bitmap->lock, flags);
1344 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1345 if (bmc == NULL)
1346 goto unlock;
1347 /* locked */
1349 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1351 if (RESYNC(*bmc)) {
1352 *bmc &= ~RESYNC_MASK;
1354 if (!NEEDED(*bmc) && aborted)
1355 *bmc |= NEEDED_MASK;
1356 else {
1357 if (*bmc <= 2) {
1358 set_page_attr(bitmap,
1359 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1360 BITMAP_PAGE_CLEAN);
1364 unlock:
1365 spin_unlock_irqrestore(&bitmap->lock, flags);
1366 bitmap->allclean = 0;
1369 void bitmap_close_sync(struct bitmap *bitmap)
1371 /* Sync has finished, and any bitmap chunks that weren't synced
1372 * properly have been aborted. It remains to us to clear the
1373 * RESYNC bit wherever it is still on
1375 sector_t sector = 0;
1376 int blocks;
1377 if (!bitmap)
1378 return;
1379 while (sector < bitmap->mddev->resync_max_sectors) {
1380 bitmap_end_sync(bitmap, sector, &blocks, 0);
1381 sector += blocks;
1385 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1387 sector_t s = 0;
1388 int blocks;
1390 if (!bitmap)
1391 return;
1392 if (sector == 0) {
1393 bitmap->last_end_sync = jiffies;
1394 return;
1396 if (time_before(jiffies, (bitmap->last_end_sync
1397 + bitmap->daemon_sleep * HZ)))
1398 return;
1399 wait_event(bitmap->mddev->recovery_wait,
1400 atomic_read(&bitmap->mddev->recovery_active) == 0);
1402 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1403 s = 0;
1404 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1405 bitmap_end_sync(bitmap, s, &blocks, 0);
1406 s += blocks;
1408 bitmap->last_end_sync = jiffies;
1411 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1413 /* For each chunk covered by any of these sectors, set the
1414 * counter to 1 and set resync_needed. They should all
1415 * be 0 at this point
1418 int secs;
1419 bitmap_counter_t *bmc;
1420 spin_lock_irq(&bitmap->lock);
1421 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1422 if (!bmc) {
1423 spin_unlock_irq(&bitmap->lock);
1424 return;
1426 if (! *bmc) {
1427 struct page *page;
1428 *bmc = 1 | (needed?NEEDED_MASK:0);
1429 bitmap_count_page(bitmap, offset, 1);
1430 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1431 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1433 spin_unlock_irq(&bitmap->lock);
1434 bitmap->allclean = 0;
1437 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1438 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1440 unsigned long chunk;
1442 for (chunk = s; chunk <= e; chunk++) {
1443 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1444 bitmap_set_memory_bits(bitmap, sec, 1);
1445 bitmap_file_set_bit(bitmap, sec);
1450 * flush out any pending updates
1452 void bitmap_flush(mddev_t *mddev)
1454 struct bitmap *bitmap = mddev->bitmap;
1455 int sleep;
1457 if (!bitmap) /* there was no bitmap */
1458 return;
1460 /* run the daemon_work three time to ensure everything is flushed
1461 * that can be
1463 sleep = bitmap->daemon_sleep;
1464 bitmap->daemon_sleep = 0;
1465 bitmap_daemon_work(bitmap);
1466 bitmap_daemon_work(bitmap);
1467 bitmap_daemon_work(bitmap);
1468 bitmap->daemon_sleep = sleep;
1469 bitmap_update_sb(bitmap);
1473 * free memory that was allocated
1475 static void bitmap_free(struct bitmap *bitmap)
1477 unsigned long k, pages;
1478 struct bitmap_page *bp;
1480 if (!bitmap) /* there was no bitmap */
1481 return;
1483 /* release the bitmap file and kill the daemon */
1484 bitmap_file_put(bitmap);
1486 bp = bitmap->bp;
1487 pages = bitmap->pages;
1489 /* free all allocated memory */
1491 if (bp) /* deallocate the page memory */
1492 for (k = 0; k < pages; k++)
1493 if (bp[k].map && !bp[k].hijacked)
1494 kfree(bp[k].map);
1495 kfree(bp);
1496 kfree(bitmap);
1498 void bitmap_destroy(mddev_t *mddev)
1500 struct bitmap *bitmap = mddev->bitmap;
1502 if (!bitmap) /* there was no bitmap */
1503 return;
1505 mddev->bitmap = NULL; /* disconnect from the md device */
1506 if (mddev->thread)
1507 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1509 bitmap_free(bitmap);
1513 * initialize the bitmap structure
1514 * if this returns an error, bitmap_destroy must be called to do clean up
1516 int bitmap_create(mddev_t *mddev)
1518 struct bitmap *bitmap;
1519 unsigned long blocks = mddev->resync_max_sectors;
1520 unsigned long chunks;
1521 unsigned long pages;
1522 struct file *file = mddev->bitmap_file;
1523 int err;
1524 sector_t start;
1526 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1528 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1529 return 0;
1531 BUG_ON(file && mddev->bitmap_offset);
1533 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1534 if (!bitmap)
1535 return -ENOMEM;
1537 spin_lock_init(&bitmap->lock);
1538 atomic_set(&bitmap->pending_writes, 0);
1539 init_waitqueue_head(&bitmap->write_wait);
1540 init_waitqueue_head(&bitmap->overflow_wait);
1542 bitmap->mddev = mddev;
1544 bitmap->file = file;
1545 bitmap->offset = mddev->bitmap_offset;
1546 if (file) {
1547 get_file(file);
1548 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1549 SYNC_FILE_RANGE_WAIT_BEFORE |
1550 SYNC_FILE_RANGE_WRITE |
1551 SYNC_FILE_RANGE_WAIT_AFTER);
1553 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1554 err = bitmap_read_sb(bitmap);
1555 if (err)
1556 goto error;
1558 bitmap->chunkshift = ffz(~bitmap->chunksize);
1560 /* now that chunksize and chunkshift are set, we can use these macros */
1561 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1562 CHUNK_BLOCK_RATIO(bitmap);
1563 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1565 BUG_ON(!pages);
1567 bitmap->chunks = chunks;
1568 bitmap->pages = pages;
1569 bitmap->missing_pages = pages;
1570 bitmap->counter_bits = COUNTER_BITS;
1572 bitmap->syncchunk = ~0UL;
1574 #ifdef INJECT_FATAL_FAULT_1
1575 bitmap->bp = NULL;
1576 #else
1577 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1578 #endif
1579 err = -ENOMEM;
1580 if (!bitmap->bp)
1581 goto error;
1583 /* now that we have some pages available, initialize the in-memory
1584 * bitmap from the on-disk bitmap */
1585 start = 0;
1586 if (mddev->degraded == 0
1587 || bitmap->events_cleared == mddev->events)
1588 /* no need to keep dirty bits to optimise a re-add of a missing device */
1589 start = mddev->recovery_cp;
1590 err = bitmap_init_from_disk(bitmap, start);
1592 if (err)
1593 goto error;
1595 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1596 pages, bmname(bitmap));
1598 mddev->bitmap = bitmap;
1600 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1602 bitmap_update_sb(bitmap);
1604 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1606 error:
1607 bitmap_free(bitmap);
1608 return err;
1611 /* the bitmap API -- for raid personalities */
1612 EXPORT_SYMBOL(bitmap_startwrite);
1613 EXPORT_SYMBOL(bitmap_endwrite);
1614 EXPORT_SYMBOL(bitmap_start_sync);
1615 EXPORT_SYMBOL(bitmap_end_sync);
1616 EXPORT_SYMBOL(bitmap_unplug);
1617 EXPORT_SYMBOL(bitmap_close_sync);
1618 EXPORT_SYMBOL(bitmap_cond_end_sync);