ACPI: pci_root: simplify list traversals
[linux-2.6/linux-acpi-2.6.git] / drivers / md / bitmap.c
blob56df1cee8fb3435a43e33b07ba1a3dadaf528680
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/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/timer.h>
25 #include <linux/sched.h>
26 #include <linux/list.h>
27 #include <linux/file.h>
28 #include <linux/mount.h>
29 #include <linux/buffer_head.h>
30 #include "md.h"
31 #include "bitmap.h"
33 /* debug macros */
35 #define DEBUG 0
37 #if DEBUG
38 /* these are for debugging purposes only! */
40 /* define one and only one of these */
41 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
42 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
43 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
44 #define INJECT_FAULTS_4 0 /* undef */
45 #define INJECT_FAULTS_5 0 /* undef */
46 #define INJECT_FAULTS_6 0
48 /* if these are defined, the driver will fail! debug only */
49 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
50 #define INJECT_FATAL_FAULT_2 0 /* undef */
51 #define INJECT_FATAL_FAULT_3 0 /* undef */
52 #endif
54 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
55 #define DPRINTK(x...) do { } while(0)
57 #ifndef PRINTK
58 # if DEBUG > 0
59 # define PRINTK(x...) printk(KERN_DEBUG x)
60 # else
61 # define PRINTK(x...)
62 # endif
63 #endif
65 static inline char * bmname(struct bitmap *bitmap)
67 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
72 * just a placeholder - calls kmalloc for bitmap pages
74 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
76 unsigned char *page;
78 #ifdef INJECT_FAULTS_1
79 page = NULL;
80 #else
81 page = kmalloc(PAGE_SIZE, GFP_NOIO);
82 #endif
83 if (!page)
84 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
85 else
86 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
87 bmname(bitmap), page);
88 return page;
92 * for now just a placeholder -- just calls kfree for bitmap pages
94 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
96 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
97 kfree(page);
101 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
103 * 1) check to see if this page is allocated, if it's not then try to alloc
104 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
105 * page pointer directly as a counter
107 * if we find our page, we increment the page's refcount so that it stays
108 * allocated while we're using it
110 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
112 unsigned char *mappage;
114 if (page >= bitmap->pages) {
115 /* This can happen if bitmap_start_sync goes beyond
116 * End-of-device while looking for a whole page.
117 * It is harmless.
119 return -EINVAL;
123 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
124 return 0;
126 if (bitmap->bp[page].map) /* page is already allocated, just return */
127 return 0;
129 if (!create)
130 return -ENOENT;
132 spin_unlock_irq(&bitmap->lock);
134 /* this page has not been allocated yet */
136 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
137 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
138 bmname(bitmap));
139 /* failed - set the hijacked flag so that we can use the
140 * pointer as a counter */
141 spin_lock_irq(&bitmap->lock);
142 if (!bitmap->bp[page].map)
143 bitmap->bp[page].hijacked = 1;
144 goto out;
147 /* got a page */
149 spin_lock_irq(&bitmap->lock);
151 /* recheck the page */
153 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
154 /* somebody beat us to getting the page */
155 bitmap_free_page(bitmap, mappage);
156 return 0;
159 /* no page was in place and we have one, so install it */
161 memset(mappage, 0, PAGE_SIZE);
162 bitmap->bp[page].map = mappage;
163 bitmap->missing_pages--;
164 out:
165 return 0;
169 /* if page is completely empty, put it back on the free list, or dealloc it */
170 /* if page was hijacked, unmark the flag so it might get alloced next time */
171 /* Note: lock should be held when calling this */
172 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
174 char *ptr;
176 if (bitmap->bp[page].count) /* page is still busy */
177 return;
179 /* page is no longer in use, it can be released */
181 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
182 bitmap->bp[page].hijacked = 0;
183 bitmap->bp[page].map = NULL;
184 return;
187 /* normal case, free the page */
189 #if 0
190 /* actually ... let's not. We will probably need the page again exactly when
191 * memory is tight and we are flusing to disk
193 return;
194 #else
195 ptr = bitmap->bp[page].map;
196 bitmap->bp[page].map = NULL;
197 bitmap->missing_pages++;
198 bitmap_free_page(bitmap, ptr);
199 return;
200 #endif
205 * bitmap file handling - read and write the bitmap file and its superblock
209 * basic page I/O operations
212 /* IO operations when bitmap is stored near all superblocks */
213 static struct page *read_sb_page(mddev_t *mddev, long offset,
214 struct page *page,
215 unsigned long index, int size)
217 /* choose a good rdev and read the page from there */
219 mdk_rdev_t *rdev;
220 sector_t target;
222 if (!page)
223 page = alloc_page(GFP_KERNEL);
224 if (!page)
225 return ERR_PTR(-ENOMEM);
227 list_for_each_entry(rdev, &mddev->disks, same_set) {
228 if (! test_bit(In_sync, &rdev->flags)
229 || test_bit(Faulty, &rdev->flags))
230 continue;
232 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
234 if (sync_page_io(rdev->bdev, target,
235 roundup(size, bdev_hardsect_size(rdev->bdev)),
236 page, READ)) {
237 page->index = index;
238 attach_page_buffers(page, NULL); /* so that free_buffer will
239 * quietly no-op */
240 return page;
243 return ERR_PTR(-EIO);
247 static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
249 /* Iterate the disks of an mddev, using rcu to protect access to the
250 * linked list, and raising the refcount of devices we return to ensure
251 * they don't disappear while in use.
252 * As devices are only added or removed when raid_disk is < 0 and
253 * nr_pending is 0 and In_sync is clear, the entries we return will
254 * still be in the same position on the list when we re-enter
255 * list_for_each_continue_rcu.
257 struct list_head *pos;
258 rcu_read_lock();
259 if (rdev == NULL)
260 /* start at the beginning */
261 pos = &mddev->disks;
262 else {
263 /* release the previous rdev and start from there. */
264 rdev_dec_pending(rdev, mddev);
265 pos = &rdev->same_set;
267 list_for_each_continue_rcu(pos, &mddev->disks) {
268 rdev = list_entry(pos, mdk_rdev_t, same_set);
269 if (rdev->raid_disk >= 0 &&
270 !test_bit(Faulty, &rdev->flags)) {
271 /* this is a usable devices */
272 atomic_inc(&rdev->nr_pending);
273 rcu_read_unlock();
274 return rdev;
277 rcu_read_unlock();
278 return NULL;
281 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
283 mdk_rdev_t *rdev = NULL;
284 mddev_t *mddev = bitmap->mddev;
286 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
287 int size = PAGE_SIZE;
288 if (page->index == bitmap->file_pages-1)
289 size = roundup(bitmap->last_page_size,
290 bdev_hardsect_size(rdev->bdev));
291 /* Just make sure we aren't corrupting data or
292 * metadata
294 if (bitmap->offset < 0) {
295 /* DATA BITMAP METADATA */
296 if (bitmap->offset
297 + (long)(page->index * (PAGE_SIZE/512))
298 + size/512 > 0)
299 /* bitmap runs in to metadata */
300 goto bad_alignment;
301 if (rdev->data_offset + mddev->dev_sectors
302 > rdev->sb_start + bitmap->offset)
303 /* data runs in to bitmap */
304 goto bad_alignment;
305 } else if (rdev->sb_start < rdev->data_offset) {
306 /* METADATA BITMAP DATA */
307 if (rdev->sb_start
308 + bitmap->offset
309 + page->index*(PAGE_SIZE/512) + size/512
310 > rdev->data_offset)
311 /* bitmap runs in to data */
312 goto bad_alignment;
313 } else {
314 /* DATA METADATA BITMAP - no problems */
316 md_super_write(mddev, rdev,
317 rdev->sb_start + bitmap->offset
318 + page->index * (PAGE_SIZE/512),
319 size,
320 page);
323 if (wait)
324 md_super_wait(mddev);
325 return 0;
327 bad_alignment:
328 rcu_read_unlock();
329 return -EINVAL;
332 static void bitmap_file_kick(struct bitmap *bitmap);
334 * write out a page to a file
336 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
338 struct buffer_head *bh;
340 if (bitmap->file == NULL) {
341 switch (write_sb_page(bitmap, page, wait)) {
342 case -EINVAL:
343 bitmap->flags |= BITMAP_WRITE_ERROR;
345 } else {
347 bh = page_buffers(page);
349 while (bh && bh->b_blocknr) {
350 atomic_inc(&bitmap->pending_writes);
351 set_buffer_locked(bh);
352 set_buffer_mapped(bh);
353 submit_bh(WRITE, bh);
354 bh = bh->b_this_page;
357 if (wait) {
358 wait_event(bitmap->write_wait,
359 atomic_read(&bitmap->pending_writes)==0);
362 if (bitmap->flags & BITMAP_WRITE_ERROR)
363 bitmap_file_kick(bitmap);
366 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
368 struct bitmap *bitmap = bh->b_private;
369 unsigned long flags;
371 if (!uptodate) {
372 spin_lock_irqsave(&bitmap->lock, flags);
373 bitmap->flags |= BITMAP_WRITE_ERROR;
374 spin_unlock_irqrestore(&bitmap->lock, flags);
376 if (atomic_dec_and_test(&bitmap->pending_writes))
377 wake_up(&bitmap->write_wait);
380 /* copied from buffer.c */
381 static void
382 __clear_page_buffers(struct page *page)
384 ClearPagePrivate(page);
385 set_page_private(page, 0);
386 page_cache_release(page);
388 static void free_buffers(struct page *page)
390 struct buffer_head *bh = page_buffers(page);
392 while (bh) {
393 struct buffer_head *next = bh->b_this_page;
394 free_buffer_head(bh);
395 bh = next;
397 __clear_page_buffers(page);
398 put_page(page);
401 /* read a page from a file.
402 * We both read the page, and attach buffers to the page to record the
403 * address of each block (using bmap). These addresses will be used
404 * to write the block later, completely bypassing the filesystem.
405 * This usage is similar to how swap files are handled, and allows us
406 * to write to a file with no concerns of memory allocation failing.
408 static struct page *read_page(struct file *file, unsigned long index,
409 struct bitmap *bitmap,
410 unsigned long count)
412 struct page *page = NULL;
413 struct inode *inode = file->f_path.dentry->d_inode;
414 struct buffer_head *bh;
415 sector_t block;
417 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
418 (unsigned long long)index << PAGE_SHIFT);
420 page = alloc_page(GFP_KERNEL);
421 if (!page)
422 page = ERR_PTR(-ENOMEM);
423 if (IS_ERR(page))
424 goto out;
426 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
427 if (!bh) {
428 put_page(page);
429 page = ERR_PTR(-ENOMEM);
430 goto out;
432 attach_page_buffers(page, bh);
433 block = index << (PAGE_SHIFT - inode->i_blkbits);
434 while (bh) {
435 if (count == 0)
436 bh->b_blocknr = 0;
437 else {
438 bh->b_blocknr = bmap(inode, block);
439 if (bh->b_blocknr == 0) {
440 /* Cannot use this file! */
441 free_buffers(page);
442 page = ERR_PTR(-EINVAL);
443 goto out;
445 bh->b_bdev = inode->i_sb->s_bdev;
446 if (count < (1<<inode->i_blkbits))
447 count = 0;
448 else
449 count -= (1<<inode->i_blkbits);
451 bh->b_end_io = end_bitmap_write;
452 bh->b_private = bitmap;
453 atomic_inc(&bitmap->pending_writes);
454 set_buffer_locked(bh);
455 set_buffer_mapped(bh);
456 submit_bh(READ, bh);
458 block++;
459 bh = bh->b_this_page;
461 page->index = index;
463 wait_event(bitmap->write_wait,
464 atomic_read(&bitmap->pending_writes)==0);
465 if (bitmap->flags & BITMAP_WRITE_ERROR) {
466 free_buffers(page);
467 page = ERR_PTR(-EIO);
469 out:
470 if (IS_ERR(page))
471 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
472 (int)PAGE_SIZE,
473 (unsigned long long)index << PAGE_SHIFT,
474 PTR_ERR(page));
475 return page;
479 * bitmap file superblock operations
482 /* update the event counter and sync the superblock to disk */
483 void bitmap_update_sb(struct bitmap *bitmap)
485 bitmap_super_t *sb;
486 unsigned long flags;
488 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
489 return;
490 spin_lock_irqsave(&bitmap->lock, flags);
491 if (!bitmap->sb_page) { /* no superblock */
492 spin_unlock_irqrestore(&bitmap->lock, flags);
493 return;
495 spin_unlock_irqrestore(&bitmap->lock, flags);
496 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
497 sb->events = cpu_to_le64(bitmap->mddev->events);
498 if (bitmap->mddev->events < bitmap->events_cleared) {
499 /* rocking back to read-only */
500 bitmap->events_cleared = bitmap->mddev->events;
501 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
503 kunmap_atomic(sb, KM_USER0);
504 write_page(bitmap, bitmap->sb_page, 1);
507 /* print out the bitmap file superblock */
508 void bitmap_print_sb(struct bitmap *bitmap)
510 bitmap_super_t *sb;
512 if (!bitmap || !bitmap->sb_page)
513 return;
514 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
515 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
516 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
517 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
518 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
519 *(__u32 *)(sb->uuid+0),
520 *(__u32 *)(sb->uuid+4),
521 *(__u32 *)(sb->uuid+8),
522 *(__u32 *)(sb->uuid+12));
523 printk(KERN_DEBUG " events: %llu\n",
524 (unsigned long long) le64_to_cpu(sb->events));
525 printk(KERN_DEBUG "events cleared: %llu\n",
526 (unsigned long long) le64_to_cpu(sb->events_cleared));
527 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
528 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
529 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
530 printk(KERN_DEBUG " sync size: %llu KB\n",
531 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
532 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
533 kunmap_atomic(sb, KM_USER0);
536 /* read the superblock from the bitmap file and initialize some bitmap fields */
537 static int bitmap_read_sb(struct bitmap *bitmap)
539 char *reason = NULL;
540 bitmap_super_t *sb;
541 unsigned long chunksize, daemon_sleep, write_behind;
542 unsigned long long events;
543 int err = -EINVAL;
545 /* page 0 is the superblock, read it... */
546 if (bitmap->file) {
547 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
548 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
550 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
551 } else {
552 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset,
553 NULL,
554 0, sizeof(bitmap_super_t));
556 if (IS_ERR(bitmap->sb_page)) {
557 err = PTR_ERR(bitmap->sb_page);
558 bitmap->sb_page = NULL;
559 return err;
562 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
564 chunksize = le32_to_cpu(sb->chunksize);
565 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
566 write_behind = le32_to_cpu(sb->write_behind);
568 /* verify that the bitmap-specific fields are valid */
569 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
570 reason = "bad magic";
571 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
572 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
573 reason = "unrecognized superblock version";
574 else if (chunksize < 512)
575 reason = "bitmap chunksize too small";
576 else if ((1 << ffz(~chunksize)) != chunksize)
577 reason = "bitmap chunksize not a power of 2";
578 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
579 reason = "daemon sleep period out of range";
580 else if (write_behind > COUNTER_MAX)
581 reason = "write-behind limit out of range (0 - 16383)";
582 if (reason) {
583 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
584 bmname(bitmap), reason);
585 goto out;
588 /* keep the array size field of the bitmap superblock up to date */
589 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
591 if (!bitmap->mddev->persistent)
592 goto success;
595 * if we have a persistent array superblock, compare the
596 * bitmap's UUID and event counter to the mddev's
598 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
599 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
600 bmname(bitmap));
601 goto out;
603 events = le64_to_cpu(sb->events);
604 if (events < bitmap->mddev->events) {
605 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
606 "-- forcing full recovery\n", bmname(bitmap), events,
607 (unsigned long long) bitmap->mddev->events);
608 sb->state |= cpu_to_le32(BITMAP_STALE);
610 success:
611 /* assign fields using values from superblock */
612 bitmap->chunksize = chunksize;
613 bitmap->daemon_sleep = daemon_sleep;
614 bitmap->daemon_lastrun = jiffies;
615 bitmap->max_write_behind = write_behind;
616 bitmap->flags |= le32_to_cpu(sb->state);
617 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
618 bitmap->flags |= BITMAP_HOSTENDIAN;
619 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
620 if (sb->state & cpu_to_le32(BITMAP_STALE))
621 bitmap->events_cleared = bitmap->mddev->events;
622 err = 0;
623 out:
624 kunmap_atomic(sb, KM_USER0);
625 if (err)
626 bitmap_print_sb(bitmap);
627 return err;
630 enum bitmap_mask_op {
631 MASK_SET,
632 MASK_UNSET
635 /* record the state of the bitmap in the superblock. Return the old value */
636 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
637 enum bitmap_mask_op op)
639 bitmap_super_t *sb;
640 unsigned long flags;
641 int old;
643 spin_lock_irqsave(&bitmap->lock, flags);
644 if (!bitmap->sb_page) { /* can't set the state */
645 spin_unlock_irqrestore(&bitmap->lock, flags);
646 return 0;
648 spin_unlock_irqrestore(&bitmap->lock, flags);
649 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
650 old = le32_to_cpu(sb->state) & bits;
651 switch (op) {
652 case MASK_SET: sb->state |= cpu_to_le32(bits);
653 break;
654 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
655 break;
656 default: BUG();
658 kunmap_atomic(sb, KM_USER0);
659 return old;
663 * general bitmap file operations
666 /* calculate the index of the page that contains this bit */
667 static inline unsigned long file_page_index(unsigned long chunk)
669 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
672 /* calculate the (bit) offset of this bit within a page */
673 static inline unsigned long file_page_offset(unsigned long chunk)
675 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
679 * return a pointer to the page in the filemap that contains the given bit
681 * this lookup is complicated by the fact that the bitmap sb might be exactly
682 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
683 * 0 or page 1
685 static inline struct page *filemap_get_page(struct bitmap *bitmap,
686 unsigned long chunk)
688 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
689 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
693 static void bitmap_file_unmap(struct bitmap *bitmap)
695 struct page **map, *sb_page;
696 unsigned long *attr;
697 int pages;
698 unsigned long flags;
700 spin_lock_irqsave(&bitmap->lock, flags);
701 map = bitmap->filemap;
702 bitmap->filemap = NULL;
703 attr = bitmap->filemap_attr;
704 bitmap->filemap_attr = NULL;
705 pages = bitmap->file_pages;
706 bitmap->file_pages = 0;
707 sb_page = bitmap->sb_page;
708 bitmap->sb_page = NULL;
709 spin_unlock_irqrestore(&bitmap->lock, flags);
711 while (pages--)
712 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
713 free_buffers(map[pages]);
714 kfree(map);
715 kfree(attr);
717 if (sb_page)
718 free_buffers(sb_page);
721 static void bitmap_file_put(struct bitmap *bitmap)
723 struct file *file;
724 unsigned long flags;
726 spin_lock_irqsave(&bitmap->lock, flags);
727 file = bitmap->file;
728 bitmap->file = NULL;
729 spin_unlock_irqrestore(&bitmap->lock, flags);
731 if (file)
732 wait_event(bitmap->write_wait,
733 atomic_read(&bitmap->pending_writes)==0);
734 bitmap_file_unmap(bitmap);
736 if (file) {
737 struct inode *inode = file->f_path.dentry->d_inode;
738 invalidate_mapping_pages(inode->i_mapping, 0, -1);
739 fput(file);
745 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
746 * then it is no longer reliable, so we stop using it and we mark the file
747 * as failed in the superblock
749 static void bitmap_file_kick(struct bitmap *bitmap)
751 char *path, *ptr = NULL;
753 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
754 bitmap_update_sb(bitmap);
756 if (bitmap->file) {
757 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
758 if (path)
759 ptr = d_path(&bitmap->file->f_path, path,
760 PAGE_SIZE);
763 printk(KERN_ALERT
764 "%s: kicking failed bitmap file %s from array!\n",
765 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
767 kfree(path);
768 } else
769 printk(KERN_ALERT
770 "%s: disabling internal bitmap due to errors\n",
771 bmname(bitmap));
774 bitmap_file_put(bitmap);
776 return;
779 enum bitmap_page_attr {
780 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
781 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
782 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
785 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
786 enum bitmap_page_attr attr)
788 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
791 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
792 enum bitmap_page_attr attr)
794 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
797 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
798 enum bitmap_page_attr attr)
800 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
804 * bitmap_file_set_bit -- called before performing a write to the md device
805 * to set (and eventually sync) a particular bit in the bitmap file
807 * we set the bit immediately, then we record the page number so that
808 * when an unplug occurs, we can flush the dirty pages out to disk
810 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
812 unsigned long bit;
813 struct page *page;
814 void *kaddr;
815 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
817 if (!bitmap->filemap) {
818 return;
821 page = filemap_get_page(bitmap, chunk);
822 if (!page) return;
823 bit = file_page_offset(chunk);
825 /* set the bit */
826 kaddr = kmap_atomic(page, KM_USER0);
827 if (bitmap->flags & BITMAP_HOSTENDIAN)
828 set_bit(bit, kaddr);
829 else
830 ext2_set_bit(bit, kaddr);
831 kunmap_atomic(kaddr, KM_USER0);
832 PRINTK("set file bit %lu page %lu\n", bit, page->index);
834 /* record page number so it gets flushed to disk when unplug occurs */
835 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
839 /* this gets called when the md device is ready to unplug its underlying
840 * (slave) device queues -- before we let any writes go down, we need to
841 * sync the dirty pages of the bitmap file to disk */
842 void bitmap_unplug(struct bitmap *bitmap)
844 unsigned long i, flags;
845 int dirty, need_write;
846 struct page *page;
847 int wait = 0;
849 if (!bitmap)
850 return;
852 /* look at each page to see if there are any set bits that need to be
853 * flushed out to disk */
854 for (i = 0; i < bitmap->file_pages; i++) {
855 spin_lock_irqsave(&bitmap->lock, flags);
856 if (!bitmap->filemap) {
857 spin_unlock_irqrestore(&bitmap->lock, flags);
858 return;
860 page = bitmap->filemap[i];
861 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
862 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
863 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
864 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
865 if (dirty)
866 wait = 1;
867 spin_unlock_irqrestore(&bitmap->lock, flags);
869 if (dirty | need_write)
870 write_page(bitmap, page, 0);
872 if (wait) { /* if any writes were performed, we need to wait on them */
873 if (bitmap->file)
874 wait_event(bitmap->write_wait,
875 atomic_read(&bitmap->pending_writes)==0);
876 else
877 md_super_wait(bitmap->mddev);
879 if (bitmap->flags & BITMAP_WRITE_ERROR)
880 bitmap_file_kick(bitmap);
883 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
884 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
885 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
886 * memory mapping of the bitmap file
887 * Special cases:
888 * if there's no bitmap file, or if the bitmap file had been
889 * previously kicked from the array, we mark all the bits as
890 * 1's in order to cause a full resync.
892 * We ignore all bits for sectors that end earlier than 'start'.
893 * This is used when reading an out-of-date bitmap...
895 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
897 unsigned long i, chunks, index, oldindex, bit;
898 struct page *page = NULL, *oldpage = NULL;
899 unsigned long num_pages, bit_cnt = 0;
900 struct file *file;
901 unsigned long bytes, offset;
902 int outofdate;
903 int ret = -ENOSPC;
904 void *paddr;
906 chunks = bitmap->chunks;
907 file = bitmap->file;
909 BUG_ON(!file && !bitmap->offset);
911 #ifdef INJECT_FAULTS_3
912 outofdate = 1;
913 #else
914 outofdate = bitmap->flags & BITMAP_STALE;
915 #endif
916 if (outofdate)
917 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
918 "recovery\n", bmname(bitmap));
920 bytes = (chunks + 7) / 8;
922 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
924 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
925 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
926 bmname(bitmap),
927 (unsigned long) i_size_read(file->f_mapping->host),
928 bytes + sizeof(bitmap_super_t));
929 goto err;
932 ret = -ENOMEM;
934 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
935 if (!bitmap->filemap)
936 goto err;
938 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
939 bitmap->filemap_attr = kzalloc(
940 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
941 GFP_KERNEL);
942 if (!bitmap->filemap_attr)
943 goto err;
945 oldindex = ~0L;
947 for (i = 0; i < chunks; i++) {
948 int b;
949 index = file_page_index(i);
950 bit = file_page_offset(i);
951 if (index != oldindex) { /* this is a new page, read it in */
952 int count;
953 /* unmap the old page, we're done with it */
954 if (index == num_pages-1)
955 count = bytes + sizeof(bitmap_super_t)
956 - index * PAGE_SIZE;
957 else
958 count = PAGE_SIZE;
959 if (index == 0) {
961 * if we're here then the superblock page
962 * contains some bits (PAGE_SIZE != sizeof sb)
963 * we've already read it in, so just use it
965 page = bitmap->sb_page;
966 offset = sizeof(bitmap_super_t);
967 if (!file)
968 read_sb_page(bitmap->mddev,
969 bitmap->offset,
970 page,
971 index, count);
972 } else if (file) {
973 page = read_page(file, index, bitmap, count);
974 offset = 0;
975 } else {
976 page = read_sb_page(bitmap->mddev, bitmap->offset,
977 NULL,
978 index, count);
979 offset = 0;
981 if (IS_ERR(page)) { /* read error */
982 ret = PTR_ERR(page);
983 goto err;
986 oldindex = index;
987 oldpage = page;
989 bitmap->filemap[bitmap->file_pages++] = page;
990 bitmap->last_page_size = count;
992 if (outofdate) {
994 * if bitmap is out of date, dirty the
995 * whole page and write it out
997 paddr = kmap_atomic(page, KM_USER0);
998 memset(paddr + offset, 0xff,
999 PAGE_SIZE - offset);
1000 kunmap_atomic(paddr, KM_USER0);
1001 write_page(bitmap, page, 1);
1003 ret = -EIO;
1004 if (bitmap->flags & BITMAP_WRITE_ERROR)
1005 goto err;
1008 paddr = kmap_atomic(page, KM_USER0);
1009 if (bitmap->flags & BITMAP_HOSTENDIAN)
1010 b = test_bit(bit, paddr);
1011 else
1012 b = ext2_test_bit(bit, paddr);
1013 kunmap_atomic(paddr, KM_USER0);
1014 if (b) {
1015 /* if the disk bit is set, set the memory bit */
1016 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1017 >= start);
1018 bitmap_set_memory_bits(bitmap,
1019 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1020 needed);
1021 bit_cnt++;
1022 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1026 /* everything went OK */
1027 ret = 0;
1028 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1030 if (bit_cnt) { /* Kick recovery if any bits were set */
1031 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1032 md_wakeup_thread(bitmap->mddev->thread);
1035 printk(KERN_INFO "%s: bitmap initialized from disk: "
1036 "read %lu/%lu pages, set %lu bits\n",
1037 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1039 return 0;
1041 err:
1042 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1043 bmname(bitmap), ret);
1044 return ret;
1047 void bitmap_write_all(struct bitmap *bitmap)
1049 /* We don't actually write all bitmap blocks here,
1050 * just flag them as needing to be written
1052 int i;
1054 for (i=0; i < bitmap->file_pages; i++)
1055 set_page_attr(bitmap, bitmap->filemap[i],
1056 BITMAP_PAGE_NEEDWRITE);
1060 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1062 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1063 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1064 bitmap->bp[page].count += inc;
1066 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1067 (unsigned long long)offset, inc, bitmap->bp[page].count);
1069 bitmap_checkfree(bitmap, page);
1071 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1072 sector_t offset, int *blocks,
1073 int create);
1076 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1077 * out to disk
1080 void bitmap_daemon_work(struct bitmap *bitmap)
1082 unsigned long j;
1083 unsigned long flags;
1084 struct page *page = NULL, *lastpage = NULL;
1085 int blocks;
1086 void *paddr;
1088 if (bitmap == NULL)
1089 return;
1090 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1091 goto done;
1093 bitmap->daemon_lastrun = jiffies;
1094 if (bitmap->allclean) {
1095 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1096 return;
1098 bitmap->allclean = 1;
1100 spin_lock_irqsave(&bitmap->lock, flags);
1101 for (j = 0; j < bitmap->chunks; j++) {
1102 bitmap_counter_t *bmc;
1103 if (!bitmap->filemap)
1104 /* error or shutdown */
1105 break;
1107 page = filemap_get_page(bitmap, j);
1109 if (page != lastpage) {
1110 /* skip this page unless it's marked as needing cleaning */
1111 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1112 int need_write = test_page_attr(bitmap, page,
1113 BITMAP_PAGE_NEEDWRITE);
1114 if (need_write)
1115 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1117 spin_unlock_irqrestore(&bitmap->lock, flags);
1118 if (need_write) {
1119 write_page(bitmap, page, 0);
1120 bitmap->allclean = 0;
1122 spin_lock_irqsave(&bitmap->lock, flags);
1123 j |= (PAGE_BITS - 1);
1124 continue;
1127 /* grab the new page, sync and release the old */
1128 if (lastpage != NULL) {
1129 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1130 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1131 spin_unlock_irqrestore(&bitmap->lock, flags);
1132 write_page(bitmap, lastpage, 0);
1133 } else {
1134 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1135 spin_unlock_irqrestore(&bitmap->lock, flags);
1137 } else
1138 spin_unlock_irqrestore(&bitmap->lock, flags);
1139 lastpage = page;
1141 /* We are possibly going to clear some bits, so make
1142 * sure that events_cleared is up-to-date.
1144 if (bitmap->need_sync) {
1145 bitmap_super_t *sb;
1146 bitmap->need_sync = 0;
1147 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1148 sb->events_cleared =
1149 cpu_to_le64(bitmap->events_cleared);
1150 kunmap_atomic(sb, KM_USER0);
1151 write_page(bitmap, bitmap->sb_page, 1);
1153 spin_lock_irqsave(&bitmap->lock, flags);
1154 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1156 bmc = bitmap_get_counter(bitmap,
1157 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1158 &blocks, 0);
1159 if (bmc) {
1161 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1163 if (*bmc)
1164 bitmap->allclean = 0;
1166 if (*bmc == 2) {
1167 *bmc=1; /* maybe clear the bit next time */
1168 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1169 } else if (*bmc == 1) {
1170 /* we can clear the bit */
1171 *bmc = 0;
1172 bitmap_count_page(bitmap,
1173 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1174 -1);
1176 /* clear the bit */
1177 paddr = kmap_atomic(page, KM_USER0);
1178 if (bitmap->flags & BITMAP_HOSTENDIAN)
1179 clear_bit(file_page_offset(j), paddr);
1180 else
1181 ext2_clear_bit(file_page_offset(j), paddr);
1182 kunmap_atomic(paddr, KM_USER0);
1184 } else
1185 j |= PAGE_COUNTER_MASK;
1187 spin_unlock_irqrestore(&bitmap->lock, flags);
1189 /* now sync the final page */
1190 if (lastpage != NULL) {
1191 spin_lock_irqsave(&bitmap->lock, flags);
1192 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1193 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1194 spin_unlock_irqrestore(&bitmap->lock, flags);
1195 write_page(bitmap, lastpage, 0);
1196 } else {
1197 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1198 spin_unlock_irqrestore(&bitmap->lock, flags);
1202 done:
1203 if (bitmap->allclean == 0)
1204 bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1207 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1208 sector_t offset, int *blocks,
1209 int create)
1211 /* If 'create', we might release the lock and reclaim it.
1212 * The lock must have been taken with interrupts enabled.
1213 * If !create, we don't release the lock.
1215 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1216 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1217 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1218 sector_t csize;
1220 if (bitmap_checkpage(bitmap, page, create) < 0) {
1221 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1222 *blocks = csize - (offset & (csize- 1));
1223 return NULL;
1225 /* now locked ... */
1227 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1228 /* should we use the first or second counter field
1229 * of the hijacked pointer? */
1230 int hi = (pageoff > PAGE_COUNTER_MASK);
1231 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1232 PAGE_COUNTER_SHIFT - 1);
1233 *blocks = csize - (offset & (csize- 1));
1234 return &((bitmap_counter_t *)
1235 &bitmap->bp[page].map)[hi];
1236 } else { /* page is allocated */
1237 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1238 *blocks = csize - (offset & (csize- 1));
1239 return (bitmap_counter_t *)
1240 &(bitmap->bp[page].map[pageoff]);
1244 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1246 if (!bitmap) return 0;
1248 if (behind) {
1249 atomic_inc(&bitmap->behind_writes);
1250 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1251 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1254 while (sectors) {
1255 int blocks;
1256 bitmap_counter_t *bmc;
1258 spin_lock_irq(&bitmap->lock);
1259 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1260 if (!bmc) {
1261 spin_unlock_irq(&bitmap->lock);
1262 return 0;
1265 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1266 DEFINE_WAIT(__wait);
1267 /* note that it is safe to do the prepare_to_wait
1268 * after the test as long as we do it before dropping
1269 * the spinlock.
1271 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1272 TASK_UNINTERRUPTIBLE);
1273 spin_unlock_irq(&bitmap->lock);
1274 blk_unplug(bitmap->mddev->queue);
1275 schedule();
1276 finish_wait(&bitmap->overflow_wait, &__wait);
1277 continue;
1280 switch(*bmc) {
1281 case 0:
1282 bitmap_file_set_bit(bitmap, offset);
1283 bitmap_count_page(bitmap,offset, 1);
1284 blk_plug_device_unlocked(bitmap->mddev->queue);
1285 /* fall through */
1286 case 1:
1287 *bmc = 2;
1290 (*bmc)++;
1292 spin_unlock_irq(&bitmap->lock);
1294 offset += blocks;
1295 if (sectors > blocks)
1296 sectors -= blocks;
1297 else sectors = 0;
1299 bitmap->allclean = 0;
1300 return 0;
1303 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1304 int success, int behind)
1306 if (!bitmap) return;
1307 if (behind) {
1308 atomic_dec(&bitmap->behind_writes);
1309 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1310 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1312 if (bitmap->mddev->degraded)
1313 /* Never clear bits or update events_cleared when degraded */
1314 success = 0;
1316 while (sectors) {
1317 int blocks;
1318 unsigned long flags;
1319 bitmap_counter_t *bmc;
1321 spin_lock_irqsave(&bitmap->lock, flags);
1322 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1323 if (!bmc) {
1324 spin_unlock_irqrestore(&bitmap->lock, flags);
1325 return;
1328 if (success &&
1329 bitmap->events_cleared < bitmap->mddev->events) {
1330 bitmap->events_cleared = bitmap->mddev->events;
1331 bitmap->need_sync = 1;
1334 if (!success && ! (*bmc & NEEDED_MASK))
1335 *bmc |= NEEDED_MASK;
1337 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1338 wake_up(&bitmap->overflow_wait);
1340 (*bmc)--;
1341 if (*bmc <= 2) {
1342 set_page_attr(bitmap,
1343 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1344 BITMAP_PAGE_CLEAN);
1346 spin_unlock_irqrestore(&bitmap->lock, flags);
1347 offset += blocks;
1348 if (sectors > blocks)
1349 sectors -= blocks;
1350 else sectors = 0;
1354 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1355 int degraded)
1357 bitmap_counter_t *bmc;
1358 int rv;
1359 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1360 *blocks = 1024;
1361 return 1; /* always resync if no bitmap */
1363 spin_lock_irq(&bitmap->lock);
1364 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1365 rv = 0;
1366 if (bmc) {
1367 /* locked */
1368 if (RESYNC(*bmc))
1369 rv = 1;
1370 else if (NEEDED(*bmc)) {
1371 rv = 1;
1372 if (!degraded) { /* don't set/clear bits if degraded */
1373 *bmc |= RESYNC_MASK;
1374 *bmc &= ~NEEDED_MASK;
1378 spin_unlock_irq(&bitmap->lock);
1379 bitmap->allclean = 0;
1380 return rv;
1383 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1384 int degraded)
1386 /* bitmap_start_sync must always report on multiples of whole
1387 * pages, otherwise resync (which is very PAGE_SIZE based) will
1388 * get confused.
1389 * So call __bitmap_start_sync repeatedly (if needed) until
1390 * At least PAGE_SIZE>>9 blocks are covered.
1391 * Return the 'or' of the result.
1393 int rv = 0;
1394 int blocks1;
1396 *blocks = 0;
1397 while (*blocks < (PAGE_SIZE>>9)) {
1398 rv |= __bitmap_start_sync(bitmap, offset,
1399 &blocks1, degraded);
1400 offset += blocks1;
1401 *blocks += blocks1;
1403 return rv;
1406 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1408 bitmap_counter_t *bmc;
1409 unsigned long flags;
1411 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1412 */ if (bitmap == NULL) {
1413 *blocks = 1024;
1414 return;
1416 spin_lock_irqsave(&bitmap->lock, flags);
1417 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1418 if (bmc == NULL)
1419 goto unlock;
1420 /* locked */
1422 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1424 if (RESYNC(*bmc)) {
1425 *bmc &= ~RESYNC_MASK;
1427 if (!NEEDED(*bmc) && aborted)
1428 *bmc |= NEEDED_MASK;
1429 else {
1430 if (*bmc <= 2) {
1431 set_page_attr(bitmap,
1432 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1433 BITMAP_PAGE_CLEAN);
1437 unlock:
1438 spin_unlock_irqrestore(&bitmap->lock, flags);
1439 bitmap->allclean = 0;
1442 void bitmap_close_sync(struct bitmap *bitmap)
1444 /* Sync has finished, and any bitmap chunks that weren't synced
1445 * properly have been aborted. It remains to us to clear the
1446 * RESYNC bit wherever it is still on
1448 sector_t sector = 0;
1449 int blocks;
1450 if (!bitmap)
1451 return;
1452 while (sector < bitmap->mddev->resync_max_sectors) {
1453 bitmap_end_sync(bitmap, sector, &blocks, 0);
1454 sector += blocks;
1458 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1460 sector_t s = 0;
1461 int blocks;
1463 if (!bitmap)
1464 return;
1465 if (sector == 0) {
1466 bitmap->last_end_sync = jiffies;
1467 return;
1469 if (time_before(jiffies, (bitmap->last_end_sync
1470 + bitmap->daemon_sleep * HZ)))
1471 return;
1472 wait_event(bitmap->mddev->recovery_wait,
1473 atomic_read(&bitmap->mddev->recovery_active) == 0);
1475 bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync;
1476 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1477 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1478 s = 0;
1479 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1480 bitmap_end_sync(bitmap, s, &blocks, 0);
1481 s += blocks;
1483 bitmap->last_end_sync = jiffies;
1484 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1487 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1489 /* For each chunk covered by any of these sectors, set the
1490 * counter to 1 and set resync_needed. They should all
1491 * be 0 at this point
1494 int secs;
1495 bitmap_counter_t *bmc;
1496 spin_lock_irq(&bitmap->lock);
1497 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1498 if (!bmc) {
1499 spin_unlock_irq(&bitmap->lock);
1500 return;
1502 if (! *bmc) {
1503 struct page *page;
1504 *bmc = 1 | (needed?NEEDED_MASK:0);
1505 bitmap_count_page(bitmap, offset, 1);
1506 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1507 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1509 spin_unlock_irq(&bitmap->lock);
1510 bitmap->allclean = 0;
1513 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1514 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1516 unsigned long chunk;
1518 for (chunk = s; chunk <= e; chunk++) {
1519 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
1520 bitmap_set_memory_bits(bitmap, sec, 1);
1521 bitmap_file_set_bit(bitmap, sec);
1526 * flush out any pending updates
1528 void bitmap_flush(mddev_t *mddev)
1530 struct bitmap *bitmap = mddev->bitmap;
1531 int sleep;
1533 if (!bitmap) /* there was no bitmap */
1534 return;
1536 /* run the daemon_work three time to ensure everything is flushed
1537 * that can be
1539 sleep = bitmap->daemon_sleep;
1540 bitmap->daemon_sleep = 0;
1541 bitmap_daemon_work(bitmap);
1542 bitmap_daemon_work(bitmap);
1543 bitmap_daemon_work(bitmap);
1544 bitmap->daemon_sleep = sleep;
1545 bitmap_update_sb(bitmap);
1549 * free memory that was allocated
1551 static void bitmap_free(struct bitmap *bitmap)
1553 unsigned long k, pages;
1554 struct bitmap_page *bp;
1556 if (!bitmap) /* there was no bitmap */
1557 return;
1559 /* release the bitmap file and kill the daemon */
1560 bitmap_file_put(bitmap);
1562 bp = bitmap->bp;
1563 pages = bitmap->pages;
1565 /* free all allocated memory */
1567 if (bp) /* deallocate the page memory */
1568 for (k = 0; k < pages; k++)
1569 if (bp[k].map && !bp[k].hijacked)
1570 kfree(bp[k].map);
1571 kfree(bp);
1572 kfree(bitmap);
1574 void bitmap_destroy(mddev_t *mddev)
1576 struct bitmap *bitmap = mddev->bitmap;
1578 if (!bitmap) /* there was no bitmap */
1579 return;
1581 mddev->bitmap = NULL; /* disconnect from the md device */
1582 if (mddev->thread)
1583 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1585 bitmap_free(bitmap);
1589 * initialize the bitmap structure
1590 * if this returns an error, bitmap_destroy must be called to do clean up
1592 int bitmap_create(mddev_t *mddev)
1594 struct bitmap *bitmap;
1595 sector_t blocks = mddev->resync_max_sectors;
1596 unsigned long chunks;
1597 unsigned long pages;
1598 struct file *file = mddev->bitmap_file;
1599 int err;
1600 sector_t start;
1602 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1604 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1605 return 0;
1607 BUG_ON(file && mddev->bitmap_offset);
1609 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1610 if (!bitmap)
1611 return -ENOMEM;
1613 spin_lock_init(&bitmap->lock);
1614 atomic_set(&bitmap->pending_writes, 0);
1615 init_waitqueue_head(&bitmap->write_wait);
1616 init_waitqueue_head(&bitmap->overflow_wait);
1618 bitmap->mddev = mddev;
1620 bitmap->file = file;
1621 bitmap->offset = mddev->bitmap_offset;
1622 if (file) {
1623 get_file(file);
1624 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1625 SYNC_FILE_RANGE_WAIT_BEFORE |
1626 SYNC_FILE_RANGE_WRITE |
1627 SYNC_FILE_RANGE_WAIT_AFTER);
1629 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1630 err = bitmap_read_sb(bitmap);
1631 if (err)
1632 goto error;
1634 bitmap->chunkshift = ffz(~bitmap->chunksize);
1636 /* now that chunksize and chunkshift are set, we can use these macros */
1637 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1638 CHUNK_BLOCK_SHIFT(bitmap);
1639 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1641 BUG_ON(!pages);
1643 bitmap->chunks = chunks;
1644 bitmap->pages = pages;
1645 bitmap->missing_pages = pages;
1646 bitmap->counter_bits = COUNTER_BITS;
1648 bitmap->syncchunk = ~0UL;
1650 #ifdef INJECT_FATAL_FAULT_1
1651 bitmap->bp = NULL;
1652 #else
1653 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1654 #endif
1655 err = -ENOMEM;
1656 if (!bitmap->bp)
1657 goto error;
1659 /* now that we have some pages available, initialize the in-memory
1660 * bitmap from the on-disk bitmap */
1661 start = 0;
1662 if (mddev->degraded == 0
1663 || bitmap->events_cleared == mddev->events)
1664 /* no need to keep dirty bits to optimise a re-add of a missing device */
1665 start = mddev->recovery_cp;
1666 err = bitmap_init_from_disk(bitmap, start);
1668 if (err)
1669 goto error;
1671 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1672 pages, bmname(bitmap));
1674 mddev->bitmap = bitmap;
1676 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1678 bitmap_update_sb(bitmap);
1680 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1682 error:
1683 bitmap_free(bitmap);
1684 return err;
1687 /* the bitmap API -- for raid personalities */
1688 EXPORT_SYMBOL(bitmap_startwrite);
1689 EXPORT_SYMBOL(bitmap_endwrite);
1690 EXPORT_SYMBOL(bitmap_start_sync);
1691 EXPORT_SYMBOL(bitmap_end_sync);
1692 EXPORT_SYMBOL(bitmap_unplug);
1693 EXPORT_SYMBOL(bitmap_close_sync);
1694 EXPORT_SYMBOL(bitmap_cond_end_sync);