md: support bitmap offset appropriate for external-metadata arrays.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / bitmap.c
blob958865445f0643a07184de100a3921d4289e3fdf
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)
111 __releases(bitmap->lock)
112 __acquires(bitmap->lock)
114 unsigned char *mappage;
116 if (page >= bitmap->pages) {
117 /* This can happen if bitmap_start_sync goes beyond
118 * End-of-device while looking for a whole page.
119 * It is harmless.
121 return -EINVAL;
125 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
126 return 0;
128 if (bitmap->bp[page].map) /* page is already allocated, just return */
129 return 0;
131 if (!create)
132 return -ENOENT;
134 spin_unlock_irq(&bitmap->lock);
136 /* this page has not been allocated yet */
138 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
139 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
140 bmname(bitmap));
141 /* failed - set the hijacked flag so that we can use the
142 * pointer as a counter */
143 spin_lock_irq(&bitmap->lock);
144 if (!bitmap->bp[page].map)
145 bitmap->bp[page].hijacked = 1;
146 goto out;
149 /* got a page */
151 spin_lock_irq(&bitmap->lock);
153 /* recheck the page */
155 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
156 /* somebody beat us to getting the page */
157 bitmap_free_page(bitmap, mappage);
158 return 0;
161 /* no page was in place and we have one, so install it */
163 memset(mappage, 0, PAGE_SIZE);
164 bitmap->bp[page].map = mappage;
165 bitmap->missing_pages--;
166 out:
167 return 0;
171 /* if page is completely empty, put it back on the free list, or dealloc it */
172 /* if page was hijacked, unmark the flag so it might get alloced next time */
173 /* Note: lock should be held when calling this */
174 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
176 char *ptr;
178 if (bitmap->bp[page].count) /* page is still busy */
179 return;
181 /* page is no longer in use, it can be released */
183 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
184 bitmap->bp[page].hijacked = 0;
185 bitmap->bp[page].map = NULL;
186 return;
189 /* normal case, free the page */
191 #if 0
192 /* actually ... let's not. We will probably need the page again exactly when
193 * memory is tight and we are flusing to disk
195 return;
196 #else
197 ptr = bitmap->bp[page].map;
198 bitmap->bp[page].map = NULL;
199 bitmap->missing_pages++;
200 bitmap_free_page(bitmap, ptr);
201 return;
202 #endif
207 * bitmap file handling - read and write the bitmap file and its superblock
211 * basic page I/O operations
214 /* IO operations when bitmap is stored near all superblocks */
215 static struct page *read_sb_page(mddev_t *mddev, loff_t offset,
216 struct page *page,
217 unsigned long index, int size)
219 /* choose a good rdev and read the page from there */
221 mdk_rdev_t *rdev;
222 sector_t target;
224 if (!page)
225 page = alloc_page(GFP_KERNEL);
226 if (!page)
227 return ERR_PTR(-ENOMEM);
229 list_for_each_entry(rdev, &mddev->disks, same_set) {
230 if (! test_bit(In_sync, &rdev->flags)
231 || test_bit(Faulty, &rdev->flags))
232 continue;
234 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
236 if (sync_page_io(rdev->bdev, target,
237 roundup(size, bdev_logical_block_size(rdev->bdev)),
238 page, READ)) {
239 page->index = index;
240 attach_page_buffers(page, NULL); /* so that free_buffer will
241 * quietly no-op */
242 return page;
245 return ERR_PTR(-EIO);
249 static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
251 /* Iterate the disks of an mddev, using rcu to protect access to the
252 * linked list, and raising the refcount of devices we return to ensure
253 * they don't disappear while in use.
254 * As devices are only added or removed when raid_disk is < 0 and
255 * nr_pending is 0 and In_sync is clear, the entries we return will
256 * still be in the same position on the list when we re-enter
257 * list_for_each_continue_rcu.
259 struct list_head *pos;
260 rcu_read_lock();
261 if (rdev == NULL)
262 /* start at the beginning */
263 pos = &mddev->disks;
264 else {
265 /* release the previous rdev and start from there. */
266 rdev_dec_pending(rdev, mddev);
267 pos = &rdev->same_set;
269 list_for_each_continue_rcu(pos, &mddev->disks) {
270 rdev = list_entry(pos, mdk_rdev_t, same_set);
271 if (rdev->raid_disk >= 0 &&
272 !test_bit(Faulty, &rdev->flags)) {
273 /* this is a usable devices */
274 atomic_inc(&rdev->nr_pending);
275 rcu_read_unlock();
276 return rdev;
279 rcu_read_unlock();
280 return NULL;
283 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
285 mdk_rdev_t *rdev = NULL;
286 mddev_t *mddev = bitmap->mddev;
288 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
289 int size = PAGE_SIZE;
290 loff_t offset = mddev->bitmap_info.offset;
291 if (page->index == bitmap->file_pages-1)
292 size = roundup(bitmap->last_page_size,
293 bdev_logical_block_size(rdev->bdev));
294 /* Just make sure we aren't corrupting data or
295 * metadata
297 if (mddev->external) {
298 /* Bitmap could be anywhere. */
299 if (rdev->sb_start + offset + (page->index *(PAGE_SIZE/512)) >
300 rdev->data_offset &&
301 rdev->sb_start + offset <
302 rdev->data_offset + mddev->dev_sectors +
303 (PAGE_SIZE/512))
304 goto bad_alignment;
305 } else if (offset < 0) {
306 /* DATA BITMAP METADATA */
307 if (offset
308 + (long)(page->index * (PAGE_SIZE/512))
309 + size/512 > 0)
310 /* bitmap runs in to metadata */
311 goto bad_alignment;
312 if (rdev->data_offset + mddev->dev_sectors
313 > rdev->sb_start + offset)
314 /* data runs in to bitmap */
315 goto bad_alignment;
316 } else if (rdev->sb_start < rdev->data_offset) {
317 /* METADATA BITMAP DATA */
318 if (rdev->sb_start
319 + offset
320 + page->index*(PAGE_SIZE/512) + size/512
321 > rdev->data_offset)
322 /* bitmap runs in to data */
323 goto bad_alignment;
324 } else {
325 /* DATA METADATA BITMAP - no problems */
327 md_super_write(mddev, rdev,
328 rdev->sb_start + offset
329 + page->index * (PAGE_SIZE/512),
330 size,
331 page);
334 if (wait)
335 md_super_wait(mddev);
336 return 0;
338 bad_alignment:
339 return -EINVAL;
342 static void bitmap_file_kick(struct bitmap *bitmap);
344 * write out a page to a file
346 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
348 struct buffer_head *bh;
350 if (bitmap->file == NULL) {
351 switch (write_sb_page(bitmap, page, wait)) {
352 case -EINVAL:
353 bitmap->flags |= BITMAP_WRITE_ERROR;
355 } else {
357 bh = page_buffers(page);
359 while (bh && bh->b_blocknr) {
360 atomic_inc(&bitmap->pending_writes);
361 set_buffer_locked(bh);
362 set_buffer_mapped(bh);
363 submit_bh(WRITE, bh);
364 bh = bh->b_this_page;
367 if (wait) {
368 wait_event(bitmap->write_wait,
369 atomic_read(&bitmap->pending_writes)==0);
372 if (bitmap->flags & BITMAP_WRITE_ERROR)
373 bitmap_file_kick(bitmap);
376 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
378 struct bitmap *bitmap = bh->b_private;
379 unsigned long flags;
381 if (!uptodate) {
382 spin_lock_irqsave(&bitmap->lock, flags);
383 bitmap->flags |= BITMAP_WRITE_ERROR;
384 spin_unlock_irqrestore(&bitmap->lock, flags);
386 if (atomic_dec_and_test(&bitmap->pending_writes))
387 wake_up(&bitmap->write_wait);
390 /* copied from buffer.c */
391 static void
392 __clear_page_buffers(struct page *page)
394 ClearPagePrivate(page);
395 set_page_private(page, 0);
396 page_cache_release(page);
398 static void free_buffers(struct page *page)
400 struct buffer_head *bh = page_buffers(page);
402 while (bh) {
403 struct buffer_head *next = bh->b_this_page;
404 free_buffer_head(bh);
405 bh = next;
407 __clear_page_buffers(page);
408 put_page(page);
411 /* read a page from a file.
412 * We both read the page, and attach buffers to the page to record the
413 * address of each block (using bmap). These addresses will be used
414 * to write the block later, completely bypassing the filesystem.
415 * This usage is similar to how swap files are handled, and allows us
416 * to write to a file with no concerns of memory allocation failing.
418 static struct page *read_page(struct file *file, unsigned long index,
419 struct bitmap *bitmap,
420 unsigned long count)
422 struct page *page = NULL;
423 struct inode *inode = file->f_path.dentry->d_inode;
424 struct buffer_head *bh;
425 sector_t block;
427 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
428 (unsigned long long)index << PAGE_SHIFT);
430 page = alloc_page(GFP_KERNEL);
431 if (!page)
432 page = ERR_PTR(-ENOMEM);
433 if (IS_ERR(page))
434 goto out;
436 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
437 if (!bh) {
438 put_page(page);
439 page = ERR_PTR(-ENOMEM);
440 goto out;
442 attach_page_buffers(page, bh);
443 block = index << (PAGE_SHIFT - inode->i_blkbits);
444 while (bh) {
445 if (count == 0)
446 bh->b_blocknr = 0;
447 else {
448 bh->b_blocknr = bmap(inode, block);
449 if (bh->b_blocknr == 0) {
450 /* Cannot use this file! */
451 free_buffers(page);
452 page = ERR_PTR(-EINVAL);
453 goto out;
455 bh->b_bdev = inode->i_sb->s_bdev;
456 if (count < (1<<inode->i_blkbits))
457 count = 0;
458 else
459 count -= (1<<inode->i_blkbits);
461 bh->b_end_io = end_bitmap_write;
462 bh->b_private = bitmap;
463 atomic_inc(&bitmap->pending_writes);
464 set_buffer_locked(bh);
465 set_buffer_mapped(bh);
466 submit_bh(READ, bh);
468 block++;
469 bh = bh->b_this_page;
471 page->index = index;
473 wait_event(bitmap->write_wait,
474 atomic_read(&bitmap->pending_writes)==0);
475 if (bitmap->flags & BITMAP_WRITE_ERROR) {
476 free_buffers(page);
477 page = ERR_PTR(-EIO);
479 out:
480 if (IS_ERR(page))
481 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
482 (int)PAGE_SIZE,
483 (unsigned long long)index << PAGE_SHIFT,
484 PTR_ERR(page));
485 return page;
489 * bitmap file superblock operations
492 /* update the event counter and sync the superblock to disk */
493 void bitmap_update_sb(struct bitmap *bitmap)
495 bitmap_super_t *sb;
496 unsigned long flags;
498 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
499 return;
500 spin_lock_irqsave(&bitmap->lock, flags);
501 if (!bitmap->sb_page) { /* no superblock */
502 spin_unlock_irqrestore(&bitmap->lock, flags);
503 return;
505 spin_unlock_irqrestore(&bitmap->lock, flags);
506 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
507 sb->events = cpu_to_le64(bitmap->mddev->events);
508 if (bitmap->mddev->events < bitmap->events_cleared) {
509 /* rocking back to read-only */
510 bitmap->events_cleared = bitmap->mddev->events;
511 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
513 kunmap_atomic(sb, KM_USER0);
514 write_page(bitmap, bitmap->sb_page, 1);
517 /* print out the bitmap file superblock */
518 void bitmap_print_sb(struct bitmap *bitmap)
520 bitmap_super_t *sb;
522 if (!bitmap || !bitmap->sb_page)
523 return;
524 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
525 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
526 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
527 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
528 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
529 *(__u32 *)(sb->uuid+0),
530 *(__u32 *)(sb->uuid+4),
531 *(__u32 *)(sb->uuid+8),
532 *(__u32 *)(sb->uuid+12));
533 printk(KERN_DEBUG " events: %llu\n",
534 (unsigned long long) le64_to_cpu(sb->events));
535 printk(KERN_DEBUG "events cleared: %llu\n",
536 (unsigned long long) le64_to_cpu(sb->events_cleared));
537 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
538 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
539 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
540 printk(KERN_DEBUG " sync size: %llu KB\n",
541 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
542 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
543 kunmap_atomic(sb, KM_USER0);
546 /* read the superblock from the bitmap file and initialize some bitmap fields */
547 static int bitmap_read_sb(struct bitmap *bitmap)
549 char *reason = NULL;
550 bitmap_super_t *sb;
551 unsigned long chunksize, daemon_sleep, write_behind;
552 unsigned long long events;
553 int err = -EINVAL;
555 /* page 0 is the superblock, read it... */
556 if (bitmap->file) {
557 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
558 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
560 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
561 } else {
562 bitmap->sb_page = read_sb_page(bitmap->mddev,
563 bitmap->mddev->bitmap_info.offset,
564 NULL,
565 0, sizeof(bitmap_super_t));
567 if (IS_ERR(bitmap->sb_page)) {
568 err = PTR_ERR(bitmap->sb_page);
569 bitmap->sb_page = NULL;
570 return err;
573 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
575 chunksize = le32_to_cpu(sb->chunksize);
576 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
577 write_behind = le32_to_cpu(sb->write_behind);
579 /* verify that the bitmap-specific fields are valid */
580 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
581 reason = "bad magic";
582 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
583 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
584 reason = "unrecognized superblock version";
585 else if (chunksize < 512)
586 reason = "bitmap chunksize too small";
587 else if ((1 << ffz(~chunksize)) != chunksize)
588 reason = "bitmap chunksize not a power of 2";
589 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
590 reason = "daemon sleep period out of range";
591 else if (write_behind > COUNTER_MAX)
592 reason = "write-behind limit out of range (0 - 16383)";
593 if (reason) {
594 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
595 bmname(bitmap), reason);
596 goto out;
599 /* keep the array size field of the bitmap superblock up to date */
600 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
602 if (!bitmap->mddev->persistent)
603 goto success;
606 * if we have a persistent array superblock, compare the
607 * bitmap's UUID and event counter to the mddev's
609 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
610 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
611 bmname(bitmap));
612 goto out;
614 events = le64_to_cpu(sb->events);
615 if (events < bitmap->mddev->events) {
616 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
617 "-- forcing full recovery\n", bmname(bitmap), events,
618 (unsigned long long) bitmap->mddev->events);
619 sb->state |= cpu_to_le32(BITMAP_STALE);
621 success:
622 /* assign fields using values from superblock */
623 bitmap->mddev->bitmap_info.chunksize = chunksize;
624 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
625 bitmap->daemon_lastrun = jiffies;
626 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
627 bitmap->flags |= le32_to_cpu(sb->state);
628 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
629 bitmap->flags |= BITMAP_HOSTENDIAN;
630 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
631 if (sb->state & cpu_to_le32(BITMAP_STALE))
632 bitmap->events_cleared = bitmap->mddev->events;
633 err = 0;
634 out:
635 kunmap_atomic(sb, KM_USER0);
636 if (err)
637 bitmap_print_sb(bitmap);
638 return err;
641 enum bitmap_mask_op {
642 MASK_SET,
643 MASK_UNSET
646 /* record the state of the bitmap in the superblock. Return the old value */
647 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
648 enum bitmap_mask_op op)
650 bitmap_super_t *sb;
651 unsigned long flags;
652 int old;
654 spin_lock_irqsave(&bitmap->lock, flags);
655 if (!bitmap->sb_page) { /* can't set the state */
656 spin_unlock_irqrestore(&bitmap->lock, flags);
657 return 0;
659 spin_unlock_irqrestore(&bitmap->lock, flags);
660 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
661 old = le32_to_cpu(sb->state) & bits;
662 switch (op) {
663 case MASK_SET: sb->state |= cpu_to_le32(bits);
664 break;
665 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
666 break;
667 default: BUG();
669 kunmap_atomic(sb, KM_USER0);
670 return old;
674 * general bitmap file operations
677 /* calculate the index of the page that contains this bit */
678 static inline unsigned long file_page_index(unsigned long chunk)
680 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
683 /* calculate the (bit) offset of this bit within a page */
684 static inline unsigned long file_page_offset(unsigned long chunk)
686 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
690 * return a pointer to the page in the filemap that contains the given bit
692 * this lookup is complicated by the fact that the bitmap sb might be exactly
693 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
694 * 0 or page 1
696 static inline struct page *filemap_get_page(struct bitmap *bitmap,
697 unsigned long chunk)
699 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
700 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
704 static void bitmap_file_unmap(struct bitmap *bitmap)
706 struct page **map, *sb_page;
707 unsigned long *attr;
708 int pages;
709 unsigned long flags;
711 spin_lock_irqsave(&bitmap->lock, flags);
712 map = bitmap->filemap;
713 bitmap->filemap = NULL;
714 attr = bitmap->filemap_attr;
715 bitmap->filemap_attr = NULL;
716 pages = bitmap->file_pages;
717 bitmap->file_pages = 0;
718 sb_page = bitmap->sb_page;
719 bitmap->sb_page = NULL;
720 spin_unlock_irqrestore(&bitmap->lock, flags);
722 while (pages--)
723 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
724 free_buffers(map[pages]);
725 kfree(map);
726 kfree(attr);
728 if (sb_page)
729 free_buffers(sb_page);
732 static void bitmap_file_put(struct bitmap *bitmap)
734 struct file *file;
735 unsigned long flags;
737 spin_lock_irqsave(&bitmap->lock, flags);
738 file = bitmap->file;
739 bitmap->file = NULL;
740 spin_unlock_irqrestore(&bitmap->lock, flags);
742 if (file)
743 wait_event(bitmap->write_wait,
744 atomic_read(&bitmap->pending_writes)==0);
745 bitmap_file_unmap(bitmap);
747 if (file) {
748 struct inode *inode = file->f_path.dentry->d_inode;
749 invalidate_mapping_pages(inode->i_mapping, 0, -1);
750 fput(file);
756 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
757 * then it is no longer reliable, so we stop using it and we mark the file
758 * as failed in the superblock
760 static void bitmap_file_kick(struct bitmap *bitmap)
762 char *path, *ptr = NULL;
764 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
765 bitmap_update_sb(bitmap);
767 if (bitmap->file) {
768 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
769 if (path)
770 ptr = d_path(&bitmap->file->f_path, path,
771 PAGE_SIZE);
774 printk(KERN_ALERT
775 "%s: kicking failed bitmap file %s from array!\n",
776 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
778 kfree(path);
779 } else
780 printk(KERN_ALERT
781 "%s: disabling internal bitmap due to errors\n",
782 bmname(bitmap));
785 bitmap_file_put(bitmap);
787 return;
790 enum bitmap_page_attr {
791 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
792 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
793 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
796 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
797 enum bitmap_page_attr attr)
799 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
802 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
803 enum bitmap_page_attr attr)
805 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
808 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
809 enum bitmap_page_attr attr)
811 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
815 * bitmap_file_set_bit -- called before performing a write to the md device
816 * to set (and eventually sync) a particular bit in the bitmap file
818 * we set the bit immediately, then we record the page number so that
819 * when an unplug occurs, we can flush the dirty pages out to disk
821 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
823 unsigned long bit;
824 struct page *page;
825 void *kaddr;
826 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
828 if (!bitmap->filemap) {
829 return;
832 page = filemap_get_page(bitmap, chunk);
833 if (!page) return;
834 bit = file_page_offset(chunk);
836 /* set the bit */
837 kaddr = kmap_atomic(page, KM_USER0);
838 if (bitmap->flags & BITMAP_HOSTENDIAN)
839 set_bit(bit, kaddr);
840 else
841 ext2_set_bit(bit, kaddr);
842 kunmap_atomic(kaddr, KM_USER0);
843 PRINTK("set file bit %lu page %lu\n", bit, page->index);
845 /* record page number so it gets flushed to disk when unplug occurs */
846 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
850 /* this gets called when the md device is ready to unplug its underlying
851 * (slave) device queues -- before we let any writes go down, we need to
852 * sync the dirty pages of the bitmap file to disk */
853 void bitmap_unplug(struct bitmap *bitmap)
855 unsigned long i, flags;
856 int dirty, need_write;
857 struct page *page;
858 int wait = 0;
860 if (!bitmap)
861 return;
863 /* look at each page to see if there are any set bits that need to be
864 * flushed out to disk */
865 for (i = 0; i < bitmap->file_pages; i++) {
866 spin_lock_irqsave(&bitmap->lock, flags);
867 if (!bitmap->filemap) {
868 spin_unlock_irqrestore(&bitmap->lock, flags);
869 return;
871 page = bitmap->filemap[i];
872 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
873 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
874 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
875 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
876 if (dirty)
877 wait = 1;
878 spin_unlock_irqrestore(&bitmap->lock, flags);
880 if (dirty | need_write)
881 write_page(bitmap, page, 0);
883 if (wait) { /* if any writes were performed, we need to wait on them */
884 if (bitmap->file)
885 wait_event(bitmap->write_wait,
886 atomic_read(&bitmap->pending_writes)==0);
887 else
888 md_super_wait(bitmap->mddev);
890 if (bitmap->flags & BITMAP_WRITE_ERROR)
891 bitmap_file_kick(bitmap);
894 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
895 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
896 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
897 * memory mapping of the bitmap file
898 * Special cases:
899 * if there's no bitmap file, or if the bitmap file had been
900 * previously kicked from the array, we mark all the bits as
901 * 1's in order to cause a full resync.
903 * We ignore all bits for sectors that end earlier than 'start'.
904 * This is used when reading an out-of-date bitmap...
906 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
908 unsigned long i, chunks, index, oldindex, bit;
909 struct page *page = NULL, *oldpage = NULL;
910 unsigned long num_pages, bit_cnt = 0;
911 struct file *file;
912 unsigned long bytes, offset;
913 int outofdate;
914 int ret = -ENOSPC;
915 void *paddr;
917 chunks = bitmap->chunks;
918 file = bitmap->file;
920 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
922 #ifdef INJECT_FAULTS_3
923 outofdate = 1;
924 #else
925 outofdate = bitmap->flags & BITMAP_STALE;
926 #endif
927 if (outofdate)
928 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
929 "recovery\n", bmname(bitmap));
931 bytes = (chunks + 7) / 8;
933 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
935 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
936 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
937 bmname(bitmap),
938 (unsigned long) i_size_read(file->f_mapping->host),
939 bytes + sizeof(bitmap_super_t));
940 goto err;
943 ret = -ENOMEM;
945 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
946 if (!bitmap->filemap)
947 goto err;
949 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
950 bitmap->filemap_attr = kzalloc(
951 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
952 GFP_KERNEL);
953 if (!bitmap->filemap_attr)
954 goto err;
956 oldindex = ~0L;
958 for (i = 0; i < chunks; i++) {
959 int b;
960 index = file_page_index(i);
961 bit = file_page_offset(i);
962 if (index != oldindex) { /* this is a new page, read it in */
963 int count;
964 /* unmap the old page, we're done with it */
965 if (index == num_pages-1)
966 count = bytes + sizeof(bitmap_super_t)
967 - index * PAGE_SIZE;
968 else
969 count = PAGE_SIZE;
970 if (index == 0) {
972 * if we're here then the superblock page
973 * contains some bits (PAGE_SIZE != sizeof sb)
974 * we've already read it in, so just use it
976 page = bitmap->sb_page;
977 offset = sizeof(bitmap_super_t);
978 if (!file)
979 read_sb_page(bitmap->mddev,
980 bitmap->mddev->bitmap_info.offset,
981 page,
982 index, count);
983 } else if (file) {
984 page = read_page(file, index, bitmap, count);
985 offset = 0;
986 } else {
987 page = read_sb_page(bitmap->mddev,
988 bitmap->mddev->bitmap_info.offset,
989 NULL,
990 index, count);
991 offset = 0;
993 if (IS_ERR(page)) { /* read error */
994 ret = PTR_ERR(page);
995 goto err;
998 oldindex = index;
999 oldpage = page;
1001 bitmap->filemap[bitmap->file_pages++] = page;
1002 bitmap->last_page_size = count;
1004 if (outofdate) {
1006 * if bitmap is out of date, dirty the
1007 * whole page and write it out
1009 paddr = kmap_atomic(page, KM_USER0);
1010 memset(paddr + offset, 0xff,
1011 PAGE_SIZE - offset);
1012 kunmap_atomic(paddr, KM_USER0);
1013 write_page(bitmap, page, 1);
1015 ret = -EIO;
1016 if (bitmap->flags & BITMAP_WRITE_ERROR)
1017 goto err;
1020 paddr = kmap_atomic(page, KM_USER0);
1021 if (bitmap->flags & BITMAP_HOSTENDIAN)
1022 b = test_bit(bit, paddr);
1023 else
1024 b = ext2_test_bit(bit, paddr);
1025 kunmap_atomic(paddr, KM_USER0);
1026 if (b) {
1027 /* if the disk bit is set, set the memory bit */
1028 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1029 >= start);
1030 bitmap_set_memory_bits(bitmap,
1031 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1032 needed);
1033 bit_cnt++;
1034 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1038 /* everything went OK */
1039 ret = 0;
1040 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1042 if (bit_cnt) { /* Kick recovery if any bits were set */
1043 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1044 md_wakeup_thread(bitmap->mddev->thread);
1047 printk(KERN_INFO "%s: bitmap initialized from disk: "
1048 "read %lu/%lu pages, set %lu bits\n",
1049 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1051 return 0;
1053 err:
1054 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1055 bmname(bitmap), ret);
1056 return ret;
1059 void bitmap_write_all(struct bitmap *bitmap)
1061 /* We don't actually write all bitmap blocks here,
1062 * just flag them as needing to be written
1064 int i;
1066 for (i=0; i < bitmap->file_pages; i++)
1067 set_page_attr(bitmap, bitmap->filemap[i],
1068 BITMAP_PAGE_NEEDWRITE);
1072 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1074 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1075 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1076 bitmap->bp[page].count += inc;
1078 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1079 (unsigned long long)offset, inc, bitmap->bp[page].count);
1081 bitmap_checkfree(bitmap, page);
1083 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1084 sector_t offset, int *blocks,
1085 int create);
1088 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1089 * out to disk
1092 void bitmap_daemon_work(mddev_t *mddev)
1094 struct bitmap *bitmap;
1095 unsigned long j;
1096 unsigned long flags;
1097 struct page *page = NULL, *lastpage = NULL;
1098 int blocks;
1099 void *paddr;
1101 /* Use a mutex to guard daemon_work against
1102 * bitmap_destroy.
1104 mutex_lock(&mddev->bitmap_info.mutex);
1105 bitmap = mddev->bitmap;
1106 if (bitmap == NULL) {
1107 mutex_unlock(&mddev->bitmap_info.mutex);
1108 return;
1110 if (time_before(jiffies, bitmap->daemon_lastrun
1111 + bitmap->mddev->bitmap_info.daemon_sleep))
1112 goto done;
1114 bitmap->daemon_lastrun = jiffies;
1115 if (bitmap->allclean) {
1116 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1117 goto done;
1119 bitmap->allclean = 1;
1121 spin_lock_irqsave(&bitmap->lock, flags);
1122 for (j = 0; j < bitmap->chunks; j++) {
1123 bitmap_counter_t *bmc;
1124 if (!bitmap->filemap)
1125 /* error or shutdown */
1126 break;
1128 page = filemap_get_page(bitmap, j);
1130 if (page != lastpage) {
1131 /* skip this page unless it's marked as needing cleaning */
1132 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1133 int need_write = test_page_attr(bitmap, page,
1134 BITMAP_PAGE_NEEDWRITE);
1135 if (need_write)
1136 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1138 spin_unlock_irqrestore(&bitmap->lock, flags);
1139 if (need_write) {
1140 write_page(bitmap, page, 0);
1141 bitmap->allclean = 0;
1143 spin_lock_irqsave(&bitmap->lock, flags);
1144 j |= (PAGE_BITS - 1);
1145 continue;
1148 /* grab the new page, sync and release the old */
1149 if (lastpage != NULL) {
1150 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1151 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1152 spin_unlock_irqrestore(&bitmap->lock, flags);
1153 write_page(bitmap, lastpage, 0);
1154 } else {
1155 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1156 spin_unlock_irqrestore(&bitmap->lock, flags);
1158 } else
1159 spin_unlock_irqrestore(&bitmap->lock, flags);
1160 lastpage = page;
1162 /* We are possibly going to clear some bits, so make
1163 * sure that events_cleared is up-to-date.
1165 if (bitmap->need_sync) {
1166 bitmap_super_t *sb;
1167 bitmap->need_sync = 0;
1168 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1169 sb->events_cleared =
1170 cpu_to_le64(bitmap->events_cleared);
1171 kunmap_atomic(sb, KM_USER0);
1172 write_page(bitmap, bitmap->sb_page, 1);
1174 spin_lock_irqsave(&bitmap->lock, flags);
1175 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1177 bmc = bitmap_get_counter(bitmap,
1178 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1179 &blocks, 0);
1180 if (bmc) {
1182 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1184 if (*bmc)
1185 bitmap->allclean = 0;
1187 if (*bmc == 2) {
1188 *bmc=1; /* maybe clear the bit next time */
1189 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1190 } else if (*bmc == 1) {
1191 /* we can clear the bit */
1192 *bmc = 0;
1193 bitmap_count_page(bitmap,
1194 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1195 -1);
1197 /* clear the bit */
1198 paddr = kmap_atomic(page, KM_USER0);
1199 if (bitmap->flags & BITMAP_HOSTENDIAN)
1200 clear_bit(file_page_offset(j), paddr);
1201 else
1202 ext2_clear_bit(file_page_offset(j), paddr);
1203 kunmap_atomic(paddr, KM_USER0);
1205 } else
1206 j |= PAGE_COUNTER_MASK;
1208 spin_unlock_irqrestore(&bitmap->lock, flags);
1210 /* now sync the final page */
1211 if (lastpage != NULL) {
1212 spin_lock_irqsave(&bitmap->lock, flags);
1213 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1214 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1215 spin_unlock_irqrestore(&bitmap->lock, flags);
1216 write_page(bitmap, lastpage, 0);
1217 } else {
1218 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1219 spin_unlock_irqrestore(&bitmap->lock, flags);
1223 done:
1224 if (bitmap->allclean == 0)
1225 bitmap->mddev->thread->timeout =
1226 bitmap->mddev->bitmap_info.daemon_sleep;
1227 mutex_unlock(&mddev->bitmap_info.mutex);
1230 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1231 sector_t offset, int *blocks,
1232 int create)
1233 __releases(bitmap->lock)
1234 __acquires(bitmap->lock)
1236 /* If 'create', we might release the lock and reclaim it.
1237 * The lock must have been taken with interrupts enabled.
1238 * If !create, we don't release the lock.
1240 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1241 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1242 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1243 sector_t csize;
1245 if (bitmap_checkpage(bitmap, page, create) < 0) {
1246 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1247 *blocks = csize - (offset & (csize- 1));
1248 return NULL;
1250 /* now locked ... */
1252 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1253 /* should we use the first or second counter field
1254 * of the hijacked pointer? */
1255 int hi = (pageoff > PAGE_COUNTER_MASK);
1256 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1257 PAGE_COUNTER_SHIFT - 1);
1258 *blocks = csize - (offset & (csize- 1));
1259 return &((bitmap_counter_t *)
1260 &bitmap->bp[page].map)[hi];
1261 } else { /* page is allocated */
1262 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1263 *blocks = csize - (offset & (csize- 1));
1264 return (bitmap_counter_t *)
1265 &(bitmap->bp[page].map[pageoff]);
1269 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1271 if (!bitmap) return 0;
1273 if (behind) {
1274 atomic_inc(&bitmap->behind_writes);
1275 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1276 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1279 while (sectors) {
1280 int blocks;
1281 bitmap_counter_t *bmc;
1283 spin_lock_irq(&bitmap->lock);
1284 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1285 if (!bmc) {
1286 spin_unlock_irq(&bitmap->lock);
1287 return 0;
1290 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1291 DEFINE_WAIT(__wait);
1292 /* note that it is safe to do the prepare_to_wait
1293 * after the test as long as we do it before dropping
1294 * the spinlock.
1296 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1297 TASK_UNINTERRUPTIBLE);
1298 spin_unlock_irq(&bitmap->lock);
1299 blk_unplug(bitmap->mddev->queue);
1300 schedule();
1301 finish_wait(&bitmap->overflow_wait, &__wait);
1302 continue;
1305 switch(*bmc) {
1306 case 0:
1307 bitmap_file_set_bit(bitmap, offset);
1308 bitmap_count_page(bitmap,offset, 1);
1309 blk_plug_device_unlocked(bitmap->mddev->queue);
1310 /* fall through */
1311 case 1:
1312 *bmc = 2;
1315 (*bmc)++;
1317 spin_unlock_irq(&bitmap->lock);
1319 offset += blocks;
1320 if (sectors > blocks)
1321 sectors -= blocks;
1322 else sectors = 0;
1324 bitmap->allclean = 0;
1325 return 0;
1328 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1329 int success, int behind)
1331 if (!bitmap) return;
1332 if (behind) {
1333 atomic_dec(&bitmap->behind_writes);
1334 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1335 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1337 if (bitmap->mddev->degraded)
1338 /* Never clear bits or update events_cleared when degraded */
1339 success = 0;
1341 while (sectors) {
1342 int blocks;
1343 unsigned long flags;
1344 bitmap_counter_t *bmc;
1346 spin_lock_irqsave(&bitmap->lock, flags);
1347 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1348 if (!bmc) {
1349 spin_unlock_irqrestore(&bitmap->lock, flags);
1350 return;
1353 if (success &&
1354 bitmap->events_cleared < bitmap->mddev->events) {
1355 bitmap->events_cleared = bitmap->mddev->events;
1356 bitmap->need_sync = 1;
1359 if (!success && ! (*bmc & NEEDED_MASK))
1360 *bmc |= NEEDED_MASK;
1362 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1363 wake_up(&bitmap->overflow_wait);
1365 (*bmc)--;
1366 if (*bmc <= 2) {
1367 set_page_attr(bitmap,
1368 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1369 BITMAP_PAGE_CLEAN);
1371 spin_unlock_irqrestore(&bitmap->lock, flags);
1372 offset += blocks;
1373 if (sectors > blocks)
1374 sectors -= blocks;
1375 else sectors = 0;
1379 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1380 int degraded)
1382 bitmap_counter_t *bmc;
1383 int rv;
1384 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1385 *blocks = 1024;
1386 return 1; /* always resync if no bitmap */
1388 spin_lock_irq(&bitmap->lock);
1389 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1390 rv = 0;
1391 if (bmc) {
1392 /* locked */
1393 if (RESYNC(*bmc))
1394 rv = 1;
1395 else if (NEEDED(*bmc)) {
1396 rv = 1;
1397 if (!degraded) { /* don't set/clear bits if degraded */
1398 *bmc |= RESYNC_MASK;
1399 *bmc &= ~NEEDED_MASK;
1403 spin_unlock_irq(&bitmap->lock);
1404 bitmap->allclean = 0;
1405 return rv;
1408 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1409 int degraded)
1411 /* bitmap_start_sync must always report on multiples of whole
1412 * pages, otherwise resync (which is very PAGE_SIZE based) will
1413 * get confused.
1414 * So call __bitmap_start_sync repeatedly (if needed) until
1415 * At least PAGE_SIZE>>9 blocks are covered.
1416 * Return the 'or' of the result.
1418 int rv = 0;
1419 int blocks1;
1421 *blocks = 0;
1422 while (*blocks < (PAGE_SIZE>>9)) {
1423 rv |= __bitmap_start_sync(bitmap, offset,
1424 &blocks1, degraded);
1425 offset += blocks1;
1426 *blocks += blocks1;
1428 return rv;
1431 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1433 bitmap_counter_t *bmc;
1434 unsigned long flags;
1436 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1437 */ if (bitmap == NULL) {
1438 *blocks = 1024;
1439 return;
1441 spin_lock_irqsave(&bitmap->lock, flags);
1442 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1443 if (bmc == NULL)
1444 goto unlock;
1445 /* locked */
1447 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1449 if (RESYNC(*bmc)) {
1450 *bmc &= ~RESYNC_MASK;
1452 if (!NEEDED(*bmc) && aborted)
1453 *bmc |= NEEDED_MASK;
1454 else {
1455 if (*bmc <= 2) {
1456 set_page_attr(bitmap,
1457 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1458 BITMAP_PAGE_CLEAN);
1462 unlock:
1463 spin_unlock_irqrestore(&bitmap->lock, flags);
1464 bitmap->allclean = 0;
1467 void bitmap_close_sync(struct bitmap *bitmap)
1469 /* Sync has finished, and any bitmap chunks that weren't synced
1470 * properly have been aborted. It remains to us to clear the
1471 * RESYNC bit wherever it is still on
1473 sector_t sector = 0;
1474 int blocks;
1475 if (!bitmap)
1476 return;
1477 while (sector < bitmap->mddev->resync_max_sectors) {
1478 bitmap_end_sync(bitmap, sector, &blocks, 0);
1479 sector += blocks;
1483 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1485 sector_t s = 0;
1486 int blocks;
1488 if (!bitmap)
1489 return;
1490 if (sector == 0) {
1491 bitmap->last_end_sync = jiffies;
1492 return;
1494 if (time_before(jiffies, (bitmap->last_end_sync
1495 + bitmap->mddev->bitmap_info.daemon_sleep)))
1496 return;
1497 wait_event(bitmap->mddev->recovery_wait,
1498 atomic_read(&bitmap->mddev->recovery_active) == 0);
1500 bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync;
1501 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1502 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1503 s = 0;
1504 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1505 bitmap_end_sync(bitmap, s, &blocks, 0);
1506 s += blocks;
1508 bitmap->last_end_sync = jiffies;
1509 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1512 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1514 /* For each chunk covered by any of these sectors, set the
1515 * counter to 1 and set resync_needed. They should all
1516 * be 0 at this point
1519 int secs;
1520 bitmap_counter_t *bmc;
1521 spin_lock_irq(&bitmap->lock);
1522 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1523 if (!bmc) {
1524 spin_unlock_irq(&bitmap->lock);
1525 return;
1527 if (! *bmc) {
1528 struct page *page;
1529 *bmc = 1 | (needed?NEEDED_MASK:0);
1530 bitmap_count_page(bitmap, offset, 1);
1531 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1532 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1534 spin_unlock_irq(&bitmap->lock);
1535 bitmap->allclean = 0;
1538 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1539 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1541 unsigned long chunk;
1543 for (chunk = s; chunk <= e; chunk++) {
1544 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
1545 bitmap_set_memory_bits(bitmap, sec, 1);
1546 bitmap_file_set_bit(bitmap, sec);
1551 * flush out any pending updates
1553 void bitmap_flush(mddev_t *mddev)
1555 struct bitmap *bitmap = mddev->bitmap;
1556 long sleep;
1558 if (!bitmap) /* there was no bitmap */
1559 return;
1561 /* run the daemon_work three time to ensure everything is flushed
1562 * that can be
1564 sleep = mddev->bitmap_info.daemon_sleep * 2;
1565 bitmap->daemon_lastrun -= sleep;
1566 bitmap_daemon_work(mddev);
1567 bitmap->daemon_lastrun -= sleep;
1568 bitmap_daemon_work(mddev);
1569 bitmap->daemon_lastrun -= sleep;
1570 bitmap_daemon_work(mddev);
1571 bitmap_update_sb(bitmap);
1575 * free memory that was allocated
1577 static void bitmap_free(struct bitmap *bitmap)
1579 unsigned long k, pages;
1580 struct bitmap_page *bp;
1582 if (!bitmap) /* there was no bitmap */
1583 return;
1585 /* release the bitmap file and kill the daemon */
1586 bitmap_file_put(bitmap);
1588 bp = bitmap->bp;
1589 pages = bitmap->pages;
1591 /* free all allocated memory */
1593 if (bp) /* deallocate the page memory */
1594 for (k = 0; k < pages; k++)
1595 if (bp[k].map && !bp[k].hijacked)
1596 kfree(bp[k].map);
1597 kfree(bp);
1598 kfree(bitmap);
1601 void bitmap_destroy(mddev_t *mddev)
1603 struct bitmap *bitmap = mddev->bitmap;
1605 if (!bitmap) /* there was no bitmap */
1606 return;
1608 mutex_lock(&mddev->bitmap_info.mutex);
1609 mddev->bitmap = NULL; /* disconnect from the md device */
1610 mutex_unlock(&mddev->bitmap_info.mutex);
1611 if (mddev->thread)
1612 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1614 bitmap_free(bitmap);
1618 * initialize the bitmap structure
1619 * if this returns an error, bitmap_destroy must be called to do clean up
1621 int bitmap_create(mddev_t *mddev)
1623 struct bitmap *bitmap;
1624 sector_t blocks = mddev->resync_max_sectors;
1625 unsigned long chunks;
1626 unsigned long pages;
1627 struct file *file = mddev->bitmap_info.file;
1628 int err;
1629 sector_t start;
1631 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1633 if (!file && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */
1634 return 0;
1636 BUG_ON(file && mddev->bitmap_info.offset);
1638 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1639 if (!bitmap)
1640 return -ENOMEM;
1642 spin_lock_init(&bitmap->lock);
1643 atomic_set(&bitmap->pending_writes, 0);
1644 init_waitqueue_head(&bitmap->write_wait);
1645 init_waitqueue_head(&bitmap->overflow_wait);
1647 bitmap->mddev = mddev;
1649 bitmap->file = file;
1650 if (file) {
1651 get_file(file);
1652 /* As future accesses to this file will use bmap,
1653 * and bypass the page cache, we must sync the file
1654 * first.
1656 vfs_fsync(file, file->f_dentry, 1);
1658 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1659 err = bitmap_read_sb(bitmap);
1660 if (err)
1661 goto error;
1663 bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize);
1665 /* now that chunksize and chunkshift are set, we can use these macros */
1666 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1667 CHUNK_BLOCK_SHIFT(bitmap);
1668 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1670 BUG_ON(!pages);
1672 bitmap->chunks = chunks;
1673 bitmap->pages = pages;
1674 bitmap->missing_pages = pages;
1675 bitmap->counter_bits = COUNTER_BITS;
1677 bitmap->syncchunk = ~0UL;
1679 #ifdef INJECT_FATAL_FAULT_1
1680 bitmap->bp = NULL;
1681 #else
1682 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1683 #endif
1684 err = -ENOMEM;
1685 if (!bitmap->bp)
1686 goto error;
1688 /* now that we have some pages available, initialize the in-memory
1689 * bitmap from the on-disk bitmap */
1690 start = 0;
1691 if (mddev->degraded == 0
1692 || bitmap->events_cleared == mddev->events)
1693 /* no need to keep dirty bits to optimise a re-add of a missing device */
1694 start = mddev->recovery_cp;
1695 err = bitmap_init_from_disk(bitmap, start);
1697 if (err)
1698 goto error;
1700 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1701 pages, bmname(bitmap));
1703 mddev->bitmap = bitmap;
1705 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1706 md_wakeup_thread(mddev->thread);
1708 bitmap_update_sb(bitmap);
1710 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1712 error:
1713 bitmap_free(bitmap);
1714 return err;
1717 /* the bitmap API -- for raid personalities */
1718 EXPORT_SYMBOL(bitmap_startwrite);
1719 EXPORT_SYMBOL(bitmap_endwrite);
1720 EXPORT_SYMBOL(bitmap_start_sync);
1721 EXPORT_SYMBOL(bitmap_end_sync);
1722 EXPORT_SYMBOL(bitmap_unplug);
1723 EXPORT_SYMBOL(bitmap_close_sync);
1724 EXPORT_SYMBOL(bitmap_cond_end_sync);