?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
fs: use kmem_cache_zalloc instead
[pv_ops_mirror.git] / drivers / md / bitmap.c
blob927cb34c480584d75803d8eb826db7e8f4a45c83
1 /*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
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
10 */
11
12 /*
13 * Still to do:
14 *
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.
17 */
18
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/raid/md.h>
30 #include <linux/raid/bitmap.h>
31
32 /* debug macros */
33
34 #define DEBUG 0
35
36 #if DEBUG
37 /* these are for debugging purposes only! */
38
39 /* define one and only one of these */
40 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
41 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
42 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
43 #define INJECT_FAULTS_4 0 /* undef */
44 #define INJECT_FAULTS_5 0 /* undef */
45 #define INJECT_FAULTS_6 0
46
47 /* if these are defined, the driver will fail! debug only */
48 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
49 #define INJECT_FATAL_FAULT_2 0 /* undef */
50 #define INJECT_FATAL_FAULT_3 0 /* undef */
51 #endif
52
53 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
54 #define DPRINTK(x...) do { } while(0)
55
56 #ifndef PRINTK
57 # if DEBUG > 0
58 # define PRINTK(x...) printk(KERN_DEBUG x)
59 # else
60 # define PRINTK(x...)
61 # endif
62 #endif
63
64 static inline char * bmname(struct bitmap *bitmap)
65 {
66 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
67 }
68
69
70 /*
71 * just a placeholder - calls kmalloc for bitmap pages
72 */
73 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
74 {
75 unsigned char *page;
76
77 #ifdef INJECT_FAULTS_1
78 page = NULL;
79 #else
80 page = kmalloc(PAGE_SIZE, GFP_NOIO);
81 #endif
82 if (!page)
83 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
84 else
85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
86 bmname(bitmap), page);
87 return page;
88 }
89
90 /*
91 * for now just a placeholder -- just calls kfree for bitmap pages
92 */
93 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
94 {
95 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
96 kfree(page);
97 }
98
99 /*
100 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
101 *
102 * 1) check to see if this page is allocated, if it's not then try to alloc
103 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
104 * page pointer directly as a counter
105 *
106 * if we find our page, we increment the page's refcount so that it stays
107 * allocated while we're using it
108 */
109 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
110 {
111 unsigned char *mappage;
112
113 if (page >= bitmap->pages) {
114 printk(KERN_ALERT
115 "%s: invalid bitmap page request: %lu (> %lu)\n",
116 bmname(bitmap), page, bitmap->pages-1);
117 return -EINVAL;
118 }
119
120
121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
122 return 0;
123
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
125 return 0;
126
127 if (!create)
128 return -ENOENT;
129
130 spin_unlock_irq(&bitmap->lock);
131
132 /* this page has not been allocated yet */
133
134 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
135 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
136 bmname(bitmap));
137 /* failed - set the hijacked flag so that we can use the
138 * pointer as a counter */
139 spin_lock_irq(&bitmap->lock);
140 if (!bitmap->bp[page].map)
141 bitmap->bp[page].hijacked = 1;
142 goto out;
143 }
144
145 /* got a page */
146
147 spin_lock_irq(&bitmap->lock);
148
149 /* recheck the page */
150
151 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
152 /* somebody beat us to getting the page */
153 bitmap_free_page(bitmap, mappage);
154 return 0;
155 }
156
157 /* no page was in place and we have one, so install it */
158
159 memset(mappage, 0, PAGE_SIZE);
160 bitmap->bp[page].map = mappage;
161 bitmap->missing_pages--;
162 out:
163 return 0;
164 }
165
166
167 /* if page is completely empty, put it back on the free list, or dealloc it */
168 /* if page was hijacked, unmark the flag so it might get alloced next time */
169 /* Note: lock should be held when calling this */
170 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
171 {
172 char *ptr;
173
174 if (bitmap->bp[page].count) /* page is still busy */
175 return;
176
177 /* page is no longer in use, it can be released */
178
179 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
180 bitmap->bp[page].hijacked = 0;
181 bitmap->bp[page].map = NULL;
182 return;
183 }
184
185 /* normal case, free the page */
186
187 #if 0
188 /* actually ... let's not. We will probably need the page again exactly when
189 * memory is tight and we are flusing to disk
190 */
191 return;
192 #else
193 ptr = bitmap->bp[page].map;
194 bitmap->bp[page].map = NULL;
195 bitmap->missing_pages++;
196 bitmap_free_page(bitmap, ptr);
197 return;
198 #endif
199 }
200
201
202 /*
203 * bitmap file handling - read and write the bitmap file and its superblock
204 */
205
206 /* copy the pathname of a file to a buffer */
207 char *file_path(struct file *file, char *buf, int count)
208 {
209 struct dentry *d;
210 struct vfsmount *v;
211
212 if (!buf)
213 return NULL;
214
215 d = file->f_path.dentry;
216 v = file->f_path.mnt;
217
218 buf = d_path(d, v, buf, count);
219
220 return IS_ERR(buf) ? NULL : buf;
221 }
222
223 /*
224 * basic page I/O operations
225 */
226
227 /* IO operations when bitmap is stored near all superblocks */
228 static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
229 {
230 /* choose a good rdev and read the page from there */
231
232 mdk_rdev_t *rdev;
233 struct list_head *tmp;
234 struct page *page = alloc_page(GFP_KERNEL);
235 sector_t target;
236
237 if (!page)
238 return ERR_PTR(-ENOMEM);
239
240 ITERATE_RDEV(mddev, rdev, tmp) {
241 if (! test_bit(In_sync, &rdev->flags)
242 || test_bit(Faulty, &rdev->flags))
243 continue;
244
245 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
246
247 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
248 page->index = index;
249 attach_page_buffers(page, NULL); /* so that free_buffer will
250 * quietly no-op */
251 return page;
252 }
253 }
254 return ERR_PTR(-EIO);
255
256 }
257
258 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
259 {
260 mdk_rdev_t *rdev;
261 struct list_head *tmp;
262 mddev_t *mddev = bitmap->mddev;
263
264 ITERATE_RDEV(mddev, rdev, tmp)
265 if (test_bit(In_sync, &rdev->flags)
266 && !test_bit(Faulty, &rdev->flags)) {
267 int size = PAGE_SIZE;
268 if (page->index == bitmap->file_pages-1)
269 size = roundup(bitmap->last_page_size,
270 bdev_hardsect_size(rdev->bdev));
271 /* Just make sure we aren't corrupting data or
272 * metadata
273 */
274 if (bitmap->offset < 0) {
275 /* DATA BITMAP METADATA */
276 if (bitmap->offset
277 + page->index * (PAGE_SIZE/512)
278 + size/512 > 0)
279 /* bitmap runs in to metadata */
280 return -EINVAL;
281 if (rdev->data_offset + mddev->size*2
282 > rdev->sb_offset*2 + bitmap->offset)
283 /* data runs in to bitmap */
284 return -EINVAL;
285 } else if (rdev->sb_offset*2 < rdev->data_offset) {
286 /* METADATA BITMAP DATA */
287 if (rdev->sb_offset*2
288 + bitmap->offset
289 + page->index*(PAGE_SIZE/512) + size/512
290 > rdev->data_offset)
291 /* bitmap runs in to data */
292 return -EINVAL;
293 } else {
294 /* DATA METADATA BITMAP - no problems */
295 }
296 md_super_write(mddev, rdev,
297 (rdev->sb_offset<<1) + bitmap->offset
298 + page->index * (PAGE_SIZE/512),
299 size,
300 page);
301 }
302
303 if (wait)
304 md_super_wait(mddev);
305 return 0;
306 }
307
308 static void bitmap_file_kick(struct bitmap *bitmap);
309 /*
310 * write out a page to a file
311 */
312 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
313 {
314 struct buffer_head *bh;
315
316 if (bitmap->file == NULL) {
317 switch (write_sb_page(bitmap, page, wait)) {
318 case -EINVAL:
319 bitmap->flags |= BITMAP_WRITE_ERROR;
320 }
321 } else {
322
323 bh = page_buffers(page);
324
325 while (bh && bh->b_blocknr) {
326 atomic_inc(&bitmap->pending_writes);
327 set_buffer_locked(bh);
328 set_buffer_mapped(bh);
329 submit_bh(WRITE, bh);
330 bh = bh->b_this_page;
331 }
332
333 if (wait) {
334 wait_event(bitmap->write_wait,
335 atomic_read(&bitmap->pending_writes)==0);
336 }
337 }
338 if (bitmap->flags & BITMAP_WRITE_ERROR)
339 bitmap_file_kick(bitmap);
340 }
341
342 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
343 {
344 struct bitmap *bitmap = bh->b_private;
345 unsigned long flags;
346
347 if (!uptodate) {
348 spin_lock_irqsave(&bitmap->lock, flags);
349 bitmap->flags |= BITMAP_WRITE_ERROR;
350 spin_unlock_irqrestore(&bitmap->lock, flags);
351 }
352 if (atomic_dec_and_test(&bitmap->pending_writes))
353 wake_up(&bitmap->write_wait);
354 }
355
356 /* copied from buffer.c */
357 static void
358 __clear_page_buffers(struct page *page)
359 {
360 ClearPagePrivate(page);
361 set_page_private(page, 0);
362 page_cache_release(page);
363 }
364 static void free_buffers(struct page *page)
365 {
366 struct buffer_head *bh = page_buffers(page);
367
368 while (bh) {
369 struct buffer_head *next = bh->b_this_page;
370 free_buffer_head(bh);
371 bh = next;
372 }
373 __clear_page_buffers(page);
374 put_page(page);
375 }
376
377 /* read a page from a file.
378 * We both read the page, and attach buffers to the page to record the
379 * address of each block (using bmap). These addresses will be used
380 * to write the block later, completely bypassing the filesystem.
381 * This usage is similar to how swap files are handled, and allows us
382 * to write to a file with no concerns of memory allocation failing.
383 */
384 static struct page *read_page(struct file *file, unsigned long index,
385 struct bitmap *bitmap,
386 unsigned long count)
387 {
388 struct page *page = NULL;
389 struct inode *inode = file->f_path.dentry->d_inode;
390 struct buffer_head *bh;
391 sector_t block;
392
393 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
394 (unsigned long long)index << PAGE_SHIFT);
395
396 page = alloc_page(GFP_KERNEL);
397 if (!page)
398 page = ERR_PTR(-ENOMEM);
399 if (IS_ERR(page))
400 goto out;
401
402 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
403 if (!bh) {
404 put_page(page);
405 page = ERR_PTR(-ENOMEM);
406 goto out;
407 }
408 attach_page_buffers(page, bh);
409 block = index << (PAGE_SHIFT - inode->i_blkbits);
410 while (bh) {
411 if (count == 0)
412 bh->b_blocknr = 0;
413 else {
414 bh->b_blocknr = bmap(inode, block);
415 if (bh->b_blocknr == 0) {
416 /* Cannot use this file! */
417 free_buffers(page);
418 page = ERR_PTR(-EINVAL);
419 goto out;
420 }
421 bh->b_bdev = inode->i_sb->s_bdev;
422 if (count < (1<<inode->i_blkbits))
423 count = 0;
424 else
425 count -= (1<<inode->i_blkbits);
426
427 bh->b_end_io = end_bitmap_write;
428 bh->b_private = bitmap;
429 atomic_inc(&bitmap->pending_writes);
430 set_buffer_locked(bh);
431 set_buffer_mapped(bh);
432 submit_bh(READ, bh);
433 }
434 block++;
435 bh = bh->b_this_page;
436 }
437 page->index = index;
438
439 wait_event(bitmap->write_wait,
440 atomic_read(&bitmap->pending_writes)==0);
441 if (bitmap->flags & BITMAP_WRITE_ERROR) {
442 free_buffers(page);
443 page = ERR_PTR(-EIO);
444 }
445 out:
446 if (IS_ERR(page))
447 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
448 (int)PAGE_SIZE,
449 (unsigned long long)index << PAGE_SHIFT,
450 PTR_ERR(page));
451 return page;
452 }
453
454 /*
455 * bitmap file superblock operations
456 */
457
458 /* update the event counter and sync the superblock to disk */
459 void bitmap_update_sb(struct bitmap *bitmap)
460 {
461 bitmap_super_t *sb;
462 unsigned long flags;
463
464 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
465 return;
466 spin_lock_irqsave(&bitmap->lock, flags);
467 if (!bitmap->sb_page) { /* no superblock */
468 spin_unlock_irqrestore(&bitmap->lock, flags);
469 return;
470 }
471 spin_unlock_irqrestore(&bitmap->lock, flags);
472 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
473 sb->events = cpu_to_le64(bitmap->mddev->events);
474 if (!bitmap->mddev->degraded)
475 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
476 kunmap_atomic(sb, KM_USER0);
477 write_page(bitmap, bitmap->sb_page, 1);
478 }
479
480 /* print out the bitmap file superblock */
481 void bitmap_print_sb(struct bitmap *bitmap)
482 {
483 bitmap_super_t *sb;
484
485 if (!bitmap || !bitmap->sb_page)
486 return;
487 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
488 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
489 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
490 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
491 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
492 *(__u32 *)(sb->uuid+0),
493 *(__u32 *)(sb->uuid+4),
494 *(__u32 *)(sb->uuid+8),
495 *(__u32 *)(sb->uuid+12));
496 printk(KERN_DEBUG " events: %llu\n",
497 (unsigned long long) le64_to_cpu(sb->events));
498 printk(KERN_DEBUG "events cleared: %llu\n",
499 (unsigned long long) le64_to_cpu(sb->events_cleared));
500 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
501 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
502 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
503 printk(KERN_DEBUG " sync size: %llu KB\n",
504 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
505 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
506 kunmap_atomic(sb, KM_USER0);
507 }
508
509 /* read the superblock from the bitmap file and initialize some bitmap fields */
510 static int bitmap_read_sb(struct bitmap *bitmap)
511 {
512 char *reason = NULL;
513 bitmap_super_t *sb;
514 unsigned long chunksize, daemon_sleep, write_behind;
515 unsigned long long events;
516 int err = -EINVAL;
517
518 /* page 0 is the superblock, read it... */
519 if (bitmap->file) {
520 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
521 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
522
523 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
524 } else {
525 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
526 }
527 if (IS_ERR(bitmap->sb_page)) {
528 err = PTR_ERR(bitmap->sb_page);
529 bitmap->sb_page = NULL;
530 return err;
531 }
532
533 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
534
535 chunksize = le32_to_cpu(sb->chunksize);
536 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
537 write_behind = le32_to_cpu(sb->write_behind);
538
539 /* verify that the bitmap-specific fields are valid */
540 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
541 reason = "bad magic";
542 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
543 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
544 reason = "unrecognized superblock version";
545 else if (chunksize < PAGE_SIZE)
546 reason = "bitmap chunksize too small";
547 else if ((1 << ffz(~chunksize)) != chunksize)
548 reason = "bitmap chunksize not a power of 2";
549 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
550 reason = "daemon sleep period out of range";
551 else if (write_behind > COUNTER_MAX)
552 reason = "write-behind limit out of range (0 - 16383)";
553 if (reason) {
554 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
555 bmname(bitmap), reason);
556 goto out;
557 }
558
559 /* keep the array size field of the bitmap superblock up to date */
560 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
561
562 if (!bitmap->mddev->persistent)
563 goto success;
564
565 /*
566 * if we have a persistent array superblock, compare the
567 * bitmap's UUID and event counter to the mddev's
568 */
569 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
570 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
571 bmname(bitmap));
572 goto out;
573 }
574 events = le64_to_cpu(sb->events);
575 if (events < bitmap->mddev->events) {
576 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
577 "-- forcing full recovery\n", bmname(bitmap), events,
578 (unsigned long long) bitmap->mddev->events);
579 sb->state |= cpu_to_le32(BITMAP_STALE);
580 }
581 success:
582 /* assign fields using values from superblock */
583 bitmap->chunksize = chunksize;
584 bitmap->daemon_sleep = daemon_sleep;
585 bitmap->daemon_lastrun = jiffies;
586 bitmap->max_write_behind = write_behind;
587 bitmap->flags |= le32_to_cpu(sb->state);
588 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
589 bitmap->flags |= BITMAP_HOSTENDIAN;
590 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
591 if (sb->state & cpu_to_le32(BITMAP_STALE))
592 bitmap->events_cleared = bitmap->mddev->events;
593 err = 0;
594 out:
595 kunmap_atomic(sb, KM_USER0);
596 if (err)
597 bitmap_print_sb(bitmap);
598 return err;
599 }
600
601 enum bitmap_mask_op {
602 MASK_SET,
603 MASK_UNSET
604 };
605
606 /* record the state of the bitmap in the superblock. Return the old value */
607 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
608 enum bitmap_mask_op op)
609 {
610 bitmap_super_t *sb;
611 unsigned long flags;
612 int old;
613
614 spin_lock_irqsave(&bitmap->lock, flags);
615 if (!bitmap->sb_page) { /* can't set the state */
616 spin_unlock_irqrestore(&bitmap->lock, flags);
617 return 0;
618 }
619 spin_unlock_irqrestore(&bitmap->lock, flags);
620 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
621 old = le32_to_cpu(sb->state) & bits;
622 switch (op) {
623 case MASK_SET: sb->state |= cpu_to_le32(bits);
624 break;
625 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
626 break;
627 default: BUG();
628 }
629 kunmap_atomic(sb, KM_USER0);
630 return old;
631 }
632
633 /*
634 * general bitmap file operations
635 */
636
637 /* calculate the index of the page that contains this bit */
638 static inline unsigned long file_page_index(unsigned long chunk)
639 {
640 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
641 }
642
643 /* calculate the (bit) offset of this bit within a page */
644 static inline unsigned long file_page_offset(unsigned long chunk)
645 {
646 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
647 }
648
649 /*
650 * return a pointer to the page in the filemap that contains the given bit
651 *
652 * this lookup is complicated by the fact that the bitmap sb might be exactly
653 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
654 * 0 or page 1
655 */
656 static inline struct page *filemap_get_page(struct bitmap *bitmap,
657 unsigned long chunk)
658 {
659 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
660 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
661 }
662
663
664 static void bitmap_file_unmap(struct bitmap *bitmap)
665 {
666 struct page **map, *sb_page;
667 unsigned long *attr;
668 int pages;
669 unsigned long flags;
670
671 spin_lock_irqsave(&bitmap->lock, flags);
672 map = bitmap->filemap;
673 bitmap->filemap = NULL;
674 attr = bitmap->filemap_attr;
675 bitmap->filemap_attr = NULL;
676 pages = bitmap->file_pages;
677 bitmap->file_pages = 0;
678 sb_page = bitmap->sb_page;
679 bitmap->sb_page = NULL;
680 spin_unlock_irqrestore(&bitmap->lock, flags);
681
682 while (pages--)
683 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
684 free_buffers(map[pages]);
685 kfree(map);
686 kfree(attr);
687
688 if (sb_page)
689 free_buffers(sb_page);
690 }
691
692 static void bitmap_file_put(struct bitmap *bitmap)
693 {
694 struct file *file;
695 unsigned long flags;
696
697 spin_lock_irqsave(&bitmap->lock, flags);
698 file = bitmap->file;
699 bitmap->file = NULL;
700 spin_unlock_irqrestore(&bitmap->lock, flags);
701
702 if (file)
703 wait_event(bitmap->write_wait,
704 atomic_read(&bitmap->pending_writes)==0);
705 bitmap_file_unmap(bitmap);
706
707 if (file) {
708 struct inode *inode = file->f_path.dentry->d_inode;
709 invalidate_mapping_pages(inode->i_mapping, 0, -1);
710 fput(file);
711 }
712 }
713
714
715 /*
716 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
717 * then it is no longer reliable, so we stop using it and we mark the file
718 * as failed in the superblock
719 */
720 static void bitmap_file_kick(struct bitmap *bitmap)
721 {
722 char *path, *ptr = NULL;
723
724 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
725 bitmap_update_sb(bitmap);
726
727 if (bitmap->file) {
728 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
729 if (path)
730 ptr = file_path(bitmap->file, path, PAGE_SIZE);
731
732 printk(KERN_ALERT
733 "%s: kicking failed bitmap file %s from array!\n",
734 bmname(bitmap), ptr ? ptr : "");
735
736 kfree(path);
737 } else
738 printk(KERN_ALERT
739 "%s: disabling internal bitmap due to errors\n",
740 bmname(bitmap));
741 }
742
743 bitmap_file_put(bitmap);
744
745 return;
746 }
747
748 enum bitmap_page_attr {
749 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
750 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
751 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
752 };
753
754 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
755 enum bitmap_page_attr attr)
756 {
757 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
758 }
759
760 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
761 enum bitmap_page_attr attr)
762 {
763 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
764 }
765
766 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
767 enum bitmap_page_attr attr)
768 {
769 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
770 }
771
772 /*
773 * bitmap_file_set_bit -- called before performing a write to the md device
774 * to set (and eventually sync) a particular bit in the bitmap file
775 *
776 * we set the bit immediately, then we record the page number so that
777 * when an unplug occurs, we can flush the dirty pages out to disk
778 */
779 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
780 {
781 unsigned long bit;
782 struct page *page;
783 void *kaddr;
784 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
785
786 if (!bitmap->filemap) {
787 return;
788 }
789
790 page = filemap_get_page(bitmap, chunk);
791 if (!page) return;
792 bit = file_page_offset(chunk);
793
794 /* set the bit */
795 kaddr = kmap_atomic(page, KM_USER0);
796 if (bitmap->flags & BITMAP_HOSTENDIAN)
797 set_bit(bit, kaddr);
798 else
799 ext2_set_bit(bit, kaddr);
800 kunmap_atomic(kaddr, KM_USER0);
801 PRINTK("set file bit %lu page %lu\n", bit, page->index);
802
803 /* record page number so it gets flushed to disk when unplug occurs */
804 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
805
806 }
807
808 /* this gets called when the md device is ready to unplug its underlying
809 * (slave) device queues -- before we let any writes go down, we need to
810 * sync the dirty pages of the bitmap file to disk */
811 void bitmap_unplug(struct bitmap *bitmap)
812 {
813 unsigned long i, flags;
814 int dirty, need_write;
815 struct page *page;
816 int wait = 0;
817
818 if (!bitmap)
819 return;
820
821 /* look at each page to see if there are any set bits that need to be
822 * flushed out to disk */
823 for (i = 0; i < bitmap->file_pages; i++) {
824 spin_lock_irqsave(&bitmap->lock, flags);
825 if (!bitmap->filemap) {
826 spin_unlock_irqrestore(&bitmap->lock, flags);
827 return;
828 }
829 page = bitmap->filemap[i];
830 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
831 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
832 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
833 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
834 if (dirty)
835 wait = 1;
836 spin_unlock_irqrestore(&bitmap->lock, flags);
837
838 if (dirty | need_write)
839 write_page(bitmap, page, 0);
840 }
841 if (wait) { /* if any writes were performed, we need to wait on them */
842 if (bitmap->file)
843 wait_event(bitmap->write_wait,
844 atomic_read(&bitmap->pending_writes)==0);
845 else
846 md_super_wait(bitmap->mddev);
847 }
848 if (bitmap->flags & BITMAP_WRITE_ERROR)
849 bitmap_file_kick(bitmap);
850 }
851
852 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
853 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
854 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
855 * memory mapping of the bitmap file
856 * Special cases:
857 * if there's no bitmap file, or if the bitmap file had been
858 * previously kicked from the array, we mark all the bits as
859 * 1's in order to cause a full resync.
860 *
861 * We ignore all bits for sectors that end earlier than 'start'.
862 * This is used when reading an out-of-date bitmap...
863 */
864 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
865 {
866 unsigned long i, chunks, index, oldindex, bit;
867 struct page *page = NULL, *oldpage = NULL;
868 unsigned long num_pages, bit_cnt = 0;
869 struct file *file;
870 unsigned long bytes, offset;
871 int outofdate;
872 int ret = -ENOSPC;
873 void *paddr;
874
875 chunks = bitmap->chunks;
876 file = bitmap->file;
877
878 BUG_ON(!file && !bitmap->offset);
879
880 #ifdef INJECT_FAULTS_3
881 outofdate = 1;
882 #else
883 outofdate = bitmap->flags & BITMAP_STALE;
884 #endif
885 if (outofdate)
886 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
887 "recovery\n", bmname(bitmap));
888
889 bytes = (chunks + 7) / 8;
890
891 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
892
893 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
894 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
895 bmname(bitmap),
896 (unsigned long) i_size_read(file->f_mapping->host),
897 bytes + sizeof(bitmap_super_t));
898 goto err;
899 }
900
901 ret = -ENOMEM;
902
903 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
904 if (!bitmap->filemap)
905 goto err;
906
907 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
908 bitmap->filemap_attr = kzalloc(
909 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
910 GFP_KERNEL);
911 if (!bitmap->filemap_attr)
912 goto err;
913
914 oldindex = ~0L;
915
916 for (i = 0; i < chunks; i++) {
917 int b;
918 index = file_page_index(i);
919 bit = file_page_offset(i);
920 if (index != oldindex) { /* this is a new page, read it in */
921 int count;
922 /* unmap the old page, we're done with it */
923 if (index == num_pages-1)
924 count = bytes + sizeof(bitmap_super_t)
925 - index * PAGE_SIZE;
926 else
927 count = PAGE_SIZE;
928 if (index == 0) {
929 /*
930 * if we're here then the superblock page
931 * contains some bits (PAGE_SIZE != sizeof sb)
932 * we've already read it in, so just use it
933 */
934 page = bitmap->sb_page;
935 offset = sizeof(bitmap_super_t);
936 } else if (file) {
937 page = read_page(file, index, bitmap, count);
938 offset = 0;
939 } else {
940 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
941 offset = 0;
942 }
943 if (IS_ERR(page)) { /* read error */
944 ret = PTR_ERR(page);
945 goto err;
946 }
947
948 oldindex = index;
949 oldpage = page;
950
951 if (outofdate) {
952 /*
953 * if bitmap is out of date, dirty the
954 * whole page and write it out
955 */
956 paddr = kmap_atomic(page, KM_USER0);
957 memset(paddr + offset, 0xff,
958 PAGE_SIZE - offset);
959 kunmap_atomic(paddr, KM_USER0);
960 write_page(bitmap, page, 1);
961
962 ret = -EIO;
963 if (bitmap->flags & BITMAP_WRITE_ERROR) {
964 /* release, page not in filemap yet */
965 put_page(page);
966 goto err;
967 }
968 }
969
970 bitmap->filemap[bitmap->file_pages++] = page;
971 bitmap->last_page_size = count;
972 }
973 paddr = kmap_atomic(page, KM_USER0);
974 if (bitmap->flags & BITMAP_HOSTENDIAN)
975 b = test_bit(bit, paddr);
976 else
977 b = ext2_test_bit(bit, paddr);
978 kunmap_atomic(paddr, KM_USER0);
979 if (b) {
980 /* if the disk bit is set, set the memory bit */
981 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
982 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
983 );
984 bit_cnt++;
985 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
986 }
987 }
988
989 /* everything went OK */
990 ret = 0;
991 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
992
993 if (bit_cnt) { /* Kick recovery if any bits were set */
994 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
995 md_wakeup_thread(bitmap->mddev->thread);
996 }
997
998 printk(KERN_INFO "%s: bitmap initialized from disk: "
999 "read %lu/%lu pages, set %lu bits\n",
1000 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1001
1002 return 0;
1003
1004 err:
1005 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1006 bmname(bitmap), ret);
1007 return ret;
1008 }
1009
1010 void bitmap_write_all(struct bitmap *bitmap)
1011 {
1012 /* We don't actually write all bitmap blocks here,
1013 * just flag them as needing to be written
1014 */
1015 int i;
1016
1017 for (i=0; i < bitmap->file_pages; i++)
1018 set_page_attr(bitmap, bitmap->filemap[i],
1019 BITMAP_PAGE_NEEDWRITE);
1020 }
1021
1022
1023 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1024 {
1025 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1026 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1027 bitmap->bp[page].count += inc;
1028 /*
1029 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1030 (unsigned long long)offset, inc, bitmap->bp[page].count);
1031 */
1032 bitmap_checkfree(bitmap, page);
1033 }
1034 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1035 sector_t offset, int *blocks,
1036 int create);
1037
1038 /*
1039 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1040 * out to disk
1041 */
1042
1043 void bitmap_daemon_work(struct bitmap *bitmap)
1044 {
1045 unsigned long j;
1046 unsigned long flags;
1047 struct page *page = NULL, *lastpage = NULL;
1048 int blocks;
1049 void *paddr;
1050
1051 if (bitmap == NULL)
1052 return;
1053 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1054 return;
1055 bitmap->daemon_lastrun = jiffies;
1056
1057 for (j = 0; j < bitmap->chunks; j++) {
1058 bitmap_counter_t *bmc;
1059 spin_lock_irqsave(&bitmap->lock, flags);
1060 if (!bitmap->filemap) {
1061 /* error or shutdown */
1062 spin_unlock_irqrestore(&bitmap->lock, flags);
1063 break;
1064 }
1065
1066 page = filemap_get_page(bitmap, j);
1067
1068 if (page != lastpage) {
1069 /* skip this page unless it's marked as needing cleaning */
1070 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1071 int need_write = test_page_attr(bitmap, page,
1072 BITMAP_PAGE_NEEDWRITE);
1073 if (need_write)
1074 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1075
1076 spin_unlock_irqrestore(&bitmap->lock, flags);
1077 if (need_write)
1078 write_page(bitmap, page, 0);
1079 continue;
1080 }
1081
1082 /* grab the new page, sync and release the old */
1083 if (lastpage != NULL) {
1084 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1085 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1086 spin_unlock_irqrestore(&bitmap->lock, flags);
1087 write_page(bitmap, lastpage, 0);
1088 } else {
1089 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1090 spin_unlock_irqrestore(&bitmap->lock, flags);
1091 }
1092 } else
1093 spin_unlock_irqrestore(&bitmap->lock, flags);
1094 lastpage = page;
1095 /*
1096 printk("bitmap clean at page %lu\n", j);
1097 */
1098 spin_lock_irqsave(&bitmap->lock, flags);
1099 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1100 }
1101 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1102 &blocks, 0);
1103 if (bmc) {
1104 /*
1105 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1106 */
1107 if (*bmc == 2) {
1108 *bmc=1; /* maybe clear the bit next time */
1109 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1110 } else if (*bmc == 1) {
1111 /* we can clear the bit */
1112 *bmc = 0;
1113 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1114 -1);
1115
1116 /* clear the bit */
1117 paddr = kmap_atomic(page, KM_USER0);
1118 if (bitmap->flags & BITMAP_HOSTENDIAN)
1119 clear_bit(file_page_offset(j), paddr);
1120 else
1121 ext2_clear_bit(file_page_offset(j), paddr);
1122 kunmap_atomic(paddr, KM_USER0);
1123 }
1124 }
1125 spin_unlock_irqrestore(&bitmap->lock, flags);
1126 }
1127
1128 /* now sync the final page */
1129 if (lastpage != NULL) {
1130 spin_lock_irqsave(&bitmap->lock, flags);
1131 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1132 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1133 spin_unlock_irqrestore(&bitmap->lock, flags);
1134 write_page(bitmap, lastpage, 0);
1135 } else {
1136 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1137 spin_unlock_irqrestore(&bitmap->lock, flags);
1138 }
1139 }
1140
1141 }
1142
1143 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1144 sector_t offset, int *blocks,
1145 int create)
1146 {
1147 /* If 'create', we might release the lock and reclaim it.
1148 * The lock must have been taken with interrupts enabled.
1149 * If !create, we don't release the lock.
1150 */
1151 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1152 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1153 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1154 sector_t csize;
1155
1156 if (bitmap_checkpage(bitmap, page, create) < 0) {
1157 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1158 *blocks = csize - (offset & (csize- 1));
1159 return NULL;
1160 }
1161 /* now locked ... */
1162
1163 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1164 /* should we use the first or second counter field
1165 * of the hijacked pointer? */
1166 int hi = (pageoff > PAGE_COUNTER_MASK);
1167 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1168 PAGE_COUNTER_SHIFT - 1);
1169 *blocks = csize - (offset & (csize- 1));
1170 return &((bitmap_counter_t *)
1171 &bitmap->bp[page].map)[hi];
1172 } else { /* page is allocated */
1173 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1174 *blocks = csize - (offset & (csize- 1));
1175 return (bitmap_counter_t *)
1176 &(bitmap->bp[page].map[pageoff]);
1177 }
1178 }
1179
1180 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1181 {
1182 if (!bitmap) return 0;
1183
1184 if (behind) {
1185 atomic_inc(&bitmap->behind_writes);
1186 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1187 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1188 }
1189
1190 while (sectors) {
1191 int blocks;
1192 bitmap_counter_t *bmc;
1193
1194 spin_lock_irq(&bitmap->lock);
1195 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1196 if (!bmc) {
1197 spin_unlock_irq(&bitmap->lock);
1198 return 0;
1199 }
1200
1201 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1202 DEFINE_WAIT(__wait);
1203 /* note that it is safe to do the prepare_to_wait
1204 * after the test as long as we do it before dropping
1205 * the spinlock.
1206 */
1207 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1208 TASK_UNINTERRUPTIBLE);
1209 spin_unlock_irq(&bitmap->lock);
1210 bitmap->mddev->queue
1211 ->unplug_fn(bitmap->mddev->queue);
1212 schedule();
1213 finish_wait(&bitmap->overflow_wait, &__wait);
1214 continue;
1215 }
1216
1217 switch(*bmc) {
1218 case 0:
1219 bitmap_file_set_bit(bitmap, offset);
1220 bitmap_count_page(bitmap,offset, 1);
1221 blk_plug_device(bitmap->mddev->queue);
1222 /* fall through */
1223 case 1:
1224 *bmc = 2;
1225 }
1226
1227 (*bmc)++;
1228
1229 spin_unlock_irq(&bitmap->lock);
1230
1231 offset += blocks;
1232 if (sectors > blocks)
1233 sectors -= blocks;
1234 else sectors = 0;
1235 }
1236 return 0;
1237 }
1238
1239 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1240 int success, int behind)
1241 {
1242 if (!bitmap) return;
1243 if (behind) {
1244 atomic_dec(&bitmap->behind_writes);
1245 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1246 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1247 }
1248
1249 while (sectors) {
1250 int blocks;
1251 unsigned long flags;
1252 bitmap_counter_t *bmc;
1253
1254 spin_lock_irqsave(&bitmap->lock, flags);
1255 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1256 if (!bmc) {
1257 spin_unlock_irqrestore(&bitmap->lock, flags);
1258 return;
1259 }
1260
1261 if (!success && ! (*bmc & NEEDED_MASK))
1262 *bmc |= NEEDED_MASK;
1263
1264 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1265 wake_up(&bitmap->overflow_wait);
1266
1267 (*bmc)--;
1268 if (*bmc <= 2) {
1269 set_page_attr(bitmap,
1270 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1271 BITMAP_PAGE_CLEAN);
1272 }
1273 spin_unlock_irqrestore(&bitmap->lock, flags);
1274 offset += blocks;
1275 if (sectors > blocks)
1276 sectors -= blocks;
1277 else sectors = 0;
1278 }
1279 }
1280
1281 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1282 int degraded)
1283 {
1284 bitmap_counter_t *bmc;
1285 int rv;
1286 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1287 *blocks = 1024;
1288 return 1; /* always resync if no bitmap */
1289 }
1290 spin_lock_irq(&bitmap->lock);
1291 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1292 rv = 0;
1293 if (bmc) {
1294 /* locked */
1295 if (RESYNC(*bmc))
1296 rv = 1;
1297 else if (NEEDED(*bmc)) {
1298 rv = 1;
1299 if (!degraded) { /* don't set/clear bits if degraded */
1300 *bmc |= RESYNC_MASK;
1301 *bmc &= ~NEEDED_MASK;
1302 }
1303 }
1304 }
1305 spin_unlock_irq(&bitmap->lock);
1306 return rv;
1307 }
1308
1309 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1310 {
1311 bitmap_counter_t *bmc;
1312 unsigned long flags;
1313 /*
1314 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1315 */ if (bitmap == NULL) {
1316 *blocks = 1024;
1317 return;
1318 }
1319 spin_lock_irqsave(&bitmap->lock, flags);
1320 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1321 if (bmc == NULL)
1322 goto unlock;
1323 /* locked */
1324 /*
1325 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1326 */
1327 if (RESYNC(*bmc)) {
1328 *bmc &= ~RESYNC_MASK;
1329
1330 if (!NEEDED(*bmc) && aborted)
1331 *bmc |= NEEDED_MASK;
1332 else {
1333 if (*bmc <= 2) {
1334 set_page_attr(bitmap,
1335 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1336 BITMAP_PAGE_CLEAN);
1337 }
1338 }
1339 }
1340 unlock:
1341 spin_unlock_irqrestore(&bitmap->lock, flags);
1342 }
1343
1344 void bitmap_close_sync(struct bitmap *bitmap)
1345 {
1346 /* Sync has finished, and any bitmap chunks that weren't synced
1347 * properly have been aborted. It remains to us to clear the
1348 * RESYNC bit wherever it is still on
1349 */
1350 sector_t sector = 0;
1351 int blocks;
1352 if (!bitmap) return;
1353 while (sector < bitmap->mddev->resync_max_sectors) {
1354 bitmap_end_sync(bitmap, sector, &blocks, 0);
1355 /*
1356 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1357 (unsigned long long)sector, blocks);
1358 */ sector += blocks;
1359 }
1360 }
1361
1362 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1363 {
1364 /* For each chunk covered by any of these sectors, set the
1365 * counter to 1 and set resync_needed. They should all
1366 * be 0 at this point
1367 */
1368
1369 int secs;
1370 bitmap_counter_t *bmc;
1371 spin_lock_irq(&bitmap->lock);
1372 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1373 if (!bmc) {
1374 spin_unlock_irq(&bitmap->lock);
1375 return;
1376 }
1377 if (! *bmc) {
1378 struct page *page;
1379 *bmc = 1 | (needed?NEEDED_MASK:0);
1380 bitmap_count_page(bitmap, offset, 1);
1381 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1382 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1383 }
1384 spin_unlock_irq(&bitmap->lock);
1385
1386 }
1387
1388 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1389 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1390 {
1391 unsigned long chunk;
1392
1393 for (chunk = s; chunk <= e; chunk++) {
1394 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1395 bitmap_set_memory_bits(bitmap, sec, 1);
1396 bitmap_file_set_bit(bitmap, sec);
1397 }
1398 }
1399
1400 /*
1401 * flush out any pending updates
1402 */
1403 void bitmap_flush(mddev_t *mddev)
1404 {
1405 struct bitmap *bitmap = mddev->bitmap;
1406 int sleep;
1407
1408 if (!bitmap) /* there was no bitmap */
1409 return;
1410
1411 /* run the daemon_work three time to ensure everything is flushed
1412 * that can be
1413 */
1414 sleep = bitmap->daemon_sleep;
1415 bitmap->daemon_sleep = 0;
1416 bitmap_daemon_work(bitmap);
1417 bitmap_daemon_work(bitmap);
1418 bitmap_daemon_work(bitmap);
1419 bitmap->daemon_sleep = sleep;
1420 bitmap_update_sb(bitmap);
1421 }
1422
1423 /*
1424 * free memory that was allocated
1425 */
1426 static void bitmap_free(struct bitmap *bitmap)
1427 {
1428 unsigned long k, pages;
1429 struct bitmap_page *bp;
1430
1431 if (!bitmap) /* there was no bitmap */
1432 return;
1433
1434 /* release the bitmap file and kill the daemon */
1435 bitmap_file_put(bitmap);
1436
1437 bp = bitmap->bp;
1438 pages = bitmap->pages;
1439
1440 /* free all allocated memory */
1441
1442 if (bp) /* deallocate the page memory */
1443 for (k = 0; k < pages; k++)
1444 if (bp[k].map && !bp[k].hijacked)
1445 kfree(bp[k].map);
1446 kfree(bp);
1447 kfree(bitmap);
1448 }
1449 void bitmap_destroy(mddev_t *mddev)
1450 {
1451 struct bitmap *bitmap = mddev->bitmap;
1452
1453 if (!bitmap) /* there was no bitmap */
1454 return;
1455
1456 mddev->bitmap = NULL; /* disconnect from the md device */
1457 if (mddev->thread)
1458 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1459
1460 bitmap_free(bitmap);
1461 }
1462
1463 /*
1464 * initialize the bitmap structure
1465 * if this returns an error, bitmap_destroy must be called to do clean up
1466 */
1467 int bitmap_create(mddev_t *mddev)
1468 {
1469 struct bitmap *bitmap;
1470 unsigned long blocks = mddev->resync_max_sectors;
1471 unsigned long chunks;
1472 unsigned long pages;
1473 struct file *file = mddev->bitmap_file;
1474 int err;
1475 sector_t start;
1476
1477 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1478
1479 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1480 return 0;
1481
1482 BUG_ON(file && mddev->bitmap_offset);
1483
1484 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1485 if (!bitmap)
1486 return -ENOMEM;
1487
1488 spin_lock_init(&bitmap->lock);
1489 atomic_set(&bitmap->pending_writes, 0);
1490 init_waitqueue_head(&bitmap->write_wait);
1491 init_waitqueue_head(&bitmap->overflow_wait);
1492
1493 bitmap->mddev = mddev;
1494
1495 bitmap->file = file;
1496 bitmap->offset = mddev->bitmap_offset;
1497 if (file) {
1498 get_file(file);
1499 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1500 SYNC_FILE_RANGE_WAIT_BEFORE |
1501 SYNC_FILE_RANGE_WRITE |
1502 SYNC_FILE_RANGE_WAIT_AFTER);
1503 }
1504 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1505 err = bitmap_read_sb(bitmap);
1506 if (err)
1507 goto error;
1508
1509 bitmap->chunkshift = ffz(~bitmap->chunksize);
1510
1511 /* now that chunksize and chunkshift are set, we can use these macros */
1512 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1513 CHUNK_BLOCK_RATIO(bitmap);
1514 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1515
1516 BUG_ON(!pages);
1517
1518 bitmap->chunks = chunks;
1519 bitmap->pages = pages;
1520 bitmap->missing_pages = pages;
1521 bitmap->counter_bits = COUNTER_BITS;
1522
1523 bitmap->syncchunk = ~0UL;
1524
1525 #ifdef INJECT_FATAL_FAULT_1
1526 bitmap->bp = NULL;
1527 #else
1528 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1529 #endif
1530 err = -ENOMEM;
1531 if (!bitmap->bp)
1532 goto error;
1533
1534 /* now that we have some pages available, initialize the in-memory
1535 * bitmap from the on-disk bitmap */
1536 start = 0;
1537 if (mddev->degraded == 0
1538 || bitmap->events_cleared == mddev->events)
1539 /* no need to keep dirty bits to optimise a re-add of a missing device */
1540 start = mddev->recovery_cp;
1541 err = bitmap_init_from_disk(bitmap, start);
1542
1543 if (err)
1544 goto error;
1545
1546 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1547 pages, bmname(bitmap));
1548
1549 mddev->bitmap = bitmap;
1550
1551 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1552
1553 bitmap_update_sb(bitmap);
1554
1555 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1556
1557 error:
1558 bitmap_free(bitmap);
1559 return err;
1560 }
1561
1562 /* the bitmap API -- for raid personalities */
1563 EXPORT_SYMBOL(bitmap_startwrite);
1564 EXPORT_SYMBOL(bitmap_endwrite);
1565 EXPORT_SYMBOL(bitmap_start_sync);
1566 EXPORT_SYMBOL(bitmap_end_sync);
1567 EXPORT_SYMBOL(bitmap_unplug);
1568 EXPORT_SYMBOL(bitmap_close_sync);
Mirror for ia64_pv_ops.git