[PATCH] md: make sure the events count in an md array never returns to zero
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / bitmap.c
CommitLineData
32a7627c
N
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
32a7627c
N
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.
32a7627c
N
17 */
18
19#include <linux/module.h>
32a7627c
N
20#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/init.h>
32a7627c
N
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
64static inline char * bmname(struct bitmap *bitmap)
65{
66 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
67}
68
32a7627c
N
69
70/*
71 * just a placeholder - calls kmalloc for bitmap pages
72 */
73static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
74{
75 unsigned char *page;
76
44456d37 77#ifdef INJECT_FAULTS_1
32a7627c
N
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
a654b9d8 85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
86 bmname(bitmap), page);
87 return page;
88}
89
90/*
91 * for now just a placeholder -- just calls kfree for bitmap pages
92 */
93static 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 */
109static 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--;
162out:
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 */
858119e1 170static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
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 */
207char *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
c649bb9c
JS
215 d = file->f_path.dentry;
216 v = file->f_path.mnt;
32a7627c
N
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
a654b9d8
N
227/* IO operations when bitmap is stored near all superblocks */
228static 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);
a654b9d8 239
ab904d63 240 ITERATE_RDEV(mddev, rdev, tmp) {
b2d444d7
N
241 if (! test_bit(In_sync, &rdev->flags)
242 || test_bit(Faulty, &rdev->flags))
ab904d63
N
243 continue;
244
a654b9d8
N
245 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
246
ab904d63
N
247 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
248 page->index = index;
ce25c31b
N
249 attach_page_buffers(page, NULL); /* so that free_buffer will
250 * quietly no-op */
ab904d63
N
251 return page;
252 }
253 }
254 return ERR_PTR(-EIO);
a654b9d8 255
a654b9d8
N
256}
257
258static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
259{
260 mdk_rdev_t *rdev;
261 struct list_head *tmp;
262
263 ITERATE_RDEV(mddev, rdev, tmp)
b2d444d7
N
264 if (test_bit(In_sync, &rdev->flags)
265 && !test_bit(Faulty, &rdev->flags))
a654b9d8
N
266 md_super_write(mddev, rdev,
267 (rdev->sb_offset<<1) + offset
268 + page->index * (PAGE_SIZE/512),
269 PAGE_SIZE,
270 page);
271
272 if (wait)
a9701a30 273 md_super_wait(mddev);
a654b9d8
N
274 return 0;
275}
276
32a7627c 277/*
a654b9d8 278 * write out a page to a file
32a7627c 279 */
77ad4bc7 280static int write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 281{
d785a06a 282 struct buffer_head *bh;
32a7627c 283
a654b9d8
N
284 if (bitmap->file == NULL)
285 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
286
d785a06a 287 bh = page_buffers(page);
c708443c 288
d785a06a
N
289 while (bh && bh->b_blocknr) {
290 atomic_inc(&bitmap->pending_writes);
291 set_buffer_locked(bh);
292 set_buffer_mapped(bh);
293 submit_bh(WRITE, bh);
294 bh = bh->b_this_page;
295 }
296
297 if (wait) {
298 wait_event(bitmap->write_wait,
299 atomic_read(&bitmap->pending_writes)==0);
300 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
8a5e9cf1 301 }
d785a06a
N
302 return 0;
303}
304
305static void end_bitmap_write(struct buffer_head *bh, int uptodate)
306{
307 struct bitmap *bitmap = bh->b_private;
308 unsigned long flags;
32a7627c 309
d785a06a
N
310 if (!uptodate) {
311 spin_lock_irqsave(&bitmap->lock, flags);
312 bitmap->flags |= BITMAP_WRITE_ERROR;
313 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 314 }
d785a06a
N
315 if (atomic_dec_and_test(&bitmap->pending_writes))
316 wake_up(&bitmap->write_wait);
317}
32a7627c 318
d785a06a
N
319/* copied from buffer.c */
320static void
321__clear_page_buffers(struct page *page)
322{
323 ClearPagePrivate(page);
324 set_page_private(page, 0);
325 page_cache_release(page);
326}
327static void free_buffers(struct page *page)
328{
329 struct buffer_head *bh = page_buffers(page);
77ad4bc7 330
d785a06a
N
331 while (bh) {
332 struct buffer_head *next = bh->b_this_page;
333 free_buffer_head(bh);
334 bh = next;
77ad4bc7 335 }
d785a06a
N
336 __clear_page_buffers(page);
337 put_page(page);
32a7627c
N
338}
339
d785a06a
N
340/* read a page from a file.
341 * We both read the page, and attach buffers to the page to record the
342 * address of each block (using bmap). These addresses will be used
343 * to write the block later, completely bypassing the filesystem.
344 * This usage is similar to how swap files are handled, and allows us
345 * to write to a file with no concerns of memory allocation failing.
346 */
32a7627c 347static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
348 struct bitmap *bitmap,
349 unsigned long count)
32a7627c 350{
32a7627c 351 struct page *page = NULL;
c649bb9c 352 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
353 struct buffer_head *bh;
354 sector_t block;
32a7627c 355
2d1f3b5d
N
356 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
357 (unsigned long long)index << PAGE_SHIFT);
32a7627c 358
d785a06a
N
359 page = alloc_page(GFP_KERNEL);
360 if (!page)
361 page = ERR_PTR(-ENOMEM);
32a7627c
N
362 if (IS_ERR(page))
363 goto out;
d785a06a 364
d785a06a
N
365 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
366 if (!bh) {
2d1f3b5d 367 put_page(page);
d785a06a 368 page = ERR_PTR(-ENOMEM);
32a7627c
N
369 goto out;
370 }
d785a06a
N
371 attach_page_buffers(page, bh);
372 block = index << (PAGE_SHIFT - inode->i_blkbits);
373 while (bh) {
374 if (count == 0)
375 bh->b_blocknr = 0;
376 else {
377 bh->b_blocknr = bmap(inode, block);
378 if (bh->b_blocknr == 0) {
379 /* Cannot use this file! */
380 free_buffers(page);
381 page = ERR_PTR(-EINVAL);
382 goto out;
383 }
384 bh->b_bdev = inode->i_sb->s_bdev;
385 if (count < (1<<inode->i_blkbits))
386 count = 0;
387 else
388 count -= (1<<inode->i_blkbits);
389
390 bh->b_end_io = end_bitmap_write;
391 bh->b_private = bitmap;
ce25c31b
N
392 atomic_inc(&bitmap->pending_writes);
393 set_buffer_locked(bh);
394 set_buffer_mapped(bh);
395 submit_bh(READ, bh);
d785a06a
N
396 }
397 block++;
398 bh = bh->b_this_page;
399 }
d785a06a 400 page->index = index;
ce25c31b
N
401
402 wait_event(bitmap->write_wait,
403 atomic_read(&bitmap->pending_writes)==0);
404 if (bitmap->flags & BITMAP_WRITE_ERROR) {
405 free_buffers(page);
406 page = ERR_PTR(-EIO);
407 }
32a7627c
N
408out:
409 if (IS_ERR(page))
410 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
411 (int)PAGE_SIZE,
412 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
413 PTR_ERR(page));
414 return page;
415}
416
417/*
418 * bitmap file superblock operations
419 */
420
421/* update the event counter and sync the superblock to disk */
422int bitmap_update_sb(struct bitmap *bitmap)
423{
424 bitmap_super_t *sb;
425 unsigned long flags;
426
427 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
428 return 0;
429 spin_lock_irqsave(&bitmap->lock, flags);
430 if (!bitmap->sb_page) { /* no superblock */
431 spin_unlock_irqrestore(&bitmap->lock, flags);
432 return 0;
433 }
32a7627c 434 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 435 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
436 sb->events = cpu_to_le64(bitmap->mddev->events);
437 if (!bitmap->mddev->degraded)
438 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
ea03aff9 439 kunmap_atomic(sb, KM_USER0);
8a5e9cf1 440 return write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
441}
442
443/* print out the bitmap file superblock */
444void bitmap_print_sb(struct bitmap *bitmap)
445{
446 bitmap_super_t *sb;
447
448 if (!bitmap || !bitmap->sb_page)
449 return;
ea03aff9 450 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 451 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
452 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
453 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
454 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
455 *(__u32 *)(sb->uuid+0),
456 *(__u32 *)(sb->uuid+4),
457 *(__u32 *)(sb->uuid+8),
458 *(__u32 *)(sb->uuid+12));
a2cff26a 459 printk(KERN_DEBUG " events: %llu\n",
32a7627c 460 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 461 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 462 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
463 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
464 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
465 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
466 printk(KERN_DEBUG " sync size: %llu KB\n",
467 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 468 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 469 kunmap_atomic(sb, KM_USER0);
32a7627c
N
470}
471
472/* read the superblock from the bitmap file and initialize some bitmap fields */
473static int bitmap_read_sb(struct bitmap *bitmap)
474{
475 char *reason = NULL;
476 bitmap_super_t *sb;
4b6d287f 477 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
478 unsigned long long events;
479 int err = -EINVAL;
480
481 /* page 0 is the superblock, read it... */
a654b9d8 482 if (bitmap->file)
d785a06a 483 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, PAGE_SIZE);
a654b9d8
N
484 else {
485 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 486 }
32a7627c
N
487 if (IS_ERR(bitmap->sb_page)) {
488 err = PTR_ERR(bitmap->sb_page);
489 bitmap->sb_page = NULL;
490 return err;
491 }
492
ea03aff9 493 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 494
32a7627c
N
495 chunksize = le32_to_cpu(sb->chunksize);
496 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 497 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
498
499 /* verify that the bitmap-specific fields are valid */
500 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
501 reason = "bad magic";
bd926c63
N
502 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
503 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 504 reason = "unrecognized superblock version";
7dd5d34c
N
505 else if (chunksize < PAGE_SIZE)
506 reason = "bitmap chunksize too small";
32a7627c
N
507 else if ((1 << ffz(~chunksize)) != chunksize)
508 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
509 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
510 reason = "daemon sleep period out of range";
4b6d287f
N
511 else if (write_behind > COUNTER_MAX)
512 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
513 if (reason) {
514 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
515 bmname(bitmap), reason);
516 goto out;
517 }
518
519 /* keep the array size field of the bitmap superblock up to date */
520 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
521
522 if (!bitmap->mddev->persistent)
523 goto success;
524
525 /*
526 * if we have a persistent array superblock, compare the
527 * bitmap's UUID and event counter to the mddev's
528 */
529 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
530 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
531 bmname(bitmap));
532 goto out;
533 }
534 events = le64_to_cpu(sb->events);
535 if (events < bitmap->mddev->events) {
536 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
537 "-- forcing full recovery\n", bmname(bitmap), events,
538 (unsigned long long) bitmap->mddev->events);
4f2e639a 539 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
540 }
541success:
542 /* assign fields using values from superblock */
543 bitmap->chunksize = chunksize;
544 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 545 bitmap->daemon_lastrun = jiffies;
4b6d287f 546 bitmap->max_write_behind = write_behind;
4f2e639a 547 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
548 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
549 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 550 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 551 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 552 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
553 err = 0;
554out:
ea03aff9 555 kunmap_atomic(sb, KM_USER0);
32a7627c
N
556 if (err)
557 bitmap_print_sb(bitmap);
558 return err;
559}
560
561enum bitmap_mask_op {
562 MASK_SET,
563 MASK_UNSET
564};
565
566/* record the state of the bitmap in the superblock */
567static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
568 enum bitmap_mask_op op)
569{
570 bitmap_super_t *sb;
571 unsigned long flags;
572
573 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 574 if (!bitmap->sb_page) { /* can't set the state */
32a7627c
N
575 spin_unlock_irqrestore(&bitmap->lock, flags);
576 return;
577 }
32a7627c 578 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 579 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 580 switch (op) {
4f2e639a 581 case MASK_SET: sb->state |= cpu_to_le32(bits);
32a7627c 582 break;
4f2e639a 583 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
32a7627c
N
584 break;
585 default: BUG();
586 }
ea03aff9 587 kunmap_atomic(sb, KM_USER0);
32a7627c
N
588}
589
590/*
591 * general bitmap file operations
592 */
593
594/* calculate the index of the page that contains this bit */
595static inline unsigned long file_page_index(unsigned long chunk)
596{
597 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
598}
599
600/* calculate the (bit) offset of this bit within a page */
601static inline unsigned long file_page_offset(unsigned long chunk)
602{
603 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
604}
605
606/*
607 * return a pointer to the page in the filemap that contains the given bit
608 *
609 * this lookup is complicated by the fact that the bitmap sb might be exactly
610 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
611 * 0 or page 1
612 */
613static inline struct page *filemap_get_page(struct bitmap *bitmap,
614 unsigned long chunk)
615{
9b1d1dac 616 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
32a7627c
N
617 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
618}
619
620
621static void bitmap_file_unmap(struct bitmap *bitmap)
622{
623 struct page **map, *sb_page;
624 unsigned long *attr;
625 int pages;
626 unsigned long flags;
627
628 spin_lock_irqsave(&bitmap->lock, flags);
629 map = bitmap->filemap;
630 bitmap->filemap = NULL;
631 attr = bitmap->filemap_attr;
632 bitmap->filemap_attr = NULL;
633 pages = bitmap->file_pages;
634 bitmap->file_pages = 0;
635 sb_page = bitmap->sb_page;
636 bitmap->sb_page = NULL;
637 spin_unlock_irqrestore(&bitmap->lock, flags);
638
639 while (pages--)
640 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
d785a06a 641 free_buffers(map[pages]);
32a7627c
N
642 kfree(map);
643 kfree(attr);
644
d785a06a
N
645 if (sb_page)
646 free_buffers(sb_page);
32a7627c
N
647}
648
649static void bitmap_file_put(struct bitmap *bitmap)
650{
651 struct file *file;
32a7627c
N
652 unsigned long flags;
653
654 spin_lock_irqsave(&bitmap->lock, flags);
655 file = bitmap->file;
656 bitmap->file = NULL;
657 spin_unlock_irqrestore(&bitmap->lock, flags);
658
d785a06a
N
659 if (file)
660 wait_event(bitmap->write_wait,
661 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
662 bitmap_file_unmap(bitmap);
663
d785a06a 664 if (file) {
c649bb9c 665 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a 666 invalidate_inode_pages(inode->i_mapping);
32a7627c 667 fput(file);
d785a06a 668 }
32a7627c
N
669}
670
671
672/*
673 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
674 * then it is no longer reliable, so we stop using it and we mark the file
675 * as failed in the superblock
676 */
677static void bitmap_file_kick(struct bitmap *bitmap)
678{
679 char *path, *ptr = NULL;
680
681 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
682 bitmap_update_sb(bitmap);
683
a654b9d8
N
684 if (bitmap->file) {
685 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
686 if (path)
687 ptr = file_path(bitmap->file, path, PAGE_SIZE);
32a7627c 688
a654b9d8
N
689 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
690 bmname(bitmap), ptr ? ptr : "");
32a7627c 691
a654b9d8
N
692 kfree(path);
693 }
32a7627c
N
694
695 bitmap_file_put(bitmap);
696
697 return;
698}
699
700enum bitmap_page_attr {
e16b68b6
N
701 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
702 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
703 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
704};
705
706static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
707 enum bitmap_page_attr attr)
708{
e16b68b6 709 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
710}
711
712static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
713 enum bitmap_page_attr attr)
714{
e16b68b6 715 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
716}
717
ec7a3197
N
718static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
719 enum bitmap_page_attr attr)
32a7627c 720{
e16b68b6 721 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
722}
723
724/*
725 * bitmap_file_set_bit -- called before performing a write to the md device
726 * to set (and eventually sync) a particular bit in the bitmap file
727 *
728 * we set the bit immediately, then we record the page number so that
729 * when an unplug occurs, we can flush the dirty pages out to disk
730 */
731static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
732{
733 unsigned long bit;
734 struct page *page;
735 void *kaddr;
736 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
737
a654b9d8 738 if (!bitmap->filemap) {
32a7627c
N
739 return;
740 }
741
742 page = filemap_get_page(bitmap, chunk);
9b1d1dac 743 if (!page) return;
32a7627c
N
744 bit = file_page_offset(chunk);
745
32a7627c
N
746 /* set the bit */
747 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
748 if (bitmap->flags & BITMAP_HOSTENDIAN)
749 set_bit(bit, kaddr);
750 else
751 ext2_set_bit(bit, kaddr);
32a7627c
N
752 kunmap_atomic(kaddr, KM_USER0);
753 PRINTK("set file bit %lu page %lu\n", bit, page->index);
754
755 /* record page number so it gets flushed to disk when unplug occurs */
756 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
757
758}
759
760/* this gets called when the md device is ready to unplug its underlying
761 * (slave) device queues -- before we let any writes go down, we need to
762 * sync the dirty pages of the bitmap file to disk */
763int bitmap_unplug(struct bitmap *bitmap)
764{
ec7a3197
N
765 unsigned long i, flags;
766 int dirty, need_write;
32a7627c
N
767 struct page *page;
768 int wait = 0;
8a5e9cf1 769 int err;
32a7627c
N
770
771 if (!bitmap)
772 return 0;
773
774 /* look at each page to see if there are any set bits that need to be
775 * flushed out to disk */
776 for (i = 0; i < bitmap->file_pages; i++) {
777 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 778 if (!bitmap->filemap) {
32a7627c
N
779 spin_unlock_irqrestore(&bitmap->lock, flags);
780 return 0;
781 }
782 page = bitmap->filemap[i];
ec7a3197
N
783 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
784 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
785 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
786 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 787 if (dirty)
32a7627c
N
788 wait = 1;
789 spin_unlock_irqrestore(&bitmap->lock, flags);
790
d785a06a 791 if (dirty | need_write)
8a5e9cf1 792 err = write_page(bitmap, page, 0);
32a7627c
N
793 }
794 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 795 if (bitmap->file)
d785a06a
N
796 wait_event(bitmap->write_wait,
797 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 798 else
a9701a30 799 md_super_wait(bitmap->mddev);
32a7627c 800 }
d785a06a
N
801 if (bitmap->flags & BITMAP_WRITE_ERROR)
802 bitmap_file_kick(bitmap);
32a7627c
N
803 return 0;
804}
805
6a07997f 806static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
807/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
808 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
809 * memory mapping of the bitmap file
810 * Special cases:
811 * if there's no bitmap file, or if the bitmap file had been
812 * previously kicked from the array, we mark all the bits as
813 * 1's in order to cause a full resync.
6a07997f
N
814 *
815 * We ignore all bits for sectors that end earlier than 'start'.
816 * This is used when reading an out-of-date bitmap...
32a7627c 817 */
6a07997f 818static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
819{
820 unsigned long i, chunks, index, oldindex, bit;
821 struct page *page = NULL, *oldpage = NULL;
822 unsigned long num_pages, bit_cnt = 0;
823 struct file *file;
d785a06a 824 unsigned long bytes, offset;
32a7627c
N
825 int outofdate;
826 int ret = -ENOSPC;
ea03aff9 827 void *paddr;
32a7627c
N
828
829 chunks = bitmap->chunks;
830 file = bitmap->file;
831
a654b9d8 832 BUG_ON(!file && !bitmap->offset);
32a7627c 833
44456d37 834#ifdef INJECT_FAULTS_3
32a7627c
N
835 outofdate = 1;
836#else
837 outofdate = bitmap->flags & BITMAP_STALE;
838#endif
839 if (outofdate)
840 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
841 "recovery\n", bmname(bitmap));
842
843 bytes = (chunks + 7) / 8;
bc7f77de 844
cdbb4cc2 845 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 846
a654b9d8 847 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
848 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
849 bmname(bitmap),
850 (unsigned long) i_size_read(file->f_mapping->host),
851 bytes + sizeof(bitmap_super_t));
852 goto out;
853 }
bc7f77de
N
854
855 ret = -ENOMEM;
856
32a7627c 857 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 858 if (!bitmap->filemap)
32a7627c 859 goto out;
32a7627c 860
e16b68b6
N
861 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
862 bitmap->filemap_attr = kzalloc(
863 (((num_pages*4/8)+sizeof(unsigned long)-1)
864 /sizeof(unsigned long))
865 *sizeof(unsigned long),
866 GFP_KERNEL);
bc7f77de 867 if (!bitmap->filemap_attr)
32a7627c 868 goto out;
32a7627c 869
32a7627c
N
870 oldindex = ~0L;
871
872 for (i = 0; i < chunks; i++) {
bd926c63 873 int b;
32a7627c
N
874 index = file_page_index(i);
875 bit = file_page_offset(i);
876 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 877 int count;
32a7627c 878 /* unmap the old page, we're done with it */
d785a06a
N
879 if (index == num_pages-1)
880 count = bytes - index * PAGE_SIZE;
881 else
882 count = PAGE_SIZE;
32a7627c
N
883 if (index == 0) {
884 /*
885 * if we're here then the superblock page
886 * contains some bits (PAGE_SIZE != sizeof sb)
887 * we've already read it in, so just use it
888 */
889 page = bitmap->sb_page;
890 offset = sizeof(bitmap_super_t);
a654b9d8 891 } else if (file) {
d785a06a 892 page = read_page(file, index, bitmap, count);
a654b9d8
N
893 offset = 0;
894 } else {
895 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
896 offset = 0;
897 }
a654b9d8
N
898 if (IS_ERR(page)) { /* read error */
899 ret = PTR_ERR(page);
900 goto out;
901 }
902
32a7627c
N
903 oldindex = index;
904 oldpage = page;
32a7627c
N
905
906 if (outofdate) {
907 /*
908 * if bitmap is out of date, dirty the
909 * whole page and write it out
910 */
ea03aff9
N
911 paddr = kmap_atomic(page, KM_USER0);
912 memset(paddr + offset, 0xff,
6a07997f 913 PAGE_SIZE - offset);
ea03aff9 914 kunmap_atomic(paddr, KM_USER0);
77ad4bc7 915 ret = write_page(bitmap, page, 1);
32a7627c 916 if (ret) {
32a7627c 917 /* release, page not in filemap yet */
2d1f3b5d 918 put_page(page);
32a7627c
N
919 goto out;
920 }
921 }
922
923 bitmap->filemap[bitmap->file_pages++] = page;
924 }
ea03aff9 925 paddr = kmap_atomic(page, KM_USER0);
bd926c63 926 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 927 b = test_bit(bit, paddr);
bd926c63 928 else
ea03aff9
N
929 b = ext2_test_bit(bit, paddr);
930 kunmap_atomic(paddr, KM_USER0);
bd926c63 931 if (b) {
32a7627c 932 /* if the disk bit is set, set the memory bit */
6a07997f
N
933 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
934 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
935 );
32a7627c 936 bit_cnt++;
6a07997f 937 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 938 }
32a7627c
N
939 }
940
941 /* everything went OK */
942 ret = 0;
943 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
944
32a7627c
N
945 if (bit_cnt) { /* Kick recovery if any bits were set */
946 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
947 md_wakeup_thread(bitmap->mddev->thread);
948 }
949
950out:
951 printk(KERN_INFO "%s: bitmap initialized from disk: "
952 "read %lu/%lu pages, set %lu bits, status: %d\n",
953 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
954
955 return ret;
956}
957
a654b9d8
N
958void bitmap_write_all(struct bitmap *bitmap)
959{
960 /* We don't actually write all bitmap blocks here,
961 * just flag them as needing to be written
962 */
ec7a3197 963 int i;
a654b9d8 964
ec7a3197
N
965 for (i=0; i < bitmap->file_pages; i++)
966 set_page_attr(bitmap, bitmap->filemap[i],
967 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
968}
969
32a7627c
N
970
971static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
972{
973 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
974 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
975 bitmap->bp[page].count += inc;
976/*
977 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
978 (unsigned long long)offset, inc, bitmap->bp[page].count);
979*/
980 bitmap_checkfree(bitmap, page);
981}
982static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
983 sector_t offset, int *blocks,
984 int create);
985
986/*
987 * bitmap daemon -- periodically wakes up to clean bits and flush pages
988 * out to disk
989 */
990
991int bitmap_daemon_work(struct bitmap *bitmap)
992{
aa3163f8 993 unsigned long j;
32a7627c
N
994 unsigned long flags;
995 struct page *page = NULL, *lastpage = NULL;
996 int err = 0;
997 int blocks;
ea03aff9 998 void *paddr;
32a7627c
N
999
1000 if (bitmap == NULL)
1001 return 0;
1002 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1003 return 0;
1004 bitmap->daemon_lastrun = jiffies;
1005
1006 for (j = 0; j < bitmap->chunks; j++) {
1007 bitmap_counter_t *bmc;
1008 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1009 if (!bitmap->filemap) {
32a7627c
N
1010 /* error or shutdown */
1011 spin_unlock_irqrestore(&bitmap->lock, flags);
1012 break;
1013 }
1014
1015 page = filemap_get_page(bitmap, j);
32a7627c
N
1016
1017 if (page != lastpage) {
aa3163f8 1018 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1019 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1020 int need_write = test_page_attr(bitmap, page,
1021 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1022 if (need_write)
aa3163f8 1023 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1024
aa3163f8 1025 spin_unlock_irqrestore(&bitmap->lock, flags);
ec7a3197 1026 if (need_write) {
8a5e9cf1 1027 switch (write_page(bitmap, page, 0)) {
8a5e9cf1
N
1028 case 0:
1029 break;
1030 default:
aa3163f8 1031 bitmap_file_kick(bitmap);
8a5e9cf1 1032 }
aa3163f8
N
1033 }
1034 continue;
1035 }
1036
32a7627c 1037 /* grab the new page, sync and release the old */
32a7627c 1038 if (lastpage != NULL) {
ec7a3197 1039 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1040 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1041 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1042 err = write_page(bitmap, lastpage, 0);
32a7627c
N
1043 } else {
1044 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1045 spin_unlock_irqrestore(&bitmap->lock, flags);
1046 }
32a7627c
N
1047 if (err)
1048 bitmap_file_kick(bitmap);
1049 } else
1050 spin_unlock_irqrestore(&bitmap->lock, flags);
1051 lastpage = page;
32a7627c
N
1052/*
1053 printk("bitmap clean at page %lu\n", j);
1054*/
1055 spin_lock_irqsave(&bitmap->lock, flags);
1056 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1057 }
1058 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1059 &blocks, 0);
1060 if (bmc) {
1061/*
1062 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1063*/
1064 if (*bmc == 2) {
1065 *bmc=1; /* maybe clear the bit next time */
1066 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1067 } else if (*bmc == 1) {
1068 /* we can clear the bit */
1069 *bmc = 0;
1070 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1071 -1);
1072
1073 /* clear the bit */
ea03aff9 1074 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1075 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1076 clear_bit(file_page_offset(j), paddr);
bd926c63 1077 else
ea03aff9
N
1078 ext2_clear_bit(file_page_offset(j), paddr);
1079 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1080 }
1081 }
1082 spin_unlock_irqrestore(&bitmap->lock, flags);
1083 }
1084
1085 /* now sync the final page */
1086 if (lastpage != NULL) {
32a7627c 1087 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1088 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1089 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1090 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1091 err = write_page(bitmap, lastpage, 0);
32a7627c
N
1092 } else {
1093 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1094 spin_unlock_irqrestore(&bitmap->lock, flags);
1095 }
32a7627c
N
1096 }
1097
1098 return err;
1099}
1100
32a7627c
N
1101static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1102 sector_t offset, int *blocks,
1103 int create)
1104{
1105 /* If 'create', we might release the lock and reclaim it.
1106 * The lock must have been taken with interrupts enabled.
1107 * If !create, we don't release the lock.
1108 */
1109 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1110 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1111 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1112 sector_t csize;
1113
1114 if (bitmap_checkpage(bitmap, page, create) < 0) {
1115 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1116 *blocks = csize - (offset & (csize- 1));
1117 return NULL;
1118 }
1119 /* now locked ... */
1120
1121 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1122 /* should we use the first or second counter field
1123 * of the hijacked pointer? */
1124 int hi = (pageoff > PAGE_COUNTER_MASK);
1125 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1126 PAGE_COUNTER_SHIFT - 1);
1127 *blocks = csize - (offset & (csize- 1));
1128 return &((bitmap_counter_t *)
1129 &bitmap->bp[page].map)[hi];
1130 } else { /* page is allocated */
1131 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1132 *blocks = csize - (offset & (csize- 1));
1133 return (bitmap_counter_t *)
1134 &(bitmap->bp[page].map[pageoff]);
1135 }
1136}
1137
4b6d287f 1138int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1139{
1140 if (!bitmap) return 0;
4b6d287f
N
1141
1142 if (behind) {
1143 atomic_inc(&bitmap->behind_writes);
1144 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1145 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1146 }
1147
32a7627c
N
1148 while (sectors) {
1149 int blocks;
1150 bitmap_counter_t *bmc;
1151
1152 spin_lock_irq(&bitmap->lock);
1153 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1154 if (!bmc) {
1155 spin_unlock_irq(&bitmap->lock);
1156 return 0;
1157 }
1158
1159 switch(*bmc) {
1160 case 0:
1161 bitmap_file_set_bit(bitmap, offset);
1162 bitmap_count_page(bitmap,offset, 1);
1163 blk_plug_device(bitmap->mddev->queue);
1164 /* fall through */
1165 case 1:
1166 *bmc = 2;
1167 }
5daf2cf1 1168 BUG_ON((*bmc & COUNTER_MAX) == COUNTER_MAX);
32a7627c
N
1169 (*bmc)++;
1170
1171 spin_unlock_irq(&bitmap->lock);
1172
1173 offset += blocks;
1174 if (sectors > blocks)
1175 sectors -= blocks;
1176 else sectors = 0;
1177 }
1178 return 0;
1179}
1180
1181void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1182 int success, int behind)
32a7627c
N
1183{
1184 if (!bitmap) return;
4b6d287f
N
1185 if (behind) {
1186 atomic_dec(&bitmap->behind_writes);
1187 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1188 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1189 }
1190
32a7627c
N
1191 while (sectors) {
1192 int blocks;
1193 unsigned long flags;
1194 bitmap_counter_t *bmc;
1195
1196 spin_lock_irqsave(&bitmap->lock, flags);
1197 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1198 if (!bmc) {
1199 spin_unlock_irqrestore(&bitmap->lock, flags);
1200 return;
1201 }
1202
1203 if (!success && ! (*bmc & NEEDED_MASK))
1204 *bmc |= NEEDED_MASK;
1205
1206 (*bmc)--;
1207 if (*bmc <= 2) {
1208 set_page_attr(bitmap,
1209 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1210 BITMAP_PAGE_CLEAN);
1211 }
1212 spin_unlock_irqrestore(&bitmap->lock, flags);
1213 offset += blocks;
1214 if (sectors > blocks)
1215 sectors -= blocks;
1216 else sectors = 0;
1217 }
1218}
1219
6a806c51
N
1220int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1221 int degraded)
32a7627c
N
1222{
1223 bitmap_counter_t *bmc;
1224 int rv;
1225 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1226 *blocks = 1024;
1227 return 1; /* always resync if no bitmap */
1228 }
1229 spin_lock_irq(&bitmap->lock);
1230 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1231 rv = 0;
1232 if (bmc) {
1233 /* locked */
1234 if (RESYNC(*bmc))
1235 rv = 1;
1236 else if (NEEDED(*bmc)) {
1237 rv = 1;
6a806c51
N
1238 if (!degraded) { /* don't set/clear bits if degraded */
1239 *bmc |= RESYNC_MASK;
1240 *bmc &= ~NEEDED_MASK;
1241 }
32a7627c
N
1242 }
1243 }
1244 spin_unlock_irq(&bitmap->lock);
1245 return rv;
1246}
1247
1248void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1249{
1250 bitmap_counter_t *bmc;
1251 unsigned long flags;
1252/*
1253 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1254*/ if (bitmap == NULL) {
1255 *blocks = 1024;
1256 return;
1257 }
1258 spin_lock_irqsave(&bitmap->lock, flags);
1259 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1260 if (bmc == NULL)
1261 goto unlock;
1262 /* locked */
1263/*
1264 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1265*/
1266 if (RESYNC(*bmc)) {
1267 *bmc &= ~RESYNC_MASK;
1268
1269 if (!NEEDED(*bmc) && aborted)
1270 *bmc |= NEEDED_MASK;
1271 else {
1272 if (*bmc <= 2) {
1273 set_page_attr(bitmap,
1274 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1275 BITMAP_PAGE_CLEAN);
1276 }
1277 }
1278 }
1279 unlock:
1280 spin_unlock_irqrestore(&bitmap->lock, flags);
1281}
1282
1283void bitmap_close_sync(struct bitmap *bitmap)
1284{
1285 /* Sync has finished, and any bitmap chunks that weren't synced
1286 * properly have been aborted. It remains to us to clear the
1287 * RESYNC bit wherever it is still on
1288 */
1289 sector_t sector = 0;
1290 int blocks;
1291 if (!bitmap) return;
1292 while (sector < bitmap->mddev->resync_max_sectors) {
1293 bitmap_end_sync(bitmap, sector, &blocks, 0);
1294/*
1295 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1296 (unsigned long long)sector, blocks);
1297*/ sector += blocks;
1298 }
1299}
1300
6a07997f 1301static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1302{
1303 /* For each chunk covered by any of these sectors, set the
193f1c93 1304 * counter to 1 and set resync_needed. They should all
32a7627c
N
1305 * be 0 at this point
1306 */
193f1c93
N
1307
1308 int secs;
1309 bitmap_counter_t *bmc;
1310 spin_lock_irq(&bitmap->lock);
1311 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1312 if (!bmc) {
32a7627c 1313 spin_unlock_irq(&bitmap->lock);
193f1c93 1314 return;
32a7627c 1315 }
193f1c93
N
1316 if (! *bmc) {
1317 struct page *page;
6a07997f 1318 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1319 bitmap_count_page(bitmap, offset, 1);
1320 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1321 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1322 }
1323 spin_unlock_irq(&bitmap->lock);
1324
32a7627c
N
1325}
1326
9b1d1dac
PC
1327/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1328void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1329{
1330 unsigned long chunk;
1331
1332 for (chunk = s; chunk <= e; chunk++) {
1333 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1334 bitmap_set_memory_bits(bitmap, sec, 1);
1335 bitmap_file_set_bit(bitmap, sec);
1336 }
1337}
1338
6b8b3e8a
N
1339/*
1340 * flush out any pending updates
1341 */
1342void bitmap_flush(mddev_t *mddev)
1343{
1344 struct bitmap *bitmap = mddev->bitmap;
1345 int sleep;
1346
1347 if (!bitmap) /* there was no bitmap */
1348 return;
1349
1350 /* run the daemon_work three time to ensure everything is flushed
1351 * that can be
1352 */
1353 sleep = bitmap->daemon_sleep;
1354 bitmap->daemon_sleep = 0;
1355 bitmap_daemon_work(bitmap);
1356 bitmap_daemon_work(bitmap);
1357 bitmap_daemon_work(bitmap);
1358 bitmap->daemon_sleep = sleep;
1359 bitmap_update_sb(bitmap);
1360}
1361
32a7627c
N
1362/*
1363 * free memory that was allocated
1364 */
3178b0db 1365static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1366{
1367 unsigned long k, pages;
1368 struct bitmap_page *bp;
32a7627c
N
1369
1370 if (!bitmap) /* there was no bitmap */
1371 return;
1372
32a7627c
N
1373 /* release the bitmap file and kill the daemon */
1374 bitmap_file_put(bitmap);
1375
1376 bp = bitmap->bp;
1377 pages = bitmap->pages;
1378
1379 /* free all allocated memory */
1380
32a7627c
N
1381 if (bp) /* deallocate the page memory */
1382 for (k = 0; k < pages; k++)
1383 if (bp[k].map && !bp[k].hijacked)
1384 kfree(bp[k].map);
1385 kfree(bp);
1386 kfree(bitmap);
1387}
3178b0db
N
1388void bitmap_destroy(mddev_t *mddev)
1389{
1390 struct bitmap *bitmap = mddev->bitmap;
1391
1392 if (!bitmap) /* there was no bitmap */
1393 return;
1394
1395 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1396 if (mddev->thread)
1397 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1398
1399 bitmap_free(bitmap);
1400}
32a7627c
N
1401
1402/*
1403 * initialize the bitmap structure
1404 * if this returns an error, bitmap_destroy must be called to do clean up
1405 */
1406int bitmap_create(mddev_t *mddev)
1407{
1408 struct bitmap *bitmap;
1409 unsigned long blocks = mddev->resync_max_sectors;
1410 unsigned long chunks;
1411 unsigned long pages;
1412 struct file *file = mddev->bitmap_file;
1413 int err;
6a07997f 1414 sector_t start;
32a7627c 1415
5f6e3c83 1416 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1417
a654b9d8 1418 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1419 return 0;
1420
a654b9d8
N
1421 BUG_ON(file && mddev->bitmap_offset);
1422
9ffae0cf 1423 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1424 if (!bitmap)
1425 return -ENOMEM;
1426
32a7627c 1427 spin_lock_init(&bitmap->lock);
ce25c31b
N
1428 atomic_set(&bitmap->pending_writes, 0);
1429 init_waitqueue_head(&bitmap->write_wait);
1430
32a7627c 1431 bitmap->mddev = mddev;
32a7627c 1432
32a7627c 1433 bitmap->file = file;
a654b9d8 1434 bitmap->offset = mddev->bitmap_offset;
ce25c31b
N
1435 if (file) {
1436 get_file(file);
1437 do_sync_file_range(file, 0, LLONG_MAX,
1438 SYNC_FILE_RANGE_WAIT_BEFORE |
1439 SYNC_FILE_RANGE_WRITE |
1440 SYNC_FILE_RANGE_WAIT_AFTER);
1441 }
32a7627c
N
1442 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1443 err = bitmap_read_sb(bitmap);
1444 if (err)
3178b0db 1445 goto error;
32a7627c 1446
a638b2dc 1447 bitmap->chunkshift = ffz(~bitmap->chunksize);
32a7627c
N
1448
1449 /* now that chunksize and chunkshift are set, we can use these macros */
1450 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1451 CHUNK_BLOCK_RATIO(bitmap);
1452 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1453
1454 BUG_ON(!pages);
1455
1456 bitmap->chunks = chunks;
1457 bitmap->pages = pages;
1458 bitmap->missing_pages = pages;
1459 bitmap->counter_bits = COUNTER_BITS;
1460
1461 bitmap->syncchunk = ~0UL;
1462
44456d37 1463#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1464 bitmap->bp = NULL;
1465#else
9ffae0cf 1466 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1467#endif
3178b0db 1468 err = -ENOMEM;
32a7627c 1469 if (!bitmap->bp)
3178b0db 1470 goto error;
32a7627c 1471
32a7627c
N
1472 /* now that we have some pages available, initialize the in-memory
1473 * bitmap from the on-disk bitmap */
6a07997f
N
1474 start = 0;
1475 if (mddev->degraded == 0
1476 || bitmap->events_cleared == mddev->events)
1477 /* no need to keep dirty bits to optimise a re-add of a missing device */
1478 start = mddev->recovery_cp;
1479 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1480
32a7627c 1481 if (err)
3178b0db 1482 goto error;
32a7627c
N
1483
1484 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1485 pages, bmname(bitmap));
1486
3178b0db
N
1487 mddev->bitmap = bitmap;
1488
b15c2e57
N
1489 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1490
32a7627c 1491 return bitmap_update_sb(bitmap);
3178b0db
N
1492
1493 error:
1494 bitmap_free(bitmap);
1495 return err;
32a7627c
N
1496}
1497
1498/* the bitmap API -- for raid personalities */
1499EXPORT_SYMBOL(bitmap_startwrite);
1500EXPORT_SYMBOL(bitmap_endwrite);
1501EXPORT_SYMBOL(bitmap_start_sync);
1502EXPORT_SYMBOL(bitmap_end_sync);
1503EXPORT_SYMBOL(bitmap_unplug);
1504EXPORT_SYMBOL(bitmap_close_sync);