md: only count actual openers as access which prevent a 'stop'
[linux-2.6-block.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
32a7627c
N
206/*
207 * basic page I/O operations
208 */
209
a654b9d8
N
210/* IO operations when bitmap is stored near all superblocks */
211static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
212{
213 /* choose a good rdev and read the page from there */
214
215 mdk_rdev_t *rdev;
216 struct list_head *tmp;
217 struct page *page = alloc_page(GFP_KERNEL);
218 sector_t target;
219
220 if (!page)
221 return ERR_PTR(-ENOMEM);
a654b9d8 222
d089c6af 223 rdev_for_each(rdev, tmp, mddev) {
b2d444d7
N
224 if (! test_bit(In_sync, &rdev->flags)
225 || test_bit(Faulty, &rdev->flags))
ab904d63
N
226 continue;
227
0f420358 228 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
a654b9d8 229
ab904d63
N
230 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
231 page->index = index;
ce25c31b
N
232 attach_page_buffers(page, NULL); /* so that free_buffer will
233 * quietly no-op */
ab904d63
N
234 return page;
235 }
236 }
237 return ERR_PTR(-EIO);
a654b9d8 238
a654b9d8
N
239}
240
ab6085c7 241static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
a654b9d8
N
242{
243 mdk_rdev_t *rdev;
244 struct list_head *tmp;
ab6085c7 245 mddev_t *mddev = bitmap->mddev;
a654b9d8 246
d089c6af 247 rdev_for_each(rdev, tmp, mddev)
b2d444d7 248 if (test_bit(In_sync, &rdev->flags)
ab6085c7
N
249 && !test_bit(Faulty, &rdev->flags)) {
250 int size = PAGE_SIZE;
251 if (page->index == bitmap->file_pages-1)
252 size = roundup(bitmap->last_page_size,
253 bdev_hardsect_size(rdev->bdev));
f0d76d70
N
254 /* Just make sure we aren't corrupting data or
255 * metadata
256 */
257 if (bitmap->offset < 0) {
258 /* DATA BITMAP METADATA */
259 if (bitmap->offset
85bfb4da 260 + (long)(page->index * (PAGE_SIZE/512))
f0d76d70
N
261 + size/512 > 0)
262 /* bitmap runs in to metadata */
263 return -EINVAL;
264 if (rdev->data_offset + mddev->size*2
0f420358 265 > rdev->sb_start + bitmap->offset)
f0d76d70
N
266 /* data runs in to bitmap */
267 return -EINVAL;
0f420358 268 } else if (rdev->sb_start < rdev->data_offset) {
f0d76d70 269 /* METADATA BITMAP DATA */
0f420358 270 if (rdev->sb_start
f0d76d70
N
271 + bitmap->offset
272 + page->index*(PAGE_SIZE/512) + size/512
273 > rdev->data_offset)
274 /* bitmap runs in to data */
275 return -EINVAL;
276 } else {
277 /* DATA METADATA BITMAP - no problems */
278 }
a654b9d8 279 md_super_write(mddev, rdev,
0f420358 280 rdev->sb_start + bitmap->offset
a654b9d8 281 + page->index * (PAGE_SIZE/512),
ab6085c7 282 size,
a654b9d8 283 page);
ab6085c7 284 }
a654b9d8
N
285
286 if (wait)
a9701a30 287 md_super_wait(mddev);
a654b9d8
N
288 return 0;
289}
290
4ad13663 291static void bitmap_file_kick(struct bitmap *bitmap);
32a7627c 292/*
a654b9d8 293 * write out a page to a file
32a7627c 294 */
4ad13663 295static void write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 296{
d785a06a 297 struct buffer_head *bh;
32a7627c 298
f0d76d70
N
299 if (bitmap->file == NULL) {
300 switch (write_sb_page(bitmap, page, wait)) {
301 case -EINVAL:
302 bitmap->flags |= BITMAP_WRITE_ERROR;
f0d76d70 303 }
4ad13663 304 } else {
a654b9d8 305
4ad13663 306 bh = page_buffers(page);
c708443c 307
4ad13663
N
308 while (bh && bh->b_blocknr) {
309 atomic_inc(&bitmap->pending_writes);
310 set_buffer_locked(bh);
311 set_buffer_mapped(bh);
312 submit_bh(WRITE, bh);
313 bh = bh->b_this_page;
314 }
d785a06a 315
4ad13663
N
316 if (wait) {
317 wait_event(bitmap->write_wait,
318 atomic_read(&bitmap->pending_writes)==0);
319 }
8a5e9cf1 320 }
4ad13663
N
321 if (bitmap->flags & BITMAP_WRITE_ERROR)
322 bitmap_file_kick(bitmap);
d785a06a
N
323}
324
325static void end_bitmap_write(struct buffer_head *bh, int uptodate)
326{
327 struct bitmap *bitmap = bh->b_private;
328 unsigned long flags;
32a7627c 329
d785a06a
N
330 if (!uptodate) {
331 spin_lock_irqsave(&bitmap->lock, flags);
332 bitmap->flags |= BITMAP_WRITE_ERROR;
333 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 334 }
d785a06a
N
335 if (atomic_dec_and_test(&bitmap->pending_writes))
336 wake_up(&bitmap->write_wait);
337}
32a7627c 338
d785a06a
N
339/* copied from buffer.c */
340static void
341__clear_page_buffers(struct page *page)
342{
343 ClearPagePrivate(page);
344 set_page_private(page, 0);
345 page_cache_release(page);
346}
347static void free_buffers(struct page *page)
348{
349 struct buffer_head *bh = page_buffers(page);
77ad4bc7 350
d785a06a
N
351 while (bh) {
352 struct buffer_head *next = bh->b_this_page;
353 free_buffer_head(bh);
354 bh = next;
77ad4bc7 355 }
d785a06a
N
356 __clear_page_buffers(page);
357 put_page(page);
32a7627c
N
358}
359
d785a06a
N
360/* read a page from a file.
361 * We both read the page, and attach buffers to the page to record the
362 * address of each block (using bmap). These addresses will be used
363 * to write the block later, completely bypassing the filesystem.
364 * This usage is similar to how swap files are handled, and allows us
365 * to write to a file with no concerns of memory allocation failing.
366 */
32a7627c 367static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
368 struct bitmap *bitmap,
369 unsigned long count)
32a7627c 370{
32a7627c 371 struct page *page = NULL;
c649bb9c 372 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
373 struct buffer_head *bh;
374 sector_t block;
32a7627c 375
2d1f3b5d
N
376 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
377 (unsigned long long)index << PAGE_SHIFT);
32a7627c 378
d785a06a
N
379 page = alloc_page(GFP_KERNEL);
380 if (!page)
381 page = ERR_PTR(-ENOMEM);
32a7627c
N
382 if (IS_ERR(page))
383 goto out;
d785a06a 384
d785a06a
N
385 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
386 if (!bh) {
2d1f3b5d 387 put_page(page);
d785a06a 388 page = ERR_PTR(-ENOMEM);
32a7627c
N
389 goto out;
390 }
d785a06a
N
391 attach_page_buffers(page, bh);
392 block = index << (PAGE_SHIFT - inode->i_blkbits);
393 while (bh) {
394 if (count == 0)
395 bh->b_blocknr = 0;
396 else {
397 bh->b_blocknr = bmap(inode, block);
398 if (bh->b_blocknr == 0) {
399 /* Cannot use this file! */
400 free_buffers(page);
401 page = ERR_PTR(-EINVAL);
402 goto out;
403 }
404 bh->b_bdev = inode->i_sb->s_bdev;
405 if (count < (1<<inode->i_blkbits))
406 count = 0;
407 else
408 count -= (1<<inode->i_blkbits);
409
410 bh->b_end_io = end_bitmap_write;
411 bh->b_private = bitmap;
ce25c31b
N
412 atomic_inc(&bitmap->pending_writes);
413 set_buffer_locked(bh);
414 set_buffer_mapped(bh);
415 submit_bh(READ, bh);
d785a06a
N
416 }
417 block++;
418 bh = bh->b_this_page;
419 }
d785a06a 420 page->index = index;
ce25c31b
N
421
422 wait_event(bitmap->write_wait,
423 atomic_read(&bitmap->pending_writes)==0);
424 if (bitmap->flags & BITMAP_WRITE_ERROR) {
425 free_buffers(page);
426 page = ERR_PTR(-EIO);
427 }
32a7627c
N
428out:
429 if (IS_ERR(page))
430 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
431 (int)PAGE_SIZE,
432 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
433 PTR_ERR(page));
434 return page;
435}
436
437/*
438 * bitmap file superblock operations
439 */
440
441/* update the event counter and sync the superblock to disk */
4ad13663 442void bitmap_update_sb(struct bitmap *bitmap)
32a7627c
N
443{
444 bitmap_super_t *sb;
445 unsigned long flags;
446
447 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
4ad13663 448 return;
32a7627c
N
449 spin_lock_irqsave(&bitmap->lock, flags);
450 if (!bitmap->sb_page) { /* no superblock */
451 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 452 return;
32a7627c 453 }
32a7627c 454 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 455 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 456 sb->events = cpu_to_le64(bitmap->mddev->events);
a0da84f3
NB
457 if (bitmap->mddev->events < bitmap->events_cleared) {
458 /* rocking back to read-only */
459 bitmap->events_cleared = bitmap->mddev->events;
460 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
461 }
ea03aff9 462 kunmap_atomic(sb, KM_USER0);
4ad13663 463 write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
464}
465
466/* print out the bitmap file superblock */
467void bitmap_print_sb(struct bitmap *bitmap)
468{
469 bitmap_super_t *sb;
470
471 if (!bitmap || !bitmap->sb_page)
472 return;
ea03aff9 473 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 474 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
475 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
476 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
477 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
478 *(__u32 *)(sb->uuid+0),
479 *(__u32 *)(sb->uuid+4),
480 *(__u32 *)(sb->uuid+8),
481 *(__u32 *)(sb->uuid+12));
a2cff26a 482 printk(KERN_DEBUG " events: %llu\n",
32a7627c 483 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 484 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 485 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
486 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
487 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
488 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
489 printk(KERN_DEBUG " sync size: %llu KB\n",
490 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 491 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 492 kunmap_atomic(sb, KM_USER0);
32a7627c
N
493}
494
495/* read the superblock from the bitmap file and initialize some bitmap fields */
496static int bitmap_read_sb(struct bitmap *bitmap)
497{
498 char *reason = NULL;
499 bitmap_super_t *sb;
4b6d287f 500 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
501 unsigned long long events;
502 int err = -EINVAL;
503
504 /* page 0 is the superblock, read it... */
f49d5e62
N
505 if (bitmap->file) {
506 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
507 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
508
509 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
510 } else {
a654b9d8 511 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 512 }
32a7627c
N
513 if (IS_ERR(bitmap->sb_page)) {
514 err = PTR_ERR(bitmap->sb_page);
515 bitmap->sb_page = NULL;
516 return err;
517 }
518
ea03aff9 519 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 520
32a7627c
N
521 chunksize = le32_to_cpu(sb->chunksize);
522 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 523 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
524
525 /* verify that the bitmap-specific fields are valid */
526 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
527 reason = "bad magic";
bd926c63
N
528 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
529 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 530 reason = "unrecognized superblock version";
7dd5d34c
N
531 else if (chunksize < PAGE_SIZE)
532 reason = "bitmap chunksize too small";
32a7627c
N
533 else if ((1 << ffz(~chunksize)) != chunksize)
534 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
535 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
536 reason = "daemon sleep period out of range";
4b6d287f
N
537 else if (write_behind > COUNTER_MAX)
538 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
539 if (reason) {
540 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
541 bmname(bitmap), reason);
542 goto out;
543 }
544
545 /* keep the array size field of the bitmap superblock up to date */
546 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
547
548 if (!bitmap->mddev->persistent)
549 goto success;
550
551 /*
552 * if we have a persistent array superblock, compare the
553 * bitmap's UUID and event counter to the mddev's
554 */
555 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
556 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
557 bmname(bitmap));
558 goto out;
559 }
560 events = le64_to_cpu(sb->events);
561 if (events < bitmap->mddev->events) {
562 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
563 "-- forcing full recovery\n", bmname(bitmap), events,
564 (unsigned long long) bitmap->mddev->events);
4f2e639a 565 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
566 }
567success:
568 /* assign fields using values from superblock */
569 bitmap->chunksize = chunksize;
570 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 571 bitmap->daemon_lastrun = jiffies;
4b6d287f 572 bitmap->max_write_behind = write_behind;
4f2e639a 573 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
574 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
575 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 576 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 577 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 578 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
579 err = 0;
580out:
ea03aff9 581 kunmap_atomic(sb, KM_USER0);
32a7627c
N
582 if (err)
583 bitmap_print_sb(bitmap);
584 return err;
585}
586
587enum bitmap_mask_op {
588 MASK_SET,
589 MASK_UNSET
590};
591
4ad13663
N
592/* record the state of the bitmap in the superblock. Return the old value */
593static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
594 enum bitmap_mask_op op)
32a7627c
N
595{
596 bitmap_super_t *sb;
597 unsigned long flags;
4ad13663 598 int old;
32a7627c
N
599
600 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 601 if (!bitmap->sb_page) { /* can't set the state */
32a7627c 602 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 603 return 0;
32a7627c 604 }
32a7627c 605 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 606 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
4ad13663 607 old = le32_to_cpu(sb->state) & bits;
32a7627c 608 switch (op) {
4f2e639a 609 case MASK_SET: sb->state |= cpu_to_le32(bits);
32a7627c 610 break;
4f2e639a 611 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
32a7627c
N
612 break;
613 default: BUG();
614 }
ea03aff9 615 kunmap_atomic(sb, KM_USER0);
4ad13663 616 return old;
32a7627c
N
617}
618
619/*
620 * general bitmap file operations
621 */
622
623/* calculate the index of the page that contains this bit */
624static inline unsigned long file_page_index(unsigned long chunk)
625{
626 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
627}
628
629/* calculate the (bit) offset of this bit within a page */
630static inline unsigned long file_page_offset(unsigned long chunk)
631{
632 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
633}
634
635/*
636 * return a pointer to the page in the filemap that contains the given bit
637 *
638 * this lookup is complicated by the fact that the bitmap sb might be exactly
639 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
640 * 0 or page 1
641 */
642static inline struct page *filemap_get_page(struct bitmap *bitmap,
643 unsigned long chunk)
644{
9b1d1dac 645 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
32a7627c
N
646 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
647}
648
649
650static void bitmap_file_unmap(struct bitmap *bitmap)
651{
652 struct page **map, *sb_page;
653 unsigned long *attr;
654 int pages;
655 unsigned long flags;
656
657 spin_lock_irqsave(&bitmap->lock, flags);
658 map = bitmap->filemap;
659 bitmap->filemap = NULL;
660 attr = bitmap->filemap_attr;
661 bitmap->filemap_attr = NULL;
662 pages = bitmap->file_pages;
663 bitmap->file_pages = 0;
664 sb_page = bitmap->sb_page;
665 bitmap->sb_page = NULL;
666 spin_unlock_irqrestore(&bitmap->lock, flags);
667
668 while (pages--)
669 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
d785a06a 670 free_buffers(map[pages]);
32a7627c
N
671 kfree(map);
672 kfree(attr);
673
d785a06a
N
674 if (sb_page)
675 free_buffers(sb_page);
32a7627c
N
676}
677
678static void bitmap_file_put(struct bitmap *bitmap)
679{
680 struct file *file;
32a7627c
N
681 unsigned long flags;
682
683 spin_lock_irqsave(&bitmap->lock, flags);
684 file = bitmap->file;
685 bitmap->file = NULL;
686 spin_unlock_irqrestore(&bitmap->lock, flags);
687
d785a06a
N
688 if (file)
689 wait_event(bitmap->write_wait,
690 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
691 bitmap_file_unmap(bitmap);
692
d785a06a 693 if (file) {
c649bb9c 694 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 695 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 696 fput(file);
d785a06a 697 }
32a7627c
N
698}
699
700
701/*
702 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
703 * then it is no longer reliable, so we stop using it and we mark the file
704 * as failed in the superblock
705 */
706static void bitmap_file_kick(struct bitmap *bitmap)
707{
708 char *path, *ptr = NULL;
709
4ad13663
N
710 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
711 bitmap_update_sb(bitmap);
32a7627c 712
4ad13663
N
713 if (bitmap->file) {
714 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
715 if (path)
6bcfd601
CH
716 ptr = d_path(&bitmap->file->f_path, path,
717 PAGE_SIZE);
718
32a7627c 719
4ad13663
N
720 printk(KERN_ALERT
721 "%s: kicking failed bitmap file %s from array!\n",
6bcfd601 722 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
32a7627c 723
4ad13663
N
724 kfree(path);
725 } else
726 printk(KERN_ALERT
727 "%s: disabling internal bitmap due to errors\n",
728 bmname(bitmap));
a654b9d8 729 }
32a7627c
N
730
731 bitmap_file_put(bitmap);
732
733 return;
734}
735
736enum bitmap_page_attr {
e16b68b6
N
737 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
738 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
739 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
740};
741
742static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
743 enum bitmap_page_attr attr)
744{
e16b68b6 745 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
746}
747
748static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
749 enum bitmap_page_attr attr)
750{
e16b68b6 751 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
752}
753
ec7a3197
N
754static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
755 enum bitmap_page_attr attr)
32a7627c 756{
e16b68b6 757 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
758}
759
760/*
761 * bitmap_file_set_bit -- called before performing a write to the md device
762 * to set (and eventually sync) a particular bit in the bitmap file
763 *
764 * we set the bit immediately, then we record the page number so that
765 * when an unplug occurs, we can flush the dirty pages out to disk
766 */
767static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
768{
769 unsigned long bit;
770 struct page *page;
771 void *kaddr;
772 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
773
a654b9d8 774 if (!bitmap->filemap) {
32a7627c
N
775 return;
776 }
777
778 page = filemap_get_page(bitmap, chunk);
9b1d1dac 779 if (!page) return;
32a7627c
N
780 bit = file_page_offset(chunk);
781
32a7627c
N
782 /* set the bit */
783 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
784 if (bitmap->flags & BITMAP_HOSTENDIAN)
785 set_bit(bit, kaddr);
786 else
787 ext2_set_bit(bit, kaddr);
32a7627c
N
788 kunmap_atomic(kaddr, KM_USER0);
789 PRINTK("set file bit %lu page %lu\n", bit, page->index);
790
791 /* record page number so it gets flushed to disk when unplug occurs */
792 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
793
794}
795
796/* this gets called when the md device is ready to unplug its underlying
797 * (slave) device queues -- before we let any writes go down, we need to
798 * sync the dirty pages of the bitmap file to disk */
4ad13663 799void bitmap_unplug(struct bitmap *bitmap)
32a7627c 800{
ec7a3197
N
801 unsigned long i, flags;
802 int dirty, need_write;
32a7627c
N
803 struct page *page;
804 int wait = 0;
805
806 if (!bitmap)
4ad13663 807 return;
32a7627c
N
808
809 /* look at each page to see if there are any set bits that need to be
810 * flushed out to disk */
811 for (i = 0; i < bitmap->file_pages; i++) {
812 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 813 if (!bitmap->filemap) {
32a7627c 814 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 815 return;
32a7627c
N
816 }
817 page = bitmap->filemap[i];
ec7a3197
N
818 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
819 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
820 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
821 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 822 if (dirty)
32a7627c
N
823 wait = 1;
824 spin_unlock_irqrestore(&bitmap->lock, flags);
825
d785a06a 826 if (dirty | need_write)
4ad13663 827 write_page(bitmap, page, 0);
32a7627c
N
828 }
829 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 830 if (bitmap->file)
d785a06a
N
831 wait_event(bitmap->write_wait,
832 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 833 else
a9701a30 834 md_super_wait(bitmap->mddev);
32a7627c 835 }
d785a06a
N
836 if (bitmap->flags & BITMAP_WRITE_ERROR)
837 bitmap_file_kick(bitmap);
32a7627c
N
838}
839
6a07997f 840static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
841/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
842 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
843 * memory mapping of the bitmap file
844 * Special cases:
845 * if there's no bitmap file, or if the bitmap file had been
846 * previously kicked from the array, we mark all the bits as
847 * 1's in order to cause a full resync.
6a07997f
N
848 *
849 * We ignore all bits for sectors that end earlier than 'start'.
850 * This is used when reading an out-of-date bitmap...
32a7627c 851 */
6a07997f 852static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
853{
854 unsigned long i, chunks, index, oldindex, bit;
855 struct page *page = NULL, *oldpage = NULL;
856 unsigned long num_pages, bit_cnt = 0;
857 struct file *file;
d785a06a 858 unsigned long bytes, offset;
32a7627c
N
859 int outofdate;
860 int ret = -ENOSPC;
ea03aff9 861 void *paddr;
32a7627c
N
862
863 chunks = bitmap->chunks;
864 file = bitmap->file;
865
a654b9d8 866 BUG_ON(!file && !bitmap->offset);
32a7627c 867
44456d37 868#ifdef INJECT_FAULTS_3
32a7627c
N
869 outofdate = 1;
870#else
871 outofdate = bitmap->flags & BITMAP_STALE;
872#endif
873 if (outofdate)
874 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
875 "recovery\n", bmname(bitmap));
876
877 bytes = (chunks + 7) / 8;
bc7f77de 878
cdbb4cc2 879 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 880
a654b9d8 881 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
882 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
883 bmname(bitmap),
884 (unsigned long) i_size_read(file->f_mapping->host),
885 bytes + sizeof(bitmap_super_t));
4ad13663 886 goto err;
32a7627c 887 }
bc7f77de
N
888
889 ret = -ENOMEM;
890
32a7627c 891 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 892 if (!bitmap->filemap)
4ad13663 893 goto err;
32a7627c 894
e16b68b6
N
895 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
896 bitmap->filemap_attr = kzalloc(
505fa2c4 897 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 898 GFP_KERNEL);
bc7f77de 899 if (!bitmap->filemap_attr)
4ad13663 900 goto err;
32a7627c 901
32a7627c
N
902 oldindex = ~0L;
903
904 for (i = 0; i < chunks; i++) {
bd926c63 905 int b;
32a7627c
N
906 index = file_page_index(i);
907 bit = file_page_offset(i);
908 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 909 int count;
32a7627c 910 /* unmap the old page, we're done with it */
d785a06a 911 if (index == num_pages-1)
f49d5e62
N
912 count = bytes + sizeof(bitmap_super_t)
913 - index * PAGE_SIZE;
d785a06a
N
914 else
915 count = PAGE_SIZE;
32a7627c
N
916 if (index == 0) {
917 /*
918 * if we're here then the superblock page
919 * contains some bits (PAGE_SIZE != sizeof sb)
920 * we've already read it in, so just use it
921 */
922 page = bitmap->sb_page;
923 offset = sizeof(bitmap_super_t);
a654b9d8 924 } else if (file) {
d785a06a 925 page = read_page(file, index, bitmap, count);
a654b9d8
N
926 offset = 0;
927 } else {
928 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
929 offset = 0;
930 }
a654b9d8
N
931 if (IS_ERR(page)) { /* read error */
932 ret = PTR_ERR(page);
4ad13663 933 goto err;
a654b9d8
N
934 }
935
32a7627c
N
936 oldindex = index;
937 oldpage = page;
32a7627c
N
938
939 if (outofdate) {
940 /*
941 * if bitmap is out of date, dirty the
942 * whole page and write it out
943 */
ea03aff9
N
944 paddr = kmap_atomic(page, KM_USER0);
945 memset(paddr + offset, 0xff,
6a07997f 946 PAGE_SIZE - offset);
ea03aff9 947 kunmap_atomic(paddr, KM_USER0);
4ad13663
N
948 write_page(bitmap, page, 1);
949
950 ret = -EIO;
951 if (bitmap->flags & BITMAP_WRITE_ERROR) {
32a7627c 952 /* release, page not in filemap yet */
2d1f3b5d 953 put_page(page);
4ad13663 954 goto err;
32a7627c
N
955 }
956 }
957
958 bitmap->filemap[bitmap->file_pages++] = page;
ab6085c7 959 bitmap->last_page_size = count;
32a7627c 960 }
ea03aff9 961 paddr = kmap_atomic(page, KM_USER0);
bd926c63 962 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 963 b = test_bit(bit, paddr);
bd926c63 964 else
ea03aff9
N
965 b = ext2_test_bit(bit, paddr);
966 kunmap_atomic(paddr, KM_USER0);
bd926c63 967 if (b) {
32a7627c 968 /* if the disk bit is set, set the memory bit */
6a07997f
N
969 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
970 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
971 );
32a7627c 972 bit_cnt++;
6a07997f 973 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 974 }
32a7627c
N
975 }
976
977 /* everything went OK */
978 ret = 0;
979 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
980
32a7627c
N
981 if (bit_cnt) { /* Kick recovery if any bits were set */
982 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
983 md_wakeup_thread(bitmap->mddev->thread);
984 }
985
32a7627c 986 printk(KERN_INFO "%s: bitmap initialized from disk: "
4ad13663
N
987 "read %lu/%lu pages, set %lu bits\n",
988 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
989
990 return 0;
32a7627c 991
4ad13663
N
992 err:
993 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
994 bmname(bitmap), ret);
32a7627c
N
995 return ret;
996}
997
a654b9d8
N
998void bitmap_write_all(struct bitmap *bitmap)
999{
1000 /* We don't actually write all bitmap blocks here,
1001 * just flag them as needing to be written
1002 */
ec7a3197 1003 int i;
a654b9d8 1004
ec7a3197
N
1005 for (i=0; i < bitmap->file_pages; i++)
1006 set_page_attr(bitmap, bitmap->filemap[i],
1007 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
1008}
1009
32a7627c
N
1010
1011static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1012{
1013 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1014 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1015 bitmap->bp[page].count += inc;
1016/*
1017 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1018 (unsigned long long)offset, inc, bitmap->bp[page].count);
1019*/
1020 bitmap_checkfree(bitmap, page);
1021}
1022static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1023 sector_t offset, int *blocks,
1024 int create);
1025
1026/*
1027 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1028 * out to disk
1029 */
1030
4ad13663 1031void bitmap_daemon_work(struct bitmap *bitmap)
32a7627c 1032{
aa3163f8 1033 unsigned long j;
32a7627c
N
1034 unsigned long flags;
1035 struct page *page = NULL, *lastpage = NULL;
32a7627c 1036 int blocks;
ea03aff9 1037 void *paddr;
32a7627c
N
1038
1039 if (bitmap == NULL)
4ad13663 1040 return;
32a7627c 1041 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
7be3dfec
N
1042 goto done;
1043
32a7627c 1044 bitmap->daemon_lastrun = jiffies;
8311c29d
N
1045 if (bitmap->allclean) {
1046 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1047 return;
1048 }
1049 bitmap->allclean = 1;
32a7627c
N
1050
1051 for (j = 0; j < bitmap->chunks; j++) {
1052 bitmap_counter_t *bmc;
1053 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1054 if (!bitmap->filemap) {
32a7627c
N
1055 /* error or shutdown */
1056 spin_unlock_irqrestore(&bitmap->lock, flags);
1057 break;
1058 }
1059
1060 page = filemap_get_page(bitmap, j);
32a7627c
N
1061
1062 if (page != lastpage) {
aa3163f8 1063 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1064 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1065 int need_write = test_page_attr(bitmap, page,
1066 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1067 if (need_write)
aa3163f8 1068 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1069
aa3163f8 1070 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1071 if (need_write) {
4ad13663 1072 write_page(bitmap, page, 0);
8311c29d
N
1073 bitmap->allclean = 0;
1074 }
aa3163f8
N
1075 continue;
1076 }
1077
32a7627c 1078 /* grab the new page, sync and release the old */
32a7627c 1079 if (lastpage != NULL) {
ec7a3197 1080 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1081 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1082 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1083 write_page(bitmap, lastpage, 0);
32a7627c
N
1084 } else {
1085 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1086 spin_unlock_irqrestore(&bitmap->lock, flags);
1087 }
32a7627c
N
1088 } else
1089 spin_unlock_irqrestore(&bitmap->lock, flags);
1090 lastpage = page;
a0da84f3
NB
1091
1092 /* We are possibly going to clear some bits, so make
1093 * sure that events_cleared is up-to-date.
1094 */
1095 if (bitmap->need_sync) {
1096 bitmap_super_t *sb;
1097 bitmap->need_sync = 0;
1098 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1099 sb->events_cleared =
1100 cpu_to_le64(bitmap->events_cleared);
1101 kunmap_atomic(sb, KM_USER0);
1102 write_page(bitmap, bitmap->sb_page, 1);
1103 }
32a7627c
N
1104 spin_lock_irqsave(&bitmap->lock, flags);
1105 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1106 }
1107 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1108 &blocks, 0);
1109 if (bmc) {
1110/*
1111 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1112*/
8311c29d
N
1113 if (*bmc)
1114 bitmap->allclean = 0;
1115
32a7627c
N
1116 if (*bmc == 2) {
1117 *bmc=1; /* maybe clear the bit next time */
1118 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1119 } else if (*bmc == 1) {
1120 /* we can clear the bit */
1121 *bmc = 0;
1122 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1123 -1);
1124
1125 /* clear the bit */
ea03aff9 1126 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1127 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1128 clear_bit(file_page_offset(j), paddr);
bd926c63 1129 else
ea03aff9
N
1130 ext2_clear_bit(file_page_offset(j), paddr);
1131 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1132 }
1133 }
1134 spin_unlock_irqrestore(&bitmap->lock, flags);
1135 }
1136
1137 /* now sync the final page */
1138 if (lastpage != NULL) {
32a7627c 1139 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1140 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1141 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1142 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1143 write_page(bitmap, lastpage, 0);
32a7627c
N
1144 } else {
1145 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1146 spin_unlock_irqrestore(&bitmap->lock, flags);
1147 }
32a7627c
N
1148 }
1149
7be3dfec 1150 done:
8311c29d
N
1151 if (bitmap->allclean == 0)
1152 bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ;
32a7627c
N
1153}
1154
32a7627c
N
1155static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1156 sector_t offset, int *blocks,
1157 int create)
1158{
1159 /* If 'create', we might release the lock and reclaim it.
1160 * The lock must have been taken with interrupts enabled.
1161 * If !create, we don't release the lock.
1162 */
1163 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1164 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1165 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1166 sector_t csize;
1167
1168 if (bitmap_checkpage(bitmap, page, create) < 0) {
1169 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1170 *blocks = csize - (offset & (csize- 1));
1171 return NULL;
1172 }
1173 /* now locked ... */
1174
1175 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1176 /* should we use the first or second counter field
1177 * of the hijacked pointer? */
1178 int hi = (pageoff > PAGE_COUNTER_MASK);
1179 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1180 PAGE_COUNTER_SHIFT - 1);
1181 *blocks = csize - (offset & (csize- 1));
1182 return &((bitmap_counter_t *)
1183 &bitmap->bp[page].map)[hi];
1184 } else { /* page is allocated */
1185 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1186 *blocks = csize - (offset & (csize- 1));
1187 return (bitmap_counter_t *)
1188 &(bitmap->bp[page].map[pageoff]);
1189 }
1190}
1191
4b6d287f 1192int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1193{
1194 if (!bitmap) return 0;
4b6d287f
N
1195
1196 if (behind) {
1197 atomic_inc(&bitmap->behind_writes);
1198 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1199 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1200 }
1201
32a7627c
N
1202 while (sectors) {
1203 int blocks;
1204 bitmap_counter_t *bmc;
1205
1206 spin_lock_irq(&bitmap->lock);
1207 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1208 if (!bmc) {
1209 spin_unlock_irq(&bitmap->lock);
1210 return 0;
1211 }
1212
da6e1a32
NB
1213 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1214 DEFINE_WAIT(__wait);
1215 /* note that it is safe to do the prepare_to_wait
1216 * after the test as long as we do it before dropping
1217 * the spinlock.
1218 */
1219 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1220 TASK_UNINTERRUPTIBLE);
1221 spin_unlock_irq(&bitmap->lock);
2ad8b1ef 1222 blk_unplug(bitmap->mddev->queue);
da6e1a32
NB
1223 schedule();
1224 finish_wait(&bitmap->overflow_wait, &__wait);
1225 continue;
1226 }
1227
32a7627c
N
1228 switch(*bmc) {
1229 case 0:
1230 bitmap_file_set_bit(bitmap, offset);
1231 bitmap_count_page(bitmap,offset, 1);
1232 blk_plug_device(bitmap->mddev->queue);
1233 /* fall through */
1234 case 1:
1235 *bmc = 2;
1236 }
da6e1a32 1237
32a7627c
N
1238 (*bmc)++;
1239
1240 spin_unlock_irq(&bitmap->lock);
1241
1242 offset += blocks;
1243 if (sectors > blocks)
1244 sectors -= blocks;
1245 else sectors = 0;
1246 }
8311c29d 1247 bitmap->allclean = 0;
32a7627c
N
1248 return 0;
1249}
1250
1251void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1252 int success, int behind)
32a7627c
N
1253{
1254 if (!bitmap) return;
4b6d287f
N
1255 if (behind) {
1256 atomic_dec(&bitmap->behind_writes);
1257 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1258 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1259 }
1260
32a7627c
N
1261 while (sectors) {
1262 int blocks;
1263 unsigned long flags;
1264 bitmap_counter_t *bmc;
1265
1266 spin_lock_irqsave(&bitmap->lock, flags);
1267 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1268 if (!bmc) {
1269 spin_unlock_irqrestore(&bitmap->lock, flags);
1270 return;
1271 }
1272
a0da84f3
NB
1273 if (success &&
1274 bitmap->events_cleared < bitmap->mddev->events) {
1275 bitmap->events_cleared = bitmap->mddev->events;
1276 bitmap->need_sync = 1;
1277 }
1278
32a7627c
N
1279 if (!success && ! (*bmc & NEEDED_MASK))
1280 *bmc |= NEEDED_MASK;
1281
da6e1a32
NB
1282 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1283 wake_up(&bitmap->overflow_wait);
1284
32a7627c
N
1285 (*bmc)--;
1286 if (*bmc <= 2) {
1287 set_page_attr(bitmap,
1288 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1289 BITMAP_PAGE_CLEAN);
1290 }
1291 spin_unlock_irqrestore(&bitmap->lock, flags);
1292 offset += blocks;
1293 if (sectors > blocks)
1294 sectors -= blocks;
1295 else sectors = 0;
1296 }
1297}
1298
6a806c51
N
1299int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1300 int degraded)
32a7627c
N
1301{
1302 bitmap_counter_t *bmc;
1303 int rv;
1304 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1305 *blocks = 1024;
1306 return 1; /* always resync if no bitmap */
1307 }
1308 spin_lock_irq(&bitmap->lock);
1309 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1310 rv = 0;
1311 if (bmc) {
1312 /* locked */
1313 if (RESYNC(*bmc))
1314 rv = 1;
1315 else if (NEEDED(*bmc)) {
1316 rv = 1;
6a806c51
N
1317 if (!degraded) { /* don't set/clear bits if degraded */
1318 *bmc |= RESYNC_MASK;
1319 *bmc &= ~NEEDED_MASK;
1320 }
32a7627c
N
1321 }
1322 }
1323 spin_unlock_irq(&bitmap->lock);
8311c29d 1324 bitmap->allclean = 0;
32a7627c
N
1325 return rv;
1326}
1327
1328void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1329{
1330 bitmap_counter_t *bmc;
1331 unsigned long flags;
1332/*
1333 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1334*/ if (bitmap == NULL) {
1335 *blocks = 1024;
1336 return;
1337 }
1338 spin_lock_irqsave(&bitmap->lock, flags);
1339 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1340 if (bmc == NULL)
1341 goto unlock;
1342 /* locked */
1343/*
1344 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1345*/
1346 if (RESYNC(*bmc)) {
1347 *bmc &= ~RESYNC_MASK;
1348
1349 if (!NEEDED(*bmc) && aborted)
1350 *bmc |= NEEDED_MASK;
1351 else {
1352 if (*bmc <= 2) {
1353 set_page_attr(bitmap,
1354 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1355 BITMAP_PAGE_CLEAN);
1356 }
1357 }
1358 }
1359 unlock:
1360 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1361 bitmap->allclean = 0;
32a7627c
N
1362}
1363
1364void bitmap_close_sync(struct bitmap *bitmap)
1365{
1366 /* Sync has finished, and any bitmap chunks that weren't synced
1367 * properly have been aborted. It remains to us to clear the
1368 * RESYNC bit wherever it is still on
1369 */
1370 sector_t sector = 0;
1371 int blocks;
b47490c9
N
1372 if (!bitmap)
1373 return;
32a7627c
N
1374 while (sector < bitmap->mddev->resync_max_sectors) {
1375 bitmap_end_sync(bitmap, sector, &blocks, 0);
b47490c9
N
1376 sector += blocks;
1377 }
1378}
1379
1380void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1381{
1382 sector_t s = 0;
1383 int blocks;
1384
1385 if (!bitmap)
1386 return;
1387 if (sector == 0) {
1388 bitmap->last_end_sync = jiffies;
1389 return;
1390 }
1391 if (time_before(jiffies, (bitmap->last_end_sync
1392 + bitmap->daemon_sleep * HZ)))
1393 return;
1394 wait_event(bitmap->mddev->recovery_wait,
1395 atomic_read(&bitmap->mddev->recovery_active) == 0);
1396
1397 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1398 s = 0;
1399 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1400 bitmap_end_sync(bitmap, s, &blocks, 0);
1401 s += blocks;
32a7627c 1402 }
b47490c9 1403 bitmap->last_end_sync = jiffies;
32a7627c
N
1404}
1405
6a07997f 1406static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1407{
1408 /* For each chunk covered by any of these sectors, set the
193f1c93 1409 * counter to 1 and set resync_needed. They should all
32a7627c
N
1410 * be 0 at this point
1411 */
193f1c93
N
1412
1413 int secs;
1414 bitmap_counter_t *bmc;
1415 spin_lock_irq(&bitmap->lock);
1416 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1417 if (!bmc) {
32a7627c 1418 spin_unlock_irq(&bitmap->lock);
193f1c93 1419 return;
32a7627c 1420 }
193f1c93
N
1421 if (! *bmc) {
1422 struct page *page;
6a07997f 1423 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1424 bitmap_count_page(bitmap, offset, 1);
1425 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1426 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1427 }
1428 spin_unlock_irq(&bitmap->lock);
8311c29d 1429 bitmap->allclean = 0;
32a7627c
N
1430}
1431
9b1d1dac
PC
1432/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1433void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1434{
1435 unsigned long chunk;
1436
1437 for (chunk = s; chunk <= e; chunk++) {
1438 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1439 bitmap_set_memory_bits(bitmap, sec, 1);
1440 bitmap_file_set_bit(bitmap, sec);
1441 }
1442}
1443
6b8b3e8a
N
1444/*
1445 * flush out any pending updates
1446 */
1447void bitmap_flush(mddev_t *mddev)
1448{
1449 struct bitmap *bitmap = mddev->bitmap;
1450 int sleep;
1451
1452 if (!bitmap) /* there was no bitmap */
1453 return;
1454
1455 /* run the daemon_work three time to ensure everything is flushed
1456 * that can be
1457 */
1458 sleep = bitmap->daemon_sleep;
1459 bitmap->daemon_sleep = 0;
1460 bitmap_daemon_work(bitmap);
1461 bitmap_daemon_work(bitmap);
1462 bitmap_daemon_work(bitmap);
1463 bitmap->daemon_sleep = sleep;
1464 bitmap_update_sb(bitmap);
1465}
1466
32a7627c
N
1467/*
1468 * free memory that was allocated
1469 */
3178b0db 1470static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1471{
1472 unsigned long k, pages;
1473 struct bitmap_page *bp;
32a7627c
N
1474
1475 if (!bitmap) /* there was no bitmap */
1476 return;
1477
32a7627c
N
1478 /* release the bitmap file and kill the daemon */
1479 bitmap_file_put(bitmap);
1480
1481 bp = bitmap->bp;
1482 pages = bitmap->pages;
1483
1484 /* free all allocated memory */
1485
32a7627c
N
1486 if (bp) /* deallocate the page memory */
1487 for (k = 0; k < pages; k++)
1488 if (bp[k].map && !bp[k].hijacked)
1489 kfree(bp[k].map);
1490 kfree(bp);
1491 kfree(bitmap);
1492}
3178b0db
N
1493void bitmap_destroy(mddev_t *mddev)
1494{
1495 struct bitmap *bitmap = mddev->bitmap;
1496
1497 if (!bitmap) /* there was no bitmap */
1498 return;
1499
1500 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1501 if (mddev->thread)
1502 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1503
1504 bitmap_free(bitmap);
1505}
32a7627c
N
1506
1507/*
1508 * initialize the bitmap structure
1509 * if this returns an error, bitmap_destroy must be called to do clean up
1510 */
1511int bitmap_create(mddev_t *mddev)
1512{
1513 struct bitmap *bitmap;
1514 unsigned long blocks = mddev->resync_max_sectors;
1515 unsigned long chunks;
1516 unsigned long pages;
1517 struct file *file = mddev->bitmap_file;
1518 int err;
6a07997f 1519 sector_t start;
32a7627c 1520
5f6e3c83 1521 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1522
a654b9d8 1523 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1524 return 0;
1525
a654b9d8
N
1526 BUG_ON(file && mddev->bitmap_offset);
1527
9ffae0cf 1528 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1529 if (!bitmap)
1530 return -ENOMEM;
1531
32a7627c 1532 spin_lock_init(&bitmap->lock);
ce25c31b
N
1533 atomic_set(&bitmap->pending_writes, 0);
1534 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1535 init_waitqueue_head(&bitmap->overflow_wait);
ce25c31b 1536
32a7627c 1537 bitmap->mddev = mddev;
32a7627c 1538
32a7627c 1539 bitmap->file = file;
a654b9d8 1540 bitmap->offset = mddev->bitmap_offset;
ce25c31b
N
1541 if (file) {
1542 get_file(file);
ef51c976
MF
1543 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1544 SYNC_FILE_RANGE_WAIT_BEFORE |
1545 SYNC_FILE_RANGE_WRITE |
1546 SYNC_FILE_RANGE_WAIT_AFTER);
ce25c31b 1547 }
32a7627c
N
1548 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1549 err = bitmap_read_sb(bitmap);
1550 if (err)
3178b0db 1551 goto error;
32a7627c 1552
a638b2dc 1553 bitmap->chunkshift = ffz(~bitmap->chunksize);
32a7627c
N
1554
1555 /* now that chunksize and chunkshift are set, we can use these macros */
1556 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1557 CHUNK_BLOCK_RATIO(bitmap);
1558 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1559
1560 BUG_ON(!pages);
1561
1562 bitmap->chunks = chunks;
1563 bitmap->pages = pages;
1564 bitmap->missing_pages = pages;
1565 bitmap->counter_bits = COUNTER_BITS;
1566
1567 bitmap->syncchunk = ~0UL;
1568
44456d37 1569#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1570 bitmap->bp = NULL;
1571#else
9ffae0cf 1572 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1573#endif
3178b0db 1574 err = -ENOMEM;
32a7627c 1575 if (!bitmap->bp)
3178b0db 1576 goto error;
32a7627c 1577
32a7627c
N
1578 /* now that we have some pages available, initialize the in-memory
1579 * bitmap from the on-disk bitmap */
6a07997f
N
1580 start = 0;
1581 if (mddev->degraded == 0
1582 || bitmap->events_cleared == mddev->events)
1583 /* no need to keep dirty bits to optimise a re-add of a missing device */
1584 start = mddev->recovery_cp;
1585 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1586
32a7627c 1587 if (err)
3178b0db 1588 goto error;
32a7627c
N
1589
1590 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1591 pages, bmname(bitmap));
1592
3178b0db
N
1593 mddev->bitmap = bitmap;
1594
b15c2e57
N
1595 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1596
4ad13663
N
1597 bitmap_update_sb(bitmap);
1598
1599 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
3178b0db
N
1600
1601 error:
1602 bitmap_free(bitmap);
1603 return err;
32a7627c
N
1604}
1605
1606/* the bitmap API -- for raid personalities */
1607EXPORT_SYMBOL(bitmap_startwrite);
1608EXPORT_SYMBOL(bitmap_endwrite);
1609EXPORT_SYMBOL(bitmap_start_sync);
1610EXPORT_SYMBOL(bitmap_end_sync);
1611EXPORT_SYMBOL(bitmap_unplug);
1612EXPORT_SYMBOL(bitmap_close_sync);
b47490c9 1613EXPORT_SYMBOL(bitmap_cond_end_sync);