md: improve the is_mddev_idle test fix
[deliverable/linux.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
ab6085c7 258static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
a654b9d8
N
259{
260 mdk_rdev_t *rdev;
261 struct list_head *tmp;
ab6085c7 262 mddev_t *mddev = bitmap->mddev;
a654b9d8
N
263
264 ITERATE_RDEV(mddev, rdev, tmp)
b2d444d7 265 if (test_bit(In_sync, &rdev->flags)
ab6085c7
N
266 && !test_bit(Faulty, &rdev->flags)) {
267 int size = PAGE_SIZE;
268 if (page->index == bitmap->file_pages-1)
269 size = roundup(bitmap->last_page_size,
270 bdev_hardsect_size(rdev->bdev));
a654b9d8 271 md_super_write(mddev, rdev,
ab6085c7 272 (rdev->sb_offset<<1) + bitmap->offset
a654b9d8 273 + page->index * (PAGE_SIZE/512),
ab6085c7 274 size,
a654b9d8 275 page);
ab6085c7 276 }
a654b9d8
N
277
278 if (wait)
a9701a30 279 md_super_wait(mddev);
a654b9d8
N
280 return 0;
281}
282
32a7627c 283/*
a654b9d8 284 * write out a page to a file
32a7627c 285 */
77ad4bc7 286static int write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 287{
d785a06a 288 struct buffer_head *bh;
32a7627c 289
a654b9d8 290 if (bitmap->file == NULL)
ab6085c7 291 return write_sb_page(bitmap, page, wait);
a654b9d8 292
d785a06a 293 bh = page_buffers(page);
c708443c 294
d785a06a
N
295 while (bh && bh->b_blocknr) {
296 atomic_inc(&bitmap->pending_writes);
297 set_buffer_locked(bh);
298 set_buffer_mapped(bh);
299 submit_bh(WRITE, bh);
300 bh = bh->b_this_page;
301 }
302
303 if (wait) {
304 wait_event(bitmap->write_wait,
305 atomic_read(&bitmap->pending_writes)==0);
306 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
8a5e9cf1 307 }
d785a06a
N
308 return 0;
309}
310
311static void end_bitmap_write(struct buffer_head *bh, int uptodate)
312{
313 struct bitmap *bitmap = bh->b_private;
314 unsigned long flags;
32a7627c 315
d785a06a
N
316 if (!uptodate) {
317 spin_lock_irqsave(&bitmap->lock, flags);
318 bitmap->flags |= BITMAP_WRITE_ERROR;
319 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 320 }
d785a06a
N
321 if (atomic_dec_and_test(&bitmap->pending_writes))
322 wake_up(&bitmap->write_wait);
323}
32a7627c 324
d785a06a
N
325/* copied from buffer.c */
326static void
327__clear_page_buffers(struct page *page)
328{
329 ClearPagePrivate(page);
330 set_page_private(page, 0);
331 page_cache_release(page);
332}
333static void free_buffers(struct page *page)
334{
335 struct buffer_head *bh = page_buffers(page);
77ad4bc7 336
d785a06a
N
337 while (bh) {
338 struct buffer_head *next = bh->b_this_page;
339 free_buffer_head(bh);
340 bh = next;
77ad4bc7 341 }
d785a06a
N
342 __clear_page_buffers(page);
343 put_page(page);
32a7627c
N
344}
345
d785a06a
N
346/* read a page from a file.
347 * We both read the page, and attach buffers to the page to record the
348 * address of each block (using bmap). These addresses will be used
349 * to write the block later, completely bypassing the filesystem.
350 * This usage is similar to how swap files are handled, and allows us
351 * to write to a file with no concerns of memory allocation failing.
352 */
32a7627c 353static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
354 struct bitmap *bitmap,
355 unsigned long count)
32a7627c 356{
32a7627c 357 struct page *page = NULL;
c649bb9c 358 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
359 struct buffer_head *bh;
360 sector_t block;
32a7627c 361
2d1f3b5d
N
362 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
363 (unsigned long long)index << PAGE_SHIFT);
32a7627c 364
d785a06a
N
365 page = alloc_page(GFP_KERNEL);
366 if (!page)
367 page = ERR_PTR(-ENOMEM);
32a7627c
N
368 if (IS_ERR(page))
369 goto out;
d785a06a 370
d785a06a
N
371 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
372 if (!bh) {
2d1f3b5d 373 put_page(page);
d785a06a 374 page = ERR_PTR(-ENOMEM);
32a7627c
N
375 goto out;
376 }
d785a06a
N
377 attach_page_buffers(page, bh);
378 block = index << (PAGE_SHIFT - inode->i_blkbits);
379 while (bh) {
380 if (count == 0)
381 bh->b_blocknr = 0;
382 else {
383 bh->b_blocknr = bmap(inode, block);
384 if (bh->b_blocknr == 0) {
385 /* Cannot use this file! */
386 free_buffers(page);
387 page = ERR_PTR(-EINVAL);
388 goto out;
389 }
390 bh->b_bdev = inode->i_sb->s_bdev;
391 if (count < (1<<inode->i_blkbits))
392 count = 0;
393 else
394 count -= (1<<inode->i_blkbits);
395
396 bh->b_end_io = end_bitmap_write;
397 bh->b_private = bitmap;
ce25c31b
N
398 atomic_inc(&bitmap->pending_writes);
399 set_buffer_locked(bh);
400 set_buffer_mapped(bh);
401 submit_bh(READ, bh);
d785a06a
N
402 }
403 block++;
404 bh = bh->b_this_page;
405 }
d785a06a 406 page->index = index;
ce25c31b
N
407
408 wait_event(bitmap->write_wait,
409 atomic_read(&bitmap->pending_writes)==0);
410 if (bitmap->flags & BITMAP_WRITE_ERROR) {
411 free_buffers(page);
412 page = ERR_PTR(-EIO);
413 }
32a7627c
N
414out:
415 if (IS_ERR(page))
416 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
417 (int)PAGE_SIZE,
418 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
419 PTR_ERR(page));
420 return page;
421}
422
423/*
424 * bitmap file superblock operations
425 */
426
427/* update the event counter and sync the superblock to disk */
428int bitmap_update_sb(struct bitmap *bitmap)
429{
430 bitmap_super_t *sb;
431 unsigned long flags;
432
433 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
434 return 0;
435 spin_lock_irqsave(&bitmap->lock, flags);
436 if (!bitmap->sb_page) { /* no superblock */
437 spin_unlock_irqrestore(&bitmap->lock, flags);
438 return 0;
439 }
32a7627c 440 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 441 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
442 sb->events = cpu_to_le64(bitmap->mddev->events);
443 if (!bitmap->mddev->degraded)
444 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
ea03aff9 445 kunmap_atomic(sb, KM_USER0);
8a5e9cf1 446 return write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
447}
448
449/* print out the bitmap file superblock */
450void bitmap_print_sb(struct bitmap *bitmap)
451{
452 bitmap_super_t *sb;
453
454 if (!bitmap || !bitmap->sb_page)
455 return;
ea03aff9 456 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 457 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
458 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
459 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
460 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
461 *(__u32 *)(sb->uuid+0),
462 *(__u32 *)(sb->uuid+4),
463 *(__u32 *)(sb->uuid+8),
464 *(__u32 *)(sb->uuid+12));
a2cff26a 465 printk(KERN_DEBUG " events: %llu\n",
32a7627c 466 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 467 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 468 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
469 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
470 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
471 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
472 printk(KERN_DEBUG " sync size: %llu KB\n",
473 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 474 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 475 kunmap_atomic(sb, KM_USER0);
32a7627c
N
476}
477
478/* read the superblock from the bitmap file and initialize some bitmap fields */
479static int bitmap_read_sb(struct bitmap *bitmap)
480{
481 char *reason = NULL;
482 bitmap_super_t *sb;
4b6d287f 483 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
484 unsigned long long events;
485 int err = -EINVAL;
486
487 /* page 0 is the superblock, read it... */
f49d5e62
N
488 if (bitmap->file) {
489 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
490 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
491
492 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
493 } else {
a654b9d8 494 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 495 }
32a7627c
N
496 if (IS_ERR(bitmap->sb_page)) {
497 err = PTR_ERR(bitmap->sb_page);
498 bitmap->sb_page = NULL;
499 return err;
500 }
501
ea03aff9 502 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 503
32a7627c
N
504 chunksize = le32_to_cpu(sb->chunksize);
505 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 506 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
507
508 /* verify that the bitmap-specific fields are valid */
509 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
510 reason = "bad magic";
bd926c63
N
511 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
512 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 513 reason = "unrecognized superblock version";
7dd5d34c
N
514 else if (chunksize < PAGE_SIZE)
515 reason = "bitmap chunksize too small";
32a7627c
N
516 else if ((1 << ffz(~chunksize)) != chunksize)
517 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
518 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
519 reason = "daemon sleep period out of range";
4b6d287f
N
520 else if (write_behind > COUNTER_MAX)
521 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
522 if (reason) {
523 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
524 bmname(bitmap), reason);
525 goto out;
526 }
527
528 /* keep the array size field of the bitmap superblock up to date */
529 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
530
531 if (!bitmap->mddev->persistent)
532 goto success;
533
534 /*
535 * if we have a persistent array superblock, compare the
536 * bitmap's UUID and event counter to the mddev's
537 */
538 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
539 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
540 bmname(bitmap));
541 goto out;
542 }
543 events = le64_to_cpu(sb->events);
544 if (events < bitmap->mddev->events) {
545 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
546 "-- forcing full recovery\n", bmname(bitmap), events,
547 (unsigned long long) bitmap->mddev->events);
4f2e639a 548 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
549 }
550success:
551 /* assign fields using values from superblock */
552 bitmap->chunksize = chunksize;
553 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 554 bitmap->daemon_lastrun = jiffies;
4b6d287f 555 bitmap->max_write_behind = write_behind;
4f2e639a 556 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
557 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
558 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 559 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 560 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 561 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
562 err = 0;
563out:
ea03aff9 564 kunmap_atomic(sb, KM_USER0);
32a7627c
N
565 if (err)
566 bitmap_print_sb(bitmap);
567 return err;
568}
569
570enum bitmap_mask_op {
571 MASK_SET,
572 MASK_UNSET
573};
574
575/* record the state of the bitmap in the superblock */
576static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
577 enum bitmap_mask_op op)
578{
579 bitmap_super_t *sb;
580 unsigned long flags;
581
582 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 583 if (!bitmap->sb_page) { /* can't set the state */
32a7627c
N
584 spin_unlock_irqrestore(&bitmap->lock, flags);
585 return;
586 }
32a7627c 587 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 588 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 589 switch (op) {
4f2e639a 590 case MASK_SET: sb->state |= cpu_to_le32(bits);
32a7627c 591 break;
4f2e639a 592 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
32a7627c
N
593 break;
594 default: BUG();
595 }
ea03aff9 596 kunmap_atomic(sb, KM_USER0);
32a7627c
N
597}
598
599/*
600 * general bitmap file operations
601 */
602
603/* calculate the index of the page that contains this bit */
604static inline unsigned long file_page_index(unsigned long chunk)
605{
606 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
607}
608
609/* calculate the (bit) offset of this bit within a page */
610static inline unsigned long file_page_offset(unsigned long chunk)
611{
612 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
613}
614
615/*
616 * return a pointer to the page in the filemap that contains the given bit
617 *
618 * this lookup is complicated by the fact that the bitmap sb might be exactly
619 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
620 * 0 or page 1
621 */
622static inline struct page *filemap_get_page(struct bitmap *bitmap,
623 unsigned long chunk)
624{
9b1d1dac 625 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
32a7627c
N
626 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
627}
628
629
630static void bitmap_file_unmap(struct bitmap *bitmap)
631{
632 struct page **map, *sb_page;
633 unsigned long *attr;
634 int pages;
635 unsigned long flags;
636
637 spin_lock_irqsave(&bitmap->lock, flags);
638 map = bitmap->filemap;
639 bitmap->filemap = NULL;
640 attr = bitmap->filemap_attr;
641 bitmap->filemap_attr = NULL;
642 pages = bitmap->file_pages;
643 bitmap->file_pages = 0;
644 sb_page = bitmap->sb_page;
645 bitmap->sb_page = NULL;
646 spin_unlock_irqrestore(&bitmap->lock, flags);
647
648 while (pages--)
649 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
d785a06a 650 free_buffers(map[pages]);
32a7627c
N
651 kfree(map);
652 kfree(attr);
653
d785a06a
N
654 if (sb_page)
655 free_buffers(sb_page);
32a7627c
N
656}
657
658static void bitmap_file_put(struct bitmap *bitmap)
659{
660 struct file *file;
32a7627c
N
661 unsigned long flags;
662
663 spin_lock_irqsave(&bitmap->lock, flags);
664 file = bitmap->file;
665 bitmap->file = NULL;
666 spin_unlock_irqrestore(&bitmap->lock, flags);
667
d785a06a
N
668 if (file)
669 wait_event(bitmap->write_wait,
670 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
671 bitmap_file_unmap(bitmap);
672
d785a06a 673 if (file) {
c649bb9c 674 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 675 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 676 fput(file);
d785a06a 677 }
32a7627c
N
678}
679
680
681/*
682 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
683 * then it is no longer reliable, so we stop using it and we mark the file
684 * as failed in the superblock
685 */
686static void bitmap_file_kick(struct bitmap *bitmap)
687{
688 char *path, *ptr = NULL;
689
690 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
691 bitmap_update_sb(bitmap);
692
a654b9d8
N
693 if (bitmap->file) {
694 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
695 if (path)
696 ptr = file_path(bitmap->file, path, PAGE_SIZE);
32a7627c 697
a654b9d8
N
698 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
699 bmname(bitmap), ptr ? ptr : "");
32a7627c 700
a654b9d8
N
701 kfree(path);
702 }
32a7627c
N
703
704 bitmap_file_put(bitmap);
705
706 return;
707}
708
709enum bitmap_page_attr {
e16b68b6
N
710 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
711 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
712 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
713};
714
715static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
716 enum bitmap_page_attr attr)
717{
e16b68b6 718 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
719}
720
721static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
722 enum bitmap_page_attr attr)
723{
e16b68b6 724 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
725}
726
ec7a3197
N
727static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
728 enum bitmap_page_attr attr)
32a7627c 729{
e16b68b6 730 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
731}
732
733/*
734 * bitmap_file_set_bit -- called before performing a write to the md device
735 * to set (and eventually sync) a particular bit in the bitmap file
736 *
737 * we set the bit immediately, then we record the page number so that
738 * when an unplug occurs, we can flush the dirty pages out to disk
739 */
740static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
741{
742 unsigned long bit;
743 struct page *page;
744 void *kaddr;
745 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
746
a654b9d8 747 if (!bitmap->filemap) {
32a7627c
N
748 return;
749 }
750
751 page = filemap_get_page(bitmap, chunk);
9b1d1dac 752 if (!page) return;
32a7627c
N
753 bit = file_page_offset(chunk);
754
32a7627c
N
755 /* set the bit */
756 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
757 if (bitmap->flags & BITMAP_HOSTENDIAN)
758 set_bit(bit, kaddr);
759 else
760 ext2_set_bit(bit, kaddr);
32a7627c
N
761 kunmap_atomic(kaddr, KM_USER0);
762 PRINTK("set file bit %lu page %lu\n", bit, page->index);
763
764 /* record page number so it gets flushed to disk when unplug occurs */
765 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
766
767}
768
769/* this gets called when the md device is ready to unplug its underlying
770 * (slave) device queues -- before we let any writes go down, we need to
771 * sync the dirty pages of the bitmap file to disk */
772int bitmap_unplug(struct bitmap *bitmap)
773{
ec7a3197
N
774 unsigned long i, flags;
775 int dirty, need_write;
32a7627c
N
776 struct page *page;
777 int wait = 0;
8a5e9cf1 778 int err;
32a7627c
N
779
780 if (!bitmap)
781 return 0;
782
783 /* look at each page to see if there are any set bits that need to be
784 * flushed out to disk */
785 for (i = 0; i < bitmap->file_pages; i++) {
786 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 787 if (!bitmap->filemap) {
32a7627c
N
788 spin_unlock_irqrestore(&bitmap->lock, flags);
789 return 0;
790 }
791 page = bitmap->filemap[i];
ec7a3197
N
792 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
793 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
794 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
795 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 796 if (dirty)
32a7627c
N
797 wait = 1;
798 spin_unlock_irqrestore(&bitmap->lock, flags);
799
d785a06a 800 if (dirty | need_write)
8a5e9cf1 801 err = write_page(bitmap, page, 0);
32a7627c
N
802 }
803 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 804 if (bitmap->file)
d785a06a
N
805 wait_event(bitmap->write_wait,
806 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 807 else
a9701a30 808 md_super_wait(bitmap->mddev);
32a7627c 809 }
d785a06a
N
810 if (bitmap->flags & BITMAP_WRITE_ERROR)
811 bitmap_file_kick(bitmap);
32a7627c
N
812 return 0;
813}
814
6a07997f 815static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
816/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
817 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
818 * memory mapping of the bitmap file
819 * Special cases:
820 * if there's no bitmap file, or if the bitmap file had been
821 * previously kicked from the array, we mark all the bits as
822 * 1's in order to cause a full resync.
6a07997f
N
823 *
824 * We ignore all bits for sectors that end earlier than 'start'.
825 * This is used when reading an out-of-date bitmap...
32a7627c 826 */
6a07997f 827static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
828{
829 unsigned long i, chunks, index, oldindex, bit;
830 struct page *page = NULL, *oldpage = NULL;
831 unsigned long num_pages, bit_cnt = 0;
832 struct file *file;
d785a06a 833 unsigned long bytes, offset;
32a7627c
N
834 int outofdate;
835 int ret = -ENOSPC;
ea03aff9 836 void *paddr;
32a7627c
N
837
838 chunks = bitmap->chunks;
839 file = bitmap->file;
840
a654b9d8 841 BUG_ON(!file && !bitmap->offset);
32a7627c 842
44456d37 843#ifdef INJECT_FAULTS_3
32a7627c
N
844 outofdate = 1;
845#else
846 outofdate = bitmap->flags & BITMAP_STALE;
847#endif
848 if (outofdate)
849 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
850 "recovery\n", bmname(bitmap));
851
852 bytes = (chunks + 7) / 8;
bc7f77de 853
cdbb4cc2 854 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 855
a654b9d8 856 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
857 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
858 bmname(bitmap),
859 (unsigned long) i_size_read(file->f_mapping->host),
860 bytes + sizeof(bitmap_super_t));
861 goto out;
862 }
bc7f77de
N
863
864 ret = -ENOMEM;
865
32a7627c 866 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 867 if (!bitmap->filemap)
32a7627c 868 goto out;
32a7627c 869
e16b68b6
N
870 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
871 bitmap->filemap_attr = kzalloc(
505fa2c4 872 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 873 GFP_KERNEL);
bc7f77de 874 if (!bitmap->filemap_attr)
32a7627c 875 goto out;
32a7627c 876
32a7627c
N
877 oldindex = ~0L;
878
879 for (i = 0; i < chunks; i++) {
bd926c63 880 int b;
32a7627c
N
881 index = file_page_index(i);
882 bit = file_page_offset(i);
883 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 884 int count;
32a7627c 885 /* unmap the old page, we're done with it */
d785a06a 886 if (index == num_pages-1)
f49d5e62
N
887 count = bytes + sizeof(bitmap_super_t)
888 - index * PAGE_SIZE;
d785a06a
N
889 else
890 count = PAGE_SIZE;
32a7627c
N
891 if (index == 0) {
892 /*
893 * if we're here then the superblock page
894 * contains some bits (PAGE_SIZE != sizeof sb)
895 * we've already read it in, so just use it
896 */
897 page = bitmap->sb_page;
898 offset = sizeof(bitmap_super_t);
a654b9d8 899 } else if (file) {
d785a06a 900 page = read_page(file, index, bitmap, count);
a654b9d8
N
901 offset = 0;
902 } else {
903 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
904 offset = 0;
905 }
a654b9d8
N
906 if (IS_ERR(page)) { /* read error */
907 ret = PTR_ERR(page);
908 goto out;
909 }
910
32a7627c
N
911 oldindex = index;
912 oldpage = page;
32a7627c
N
913
914 if (outofdate) {
915 /*
916 * if bitmap is out of date, dirty the
917 * whole page and write it out
918 */
ea03aff9
N
919 paddr = kmap_atomic(page, KM_USER0);
920 memset(paddr + offset, 0xff,
6a07997f 921 PAGE_SIZE - offset);
ea03aff9 922 kunmap_atomic(paddr, KM_USER0);
77ad4bc7 923 ret = write_page(bitmap, page, 1);
32a7627c 924 if (ret) {
32a7627c 925 /* release, page not in filemap yet */
2d1f3b5d 926 put_page(page);
32a7627c
N
927 goto out;
928 }
929 }
930
931 bitmap->filemap[bitmap->file_pages++] = page;
ab6085c7 932 bitmap->last_page_size = count;
32a7627c 933 }
ea03aff9 934 paddr = kmap_atomic(page, KM_USER0);
bd926c63 935 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 936 b = test_bit(bit, paddr);
bd926c63 937 else
ea03aff9
N
938 b = ext2_test_bit(bit, paddr);
939 kunmap_atomic(paddr, KM_USER0);
bd926c63 940 if (b) {
32a7627c 941 /* if the disk bit is set, set the memory bit */
6a07997f
N
942 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
943 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
944 );
32a7627c 945 bit_cnt++;
6a07997f 946 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 947 }
32a7627c
N
948 }
949
950 /* everything went OK */
951 ret = 0;
952 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
953
32a7627c
N
954 if (bit_cnt) { /* Kick recovery if any bits were set */
955 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
956 md_wakeup_thread(bitmap->mddev->thread);
957 }
958
959out:
960 printk(KERN_INFO "%s: bitmap initialized from disk: "
961 "read %lu/%lu pages, set %lu bits, status: %d\n",
962 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
963
964 return ret;
965}
966
a654b9d8
N
967void bitmap_write_all(struct bitmap *bitmap)
968{
969 /* We don't actually write all bitmap blocks here,
970 * just flag them as needing to be written
971 */
ec7a3197 972 int i;
a654b9d8 973
ec7a3197
N
974 for (i=0; i < bitmap->file_pages; i++)
975 set_page_attr(bitmap, bitmap->filemap[i],
976 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
977}
978
32a7627c
N
979
980static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
981{
982 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
983 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
984 bitmap->bp[page].count += inc;
985/*
986 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
987 (unsigned long long)offset, inc, bitmap->bp[page].count);
988*/
989 bitmap_checkfree(bitmap, page);
990}
991static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
992 sector_t offset, int *blocks,
993 int create);
994
995/*
996 * bitmap daemon -- periodically wakes up to clean bits and flush pages
997 * out to disk
998 */
999
1000int bitmap_daemon_work(struct bitmap *bitmap)
1001{
aa3163f8 1002 unsigned long j;
32a7627c
N
1003 unsigned long flags;
1004 struct page *page = NULL, *lastpage = NULL;
1005 int err = 0;
1006 int blocks;
ea03aff9 1007 void *paddr;
32a7627c
N
1008
1009 if (bitmap == NULL)
1010 return 0;
1011 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1012 return 0;
1013 bitmap->daemon_lastrun = jiffies;
1014
1015 for (j = 0; j < bitmap->chunks; j++) {
1016 bitmap_counter_t *bmc;
1017 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1018 if (!bitmap->filemap) {
32a7627c
N
1019 /* error or shutdown */
1020 spin_unlock_irqrestore(&bitmap->lock, flags);
1021 break;
1022 }
1023
1024 page = filemap_get_page(bitmap, j);
32a7627c
N
1025
1026 if (page != lastpage) {
aa3163f8 1027 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1028 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1029 int need_write = test_page_attr(bitmap, page,
1030 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1031 if (need_write)
aa3163f8 1032 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1033
aa3163f8 1034 spin_unlock_irqrestore(&bitmap->lock, flags);
ec7a3197 1035 if (need_write) {
8a5e9cf1 1036 switch (write_page(bitmap, page, 0)) {
8a5e9cf1
N
1037 case 0:
1038 break;
1039 default:
aa3163f8 1040 bitmap_file_kick(bitmap);
8a5e9cf1 1041 }
aa3163f8
N
1042 }
1043 continue;
1044 }
1045
32a7627c 1046 /* grab the new page, sync and release the old */
32a7627c 1047 if (lastpage != NULL) {
ec7a3197 1048 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1049 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1050 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1051 err = write_page(bitmap, lastpage, 0);
32a7627c
N
1052 } else {
1053 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1054 spin_unlock_irqrestore(&bitmap->lock, flags);
1055 }
32a7627c
N
1056 if (err)
1057 bitmap_file_kick(bitmap);
1058 } else
1059 spin_unlock_irqrestore(&bitmap->lock, flags);
1060 lastpage = page;
32a7627c
N
1061/*
1062 printk("bitmap clean at page %lu\n", j);
1063*/
1064 spin_lock_irqsave(&bitmap->lock, flags);
1065 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1066 }
1067 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1068 &blocks, 0);
1069 if (bmc) {
1070/*
1071 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1072*/
1073 if (*bmc == 2) {
1074 *bmc=1; /* maybe clear the bit next time */
1075 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1076 } else if (*bmc == 1) {
1077 /* we can clear the bit */
1078 *bmc = 0;
1079 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1080 -1);
1081
1082 /* clear the bit */
ea03aff9 1083 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1084 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1085 clear_bit(file_page_offset(j), paddr);
bd926c63 1086 else
ea03aff9
N
1087 ext2_clear_bit(file_page_offset(j), paddr);
1088 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1089 }
1090 }
1091 spin_unlock_irqrestore(&bitmap->lock, flags);
1092 }
1093
1094 /* now sync the final page */
1095 if (lastpage != NULL) {
32a7627c 1096 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1097 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1098 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1099 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1100 err = write_page(bitmap, lastpage, 0);
32a7627c
N
1101 } else {
1102 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1103 spin_unlock_irqrestore(&bitmap->lock, flags);
1104 }
32a7627c
N
1105 }
1106
1107 return err;
1108}
1109
32a7627c
N
1110static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1111 sector_t offset, int *blocks,
1112 int create)
1113{
1114 /* If 'create', we might release the lock and reclaim it.
1115 * The lock must have been taken with interrupts enabled.
1116 * If !create, we don't release the lock.
1117 */
1118 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1119 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1120 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1121 sector_t csize;
1122
1123 if (bitmap_checkpage(bitmap, page, create) < 0) {
1124 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1125 *blocks = csize - (offset & (csize- 1));
1126 return NULL;
1127 }
1128 /* now locked ... */
1129
1130 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1131 /* should we use the first or second counter field
1132 * of the hijacked pointer? */
1133 int hi = (pageoff > PAGE_COUNTER_MASK);
1134 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1135 PAGE_COUNTER_SHIFT - 1);
1136 *blocks = csize - (offset & (csize- 1));
1137 return &((bitmap_counter_t *)
1138 &bitmap->bp[page].map)[hi];
1139 } else { /* page is allocated */
1140 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1141 *blocks = csize - (offset & (csize- 1));
1142 return (bitmap_counter_t *)
1143 &(bitmap->bp[page].map[pageoff]);
1144 }
1145}
1146
4b6d287f 1147int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1148{
1149 if (!bitmap) return 0;
4b6d287f
N
1150
1151 if (behind) {
1152 atomic_inc(&bitmap->behind_writes);
1153 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1154 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1155 }
1156
32a7627c
N
1157 while (sectors) {
1158 int blocks;
1159 bitmap_counter_t *bmc;
1160
1161 spin_lock_irq(&bitmap->lock);
1162 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1163 if (!bmc) {
1164 spin_unlock_irq(&bitmap->lock);
1165 return 0;
1166 }
1167
da6e1a32
NB
1168 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1169 DEFINE_WAIT(__wait);
1170 /* note that it is safe to do the prepare_to_wait
1171 * after the test as long as we do it before dropping
1172 * the spinlock.
1173 */
1174 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1175 TASK_UNINTERRUPTIBLE);
1176 spin_unlock_irq(&bitmap->lock);
1177 bitmap->mddev->queue
1178 ->unplug_fn(bitmap->mddev->queue);
1179 schedule();
1180 finish_wait(&bitmap->overflow_wait, &__wait);
1181 continue;
1182 }
1183
32a7627c
N
1184 switch(*bmc) {
1185 case 0:
1186 bitmap_file_set_bit(bitmap, offset);
1187 bitmap_count_page(bitmap,offset, 1);
1188 blk_plug_device(bitmap->mddev->queue);
1189 /* fall through */
1190 case 1:
1191 *bmc = 2;
1192 }
da6e1a32 1193
32a7627c
N
1194 (*bmc)++;
1195
1196 spin_unlock_irq(&bitmap->lock);
1197
1198 offset += blocks;
1199 if (sectors > blocks)
1200 sectors -= blocks;
1201 else sectors = 0;
1202 }
1203 return 0;
1204}
1205
1206void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1207 int success, int behind)
32a7627c
N
1208{
1209 if (!bitmap) return;
4b6d287f
N
1210 if (behind) {
1211 atomic_dec(&bitmap->behind_writes);
1212 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1213 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1214 }
1215
32a7627c
N
1216 while (sectors) {
1217 int blocks;
1218 unsigned long flags;
1219 bitmap_counter_t *bmc;
1220
1221 spin_lock_irqsave(&bitmap->lock, flags);
1222 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1223 if (!bmc) {
1224 spin_unlock_irqrestore(&bitmap->lock, flags);
1225 return;
1226 }
1227
1228 if (!success && ! (*bmc & NEEDED_MASK))
1229 *bmc |= NEEDED_MASK;
1230
da6e1a32
NB
1231 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1232 wake_up(&bitmap->overflow_wait);
1233
32a7627c
N
1234 (*bmc)--;
1235 if (*bmc <= 2) {
1236 set_page_attr(bitmap,
1237 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1238 BITMAP_PAGE_CLEAN);
1239 }
1240 spin_unlock_irqrestore(&bitmap->lock, flags);
1241 offset += blocks;
1242 if (sectors > blocks)
1243 sectors -= blocks;
1244 else sectors = 0;
1245 }
1246}
1247
6a806c51
N
1248int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1249 int degraded)
32a7627c
N
1250{
1251 bitmap_counter_t *bmc;
1252 int rv;
1253 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1254 *blocks = 1024;
1255 return 1; /* always resync if no bitmap */
1256 }
1257 spin_lock_irq(&bitmap->lock);
1258 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1259 rv = 0;
1260 if (bmc) {
1261 /* locked */
1262 if (RESYNC(*bmc))
1263 rv = 1;
1264 else if (NEEDED(*bmc)) {
1265 rv = 1;
6a806c51
N
1266 if (!degraded) { /* don't set/clear bits if degraded */
1267 *bmc |= RESYNC_MASK;
1268 *bmc &= ~NEEDED_MASK;
1269 }
32a7627c
N
1270 }
1271 }
1272 spin_unlock_irq(&bitmap->lock);
1273 return rv;
1274}
1275
1276void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1277{
1278 bitmap_counter_t *bmc;
1279 unsigned long flags;
1280/*
1281 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1282*/ if (bitmap == NULL) {
1283 *blocks = 1024;
1284 return;
1285 }
1286 spin_lock_irqsave(&bitmap->lock, flags);
1287 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1288 if (bmc == NULL)
1289 goto unlock;
1290 /* locked */
1291/*
1292 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1293*/
1294 if (RESYNC(*bmc)) {
1295 *bmc &= ~RESYNC_MASK;
1296
1297 if (!NEEDED(*bmc) && aborted)
1298 *bmc |= NEEDED_MASK;
1299 else {
1300 if (*bmc <= 2) {
1301 set_page_attr(bitmap,
1302 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1303 BITMAP_PAGE_CLEAN);
1304 }
1305 }
1306 }
1307 unlock:
1308 spin_unlock_irqrestore(&bitmap->lock, flags);
1309}
1310
1311void bitmap_close_sync(struct bitmap *bitmap)
1312{
1313 /* Sync has finished, and any bitmap chunks that weren't synced
1314 * properly have been aborted. It remains to us to clear the
1315 * RESYNC bit wherever it is still on
1316 */
1317 sector_t sector = 0;
1318 int blocks;
1319 if (!bitmap) return;
1320 while (sector < bitmap->mddev->resync_max_sectors) {
1321 bitmap_end_sync(bitmap, sector, &blocks, 0);
1322/*
1323 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1324 (unsigned long long)sector, blocks);
1325*/ sector += blocks;
1326 }
1327}
1328
6a07997f 1329static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1330{
1331 /* For each chunk covered by any of these sectors, set the
193f1c93 1332 * counter to 1 and set resync_needed. They should all
32a7627c
N
1333 * be 0 at this point
1334 */
193f1c93
N
1335
1336 int secs;
1337 bitmap_counter_t *bmc;
1338 spin_lock_irq(&bitmap->lock);
1339 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1340 if (!bmc) {
32a7627c 1341 spin_unlock_irq(&bitmap->lock);
193f1c93 1342 return;
32a7627c 1343 }
193f1c93
N
1344 if (! *bmc) {
1345 struct page *page;
6a07997f 1346 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1347 bitmap_count_page(bitmap, offset, 1);
1348 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1349 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1350 }
1351 spin_unlock_irq(&bitmap->lock);
1352
32a7627c
N
1353}
1354
9b1d1dac
PC
1355/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1356void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1357{
1358 unsigned long chunk;
1359
1360 for (chunk = s; chunk <= e; chunk++) {
1361 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1362 bitmap_set_memory_bits(bitmap, sec, 1);
1363 bitmap_file_set_bit(bitmap, sec);
1364 }
1365}
1366
6b8b3e8a
N
1367/*
1368 * flush out any pending updates
1369 */
1370void bitmap_flush(mddev_t *mddev)
1371{
1372 struct bitmap *bitmap = mddev->bitmap;
1373 int sleep;
1374
1375 if (!bitmap) /* there was no bitmap */
1376 return;
1377
1378 /* run the daemon_work three time to ensure everything is flushed
1379 * that can be
1380 */
1381 sleep = bitmap->daemon_sleep;
1382 bitmap->daemon_sleep = 0;
1383 bitmap_daemon_work(bitmap);
1384 bitmap_daemon_work(bitmap);
1385 bitmap_daemon_work(bitmap);
1386 bitmap->daemon_sleep = sleep;
1387 bitmap_update_sb(bitmap);
1388}
1389
32a7627c
N
1390/*
1391 * free memory that was allocated
1392 */
3178b0db 1393static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1394{
1395 unsigned long k, pages;
1396 struct bitmap_page *bp;
32a7627c
N
1397
1398 if (!bitmap) /* there was no bitmap */
1399 return;
1400
32a7627c
N
1401 /* release the bitmap file and kill the daemon */
1402 bitmap_file_put(bitmap);
1403
1404 bp = bitmap->bp;
1405 pages = bitmap->pages;
1406
1407 /* free all allocated memory */
1408
32a7627c
N
1409 if (bp) /* deallocate the page memory */
1410 for (k = 0; k < pages; k++)
1411 if (bp[k].map && !bp[k].hijacked)
1412 kfree(bp[k].map);
1413 kfree(bp);
1414 kfree(bitmap);
1415}
3178b0db
N
1416void bitmap_destroy(mddev_t *mddev)
1417{
1418 struct bitmap *bitmap = mddev->bitmap;
1419
1420 if (!bitmap) /* there was no bitmap */
1421 return;
1422
1423 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1424 if (mddev->thread)
1425 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1426
1427 bitmap_free(bitmap);
1428}
32a7627c
N
1429
1430/*
1431 * initialize the bitmap structure
1432 * if this returns an error, bitmap_destroy must be called to do clean up
1433 */
1434int bitmap_create(mddev_t *mddev)
1435{
1436 struct bitmap *bitmap;
1437 unsigned long blocks = mddev->resync_max_sectors;
1438 unsigned long chunks;
1439 unsigned long pages;
1440 struct file *file = mddev->bitmap_file;
1441 int err;
6a07997f 1442 sector_t start;
32a7627c 1443
5f6e3c83 1444 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1445
a654b9d8 1446 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1447 return 0;
1448
a654b9d8
N
1449 BUG_ON(file && mddev->bitmap_offset);
1450
9ffae0cf 1451 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1452 if (!bitmap)
1453 return -ENOMEM;
1454
32a7627c 1455 spin_lock_init(&bitmap->lock);
ce25c31b
N
1456 atomic_set(&bitmap->pending_writes, 0);
1457 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1458 init_waitqueue_head(&bitmap->overflow_wait);
ce25c31b 1459
32a7627c 1460 bitmap->mddev = mddev;
32a7627c 1461
32a7627c 1462 bitmap->file = file;
a654b9d8 1463 bitmap->offset = mddev->bitmap_offset;
ce25c31b
N
1464 if (file) {
1465 get_file(file);
ef51c976
MF
1466 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1467 SYNC_FILE_RANGE_WAIT_BEFORE |
1468 SYNC_FILE_RANGE_WRITE |
1469 SYNC_FILE_RANGE_WAIT_AFTER);
ce25c31b 1470 }
32a7627c
N
1471 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1472 err = bitmap_read_sb(bitmap);
1473 if (err)
3178b0db 1474 goto error;
32a7627c 1475
a638b2dc 1476 bitmap->chunkshift = ffz(~bitmap->chunksize);
32a7627c
N
1477
1478 /* now that chunksize and chunkshift are set, we can use these macros */
1479 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1480 CHUNK_BLOCK_RATIO(bitmap);
1481 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1482
1483 BUG_ON(!pages);
1484
1485 bitmap->chunks = chunks;
1486 bitmap->pages = pages;
1487 bitmap->missing_pages = pages;
1488 bitmap->counter_bits = COUNTER_BITS;
1489
1490 bitmap->syncchunk = ~0UL;
1491
44456d37 1492#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1493 bitmap->bp = NULL;
1494#else
9ffae0cf 1495 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1496#endif
3178b0db 1497 err = -ENOMEM;
32a7627c 1498 if (!bitmap->bp)
3178b0db 1499 goto error;
32a7627c 1500
32a7627c
N
1501 /* now that we have some pages available, initialize the in-memory
1502 * bitmap from the on-disk bitmap */
6a07997f
N
1503 start = 0;
1504 if (mddev->degraded == 0
1505 || bitmap->events_cleared == mddev->events)
1506 /* no need to keep dirty bits to optimise a re-add of a missing device */
1507 start = mddev->recovery_cp;
1508 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1509
32a7627c 1510 if (err)
3178b0db 1511 goto error;
32a7627c
N
1512
1513 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1514 pages, bmname(bitmap));
1515
3178b0db
N
1516 mddev->bitmap = bitmap;
1517
b15c2e57
N
1518 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1519
32a7627c 1520 return bitmap_update_sb(bitmap);
3178b0db
N
1521
1522 error:
1523 bitmap_free(bitmap);
1524 return err;
32a7627c
N
1525}
1526
1527/* the bitmap API -- for raid personalities */
1528EXPORT_SYMBOL(bitmap_startwrite);
1529EXPORT_SYMBOL(bitmap_endwrite);
1530EXPORT_SYMBOL(bitmap_start_sync);
1531EXPORT_SYMBOL(bitmap_end_sync);
1532EXPORT_SYMBOL(bitmap_unplug);
1533EXPORT_SYMBOL(bitmap_close_sync);
This page took 0.327652 seconds and 5 git commands to generate.