md: avoid overflow in raid0 calculation with large components
[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
258static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
259{
260 mdk_rdev_t *rdev;
261 struct list_head *tmp;
262
263 ITERATE_RDEV(mddev, rdev, tmp)
b2d444d7
N
264 if (test_bit(In_sync, &rdev->flags)
265 && !test_bit(Faulty, &rdev->flags))
a654b9d8
N
266 md_super_write(mddev, rdev,
267 (rdev->sb_offset<<1) + offset
268 + page->index * (PAGE_SIZE/512),
269 PAGE_SIZE,
270 page);
271
272 if (wait)
a9701a30 273 md_super_wait(mddev);
a654b9d8
N
274 return 0;
275}
276
32a7627c 277/*
a654b9d8 278 * write out a page to a file
32a7627c 279 */
77ad4bc7 280static int write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 281{
d785a06a 282 struct buffer_head *bh;
32a7627c 283
a654b9d8
N
284 if (bitmap->file == NULL)
285 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
286
d785a06a 287 bh = page_buffers(page);
c708443c 288
d785a06a
N
289 while (bh && bh->b_blocknr) {
290 atomic_inc(&bitmap->pending_writes);
291 set_buffer_locked(bh);
292 set_buffer_mapped(bh);
293 submit_bh(WRITE, bh);
294 bh = bh->b_this_page;
295 }
296
297 if (wait) {
298 wait_event(bitmap->write_wait,
299 atomic_read(&bitmap->pending_writes)==0);
300 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
8a5e9cf1 301 }
d785a06a
N
302 return 0;
303}
304
305static void end_bitmap_write(struct buffer_head *bh, int uptodate)
306{
307 struct bitmap *bitmap = bh->b_private;
308 unsigned long flags;
32a7627c 309
d785a06a
N
310 if (!uptodate) {
311 spin_lock_irqsave(&bitmap->lock, flags);
312 bitmap->flags |= BITMAP_WRITE_ERROR;
313 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 314 }
d785a06a
N
315 if (atomic_dec_and_test(&bitmap->pending_writes))
316 wake_up(&bitmap->write_wait);
317}
32a7627c 318
d785a06a
N
319/* copied from buffer.c */
320static void
321__clear_page_buffers(struct page *page)
322{
323 ClearPagePrivate(page);
324 set_page_private(page, 0);
325 page_cache_release(page);
326}
327static void free_buffers(struct page *page)
328{
329 struct buffer_head *bh = page_buffers(page);
77ad4bc7 330
d785a06a
N
331 while (bh) {
332 struct buffer_head *next = bh->b_this_page;
333 free_buffer_head(bh);
334 bh = next;
77ad4bc7 335 }
d785a06a
N
336 __clear_page_buffers(page);
337 put_page(page);
32a7627c
N
338}
339
d785a06a
N
340/* read a page from a file.
341 * We both read the page, and attach buffers to the page to record the
342 * address of each block (using bmap). These addresses will be used
343 * to write the block later, completely bypassing the filesystem.
344 * This usage is similar to how swap files are handled, and allows us
345 * to write to a file with no concerns of memory allocation failing.
346 */
32a7627c 347static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
348 struct bitmap *bitmap,
349 unsigned long count)
32a7627c 350{
32a7627c 351 struct page *page = NULL;
c649bb9c 352 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
353 struct buffer_head *bh;
354 sector_t block;
32a7627c 355
2d1f3b5d
N
356 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
357 (unsigned long long)index << PAGE_SHIFT);
32a7627c 358
d785a06a
N
359 page = alloc_page(GFP_KERNEL);
360 if (!page)
361 page = ERR_PTR(-ENOMEM);
32a7627c
N
362 if (IS_ERR(page))
363 goto out;
d785a06a 364
d785a06a
N
365 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
366 if (!bh) {
2d1f3b5d 367 put_page(page);
d785a06a 368 page = ERR_PTR(-ENOMEM);
32a7627c
N
369 goto out;
370 }
d785a06a
N
371 attach_page_buffers(page, bh);
372 block = index << (PAGE_SHIFT - inode->i_blkbits);
373 while (bh) {
374 if (count == 0)
375 bh->b_blocknr = 0;
376 else {
377 bh->b_blocknr = bmap(inode, block);
378 if (bh->b_blocknr == 0) {
379 /* Cannot use this file! */
380 free_buffers(page);
381 page = ERR_PTR(-EINVAL);
382 goto out;
383 }
384 bh->b_bdev = inode->i_sb->s_bdev;
385 if (count < (1<<inode->i_blkbits))
386 count = 0;
387 else
388 count -= (1<<inode->i_blkbits);
389
390 bh->b_end_io = end_bitmap_write;
391 bh->b_private = bitmap;
ce25c31b
N
392 atomic_inc(&bitmap->pending_writes);
393 set_buffer_locked(bh);
394 set_buffer_mapped(bh);
395 submit_bh(READ, bh);
d785a06a
N
396 }
397 block++;
398 bh = bh->b_this_page;
399 }
d785a06a 400 page->index = index;
ce25c31b
N
401
402 wait_event(bitmap->write_wait,
403 atomic_read(&bitmap->pending_writes)==0);
404 if (bitmap->flags & BITMAP_WRITE_ERROR) {
405 free_buffers(page);
406 page = ERR_PTR(-EIO);
407 }
32a7627c
N
408out:
409 if (IS_ERR(page))
410 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
411 (int)PAGE_SIZE,
412 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
413 PTR_ERR(page));
414 return page;
415}
416
417/*
418 * bitmap file superblock operations
419 */
420
421/* update the event counter and sync the superblock to disk */
422int bitmap_update_sb(struct bitmap *bitmap)
423{
424 bitmap_super_t *sb;
425 unsigned long flags;
426
427 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
428 return 0;
429 spin_lock_irqsave(&bitmap->lock, flags);
430 if (!bitmap->sb_page) { /* no superblock */
431 spin_unlock_irqrestore(&bitmap->lock, flags);
432 return 0;
433 }
32a7627c 434 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 435 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
436 sb->events = cpu_to_le64(bitmap->mddev->events);
437 if (!bitmap->mddev->degraded)
438 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
ea03aff9 439 kunmap_atomic(sb, KM_USER0);
8a5e9cf1 440 return write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
441}
442
443/* print out the bitmap file superblock */
444void bitmap_print_sb(struct bitmap *bitmap)
445{
446 bitmap_super_t *sb;
447
448 if (!bitmap || !bitmap->sb_page)
449 return;
ea03aff9 450 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 451 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
452 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
453 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
454 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
455 *(__u32 *)(sb->uuid+0),
456 *(__u32 *)(sb->uuid+4),
457 *(__u32 *)(sb->uuid+8),
458 *(__u32 *)(sb->uuid+12));
a2cff26a 459 printk(KERN_DEBUG " events: %llu\n",
32a7627c 460 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 461 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 462 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
463 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
464 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
465 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
466 printk(KERN_DEBUG " sync size: %llu KB\n",
467 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 468 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 469 kunmap_atomic(sb, KM_USER0);
32a7627c
N
470}
471
472/* read the superblock from the bitmap file and initialize some bitmap fields */
473static int bitmap_read_sb(struct bitmap *bitmap)
474{
475 char *reason = NULL;
476 bitmap_super_t *sb;
4b6d287f 477 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
478 unsigned long long events;
479 int err = -EINVAL;
480
481 /* page 0 is the superblock, read it... */
f49d5e62
N
482 if (bitmap->file) {
483 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
484 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
485
486 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
487 } else {
a654b9d8 488 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 489 }
32a7627c
N
490 if (IS_ERR(bitmap->sb_page)) {
491 err = PTR_ERR(bitmap->sb_page);
492 bitmap->sb_page = NULL;
493 return err;
494 }
495
ea03aff9 496 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 497
32a7627c
N
498 chunksize = le32_to_cpu(sb->chunksize);
499 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 500 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
501
502 /* verify that the bitmap-specific fields are valid */
503 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
504 reason = "bad magic";
bd926c63
N
505 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
506 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 507 reason = "unrecognized superblock version";
7dd5d34c
N
508 else if (chunksize < PAGE_SIZE)
509 reason = "bitmap chunksize too small";
32a7627c
N
510 else if ((1 << ffz(~chunksize)) != chunksize)
511 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
512 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
513 reason = "daemon sleep period out of range";
4b6d287f
N
514 else if (write_behind > COUNTER_MAX)
515 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
516 if (reason) {
517 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
518 bmname(bitmap), reason);
519 goto out;
520 }
521
522 /* keep the array size field of the bitmap superblock up to date */
523 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
524
525 if (!bitmap->mddev->persistent)
526 goto success;
527
528 /*
529 * if we have a persistent array superblock, compare the
530 * bitmap's UUID and event counter to the mddev's
531 */
532 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
533 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
534 bmname(bitmap));
535 goto out;
536 }
537 events = le64_to_cpu(sb->events);
538 if (events < bitmap->mddev->events) {
539 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
540 "-- forcing full recovery\n", bmname(bitmap), events,
541 (unsigned long long) bitmap->mddev->events);
4f2e639a 542 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
543 }
544success:
545 /* assign fields using values from superblock */
546 bitmap->chunksize = chunksize;
547 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 548 bitmap->daemon_lastrun = jiffies;
4b6d287f 549 bitmap->max_write_behind = write_behind;
4f2e639a 550 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
551 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
552 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 553 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 554 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 555 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
556 err = 0;
557out:
ea03aff9 558 kunmap_atomic(sb, KM_USER0);
32a7627c
N
559 if (err)
560 bitmap_print_sb(bitmap);
561 return err;
562}
563
564enum bitmap_mask_op {
565 MASK_SET,
566 MASK_UNSET
567};
568
569/* record the state of the bitmap in the superblock */
570static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
571 enum bitmap_mask_op op)
572{
573 bitmap_super_t *sb;
574 unsigned long flags;
575
576 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 577 if (!bitmap->sb_page) { /* can't set the state */
32a7627c
N
578 spin_unlock_irqrestore(&bitmap->lock, flags);
579 return;
580 }
32a7627c 581 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 582 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 583 switch (op) {
4f2e639a 584 case MASK_SET: sb->state |= cpu_to_le32(bits);
32a7627c 585 break;
4f2e639a 586 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
32a7627c
N
587 break;
588 default: BUG();
589 }
ea03aff9 590 kunmap_atomic(sb, KM_USER0);
32a7627c
N
591}
592
593/*
594 * general bitmap file operations
595 */
596
597/* calculate the index of the page that contains this bit */
598static inline unsigned long file_page_index(unsigned long chunk)
599{
600 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
601}
602
603/* calculate the (bit) offset of this bit within a page */
604static inline unsigned long file_page_offset(unsigned long chunk)
605{
606 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
607}
608
609/*
610 * return a pointer to the page in the filemap that contains the given bit
611 *
612 * this lookup is complicated by the fact that the bitmap sb might be exactly
613 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
614 * 0 or page 1
615 */
616static inline struct page *filemap_get_page(struct bitmap *bitmap,
617 unsigned long chunk)
618{
9b1d1dac 619 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
32a7627c
N
620 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
621}
622
623
624static void bitmap_file_unmap(struct bitmap *bitmap)
625{
626 struct page **map, *sb_page;
627 unsigned long *attr;
628 int pages;
629 unsigned long flags;
630
631 spin_lock_irqsave(&bitmap->lock, flags);
632 map = bitmap->filemap;
633 bitmap->filemap = NULL;
634 attr = bitmap->filemap_attr;
635 bitmap->filemap_attr = NULL;
636 pages = bitmap->file_pages;
637 bitmap->file_pages = 0;
638 sb_page = bitmap->sb_page;
639 bitmap->sb_page = NULL;
640 spin_unlock_irqrestore(&bitmap->lock, flags);
641
642 while (pages--)
643 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
d785a06a 644 free_buffers(map[pages]);
32a7627c
N
645 kfree(map);
646 kfree(attr);
647
d785a06a
N
648 if (sb_page)
649 free_buffers(sb_page);
32a7627c
N
650}
651
652static void bitmap_file_put(struct bitmap *bitmap)
653{
654 struct file *file;
32a7627c
N
655 unsigned long flags;
656
657 spin_lock_irqsave(&bitmap->lock, flags);
658 file = bitmap->file;
659 bitmap->file = NULL;
660 spin_unlock_irqrestore(&bitmap->lock, flags);
661
d785a06a
N
662 if (file)
663 wait_event(bitmap->write_wait,
664 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
665 bitmap_file_unmap(bitmap);
666
d785a06a 667 if (file) {
c649bb9c 668 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 669 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 670 fput(file);
d785a06a 671 }
32a7627c
N
672}
673
674
675/*
676 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
677 * then it is no longer reliable, so we stop using it and we mark the file
678 * as failed in the superblock
679 */
680static void bitmap_file_kick(struct bitmap *bitmap)
681{
682 char *path, *ptr = NULL;
683
684 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
685 bitmap_update_sb(bitmap);
686
a654b9d8
N
687 if (bitmap->file) {
688 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
689 if (path)
690 ptr = file_path(bitmap->file, path, PAGE_SIZE);
32a7627c 691
a654b9d8
N
692 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
693 bmname(bitmap), ptr ? ptr : "");
32a7627c 694
a654b9d8
N
695 kfree(path);
696 }
32a7627c
N
697
698 bitmap_file_put(bitmap);
699
700 return;
701}
702
703enum bitmap_page_attr {
e16b68b6
N
704 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
705 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
706 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
707};
708
709static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
710 enum bitmap_page_attr attr)
711{
e16b68b6 712 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
713}
714
715static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
716 enum bitmap_page_attr attr)
717{
e16b68b6 718 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
719}
720
ec7a3197
N
721static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
722 enum bitmap_page_attr attr)
32a7627c 723{
e16b68b6 724 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
725}
726
727/*
728 * bitmap_file_set_bit -- called before performing a write to the md device
729 * to set (and eventually sync) a particular bit in the bitmap file
730 *
731 * we set the bit immediately, then we record the page number so that
732 * when an unplug occurs, we can flush the dirty pages out to disk
733 */
734static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
735{
736 unsigned long bit;
737 struct page *page;
738 void *kaddr;
739 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
740
a654b9d8 741 if (!bitmap->filemap) {
32a7627c
N
742 return;
743 }
744
745 page = filemap_get_page(bitmap, chunk);
9b1d1dac 746 if (!page) return;
32a7627c
N
747 bit = file_page_offset(chunk);
748
32a7627c
N
749 /* set the bit */
750 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
751 if (bitmap->flags & BITMAP_HOSTENDIAN)
752 set_bit(bit, kaddr);
753 else
754 ext2_set_bit(bit, kaddr);
32a7627c
N
755 kunmap_atomic(kaddr, KM_USER0);
756 PRINTK("set file bit %lu page %lu\n", bit, page->index);
757
758 /* record page number so it gets flushed to disk when unplug occurs */
759 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
760
761}
762
763/* this gets called when the md device is ready to unplug its underlying
764 * (slave) device queues -- before we let any writes go down, we need to
765 * sync the dirty pages of the bitmap file to disk */
766int bitmap_unplug(struct bitmap *bitmap)
767{
ec7a3197
N
768 unsigned long i, flags;
769 int dirty, need_write;
32a7627c
N
770 struct page *page;
771 int wait = 0;
8a5e9cf1 772 int err;
32a7627c
N
773
774 if (!bitmap)
775 return 0;
776
777 /* look at each page to see if there are any set bits that need to be
778 * flushed out to disk */
779 for (i = 0; i < bitmap->file_pages; i++) {
780 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 781 if (!bitmap->filemap) {
32a7627c
N
782 spin_unlock_irqrestore(&bitmap->lock, flags);
783 return 0;
784 }
785 page = bitmap->filemap[i];
ec7a3197
N
786 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
787 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
788 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
789 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 790 if (dirty)
32a7627c
N
791 wait = 1;
792 spin_unlock_irqrestore(&bitmap->lock, flags);
793
d785a06a 794 if (dirty | need_write)
8a5e9cf1 795 err = write_page(bitmap, page, 0);
32a7627c
N
796 }
797 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 798 if (bitmap->file)
d785a06a
N
799 wait_event(bitmap->write_wait,
800 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 801 else
a9701a30 802 md_super_wait(bitmap->mddev);
32a7627c 803 }
d785a06a
N
804 if (bitmap->flags & BITMAP_WRITE_ERROR)
805 bitmap_file_kick(bitmap);
32a7627c
N
806 return 0;
807}
808
6a07997f 809static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
810/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
811 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
812 * memory mapping of the bitmap file
813 * Special cases:
814 * if there's no bitmap file, or if the bitmap file had been
815 * previously kicked from the array, we mark all the bits as
816 * 1's in order to cause a full resync.
6a07997f
N
817 *
818 * We ignore all bits for sectors that end earlier than 'start'.
819 * This is used when reading an out-of-date bitmap...
32a7627c 820 */
6a07997f 821static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
822{
823 unsigned long i, chunks, index, oldindex, bit;
824 struct page *page = NULL, *oldpage = NULL;
825 unsigned long num_pages, bit_cnt = 0;
826 struct file *file;
d785a06a 827 unsigned long bytes, offset;
32a7627c
N
828 int outofdate;
829 int ret = -ENOSPC;
ea03aff9 830 void *paddr;
32a7627c
N
831
832 chunks = bitmap->chunks;
833 file = bitmap->file;
834
a654b9d8 835 BUG_ON(!file && !bitmap->offset);
32a7627c 836
44456d37 837#ifdef INJECT_FAULTS_3
32a7627c
N
838 outofdate = 1;
839#else
840 outofdate = bitmap->flags & BITMAP_STALE;
841#endif
842 if (outofdate)
843 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
844 "recovery\n", bmname(bitmap));
845
846 bytes = (chunks + 7) / 8;
bc7f77de 847
cdbb4cc2 848 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 849
a654b9d8 850 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
851 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
852 bmname(bitmap),
853 (unsigned long) i_size_read(file->f_mapping->host),
854 bytes + sizeof(bitmap_super_t));
855 goto out;
856 }
bc7f77de
N
857
858 ret = -ENOMEM;
859
32a7627c 860 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 861 if (!bitmap->filemap)
32a7627c 862 goto out;
32a7627c 863
e16b68b6
N
864 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
865 bitmap->filemap_attr = kzalloc(
505fa2c4 866 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 867 GFP_KERNEL);
bc7f77de 868 if (!bitmap->filemap_attr)
32a7627c 869 goto out;
32a7627c 870
32a7627c
N
871 oldindex = ~0L;
872
873 for (i = 0; i < chunks; i++) {
bd926c63 874 int b;
32a7627c
N
875 index = file_page_index(i);
876 bit = file_page_offset(i);
877 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 878 int count;
32a7627c 879 /* unmap the old page, we're done with it */
d785a06a 880 if (index == num_pages-1)
f49d5e62
N
881 count = bytes + sizeof(bitmap_super_t)
882 - index * PAGE_SIZE;
d785a06a
N
883 else
884 count = PAGE_SIZE;
32a7627c
N
885 if (index == 0) {
886 /*
887 * if we're here then the superblock page
888 * contains some bits (PAGE_SIZE != sizeof sb)
889 * we've already read it in, so just use it
890 */
891 page = bitmap->sb_page;
892 offset = sizeof(bitmap_super_t);
a654b9d8 893 } else if (file) {
d785a06a 894 page = read_page(file, index, bitmap, count);
a654b9d8
N
895 offset = 0;
896 } else {
897 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
898 offset = 0;
899 }
a654b9d8
N
900 if (IS_ERR(page)) { /* read error */
901 ret = PTR_ERR(page);
902 goto out;
903 }
904
32a7627c
N
905 oldindex = index;
906 oldpage = page;
32a7627c
N
907
908 if (outofdate) {
909 /*
910 * if bitmap is out of date, dirty the
911 * whole page and write it out
912 */
ea03aff9
N
913 paddr = kmap_atomic(page, KM_USER0);
914 memset(paddr + offset, 0xff,
6a07997f 915 PAGE_SIZE - offset);
ea03aff9 916 kunmap_atomic(paddr, KM_USER0);
77ad4bc7 917 ret = write_page(bitmap, page, 1);
32a7627c 918 if (ret) {
32a7627c 919 /* release, page not in filemap yet */
2d1f3b5d 920 put_page(page);
32a7627c
N
921 goto out;
922 }
923 }
924
925 bitmap->filemap[bitmap->file_pages++] = page;
926 }
ea03aff9 927 paddr = kmap_atomic(page, KM_USER0);
bd926c63 928 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 929 b = test_bit(bit, paddr);
bd926c63 930 else
ea03aff9
N
931 b = ext2_test_bit(bit, paddr);
932 kunmap_atomic(paddr, KM_USER0);
bd926c63 933 if (b) {
32a7627c 934 /* if the disk bit is set, set the memory bit */
6a07997f
N
935 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
936 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
937 );
32a7627c 938 bit_cnt++;
6a07997f 939 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 940 }
32a7627c
N
941 }
942
943 /* everything went OK */
944 ret = 0;
945 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
946
32a7627c
N
947 if (bit_cnt) { /* Kick recovery if any bits were set */
948 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
949 md_wakeup_thread(bitmap->mddev->thread);
950 }
951
952out:
953 printk(KERN_INFO "%s: bitmap initialized from disk: "
954 "read %lu/%lu pages, set %lu bits, status: %d\n",
955 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
956
957 return ret;
958}
959
a654b9d8
N
960void bitmap_write_all(struct bitmap *bitmap)
961{
962 /* We don't actually write all bitmap blocks here,
963 * just flag them as needing to be written
964 */
ec7a3197 965 int i;
a654b9d8 966
ec7a3197
N
967 for (i=0; i < bitmap->file_pages; i++)
968 set_page_attr(bitmap, bitmap->filemap[i],
969 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
970}
971
32a7627c
N
972
973static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
974{
975 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
976 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
977 bitmap->bp[page].count += inc;
978/*
979 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
980 (unsigned long long)offset, inc, bitmap->bp[page].count);
981*/
982 bitmap_checkfree(bitmap, page);
983}
984static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
985 sector_t offset, int *blocks,
986 int create);
987
988/*
989 * bitmap daemon -- periodically wakes up to clean bits and flush pages
990 * out to disk
991 */
992
993int bitmap_daemon_work(struct bitmap *bitmap)
994{
aa3163f8 995 unsigned long j;
32a7627c
N
996 unsigned long flags;
997 struct page *page = NULL, *lastpage = NULL;
998 int err = 0;
999 int blocks;
ea03aff9 1000 void *paddr;
32a7627c
N
1001
1002 if (bitmap == NULL)
1003 return 0;
1004 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1005 return 0;
1006 bitmap->daemon_lastrun = jiffies;
1007
1008 for (j = 0; j < bitmap->chunks; j++) {
1009 bitmap_counter_t *bmc;
1010 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1011 if (!bitmap->filemap) {
32a7627c
N
1012 /* error or shutdown */
1013 spin_unlock_irqrestore(&bitmap->lock, flags);
1014 break;
1015 }
1016
1017 page = filemap_get_page(bitmap, j);
32a7627c
N
1018
1019 if (page != lastpage) {
aa3163f8 1020 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1021 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1022 int need_write = test_page_attr(bitmap, page,
1023 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1024 if (need_write)
aa3163f8 1025 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1026
aa3163f8 1027 spin_unlock_irqrestore(&bitmap->lock, flags);
ec7a3197 1028 if (need_write) {
8a5e9cf1 1029 switch (write_page(bitmap, page, 0)) {
8a5e9cf1
N
1030 case 0:
1031 break;
1032 default:
aa3163f8 1033 bitmap_file_kick(bitmap);
8a5e9cf1 1034 }
aa3163f8
N
1035 }
1036 continue;
1037 }
1038
32a7627c 1039 /* grab the new page, sync and release the old */
32a7627c 1040 if (lastpage != NULL) {
ec7a3197 1041 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1042 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1043 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1044 err = write_page(bitmap, lastpage, 0);
32a7627c
N
1045 } else {
1046 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1047 spin_unlock_irqrestore(&bitmap->lock, flags);
1048 }
32a7627c
N
1049 if (err)
1050 bitmap_file_kick(bitmap);
1051 } else
1052 spin_unlock_irqrestore(&bitmap->lock, flags);
1053 lastpage = page;
32a7627c
N
1054/*
1055 printk("bitmap clean at page %lu\n", j);
1056*/
1057 spin_lock_irqsave(&bitmap->lock, flags);
1058 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1059 }
1060 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1061 &blocks, 0);
1062 if (bmc) {
1063/*
1064 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1065*/
1066 if (*bmc == 2) {
1067 *bmc=1; /* maybe clear the bit next time */
1068 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1069 } else if (*bmc == 1) {
1070 /* we can clear the bit */
1071 *bmc = 0;
1072 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1073 -1);
1074
1075 /* clear the bit */
ea03aff9 1076 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1077 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1078 clear_bit(file_page_offset(j), paddr);
bd926c63 1079 else
ea03aff9
N
1080 ext2_clear_bit(file_page_offset(j), paddr);
1081 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1082 }
1083 }
1084 spin_unlock_irqrestore(&bitmap->lock, flags);
1085 }
1086
1087 /* now sync the final page */
1088 if (lastpage != NULL) {
32a7627c 1089 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1090 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1091 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1092 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1093 err = write_page(bitmap, lastpage, 0);
32a7627c
N
1094 } else {
1095 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1096 spin_unlock_irqrestore(&bitmap->lock, flags);
1097 }
32a7627c
N
1098 }
1099
1100 return err;
1101}
1102
32a7627c
N
1103static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1104 sector_t offset, int *blocks,
1105 int create)
1106{
1107 /* If 'create', we might release the lock and reclaim it.
1108 * The lock must have been taken with interrupts enabled.
1109 * If !create, we don't release the lock.
1110 */
1111 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1112 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1113 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1114 sector_t csize;
1115
1116 if (bitmap_checkpage(bitmap, page, create) < 0) {
1117 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1118 *blocks = csize - (offset & (csize- 1));
1119 return NULL;
1120 }
1121 /* now locked ... */
1122
1123 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1124 /* should we use the first or second counter field
1125 * of the hijacked pointer? */
1126 int hi = (pageoff > PAGE_COUNTER_MASK);
1127 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1128 PAGE_COUNTER_SHIFT - 1);
1129 *blocks = csize - (offset & (csize- 1));
1130 return &((bitmap_counter_t *)
1131 &bitmap->bp[page].map)[hi];
1132 } else { /* page is allocated */
1133 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1134 *blocks = csize - (offset & (csize- 1));
1135 return (bitmap_counter_t *)
1136 &(bitmap->bp[page].map[pageoff]);
1137 }
1138}
1139
4b6d287f 1140int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1141{
1142 if (!bitmap) return 0;
4b6d287f
N
1143
1144 if (behind) {
1145 atomic_inc(&bitmap->behind_writes);
1146 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1147 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1148 }
1149
32a7627c
N
1150 while (sectors) {
1151 int blocks;
1152 bitmap_counter_t *bmc;
1153
1154 spin_lock_irq(&bitmap->lock);
1155 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1156 if (!bmc) {
1157 spin_unlock_irq(&bitmap->lock);
1158 return 0;
1159 }
1160
da6e1a32
NB
1161 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1162 DEFINE_WAIT(__wait);
1163 /* note that it is safe to do the prepare_to_wait
1164 * after the test as long as we do it before dropping
1165 * the spinlock.
1166 */
1167 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1168 TASK_UNINTERRUPTIBLE);
1169 spin_unlock_irq(&bitmap->lock);
1170 bitmap->mddev->queue
1171 ->unplug_fn(bitmap->mddev->queue);
1172 schedule();
1173 finish_wait(&bitmap->overflow_wait, &__wait);
1174 continue;
1175 }
1176
32a7627c
N
1177 switch(*bmc) {
1178 case 0:
1179 bitmap_file_set_bit(bitmap, offset);
1180 bitmap_count_page(bitmap,offset, 1);
1181 blk_plug_device(bitmap->mddev->queue);
1182 /* fall through */
1183 case 1:
1184 *bmc = 2;
1185 }
da6e1a32 1186
32a7627c
N
1187 (*bmc)++;
1188
1189 spin_unlock_irq(&bitmap->lock);
1190
1191 offset += blocks;
1192 if (sectors > blocks)
1193 sectors -= blocks;
1194 else sectors = 0;
1195 }
1196 return 0;
1197}
1198
1199void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1200 int success, int behind)
32a7627c
N
1201{
1202 if (!bitmap) return;
4b6d287f
N
1203 if (behind) {
1204 atomic_dec(&bitmap->behind_writes);
1205 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1206 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1207 }
1208
32a7627c
N
1209 while (sectors) {
1210 int blocks;
1211 unsigned long flags;
1212 bitmap_counter_t *bmc;
1213
1214 spin_lock_irqsave(&bitmap->lock, flags);
1215 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1216 if (!bmc) {
1217 spin_unlock_irqrestore(&bitmap->lock, flags);
1218 return;
1219 }
1220
1221 if (!success && ! (*bmc & NEEDED_MASK))
1222 *bmc |= NEEDED_MASK;
1223
da6e1a32
NB
1224 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1225 wake_up(&bitmap->overflow_wait);
1226
32a7627c
N
1227 (*bmc)--;
1228 if (*bmc <= 2) {
1229 set_page_attr(bitmap,
1230 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1231 BITMAP_PAGE_CLEAN);
1232 }
1233 spin_unlock_irqrestore(&bitmap->lock, flags);
1234 offset += blocks;
1235 if (sectors > blocks)
1236 sectors -= blocks;
1237 else sectors = 0;
1238 }
1239}
1240
6a806c51
N
1241int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1242 int degraded)
32a7627c
N
1243{
1244 bitmap_counter_t *bmc;
1245 int rv;
1246 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1247 *blocks = 1024;
1248 return 1; /* always resync if no bitmap */
1249 }
1250 spin_lock_irq(&bitmap->lock);
1251 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1252 rv = 0;
1253 if (bmc) {
1254 /* locked */
1255 if (RESYNC(*bmc))
1256 rv = 1;
1257 else if (NEEDED(*bmc)) {
1258 rv = 1;
6a806c51
N
1259 if (!degraded) { /* don't set/clear bits if degraded */
1260 *bmc |= RESYNC_MASK;
1261 *bmc &= ~NEEDED_MASK;
1262 }
32a7627c
N
1263 }
1264 }
1265 spin_unlock_irq(&bitmap->lock);
1266 return rv;
1267}
1268
1269void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1270{
1271 bitmap_counter_t *bmc;
1272 unsigned long flags;
1273/*
1274 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1275*/ if (bitmap == NULL) {
1276 *blocks = 1024;
1277 return;
1278 }
1279 spin_lock_irqsave(&bitmap->lock, flags);
1280 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1281 if (bmc == NULL)
1282 goto unlock;
1283 /* locked */
1284/*
1285 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1286*/
1287 if (RESYNC(*bmc)) {
1288 *bmc &= ~RESYNC_MASK;
1289
1290 if (!NEEDED(*bmc) && aborted)
1291 *bmc |= NEEDED_MASK;
1292 else {
1293 if (*bmc <= 2) {
1294 set_page_attr(bitmap,
1295 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1296 BITMAP_PAGE_CLEAN);
1297 }
1298 }
1299 }
1300 unlock:
1301 spin_unlock_irqrestore(&bitmap->lock, flags);
1302}
1303
1304void bitmap_close_sync(struct bitmap *bitmap)
1305{
1306 /* Sync has finished, and any bitmap chunks that weren't synced
1307 * properly have been aborted. It remains to us to clear the
1308 * RESYNC bit wherever it is still on
1309 */
1310 sector_t sector = 0;
1311 int blocks;
1312 if (!bitmap) return;
1313 while (sector < bitmap->mddev->resync_max_sectors) {
1314 bitmap_end_sync(bitmap, sector, &blocks, 0);
1315/*
1316 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1317 (unsigned long long)sector, blocks);
1318*/ sector += blocks;
1319 }
1320}
1321
6a07997f 1322static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1323{
1324 /* For each chunk covered by any of these sectors, set the
193f1c93 1325 * counter to 1 and set resync_needed. They should all
32a7627c
N
1326 * be 0 at this point
1327 */
193f1c93
N
1328
1329 int secs;
1330 bitmap_counter_t *bmc;
1331 spin_lock_irq(&bitmap->lock);
1332 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1333 if (!bmc) {
32a7627c 1334 spin_unlock_irq(&bitmap->lock);
193f1c93 1335 return;
32a7627c 1336 }
193f1c93
N
1337 if (! *bmc) {
1338 struct page *page;
6a07997f 1339 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1340 bitmap_count_page(bitmap, offset, 1);
1341 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1342 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1343 }
1344 spin_unlock_irq(&bitmap->lock);
1345
32a7627c
N
1346}
1347
9b1d1dac
PC
1348/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1349void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1350{
1351 unsigned long chunk;
1352
1353 for (chunk = s; chunk <= e; chunk++) {
1354 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1355 bitmap_set_memory_bits(bitmap, sec, 1);
1356 bitmap_file_set_bit(bitmap, sec);
1357 }
1358}
1359
6b8b3e8a
N
1360/*
1361 * flush out any pending updates
1362 */
1363void bitmap_flush(mddev_t *mddev)
1364{
1365 struct bitmap *bitmap = mddev->bitmap;
1366 int sleep;
1367
1368 if (!bitmap) /* there was no bitmap */
1369 return;
1370
1371 /* run the daemon_work three time to ensure everything is flushed
1372 * that can be
1373 */
1374 sleep = bitmap->daemon_sleep;
1375 bitmap->daemon_sleep = 0;
1376 bitmap_daemon_work(bitmap);
1377 bitmap_daemon_work(bitmap);
1378 bitmap_daemon_work(bitmap);
1379 bitmap->daemon_sleep = sleep;
1380 bitmap_update_sb(bitmap);
1381}
1382
32a7627c
N
1383/*
1384 * free memory that was allocated
1385 */
3178b0db 1386static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1387{
1388 unsigned long k, pages;
1389 struct bitmap_page *bp;
32a7627c
N
1390
1391 if (!bitmap) /* there was no bitmap */
1392 return;
1393
32a7627c
N
1394 /* release the bitmap file and kill the daemon */
1395 bitmap_file_put(bitmap);
1396
1397 bp = bitmap->bp;
1398 pages = bitmap->pages;
1399
1400 /* free all allocated memory */
1401
32a7627c
N
1402 if (bp) /* deallocate the page memory */
1403 for (k = 0; k < pages; k++)
1404 if (bp[k].map && !bp[k].hijacked)
1405 kfree(bp[k].map);
1406 kfree(bp);
1407 kfree(bitmap);
1408}
3178b0db
N
1409void bitmap_destroy(mddev_t *mddev)
1410{
1411 struct bitmap *bitmap = mddev->bitmap;
1412
1413 if (!bitmap) /* there was no bitmap */
1414 return;
1415
1416 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1417 if (mddev->thread)
1418 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1419
1420 bitmap_free(bitmap);
1421}
32a7627c
N
1422
1423/*
1424 * initialize the bitmap structure
1425 * if this returns an error, bitmap_destroy must be called to do clean up
1426 */
1427int bitmap_create(mddev_t *mddev)
1428{
1429 struct bitmap *bitmap;
1430 unsigned long blocks = mddev->resync_max_sectors;
1431 unsigned long chunks;
1432 unsigned long pages;
1433 struct file *file = mddev->bitmap_file;
1434 int err;
6a07997f 1435 sector_t start;
32a7627c 1436
5f6e3c83 1437 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1438
a654b9d8 1439 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1440 return 0;
1441
a654b9d8
N
1442 BUG_ON(file && mddev->bitmap_offset);
1443
9ffae0cf 1444 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1445 if (!bitmap)
1446 return -ENOMEM;
1447
32a7627c 1448 spin_lock_init(&bitmap->lock);
ce25c31b
N
1449 atomic_set(&bitmap->pending_writes, 0);
1450 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1451 init_waitqueue_head(&bitmap->overflow_wait);
ce25c31b 1452
32a7627c 1453 bitmap->mddev = mddev;
32a7627c 1454
32a7627c 1455 bitmap->file = file;
a654b9d8 1456 bitmap->offset = mddev->bitmap_offset;
ce25c31b
N
1457 if (file) {
1458 get_file(file);
ef51c976
MF
1459 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1460 SYNC_FILE_RANGE_WAIT_BEFORE |
1461 SYNC_FILE_RANGE_WRITE |
1462 SYNC_FILE_RANGE_WAIT_AFTER);
ce25c31b 1463 }
32a7627c
N
1464 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1465 err = bitmap_read_sb(bitmap);
1466 if (err)
3178b0db 1467 goto error;
32a7627c 1468
a638b2dc 1469 bitmap->chunkshift = ffz(~bitmap->chunksize);
32a7627c
N
1470
1471 /* now that chunksize and chunkshift are set, we can use these macros */
1472 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1473 CHUNK_BLOCK_RATIO(bitmap);
1474 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1475
1476 BUG_ON(!pages);
1477
1478 bitmap->chunks = chunks;
1479 bitmap->pages = pages;
1480 bitmap->missing_pages = pages;
1481 bitmap->counter_bits = COUNTER_BITS;
1482
1483 bitmap->syncchunk = ~0UL;
1484
44456d37 1485#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1486 bitmap->bp = NULL;
1487#else
9ffae0cf 1488 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1489#endif
3178b0db 1490 err = -ENOMEM;
32a7627c 1491 if (!bitmap->bp)
3178b0db 1492 goto error;
32a7627c 1493
32a7627c
N
1494 /* now that we have some pages available, initialize the in-memory
1495 * bitmap from the on-disk bitmap */
6a07997f
N
1496 start = 0;
1497 if (mddev->degraded == 0
1498 || bitmap->events_cleared == mddev->events)
1499 /* no need to keep dirty bits to optimise a re-add of a missing device */
1500 start = mddev->recovery_cp;
1501 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1502
32a7627c 1503 if (err)
3178b0db 1504 goto error;
32a7627c
N
1505
1506 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1507 pages, bmname(bitmap));
1508
3178b0db
N
1509 mddev->bitmap = bitmap;
1510
b15c2e57
N
1511 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1512
32a7627c 1513 return bitmap_update_sb(bitmap);
3178b0db
N
1514
1515 error:
1516 bitmap_free(bitmap);
1517 return err;
32a7627c
N
1518}
1519
1520/* the bitmap API -- for raid personalities */
1521EXPORT_SYMBOL(bitmap_startwrite);
1522EXPORT_SYMBOL(bitmap_endwrite);
1523EXPORT_SYMBOL(bitmap_start_sync);
1524EXPORT_SYMBOL(bitmap_end_sync);
1525EXPORT_SYMBOL(bitmap_unplug);
1526EXPORT_SYMBOL(bitmap_close_sync);
This page took 0.308433 seconds and 5 git commands to generate.