[PATCH] md/bitmap: change md/bitmap file handling to use bmap to file blocks
[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>
23#include <linux/config.h>
24#include <linux/timer.h>
25#include <linux/sched.h>
26#include <linux/list.h>
27#include <linux/file.h>
28#include <linux/mount.h>
29#include <linux/buffer_head.h>
30#include <linux/raid/md.h>
31#include <linux/raid/bitmap.h>
32
33/* debug macros */
34
35#define DEBUG 0
36
37#if DEBUG
38/* these are for debugging purposes only! */
39
40/* define one and only one of these */
41#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
42#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
43#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
44#define INJECT_FAULTS_4 0 /* undef */
45#define INJECT_FAULTS_5 0 /* undef */
46#define INJECT_FAULTS_6 0
47
48/* if these are defined, the driver will fail! debug only */
49#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
50#define INJECT_FATAL_FAULT_2 0 /* undef */
51#define INJECT_FATAL_FAULT_3 0 /* undef */
52#endif
53
54//#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
55#define DPRINTK(x...) do { } while(0)
56
57#ifndef PRINTK
58# if DEBUG > 0
59# define PRINTK(x...) printk(KERN_DEBUG x)
60# else
61# define PRINTK(x...)
62# endif
63#endif
64
65static inline char * bmname(struct bitmap *bitmap)
66{
67 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
68}
69
32a7627c
N
70
71/*
72 * just a placeholder - calls kmalloc for bitmap pages
73 */
74static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
75{
76 unsigned char *page;
77
44456d37 78#ifdef INJECT_FAULTS_1
32a7627c
N
79 page = NULL;
80#else
81 page = kmalloc(PAGE_SIZE, GFP_NOIO);
82#endif
83 if (!page)
84 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
85 else
a654b9d8 86 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
87 bmname(bitmap), page);
88 return page;
89}
90
91/*
92 * for now just a placeholder -- just calls kfree for bitmap pages
93 */
94static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
95{
96 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
97 kfree(page);
98}
99
100/*
101 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
102 *
103 * 1) check to see if this page is allocated, if it's not then try to alloc
104 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
105 * page pointer directly as a counter
106 *
107 * if we find our page, we increment the page's refcount so that it stays
108 * allocated while we're using it
109 */
110static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
111{
112 unsigned char *mappage;
113
114 if (page >= bitmap->pages) {
115 printk(KERN_ALERT
116 "%s: invalid bitmap page request: %lu (> %lu)\n",
117 bmname(bitmap), page, bitmap->pages-1);
118 return -EINVAL;
119 }
120
121
122 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
123 return 0;
124
125 if (bitmap->bp[page].map) /* page is already allocated, just return */
126 return 0;
127
128 if (!create)
129 return -ENOENT;
130
131 spin_unlock_irq(&bitmap->lock);
132
133 /* this page has not been allocated yet */
134
135 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
136 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
137 bmname(bitmap));
138 /* failed - set the hijacked flag so that we can use the
139 * pointer as a counter */
140 spin_lock_irq(&bitmap->lock);
141 if (!bitmap->bp[page].map)
142 bitmap->bp[page].hijacked = 1;
143 goto out;
144 }
145
146 /* got a page */
147
148 spin_lock_irq(&bitmap->lock);
149
150 /* recheck the page */
151
152 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
153 /* somebody beat us to getting the page */
154 bitmap_free_page(bitmap, mappage);
155 return 0;
156 }
157
158 /* no page was in place and we have one, so install it */
159
160 memset(mappage, 0, PAGE_SIZE);
161 bitmap->bp[page].map = mappage;
162 bitmap->missing_pages--;
163out:
164 return 0;
165}
166
167
168/* if page is completely empty, put it back on the free list, or dealloc it */
169/* if page was hijacked, unmark the flag so it might get alloced next time */
170/* Note: lock should be held when calling this */
858119e1 171static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
172{
173 char *ptr;
174
175 if (bitmap->bp[page].count) /* page is still busy */
176 return;
177
178 /* page is no longer in use, it can be released */
179
180 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
181 bitmap->bp[page].hijacked = 0;
182 bitmap->bp[page].map = NULL;
183 return;
184 }
185
186 /* normal case, free the page */
187
188#if 0
189/* actually ... let's not. We will probably need the page again exactly when
190 * memory is tight and we are flusing to disk
191 */
192 return;
193#else
194 ptr = bitmap->bp[page].map;
195 bitmap->bp[page].map = NULL;
196 bitmap->missing_pages++;
197 bitmap_free_page(bitmap, ptr);
198 return;
199#endif
200}
201
202
203/*
204 * bitmap file handling - read and write the bitmap file and its superblock
205 */
206
207/* copy the pathname of a file to a buffer */
208char *file_path(struct file *file, char *buf, int count)
209{
210 struct dentry *d;
211 struct vfsmount *v;
212
213 if (!buf)
214 return NULL;
215
216 d = file->f_dentry;
217 v = file->f_vfsmnt;
218
219 buf = d_path(d, v, buf, count);
220
221 return IS_ERR(buf) ? NULL : buf;
222}
223
224/*
225 * basic page I/O operations
226 */
227
a654b9d8
N
228/* IO operations when bitmap is stored near all superblocks */
229static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
230{
231 /* choose a good rdev and read the page from there */
232
233 mdk_rdev_t *rdev;
234 struct list_head *tmp;
235 struct page *page = alloc_page(GFP_KERNEL);
236 sector_t target;
237
238 if (!page)
239 return ERR_PTR(-ENOMEM);
a654b9d8 240
ab904d63 241 ITERATE_RDEV(mddev, rdev, tmp) {
b2d444d7
N
242 if (! test_bit(In_sync, &rdev->flags)
243 || test_bit(Faulty, &rdev->flags))
ab904d63
N
244 continue;
245
a654b9d8
N
246 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
247
ab904d63
N
248 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
249 page->index = index;
250 return page;
251 }
252 }
253 return ERR_PTR(-EIO);
a654b9d8 254
a654b9d8
N
255}
256
257static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
258{
259 mdk_rdev_t *rdev;
260 struct list_head *tmp;
261
262 ITERATE_RDEV(mddev, rdev, tmp)
b2d444d7
N
263 if (test_bit(In_sync, &rdev->flags)
264 && !test_bit(Faulty, &rdev->flags))
a654b9d8
N
265 md_super_write(mddev, rdev,
266 (rdev->sb_offset<<1) + offset
267 + page->index * (PAGE_SIZE/512),
268 PAGE_SIZE,
269 page);
270
271 if (wait)
a9701a30 272 md_super_wait(mddev);
a654b9d8
N
273 return 0;
274}
275
32a7627c 276/*
a654b9d8 277 * write out a page to a file
32a7627c 278 */
77ad4bc7 279static int write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 280{
d785a06a 281 struct buffer_head *bh;
32a7627c 282
a654b9d8
N
283 if (bitmap->file == NULL)
284 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
285
d785a06a 286 bh = page_buffers(page);
c708443c 287
d785a06a
N
288 while (bh && bh->b_blocknr) {
289 atomic_inc(&bitmap->pending_writes);
290 set_buffer_locked(bh);
291 set_buffer_mapped(bh);
292 submit_bh(WRITE, bh);
293 bh = bh->b_this_page;
294 }
295
296 if (wait) {
297 wait_event(bitmap->write_wait,
298 atomic_read(&bitmap->pending_writes)==0);
299 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
8a5e9cf1 300 }
d785a06a
N
301 return 0;
302}
303
304static void end_bitmap_write(struct buffer_head *bh, int uptodate)
305{
306 struct bitmap *bitmap = bh->b_private;
307 unsigned long flags;
32a7627c 308
d785a06a
N
309 if (!uptodate) {
310 spin_lock_irqsave(&bitmap->lock, flags);
311 bitmap->flags |= BITMAP_WRITE_ERROR;
312 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 313 }
d785a06a
N
314 if (atomic_dec_and_test(&bitmap->pending_writes))
315 wake_up(&bitmap->write_wait);
316}
32a7627c 317
d785a06a
N
318/* copied from buffer.c */
319static void
320__clear_page_buffers(struct page *page)
321{
322 ClearPagePrivate(page);
323 set_page_private(page, 0);
324 page_cache_release(page);
325}
326static void free_buffers(struct page *page)
327{
328 struct buffer_head *bh = page_buffers(page);
77ad4bc7 329
d785a06a
N
330 while (bh) {
331 struct buffer_head *next = bh->b_this_page;
332 free_buffer_head(bh);
333 bh = next;
77ad4bc7 334 }
d785a06a
N
335 __clear_page_buffers(page);
336 put_page(page);
32a7627c
N
337}
338
d785a06a
N
339/* read a page from a file.
340 * We both read the page, and attach buffers to the page to record the
341 * address of each block (using bmap). These addresses will be used
342 * to write the block later, completely bypassing the filesystem.
343 * This usage is similar to how swap files are handled, and allows us
344 * to write to a file with no concerns of memory allocation failing.
345 */
32a7627c 346static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
347 struct bitmap *bitmap,
348 unsigned long count)
32a7627c 349{
32a7627c 350 struct page *page = NULL;
d785a06a
N
351 struct inode *inode = file->f_dentry->d_inode;
352 loff_t pos = index << PAGE_SHIFT;
353 int ret;
354 struct buffer_head *bh;
355 sector_t block;
356 mm_segment_t oldfs;
32a7627c 357
2d1f3b5d
N
358 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
359 (unsigned long long)index << PAGE_SHIFT);
32a7627c 360
d785a06a
N
361 page = alloc_page(GFP_KERNEL);
362 if (!page)
363 page = ERR_PTR(-ENOMEM);
32a7627c
N
364 if (IS_ERR(page))
365 goto out;
d785a06a
N
366
367 oldfs = get_fs();
368 set_fs(KERNEL_DS);
369 ret = vfs_read(file, (char __user*) page_address(page), count, &pos);
370 set_fs(oldfs);
371
372 if (ret >= 0 && ret != count)
373 ret = -EIO;
374 if (ret < 0) {
375 put_page(page);
376 page = ERR_PTR(ret);
377 goto out;
378 }
379 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
380 if (!bh) {
2d1f3b5d 381 put_page(page);
d785a06a 382 page = ERR_PTR(-ENOMEM);
32a7627c
N
383 goto out;
384 }
d785a06a
N
385 attach_page_buffers(page, bh);
386 block = index << (PAGE_SHIFT - inode->i_blkbits);
387 while (bh) {
388 if (count == 0)
389 bh->b_blocknr = 0;
390 else {
391 bh->b_blocknr = bmap(inode, block);
392 if (bh->b_blocknr == 0) {
393 /* Cannot use this file! */
394 free_buffers(page);
395 page = ERR_PTR(-EINVAL);
396 goto out;
397 }
398 bh->b_bdev = inode->i_sb->s_bdev;
399 if (count < (1<<inode->i_blkbits))
400 count = 0;
401 else
402 count -= (1<<inode->i_blkbits);
403
404 bh->b_end_io = end_bitmap_write;
405 bh->b_private = bitmap;
406 }
407 block++;
408 bh = bh->b_this_page;
409 }
32a7627c 410
d785a06a 411 page->index = index;
32a7627c
N
412out:
413 if (IS_ERR(page))
414 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
415 (int)PAGE_SIZE,
416 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
417 PTR_ERR(page));
418 return page;
419}
420
421/*
422 * bitmap file superblock operations
423 */
424
425/* update the event counter and sync the superblock to disk */
426int bitmap_update_sb(struct bitmap *bitmap)
427{
428 bitmap_super_t *sb;
429 unsigned long flags;
430
431 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
432 return 0;
433 spin_lock_irqsave(&bitmap->lock, flags);
434 if (!bitmap->sb_page) { /* no superblock */
435 spin_unlock_irqrestore(&bitmap->lock, flags);
436 return 0;
437 }
32a7627c 438 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 439 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
440 sb->events = cpu_to_le64(bitmap->mddev->events);
441 if (!bitmap->mddev->degraded)
442 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
ea03aff9 443 kunmap_atomic(sb, KM_USER0);
8a5e9cf1 444 return write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
445}
446
447/* print out the bitmap file superblock */
448void bitmap_print_sb(struct bitmap *bitmap)
449{
450 bitmap_super_t *sb;
451
452 if (!bitmap || !bitmap->sb_page)
453 return;
ea03aff9 454 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 455 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
456 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
457 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
458 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
459 *(__u32 *)(sb->uuid+0),
460 *(__u32 *)(sb->uuid+4),
461 *(__u32 *)(sb->uuid+8),
462 *(__u32 *)(sb->uuid+12));
a2cff26a 463 printk(KERN_DEBUG " events: %llu\n",
32a7627c 464 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 465 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 466 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
467 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
468 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
469 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
470 printk(KERN_DEBUG " sync size: %llu KB\n",
471 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 472 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 473 kunmap_atomic(sb, KM_USER0);
32a7627c
N
474}
475
476/* read the superblock from the bitmap file and initialize some bitmap fields */
477static int bitmap_read_sb(struct bitmap *bitmap)
478{
479 char *reason = NULL;
480 bitmap_super_t *sb;
4b6d287f 481 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
482 unsigned long long events;
483 int err = -EINVAL;
484
485 /* page 0 is the superblock, read it... */
a654b9d8 486 if (bitmap->file)
d785a06a 487 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, PAGE_SIZE);
a654b9d8
N
488 else {
489 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 490 }
32a7627c
N
491 if (IS_ERR(bitmap->sb_page)) {
492 err = PTR_ERR(bitmap->sb_page);
493 bitmap->sb_page = NULL;
494 return err;
495 }
496
ea03aff9 497 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 498
32a7627c
N
499 chunksize = le32_to_cpu(sb->chunksize);
500 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 501 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
502
503 /* verify that the bitmap-specific fields are valid */
504 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
505 reason = "bad magic";
bd926c63
N
506 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
507 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 508 reason = "unrecognized superblock version";
7dd5d34c
N
509 else if (chunksize < PAGE_SIZE)
510 reason = "bitmap chunksize too small";
32a7627c
N
511 else if ((1 << ffz(~chunksize)) != chunksize)
512 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
513 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
514 reason = "daemon sleep period out of range";
4b6d287f
N
515 else if (write_behind > COUNTER_MAX)
516 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
517 if (reason) {
518 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
519 bmname(bitmap), reason);
520 goto out;
521 }
522
523 /* keep the array size field of the bitmap superblock up to date */
524 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
525
526 if (!bitmap->mddev->persistent)
527 goto success;
528
529 /*
530 * if we have a persistent array superblock, compare the
531 * bitmap's UUID and event counter to the mddev's
532 */
533 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
534 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
535 bmname(bitmap));
536 goto out;
537 }
538 events = le64_to_cpu(sb->events);
539 if (events < bitmap->mddev->events) {
540 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
541 "-- forcing full recovery\n", bmname(bitmap), events,
542 (unsigned long long) bitmap->mddev->events);
543 sb->state |= BITMAP_STALE;
544 }
545success:
546 /* assign fields using values from superblock */
547 bitmap->chunksize = chunksize;
548 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 549 bitmap->daemon_lastrun = jiffies;
4b6d287f 550 bitmap->max_write_behind = write_behind;
32a7627c 551 bitmap->flags |= sb->state;
bd926c63
N
552 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
553 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 554 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
6a07997f
N
555 if (sb->state & BITMAP_STALE)
556 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
557 err = 0;
558out:
ea03aff9 559 kunmap_atomic(sb, KM_USER0);
32a7627c
N
560 if (err)
561 bitmap_print_sb(bitmap);
562 return err;
563}
564
565enum bitmap_mask_op {
566 MASK_SET,
567 MASK_UNSET
568};
569
570/* record the state of the bitmap in the superblock */
571static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
572 enum bitmap_mask_op op)
573{
574 bitmap_super_t *sb;
575 unsigned long flags;
576
577 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 578 if (!bitmap->sb_page) { /* can't set the state */
32a7627c
N
579 spin_unlock_irqrestore(&bitmap->lock, flags);
580 return;
581 }
32a7627c 582 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 583 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
584 switch (op) {
585 case MASK_SET: sb->state |= bits;
586 break;
587 case MASK_UNSET: sb->state &= ~bits;
588 break;
589 default: BUG();
590 }
ea03aff9 591 kunmap_atomic(sb, KM_USER0);
32a7627c
N
592}
593
594/*
595 * general bitmap file operations
596 */
597
598/* calculate the index of the page that contains this bit */
599static inline unsigned long file_page_index(unsigned long chunk)
600{
601 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
602}
603
604/* calculate the (bit) offset of this bit within a page */
605static inline unsigned long file_page_offset(unsigned long chunk)
606{
607 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
608}
609
610/*
611 * return a pointer to the page in the filemap that contains the given bit
612 *
613 * this lookup is complicated by the fact that the bitmap sb might be exactly
614 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
615 * 0 or page 1
616 */
617static inline struct page *filemap_get_page(struct bitmap *bitmap,
618 unsigned long chunk)
619{
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
N
667 if (file) {
668 struct inode *inode = file->f_dentry->d_inode;
669 invalidate_inode_pages(inode->i_mapping);
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);
746 bit = file_page_offset(chunk);
747
32a7627c
N
748 /* set the bit */
749 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
750 if (bitmap->flags & BITMAP_HOSTENDIAN)
751 set_bit(bit, kaddr);
752 else
753 ext2_set_bit(bit, kaddr);
32a7627c
N
754 kunmap_atomic(kaddr, KM_USER0);
755 PRINTK("set file bit %lu page %lu\n", bit, page->index);
756
757 /* record page number so it gets flushed to disk when unplug occurs */
758 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
759
760}
761
762/* this gets called when the md device is ready to unplug its underlying
763 * (slave) device queues -- before we let any writes go down, we need to
764 * sync the dirty pages of the bitmap file to disk */
765int bitmap_unplug(struct bitmap *bitmap)
766{
ec7a3197
N
767 unsigned long i, flags;
768 int dirty, need_write;
32a7627c
N
769 struct page *page;
770 int wait = 0;
8a5e9cf1 771 int err;
32a7627c
N
772
773 if (!bitmap)
774 return 0;
775
776 /* look at each page to see if there are any set bits that need to be
777 * flushed out to disk */
778 for (i = 0; i < bitmap->file_pages; i++) {
779 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 780 if (!bitmap->filemap) {
32a7627c
N
781 spin_unlock_irqrestore(&bitmap->lock, flags);
782 return 0;
783 }
784 page = bitmap->filemap[i];
ec7a3197
N
785 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
786 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
787 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
788 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 789 if (dirty)
32a7627c
N
790 wait = 1;
791 spin_unlock_irqrestore(&bitmap->lock, flags);
792
d785a06a 793 if (dirty | need_write)
8a5e9cf1 794 err = write_page(bitmap, page, 0);
32a7627c
N
795 }
796 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 797 if (bitmap->file)
d785a06a
N
798 wait_event(bitmap->write_wait,
799 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 800 else
a9701a30 801 md_super_wait(bitmap->mddev);
32a7627c 802 }
d785a06a
N
803 if (bitmap->flags & BITMAP_WRITE_ERROR)
804 bitmap_file_kick(bitmap);
32a7627c
N
805 return 0;
806}
807
6a07997f 808static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
809/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
810 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
811 * memory mapping of the bitmap file
812 * Special cases:
813 * if there's no bitmap file, or if the bitmap file had been
814 * previously kicked from the array, we mark all the bits as
815 * 1's in order to cause a full resync.
6a07997f
N
816 *
817 * We ignore all bits for sectors that end earlier than 'start'.
818 * This is used when reading an out-of-date bitmap...
32a7627c 819 */
6a07997f 820static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
821{
822 unsigned long i, chunks, index, oldindex, bit;
823 struct page *page = NULL, *oldpage = NULL;
824 unsigned long num_pages, bit_cnt = 0;
825 struct file *file;
d785a06a 826 unsigned long bytes, offset;
32a7627c
N
827 int outofdate;
828 int ret = -ENOSPC;
ea03aff9 829 void *paddr;
32a7627c
N
830
831 chunks = bitmap->chunks;
832 file = bitmap->file;
833
a654b9d8 834 BUG_ON(!file && !bitmap->offset);
32a7627c 835
44456d37 836#ifdef INJECT_FAULTS_3
32a7627c
N
837 outofdate = 1;
838#else
839 outofdate = bitmap->flags & BITMAP_STALE;
840#endif
841 if (outofdate)
842 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
843 "recovery\n", bmname(bitmap));
844
845 bytes = (chunks + 7) / 8;
bc7f77de 846
cdbb4cc2 847 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 848
a654b9d8 849 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
850 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
851 bmname(bitmap),
852 (unsigned long) i_size_read(file->f_mapping->host),
853 bytes + sizeof(bitmap_super_t));
854 goto out;
855 }
bc7f77de
N
856
857 ret = -ENOMEM;
858
32a7627c 859 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 860 if (!bitmap->filemap)
32a7627c 861 goto out;
32a7627c 862
e16b68b6
N
863 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
864 bitmap->filemap_attr = kzalloc(
865 (((num_pages*4/8)+sizeof(unsigned long)-1)
866 /sizeof(unsigned long))
867 *sizeof(unsigned long),
868 GFP_KERNEL);
bc7f77de 869 if (!bitmap->filemap_attr)
32a7627c 870 goto out;
32a7627c 871
32a7627c
N
872 oldindex = ~0L;
873
874 for (i = 0; i < chunks; i++) {
bd926c63 875 int b;
32a7627c
N
876 index = file_page_index(i);
877 bit = file_page_offset(i);
878 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 879 int count;
32a7627c 880 /* unmap the old page, we're done with it */
d785a06a
N
881 if (index == num_pages-1)
882 count = bytes - index * PAGE_SIZE;
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
1161 switch(*bmc) {
1162 case 0:
1163 bitmap_file_set_bit(bitmap, offset);
1164 bitmap_count_page(bitmap,offset, 1);
1165 blk_plug_device(bitmap->mddev->queue);
1166 /* fall through */
1167 case 1:
1168 *bmc = 2;
1169 }
5daf2cf1 1170 BUG_ON((*bmc & COUNTER_MAX) == COUNTER_MAX);
32a7627c
N
1171 (*bmc)++;
1172
1173 spin_unlock_irq(&bitmap->lock);
1174
1175 offset += blocks;
1176 if (sectors > blocks)
1177 sectors -= blocks;
1178 else sectors = 0;
1179 }
1180 return 0;
1181}
1182
1183void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1184 int success, int behind)
32a7627c
N
1185{
1186 if (!bitmap) return;
4b6d287f
N
1187 if (behind) {
1188 atomic_dec(&bitmap->behind_writes);
1189 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1190 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1191 }
1192
32a7627c
N
1193 while (sectors) {
1194 int blocks;
1195 unsigned long flags;
1196 bitmap_counter_t *bmc;
1197
1198 spin_lock_irqsave(&bitmap->lock, flags);
1199 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1200 if (!bmc) {
1201 spin_unlock_irqrestore(&bitmap->lock, flags);
1202 return;
1203 }
1204
1205 if (!success && ! (*bmc & NEEDED_MASK))
1206 *bmc |= NEEDED_MASK;
1207
1208 (*bmc)--;
1209 if (*bmc <= 2) {
1210 set_page_attr(bitmap,
1211 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1212 BITMAP_PAGE_CLEAN);
1213 }
1214 spin_unlock_irqrestore(&bitmap->lock, flags);
1215 offset += blocks;
1216 if (sectors > blocks)
1217 sectors -= blocks;
1218 else sectors = 0;
1219 }
1220}
1221
6a806c51
N
1222int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1223 int degraded)
32a7627c
N
1224{
1225 bitmap_counter_t *bmc;
1226 int rv;
1227 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1228 *blocks = 1024;
1229 return 1; /* always resync if no bitmap */
1230 }
1231 spin_lock_irq(&bitmap->lock);
1232 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1233 rv = 0;
1234 if (bmc) {
1235 /* locked */
1236 if (RESYNC(*bmc))
1237 rv = 1;
1238 else if (NEEDED(*bmc)) {
1239 rv = 1;
6a806c51
N
1240 if (!degraded) { /* don't set/clear bits if degraded */
1241 *bmc |= RESYNC_MASK;
1242 *bmc &= ~NEEDED_MASK;
1243 }
32a7627c
N
1244 }
1245 }
1246 spin_unlock_irq(&bitmap->lock);
1247 return rv;
1248}
1249
1250void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1251{
1252 bitmap_counter_t *bmc;
1253 unsigned long flags;
1254/*
1255 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1256*/ if (bitmap == NULL) {
1257 *blocks = 1024;
1258 return;
1259 }
1260 spin_lock_irqsave(&bitmap->lock, flags);
1261 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1262 if (bmc == NULL)
1263 goto unlock;
1264 /* locked */
1265/*
1266 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1267*/
1268 if (RESYNC(*bmc)) {
1269 *bmc &= ~RESYNC_MASK;
1270
1271 if (!NEEDED(*bmc) && aborted)
1272 *bmc |= NEEDED_MASK;
1273 else {
1274 if (*bmc <= 2) {
1275 set_page_attr(bitmap,
1276 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1277 BITMAP_PAGE_CLEAN);
1278 }
1279 }
1280 }
1281 unlock:
1282 spin_unlock_irqrestore(&bitmap->lock, flags);
1283}
1284
1285void bitmap_close_sync(struct bitmap *bitmap)
1286{
1287 /* Sync has finished, and any bitmap chunks that weren't synced
1288 * properly have been aborted. It remains to us to clear the
1289 * RESYNC bit wherever it is still on
1290 */
1291 sector_t sector = 0;
1292 int blocks;
1293 if (!bitmap) return;
1294 while (sector < bitmap->mddev->resync_max_sectors) {
1295 bitmap_end_sync(bitmap, sector, &blocks, 0);
1296/*
1297 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1298 (unsigned long long)sector, blocks);
1299*/ sector += blocks;
1300 }
1301}
1302
6a07997f 1303static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1304{
1305 /* For each chunk covered by any of these sectors, set the
193f1c93 1306 * counter to 1 and set resync_needed. They should all
32a7627c
N
1307 * be 0 at this point
1308 */
193f1c93
N
1309
1310 int secs;
1311 bitmap_counter_t *bmc;
1312 spin_lock_irq(&bitmap->lock);
1313 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1314 if (!bmc) {
32a7627c 1315 spin_unlock_irq(&bitmap->lock);
193f1c93 1316 return;
32a7627c 1317 }
193f1c93
N
1318 if (! *bmc) {
1319 struct page *page;
6a07997f 1320 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1321 bitmap_count_page(bitmap, offset, 1);
1322 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1323 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1324 }
1325 spin_unlock_irq(&bitmap->lock);
1326
32a7627c
N
1327}
1328
6b8b3e8a
N
1329/*
1330 * flush out any pending updates
1331 */
1332void bitmap_flush(mddev_t *mddev)
1333{
1334 struct bitmap *bitmap = mddev->bitmap;
1335 int sleep;
1336
1337 if (!bitmap) /* there was no bitmap */
1338 return;
1339
1340 /* run the daemon_work three time to ensure everything is flushed
1341 * that can be
1342 */
1343 sleep = bitmap->daemon_sleep;
1344 bitmap->daemon_sleep = 0;
1345 bitmap_daemon_work(bitmap);
1346 bitmap_daemon_work(bitmap);
1347 bitmap_daemon_work(bitmap);
1348 bitmap->daemon_sleep = sleep;
1349 bitmap_update_sb(bitmap);
1350}
1351
32a7627c
N
1352/*
1353 * free memory that was allocated
1354 */
3178b0db 1355static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1356{
1357 unsigned long k, pages;
1358 struct bitmap_page *bp;
32a7627c
N
1359
1360 if (!bitmap) /* there was no bitmap */
1361 return;
1362
32a7627c
N
1363 /* release the bitmap file and kill the daemon */
1364 bitmap_file_put(bitmap);
1365
1366 bp = bitmap->bp;
1367 pages = bitmap->pages;
1368
1369 /* free all allocated memory */
1370
32a7627c
N
1371 if (bp) /* deallocate the page memory */
1372 for (k = 0; k < pages; k++)
1373 if (bp[k].map && !bp[k].hijacked)
1374 kfree(bp[k].map);
1375 kfree(bp);
1376 kfree(bitmap);
1377}
3178b0db
N
1378void bitmap_destroy(mddev_t *mddev)
1379{
1380 struct bitmap *bitmap = mddev->bitmap;
1381
1382 if (!bitmap) /* there was no bitmap */
1383 return;
1384
1385 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1386 if (mddev->thread)
1387 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1388
1389 bitmap_free(bitmap);
1390}
32a7627c
N
1391
1392/*
1393 * initialize the bitmap structure
1394 * if this returns an error, bitmap_destroy must be called to do clean up
1395 */
1396int bitmap_create(mddev_t *mddev)
1397{
1398 struct bitmap *bitmap;
1399 unsigned long blocks = mddev->resync_max_sectors;
1400 unsigned long chunks;
1401 unsigned long pages;
1402 struct file *file = mddev->bitmap_file;
1403 int err;
6a07997f 1404 sector_t start;
32a7627c
N
1405
1406 BUG_ON(sizeof(bitmap_super_t) != 256);
1407
a654b9d8 1408 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1409 return 0;
1410
a654b9d8
N
1411 BUG_ON(file && mddev->bitmap_offset);
1412
9ffae0cf 1413 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1414 if (!bitmap)
1415 return -ENOMEM;
1416
32a7627c
N
1417 spin_lock_init(&bitmap->lock);
1418 bitmap->mddev = mddev;
32a7627c 1419
32a7627c 1420 bitmap->file = file;
a654b9d8
N
1421 bitmap->offset = mddev->bitmap_offset;
1422 if (file) get_file(file);
d785a06a
N
1423
1424 /* Ensure we read fresh data */
1425 invalidate_inode_pages(file->f_dentry->d_inode->i_mapping);
1426
32a7627c
N
1427 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1428 err = bitmap_read_sb(bitmap);
1429 if (err)
3178b0db 1430 goto error;
32a7627c
N
1431
1432 bitmap->chunkshift = find_first_bit(&bitmap->chunksize,
1433 sizeof(bitmap->chunksize));
1434
1435 /* now that chunksize and chunkshift are set, we can use these macros */
1436 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1437 CHUNK_BLOCK_RATIO(bitmap);
1438 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1439
1440 BUG_ON(!pages);
1441
1442 bitmap->chunks = chunks;
1443 bitmap->pages = pages;
1444 bitmap->missing_pages = pages;
1445 bitmap->counter_bits = COUNTER_BITS;
1446
1447 bitmap->syncchunk = ~0UL;
1448
d785a06a
N
1449 atomic_set(&bitmap->pending_writes, 0);
1450 init_waitqueue_head(&bitmap->write_wait);
1451
44456d37 1452#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1453 bitmap->bp = NULL;
1454#else
9ffae0cf 1455 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1456#endif
3178b0db 1457 err = -ENOMEM;
32a7627c 1458 if (!bitmap->bp)
3178b0db 1459 goto error;
32a7627c 1460
32a7627c
N
1461 /* now that we have some pages available, initialize the in-memory
1462 * bitmap from the on-disk bitmap */
6a07997f
N
1463 start = 0;
1464 if (mddev->degraded == 0
1465 || bitmap->events_cleared == mddev->events)
1466 /* no need to keep dirty bits to optimise a re-add of a missing device */
1467 start = mddev->recovery_cp;
1468 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1469
32a7627c 1470 if (err)
3178b0db 1471 goto error;
32a7627c
N
1472
1473 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1474 pages, bmname(bitmap));
1475
3178b0db
N
1476 mddev->bitmap = bitmap;
1477
b15c2e57
N
1478 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1479
32a7627c 1480 return bitmap_update_sb(bitmap);
3178b0db
N
1481
1482 error:
1483 bitmap_free(bitmap);
1484 return err;
32a7627c
N
1485}
1486
1487/* the bitmap API -- for raid personalities */
1488EXPORT_SYMBOL(bitmap_startwrite);
1489EXPORT_SYMBOL(bitmap_endwrite);
1490EXPORT_SYMBOL(bitmap_start_sync);
1491EXPORT_SYMBOL(bitmap_end_sync);
1492EXPORT_SYMBOL(bitmap_unplug);
1493EXPORT_SYMBOL(bitmap_close_sync);
This page took 0.209915 seconds and 5 git commands to generate.