]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/md/bitmap.c
use bio_endio instead of a call to bi_end_io
[net-next-2.6.git] / drivers / md / bitmap.c
CommitLineData
32a7627c
N
1/*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
32a7627c
N
10 */
11
12/*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
16 * wait if count gets too high, wake when it drops to half.
32a7627c
N
17 */
18
19#include <linux/module.h>
32a7627c
N
20#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/init.h>
32a7627c
N
23#include <linux/timer.h>
24#include <linux/sched.h>
25#include <linux/list.h>
26#include <linux/file.h>
27#include <linux/mount.h>
28#include <linux/buffer_head.h>
29#include <linux/raid/md.h>
30#include <linux/raid/bitmap.h>
31
32/* debug macros */
33
34#define DEBUG 0
35
36#if DEBUG
37/* these are for debugging purposes only! */
38
39/* define one and only one of these */
40#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
41#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
42#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
43#define INJECT_FAULTS_4 0 /* undef */
44#define INJECT_FAULTS_5 0 /* undef */
45#define INJECT_FAULTS_6 0
46
47/* if these are defined, the driver will fail! debug only */
48#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
49#define INJECT_FATAL_FAULT_2 0 /* undef */
50#define INJECT_FATAL_FAULT_3 0 /* undef */
51#endif
52
53//#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
54#define DPRINTK(x...) do { } while(0)
55
56#ifndef PRINTK
57# if DEBUG > 0
58# define PRINTK(x...) printk(KERN_DEBUG x)
59# else
60# define PRINTK(x...)
61# endif
62#endif
63
64static inline char * bmname(struct bitmap *bitmap)
65{
66 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
67}
68
32a7627c
N
69
70/*
71 * just a placeholder - calls kmalloc for bitmap pages
72 */
73static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
74{
75 unsigned char *page;
76
44456d37 77#ifdef INJECT_FAULTS_1
32a7627c
N
78 page = NULL;
79#else
80 page = kmalloc(PAGE_SIZE, GFP_NOIO);
81#endif
82 if (!page)
83 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
84 else
a654b9d8 85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
86 bmname(bitmap), page);
87 return page;
88}
89
90/*
91 * for now just a placeholder -- just calls kfree for bitmap pages
92 */
93static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
94{
95 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
96 kfree(page);
97}
98
99/*
100 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
101 *
102 * 1) check to see if this page is allocated, if it's not then try to alloc
103 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
104 * page pointer directly as a counter
105 *
106 * if we find our page, we increment the page's refcount so that it stays
107 * allocated while we're using it
108 */
109static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
110{
111 unsigned char *mappage;
112
113 if (page >= bitmap->pages) {
114 printk(KERN_ALERT
115 "%s: invalid bitmap page request: %lu (> %lu)\n",
116 bmname(bitmap), page, bitmap->pages-1);
117 return -EINVAL;
118 }
119
120
121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
122 return 0;
123
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
125 return 0;
126
127 if (!create)
128 return -ENOENT;
129
130 spin_unlock_irq(&bitmap->lock);
131
132 /* this page has not been allocated yet */
133
134 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
135 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
136 bmname(bitmap));
137 /* failed - set the hijacked flag so that we can use the
138 * pointer as a counter */
139 spin_lock_irq(&bitmap->lock);
140 if (!bitmap->bp[page].map)
141 bitmap->bp[page].hijacked = 1;
142 goto out;
143 }
144
145 /* got a page */
146
147 spin_lock_irq(&bitmap->lock);
148
149 /* recheck the page */
150
151 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
152 /* somebody beat us to getting the page */
153 bitmap_free_page(bitmap, mappage);
154 return 0;
155 }
156
157 /* no page was in place and we have one, so install it */
158
159 memset(mappage, 0, PAGE_SIZE);
160 bitmap->bp[page].map = mappage;
161 bitmap->missing_pages--;
162out:
163 return 0;
164}
165
166
167/* if page is completely empty, put it back on the free list, or dealloc it */
168/* if page was hijacked, unmark the flag so it might get alloced next time */
169/* Note: lock should be held when calling this */
858119e1 170static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
171{
172 char *ptr;
173
174 if (bitmap->bp[page].count) /* page is still busy */
175 return;
176
177 /* page is no longer in use, it can be released */
178
179 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
180 bitmap->bp[page].hijacked = 0;
181 bitmap->bp[page].map = NULL;
182 return;
183 }
184
185 /* normal case, free the page */
186
187#if 0
188/* actually ... let's not. We will probably need the page again exactly when
189 * memory is tight and we are flusing to disk
190 */
191 return;
192#else
193 ptr = bitmap->bp[page].map;
194 bitmap->bp[page].map = NULL;
195 bitmap->missing_pages++;
196 bitmap_free_page(bitmap, ptr);
197 return;
198#endif
199}
200
201
202/*
203 * bitmap file handling - read and write the bitmap file and its superblock
204 */
205
32a7627c
N
206/*
207 * basic page I/O operations
208 */
209
a654b9d8
N
210/* IO operations when bitmap is stored near all superblocks */
211static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
212{
213 /* choose a good rdev and read the page from there */
214
215 mdk_rdev_t *rdev;
216 struct list_head *tmp;
217 struct page *page = alloc_page(GFP_KERNEL);
218 sector_t target;
219
220 if (!page)
221 return ERR_PTR(-ENOMEM);
a654b9d8 222
d089c6af 223 rdev_for_each(rdev, tmp, mddev) {
b2d444d7
N
224 if (! test_bit(In_sync, &rdev->flags)
225 || test_bit(Faulty, &rdev->flags))
ab904d63
N
226 continue;
227
a654b9d8
N
228 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
229
ab904d63
N
230 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
231 page->index = index;
ce25c31b
N
232 attach_page_buffers(page, NULL); /* so that free_buffer will
233 * quietly no-op */
ab904d63
N
234 return page;
235 }
236 }
237 return ERR_PTR(-EIO);
a654b9d8 238
a654b9d8
N
239}
240
ab6085c7 241static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
a654b9d8
N
242{
243 mdk_rdev_t *rdev;
244 struct list_head *tmp;
ab6085c7 245 mddev_t *mddev = bitmap->mddev;
a654b9d8 246
d089c6af 247 rdev_for_each(rdev, tmp, mddev)
b2d444d7 248 if (test_bit(In_sync, &rdev->flags)
ab6085c7
N
249 && !test_bit(Faulty, &rdev->flags)) {
250 int size = PAGE_SIZE;
251 if (page->index == bitmap->file_pages-1)
252 size = roundup(bitmap->last_page_size,
253 bdev_hardsect_size(rdev->bdev));
f0d76d70
N
254 /* Just make sure we aren't corrupting data or
255 * metadata
256 */
257 if (bitmap->offset < 0) {
258 /* DATA BITMAP METADATA */
259 if (bitmap->offset
85bfb4da 260 + (long)(page->index * (PAGE_SIZE/512))
f0d76d70
N
261 + size/512 > 0)
262 /* bitmap runs in to metadata */
263 return -EINVAL;
264 if (rdev->data_offset + mddev->size*2
265 > rdev->sb_offset*2 + bitmap->offset)
266 /* data runs in to bitmap */
267 return -EINVAL;
268 } else if (rdev->sb_offset*2 < rdev->data_offset) {
269 /* METADATA BITMAP DATA */
270 if (rdev->sb_offset*2
271 + bitmap->offset
272 + page->index*(PAGE_SIZE/512) + size/512
273 > rdev->data_offset)
274 /* bitmap runs in to data */
275 return -EINVAL;
276 } else {
277 /* DATA METADATA BITMAP - no problems */
278 }
a654b9d8 279 md_super_write(mddev, rdev,
ab6085c7 280 (rdev->sb_offset<<1) + bitmap->offset
a654b9d8 281 + page->index * (PAGE_SIZE/512),
ab6085c7 282 size,
a654b9d8 283 page);
ab6085c7 284 }
a654b9d8
N
285
286 if (wait)
a9701a30 287 md_super_wait(mddev);
a654b9d8
N
288 return 0;
289}
290
4ad13663 291static void bitmap_file_kick(struct bitmap *bitmap);
32a7627c 292/*
a654b9d8 293 * write out a page to a file
32a7627c 294 */
4ad13663 295static void write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 296{
d785a06a 297 struct buffer_head *bh;
32a7627c 298
f0d76d70
N
299 if (bitmap->file == NULL) {
300 switch (write_sb_page(bitmap, page, wait)) {
301 case -EINVAL:
302 bitmap->flags |= BITMAP_WRITE_ERROR;
f0d76d70 303 }
4ad13663 304 } else {
a654b9d8 305
4ad13663 306 bh = page_buffers(page);
c708443c 307
4ad13663
N
308 while (bh && bh->b_blocknr) {
309 atomic_inc(&bitmap->pending_writes);
310 set_buffer_locked(bh);
311 set_buffer_mapped(bh);
312 submit_bh(WRITE, bh);
313 bh = bh->b_this_page;
314 }
d785a06a 315
4ad13663
N
316 if (wait) {
317 wait_event(bitmap->write_wait,
318 atomic_read(&bitmap->pending_writes)==0);
319 }
8a5e9cf1 320 }
4ad13663
N
321 if (bitmap->flags & BITMAP_WRITE_ERROR)
322 bitmap_file_kick(bitmap);
d785a06a
N
323}
324
325static void end_bitmap_write(struct buffer_head *bh, int uptodate)
326{
327 struct bitmap *bitmap = bh->b_private;
328 unsigned long flags;
32a7627c 329
d785a06a
N
330 if (!uptodate) {
331 spin_lock_irqsave(&bitmap->lock, flags);
332 bitmap->flags |= BITMAP_WRITE_ERROR;
333 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 334 }
d785a06a
N
335 if (atomic_dec_and_test(&bitmap->pending_writes))
336 wake_up(&bitmap->write_wait);
337}
32a7627c 338
d785a06a
N
339/* copied from buffer.c */
340static void
341__clear_page_buffers(struct page *page)
342{
343 ClearPagePrivate(page);
344 set_page_private(page, 0);
345 page_cache_release(page);
346}
347static void free_buffers(struct page *page)
348{
349 struct buffer_head *bh = page_buffers(page);
77ad4bc7 350
d785a06a
N
351 while (bh) {
352 struct buffer_head *next = bh->b_this_page;
353 free_buffer_head(bh);
354 bh = next;
77ad4bc7 355 }
d785a06a
N
356 __clear_page_buffers(page);
357 put_page(page);
32a7627c
N
358}
359
d785a06a
N
360/* read a page from a file.
361 * We both read the page, and attach buffers to the page to record the
362 * address of each block (using bmap). These addresses will be used
363 * to write the block later, completely bypassing the filesystem.
364 * This usage is similar to how swap files are handled, and allows us
365 * to write to a file with no concerns of memory allocation failing.
366 */
32a7627c 367static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
368 struct bitmap *bitmap,
369 unsigned long count)
32a7627c 370{
32a7627c 371 struct page *page = NULL;
c649bb9c 372 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
373 struct buffer_head *bh;
374 sector_t block;
32a7627c 375
2d1f3b5d
N
376 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
377 (unsigned long long)index << PAGE_SHIFT);
32a7627c 378
d785a06a
N
379 page = alloc_page(GFP_KERNEL);
380 if (!page)
381 page = ERR_PTR(-ENOMEM);
32a7627c
N
382 if (IS_ERR(page))
383 goto out;
d785a06a 384
d785a06a
N
385 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
386 if (!bh) {
2d1f3b5d 387 put_page(page);
d785a06a 388 page = ERR_PTR(-ENOMEM);
32a7627c
N
389 goto out;
390 }
d785a06a
N
391 attach_page_buffers(page, bh);
392 block = index << (PAGE_SHIFT - inode->i_blkbits);
393 while (bh) {
394 if (count == 0)
395 bh->b_blocknr = 0;
396 else {
397 bh->b_blocknr = bmap(inode, block);
398 if (bh->b_blocknr == 0) {
399 /* Cannot use this file! */
400 free_buffers(page);
401 page = ERR_PTR(-EINVAL);
402 goto out;
403 }
404 bh->b_bdev = inode->i_sb->s_bdev;
405 if (count < (1<<inode->i_blkbits))
406 count = 0;
407 else
408 count -= (1<<inode->i_blkbits);
409
410 bh->b_end_io = end_bitmap_write;
411 bh->b_private = bitmap;
ce25c31b
N
412 atomic_inc(&bitmap->pending_writes);
413 set_buffer_locked(bh);
414 set_buffer_mapped(bh);
415 submit_bh(READ, bh);
d785a06a
N
416 }
417 block++;
418 bh = bh->b_this_page;
419 }
d785a06a 420 page->index = index;
ce25c31b
N
421
422 wait_event(bitmap->write_wait,
423 atomic_read(&bitmap->pending_writes)==0);
424 if (bitmap->flags & BITMAP_WRITE_ERROR) {
425 free_buffers(page);
426 page = ERR_PTR(-EIO);
427 }
32a7627c
N
428out:
429 if (IS_ERR(page))
430 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
431 (int)PAGE_SIZE,
432 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
433 PTR_ERR(page));
434 return page;
435}
436
437/*
438 * bitmap file superblock operations
439 */
440
441/* update the event counter and sync the superblock to disk */
4ad13663 442void bitmap_update_sb(struct bitmap *bitmap)
32a7627c
N
443{
444 bitmap_super_t *sb;
445 unsigned long flags;
446
447 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
4ad13663 448 return;
32a7627c
N
449 spin_lock_irqsave(&bitmap->lock, flags);
450 if (!bitmap->sb_page) { /* no superblock */
451 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 452 return;
32a7627c 453 }
32a7627c 454 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 455 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
456 sb->events = cpu_to_le64(bitmap->mddev->events);
457 if (!bitmap->mddev->degraded)
458 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
ea03aff9 459 kunmap_atomic(sb, KM_USER0);
4ad13663 460 write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
461}
462
463/* print out the bitmap file superblock */
464void bitmap_print_sb(struct bitmap *bitmap)
465{
466 bitmap_super_t *sb;
467
468 if (!bitmap || !bitmap->sb_page)
469 return;
ea03aff9 470 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 471 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
472 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
473 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
474 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
475 *(__u32 *)(sb->uuid+0),
476 *(__u32 *)(sb->uuid+4),
477 *(__u32 *)(sb->uuid+8),
478 *(__u32 *)(sb->uuid+12));
a2cff26a 479 printk(KERN_DEBUG " events: %llu\n",
32a7627c 480 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 481 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 482 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
483 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
484 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
485 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
486 printk(KERN_DEBUG " sync size: %llu KB\n",
487 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 488 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 489 kunmap_atomic(sb, KM_USER0);
32a7627c
N
490}
491
492/* read the superblock from the bitmap file and initialize some bitmap fields */
493static int bitmap_read_sb(struct bitmap *bitmap)
494{
495 char *reason = NULL;
496 bitmap_super_t *sb;
4b6d287f 497 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
498 unsigned long long events;
499 int err = -EINVAL;
500
501 /* page 0 is the superblock, read it... */
f49d5e62
N
502 if (bitmap->file) {
503 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
504 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
505
506 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
507 } else {
a654b9d8 508 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 509 }
32a7627c
N
510 if (IS_ERR(bitmap->sb_page)) {
511 err = PTR_ERR(bitmap->sb_page);
512 bitmap->sb_page = NULL;
513 return err;
514 }
515
ea03aff9 516 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 517
32a7627c
N
518 chunksize = le32_to_cpu(sb->chunksize);
519 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 520 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
521
522 /* verify that the bitmap-specific fields are valid */
523 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
524 reason = "bad magic";
bd926c63
N
525 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
526 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 527 reason = "unrecognized superblock version";
7dd5d34c
N
528 else if (chunksize < PAGE_SIZE)
529 reason = "bitmap chunksize too small";
32a7627c
N
530 else if ((1 << ffz(~chunksize)) != chunksize)
531 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
532 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
533 reason = "daemon sleep period out of range";
4b6d287f
N
534 else if (write_behind > COUNTER_MAX)
535 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
536 if (reason) {
537 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
538 bmname(bitmap), reason);
539 goto out;
540 }
541
542 /* keep the array size field of the bitmap superblock up to date */
543 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
544
545 if (!bitmap->mddev->persistent)
546 goto success;
547
548 /*
549 * if we have a persistent array superblock, compare the
550 * bitmap's UUID and event counter to the mddev's
551 */
552 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
553 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
554 bmname(bitmap));
555 goto out;
556 }
557 events = le64_to_cpu(sb->events);
558 if (events < bitmap->mddev->events) {
559 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
560 "-- forcing full recovery\n", bmname(bitmap), events,
561 (unsigned long long) bitmap->mddev->events);
4f2e639a 562 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
563 }
564success:
565 /* assign fields using values from superblock */
566 bitmap->chunksize = chunksize;
567 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 568 bitmap->daemon_lastrun = jiffies;
4b6d287f 569 bitmap->max_write_behind = write_behind;
4f2e639a 570 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
571 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
572 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 573 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 574 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 575 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
576 err = 0;
577out:
ea03aff9 578 kunmap_atomic(sb, KM_USER0);
32a7627c
N
579 if (err)
580 bitmap_print_sb(bitmap);
581 return err;
582}
583
584enum bitmap_mask_op {
585 MASK_SET,
586 MASK_UNSET
587};
588
4ad13663
N
589/* record the state of the bitmap in the superblock. Return the old value */
590static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
591 enum bitmap_mask_op op)
32a7627c
N
592{
593 bitmap_super_t *sb;
594 unsigned long flags;
4ad13663 595 int old;
32a7627c
N
596
597 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 598 if (!bitmap->sb_page) { /* can't set the state */
32a7627c 599 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 600 return 0;
32a7627c 601 }
32a7627c 602 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 603 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
4ad13663 604 old = le32_to_cpu(sb->state) & bits;
32a7627c 605 switch (op) {
4f2e639a 606 case MASK_SET: sb->state |= cpu_to_le32(bits);
32a7627c 607 break;
4f2e639a 608 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
32a7627c
N
609 break;
610 default: BUG();
611 }
ea03aff9 612 kunmap_atomic(sb, KM_USER0);
4ad13663 613 return old;
32a7627c
N
614}
615
616/*
617 * general bitmap file operations
618 */
619
620/* calculate the index of the page that contains this bit */
621static inline unsigned long file_page_index(unsigned long chunk)
622{
623 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
624}
625
626/* calculate the (bit) offset of this bit within a page */
627static inline unsigned long file_page_offset(unsigned long chunk)
628{
629 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
630}
631
632/*
633 * return a pointer to the page in the filemap that contains the given bit
634 *
635 * this lookup is complicated by the fact that the bitmap sb might be exactly
636 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
637 * 0 or page 1
638 */
639static inline struct page *filemap_get_page(struct bitmap *bitmap,
640 unsigned long chunk)
641{
9b1d1dac 642 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
32a7627c
N
643 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
644}
645
646
647static void bitmap_file_unmap(struct bitmap *bitmap)
648{
649 struct page **map, *sb_page;
650 unsigned long *attr;
651 int pages;
652 unsigned long flags;
653
654 spin_lock_irqsave(&bitmap->lock, flags);
655 map = bitmap->filemap;
656 bitmap->filemap = NULL;
657 attr = bitmap->filemap_attr;
658 bitmap->filemap_attr = NULL;
659 pages = bitmap->file_pages;
660 bitmap->file_pages = 0;
661 sb_page = bitmap->sb_page;
662 bitmap->sb_page = NULL;
663 spin_unlock_irqrestore(&bitmap->lock, flags);
664
665 while (pages--)
666 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
d785a06a 667 free_buffers(map[pages]);
32a7627c
N
668 kfree(map);
669 kfree(attr);
670
d785a06a
N
671 if (sb_page)
672 free_buffers(sb_page);
32a7627c
N
673}
674
675static void bitmap_file_put(struct bitmap *bitmap)
676{
677 struct file *file;
32a7627c
N
678 unsigned long flags;
679
680 spin_lock_irqsave(&bitmap->lock, flags);
681 file = bitmap->file;
682 bitmap->file = NULL;
683 spin_unlock_irqrestore(&bitmap->lock, flags);
684
d785a06a
N
685 if (file)
686 wait_event(bitmap->write_wait,
687 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
688 bitmap_file_unmap(bitmap);
689
d785a06a 690 if (file) {
c649bb9c 691 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 692 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 693 fput(file);
d785a06a 694 }
32a7627c
N
695}
696
697
698/*
699 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
700 * then it is no longer reliable, so we stop using it and we mark the file
701 * as failed in the superblock
702 */
703static void bitmap_file_kick(struct bitmap *bitmap)
704{
705 char *path, *ptr = NULL;
706
4ad13663
N
707 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
708 bitmap_update_sb(bitmap);
32a7627c 709
4ad13663
N
710 if (bitmap->file) {
711 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
712 if (path)
6bcfd601
CH
713 ptr = d_path(&bitmap->file->f_path, path,
714 PAGE_SIZE);
715
32a7627c 716
4ad13663
N
717 printk(KERN_ALERT
718 "%s: kicking failed bitmap file %s from array!\n",
6bcfd601 719 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
32a7627c 720
4ad13663
N
721 kfree(path);
722 } else
723 printk(KERN_ALERT
724 "%s: disabling internal bitmap due to errors\n",
725 bmname(bitmap));
a654b9d8 726 }
32a7627c
N
727
728 bitmap_file_put(bitmap);
729
730 return;
731}
732
733enum bitmap_page_attr {
e16b68b6
N
734 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
735 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
736 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
737};
738
739static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
740 enum bitmap_page_attr attr)
741{
e16b68b6 742 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
743}
744
745static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
746 enum bitmap_page_attr attr)
747{
e16b68b6 748 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
749}
750
ec7a3197
N
751static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
752 enum bitmap_page_attr attr)
32a7627c 753{
e16b68b6 754 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
755}
756
757/*
758 * bitmap_file_set_bit -- called before performing a write to the md device
759 * to set (and eventually sync) a particular bit in the bitmap file
760 *
761 * we set the bit immediately, then we record the page number so that
762 * when an unplug occurs, we can flush the dirty pages out to disk
763 */
764static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
765{
766 unsigned long bit;
767 struct page *page;
768 void *kaddr;
769 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
770
a654b9d8 771 if (!bitmap->filemap) {
32a7627c
N
772 return;
773 }
774
775 page = filemap_get_page(bitmap, chunk);
9b1d1dac 776 if (!page) return;
32a7627c
N
777 bit = file_page_offset(chunk);
778
32a7627c
N
779 /* set the bit */
780 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
781 if (bitmap->flags & BITMAP_HOSTENDIAN)
782 set_bit(bit, kaddr);
783 else
784 ext2_set_bit(bit, kaddr);
32a7627c
N
785 kunmap_atomic(kaddr, KM_USER0);
786 PRINTK("set file bit %lu page %lu\n", bit, page->index);
787
788 /* record page number so it gets flushed to disk when unplug occurs */
789 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
790
791}
792
793/* this gets called when the md device is ready to unplug its underlying
794 * (slave) device queues -- before we let any writes go down, we need to
795 * sync the dirty pages of the bitmap file to disk */
4ad13663 796void bitmap_unplug(struct bitmap *bitmap)
32a7627c 797{
ec7a3197
N
798 unsigned long i, flags;
799 int dirty, need_write;
32a7627c
N
800 struct page *page;
801 int wait = 0;
802
803 if (!bitmap)
4ad13663 804 return;
32a7627c
N
805
806 /* look at each page to see if there are any set bits that need to be
807 * flushed out to disk */
808 for (i = 0; i < bitmap->file_pages; i++) {
809 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 810 if (!bitmap->filemap) {
32a7627c 811 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 812 return;
32a7627c
N
813 }
814 page = bitmap->filemap[i];
ec7a3197
N
815 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
816 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
817 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
818 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 819 if (dirty)
32a7627c
N
820 wait = 1;
821 spin_unlock_irqrestore(&bitmap->lock, flags);
822
d785a06a 823 if (dirty | need_write)
4ad13663 824 write_page(bitmap, page, 0);
32a7627c
N
825 }
826 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 827 if (bitmap->file)
d785a06a
N
828 wait_event(bitmap->write_wait,
829 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 830 else
a9701a30 831 md_super_wait(bitmap->mddev);
32a7627c 832 }
d785a06a
N
833 if (bitmap->flags & BITMAP_WRITE_ERROR)
834 bitmap_file_kick(bitmap);
32a7627c
N
835}
836
6a07997f 837static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
838/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
839 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
840 * memory mapping of the bitmap file
841 * Special cases:
842 * if there's no bitmap file, or if the bitmap file had been
843 * previously kicked from the array, we mark all the bits as
844 * 1's in order to cause a full resync.
6a07997f
N
845 *
846 * We ignore all bits for sectors that end earlier than 'start'.
847 * This is used when reading an out-of-date bitmap...
32a7627c 848 */
6a07997f 849static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
850{
851 unsigned long i, chunks, index, oldindex, bit;
852 struct page *page = NULL, *oldpage = NULL;
853 unsigned long num_pages, bit_cnt = 0;
854 struct file *file;
d785a06a 855 unsigned long bytes, offset;
32a7627c
N
856 int outofdate;
857 int ret = -ENOSPC;
ea03aff9 858 void *paddr;
32a7627c
N
859
860 chunks = bitmap->chunks;
861 file = bitmap->file;
862
a654b9d8 863 BUG_ON(!file && !bitmap->offset);
32a7627c 864
44456d37 865#ifdef INJECT_FAULTS_3
32a7627c
N
866 outofdate = 1;
867#else
868 outofdate = bitmap->flags & BITMAP_STALE;
869#endif
870 if (outofdate)
871 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
872 "recovery\n", bmname(bitmap));
873
874 bytes = (chunks + 7) / 8;
bc7f77de 875
cdbb4cc2 876 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 877
a654b9d8 878 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
879 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
880 bmname(bitmap),
881 (unsigned long) i_size_read(file->f_mapping->host),
882 bytes + sizeof(bitmap_super_t));
4ad13663 883 goto err;
32a7627c 884 }
bc7f77de
N
885
886 ret = -ENOMEM;
887
32a7627c 888 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 889 if (!bitmap->filemap)
4ad13663 890 goto err;
32a7627c 891
e16b68b6
N
892 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
893 bitmap->filemap_attr = kzalloc(
505fa2c4 894 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 895 GFP_KERNEL);
bc7f77de 896 if (!bitmap->filemap_attr)
4ad13663 897 goto err;
32a7627c 898
32a7627c
N
899 oldindex = ~0L;
900
901 for (i = 0; i < chunks; i++) {
bd926c63 902 int b;
32a7627c
N
903 index = file_page_index(i);
904 bit = file_page_offset(i);
905 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 906 int count;
32a7627c 907 /* unmap the old page, we're done with it */
d785a06a 908 if (index == num_pages-1)
f49d5e62
N
909 count = bytes + sizeof(bitmap_super_t)
910 - index * PAGE_SIZE;
d785a06a
N
911 else
912 count = PAGE_SIZE;
32a7627c
N
913 if (index == 0) {
914 /*
915 * if we're here then the superblock page
916 * contains some bits (PAGE_SIZE != sizeof sb)
917 * we've already read it in, so just use it
918 */
919 page = bitmap->sb_page;
920 offset = sizeof(bitmap_super_t);
a654b9d8 921 } else if (file) {
d785a06a 922 page = read_page(file, index, bitmap, count);
a654b9d8
N
923 offset = 0;
924 } else {
925 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
926 offset = 0;
927 }
a654b9d8
N
928 if (IS_ERR(page)) { /* read error */
929 ret = PTR_ERR(page);
4ad13663 930 goto err;
a654b9d8
N
931 }
932
32a7627c
N
933 oldindex = index;
934 oldpage = page;
32a7627c
N
935
936 if (outofdate) {
937 /*
938 * if bitmap is out of date, dirty the
939 * whole page and write it out
940 */
ea03aff9
N
941 paddr = kmap_atomic(page, KM_USER0);
942 memset(paddr + offset, 0xff,
6a07997f 943 PAGE_SIZE - offset);
ea03aff9 944 kunmap_atomic(paddr, KM_USER0);
4ad13663
N
945 write_page(bitmap, page, 1);
946
947 ret = -EIO;
948 if (bitmap->flags & BITMAP_WRITE_ERROR) {
32a7627c 949 /* release, page not in filemap yet */
2d1f3b5d 950 put_page(page);
4ad13663 951 goto err;
32a7627c
N
952 }
953 }
954
955 bitmap->filemap[bitmap->file_pages++] = page;
ab6085c7 956 bitmap->last_page_size = count;
32a7627c 957 }
ea03aff9 958 paddr = kmap_atomic(page, KM_USER0);
bd926c63 959 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 960 b = test_bit(bit, paddr);
bd926c63 961 else
ea03aff9
N
962 b = ext2_test_bit(bit, paddr);
963 kunmap_atomic(paddr, KM_USER0);
bd926c63 964 if (b) {
32a7627c 965 /* if the disk bit is set, set the memory bit */
6a07997f
N
966 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
967 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
968 );
32a7627c 969 bit_cnt++;
6a07997f 970 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 971 }
32a7627c
N
972 }
973
974 /* everything went OK */
975 ret = 0;
976 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
977
32a7627c
N
978 if (bit_cnt) { /* Kick recovery if any bits were set */
979 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
980 md_wakeup_thread(bitmap->mddev->thread);
981 }
982
32a7627c 983 printk(KERN_INFO "%s: bitmap initialized from disk: "
4ad13663
N
984 "read %lu/%lu pages, set %lu bits\n",
985 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
986
987 return 0;
32a7627c 988
4ad13663
N
989 err:
990 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
991 bmname(bitmap), ret);
32a7627c
N
992 return ret;
993}
994
a654b9d8
N
995void bitmap_write_all(struct bitmap *bitmap)
996{
997 /* We don't actually write all bitmap blocks here,
998 * just flag them as needing to be written
999 */
ec7a3197 1000 int i;
a654b9d8 1001
ec7a3197
N
1002 for (i=0; i < bitmap->file_pages; i++)
1003 set_page_attr(bitmap, bitmap->filemap[i],
1004 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
1005}
1006
32a7627c
N
1007
1008static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1009{
1010 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1011 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1012 bitmap->bp[page].count += inc;
1013/*
1014 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1015 (unsigned long long)offset, inc, bitmap->bp[page].count);
1016*/
1017 bitmap_checkfree(bitmap, page);
1018}
1019static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1020 sector_t offset, int *blocks,
1021 int create);
1022
1023/*
1024 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1025 * out to disk
1026 */
1027
4ad13663 1028void bitmap_daemon_work(struct bitmap *bitmap)
32a7627c 1029{
aa3163f8 1030 unsigned long j;
32a7627c
N
1031 unsigned long flags;
1032 struct page *page = NULL, *lastpage = NULL;
32a7627c 1033 int blocks;
ea03aff9 1034 void *paddr;
32a7627c
N
1035
1036 if (bitmap == NULL)
4ad13663 1037 return;
32a7627c 1038 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
7be3dfec
N
1039 goto done;
1040
32a7627c 1041 bitmap->daemon_lastrun = jiffies;
8311c29d
N
1042 if (bitmap->allclean) {
1043 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1044 return;
1045 }
1046 bitmap->allclean = 1;
32a7627c
N
1047
1048 for (j = 0; j < bitmap->chunks; j++) {
1049 bitmap_counter_t *bmc;
1050 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1051 if (!bitmap->filemap) {
32a7627c
N
1052 /* error or shutdown */
1053 spin_unlock_irqrestore(&bitmap->lock, flags);
1054 break;
1055 }
1056
1057 page = filemap_get_page(bitmap, j);
32a7627c
N
1058
1059 if (page != lastpage) {
aa3163f8 1060 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1061 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1062 int need_write = test_page_attr(bitmap, page,
1063 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1064 if (need_write)
aa3163f8 1065 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1066
aa3163f8 1067 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1068 if (need_write) {
4ad13663 1069 write_page(bitmap, page, 0);
8311c29d
N
1070 bitmap->allclean = 0;
1071 }
aa3163f8
N
1072 continue;
1073 }
1074
32a7627c 1075 /* grab the new page, sync and release the old */
32a7627c 1076 if (lastpage != NULL) {
ec7a3197 1077 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1078 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1079 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1080 write_page(bitmap, lastpage, 0);
32a7627c
N
1081 } else {
1082 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1083 spin_unlock_irqrestore(&bitmap->lock, flags);
1084 }
32a7627c
N
1085 } else
1086 spin_unlock_irqrestore(&bitmap->lock, flags);
1087 lastpage = page;
32a7627c
N
1088/*
1089 printk("bitmap clean at page %lu\n", j);
1090*/
1091 spin_lock_irqsave(&bitmap->lock, flags);
1092 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1093 }
1094 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1095 &blocks, 0);
1096 if (bmc) {
1097/*
1098 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1099*/
8311c29d
N
1100 if (*bmc)
1101 bitmap->allclean = 0;
1102
32a7627c
N
1103 if (*bmc == 2) {
1104 *bmc=1; /* maybe clear the bit next time */
1105 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1106 } else if (*bmc == 1) {
1107 /* we can clear the bit */
1108 *bmc = 0;
1109 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1110 -1);
1111
1112 /* clear the bit */
ea03aff9 1113 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1114 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1115 clear_bit(file_page_offset(j), paddr);
bd926c63 1116 else
ea03aff9
N
1117 ext2_clear_bit(file_page_offset(j), paddr);
1118 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1119 }
1120 }
1121 spin_unlock_irqrestore(&bitmap->lock, flags);
1122 }
1123
1124 /* now sync the final page */
1125 if (lastpage != NULL) {
32a7627c 1126 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1127 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1128 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1129 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1130 write_page(bitmap, lastpage, 0);
32a7627c
N
1131 } else {
1132 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1133 spin_unlock_irqrestore(&bitmap->lock, flags);
1134 }
32a7627c
N
1135 }
1136
7be3dfec 1137 done:
8311c29d
N
1138 if (bitmap->allclean == 0)
1139 bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ;
32a7627c
N
1140}
1141
32a7627c
N
1142static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1143 sector_t offset, int *blocks,
1144 int create)
1145{
1146 /* If 'create', we might release the lock and reclaim it.
1147 * The lock must have been taken with interrupts enabled.
1148 * If !create, we don't release the lock.
1149 */
1150 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1151 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1152 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1153 sector_t csize;
1154
1155 if (bitmap_checkpage(bitmap, page, create) < 0) {
1156 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1157 *blocks = csize - (offset & (csize- 1));
1158 return NULL;
1159 }
1160 /* now locked ... */
1161
1162 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1163 /* should we use the first or second counter field
1164 * of the hijacked pointer? */
1165 int hi = (pageoff > PAGE_COUNTER_MASK);
1166 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1167 PAGE_COUNTER_SHIFT - 1);
1168 *blocks = csize - (offset & (csize- 1));
1169 return &((bitmap_counter_t *)
1170 &bitmap->bp[page].map)[hi];
1171 } else { /* page is allocated */
1172 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1173 *blocks = csize - (offset & (csize- 1));
1174 return (bitmap_counter_t *)
1175 &(bitmap->bp[page].map[pageoff]);
1176 }
1177}
1178
4b6d287f 1179int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1180{
1181 if (!bitmap) return 0;
4b6d287f
N
1182
1183 if (behind) {
1184 atomic_inc(&bitmap->behind_writes);
1185 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1186 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1187 }
1188
32a7627c
N
1189 while (sectors) {
1190 int blocks;
1191 bitmap_counter_t *bmc;
1192
1193 spin_lock_irq(&bitmap->lock);
1194 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1195 if (!bmc) {
1196 spin_unlock_irq(&bitmap->lock);
1197 return 0;
1198 }
1199
da6e1a32
NB
1200 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1201 DEFINE_WAIT(__wait);
1202 /* note that it is safe to do the prepare_to_wait
1203 * after the test as long as we do it before dropping
1204 * the spinlock.
1205 */
1206 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1207 TASK_UNINTERRUPTIBLE);
1208 spin_unlock_irq(&bitmap->lock);
2ad8b1ef 1209 blk_unplug(bitmap->mddev->queue);
da6e1a32
NB
1210 schedule();
1211 finish_wait(&bitmap->overflow_wait, &__wait);
1212 continue;
1213 }
1214
32a7627c
N
1215 switch(*bmc) {
1216 case 0:
1217 bitmap_file_set_bit(bitmap, offset);
1218 bitmap_count_page(bitmap,offset, 1);
1219 blk_plug_device(bitmap->mddev->queue);
1220 /* fall through */
1221 case 1:
1222 *bmc = 2;
1223 }
da6e1a32 1224
32a7627c
N
1225 (*bmc)++;
1226
1227 spin_unlock_irq(&bitmap->lock);
1228
1229 offset += blocks;
1230 if (sectors > blocks)
1231 sectors -= blocks;
1232 else sectors = 0;
1233 }
8311c29d 1234 bitmap->allclean = 0;
32a7627c
N
1235 return 0;
1236}
1237
1238void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1239 int success, int behind)
32a7627c
N
1240{
1241 if (!bitmap) return;
4b6d287f
N
1242 if (behind) {
1243 atomic_dec(&bitmap->behind_writes);
1244 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1245 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1246 }
1247
32a7627c
N
1248 while (sectors) {
1249 int blocks;
1250 unsigned long flags;
1251 bitmap_counter_t *bmc;
1252
1253 spin_lock_irqsave(&bitmap->lock, flags);
1254 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1255 if (!bmc) {
1256 spin_unlock_irqrestore(&bitmap->lock, flags);
1257 return;
1258 }
1259
1260 if (!success && ! (*bmc & NEEDED_MASK))
1261 *bmc |= NEEDED_MASK;
1262
da6e1a32
NB
1263 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1264 wake_up(&bitmap->overflow_wait);
1265
32a7627c
N
1266 (*bmc)--;
1267 if (*bmc <= 2) {
1268 set_page_attr(bitmap,
1269 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1270 BITMAP_PAGE_CLEAN);
1271 }
1272 spin_unlock_irqrestore(&bitmap->lock, flags);
1273 offset += blocks;
1274 if (sectors > blocks)
1275 sectors -= blocks;
1276 else sectors = 0;
1277 }
1278}
1279
6a806c51
N
1280int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1281 int degraded)
32a7627c
N
1282{
1283 bitmap_counter_t *bmc;
1284 int rv;
1285 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1286 *blocks = 1024;
1287 return 1; /* always resync if no bitmap */
1288 }
1289 spin_lock_irq(&bitmap->lock);
1290 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1291 rv = 0;
1292 if (bmc) {
1293 /* locked */
1294 if (RESYNC(*bmc))
1295 rv = 1;
1296 else if (NEEDED(*bmc)) {
1297 rv = 1;
6a806c51
N
1298 if (!degraded) { /* don't set/clear bits if degraded */
1299 *bmc |= RESYNC_MASK;
1300 *bmc &= ~NEEDED_MASK;
1301 }
32a7627c
N
1302 }
1303 }
1304 spin_unlock_irq(&bitmap->lock);
8311c29d 1305 bitmap->allclean = 0;
32a7627c
N
1306 return rv;
1307}
1308
1309void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1310{
1311 bitmap_counter_t *bmc;
1312 unsigned long flags;
1313/*
1314 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1315*/ if (bitmap == NULL) {
1316 *blocks = 1024;
1317 return;
1318 }
1319 spin_lock_irqsave(&bitmap->lock, flags);
1320 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1321 if (bmc == NULL)
1322 goto unlock;
1323 /* locked */
1324/*
1325 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1326*/
1327 if (RESYNC(*bmc)) {
1328 *bmc &= ~RESYNC_MASK;
1329
1330 if (!NEEDED(*bmc) && aborted)
1331 *bmc |= NEEDED_MASK;
1332 else {
1333 if (*bmc <= 2) {
1334 set_page_attr(bitmap,
1335 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1336 BITMAP_PAGE_CLEAN);
1337 }
1338 }
1339 }
1340 unlock:
1341 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1342 bitmap->allclean = 0;
32a7627c
N
1343}
1344
1345void bitmap_close_sync(struct bitmap *bitmap)
1346{
1347 /* Sync has finished, and any bitmap chunks that weren't synced
1348 * properly have been aborted. It remains to us to clear the
1349 * RESYNC bit wherever it is still on
1350 */
1351 sector_t sector = 0;
1352 int blocks;
b47490c9
N
1353 if (!bitmap)
1354 return;
32a7627c
N
1355 while (sector < bitmap->mddev->resync_max_sectors) {
1356 bitmap_end_sync(bitmap, sector, &blocks, 0);
b47490c9
N
1357 sector += blocks;
1358 }
1359}
1360
1361void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1362{
1363 sector_t s = 0;
1364 int blocks;
1365
1366 if (!bitmap)
1367 return;
1368 if (sector == 0) {
1369 bitmap->last_end_sync = jiffies;
1370 return;
1371 }
1372 if (time_before(jiffies, (bitmap->last_end_sync
1373 + bitmap->daemon_sleep * HZ)))
1374 return;
1375 wait_event(bitmap->mddev->recovery_wait,
1376 atomic_read(&bitmap->mddev->recovery_active) == 0);
1377
1378 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1379 s = 0;
1380 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1381 bitmap_end_sync(bitmap, s, &blocks, 0);
1382 s += blocks;
32a7627c 1383 }
b47490c9 1384 bitmap->last_end_sync = jiffies;
32a7627c
N
1385}
1386
6a07997f 1387static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1388{
1389 /* For each chunk covered by any of these sectors, set the
193f1c93 1390 * counter to 1 and set resync_needed. They should all
32a7627c
N
1391 * be 0 at this point
1392 */
193f1c93
N
1393
1394 int secs;
1395 bitmap_counter_t *bmc;
1396 spin_lock_irq(&bitmap->lock);
1397 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1398 if (!bmc) {
32a7627c 1399 spin_unlock_irq(&bitmap->lock);
193f1c93 1400 return;
32a7627c 1401 }
193f1c93
N
1402 if (! *bmc) {
1403 struct page *page;
6a07997f 1404 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1405 bitmap_count_page(bitmap, offset, 1);
1406 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1407 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1408 }
1409 spin_unlock_irq(&bitmap->lock);
8311c29d 1410 bitmap->allclean = 0;
32a7627c
N
1411}
1412
9b1d1dac
PC
1413/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1414void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1415{
1416 unsigned long chunk;
1417
1418 for (chunk = s; chunk <= e; chunk++) {
1419 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1420 bitmap_set_memory_bits(bitmap, sec, 1);
1421 bitmap_file_set_bit(bitmap, sec);
1422 }
1423}
1424
6b8b3e8a
N
1425/*
1426 * flush out any pending updates
1427 */
1428void bitmap_flush(mddev_t *mddev)
1429{
1430 struct bitmap *bitmap = mddev->bitmap;
1431 int sleep;
1432
1433 if (!bitmap) /* there was no bitmap */
1434 return;
1435
1436 /* run the daemon_work three time to ensure everything is flushed
1437 * that can be
1438 */
1439 sleep = bitmap->daemon_sleep;
1440 bitmap->daemon_sleep = 0;
1441 bitmap_daemon_work(bitmap);
1442 bitmap_daemon_work(bitmap);
1443 bitmap_daemon_work(bitmap);
1444 bitmap->daemon_sleep = sleep;
1445 bitmap_update_sb(bitmap);
1446}
1447
32a7627c
N
1448/*
1449 * free memory that was allocated
1450 */
3178b0db 1451static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1452{
1453 unsigned long k, pages;
1454 struct bitmap_page *bp;
32a7627c
N
1455
1456 if (!bitmap) /* there was no bitmap */
1457 return;
1458
32a7627c
N
1459 /* release the bitmap file and kill the daemon */
1460 bitmap_file_put(bitmap);
1461
1462 bp = bitmap->bp;
1463 pages = bitmap->pages;
1464
1465 /* free all allocated memory */
1466
32a7627c
N
1467 if (bp) /* deallocate the page memory */
1468 for (k = 0; k < pages; k++)
1469 if (bp[k].map && !bp[k].hijacked)
1470 kfree(bp[k].map);
1471 kfree(bp);
1472 kfree(bitmap);
1473}
3178b0db
N
1474void bitmap_destroy(mddev_t *mddev)
1475{
1476 struct bitmap *bitmap = mddev->bitmap;
1477
1478 if (!bitmap) /* there was no bitmap */
1479 return;
1480
1481 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1482 if (mddev->thread)
1483 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1484
1485 bitmap_free(bitmap);
1486}
32a7627c
N
1487
1488/*
1489 * initialize the bitmap structure
1490 * if this returns an error, bitmap_destroy must be called to do clean up
1491 */
1492int bitmap_create(mddev_t *mddev)
1493{
1494 struct bitmap *bitmap;
1495 unsigned long blocks = mddev->resync_max_sectors;
1496 unsigned long chunks;
1497 unsigned long pages;
1498 struct file *file = mddev->bitmap_file;
1499 int err;
6a07997f 1500 sector_t start;
32a7627c 1501
5f6e3c83 1502 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1503
a654b9d8 1504 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1505 return 0;
1506
a654b9d8
N
1507 BUG_ON(file && mddev->bitmap_offset);
1508
9ffae0cf 1509 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1510 if (!bitmap)
1511 return -ENOMEM;
1512
32a7627c 1513 spin_lock_init(&bitmap->lock);
ce25c31b
N
1514 atomic_set(&bitmap->pending_writes, 0);
1515 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1516 init_waitqueue_head(&bitmap->overflow_wait);
ce25c31b 1517
32a7627c 1518 bitmap->mddev = mddev;
32a7627c 1519
32a7627c 1520 bitmap->file = file;
a654b9d8 1521 bitmap->offset = mddev->bitmap_offset;
ce25c31b
N
1522 if (file) {
1523 get_file(file);
ef51c976
MF
1524 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1525 SYNC_FILE_RANGE_WAIT_BEFORE |
1526 SYNC_FILE_RANGE_WRITE |
1527 SYNC_FILE_RANGE_WAIT_AFTER);
ce25c31b 1528 }
32a7627c
N
1529 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1530 err = bitmap_read_sb(bitmap);
1531 if (err)
3178b0db 1532 goto error;
32a7627c 1533
a638b2dc 1534 bitmap->chunkshift = ffz(~bitmap->chunksize);
32a7627c
N
1535
1536 /* now that chunksize and chunkshift are set, we can use these macros */
1537 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1538 CHUNK_BLOCK_RATIO(bitmap);
1539 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1540
1541 BUG_ON(!pages);
1542
1543 bitmap->chunks = chunks;
1544 bitmap->pages = pages;
1545 bitmap->missing_pages = pages;
1546 bitmap->counter_bits = COUNTER_BITS;
1547
1548 bitmap->syncchunk = ~0UL;
1549
44456d37 1550#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1551 bitmap->bp = NULL;
1552#else
9ffae0cf 1553 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1554#endif
3178b0db 1555 err = -ENOMEM;
32a7627c 1556 if (!bitmap->bp)
3178b0db 1557 goto error;
32a7627c 1558
32a7627c
N
1559 /* now that we have some pages available, initialize the in-memory
1560 * bitmap from the on-disk bitmap */
6a07997f
N
1561 start = 0;
1562 if (mddev->degraded == 0
1563 || bitmap->events_cleared == mddev->events)
1564 /* no need to keep dirty bits to optimise a re-add of a missing device */
1565 start = mddev->recovery_cp;
1566 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1567
32a7627c 1568 if (err)
3178b0db 1569 goto error;
32a7627c
N
1570
1571 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1572 pages, bmname(bitmap));
1573
3178b0db
N
1574 mddev->bitmap = bitmap;
1575
b15c2e57
N
1576 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1577
4ad13663
N
1578 bitmap_update_sb(bitmap);
1579
1580 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
3178b0db
N
1581
1582 error:
1583 bitmap_free(bitmap);
1584 return err;
32a7627c
N
1585}
1586
1587/* the bitmap API -- for raid personalities */
1588EXPORT_SYMBOL(bitmap_startwrite);
1589EXPORT_SYMBOL(bitmap_endwrite);
1590EXPORT_SYMBOL(bitmap_start_sync);
1591EXPORT_SYMBOL(bitmap_end_sync);
1592EXPORT_SYMBOL(bitmap_unplug);
1593EXPORT_SYMBOL(bitmap_close_sync);
b47490c9 1594EXPORT_SYMBOL(bitmap_cond_end_sync);