resubmit: cciss: procfs updates to display info about many
[linux-2.6-block.git] / drivers / md / bitmap.c
CommitLineData
32a7627c
N
1/*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
32a7627c
N
10 */
11
12/*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
16 * wait if count gets too high, wake when it drops to half.
32a7627c
N
17 */
18
19#include <linux/module.h>
32a7627c
N
20#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/init.h>
32a7627c
N
23#include <linux/timer.h>
24#include <linux/sched.h>
25#include <linux/list.h>
26#include <linux/file.h>
27#include <linux/mount.h>
28#include <linux/buffer_head.h>
29#include <linux/raid/md.h>
30#include <linux/raid/bitmap.h>
31
32/* debug macros */
33
34#define DEBUG 0
35
36#if DEBUG
37/* these are for debugging purposes only! */
38
39/* define one and only one of these */
40#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
41#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
42#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
43#define INJECT_FAULTS_4 0 /* undef */
44#define INJECT_FAULTS_5 0 /* undef */
45#define INJECT_FAULTS_6 0
46
47/* if these are defined, the driver will fail! debug only */
48#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
49#define INJECT_FATAL_FAULT_2 0 /* undef */
50#define INJECT_FATAL_FAULT_3 0 /* undef */
51#endif
52
53//#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
54#define DPRINTK(x...) do { } while(0)
55
56#ifndef PRINTK
57# if DEBUG > 0
58# define PRINTK(x...) printk(KERN_DEBUG x)
59# else
60# define PRINTK(x...)
61# endif
62#endif
63
64static inline char * bmname(struct bitmap *bitmap)
65{
66 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
67}
68
32a7627c
N
69
70/*
71 * just a placeholder - calls kmalloc for bitmap pages
72 */
73static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
74{
75 unsigned char *page;
76
44456d37 77#ifdef INJECT_FAULTS_1
32a7627c
N
78 page = NULL;
79#else
80 page = kmalloc(PAGE_SIZE, GFP_NOIO);
81#endif
82 if (!page)
83 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
84 else
a654b9d8 85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
86 bmname(bitmap), page);
87 return page;
88}
89
90/*
91 * for now just a placeholder -- just calls kfree for bitmap pages
92 */
93static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
94{
95 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
96 kfree(page);
97}
98
99/*
100 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
101 *
102 * 1) check to see if this page is allocated, if it's not then try to alloc
103 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
104 * page pointer directly as a counter
105 *
106 * if we find our page, we increment the page's refcount so that it stays
107 * allocated while we're using it
108 */
109static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
110{
111 unsigned char *mappage;
112
113 if (page >= bitmap->pages) {
114 printk(KERN_ALERT
115 "%s: invalid bitmap page request: %lu (> %lu)\n",
116 bmname(bitmap), page, bitmap->pages-1);
117 return -EINVAL;
118 }
119
120
121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
122 return 0;
123
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
125 return 0;
126
127 if (!create)
128 return -ENOENT;
129
130 spin_unlock_irq(&bitmap->lock);
131
132 /* this page has not been allocated yet */
133
134 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
135 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
136 bmname(bitmap));
137 /* failed - set the hijacked flag so that we can use the
138 * pointer as a counter */
139 spin_lock_irq(&bitmap->lock);
140 if (!bitmap->bp[page].map)
141 bitmap->bp[page].hijacked = 1;
142 goto out;
143 }
144
145 /* got a page */
146
147 spin_lock_irq(&bitmap->lock);
148
149 /* recheck the page */
150
151 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
152 /* somebody beat us to getting the page */
153 bitmap_free_page(bitmap, mappage);
154 return 0;
155 }
156
157 /* no page was in place and we have one, so install it */
158
159 memset(mappage, 0, PAGE_SIZE);
160 bitmap->bp[page].map = mappage;
161 bitmap->missing_pages--;
162out:
163 return 0;
164}
165
166
167/* if page is completely empty, put it back on the free list, or dealloc it */
168/* if page was hijacked, unmark the flag so it might get alloced next time */
169/* Note: lock should be held when calling this */
858119e1 170static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
171{
172 char *ptr;
173
174 if (bitmap->bp[page].count) /* page is still busy */
175 return;
176
177 /* page is no longer in use, it can be released */
178
179 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
180 bitmap->bp[page].hijacked = 0;
181 bitmap->bp[page].map = NULL;
182 return;
183 }
184
185 /* normal case, free the page */
186
187#if 0
188/* actually ... let's not. We will probably need the page again exactly when
189 * memory is tight and we are flusing to disk
190 */
191 return;
192#else
193 ptr = bitmap->bp[page].map;
194 bitmap->bp[page].map = NULL;
195 bitmap->missing_pages++;
196 bitmap_free_page(bitmap, ptr);
197 return;
198#endif
199}
200
201
202/*
203 * bitmap file handling - read and write the bitmap file and its superblock
204 */
205
206/* copy the pathname of a file to a buffer */
207char *file_path(struct file *file, char *buf, int count)
208{
32a7627c
N
209 if (!buf)
210 return NULL;
211
cf28b486 212 buf = d_path(&file->f_path, buf, count);
32a7627c
N
213
214 return IS_ERR(buf) ? NULL : buf;
215}
216
217/*
218 * basic page I/O operations
219 */
220
a654b9d8
N
221/* IO operations when bitmap is stored near all superblocks */
222static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
223{
224 /* choose a good rdev and read the page from there */
225
226 mdk_rdev_t *rdev;
227 struct list_head *tmp;
228 struct page *page = alloc_page(GFP_KERNEL);
229 sector_t target;
230
231 if (!page)
232 return ERR_PTR(-ENOMEM);
a654b9d8 233
d089c6af 234 rdev_for_each(rdev, tmp, mddev) {
b2d444d7
N
235 if (! test_bit(In_sync, &rdev->flags)
236 || test_bit(Faulty, &rdev->flags))
ab904d63
N
237 continue;
238
a654b9d8
N
239 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
240
ab904d63
N
241 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
242 page->index = index;
ce25c31b
N
243 attach_page_buffers(page, NULL); /* so that free_buffer will
244 * quietly no-op */
ab904d63
N
245 return page;
246 }
247 }
248 return ERR_PTR(-EIO);
a654b9d8 249
a654b9d8
N
250}
251
ab6085c7 252static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
a654b9d8
N
253{
254 mdk_rdev_t *rdev;
255 struct list_head *tmp;
ab6085c7 256 mddev_t *mddev = bitmap->mddev;
a654b9d8 257
d089c6af 258 rdev_for_each(rdev, tmp, mddev)
b2d444d7 259 if (test_bit(In_sync, &rdev->flags)
ab6085c7
N
260 && !test_bit(Faulty, &rdev->flags)) {
261 int size = PAGE_SIZE;
262 if (page->index == bitmap->file_pages-1)
263 size = roundup(bitmap->last_page_size,
264 bdev_hardsect_size(rdev->bdev));
f0d76d70
N
265 /* Just make sure we aren't corrupting data or
266 * metadata
267 */
268 if (bitmap->offset < 0) {
269 /* DATA BITMAP METADATA */
270 if (bitmap->offset
85bfb4da 271 + (long)(page->index * (PAGE_SIZE/512))
f0d76d70
N
272 + size/512 > 0)
273 /* bitmap runs in to metadata */
274 return -EINVAL;
275 if (rdev->data_offset + mddev->size*2
276 > rdev->sb_offset*2 + bitmap->offset)
277 /* data runs in to bitmap */
278 return -EINVAL;
279 } else if (rdev->sb_offset*2 < rdev->data_offset) {
280 /* METADATA BITMAP DATA */
281 if (rdev->sb_offset*2
282 + bitmap->offset
283 + page->index*(PAGE_SIZE/512) + size/512
284 > rdev->data_offset)
285 /* bitmap runs in to data */
286 return -EINVAL;
287 } else {
288 /* DATA METADATA BITMAP - no problems */
289 }
a654b9d8 290 md_super_write(mddev, rdev,
ab6085c7 291 (rdev->sb_offset<<1) + bitmap->offset
a654b9d8 292 + page->index * (PAGE_SIZE/512),
ab6085c7 293 size,
a654b9d8 294 page);
ab6085c7 295 }
a654b9d8
N
296
297 if (wait)
a9701a30 298 md_super_wait(mddev);
a654b9d8
N
299 return 0;
300}
301
4ad13663 302static void bitmap_file_kick(struct bitmap *bitmap);
32a7627c 303/*
a654b9d8 304 * write out a page to a file
32a7627c 305 */
4ad13663 306static void write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 307{
d785a06a 308 struct buffer_head *bh;
32a7627c 309
f0d76d70
N
310 if (bitmap->file == NULL) {
311 switch (write_sb_page(bitmap, page, wait)) {
312 case -EINVAL:
313 bitmap->flags |= BITMAP_WRITE_ERROR;
f0d76d70 314 }
4ad13663 315 } else {
a654b9d8 316
4ad13663 317 bh = page_buffers(page);
c708443c 318
4ad13663
N
319 while (bh && bh->b_blocknr) {
320 atomic_inc(&bitmap->pending_writes);
321 set_buffer_locked(bh);
322 set_buffer_mapped(bh);
323 submit_bh(WRITE, bh);
324 bh = bh->b_this_page;
325 }
d785a06a 326
4ad13663
N
327 if (wait) {
328 wait_event(bitmap->write_wait,
329 atomic_read(&bitmap->pending_writes)==0);
330 }
8a5e9cf1 331 }
4ad13663
N
332 if (bitmap->flags & BITMAP_WRITE_ERROR)
333 bitmap_file_kick(bitmap);
d785a06a
N
334}
335
336static void end_bitmap_write(struct buffer_head *bh, int uptodate)
337{
338 struct bitmap *bitmap = bh->b_private;
339 unsigned long flags;
32a7627c 340
d785a06a
N
341 if (!uptodate) {
342 spin_lock_irqsave(&bitmap->lock, flags);
343 bitmap->flags |= BITMAP_WRITE_ERROR;
344 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 345 }
d785a06a
N
346 if (atomic_dec_and_test(&bitmap->pending_writes))
347 wake_up(&bitmap->write_wait);
348}
32a7627c 349
d785a06a
N
350/* copied from buffer.c */
351static void
352__clear_page_buffers(struct page *page)
353{
354 ClearPagePrivate(page);
355 set_page_private(page, 0);
356 page_cache_release(page);
357}
358static void free_buffers(struct page *page)
359{
360 struct buffer_head *bh = page_buffers(page);
77ad4bc7 361
d785a06a
N
362 while (bh) {
363 struct buffer_head *next = bh->b_this_page;
364 free_buffer_head(bh);
365 bh = next;
77ad4bc7 366 }
d785a06a
N
367 __clear_page_buffers(page);
368 put_page(page);
32a7627c
N
369}
370
d785a06a
N
371/* read a page from a file.
372 * We both read the page, and attach buffers to the page to record the
373 * address of each block (using bmap). These addresses will be used
374 * to write the block later, completely bypassing the filesystem.
375 * This usage is similar to how swap files are handled, and allows us
376 * to write to a file with no concerns of memory allocation failing.
377 */
32a7627c 378static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
379 struct bitmap *bitmap,
380 unsigned long count)
32a7627c 381{
32a7627c 382 struct page *page = NULL;
c649bb9c 383 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
384 struct buffer_head *bh;
385 sector_t block;
32a7627c 386
2d1f3b5d
N
387 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
388 (unsigned long long)index << PAGE_SHIFT);
32a7627c 389
d785a06a
N
390 page = alloc_page(GFP_KERNEL);
391 if (!page)
392 page = ERR_PTR(-ENOMEM);
32a7627c
N
393 if (IS_ERR(page))
394 goto out;
d785a06a 395
d785a06a
N
396 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
397 if (!bh) {
2d1f3b5d 398 put_page(page);
d785a06a 399 page = ERR_PTR(-ENOMEM);
32a7627c
N
400 goto out;
401 }
d785a06a
N
402 attach_page_buffers(page, bh);
403 block = index << (PAGE_SHIFT - inode->i_blkbits);
404 while (bh) {
405 if (count == 0)
406 bh->b_blocknr = 0;
407 else {
408 bh->b_blocknr = bmap(inode, block);
409 if (bh->b_blocknr == 0) {
410 /* Cannot use this file! */
411 free_buffers(page);
412 page = ERR_PTR(-EINVAL);
413 goto out;
414 }
415 bh->b_bdev = inode->i_sb->s_bdev;
416 if (count < (1<<inode->i_blkbits))
417 count = 0;
418 else
419 count -= (1<<inode->i_blkbits);
420
421 bh->b_end_io = end_bitmap_write;
422 bh->b_private = bitmap;
ce25c31b
N
423 atomic_inc(&bitmap->pending_writes);
424 set_buffer_locked(bh);
425 set_buffer_mapped(bh);
426 submit_bh(READ, bh);
d785a06a
N
427 }
428 block++;
429 bh = bh->b_this_page;
430 }
d785a06a 431 page->index = index;
ce25c31b
N
432
433 wait_event(bitmap->write_wait,
434 atomic_read(&bitmap->pending_writes)==0);
435 if (bitmap->flags & BITMAP_WRITE_ERROR) {
436 free_buffers(page);
437 page = ERR_PTR(-EIO);
438 }
32a7627c
N
439out:
440 if (IS_ERR(page))
441 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
442 (int)PAGE_SIZE,
443 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
444 PTR_ERR(page));
445 return page;
446}
447
448/*
449 * bitmap file superblock operations
450 */
451
452/* update the event counter and sync the superblock to disk */
4ad13663 453void bitmap_update_sb(struct bitmap *bitmap)
32a7627c
N
454{
455 bitmap_super_t *sb;
456 unsigned long flags;
457
458 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
4ad13663 459 return;
32a7627c
N
460 spin_lock_irqsave(&bitmap->lock, flags);
461 if (!bitmap->sb_page) { /* no superblock */
462 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 463 return;
32a7627c 464 }
32a7627c 465 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 466 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
467 sb->events = cpu_to_le64(bitmap->mddev->events);
468 if (!bitmap->mddev->degraded)
469 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
ea03aff9 470 kunmap_atomic(sb, KM_USER0);
4ad13663 471 write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
472}
473
474/* print out the bitmap file superblock */
475void bitmap_print_sb(struct bitmap *bitmap)
476{
477 bitmap_super_t *sb;
478
479 if (!bitmap || !bitmap->sb_page)
480 return;
ea03aff9 481 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 482 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
483 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
484 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
485 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
486 *(__u32 *)(sb->uuid+0),
487 *(__u32 *)(sb->uuid+4),
488 *(__u32 *)(sb->uuid+8),
489 *(__u32 *)(sb->uuid+12));
a2cff26a 490 printk(KERN_DEBUG " events: %llu\n",
32a7627c 491 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 492 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 493 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
494 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
495 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
496 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
497 printk(KERN_DEBUG " sync size: %llu KB\n",
498 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 499 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 500 kunmap_atomic(sb, KM_USER0);
32a7627c
N
501}
502
503/* read the superblock from the bitmap file and initialize some bitmap fields */
504static int bitmap_read_sb(struct bitmap *bitmap)
505{
506 char *reason = NULL;
507 bitmap_super_t *sb;
4b6d287f 508 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
509 unsigned long long events;
510 int err = -EINVAL;
511
512 /* page 0 is the superblock, read it... */
f49d5e62
N
513 if (bitmap->file) {
514 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
515 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
516
517 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
518 } else {
a654b9d8 519 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 520 }
32a7627c
N
521 if (IS_ERR(bitmap->sb_page)) {
522 err = PTR_ERR(bitmap->sb_page);
523 bitmap->sb_page = NULL;
524 return err;
525 }
526
ea03aff9 527 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 528
32a7627c
N
529 chunksize = le32_to_cpu(sb->chunksize);
530 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 531 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
532
533 /* verify that the bitmap-specific fields are valid */
534 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
535 reason = "bad magic";
bd926c63
N
536 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
537 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 538 reason = "unrecognized superblock version";
7dd5d34c
N
539 else if (chunksize < PAGE_SIZE)
540 reason = "bitmap chunksize too small";
32a7627c
N
541 else if ((1 << ffz(~chunksize)) != chunksize)
542 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
543 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
544 reason = "daemon sleep period out of range";
4b6d287f
N
545 else if (write_behind > COUNTER_MAX)
546 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
547 if (reason) {
548 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
549 bmname(bitmap), reason);
550 goto out;
551 }
552
553 /* keep the array size field of the bitmap superblock up to date */
554 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
555
556 if (!bitmap->mddev->persistent)
557 goto success;
558
559 /*
560 * if we have a persistent array superblock, compare the
561 * bitmap's UUID and event counter to the mddev's
562 */
563 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
564 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
565 bmname(bitmap));
566 goto out;
567 }
568 events = le64_to_cpu(sb->events);
569 if (events < bitmap->mddev->events) {
570 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
571 "-- forcing full recovery\n", bmname(bitmap), events,
572 (unsigned long long) bitmap->mddev->events);
4f2e639a 573 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
574 }
575success:
576 /* assign fields using values from superblock */
577 bitmap->chunksize = chunksize;
578 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 579 bitmap->daemon_lastrun = jiffies;
4b6d287f 580 bitmap->max_write_behind = write_behind;
4f2e639a 581 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
582 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
583 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 584 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 585 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 586 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
587 err = 0;
588out:
ea03aff9 589 kunmap_atomic(sb, KM_USER0);
32a7627c
N
590 if (err)
591 bitmap_print_sb(bitmap);
592 return err;
593}
594
595enum bitmap_mask_op {
596 MASK_SET,
597 MASK_UNSET
598};
599
4ad13663
N
600/* record the state of the bitmap in the superblock. Return the old value */
601static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
602 enum bitmap_mask_op op)
32a7627c
N
603{
604 bitmap_super_t *sb;
605 unsigned long flags;
4ad13663 606 int old;
32a7627c
N
607
608 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 609 if (!bitmap->sb_page) { /* can't set the state */
32a7627c 610 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 611 return 0;
32a7627c 612 }
32a7627c 613 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 614 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
4ad13663 615 old = le32_to_cpu(sb->state) & bits;
32a7627c 616 switch (op) {
4f2e639a 617 case MASK_SET: sb->state |= cpu_to_le32(bits);
32a7627c 618 break;
4f2e639a 619 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
32a7627c
N
620 break;
621 default: BUG();
622 }
ea03aff9 623 kunmap_atomic(sb, KM_USER0);
4ad13663 624 return old;
32a7627c
N
625}
626
627/*
628 * general bitmap file operations
629 */
630
631/* calculate the index of the page that contains this bit */
632static inline unsigned long file_page_index(unsigned long chunk)
633{
634 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
635}
636
637/* calculate the (bit) offset of this bit within a page */
638static inline unsigned long file_page_offset(unsigned long chunk)
639{
640 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
641}
642
643/*
644 * return a pointer to the page in the filemap that contains the given bit
645 *
646 * this lookup is complicated by the fact that the bitmap sb might be exactly
647 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
648 * 0 or page 1
649 */
650static inline struct page *filemap_get_page(struct bitmap *bitmap,
651 unsigned long chunk)
652{
9b1d1dac 653 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
32a7627c
N
654 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
655}
656
657
658static void bitmap_file_unmap(struct bitmap *bitmap)
659{
660 struct page **map, *sb_page;
661 unsigned long *attr;
662 int pages;
663 unsigned long flags;
664
665 spin_lock_irqsave(&bitmap->lock, flags);
666 map = bitmap->filemap;
667 bitmap->filemap = NULL;
668 attr = bitmap->filemap_attr;
669 bitmap->filemap_attr = NULL;
670 pages = bitmap->file_pages;
671 bitmap->file_pages = 0;
672 sb_page = bitmap->sb_page;
673 bitmap->sb_page = NULL;
674 spin_unlock_irqrestore(&bitmap->lock, flags);
675
676 while (pages--)
677 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
d785a06a 678 free_buffers(map[pages]);
32a7627c
N
679 kfree(map);
680 kfree(attr);
681
d785a06a
N
682 if (sb_page)
683 free_buffers(sb_page);
32a7627c
N
684}
685
686static void bitmap_file_put(struct bitmap *bitmap)
687{
688 struct file *file;
32a7627c
N
689 unsigned long flags;
690
691 spin_lock_irqsave(&bitmap->lock, flags);
692 file = bitmap->file;
693 bitmap->file = NULL;
694 spin_unlock_irqrestore(&bitmap->lock, flags);
695
d785a06a
N
696 if (file)
697 wait_event(bitmap->write_wait,
698 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
699 bitmap_file_unmap(bitmap);
700
d785a06a 701 if (file) {
c649bb9c 702 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 703 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 704 fput(file);
d785a06a 705 }
32a7627c
N
706}
707
708
709/*
710 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
711 * then it is no longer reliable, so we stop using it and we mark the file
712 * as failed in the superblock
713 */
714static void bitmap_file_kick(struct bitmap *bitmap)
715{
716 char *path, *ptr = NULL;
717
4ad13663
N
718 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
719 bitmap_update_sb(bitmap);
32a7627c 720
4ad13663
N
721 if (bitmap->file) {
722 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
723 if (path)
724 ptr = file_path(bitmap->file, path, PAGE_SIZE);
32a7627c 725
4ad13663
N
726 printk(KERN_ALERT
727 "%s: kicking failed bitmap file %s from array!\n",
728 bmname(bitmap), ptr ? ptr : "");
32a7627c 729
4ad13663
N
730 kfree(path);
731 } else
732 printk(KERN_ALERT
733 "%s: disabling internal bitmap due to errors\n",
734 bmname(bitmap));
a654b9d8 735 }
32a7627c
N
736
737 bitmap_file_put(bitmap);
738
739 return;
740}
741
742enum bitmap_page_attr {
e16b68b6
N
743 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
744 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
745 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
746};
747
748static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
749 enum bitmap_page_attr attr)
750{
e16b68b6 751 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
752}
753
754static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
755 enum bitmap_page_attr attr)
756{
e16b68b6 757 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
758}
759
ec7a3197
N
760static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
761 enum bitmap_page_attr attr)
32a7627c 762{
e16b68b6 763 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
764}
765
766/*
767 * bitmap_file_set_bit -- called before performing a write to the md device
768 * to set (and eventually sync) a particular bit in the bitmap file
769 *
770 * we set the bit immediately, then we record the page number so that
771 * when an unplug occurs, we can flush the dirty pages out to disk
772 */
773static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
774{
775 unsigned long bit;
776 struct page *page;
777 void *kaddr;
778 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
779
a654b9d8 780 if (!bitmap->filemap) {
32a7627c
N
781 return;
782 }
783
784 page = filemap_get_page(bitmap, chunk);
9b1d1dac 785 if (!page) return;
32a7627c
N
786 bit = file_page_offset(chunk);
787
32a7627c
N
788 /* set the bit */
789 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
790 if (bitmap->flags & BITMAP_HOSTENDIAN)
791 set_bit(bit, kaddr);
792 else
793 ext2_set_bit(bit, kaddr);
32a7627c
N
794 kunmap_atomic(kaddr, KM_USER0);
795 PRINTK("set file bit %lu page %lu\n", bit, page->index);
796
797 /* record page number so it gets flushed to disk when unplug occurs */
798 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
799
800}
801
802/* this gets called when the md device is ready to unplug its underlying
803 * (slave) device queues -- before we let any writes go down, we need to
804 * sync the dirty pages of the bitmap file to disk */
4ad13663 805void bitmap_unplug(struct bitmap *bitmap)
32a7627c 806{
ec7a3197
N
807 unsigned long i, flags;
808 int dirty, need_write;
32a7627c
N
809 struct page *page;
810 int wait = 0;
811
812 if (!bitmap)
4ad13663 813 return;
32a7627c
N
814
815 /* look at each page to see if there are any set bits that need to be
816 * flushed out to disk */
817 for (i = 0; i < bitmap->file_pages; i++) {
818 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 819 if (!bitmap->filemap) {
32a7627c 820 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 821 return;
32a7627c
N
822 }
823 page = bitmap->filemap[i];
ec7a3197
N
824 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
825 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
826 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
827 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 828 if (dirty)
32a7627c
N
829 wait = 1;
830 spin_unlock_irqrestore(&bitmap->lock, flags);
831
d785a06a 832 if (dirty | need_write)
4ad13663 833 write_page(bitmap, page, 0);
32a7627c
N
834 }
835 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 836 if (bitmap->file)
d785a06a
N
837 wait_event(bitmap->write_wait,
838 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 839 else
a9701a30 840 md_super_wait(bitmap->mddev);
32a7627c 841 }
d785a06a
N
842 if (bitmap->flags & BITMAP_WRITE_ERROR)
843 bitmap_file_kick(bitmap);
32a7627c
N
844}
845
6a07997f 846static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
847/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
848 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
849 * memory mapping of the bitmap file
850 * Special cases:
851 * if there's no bitmap file, or if the bitmap file had been
852 * previously kicked from the array, we mark all the bits as
853 * 1's in order to cause a full resync.
6a07997f
N
854 *
855 * We ignore all bits for sectors that end earlier than 'start'.
856 * This is used when reading an out-of-date bitmap...
32a7627c 857 */
6a07997f 858static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
859{
860 unsigned long i, chunks, index, oldindex, bit;
861 struct page *page = NULL, *oldpage = NULL;
862 unsigned long num_pages, bit_cnt = 0;
863 struct file *file;
d785a06a 864 unsigned long bytes, offset;
32a7627c
N
865 int outofdate;
866 int ret = -ENOSPC;
ea03aff9 867 void *paddr;
32a7627c
N
868
869 chunks = bitmap->chunks;
870 file = bitmap->file;
871
a654b9d8 872 BUG_ON(!file && !bitmap->offset);
32a7627c 873
44456d37 874#ifdef INJECT_FAULTS_3
32a7627c
N
875 outofdate = 1;
876#else
877 outofdate = bitmap->flags & BITMAP_STALE;
878#endif
879 if (outofdate)
880 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
881 "recovery\n", bmname(bitmap));
882
883 bytes = (chunks + 7) / 8;
bc7f77de 884
cdbb4cc2 885 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 886
a654b9d8 887 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
888 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
889 bmname(bitmap),
890 (unsigned long) i_size_read(file->f_mapping->host),
891 bytes + sizeof(bitmap_super_t));
4ad13663 892 goto err;
32a7627c 893 }
bc7f77de
N
894
895 ret = -ENOMEM;
896
32a7627c 897 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 898 if (!bitmap->filemap)
4ad13663 899 goto err;
32a7627c 900
e16b68b6
N
901 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
902 bitmap->filemap_attr = kzalloc(
505fa2c4 903 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 904 GFP_KERNEL);
bc7f77de 905 if (!bitmap->filemap_attr)
4ad13663 906 goto err;
32a7627c 907
32a7627c
N
908 oldindex = ~0L;
909
910 for (i = 0; i < chunks; i++) {
bd926c63 911 int b;
32a7627c
N
912 index = file_page_index(i);
913 bit = file_page_offset(i);
914 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 915 int count;
32a7627c 916 /* unmap the old page, we're done with it */
d785a06a 917 if (index == num_pages-1)
f49d5e62
N
918 count = bytes + sizeof(bitmap_super_t)
919 - index * PAGE_SIZE;
d785a06a
N
920 else
921 count = PAGE_SIZE;
32a7627c
N
922 if (index == 0) {
923 /*
924 * if we're here then the superblock page
925 * contains some bits (PAGE_SIZE != sizeof sb)
926 * we've already read it in, so just use it
927 */
928 page = bitmap->sb_page;
929 offset = sizeof(bitmap_super_t);
a654b9d8 930 } else if (file) {
d785a06a 931 page = read_page(file, index, bitmap, count);
a654b9d8
N
932 offset = 0;
933 } else {
934 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
935 offset = 0;
936 }
a654b9d8
N
937 if (IS_ERR(page)) { /* read error */
938 ret = PTR_ERR(page);
4ad13663 939 goto err;
a654b9d8
N
940 }
941
32a7627c
N
942 oldindex = index;
943 oldpage = page;
32a7627c
N
944
945 if (outofdate) {
946 /*
947 * if bitmap is out of date, dirty the
948 * whole page and write it out
949 */
ea03aff9
N
950 paddr = kmap_atomic(page, KM_USER0);
951 memset(paddr + offset, 0xff,
6a07997f 952 PAGE_SIZE - offset);
ea03aff9 953 kunmap_atomic(paddr, KM_USER0);
4ad13663
N
954 write_page(bitmap, page, 1);
955
956 ret = -EIO;
957 if (bitmap->flags & BITMAP_WRITE_ERROR) {
32a7627c 958 /* release, page not in filemap yet */
2d1f3b5d 959 put_page(page);
4ad13663 960 goto err;
32a7627c
N
961 }
962 }
963
964 bitmap->filemap[bitmap->file_pages++] = page;
ab6085c7 965 bitmap->last_page_size = count;
32a7627c 966 }
ea03aff9 967 paddr = kmap_atomic(page, KM_USER0);
bd926c63 968 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 969 b = test_bit(bit, paddr);
bd926c63 970 else
ea03aff9
N
971 b = ext2_test_bit(bit, paddr);
972 kunmap_atomic(paddr, KM_USER0);
bd926c63 973 if (b) {
32a7627c 974 /* if the disk bit is set, set the memory bit */
6a07997f
N
975 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
976 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
977 );
32a7627c 978 bit_cnt++;
6a07997f 979 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 980 }
32a7627c
N
981 }
982
983 /* everything went OK */
984 ret = 0;
985 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
986
32a7627c
N
987 if (bit_cnt) { /* Kick recovery if any bits were set */
988 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
989 md_wakeup_thread(bitmap->mddev->thread);
990 }
991
32a7627c 992 printk(KERN_INFO "%s: bitmap initialized from disk: "
4ad13663
N
993 "read %lu/%lu pages, set %lu bits\n",
994 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
995
996 return 0;
32a7627c 997
4ad13663
N
998 err:
999 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1000 bmname(bitmap), ret);
32a7627c
N
1001 return ret;
1002}
1003
a654b9d8
N
1004void bitmap_write_all(struct bitmap *bitmap)
1005{
1006 /* We don't actually write all bitmap blocks here,
1007 * just flag them as needing to be written
1008 */
ec7a3197 1009 int i;
a654b9d8 1010
ec7a3197
N
1011 for (i=0; i < bitmap->file_pages; i++)
1012 set_page_attr(bitmap, bitmap->filemap[i],
1013 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
1014}
1015
32a7627c
N
1016
1017static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1018{
1019 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1020 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1021 bitmap->bp[page].count += inc;
1022/*
1023 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1024 (unsigned long long)offset, inc, bitmap->bp[page].count);
1025*/
1026 bitmap_checkfree(bitmap, page);
1027}
1028static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1029 sector_t offset, int *blocks,
1030 int create);
1031
1032/*
1033 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1034 * out to disk
1035 */
1036
4ad13663 1037void bitmap_daemon_work(struct bitmap *bitmap)
32a7627c 1038{
aa3163f8 1039 unsigned long j;
32a7627c
N
1040 unsigned long flags;
1041 struct page *page = NULL, *lastpage = NULL;
32a7627c 1042 int blocks;
ea03aff9 1043 void *paddr;
32a7627c
N
1044
1045 if (bitmap == NULL)
4ad13663 1046 return;
32a7627c 1047 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
4ad13663 1048 return;
32a7627c
N
1049 bitmap->daemon_lastrun = jiffies;
1050
1051 for (j = 0; j < bitmap->chunks; j++) {
1052 bitmap_counter_t *bmc;
1053 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1054 if (!bitmap->filemap) {
32a7627c
N
1055 /* error or shutdown */
1056 spin_unlock_irqrestore(&bitmap->lock, flags);
1057 break;
1058 }
1059
1060 page = filemap_get_page(bitmap, j);
32a7627c
N
1061
1062 if (page != lastpage) {
aa3163f8 1063 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1064 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1065 int need_write = test_page_attr(bitmap, page,
1066 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1067 if (need_write)
aa3163f8 1068 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1069
aa3163f8 1070 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663
N
1071 if (need_write)
1072 write_page(bitmap, page, 0);
aa3163f8
N
1073 continue;
1074 }
1075
32a7627c 1076 /* grab the new page, sync and release the old */
32a7627c 1077 if (lastpage != NULL) {
ec7a3197 1078 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1079 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1080 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1081 write_page(bitmap, lastpage, 0);
32a7627c
N
1082 } else {
1083 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1084 spin_unlock_irqrestore(&bitmap->lock, flags);
1085 }
32a7627c
N
1086 } else
1087 spin_unlock_irqrestore(&bitmap->lock, flags);
1088 lastpage = page;
32a7627c
N
1089/*
1090 printk("bitmap clean at page %lu\n", j);
1091*/
1092 spin_lock_irqsave(&bitmap->lock, flags);
1093 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1094 }
1095 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1096 &blocks, 0);
1097 if (bmc) {
1098/*
1099 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1100*/
1101 if (*bmc == 2) {
1102 *bmc=1; /* maybe clear the bit next time */
1103 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1104 } else if (*bmc == 1) {
1105 /* we can clear the bit */
1106 *bmc = 0;
1107 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1108 -1);
1109
1110 /* clear the bit */
ea03aff9 1111 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1112 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1113 clear_bit(file_page_offset(j), paddr);
bd926c63 1114 else
ea03aff9
N
1115 ext2_clear_bit(file_page_offset(j), paddr);
1116 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1117 }
1118 }
1119 spin_unlock_irqrestore(&bitmap->lock, flags);
1120 }
1121
1122 /* now sync the final page */
1123 if (lastpage != NULL) {
32a7627c 1124 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1125 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1126 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1127 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1128 write_page(bitmap, lastpage, 0);
32a7627c
N
1129 } else {
1130 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1131 spin_unlock_irqrestore(&bitmap->lock, flags);
1132 }
32a7627c
N
1133 }
1134
32a7627c
N
1135}
1136
32a7627c
N
1137static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1138 sector_t offset, int *blocks,
1139 int create)
1140{
1141 /* If 'create', we might release the lock and reclaim it.
1142 * The lock must have been taken with interrupts enabled.
1143 * If !create, we don't release the lock.
1144 */
1145 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1146 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1147 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1148 sector_t csize;
1149
1150 if (bitmap_checkpage(bitmap, page, create) < 0) {
1151 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1152 *blocks = csize - (offset & (csize- 1));
1153 return NULL;
1154 }
1155 /* now locked ... */
1156
1157 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1158 /* should we use the first or second counter field
1159 * of the hijacked pointer? */
1160 int hi = (pageoff > PAGE_COUNTER_MASK);
1161 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1162 PAGE_COUNTER_SHIFT - 1);
1163 *blocks = csize - (offset & (csize- 1));
1164 return &((bitmap_counter_t *)
1165 &bitmap->bp[page].map)[hi];
1166 } else { /* page is allocated */
1167 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1168 *blocks = csize - (offset & (csize- 1));
1169 return (bitmap_counter_t *)
1170 &(bitmap->bp[page].map[pageoff]);
1171 }
1172}
1173
4b6d287f 1174int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1175{
1176 if (!bitmap) return 0;
4b6d287f
N
1177
1178 if (behind) {
1179 atomic_inc(&bitmap->behind_writes);
1180 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1181 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1182 }
1183
32a7627c
N
1184 while (sectors) {
1185 int blocks;
1186 bitmap_counter_t *bmc;
1187
1188 spin_lock_irq(&bitmap->lock);
1189 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1190 if (!bmc) {
1191 spin_unlock_irq(&bitmap->lock);
1192 return 0;
1193 }
1194
da6e1a32
NB
1195 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1196 DEFINE_WAIT(__wait);
1197 /* note that it is safe to do the prepare_to_wait
1198 * after the test as long as we do it before dropping
1199 * the spinlock.
1200 */
1201 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1202 TASK_UNINTERRUPTIBLE);
1203 spin_unlock_irq(&bitmap->lock);
2ad8b1ef 1204 blk_unplug(bitmap->mddev->queue);
da6e1a32
NB
1205 schedule();
1206 finish_wait(&bitmap->overflow_wait, &__wait);
1207 continue;
1208 }
1209
32a7627c
N
1210 switch(*bmc) {
1211 case 0:
1212 bitmap_file_set_bit(bitmap, offset);
1213 bitmap_count_page(bitmap,offset, 1);
1214 blk_plug_device(bitmap->mddev->queue);
1215 /* fall through */
1216 case 1:
1217 *bmc = 2;
1218 }
da6e1a32 1219
32a7627c
N
1220 (*bmc)++;
1221
1222 spin_unlock_irq(&bitmap->lock);
1223
1224 offset += blocks;
1225 if (sectors > blocks)
1226 sectors -= blocks;
1227 else sectors = 0;
1228 }
1229 return 0;
1230}
1231
1232void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1233 int success, int behind)
32a7627c
N
1234{
1235 if (!bitmap) return;
4b6d287f
N
1236 if (behind) {
1237 atomic_dec(&bitmap->behind_writes);
1238 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1239 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1240 }
1241
32a7627c
N
1242 while (sectors) {
1243 int blocks;
1244 unsigned long flags;
1245 bitmap_counter_t *bmc;
1246
1247 spin_lock_irqsave(&bitmap->lock, flags);
1248 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1249 if (!bmc) {
1250 spin_unlock_irqrestore(&bitmap->lock, flags);
1251 return;
1252 }
1253
1254 if (!success && ! (*bmc & NEEDED_MASK))
1255 *bmc |= NEEDED_MASK;
1256
da6e1a32
NB
1257 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1258 wake_up(&bitmap->overflow_wait);
1259
32a7627c
N
1260 (*bmc)--;
1261 if (*bmc <= 2) {
1262 set_page_attr(bitmap,
1263 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1264 BITMAP_PAGE_CLEAN);
1265 }
1266 spin_unlock_irqrestore(&bitmap->lock, flags);
1267 offset += blocks;
1268 if (sectors > blocks)
1269 sectors -= blocks;
1270 else sectors = 0;
1271 }
1272}
1273
6a806c51
N
1274int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1275 int degraded)
32a7627c
N
1276{
1277 bitmap_counter_t *bmc;
1278 int rv;
1279 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1280 *blocks = 1024;
1281 return 1; /* always resync if no bitmap */
1282 }
1283 spin_lock_irq(&bitmap->lock);
1284 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1285 rv = 0;
1286 if (bmc) {
1287 /* locked */
1288 if (RESYNC(*bmc))
1289 rv = 1;
1290 else if (NEEDED(*bmc)) {
1291 rv = 1;
6a806c51
N
1292 if (!degraded) { /* don't set/clear bits if degraded */
1293 *bmc |= RESYNC_MASK;
1294 *bmc &= ~NEEDED_MASK;
1295 }
32a7627c
N
1296 }
1297 }
1298 spin_unlock_irq(&bitmap->lock);
1299 return rv;
1300}
1301
1302void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1303{
1304 bitmap_counter_t *bmc;
1305 unsigned long flags;
1306/*
1307 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1308*/ if (bitmap == NULL) {
1309 *blocks = 1024;
1310 return;
1311 }
1312 spin_lock_irqsave(&bitmap->lock, flags);
1313 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1314 if (bmc == NULL)
1315 goto unlock;
1316 /* locked */
1317/*
1318 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1319*/
1320 if (RESYNC(*bmc)) {
1321 *bmc &= ~RESYNC_MASK;
1322
1323 if (!NEEDED(*bmc) && aborted)
1324 *bmc |= NEEDED_MASK;
1325 else {
1326 if (*bmc <= 2) {
1327 set_page_attr(bitmap,
1328 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1329 BITMAP_PAGE_CLEAN);
1330 }
1331 }
1332 }
1333 unlock:
1334 spin_unlock_irqrestore(&bitmap->lock, flags);
1335}
1336
1337void bitmap_close_sync(struct bitmap *bitmap)
1338{
1339 /* Sync has finished, and any bitmap chunks that weren't synced
1340 * properly have been aborted. It remains to us to clear the
1341 * RESYNC bit wherever it is still on
1342 */
1343 sector_t sector = 0;
1344 int blocks;
b47490c9
N
1345 if (!bitmap)
1346 return;
32a7627c
N
1347 while (sector < bitmap->mddev->resync_max_sectors) {
1348 bitmap_end_sync(bitmap, sector, &blocks, 0);
b47490c9
N
1349 sector += blocks;
1350 }
1351}
1352
1353void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1354{
1355 sector_t s = 0;
1356 int blocks;
1357
1358 if (!bitmap)
1359 return;
1360 if (sector == 0) {
1361 bitmap->last_end_sync = jiffies;
1362 return;
1363 }
1364 if (time_before(jiffies, (bitmap->last_end_sync
1365 + bitmap->daemon_sleep * HZ)))
1366 return;
1367 wait_event(bitmap->mddev->recovery_wait,
1368 atomic_read(&bitmap->mddev->recovery_active) == 0);
1369
1370 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1371 s = 0;
1372 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1373 bitmap_end_sync(bitmap, s, &blocks, 0);
1374 s += blocks;
32a7627c 1375 }
b47490c9 1376 bitmap->last_end_sync = jiffies;
32a7627c
N
1377}
1378
6a07997f 1379static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1380{
1381 /* For each chunk covered by any of these sectors, set the
193f1c93 1382 * counter to 1 and set resync_needed. They should all
32a7627c
N
1383 * be 0 at this point
1384 */
193f1c93
N
1385
1386 int secs;
1387 bitmap_counter_t *bmc;
1388 spin_lock_irq(&bitmap->lock);
1389 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1390 if (!bmc) {
32a7627c 1391 spin_unlock_irq(&bitmap->lock);
193f1c93 1392 return;
32a7627c 1393 }
193f1c93
N
1394 if (! *bmc) {
1395 struct page *page;
6a07997f 1396 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1397 bitmap_count_page(bitmap, offset, 1);
1398 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1399 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1400 }
1401 spin_unlock_irq(&bitmap->lock);
1402
32a7627c
N
1403}
1404
9b1d1dac
PC
1405/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1406void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1407{
1408 unsigned long chunk;
1409
1410 for (chunk = s; chunk <= e; chunk++) {
1411 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1412 bitmap_set_memory_bits(bitmap, sec, 1);
1413 bitmap_file_set_bit(bitmap, sec);
1414 }
1415}
1416
6b8b3e8a
N
1417/*
1418 * flush out any pending updates
1419 */
1420void bitmap_flush(mddev_t *mddev)
1421{
1422 struct bitmap *bitmap = mddev->bitmap;
1423 int sleep;
1424
1425 if (!bitmap) /* there was no bitmap */
1426 return;
1427
1428 /* run the daemon_work three time to ensure everything is flushed
1429 * that can be
1430 */
1431 sleep = bitmap->daemon_sleep;
1432 bitmap->daemon_sleep = 0;
1433 bitmap_daemon_work(bitmap);
1434 bitmap_daemon_work(bitmap);
1435 bitmap_daemon_work(bitmap);
1436 bitmap->daemon_sleep = sleep;
1437 bitmap_update_sb(bitmap);
1438}
1439
32a7627c
N
1440/*
1441 * free memory that was allocated
1442 */
3178b0db 1443static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1444{
1445 unsigned long k, pages;
1446 struct bitmap_page *bp;
32a7627c
N
1447
1448 if (!bitmap) /* there was no bitmap */
1449 return;
1450
32a7627c
N
1451 /* release the bitmap file and kill the daemon */
1452 bitmap_file_put(bitmap);
1453
1454 bp = bitmap->bp;
1455 pages = bitmap->pages;
1456
1457 /* free all allocated memory */
1458
32a7627c
N
1459 if (bp) /* deallocate the page memory */
1460 for (k = 0; k < pages; k++)
1461 if (bp[k].map && !bp[k].hijacked)
1462 kfree(bp[k].map);
1463 kfree(bp);
1464 kfree(bitmap);
1465}
3178b0db
N
1466void bitmap_destroy(mddev_t *mddev)
1467{
1468 struct bitmap *bitmap = mddev->bitmap;
1469
1470 if (!bitmap) /* there was no bitmap */
1471 return;
1472
1473 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1474 if (mddev->thread)
1475 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1476
1477 bitmap_free(bitmap);
1478}
32a7627c
N
1479
1480/*
1481 * initialize the bitmap structure
1482 * if this returns an error, bitmap_destroy must be called to do clean up
1483 */
1484int bitmap_create(mddev_t *mddev)
1485{
1486 struct bitmap *bitmap;
1487 unsigned long blocks = mddev->resync_max_sectors;
1488 unsigned long chunks;
1489 unsigned long pages;
1490 struct file *file = mddev->bitmap_file;
1491 int err;
6a07997f 1492 sector_t start;
32a7627c 1493
5f6e3c83 1494 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1495
a654b9d8 1496 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1497 return 0;
1498
a654b9d8
N
1499 BUG_ON(file && mddev->bitmap_offset);
1500
9ffae0cf 1501 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1502 if (!bitmap)
1503 return -ENOMEM;
1504
32a7627c 1505 spin_lock_init(&bitmap->lock);
ce25c31b
N
1506 atomic_set(&bitmap->pending_writes, 0);
1507 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1508 init_waitqueue_head(&bitmap->overflow_wait);
ce25c31b 1509
32a7627c 1510 bitmap->mddev = mddev;
32a7627c 1511
32a7627c 1512 bitmap->file = file;
a654b9d8 1513 bitmap->offset = mddev->bitmap_offset;
ce25c31b
N
1514 if (file) {
1515 get_file(file);
ef51c976
MF
1516 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1517 SYNC_FILE_RANGE_WAIT_BEFORE |
1518 SYNC_FILE_RANGE_WRITE |
1519 SYNC_FILE_RANGE_WAIT_AFTER);
ce25c31b 1520 }
32a7627c
N
1521 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1522 err = bitmap_read_sb(bitmap);
1523 if (err)
3178b0db 1524 goto error;
32a7627c 1525
a638b2dc 1526 bitmap->chunkshift = ffz(~bitmap->chunksize);
32a7627c
N
1527
1528 /* now that chunksize and chunkshift are set, we can use these macros */
1529 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1530 CHUNK_BLOCK_RATIO(bitmap);
1531 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1532
1533 BUG_ON(!pages);
1534
1535 bitmap->chunks = chunks;
1536 bitmap->pages = pages;
1537 bitmap->missing_pages = pages;
1538 bitmap->counter_bits = COUNTER_BITS;
1539
1540 bitmap->syncchunk = ~0UL;
1541
44456d37 1542#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1543 bitmap->bp = NULL;
1544#else
9ffae0cf 1545 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1546#endif
3178b0db 1547 err = -ENOMEM;
32a7627c 1548 if (!bitmap->bp)
3178b0db 1549 goto error;
32a7627c 1550
32a7627c
N
1551 /* now that we have some pages available, initialize the in-memory
1552 * bitmap from the on-disk bitmap */
6a07997f
N
1553 start = 0;
1554 if (mddev->degraded == 0
1555 || bitmap->events_cleared == mddev->events)
1556 /* no need to keep dirty bits to optimise a re-add of a missing device */
1557 start = mddev->recovery_cp;
1558 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1559
32a7627c 1560 if (err)
3178b0db 1561 goto error;
32a7627c
N
1562
1563 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1564 pages, bmname(bitmap));
1565
3178b0db
N
1566 mddev->bitmap = bitmap;
1567
b15c2e57
N
1568 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1569
4ad13663
N
1570 bitmap_update_sb(bitmap);
1571
1572 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
3178b0db
N
1573
1574 error:
1575 bitmap_free(bitmap);
1576 return err;
32a7627c
N
1577}
1578
1579/* the bitmap API -- for raid personalities */
1580EXPORT_SYMBOL(bitmap_startwrite);
1581EXPORT_SYMBOL(bitmap_endwrite);
1582EXPORT_SYMBOL(bitmap_start_sync);
1583EXPORT_SYMBOL(bitmap_end_sync);
1584EXPORT_SYMBOL(bitmap_unplug);
1585EXPORT_SYMBOL(bitmap_close_sync);
b47490c9 1586EXPORT_SYMBOL(bitmap_cond_end_sync);