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