]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/md/bitmap.c
Merge branch 'for-linus' of git://git.open-osd.org/linux-open-osd
[net-next-2.6.git] / drivers / md / bitmap.c
CommitLineData
32a7627c
N
1/*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
32a7627c
N
10 */
11
12/*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
32a7627c
N
16 */
17
bff61975 18#include <linux/blkdev.h>
32a7627c 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>
43b2e5d8 29#include "md.h"
ef740c37 30#include "bitmap.h"
32a7627c 31
e384e585 32#include <linux/dm-dirty-log.h>
32a7627c
N
33/* debug macros */
34
35#define DEBUG 0
36
37#if DEBUG
38/* these are for debugging purposes only! */
39
40/* define one and only one of these */
41#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
42#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
43#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
44#define INJECT_FAULTS_4 0 /* undef */
45#define INJECT_FAULTS_5 0 /* undef */
46#define INJECT_FAULTS_6 0
47
48/* if these are defined, the driver will fail! debug only */
49#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
50#define INJECT_FATAL_FAULT_2 0 /* undef */
51#define INJECT_FATAL_FAULT_3 0 /* undef */
52#endif
53
32a7627c
N
54#ifndef PRINTK
55# if DEBUG > 0
56# define PRINTK(x...) printk(KERN_DEBUG x)
57# else
58# define PRINTK(x...)
59# endif
60#endif
61
ac2f40be 62static inline char *bmname(struct bitmap *bitmap)
32a7627c
N
63{
64 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
65}
66
32a7627c
N
67/*
68 * just a placeholder - calls kmalloc for bitmap pages
69 */
70static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
71{
72 unsigned char *page;
73
44456d37 74#ifdef INJECT_FAULTS_1
32a7627c
N
75 page = NULL;
76#else
ac2f40be 77 page = kzalloc(PAGE_SIZE, GFP_NOIO);
32a7627c
N
78#endif
79 if (!page)
80 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
81 else
a654b9d8 82 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
83 bmname(bitmap), page);
84 return page;
85}
86
87/*
88 * for now just a placeholder -- just calls kfree for bitmap pages
89 */
90static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
91{
92 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
93 kfree(page);
94}
95
96/*
97 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
98 *
99 * 1) check to see if this page is allocated, if it's not then try to alloc
100 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
101 * page pointer directly as a counter
102 *
103 * if we find our page, we increment the page's refcount so that it stays
104 * allocated while we're using it
105 */
ac2f40be
N
106static int bitmap_checkpage(struct bitmap *bitmap,
107 unsigned long page, int create)
ee305ace
N
108__releases(bitmap->lock)
109__acquires(bitmap->lock)
32a7627c
N
110{
111 unsigned char *mappage;
112
113 if (page >= bitmap->pages) {
1187cf0a
N
114 /* This can happen if bitmap_start_sync goes beyond
115 * End-of-device while looking for a whole page.
116 * It is harmless.
117 */
32a7627c
N
118 return -EINVAL;
119 }
120
32a7627c
N
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
32a7627c
N
130 /* this page has not been allocated yet */
131
ac2f40be
N
132 spin_unlock_irq(&bitmap->lock);
133 mappage = bitmap_alloc_page(bitmap);
134 spin_lock_irq(&bitmap->lock);
135
136 if (mappage == NULL) {
32a7627c
N
137 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
138 bmname(bitmap));
139 /* failed - set the hijacked flag so that we can use the
140 * pointer as a counter */
32a7627c
N
141 if (!bitmap->bp[page].map)
142 bitmap->bp[page].hijacked = 1;
ac2f40be
N
143 } else if (bitmap->bp[page].map ||
144 bitmap->bp[page].hijacked) {
32a7627c
N
145 /* somebody beat us to getting the page */
146 bitmap_free_page(bitmap, mappage);
147 return 0;
ac2f40be 148 } else {
32a7627c 149
ac2f40be 150 /* no page was in place and we have one, so install it */
32a7627c 151
ac2f40be
N
152 bitmap->bp[page].map = mappage;
153 bitmap->missing_pages--;
154 }
32a7627c
N
155 return 0;
156}
157
32a7627c
N
158/* if page is completely empty, put it back on the free list, or dealloc it */
159/* if page was hijacked, unmark the flag so it might get alloced next time */
160/* Note: lock should be held when calling this */
858119e1 161static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
162{
163 char *ptr;
164
165 if (bitmap->bp[page].count) /* page is still busy */
166 return;
167
168 /* page is no longer in use, it can be released */
169
170 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
171 bitmap->bp[page].hijacked = 0;
172 bitmap->bp[page].map = NULL;
ac2f40be
N
173 } else {
174 /* normal case, free the page */
175 ptr = bitmap->bp[page].map;
176 bitmap->bp[page].map = NULL;
177 bitmap->missing_pages++;
178 bitmap_free_page(bitmap, ptr);
32a7627c 179 }
32a7627c
N
180}
181
32a7627c
N
182/*
183 * bitmap file handling - read and write the bitmap file and its superblock
184 */
185
32a7627c
N
186/*
187 * basic page I/O operations
188 */
189
a654b9d8 190/* IO operations when bitmap is stored near all superblocks */
f6af949c 191static struct page *read_sb_page(mddev_t *mddev, loff_t offset,
a2ed9615
N
192 struct page *page,
193 unsigned long index, int size)
a654b9d8
N
194{
195 /* choose a good rdev and read the page from there */
196
197 mdk_rdev_t *rdev;
a654b9d8 198 sector_t target;
ac2f40be 199 int did_alloc = 0;
a654b9d8 200
ac2f40be 201 if (!page) {
a2ed9615 202 page = alloc_page(GFP_KERNEL);
ac2f40be
N
203 if (!page)
204 return ERR_PTR(-ENOMEM);
205 did_alloc = 1;
206 }
a654b9d8 207
159ec1fc 208 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7
N
209 if (! test_bit(In_sync, &rdev->flags)
210 || test_bit(Faulty, &rdev->flags))
ab904d63
N
211 continue;
212
0f420358 213 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
a654b9d8 214
a2ed9615 215 if (sync_page_io(rdev->bdev, target,
e1defc4f 216 roundup(size, bdev_logical_block_size(rdev->bdev)),
a2ed9615 217 page, READ)) {
ab904d63 218 page->index = index;
ce25c31b
N
219 attach_page_buffers(page, NULL); /* so that free_buffer will
220 * quietly no-op */
ab904d63
N
221 return page;
222 }
223 }
ac2f40be
N
224 if (did_alloc)
225 put_page(page);
ab904d63 226 return ERR_PTR(-EIO);
a654b9d8 227
a654b9d8
N
228}
229
b2d2c4ce
N
230static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
231{
232 /* Iterate the disks of an mddev, using rcu to protect access to the
233 * linked list, and raising the refcount of devices we return to ensure
234 * they don't disappear while in use.
235 * As devices are only added or removed when raid_disk is < 0 and
236 * nr_pending is 0 and In_sync is clear, the entries we return will
237 * still be in the same position on the list when we re-enter
238 * list_for_each_continue_rcu.
239 */
240 struct list_head *pos;
241 rcu_read_lock();
242 if (rdev == NULL)
243 /* start at the beginning */
244 pos = &mddev->disks;
245 else {
246 /* release the previous rdev and start from there. */
247 rdev_dec_pending(rdev, mddev);
248 pos = &rdev->same_set;
249 }
250 list_for_each_continue_rcu(pos, &mddev->disks) {
251 rdev = list_entry(pos, mdk_rdev_t, same_set);
252 if (rdev->raid_disk >= 0 &&
b2d2c4ce
N
253 !test_bit(Faulty, &rdev->flags)) {
254 /* this is a usable devices */
255 atomic_inc(&rdev->nr_pending);
256 rcu_read_unlock();
257 return rdev;
258 }
259 }
260 rcu_read_unlock();
261 return NULL;
262}
263
ab6085c7 264static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
a654b9d8 265{
b2d2c4ce 266 mdk_rdev_t *rdev = NULL;
ab6085c7 267 mddev_t *mddev = bitmap->mddev;
a654b9d8 268
b2d2c4ce 269 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
ac2f40be
N
270 int size = PAGE_SIZE;
271 loff_t offset = mddev->bitmap_info.offset;
272 if (page->index == bitmap->file_pages-1)
273 size = roundup(bitmap->last_page_size,
274 bdev_logical_block_size(rdev->bdev));
275 /* Just make sure we aren't corrupting data or
276 * metadata
277 */
278 if (mddev->external) {
279 /* Bitmap could be anywhere. */
280 if (rdev->sb_start + offset + (page->index
281 * (PAGE_SIZE/512))
282 > rdev->data_offset
283 &&
284 rdev->sb_start + offset
285 < (rdev->data_offset + mddev->dev_sectors
286 + (PAGE_SIZE/512)))
287 goto bad_alignment;
288 } else if (offset < 0) {
289 /* DATA BITMAP METADATA */
290 if (offset
291 + (long)(page->index * (PAGE_SIZE/512))
292 + size/512 > 0)
293 /* bitmap runs in to metadata */
294 goto bad_alignment;
295 if (rdev->data_offset + mddev->dev_sectors
296 > rdev->sb_start + offset)
297 /* data runs in to bitmap */
298 goto bad_alignment;
299 } else if (rdev->sb_start < rdev->data_offset) {
300 /* METADATA BITMAP DATA */
301 if (rdev->sb_start
302 + offset
303 + page->index*(PAGE_SIZE/512) + size/512
304 > rdev->data_offset)
305 /* bitmap runs in to data */
306 goto bad_alignment;
307 } else {
308 /* DATA METADATA BITMAP - no problems */
309 }
310 md_super_write(mddev, rdev,
311 rdev->sb_start + offset
312 + page->index * (PAGE_SIZE/512),
313 size,
314 page);
b2d2c4ce 315 }
a654b9d8
N
316
317 if (wait)
a9701a30 318 md_super_wait(mddev);
a654b9d8 319 return 0;
4b80991c
N
320
321 bad_alignment:
4b80991c 322 return -EINVAL;
a654b9d8
N
323}
324
4ad13663 325static void bitmap_file_kick(struct bitmap *bitmap);
32a7627c 326/*
a654b9d8 327 * write out a page to a file
32a7627c 328 */
4ad13663 329static void write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 330{
d785a06a 331 struct buffer_head *bh;
32a7627c 332
f0d76d70
N
333 if (bitmap->file == NULL) {
334 switch (write_sb_page(bitmap, page, wait)) {
335 case -EINVAL:
336 bitmap->flags |= BITMAP_WRITE_ERROR;
f0d76d70 337 }
4ad13663 338 } else {
a654b9d8 339
4ad13663 340 bh = page_buffers(page);
c708443c 341
4ad13663
N
342 while (bh && bh->b_blocknr) {
343 atomic_inc(&bitmap->pending_writes);
344 set_buffer_locked(bh);
345 set_buffer_mapped(bh);
346 submit_bh(WRITE, bh);
347 bh = bh->b_this_page;
348 }
d785a06a 349
ac2f40be 350 if (wait)
4ad13663
N
351 wait_event(bitmap->write_wait,
352 atomic_read(&bitmap->pending_writes)==0);
8a5e9cf1 353 }
4ad13663
N
354 if (bitmap->flags & BITMAP_WRITE_ERROR)
355 bitmap_file_kick(bitmap);
d785a06a
N
356}
357
358static void end_bitmap_write(struct buffer_head *bh, int uptodate)
359{
360 struct bitmap *bitmap = bh->b_private;
361 unsigned long flags;
32a7627c 362
d785a06a
N
363 if (!uptodate) {
364 spin_lock_irqsave(&bitmap->lock, flags);
365 bitmap->flags |= BITMAP_WRITE_ERROR;
366 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 367 }
d785a06a
N
368 if (atomic_dec_and_test(&bitmap->pending_writes))
369 wake_up(&bitmap->write_wait);
370}
32a7627c 371
d785a06a
N
372/* copied from buffer.c */
373static void
374__clear_page_buffers(struct page *page)
375{
376 ClearPagePrivate(page);
377 set_page_private(page, 0);
378 page_cache_release(page);
379}
380static void free_buffers(struct page *page)
381{
382 struct buffer_head *bh = page_buffers(page);
77ad4bc7 383
d785a06a
N
384 while (bh) {
385 struct buffer_head *next = bh->b_this_page;
386 free_buffer_head(bh);
387 bh = next;
77ad4bc7 388 }
d785a06a
N
389 __clear_page_buffers(page);
390 put_page(page);
32a7627c
N
391}
392
d785a06a
N
393/* read a page from a file.
394 * We both read the page, and attach buffers to the page to record the
395 * address of each block (using bmap). These addresses will be used
396 * to write the block later, completely bypassing the filesystem.
397 * This usage is similar to how swap files are handled, and allows us
398 * to write to a file with no concerns of memory allocation failing.
399 */
32a7627c 400static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
401 struct bitmap *bitmap,
402 unsigned long count)
32a7627c 403{
32a7627c 404 struct page *page = NULL;
c649bb9c 405 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
406 struct buffer_head *bh;
407 sector_t block;
32a7627c 408
ac2f40be 409 PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
2d1f3b5d 410 (unsigned long long)index << PAGE_SHIFT);
32a7627c 411
d785a06a
N
412 page = alloc_page(GFP_KERNEL);
413 if (!page)
414 page = ERR_PTR(-ENOMEM);
32a7627c
N
415 if (IS_ERR(page))
416 goto out;
d785a06a 417
d785a06a
N
418 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
419 if (!bh) {
2d1f3b5d 420 put_page(page);
d785a06a 421 page = ERR_PTR(-ENOMEM);
32a7627c
N
422 goto out;
423 }
d785a06a
N
424 attach_page_buffers(page, bh);
425 block = index << (PAGE_SHIFT - inode->i_blkbits);
426 while (bh) {
427 if (count == 0)
428 bh->b_blocknr = 0;
429 else {
430 bh->b_blocknr = bmap(inode, block);
431 if (bh->b_blocknr == 0) {
432 /* Cannot use this file! */
433 free_buffers(page);
434 page = ERR_PTR(-EINVAL);
435 goto out;
436 }
437 bh->b_bdev = inode->i_sb->s_bdev;
438 if (count < (1<<inode->i_blkbits))
439 count = 0;
440 else
441 count -= (1<<inode->i_blkbits);
442
443 bh->b_end_io = end_bitmap_write;
444 bh->b_private = bitmap;
ce25c31b
N
445 atomic_inc(&bitmap->pending_writes);
446 set_buffer_locked(bh);
447 set_buffer_mapped(bh);
448 submit_bh(READ, bh);
d785a06a
N
449 }
450 block++;
451 bh = bh->b_this_page;
452 }
d785a06a 453 page->index = index;
ce25c31b
N
454
455 wait_event(bitmap->write_wait,
456 atomic_read(&bitmap->pending_writes)==0);
457 if (bitmap->flags & BITMAP_WRITE_ERROR) {
458 free_buffers(page);
459 page = ERR_PTR(-EIO);
460 }
32a7627c
N
461out:
462 if (IS_ERR(page))
ac2f40be 463 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n",
2d1f3b5d
N
464 (int)PAGE_SIZE,
465 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
466 PTR_ERR(page));
467 return page;
468}
469
470/*
471 * bitmap file superblock operations
472 */
473
474/* update the event counter and sync the superblock to disk */
4ad13663 475void bitmap_update_sb(struct bitmap *bitmap)
32a7627c
N
476{
477 bitmap_super_t *sb;
478 unsigned long flags;
479
480 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
4ad13663 481 return;
ece5cff0
N
482 if (bitmap->mddev->bitmap_info.external)
483 return;
32a7627c
N
484 spin_lock_irqsave(&bitmap->lock, flags);
485 if (!bitmap->sb_page) { /* no superblock */
486 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 487 return;
32a7627c 488 }
32a7627c 489 spin_unlock_irqrestore(&bitmap->lock, flags);
7b92813c 490 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 491 sb->events = cpu_to_le64(bitmap->mddev->events);
a0da84f3
NB
492 if (bitmap->mddev->events < bitmap->events_cleared) {
493 /* rocking back to read-only */
494 bitmap->events_cleared = bitmap->mddev->events;
495 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
496 }
43a70507
N
497 /* Just in case these have been changed via sysfs: */
498 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
499 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
ea03aff9 500 kunmap_atomic(sb, KM_USER0);
4ad13663 501 write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
502}
503
504/* print out the bitmap file superblock */
505void bitmap_print_sb(struct bitmap *bitmap)
506{
507 bitmap_super_t *sb;
508
509 if (!bitmap || !bitmap->sb_page)
510 return;
7b92813c 511 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 512 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
513 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
514 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
515 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
516 *(__u32 *)(sb->uuid+0),
517 *(__u32 *)(sb->uuid+4),
518 *(__u32 *)(sb->uuid+8),
519 *(__u32 *)(sb->uuid+12));
a2cff26a 520 printk(KERN_DEBUG " events: %llu\n",
32a7627c 521 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 522 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 523 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
524 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
525 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
526 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
527 printk(KERN_DEBUG " sync size: %llu KB\n",
528 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 529 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 530 kunmap_atomic(sb, KM_USER0);
32a7627c
N
531}
532
533/* read the superblock from the bitmap file and initialize some bitmap fields */
534static int bitmap_read_sb(struct bitmap *bitmap)
535{
536 char *reason = NULL;
537 bitmap_super_t *sb;
4b6d287f 538 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
539 unsigned long long events;
540 int err = -EINVAL;
541
542 /* page 0 is the superblock, read it... */
f49d5e62
N
543 if (bitmap->file) {
544 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
545 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
546
547 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
548 } else {
42a04b50
N
549 bitmap->sb_page = read_sb_page(bitmap->mddev,
550 bitmap->mddev->bitmap_info.offset,
a2ed9615
N
551 NULL,
552 0, sizeof(bitmap_super_t));
a654b9d8 553 }
32a7627c
N
554 if (IS_ERR(bitmap->sb_page)) {
555 err = PTR_ERR(bitmap->sb_page);
556 bitmap->sb_page = NULL;
557 return err;
558 }
559
7b92813c 560 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 561
32a7627c 562 chunksize = le32_to_cpu(sb->chunksize);
1b04be96 563 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
4b6d287f 564 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
565
566 /* verify that the bitmap-specific fields are valid */
567 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
568 reason = "bad magic";
bd926c63
N
569 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
570 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 571 reason = "unrecognized superblock version";
1187cf0a 572 else if (chunksize < 512)
7dd5d34c 573 reason = "bitmap chunksize too small";
32a7627c
N
574 else if ((1 << ffz(~chunksize)) != chunksize)
575 reason = "bitmap chunksize not a power of 2";
1b04be96 576 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
7dd5d34c 577 reason = "daemon sleep period out of range";
4b6d287f
N
578 else if (write_behind > COUNTER_MAX)
579 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
580 if (reason) {
581 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
582 bmname(bitmap), reason);
583 goto out;
584 }
585
586 /* keep the array size field of the bitmap superblock up to date */
587 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
588
589 if (!bitmap->mddev->persistent)
590 goto success;
591
592 /*
593 * if we have a persistent array superblock, compare the
594 * bitmap's UUID and event counter to the mddev's
595 */
596 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
597 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
598 bmname(bitmap));
599 goto out;
600 }
601 events = le64_to_cpu(sb->events);
602 if (events < bitmap->mddev->events) {
603 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
604 "-- forcing full recovery\n", bmname(bitmap), events,
605 (unsigned long long) bitmap->mddev->events);
4f2e639a 606 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
607 }
608success:
609 /* assign fields using values from superblock */
42a04b50
N
610 bitmap->mddev->bitmap_info.chunksize = chunksize;
611 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
42a04b50 612 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
4f2e639a 613 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
614 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
615 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 616 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 617 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 618 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
619 err = 0;
620out:
ea03aff9 621 kunmap_atomic(sb, KM_USER0);
32a7627c
N
622 if (err)
623 bitmap_print_sb(bitmap);
624 return err;
625}
626
627enum bitmap_mask_op {
628 MASK_SET,
629 MASK_UNSET
630};
631
4ad13663
N
632/* record the state of the bitmap in the superblock. Return the old value */
633static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
634 enum bitmap_mask_op op)
32a7627c
N
635{
636 bitmap_super_t *sb;
637 unsigned long flags;
4ad13663 638 int old;
32a7627c
N
639
640 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 641 if (!bitmap->sb_page) { /* can't set the state */
32a7627c 642 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 643 return 0;
32a7627c 644 }
32a7627c 645 spin_unlock_irqrestore(&bitmap->lock, flags);
7b92813c 646 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
4ad13663 647 old = le32_to_cpu(sb->state) & bits;
32a7627c 648 switch (op) {
ac2f40be
N
649 case MASK_SET:
650 sb->state |= cpu_to_le32(bits);
651 break;
652 case MASK_UNSET:
653 sb->state &= cpu_to_le32(~bits);
654 break;
655 default:
656 BUG();
32a7627c 657 }
ea03aff9 658 kunmap_atomic(sb, KM_USER0);
4ad13663 659 return old;
32a7627c
N
660}
661
662/*
663 * general bitmap file operations
664 */
665
ece5cff0
N
666/*
667 * on-disk bitmap:
668 *
669 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
670 * file a page at a time. There's a superblock at the start of the file.
671 */
32a7627c 672/* calculate the index of the page that contains this bit */
ece5cff0 673static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
32a7627c 674{
ece5cff0
N
675 if (!bitmap->mddev->bitmap_info.external)
676 chunk += sizeof(bitmap_super_t) << 3;
677 return chunk >> PAGE_BIT_SHIFT;
32a7627c
N
678}
679
680/* calculate the (bit) offset of this bit within a page */
ece5cff0 681static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
32a7627c 682{
ece5cff0
N
683 if (!bitmap->mddev->bitmap_info.external)
684 chunk += sizeof(bitmap_super_t) << 3;
685 return chunk & (PAGE_BITS - 1);
32a7627c
N
686}
687
688/*
689 * return a pointer to the page in the filemap that contains the given bit
690 *
691 * this lookup is complicated by the fact that the bitmap sb might be exactly
692 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
693 * 0 or page 1
694 */
695static inline struct page *filemap_get_page(struct bitmap *bitmap,
696 unsigned long chunk)
697{
e384e585
N
698 if (bitmap->filemap == NULL)
699 return NULL;
ac2f40be
N
700 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
701 return NULL;
ece5cff0
N
702 return bitmap->filemap[file_page_index(bitmap, chunk)
703 - file_page_index(bitmap, 0)];
32a7627c
N
704}
705
32a7627c
N
706static void bitmap_file_unmap(struct bitmap *bitmap)
707{
708 struct page **map, *sb_page;
709 unsigned long *attr;
710 int pages;
711 unsigned long flags;
712
713 spin_lock_irqsave(&bitmap->lock, flags);
714 map = bitmap->filemap;
715 bitmap->filemap = NULL;
716 attr = bitmap->filemap_attr;
717 bitmap->filemap_attr = NULL;
718 pages = bitmap->file_pages;
719 bitmap->file_pages = 0;
720 sb_page = bitmap->sb_page;
721 bitmap->sb_page = NULL;
722 spin_unlock_irqrestore(&bitmap->lock, flags);
723
724 while (pages--)
ece5cff0 725 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
d785a06a 726 free_buffers(map[pages]);
32a7627c
N
727 kfree(map);
728 kfree(attr);
729
d785a06a
N
730 if (sb_page)
731 free_buffers(sb_page);
32a7627c
N
732}
733
734static void bitmap_file_put(struct bitmap *bitmap)
735{
736 struct file *file;
32a7627c
N
737 unsigned long flags;
738
739 spin_lock_irqsave(&bitmap->lock, flags);
740 file = bitmap->file;
741 bitmap->file = NULL;
742 spin_unlock_irqrestore(&bitmap->lock, flags);
743
d785a06a
N
744 if (file)
745 wait_event(bitmap->write_wait,
746 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
747 bitmap_file_unmap(bitmap);
748
d785a06a 749 if (file) {
c649bb9c 750 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 751 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 752 fput(file);
d785a06a 753 }
32a7627c
N
754}
755
32a7627c
N
756/*
757 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
758 * then it is no longer reliable, so we stop using it and we mark the file
759 * as failed in the superblock
760 */
761static void bitmap_file_kick(struct bitmap *bitmap)
762{
763 char *path, *ptr = NULL;
764
4ad13663
N
765 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
766 bitmap_update_sb(bitmap);
32a7627c 767
4ad13663
N
768 if (bitmap->file) {
769 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
770 if (path)
6bcfd601
CH
771 ptr = d_path(&bitmap->file->f_path, path,
772 PAGE_SIZE);
773
4ad13663
N
774 printk(KERN_ALERT
775 "%s: kicking failed bitmap file %s from array!\n",
6bcfd601 776 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
32a7627c 777
4ad13663
N
778 kfree(path);
779 } else
780 printk(KERN_ALERT
781 "%s: disabling internal bitmap due to errors\n",
782 bmname(bitmap));
a654b9d8 783 }
32a7627c
N
784
785 bitmap_file_put(bitmap);
786
787 return;
788}
789
790enum bitmap_page_attr {
ac2f40be
N
791 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
792 BITMAP_PAGE_CLEAN = 1, /* there are bits that might need to be cleared */
793 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
32a7627c
N
794};
795
796static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
797 enum bitmap_page_attr attr)
798{
e384e585
N
799 if (page)
800 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
801 else
802 __set_bit(attr, &bitmap->logattrs);
32a7627c
N
803}
804
805static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
806 enum bitmap_page_attr attr)
807{
e384e585
N
808 if (page)
809 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
810 else
811 __clear_bit(attr, &bitmap->logattrs);
32a7627c
N
812}
813
ec7a3197
N
814static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
815 enum bitmap_page_attr attr)
32a7627c 816{
e384e585
N
817 if (page)
818 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
819 else
820 return test_bit(attr, &bitmap->logattrs);
32a7627c
N
821}
822
823/*
824 * bitmap_file_set_bit -- called before performing a write to the md device
825 * to set (and eventually sync) a particular bit in the bitmap file
826 *
827 * we set the bit immediately, then we record the page number so that
828 * when an unplug occurs, we can flush the dirty pages out to disk
829 */
830static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
831{
832 unsigned long bit;
e384e585 833 struct page *page = NULL;
32a7627c
N
834 void *kaddr;
835 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
836
e384e585
N
837 if (!bitmap->filemap) {
838 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
839 if (log)
840 log->type->mark_region(log, chunk);
841 } else {
32a7627c 842
e384e585
N
843 page = filemap_get_page(bitmap, chunk);
844 if (!page)
845 return;
846 bit = file_page_offset(bitmap, chunk);
32a7627c 847
e384e585
N
848 /* set the bit */
849 kaddr = kmap_atomic(page, KM_USER0);
850 if (bitmap->flags & BITMAP_HOSTENDIAN)
851 set_bit(bit, kaddr);
852 else
853 ext2_set_bit(bit, kaddr);
854 kunmap_atomic(kaddr, KM_USER0);
855 PRINTK("set file bit %lu page %lu\n", bit, page->index);
856 }
32a7627c
N
857 /* record page number so it gets flushed to disk when unplug occurs */
858 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
32a7627c
N
859}
860
861/* this gets called when the md device is ready to unplug its underlying
862 * (slave) device queues -- before we let any writes go down, we need to
863 * sync the dirty pages of the bitmap file to disk */
4ad13663 864void bitmap_unplug(struct bitmap *bitmap)
32a7627c 865{
ec7a3197
N
866 unsigned long i, flags;
867 int dirty, need_write;
32a7627c
N
868 struct page *page;
869 int wait = 0;
870
871 if (!bitmap)
4ad13663 872 return;
e384e585
N
873 if (!bitmap->filemap) {
874 /* Must be using a dirty_log */
875 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
876 dirty = test_and_clear_bit(BITMAP_PAGE_DIRTY, &bitmap->logattrs);
877 need_write = test_and_clear_bit(BITMAP_PAGE_NEEDWRITE, &bitmap->logattrs);
878 if (dirty || need_write)
879 if (log->type->flush(log))
880 bitmap->flags |= BITMAP_WRITE_ERROR;
881 goto out;
882 }
32a7627c
N
883
884 /* look at each page to see if there are any set bits that need to be
885 * flushed out to disk */
886 for (i = 0; i < bitmap->file_pages; i++) {
887 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 888 if (!bitmap->filemap) {
32a7627c 889 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 890 return;
32a7627c
N
891 }
892 page = bitmap->filemap[i];
ec7a3197
N
893 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
894 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
895 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
896 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 897 if (dirty)
32a7627c
N
898 wait = 1;
899 spin_unlock_irqrestore(&bitmap->lock, flags);
900
ac2f40be 901 if (dirty || need_write)
4ad13663 902 write_page(bitmap, page, 0);
32a7627c
N
903 }
904 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 905 if (bitmap->file)
d785a06a
N
906 wait_event(bitmap->write_wait,
907 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 908 else
a9701a30 909 md_super_wait(bitmap->mddev);
32a7627c 910 }
e384e585 911out:
d785a06a
N
912 if (bitmap->flags & BITMAP_WRITE_ERROR)
913 bitmap_file_kick(bitmap);
32a7627c 914}
ac2f40be 915EXPORT_SYMBOL(bitmap_unplug);
32a7627c 916
6a07997f 917static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
918/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
919 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
920 * memory mapping of the bitmap file
921 * Special cases:
922 * if there's no bitmap file, or if the bitmap file had been
923 * previously kicked from the array, we mark all the bits as
924 * 1's in order to cause a full resync.
6a07997f
N
925 *
926 * We ignore all bits for sectors that end earlier than 'start'.
927 * This is used when reading an out-of-date bitmap...
32a7627c 928 */
6a07997f 929static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
930{
931 unsigned long i, chunks, index, oldindex, bit;
932 struct page *page = NULL, *oldpage = NULL;
933 unsigned long num_pages, bit_cnt = 0;
934 struct file *file;
d785a06a 935 unsigned long bytes, offset;
32a7627c
N
936 int outofdate;
937 int ret = -ENOSPC;
ea03aff9 938 void *paddr;
32a7627c
N
939
940 chunks = bitmap->chunks;
941 file = bitmap->file;
942
42a04b50 943 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
32a7627c 944
44456d37 945#ifdef INJECT_FAULTS_3
32a7627c
N
946 outofdate = 1;
947#else
948 outofdate = bitmap->flags & BITMAP_STALE;
949#endif
950 if (outofdate)
951 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
952 "recovery\n", bmname(bitmap));
953
e384e585 954 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
ece5cff0
N
955 if (!bitmap->mddev->bitmap_info.external)
956 bytes += sizeof(bitmap_super_t);
bc7f77de 957
e384e585 958 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
bc7f77de 959
ece5cff0 960 if (file && i_size_read(file->f_mapping->host) < bytes) {
32a7627c
N
961 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
962 bmname(bitmap),
963 (unsigned long) i_size_read(file->f_mapping->host),
ece5cff0 964 bytes);
4ad13663 965 goto err;
32a7627c 966 }
bc7f77de
N
967
968 ret = -ENOMEM;
969
32a7627c 970 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 971 if (!bitmap->filemap)
4ad13663 972 goto err;
32a7627c 973
e16b68b6
N
974 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
975 bitmap->filemap_attr = kzalloc(
ac2f40be 976 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 977 GFP_KERNEL);
bc7f77de 978 if (!bitmap->filemap_attr)
4ad13663 979 goto err;
32a7627c 980
32a7627c
N
981 oldindex = ~0L;
982
983 for (i = 0; i < chunks; i++) {
bd926c63 984 int b;
ece5cff0
N
985 index = file_page_index(bitmap, i);
986 bit = file_page_offset(bitmap, i);
32a7627c 987 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 988 int count;
32a7627c 989 /* unmap the old page, we're done with it */
d785a06a 990 if (index == num_pages-1)
ece5cff0 991 count = bytes - index * PAGE_SIZE;
d785a06a
N
992 else
993 count = PAGE_SIZE;
ece5cff0 994 if (index == 0 && bitmap->sb_page) {
32a7627c
N
995 /*
996 * if we're here then the superblock page
997 * contains some bits (PAGE_SIZE != sizeof sb)
998 * we've already read it in, so just use it
999 */
1000 page = bitmap->sb_page;
1001 offset = sizeof(bitmap_super_t);
53845270 1002 if (!file)
5c04f551
VK
1003 page = read_sb_page(
1004 bitmap->mddev,
1005 bitmap->mddev->bitmap_info.offset,
1006 page,
1007 index, count);
a654b9d8 1008 } else if (file) {
d785a06a 1009 page = read_page(file, index, bitmap, count);
a654b9d8
N
1010 offset = 0;
1011 } else {
42a04b50
N
1012 page = read_sb_page(bitmap->mddev,
1013 bitmap->mddev->bitmap_info.offset,
a2ed9615
N
1014 NULL,
1015 index, count);
32a7627c
N
1016 offset = 0;
1017 }
a654b9d8
N
1018 if (IS_ERR(page)) { /* read error */
1019 ret = PTR_ERR(page);
4ad13663 1020 goto err;
a654b9d8
N
1021 }
1022
32a7627c
N
1023 oldindex = index;
1024 oldpage = page;
32a7627c 1025
b74fd282
N
1026 bitmap->filemap[bitmap->file_pages++] = page;
1027 bitmap->last_page_size = count;
1028
32a7627c
N
1029 if (outofdate) {
1030 /*
1031 * if bitmap is out of date, dirty the
ac2f40be 1032 * whole page and write it out
32a7627c 1033 */
ea03aff9
N
1034 paddr = kmap_atomic(page, KM_USER0);
1035 memset(paddr + offset, 0xff,
6a07997f 1036 PAGE_SIZE - offset);
ea03aff9 1037 kunmap_atomic(paddr, KM_USER0);
4ad13663
N
1038 write_page(bitmap, page, 1);
1039
1040 ret = -EIO;
b74fd282 1041 if (bitmap->flags & BITMAP_WRITE_ERROR)
4ad13663 1042 goto err;
32a7627c 1043 }
32a7627c 1044 }
ea03aff9 1045 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1046 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1047 b = test_bit(bit, paddr);
bd926c63 1048 else
ea03aff9
N
1049 b = ext2_test_bit(bit, paddr);
1050 kunmap_atomic(paddr, KM_USER0);
bd926c63 1051 if (b) {
32a7627c 1052 /* if the disk bit is set, set the memory bit */
db305e50
N
1053 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1054 >= start);
1055 bitmap_set_memory_bits(bitmap,
1056 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1057 needed);
32a7627c 1058 bit_cnt++;
6a07997f 1059 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 1060 }
32a7627c
N
1061 }
1062
ac2f40be 1063 /* everything went OK */
32a7627c
N
1064 ret = 0;
1065 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1066
32a7627c
N
1067 if (bit_cnt) { /* Kick recovery if any bits were set */
1068 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1069 md_wakeup_thread(bitmap->mddev->thread);
1070 }
1071
32a7627c 1072 printk(KERN_INFO "%s: bitmap initialized from disk: "
4ad13663
N
1073 "read %lu/%lu pages, set %lu bits\n",
1074 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1075
1076 return 0;
32a7627c 1077
4ad13663
N
1078 err:
1079 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1080 bmname(bitmap), ret);
32a7627c
N
1081 return ret;
1082}
1083
a654b9d8
N
1084void bitmap_write_all(struct bitmap *bitmap)
1085{
1086 /* We don't actually write all bitmap blocks here,
1087 * just flag them as needing to be written
1088 */
ec7a3197 1089 int i;
a654b9d8 1090
ac2f40be 1091 for (i = 0; i < bitmap->file_pages; i++)
ec7a3197
N
1092 set_page_attr(bitmap, bitmap->filemap[i],
1093 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
1094}
1095
32a7627c
N
1096static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1097{
1098 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1099 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1100 bitmap->bp[page].count += inc;
32a7627c
N
1101 bitmap_checkfree(bitmap, page);
1102}
1103static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1104 sector_t offset, int *blocks,
1105 int create);
1106
1107/*
1108 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1109 * out to disk
1110 */
1111
aa5cbd10 1112void bitmap_daemon_work(mddev_t *mddev)
32a7627c 1113{
aa5cbd10 1114 struct bitmap *bitmap;
aa3163f8 1115 unsigned long j;
32a7627c
N
1116 unsigned long flags;
1117 struct page *page = NULL, *lastpage = NULL;
32a7627c 1118 int blocks;
ea03aff9 1119 void *paddr;
e384e585 1120 struct dm_dirty_log *log = mddev->bitmap_info.log;
32a7627c 1121
aa5cbd10
N
1122 /* Use a mutex to guard daemon_work against
1123 * bitmap_destroy.
1124 */
c3d9714e 1125 mutex_lock(&mddev->bitmap_info.mutex);
aa5cbd10
N
1126 bitmap = mddev->bitmap;
1127 if (bitmap == NULL) {
c3d9714e 1128 mutex_unlock(&mddev->bitmap_info.mutex);
4ad13663 1129 return;
aa5cbd10 1130 }
42a04b50 1131 if (time_before(jiffies, bitmap->daemon_lastrun
1b04be96 1132 + bitmap->mddev->bitmap_info.daemon_sleep))
7be3dfec
N
1133 goto done;
1134
32a7627c 1135 bitmap->daemon_lastrun = jiffies;
8311c29d
N
1136 if (bitmap->allclean) {
1137 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
aa5cbd10 1138 goto done;
8311c29d
N
1139 }
1140 bitmap->allclean = 1;
32a7627c 1141
be512691 1142 spin_lock_irqsave(&bitmap->lock, flags);
32a7627c
N
1143 for (j = 0; j < bitmap->chunks; j++) {
1144 bitmap_counter_t *bmc;
e384e585
N
1145 if (!bitmap->filemap) {
1146 if (!log)
1147 /* error or shutdown */
1148 break;
1149 } else
1150 page = filemap_get_page(bitmap, j);
32a7627c
N
1151
1152 if (page != lastpage) {
aa3163f8 1153 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1154 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1155 int need_write = test_page_attr(bitmap, page,
1156 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1157 if (need_write)
aa3163f8 1158 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1159
aa3163f8 1160 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1161 if (need_write) {
4ad13663 1162 write_page(bitmap, page, 0);
8311c29d
N
1163 bitmap->allclean = 0;
1164 }
be512691
N
1165 spin_lock_irqsave(&bitmap->lock, flags);
1166 j |= (PAGE_BITS - 1);
aa3163f8
N
1167 continue;
1168 }
1169
32a7627c 1170 /* grab the new page, sync and release the old */
32a7627c 1171 if (lastpage != NULL) {
ec7a3197 1172 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1173 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1174 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1175 write_page(bitmap, lastpage, 0);
32a7627c
N
1176 } else {
1177 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1178 spin_unlock_irqrestore(&bitmap->lock, flags);
1179 }
32a7627c
N
1180 } else
1181 spin_unlock_irqrestore(&bitmap->lock, flags);
1182 lastpage = page;
a0da84f3
NB
1183
1184 /* We are possibly going to clear some bits, so make
1185 * sure that events_cleared is up-to-date.
1186 */
ece5cff0
N
1187 if (bitmap->need_sync &&
1188 bitmap->mddev->bitmap_info.external == 0) {
a0da84f3
NB
1189 bitmap_super_t *sb;
1190 bitmap->need_sync = 0;
1191 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1192 sb->events_cleared =
1193 cpu_to_le64(bitmap->events_cleared);
1194 kunmap_atomic(sb, KM_USER0);
1195 write_page(bitmap, bitmap->sb_page, 1);
1196 }
32a7627c 1197 spin_lock_irqsave(&bitmap->lock, flags);
ece5cff0
N
1198 if (!bitmap->need_sync)
1199 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 1200 }
db305e50
N
1201 bmc = bitmap_get_counter(bitmap,
1202 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1203 &blocks, 0);
32a7627c 1204 if (bmc) {
8311c29d
N
1205 if (*bmc)
1206 bitmap->allclean = 0;
1207
32a7627c 1208 if (*bmc == 2) {
ac2f40be 1209 *bmc = 1; /* maybe clear the bit next time */
32a7627c 1210 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
ece5cff0 1211 } else if (*bmc == 1 && !bitmap->need_sync) {
32a7627c
N
1212 /* we can clear the bit */
1213 *bmc = 0;
db305e50
N
1214 bitmap_count_page(bitmap,
1215 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
32a7627c
N
1216 -1);
1217
1218 /* clear the bit */
e384e585
N
1219 if (page) {
1220 paddr = kmap_atomic(page, KM_USER0);
1221 if (bitmap->flags & BITMAP_HOSTENDIAN)
1222 clear_bit(file_page_offset(bitmap, j),
1223 paddr);
1224 else
1225 ext2_clear_bit(file_page_offset(bitmap, j),
1226 paddr);
1227 kunmap_atomic(paddr, KM_USER0);
1228 } else
1229 log->type->clear_region(log, j);
32a7627c 1230 }
be512691
N
1231 } else
1232 j |= PAGE_COUNTER_MASK;
32a7627c 1233 }
be512691 1234 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c
N
1235
1236 /* now sync the final page */
e384e585 1237 if (lastpage != NULL || log != NULL) {
32a7627c 1238 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1239 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1240 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1241 spin_unlock_irqrestore(&bitmap->lock, flags);
e384e585
N
1242 if (lastpage)
1243 write_page(bitmap, lastpage, 0);
1244 else
1245 if (log->type->flush(log))
1246 bitmap->flags |= BITMAP_WRITE_ERROR;
32a7627c
N
1247 } else {
1248 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1249 spin_unlock_irqrestore(&bitmap->lock, flags);
1250 }
32a7627c
N
1251 }
1252
7be3dfec 1253 done:
8311c29d 1254 if (bitmap->allclean == 0)
ac2f40be 1255 bitmap->mddev->thread->timeout =
1b04be96 1256 bitmap->mddev->bitmap_info.daemon_sleep;
c3d9714e 1257 mutex_unlock(&mddev->bitmap_info.mutex);
32a7627c
N
1258}
1259
32a7627c
N
1260static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1261 sector_t offset, int *blocks,
1262 int create)
ee305ace
N
1263__releases(bitmap->lock)
1264__acquires(bitmap->lock)
32a7627c
N
1265{
1266 /* If 'create', we might release the lock and reclaim it.
1267 * The lock must have been taken with interrupts enabled.
1268 * If !create, we don't release the lock.
1269 */
1270 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1271 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1272 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1273 sector_t csize;
ef425673 1274 int err;
32a7627c 1275
ef425673
N
1276 err = bitmap_checkpage(bitmap, page, create);
1277
1278 if (bitmap->bp[page].hijacked ||
1279 bitmap->bp[page].map == NULL)
1280 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1281 PAGE_COUNTER_SHIFT - 1);
1282 else
32a7627c 1283 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
ef425673
N
1284 *blocks = csize - (offset & (csize - 1));
1285
1286 if (err < 0)
32a7627c 1287 return NULL;
ef425673 1288
32a7627c
N
1289 /* now locked ... */
1290
1291 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1292 /* should we use the first or second counter field
1293 * of the hijacked pointer? */
1294 int hi = (pageoff > PAGE_COUNTER_MASK);
32a7627c
N
1295 return &((bitmap_counter_t *)
1296 &bitmap->bp[page].map)[hi];
ef425673 1297 } else /* page is allocated */
32a7627c
N
1298 return (bitmap_counter_t *)
1299 &(bitmap->bp[page].map[pageoff]);
32a7627c
N
1300}
1301
4b6d287f 1302int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c 1303{
ac2f40be
N
1304 if (!bitmap)
1305 return 0;
4b6d287f
N
1306
1307 if (behind) {
696fcd53 1308 int bw;
4b6d287f 1309 atomic_inc(&bitmap->behind_writes);
696fcd53
PC
1310 bw = atomic_read(&bitmap->behind_writes);
1311 if (bw > bitmap->behind_writes_used)
1312 bitmap->behind_writes_used = bw;
1313
4b6d287f 1314 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
696fcd53 1315 bw, bitmap->max_write_behind);
4b6d287f
N
1316 }
1317
32a7627c
N
1318 while (sectors) {
1319 int blocks;
1320 bitmap_counter_t *bmc;
1321
1322 spin_lock_irq(&bitmap->lock);
1323 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1324 if (!bmc) {
1325 spin_unlock_irq(&bitmap->lock);
1326 return 0;
1327 }
1328
da6e1a32
NB
1329 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1330 DEFINE_WAIT(__wait);
1331 /* note that it is safe to do the prepare_to_wait
1332 * after the test as long as we do it before dropping
1333 * the spinlock.
1334 */
1335 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1336 TASK_UNINTERRUPTIBLE);
1337 spin_unlock_irq(&bitmap->lock);
b63d7c2e 1338 md_unplug(bitmap->mddev);
da6e1a32
NB
1339 schedule();
1340 finish_wait(&bitmap->overflow_wait, &__wait);
1341 continue;
1342 }
1343
ac2f40be 1344 switch (*bmc) {
32a7627c
N
1345 case 0:
1346 bitmap_file_set_bit(bitmap, offset);
ac2f40be 1347 bitmap_count_page(bitmap, offset, 1);
32a7627c
N
1348 /* fall through */
1349 case 1:
1350 *bmc = 2;
1351 }
da6e1a32 1352
32a7627c
N
1353 (*bmc)++;
1354
1355 spin_unlock_irq(&bitmap->lock);
1356
1357 offset += blocks;
1358 if (sectors > blocks)
1359 sectors -= blocks;
ac2f40be
N
1360 else
1361 sectors = 0;
32a7627c 1362 }
8311c29d 1363 bitmap->allclean = 0;
32a7627c
N
1364 return 0;
1365}
ac2f40be 1366EXPORT_SYMBOL(bitmap_startwrite);
32a7627c
N
1367
1368void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1369 int success, int behind)
32a7627c 1370{
ac2f40be
N
1371 if (!bitmap)
1372 return;
4b6d287f 1373 if (behind) {
e555190d
N
1374 if (atomic_dec_and_test(&bitmap->behind_writes))
1375 wake_up(&bitmap->behind_wait);
4b6d287f
N
1376 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1377 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1378 }
d0a4bb49
N
1379 if (bitmap->mddev->degraded)
1380 /* Never clear bits or update events_cleared when degraded */
1381 success = 0;
4b6d287f 1382
32a7627c
N
1383 while (sectors) {
1384 int blocks;
1385 unsigned long flags;
1386 bitmap_counter_t *bmc;
1387
1388 spin_lock_irqsave(&bitmap->lock, flags);
1389 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1390 if (!bmc) {
1391 spin_unlock_irqrestore(&bitmap->lock, flags);
1392 return;
1393 }
1394
a0da84f3
NB
1395 if (success &&
1396 bitmap->events_cleared < bitmap->mddev->events) {
1397 bitmap->events_cleared = bitmap->mddev->events;
1398 bitmap->need_sync = 1;
5ff5afff 1399 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
a0da84f3
NB
1400 }
1401
32a7627c
N
1402 if (!success && ! (*bmc & NEEDED_MASK))
1403 *bmc |= NEEDED_MASK;
1404
da6e1a32
NB
1405 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1406 wake_up(&bitmap->overflow_wait);
1407
32a7627c 1408 (*bmc)--;
ac2f40be 1409 if (*bmc <= 2)
32a7627c 1410 set_page_attr(bitmap,
e384e585
N
1411 filemap_get_page(
1412 bitmap,
1413 offset >> CHUNK_BLOCK_SHIFT(bitmap)),
32a7627c 1414 BITMAP_PAGE_CLEAN);
ac2f40be 1415
32a7627c
N
1416 spin_unlock_irqrestore(&bitmap->lock, flags);
1417 offset += blocks;
1418 if (sectors > blocks)
1419 sectors -= blocks;
ac2f40be
N
1420 else
1421 sectors = 0;
32a7627c
N
1422 }
1423}
ac2f40be 1424EXPORT_SYMBOL(bitmap_endwrite);
32a7627c 1425
1187cf0a
N
1426static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1427 int degraded)
32a7627c
N
1428{
1429 bitmap_counter_t *bmc;
1430 int rv;
1431 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1432 *blocks = 1024;
1433 return 1; /* always resync if no bitmap */
1434 }
1435 spin_lock_irq(&bitmap->lock);
1436 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1437 rv = 0;
1438 if (bmc) {
1439 /* locked */
1440 if (RESYNC(*bmc))
1441 rv = 1;
1442 else if (NEEDED(*bmc)) {
1443 rv = 1;
6a806c51
N
1444 if (!degraded) { /* don't set/clear bits if degraded */
1445 *bmc |= RESYNC_MASK;
1446 *bmc &= ~NEEDED_MASK;
1447 }
32a7627c
N
1448 }
1449 }
1450 spin_unlock_irq(&bitmap->lock);
8311c29d 1451 bitmap->allclean = 0;
32a7627c
N
1452 return rv;
1453}
1454
1187cf0a
N
1455int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1456 int degraded)
1457{
1458 /* bitmap_start_sync must always report on multiples of whole
1459 * pages, otherwise resync (which is very PAGE_SIZE based) will
1460 * get confused.
1461 * So call __bitmap_start_sync repeatedly (if needed) until
1462 * At least PAGE_SIZE>>9 blocks are covered.
1463 * Return the 'or' of the result.
1464 */
1465 int rv = 0;
1466 int blocks1;
1467
1468 *blocks = 0;
1469 while (*blocks < (PAGE_SIZE>>9)) {
1470 rv |= __bitmap_start_sync(bitmap, offset,
1471 &blocks1, degraded);
1472 offset += blocks1;
1473 *blocks += blocks1;
1474 }
1475 return rv;
1476}
ac2f40be 1477EXPORT_SYMBOL(bitmap_start_sync);
1187cf0a 1478
32a7627c
N
1479void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1480{
1481 bitmap_counter_t *bmc;
1482 unsigned long flags;
ac2f40be
N
1483
1484 if (bitmap == NULL) {
32a7627c
N
1485 *blocks = 1024;
1486 return;
1487 }
1488 spin_lock_irqsave(&bitmap->lock, flags);
1489 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1490 if (bmc == NULL)
1491 goto unlock;
1492 /* locked */
32a7627c
N
1493 if (RESYNC(*bmc)) {
1494 *bmc &= ~RESYNC_MASK;
1495
1496 if (!NEEDED(*bmc) && aborted)
1497 *bmc |= NEEDED_MASK;
1498 else {
ac2f40be 1499 if (*bmc <= 2)
32a7627c
N
1500 set_page_attr(bitmap,
1501 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1502 BITMAP_PAGE_CLEAN);
32a7627c
N
1503 }
1504 }
1505 unlock:
1506 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1507 bitmap->allclean = 0;
32a7627c 1508}
ac2f40be 1509EXPORT_SYMBOL(bitmap_end_sync);
32a7627c
N
1510
1511void bitmap_close_sync(struct bitmap *bitmap)
1512{
1513 /* Sync has finished, and any bitmap chunks that weren't synced
1514 * properly have been aborted. It remains to us to clear the
1515 * RESYNC bit wherever it is still on
1516 */
1517 sector_t sector = 0;
1518 int blocks;
b47490c9
N
1519 if (!bitmap)
1520 return;
32a7627c
N
1521 while (sector < bitmap->mddev->resync_max_sectors) {
1522 bitmap_end_sync(bitmap, sector, &blocks, 0);
b47490c9
N
1523 sector += blocks;
1524 }
1525}
ac2f40be 1526EXPORT_SYMBOL(bitmap_close_sync);
b47490c9
N
1527
1528void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1529{
1530 sector_t s = 0;
1531 int blocks;
1532
1533 if (!bitmap)
1534 return;
1535 if (sector == 0) {
1536 bitmap->last_end_sync = jiffies;
1537 return;
1538 }
1539 if (time_before(jiffies, (bitmap->last_end_sync
1b04be96 1540 + bitmap->mddev->bitmap_info.daemon_sleep)))
b47490c9
N
1541 return;
1542 wait_event(bitmap->mddev->recovery_wait,
1543 atomic_read(&bitmap->mddev->recovery_active) == 0);
1544
97e4f42d 1545 bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync;
070dc6dd 1546 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
b47490c9
N
1547 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1548 s = 0;
1549 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1550 bitmap_end_sync(bitmap, s, &blocks, 0);
1551 s += blocks;
32a7627c 1552 }
b47490c9 1553 bitmap->last_end_sync = jiffies;
acb180b0 1554 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
32a7627c 1555}
ac2f40be 1556EXPORT_SYMBOL(bitmap_cond_end_sync);
32a7627c 1557
6a07997f 1558static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1559{
1560 /* For each chunk covered by any of these sectors, set the
193f1c93 1561 * counter to 1 and set resync_needed. They should all
32a7627c
N
1562 * be 0 at this point
1563 */
193f1c93
N
1564
1565 int secs;
1566 bitmap_counter_t *bmc;
1567 spin_lock_irq(&bitmap->lock);
1568 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1569 if (!bmc) {
32a7627c 1570 spin_unlock_irq(&bitmap->lock);
193f1c93 1571 return;
32a7627c 1572 }
ac2f40be 1573 if (!*bmc) {
193f1c93 1574 struct page *page;
ac2f40be 1575 *bmc = 1 | (needed ? NEEDED_MASK : 0);
193f1c93
N
1576 bitmap_count_page(bitmap, offset, 1);
1577 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1578 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1579 }
1580 spin_unlock_irq(&bitmap->lock);
8311c29d 1581 bitmap->allclean = 0;
32a7627c
N
1582}
1583
9b1d1dac
PC
1584/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1585void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1586{
1587 unsigned long chunk;
1588
1589 for (chunk = s; chunk <= e; chunk++) {
db305e50 1590 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
9b1d1dac
PC
1591 bitmap_set_memory_bits(bitmap, sec, 1);
1592 bitmap_file_set_bit(bitmap, sec);
ffa23322
N
1593 if (sec < bitmap->mddev->recovery_cp)
1594 /* We are asserting that the array is dirty,
1595 * so move the recovery_cp address back so
1596 * that it is obvious that it is dirty
1597 */
1598 bitmap->mddev->recovery_cp = sec;
9b1d1dac
PC
1599 }
1600}
1601
6b8b3e8a
N
1602/*
1603 * flush out any pending updates
1604 */
1605void bitmap_flush(mddev_t *mddev)
1606{
1607 struct bitmap *bitmap = mddev->bitmap;
42a04b50 1608 long sleep;
6b8b3e8a
N
1609
1610 if (!bitmap) /* there was no bitmap */
1611 return;
1612
1613 /* run the daemon_work three time to ensure everything is flushed
1614 * that can be
1615 */
1b04be96 1616 sleep = mddev->bitmap_info.daemon_sleep * 2;
42a04b50 1617 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1618 bitmap_daemon_work(mddev);
42a04b50 1619 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1620 bitmap_daemon_work(mddev);
42a04b50 1621 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1622 bitmap_daemon_work(mddev);
6b8b3e8a
N
1623 bitmap_update_sb(bitmap);
1624}
1625
32a7627c
N
1626/*
1627 * free memory that was allocated
1628 */
3178b0db 1629static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1630{
1631 unsigned long k, pages;
1632 struct bitmap_page *bp;
32a7627c
N
1633
1634 if (!bitmap) /* there was no bitmap */
1635 return;
1636
32a7627c
N
1637 /* release the bitmap file and kill the daemon */
1638 bitmap_file_put(bitmap);
1639
1640 bp = bitmap->bp;
1641 pages = bitmap->pages;
1642
1643 /* free all allocated memory */
1644
32a7627c
N
1645 if (bp) /* deallocate the page memory */
1646 for (k = 0; k < pages; k++)
1647 if (bp[k].map && !bp[k].hijacked)
1648 kfree(bp[k].map);
1649 kfree(bp);
1650 kfree(bitmap);
1651}
aa5cbd10 1652
3178b0db
N
1653void bitmap_destroy(mddev_t *mddev)
1654{
1655 struct bitmap *bitmap = mddev->bitmap;
1656
1657 if (!bitmap) /* there was no bitmap */
1658 return;
1659
c3d9714e 1660 mutex_lock(&mddev->bitmap_info.mutex);
3178b0db 1661 mddev->bitmap = NULL; /* disconnect from the md device */
c3d9714e 1662 mutex_unlock(&mddev->bitmap_info.mutex);
b15c2e57
N
1663 if (mddev->thread)
1664 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db 1665
ece5cff0
N
1666 if (bitmap->sysfs_can_clear)
1667 sysfs_put(bitmap->sysfs_can_clear);
1668
3178b0db
N
1669 bitmap_free(bitmap);
1670}
32a7627c
N
1671
1672/*
1673 * initialize the bitmap structure
1674 * if this returns an error, bitmap_destroy must be called to do clean up
1675 */
1676int bitmap_create(mddev_t *mddev)
1677{
1678 struct bitmap *bitmap;
1f593903 1679 sector_t blocks = mddev->resync_max_sectors;
32a7627c
N
1680 unsigned long chunks;
1681 unsigned long pages;
c3d9714e 1682 struct file *file = mddev->bitmap_info.file;
32a7627c 1683 int err;
5ff5afff 1684 struct sysfs_dirent *bm = NULL;
32a7627c 1685
5f6e3c83 1686 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1687
e384e585
N
1688 if (!file
1689 && !mddev->bitmap_info.offset
1690 && !mddev->bitmap_info.log) /* bitmap disabled, nothing to do */
32a7627c
N
1691 return 0;
1692
c3d9714e 1693 BUG_ON(file && mddev->bitmap_info.offset);
e384e585 1694 BUG_ON(mddev->bitmap_info.offset && mddev->bitmap_info.log);
a654b9d8 1695
9ffae0cf 1696 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1697 if (!bitmap)
1698 return -ENOMEM;
1699
32a7627c 1700 spin_lock_init(&bitmap->lock);
ce25c31b
N
1701 atomic_set(&bitmap->pending_writes, 0);
1702 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1703 init_waitqueue_head(&bitmap->overflow_wait);
e555190d 1704 init_waitqueue_head(&bitmap->behind_wait);
ce25c31b 1705
32a7627c 1706 bitmap->mddev = mddev;
32a7627c 1707
5ff5afff
N
1708 if (mddev->kobj.sd)
1709 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
ece5cff0 1710 if (bm) {
3ff195b0 1711 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
ece5cff0
N
1712 sysfs_put(bm);
1713 } else
1714 bitmap->sysfs_can_clear = NULL;
1715
32a7627c 1716 bitmap->file = file;
ce25c31b
N
1717 if (file) {
1718 get_file(file);
ae8fa283
N
1719 /* As future accesses to this file will use bmap,
1720 * and bypass the page cache, we must sync the file
1721 * first.
1722 */
8018ab05 1723 vfs_fsync(file, 1);
ce25c31b 1724 }
42a04b50 1725 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
ece5cff0
N
1726 if (!mddev->bitmap_info.external)
1727 err = bitmap_read_sb(bitmap);
1728 else {
1729 err = 0;
1730 if (mddev->bitmap_info.chunksize == 0 ||
1731 mddev->bitmap_info.daemon_sleep == 0)
1732 /* chunksize and time_base need to be
1733 * set first. */
1734 err = -EINVAL;
1735 }
32a7627c 1736 if (err)
3178b0db 1737 goto error;
32a7627c 1738
624ce4f5 1739 bitmap->daemon_lastrun = jiffies;
42a04b50 1740 bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize);
32a7627c
N
1741
1742 /* now that chunksize and chunkshift are set, we can use these macros */
ac2f40be 1743 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1f593903 1744 CHUNK_BLOCK_SHIFT(bitmap);
ac2f40be 1745 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
32a7627c
N
1746
1747 BUG_ON(!pages);
1748
1749 bitmap->chunks = chunks;
1750 bitmap->pages = pages;
1751 bitmap->missing_pages = pages;
1752 bitmap->counter_bits = COUNTER_BITS;
1753
1754 bitmap->syncchunk = ~0UL;
1755
44456d37 1756#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1757 bitmap->bp = NULL;
1758#else
9ffae0cf 1759 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1760#endif
3178b0db 1761 err = -ENOMEM;
32a7627c 1762 if (!bitmap->bp)
3178b0db 1763 goto error;
32a7627c 1764
69e51b44
N
1765 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1766 pages, bmname(bitmap));
1767
1768 mddev->bitmap = bitmap;
1769
1770
1771 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1772
1773 error:
1774 bitmap_free(bitmap);
1775 return err;
1776}
1777
1778int bitmap_load(mddev_t *mddev)
1779{
1780 int err = 0;
1781 sector_t sector = 0;
1782 struct bitmap *bitmap = mddev->bitmap;
1783
1784 if (!bitmap)
1785 goto out;
1786
1787 /* Clear out old bitmap info first: Either there is none, or we
1788 * are resuming after someone else has possibly changed things,
1789 * so we should forget old cached info.
1790 * All chunks should be clean, but some might need_sync.
1791 */
1792 while (sector < mddev->resync_max_sectors) {
1793 int blocks;
1794 bitmap_start_sync(bitmap, sector, &blocks, 0);
1795 sector += blocks;
1796 }
1797 bitmap_close_sync(bitmap);
1798
e384e585
N
1799 if (mddev->bitmap_info.log) {
1800 unsigned long i;
1801 struct dm_dirty_log *log = mddev->bitmap_info.log;
1802 for (i = 0; i < bitmap->chunks; i++)
1803 if (!log->type->in_sync(log, i, 1))
1804 bitmap_set_memory_bits(bitmap,
1805 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1806 1);
69e51b44
N
1807 } else {
1808 sector_t start = 0;
1809 if (mddev->degraded == 0
1810 || bitmap->events_cleared == mddev->events)
1811 /* no need to keep dirty bits to optimise a
1812 * re-add of a missing device */
1813 start = mddev->recovery_cp;
193f1c93 1814
69e51b44
N
1815 err = bitmap_init_from_disk(bitmap, start);
1816 }
32a7627c 1817 if (err)
69e51b44 1818 goto out;
3178b0db 1819
1b04be96 1820 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
9cd30fdc 1821 md_wakeup_thread(mddev->thread);
b15c2e57 1822
4ad13663
N
1823 bitmap_update_sb(bitmap);
1824
69e51b44
N
1825 if (bitmap->flags & BITMAP_WRITE_ERROR)
1826 err = -EIO;
1827out:
3178b0db 1828 return err;
32a7627c 1829}
69e51b44 1830EXPORT_SYMBOL_GPL(bitmap_load);
32a7627c 1831
43a70507
N
1832static ssize_t
1833location_show(mddev_t *mddev, char *page)
1834{
1835 ssize_t len;
ac2f40be 1836 if (mddev->bitmap_info.file)
43a70507 1837 len = sprintf(page, "file");
ac2f40be 1838 else if (mddev->bitmap_info.offset)
43a70507 1839 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
ac2f40be 1840 else
43a70507
N
1841 len = sprintf(page, "none");
1842 len += sprintf(page+len, "\n");
1843 return len;
1844}
1845
1846static ssize_t
1847location_store(mddev_t *mddev, const char *buf, size_t len)
1848{
1849
1850 if (mddev->pers) {
1851 if (!mddev->pers->quiesce)
1852 return -EBUSY;
1853 if (mddev->recovery || mddev->sync_thread)
1854 return -EBUSY;
1855 }
1856
1857 if (mddev->bitmap || mddev->bitmap_info.file ||
1858 mddev->bitmap_info.offset) {
1859 /* bitmap already configured. Only option is to clear it */
1860 if (strncmp(buf, "none", 4) != 0)
1861 return -EBUSY;
1862 if (mddev->pers) {
1863 mddev->pers->quiesce(mddev, 1);
1864 bitmap_destroy(mddev);
1865 mddev->pers->quiesce(mddev, 0);
1866 }
1867 mddev->bitmap_info.offset = 0;
1868 if (mddev->bitmap_info.file) {
1869 struct file *f = mddev->bitmap_info.file;
1870 mddev->bitmap_info.file = NULL;
1871 restore_bitmap_write_access(f);
1872 fput(f);
1873 }
1874 } else {
1875 /* No bitmap, OK to set a location */
1876 long long offset;
1877 if (strncmp(buf, "none", 4) == 0)
1878 /* nothing to be done */;
1879 else if (strncmp(buf, "file:", 5) == 0) {
1880 /* Not supported yet */
1881 return -EINVAL;
1882 } else {
1883 int rv;
1884 if (buf[0] == '+')
1885 rv = strict_strtoll(buf+1, 10, &offset);
1886 else
1887 rv = strict_strtoll(buf, 10, &offset);
1888 if (rv)
1889 return rv;
1890 if (offset == 0)
1891 return -EINVAL;
ece5cff0
N
1892 if (mddev->bitmap_info.external == 0 &&
1893 mddev->major_version == 0 &&
43a70507
N
1894 offset != mddev->bitmap_info.default_offset)
1895 return -EINVAL;
1896 mddev->bitmap_info.offset = offset;
1897 if (mddev->pers) {
1898 mddev->pers->quiesce(mddev, 1);
1899 rv = bitmap_create(mddev);
1900 if (rv) {
1901 bitmap_destroy(mddev);
1902 mddev->bitmap_info.offset = 0;
1903 }
1904 mddev->pers->quiesce(mddev, 0);
1905 if (rv)
1906 return rv;
1907 }
1908 }
1909 }
1910 if (!mddev->external) {
1911 /* Ensure new bitmap info is stored in
1912 * metadata promptly.
1913 */
1914 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1915 md_wakeup_thread(mddev->thread);
1916 }
1917 return len;
1918}
1919
1920static struct md_sysfs_entry bitmap_location =
1921__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1922
1923static ssize_t
1924timeout_show(mddev_t *mddev, char *page)
1925{
1926 ssize_t len;
1927 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1928 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
ac2f40be 1929
43a70507
N
1930 len = sprintf(page, "%lu", secs);
1931 if (jifs)
1932 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1933 len += sprintf(page+len, "\n");
1934 return len;
1935}
1936
1937static ssize_t
1938timeout_store(mddev_t *mddev, const char *buf, size_t len)
1939{
1940 /* timeout can be set at any time */
1941 unsigned long timeout;
1942 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1943 if (rv)
1944 return rv;
1945
1946 /* just to make sure we don't overflow... */
1947 if (timeout >= LONG_MAX / HZ)
1948 return -EINVAL;
1949
1950 timeout = timeout * HZ / 10000;
1951
1952 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1953 timeout = MAX_SCHEDULE_TIMEOUT-1;
1954 if (timeout < 1)
1955 timeout = 1;
1956 mddev->bitmap_info.daemon_sleep = timeout;
1957 if (mddev->thread) {
1958 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1959 * the bitmap is all clean and we don't need to
1960 * adjust the timeout right now
1961 */
1962 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1963 mddev->thread->timeout = timeout;
1964 md_wakeup_thread(mddev->thread);
1965 }
1966 }
1967 return len;
1968}
1969
1970static struct md_sysfs_entry bitmap_timeout =
1971__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1972
1973static ssize_t
1974backlog_show(mddev_t *mddev, char *page)
1975{
1976 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1977}
1978
1979static ssize_t
1980backlog_store(mddev_t *mddev, const char *buf, size_t len)
1981{
1982 unsigned long backlog;
1983 int rv = strict_strtoul(buf, 10, &backlog);
1984 if (rv)
1985 return rv;
1986 if (backlog > COUNTER_MAX)
1987 return -EINVAL;
1988 mddev->bitmap_info.max_write_behind = backlog;
1989 return len;
1990}
1991
1992static struct md_sysfs_entry bitmap_backlog =
1993__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
1994
1995static ssize_t
1996chunksize_show(mddev_t *mddev, char *page)
1997{
1998 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
1999}
2000
2001static ssize_t
2002chunksize_store(mddev_t *mddev, const char *buf, size_t len)
2003{
2004 /* Can only be changed when no bitmap is active */
2005 int rv;
2006 unsigned long csize;
2007 if (mddev->bitmap)
2008 return -EBUSY;
2009 rv = strict_strtoul(buf, 10, &csize);
2010 if (rv)
2011 return rv;
2012 if (csize < 512 ||
2013 !is_power_of_2(csize))
2014 return -EINVAL;
2015 mddev->bitmap_info.chunksize = csize;
2016 return len;
2017}
2018
2019static struct md_sysfs_entry bitmap_chunksize =
2020__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2021
ece5cff0
N
2022static ssize_t metadata_show(mddev_t *mddev, char *page)
2023{
2024 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2025 ? "external" : "internal"));
2026}
2027
2028static ssize_t metadata_store(mddev_t *mddev, const char *buf, size_t len)
2029{
2030 if (mddev->bitmap ||
2031 mddev->bitmap_info.file ||
2032 mddev->bitmap_info.offset)
2033 return -EBUSY;
2034 if (strncmp(buf, "external", 8) == 0)
2035 mddev->bitmap_info.external = 1;
2036 else if (strncmp(buf, "internal", 8) == 0)
2037 mddev->bitmap_info.external = 0;
2038 else
2039 return -EINVAL;
2040 return len;
2041}
2042
2043static struct md_sysfs_entry bitmap_metadata =
2044__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2045
2046static ssize_t can_clear_show(mddev_t *mddev, char *page)
2047{
2048 int len;
2049 if (mddev->bitmap)
2050 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2051 "false" : "true"));
2052 else
2053 len = sprintf(page, "\n");
2054 return len;
2055}
2056
2057static ssize_t can_clear_store(mddev_t *mddev, const char *buf, size_t len)
2058{
2059 if (mddev->bitmap == NULL)
2060 return -ENOENT;
2061 if (strncmp(buf, "false", 5) == 0)
2062 mddev->bitmap->need_sync = 1;
2063 else if (strncmp(buf, "true", 4) == 0) {
2064 if (mddev->degraded)
2065 return -EBUSY;
2066 mddev->bitmap->need_sync = 0;
2067 } else
2068 return -EINVAL;
2069 return len;
2070}
2071
2072static struct md_sysfs_entry bitmap_can_clear =
2073__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2074
696fcd53
PC
2075static ssize_t
2076behind_writes_used_show(mddev_t *mddev, char *page)
2077{
2078 if (mddev->bitmap == NULL)
2079 return sprintf(page, "0\n");
2080 return sprintf(page, "%lu\n",
2081 mddev->bitmap->behind_writes_used);
2082}
2083
2084static ssize_t
2085behind_writes_used_reset(mddev_t *mddev, const char *buf, size_t len)
2086{
2087 if (mddev->bitmap)
2088 mddev->bitmap->behind_writes_used = 0;
2089 return len;
2090}
2091
2092static struct md_sysfs_entry max_backlog_used =
2093__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2094 behind_writes_used_show, behind_writes_used_reset);
2095
43a70507
N
2096static struct attribute *md_bitmap_attrs[] = {
2097 &bitmap_location.attr,
2098 &bitmap_timeout.attr,
2099 &bitmap_backlog.attr,
2100 &bitmap_chunksize.attr,
ece5cff0
N
2101 &bitmap_metadata.attr,
2102 &bitmap_can_clear.attr,
696fcd53 2103 &max_backlog_used.attr,
43a70507
N
2104 NULL
2105};
2106struct attribute_group md_bitmap_group = {
2107 .name = "bitmap",
2108 .attrs = md_bitmap_attrs,
2109};
2110