MD: add sync_super to mddev_t struct
[linux-2.6-block.git] / drivers / md / bitmap.c
CommitLineData
32a7627c
N
1/*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
32a7627c
N
10 */
11
12/*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
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
ccebd4c4 213 target = offset + index * (PAGE_SIZE/512);
a654b9d8 214
2b193363 215 if (sync_page_io(rdev, target,
e1defc4f 216 roundup(size, bdev_logical_block_size(rdev->bdev)),
ccebd4c4 217 page, READ, true)) {
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;
a6ff7e08 267 struct block_device *bdev;
ab6085c7 268 mddev_t *mddev = bitmap->mddev;
a654b9d8 269
b2d2c4ce 270 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
ac2f40be
N
271 int size = PAGE_SIZE;
272 loff_t offset = mddev->bitmap_info.offset;
a6ff7e08
JB
273
274 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
275
ac2f40be
N
276 if (page->index == bitmap->file_pages-1)
277 size = roundup(bitmap->last_page_size,
a6ff7e08 278 bdev_logical_block_size(bdev));
ac2f40be
N
279 /* Just make sure we aren't corrupting data or
280 * metadata
281 */
282 if (mddev->external) {
283 /* Bitmap could be anywhere. */
284 if (rdev->sb_start + offset + (page->index
285 * (PAGE_SIZE/512))
286 > rdev->data_offset
287 &&
288 rdev->sb_start + offset
289 < (rdev->data_offset + mddev->dev_sectors
290 + (PAGE_SIZE/512)))
291 goto bad_alignment;
292 } else if (offset < 0) {
293 /* DATA BITMAP METADATA */
294 if (offset
295 + (long)(page->index * (PAGE_SIZE/512))
296 + size/512 > 0)
297 /* bitmap runs in to metadata */
298 goto bad_alignment;
299 if (rdev->data_offset + mddev->dev_sectors
300 > rdev->sb_start + offset)
301 /* data runs in to bitmap */
302 goto bad_alignment;
303 } else if (rdev->sb_start < rdev->data_offset) {
304 /* METADATA BITMAP DATA */
305 if (rdev->sb_start
306 + offset
307 + page->index*(PAGE_SIZE/512) + size/512
308 > rdev->data_offset)
309 /* bitmap runs in to data */
310 goto bad_alignment;
311 } else {
312 /* DATA METADATA BITMAP - no problems */
313 }
314 md_super_write(mddev, rdev,
315 rdev->sb_start + offset
316 + page->index * (PAGE_SIZE/512),
317 size,
318 page);
b2d2c4ce 319 }
a654b9d8
N
320
321 if (wait)
a9701a30 322 md_super_wait(mddev);
a654b9d8 323 return 0;
4b80991c
N
324
325 bad_alignment:
4b80991c 326 return -EINVAL;
a654b9d8
N
327}
328
4ad13663 329static void bitmap_file_kick(struct bitmap *bitmap);
32a7627c 330/*
a654b9d8 331 * write out a page to a file
32a7627c 332 */
4ad13663 333static void write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 334{
d785a06a 335 struct buffer_head *bh;
32a7627c 336
f0d76d70
N
337 if (bitmap->file == NULL) {
338 switch (write_sb_page(bitmap, page, wait)) {
339 case -EINVAL:
340 bitmap->flags |= BITMAP_WRITE_ERROR;
f0d76d70 341 }
4ad13663 342 } else {
a654b9d8 343
4ad13663 344 bh = page_buffers(page);
c708443c 345
4ad13663
N
346 while (bh && bh->b_blocknr) {
347 atomic_inc(&bitmap->pending_writes);
348 set_buffer_locked(bh);
349 set_buffer_mapped(bh);
721a9602 350 submit_bh(WRITE | REQ_SYNC, bh);
4ad13663
N
351 bh = bh->b_this_page;
352 }
d785a06a 353
ac2f40be 354 if (wait)
4ad13663
N
355 wait_event(bitmap->write_wait,
356 atomic_read(&bitmap->pending_writes)==0);
8a5e9cf1 357 }
4ad13663
N
358 if (bitmap->flags & BITMAP_WRITE_ERROR)
359 bitmap_file_kick(bitmap);
d785a06a
N
360}
361
362static void end_bitmap_write(struct buffer_head *bh, int uptodate)
363{
364 struct bitmap *bitmap = bh->b_private;
365 unsigned long flags;
32a7627c 366
d785a06a
N
367 if (!uptodate) {
368 spin_lock_irqsave(&bitmap->lock, flags);
369 bitmap->flags |= BITMAP_WRITE_ERROR;
370 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 371 }
d785a06a
N
372 if (atomic_dec_and_test(&bitmap->pending_writes))
373 wake_up(&bitmap->write_wait);
374}
32a7627c 375
d785a06a
N
376/* copied from buffer.c */
377static void
378__clear_page_buffers(struct page *page)
379{
380 ClearPagePrivate(page);
381 set_page_private(page, 0);
382 page_cache_release(page);
383}
384static void free_buffers(struct page *page)
385{
386 struct buffer_head *bh = page_buffers(page);
77ad4bc7 387
d785a06a
N
388 while (bh) {
389 struct buffer_head *next = bh->b_this_page;
390 free_buffer_head(bh);
391 bh = next;
77ad4bc7 392 }
d785a06a
N
393 __clear_page_buffers(page);
394 put_page(page);
32a7627c
N
395}
396
d785a06a
N
397/* read a page from a file.
398 * We both read the page, and attach buffers to the page to record the
399 * address of each block (using bmap). These addresses will be used
400 * to write the block later, completely bypassing the filesystem.
401 * This usage is similar to how swap files are handled, and allows us
402 * to write to a file with no concerns of memory allocation failing.
403 */
32a7627c 404static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
405 struct bitmap *bitmap,
406 unsigned long count)
32a7627c 407{
32a7627c 408 struct page *page = NULL;
c649bb9c 409 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
410 struct buffer_head *bh;
411 sector_t block;
32a7627c 412
ac2f40be 413 PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
2d1f3b5d 414 (unsigned long long)index << PAGE_SHIFT);
32a7627c 415
d785a06a
N
416 page = alloc_page(GFP_KERNEL);
417 if (!page)
418 page = ERR_PTR(-ENOMEM);
32a7627c
N
419 if (IS_ERR(page))
420 goto out;
d785a06a 421
d785a06a
N
422 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
423 if (!bh) {
2d1f3b5d 424 put_page(page);
d785a06a 425 page = ERR_PTR(-ENOMEM);
32a7627c
N
426 goto out;
427 }
d785a06a
N
428 attach_page_buffers(page, bh);
429 block = index << (PAGE_SHIFT - inode->i_blkbits);
430 while (bh) {
431 if (count == 0)
432 bh->b_blocknr = 0;
433 else {
434 bh->b_blocknr = bmap(inode, block);
435 if (bh->b_blocknr == 0) {
436 /* Cannot use this file! */
437 free_buffers(page);
438 page = ERR_PTR(-EINVAL);
439 goto out;
440 }
441 bh->b_bdev = inode->i_sb->s_bdev;
442 if (count < (1<<inode->i_blkbits))
443 count = 0;
444 else
445 count -= (1<<inode->i_blkbits);
446
447 bh->b_end_io = end_bitmap_write;
448 bh->b_private = bitmap;
ce25c31b
N
449 atomic_inc(&bitmap->pending_writes);
450 set_buffer_locked(bh);
451 set_buffer_mapped(bh);
452 submit_bh(READ, bh);
d785a06a
N
453 }
454 block++;
455 bh = bh->b_this_page;
456 }
d785a06a 457 page->index = index;
ce25c31b
N
458
459 wait_event(bitmap->write_wait,
460 atomic_read(&bitmap->pending_writes)==0);
461 if (bitmap->flags & BITMAP_WRITE_ERROR) {
462 free_buffers(page);
463 page = ERR_PTR(-EIO);
464 }
32a7627c
N
465out:
466 if (IS_ERR(page))
ac2f40be 467 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n",
2d1f3b5d
N
468 (int)PAGE_SIZE,
469 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
470 PTR_ERR(page));
471 return page;
472}
473
474/*
475 * bitmap file superblock operations
476 */
477
478/* update the event counter and sync the superblock to disk */
4ad13663 479void bitmap_update_sb(struct bitmap *bitmap)
32a7627c
N
480{
481 bitmap_super_t *sb;
482 unsigned long flags;
483
484 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
4ad13663 485 return;
ece5cff0
N
486 if (bitmap->mddev->bitmap_info.external)
487 return;
32a7627c
N
488 spin_lock_irqsave(&bitmap->lock, flags);
489 if (!bitmap->sb_page) { /* no superblock */
490 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 491 return;
32a7627c 492 }
32a7627c 493 spin_unlock_irqrestore(&bitmap->lock, flags);
7b92813c 494 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 495 sb->events = cpu_to_le64(bitmap->mddev->events);
8258c532 496 if (bitmap->mddev->events < bitmap->events_cleared)
a0da84f3
NB
497 /* rocking back to read-only */
498 bitmap->events_cleared = bitmap->mddev->events;
8258c532
N
499 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
500 sb->state = cpu_to_le32(bitmap->flags);
43a70507
N
501 /* Just in case these have been changed via sysfs: */
502 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
503 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
ea03aff9 504 kunmap_atomic(sb, KM_USER0);
4ad13663 505 write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
506}
507
508/* print out the bitmap file superblock */
509void bitmap_print_sb(struct bitmap *bitmap)
510{
511 bitmap_super_t *sb;
512
513 if (!bitmap || !bitmap->sb_page)
514 return;
7b92813c 515 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 516 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
517 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
518 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
519 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
520 *(__u32 *)(sb->uuid+0),
521 *(__u32 *)(sb->uuid+4),
522 *(__u32 *)(sb->uuid+8),
523 *(__u32 *)(sb->uuid+12));
a2cff26a 524 printk(KERN_DEBUG " events: %llu\n",
32a7627c 525 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 526 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 527 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
528 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
529 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
530 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
531 printk(KERN_DEBUG " sync size: %llu KB\n",
532 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 533 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 534 kunmap_atomic(sb, KM_USER0);
32a7627c
N
535}
536
537/* read the superblock from the bitmap file and initialize some bitmap fields */
538static int bitmap_read_sb(struct bitmap *bitmap)
539{
540 char *reason = NULL;
541 bitmap_super_t *sb;
4b6d287f 542 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
543 unsigned long long events;
544 int err = -EINVAL;
545
546 /* page 0 is the superblock, read it... */
f49d5e62
N
547 if (bitmap->file) {
548 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
549 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
550
551 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
552 } else {
42a04b50
N
553 bitmap->sb_page = read_sb_page(bitmap->mddev,
554 bitmap->mddev->bitmap_info.offset,
a2ed9615
N
555 NULL,
556 0, sizeof(bitmap_super_t));
a654b9d8 557 }
32a7627c
N
558 if (IS_ERR(bitmap->sb_page)) {
559 err = PTR_ERR(bitmap->sb_page);
560 bitmap->sb_page = NULL;
561 return err;
562 }
563
7b92813c 564 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 565
32a7627c 566 chunksize = le32_to_cpu(sb->chunksize);
1b04be96 567 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
4b6d287f 568 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
569
570 /* verify that the bitmap-specific fields are valid */
571 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
572 reason = "bad magic";
bd926c63
N
573 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
574 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 575 reason = "unrecognized superblock version";
1187cf0a 576 else if (chunksize < 512)
7dd5d34c 577 reason = "bitmap chunksize too small";
32a7627c
N
578 else if ((1 << ffz(~chunksize)) != chunksize)
579 reason = "bitmap chunksize not a power of 2";
1b04be96 580 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
7dd5d34c 581 reason = "daemon sleep period out of range";
4b6d287f
N
582 else if (write_behind > COUNTER_MAX)
583 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
584 if (reason) {
585 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
586 bmname(bitmap), reason);
587 goto out;
588 }
589
590 /* keep the array size field of the bitmap superblock up to date */
591 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
592
593 if (!bitmap->mddev->persistent)
594 goto success;
595
596 /*
597 * if we have a persistent array superblock, compare the
598 * bitmap's UUID and event counter to the mddev's
599 */
600 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
601 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
602 bmname(bitmap));
603 goto out;
604 }
605 events = le64_to_cpu(sb->events);
606 if (events < bitmap->mddev->events) {
607 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
608 "-- forcing full recovery\n", bmname(bitmap), events,
609 (unsigned long long) bitmap->mddev->events);
4f2e639a 610 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
611 }
612success:
613 /* assign fields using values from superblock */
42a04b50
N
614 bitmap->mddev->bitmap_info.chunksize = chunksize;
615 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
42a04b50 616 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
4f2e639a 617 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
618 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
619 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 620 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
8258c532 621 if (bitmap->flags & BITMAP_STALE)
6a07997f 622 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
623 err = 0;
624out:
ea03aff9 625 kunmap_atomic(sb, KM_USER0);
32a7627c
N
626 if (err)
627 bitmap_print_sb(bitmap);
628 return err;
629}
630
631enum bitmap_mask_op {
632 MASK_SET,
633 MASK_UNSET
634};
635
4ad13663
N
636/* record the state of the bitmap in the superblock. Return the old value */
637static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
638 enum bitmap_mask_op op)
32a7627c
N
639{
640 bitmap_super_t *sb;
641 unsigned long flags;
4ad13663 642 int old;
32a7627c
N
643
644 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 645 if (!bitmap->sb_page) { /* can't set the state */
32a7627c 646 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 647 return 0;
32a7627c 648 }
32a7627c 649 spin_unlock_irqrestore(&bitmap->lock, flags);
7b92813c 650 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
4ad13663 651 old = le32_to_cpu(sb->state) & bits;
32a7627c 652 switch (op) {
ac2f40be
N
653 case MASK_SET:
654 sb->state |= cpu_to_le32(bits);
8258c532 655 bitmap->flags |= bits;
ac2f40be
N
656 break;
657 case MASK_UNSET:
658 sb->state &= cpu_to_le32(~bits);
8258c532 659 bitmap->flags &= ~bits;
ac2f40be
N
660 break;
661 default:
662 BUG();
32a7627c 663 }
ea03aff9 664 kunmap_atomic(sb, KM_USER0);
4ad13663 665 return old;
32a7627c
N
666}
667
668/*
669 * general bitmap file operations
670 */
671
ece5cff0
N
672/*
673 * on-disk bitmap:
674 *
675 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
676 * file a page at a time. There's a superblock at the start of the file.
677 */
32a7627c 678/* calculate the index of the page that contains this bit */
ece5cff0 679static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
32a7627c 680{
ece5cff0
N
681 if (!bitmap->mddev->bitmap_info.external)
682 chunk += sizeof(bitmap_super_t) << 3;
683 return chunk >> PAGE_BIT_SHIFT;
32a7627c
N
684}
685
686/* calculate the (bit) offset of this bit within a page */
ece5cff0 687static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
32a7627c 688{
ece5cff0
N
689 if (!bitmap->mddev->bitmap_info.external)
690 chunk += sizeof(bitmap_super_t) << 3;
691 return chunk & (PAGE_BITS - 1);
32a7627c
N
692}
693
694/*
695 * return a pointer to the page in the filemap that contains the given bit
696 *
697 * this lookup is complicated by the fact that the bitmap sb might be exactly
698 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
699 * 0 or page 1
700 */
701static inline struct page *filemap_get_page(struct bitmap *bitmap,
702 unsigned long chunk)
703{
e384e585
N
704 if (bitmap->filemap == NULL)
705 return NULL;
ac2f40be
N
706 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
707 return NULL;
ece5cff0
N
708 return bitmap->filemap[file_page_index(bitmap, chunk)
709 - file_page_index(bitmap, 0)];
32a7627c
N
710}
711
32a7627c
N
712static void bitmap_file_unmap(struct bitmap *bitmap)
713{
714 struct page **map, *sb_page;
715 unsigned long *attr;
716 int pages;
717 unsigned long flags;
718
719 spin_lock_irqsave(&bitmap->lock, flags);
720 map = bitmap->filemap;
721 bitmap->filemap = NULL;
722 attr = bitmap->filemap_attr;
723 bitmap->filemap_attr = NULL;
724 pages = bitmap->file_pages;
725 bitmap->file_pages = 0;
726 sb_page = bitmap->sb_page;
727 bitmap->sb_page = NULL;
728 spin_unlock_irqrestore(&bitmap->lock, flags);
729
730 while (pages--)
ece5cff0 731 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
d785a06a 732 free_buffers(map[pages]);
32a7627c
N
733 kfree(map);
734 kfree(attr);
735
d785a06a
N
736 if (sb_page)
737 free_buffers(sb_page);
32a7627c
N
738}
739
740static void bitmap_file_put(struct bitmap *bitmap)
741{
742 struct file *file;
32a7627c
N
743 unsigned long flags;
744
745 spin_lock_irqsave(&bitmap->lock, flags);
746 file = bitmap->file;
747 bitmap->file = NULL;
748 spin_unlock_irqrestore(&bitmap->lock, flags);
749
d785a06a
N
750 if (file)
751 wait_event(bitmap->write_wait,
752 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
753 bitmap_file_unmap(bitmap);
754
d785a06a 755 if (file) {
c649bb9c 756 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 757 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 758 fput(file);
d785a06a 759 }
32a7627c
N
760}
761
32a7627c
N
762/*
763 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
764 * then it is no longer reliable, so we stop using it and we mark the file
765 * as failed in the superblock
766 */
767static void bitmap_file_kick(struct bitmap *bitmap)
768{
769 char *path, *ptr = NULL;
770
4ad13663
N
771 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
772 bitmap_update_sb(bitmap);
32a7627c 773
4ad13663
N
774 if (bitmap->file) {
775 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
776 if (path)
6bcfd601
CH
777 ptr = d_path(&bitmap->file->f_path, path,
778 PAGE_SIZE);
779
4ad13663
N
780 printk(KERN_ALERT
781 "%s: kicking failed bitmap file %s from array!\n",
6bcfd601 782 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
32a7627c 783
4ad13663
N
784 kfree(path);
785 } else
786 printk(KERN_ALERT
787 "%s: disabling internal bitmap due to errors\n",
788 bmname(bitmap));
a654b9d8 789 }
32a7627c
N
790
791 bitmap_file_put(bitmap);
792
793 return;
794}
795
796enum bitmap_page_attr {
ac2f40be
N
797 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
798 BITMAP_PAGE_CLEAN = 1, /* there are bits that might need to be cleared */
799 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
32a7627c
N
800};
801
802static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
803 enum bitmap_page_attr attr)
804{
e384e585
N
805 if (page)
806 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
807 else
808 __set_bit(attr, &bitmap->logattrs);
32a7627c
N
809}
810
811static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
812 enum bitmap_page_attr attr)
813{
e384e585
N
814 if (page)
815 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
816 else
817 __clear_bit(attr, &bitmap->logattrs);
32a7627c
N
818}
819
ec7a3197
N
820static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
821 enum bitmap_page_attr attr)
32a7627c 822{
e384e585
N
823 if (page)
824 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
825 else
826 return test_bit(attr, &bitmap->logattrs);
32a7627c
N
827}
828
829/*
830 * bitmap_file_set_bit -- called before performing a write to the md device
831 * to set (and eventually sync) a particular bit in the bitmap file
832 *
833 * we set the bit immediately, then we record the page number so that
834 * when an unplug occurs, we can flush the dirty pages out to disk
835 */
836static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
837{
838 unsigned long bit;
e384e585 839 struct page *page = NULL;
32a7627c
N
840 void *kaddr;
841 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
842
e384e585
N
843 if (!bitmap->filemap) {
844 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
845 if (log)
846 log->type->mark_region(log, chunk);
847 } else {
32a7627c 848
e384e585
N
849 page = filemap_get_page(bitmap, chunk);
850 if (!page)
851 return;
852 bit = file_page_offset(bitmap, chunk);
32a7627c 853
e384e585
N
854 /* set the bit */
855 kaddr = kmap_atomic(page, KM_USER0);
856 if (bitmap->flags & BITMAP_HOSTENDIAN)
857 set_bit(bit, kaddr);
858 else
6b33aff3 859 __test_and_set_bit_le(bit, kaddr);
e384e585
N
860 kunmap_atomic(kaddr, KM_USER0);
861 PRINTK("set file bit %lu page %lu\n", bit, page->index);
862 }
32a7627c
N
863 /* record page number so it gets flushed to disk when unplug occurs */
864 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
32a7627c
N
865}
866
867/* this gets called when the md device is ready to unplug its underlying
868 * (slave) device queues -- before we let any writes go down, we need to
869 * sync the dirty pages of the bitmap file to disk */
4ad13663 870void bitmap_unplug(struct bitmap *bitmap)
32a7627c 871{
ec7a3197
N
872 unsigned long i, flags;
873 int dirty, need_write;
32a7627c
N
874 struct page *page;
875 int wait = 0;
876
877 if (!bitmap)
4ad13663 878 return;
e384e585
N
879 if (!bitmap->filemap) {
880 /* Must be using a dirty_log */
881 struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log;
882 dirty = test_and_clear_bit(BITMAP_PAGE_DIRTY, &bitmap->logattrs);
883 need_write = test_and_clear_bit(BITMAP_PAGE_NEEDWRITE, &bitmap->logattrs);
884 if (dirty || need_write)
885 if (log->type->flush(log))
886 bitmap->flags |= BITMAP_WRITE_ERROR;
887 goto out;
888 }
32a7627c
N
889
890 /* look at each page to see if there are any set bits that need to be
891 * flushed out to disk */
892 for (i = 0; i < bitmap->file_pages; i++) {
893 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 894 if (!bitmap->filemap) {
32a7627c 895 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 896 return;
32a7627c
N
897 }
898 page = bitmap->filemap[i];
ec7a3197
N
899 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
900 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
901 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
902 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 903 if (dirty)
32a7627c
N
904 wait = 1;
905 spin_unlock_irqrestore(&bitmap->lock, flags);
906
ac2f40be 907 if (dirty || need_write)
4ad13663 908 write_page(bitmap, page, 0);
32a7627c
N
909 }
910 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 911 if (bitmap->file)
d785a06a
N
912 wait_event(bitmap->write_wait,
913 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 914 else
a9701a30 915 md_super_wait(bitmap->mddev);
32a7627c 916 }
e384e585 917out:
d785a06a
N
918 if (bitmap->flags & BITMAP_WRITE_ERROR)
919 bitmap_file_kick(bitmap);
32a7627c 920}
ac2f40be 921EXPORT_SYMBOL(bitmap_unplug);
32a7627c 922
6a07997f 923static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
924/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
925 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
926 * memory mapping of the bitmap file
927 * Special cases:
928 * if there's no bitmap file, or if the bitmap file had been
929 * previously kicked from the array, we mark all the bits as
930 * 1's in order to cause a full resync.
6a07997f
N
931 *
932 * We ignore all bits for sectors that end earlier than 'start'.
933 * This is used when reading an out-of-date bitmap...
32a7627c 934 */
6a07997f 935static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
936{
937 unsigned long i, chunks, index, oldindex, bit;
938 struct page *page = NULL, *oldpage = NULL;
939 unsigned long num_pages, bit_cnt = 0;
940 struct file *file;
d785a06a 941 unsigned long bytes, offset;
32a7627c
N
942 int outofdate;
943 int ret = -ENOSPC;
ea03aff9 944 void *paddr;
32a7627c
N
945
946 chunks = bitmap->chunks;
947 file = bitmap->file;
948
42a04b50 949 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
32a7627c 950
44456d37 951#ifdef INJECT_FAULTS_3
32a7627c
N
952 outofdate = 1;
953#else
954 outofdate = bitmap->flags & BITMAP_STALE;
955#endif
956 if (outofdate)
957 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
958 "recovery\n", bmname(bitmap));
959
e384e585 960 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
ece5cff0
N
961 if (!bitmap->mddev->bitmap_info.external)
962 bytes += sizeof(bitmap_super_t);
bc7f77de 963
e384e585 964 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
bc7f77de 965
ece5cff0 966 if (file && i_size_read(file->f_mapping->host) < bytes) {
32a7627c
N
967 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
968 bmname(bitmap),
969 (unsigned long) i_size_read(file->f_mapping->host),
ece5cff0 970 bytes);
4ad13663 971 goto err;
32a7627c 972 }
bc7f77de
N
973
974 ret = -ENOMEM;
975
32a7627c 976 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 977 if (!bitmap->filemap)
4ad13663 978 goto err;
32a7627c 979
e16b68b6
N
980 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
981 bitmap->filemap_attr = kzalloc(
ac2f40be 982 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 983 GFP_KERNEL);
bc7f77de 984 if (!bitmap->filemap_attr)
4ad13663 985 goto err;
32a7627c 986
32a7627c
N
987 oldindex = ~0L;
988
989 for (i = 0; i < chunks; i++) {
bd926c63 990 int b;
ece5cff0
N
991 index = file_page_index(bitmap, i);
992 bit = file_page_offset(bitmap, i);
32a7627c 993 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 994 int count;
32a7627c 995 /* unmap the old page, we're done with it */
d785a06a 996 if (index == num_pages-1)
ece5cff0 997 count = bytes - index * PAGE_SIZE;
d785a06a
N
998 else
999 count = PAGE_SIZE;
ece5cff0 1000 if (index == 0 && bitmap->sb_page) {
32a7627c
N
1001 /*
1002 * if we're here then the superblock page
1003 * contains some bits (PAGE_SIZE != sizeof sb)
1004 * we've already read it in, so just use it
1005 */
1006 page = bitmap->sb_page;
1007 offset = sizeof(bitmap_super_t);
53845270 1008 if (!file)
5c04f551
VK
1009 page = read_sb_page(
1010 bitmap->mddev,
1011 bitmap->mddev->bitmap_info.offset,
1012 page,
1013 index, count);
a654b9d8 1014 } else if (file) {
d785a06a 1015 page = read_page(file, index, bitmap, count);
a654b9d8
N
1016 offset = 0;
1017 } else {
42a04b50
N
1018 page = read_sb_page(bitmap->mddev,
1019 bitmap->mddev->bitmap_info.offset,
a2ed9615
N
1020 NULL,
1021 index, count);
32a7627c
N
1022 offset = 0;
1023 }
a654b9d8
N
1024 if (IS_ERR(page)) { /* read error */
1025 ret = PTR_ERR(page);
4ad13663 1026 goto err;
a654b9d8
N
1027 }
1028
32a7627c
N
1029 oldindex = index;
1030 oldpage = page;
32a7627c 1031
b74fd282
N
1032 bitmap->filemap[bitmap->file_pages++] = page;
1033 bitmap->last_page_size = count;
1034
32a7627c
N
1035 if (outofdate) {
1036 /*
1037 * if bitmap is out of date, dirty the
ac2f40be 1038 * whole page and write it out
32a7627c 1039 */
ea03aff9
N
1040 paddr = kmap_atomic(page, KM_USER0);
1041 memset(paddr + offset, 0xff,
6a07997f 1042 PAGE_SIZE - offset);
ea03aff9 1043 kunmap_atomic(paddr, KM_USER0);
4ad13663
N
1044 write_page(bitmap, page, 1);
1045
1046 ret = -EIO;
b74fd282 1047 if (bitmap->flags & BITMAP_WRITE_ERROR)
4ad13663 1048 goto err;
32a7627c 1049 }
32a7627c 1050 }
ea03aff9 1051 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1052 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1053 b = test_bit(bit, paddr);
bd926c63 1054 else
6b33aff3 1055 b = test_bit_le(bit, paddr);
ea03aff9 1056 kunmap_atomic(paddr, KM_USER0);
bd926c63 1057 if (b) {
32a7627c 1058 /* if the disk bit is set, set the memory bit */
db305e50
N
1059 int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1060 >= start);
1061 bitmap_set_memory_bits(bitmap,
1062 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1063 needed);
32a7627c 1064 bit_cnt++;
6a07997f 1065 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 1066 }
32a7627c
N
1067 }
1068
ac2f40be 1069 /* everything went OK */
32a7627c
N
1070 ret = 0;
1071 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1072
32a7627c
N
1073 if (bit_cnt) { /* Kick recovery if any bits were set */
1074 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1075 md_wakeup_thread(bitmap->mddev->thread);
1076 }
1077
32a7627c 1078 printk(KERN_INFO "%s: bitmap initialized from disk: "
4ad13663
N
1079 "read %lu/%lu pages, set %lu bits\n",
1080 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1081
1082 return 0;
32a7627c 1083
4ad13663
N
1084 err:
1085 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1086 bmname(bitmap), ret);
32a7627c
N
1087 return ret;
1088}
1089
a654b9d8
N
1090void bitmap_write_all(struct bitmap *bitmap)
1091{
1092 /* We don't actually write all bitmap blocks here,
1093 * just flag them as needing to be written
1094 */
ec7a3197 1095 int i;
a654b9d8 1096
ac2f40be 1097 for (i = 0; i < bitmap->file_pages; i++)
ec7a3197
N
1098 set_page_attr(bitmap, bitmap->filemap[i],
1099 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
1100}
1101
32a7627c
N
1102static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1103{
1104 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1105 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1106 bitmap->bp[page].count += inc;
32a7627c
N
1107 bitmap_checkfree(bitmap, page);
1108}
1109static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
57dab0bd 1110 sector_t offset, sector_t *blocks,
32a7627c
N
1111 int create);
1112
1113/*
1114 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1115 * out to disk
1116 */
1117
aa5cbd10 1118void bitmap_daemon_work(mddev_t *mddev)
32a7627c 1119{
aa5cbd10 1120 struct bitmap *bitmap;
aa3163f8 1121 unsigned long j;
32a7627c
N
1122 unsigned long flags;
1123 struct page *page = NULL, *lastpage = NULL;
57dab0bd 1124 sector_t blocks;
ea03aff9 1125 void *paddr;
e384e585 1126 struct dm_dirty_log *log = mddev->bitmap_info.log;
32a7627c 1127
aa5cbd10
N
1128 /* Use a mutex to guard daemon_work against
1129 * bitmap_destroy.
1130 */
c3d9714e 1131 mutex_lock(&mddev->bitmap_info.mutex);
aa5cbd10
N
1132 bitmap = mddev->bitmap;
1133 if (bitmap == NULL) {
c3d9714e 1134 mutex_unlock(&mddev->bitmap_info.mutex);
4ad13663 1135 return;
aa5cbd10 1136 }
42a04b50 1137 if (time_before(jiffies, bitmap->daemon_lastrun
1b04be96 1138 + bitmap->mddev->bitmap_info.daemon_sleep))
7be3dfec
N
1139 goto done;
1140
32a7627c 1141 bitmap->daemon_lastrun = jiffies;
8311c29d
N
1142 if (bitmap->allclean) {
1143 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
aa5cbd10 1144 goto done;
8311c29d
N
1145 }
1146 bitmap->allclean = 1;
32a7627c 1147
be512691 1148 spin_lock_irqsave(&bitmap->lock, flags);
32a7627c
N
1149 for (j = 0; j < bitmap->chunks; j++) {
1150 bitmap_counter_t *bmc;
e384e585
N
1151 if (!bitmap->filemap) {
1152 if (!log)
1153 /* error or shutdown */
1154 break;
1155 } else
1156 page = filemap_get_page(bitmap, j);
32a7627c
N
1157
1158 if (page != lastpage) {
aa3163f8 1159 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1160 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1161 int need_write = test_page_attr(bitmap, page,
1162 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1163 if (need_write)
aa3163f8 1164 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1165
aa3163f8 1166 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1167 if (need_write) {
4ad13663 1168 write_page(bitmap, page, 0);
8311c29d
N
1169 bitmap->allclean = 0;
1170 }
be512691
N
1171 spin_lock_irqsave(&bitmap->lock, flags);
1172 j |= (PAGE_BITS - 1);
aa3163f8
N
1173 continue;
1174 }
1175
32a7627c 1176 /* grab the new page, sync and release the old */
32a7627c 1177 if (lastpage != NULL) {
ec7a3197 1178 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1179 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1180 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1181 write_page(bitmap, lastpage, 0);
32a7627c
N
1182 } else {
1183 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1184 spin_unlock_irqrestore(&bitmap->lock, flags);
1185 }
32a7627c
N
1186 } else
1187 spin_unlock_irqrestore(&bitmap->lock, flags);
1188 lastpage = page;
a0da84f3
NB
1189
1190 /* We are possibly going to clear some bits, so make
1191 * sure that events_cleared is up-to-date.
1192 */
ece5cff0
N
1193 if (bitmap->need_sync &&
1194 bitmap->mddev->bitmap_info.external == 0) {
a0da84f3
NB
1195 bitmap_super_t *sb;
1196 bitmap->need_sync = 0;
1197 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1198 sb->events_cleared =
1199 cpu_to_le64(bitmap->events_cleared);
1200 kunmap_atomic(sb, KM_USER0);
1201 write_page(bitmap, bitmap->sb_page, 1);
1202 }
32a7627c 1203 spin_lock_irqsave(&bitmap->lock, flags);
ece5cff0
N
1204 if (!bitmap->need_sync)
1205 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 1206 }
db305e50
N
1207 bmc = bitmap_get_counter(bitmap,
1208 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1209 &blocks, 0);
32a7627c 1210 if (bmc) {
8311c29d
N
1211 if (*bmc)
1212 bitmap->allclean = 0;
1213
32a7627c 1214 if (*bmc == 2) {
ac2f40be 1215 *bmc = 1; /* maybe clear the bit next time */
32a7627c 1216 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
ece5cff0 1217 } else if (*bmc == 1 && !bitmap->need_sync) {
32a7627c
N
1218 /* we can clear the bit */
1219 *bmc = 0;
db305e50
N
1220 bitmap_count_page(bitmap,
1221 (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
32a7627c
N
1222 -1);
1223
1224 /* clear the bit */
e384e585
N
1225 if (page) {
1226 paddr = kmap_atomic(page, KM_USER0);
1227 if (bitmap->flags & BITMAP_HOSTENDIAN)
1228 clear_bit(file_page_offset(bitmap, j),
1229 paddr);
1230 else
6b33aff3 1231 __test_and_clear_bit_le(file_page_offset(bitmap, j),
e384e585
N
1232 paddr);
1233 kunmap_atomic(paddr, KM_USER0);
1234 } else
1235 log->type->clear_region(log, j);
32a7627c 1236 }
be512691
N
1237 } else
1238 j |= PAGE_COUNTER_MASK;
32a7627c 1239 }
be512691 1240 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c
N
1241
1242 /* now sync the final page */
e384e585 1243 if (lastpage != NULL || log != NULL) {
32a7627c 1244 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1245 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1246 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1247 spin_unlock_irqrestore(&bitmap->lock, flags);
e384e585
N
1248 if (lastpage)
1249 write_page(bitmap, lastpage, 0);
1250 else
1251 if (log->type->flush(log))
1252 bitmap->flags |= BITMAP_WRITE_ERROR;
32a7627c
N
1253 } else {
1254 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1255 spin_unlock_irqrestore(&bitmap->lock, flags);
1256 }
32a7627c
N
1257 }
1258
7be3dfec 1259 done:
8311c29d 1260 if (bitmap->allclean == 0)
ac2f40be 1261 bitmap->mddev->thread->timeout =
1b04be96 1262 bitmap->mddev->bitmap_info.daemon_sleep;
c3d9714e 1263 mutex_unlock(&mddev->bitmap_info.mutex);
32a7627c
N
1264}
1265
32a7627c 1266static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
57dab0bd 1267 sector_t offset, sector_t *blocks,
32a7627c 1268 int create)
ee305ace
N
1269__releases(bitmap->lock)
1270__acquires(bitmap->lock)
32a7627c
N
1271{
1272 /* If 'create', we might release the lock and reclaim it.
1273 * The lock must have been taken with interrupts enabled.
1274 * If !create, we don't release the lock.
1275 */
1276 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1277 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1278 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1279 sector_t csize;
ef425673 1280 int err;
32a7627c 1281
ef425673
N
1282 err = bitmap_checkpage(bitmap, page, create);
1283
1284 if (bitmap->bp[page].hijacked ||
1285 bitmap->bp[page].map == NULL)
1286 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1287 PAGE_COUNTER_SHIFT - 1);
1288 else
32a7627c 1289 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
ef425673
N
1290 *blocks = csize - (offset & (csize - 1));
1291
1292 if (err < 0)
32a7627c 1293 return NULL;
ef425673 1294
32a7627c
N
1295 /* now locked ... */
1296
1297 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1298 /* should we use the first or second counter field
1299 * of the hijacked pointer? */
1300 int hi = (pageoff > PAGE_COUNTER_MASK);
32a7627c
N
1301 return &((bitmap_counter_t *)
1302 &bitmap->bp[page].map)[hi];
ef425673 1303 } else /* page is allocated */
32a7627c
N
1304 return (bitmap_counter_t *)
1305 &(bitmap->bp[page].map[pageoff]);
32a7627c
N
1306}
1307
4b6d287f 1308int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c 1309{
ac2f40be
N
1310 if (!bitmap)
1311 return 0;
4b6d287f
N
1312
1313 if (behind) {
696fcd53 1314 int bw;
4b6d287f 1315 atomic_inc(&bitmap->behind_writes);
696fcd53
PC
1316 bw = atomic_read(&bitmap->behind_writes);
1317 if (bw > bitmap->behind_writes_used)
1318 bitmap->behind_writes_used = bw;
1319
4b6d287f 1320 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
696fcd53 1321 bw, bitmap->max_write_behind);
4b6d287f
N
1322 }
1323
32a7627c 1324 while (sectors) {
57dab0bd 1325 sector_t blocks;
32a7627c
N
1326 bitmap_counter_t *bmc;
1327
1328 spin_lock_irq(&bitmap->lock);
1329 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1330 if (!bmc) {
1331 spin_unlock_irq(&bitmap->lock);
1332 return 0;
1333 }
1334
da6e1a32
NB
1335 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1336 DEFINE_WAIT(__wait);
1337 /* note that it is safe to do the prepare_to_wait
1338 * after the test as long as we do it before dropping
1339 * the spinlock.
1340 */
1341 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1342 TASK_UNINTERRUPTIBLE);
1343 spin_unlock_irq(&bitmap->lock);
7eaceacc 1344 io_schedule();
da6e1a32
NB
1345 finish_wait(&bitmap->overflow_wait, &__wait);
1346 continue;
1347 }
1348
ac2f40be 1349 switch (*bmc) {
32a7627c
N
1350 case 0:
1351 bitmap_file_set_bit(bitmap, offset);
ac2f40be 1352 bitmap_count_page(bitmap, offset, 1);
32a7627c
N
1353 /* fall through */
1354 case 1:
1355 *bmc = 2;
1356 }
da6e1a32 1357
32a7627c
N
1358 (*bmc)++;
1359
1360 spin_unlock_irq(&bitmap->lock);
1361
1362 offset += blocks;
1363 if (sectors > blocks)
1364 sectors -= blocks;
ac2f40be
N
1365 else
1366 sectors = 0;
32a7627c 1367 }
8311c29d 1368 bitmap->allclean = 0;
32a7627c
N
1369 return 0;
1370}
ac2f40be 1371EXPORT_SYMBOL(bitmap_startwrite);
32a7627c
N
1372
1373void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1374 int success, int behind)
32a7627c 1375{
ac2f40be
N
1376 if (!bitmap)
1377 return;
4b6d287f 1378 if (behind) {
e555190d
N
1379 if (atomic_dec_and_test(&bitmap->behind_writes))
1380 wake_up(&bitmap->behind_wait);
4b6d287f
N
1381 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1382 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1383 }
d0a4bb49
N
1384 if (bitmap->mddev->degraded)
1385 /* Never clear bits or update events_cleared when degraded */
1386 success = 0;
4b6d287f 1387
32a7627c 1388 while (sectors) {
57dab0bd 1389 sector_t blocks;
32a7627c
N
1390 unsigned long flags;
1391 bitmap_counter_t *bmc;
1392
1393 spin_lock_irqsave(&bitmap->lock, flags);
1394 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1395 if (!bmc) {
1396 spin_unlock_irqrestore(&bitmap->lock, flags);
1397 return;
1398 }
1399
a0da84f3
NB
1400 if (success &&
1401 bitmap->events_cleared < bitmap->mddev->events) {
1402 bitmap->events_cleared = bitmap->mddev->events;
1403 bitmap->need_sync = 1;
5ff5afff 1404 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
a0da84f3
NB
1405 }
1406
32a7627c
N
1407 if (!success && ! (*bmc & NEEDED_MASK))
1408 *bmc |= NEEDED_MASK;
1409
da6e1a32
NB
1410 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1411 wake_up(&bitmap->overflow_wait);
1412
32a7627c 1413 (*bmc)--;
ac2f40be 1414 if (*bmc <= 2)
32a7627c 1415 set_page_attr(bitmap,
e384e585
N
1416 filemap_get_page(
1417 bitmap,
1418 offset >> CHUNK_BLOCK_SHIFT(bitmap)),
32a7627c 1419 BITMAP_PAGE_CLEAN);
ac2f40be 1420
32a7627c
N
1421 spin_unlock_irqrestore(&bitmap->lock, flags);
1422 offset += blocks;
1423 if (sectors > blocks)
1424 sectors -= blocks;
ac2f40be
N
1425 else
1426 sectors = 0;
32a7627c
N
1427 }
1428}
ac2f40be 1429EXPORT_SYMBOL(bitmap_endwrite);
32a7627c 1430
57dab0bd 1431static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1187cf0a 1432 int degraded)
32a7627c
N
1433{
1434 bitmap_counter_t *bmc;
1435 int rv;
1436 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1437 *blocks = 1024;
1438 return 1; /* always resync if no bitmap */
1439 }
1440 spin_lock_irq(&bitmap->lock);
1441 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1442 rv = 0;
1443 if (bmc) {
1444 /* locked */
1445 if (RESYNC(*bmc))
1446 rv = 1;
1447 else if (NEEDED(*bmc)) {
1448 rv = 1;
6a806c51
N
1449 if (!degraded) { /* don't set/clear bits if degraded */
1450 *bmc |= RESYNC_MASK;
1451 *bmc &= ~NEEDED_MASK;
1452 }
32a7627c
N
1453 }
1454 }
1455 spin_unlock_irq(&bitmap->lock);
8311c29d 1456 bitmap->allclean = 0;
32a7627c
N
1457 return rv;
1458}
1459
57dab0bd 1460int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1187cf0a
N
1461 int degraded)
1462{
1463 /* bitmap_start_sync must always report on multiples of whole
1464 * pages, otherwise resync (which is very PAGE_SIZE based) will
1465 * get confused.
1466 * So call __bitmap_start_sync repeatedly (if needed) until
1467 * At least PAGE_SIZE>>9 blocks are covered.
1468 * Return the 'or' of the result.
1469 */
1470 int rv = 0;
57dab0bd 1471 sector_t blocks1;
1187cf0a
N
1472
1473 *blocks = 0;
1474 while (*blocks < (PAGE_SIZE>>9)) {
1475 rv |= __bitmap_start_sync(bitmap, offset,
1476 &blocks1, degraded);
1477 offset += blocks1;
1478 *blocks += blocks1;
1479 }
1480 return rv;
1481}
ac2f40be 1482EXPORT_SYMBOL(bitmap_start_sync);
1187cf0a 1483
57dab0bd 1484void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
32a7627c
N
1485{
1486 bitmap_counter_t *bmc;
1487 unsigned long flags;
ac2f40be
N
1488
1489 if (bitmap == NULL) {
32a7627c
N
1490 *blocks = 1024;
1491 return;
1492 }
1493 spin_lock_irqsave(&bitmap->lock, flags);
1494 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1495 if (bmc == NULL)
1496 goto unlock;
1497 /* locked */
32a7627c
N
1498 if (RESYNC(*bmc)) {
1499 *bmc &= ~RESYNC_MASK;
1500
1501 if (!NEEDED(*bmc) && aborted)
1502 *bmc |= NEEDED_MASK;
1503 else {
ac2f40be 1504 if (*bmc <= 2)
32a7627c
N
1505 set_page_attr(bitmap,
1506 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1507 BITMAP_PAGE_CLEAN);
32a7627c
N
1508 }
1509 }
1510 unlock:
1511 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1512 bitmap->allclean = 0;
32a7627c 1513}
ac2f40be 1514EXPORT_SYMBOL(bitmap_end_sync);
32a7627c
N
1515
1516void bitmap_close_sync(struct bitmap *bitmap)
1517{
1518 /* Sync has finished, and any bitmap chunks that weren't synced
1519 * properly have been aborted. It remains to us to clear the
1520 * RESYNC bit wherever it is still on
1521 */
1522 sector_t sector = 0;
57dab0bd 1523 sector_t blocks;
b47490c9
N
1524 if (!bitmap)
1525 return;
32a7627c
N
1526 while (sector < bitmap->mddev->resync_max_sectors) {
1527 bitmap_end_sync(bitmap, sector, &blocks, 0);
b47490c9
N
1528 sector += blocks;
1529 }
1530}
ac2f40be 1531EXPORT_SYMBOL(bitmap_close_sync);
b47490c9
N
1532
1533void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1534{
1535 sector_t s = 0;
57dab0bd 1536 sector_t blocks;
b47490c9
N
1537
1538 if (!bitmap)
1539 return;
1540 if (sector == 0) {
1541 bitmap->last_end_sync = jiffies;
1542 return;
1543 }
1544 if (time_before(jiffies, (bitmap->last_end_sync
1b04be96 1545 + bitmap->mddev->bitmap_info.daemon_sleep)))
b47490c9
N
1546 return;
1547 wait_event(bitmap->mddev->recovery_wait,
1548 atomic_read(&bitmap->mddev->recovery_active) == 0);
1549
75d3da43 1550 bitmap->mddev->curr_resync_completed = sector;
070dc6dd 1551 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
b47490c9
N
1552 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1553 s = 0;
1554 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1555 bitmap_end_sync(bitmap, s, &blocks, 0);
1556 s += blocks;
32a7627c 1557 }
b47490c9 1558 bitmap->last_end_sync = jiffies;
acb180b0 1559 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
32a7627c 1560}
ac2f40be 1561EXPORT_SYMBOL(bitmap_cond_end_sync);
32a7627c 1562
6a07997f 1563static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1564{
1565 /* For each chunk covered by any of these sectors, set the
193f1c93 1566 * counter to 1 and set resync_needed. They should all
32a7627c
N
1567 * be 0 at this point
1568 */
193f1c93 1569
57dab0bd 1570 sector_t secs;
193f1c93
N
1571 bitmap_counter_t *bmc;
1572 spin_lock_irq(&bitmap->lock);
1573 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1574 if (!bmc) {
32a7627c 1575 spin_unlock_irq(&bitmap->lock);
193f1c93 1576 return;
32a7627c 1577 }
ac2f40be 1578 if (!*bmc) {
193f1c93 1579 struct page *page;
ac2f40be 1580 *bmc = 1 | (needed ? NEEDED_MASK : 0);
193f1c93
N
1581 bitmap_count_page(bitmap, offset, 1);
1582 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1583 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1584 }
1585 spin_unlock_irq(&bitmap->lock);
8311c29d 1586 bitmap->allclean = 0;
32a7627c
N
1587}
1588
9b1d1dac
PC
1589/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1590void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1591{
1592 unsigned long chunk;
1593
1594 for (chunk = s; chunk <= e; chunk++) {
db305e50 1595 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
9b1d1dac
PC
1596 bitmap_set_memory_bits(bitmap, sec, 1);
1597 bitmap_file_set_bit(bitmap, sec);
ffa23322
N
1598 if (sec < bitmap->mddev->recovery_cp)
1599 /* We are asserting that the array is dirty,
1600 * so move the recovery_cp address back so
1601 * that it is obvious that it is dirty
1602 */
1603 bitmap->mddev->recovery_cp = sec;
9b1d1dac
PC
1604 }
1605}
1606
6b8b3e8a
N
1607/*
1608 * flush out any pending updates
1609 */
1610void bitmap_flush(mddev_t *mddev)
1611{
1612 struct bitmap *bitmap = mddev->bitmap;
42a04b50 1613 long sleep;
6b8b3e8a
N
1614
1615 if (!bitmap) /* there was no bitmap */
1616 return;
1617
1618 /* run the daemon_work three time to ensure everything is flushed
1619 * that can be
1620 */
1b04be96 1621 sleep = mddev->bitmap_info.daemon_sleep * 2;
42a04b50 1622 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1623 bitmap_daemon_work(mddev);
42a04b50 1624 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1625 bitmap_daemon_work(mddev);
42a04b50 1626 bitmap->daemon_lastrun -= sleep;
aa5cbd10 1627 bitmap_daemon_work(mddev);
6b8b3e8a
N
1628 bitmap_update_sb(bitmap);
1629}
1630
32a7627c
N
1631/*
1632 * free memory that was allocated
1633 */
3178b0db 1634static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1635{
1636 unsigned long k, pages;
1637 struct bitmap_page *bp;
32a7627c
N
1638
1639 if (!bitmap) /* there was no bitmap */
1640 return;
1641
32a7627c
N
1642 /* release the bitmap file and kill the daemon */
1643 bitmap_file_put(bitmap);
1644
1645 bp = bitmap->bp;
1646 pages = bitmap->pages;
1647
1648 /* free all allocated memory */
1649
32a7627c
N
1650 if (bp) /* deallocate the page memory */
1651 for (k = 0; k < pages; k++)
1652 if (bp[k].map && !bp[k].hijacked)
1653 kfree(bp[k].map);
1654 kfree(bp);
1655 kfree(bitmap);
1656}
aa5cbd10 1657
3178b0db
N
1658void bitmap_destroy(mddev_t *mddev)
1659{
1660 struct bitmap *bitmap = mddev->bitmap;
1661
1662 if (!bitmap) /* there was no bitmap */
1663 return;
1664
c3d9714e 1665 mutex_lock(&mddev->bitmap_info.mutex);
3178b0db 1666 mddev->bitmap = NULL; /* disconnect from the md device */
c3d9714e 1667 mutex_unlock(&mddev->bitmap_info.mutex);
b15c2e57
N
1668 if (mddev->thread)
1669 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db 1670
ece5cff0
N
1671 if (bitmap->sysfs_can_clear)
1672 sysfs_put(bitmap->sysfs_can_clear);
1673
3178b0db
N
1674 bitmap_free(bitmap);
1675}
32a7627c
N
1676
1677/*
1678 * initialize the bitmap structure
1679 * if this returns an error, bitmap_destroy must be called to do clean up
1680 */
1681int bitmap_create(mddev_t *mddev)
1682{
1683 struct bitmap *bitmap;
1f593903 1684 sector_t blocks = mddev->resync_max_sectors;
32a7627c
N
1685 unsigned long chunks;
1686 unsigned long pages;
c3d9714e 1687 struct file *file = mddev->bitmap_info.file;
32a7627c 1688 int err;
5ff5afff 1689 struct sysfs_dirent *bm = NULL;
32a7627c 1690
5f6e3c83 1691 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1692
e384e585
N
1693 if (!file
1694 && !mddev->bitmap_info.offset
1695 && !mddev->bitmap_info.log) /* bitmap disabled, nothing to do */
32a7627c
N
1696 return 0;
1697
c3d9714e 1698 BUG_ON(file && mddev->bitmap_info.offset);
e384e585 1699 BUG_ON(mddev->bitmap_info.offset && mddev->bitmap_info.log);
a654b9d8 1700
9ffae0cf 1701 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1702 if (!bitmap)
1703 return -ENOMEM;
1704
32a7627c 1705 spin_lock_init(&bitmap->lock);
ce25c31b
N
1706 atomic_set(&bitmap->pending_writes, 0);
1707 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1708 init_waitqueue_head(&bitmap->overflow_wait);
e555190d 1709 init_waitqueue_head(&bitmap->behind_wait);
ce25c31b 1710
32a7627c 1711 bitmap->mddev = mddev;
32a7627c 1712
5ff5afff
N
1713 if (mddev->kobj.sd)
1714 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
ece5cff0 1715 if (bm) {
3ff195b0 1716 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
ece5cff0
N
1717 sysfs_put(bm);
1718 } else
1719 bitmap->sysfs_can_clear = NULL;
1720
32a7627c 1721 bitmap->file = file;
ce25c31b
N
1722 if (file) {
1723 get_file(file);
ae8fa283
N
1724 /* As future accesses to this file will use bmap,
1725 * and bypass the page cache, we must sync the file
1726 * first.
1727 */
8018ab05 1728 vfs_fsync(file, 1);
ce25c31b 1729 }
42a04b50 1730 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
ece5cff0
N
1731 if (!mddev->bitmap_info.external)
1732 err = bitmap_read_sb(bitmap);
1733 else {
1734 err = 0;
1735 if (mddev->bitmap_info.chunksize == 0 ||
1736 mddev->bitmap_info.daemon_sleep == 0)
1737 /* chunksize and time_base need to be
1738 * set first. */
1739 err = -EINVAL;
1740 }
32a7627c 1741 if (err)
3178b0db 1742 goto error;
32a7627c 1743
624ce4f5 1744 bitmap->daemon_lastrun = jiffies;
42a04b50 1745 bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize);
32a7627c
N
1746
1747 /* now that chunksize and chunkshift are set, we can use these macros */
ac2f40be 1748 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1f593903 1749 CHUNK_BLOCK_SHIFT(bitmap);
ac2f40be 1750 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
32a7627c
N
1751
1752 BUG_ON(!pages);
1753
1754 bitmap->chunks = chunks;
1755 bitmap->pages = pages;
1756 bitmap->missing_pages = pages;
1757 bitmap->counter_bits = COUNTER_BITS;
1758
1759 bitmap->syncchunk = ~0UL;
1760
44456d37 1761#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1762 bitmap->bp = NULL;
1763#else
9ffae0cf 1764 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1765#endif
3178b0db 1766 err = -ENOMEM;
32a7627c 1767 if (!bitmap->bp)
3178b0db 1768 goto error;
32a7627c 1769
69e51b44
N
1770 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1771 pages, bmname(bitmap));
1772
1773 mddev->bitmap = bitmap;
1774
1775
1776 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1777
1778 error:
1779 bitmap_free(bitmap);
1780 return err;
1781}
1782
1783int bitmap_load(mddev_t *mddev)
1784{
1785 int err = 0;
1786 sector_t sector = 0;
1787 struct bitmap *bitmap = mddev->bitmap;
1788
1789 if (!bitmap)
1790 goto out;
1791
1792 /* Clear out old bitmap info first: Either there is none, or we
1793 * are resuming after someone else has possibly changed things,
1794 * so we should forget old cached info.
1795 * All chunks should be clean, but some might need_sync.
1796 */
1797 while (sector < mddev->resync_max_sectors) {
57dab0bd 1798 sector_t blocks;
69e51b44
N
1799 bitmap_start_sync(bitmap, sector, &blocks, 0);
1800 sector += blocks;
1801 }
1802 bitmap_close_sync(bitmap);
1803
e384e585
N
1804 if (mddev->bitmap_info.log) {
1805 unsigned long i;
1806 struct dm_dirty_log *log = mddev->bitmap_info.log;
1807 for (i = 0; i < bitmap->chunks; i++)
1808 if (!log->type->in_sync(log, i, 1))
1809 bitmap_set_memory_bits(bitmap,
1810 (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1811 1);
69e51b44
N
1812 } else {
1813 sector_t start = 0;
1814 if (mddev->degraded == 0
1815 || bitmap->events_cleared == mddev->events)
1816 /* no need to keep dirty bits to optimise a
1817 * re-add of a missing device */
1818 start = mddev->recovery_cp;
193f1c93 1819
69e51b44
N
1820 err = bitmap_init_from_disk(bitmap, start);
1821 }
32a7627c 1822 if (err)
69e51b44 1823 goto out;
3178b0db 1824
1b04be96 1825 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
9cd30fdc 1826 md_wakeup_thread(mddev->thread);
b15c2e57 1827
4ad13663
N
1828 bitmap_update_sb(bitmap);
1829
69e51b44
N
1830 if (bitmap->flags & BITMAP_WRITE_ERROR)
1831 err = -EIO;
1832out:
3178b0db 1833 return err;
32a7627c 1834}
69e51b44 1835EXPORT_SYMBOL_GPL(bitmap_load);
32a7627c 1836
43a70507
N
1837static ssize_t
1838location_show(mddev_t *mddev, char *page)
1839{
1840 ssize_t len;
ac2f40be 1841 if (mddev->bitmap_info.file)
43a70507 1842 len = sprintf(page, "file");
ac2f40be 1843 else if (mddev->bitmap_info.offset)
43a70507 1844 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
ac2f40be 1845 else
43a70507
N
1846 len = sprintf(page, "none");
1847 len += sprintf(page+len, "\n");
1848 return len;
1849}
1850
1851static ssize_t
1852location_store(mddev_t *mddev, const char *buf, size_t len)
1853{
1854
1855 if (mddev->pers) {
1856 if (!mddev->pers->quiesce)
1857 return -EBUSY;
1858 if (mddev->recovery || mddev->sync_thread)
1859 return -EBUSY;
1860 }
1861
1862 if (mddev->bitmap || mddev->bitmap_info.file ||
1863 mddev->bitmap_info.offset) {
1864 /* bitmap already configured. Only option is to clear it */
1865 if (strncmp(buf, "none", 4) != 0)
1866 return -EBUSY;
1867 if (mddev->pers) {
1868 mddev->pers->quiesce(mddev, 1);
1869 bitmap_destroy(mddev);
1870 mddev->pers->quiesce(mddev, 0);
1871 }
1872 mddev->bitmap_info.offset = 0;
1873 if (mddev->bitmap_info.file) {
1874 struct file *f = mddev->bitmap_info.file;
1875 mddev->bitmap_info.file = NULL;
1876 restore_bitmap_write_access(f);
1877 fput(f);
1878 }
1879 } else {
1880 /* No bitmap, OK to set a location */
1881 long long offset;
1882 if (strncmp(buf, "none", 4) == 0)
1883 /* nothing to be done */;
1884 else if (strncmp(buf, "file:", 5) == 0) {
1885 /* Not supported yet */
1886 return -EINVAL;
1887 } else {
1888 int rv;
1889 if (buf[0] == '+')
1890 rv = strict_strtoll(buf+1, 10, &offset);
1891 else
1892 rv = strict_strtoll(buf, 10, &offset);
1893 if (rv)
1894 return rv;
1895 if (offset == 0)
1896 return -EINVAL;
ece5cff0
N
1897 if (mddev->bitmap_info.external == 0 &&
1898 mddev->major_version == 0 &&
43a70507
N
1899 offset != mddev->bitmap_info.default_offset)
1900 return -EINVAL;
1901 mddev->bitmap_info.offset = offset;
1902 if (mddev->pers) {
1903 mddev->pers->quiesce(mddev, 1);
1904 rv = bitmap_create(mddev);
1905 if (rv) {
1906 bitmap_destroy(mddev);
1907 mddev->bitmap_info.offset = 0;
1908 }
1909 mddev->pers->quiesce(mddev, 0);
1910 if (rv)
1911 return rv;
1912 }
1913 }
1914 }
1915 if (!mddev->external) {
1916 /* Ensure new bitmap info is stored in
1917 * metadata promptly.
1918 */
1919 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1920 md_wakeup_thread(mddev->thread);
1921 }
1922 return len;
1923}
1924
1925static struct md_sysfs_entry bitmap_location =
1926__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1927
1928static ssize_t
1929timeout_show(mddev_t *mddev, char *page)
1930{
1931 ssize_t len;
1932 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1933 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
ac2f40be 1934
43a70507
N
1935 len = sprintf(page, "%lu", secs);
1936 if (jifs)
1937 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1938 len += sprintf(page+len, "\n");
1939 return len;
1940}
1941
1942static ssize_t
1943timeout_store(mddev_t *mddev, const char *buf, size_t len)
1944{
1945 /* timeout can be set at any time */
1946 unsigned long timeout;
1947 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1948 if (rv)
1949 return rv;
1950
1951 /* just to make sure we don't overflow... */
1952 if (timeout >= LONG_MAX / HZ)
1953 return -EINVAL;
1954
1955 timeout = timeout * HZ / 10000;
1956
1957 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1958 timeout = MAX_SCHEDULE_TIMEOUT-1;
1959 if (timeout < 1)
1960 timeout = 1;
1961 mddev->bitmap_info.daemon_sleep = timeout;
1962 if (mddev->thread) {
1963 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1964 * the bitmap is all clean and we don't need to
1965 * adjust the timeout right now
1966 */
1967 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1968 mddev->thread->timeout = timeout;
1969 md_wakeup_thread(mddev->thread);
1970 }
1971 }
1972 return len;
1973}
1974
1975static struct md_sysfs_entry bitmap_timeout =
1976__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1977
1978static ssize_t
1979backlog_show(mddev_t *mddev, char *page)
1980{
1981 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1982}
1983
1984static ssize_t
1985backlog_store(mddev_t *mddev, const char *buf, size_t len)
1986{
1987 unsigned long backlog;
1988 int rv = strict_strtoul(buf, 10, &backlog);
1989 if (rv)
1990 return rv;
1991 if (backlog > COUNTER_MAX)
1992 return -EINVAL;
1993 mddev->bitmap_info.max_write_behind = backlog;
1994 return len;
1995}
1996
1997static struct md_sysfs_entry bitmap_backlog =
1998__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
1999
2000static ssize_t
2001chunksize_show(mddev_t *mddev, char *page)
2002{
2003 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2004}
2005
2006static ssize_t
2007chunksize_store(mddev_t *mddev, const char *buf, size_t len)
2008{
2009 /* Can only be changed when no bitmap is active */
2010 int rv;
2011 unsigned long csize;
2012 if (mddev->bitmap)
2013 return -EBUSY;
2014 rv = strict_strtoul(buf, 10, &csize);
2015 if (rv)
2016 return rv;
2017 if (csize < 512 ||
2018 !is_power_of_2(csize))
2019 return -EINVAL;
2020 mddev->bitmap_info.chunksize = csize;
2021 return len;
2022}
2023
2024static struct md_sysfs_entry bitmap_chunksize =
2025__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2026
ece5cff0
N
2027static ssize_t metadata_show(mddev_t *mddev, char *page)
2028{
2029 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2030 ? "external" : "internal"));
2031}
2032
2033static ssize_t metadata_store(mddev_t *mddev, const char *buf, size_t len)
2034{
2035 if (mddev->bitmap ||
2036 mddev->bitmap_info.file ||
2037 mddev->bitmap_info.offset)
2038 return -EBUSY;
2039 if (strncmp(buf, "external", 8) == 0)
2040 mddev->bitmap_info.external = 1;
2041 else if (strncmp(buf, "internal", 8) == 0)
2042 mddev->bitmap_info.external = 0;
2043 else
2044 return -EINVAL;
2045 return len;
2046}
2047
2048static struct md_sysfs_entry bitmap_metadata =
2049__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2050
2051static ssize_t can_clear_show(mddev_t *mddev, char *page)
2052{
2053 int len;
2054 if (mddev->bitmap)
2055 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2056 "false" : "true"));
2057 else
2058 len = sprintf(page, "\n");
2059 return len;
2060}
2061
2062static ssize_t can_clear_store(mddev_t *mddev, const char *buf, size_t len)
2063{
2064 if (mddev->bitmap == NULL)
2065 return -ENOENT;
2066 if (strncmp(buf, "false", 5) == 0)
2067 mddev->bitmap->need_sync = 1;
2068 else if (strncmp(buf, "true", 4) == 0) {
2069 if (mddev->degraded)
2070 return -EBUSY;
2071 mddev->bitmap->need_sync = 0;
2072 } else
2073 return -EINVAL;
2074 return len;
2075}
2076
2077static struct md_sysfs_entry bitmap_can_clear =
2078__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2079
696fcd53
PC
2080static ssize_t
2081behind_writes_used_show(mddev_t *mddev, char *page)
2082{
2083 if (mddev->bitmap == NULL)
2084 return sprintf(page, "0\n");
2085 return sprintf(page, "%lu\n",
2086 mddev->bitmap->behind_writes_used);
2087}
2088
2089static ssize_t
2090behind_writes_used_reset(mddev_t *mddev, const char *buf, size_t len)
2091{
2092 if (mddev->bitmap)
2093 mddev->bitmap->behind_writes_used = 0;
2094 return len;
2095}
2096
2097static struct md_sysfs_entry max_backlog_used =
2098__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2099 behind_writes_used_show, behind_writes_used_reset);
2100
43a70507
N
2101static struct attribute *md_bitmap_attrs[] = {
2102 &bitmap_location.attr,
2103 &bitmap_timeout.attr,
2104 &bitmap_backlog.attr,
2105 &bitmap_chunksize.attr,
ece5cff0
N
2106 &bitmap_metadata.attr,
2107 &bitmap_can_clear.attr,
696fcd53 2108 &max_backlog_used.attr,
43a70507
N
2109 NULL
2110};
2111struct attribute_group md_bitmap_group = {
2112 .name = "bitmap",
2113 .attrs = md_bitmap_attrs,
2114};
2115