]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - drivers/md/bitmap.c
Remove invalidate_partition call from do_md_stop.
[mirror_ubuntu-eoan-kernel.git] / drivers / md / bitmap.c
CommitLineData
32a7627c
N
1/*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
32a7627c
N
10 */
11
12/*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
16 * wait if count gets too high, wake when it drops to half.
32a7627c
N
17 */
18
19#include <linux/module.h>
32a7627c
N
20#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/init.h>
32a7627c
N
23#include <linux/timer.h>
24#include <linux/sched.h>
25#include <linux/list.h>
26#include <linux/file.h>
27#include <linux/mount.h>
28#include <linux/buffer_head.h>
29#include <linux/raid/md.h>
30#include <linux/raid/bitmap.h>
31
32/* debug macros */
33
34#define DEBUG 0
35
36#if DEBUG
37/* these are for debugging purposes only! */
38
39/* define one and only one of these */
40#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
41#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
42#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
43#define INJECT_FAULTS_4 0 /* undef */
44#define INJECT_FAULTS_5 0 /* undef */
45#define INJECT_FAULTS_6 0
46
47/* if these are defined, the driver will fail! debug only */
48#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
49#define INJECT_FATAL_FAULT_2 0 /* undef */
50#define INJECT_FATAL_FAULT_3 0 /* undef */
51#endif
52
53//#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
54#define DPRINTK(x...) do { } while(0)
55
56#ifndef PRINTK
57# if DEBUG > 0
58# define PRINTK(x...) printk(KERN_DEBUG x)
59# else
60# define PRINTK(x...)
61# endif
62#endif
63
64static inline char * bmname(struct bitmap *bitmap)
65{
66 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
67}
68
32a7627c
N
69
70/*
71 * just a placeholder - calls kmalloc for bitmap pages
72 */
73static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
74{
75 unsigned char *page;
76
44456d37 77#ifdef INJECT_FAULTS_1
32a7627c
N
78 page = NULL;
79#else
80 page = kmalloc(PAGE_SIZE, GFP_NOIO);
81#endif
82 if (!page)
83 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
84 else
a654b9d8 85 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
86 bmname(bitmap), page);
87 return page;
88}
89
90/*
91 * for now just a placeholder -- just calls kfree for bitmap pages
92 */
93static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
94{
95 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
96 kfree(page);
97}
98
99/*
100 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
101 *
102 * 1) check to see if this page is allocated, if it's not then try to alloc
103 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
104 * page pointer directly as a counter
105 *
106 * if we find our page, we increment the page's refcount so that it stays
107 * allocated while we're using it
108 */
109static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
110{
111 unsigned char *mappage;
112
113 if (page >= bitmap->pages) {
114 printk(KERN_ALERT
115 "%s: invalid bitmap page request: %lu (> %lu)\n",
116 bmname(bitmap), page, bitmap->pages-1);
117 return -EINVAL;
118 }
119
120
121 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
122 return 0;
123
124 if (bitmap->bp[page].map) /* page is already allocated, just return */
125 return 0;
126
127 if (!create)
128 return -ENOENT;
129
130 spin_unlock_irq(&bitmap->lock);
131
132 /* this page has not been allocated yet */
133
134 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
135 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
136 bmname(bitmap));
137 /* failed - set the hijacked flag so that we can use the
138 * pointer as a counter */
139 spin_lock_irq(&bitmap->lock);
140 if (!bitmap->bp[page].map)
141 bitmap->bp[page].hijacked = 1;
142 goto out;
143 }
144
145 /* got a page */
146
147 spin_lock_irq(&bitmap->lock);
148
149 /* recheck the page */
150
151 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
152 /* somebody beat us to getting the page */
153 bitmap_free_page(bitmap, mappage);
154 return 0;
155 }
156
157 /* no page was in place and we have one, so install it */
158
159 memset(mappage, 0, PAGE_SIZE);
160 bitmap->bp[page].map = mappage;
161 bitmap->missing_pages--;
162out:
163 return 0;
164}
165
166
167/* if page is completely empty, put it back on the free list, or dealloc it */
168/* if page was hijacked, unmark the flag so it might get alloced next time */
169/* Note: lock should be held when calling this */
858119e1 170static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
171{
172 char *ptr;
173
174 if (bitmap->bp[page].count) /* page is still busy */
175 return;
176
177 /* page is no longer in use, it can be released */
178
179 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
180 bitmap->bp[page].hijacked = 0;
181 bitmap->bp[page].map = NULL;
182 return;
183 }
184
185 /* normal case, free the page */
186
187#if 0
188/* actually ... let's not. We will probably need the page again exactly when
189 * memory is tight and we are flusing to disk
190 */
191 return;
192#else
193 ptr = bitmap->bp[page].map;
194 bitmap->bp[page].map = NULL;
195 bitmap->missing_pages++;
196 bitmap_free_page(bitmap, ptr);
197 return;
198#endif
199}
200
201
202/*
203 * bitmap file handling - read and write the bitmap file and its superblock
204 */
205
32a7627c
N
206/*
207 * basic page I/O operations
208 */
209
a654b9d8
N
210/* IO operations when bitmap is stored near all superblocks */
211static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
212{
213 /* choose a good rdev and read the page from there */
214
215 mdk_rdev_t *rdev;
216 struct list_head *tmp;
217 struct page *page = alloc_page(GFP_KERNEL);
218 sector_t target;
219
220 if (!page)
221 return ERR_PTR(-ENOMEM);
a654b9d8 222
d089c6af 223 rdev_for_each(rdev, tmp, mddev) {
b2d444d7
N
224 if (! test_bit(In_sync, &rdev->flags)
225 || test_bit(Faulty, &rdev->flags))
ab904d63
N
226 continue;
227
0f420358 228 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
a654b9d8 229
ab904d63
N
230 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
231 page->index = index;
ce25c31b
N
232 attach_page_buffers(page, NULL); /* so that free_buffer will
233 * quietly no-op */
ab904d63
N
234 return page;
235 }
236 }
237 return ERR_PTR(-EIO);
a654b9d8 238
a654b9d8
N
239}
240
ab6085c7 241static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
a654b9d8
N
242{
243 mdk_rdev_t *rdev;
ab6085c7 244 mddev_t *mddev = bitmap->mddev;
a654b9d8 245
4b80991c
N
246 rcu_read_lock();
247 rdev_for_each_rcu(rdev, mddev)
b2d444d7 248 if (test_bit(In_sync, &rdev->flags)
ab6085c7
N
249 && !test_bit(Faulty, &rdev->flags)) {
250 int size = PAGE_SIZE;
251 if (page->index == bitmap->file_pages-1)
252 size = roundup(bitmap->last_page_size,
253 bdev_hardsect_size(rdev->bdev));
f0d76d70
N
254 /* Just make sure we aren't corrupting data or
255 * metadata
256 */
257 if (bitmap->offset < 0) {
258 /* DATA BITMAP METADATA */
259 if (bitmap->offset
85bfb4da 260 + (long)(page->index * (PAGE_SIZE/512))
f0d76d70
N
261 + size/512 > 0)
262 /* bitmap runs in to metadata */
4b80991c 263 goto bad_alignment;
f0d76d70 264 if (rdev->data_offset + mddev->size*2
0f420358 265 > rdev->sb_start + bitmap->offset)
f0d76d70 266 /* data runs in to bitmap */
4b80991c 267 goto bad_alignment;
0f420358 268 } else if (rdev->sb_start < rdev->data_offset) {
f0d76d70 269 /* METADATA BITMAP DATA */
0f420358 270 if (rdev->sb_start
f0d76d70
N
271 + bitmap->offset
272 + page->index*(PAGE_SIZE/512) + size/512
273 > rdev->data_offset)
274 /* bitmap runs in to data */
4b80991c 275 goto bad_alignment;
f0d76d70
N
276 } else {
277 /* DATA METADATA BITMAP - no problems */
278 }
a654b9d8 279 md_super_write(mddev, rdev,
0f420358 280 rdev->sb_start + bitmap->offset
a654b9d8 281 + page->index * (PAGE_SIZE/512),
ab6085c7 282 size,
a654b9d8 283 page);
ab6085c7 284 }
4b80991c 285 rcu_read_unlock();
a654b9d8
N
286
287 if (wait)
a9701a30 288 md_super_wait(mddev);
a654b9d8 289 return 0;
4b80991c
N
290
291 bad_alignment:
292 rcu_read_unlock();
293 return -EINVAL;
a654b9d8
N
294}
295
4ad13663 296static void bitmap_file_kick(struct bitmap *bitmap);
32a7627c 297/*
a654b9d8 298 * write out a page to a file
32a7627c 299 */
4ad13663 300static void write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c 301{
d785a06a 302 struct buffer_head *bh;
32a7627c 303
f0d76d70
N
304 if (bitmap->file == NULL) {
305 switch (write_sb_page(bitmap, page, wait)) {
306 case -EINVAL:
307 bitmap->flags |= BITMAP_WRITE_ERROR;
f0d76d70 308 }
4ad13663 309 } else {
a654b9d8 310
4ad13663 311 bh = page_buffers(page);
c708443c 312
4ad13663
N
313 while (bh && bh->b_blocknr) {
314 atomic_inc(&bitmap->pending_writes);
315 set_buffer_locked(bh);
316 set_buffer_mapped(bh);
317 submit_bh(WRITE, bh);
318 bh = bh->b_this_page;
319 }
d785a06a 320
4ad13663
N
321 if (wait) {
322 wait_event(bitmap->write_wait,
323 atomic_read(&bitmap->pending_writes)==0);
324 }
8a5e9cf1 325 }
4ad13663
N
326 if (bitmap->flags & BITMAP_WRITE_ERROR)
327 bitmap_file_kick(bitmap);
d785a06a
N
328}
329
330static void end_bitmap_write(struct buffer_head *bh, int uptodate)
331{
332 struct bitmap *bitmap = bh->b_private;
333 unsigned long flags;
32a7627c 334
d785a06a
N
335 if (!uptodate) {
336 spin_lock_irqsave(&bitmap->lock, flags);
337 bitmap->flags |= BITMAP_WRITE_ERROR;
338 spin_unlock_irqrestore(&bitmap->lock, flags);
32a7627c 339 }
d785a06a
N
340 if (atomic_dec_and_test(&bitmap->pending_writes))
341 wake_up(&bitmap->write_wait);
342}
32a7627c 343
d785a06a
N
344/* copied from buffer.c */
345static void
346__clear_page_buffers(struct page *page)
347{
348 ClearPagePrivate(page);
349 set_page_private(page, 0);
350 page_cache_release(page);
351}
352static void free_buffers(struct page *page)
353{
354 struct buffer_head *bh = page_buffers(page);
77ad4bc7 355
d785a06a
N
356 while (bh) {
357 struct buffer_head *next = bh->b_this_page;
358 free_buffer_head(bh);
359 bh = next;
77ad4bc7 360 }
d785a06a
N
361 __clear_page_buffers(page);
362 put_page(page);
32a7627c
N
363}
364
d785a06a
N
365/* read a page from a file.
366 * We both read the page, and attach buffers to the page to record the
367 * address of each block (using bmap). These addresses will be used
368 * to write the block later, completely bypassing the filesystem.
369 * This usage is similar to how swap files are handled, and allows us
370 * to write to a file with no concerns of memory allocation failing.
371 */
32a7627c 372static struct page *read_page(struct file *file, unsigned long index,
d785a06a
N
373 struct bitmap *bitmap,
374 unsigned long count)
32a7627c 375{
32a7627c 376 struct page *page = NULL;
c649bb9c 377 struct inode *inode = file->f_path.dentry->d_inode;
d785a06a
N
378 struct buffer_head *bh;
379 sector_t block;
32a7627c 380
2d1f3b5d
N
381 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
382 (unsigned long long)index << PAGE_SHIFT);
32a7627c 383
d785a06a
N
384 page = alloc_page(GFP_KERNEL);
385 if (!page)
386 page = ERR_PTR(-ENOMEM);
32a7627c
N
387 if (IS_ERR(page))
388 goto out;
d785a06a 389
d785a06a
N
390 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
391 if (!bh) {
2d1f3b5d 392 put_page(page);
d785a06a 393 page = ERR_PTR(-ENOMEM);
32a7627c
N
394 goto out;
395 }
d785a06a
N
396 attach_page_buffers(page, bh);
397 block = index << (PAGE_SHIFT - inode->i_blkbits);
398 while (bh) {
399 if (count == 0)
400 bh->b_blocknr = 0;
401 else {
402 bh->b_blocknr = bmap(inode, block);
403 if (bh->b_blocknr == 0) {
404 /* Cannot use this file! */
405 free_buffers(page);
406 page = ERR_PTR(-EINVAL);
407 goto out;
408 }
409 bh->b_bdev = inode->i_sb->s_bdev;
410 if (count < (1<<inode->i_blkbits))
411 count = 0;
412 else
413 count -= (1<<inode->i_blkbits);
414
415 bh->b_end_io = end_bitmap_write;
416 bh->b_private = bitmap;
ce25c31b
N
417 atomic_inc(&bitmap->pending_writes);
418 set_buffer_locked(bh);
419 set_buffer_mapped(bh);
420 submit_bh(READ, bh);
d785a06a
N
421 }
422 block++;
423 bh = bh->b_this_page;
424 }
d785a06a 425 page->index = index;
ce25c31b
N
426
427 wait_event(bitmap->write_wait,
428 atomic_read(&bitmap->pending_writes)==0);
429 if (bitmap->flags & BITMAP_WRITE_ERROR) {
430 free_buffers(page);
431 page = ERR_PTR(-EIO);
432 }
32a7627c
N
433out:
434 if (IS_ERR(page))
435 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
436 (int)PAGE_SIZE,
437 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
438 PTR_ERR(page));
439 return page;
440}
441
442/*
443 * bitmap file superblock operations
444 */
445
446/* update the event counter and sync the superblock to disk */
4ad13663 447void bitmap_update_sb(struct bitmap *bitmap)
32a7627c
N
448{
449 bitmap_super_t *sb;
450 unsigned long flags;
451
452 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
4ad13663 453 return;
32a7627c
N
454 spin_lock_irqsave(&bitmap->lock, flags);
455 if (!bitmap->sb_page) { /* no superblock */
456 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 457 return;
32a7627c 458 }
32a7627c 459 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 460 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 461 sb->events = cpu_to_le64(bitmap->mddev->events);
a0da84f3
NB
462 if (bitmap->mddev->events < bitmap->events_cleared) {
463 /* rocking back to read-only */
464 bitmap->events_cleared = bitmap->mddev->events;
465 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
466 }
ea03aff9 467 kunmap_atomic(sb, KM_USER0);
4ad13663 468 write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
469}
470
471/* print out the bitmap file superblock */
472void bitmap_print_sb(struct bitmap *bitmap)
473{
474 bitmap_super_t *sb;
475
476 if (!bitmap || !bitmap->sb_page)
477 return;
ea03aff9 478 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 479 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
480 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
481 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
482 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
483 *(__u32 *)(sb->uuid+0),
484 *(__u32 *)(sb->uuid+4),
485 *(__u32 *)(sb->uuid+8),
486 *(__u32 *)(sb->uuid+12));
a2cff26a 487 printk(KERN_DEBUG " events: %llu\n",
32a7627c 488 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 489 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 490 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
491 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
492 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
493 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
494 printk(KERN_DEBUG " sync size: %llu KB\n",
495 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 496 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 497 kunmap_atomic(sb, KM_USER0);
32a7627c
N
498}
499
500/* read the superblock from the bitmap file and initialize some bitmap fields */
501static int bitmap_read_sb(struct bitmap *bitmap)
502{
503 char *reason = NULL;
504 bitmap_super_t *sb;
4b6d287f 505 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
506 unsigned long long events;
507 int err = -EINVAL;
508
509 /* page 0 is the superblock, read it... */
f49d5e62
N
510 if (bitmap->file) {
511 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
512 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
513
514 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
515 } else {
a654b9d8 516 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
a654b9d8 517 }
32a7627c
N
518 if (IS_ERR(bitmap->sb_page)) {
519 err = PTR_ERR(bitmap->sb_page);
520 bitmap->sb_page = NULL;
521 return err;
522 }
523
ea03aff9 524 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 525
32a7627c
N
526 chunksize = le32_to_cpu(sb->chunksize);
527 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 528 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
529
530 /* verify that the bitmap-specific fields are valid */
531 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
532 reason = "bad magic";
bd926c63
N
533 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
534 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 535 reason = "unrecognized superblock version";
7dd5d34c
N
536 else if (chunksize < PAGE_SIZE)
537 reason = "bitmap chunksize too small";
32a7627c
N
538 else if ((1 << ffz(~chunksize)) != chunksize)
539 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
540 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
541 reason = "daemon sleep period out of range";
4b6d287f
N
542 else if (write_behind > COUNTER_MAX)
543 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
544 if (reason) {
545 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
546 bmname(bitmap), reason);
547 goto out;
548 }
549
550 /* keep the array size field of the bitmap superblock up to date */
551 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
552
553 if (!bitmap->mddev->persistent)
554 goto success;
555
556 /*
557 * if we have a persistent array superblock, compare the
558 * bitmap's UUID and event counter to the mddev's
559 */
560 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
561 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
562 bmname(bitmap));
563 goto out;
564 }
565 events = le64_to_cpu(sb->events);
566 if (events < bitmap->mddev->events) {
567 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
568 "-- forcing full recovery\n", bmname(bitmap), events,
569 (unsigned long long) bitmap->mddev->events);
4f2e639a 570 sb->state |= cpu_to_le32(BITMAP_STALE);
32a7627c
N
571 }
572success:
573 /* assign fields using values from superblock */
574 bitmap->chunksize = chunksize;
575 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 576 bitmap->daemon_lastrun = jiffies;
4b6d287f 577 bitmap->max_write_behind = write_behind;
4f2e639a 578 bitmap->flags |= le32_to_cpu(sb->state);
bd926c63
N
579 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
580 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 581 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
4f2e639a 582 if (sb->state & cpu_to_le32(BITMAP_STALE))
6a07997f 583 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
584 err = 0;
585out:
ea03aff9 586 kunmap_atomic(sb, KM_USER0);
32a7627c
N
587 if (err)
588 bitmap_print_sb(bitmap);
589 return err;
590}
591
592enum bitmap_mask_op {
593 MASK_SET,
594 MASK_UNSET
595};
596
4ad13663
N
597/* record the state of the bitmap in the superblock. Return the old value */
598static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
599 enum bitmap_mask_op op)
32a7627c
N
600{
601 bitmap_super_t *sb;
602 unsigned long flags;
4ad13663 603 int old;
32a7627c
N
604
605 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 606 if (!bitmap->sb_page) { /* can't set the state */
32a7627c 607 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 608 return 0;
32a7627c 609 }
32a7627c 610 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 611 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
4ad13663 612 old = le32_to_cpu(sb->state) & bits;
32a7627c 613 switch (op) {
4f2e639a 614 case MASK_SET: sb->state |= cpu_to_le32(bits);
32a7627c 615 break;
4f2e639a 616 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
32a7627c
N
617 break;
618 default: BUG();
619 }
ea03aff9 620 kunmap_atomic(sb, KM_USER0);
4ad13663 621 return old;
32a7627c
N
622}
623
624/*
625 * general bitmap file operations
626 */
627
628/* calculate the index of the page that contains this bit */
629static inline unsigned long file_page_index(unsigned long chunk)
630{
631 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
632}
633
634/* calculate the (bit) offset of this bit within a page */
635static inline unsigned long file_page_offset(unsigned long chunk)
636{
637 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
638}
639
640/*
641 * return a pointer to the page in the filemap that contains the given bit
642 *
643 * this lookup is complicated by the fact that the bitmap sb might be exactly
644 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
645 * 0 or page 1
646 */
647static inline struct page *filemap_get_page(struct bitmap *bitmap,
648 unsigned long chunk)
649{
9b1d1dac 650 if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
32a7627c
N
651 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
652}
653
654
655static void bitmap_file_unmap(struct bitmap *bitmap)
656{
657 struct page **map, *sb_page;
658 unsigned long *attr;
659 int pages;
660 unsigned long flags;
661
662 spin_lock_irqsave(&bitmap->lock, flags);
663 map = bitmap->filemap;
664 bitmap->filemap = NULL;
665 attr = bitmap->filemap_attr;
666 bitmap->filemap_attr = NULL;
667 pages = bitmap->file_pages;
668 bitmap->file_pages = 0;
669 sb_page = bitmap->sb_page;
670 bitmap->sb_page = NULL;
671 spin_unlock_irqrestore(&bitmap->lock, flags);
672
673 while (pages--)
674 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
d785a06a 675 free_buffers(map[pages]);
32a7627c
N
676 kfree(map);
677 kfree(attr);
678
d785a06a
N
679 if (sb_page)
680 free_buffers(sb_page);
32a7627c
N
681}
682
683static void bitmap_file_put(struct bitmap *bitmap)
684{
685 struct file *file;
32a7627c
N
686 unsigned long flags;
687
688 spin_lock_irqsave(&bitmap->lock, flags);
689 file = bitmap->file;
690 bitmap->file = NULL;
691 spin_unlock_irqrestore(&bitmap->lock, flags);
692
d785a06a
N
693 if (file)
694 wait_event(bitmap->write_wait,
695 atomic_read(&bitmap->pending_writes)==0);
32a7627c
N
696 bitmap_file_unmap(bitmap);
697
d785a06a 698 if (file) {
c649bb9c 699 struct inode *inode = file->f_path.dentry->d_inode;
fc0ecff6 700 invalidate_mapping_pages(inode->i_mapping, 0, -1);
32a7627c 701 fput(file);
d785a06a 702 }
32a7627c
N
703}
704
705
706/*
707 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
708 * then it is no longer reliable, so we stop using it and we mark the file
709 * as failed in the superblock
710 */
711static void bitmap_file_kick(struct bitmap *bitmap)
712{
713 char *path, *ptr = NULL;
714
4ad13663
N
715 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
716 bitmap_update_sb(bitmap);
32a7627c 717
4ad13663
N
718 if (bitmap->file) {
719 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
720 if (path)
6bcfd601
CH
721 ptr = d_path(&bitmap->file->f_path, path,
722 PAGE_SIZE);
723
32a7627c 724
4ad13663
N
725 printk(KERN_ALERT
726 "%s: kicking failed bitmap file %s from array!\n",
6bcfd601 727 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
32a7627c 728
4ad13663
N
729 kfree(path);
730 } else
731 printk(KERN_ALERT
732 "%s: disabling internal bitmap due to errors\n",
733 bmname(bitmap));
a654b9d8 734 }
32a7627c
N
735
736 bitmap_file_put(bitmap);
737
738 return;
739}
740
741enum bitmap_page_attr {
e16b68b6
N
742 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
743 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
744 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
745};
746
747static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
748 enum bitmap_page_attr attr)
749{
e16b68b6 750 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
751}
752
753static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
754 enum bitmap_page_attr attr)
755{
e16b68b6 756 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
757}
758
ec7a3197
N
759static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
760 enum bitmap_page_attr attr)
32a7627c 761{
e16b68b6 762 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
763}
764
765/*
766 * bitmap_file_set_bit -- called before performing a write to the md device
767 * to set (and eventually sync) a particular bit in the bitmap file
768 *
769 * we set the bit immediately, then we record the page number so that
770 * when an unplug occurs, we can flush the dirty pages out to disk
771 */
772static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
773{
774 unsigned long bit;
775 struct page *page;
776 void *kaddr;
777 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
778
a654b9d8 779 if (!bitmap->filemap) {
32a7627c
N
780 return;
781 }
782
783 page = filemap_get_page(bitmap, chunk);
9b1d1dac 784 if (!page) return;
32a7627c
N
785 bit = file_page_offset(chunk);
786
32a7627c
N
787 /* set the bit */
788 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
789 if (bitmap->flags & BITMAP_HOSTENDIAN)
790 set_bit(bit, kaddr);
791 else
792 ext2_set_bit(bit, kaddr);
32a7627c
N
793 kunmap_atomic(kaddr, KM_USER0);
794 PRINTK("set file bit %lu page %lu\n", bit, page->index);
795
796 /* record page number so it gets flushed to disk when unplug occurs */
797 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
798
799}
800
801/* this gets called when the md device is ready to unplug its underlying
802 * (slave) device queues -- before we let any writes go down, we need to
803 * sync the dirty pages of the bitmap file to disk */
4ad13663 804void bitmap_unplug(struct bitmap *bitmap)
32a7627c 805{
ec7a3197
N
806 unsigned long i, flags;
807 int dirty, need_write;
32a7627c
N
808 struct page *page;
809 int wait = 0;
810
811 if (!bitmap)
4ad13663 812 return;
32a7627c
N
813
814 /* look at each page to see if there are any set bits that need to be
815 * flushed out to disk */
816 for (i = 0; i < bitmap->file_pages; i++) {
817 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 818 if (!bitmap->filemap) {
32a7627c 819 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 820 return;
32a7627c
N
821 }
822 page = bitmap->filemap[i];
ec7a3197
N
823 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
824 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
825 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
826 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 827 if (dirty)
32a7627c
N
828 wait = 1;
829 spin_unlock_irqrestore(&bitmap->lock, flags);
830
d785a06a 831 if (dirty | need_write)
4ad13663 832 write_page(bitmap, page, 0);
32a7627c
N
833 }
834 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0 835 if (bitmap->file)
d785a06a
N
836 wait_event(bitmap->write_wait,
837 atomic_read(&bitmap->pending_writes)==0);
0b79ccf0 838 else
a9701a30 839 md_super_wait(bitmap->mddev);
32a7627c 840 }
d785a06a
N
841 if (bitmap->flags & BITMAP_WRITE_ERROR)
842 bitmap_file_kick(bitmap);
32a7627c
N
843}
844
6a07997f 845static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
846/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
847 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
848 * memory mapping of the bitmap file
849 * Special cases:
850 * if there's no bitmap file, or if the bitmap file had been
851 * previously kicked from the array, we mark all the bits as
852 * 1's in order to cause a full resync.
6a07997f
N
853 *
854 * We ignore all bits for sectors that end earlier than 'start'.
855 * This is used when reading an out-of-date bitmap...
32a7627c 856 */
6a07997f 857static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
858{
859 unsigned long i, chunks, index, oldindex, bit;
860 struct page *page = NULL, *oldpage = NULL;
861 unsigned long num_pages, bit_cnt = 0;
862 struct file *file;
d785a06a 863 unsigned long bytes, offset;
32a7627c
N
864 int outofdate;
865 int ret = -ENOSPC;
ea03aff9 866 void *paddr;
32a7627c
N
867
868 chunks = bitmap->chunks;
869 file = bitmap->file;
870
a654b9d8 871 BUG_ON(!file && !bitmap->offset);
32a7627c 872
44456d37 873#ifdef INJECT_FAULTS_3
32a7627c
N
874 outofdate = 1;
875#else
876 outofdate = bitmap->flags & BITMAP_STALE;
877#endif
878 if (outofdate)
879 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
880 "recovery\n", bmname(bitmap));
881
882 bytes = (chunks + 7) / 8;
bc7f77de 883
cdbb4cc2 884 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 885
a654b9d8 886 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
887 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
888 bmname(bitmap),
889 (unsigned long) i_size_read(file->f_mapping->host),
890 bytes + sizeof(bitmap_super_t));
4ad13663 891 goto err;
32a7627c 892 }
bc7f77de
N
893
894 ret = -ENOMEM;
895
32a7627c 896 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 897 if (!bitmap->filemap)
4ad13663 898 goto err;
32a7627c 899
e16b68b6
N
900 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
901 bitmap->filemap_attr = kzalloc(
505fa2c4 902 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
e16b68b6 903 GFP_KERNEL);
bc7f77de 904 if (!bitmap->filemap_attr)
4ad13663 905 goto err;
32a7627c 906
32a7627c
N
907 oldindex = ~0L;
908
909 for (i = 0; i < chunks; i++) {
bd926c63 910 int b;
32a7627c
N
911 index = file_page_index(i);
912 bit = file_page_offset(i);
913 if (index != oldindex) { /* this is a new page, read it in */
d785a06a 914 int count;
32a7627c 915 /* unmap the old page, we're done with it */
d785a06a 916 if (index == num_pages-1)
f49d5e62
N
917 count = bytes + sizeof(bitmap_super_t)
918 - index * PAGE_SIZE;
d785a06a
N
919 else
920 count = PAGE_SIZE;
32a7627c
N
921 if (index == 0) {
922 /*
923 * if we're here then the superblock page
924 * contains some bits (PAGE_SIZE != sizeof sb)
925 * we've already read it in, so just use it
926 */
927 page = bitmap->sb_page;
928 offset = sizeof(bitmap_super_t);
a654b9d8 929 } else if (file) {
d785a06a 930 page = read_page(file, index, bitmap, count);
a654b9d8
N
931 offset = 0;
932 } else {
933 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
934 offset = 0;
935 }
a654b9d8
N
936 if (IS_ERR(page)) { /* read error */
937 ret = PTR_ERR(page);
4ad13663 938 goto err;
a654b9d8
N
939 }
940
32a7627c
N
941 oldindex = index;
942 oldpage = page;
32a7627c
N
943
944 if (outofdate) {
945 /*
946 * if bitmap is out of date, dirty the
947 * whole page and write it out
948 */
ea03aff9
N
949 paddr = kmap_atomic(page, KM_USER0);
950 memset(paddr + offset, 0xff,
6a07997f 951 PAGE_SIZE - offset);
ea03aff9 952 kunmap_atomic(paddr, KM_USER0);
4ad13663
N
953 write_page(bitmap, page, 1);
954
955 ret = -EIO;
956 if (bitmap->flags & BITMAP_WRITE_ERROR) {
32a7627c 957 /* release, page not in filemap yet */
2d1f3b5d 958 put_page(page);
4ad13663 959 goto err;
32a7627c
N
960 }
961 }
962
963 bitmap->filemap[bitmap->file_pages++] = page;
ab6085c7 964 bitmap->last_page_size = count;
32a7627c 965 }
ea03aff9 966 paddr = kmap_atomic(page, KM_USER0);
bd926c63 967 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 968 b = test_bit(bit, paddr);
bd926c63 969 else
ea03aff9
N
970 b = ext2_test_bit(bit, paddr);
971 kunmap_atomic(paddr, KM_USER0);
bd926c63 972 if (b) {
32a7627c 973 /* if the disk bit is set, set the memory bit */
6a07997f
N
974 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
975 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
976 );
32a7627c 977 bit_cnt++;
6a07997f 978 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 979 }
32a7627c
N
980 }
981
982 /* everything went OK */
983 ret = 0;
984 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
985
32a7627c
N
986 if (bit_cnt) { /* Kick recovery if any bits were set */
987 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
988 md_wakeup_thread(bitmap->mddev->thread);
989 }
990
32a7627c 991 printk(KERN_INFO "%s: bitmap initialized from disk: "
4ad13663
N
992 "read %lu/%lu pages, set %lu bits\n",
993 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
994
995 return 0;
32a7627c 996
4ad13663
N
997 err:
998 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
999 bmname(bitmap), ret);
32a7627c
N
1000 return ret;
1001}
1002
a654b9d8
N
1003void bitmap_write_all(struct bitmap *bitmap)
1004{
1005 /* We don't actually write all bitmap blocks here,
1006 * just flag them as needing to be written
1007 */
ec7a3197 1008 int i;
a654b9d8 1009
ec7a3197
N
1010 for (i=0; i < bitmap->file_pages; i++)
1011 set_page_attr(bitmap, bitmap->filemap[i],
1012 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
1013}
1014
32a7627c
N
1015
1016static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1017{
1018 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1019 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1020 bitmap->bp[page].count += inc;
1021/*
1022 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1023 (unsigned long long)offset, inc, bitmap->bp[page].count);
1024*/
1025 bitmap_checkfree(bitmap, page);
1026}
1027static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1028 sector_t offset, int *blocks,
1029 int create);
1030
1031/*
1032 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1033 * out to disk
1034 */
1035
4ad13663 1036void bitmap_daemon_work(struct bitmap *bitmap)
32a7627c 1037{
aa3163f8 1038 unsigned long j;
32a7627c
N
1039 unsigned long flags;
1040 struct page *page = NULL, *lastpage = NULL;
32a7627c 1041 int blocks;
ea03aff9 1042 void *paddr;
32a7627c
N
1043
1044 if (bitmap == NULL)
4ad13663 1045 return;
32a7627c 1046 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
7be3dfec
N
1047 goto done;
1048
32a7627c 1049 bitmap->daemon_lastrun = jiffies;
8311c29d
N
1050 if (bitmap->allclean) {
1051 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1052 return;
1053 }
1054 bitmap->allclean = 1;
32a7627c
N
1055
1056 for (j = 0; j < bitmap->chunks; j++) {
1057 bitmap_counter_t *bmc;
1058 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1059 if (!bitmap->filemap) {
32a7627c
N
1060 /* error or shutdown */
1061 spin_unlock_irqrestore(&bitmap->lock, flags);
1062 break;
1063 }
1064
1065 page = filemap_get_page(bitmap, j);
32a7627c
N
1066
1067 if (page != lastpage) {
aa3163f8 1068 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
1069 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1070 int need_write = test_page_attr(bitmap, page,
1071 BITMAP_PAGE_NEEDWRITE);
a647e4bc 1072 if (need_write)
aa3163f8 1073 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 1074
aa3163f8 1075 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1076 if (need_write) {
4ad13663 1077 write_page(bitmap, page, 0);
8311c29d
N
1078 bitmap->allclean = 0;
1079 }
aa3163f8
N
1080 continue;
1081 }
1082
32a7627c 1083 /* grab the new page, sync and release the old */
32a7627c 1084 if (lastpage != NULL) {
ec7a3197 1085 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1086 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1087 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1088 write_page(bitmap, lastpage, 0);
32a7627c
N
1089 } else {
1090 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1091 spin_unlock_irqrestore(&bitmap->lock, flags);
1092 }
32a7627c
N
1093 } else
1094 spin_unlock_irqrestore(&bitmap->lock, flags);
1095 lastpage = page;
a0da84f3
NB
1096
1097 /* We are possibly going to clear some bits, so make
1098 * sure that events_cleared is up-to-date.
1099 */
1100 if (bitmap->need_sync) {
1101 bitmap_super_t *sb;
1102 bitmap->need_sync = 0;
1103 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1104 sb->events_cleared =
1105 cpu_to_le64(bitmap->events_cleared);
1106 kunmap_atomic(sb, KM_USER0);
1107 write_page(bitmap, bitmap->sb_page, 1);
1108 }
32a7627c
N
1109 spin_lock_irqsave(&bitmap->lock, flags);
1110 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1111 }
1112 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1113 &blocks, 0);
1114 if (bmc) {
1115/*
1116 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1117*/
8311c29d
N
1118 if (*bmc)
1119 bitmap->allclean = 0;
1120
32a7627c
N
1121 if (*bmc == 2) {
1122 *bmc=1; /* maybe clear the bit next time */
1123 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1124 } else if (*bmc == 1) {
1125 /* we can clear the bit */
1126 *bmc = 0;
1127 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1128 -1);
1129
1130 /* clear the bit */
ea03aff9 1131 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1132 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1133 clear_bit(file_page_offset(j), paddr);
bd926c63 1134 else
ea03aff9
N
1135 ext2_clear_bit(file_page_offset(j), paddr);
1136 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1137 }
1138 }
1139 spin_unlock_irqrestore(&bitmap->lock, flags);
1140 }
1141
1142 /* now sync the final page */
1143 if (lastpage != NULL) {
32a7627c 1144 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1145 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1146 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1147 spin_unlock_irqrestore(&bitmap->lock, flags);
4ad13663 1148 write_page(bitmap, lastpage, 0);
32a7627c
N
1149 } else {
1150 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1151 spin_unlock_irqrestore(&bitmap->lock, flags);
1152 }
32a7627c
N
1153 }
1154
7be3dfec 1155 done:
8311c29d
N
1156 if (bitmap->allclean == 0)
1157 bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ;
32a7627c
N
1158}
1159
32a7627c
N
1160static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1161 sector_t offset, int *blocks,
1162 int create)
1163{
1164 /* If 'create', we might release the lock and reclaim it.
1165 * The lock must have been taken with interrupts enabled.
1166 * If !create, we don't release the lock.
1167 */
1168 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1169 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1170 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1171 sector_t csize;
1172
1173 if (bitmap_checkpage(bitmap, page, create) < 0) {
1174 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1175 *blocks = csize - (offset & (csize- 1));
1176 return NULL;
1177 }
1178 /* now locked ... */
1179
1180 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1181 /* should we use the first or second counter field
1182 * of the hijacked pointer? */
1183 int hi = (pageoff > PAGE_COUNTER_MASK);
1184 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1185 PAGE_COUNTER_SHIFT - 1);
1186 *blocks = csize - (offset & (csize- 1));
1187 return &((bitmap_counter_t *)
1188 &bitmap->bp[page].map)[hi];
1189 } else { /* page is allocated */
1190 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1191 *blocks = csize - (offset & (csize- 1));
1192 return (bitmap_counter_t *)
1193 &(bitmap->bp[page].map[pageoff]);
1194 }
1195}
1196
4b6d287f 1197int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1198{
1199 if (!bitmap) return 0;
4b6d287f
N
1200
1201 if (behind) {
1202 atomic_inc(&bitmap->behind_writes);
1203 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1204 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1205 }
1206
32a7627c
N
1207 while (sectors) {
1208 int blocks;
1209 bitmap_counter_t *bmc;
1210
1211 spin_lock_irq(&bitmap->lock);
1212 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1213 if (!bmc) {
1214 spin_unlock_irq(&bitmap->lock);
1215 return 0;
1216 }
1217
da6e1a32
NB
1218 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1219 DEFINE_WAIT(__wait);
1220 /* note that it is safe to do the prepare_to_wait
1221 * after the test as long as we do it before dropping
1222 * the spinlock.
1223 */
1224 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1225 TASK_UNINTERRUPTIBLE);
1226 spin_unlock_irq(&bitmap->lock);
2ad8b1ef 1227 blk_unplug(bitmap->mddev->queue);
da6e1a32
NB
1228 schedule();
1229 finish_wait(&bitmap->overflow_wait, &__wait);
1230 continue;
1231 }
1232
32a7627c
N
1233 switch(*bmc) {
1234 case 0:
1235 bitmap_file_set_bit(bitmap, offset);
1236 bitmap_count_page(bitmap,offset, 1);
93769f58 1237 blk_plug_device_unlocked(bitmap->mddev->queue);
32a7627c
N
1238 /* fall through */
1239 case 1:
1240 *bmc = 2;
1241 }
da6e1a32 1242
32a7627c
N
1243 (*bmc)++;
1244
1245 spin_unlock_irq(&bitmap->lock);
1246
1247 offset += blocks;
1248 if (sectors > blocks)
1249 sectors -= blocks;
1250 else sectors = 0;
1251 }
8311c29d 1252 bitmap->allclean = 0;
32a7627c
N
1253 return 0;
1254}
1255
1256void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1257 int success, int behind)
32a7627c
N
1258{
1259 if (!bitmap) return;
4b6d287f
N
1260 if (behind) {
1261 atomic_dec(&bitmap->behind_writes);
1262 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1263 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1264 }
1265
32a7627c
N
1266 while (sectors) {
1267 int blocks;
1268 unsigned long flags;
1269 bitmap_counter_t *bmc;
1270
1271 spin_lock_irqsave(&bitmap->lock, flags);
1272 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1273 if (!bmc) {
1274 spin_unlock_irqrestore(&bitmap->lock, flags);
1275 return;
1276 }
1277
a0da84f3
NB
1278 if (success &&
1279 bitmap->events_cleared < bitmap->mddev->events) {
1280 bitmap->events_cleared = bitmap->mddev->events;
1281 bitmap->need_sync = 1;
1282 }
1283
32a7627c
N
1284 if (!success && ! (*bmc & NEEDED_MASK))
1285 *bmc |= NEEDED_MASK;
1286
da6e1a32
NB
1287 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1288 wake_up(&bitmap->overflow_wait);
1289
32a7627c
N
1290 (*bmc)--;
1291 if (*bmc <= 2) {
1292 set_page_attr(bitmap,
1293 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1294 BITMAP_PAGE_CLEAN);
1295 }
1296 spin_unlock_irqrestore(&bitmap->lock, flags);
1297 offset += blocks;
1298 if (sectors > blocks)
1299 sectors -= blocks;
1300 else sectors = 0;
1301 }
1302}
1303
6a806c51
N
1304int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1305 int degraded)
32a7627c
N
1306{
1307 bitmap_counter_t *bmc;
1308 int rv;
1309 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1310 *blocks = 1024;
1311 return 1; /* always resync if no bitmap */
1312 }
1313 spin_lock_irq(&bitmap->lock);
1314 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1315 rv = 0;
1316 if (bmc) {
1317 /* locked */
1318 if (RESYNC(*bmc))
1319 rv = 1;
1320 else if (NEEDED(*bmc)) {
1321 rv = 1;
6a806c51
N
1322 if (!degraded) { /* don't set/clear bits if degraded */
1323 *bmc |= RESYNC_MASK;
1324 *bmc &= ~NEEDED_MASK;
1325 }
32a7627c
N
1326 }
1327 }
1328 spin_unlock_irq(&bitmap->lock);
8311c29d 1329 bitmap->allclean = 0;
32a7627c
N
1330 return rv;
1331}
1332
1333void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1334{
1335 bitmap_counter_t *bmc;
1336 unsigned long flags;
1337/*
1338 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1339*/ if (bitmap == NULL) {
1340 *blocks = 1024;
1341 return;
1342 }
1343 spin_lock_irqsave(&bitmap->lock, flags);
1344 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1345 if (bmc == NULL)
1346 goto unlock;
1347 /* locked */
1348/*
1349 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1350*/
1351 if (RESYNC(*bmc)) {
1352 *bmc &= ~RESYNC_MASK;
1353
1354 if (!NEEDED(*bmc) && aborted)
1355 *bmc |= NEEDED_MASK;
1356 else {
1357 if (*bmc <= 2) {
1358 set_page_attr(bitmap,
1359 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1360 BITMAP_PAGE_CLEAN);
1361 }
1362 }
1363 }
1364 unlock:
1365 spin_unlock_irqrestore(&bitmap->lock, flags);
8311c29d 1366 bitmap->allclean = 0;
32a7627c
N
1367}
1368
1369void bitmap_close_sync(struct bitmap *bitmap)
1370{
1371 /* Sync has finished, and any bitmap chunks that weren't synced
1372 * properly have been aborted. It remains to us to clear the
1373 * RESYNC bit wherever it is still on
1374 */
1375 sector_t sector = 0;
1376 int blocks;
b47490c9
N
1377 if (!bitmap)
1378 return;
32a7627c
N
1379 while (sector < bitmap->mddev->resync_max_sectors) {
1380 bitmap_end_sync(bitmap, sector, &blocks, 0);
b47490c9
N
1381 sector += blocks;
1382 }
1383}
1384
1385void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1386{
1387 sector_t s = 0;
1388 int blocks;
1389
1390 if (!bitmap)
1391 return;
1392 if (sector == 0) {
1393 bitmap->last_end_sync = jiffies;
1394 return;
1395 }
1396 if (time_before(jiffies, (bitmap->last_end_sync
1397 + bitmap->daemon_sleep * HZ)))
1398 return;
1399 wait_event(bitmap->mddev->recovery_wait,
1400 atomic_read(&bitmap->mddev->recovery_active) == 0);
1401
1402 sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1403 s = 0;
1404 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1405 bitmap_end_sync(bitmap, s, &blocks, 0);
1406 s += blocks;
32a7627c 1407 }
b47490c9 1408 bitmap->last_end_sync = jiffies;
32a7627c
N
1409}
1410
6a07997f 1411static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1412{
1413 /* For each chunk covered by any of these sectors, set the
193f1c93 1414 * counter to 1 and set resync_needed. They should all
32a7627c
N
1415 * be 0 at this point
1416 */
193f1c93
N
1417
1418 int secs;
1419 bitmap_counter_t *bmc;
1420 spin_lock_irq(&bitmap->lock);
1421 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1422 if (!bmc) {
32a7627c 1423 spin_unlock_irq(&bitmap->lock);
193f1c93 1424 return;
32a7627c 1425 }
193f1c93
N
1426 if (! *bmc) {
1427 struct page *page;
6a07997f 1428 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1429 bitmap_count_page(bitmap, offset, 1);
1430 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1431 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1432 }
1433 spin_unlock_irq(&bitmap->lock);
8311c29d 1434 bitmap->allclean = 0;
32a7627c
N
1435}
1436
9b1d1dac
PC
1437/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1438void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1439{
1440 unsigned long chunk;
1441
1442 for (chunk = s; chunk <= e; chunk++) {
1443 sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap);
1444 bitmap_set_memory_bits(bitmap, sec, 1);
1445 bitmap_file_set_bit(bitmap, sec);
1446 }
1447}
1448
6b8b3e8a
N
1449/*
1450 * flush out any pending updates
1451 */
1452void bitmap_flush(mddev_t *mddev)
1453{
1454 struct bitmap *bitmap = mddev->bitmap;
1455 int sleep;
1456
1457 if (!bitmap) /* there was no bitmap */
1458 return;
1459
1460 /* run the daemon_work three time to ensure everything is flushed
1461 * that can be
1462 */
1463 sleep = bitmap->daemon_sleep;
1464 bitmap->daemon_sleep = 0;
1465 bitmap_daemon_work(bitmap);
1466 bitmap_daemon_work(bitmap);
1467 bitmap_daemon_work(bitmap);
1468 bitmap->daemon_sleep = sleep;
1469 bitmap_update_sb(bitmap);
1470}
1471
32a7627c
N
1472/*
1473 * free memory that was allocated
1474 */
3178b0db 1475static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1476{
1477 unsigned long k, pages;
1478 struct bitmap_page *bp;
32a7627c
N
1479
1480 if (!bitmap) /* there was no bitmap */
1481 return;
1482
32a7627c
N
1483 /* release the bitmap file and kill the daemon */
1484 bitmap_file_put(bitmap);
1485
1486 bp = bitmap->bp;
1487 pages = bitmap->pages;
1488
1489 /* free all allocated memory */
1490
32a7627c
N
1491 if (bp) /* deallocate the page memory */
1492 for (k = 0; k < pages; k++)
1493 if (bp[k].map && !bp[k].hijacked)
1494 kfree(bp[k].map);
1495 kfree(bp);
1496 kfree(bitmap);
1497}
3178b0db
N
1498void bitmap_destroy(mddev_t *mddev)
1499{
1500 struct bitmap *bitmap = mddev->bitmap;
1501
1502 if (!bitmap) /* there was no bitmap */
1503 return;
1504
1505 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1506 if (mddev->thread)
1507 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1508
1509 bitmap_free(bitmap);
1510}
32a7627c
N
1511
1512/*
1513 * initialize the bitmap structure
1514 * if this returns an error, bitmap_destroy must be called to do clean up
1515 */
1516int bitmap_create(mddev_t *mddev)
1517{
1518 struct bitmap *bitmap;
1519 unsigned long blocks = mddev->resync_max_sectors;
1520 unsigned long chunks;
1521 unsigned long pages;
1522 struct file *file = mddev->bitmap_file;
1523 int err;
6a07997f 1524 sector_t start;
32a7627c 1525
5f6e3c83 1526 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
32a7627c 1527
a654b9d8 1528 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1529 return 0;
1530
a654b9d8
N
1531 BUG_ON(file && mddev->bitmap_offset);
1532
9ffae0cf 1533 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1534 if (!bitmap)
1535 return -ENOMEM;
1536
32a7627c 1537 spin_lock_init(&bitmap->lock);
ce25c31b
N
1538 atomic_set(&bitmap->pending_writes, 0);
1539 init_waitqueue_head(&bitmap->write_wait);
da6e1a32 1540 init_waitqueue_head(&bitmap->overflow_wait);
ce25c31b 1541
32a7627c 1542 bitmap->mddev = mddev;
32a7627c 1543
32a7627c 1544 bitmap->file = file;
a654b9d8 1545 bitmap->offset = mddev->bitmap_offset;
ce25c31b
N
1546 if (file) {
1547 get_file(file);
ef51c976
MF
1548 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1549 SYNC_FILE_RANGE_WAIT_BEFORE |
1550 SYNC_FILE_RANGE_WRITE |
1551 SYNC_FILE_RANGE_WAIT_AFTER);
ce25c31b 1552 }
32a7627c
N
1553 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1554 err = bitmap_read_sb(bitmap);
1555 if (err)
3178b0db 1556 goto error;
32a7627c 1557
a638b2dc 1558 bitmap->chunkshift = ffz(~bitmap->chunksize);
32a7627c
N
1559
1560 /* now that chunksize and chunkshift are set, we can use these macros */
1561 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1562 CHUNK_BLOCK_RATIO(bitmap);
1563 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1564
1565 BUG_ON(!pages);
1566
1567 bitmap->chunks = chunks;
1568 bitmap->pages = pages;
1569 bitmap->missing_pages = pages;
1570 bitmap->counter_bits = COUNTER_BITS;
1571
1572 bitmap->syncchunk = ~0UL;
1573
44456d37 1574#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1575 bitmap->bp = NULL;
1576#else
9ffae0cf 1577 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1578#endif
3178b0db 1579 err = -ENOMEM;
32a7627c 1580 if (!bitmap->bp)
3178b0db 1581 goto error;
32a7627c 1582
32a7627c
N
1583 /* now that we have some pages available, initialize the in-memory
1584 * bitmap from the on-disk bitmap */
6a07997f
N
1585 start = 0;
1586 if (mddev->degraded == 0
1587 || bitmap->events_cleared == mddev->events)
1588 /* no need to keep dirty bits to optimise a re-add of a missing device */
1589 start = mddev->recovery_cp;
1590 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1591
32a7627c 1592 if (err)
3178b0db 1593 goto error;
32a7627c
N
1594
1595 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1596 pages, bmname(bitmap));
1597
3178b0db
N
1598 mddev->bitmap = bitmap;
1599
b15c2e57
N
1600 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1601
4ad13663
N
1602 bitmap_update_sb(bitmap);
1603
1604 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
3178b0db
N
1605
1606 error:
1607 bitmap_free(bitmap);
1608 return err;
32a7627c
N
1609}
1610
1611/* the bitmap API -- for raid personalities */
1612EXPORT_SYMBOL(bitmap_startwrite);
1613EXPORT_SYMBOL(bitmap_endwrite);
1614EXPORT_SYMBOL(bitmap_start_sync);
1615EXPORT_SYMBOL(bitmap_end_sync);
1616EXPORT_SYMBOL(bitmap_unplug);
1617EXPORT_SYMBOL(bitmap_close_sync);
b47490c9 1618EXPORT_SYMBOL(bitmap_cond_end_sync);