[PATCH] md/bitmap: tidy up i_writecount handling in md/bitmap
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
23#include <linux/config.h>
24#include <linux/timer.h>
25#include <linux/sched.h>
26#include <linux/list.h>
27#include <linux/file.h>
28#include <linux/mount.h>
29#include <linux/buffer_head.h>
30#include <linux/raid/md.h>
31#include <linux/raid/bitmap.h>
32
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
54//#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
55#define DPRINTK(x...) do { } while(0)
56
57#ifndef PRINTK
58# if DEBUG > 0
59# define PRINTK(x...) printk(KERN_DEBUG x)
60# else
61# define PRINTK(x...)
62# endif
63#endif
64
65static inline char * bmname(struct bitmap *bitmap)
66{
67 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
68}
69
32a7627c 70#define WRITE_POOL_SIZE 256
32a7627c
N
71
72/*
73 * just a placeholder - calls kmalloc for bitmap pages
74 */
75static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
76{
77 unsigned char *page;
78
44456d37 79#ifdef INJECT_FAULTS_1
32a7627c
N
80 page = NULL;
81#else
82 page = kmalloc(PAGE_SIZE, GFP_NOIO);
83#endif
84 if (!page)
85 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
86 else
a654b9d8 87 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
88 bmname(bitmap), page);
89 return page;
90}
91
92/*
93 * for now just a placeholder -- just calls kfree for bitmap pages
94 */
95static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
96{
97 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
98 kfree(page);
99}
100
101/*
102 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
103 *
104 * 1) check to see if this page is allocated, if it's not then try to alloc
105 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
106 * page pointer directly as a counter
107 *
108 * if we find our page, we increment the page's refcount so that it stays
109 * allocated while we're using it
110 */
111static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
112{
113 unsigned char *mappage;
114
115 if (page >= bitmap->pages) {
116 printk(KERN_ALERT
117 "%s: invalid bitmap page request: %lu (> %lu)\n",
118 bmname(bitmap), page, bitmap->pages-1);
119 return -EINVAL;
120 }
121
122
123 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
124 return 0;
125
126 if (bitmap->bp[page].map) /* page is already allocated, just return */
127 return 0;
128
129 if (!create)
130 return -ENOENT;
131
132 spin_unlock_irq(&bitmap->lock);
133
134 /* this page has not been allocated yet */
135
136 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
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 */
141 spin_lock_irq(&bitmap->lock);
142 if (!bitmap->bp[page].map)
143 bitmap->bp[page].hijacked = 1;
144 goto out;
145 }
146
147 /* got a page */
148
149 spin_lock_irq(&bitmap->lock);
150
151 /* recheck the page */
152
153 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
154 /* somebody beat us to getting the page */
155 bitmap_free_page(bitmap, mappage);
156 return 0;
157 }
158
159 /* no page was in place and we have one, so install it */
160
161 memset(mappage, 0, PAGE_SIZE);
162 bitmap->bp[page].map = mappage;
163 bitmap->missing_pages--;
164out:
165 return 0;
166}
167
168
169/* if page is completely empty, put it back on the free list, or dealloc it */
170/* if page was hijacked, unmark the flag so it might get alloced next time */
171/* Note: lock should be held when calling this */
858119e1 172static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
32a7627c
N
173{
174 char *ptr;
175
176 if (bitmap->bp[page].count) /* page is still busy */
177 return;
178
179 /* page is no longer in use, it can be released */
180
181 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
182 bitmap->bp[page].hijacked = 0;
183 bitmap->bp[page].map = NULL;
184 return;
185 }
186
187 /* normal case, free the page */
188
189#if 0
190/* actually ... let's not. We will probably need the page again exactly when
191 * memory is tight and we are flusing to disk
192 */
193 return;
194#else
195 ptr = bitmap->bp[page].map;
196 bitmap->bp[page].map = NULL;
197 bitmap->missing_pages++;
198 bitmap_free_page(bitmap, ptr);
199 return;
200#endif
201}
202
203
204/*
205 * bitmap file handling - read and write the bitmap file and its superblock
206 */
207
208/* copy the pathname of a file to a buffer */
209char *file_path(struct file *file, char *buf, int count)
210{
211 struct dentry *d;
212 struct vfsmount *v;
213
214 if (!buf)
215 return NULL;
216
217 d = file->f_dentry;
218 v = file->f_vfsmnt;
219
220 buf = d_path(d, v, buf, count);
221
222 return IS_ERR(buf) ? NULL : buf;
223}
224
225/*
226 * basic page I/O operations
227 */
228
a654b9d8
N
229/* IO operations when bitmap is stored near all superblocks */
230static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
231{
232 /* choose a good rdev and read the page from there */
233
234 mdk_rdev_t *rdev;
235 struct list_head *tmp;
236 struct page *page = alloc_page(GFP_KERNEL);
237 sector_t target;
238
239 if (!page)
240 return ERR_PTR(-ENOMEM);
a654b9d8 241
ab904d63 242 ITERATE_RDEV(mddev, rdev, tmp) {
b2d444d7
N
243 if (! test_bit(In_sync, &rdev->flags)
244 || test_bit(Faulty, &rdev->flags))
ab904d63
N
245 continue;
246
a654b9d8
N
247 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
248
ab904d63
N
249 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
250 page->index = index;
251 return page;
252 }
253 }
254 return ERR_PTR(-EIO);
a654b9d8 255
a654b9d8
N
256}
257
258static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
259{
260 mdk_rdev_t *rdev;
261 struct list_head *tmp;
262
263 ITERATE_RDEV(mddev, rdev, tmp)
b2d444d7
N
264 if (test_bit(In_sync, &rdev->flags)
265 && !test_bit(Faulty, &rdev->flags))
a654b9d8
N
266 md_super_write(mddev, rdev,
267 (rdev->sb_offset<<1) + offset
268 + page->index * (PAGE_SIZE/512),
269 PAGE_SIZE,
270 page);
271
272 if (wait)
a9701a30 273 md_super_wait(mddev);
a654b9d8
N
274 return 0;
275}
276
32a7627c 277/*
a654b9d8 278 * write out a page to a file
32a7627c 279 */
77ad4bc7 280static int write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c
N
281{
282 int ret = -ENOMEM;
283
a654b9d8
N
284 if (bitmap->file == NULL)
285 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
286
c708443c
N
287 flush_dcache_page(page); /* make sure visible to anyone reading the file */
288
8a5e9cf1
N
289 if (wait)
290 lock_page(page);
291 else {
292 if (TestSetPageLocked(page))
293 return -EAGAIN; /* already locked */
294 if (PageWriteback(page)) {
295 unlock_page(page);
296 return -EAGAIN;
297 }
298 }
32a7627c 299
34ef75f0 300 ret = page->mapping->a_ops->prepare_write(bitmap->file, page, 0, PAGE_SIZE);
32a7627c 301 if (!ret)
34ef75f0 302 ret = page->mapping->a_ops->commit_write(bitmap->file, page, 0,
32a7627c
N
303 PAGE_SIZE);
304 if (ret) {
32a7627c
N
305 unlock_page(page);
306 return ret;
307 }
308
309 set_page_dirty(page); /* force it to be written out */
77ad4bc7
N
310
311 if (!wait) {
0b79ccf0 312 /* add to list to be waited for */
77ad4bc7
N
313 struct page_list *item = mempool_alloc(bitmap->write_pool, GFP_NOIO);
314 item->page = page;
77ad4bc7
N
315 spin_lock(&bitmap->write_lock);
316 list_add(&item->list, &bitmap->complete_pages);
317 spin_unlock(&bitmap->write_lock);
77ad4bc7 318 }
32a7627c
N
319 return write_one_page(page, wait);
320}
321
322/* read a page from a file, pinning it into cache, and return bytes_read */
323static struct page *read_page(struct file *file, unsigned long index,
324 unsigned long *bytes_read)
325{
326 struct inode *inode = file->f_mapping->host;
327 struct page *page = NULL;
328 loff_t isize = i_size_read(inode);
2d1f3b5d 329 unsigned long end_index = isize >> PAGE_SHIFT;
32a7627c 330
2d1f3b5d
N
331 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
332 (unsigned long long)index << PAGE_SHIFT);
32a7627c
N
333
334 page = read_cache_page(inode->i_mapping, index,
335 (filler_t *)inode->i_mapping->a_ops->readpage, file);
336 if (IS_ERR(page))
337 goto out;
338 wait_on_page_locked(page);
339 if (!PageUptodate(page) || PageError(page)) {
2d1f3b5d 340 put_page(page);
32a7627c
N
341 page = ERR_PTR(-EIO);
342 goto out;
343 }
344
345 if (index > end_index) /* we have read beyond EOF */
346 *bytes_read = 0;
347 else if (index == end_index) /* possible short read */
2d1f3b5d 348 *bytes_read = isize & ~PAGE_MASK;
32a7627c 349 else
2d1f3b5d 350 *bytes_read = PAGE_SIZE; /* got a full page */
32a7627c
N
351out:
352 if (IS_ERR(page))
353 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
2d1f3b5d
N
354 (int)PAGE_SIZE,
355 (unsigned long long)index << PAGE_SHIFT,
32a7627c
N
356 PTR_ERR(page));
357 return page;
358}
359
360/*
361 * bitmap file superblock operations
362 */
363
364/* update the event counter and sync the superblock to disk */
365int bitmap_update_sb(struct bitmap *bitmap)
366{
367 bitmap_super_t *sb;
368 unsigned long flags;
369
370 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
371 return 0;
372 spin_lock_irqsave(&bitmap->lock, flags);
373 if (!bitmap->sb_page) { /* no superblock */
374 spin_unlock_irqrestore(&bitmap->lock, flags);
375 return 0;
376 }
32a7627c 377 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 378 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
379 sb->events = cpu_to_le64(bitmap->mddev->events);
380 if (!bitmap->mddev->degraded)
381 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
ea03aff9 382 kunmap_atomic(sb, KM_USER0);
8a5e9cf1 383 return write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
384}
385
386/* print out the bitmap file superblock */
387void bitmap_print_sb(struct bitmap *bitmap)
388{
389 bitmap_super_t *sb;
390
391 if (!bitmap || !bitmap->sb_page)
392 return;
ea03aff9 393 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c 394 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
395 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
396 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
397 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
398 *(__u32 *)(sb->uuid+0),
399 *(__u32 *)(sb->uuid+4),
400 *(__u32 *)(sb->uuid+8),
401 *(__u32 *)(sb->uuid+12));
a2cff26a 402 printk(KERN_DEBUG " events: %llu\n",
32a7627c 403 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 404 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 405 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
406 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
407 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
408 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
409 printk(KERN_DEBUG " sync size: %llu KB\n",
410 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 411 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
ea03aff9 412 kunmap_atomic(sb, KM_USER0);
32a7627c
N
413}
414
415/* read the superblock from the bitmap file and initialize some bitmap fields */
416static int bitmap_read_sb(struct bitmap *bitmap)
417{
418 char *reason = NULL;
419 bitmap_super_t *sb;
4b6d287f 420 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
421 unsigned long bytes_read;
422 unsigned long long events;
423 int err = -EINVAL;
424
425 /* page 0 is the superblock, read it... */
a654b9d8
N
426 if (bitmap->file)
427 bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read);
428 else {
429 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
430 bytes_read = PAGE_SIZE;
431 }
32a7627c
N
432 if (IS_ERR(bitmap->sb_page)) {
433 err = PTR_ERR(bitmap->sb_page);
434 bitmap->sb_page = NULL;
435 return err;
436 }
437
ea03aff9 438 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
439
440 if (bytes_read < sizeof(*sb)) { /* short read */
441 printk(KERN_INFO "%s: bitmap file superblock truncated\n",
442 bmname(bitmap));
443 err = -ENOSPC;
444 goto out;
445 }
446
447 chunksize = le32_to_cpu(sb->chunksize);
448 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 449 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
450
451 /* verify that the bitmap-specific fields are valid */
452 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
453 reason = "bad magic";
bd926c63
N
454 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
455 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c 456 reason = "unrecognized superblock version";
7dd5d34c
N
457 else if (chunksize < PAGE_SIZE)
458 reason = "bitmap chunksize too small";
32a7627c
N
459 else if ((1 << ffz(~chunksize)) != chunksize)
460 reason = "bitmap chunksize not a power of 2";
7dd5d34c
N
461 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
462 reason = "daemon sleep period out of range";
4b6d287f
N
463 else if (write_behind > COUNTER_MAX)
464 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
465 if (reason) {
466 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
467 bmname(bitmap), reason);
468 goto out;
469 }
470
471 /* keep the array size field of the bitmap superblock up to date */
472 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
473
474 if (!bitmap->mddev->persistent)
475 goto success;
476
477 /*
478 * if we have a persistent array superblock, compare the
479 * bitmap's UUID and event counter to the mddev's
480 */
481 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
482 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
483 bmname(bitmap));
484 goto out;
485 }
486 events = le64_to_cpu(sb->events);
487 if (events < bitmap->mddev->events) {
488 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
489 "-- forcing full recovery\n", bmname(bitmap), events,
490 (unsigned long long) bitmap->mddev->events);
491 sb->state |= BITMAP_STALE;
492 }
493success:
494 /* assign fields using values from superblock */
495 bitmap->chunksize = chunksize;
496 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 497 bitmap->daemon_lastrun = jiffies;
4b6d287f 498 bitmap->max_write_behind = write_behind;
32a7627c 499 bitmap->flags |= sb->state;
bd926c63
N
500 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
501 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 502 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
6a07997f
N
503 if (sb->state & BITMAP_STALE)
504 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
505 err = 0;
506out:
ea03aff9 507 kunmap_atomic(sb, KM_USER0);
32a7627c
N
508 if (err)
509 bitmap_print_sb(bitmap);
510 return err;
511}
512
513enum bitmap_mask_op {
514 MASK_SET,
515 MASK_UNSET
516};
517
518/* record the state of the bitmap in the superblock */
519static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
520 enum bitmap_mask_op op)
521{
522 bitmap_super_t *sb;
523 unsigned long flags;
524
525 spin_lock_irqsave(&bitmap->lock, flags);
7e317655 526 if (!bitmap->sb_page) { /* can't set the state */
32a7627c
N
527 spin_unlock_irqrestore(&bitmap->lock, flags);
528 return;
529 }
32a7627c 530 spin_unlock_irqrestore(&bitmap->lock, flags);
ea03aff9 531 sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
32a7627c
N
532 switch (op) {
533 case MASK_SET: sb->state |= bits;
534 break;
535 case MASK_UNSET: sb->state &= ~bits;
536 break;
537 default: BUG();
538 }
ea03aff9 539 kunmap_atomic(sb, KM_USER0);
32a7627c
N
540}
541
542/*
543 * general bitmap file operations
544 */
545
546/* calculate the index of the page that contains this bit */
547static inline unsigned long file_page_index(unsigned long chunk)
548{
549 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
550}
551
552/* calculate the (bit) offset of this bit within a page */
553static inline unsigned long file_page_offset(unsigned long chunk)
554{
555 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
556}
557
558/*
559 * return a pointer to the page in the filemap that contains the given bit
560 *
561 * this lookup is complicated by the fact that the bitmap sb might be exactly
562 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
563 * 0 or page 1
564 */
565static inline struct page *filemap_get_page(struct bitmap *bitmap,
566 unsigned long chunk)
567{
568 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
569}
570
571
572static void bitmap_file_unmap(struct bitmap *bitmap)
573{
574 struct page **map, *sb_page;
575 unsigned long *attr;
576 int pages;
577 unsigned long flags;
578
579 spin_lock_irqsave(&bitmap->lock, flags);
580 map = bitmap->filemap;
581 bitmap->filemap = NULL;
582 attr = bitmap->filemap_attr;
583 bitmap->filemap_attr = NULL;
584 pages = bitmap->file_pages;
585 bitmap->file_pages = 0;
586 sb_page = bitmap->sb_page;
587 bitmap->sb_page = NULL;
588 spin_unlock_irqrestore(&bitmap->lock, flags);
589
590 while (pages--)
591 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
2d1f3b5d 592 put_page(map[pages]);
32a7627c
N
593 kfree(map);
594 kfree(attr);
595
1345b1d8 596 safe_put_page(sb_page);
32a7627c
N
597}
598
32a7627c 599/* dequeue the next item in a page list -- don't call from irq context */
77ad4bc7 600static struct page_list *dequeue_page(struct bitmap *bitmap)
32a7627c
N
601{
602 struct page_list *item = NULL;
77ad4bc7 603 struct list_head *head = &bitmap->complete_pages;
32a7627c
N
604
605 spin_lock(&bitmap->write_lock);
606 if (list_empty(head))
607 goto out;
608 item = list_entry(head->prev, struct page_list, list);
609 list_del(head->prev);
610out:
611 spin_unlock(&bitmap->write_lock);
612 return item;
613}
614
615static void drain_write_queues(struct bitmap *bitmap)
616{
32a7627c 617 struct page_list *item;
32a7627c 618
77ad4bc7
N
619 while ((item = dequeue_page(bitmap))) {
620 /* don't bother to wait */
77ad4bc7 621 mempool_free(item, bitmap->write_pool);
32a7627c 622 }
32a7627c
N
623}
624
625static void bitmap_file_put(struct bitmap *bitmap)
626{
627 struct file *file;
32a7627c
N
628 unsigned long flags;
629
630 spin_lock_irqsave(&bitmap->lock, flags);
631 file = bitmap->file;
632 bitmap->file = NULL;
633 spin_unlock_irqrestore(&bitmap->lock, flags);
634
32a7627c
N
635 drain_write_queues(bitmap);
636
637 bitmap_file_unmap(bitmap);
638
acc55e22 639 if (file)
32a7627c 640 fput(file);
32a7627c
N
641}
642
643
644/*
645 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
646 * then it is no longer reliable, so we stop using it and we mark the file
647 * as failed in the superblock
648 */
649static void bitmap_file_kick(struct bitmap *bitmap)
650{
651 char *path, *ptr = NULL;
652
653 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
654 bitmap_update_sb(bitmap);
655
a654b9d8
N
656 if (bitmap->file) {
657 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
658 if (path)
659 ptr = file_path(bitmap->file, path, PAGE_SIZE);
32a7627c 660
a654b9d8
N
661 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
662 bmname(bitmap), ptr ? ptr : "");
32a7627c 663
a654b9d8
N
664 kfree(path);
665 }
32a7627c
N
666
667 bitmap_file_put(bitmap);
668
669 return;
670}
671
672enum bitmap_page_attr {
e16b68b6
N
673 BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
674 BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
675 BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
32a7627c
N
676};
677
678static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
679 enum bitmap_page_attr attr)
680{
e16b68b6 681 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
682}
683
684static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
685 enum bitmap_page_attr attr)
686{
e16b68b6 687 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
688}
689
ec7a3197
N
690static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
691 enum bitmap_page_attr attr)
32a7627c 692{
e16b68b6 693 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
32a7627c
N
694}
695
696/*
697 * bitmap_file_set_bit -- called before performing a write to the md device
698 * to set (and eventually sync) a particular bit in the bitmap file
699 *
700 * we set the bit immediately, then we record the page number so that
701 * when an unplug occurs, we can flush the dirty pages out to disk
702 */
703static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
704{
705 unsigned long bit;
706 struct page *page;
707 void *kaddr;
708 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
709
a654b9d8 710 if (!bitmap->filemap) {
32a7627c
N
711 return;
712 }
713
714 page = filemap_get_page(bitmap, chunk);
715 bit = file_page_offset(chunk);
716
32a7627c
N
717 /* set the bit */
718 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
719 if (bitmap->flags & BITMAP_HOSTENDIAN)
720 set_bit(bit, kaddr);
721 else
722 ext2_set_bit(bit, kaddr);
32a7627c
N
723 kunmap_atomic(kaddr, KM_USER0);
724 PRINTK("set file bit %lu page %lu\n", bit, page->index);
725
726 /* record page number so it gets flushed to disk when unplug occurs */
727 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
728
729}
730
0b79ccf0
N
731static void bitmap_writeback(struct bitmap *bitmap);
732
32a7627c
N
733/* this gets called when the md device is ready to unplug its underlying
734 * (slave) device queues -- before we let any writes go down, we need to
735 * sync the dirty pages of the bitmap file to disk */
736int bitmap_unplug(struct bitmap *bitmap)
737{
ec7a3197
N
738 unsigned long i, flags;
739 int dirty, need_write;
32a7627c
N
740 struct page *page;
741 int wait = 0;
8a5e9cf1 742 int err;
32a7627c
N
743
744 if (!bitmap)
745 return 0;
746
747 /* look at each page to see if there are any set bits that need to be
748 * flushed out to disk */
749 for (i = 0; i < bitmap->file_pages; i++) {
750 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 751 if (!bitmap->filemap) {
32a7627c
N
752 spin_unlock_irqrestore(&bitmap->lock, flags);
753 return 0;
754 }
755 page = bitmap->filemap[i];
ec7a3197
N
756 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
757 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
32a7627c
N
758 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
759 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
ec7a3197 760 if (dirty)
32a7627c
N
761 wait = 1;
762 spin_unlock_irqrestore(&bitmap->lock, flags);
763
ec7a3197 764 if (dirty | need_write) {
8a5e9cf1
N
765 err = write_page(bitmap, page, 0);
766 if (err == -EAGAIN) {
ec7a3197 767 if (dirty)
8a5e9cf1
N
768 err = write_page(bitmap, page, 1);
769 else
770 err = 0;
771 }
772 if (err)
bfb39fba 773 return 1;
8a5e9cf1 774 }
32a7627c
N
775 }
776 if (wait) { /* if any writes were performed, we need to wait on them */
0b79ccf0
N
777 if (bitmap->file)
778 bitmap_writeback(bitmap);
779 else
a9701a30 780 md_super_wait(bitmap->mddev);
32a7627c
N
781 }
782 return 0;
783}
784
6a07997f 785static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
786/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
787 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
788 * memory mapping of the bitmap file
789 * Special cases:
790 * if there's no bitmap file, or if the bitmap file had been
791 * previously kicked from the array, we mark all the bits as
792 * 1's in order to cause a full resync.
6a07997f
N
793 *
794 * We ignore all bits for sectors that end earlier than 'start'.
795 * This is used when reading an out-of-date bitmap...
32a7627c 796 */
6a07997f 797static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
798{
799 unsigned long i, chunks, index, oldindex, bit;
800 struct page *page = NULL, *oldpage = NULL;
801 unsigned long num_pages, bit_cnt = 0;
802 struct file *file;
803 unsigned long bytes, offset, dummy;
804 int outofdate;
805 int ret = -ENOSPC;
ea03aff9 806 void *paddr;
32a7627c
N
807
808 chunks = bitmap->chunks;
809 file = bitmap->file;
810
a654b9d8 811 BUG_ON(!file && !bitmap->offset);
32a7627c 812
44456d37 813#ifdef INJECT_FAULTS_3
32a7627c
N
814 outofdate = 1;
815#else
816 outofdate = bitmap->flags & BITMAP_STALE;
817#endif
818 if (outofdate)
819 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
820 "recovery\n", bmname(bitmap));
821
822 bytes = (chunks + 7) / 8;
bc7f77de 823
cdbb4cc2 824 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 825
a654b9d8 826 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
827 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
828 bmname(bitmap),
829 (unsigned long) i_size_read(file->f_mapping->host),
830 bytes + sizeof(bitmap_super_t));
831 goto out;
832 }
bc7f77de
N
833
834 ret = -ENOMEM;
835
32a7627c 836 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 837 if (!bitmap->filemap)
32a7627c 838 goto out;
32a7627c 839
e16b68b6
N
840 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
841 bitmap->filemap_attr = kzalloc(
842 (((num_pages*4/8)+sizeof(unsigned long)-1)
843 /sizeof(unsigned long))
844 *sizeof(unsigned long),
845 GFP_KERNEL);
bc7f77de 846 if (!bitmap->filemap_attr)
32a7627c 847 goto out;
32a7627c 848
32a7627c
N
849 oldindex = ~0L;
850
851 for (i = 0; i < chunks; i++) {
bd926c63 852 int b;
32a7627c
N
853 index = file_page_index(i);
854 bit = file_page_offset(i);
855 if (index != oldindex) { /* this is a new page, read it in */
856 /* unmap the old page, we're done with it */
32a7627c
N
857 if (index == 0) {
858 /*
859 * if we're here then the superblock page
860 * contains some bits (PAGE_SIZE != sizeof sb)
861 * we've already read it in, so just use it
862 */
863 page = bitmap->sb_page;
864 offset = sizeof(bitmap_super_t);
a654b9d8 865 } else if (file) {
32a7627c 866 page = read_page(file, index, &dummy);
a654b9d8
N
867 offset = 0;
868 } else {
869 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
870 offset = 0;
871 }
a654b9d8
N
872 if (IS_ERR(page)) { /* read error */
873 ret = PTR_ERR(page);
874 goto out;
875 }
876
32a7627c
N
877 oldindex = index;
878 oldpage = page;
32a7627c
N
879
880 if (outofdate) {
881 /*
882 * if bitmap is out of date, dirty the
883 * whole page and write it out
884 */
ea03aff9
N
885 paddr = kmap_atomic(page, KM_USER0);
886 memset(paddr + offset, 0xff,
6a07997f 887 PAGE_SIZE - offset);
ea03aff9 888 kunmap_atomic(paddr, KM_USER0);
77ad4bc7 889 ret = write_page(bitmap, page, 1);
32a7627c 890 if (ret) {
32a7627c 891 /* release, page not in filemap yet */
2d1f3b5d 892 put_page(page);
32a7627c
N
893 goto out;
894 }
895 }
896
897 bitmap->filemap[bitmap->file_pages++] = page;
898 }
ea03aff9 899 paddr = kmap_atomic(page, KM_USER0);
bd926c63 900 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 901 b = test_bit(bit, paddr);
bd926c63 902 else
ea03aff9
N
903 b = ext2_test_bit(bit, paddr);
904 kunmap_atomic(paddr, KM_USER0);
bd926c63 905 if (b) {
32a7627c 906 /* if the disk bit is set, set the memory bit */
6a07997f
N
907 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
908 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
909 );
32a7627c 910 bit_cnt++;
6a07997f 911 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 912 }
32a7627c
N
913 }
914
915 /* everything went OK */
916 ret = 0;
917 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
918
32a7627c
N
919 if (bit_cnt) { /* Kick recovery if any bits were set */
920 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
921 md_wakeup_thread(bitmap->mddev->thread);
922 }
923
924out:
925 printk(KERN_INFO "%s: bitmap initialized from disk: "
926 "read %lu/%lu pages, set %lu bits, status: %d\n",
927 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
928
929 return ret;
930}
931
a654b9d8
N
932void bitmap_write_all(struct bitmap *bitmap)
933{
934 /* We don't actually write all bitmap blocks here,
935 * just flag them as needing to be written
936 */
ec7a3197 937 int i;
a654b9d8 938
ec7a3197
N
939 for (i=0; i < bitmap->file_pages; i++)
940 set_page_attr(bitmap, bitmap->filemap[i],
941 BITMAP_PAGE_NEEDWRITE);
a654b9d8
N
942}
943
32a7627c
N
944
945static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
946{
947 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
948 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
949 bitmap->bp[page].count += inc;
950/*
951 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
952 (unsigned long long)offset, inc, bitmap->bp[page].count);
953*/
954 bitmap_checkfree(bitmap, page);
955}
956static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
957 sector_t offset, int *blocks,
958 int create);
959
960/*
961 * bitmap daemon -- periodically wakes up to clean bits and flush pages
962 * out to disk
963 */
964
965int bitmap_daemon_work(struct bitmap *bitmap)
966{
aa3163f8 967 unsigned long j;
32a7627c
N
968 unsigned long flags;
969 struct page *page = NULL, *lastpage = NULL;
970 int err = 0;
971 int blocks;
ea03aff9 972 void *paddr;
32a7627c
N
973
974 if (bitmap == NULL)
975 return 0;
976 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
977 return 0;
978 bitmap->daemon_lastrun = jiffies;
979
980 for (j = 0; j < bitmap->chunks; j++) {
981 bitmap_counter_t *bmc;
982 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 983 if (!bitmap->filemap) {
32a7627c
N
984 /* error or shutdown */
985 spin_unlock_irqrestore(&bitmap->lock, flags);
986 break;
987 }
988
989 page = filemap_get_page(bitmap, j);
32a7627c
N
990
991 if (page != lastpage) {
aa3163f8 992 /* skip this page unless it's marked as needing cleaning */
ec7a3197
N
993 if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
994 int need_write = test_page_attr(bitmap, page,
995 BITMAP_PAGE_NEEDWRITE);
a647e4bc 996 if (need_write)
aa3163f8 997 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
a647e4bc 998
aa3163f8 999 spin_unlock_irqrestore(&bitmap->lock, flags);
ec7a3197 1000 if (need_write) {
8a5e9cf1
N
1001 switch (write_page(bitmap, page, 0)) {
1002 case -EAGAIN:
1003 set_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1004 break;
1005 case 0:
1006 break;
1007 default:
aa3163f8 1008 bitmap_file_kick(bitmap);
8a5e9cf1 1009 }
aa3163f8
N
1010 }
1011 continue;
1012 }
1013
32a7627c 1014 /* grab the new page, sync and release the old */
32a7627c 1015 if (lastpage != NULL) {
ec7a3197 1016 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1017 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1018 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1019 err = write_page(bitmap, lastpage, 0);
8a5e9cf1
N
1020 if (err == -EAGAIN) {
1021 err = 0;
1022 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1023 }
32a7627c
N
1024 } else {
1025 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1026 spin_unlock_irqrestore(&bitmap->lock, flags);
1027 }
32a7627c
N
1028 if (err)
1029 bitmap_file_kick(bitmap);
1030 } else
1031 spin_unlock_irqrestore(&bitmap->lock, flags);
1032 lastpage = page;
32a7627c
N
1033/*
1034 printk("bitmap clean at page %lu\n", j);
1035*/
1036 spin_lock_irqsave(&bitmap->lock, flags);
1037 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1038 }
1039 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1040 &blocks, 0);
1041 if (bmc) {
1042/*
1043 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1044*/
1045 if (*bmc == 2) {
1046 *bmc=1; /* maybe clear the bit next time */
1047 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1048 } else if (*bmc == 1) {
1049 /* we can clear the bit */
1050 *bmc = 0;
1051 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1052 -1);
1053
1054 /* clear the bit */
ea03aff9 1055 paddr = kmap_atomic(page, KM_USER0);
bd926c63 1056 if (bitmap->flags & BITMAP_HOSTENDIAN)
ea03aff9 1057 clear_bit(file_page_offset(j), paddr);
bd926c63 1058 else
ea03aff9
N
1059 ext2_clear_bit(file_page_offset(j), paddr);
1060 kunmap_atomic(paddr, KM_USER0);
32a7627c
N
1061 }
1062 }
1063 spin_unlock_irqrestore(&bitmap->lock, flags);
1064 }
1065
1066 /* now sync the final page */
1067 if (lastpage != NULL) {
32a7627c 1068 spin_lock_irqsave(&bitmap->lock, flags);
ec7a3197 1069 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
32a7627c
N
1070 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1071 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1072 err = write_page(bitmap, lastpage, 0);
8a5e9cf1
N
1073 if (err == -EAGAIN) {
1074 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1075 err = 0;
1076 }
32a7627c
N
1077 } else {
1078 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1079 spin_unlock_irqrestore(&bitmap->lock, flags);
1080 }
32a7627c
N
1081 }
1082
1083 return err;
1084}
1085
0b79ccf0 1086static void bitmap_writeback(struct bitmap *bitmap)
32a7627c 1087{
32a7627c
N
1088 struct page *page;
1089 struct page_list *item;
1090 int err = 0;
1091
77ad4bc7
N
1092 PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap));
1093 /* wait on bitmap page writebacks */
1094 while ((item = dequeue_page(bitmap))) {
1095 page = item->page;
1096 mempool_free(item, bitmap->write_pool);
1097 PRINTK("wait on page writeback: %p\n", page);
1098 wait_on_page_writeback(page);
1099 PRINTK("finished page writeback: %p\n", page);
1100
1101 err = PageError(page);
77ad4bc7
N
1102 if (err) {
1103 printk(KERN_WARNING "%s: bitmap file writeback "
1104 "failed (page %lu): %d\n",
1105 bmname(bitmap), page->index, err);
1106 bitmap_file_kick(bitmap);
0b79ccf0 1107 break;
32a7627c
N
1108 }
1109 }
32a7627c
N
1110}
1111
1112static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1113 sector_t offset, int *blocks,
1114 int create)
1115{
1116 /* If 'create', we might release the lock and reclaim it.
1117 * The lock must have been taken with interrupts enabled.
1118 * If !create, we don't release the lock.
1119 */
1120 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1121 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1122 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1123 sector_t csize;
1124
1125 if (bitmap_checkpage(bitmap, page, create) < 0) {
1126 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1127 *blocks = csize - (offset & (csize- 1));
1128 return NULL;
1129 }
1130 /* now locked ... */
1131
1132 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1133 /* should we use the first or second counter field
1134 * of the hijacked pointer? */
1135 int hi = (pageoff > PAGE_COUNTER_MASK);
1136 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1137 PAGE_COUNTER_SHIFT - 1);
1138 *blocks = csize - (offset & (csize- 1));
1139 return &((bitmap_counter_t *)
1140 &bitmap->bp[page].map)[hi];
1141 } else { /* page is allocated */
1142 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1143 *blocks = csize - (offset & (csize- 1));
1144 return (bitmap_counter_t *)
1145 &(bitmap->bp[page].map[pageoff]);
1146 }
1147}
1148
4b6d287f 1149int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1150{
1151 if (!bitmap) return 0;
4b6d287f
N
1152
1153 if (behind) {
1154 atomic_inc(&bitmap->behind_writes);
1155 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1156 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1157 }
1158
32a7627c
N
1159 while (sectors) {
1160 int blocks;
1161 bitmap_counter_t *bmc;
1162
1163 spin_lock_irq(&bitmap->lock);
1164 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1165 if (!bmc) {
1166 spin_unlock_irq(&bitmap->lock);
1167 return 0;
1168 }
1169
1170 switch(*bmc) {
1171 case 0:
1172 bitmap_file_set_bit(bitmap, offset);
1173 bitmap_count_page(bitmap,offset, 1);
1174 blk_plug_device(bitmap->mddev->queue);
1175 /* fall through */
1176 case 1:
1177 *bmc = 2;
1178 }
5daf2cf1 1179 BUG_ON((*bmc & COUNTER_MAX) == COUNTER_MAX);
32a7627c
N
1180 (*bmc)++;
1181
1182 spin_unlock_irq(&bitmap->lock);
1183
1184 offset += blocks;
1185 if (sectors > blocks)
1186 sectors -= blocks;
1187 else sectors = 0;
1188 }
1189 return 0;
1190}
1191
1192void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1193 int success, int behind)
32a7627c
N
1194{
1195 if (!bitmap) return;
4b6d287f
N
1196 if (behind) {
1197 atomic_dec(&bitmap->behind_writes);
1198 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1199 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1200 }
1201
32a7627c
N
1202 while (sectors) {
1203 int blocks;
1204 unsigned long flags;
1205 bitmap_counter_t *bmc;
1206
1207 spin_lock_irqsave(&bitmap->lock, flags);
1208 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1209 if (!bmc) {
1210 spin_unlock_irqrestore(&bitmap->lock, flags);
1211 return;
1212 }
1213
1214 if (!success && ! (*bmc & NEEDED_MASK))
1215 *bmc |= NEEDED_MASK;
1216
1217 (*bmc)--;
1218 if (*bmc <= 2) {
1219 set_page_attr(bitmap,
1220 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1221 BITMAP_PAGE_CLEAN);
1222 }
1223 spin_unlock_irqrestore(&bitmap->lock, flags);
1224 offset += blocks;
1225 if (sectors > blocks)
1226 sectors -= blocks;
1227 else sectors = 0;
1228 }
1229}
1230
6a806c51
N
1231int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1232 int degraded)
32a7627c
N
1233{
1234 bitmap_counter_t *bmc;
1235 int rv;
1236 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1237 *blocks = 1024;
1238 return 1; /* always resync if no bitmap */
1239 }
1240 spin_lock_irq(&bitmap->lock);
1241 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1242 rv = 0;
1243 if (bmc) {
1244 /* locked */
1245 if (RESYNC(*bmc))
1246 rv = 1;
1247 else if (NEEDED(*bmc)) {
1248 rv = 1;
6a806c51
N
1249 if (!degraded) { /* don't set/clear bits if degraded */
1250 *bmc |= RESYNC_MASK;
1251 *bmc &= ~NEEDED_MASK;
1252 }
32a7627c
N
1253 }
1254 }
1255 spin_unlock_irq(&bitmap->lock);
1256 return rv;
1257}
1258
1259void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1260{
1261 bitmap_counter_t *bmc;
1262 unsigned long flags;
1263/*
1264 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1265*/ if (bitmap == NULL) {
1266 *blocks = 1024;
1267 return;
1268 }
1269 spin_lock_irqsave(&bitmap->lock, flags);
1270 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1271 if (bmc == NULL)
1272 goto unlock;
1273 /* locked */
1274/*
1275 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1276*/
1277 if (RESYNC(*bmc)) {
1278 *bmc &= ~RESYNC_MASK;
1279
1280 if (!NEEDED(*bmc) && aborted)
1281 *bmc |= NEEDED_MASK;
1282 else {
1283 if (*bmc <= 2) {
1284 set_page_attr(bitmap,
1285 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1286 BITMAP_PAGE_CLEAN);
1287 }
1288 }
1289 }
1290 unlock:
1291 spin_unlock_irqrestore(&bitmap->lock, flags);
1292}
1293
1294void bitmap_close_sync(struct bitmap *bitmap)
1295{
1296 /* Sync has finished, and any bitmap chunks that weren't synced
1297 * properly have been aborted. It remains to us to clear the
1298 * RESYNC bit wherever it is still on
1299 */
1300 sector_t sector = 0;
1301 int blocks;
1302 if (!bitmap) return;
1303 while (sector < bitmap->mddev->resync_max_sectors) {
1304 bitmap_end_sync(bitmap, sector, &blocks, 0);
1305/*
1306 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1307 (unsigned long long)sector, blocks);
1308*/ sector += blocks;
1309 }
1310}
1311
6a07997f 1312static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1313{
1314 /* For each chunk covered by any of these sectors, set the
193f1c93 1315 * counter to 1 and set resync_needed. They should all
32a7627c
N
1316 * be 0 at this point
1317 */
193f1c93
N
1318
1319 int secs;
1320 bitmap_counter_t *bmc;
1321 spin_lock_irq(&bitmap->lock);
1322 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1323 if (!bmc) {
32a7627c 1324 spin_unlock_irq(&bitmap->lock);
193f1c93 1325 return;
32a7627c 1326 }
193f1c93
N
1327 if (! *bmc) {
1328 struct page *page;
6a07997f 1329 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1330 bitmap_count_page(bitmap, offset, 1);
1331 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1332 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1333 }
1334 spin_unlock_irq(&bitmap->lock);
1335
32a7627c
N
1336}
1337
6b8b3e8a
N
1338/*
1339 * flush out any pending updates
1340 */
1341void bitmap_flush(mddev_t *mddev)
1342{
1343 struct bitmap *bitmap = mddev->bitmap;
1344 int sleep;
1345
1346 if (!bitmap) /* there was no bitmap */
1347 return;
1348
1349 /* run the daemon_work three time to ensure everything is flushed
1350 * that can be
1351 */
1352 sleep = bitmap->daemon_sleep;
1353 bitmap->daemon_sleep = 0;
1354 bitmap_daemon_work(bitmap);
1355 bitmap_daemon_work(bitmap);
1356 bitmap_daemon_work(bitmap);
1357 bitmap->daemon_sleep = sleep;
1358 bitmap_update_sb(bitmap);
1359}
1360
32a7627c
N
1361/*
1362 * free memory that was allocated
1363 */
3178b0db 1364static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1365{
1366 unsigned long k, pages;
1367 struct bitmap_page *bp;
32a7627c
N
1368
1369 if (!bitmap) /* there was no bitmap */
1370 return;
1371
32a7627c
N
1372 /* release the bitmap file and kill the daemon */
1373 bitmap_file_put(bitmap);
1374
1375 bp = bitmap->bp;
1376 pages = bitmap->pages;
1377
1378 /* free all allocated memory */
1379
1380 mempool_destroy(bitmap->write_pool);
1381
1382 if (bp) /* deallocate the page memory */
1383 for (k = 0; k < pages; k++)
1384 if (bp[k].map && !bp[k].hijacked)
1385 kfree(bp[k].map);
1386 kfree(bp);
1387 kfree(bitmap);
1388}
3178b0db
N
1389void bitmap_destroy(mddev_t *mddev)
1390{
1391 struct bitmap *bitmap = mddev->bitmap;
1392
1393 if (!bitmap) /* there was no bitmap */
1394 return;
1395
1396 mddev->bitmap = NULL; /* disconnect from the md device */
b15c2e57
N
1397 if (mddev->thread)
1398 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
3178b0db
N
1399
1400 bitmap_free(bitmap);
1401}
32a7627c
N
1402
1403/*
1404 * initialize the bitmap structure
1405 * if this returns an error, bitmap_destroy must be called to do clean up
1406 */
1407int bitmap_create(mddev_t *mddev)
1408{
1409 struct bitmap *bitmap;
1410 unsigned long blocks = mddev->resync_max_sectors;
1411 unsigned long chunks;
1412 unsigned long pages;
1413 struct file *file = mddev->bitmap_file;
1414 int err;
6a07997f 1415 sector_t start;
32a7627c
N
1416
1417 BUG_ON(sizeof(bitmap_super_t) != 256);
1418
a654b9d8 1419 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1420 return 0;
1421
a654b9d8
N
1422 BUG_ON(file && mddev->bitmap_offset);
1423
9ffae0cf 1424 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
32a7627c
N
1425 if (!bitmap)
1426 return -ENOMEM;
1427
32a7627c
N
1428 spin_lock_init(&bitmap->lock);
1429 bitmap->mddev = mddev;
32a7627c
N
1430
1431 spin_lock_init(&bitmap->write_lock);
32a7627c 1432 INIT_LIST_HEAD(&bitmap->complete_pages);
0eaae62a
MD
1433 bitmap->write_pool = mempool_create_kmalloc_pool(WRITE_POOL_SIZE,
1434 sizeof(struct page_list));
3178b0db 1435 err = -ENOMEM;
32a7627c 1436 if (!bitmap->write_pool)
3178b0db 1437 goto error;
32a7627c
N
1438
1439 bitmap->file = file;
a654b9d8
N
1440 bitmap->offset = mddev->bitmap_offset;
1441 if (file) get_file(file);
32a7627c
N
1442 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1443 err = bitmap_read_sb(bitmap);
1444 if (err)
3178b0db 1445 goto error;
32a7627c
N
1446
1447 bitmap->chunkshift = find_first_bit(&bitmap->chunksize,
1448 sizeof(bitmap->chunksize));
1449
1450 /* now that chunksize and chunkshift are set, we can use these macros */
1451 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1452 CHUNK_BLOCK_RATIO(bitmap);
1453 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1454
1455 BUG_ON(!pages);
1456
1457 bitmap->chunks = chunks;
1458 bitmap->pages = pages;
1459 bitmap->missing_pages = pages;
1460 bitmap->counter_bits = COUNTER_BITS;
1461
1462 bitmap->syncchunk = ~0UL;
1463
44456d37 1464#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1465 bitmap->bp = NULL;
1466#else
9ffae0cf 1467 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
32a7627c 1468#endif
3178b0db 1469 err = -ENOMEM;
32a7627c 1470 if (!bitmap->bp)
3178b0db 1471 goto error;
32a7627c 1472
32a7627c
N
1473 /* now that we have some pages available, initialize the in-memory
1474 * bitmap from the on-disk bitmap */
6a07997f
N
1475 start = 0;
1476 if (mddev->degraded == 0
1477 || bitmap->events_cleared == mddev->events)
1478 /* no need to keep dirty bits to optimise a re-add of a missing device */
1479 start = mddev->recovery_cp;
1480 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1481
32a7627c 1482 if (err)
3178b0db 1483 goto error;
32a7627c
N
1484
1485 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1486 pages, bmname(bitmap));
1487
3178b0db
N
1488 mddev->bitmap = bitmap;
1489
b15c2e57
N
1490 mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1491
32a7627c 1492 return bitmap_update_sb(bitmap);
3178b0db
N
1493
1494 error:
1495 bitmap_free(bitmap);
1496 return err;
32a7627c
N
1497}
1498
1499/* the bitmap API -- for raid personalities */
1500EXPORT_SYMBOL(bitmap_startwrite);
1501EXPORT_SYMBOL(bitmap_endwrite);
1502EXPORT_SYMBOL(bitmap_start_sync);
1503EXPORT_SYMBOL(bitmap_end_sync);
1504EXPORT_SYMBOL(bitmap_unplug);
1505EXPORT_SYMBOL(bitmap_close_sync);
1506EXPORT_SYMBOL(bitmap_daemon_work);