]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/md/bitmap.c
[PATCH] md: make md on-disk bitmaps not host-endian
[mirror_ubuntu-artful-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
10 * - added bitmap daemon (to asynchronously clear bitmap bits from disk)
11 */
12
13/*
14 * Still to do:
15 *
16 * flush after percent set rather than just time based. (maybe both).
17 * wait if count gets too high, wake when it drops to half.
18 * allow bitmap to be mirrored with superblock (before or after...)
19 * allow hot-add to re-instate a current device.
20 * allow hot-add of bitmap after quiessing device
21 */
22
23#include <linux/module.h>
32a7627c
N
24#include <linux/errno.h>
25#include <linux/slab.h>
26#include <linux/init.h>
27#include <linux/config.h>
28#include <linux/timer.h>
29#include <linux/sched.h>
30#include <linux/list.h>
31#include <linux/file.h>
32#include <linux/mount.h>
33#include <linux/buffer_head.h>
34#include <linux/raid/md.h>
35#include <linux/raid/bitmap.h>
36
37/* debug macros */
38
39#define DEBUG 0
40
41#if DEBUG
42/* these are for debugging purposes only! */
43
44/* define one and only one of these */
45#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
46#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
47#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
48#define INJECT_FAULTS_4 0 /* undef */
49#define INJECT_FAULTS_5 0 /* undef */
50#define INJECT_FAULTS_6 0
51
52/* if these are defined, the driver will fail! debug only */
53#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
54#define INJECT_FATAL_FAULT_2 0 /* undef */
55#define INJECT_FATAL_FAULT_3 0 /* undef */
56#endif
57
58//#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
59#define DPRINTK(x...) do { } while(0)
60
61#ifndef PRINTK
62# if DEBUG > 0
63# define PRINTK(x...) printk(KERN_DEBUG x)
64# else
65# define PRINTK(x...)
66# endif
67#endif
68
69static inline char * bmname(struct bitmap *bitmap)
70{
71 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
72}
73
74
75/*
76 * test if the bitmap is active
77 */
78int bitmap_active(struct bitmap *bitmap)
79{
80 unsigned long flags;
81 int res = 0;
82
83 if (!bitmap)
84 return res;
85 spin_lock_irqsave(&bitmap->lock, flags);
86 res = bitmap->flags & BITMAP_ACTIVE;
87 spin_unlock_irqrestore(&bitmap->lock, flags);
88 return res;
89}
90
91#define WRITE_POOL_SIZE 256
92/* mempool for queueing pending writes on the bitmap file */
b4e3ca1a 93static void *write_pool_alloc(gfp_t gfp_flags, void *data)
32a7627c
N
94{
95 return kmalloc(sizeof(struct page_list), gfp_flags);
96}
97
98static void write_pool_free(void *ptr, void *data)
99{
100 kfree(ptr);
101}
102
103/*
104 * just a placeholder - calls kmalloc for bitmap pages
105 */
106static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
107{
108 unsigned char *page;
109
44456d37 110#ifdef INJECT_FAULTS_1
32a7627c
N
111 page = NULL;
112#else
113 page = kmalloc(PAGE_SIZE, GFP_NOIO);
114#endif
115 if (!page)
116 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
117 else
a654b9d8 118 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
32a7627c
N
119 bmname(bitmap), page);
120 return page;
121}
122
123/*
124 * for now just a placeholder -- just calls kfree for bitmap pages
125 */
126static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
127{
128 PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
129 kfree(page);
130}
131
132/*
133 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
134 *
135 * 1) check to see if this page is allocated, if it's not then try to alloc
136 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
137 * page pointer directly as a counter
138 *
139 * if we find our page, we increment the page's refcount so that it stays
140 * allocated while we're using it
141 */
142static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
143{
144 unsigned char *mappage;
145
146 if (page >= bitmap->pages) {
147 printk(KERN_ALERT
148 "%s: invalid bitmap page request: %lu (> %lu)\n",
149 bmname(bitmap), page, bitmap->pages-1);
150 return -EINVAL;
151 }
152
153
154 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
155 return 0;
156
157 if (bitmap->bp[page].map) /* page is already allocated, just return */
158 return 0;
159
160 if (!create)
161 return -ENOENT;
162
163 spin_unlock_irq(&bitmap->lock);
164
165 /* this page has not been allocated yet */
166
167 if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
168 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
169 bmname(bitmap));
170 /* failed - set the hijacked flag so that we can use the
171 * pointer as a counter */
172 spin_lock_irq(&bitmap->lock);
173 if (!bitmap->bp[page].map)
174 bitmap->bp[page].hijacked = 1;
175 goto out;
176 }
177
178 /* got a page */
179
180 spin_lock_irq(&bitmap->lock);
181
182 /* recheck the page */
183
184 if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
185 /* somebody beat us to getting the page */
186 bitmap_free_page(bitmap, mappage);
187 return 0;
188 }
189
190 /* no page was in place and we have one, so install it */
191
192 memset(mappage, 0, PAGE_SIZE);
193 bitmap->bp[page].map = mappage;
194 bitmap->missing_pages--;
195out:
196 return 0;
197}
198
199
200/* if page is completely empty, put it back on the free list, or dealloc it */
201/* if page was hijacked, unmark the flag so it might get alloced next time */
202/* Note: lock should be held when calling this */
203static inline void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
204{
205 char *ptr;
206
207 if (bitmap->bp[page].count) /* page is still busy */
208 return;
209
210 /* page is no longer in use, it can be released */
211
212 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
213 bitmap->bp[page].hijacked = 0;
214 bitmap->bp[page].map = NULL;
215 return;
216 }
217
218 /* normal case, free the page */
219
220#if 0
221/* actually ... let's not. We will probably need the page again exactly when
222 * memory is tight and we are flusing to disk
223 */
224 return;
225#else
226 ptr = bitmap->bp[page].map;
227 bitmap->bp[page].map = NULL;
228 bitmap->missing_pages++;
229 bitmap_free_page(bitmap, ptr);
230 return;
231#endif
232}
233
234
235/*
236 * bitmap file handling - read and write the bitmap file and its superblock
237 */
238
239/* copy the pathname of a file to a buffer */
240char *file_path(struct file *file, char *buf, int count)
241{
242 struct dentry *d;
243 struct vfsmount *v;
244
245 if (!buf)
246 return NULL;
247
248 d = file->f_dentry;
249 v = file->f_vfsmnt;
250
251 buf = d_path(d, v, buf, count);
252
253 return IS_ERR(buf) ? NULL : buf;
254}
255
256/*
257 * basic page I/O operations
258 */
259
a654b9d8
N
260/* IO operations when bitmap is stored near all superblocks */
261static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index)
262{
263 /* choose a good rdev and read the page from there */
264
265 mdk_rdev_t *rdev;
266 struct list_head *tmp;
267 struct page *page = alloc_page(GFP_KERNEL);
268 sector_t target;
269
270 if (!page)
271 return ERR_PTR(-ENOMEM);
a654b9d8 272
ab904d63 273 ITERATE_RDEV(mddev, rdev, tmp) {
b2d444d7
N
274 if (! test_bit(In_sync, &rdev->flags)
275 || test_bit(Faulty, &rdev->flags))
ab904d63
N
276 continue;
277
a654b9d8
N
278 target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512);
279
ab904d63
N
280 if (sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)) {
281 page->index = index;
282 return page;
283 }
284 }
285 return ERR_PTR(-EIO);
a654b9d8 286
a654b9d8
N
287}
288
289static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait)
290{
291 mdk_rdev_t *rdev;
292 struct list_head *tmp;
293
294 ITERATE_RDEV(mddev, rdev, tmp)
b2d444d7
N
295 if (test_bit(In_sync, &rdev->flags)
296 && !test_bit(Faulty, &rdev->flags))
a654b9d8
N
297 md_super_write(mddev, rdev,
298 (rdev->sb_offset<<1) + offset
299 + page->index * (PAGE_SIZE/512),
300 PAGE_SIZE,
301 page);
302
303 if (wait)
304 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
305 return 0;
306}
307
32a7627c 308/*
a654b9d8 309 * write out a page to a file
32a7627c 310 */
77ad4bc7 311static int write_page(struct bitmap *bitmap, struct page *page, int wait)
32a7627c
N
312{
313 int ret = -ENOMEM;
314
a654b9d8
N
315 if (bitmap->file == NULL)
316 return write_sb_page(bitmap->mddev, bitmap->offset, page, wait);
317
8a5e9cf1
N
318 if (wait)
319 lock_page(page);
320 else {
321 if (TestSetPageLocked(page))
322 return -EAGAIN; /* already locked */
323 if (PageWriteback(page)) {
324 unlock_page(page);
325 return -EAGAIN;
326 }
327 }
32a7627c 328
32a7627c
N
329 ret = page->mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE);
330 if (!ret)
331 ret = page->mapping->a_ops->commit_write(NULL, page, 0,
332 PAGE_SIZE);
333 if (ret) {
32a7627c
N
334 unlock_page(page);
335 return ret;
336 }
337
338 set_page_dirty(page); /* force it to be written out */
77ad4bc7
N
339
340 if (!wait) {
341 /* add to list to be waited for by daemon */
342 struct page_list *item = mempool_alloc(bitmap->write_pool, GFP_NOIO);
343 item->page = page;
344 page_cache_get(page);
345 spin_lock(&bitmap->write_lock);
346 list_add(&item->list, &bitmap->complete_pages);
347 spin_unlock(&bitmap->write_lock);
348 md_wakeup_thread(bitmap->writeback_daemon);
349 }
32a7627c
N
350 return write_one_page(page, wait);
351}
352
353/* read a page from a file, pinning it into cache, and return bytes_read */
354static struct page *read_page(struct file *file, unsigned long index,
355 unsigned long *bytes_read)
356{
357 struct inode *inode = file->f_mapping->host;
358 struct page *page = NULL;
359 loff_t isize = i_size_read(inode);
360 unsigned long end_index = isize >> PAGE_CACHE_SHIFT;
361
362 PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_CACHE_SIZE,
363 (unsigned long long)index << PAGE_CACHE_SHIFT);
364
365 page = read_cache_page(inode->i_mapping, index,
366 (filler_t *)inode->i_mapping->a_ops->readpage, file);
367 if (IS_ERR(page))
368 goto out;
369 wait_on_page_locked(page);
370 if (!PageUptodate(page) || PageError(page)) {
371 page_cache_release(page);
372 page = ERR_PTR(-EIO);
373 goto out;
374 }
375
376 if (index > end_index) /* we have read beyond EOF */
377 *bytes_read = 0;
378 else if (index == end_index) /* possible short read */
379 *bytes_read = isize & ~PAGE_CACHE_MASK;
380 else
381 *bytes_read = PAGE_CACHE_SIZE; /* got a full page */
382out:
383 if (IS_ERR(page))
384 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
385 (int)PAGE_CACHE_SIZE,
386 (unsigned long long)index << PAGE_CACHE_SHIFT,
387 PTR_ERR(page));
388 return page;
389}
390
391/*
392 * bitmap file superblock operations
393 */
394
395/* update the event counter and sync the superblock to disk */
396int bitmap_update_sb(struct bitmap *bitmap)
397{
398 bitmap_super_t *sb;
399 unsigned long flags;
400
401 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
402 return 0;
403 spin_lock_irqsave(&bitmap->lock, flags);
404 if (!bitmap->sb_page) { /* no superblock */
405 spin_unlock_irqrestore(&bitmap->lock, flags);
406 return 0;
407 }
32a7627c
N
408 spin_unlock_irqrestore(&bitmap->lock, flags);
409 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
410 sb->events = cpu_to_le64(bitmap->mddev->events);
411 if (!bitmap->mddev->degraded)
412 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
413 kunmap(bitmap->sb_page);
8a5e9cf1 414 return write_page(bitmap, bitmap->sb_page, 1);
32a7627c
N
415}
416
417/* print out the bitmap file superblock */
418void bitmap_print_sb(struct bitmap *bitmap)
419{
420 bitmap_super_t *sb;
421
422 if (!bitmap || !bitmap->sb_page)
423 return;
424 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
425 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
a2cff26a
N
426 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
427 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
428 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
32a7627c
N
429 *(__u32 *)(sb->uuid+0),
430 *(__u32 *)(sb->uuid+4),
431 *(__u32 *)(sb->uuid+8),
432 *(__u32 *)(sb->uuid+12));
a2cff26a 433 printk(KERN_DEBUG " events: %llu\n",
32a7627c 434 (unsigned long long) le64_to_cpu(sb->events));
a2cff26a 435 printk(KERN_DEBUG "events cleared: %llu\n",
32a7627c 436 (unsigned long long) le64_to_cpu(sb->events_cleared));
a2cff26a
N
437 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
438 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
439 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
440 printk(KERN_DEBUG " sync size: %llu KB\n",
441 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
4b6d287f 442 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
32a7627c
N
443 kunmap(bitmap->sb_page);
444}
445
446/* read the superblock from the bitmap file and initialize some bitmap fields */
447static int bitmap_read_sb(struct bitmap *bitmap)
448{
449 char *reason = NULL;
450 bitmap_super_t *sb;
4b6d287f 451 unsigned long chunksize, daemon_sleep, write_behind;
32a7627c
N
452 unsigned long bytes_read;
453 unsigned long long events;
454 int err = -EINVAL;
455
456 /* page 0 is the superblock, read it... */
a654b9d8
N
457 if (bitmap->file)
458 bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read);
459 else {
460 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0);
461 bytes_read = PAGE_SIZE;
462 }
32a7627c
N
463 if (IS_ERR(bitmap->sb_page)) {
464 err = PTR_ERR(bitmap->sb_page);
465 bitmap->sb_page = NULL;
466 return err;
467 }
468
469 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
470
471 if (bytes_read < sizeof(*sb)) { /* short read */
472 printk(KERN_INFO "%s: bitmap file superblock truncated\n",
473 bmname(bitmap));
474 err = -ENOSPC;
475 goto out;
476 }
477
478 chunksize = le32_to_cpu(sb->chunksize);
479 daemon_sleep = le32_to_cpu(sb->daemon_sleep);
4b6d287f 480 write_behind = le32_to_cpu(sb->write_behind);
32a7627c
N
481
482 /* verify that the bitmap-specific fields are valid */
483 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
484 reason = "bad magic";
bd926c63
N
485 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
486 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
32a7627c
N
487 reason = "unrecognized superblock version";
488 else if (chunksize < 512 || chunksize > (1024 * 1024 * 4))
489 reason = "bitmap chunksize out of range (512B - 4MB)";
490 else if ((1 << ffz(~chunksize)) != chunksize)
491 reason = "bitmap chunksize not a power of 2";
492 else if (daemon_sleep < 1 || daemon_sleep > 15)
4b6d287f
N
493 reason = "daemon sleep period out of range (1-15s)";
494 else if (write_behind > COUNTER_MAX)
495 reason = "write-behind limit out of range (0 - 16383)";
32a7627c
N
496 if (reason) {
497 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
498 bmname(bitmap), reason);
499 goto out;
500 }
501
502 /* keep the array size field of the bitmap superblock up to date */
503 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
504
505 if (!bitmap->mddev->persistent)
506 goto success;
507
508 /*
509 * if we have a persistent array superblock, compare the
510 * bitmap's UUID and event counter to the mddev's
511 */
512 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
513 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
514 bmname(bitmap));
515 goto out;
516 }
517 events = le64_to_cpu(sb->events);
518 if (events < bitmap->mddev->events) {
519 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
520 "-- forcing full recovery\n", bmname(bitmap), events,
521 (unsigned long long) bitmap->mddev->events);
522 sb->state |= BITMAP_STALE;
523 }
524success:
525 /* assign fields using values from superblock */
526 bitmap->chunksize = chunksize;
527 bitmap->daemon_sleep = daemon_sleep;
585f0dd5 528 bitmap->daemon_lastrun = jiffies;
4b6d287f 529 bitmap->max_write_behind = write_behind;
32a7627c 530 bitmap->flags |= sb->state;
bd926c63
N
531 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
532 bitmap->flags |= BITMAP_HOSTENDIAN;
32a7627c 533 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
6a07997f
N
534 if (sb->state & BITMAP_STALE)
535 bitmap->events_cleared = bitmap->mddev->events;
32a7627c
N
536 err = 0;
537out:
538 kunmap(bitmap->sb_page);
539 if (err)
540 bitmap_print_sb(bitmap);
541 return err;
542}
543
544enum bitmap_mask_op {
545 MASK_SET,
546 MASK_UNSET
547};
548
549/* record the state of the bitmap in the superblock */
550static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
551 enum bitmap_mask_op op)
552{
553 bitmap_super_t *sb;
554 unsigned long flags;
555
556 spin_lock_irqsave(&bitmap->lock, flags);
557 if (!bitmap || !bitmap->sb_page) { /* can't set the state */
558 spin_unlock_irqrestore(&bitmap->lock, flags);
559 return;
560 }
561 page_cache_get(bitmap->sb_page);
562 spin_unlock_irqrestore(&bitmap->lock, flags);
563 sb = (bitmap_super_t *)kmap(bitmap->sb_page);
564 switch (op) {
565 case MASK_SET: sb->state |= bits;
566 break;
567 case MASK_UNSET: sb->state &= ~bits;
568 break;
569 default: BUG();
570 }
571 kunmap(bitmap->sb_page);
572 page_cache_release(bitmap->sb_page);
573}
574
575/*
576 * general bitmap file operations
577 */
578
579/* calculate the index of the page that contains this bit */
580static inline unsigned long file_page_index(unsigned long chunk)
581{
582 return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
583}
584
585/* calculate the (bit) offset of this bit within a page */
586static inline unsigned long file_page_offset(unsigned long chunk)
587{
588 return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
589}
590
591/*
592 * return a pointer to the page in the filemap that contains the given bit
593 *
594 * this lookup is complicated by the fact that the bitmap sb might be exactly
595 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
596 * 0 or page 1
597 */
598static inline struct page *filemap_get_page(struct bitmap *bitmap,
599 unsigned long chunk)
600{
601 return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
602}
603
604
605static void bitmap_file_unmap(struct bitmap *bitmap)
606{
607 struct page **map, *sb_page;
608 unsigned long *attr;
609 int pages;
610 unsigned long flags;
611
612 spin_lock_irqsave(&bitmap->lock, flags);
613 map = bitmap->filemap;
614 bitmap->filemap = NULL;
615 attr = bitmap->filemap_attr;
616 bitmap->filemap_attr = NULL;
617 pages = bitmap->file_pages;
618 bitmap->file_pages = 0;
619 sb_page = bitmap->sb_page;
620 bitmap->sb_page = NULL;
621 spin_unlock_irqrestore(&bitmap->lock, flags);
622
623 while (pages--)
624 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
625 page_cache_release(map[pages]);
626 kfree(map);
627 kfree(attr);
628
629 if (sb_page)
630 page_cache_release(sb_page);
631}
632
500af87a 633static void bitmap_stop_daemon(struct bitmap *bitmap);
32a7627c
N
634
635/* dequeue the next item in a page list -- don't call from irq context */
77ad4bc7 636static struct page_list *dequeue_page(struct bitmap *bitmap)
32a7627c
N
637{
638 struct page_list *item = NULL;
77ad4bc7 639 struct list_head *head = &bitmap->complete_pages;
32a7627c
N
640
641 spin_lock(&bitmap->write_lock);
642 if (list_empty(head))
643 goto out;
644 item = list_entry(head->prev, struct page_list, list);
645 list_del(head->prev);
646out:
647 spin_unlock(&bitmap->write_lock);
648 return item;
649}
650
651static void drain_write_queues(struct bitmap *bitmap)
652{
32a7627c 653 struct page_list *item;
32a7627c 654
77ad4bc7
N
655 while ((item = dequeue_page(bitmap))) {
656 /* don't bother to wait */
657 page_cache_release(item->page);
658 mempool_free(item, bitmap->write_pool);
32a7627c
N
659 }
660
32a7627c 661 wake_up(&bitmap->write_wait);
32a7627c
N
662}
663
664static void bitmap_file_put(struct bitmap *bitmap)
665{
666 struct file *file;
667 struct inode *inode;
668 unsigned long flags;
669
670 spin_lock_irqsave(&bitmap->lock, flags);
671 file = bitmap->file;
672 bitmap->file = NULL;
673 spin_unlock_irqrestore(&bitmap->lock, flags);
674
500af87a 675 bitmap_stop_daemon(bitmap);
32a7627c
N
676
677 drain_write_queues(bitmap);
678
679 bitmap_file_unmap(bitmap);
680
681 if (file) {
682 inode = file->f_mapping->host;
683 spin_lock(&inode->i_lock);
684 atomic_set(&inode->i_writecount, 1); /* allow writes again */
685 spin_unlock(&inode->i_lock);
686 fput(file);
687 }
688}
689
690
691/*
692 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
693 * then it is no longer reliable, so we stop using it and we mark the file
694 * as failed in the superblock
695 */
696static void bitmap_file_kick(struct bitmap *bitmap)
697{
698 char *path, *ptr = NULL;
699
700 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET);
701 bitmap_update_sb(bitmap);
702
a654b9d8
N
703 if (bitmap->file) {
704 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
705 if (path)
706 ptr = file_path(bitmap->file, path, PAGE_SIZE);
32a7627c 707
a654b9d8
N
708 printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n",
709 bmname(bitmap), ptr ? ptr : "");
32a7627c 710
a654b9d8
N
711 kfree(path);
712 }
32a7627c
N
713
714 bitmap_file_put(bitmap);
715
716 return;
717}
718
719enum bitmap_page_attr {
720 BITMAP_PAGE_DIRTY = 1, // there are set bits that need to be synced
721 BITMAP_PAGE_CLEAN = 2, // there are bits that might need to be cleared
722 BITMAP_PAGE_NEEDWRITE=4, // there are cleared bits that need to be synced
723};
724
725static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
726 enum bitmap_page_attr attr)
727{
728 bitmap->filemap_attr[page->index] |= attr;
729}
730
731static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
732 enum bitmap_page_attr attr)
733{
734 bitmap->filemap_attr[page->index] &= ~attr;
735}
736
737static inline unsigned long get_page_attr(struct bitmap *bitmap, struct page *page)
738{
739 return bitmap->filemap_attr[page->index];
740}
741
742/*
743 * bitmap_file_set_bit -- called before performing a write to the md device
744 * to set (and eventually sync) a particular bit in the bitmap file
745 *
746 * we set the bit immediately, then we record the page number so that
747 * when an unplug occurs, we can flush the dirty pages out to disk
748 */
749static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
750{
751 unsigned long bit;
752 struct page *page;
753 void *kaddr;
754 unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
755
a654b9d8 756 if (!bitmap->filemap) {
32a7627c
N
757 return;
758 }
759
760 page = filemap_get_page(bitmap, chunk);
761 bit = file_page_offset(chunk);
762
763
764 /* make sure the page stays cached until it gets written out */
765 if (! (get_page_attr(bitmap, page) & BITMAP_PAGE_DIRTY))
766 page_cache_get(page);
767
768 /* set the bit */
769 kaddr = kmap_atomic(page, KM_USER0);
bd926c63
N
770 if (bitmap->flags & BITMAP_HOSTENDIAN)
771 set_bit(bit, kaddr);
772 else
773 ext2_set_bit(bit, kaddr);
32a7627c
N
774 kunmap_atomic(kaddr, KM_USER0);
775 PRINTK("set file bit %lu page %lu\n", bit, page->index);
776
777 /* record page number so it gets flushed to disk when unplug occurs */
778 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
779
780}
781
782/* this gets called when the md device is ready to unplug its underlying
783 * (slave) device queues -- before we let any writes go down, we need to
784 * sync the dirty pages of the bitmap file to disk */
785int bitmap_unplug(struct bitmap *bitmap)
786{
787 unsigned long i, attr, flags;
788 struct page *page;
789 int wait = 0;
8a5e9cf1 790 int err;
32a7627c
N
791
792 if (!bitmap)
793 return 0;
794
795 /* look at each page to see if there are any set bits that need to be
796 * flushed out to disk */
797 for (i = 0; i < bitmap->file_pages; i++) {
798 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 799 if (!bitmap->filemap) {
32a7627c
N
800 spin_unlock_irqrestore(&bitmap->lock, flags);
801 return 0;
802 }
803 page = bitmap->filemap[i];
804 attr = get_page_attr(bitmap, page);
805 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
806 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
807 if ((attr & BITMAP_PAGE_DIRTY))
808 wait = 1;
809 spin_unlock_irqrestore(&bitmap->lock, flags);
810
8a5e9cf1
N
811 if (attr & (BITMAP_PAGE_DIRTY | BITMAP_PAGE_NEEDWRITE)) {
812 err = write_page(bitmap, page, 0);
813 if (err == -EAGAIN) {
814 if (attr & BITMAP_PAGE_DIRTY)
815 err = write_page(bitmap, page, 1);
816 else
817 err = 0;
818 }
819 if (err)
bfb39fba 820 return 1;
8a5e9cf1 821 }
32a7627c
N
822 }
823 if (wait) { /* if any writes were performed, we need to wait on them */
a654b9d8
N
824 if (bitmap->file) {
825 spin_lock_irq(&bitmap->write_lock);
826 wait_event_lock_irq(bitmap->write_wait,
827 list_empty(&bitmap->complete_pages), bitmap->write_lock,
828 wake_up_process(bitmap->writeback_daemon->tsk));
829 spin_unlock_irq(&bitmap->write_lock);
830 } else
831 wait_event(bitmap->mddev->sb_wait,
832 atomic_read(&bitmap->mddev->pending_writes)==0);
32a7627c
N
833 }
834 return 0;
835}
836
6a07997f 837static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
32a7627c
N
838/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
839 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
840 * memory mapping of the bitmap file
841 * Special cases:
842 * if there's no bitmap file, or if the bitmap file had been
843 * previously kicked from the array, we mark all the bits as
844 * 1's in order to cause a full resync.
6a07997f
N
845 *
846 * We ignore all bits for sectors that end earlier than 'start'.
847 * This is used when reading an out-of-date bitmap...
32a7627c 848 */
6a07997f 849static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
32a7627c
N
850{
851 unsigned long i, chunks, index, oldindex, bit;
852 struct page *page = NULL, *oldpage = NULL;
853 unsigned long num_pages, bit_cnt = 0;
854 struct file *file;
855 unsigned long bytes, offset, dummy;
856 int outofdate;
857 int ret = -ENOSPC;
858
859 chunks = bitmap->chunks;
860 file = bitmap->file;
861
a654b9d8 862 BUG_ON(!file && !bitmap->offset);
32a7627c 863
44456d37 864#ifdef INJECT_FAULTS_3
32a7627c
N
865 outofdate = 1;
866#else
867 outofdate = bitmap->flags & BITMAP_STALE;
868#endif
869 if (outofdate)
870 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
871 "recovery\n", bmname(bitmap));
872
873 bytes = (chunks + 7) / 8;
bc7f77de 874
cdbb4cc2 875 num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
bc7f77de 876
a654b9d8 877 if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
32a7627c
N
878 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
879 bmname(bitmap),
880 (unsigned long) i_size_read(file->f_mapping->host),
881 bytes + sizeof(bitmap_super_t));
882 goto out;
883 }
bc7f77de
N
884
885 ret = -ENOMEM;
886
32a7627c 887 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
bc7f77de 888 if (!bitmap->filemap)
32a7627c 889 goto out;
32a7627c
N
890
891 bitmap->filemap_attr = kmalloc(sizeof(long) * num_pages, GFP_KERNEL);
bc7f77de 892 if (!bitmap->filemap_attr)
32a7627c 893 goto out;
32a7627c
N
894
895 memset(bitmap->filemap_attr, 0, sizeof(long) * num_pages);
896
897 oldindex = ~0L;
898
899 for (i = 0; i < chunks; i++) {
bd926c63 900 int b;
32a7627c
N
901 index = file_page_index(i);
902 bit = file_page_offset(i);
903 if (index != oldindex) { /* this is a new page, read it in */
904 /* unmap the old page, we're done with it */
905 if (oldpage != NULL)
906 kunmap(oldpage);
907 if (index == 0) {
908 /*
909 * if we're here then the superblock page
910 * contains some bits (PAGE_SIZE != sizeof sb)
911 * we've already read it in, so just use it
912 */
913 page = bitmap->sb_page;
914 offset = sizeof(bitmap_super_t);
a654b9d8 915 } else if (file) {
32a7627c 916 page = read_page(file, index, &dummy);
a654b9d8
N
917 offset = 0;
918 } else {
919 page = read_sb_page(bitmap->mddev, bitmap->offset, index);
32a7627c
N
920 offset = 0;
921 }
a654b9d8
N
922 if (IS_ERR(page)) { /* read error */
923 ret = PTR_ERR(page);
924 goto out;
925 }
926
32a7627c
N
927 oldindex = index;
928 oldpage = page;
929 kmap(page);
930
931 if (outofdate) {
932 /*
933 * if bitmap is out of date, dirty the
934 * whole page and write it out
935 */
936 memset(page_address(page) + offset, 0xff,
6a07997f 937 PAGE_SIZE - offset);
77ad4bc7 938 ret = write_page(bitmap, page, 1);
32a7627c
N
939 if (ret) {
940 kunmap(page);
941 /* release, page not in filemap yet */
942 page_cache_release(page);
943 goto out;
944 }
945 }
946
947 bitmap->filemap[bitmap->file_pages++] = page;
948 }
bd926c63
N
949 if (bitmap->flags & BITMAP_HOSTENDIAN)
950 b = test_bit(bit, page_address(page));
951 else
952 b = ext2_test_bit(bit, page_address(page));
953 if (b) {
32a7627c 954 /* if the disk bit is set, set the memory bit */
6a07997f
N
955 bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
956 ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
957 );
32a7627c 958 bit_cnt++;
6a07997f 959 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
32a7627c 960 }
32a7627c
N
961 }
962
963 /* everything went OK */
964 ret = 0;
965 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
966
967 if (page) /* unmap the last page */
968 kunmap(page);
969
970 if (bit_cnt) { /* Kick recovery if any bits were set */
971 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
972 md_wakeup_thread(bitmap->mddev->thread);
973 }
974
975out:
976 printk(KERN_INFO "%s: bitmap initialized from disk: "
977 "read %lu/%lu pages, set %lu bits, status: %d\n",
978 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
979
980 return ret;
981}
982
a654b9d8
N
983void bitmap_write_all(struct bitmap *bitmap)
984{
985 /* We don't actually write all bitmap blocks here,
986 * just flag them as needing to be written
987 */
988
989 unsigned long chunks = bitmap->chunks;
990 unsigned long bytes = (chunks+7)/8 + sizeof(bitmap_super_t);
991 unsigned long num_pages = (bytes + PAGE_SIZE-1) / PAGE_SIZE;
992 while (num_pages--)
993 bitmap->filemap_attr[num_pages] |= BITMAP_PAGE_NEEDWRITE;
994}
995
32a7627c
N
996
997static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
998{
999 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1000 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1001 bitmap->bp[page].count += inc;
1002/*
1003 if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1004 (unsigned long long)offset, inc, bitmap->bp[page].count);
1005*/
1006 bitmap_checkfree(bitmap, page);
1007}
1008static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1009 sector_t offset, int *blocks,
1010 int create);
1011
1012/*
1013 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1014 * out to disk
1015 */
1016
1017int bitmap_daemon_work(struct bitmap *bitmap)
1018{
aa3163f8 1019 unsigned long j;
32a7627c
N
1020 unsigned long flags;
1021 struct page *page = NULL, *lastpage = NULL;
1022 int err = 0;
1023 int blocks;
1024 int attr;
1025
1026 if (bitmap == NULL)
1027 return 0;
1028 if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1029 return 0;
1030 bitmap->daemon_lastrun = jiffies;
1031
1032 for (j = 0; j < bitmap->chunks; j++) {
1033 bitmap_counter_t *bmc;
1034 spin_lock_irqsave(&bitmap->lock, flags);
a654b9d8 1035 if (!bitmap->filemap) {
32a7627c
N
1036 /* error or shutdown */
1037 spin_unlock_irqrestore(&bitmap->lock, flags);
1038 break;
1039 }
1040
1041 page = filemap_get_page(bitmap, j);
32a7627c
N
1042
1043 if (page != lastpage) {
aa3163f8
N
1044 /* skip this page unless it's marked as needing cleaning */
1045 if (!((attr=get_page_attr(bitmap, page)) & BITMAP_PAGE_CLEAN)) {
1046 if (attr & BITMAP_PAGE_NEEDWRITE) {
1047 page_cache_get(page);
1048 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1049 }
1050 spin_unlock_irqrestore(&bitmap->lock, flags);
1051 if (attr & BITMAP_PAGE_NEEDWRITE) {
8a5e9cf1
N
1052 switch (write_page(bitmap, page, 0)) {
1053 case -EAGAIN:
1054 set_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1055 break;
1056 case 0:
1057 break;
1058 default:
aa3163f8 1059 bitmap_file_kick(bitmap);
8a5e9cf1 1060 }
aa3163f8
N
1061 page_cache_release(page);
1062 }
1063 continue;
1064 }
1065
32a7627c
N
1066 /* grab the new page, sync and release the old */
1067 page_cache_get(page);
1068 if (lastpage != NULL) {
1069 if (get_page_attr(bitmap, lastpage) & BITMAP_PAGE_NEEDWRITE) {
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 err = 0;
1075 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1076 }
32a7627c
N
1077 } else {
1078 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1079 spin_unlock_irqrestore(&bitmap->lock, flags);
1080 }
1081 kunmap(lastpage);
1082 page_cache_release(lastpage);
1083 if (err)
1084 bitmap_file_kick(bitmap);
1085 } else
1086 spin_unlock_irqrestore(&bitmap->lock, flags);
1087 lastpage = page;
1088 kmap(page);
1089/*
1090 printk("bitmap clean at page %lu\n", j);
1091*/
1092 spin_lock_irqsave(&bitmap->lock, flags);
1093 clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1094 }
1095 bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1096 &blocks, 0);
1097 if (bmc) {
1098/*
1099 if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1100*/
1101 if (*bmc == 2) {
1102 *bmc=1; /* maybe clear the bit next time */
1103 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1104 } else if (*bmc == 1) {
1105 /* we can clear the bit */
1106 *bmc = 0;
1107 bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap),
1108 -1);
1109
1110 /* clear the bit */
bd926c63
N
1111 if (bitmap->flags & BITMAP_HOSTENDIAN)
1112 clear_bit(file_page_offset(j), page_address(page));
1113 else
1114 ext2_clear_bit(file_page_offset(j), page_address(page));
32a7627c
N
1115 }
1116 }
1117 spin_unlock_irqrestore(&bitmap->lock, flags);
1118 }
1119
1120 /* now sync the final page */
1121 if (lastpage != NULL) {
1122 kunmap(lastpage);
1123 spin_lock_irqsave(&bitmap->lock, flags);
1124 if (get_page_attr(bitmap, lastpage) &BITMAP_PAGE_NEEDWRITE) {
1125 clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1126 spin_unlock_irqrestore(&bitmap->lock, flags);
77ad4bc7 1127 err = write_page(bitmap, lastpage, 0);
8a5e9cf1
N
1128 if (err == -EAGAIN) {
1129 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1130 err = 0;
1131 }
32a7627c
N
1132 } else {
1133 set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1134 spin_unlock_irqrestore(&bitmap->lock, flags);
1135 }
1136
1137 page_cache_release(lastpage);
1138 }
1139
1140 return err;
1141}
1142
1143static void daemon_exit(struct bitmap *bitmap, mdk_thread_t **daemon)
1144{
1145 mdk_thread_t *dmn;
1146 unsigned long flags;
1147
1148 /* if no one is waiting on us, we'll free the md thread struct
1149 * and exit, otherwise we let the waiter clean things up */
1150 spin_lock_irqsave(&bitmap->lock, flags);
1151 if ((dmn = *daemon)) { /* no one is waiting, cleanup and exit */
1152 *daemon = NULL;
1153 spin_unlock_irqrestore(&bitmap->lock, flags);
1154 kfree(dmn);
1155 complete_and_exit(NULL, 0); /* do_exit not exported */
1156 }
1157 spin_unlock_irqrestore(&bitmap->lock, flags);
1158}
1159
1160static void bitmap_writeback_daemon(mddev_t *mddev)
1161{
1162 struct bitmap *bitmap = mddev->bitmap;
1163 struct page *page;
1164 struct page_list *item;
1165 int err = 0;
1166
77ad4bc7
N
1167 if (signal_pending(current)) {
1168 printk(KERN_INFO
1169 "%s: bitmap writeback daemon got signal, exiting...\n",
1170 bmname(bitmap));
1171 err = -EINTR;
1172 goto out;
1173 }
9ba00538
N
1174 if (bitmap == NULL)
1175 /* about to be stopped. */
1176 return;
32a7627c 1177
77ad4bc7
N
1178 PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap));
1179 /* wait on bitmap page writebacks */
1180 while ((item = dequeue_page(bitmap))) {
1181 page = item->page;
1182 mempool_free(item, bitmap->write_pool);
1183 PRINTK("wait on page writeback: %p\n", page);
1184 wait_on_page_writeback(page);
1185 PRINTK("finished page writeback: %p\n", page);
1186
1187 err = PageError(page);
1188 page_cache_release(page);
1189 if (err) {
1190 printk(KERN_WARNING "%s: bitmap file writeback "
1191 "failed (page %lu): %d\n",
1192 bmname(bitmap), page->index, err);
1193 bitmap_file_kick(bitmap);
1194 goto out;
32a7627c
N
1195 }
1196 }
77ad4bc7
N
1197 out:
1198 wake_up(&bitmap->write_wait);
32a7627c
N
1199 if (err) {
1200 printk(KERN_INFO "%s: bitmap writeback daemon exiting (%d)\n",
77ad4bc7 1201 bmname(bitmap), err);
32a7627c
N
1202 daemon_exit(bitmap, &bitmap->writeback_daemon);
1203 }
32a7627c
N
1204}
1205
500af87a 1206static mdk_thread_t *bitmap_start_daemon(struct bitmap *bitmap,
32a7627c
N
1207 void (*func)(mddev_t *), char *name)
1208{
1209 mdk_thread_t *daemon;
32a7627c
N
1210 char namebuf[32];
1211
44456d37 1212#ifdef INJECT_FATAL_FAULT_2
32a7627c
N
1213 daemon = NULL;
1214#else
1215 sprintf(namebuf, "%%s_%s", name);
1216 daemon = md_register_thread(func, bitmap->mddev, namebuf);
1217#endif
1218 if (!daemon) {
1219 printk(KERN_ERR "%s: failed to start bitmap daemon\n",
1220 bmname(bitmap));
500af87a 1221 return ERR_PTR(-ECHILD);
32a7627c
N
1222 }
1223
32a7627c
N
1224 md_wakeup_thread(daemon); /* start it running */
1225
1226 PRINTK("%s: %s daemon (pid %d) started...\n",
d80a138c 1227 bmname(bitmap), name, daemon->tsk->pid);
32a7627c 1228
500af87a 1229 return daemon;
32a7627c
N
1230}
1231
500af87a 1232static void bitmap_stop_daemon(struct bitmap *bitmap)
32a7627c 1233{
500af87a
N
1234 /* the daemon can't stop itself... it'll just exit instead... */
1235 if (bitmap->writeback_daemon && ! IS_ERR(bitmap->writeback_daemon) &&
1236 current->pid != bitmap->writeback_daemon->tsk->pid) {
1237 mdk_thread_t *daemon;
1238 unsigned long flags;
32a7627c 1239
500af87a
N
1240 spin_lock_irqsave(&bitmap->lock, flags);
1241 daemon = bitmap->writeback_daemon;
1242 bitmap->writeback_daemon = NULL;
1243 spin_unlock_irqrestore(&bitmap->lock, flags);
1244 if (daemon && ! IS_ERR(daemon))
1245 md_unregister_thread(daemon); /* destroy the thread */
1246 }
32a7627c
N
1247}
1248
1249static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1250 sector_t offset, int *blocks,
1251 int create)
1252{
1253 /* If 'create', we might release the lock and reclaim it.
1254 * The lock must have been taken with interrupts enabled.
1255 * If !create, we don't release the lock.
1256 */
1257 sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1258 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1259 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1260 sector_t csize;
1261
1262 if (bitmap_checkpage(bitmap, page, create) < 0) {
1263 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1264 *blocks = csize - (offset & (csize- 1));
1265 return NULL;
1266 }
1267 /* now locked ... */
1268
1269 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1270 /* should we use the first or second counter field
1271 * of the hijacked pointer? */
1272 int hi = (pageoff > PAGE_COUNTER_MASK);
1273 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1274 PAGE_COUNTER_SHIFT - 1);
1275 *blocks = csize - (offset & (csize- 1));
1276 return &((bitmap_counter_t *)
1277 &bitmap->bp[page].map)[hi];
1278 } else { /* page is allocated */
1279 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1280 *blocks = csize - (offset & (csize- 1));
1281 return (bitmap_counter_t *)
1282 &(bitmap->bp[page].map[pageoff]);
1283 }
1284}
1285
4b6d287f 1286int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
32a7627c
N
1287{
1288 if (!bitmap) return 0;
4b6d287f
N
1289
1290 if (behind) {
1291 atomic_inc(&bitmap->behind_writes);
1292 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1293 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1294 }
1295
32a7627c
N
1296 while (sectors) {
1297 int blocks;
1298 bitmap_counter_t *bmc;
1299
1300 spin_lock_irq(&bitmap->lock);
1301 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1302 if (!bmc) {
1303 spin_unlock_irq(&bitmap->lock);
1304 return 0;
1305 }
1306
1307 switch(*bmc) {
1308 case 0:
1309 bitmap_file_set_bit(bitmap, offset);
1310 bitmap_count_page(bitmap,offset, 1);
1311 blk_plug_device(bitmap->mddev->queue);
1312 /* fall through */
1313 case 1:
1314 *bmc = 2;
1315 }
1316 if ((*bmc & COUNTER_MAX) == COUNTER_MAX) BUG();
1317 (*bmc)++;
1318
1319 spin_unlock_irq(&bitmap->lock);
1320
1321 offset += blocks;
1322 if (sectors > blocks)
1323 sectors -= blocks;
1324 else sectors = 0;
1325 }
1326 return 0;
1327}
1328
1329void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
4b6d287f 1330 int success, int behind)
32a7627c
N
1331{
1332 if (!bitmap) return;
4b6d287f
N
1333 if (behind) {
1334 atomic_dec(&bitmap->behind_writes);
1335 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1336 atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1337 }
1338
32a7627c
N
1339 while (sectors) {
1340 int blocks;
1341 unsigned long flags;
1342 bitmap_counter_t *bmc;
1343
1344 spin_lock_irqsave(&bitmap->lock, flags);
1345 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1346 if (!bmc) {
1347 spin_unlock_irqrestore(&bitmap->lock, flags);
1348 return;
1349 }
1350
1351 if (!success && ! (*bmc & NEEDED_MASK))
1352 *bmc |= NEEDED_MASK;
1353
1354 (*bmc)--;
1355 if (*bmc <= 2) {
1356 set_page_attr(bitmap,
1357 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1358 BITMAP_PAGE_CLEAN);
1359 }
1360 spin_unlock_irqrestore(&bitmap->lock, flags);
1361 offset += blocks;
1362 if (sectors > blocks)
1363 sectors -= blocks;
1364 else sectors = 0;
1365 }
1366}
1367
6a806c51
N
1368int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1369 int degraded)
32a7627c
N
1370{
1371 bitmap_counter_t *bmc;
1372 int rv;
1373 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1374 *blocks = 1024;
1375 return 1; /* always resync if no bitmap */
1376 }
1377 spin_lock_irq(&bitmap->lock);
1378 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1379 rv = 0;
1380 if (bmc) {
1381 /* locked */
1382 if (RESYNC(*bmc))
1383 rv = 1;
1384 else if (NEEDED(*bmc)) {
1385 rv = 1;
6a806c51
N
1386 if (!degraded) { /* don't set/clear bits if degraded */
1387 *bmc |= RESYNC_MASK;
1388 *bmc &= ~NEEDED_MASK;
1389 }
32a7627c
N
1390 }
1391 }
1392 spin_unlock_irq(&bitmap->lock);
1393 return rv;
1394}
1395
1396void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1397{
1398 bitmap_counter_t *bmc;
1399 unsigned long flags;
1400/*
1401 if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1402*/ if (bitmap == NULL) {
1403 *blocks = 1024;
1404 return;
1405 }
1406 spin_lock_irqsave(&bitmap->lock, flags);
1407 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1408 if (bmc == NULL)
1409 goto unlock;
1410 /* locked */
1411/*
1412 if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1413*/
1414 if (RESYNC(*bmc)) {
1415 *bmc &= ~RESYNC_MASK;
1416
1417 if (!NEEDED(*bmc) && aborted)
1418 *bmc |= NEEDED_MASK;
1419 else {
1420 if (*bmc <= 2) {
1421 set_page_attr(bitmap,
1422 filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1423 BITMAP_PAGE_CLEAN);
1424 }
1425 }
1426 }
1427 unlock:
1428 spin_unlock_irqrestore(&bitmap->lock, flags);
1429}
1430
1431void bitmap_close_sync(struct bitmap *bitmap)
1432{
1433 /* Sync has finished, and any bitmap chunks that weren't synced
1434 * properly have been aborted. It remains to us to clear the
1435 * RESYNC bit wherever it is still on
1436 */
1437 sector_t sector = 0;
1438 int blocks;
1439 if (!bitmap) return;
1440 while (sector < bitmap->mddev->resync_max_sectors) {
1441 bitmap_end_sync(bitmap, sector, &blocks, 0);
1442/*
1443 if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n",
1444 (unsigned long long)sector, blocks);
1445*/ sector += blocks;
1446 }
1447}
1448
6a07997f 1449static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
32a7627c
N
1450{
1451 /* For each chunk covered by any of these sectors, set the
193f1c93 1452 * counter to 1 and set resync_needed. They should all
32a7627c
N
1453 * be 0 at this point
1454 */
193f1c93
N
1455
1456 int secs;
1457 bitmap_counter_t *bmc;
1458 spin_lock_irq(&bitmap->lock);
1459 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1460 if (!bmc) {
32a7627c 1461 spin_unlock_irq(&bitmap->lock);
193f1c93 1462 return;
32a7627c 1463 }
193f1c93
N
1464 if (! *bmc) {
1465 struct page *page;
6a07997f 1466 *bmc = 1 | (needed?NEEDED_MASK:0);
193f1c93
N
1467 bitmap_count_page(bitmap, offset, 1);
1468 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1469 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1470 }
1471 spin_unlock_irq(&bitmap->lock);
1472
32a7627c
N
1473}
1474
6b8b3e8a
N
1475/*
1476 * flush out any pending updates
1477 */
1478void bitmap_flush(mddev_t *mddev)
1479{
1480 struct bitmap *bitmap = mddev->bitmap;
1481 int sleep;
1482
1483 if (!bitmap) /* there was no bitmap */
1484 return;
1485
1486 /* run the daemon_work three time to ensure everything is flushed
1487 * that can be
1488 */
1489 sleep = bitmap->daemon_sleep;
1490 bitmap->daemon_sleep = 0;
1491 bitmap_daemon_work(bitmap);
1492 bitmap_daemon_work(bitmap);
1493 bitmap_daemon_work(bitmap);
1494 bitmap->daemon_sleep = sleep;
1495 bitmap_update_sb(bitmap);
1496}
1497
32a7627c
N
1498/*
1499 * free memory that was allocated
1500 */
3178b0db 1501static void bitmap_free(struct bitmap *bitmap)
32a7627c
N
1502{
1503 unsigned long k, pages;
1504 struct bitmap_page *bp;
32a7627c
N
1505
1506 if (!bitmap) /* there was no bitmap */
1507 return;
1508
32a7627c
N
1509 /* release the bitmap file and kill the daemon */
1510 bitmap_file_put(bitmap);
1511
1512 bp = bitmap->bp;
1513 pages = bitmap->pages;
1514
1515 /* free all allocated memory */
1516
1517 mempool_destroy(bitmap->write_pool);
1518
1519 if (bp) /* deallocate the page memory */
1520 for (k = 0; k < pages; k++)
1521 if (bp[k].map && !bp[k].hijacked)
1522 kfree(bp[k].map);
1523 kfree(bp);
1524 kfree(bitmap);
1525}
3178b0db
N
1526void bitmap_destroy(mddev_t *mddev)
1527{
1528 struct bitmap *bitmap = mddev->bitmap;
1529
1530 if (!bitmap) /* there was no bitmap */
1531 return;
1532
1533 mddev->bitmap = NULL; /* disconnect from the md device */
1534
1535 bitmap_free(bitmap);
1536}
32a7627c
N
1537
1538/*
1539 * initialize the bitmap structure
1540 * if this returns an error, bitmap_destroy must be called to do clean up
1541 */
1542int bitmap_create(mddev_t *mddev)
1543{
1544 struct bitmap *bitmap;
1545 unsigned long blocks = mddev->resync_max_sectors;
1546 unsigned long chunks;
1547 unsigned long pages;
1548 struct file *file = mddev->bitmap_file;
1549 int err;
6a07997f 1550 sector_t start;
32a7627c
N
1551
1552 BUG_ON(sizeof(bitmap_super_t) != 256);
1553
a654b9d8 1554 if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
32a7627c
N
1555 return 0;
1556
a654b9d8
N
1557 BUG_ON(file && mddev->bitmap_offset);
1558
32a7627c
N
1559 bitmap = kmalloc(sizeof(*bitmap), GFP_KERNEL);
1560 if (!bitmap)
1561 return -ENOMEM;
1562
1563 memset(bitmap, 0, sizeof(*bitmap));
1564
1565 spin_lock_init(&bitmap->lock);
1566 bitmap->mddev = mddev;
32a7627c
N
1567
1568 spin_lock_init(&bitmap->write_lock);
32a7627c
N
1569 INIT_LIST_HEAD(&bitmap->complete_pages);
1570 init_waitqueue_head(&bitmap->write_wait);
1571 bitmap->write_pool = mempool_create(WRITE_POOL_SIZE, write_pool_alloc,
1572 write_pool_free, NULL);
3178b0db 1573 err = -ENOMEM;
32a7627c 1574 if (!bitmap->write_pool)
3178b0db 1575 goto error;
32a7627c
N
1576
1577 bitmap->file = file;
a654b9d8
N
1578 bitmap->offset = mddev->bitmap_offset;
1579 if (file) get_file(file);
32a7627c
N
1580 /* read superblock from bitmap file (this sets bitmap->chunksize) */
1581 err = bitmap_read_sb(bitmap);
1582 if (err)
3178b0db 1583 goto error;
32a7627c
N
1584
1585 bitmap->chunkshift = find_first_bit(&bitmap->chunksize,
1586 sizeof(bitmap->chunksize));
1587
1588 /* now that chunksize and chunkshift are set, we can use these macros */
1589 chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) /
1590 CHUNK_BLOCK_RATIO(bitmap);
1591 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1592
1593 BUG_ON(!pages);
1594
1595 bitmap->chunks = chunks;
1596 bitmap->pages = pages;
1597 bitmap->missing_pages = pages;
1598 bitmap->counter_bits = COUNTER_BITS;
1599
1600 bitmap->syncchunk = ~0UL;
1601
44456d37 1602#ifdef INJECT_FATAL_FAULT_1
32a7627c
N
1603 bitmap->bp = NULL;
1604#else
1605 bitmap->bp = kmalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1606#endif
3178b0db 1607 err = -ENOMEM;
32a7627c 1608 if (!bitmap->bp)
3178b0db 1609 goto error;
32a7627c
N
1610 memset(bitmap->bp, 0, pages * sizeof(*bitmap->bp));
1611
1612 bitmap->flags |= BITMAP_ACTIVE;
1613
1614 /* now that we have some pages available, initialize the in-memory
1615 * bitmap from the on-disk bitmap */
6a07997f
N
1616 start = 0;
1617 if (mddev->degraded == 0
1618 || bitmap->events_cleared == mddev->events)
1619 /* no need to keep dirty bits to optimise a re-add of a missing device */
1620 start = mddev->recovery_cp;
1621 err = bitmap_init_from_disk(bitmap, start);
193f1c93 1622
32a7627c 1623 if (err)
3178b0db 1624 goto error;
32a7627c
N
1625
1626 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1627 pages, bmname(bitmap));
1628
3178b0db
N
1629 mddev->bitmap = bitmap;
1630
500af87a
N
1631 if (file)
1632 /* kick off the bitmap writeback daemon */
1633 bitmap->writeback_daemon =
1634 bitmap_start_daemon(bitmap,
1635 bitmap_writeback_daemon,
1636 "bitmap_wb");
1637
1638 if (IS_ERR(bitmap->writeback_daemon))
1639 return PTR_ERR(bitmap->writeback_daemon);
32a7627c 1640 return bitmap_update_sb(bitmap);
3178b0db
N
1641
1642 error:
1643 bitmap_free(bitmap);
1644 return err;
32a7627c
N
1645}
1646
1647/* the bitmap API -- for raid personalities */
1648EXPORT_SYMBOL(bitmap_startwrite);
1649EXPORT_SYMBOL(bitmap_endwrite);
1650EXPORT_SYMBOL(bitmap_start_sync);
1651EXPORT_SYMBOL(bitmap_end_sync);
1652EXPORT_SYMBOL(bitmap_unplug);
1653EXPORT_SYMBOL(bitmap_close_sync);
1654EXPORT_SYMBOL(bitmap_daemon_work);