md/bitmap: remove bitmap_mask_state
[deliverable/linux.git] / drivers / md / bitmap.c
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 */
11
12 /*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
16 */
17
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
30 #include "md.h"
31 #include "bitmap.h"
32
33 static inline char *bmname(struct bitmap *bitmap)
34 {
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
36 }
37
38 /*
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
40 *
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
44 *
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
47 */
48 static int bitmap_checkpage(struct bitmap *bitmap,
49 unsigned long page, int create)
50 __releases(bitmap->lock)
51 __acquires(bitmap->lock)
52 {
53 unsigned char *mappage;
54
55 if (page >= bitmap->pages) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
58 * It is harmless.
59 */
60 return -EINVAL;
61 }
62
63 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
64 return 0;
65
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
67 return 0;
68
69 if (!create)
70 return -ENOENT;
71
72 /* this page has not been allocated yet */
73
74 spin_unlock_irq(&bitmap->lock);
75 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
76 spin_lock_irq(&bitmap->lock);
77
78 if (mappage == NULL) {
79 pr_debug("%s: bitmap map page allocation failed, hijacking\n",
80 bmname(bitmap));
81 /* failed - set the hijacked flag so that we can use the
82 * pointer as a counter */
83 if (!bitmap->bp[page].map)
84 bitmap->bp[page].hijacked = 1;
85 } else if (bitmap->bp[page].map ||
86 bitmap->bp[page].hijacked) {
87 /* somebody beat us to getting the page */
88 kfree(mappage);
89 return 0;
90 } else {
91
92 /* no page was in place and we have one, so install it */
93
94 bitmap->bp[page].map = mappage;
95 bitmap->missing_pages--;
96 }
97 return 0;
98 }
99
100 /* if page is completely empty, put it back on the free list, or dealloc it */
101 /* if page was hijacked, unmark the flag so it might get alloced next time */
102 /* Note: lock should be held when calling this */
103 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
104 {
105 char *ptr;
106
107 if (bitmap->bp[page].count) /* page is still busy */
108 return;
109
110 /* page is no longer in use, it can be released */
111
112 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
113 bitmap->bp[page].hijacked = 0;
114 bitmap->bp[page].map = NULL;
115 } else {
116 /* normal case, free the page */
117 ptr = bitmap->bp[page].map;
118 bitmap->bp[page].map = NULL;
119 bitmap->missing_pages++;
120 kfree(ptr);
121 }
122 }
123
124 /*
125 * bitmap file handling - read and write the bitmap file and its superblock
126 */
127
128 /*
129 * basic page I/O operations
130 */
131
132 /* IO operations when bitmap is stored near all superblocks */
133 static int read_sb_page(struct mddev *mddev, loff_t offset,
134 struct page *page,
135 unsigned long index, int size)
136 {
137 /* choose a good rdev and read the page from there */
138
139 struct md_rdev *rdev;
140 sector_t target;
141
142 rdev_for_each(rdev, mddev) {
143 if (! test_bit(In_sync, &rdev->flags)
144 || test_bit(Faulty, &rdev->flags))
145 continue;
146
147 target = offset + index * (PAGE_SIZE/512);
148
149 if (sync_page_io(rdev, target,
150 roundup(size, bdev_logical_block_size(rdev->bdev)),
151 page, READ, true)) {
152 page->index = index;
153 return 0;
154 }
155 }
156 return -EIO;
157 }
158
159 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
160 {
161 /* Iterate the disks of an mddev, using rcu to protect access to the
162 * linked list, and raising the refcount of devices we return to ensure
163 * they don't disappear while in use.
164 * As devices are only added or removed when raid_disk is < 0 and
165 * nr_pending is 0 and In_sync is clear, the entries we return will
166 * still be in the same position on the list when we re-enter
167 * list_for_each_continue_rcu.
168 */
169 struct list_head *pos;
170 rcu_read_lock();
171 if (rdev == NULL)
172 /* start at the beginning */
173 pos = &mddev->disks;
174 else {
175 /* release the previous rdev and start from there. */
176 rdev_dec_pending(rdev, mddev);
177 pos = &rdev->same_set;
178 }
179 list_for_each_continue_rcu(pos, &mddev->disks) {
180 rdev = list_entry(pos, struct md_rdev, same_set);
181 if (rdev->raid_disk >= 0 &&
182 !test_bit(Faulty, &rdev->flags)) {
183 /* this is a usable devices */
184 atomic_inc(&rdev->nr_pending);
185 rcu_read_unlock();
186 return rdev;
187 }
188 }
189 rcu_read_unlock();
190 return NULL;
191 }
192
193 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
194 {
195 struct md_rdev *rdev = NULL;
196 struct block_device *bdev;
197 struct mddev *mddev = bitmap->mddev;
198 struct bitmap_storage *store = &bitmap->storage;
199
200 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
201 int size = PAGE_SIZE;
202 loff_t offset = mddev->bitmap_info.offset;
203
204 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
205
206 if (page->index == store->file_pages-1) {
207 int last_page_size = store->bytes & (PAGE_SIZE-1);
208 if (last_page_size == 0)
209 last_page_size = PAGE_SIZE;
210 size = roundup(last_page_size,
211 bdev_logical_block_size(bdev));
212 }
213 /* Just make sure we aren't corrupting data or
214 * metadata
215 */
216 if (mddev->external) {
217 /* Bitmap could be anywhere. */
218 if (rdev->sb_start + offset + (page->index
219 * (PAGE_SIZE/512))
220 > rdev->data_offset
221 &&
222 rdev->sb_start + offset
223 < (rdev->data_offset + mddev->dev_sectors
224 + (PAGE_SIZE/512)))
225 goto bad_alignment;
226 } else if (offset < 0) {
227 /* DATA BITMAP METADATA */
228 if (offset
229 + (long)(page->index * (PAGE_SIZE/512))
230 + size/512 > 0)
231 /* bitmap runs in to metadata */
232 goto bad_alignment;
233 if (rdev->data_offset + mddev->dev_sectors
234 > rdev->sb_start + offset)
235 /* data runs in to bitmap */
236 goto bad_alignment;
237 } else if (rdev->sb_start < rdev->data_offset) {
238 /* METADATA BITMAP DATA */
239 if (rdev->sb_start
240 + offset
241 + page->index*(PAGE_SIZE/512) + size/512
242 > rdev->data_offset)
243 /* bitmap runs in to data */
244 goto bad_alignment;
245 } else {
246 /* DATA METADATA BITMAP - no problems */
247 }
248 md_super_write(mddev, rdev,
249 rdev->sb_start + offset
250 + page->index * (PAGE_SIZE/512),
251 size,
252 page);
253 }
254
255 if (wait)
256 md_super_wait(mddev);
257 return 0;
258
259 bad_alignment:
260 return -EINVAL;
261 }
262
263 static void bitmap_file_kick(struct bitmap *bitmap);
264 /*
265 * write out a page to a file
266 */
267 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
268 {
269 struct buffer_head *bh;
270
271 if (bitmap->storage.file == NULL) {
272 switch (write_sb_page(bitmap, page, wait)) {
273 case -EINVAL:
274 bitmap->flags |= BITMAP_WRITE_ERROR;
275 }
276 } else {
277
278 bh = page_buffers(page);
279
280 while (bh && bh->b_blocknr) {
281 atomic_inc(&bitmap->pending_writes);
282 set_buffer_locked(bh);
283 set_buffer_mapped(bh);
284 submit_bh(WRITE | REQ_SYNC, bh);
285 bh = bh->b_this_page;
286 }
287
288 if (wait)
289 wait_event(bitmap->write_wait,
290 atomic_read(&bitmap->pending_writes)==0);
291 }
292 if (bitmap->flags & BITMAP_WRITE_ERROR)
293 bitmap_file_kick(bitmap);
294 }
295
296 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
297 {
298 struct bitmap *bitmap = bh->b_private;
299 unsigned long flags;
300
301 if (!uptodate) {
302 spin_lock_irqsave(&bitmap->lock, flags);
303 bitmap->flags |= BITMAP_WRITE_ERROR;
304 spin_unlock_irqrestore(&bitmap->lock, flags);
305 }
306 if (atomic_dec_and_test(&bitmap->pending_writes))
307 wake_up(&bitmap->write_wait);
308 }
309
310 /* copied from buffer.c */
311 static void
312 __clear_page_buffers(struct page *page)
313 {
314 ClearPagePrivate(page);
315 set_page_private(page, 0);
316 page_cache_release(page);
317 }
318 static void free_buffers(struct page *page)
319 {
320 struct buffer_head *bh;
321
322 if (!PagePrivate(page))
323 return;
324
325 bh = page_buffers(page);
326 while (bh) {
327 struct buffer_head *next = bh->b_this_page;
328 free_buffer_head(bh);
329 bh = next;
330 }
331 __clear_page_buffers(page);
332 put_page(page);
333 }
334
335 /* read a page from a file.
336 * We both read the page, and attach buffers to the page to record the
337 * address of each block (using bmap). These addresses will be used
338 * to write the block later, completely bypassing the filesystem.
339 * This usage is similar to how swap files are handled, and allows us
340 * to write to a file with no concerns of memory allocation failing.
341 */
342 static int read_page(struct file *file, unsigned long index,
343 struct bitmap *bitmap,
344 unsigned long count,
345 struct page *page)
346 {
347 int ret = 0;
348 struct inode *inode = file->f_path.dentry->d_inode;
349 struct buffer_head *bh;
350 sector_t block;
351
352 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
353 (unsigned long long)index << PAGE_SHIFT);
354
355 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
356 if (!bh) {
357 ret = -ENOMEM;
358 goto out;
359 }
360 attach_page_buffers(page, bh);
361 block = index << (PAGE_SHIFT - inode->i_blkbits);
362 while (bh) {
363 if (count == 0)
364 bh->b_blocknr = 0;
365 else {
366 bh->b_blocknr = bmap(inode, block);
367 if (bh->b_blocknr == 0) {
368 /* Cannot use this file! */
369 ret = -EINVAL;
370 goto out;
371 }
372 bh->b_bdev = inode->i_sb->s_bdev;
373 if (count < (1<<inode->i_blkbits))
374 count = 0;
375 else
376 count -= (1<<inode->i_blkbits);
377
378 bh->b_end_io = end_bitmap_write;
379 bh->b_private = bitmap;
380 atomic_inc(&bitmap->pending_writes);
381 set_buffer_locked(bh);
382 set_buffer_mapped(bh);
383 submit_bh(READ, bh);
384 }
385 block++;
386 bh = bh->b_this_page;
387 }
388 page->index = index;
389
390 wait_event(bitmap->write_wait,
391 atomic_read(&bitmap->pending_writes)==0);
392 if (bitmap->flags & BITMAP_WRITE_ERROR)
393 ret = -EIO;
394 out:
395 if (ret)
396 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
397 (int)PAGE_SIZE,
398 (unsigned long long)index << PAGE_SHIFT,
399 ret);
400 return ret;
401 }
402
403 /*
404 * bitmap file superblock operations
405 */
406
407 /* update the event counter and sync the superblock to disk */
408 void bitmap_update_sb(struct bitmap *bitmap)
409 {
410 bitmap_super_t *sb;
411
412 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
413 return;
414 if (bitmap->mddev->bitmap_info.external)
415 return;
416 if (!bitmap->storage.sb_page) /* no superblock */
417 return;
418 sb = kmap_atomic(bitmap->storage.sb_page);
419 sb->events = cpu_to_le64(bitmap->mddev->events);
420 if (bitmap->mddev->events < bitmap->events_cleared)
421 /* rocking back to read-only */
422 bitmap->events_cleared = bitmap->mddev->events;
423 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
424 sb->state = cpu_to_le32(bitmap->flags);
425 /* Just in case these have been changed via sysfs: */
426 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
427 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
428 kunmap_atomic(sb);
429 write_page(bitmap, bitmap->storage.sb_page, 1);
430 }
431
432 /* print out the bitmap file superblock */
433 void bitmap_print_sb(struct bitmap *bitmap)
434 {
435 bitmap_super_t *sb;
436
437 if (!bitmap || !bitmap->storage.sb_page)
438 return;
439 sb = kmap_atomic(bitmap->storage.sb_page);
440 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
441 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
442 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
443 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
444 *(__u32 *)(sb->uuid+0),
445 *(__u32 *)(sb->uuid+4),
446 *(__u32 *)(sb->uuid+8),
447 *(__u32 *)(sb->uuid+12));
448 printk(KERN_DEBUG " events: %llu\n",
449 (unsigned long long) le64_to_cpu(sb->events));
450 printk(KERN_DEBUG "events cleared: %llu\n",
451 (unsigned long long) le64_to_cpu(sb->events_cleared));
452 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
453 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
454 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
455 printk(KERN_DEBUG " sync size: %llu KB\n",
456 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
457 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
458 kunmap_atomic(sb);
459 }
460
461 /*
462 * bitmap_new_disk_sb
463 * @bitmap
464 *
465 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
466 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
467 * This function verifies 'bitmap_info' and populates the on-disk bitmap
468 * structure, which is to be written to disk.
469 *
470 * Returns: 0 on success, -Exxx on error
471 */
472 static int bitmap_new_disk_sb(struct bitmap *bitmap)
473 {
474 bitmap_super_t *sb;
475 unsigned long chunksize, daemon_sleep, write_behind;
476 int err = -EINVAL;
477
478 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
479 if (IS_ERR(bitmap->storage.sb_page)) {
480 err = PTR_ERR(bitmap->storage.sb_page);
481 bitmap->storage.sb_page = NULL;
482 return err;
483 }
484 bitmap->storage.sb_page->index = 0;
485
486 sb = kmap_atomic(bitmap->storage.sb_page);
487
488 sb->magic = cpu_to_le32(BITMAP_MAGIC);
489 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
490
491 chunksize = bitmap->mddev->bitmap_info.chunksize;
492 BUG_ON(!chunksize);
493 if (!is_power_of_2(chunksize)) {
494 kunmap_atomic(sb);
495 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
496 return -EINVAL;
497 }
498 sb->chunksize = cpu_to_le32(chunksize);
499
500 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
501 if (!daemon_sleep ||
502 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
503 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
504 daemon_sleep = 5 * HZ;
505 }
506 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
507 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
508
509 /*
510 * FIXME: write_behind for RAID1. If not specified, what
511 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
512 */
513 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
514 if (write_behind > COUNTER_MAX)
515 write_behind = COUNTER_MAX / 2;
516 sb->write_behind = cpu_to_le32(write_behind);
517 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
518
519 /* keep the array size field of the bitmap superblock up to date */
520 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
521
522 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
523
524 bitmap->flags |= BITMAP_STALE;
525 sb->state |= cpu_to_le32(BITMAP_STALE);
526 bitmap->events_cleared = bitmap->mddev->events;
527 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
528
529 kunmap_atomic(sb);
530
531 return 0;
532 }
533
534 /* read the superblock from the bitmap file and initialize some bitmap fields */
535 static int bitmap_read_sb(struct bitmap *bitmap)
536 {
537 char *reason = NULL;
538 bitmap_super_t *sb;
539 unsigned long chunksize, daemon_sleep, write_behind;
540 unsigned long long events;
541 int err = -EINVAL;
542 struct page *sb_page;
543
544 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
545 chunksize = 128 * 1024 * 1024;
546 daemon_sleep = 5 * HZ;
547 write_behind = 0;
548 bitmap->flags = BITMAP_STALE;
549 err = 0;
550 goto out_no_sb;
551 }
552 /* page 0 is the superblock, read it... */
553 sb_page = alloc_page(GFP_KERNEL);
554 if (!sb_page)
555 return -ENOMEM;
556 bitmap->storage.sb_page = sb_page;
557
558 if (bitmap->storage.file) {
559 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
560 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
561
562 err = read_page(bitmap->storage.file, 0,
563 bitmap, bytes, sb_page);
564 } else {
565 err = read_sb_page(bitmap->mddev,
566 bitmap->mddev->bitmap_info.offset,
567 sb_page,
568 0, sizeof(bitmap_super_t));
569 }
570 if (err)
571 return err;
572
573 sb = kmap_atomic(sb_page);
574
575 chunksize = le32_to_cpu(sb->chunksize);
576 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
577 write_behind = le32_to_cpu(sb->write_behind);
578
579 /* verify that the bitmap-specific fields are valid */
580 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
581 reason = "bad magic";
582 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
583 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
584 reason = "unrecognized superblock version";
585 else if (chunksize < 512)
586 reason = "bitmap chunksize too small";
587 else if (!is_power_of_2(chunksize))
588 reason = "bitmap chunksize not a power of 2";
589 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
590 reason = "daemon sleep period out of range";
591 else if (write_behind > COUNTER_MAX)
592 reason = "write-behind limit out of range (0 - 16383)";
593 if (reason) {
594 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
595 bmname(bitmap), reason);
596 goto out;
597 }
598
599 /* keep the array size field of the bitmap superblock up to date */
600 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
601
602 if (bitmap->mddev->persistent) {
603 /*
604 * We have a persistent array superblock, so compare the
605 * bitmap's UUID and event counter to the mddev's
606 */
607 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
608 printk(KERN_INFO
609 "%s: bitmap superblock UUID mismatch\n",
610 bmname(bitmap));
611 goto out;
612 }
613 events = le64_to_cpu(sb->events);
614 if (events < bitmap->mddev->events) {
615 printk(KERN_INFO
616 "%s: bitmap file is out of date (%llu < %llu) "
617 "-- forcing full recovery\n",
618 bmname(bitmap), events,
619 (unsigned long long) bitmap->mddev->events);
620 sb->state |= cpu_to_le32(BITMAP_STALE);
621 }
622 }
623
624 /* assign fields using values from superblock */
625 bitmap->flags |= le32_to_cpu(sb->state);
626 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
627 bitmap->flags |= BITMAP_HOSTENDIAN;
628 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
629 err = 0;
630 out:
631 kunmap_atomic(sb);
632 out_no_sb:
633 if (bitmap->flags & BITMAP_STALE)
634 bitmap->events_cleared = bitmap->mddev->events;
635 bitmap->mddev->bitmap_info.chunksize = chunksize;
636 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
637 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
638 if (err)
639 bitmap_print_sb(bitmap);
640 return err;
641 }
642
643 /*
644 * general bitmap file operations
645 */
646
647 /*
648 * on-disk bitmap:
649 *
650 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
651 * file a page at a time. There's a superblock at the start of the file.
652 */
653 /* calculate the index of the page that contains this bit */
654 static inline unsigned long file_page_index(struct bitmap_storage *store,
655 unsigned long chunk)
656 {
657 if (store->sb_page)
658 chunk += sizeof(bitmap_super_t) << 3;
659 return chunk >> PAGE_BIT_SHIFT;
660 }
661
662 /* calculate the (bit) offset of this bit within a page */
663 static inline unsigned long file_page_offset(struct bitmap_storage *store,
664 unsigned long chunk)
665 {
666 if (store->sb_page)
667 chunk += sizeof(bitmap_super_t) << 3;
668 return chunk & (PAGE_BITS - 1);
669 }
670
671 /*
672 * return a pointer to the page in the filemap that contains the given bit
673 *
674 * this lookup is complicated by the fact that the bitmap sb might be exactly
675 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
676 * 0 or page 1
677 */
678 static inline struct page *filemap_get_page(struct bitmap_storage *store,
679 unsigned long chunk)
680 {
681 if (file_page_index(store, chunk) >= store->file_pages)
682 return NULL;
683 return store->filemap[file_page_index(store, chunk)
684 - file_page_index(store, 0)];
685 }
686
687 static int bitmap_storage_alloc(struct bitmap_storage *store,
688 unsigned long chunks, int with_super)
689 {
690 int pnum;
691 unsigned long num_pages;
692 unsigned long bytes;
693
694 bytes = DIV_ROUND_UP(chunks, 8);
695 if (with_super)
696 bytes += sizeof(bitmap_super_t);
697
698 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
699
700 store->filemap = kmalloc(sizeof(struct page *)
701 * num_pages, GFP_KERNEL);
702 if (!store->filemap)
703 return -ENOMEM;
704
705 if (with_super && !store->sb_page) {
706 store->sb_page = alloc_page(GFP_KERNEL);
707 if (store->sb_page == NULL)
708 return -ENOMEM;
709 store->sb_page->index = 0;
710 }
711 pnum = 0;
712 if (store->sb_page) {
713 store->filemap[0] = store->sb_page;
714 pnum = 1;
715 }
716 for ( ; pnum < num_pages; pnum++) {
717 store->filemap[pnum] = alloc_page(GFP_KERNEL);
718 if (!store->filemap[pnum]) {
719 store->file_pages = pnum;
720 return -ENOMEM;
721 }
722 store->filemap[pnum]->index = pnum;
723 }
724 store->file_pages = pnum;
725
726 /* We need 4 bits per page, rounded up to a multiple
727 * of sizeof(unsigned long) */
728 store->filemap_attr = kzalloc(
729 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
730 GFP_KERNEL);
731 if (!store->filemap_attr)
732 return -ENOMEM;
733
734 store->bytes = bytes;
735
736 return 0;
737 }
738
739 static void bitmap_file_unmap(struct bitmap *bitmap)
740 {
741 struct page **map, *sb_page;
742 unsigned long *attr;
743 int pages;
744 unsigned long flags;
745 struct bitmap_storage *store = &bitmap->storage;
746
747 spin_lock_irqsave(&bitmap->lock, flags);
748 map = store->filemap;
749 store->filemap = NULL;
750 attr = store->filemap_attr;
751 store->filemap_attr = NULL;
752 pages = store->file_pages;
753 store->file_pages = 0;
754 sb_page = store->sb_page;
755 store->sb_page = NULL;
756 spin_unlock_irqrestore(&bitmap->lock, flags);
757
758 while (pages--)
759 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
760 free_buffers(map[pages]);
761 kfree(map);
762 kfree(attr);
763
764 if (sb_page)
765 free_buffers(sb_page);
766 }
767
768 static void bitmap_file_put(struct bitmap *bitmap)
769 {
770 struct file *file;
771 unsigned long flags;
772
773 spin_lock_irqsave(&bitmap->lock, flags);
774 file = bitmap->storage.file;
775 bitmap->storage.file = NULL;
776 spin_unlock_irqrestore(&bitmap->lock, flags);
777
778 if (file)
779 wait_event(bitmap->write_wait,
780 atomic_read(&bitmap->pending_writes)==0);
781 bitmap_file_unmap(bitmap);
782
783 if (file) {
784 struct inode *inode = file->f_path.dentry->d_inode;
785 invalidate_mapping_pages(inode->i_mapping, 0, -1);
786 fput(file);
787 }
788 }
789
790 /*
791 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
792 * then it is no longer reliable, so we stop using it and we mark the file
793 * as failed in the superblock
794 */
795 static void bitmap_file_kick(struct bitmap *bitmap)
796 {
797 char *path, *ptr = NULL;
798
799 if (!(bitmap->flags & BITMAP_STALE)) {
800 bitmap->flags |= BITMAP_STALE;
801 bitmap_update_sb(bitmap);
802
803 if (bitmap->storage.file) {
804 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
805 if (path)
806 ptr = d_path(&bitmap->storage.file->f_path,
807 path, PAGE_SIZE);
808
809 printk(KERN_ALERT
810 "%s: kicking failed bitmap file %s from array!\n",
811 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
812
813 kfree(path);
814 } else
815 printk(KERN_ALERT
816 "%s: disabling internal bitmap due to errors\n",
817 bmname(bitmap));
818 }
819
820 bitmap_file_put(bitmap);
821
822 return;
823 }
824
825 enum bitmap_page_attr {
826 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
827 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
828 * i.e. counter is 1 or 2. */
829 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
830 };
831
832 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
833 enum bitmap_page_attr attr)
834 {
835 __set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
836 }
837
838 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
839 enum bitmap_page_attr attr)
840 {
841 __clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
842 }
843
844 static inline unsigned long test_page_attr(struct bitmap *bitmap, int pnum,
845 enum bitmap_page_attr attr)
846 {
847 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
848 }
849
850 /*
851 * bitmap_file_set_bit -- called before performing a write to the md device
852 * to set (and eventually sync) a particular bit in the bitmap file
853 *
854 * we set the bit immediately, then we record the page number so that
855 * when an unplug occurs, we can flush the dirty pages out to disk
856 */
857 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
858 {
859 unsigned long bit;
860 struct page *page;
861 void *kaddr;
862 unsigned long chunk = block >> bitmap->chunkshift;
863
864 page = filemap_get_page(&bitmap->storage, chunk);
865 if (!page)
866 return;
867 bit = file_page_offset(&bitmap->storage, chunk);
868
869 /* set the bit */
870 kaddr = kmap_atomic(page);
871 if (bitmap->flags & BITMAP_HOSTENDIAN)
872 set_bit(bit, kaddr);
873 else
874 __set_bit_le(bit, kaddr);
875 kunmap_atomic(kaddr);
876 pr_debug("set file bit %lu page %lu\n", bit, page->index);
877 /* record page number so it gets flushed to disk when unplug occurs */
878 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
879 }
880
881 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
882 {
883 unsigned long bit;
884 struct page *page;
885 void *paddr;
886 unsigned long chunk = block >> bitmap->chunkshift;
887
888 page = filemap_get_page(&bitmap->storage, chunk);
889 if (!page)
890 return;
891 bit = file_page_offset(&bitmap->storage, chunk);
892 paddr = kmap_atomic(page);
893 if (bitmap->flags & BITMAP_HOSTENDIAN)
894 clear_bit(bit, paddr);
895 else
896 __clear_bit_le(bit, paddr);
897 kunmap_atomic(paddr);
898 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
899 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
900 bitmap->allclean = 0;
901 }
902 }
903
904 /* this gets called when the md device is ready to unplug its underlying
905 * (slave) device queues -- before we let any writes go down, we need to
906 * sync the dirty pages of the bitmap file to disk */
907 void bitmap_unplug(struct bitmap *bitmap)
908 {
909 unsigned long i, flags;
910 int dirty, need_write;
911 int wait = 0;
912
913 if (!bitmap || !bitmap->storage.filemap)
914 return;
915
916 /* look at each page to see if there are any set bits that need to be
917 * flushed out to disk */
918 for (i = 0; i < bitmap->storage.file_pages; i++) {
919 spin_lock_irqsave(&bitmap->lock, flags);
920 if (!bitmap->storage.filemap) {
921 spin_unlock_irqrestore(&bitmap->lock, flags);
922 return;
923 }
924 dirty = test_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
925 need_write = test_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
926 clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
927 clear_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
928 if (dirty || need_write)
929 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
930 if (dirty)
931 wait = 1;
932 spin_unlock_irqrestore(&bitmap->lock, flags);
933
934 if (dirty || need_write)
935 write_page(bitmap, bitmap->storage.filemap[i], 0);
936 }
937 if (wait) { /* if any writes were performed, we need to wait on them */
938 if (bitmap->storage.file)
939 wait_event(bitmap->write_wait,
940 atomic_read(&bitmap->pending_writes)==0);
941 else
942 md_super_wait(bitmap->mddev);
943 }
944 if (bitmap->flags & BITMAP_WRITE_ERROR)
945 bitmap_file_kick(bitmap);
946 }
947 EXPORT_SYMBOL(bitmap_unplug);
948
949 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
950 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
951 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
952 * memory mapping of the bitmap file
953 * Special cases:
954 * if there's no bitmap file, or if the bitmap file had been
955 * previously kicked from the array, we mark all the bits as
956 * 1's in order to cause a full resync.
957 *
958 * We ignore all bits for sectors that end earlier than 'start'.
959 * This is used when reading an out-of-date bitmap...
960 */
961 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
962 {
963 unsigned long i, chunks, index, oldindex, bit;
964 struct page *page = NULL;
965 unsigned long bit_cnt = 0;
966 struct file *file;
967 unsigned long offset;
968 int outofdate;
969 int ret = -ENOSPC;
970 void *paddr;
971 struct bitmap_storage *store = &bitmap->storage;
972
973 chunks = bitmap->chunks;
974 file = store->file;
975
976 if (!file && !bitmap->mddev->bitmap_info.offset) {
977 /* No permanent bitmap - fill with '1s'. */
978 store->filemap = NULL;
979 store->file_pages = 0;
980 for (i = 0; i < chunks ; i++) {
981 /* if the disk bit is set, set the memory bit */
982 int needed = ((sector_t)(i+1) << (bitmap->chunkshift)
983 >= start);
984 bitmap_set_memory_bits(bitmap,
985 (sector_t)i << bitmap->chunkshift,
986 needed);
987 }
988 return 0;
989 }
990
991 outofdate = bitmap->flags & BITMAP_STALE;
992 if (outofdate)
993 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
994 "recovery\n", bmname(bitmap));
995
996 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
997 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
998 bmname(bitmap),
999 (unsigned long) i_size_read(file->f_mapping->host),
1000 store->bytes);
1001 goto err;
1002 }
1003
1004 oldindex = ~0L;
1005 offset = 0;
1006 if (!bitmap->mddev->bitmap_info.external)
1007 offset = sizeof(bitmap_super_t);
1008
1009 for (i = 0; i < chunks; i++) {
1010 int b;
1011 index = file_page_index(&bitmap->storage, i);
1012 bit = file_page_offset(&bitmap->storage, i);
1013 if (index != oldindex) { /* this is a new page, read it in */
1014 int count;
1015 /* unmap the old page, we're done with it */
1016 if (index == store->file_pages-1)
1017 count = store->bytes - index * PAGE_SIZE;
1018 else
1019 count = PAGE_SIZE;
1020 page = store->filemap[index];
1021 if (file)
1022 ret = read_page(file, index, bitmap,
1023 count, page);
1024 else
1025 ret = read_sb_page(
1026 bitmap->mddev,
1027 bitmap->mddev->bitmap_info.offset,
1028 page,
1029 index, count);
1030
1031 if (ret)
1032 goto err;
1033
1034 oldindex = index;
1035
1036 if (outofdate) {
1037 /*
1038 * if bitmap is out of date, dirty the
1039 * whole page and write it out
1040 */
1041 paddr = kmap_atomic(page);
1042 memset(paddr + offset, 0xff,
1043 PAGE_SIZE - offset);
1044 kunmap_atomic(paddr);
1045 write_page(bitmap, page, 1);
1046
1047 ret = -EIO;
1048 if (bitmap->flags & BITMAP_WRITE_ERROR)
1049 goto err;
1050 }
1051 }
1052 paddr = kmap_atomic(page);
1053 if (bitmap->flags & BITMAP_HOSTENDIAN)
1054 b = test_bit(bit, paddr);
1055 else
1056 b = test_bit_le(bit, paddr);
1057 kunmap_atomic(paddr);
1058 if (b) {
1059 /* if the disk bit is set, set the memory bit */
1060 int needed = ((sector_t)(i+1) << bitmap->chunkshift
1061 >= start);
1062 bitmap_set_memory_bits(bitmap,
1063 (sector_t)i << bitmap->chunkshift,
1064 needed);
1065 bit_cnt++;
1066 }
1067 offset = 0;
1068 }
1069
1070 printk(KERN_INFO "%s: bitmap initialized from disk: "
1071 "read %lu pages, set %lu of %lu bits\n",
1072 bmname(bitmap), store->file_pages,
1073 bit_cnt, chunks);
1074
1075 return 0;
1076
1077 err:
1078 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1079 bmname(bitmap), ret);
1080 return ret;
1081 }
1082
1083 void bitmap_write_all(struct bitmap *bitmap)
1084 {
1085 /* We don't actually write all bitmap blocks here,
1086 * just flag them as needing to be written
1087 */
1088 int i;
1089
1090 if (!bitmap || !bitmap->storage.filemap)
1091 return;
1092 if (bitmap->storage.file)
1093 /* Only one copy, so nothing needed */
1094 return;
1095
1096 spin_lock_irq(&bitmap->lock);
1097 for (i = 0; i < bitmap->storage.file_pages; i++)
1098 set_page_attr(bitmap, i,
1099 BITMAP_PAGE_NEEDWRITE);
1100 bitmap->allclean = 0;
1101 spin_unlock_irq(&bitmap->lock);
1102 }
1103
1104 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1105 {
1106 sector_t chunk = offset >> bitmap->chunkshift;
1107 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1108 bitmap->bp[page].count += inc;
1109 bitmap_checkfree(bitmap, page);
1110 }
1111
1112 static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset)
1113 {
1114 sector_t chunk = offset >> bitmap->chunkshift;
1115 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1116 struct bitmap_page *bp = &bitmap->bp[page];
1117
1118 if (!bp->pending)
1119 bp->pending = 1;
1120 }
1121
1122 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1123 sector_t offset, sector_t *blocks,
1124 int create);
1125
1126 /*
1127 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1128 * out to disk
1129 */
1130
1131 void bitmap_daemon_work(struct mddev *mddev)
1132 {
1133 struct bitmap *bitmap;
1134 unsigned long j;
1135 unsigned long nextpage;
1136 unsigned long flags;
1137 sector_t blocks;
1138
1139 /* Use a mutex to guard daemon_work against
1140 * bitmap_destroy.
1141 */
1142 mutex_lock(&mddev->bitmap_info.mutex);
1143 bitmap = mddev->bitmap;
1144 if (bitmap == NULL) {
1145 mutex_unlock(&mddev->bitmap_info.mutex);
1146 return;
1147 }
1148 if (time_before(jiffies, bitmap->daemon_lastrun
1149 + mddev->bitmap_info.daemon_sleep))
1150 goto done;
1151
1152 bitmap->daemon_lastrun = jiffies;
1153 if (bitmap->allclean) {
1154 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1155 goto done;
1156 }
1157 bitmap->allclean = 1;
1158
1159 /* Any file-page which is PENDING now needs to be written.
1160 * So set NEEDWRITE now, then after we make any last-minute changes
1161 * we will write it.
1162 */
1163 spin_lock_irqsave(&bitmap->lock, flags);
1164 for (j = 0; j < bitmap->storage.file_pages; j++)
1165 if (test_page_attr(bitmap, j,
1166 BITMAP_PAGE_PENDING)) {
1167 set_page_attr(bitmap, j,
1168 BITMAP_PAGE_NEEDWRITE);
1169 clear_page_attr(bitmap, j,
1170 BITMAP_PAGE_PENDING);
1171 }
1172
1173 if (bitmap->need_sync &&
1174 mddev->bitmap_info.external == 0) {
1175 /* Arrange for superblock update as well as
1176 * other changes */
1177 bitmap_super_t *sb;
1178 bitmap->need_sync = 0;
1179 if (bitmap->storage.filemap) {
1180 sb = kmap_atomic(bitmap->storage.sb_page);
1181 sb->events_cleared =
1182 cpu_to_le64(bitmap->events_cleared);
1183 kunmap_atomic(sb);
1184 set_page_attr(bitmap, 0,
1185 BITMAP_PAGE_NEEDWRITE);
1186 }
1187 }
1188 /* Now look at the bitmap counters and if any are '2' or '1',
1189 * decrement and handle accordingly.
1190 */
1191 nextpage = 0;
1192 for (j = 0; j < bitmap->chunks; j++) {
1193 bitmap_counter_t *bmc;
1194 sector_t block = (sector_t)j << bitmap->chunkshift;
1195
1196 if (j == nextpage) {
1197 nextpage += PAGE_COUNTER_RATIO;
1198 if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1199 j |= PAGE_COUNTER_MASK;
1200 continue;
1201 }
1202 bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1203 }
1204 bmc = bitmap_get_counter(bitmap,
1205 block,
1206 &blocks, 0);
1207
1208 if (!bmc) {
1209 j |= PAGE_COUNTER_MASK;
1210 continue;
1211 }
1212 if (*bmc == 1 && !bitmap->need_sync) {
1213 /* We can clear the bit */
1214 *bmc = 0;
1215 bitmap_count_page(bitmap, block, -1);
1216 bitmap_file_clear_bit(bitmap, block);
1217 } else if (*bmc && *bmc <= 2) {
1218 *bmc = 1;
1219 bitmap_set_pending(bitmap, block);
1220 bitmap->allclean = 0;
1221 }
1222 }
1223
1224 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1225 * DIRTY pages need to be written by bitmap_unplug so it can wait
1226 * for them.
1227 * If we find any DIRTY page we stop there and let bitmap_unplug
1228 * handle all the rest. This is important in the case where
1229 * the first blocking holds the superblock and it has been updated.
1230 * We mustn't write any other blocks before the superblock.
1231 */
1232 for (j = 0; j < bitmap->storage.file_pages; j++) {
1233
1234 if (test_page_attr(bitmap, j,
1235 BITMAP_PAGE_DIRTY))
1236 /* bitmap_unplug will handle the rest */
1237 break;
1238 if (test_page_attr(bitmap, j,
1239 BITMAP_PAGE_NEEDWRITE)) {
1240 clear_page_attr(bitmap, j,
1241 BITMAP_PAGE_NEEDWRITE);
1242 spin_unlock_irqrestore(&bitmap->lock, flags);
1243 write_page(bitmap, bitmap->storage.filemap[j], 0);
1244 spin_lock_irqsave(&bitmap->lock, flags);
1245 if (!bitmap->storage.filemap)
1246 break;
1247 }
1248 }
1249 spin_unlock_irqrestore(&bitmap->lock, flags);
1250
1251 done:
1252 if (bitmap->allclean == 0)
1253 mddev->thread->timeout =
1254 mddev->bitmap_info.daemon_sleep;
1255 mutex_unlock(&mddev->bitmap_info.mutex);
1256 }
1257
1258 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1259 sector_t offset, sector_t *blocks,
1260 int create)
1261 __releases(bitmap->lock)
1262 __acquires(bitmap->lock)
1263 {
1264 /* If 'create', we might release the lock and reclaim it.
1265 * The lock must have been taken with interrupts enabled.
1266 * If !create, we don't release the lock.
1267 */
1268 sector_t chunk = offset >> bitmap->chunkshift;
1269 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1270 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1271 sector_t csize;
1272 int err;
1273
1274 err = bitmap_checkpage(bitmap, page, create);
1275
1276 if (bitmap->bp[page].hijacked ||
1277 bitmap->bp[page].map == NULL)
1278 csize = ((sector_t)1) << (bitmap->chunkshift +
1279 PAGE_COUNTER_SHIFT - 1);
1280 else
1281 csize = ((sector_t)1) << bitmap->chunkshift;
1282 *blocks = csize - (offset & (csize - 1));
1283
1284 if (err < 0)
1285 return NULL;
1286
1287 /* now locked ... */
1288
1289 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1290 /* should we use the first or second counter field
1291 * of the hijacked pointer? */
1292 int hi = (pageoff > PAGE_COUNTER_MASK);
1293 return &((bitmap_counter_t *)
1294 &bitmap->bp[page].map)[hi];
1295 } else /* page is allocated */
1296 return (bitmap_counter_t *)
1297 &(bitmap->bp[page].map[pageoff]);
1298 }
1299
1300 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1301 {
1302 if (!bitmap)
1303 return 0;
1304
1305 if (behind) {
1306 int bw;
1307 atomic_inc(&bitmap->behind_writes);
1308 bw = atomic_read(&bitmap->behind_writes);
1309 if (bw > bitmap->behind_writes_used)
1310 bitmap->behind_writes_used = bw;
1311
1312 pr_debug("inc write-behind count %d/%lu\n",
1313 bw, bitmap->mddev->bitmap_info.max_write_behind);
1314 }
1315
1316 while (sectors) {
1317 sector_t blocks;
1318 bitmap_counter_t *bmc;
1319
1320 spin_lock_irq(&bitmap->lock);
1321 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1322 if (!bmc) {
1323 spin_unlock_irq(&bitmap->lock);
1324 return 0;
1325 }
1326
1327 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1328 DEFINE_WAIT(__wait);
1329 /* note that it is safe to do the prepare_to_wait
1330 * after the test as long as we do it before dropping
1331 * the spinlock.
1332 */
1333 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1334 TASK_UNINTERRUPTIBLE);
1335 spin_unlock_irq(&bitmap->lock);
1336 io_schedule();
1337 finish_wait(&bitmap->overflow_wait, &__wait);
1338 continue;
1339 }
1340
1341 switch (*bmc) {
1342 case 0:
1343 bitmap_file_set_bit(bitmap, offset);
1344 bitmap_count_page(bitmap, offset, 1);
1345 /* fall through */
1346 case 1:
1347 *bmc = 2;
1348 }
1349
1350 (*bmc)++;
1351
1352 spin_unlock_irq(&bitmap->lock);
1353
1354 offset += blocks;
1355 if (sectors > blocks)
1356 sectors -= blocks;
1357 else
1358 sectors = 0;
1359 }
1360 return 0;
1361 }
1362 EXPORT_SYMBOL(bitmap_startwrite);
1363
1364 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1365 int success, int behind)
1366 {
1367 if (!bitmap)
1368 return;
1369 if (behind) {
1370 if (atomic_dec_and_test(&bitmap->behind_writes))
1371 wake_up(&bitmap->behind_wait);
1372 pr_debug("dec write-behind count %d/%lu\n",
1373 atomic_read(&bitmap->behind_writes),
1374 bitmap->mddev->bitmap_info.max_write_behind);
1375 }
1376
1377 while (sectors) {
1378 sector_t blocks;
1379 unsigned long flags;
1380 bitmap_counter_t *bmc;
1381
1382 spin_lock_irqsave(&bitmap->lock, flags);
1383 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1384 if (!bmc) {
1385 spin_unlock_irqrestore(&bitmap->lock, flags);
1386 return;
1387 }
1388
1389 if (success && !bitmap->mddev->degraded &&
1390 bitmap->events_cleared < bitmap->mddev->events) {
1391 bitmap->events_cleared = bitmap->mddev->events;
1392 bitmap->need_sync = 1;
1393 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1394 }
1395
1396 if (!success && !NEEDED(*bmc))
1397 *bmc |= NEEDED_MASK;
1398
1399 if (COUNTER(*bmc) == COUNTER_MAX)
1400 wake_up(&bitmap->overflow_wait);
1401
1402 (*bmc)--;
1403 if (*bmc <= 2) {
1404 bitmap_set_pending(bitmap, offset);
1405 bitmap->allclean = 0;
1406 }
1407 spin_unlock_irqrestore(&bitmap->lock, flags);
1408 offset += blocks;
1409 if (sectors > blocks)
1410 sectors -= blocks;
1411 else
1412 sectors = 0;
1413 }
1414 }
1415 EXPORT_SYMBOL(bitmap_endwrite);
1416
1417 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1418 int degraded)
1419 {
1420 bitmap_counter_t *bmc;
1421 int rv;
1422 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1423 *blocks = 1024;
1424 return 1; /* always resync if no bitmap */
1425 }
1426 spin_lock_irq(&bitmap->lock);
1427 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1428 rv = 0;
1429 if (bmc) {
1430 /* locked */
1431 if (RESYNC(*bmc))
1432 rv = 1;
1433 else if (NEEDED(*bmc)) {
1434 rv = 1;
1435 if (!degraded) { /* don't set/clear bits if degraded */
1436 *bmc |= RESYNC_MASK;
1437 *bmc &= ~NEEDED_MASK;
1438 }
1439 }
1440 }
1441 spin_unlock_irq(&bitmap->lock);
1442 return rv;
1443 }
1444
1445 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1446 int degraded)
1447 {
1448 /* bitmap_start_sync must always report on multiples of whole
1449 * pages, otherwise resync (which is very PAGE_SIZE based) will
1450 * get confused.
1451 * So call __bitmap_start_sync repeatedly (if needed) until
1452 * At least PAGE_SIZE>>9 blocks are covered.
1453 * Return the 'or' of the result.
1454 */
1455 int rv = 0;
1456 sector_t blocks1;
1457
1458 *blocks = 0;
1459 while (*blocks < (PAGE_SIZE>>9)) {
1460 rv |= __bitmap_start_sync(bitmap, offset,
1461 &blocks1, degraded);
1462 offset += blocks1;
1463 *blocks += blocks1;
1464 }
1465 return rv;
1466 }
1467 EXPORT_SYMBOL(bitmap_start_sync);
1468
1469 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1470 {
1471 bitmap_counter_t *bmc;
1472 unsigned long flags;
1473
1474 if (bitmap == NULL) {
1475 *blocks = 1024;
1476 return;
1477 }
1478 spin_lock_irqsave(&bitmap->lock, flags);
1479 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1480 if (bmc == NULL)
1481 goto unlock;
1482 /* locked */
1483 if (RESYNC(*bmc)) {
1484 *bmc &= ~RESYNC_MASK;
1485
1486 if (!NEEDED(*bmc) && aborted)
1487 *bmc |= NEEDED_MASK;
1488 else {
1489 if (*bmc <= 2) {
1490 bitmap_set_pending(bitmap, offset);
1491 bitmap->allclean = 0;
1492 }
1493 }
1494 }
1495 unlock:
1496 spin_unlock_irqrestore(&bitmap->lock, flags);
1497 }
1498 EXPORT_SYMBOL(bitmap_end_sync);
1499
1500 void bitmap_close_sync(struct bitmap *bitmap)
1501 {
1502 /* Sync has finished, and any bitmap chunks that weren't synced
1503 * properly have been aborted. It remains to us to clear the
1504 * RESYNC bit wherever it is still on
1505 */
1506 sector_t sector = 0;
1507 sector_t blocks;
1508 if (!bitmap)
1509 return;
1510 while (sector < bitmap->mddev->resync_max_sectors) {
1511 bitmap_end_sync(bitmap, sector, &blocks, 0);
1512 sector += blocks;
1513 }
1514 }
1515 EXPORT_SYMBOL(bitmap_close_sync);
1516
1517 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1518 {
1519 sector_t s = 0;
1520 sector_t blocks;
1521
1522 if (!bitmap)
1523 return;
1524 if (sector == 0) {
1525 bitmap->last_end_sync = jiffies;
1526 return;
1527 }
1528 if (time_before(jiffies, (bitmap->last_end_sync
1529 + bitmap->mddev->bitmap_info.daemon_sleep)))
1530 return;
1531 wait_event(bitmap->mddev->recovery_wait,
1532 atomic_read(&bitmap->mddev->recovery_active) == 0);
1533
1534 bitmap->mddev->curr_resync_completed = sector;
1535 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1536 sector &= ~((1ULL << bitmap->chunkshift) - 1);
1537 s = 0;
1538 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1539 bitmap_end_sync(bitmap, s, &blocks, 0);
1540 s += blocks;
1541 }
1542 bitmap->last_end_sync = jiffies;
1543 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1544 }
1545 EXPORT_SYMBOL(bitmap_cond_end_sync);
1546
1547 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1548 {
1549 /* For each chunk covered by any of these sectors, set the
1550 * counter to 2 and possibly set resync_needed. They should all
1551 * be 0 at this point
1552 */
1553
1554 sector_t secs;
1555 bitmap_counter_t *bmc;
1556 spin_lock_irq(&bitmap->lock);
1557 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1558 if (!bmc) {
1559 spin_unlock_irq(&bitmap->lock);
1560 return;
1561 }
1562 if (!*bmc) {
1563 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1564 bitmap_count_page(bitmap, offset, 1);
1565 bitmap_set_pending(bitmap, offset);
1566 bitmap->allclean = 0;
1567 }
1568 spin_unlock_irq(&bitmap->lock);
1569 }
1570
1571 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1572 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1573 {
1574 unsigned long chunk;
1575
1576 for (chunk = s; chunk <= e; chunk++) {
1577 sector_t sec = (sector_t)chunk << bitmap->chunkshift;
1578 bitmap_set_memory_bits(bitmap, sec, 1);
1579 spin_lock_irq(&bitmap->lock);
1580 bitmap_file_set_bit(bitmap, sec);
1581 spin_unlock_irq(&bitmap->lock);
1582 if (sec < bitmap->mddev->recovery_cp)
1583 /* We are asserting that the array is dirty,
1584 * so move the recovery_cp address back so
1585 * that it is obvious that it is dirty
1586 */
1587 bitmap->mddev->recovery_cp = sec;
1588 }
1589 }
1590
1591 /*
1592 * flush out any pending updates
1593 */
1594 void bitmap_flush(struct mddev *mddev)
1595 {
1596 struct bitmap *bitmap = mddev->bitmap;
1597 long sleep;
1598
1599 if (!bitmap) /* there was no bitmap */
1600 return;
1601
1602 /* run the daemon_work three time to ensure everything is flushed
1603 * that can be
1604 */
1605 sleep = mddev->bitmap_info.daemon_sleep * 2;
1606 bitmap->daemon_lastrun -= sleep;
1607 bitmap_daemon_work(mddev);
1608 bitmap->daemon_lastrun -= sleep;
1609 bitmap_daemon_work(mddev);
1610 bitmap->daemon_lastrun -= sleep;
1611 bitmap_daemon_work(mddev);
1612 bitmap_update_sb(bitmap);
1613 }
1614
1615 /*
1616 * free memory that was allocated
1617 */
1618 static void bitmap_free(struct bitmap *bitmap)
1619 {
1620 unsigned long k, pages;
1621 struct bitmap_page *bp;
1622
1623 if (!bitmap) /* there was no bitmap */
1624 return;
1625
1626 /* release the bitmap file and kill the daemon */
1627 bitmap_file_put(bitmap);
1628
1629 bp = bitmap->bp;
1630 pages = bitmap->pages;
1631
1632 /* free all allocated memory */
1633
1634 if (bp) /* deallocate the page memory */
1635 for (k = 0; k < pages; k++)
1636 if (bp[k].map && !bp[k].hijacked)
1637 kfree(bp[k].map);
1638 kfree(bp);
1639 kfree(bitmap);
1640 }
1641
1642 void bitmap_destroy(struct mddev *mddev)
1643 {
1644 struct bitmap *bitmap = mddev->bitmap;
1645
1646 if (!bitmap) /* there was no bitmap */
1647 return;
1648
1649 mutex_lock(&mddev->bitmap_info.mutex);
1650 mddev->bitmap = NULL; /* disconnect from the md device */
1651 mutex_unlock(&mddev->bitmap_info.mutex);
1652 if (mddev->thread)
1653 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1654
1655 if (bitmap->sysfs_can_clear)
1656 sysfs_put(bitmap->sysfs_can_clear);
1657
1658 bitmap_free(bitmap);
1659 }
1660
1661 /*
1662 * initialize the bitmap structure
1663 * if this returns an error, bitmap_destroy must be called to do clean up
1664 */
1665 int bitmap_create(struct mddev *mddev)
1666 {
1667 struct bitmap *bitmap;
1668 sector_t blocks = mddev->resync_max_sectors;
1669 unsigned long chunks;
1670 unsigned long pages;
1671 struct file *file = mddev->bitmap_info.file;
1672 int err;
1673 struct sysfs_dirent *bm = NULL;
1674
1675 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1676
1677 BUG_ON(file && mddev->bitmap_info.offset);
1678
1679 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1680 if (!bitmap)
1681 return -ENOMEM;
1682
1683 spin_lock_init(&bitmap->lock);
1684 atomic_set(&bitmap->pending_writes, 0);
1685 init_waitqueue_head(&bitmap->write_wait);
1686 init_waitqueue_head(&bitmap->overflow_wait);
1687 init_waitqueue_head(&bitmap->behind_wait);
1688
1689 bitmap->mddev = mddev;
1690
1691 if (mddev->kobj.sd)
1692 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1693 if (bm) {
1694 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1695 sysfs_put(bm);
1696 } else
1697 bitmap->sysfs_can_clear = NULL;
1698
1699 bitmap->storage.file = file;
1700 if (file) {
1701 get_file(file);
1702 /* As future accesses to this file will use bmap,
1703 * and bypass the page cache, we must sync the file
1704 * first.
1705 */
1706 vfs_fsync(file, 1);
1707 }
1708 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1709 if (!mddev->bitmap_info.external) {
1710 /*
1711 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1712 * instructing us to create a new on-disk bitmap instance.
1713 */
1714 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1715 err = bitmap_new_disk_sb(bitmap);
1716 else
1717 err = bitmap_read_sb(bitmap);
1718 } else {
1719 err = 0;
1720 if (mddev->bitmap_info.chunksize == 0 ||
1721 mddev->bitmap_info.daemon_sleep == 0)
1722 /* chunksize and time_base need to be
1723 * set first. */
1724 err = -EINVAL;
1725 }
1726 if (err)
1727 goto error;
1728
1729 bitmap->daemon_lastrun = jiffies;
1730 bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize)
1731 - BITMAP_BLOCK_SHIFT);
1732
1733 chunks = (blocks + (1 << bitmap->chunkshift) - 1) >>
1734 bitmap->chunkshift;
1735 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1736
1737 BUG_ON(!pages);
1738
1739 bitmap->chunks = chunks;
1740 bitmap->pages = pages;
1741 bitmap->missing_pages = pages;
1742
1743 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1744
1745 err = -ENOMEM;
1746 if (!bitmap->bp)
1747 goto error;
1748
1749 if (file || mddev->bitmap_info.offset) {
1750 err = bitmap_storage_alloc(&bitmap->storage, bitmap->chunks,
1751 !mddev->bitmap_info.external);
1752 if (err)
1753 goto error;
1754 }
1755 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1756 pages, bmname(bitmap));
1757
1758 mddev->bitmap = bitmap;
1759
1760
1761 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1762
1763 error:
1764 bitmap_free(bitmap);
1765 return err;
1766 }
1767
1768 int bitmap_load(struct mddev *mddev)
1769 {
1770 int err = 0;
1771 sector_t start = 0;
1772 sector_t sector = 0;
1773 struct bitmap *bitmap = mddev->bitmap;
1774
1775 if (!bitmap)
1776 goto out;
1777
1778 /* Clear out old bitmap info first: Either there is none, or we
1779 * are resuming after someone else has possibly changed things,
1780 * so we should forget old cached info.
1781 * All chunks should be clean, but some might need_sync.
1782 */
1783 while (sector < mddev->resync_max_sectors) {
1784 sector_t blocks;
1785 bitmap_start_sync(bitmap, sector, &blocks, 0);
1786 sector += blocks;
1787 }
1788 bitmap_close_sync(bitmap);
1789
1790 if (mddev->degraded == 0
1791 || bitmap->events_cleared == mddev->events)
1792 /* no need to keep dirty bits to optimise a
1793 * re-add of a missing device */
1794 start = mddev->recovery_cp;
1795
1796 mutex_lock(&mddev->bitmap_info.mutex);
1797 err = bitmap_init_from_disk(bitmap, start);
1798 mutex_unlock(&mddev->bitmap_info.mutex);
1799
1800 if (err)
1801 goto out;
1802 bitmap->flags &= ~BITMAP_STALE;
1803
1804 /* Kick recovery in case any bits were set */
1805 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1806
1807 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1808 md_wakeup_thread(mddev->thread);
1809
1810 bitmap_update_sb(bitmap);
1811
1812 if (bitmap->flags & BITMAP_WRITE_ERROR)
1813 err = -EIO;
1814 out:
1815 return err;
1816 }
1817 EXPORT_SYMBOL_GPL(bitmap_load);
1818
1819 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1820 {
1821 unsigned long chunk_kb;
1822 unsigned long flags;
1823
1824 if (!bitmap)
1825 return;
1826
1827 spin_lock_irqsave(&bitmap->lock, flags);
1828 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1829 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1830 "%lu%s chunk",
1831 bitmap->pages - bitmap->missing_pages,
1832 bitmap->pages,
1833 (bitmap->pages - bitmap->missing_pages)
1834 << (PAGE_SHIFT - 10),
1835 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1836 chunk_kb ? "KB" : "B");
1837 if (bitmap->storage.file) {
1838 seq_printf(seq, ", file: ");
1839 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1840 }
1841
1842 seq_printf(seq, "\n");
1843 spin_unlock_irqrestore(&bitmap->lock, flags);
1844 }
1845
1846 static ssize_t
1847 location_show(struct mddev *mddev, char *page)
1848 {
1849 ssize_t len;
1850 if (mddev->bitmap_info.file)
1851 len = sprintf(page, "file");
1852 else if (mddev->bitmap_info.offset)
1853 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1854 else
1855 len = sprintf(page, "none");
1856 len += sprintf(page+len, "\n");
1857 return len;
1858 }
1859
1860 static ssize_t
1861 location_store(struct mddev *mddev, const char *buf, size_t len)
1862 {
1863
1864 if (mddev->pers) {
1865 if (!mddev->pers->quiesce)
1866 return -EBUSY;
1867 if (mddev->recovery || mddev->sync_thread)
1868 return -EBUSY;
1869 }
1870
1871 if (mddev->bitmap || mddev->bitmap_info.file ||
1872 mddev->bitmap_info.offset) {
1873 /* bitmap already configured. Only option is to clear it */
1874 if (strncmp(buf, "none", 4) != 0)
1875 return -EBUSY;
1876 if (mddev->pers) {
1877 mddev->pers->quiesce(mddev, 1);
1878 bitmap_destroy(mddev);
1879 mddev->pers->quiesce(mddev, 0);
1880 }
1881 mddev->bitmap_info.offset = 0;
1882 if (mddev->bitmap_info.file) {
1883 struct file *f = mddev->bitmap_info.file;
1884 mddev->bitmap_info.file = NULL;
1885 restore_bitmap_write_access(f);
1886 fput(f);
1887 }
1888 } else {
1889 /* No bitmap, OK to set a location */
1890 long long offset;
1891 if (strncmp(buf, "none", 4) == 0)
1892 /* nothing to be done */;
1893 else if (strncmp(buf, "file:", 5) == 0) {
1894 /* Not supported yet */
1895 return -EINVAL;
1896 } else {
1897 int rv;
1898 if (buf[0] == '+')
1899 rv = strict_strtoll(buf+1, 10, &offset);
1900 else
1901 rv = strict_strtoll(buf, 10, &offset);
1902 if (rv)
1903 return rv;
1904 if (offset == 0)
1905 return -EINVAL;
1906 if (mddev->bitmap_info.external == 0 &&
1907 mddev->major_version == 0 &&
1908 offset != mddev->bitmap_info.default_offset)
1909 return -EINVAL;
1910 mddev->bitmap_info.offset = offset;
1911 if (mddev->pers) {
1912 mddev->pers->quiesce(mddev, 1);
1913 rv = bitmap_create(mddev);
1914 if (!rv)
1915 rv = bitmap_load(mddev);
1916 if (rv) {
1917 bitmap_destroy(mddev);
1918 mddev->bitmap_info.offset = 0;
1919 }
1920 mddev->pers->quiesce(mddev, 0);
1921 if (rv)
1922 return rv;
1923 }
1924 }
1925 }
1926 if (!mddev->external) {
1927 /* Ensure new bitmap info is stored in
1928 * metadata promptly.
1929 */
1930 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1931 md_wakeup_thread(mddev->thread);
1932 }
1933 return len;
1934 }
1935
1936 static struct md_sysfs_entry bitmap_location =
1937 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1938
1939 /* 'bitmap/space' is the space available at 'location' for the
1940 * bitmap. This allows the kernel to know when it is safe to
1941 * resize the bitmap to match a resized array.
1942 */
1943 static ssize_t
1944 space_show(struct mddev *mddev, char *page)
1945 {
1946 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
1947 }
1948
1949 static ssize_t
1950 space_store(struct mddev *mddev, const char *buf, size_t len)
1951 {
1952 unsigned long sectors;
1953 int rv;
1954
1955 rv = kstrtoul(buf, 10, &sectors);
1956 if (rv)
1957 return rv;
1958
1959 if (sectors == 0)
1960 return -EINVAL;
1961
1962 if (mddev->bitmap &&
1963 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
1964 return -EFBIG; /* Bitmap is too big for this small space */
1965
1966 /* could make sure it isn't too big, but that isn't really
1967 * needed - user-space should be careful.
1968 */
1969 mddev->bitmap_info.space = sectors;
1970 return len;
1971 }
1972
1973 static struct md_sysfs_entry bitmap_space =
1974 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
1975
1976 static ssize_t
1977 timeout_show(struct mddev *mddev, char *page)
1978 {
1979 ssize_t len;
1980 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1981 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1982
1983 len = sprintf(page, "%lu", secs);
1984 if (jifs)
1985 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1986 len += sprintf(page+len, "\n");
1987 return len;
1988 }
1989
1990 static ssize_t
1991 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1992 {
1993 /* timeout can be set at any time */
1994 unsigned long timeout;
1995 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1996 if (rv)
1997 return rv;
1998
1999 /* just to make sure we don't overflow... */
2000 if (timeout >= LONG_MAX / HZ)
2001 return -EINVAL;
2002
2003 timeout = timeout * HZ / 10000;
2004
2005 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2006 timeout = MAX_SCHEDULE_TIMEOUT-1;
2007 if (timeout < 1)
2008 timeout = 1;
2009 mddev->bitmap_info.daemon_sleep = timeout;
2010 if (mddev->thread) {
2011 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2012 * the bitmap is all clean and we don't need to
2013 * adjust the timeout right now
2014 */
2015 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2016 mddev->thread->timeout = timeout;
2017 md_wakeup_thread(mddev->thread);
2018 }
2019 }
2020 return len;
2021 }
2022
2023 static struct md_sysfs_entry bitmap_timeout =
2024 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2025
2026 static ssize_t
2027 backlog_show(struct mddev *mddev, char *page)
2028 {
2029 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2030 }
2031
2032 static ssize_t
2033 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2034 {
2035 unsigned long backlog;
2036 int rv = strict_strtoul(buf, 10, &backlog);
2037 if (rv)
2038 return rv;
2039 if (backlog > COUNTER_MAX)
2040 return -EINVAL;
2041 mddev->bitmap_info.max_write_behind = backlog;
2042 return len;
2043 }
2044
2045 static struct md_sysfs_entry bitmap_backlog =
2046 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2047
2048 static ssize_t
2049 chunksize_show(struct mddev *mddev, char *page)
2050 {
2051 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2052 }
2053
2054 static ssize_t
2055 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2056 {
2057 /* Can only be changed when no bitmap is active */
2058 int rv;
2059 unsigned long csize;
2060 if (mddev->bitmap)
2061 return -EBUSY;
2062 rv = strict_strtoul(buf, 10, &csize);
2063 if (rv)
2064 return rv;
2065 if (csize < 512 ||
2066 !is_power_of_2(csize))
2067 return -EINVAL;
2068 mddev->bitmap_info.chunksize = csize;
2069 return len;
2070 }
2071
2072 static struct md_sysfs_entry bitmap_chunksize =
2073 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2074
2075 static ssize_t metadata_show(struct mddev *mddev, char *page)
2076 {
2077 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2078 ? "external" : "internal"));
2079 }
2080
2081 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2082 {
2083 if (mddev->bitmap ||
2084 mddev->bitmap_info.file ||
2085 mddev->bitmap_info.offset)
2086 return -EBUSY;
2087 if (strncmp(buf, "external", 8) == 0)
2088 mddev->bitmap_info.external = 1;
2089 else if (strncmp(buf, "internal", 8) == 0)
2090 mddev->bitmap_info.external = 0;
2091 else
2092 return -EINVAL;
2093 return len;
2094 }
2095
2096 static struct md_sysfs_entry bitmap_metadata =
2097 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2098
2099 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2100 {
2101 int len;
2102 if (mddev->bitmap)
2103 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2104 "false" : "true"));
2105 else
2106 len = sprintf(page, "\n");
2107 return len;
2108 }
2109
2110 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2111 {
2112 if (mddev->bitmap == NULL)
2113 return -ENOENT;
2114 if (strncmp(buf, "false", 5) == 0)
2115 mddev->bitmap->need_sync = 1;
2116 else if (strncmp(buf, "true", 4) == 0) {
2117 if (mddev->degraded)
2118 return -EBUSY;
2119 mddev->bitmap->need_sync = 0;
2120 } else
2121 return -EINVAL;
2122 return len;
2123 }
2124
2125 static struct md_sysfs_entry bitmap_can_clear =
2126 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2127
2128 static ssize_t
2129 behind_writes_used_show(struct mddev *mddev, char *page)
2130 {
2131 if (mddev->bitmap == NULL)
2132 return sprintf(page, "0\n");
2133 return sprintf(page, "%lu\n",
2134 mddev->bitmap->behind_writes_used);
2135 }
2136
2137 static ssize_t
2138 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2139 {
2140 if (mddev->bitmap)
2141 mddev->bitmap->behind_writes_used = 0;
2142 return len;
2143 }
2144
2145 static struct md_sysfs_entry max_backlog_used =
2146 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2147 behind_writes_used_show, behind_writes_used_reset);
2148
2149 static struct attribute *md_bitmap_attrs[] = {
2150 &bitmap_location.attr,
2151 &bitmap_space.attr,
2152 &bitmap_timeout.attr,
2153 &bitmap_backlog.attr,
2154 &bitmap_chunksize.attr,
2155 &bitmap_metadata.attr,
2156 &bitmap_can_clear.attr,
2157 &max_backlog_used.attr,
2158 NULL
2159 };
2160 struct attribute_group md_bitmap_group = {
2161 .name = "bitmap",
2162 .attrs = md_bitmap_attrs,
2163 };
2164
This page took 0.075726 seconds and 6 git commands to generate.