md/bitmap: create a 'struct bitmap_counts' substructure of 'struct bitmap'
[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_counts *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("md/bitmap: map page allocation failed, hijacking\n");
80 /* failed - set the hijacked flag so that we can use the
81 * pointer as a counter */
82 if (!bitmap->bp[page].map)
83 bitmap->bp[page].hijacked = 1;
84 } else if (bitmap->bp[page].map ||
85 bitmap->bp[page].hijacked) {
86 /* somebody beat us to getting the page */
87 kfree(mappage);
88 return 0;
89 } else {
90
91 /* no page was in place and we have one, so install it */
92
93 bitmap->bp[page].map = mappage;
94 bitmap->missing_pages--;
95 }
96 return 0;
97 }
98
99 /* if page is completely empty, put it back on the free list, or dealloc it */
100 /* if page was hijacked, unmark the flag so it might get alloced next time */
101 /* Note: lock should be held when calling this */
102 static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
103 {
104 char *ptr;
105
106 if (bitmap->bp[page].count) /* page is still busy */
107 return;
108
109 /* page is no longer in use, it can be released */
110
111 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
112 bitmap->bp[page].hijacked = 0;
113 bitmap->bp[page].map = NULL;
114 } else {
115 /* normal case, free the page */
116 ptr = bitmap->bp[page].map;
117 bitmap->bp[page].map = NULL;
118 bitmap->missing_pages++;
119 kfree(ptr);
120 }
121 }
122
123 /*
124 * bitmap file handling - read and write the bitmap file and its superblock
125 */
126
127 /*
128 * basic page I/O operations
129 */
130
131 /* IO operations when bitmap is stored near all superblocks */
132 static int read_sb_page(struct mddev *mddev, loff_t offset,
133 struct page *page,
134 unsigned long index, int size)
135 {
136 /* choose a good rdev and read the page from there */
137
138 struct md_rdev *rdev;
139 sector_t target;
140
141 rdev_for_each(rdev, mddev) {
142 if (! test_bit(In_sync, &rdev->flags)
143 || test_bit(Faulty, &rdev->flags))
144 continue;
145
146 target = offset + index * (PAGE_SIZE/512);
147
148 if (sync_page_io(rdev, target,
149 roundup(size, bdev_logical_block_size(rdev->bdev)),
150 page, READ, true)) {
151 page->index = index;
152 return 0;
153 }
154 }
155 return -EIO;
156 }
157
158 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
159 {
160 /* Iterate the disks of an mddev, using rcu to protect access to the
161 * linked list, and raising the refcount of devices we return to ensure
162 * they don't disappear while in use.
163 * As devices are only added or removed when raid_disk is < 0 and
164 * nr_pending is 0 and In_sync is clear, the entries we return will
165 * still be in the same position on the list when we re-enter
166 * list_for_each_continue_rcu.
167 */
168 struct list_head *pos;
169 rcu_read_lock();
170 if (rdev == NULL)
171 /* start at the beginning */
172 pos = &mddev->disks;
173 else {
174 /* release the previous rdev and start from there. */
175 rdev_dec_pending(rdev, mddev);
176 pos = &rdev->same_set;
177 }
178 list_for_each_continue_rcu(pos, &mddev->disks) {
179 rdev = list_entry(pos, struct md_rdev, same_set);
180 if (rdev->raid_disk >= 0 &&
181 !test_bit(Faulty, &rdev->flags)) {
182 /* this is a usable devices */
183 atomic_inc(&rdev->nr_pending);
184 rcu_read_unlock();
185 return rdev;
186 }
187 }
188 rcu_read_unlock();
189 return NULL;
190 }
191
192 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
193 {
194 struct md_rdev *rdev = NULL;
195 struct block_device *bdev;
196 struct mddev *mddev = bitmap->mddev;
197 struct bitmap_storage *store = &bitmap->storage;
198
199 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
200 int size = PAGE_SIZE;
201 loff_t offset = mddev->bitmap_info.offset;
202
203 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
204
205 if (page->index == store->file_pages-1) {
206 int last_page_size = store->bytes & (PAGE_SIZE-1);
207 if (last_page_size == 0)
208 last_page_size = PAGE_SIZE;
209 size = roundup(last_page_size,
210 bdev_logical_block_size(bdev));
211 }
212 /* Just make sure we aren't corrupting data or
213 * metadata
214 */
215 if (mddev->external) {
216 /* Bitmap could be anywhere. */
217 if (rdev->sb_start + offset + (page->index
218 * (PAGE_SIZE/512))
219 > rdev->data_offset
220 &&
221 rdev->sb_start + offset
222 < (rdev->data_offset + mddev->dev_sectors
223 + (PAGE_SIZE/512)))
224 goto bad_alignment;
225 } else if (offset < 0) {
226 /* DATA BITMAP METADATA */
227 if (offset
228 + (long)(page->index * (PAGE_SIZE/512))
229 + size/512 > 0)
230 /* bitmap runs in to metadata */
231 goto bad_alignment;
232 if (rdev->data_offset + mddev->dev_sectors
233 > rdev->sb_start + offset)
234 /* data runs in to bitmap */
235 goto bad_alignment;
236 } else if (rdev->sb_start < rdev->data_offset) {
237 /* METADATA BITMAP DATA */
238 if (rdev->sb_start
239 + offset
240 + page->index*(PAGE_SIZE/512) + size/512
241 > rdev->data_offset)
242 /* bitmap runs in to data */
243 goto bad_alignment;
244 } else {
245 /* DATA METADATA BITMAP - no problems */
246 }
247 md_super_write(mddev, rdev,
248 rdev->sb_start + offset
249 + page->index * (PAGE_SIZE/512),
250 size,
251 page);
252 }
253
254 if (wait)
255 md_super_wait(mddev);
256 return 0;
257
258 bad_alignment:
259 return -EINVAL;
260 }
261
262 static void bitmap_file_kick(struct bitmap *bitmap);
263 /*
264 * write out a page to a file
265 */
266 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
267 {
268 struct buffer_head *bh;
269
270 if (bitmap->storage.file == NULL) {
271 switch (write_sb_page(bitmap, page, wait)) {
272 case -EINVAL:
273 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
274 }
275 } else {
276
277 bh = page_buffers(page);
278
279 while (bh && bh->b_blocknr) {
280 atomic_inc(&bitmap->pending_writes);
281 set_buffer_locked(bh);
282 set_buffer_mapped(bh);
283 submit_bh(WRITE | REQ_SYNC, bh);
284 bh = bh->b_this_page;
285 }
286
287 if (wait)
288 wait_event(bitmap->write_wait,
289 atomic_read(&bitmap->pending_writes)==0);
290 }
291 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
292 bitmap_file_kick(bitmap);
293 }
294
295 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
296 {
297 struct bitmap *bitmap = bh->b_private;
298
299 if (!uptodate)
300 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
301 if (atomic_dec_and_test(&bitmap->pending_writes))
302 wake_up(&bitmap->write_wait);
303 }
304
305 /* copied from buffer.c */
306 static void
307 __clear_page_buffers(struct page *page)
308 {
309 ClearPagePrivate(page);
310 set_page_private(page, 0);
311 page_cache_release(page);
312 }
313 static void free_buffers(struct page *page)
314 {
315 struct buffer_head *bh;
316
317 if (!PagePrivate(page))
318 return;
319
320 bh = page_buffers(page);
321 while (bh) {
322 struct buffer_head *next = bh->b_this_page;
323 free_buffer_head(bh);
324 bh = next;
325 }
326 __clear_page_buffers(page);
327 put_page(page);
328 }
329
330 /* read a page from a file.
331 * We both read the page, and attach buffers to the page to record the
332 * address of each block (using bmap). These addresses will be used
333 * to write the block later, completely bypassing the filesystem.
334 * This usage is similar to how swap files are handled, and allows us
335 * to write to a file with no concerns of memory allocation failing.
336 */
337 static int read_page(struct file *file, unsigned long index,
338 struct bitmap *bitmap,
339 unsigned long count,
340 struct page *page)
341 {
342 int ret = 0;
343 struct inode *inode = file->f_path.dentry->d_inode;
344 struct buffer_head *bh;
345 sector_t block;
346
347 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
348 (unsigned long long)index << PAGE_SHIFT);
349
350 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
351 if (!bh) {
352 ret = -ENOMEM;
353 goto out;
354 }
355 attach_page_buffers(page, bh);
356 block = index << (PAGE_SHIFT - inode->i_blkbits);
357 while (bh) {
358 if (count == 0)
359 bh->b_blocknr = 0;
360 else {
361 bh->b_blocknr = bmap(inode, block);
362 if (bh->b_blocknr == 0) {
363 /* Cannot use this file! */
364 ret = -EINVAL;
365 goto out;
366 }
367 bh->b_bdev = inode->i_sb->s_bdev;
368 if (count < (1<<inode->i_blkbits))
369 count = 0;
370 else
371 count -= (1<<inode->i_blkbits);
372
373 bh->b_end_io = end_bitmap_write;
374 bh->b_private = bitmap;
375 atomic_inc(&bitmap->pending_writes);
376 set_buffer_locked(bh);
377 set_buffer_mapped(bh);
378 submit_bh(READ, bh);
379 }
380 block++;
381 bh = bh->b_this_page;
382 }
383 page->index = index;
384
385 wait_event(bitmap->write_wait,
386 atomic_read(&bitmap->pending_writes)==0);
387 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
388 ret = -EIO;
389 out:
390 if (ret)
391 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
392 (int)PAGE_SIZE,
393 (unsigned long long)index << PAGE_SHIFT,
394 ret);
395 return ret;
396 }
397
398 /*
399 * bitmap file superblock operations
400 */
401
402 /* update the event counter and sync the superblock to disk */
403 void bitmap_update_sb(struct bitmap *bitmap)
404 {
405 bitmap_super_t *sb;
406
407 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
408 return;
409 if (bitmap->mddev->bitmap_info.external)
410 return;
411 if (!bitmap->storage.sb_page) /* no superblock */
412 return;
413 sb = kmap_atomic(bitmap->storage.sb_page);
414 sb->events = cpu_to_le64(bitmap->mddev->events);
415 if (bitmap->mddev->events < bitmap->events_cleared)
416 /* rocking back to read-only */
417 bitmap->events_cleared = bitmap->mddev->events;
418 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
419 sb->state = cpu_to_le32(bitmap->flags);
420 /* Just in case these have been changed via sysfs: */
421 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
422 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
423 kunmap_atomic(sb);
424 write_page(bitmap, bitmap->storage.sb_page, 1);
425 }
426
427 /* print out the bitmap file superblock */
428 void bitmap_print_sb(struct bitmap *bitmap)
429 {
430 bitmap_super_t *sb;
431
432 if (!bitmap || !bitmap->storage.sb_page)
433 return;
434 sb = kmap_atomic(bitmap->storage.sb_page);
435 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
436 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
437 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
438 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
439 *(__u32 *)(sb->uuid+0),
440 *(__u32 *)(sb->uuid+4),
441 *(__u32 *)(sb->uuid+8),
442 *(__u32 *)(sb->uuid+12));
443 printk(KERN_DEBUG " events: %llu\n",
444 (unsigned long long) le64_to_cpu(sb->events));
445 printk(KERN_DEBUG "events cleared: %llu\n",
446 (unsigned long long) le64_to_cpu(sb->events_cleared));
447 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
448 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
449 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
450 printk(KERN_DEBUG " sync size: %llu KB\n",
451 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
452 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
453 kunmap_atomic(sb);
454 }
455
456 /*
457 * bitmap_new_disk_sb
458 * @bitmap
459 *
460 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
461 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
462 * This function verifies 'bitmap_info' and populates the on-disk bitmap
463 * structure, which is to be written to disk.
464 *
465 * Returns: 0 on success, -Exxx on error
466 */
467 static int bitmap_new_disk_sb(struct bitmap *bitmap)
468 {
469 bitmap_super_t *sb;
470 unsigned long chunksize, daemon_sleep, write_behind;
471 int err = -EINVAL;
472
473 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
474 if (IS_ERR(bitmap->storage.sb_page)) {
475 err = PTR_ERR(bitmap->storage.sb_page);
476 bitmap->storage.sb_page = NULL;
477 return err;
478 }
479 bitmap->storage.sb_page->index = 0;
480
481 sb = kmap_atomic(bitmap->storage.sb_page);
482
483 sb->magic = cpu_to_le32(BITMAP_MAGIC);
484 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
485
486 chunksize = bitmap->mddev->bitmap_info.chunksize;
487 BUG_ON(!chunksize);
488 if (!is_power_of_2(chunksize)) {
489 kunmap_atomic(sb);
490 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
491 return -EINVAL;
492 }
493 sb->chunksize = cpu_to_le32(chunksize);
494
495 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
496 if (!daemon_sleep ||
497 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
498 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
499 daemon_sleep = 5 * HZ;
500 }
501 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
502 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
503
504 /*
505 * FIXME: write_behind for RAID1. If not specified, what
506 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
507 */
508 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
509 if (write_behind > COUNTER_MAX)
510 write_behind = COUNTER_MAX / 2;
511 sb->write_behind = cpu_to_le32(write_behind);
512 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
513
514 /* keep the array size field of the bitmap superblock up to date */
515 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
516
517 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
518
519 set_bit(BITMAP_STALE, &bitmap->flags);
520 sb->state = cpu_to_le32(bitmap->flags);
521 bitmap->events_cleared = bitmap->mddev->events;
522 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
523
524 kunmap_atomic(sb);
525
526 return 0;
527 }
528
529 /* read the superblock from the bitmap file and initialize some bitmap fields */
530 static int bitmap_read_sb(struct bitmap *bitmap)
531 {
532 char *reason = NULL;
533 bitmap_super_t *sb;
534 unsigned long chunksize, daemon_sleep, write_behind;
535 unsigned long long events;
536 int err = -EINVAL;
537 struct page *sb_page;
538
539 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
540 chunksize = 128 * 1024 * 1024;
541 daemon_sleep = 5 * HZ;
542 write_behind = 0;
543 set_bit(BITMAP_STALE, &bitmap->flags);
544 err = 0;
545 goto out_no_sb;
546 }
547 /* page 0 is the superblock, read it... */
548 sb_page = alloc_page(GFP_KERNEL);
549 if (!sb_page)
550 return -ENOMEM;
551 bitmap->storage.sb_page = sb_page;
552
553 if (bitmap->storage.file) {
554 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
555 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
556
557 err = read_page(bitmap->storage.file, 0,
558 bitmap, bytes, sb_page);
559 } else {
560 err = read_sb_page(bitmap->mddev,
561 bitmap->mddev->bitmap_info.offset,
562 sb_page,
563 0, sizeof(bitmap_super_t));
564 }
565 if (err)
566 return err;
567
568 sb = kmap_atomic(sb_page);
569
570 chunksize = le32_to_cpu(sb->chunksize);
571 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
572 write_behind = le32_to_cpu(sb->write_behind);
573
574 /* verify that the bitmap-specific fields are valid */
575 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
576 reason = "bad magic";
577 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
578 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
579 reason = "unrecognized superblock version";
580 else if (chunksize < 512)
581 reason = "bitmap chunksize too small";
582 else if (!is_power_of_2(chunksize))
583 reason = "bitmap chunksize not a power of 2";
584 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
585 reason = "daemon sleep period out of range";
586 else if (write_behind > COUNTER_MAX)
587 reason = "write-behind limit out of range (0 - 16383)";
588 if (reason) {
589 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
590 bmname(bitmap), reason);
591 goto out;
592 }
593
594 /* keep the array size field of the bitmap superblock up to date */
595 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
596
597 if (bitmap->mddev->persistent) {
598 /*
599 * We have a persistent array superblock, so compare the
600 * bitmap's UUID and event counter to the mddev's
601 */
602 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
603 printk(KERN_INFO
604 "%s: bitmap superblock UUID mismatch\n",
605 bmname(bitmap));
606 goto out;
607 }
608 events = le64_to_cpu(sb->events);
609 if (events < bitmap->mddev->events) {
610 printk(KERN_INFO
611 "%s: bitmap file is out of date (%llu < %llu) "
612 "-- forcing full recovery\n",
613 bmname(bitmap), events,
614 (unsigned long long) bitmap->mddev->events);
615 set_bit(BITMAP_STALE, &bitmap->flags);
616 }
617 }
618
619 /* assign fields using values from superblock */
620 bitmap->flags |= le32_to_cpu(sb->state);
621 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
622 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
623 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
624 err = 0;
625 out:
626 kunmap_atomic(sb);
627 out_no_sb:
628 if (test_bit(BITMAP_STALE, &bitmap->flags))
629 bitmap->events_cleared = bitmap->mddev->events;
630 bitmap->mddev->bitmap_info.chunksize = chunksize;
631 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
632 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
633 if (err)
634 bitmap_print_sb(bitmap);
635 return err;
636 }
637
638 /*
639 * general bitmap file operations
640 */
641
642 /*
643 * on-disk bitmap:
644 *
645 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
646 * file a page at a time. There's a superblock at the start of the file.
647 */
648 /* calculate the index of the page that contains this bit */
649 static inline unsigned long file_page_index(struct bitmap_storage *store,
650 unsigned long chunk)
651 {
652 if (store->sb_page)
653 chunk += sizeof(bitmap_super_t) << 3;
654 return chunk >> PAGE_BIT_SHIFT;
655 }
656
657 /* calculate the (bit) offset of this bit within a page */
658 static inline unsigned long file_page_offset(struct bitmap_storage *store,
659 unsigned long chunk)
660 {
661 if (store->sb_page)
662 chunk += sizeof(bitmap_super_t) << 3;
663 return chunk & (PAGE_BITS - 1);
664 }
665
666 /*
667 * return a pointer to the page in the filemap that contains the given bit
668 *
669 * this lookup is complicated by the fact that the bitmap sb might be exactly
670 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
671 * 0 or page 1
672 */
673 static inline struct page *filemap_get_page(struct bitmap_storage *store,
674 unsigned long chunk)
675 {
676 if (file_page_index(store, chunk) >= store->file_pages)
677 return NULL;
678 return store->filemap[file_page_index(store, chunk)
679 - file_page_index(store, 0)];
680 }
681
682 static int bitmap_storage_alloc(struct bitmap_storage *store,
683 unsigned long chunks, int with_super)
684 {
685 int pnum;
686 unsigned long num_pages;
687 unsigned long bytes;
688
689 bytes = DIV_ROUND_UP(chunks, 8);
690 if (with_super)
691 bytes += sizeof(bitmap_super_t);
692
693 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
694
695 store->filemap = kmalloc(sizeof(struct page *)
696 * num_pages, GFP_KERNEL);
697 if (!store->filemap)
698 return -ENOMEM;
699
700 if (with_super && !store->sb_page) {
701 store->sb_page = alloc_page(GFP_KERNEL);
702 if (store->sb_page == NULL)
703 return -ENOMEM;
704 store->sb_page->index = 0;
705 }
706 pnum = 0;
707 if (store->sb_page) {
708 store->filemap[0] = store->sb_page;
709 pnum = 1;
710 }
711 for ( ; pnum < num_pages; pnum++) {
712 store->filemap[pnum] = alloc_page(GFP_KERNEL);
713 if (!store->filemap[pnum]) {
714 store->file_pages = pnum;
715 return -ENOMEM;
716 }
717 store->filemap[pnum]->index = pnum;
718 }
719 store->file_pages = pnum;
720
721 /* We need 4 bits per page, rounded up to a multiple
722 * of sizeof(unsigned long) */
723 store->filemap_attr = kzalloc(
724 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
725 GFP_KERNEL);
726 if (!store->filemap_attr)
727 return -ENOMEM;
728
729 store->bytes = bytes;
730
731 return 0;
732 }
733
734 static void bitmap_file_unmap(struct bitmap_storage *store)
735 {
736 struct page **map, *sb_page;
737 int pages;
738 struct file *file;
739
740 file = store->file;
741 map = store->filemap;
742 pages = store->file_pages;
743 sb_page = store->sb_page;
744
745 while (pages--)
746 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
747 free_buffers(map[pages]);
748 kfree(map);
749 kfree(store->filemap_attr);
750
751 if (sb_page)
752 free_buffers(sb_page);
753
754 if (file) {
755 struct inode *inode = file->f_path.dentry->d_inode;
756 invalidate_mapping_pages(inode->i_mapping, 0, -1);
757 fput(file);
758 }
759 }
760
761 /*
762 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
763 * then it is no longer reliable, so we stop using it and we mark the file
764 * as failed in the superblock
765 */
766 static void bitmap_file_kick(struct bitmap *bitmap)
767 {
768 char *path, *ptr = NULL;
769
770 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
771 bitmap_update_sb(bitmap);
772
773 if (bitmap->storage.file) {
774 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
775 if (path)
776 ptr = d_path(&bitmap->storage.file->f_path,
777 path, PAGE_SIZE);
778
779 printk(KERN_ALERT
780 "%s: kicking failed bitmap file %s from array!\n",
781 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
782
783 kfree(path);
784 } else
785 printk(KERN_ALERT
786 "%s: disabling internal bitmap due to errors\n",
787 bmname(bitmap));
788 }
789 }
790
791 enum bitmap_page_attr {
792 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
793 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
794 * i.e. counter is 1 or 2. */
795 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
796 };
797
798 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
799 enum bitmap_page_attr attr)
800 {
801 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
802 }
803
804 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
805 enum bitmap_page_attr attr)
806 {
807 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
808 }
809
810 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
811 enum bitmap_page_attr attr)
812 {
813 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
814 }
815
816 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
817 enum bitmap_page_attr attr)
818 {
819 return test_and_clear_bit((pnum<<2) + attr,
820 bitmap->storage.filemap_attr);
821 }
822 /*
823 * bitmap_file_set_bit -- called before performing a write to the md device
824 * to set (and eventually sync) a particular bit in the bitmap file
825 *
826 * we set the bit immediately, then we record the page number so that
827 * when an unplug occurs, we can flush the dirty pages out to disk
828 */
829 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
830 {
831 unsigned long bit;
832 struct page *page;
833 void *kaddr;
834 unsigned long chunk = block >> bitmap->counts.chunkshift;
835
836 page = filemap_get_page(&bitmap->storage, chunk);
837 if (!page)
838 return;
839 bit = file_page_offset(&bitmap->storage, chunk);
840
841 /* set the bit */
842 kaddr = kmap_atomic(page);
843 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
844 set_bit(bit, kaddr);
845 else
846 test_and_set_bit_le(bit, kaddr);
847 kunmap_atomic(kaddr);
848 pr_debug("set file bit %lu page %lu\n", bit, page->index);
849 /* record page number so it gets flushed to disk when unplug occurs */
850 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
851 }
852
853 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
854 {
855 unsigned long bit;
856 struct page *page;
857 void *paddr;
858 unsigned long chunk = block >> bitmap->counts.chunkshift;
859
860 page = filemap_get_page(&bitmap->storage, chunk);
861 if (!page)
862 return;
863 bit = file_page_offset(&bitmap->storage, chunk);
864 paddr = kmap_atomic(page);
865 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
866 clear_bit(bit, paddr);
867 else
868 test_and_clear_bit_le(bit, paddr);
869 kunmap_atomic(paddr);
870 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
871 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
872 bitmap->allclean = 0;
873 }
874 }
875
876 /* this gets called when the md device is ready to unplug its underlying
877 * (slave) device queues -- before we let any writes go down, we need to
878 * sync the dirty pages of the bitmap file to disk */
879 void bitmap_unplug(struct bitmap *bitmap)
880 {
881 unsigned long i;
882 int dirty, need_write;
883 int wait = 0;
884
885 if (!bitmap || !bitmap->storage.filemap ||
886 test_bit(BITMAP_STALE, &bitmap->flags))
887 return;
888
889 /* look at each page to see if there are any set bits that need to be
890 * flushed out to disk */
891 for (i = 0; i < bitmap->storage.file_pages; i++) {
892 if (!bitmap->storage.filemap)
893 return;
894 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
895 need_write = test_and_clear_page_attr(bitmap, i,
896 BITMAP_PAGE_NEEDWRITE);
897 if (dirty || need_write) {
898 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
899 write_page(bitmap, bitmap->storage.filemap[i], 0);
900 }
901 if (dirty)
902 wait = 1;
903 }
904 if (wait) { /* if any writes were performed, we need to wait on them */
905 if (bitmap->storage.file)
906 wait_event(bitmap->write_wait,
907 atomic_read(&bitmap->pending_writes)==0);
908 else
909 md_super_wait(bitmap->mddev);
910 }
911 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
912 bitmap_file_kick(bitmap);
913 }
914 EXPORT_SYMBOL(bitmap_unplug);
915
916 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
917 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
918 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
919 * memory mapping of the bitmap file
920 * Special cases:
921 * if there's no bitmap file, or if the bitmap file had been
922 * previously kicked from the array, we mark all the bits as
923 * 1's in order to cause a full resync.
924 *
925 * We ignore all bits for sectors that end earlier than 'start'.
926 * This is used when reading an out-of-date bitmap...
927 */
928 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
929 {
930 unsigned long i, chunks, index, oldindex, bit;
931 struct page *page = NULL;
932 unsigned long bit_cnt = 0;
933 struct file *file;
934 unsigned long offset;
935 int outofdate;
936 int ret = -ENOSPC;
937 void *paddr;
938 struct bitmap_storage *store = &bitmap->storage;
939
940 chunks = bitmap->counts.chunks;
941 file = store->file;
942
943 if (!file && !bitmap->mddev->bitmap_info.offset) {
944 /* No permanent bitmap - fill with '1s'. */
945 store->filemap = NULL;
946 store->file_pages = 0;
947 for (i = 0; i < chunks ; i++) {
948 /* if the disk bit is set, set the memory bit */
949 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
950 >= start);
951 bitmap_set_memory_bits(bitmap,
952 (sector_t)i << bitmap->counts.chunkshift,
953 needed);
954 }
955 return 0;
956 }
957
958 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
959 if (outofdate)
960 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
961 "recovery\n", bmname(bitmap));
962
963 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
964 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
965 bmname(bitmap),
966 (unsigned long) i_size_read(file->f_mapping->host),
967 store->bytes);
968 goto err;
969 }
970
971 oldindex = ~0L;
972 offset = 0;
973 if (!bitmap->mddev->bitmap_info.external)
974 offset = sizeof(bitmap_super_t);
975
976 for (i = 0; i < chunks; i++) {
977 int b;
978 index = file_page_index(&bitmap->storage, i);
979 bit = file_page_offset(&bitmap->storage, i);
980 if (index != oldindex) { /* this is a new page, read it in */
981 int count;
982 /* unmap the old page, we're done with it */
983 if (index == store->file_pages-1)
984 count = store->bytes - index * PAGE_SIZE;
985 else
986 count = PAGE_SIZE;
987 page = store->filemap[index];
988 if (file)
989 ret = read_page(file, index, bitmap,
990 count, page);
991 else
992 ret = read_sb_page(
993 bitmap->mddev,
994 bitmap->mddev->bitmap_info.offset,
995 page,
996 index, count);
997
998 if (ret)
999 goto err;
1000
1001 oldindex = index;
1002
1003 if (outofdate) {
1004 /*
1005 * if bitmap is out of date, dirty the
1006 * whole page and write it out
1007 */
1008 paddr = kmap_atomic(page);
1009 memset(paddr + offset, 0xff,
1010 PAGE_SIZE - offset);
1011 kunmap_atomic(paddr);
1012 write_page(bitmap, page, 1);
1013
1014 ret = -EIO;
1015 if (test_bit(BITMAP_WRITE_ERROR,
1016 &bitmap->flags))
1017 goto err;
1018 }
1019 }
1020 paddr = kmap_atomic(page);
1021 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1022 b = test_bit(bit, paddr);
1023 else
1024 b = test_bit_le(bit, paddr);
1025 kunmap_atomic(paddr);
1026 if (b) {
1027 /* if the disk bit is set, set the memory bit */
1028 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1029 >= start);
1030 bitmap_set_memory_bits(bitmap,
1031 (sector_t)i << bitmap->counts.chunkshift,
1032 needed);
1033 bit_cnt++;
1034 }
1035 offset = 0;
1036 }
1037
1038 printk(KERN_INFO "%s: bitmap initialized from disk: "
1039 "read %lu pages, set %lu of %lu bits\n",
1040 bmname(bitmap), store->file_pages,
1041 bit_cnt, chunks);
1042
1043 return 0;
1044
1045 err:
1046 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1047 bmname(bitmap), ret);
1048 return ret;
1049 }
1050
1051 void bitmap_write_all(struct bitmap *bitmap)
1052 {
1053 /* We don't actually write all bitmap blocks here,
1054 * just flag them as needing to be written
1055 */
1056 int i;
1057
1058 if (!bitmap || !bitmap->storage.filemap)
1059 return;
1060 if (bitmap->storage.file)
1061 /* Only one copy, so nothing needed */
1062 return;
1063
1064 for (i = 0; i < bitmap->storage.file_pages; i++)
1065 set_page_attr(bitmap, i,
1066 BITMAP_PAGE_NEEDWRITE);
1067 bitmap->allclean = 0;
1068 }
1069
1070 static void bitmap_count_page(struct bitmap_counts *bitmap,
1071 sector_t offset, int inc)
1072 {
1073 sector_t chunk = offset >> bitmap->chunkshift;
1074 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1075 bitmap->bp[page].count += inc;
1076 bitmap_checkfree(bitmap, page);
1077 }
1078
1079 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1080 {
1081 sector_t chunk = offset >> bitmap->chunkshift;
1082 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1083 struct bitmap_page *bp = &bitmap->bp[page];
1084
1085 if (!bp->pending)
1086 bp->pending = 1;
1087 }
1088
1089 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1090 sector_t offset, sector_t *blocks,
1091 int create);
1092
1093 /*
1094 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1095 * out to disk
1096 */
1097
1098 void bitmap_daemon_work(struct mddev *mddev)
1099 {
1100 struct bitmap *bitmap;
1101 unsigned long j;
1102 unsigned long nextpage;
1103 sector_t blocks;
1104 struct bitmap_counts *counts;
1105
1106 /* Use a mutex to guard daemon_work against
1107 * bitmap_destroy.
1108 */
1109 mutex_lock(&mddev->bitmap_info.mutex);
1110 bitmap = mddev->bitmap;
1111 if (bitmap == NULL) {
1112 mutex_unlock(&mddev->bitmap_info.mutex);
1113 return;
1114 }
1115 if (time_before(jiffies, bitmap->daemon_lastrun
1116 + mddev->bitmap_info.daemon_sleep))
1117 goto done;
1118
1119 bitmap->daemon_lastrun = jiffies;
1120 if (bitmap->allclean) {
1121 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1122 goto done;
1123 }
1124 bitmap->allclean = 1;
1125
1126 /* Any file-page which is PENDING now needs to be written.
1127 * So set NEEDWRITE now, then after we make any last-minute changes
1128 * we will write it.
1129 */
1130 for (j = 0; j < bitmap->storage.file_pages; j++)
1131 if (test_and_clear_page_attr(bitmap, j,
1132 BITMAP_PAGE_PENDING))
1133 set_page_attr(bitmap, j,
1134 BITMAP_PAGE_NEEDWRITE);
1135
1136 if (bitmap->need_sync &&
1137 mddev->bitmap_info.external == 0) {
1138 /* Arrange for superblock update as well as
1139 * other changes */
1140 bitmap_super_t *sb;
1141 bitmap->need_sync = 0;
1142 if (bitmap->storage.filemap) {
1143 sb = kmap_atomic(bitmap->storage.sb_page);
1144 sb->events_cleared =
1145 cpu_to_le64(bitmap->events_cleared);
1146 kunmap_atomic(sb);
1147 set_page_attr(bitmap, 0,
1148 BITMAP_PAGE_NEEDWRITE);
1149 }
1150 }
1151 /* Now look at the bitmap counters and if any are '2' or '1',
1152 * decrement and handle accordingly.
1153 */
1154 counts = &bitmap->counts;
1155 spin_lock_irq(&counts->lock);
1156 nextpage = 0;
1157 for (j = 0; j < counts->chunks; j++) {
1158 bitmap_counter_t *bmc;
1159 sector_t block = (sector_t)j << counts->chunkshift;
1160
1161 if (j == nextpage) {
1162 nextpage += PAGE_COUNTER_RATIO;
1163 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1164 j |= PAGE_COUNTER_MASK;
1165 continue;
1166 }
1167 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1168 }
1169 bmc = bitmap_get_counter(counts,
1170 block,
1171 &blocks, 0);
1172
1173 if (!bmc) {
1174 j |= PAGE_COUNTER_MASK;
1175 continue;
1176 }
1177 if (*bmc == 1 && !bitmap->need_sync) {
1178 /* We can clear the bit */
1179 *bmc = 0;
1180 bitmap_count_page(counts, block, -1);
1181 bitmap_file_clear_bit(bitmap, block);
1182 } else if (*bmc && *bmc <= 2) {
1183 *bmc = 1;
1184 bitmap_set_pending(counts, block);
1185 bitmap->allclean = 0;
1186 }
1187 }
1188 spin_unlock_irq(&counts->lock);
1189
1190 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1191 * DIRTY pages need to be written by bitmap_unplug so it can wait
1192 * for them.
1193 * If we find any DIRTY page we stop there and let bitmap_unplug
1194 * handle all the rest. This is important in the case where
1195 * the first blocking holds the superblock and it has been updated.
1196 * We mustn't write any other blocks before the superblock.
1197 */
1198 for (j = 0;
1199 j < bitmap->storage.file_pages
1200 && !test_bit(BITMAP_STALE, &bitmap->flags);
1201 j++) {
1202
1203 if (test_page_attr(bitmap, j,
1204 BITMAP_PAGE_DIRTY))
1205 /* bitmap_unplug will handle the rest */
1206 break;
1207 if (test_and_clear_page_attr(bitmap, j,
1208 BITMAP_PAGE_NEEDWRITE)) {
1209 write_page(bitmap, bitmap->storage.filemap[j], 0);
1210 }
1211 }
1212
1213 done:
1214 if (bitmap->allclean == 0)
1215 mddev->thread->timeout =
1216 mddev->bitmap_info.daemon_sleep;
1217 mutex_unlock(&mddev->bitmap_info.mutex);
1218 }
1219
1220 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1221 sector_t offset, sector_t *blocks,
1222 int create)
1223 __releases(bitmap->lock)
1224 __acquires(bitmap->lock)
1225 {
1226 /* If 'create', we might release the lock and reclaim it.
1227 * The lock must have been taken with interrupts enabled.
1228 * If !create, we don't release the lock.
1229 */
1230 sector_t chunk = offset >> bitmap->chunkshift;
1231 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1232 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1233 sector_t csize;
1234 int err;
1235
1236 err = bitmap_checkpage(bitmap, page, create);
1237
1238 if (bitmap->bp[page].hijacked ||
1239 bitmap->bp[page].map == NULL)
1240 csize = ((sector_t)1) << (bitmap->chunkshift +
1241 PAGE_COUNTER_SHIFT - 1);
1242 else
1243 csize = ((sector_t)1) << bitmap->chunkshift;
1244 *blocks = csize - (offset & (csize - 1));
1245
1246 if (err < 0)
1247 return NULL;
1248
1249 /* now locked ... */
1250
1251 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1252 /* should we use the first or second counter field
1253 * of the hijacked pointer? */
1254 int hi = (pageoff > PAGE_COUNTER_MASK);
1255 return &((bitmap_counter_t *)
1256 &bitmap->bp[page].map)[hi];
1257 } else /* page is allocated */
1258 return (bitmap_counter_t *)
1259 &(bitmap->bp[page].map[pageoff]);
1260 }
1261
1262 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1263 {
1264 if (!bitmap)
1265 return 0;
1266
1267 if (behind) {
1268 int bw;
1269 atomic_inc(&bitmap->behind_writes);
1270 bw = atomic_read(&bitmap->behind_writes);
1271 if (bw > bitmap->behind_writes_used)
1272 bitmap->behind_writes_used = bw;
1273
1274 pr_debug("inc write-behind count %d/%lu\n",
1275 bw, bitmap->mddev->bitmap_info.max_write_behind);
1276 }
1277
1278 while (sectors) {
1279 sector_t blocks;
1280 bitmap_counter_t *bmc;
1281
1282 spin_lock_irq(&bitmap->counts.lock);
1283 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1284 if (!bmc) {
1285 spin_unlock_irq(&bitmap->counts.lock);
1286 return 0;
1287 }
1288
1289 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1290 DEFINE_WAIT(__wait);
1291 /* note that it is safe to do the prepare_to_wait
1292 * after the test as long as we do it before dropping
1293 * the spinlock.
1294 */
1295 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1296 TASK_UNINTERRUPTIBLE);
1297 spin_unlock_irq(&bitmap->counts.lock);
1298 io_schedule();
1299 finish_wait(&bitmap->overflow_wait, &__wait);
1300 continue;
1301 }
1302
1303 switch (*bmc) {
1304 case 0:
1305 bitmap_file_set_bit(bitmap, offset);
1306 bitmap_count_page(&bitmap->counts, offset, 1);
1307 /* fall through */
1308 case 1:
1309 *bmc = 2;
1310 }
1311
1312 (*bmc)++;
1313
1314 spin_unlock_irq(&bitmap->counts.lock);
1315
1316 offset += blocks;
1317 if (sectors > blocks)
1318 sectors -= blocks;
1319 else
1320 sectors = 0;
1321 }
1322 return 0;
1323 }
1324 EXPORT_SYMBOL(bitmap_startwrite);
1325
1326 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1327 int success, int behind)
1328 {
1329 if (!bitmap)
1330 return;
1331 if (behind) {
1332 if (atomic_dec_and_test(&bitmap->behind_writes))
1333 wake_up(&bitmap->behind_wait);
1334 pr_debug("dec write-behind count %d/%lu\n",
1335 atomic_read(&bitmap->behind_writes),
1336 bitmap->mddev->bitmap_info.max_write_behind);
1337 }
1338
1339 while (sectors) {
1340 sector_t blocks;
1341 unsigned long flags;
1342 bitmap_counter_t *bmc;
1343
1344 spin_lock_irqsave(&bitmap->counts.lock, flags);
1345 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1346 if (!bmc) {
1347 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1348 return;
1349 }
1350
1351 if (success && !bitmap->mddev->degraded &&
1352 bitmap->events_cleared < bitmap->mddev->events) {
1353 bitmap->events_cleared = bitmap->mddev->events;
1354 bitmap->need_sync = 1;
1355 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1356 }
1357
1358 if (!success && !NEEDED(*bmc))
1359 *bmc |= NEEDED_MASK;
1360
1361 if (COUNTER(*bmc) == COUNTER_MAX)
1362 wake_up(&bitmap->overflow_wait);
1363
1364 (*bmc)--;
1365 if (*bmc <= 2) {
1366 bitmap_set_pending(&bitmap->counts, offset);
1367 bitmap->allclean = 0;
1368 }
1369 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1370 offset += blocks;
1371 if (sectors > blocks)
1372 sectors -= blocks;
1373 else
1374 sectors = 0;
1375 }
1376 }
1377 EXPORT_SYMBOL(bitmap_endwrite);
1378
1379 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1380 int degraded)
1381 {
1382 bitmap_counter_t *bmc;
1383 int rv;
1384 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1385 *blocks = 1024;
1386 return 1; /* always resync if no bitmap */
1387 }
1388 spin_lock_irq(&bitmap->counts.lock);
1389 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1390 rv = 0;
1391 if (bmc) {
1392 /* locked */
1393 if (RESYNC(*bmc))
1394 rv = 1;
1395 else if (NEEDED(*bmc)) {
1396 rv = 1;
1397 if (!degraded) { /* don't set/clear bits if degraded */
1398 *bmc |= RESYNC_MASK;
1399 *bmc &= ~NEEDED_MASK;
1400 }
1401 }
1402 }
1403 spin_unlock_irq(&bitmap->counts.lock);
1404 return rv;
1405 }
1406
1407 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1408 int degraded)
1409 {
1410 /* bitmap_start_sync must always report on multiples of whole
1411 * pages, otherwise resync (which is very PAGE_SIZE based) will
1412 * get confused.
1413 * So call __bitmap_start_sync repeatedly (if needed) until
1414 * At least PAGE_SIZE>>9 blocks are covered.
1415 * Return the 'or' of the result.
1416 */
1417 int rv = 0;
1418 sector_t blocks1;
1419
1420 *blocks = 0;
1421 while (*blocks < (PAGE_SIZE>>9)) {
1422 rv |= __bitmap_start_sync(bitmap, offset,
1423 &blocks1, degraded);
1424 offset += blocks1;
1425 *blocks += blocks1;
1426 }
1427 return rv;
1428 }
1429 EXPORT_SYMBOL(bitmap_start_sync);
1430
1431 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1432 {
1433 bitmap_counter_t *bmc;
1434 unsigned long flags;
1435
1436 if (bitmap == NULL) {
1437 *blocks = 1024;
1438 return;
1439 }
1440 spin_lock_irqsave(&bitmap->counts.lock, flags);
1441 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1442 if (bmc == NULL)
1443 goto unlock;
1444 /* locked */
1445 if (RESYNC(*bmc)) {
1446 *bmc &= ~RESYNC_MASK;
1447
1448 if (!NEEDED(*bmc) && aborted)
1449 *bmc |= NEEDED_MASK;
1450 else {
1451 if (*bmc <= 2) {
1452 bitmap_set_pending(&bitmap->counts, offset);
1453 bitmap->allclean = 0;
1454 }
1455 }
1456 }
1457 unlock:
1458 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1459 }
1460 EXPORT_SYMBOL(bitmap_end_sync);
1461
1462 void bitmap_close_sync(struct bitmap *bitmap)
1463 {
1464 /* Sync has finished, and any bitmap chunks that weren't synced
1465 * properly have been aborted. It remains to us to clear the
1466 * RESYNC bit wherever it is still on
1467 */
1468 sector_t sector = 0;
1469 sector_t blocks;
1470 if (!bitmap)
1471 return;
1472 while (sector < bitmap->mddev->resync_max_sectors) {
1473 bitmap_end_sync(bitmap, sector, &blocks, 0);
1474 sector += blocks;
1475 }
1476 }
1477 EXPORT_SYMBOL(bitmap_close_sync);
1478
1479 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1480 {
1481 sector_t s = 0;
1482 sector_t blocks;
1483
1484 if (!bitmap)
1485 return;
1486 if (sector == 0) {
1487 bitmap->last_end_sync = jiffies;
1488 return;
1489 }
1490 if (time_before(jiffies, (bitmap->last_end_sync
1491 + bitmap->mddev->bitmap_info.daemon_sleep)))
1492 return;
1493 wait_event(bitmap->mddev->recovery_wait,
1494 atomic_read(&bitmap->mddev->recovery_active) == 0);
1495
1496 bitmap->mddev->curr_resync_completed = sector;
1497 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1498 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1499 s = 0;
1500 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1501 bitmap_end_sync(bitmap, s, &blocks, 0);
1502 s += blocks;
1503 }
1504 bitmap->last_end_sync = jiffies;
1505 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1506 }
1507 EXPORT_SYMBOL(bitmap_cond_end_sync);
1508
1509 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1510 {
1511 /* For each chunk covered by any of these sectors, set the
1512 * counter to 2 and possibly set resync_needed. They should all
1513 * be 0 at this point
1514 */
1515
1516 sector_t secs;
1517 bitmap_counter_t *bmc;
1518 spin_lock_irq(&bitmap->counts.lock);
1519 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1520 if (!bmc) {
1521 spin_unlock_irq(&bitmap->counts.lock);
1522 return;
1523 }
1524 if (!*bmc) {
1525 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1526 bitmap_count_page(&bitmap->counts, offset, 1);
1527 bitmap_set_pending(&bitmap->counts, offset);
1528 bitmap->allclean = 0;
1529 }
1530 spin_unlock_irq(&bitmap->counts.lock);
1531 }
1532
1533 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1534 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1535 {
1536 unsigned long chunk;
1537
1538 for (chunk = s; chunk <= e; chunk++) {
1539 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1540 bitmap_set_memory_bits(bitmap, sec, 1);
1541 bitmap_file_set_bit(bitmap, sec);
1542 if (sec < bitmap->mddev->recovery_cp)
1543 /* We are asserting that the array is dirty,
1544 * so move the recovery_cp address back so
1545 * that it is obvious that it is dirty
1546 */
1547 bitmap->mddev->recovery_cp = sec;
1548 }
1549 }
1550
1551 /*
1552 * flush out any pending updates
1553 */
1554 void bitmap_flush(struct mddev *mddev)
1555 {
1556 struct bitmap *bitmap = mddev->bitmap;
1557 long sleep;
1558
1559 if (!bitmap) /* there was no bitmap */
1560 return;
1561
1562 /* run the daemon_work three time to ensure everything is flushed
1563 * that can be
1564 */
1565 sleep = mddev->bitmap_info.daemon_sleep * 2;
1566 bitmap->daemon_lastrun -= sleep;
1567 bitmap_daemon_work(mddev);
1568 bitmap->daemon_lastrun -= sleep;
1569 bitmap_daemon_work(mddev);
1570 bitmap->daemon_lastrun -= sleep;
1571 bitmap_daemon_work(mddev);
1572 bitmap_update_sb(bitmap);
1573 }
1574
1575 /*
1576 * free memory that was allocated
1577 */
1578 static void bitmap_free(struct bitmap *bitmap)
1579 {
1580 unsigned long k, pages;
1581 struct bitmap_page *bp;
1582
1583 if (!bitmap) /* there was no bitmap */
1584 return;
1585
1586 /* Shouldn't be needed - but just in case.... */
1587 wait_event(bitmap->write_wait,
1588 atomic_read(&bitmap->pending_writes) == 0);
1589
1590 /* release the bitmap file */
1591 bitmap_file_unmap(&bitmap->storage);
1592
1593 bp = bitmap->counts.bp;
1594 pages = bitmap->counts.pages;
1595
1596 /* free all allocated memory */
1597
1598 if (bp) /* deallocate the page memory */
1599 for (k = 0; k < pages; k++)
1600 if (bp[k].map && !bp[k].hijacked)
1601 kfree(bp[k].map);
1602 kfree(bp);
1603 kfree(bitmap);
1604 }
1605
1606 void bitmap_destroy(struct mddev *mddev)
1607 {
1608 struct bitmap *bitmap = mddev->bitmap;
1609
1610 if (!bitmap) /* there was no bitmap */
1611 return;
1612
1613 mutex_lock(&mddev->bitmap_info.mutex);
1614 mddev->bitmap = NULL; /* disconnect from the md device */
1615 mutex_unlock(&mddev->bitmap_info.mutex);
1616 if (mddev->thread)
1617 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1618
1619 if (bitmap->sysfs_can_clear)
1620 sysfs_put(bitmap->sysfs_can_clear);
1621
1622 bitmap_free(bitmap);
1623 }
1624
1625 /*
1626 * initialize the bitmap structure
1627 * if this returns an error, bitmap_destroy must be called to do clean up
1628 */
1629 int bitmap_create(struct mddev *mddev)
1630 {
1631 struct bitmap *bitmap;
1632 sector_t blocks = mddev->resync_max_sectors;
1633 unsigned long chunks;
1634 unsigned long pages;
1635 struct file *file = mddev->bitmap_info.file;
1636 int err;
1637 struct sysfs_dirent *bm = NULL;
1638
1639 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1640
1641 BUG_ON(file && mddev->bitmap_info.offset);
1642
1643 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1644 if (!bitmap)
1645 return -ENOMEM;
1646
1647 spin_lock_init(&bitmap->counts.lock);
1648 atomic_set(&bitmap->pending_writes, 0);
1649 init_waitqueue_head(&bitmap->write_wait);
1650 init_waitqueue_head(&bitmap->overflow_wait);
1651 init_waitqueue_head(&bitmap->behind_wait);
1652
1653 bitmap->mddev = mddev;
1654
1655 if (mddev->kobj.sd)
1656 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1657 if (bm) {
1658 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1659 sysfs_put(bm);
1660 } else
1661 bitmap->sysfs_can_clear = NULL;
1662
1663 bitmap->storage.file = file;
1664 if (file) {
1665 get_file(file);
1666 /* As future accesses to this file will use bmap,
1667 * and bypass the page cache, we must sync the file
1668 * first.
1669 */
1670 vfs_fsync(file, 1);
1671 }
1672 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1673 if (!mddev->bitmap_info.external) {
1674 /*
1675 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1676 * instructing us to create a new on-disk bitmap instance.
1677 */
1678 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1679 err = bitmap_new_disk_sb(bitmap);
1680 else
1681 err = bitmap_read_sb(bitmap);
1682 } else {
1683 err = 0;
1684 if (mddev->bitmap_info.chunksize == 0 ||
1685 mddev->bitmap_info.daemon_sleep == 0)
1686 /* chunksize and time_base need to be
1687 * set first. */
1688 err = -EINVAL;
1689 }
1690 if (err)
1691 goto error;
1692
1693 bitmap->daemon_lastrun = jiffies;
1694 bitmap->counts.chunkshift = (ffz(~mddev->bitmap_info.chunksize)
1695 - BITMAP_BLOCK_SHIFT);
1696
1697 chunks = (blocks + (1 << bitmap->counts.chunkshift) - 1) >>
1698 bitmap->counts.chunkshift;
1699 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1700
1701 BUG_ON(!pages);
1702
1703 bitmap->counts.chunks = chunks;
1704 bitmap->counts.pages = pages;
1705 bitmap->counts.missing_pages = pages;
1706
1707 bitmap->counts.bp = kzalloc(pages * sizeof(*bitmap->counts.bp),
1708 GFP_KERNEL);
1709
1710 err = -ENOMEM;
1711 if (!bitmap->counts.bp)
1712 goto error;
1713
1714 if (file || mddev->bitmap_info.offset) {
1715 err = bitmap_storage_alloc(&bitmap->storage, bitmap->counts.chunks,
1716 !mddev->bitmap_info.external);
1717 if (err)
1718 goto error;
1719 }
1720 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1721 pages, bmname(bitmap));
1722
1723 mddev->bitmap = bitmap;
1724
1725
1726 return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1727
1728 error:
1729 bitmap_free(bitmap);
1730 return err;
1731 }
1732
1733 int bitmap_load(struct mddev *mddev)
1734 {
1735 int err = 0;
1736 sector_t start = 0;
1737 sector_t sector = 0;
1738 struct bitmap *bitmap = mddev->bitmap;
1739
1740 if (!bitmap)
1741 goto out;
1742
1743 /* Clear out old bitmap info first: Either there is none, or we
1744 * are resuming after someone else has possibly changed things,
1745 * so we should forget old cached info.
1746 * All chunks should be clean, but some might need_sync.
1747 */
1748 while (sector < mddev->resync_max_sectors) {
1749 sector_t blocks;
1750 bitmap_start_sync(bitmap, sector, &blocks, 0);
1751 sector += blocks;
1752 }
1753 bitmap_close_sync(bitmap);
1754
1755 if (mddev->degraded == 0
1756 || bitmap->events_cleared == mddev->events)
1757 /* no need to keep dirty bits to optimise a
1758 * re-add of a missing device */
1759 start = mddev->recovery_cp;
1760
1761 mutex_lock(&mddev->bitmap_info.mutex);
1762 err = bitmap_init_from_disk(bitmap, start);
1763 mutex_unlock(&mddev->bitmap_info.mutex);
1764
1765 if (err)
1766 goto out;
1767 clear_bit(BITMAP_STALE, &bitmap->flags);
1768
1769 /* Kick recovery in case any bits were set */
1770 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1771
1772 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1773 md_wakeup_thread(mddev->thread);
1774
1775 bitmap_update_sb(bitmap);
1776
1777 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1778 err = -EIO;
1779 out:
1780 return err;
1781 }
1782 EXPORT_SYMBOL_GPL(bitmap_load);
1783
1784 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1785 {
1786 unsigned long chunk_kb;
1787 struct bitmap_counts *counts;
1788
1789 if (!bitmap)
1790 return;
1791
1792 counts = &bitmap->counts;
1793
1794 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1795 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1796 "%lu%s chunk",
1797 counts->pages - counts->missing_pages,
1798 counts->pages,
1799 (counts->pages - counts->missing_pages)
1800 << (PAGE_SHIFT - 10),
1801 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1802 chunk_kb ? "KB" : "B");
1803 if (bitmap->storage.file) {
1804 seq_printf(seq, ", file: ");
1805 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1806 }
1807
1808 seq_printf(seq, "\n");
1809 }
1810
1811 static ssize_t
1812 location_show(struct mddev *mddev, char *page)
1813 {
1814 ssize_t len;
1815 if (mddev->bitmap_info.file)
1816 len = sprintf(page, "file");
1817 else if (mddev->bitmap_info.offset)
1818 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1819 else
1820 len = sprintf(page, "none");
1821 len += sprintf(page+len, "\n");
1822 return len;
1823 }
1824
1825 static ssize_t
1826 location_store(struct mddev *mddev, const char *buf, size_t len)
1827 {
1828
1829 if (mddev->pers) {
1830 if (!mddev->pers->quiesce)
1831 return -EBUSY;
1832 if (mddev->recovery || mddev->sync_thread)
1833 return -EBUSY;
1834 }
1835
1836 if (mddev->bitmap || mddev->bitmap_info.file ||
1837 mddev->bitmap_info.offset) {
1838 /* bitmap already configured. Only option is to clear it */
1839 if (strncmp(buf, "none", 4) != 0)
1840 return -EBUSY;
1841 if (mddev->pers) {
1842 mddev->pers->quiesce(mddev, 1);
1843 bitmap_destroy(mddev);
1844 mddev->pers->quiesce(mddev, 0);
1845 }
1846 mddev->bitmap_info.offset = 0;
1847 if (mddev->bitmap_info.file) {
1848 struct file *f = mddev->bitmap_info.file;
1849 mddev->bitmap_info.file = NULL;
1850 restore_bitmap_write_access(f);
1851 fput(f);
1852 }
1853 } else {
1854 /* No bitmap, OK to set a location */
1855 long long offset;
1856 if (strncmp(buf, "none", 4) == 0)
1857 /* nothing to be done */;
1858 else if (strncmp(buf, "file:", 5) == 0) {
1859 /* Not supported yet */
1860 return -EINVAL;
1861 } else {
1862 int rv;
1863 if (buf[0] == '+')
1864 rv = strict_strtoll(buf+1, 10, &offset);
1865 else
1866 rv = strict_strtoll(buf, 10, &offset);
1867 if (rv)
1868 return rv;
1869 if (offset == 0)
1870 return -EINVAL;
1871 if (mddev->bitmap_info.external == 0 &&
1872 mddev->major_version == 0 &&
1873 offset != mddev->bitmap_info.default_offset)
1874 return -EINVAL;
1875 mddev->bitmap_info.offset = offset;
1876 if (mddev->pers) {
1877 mddev->pers->quiesce(mddev, 1);
1878 rv = bitmap_create(mddev);
1879 if (!rv)
1880 rv = bitmap_load(mddev);
1881 if (rv) {
1882 bitmap_destroy(mddev);
1883 mddev->bitmap_info.offset = 0;
1884 }
1885 mddev->pers->quiesce(mddev, 0);
1886 if (rv)
1887 return rv;
1888 }
1889 }
1890 }
1891 if (!mddev->external) {
1892 /* Ensure new bitmap info is stored in
1893 * metadata promptly.
1894 */
1895 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1896 md_wakeup_thread(mddev->thread);
1897 }
1898 return len;
1899 }
1900
1901 static struct md_sysfs_entry bitmap_location =
1902 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1903
1904 /* 'bitmap/space' is the space available at 'location' for the
1905 * bitmap. This allows the kernel to know when it is safe to
1906 * resize the bitmap to match a resized array.
1907 */
1908 static ssize_t
1909 space_show(struct mddev *mddev, char *page)
1910 {
1911 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
1912 }
1913
1914 static ssize_t
1915 space_store(struct mddev *mddev, const char *buf, size_t len)
1916 {
1917 unsigned long sectors;
1918 int rv;
1919
1920 rv = kstrtoul(buf, 10, &sectors);
1921 if (rv)
1922 return rv;
1923
1924 if (sectors == 0)
1925 return -EINVAL;
1926
1927 if (mddev->bitmap &&
1928 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
1929 return -EFBIG; /* Bitmap is too big for this small space */
1930
1931 /* could make sure it isn't too big, but that isn't really
1932 * needed - user-space should be careful.
1933 */
1934 mddev->bitmap_info.space = sectors;
1935 return len;
1936 }
1937
1938 static struct md_sysfs_entry bitmap_space =
1939 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
1940
1941 static ssize_t
1942 timeout_show(struct mddev *mddev, char *page)
1943 {
1944 ssize_t len;
1945 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1946 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1947
1948 len = sprintf(page, "%lu", secs);
1949 if (jifs)
1950 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1951 len += sprintf(page+len, "\n");
1952 return len;
1953 }
1954
1955 static ssize_t
1956 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1957 {
1958 /* timeout can be set at any time */
1959 unsigned long timeout;
1960 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1961 if (rv)
1962 return rv;
1963
1964 /* just to make sure we don't overflow... */
1965 if (timeout >= LONG_MAX / HZ)
1966 return -EINVAL;
1967
1968 timeout = timeout * HZ / 10000;
1969
1970 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1971 timeout = MAX_SCHEDULE_TIMEOUT-1;
1972 if (timeout < 1)
1973 timeout = 1;
1974 mddev->bitmap_info.daemon_sleep = timeout;
1975 if (mddev->thread) {
1976 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1977 * the bitmap is all clean and we don't need to
1978 * adjust the timeout right now
1979 */
1980 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1981 mddev->thread->timeout = timeout;
1982 md_wakeup_thread(mddev->thread);
1983 }
1984 }
1985 return len;
1986 }
1987
1988 static struct md_sysfs_entry bitmap_timeout =
1989 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1990
1991 static ssize_t
1992 backlog_show(struct mddev *mddev, char *page)
1993 {
1994 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1995 }
1996
1997 static ssize_t
1998 backlog_store(struct mddev *mddev, const char *buf, size_t len)
1999 {
2000 unsigned long backlog;
2001 int rv = strict_strtoul(buf, 10, &backlog);
2002 if (rv)
2003 return rv;
2004 if (backlog > COUNTER_MAX)
2005 return -EINVAL;
2006 mddev->bitmap_info.max_write_behind = backlog;
2007 return len;
2008 }
2009
2010 static struct md_sysfs_entry bitmap_backlog =
2011 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2012
2013 static ssize_t
2014 chunksize_show(struct mddev *mddev, char *page)
2015 {
2016 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2017 }
2018
2019 static ssize_t
2020 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2021 {
2022 /* Can only be changed when no bitmap is active */
2023 int rv;
2024 unsigned long csize;
2025 if (mddev->bitmap)
2026 return -EBUSY;
2027 rv = strict_strtoul(buf, 10, &csize);
2028 if (rv)
2029 return rv;
2030 if (csize < 512 ||
2031 !is_power_of_2(csize))
2032 return -EINVAL;
2033 mddev->bitmap_info.chunksize = csize;
2034 return len;
2035 }
2036
2037 static struct md_sysfs_entry bitmap_chunksize =
2038 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2039
2040 static ssize_t metadata_show(struct mddev *mddev, char *page)
2041 {
2042 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2043 ? "external" : "internal"));
2044 }
2045
2046 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2047 {
2048 if (mddev->bitmap ||
2049 mddev->bitmap_info.file ||
2050 mddev->bitmap_info.offset)
2051 return -EBUSY;
2052 if (strncmp(buf, "external", 8) == 0)
2053 mddev->bitmap_info.external = 1;
2054 else if (strncmp(buf, "internal", 8) == 0)
2055 mddev->bitmap_info.external = 0;
2056 else
2057 return -EINVAL;
2058 return len;
2059 }
2060
2061 static struct md_sysfs_entry bitmap_metadata =
2062 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2063
2064 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2065 {
2066 int len;
2067 if (mddev->bitmap)
2068 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2069 "false" : "true"));
2070 else
2071 len = sprintf(page, "\n");
2072 return len;
2073 }
2074
2075 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2076 {
2077 if (mddev->bitmap == NULL)
2078 return -ENOENT;
2079 if (strncmp(buf, "false", 5) == 0)
2080 mddev->bitmap->need_sync = 1;
2081 else if (strncmp(buf, "true", 4) == 0) {
2082 if (mddev->degraded)
2083 return -EBUSY;
2084 mddev->bitmap->need_sync = 0;
2085 } else
2086 return -EINVAL;
2087 return len;
2088 }
2089
2090 static struct md_sysfs_entry bitmap_can_clear =
2091 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2092
2093 static ssize_t
2094 behind_writes_used_show(struct mddev *mddev, char *page)
2095 {
2096 if (mddev->bitmap == NULL)
2097 return sprintf(page, "0\n");
2098 return sprintf(page, "%lu\n",
2099 mddev->bitmap->behind_writes_used);
2100 }
2101
2102 static ssize_t
2103 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2104 {
2105 if (mddev->bitmap)
2106 mddev->bitmap->behind_writes_used = 0;
2107 return len;
2108 }
2109
2110 static struct md_sysfs_entry max_backlog_used =
2111 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2112 behind_writes_used_show, behind_writes_used_reset);
2113
2114 static struct attribute *md_bitmap_attrs[] = {
2115 &bitmap_location.attr,
2116 &bitmap_space.attr,
2117 &bitmap_timeout.attr,
2118 &bitmap_backlog.attr,
2119 &bitmap_chunksize.attr,
2120 &bitmap_metadata.attr,
2121 &bitmap_can_clear.attr,
2122 &max_backlog_used.attr,
2123 NULL
2124 };
2125 struct attribute_group md_bitmap_group = {
2126 .name = "bitmap",
2127 .attrs = md_bitmap_attrs,
2128 };
2129
This page took 0.142926 seconds and 6 git commands to generate.