md/bitmap: merge bitmap_file_unmap and bitmap_file_put.
[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 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
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 (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
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
300 if (!uptodate)
301 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
302 if (atomic_dec_and_test(&bitmap->pending_writes))
303 wake_up(&bitmap->write_wait);
304 }
305
306 /* copied from buffer.c */
307 static void
308 __clear_page_buffers(struct page *page)
309 {
310 ClearPagePrivate(page);
311 set_page_private(page, 0);
312 page_cache_release(page);
313 }
314 static void free_buffers(struct page *page)
315 {
316 struct buffer_head *bh;
317
318 if (!PagePrivate(page))
319 return;
320
321 bh = page_buffers(page);
322 while (bh) {
323 struct buffer_head *next = bh->b_this_page;
324 free_buffer_head(bh);
325 bh = next;
326 }
327 __clear_page_buffers(page);
328 put_page(page);
329 }
330
331 /* read a page from a file.
332 * We both read the page, and attach buffers to the page to record the
333 * address of each block (using bmap). These addresses will be used
334 * to write the block later, completely bypassing the filesystem.
335 * This usage is similar to how swap files are handled, and allows us
336 * to write to a file with no concerns of memory allocation failing.
337 */
338 static int read_page(struct file *file, unsigned long index,
339 struct bitmap *bitmap,
340 unsigned long count,
341 struct page *page)
342 {
343 int ret = 0;
344 struct inode *inode = file->f_path.dentry->d_inode;
345 struct buffer_head *bh;
346 sector_t block;
347
348 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
349 (unsigned long long)index << PAGE_SHIFT);
350
351 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
352 if (!bh) {
353 ret = -ENOMEM;
354 goto out;
355 }
356 attach_page_buffers(page, bh);
357 block = index << (PAGE_SHIFT - inode->i_blkbits);
358 while (bh) {
359 if (count == 0)
360 bh->b_blocknr = 0;
361 else {
362 bh->b_blocknr = bmap(inode, block);
363 if (bh->b_blocknr == 0) {
364 /* Cannot use this file! */
365 ret = -EINVAL;
366 goto out;
367 }
368 bh->b_bdev = inode->i_sb->s_bdev;
369 if (count < (1<<inode->i_blkbits))
370 count = 0;
371 else
372 count -= (1<<inode->i_blkbits);
373
374 bh->b_end_io = end_bitmap_write;
375 bh->b_private = bitmap;
376 atomic_inc(&bitmap->pending_writes);
377 set_buffer_locked(bh);
378 set_buffer_mapped(bh);
379 submit_bh(READ, bh);
380 }
381 block++;
382 bh = bh->b_this_page;
383 }
384 page->index = index;
385
386 wait_event(bitmap->write_wait,
387 atomic_read(&bitmap->pending_writes)==0);
388 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
389 ret = -EIO;
390 out:
391 if (ret)
392 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
393 (int)PAGE_SIZE,
394 (unsigned long long)index << PAGE_SHIFT,
395 ret);
396 return ret;
397 }
398
399 /*
400 * bitmap file superblock operations
401 */
402
403 /* update the event counter and sync the superblock to disk */
404 void bitmap_update_sb(struct bitmap *bitmap)
405 {
406 bitmap_super_t *sb;
407
408 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
409 return;
410 if (bitmap->mddev->bitmap_info.external)
411 return;
412 if (!bitmap->storage.sb_page) /* no superblock */
413 return;
414 sb = kmap_atomic(bitmap->storage.sb_page);
415 sb->events = cpu_to_le64(bitmap->mddev->events);
416 if (bitmap->mddev->events < bitmap->events_cleared)
417 /* rocking back to read-only */
418 bitmap->events_cleared = bitmap->mddev->events;
419 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
420 sb->state = cpu_to_le32(bitmap->flags);
421 /* Just in case these have been changed via sysfs: */
422 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
423 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
424 kunmap_atomic(sb);
425 write_page(bitmap, bitmap->storage.sb_page, 1);
426 }
427
428 /* print out the bitmap file superblock */
429 void bitmap_print_sb(struct bitmap *bitmap)
430 {
431 bitmap_super_t *sb;
432
433 if (!bitmap || !bitmap->storage.sb_page)
434 return;
435 sb = kmap_atomic(bitmap->storage.sb_page);
436 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
437 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
438 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
439 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
440 *(__u32 *)(sb->uuid+0),
441 *(__u32 *)(sb->uuid+4),
442 *(__u32 *)(sb->uuid+8),
443 *(__u32 *)(sb->uuid+12));
444 printk(KERN_DEBUG " events: %llu\n",
445 (unsigned long long) le64_to_cpu(sb->events));
446 printk(KERN_DEBUG "events cleared: %llu\n",
447 (unsigned long long) le64_to_cpu(sb->events_cleared));
448 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
449 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
450 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
451 printk(KERN_DEBUG " sync size: %llu KB\n",
452 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
453 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
454 kunmap_atomic(sb);
455 }
456
457 /*
458 * bitmap_new_disk_sb
459 * @bitmap
460 *
461 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
462 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
463 * This function verifies 'bitmap_info' and populates the on-disk bitmap
464 * structure, which is to be written to disk.
465 *
466 * Returns: 0 on success, -Exxx on error
467 */
468 static int bitmap_new_disk_sb(struct bitmap *bitmap)
469 {
470 bitmap_super_t *sb;
471 unsigned long chunksize, daemon_sleep, write_behind;
472 int err = -EINVAL;
473
474 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
475 if (IS_ERR(bitmap->storage.sb_page)) {
476 err = PTR_ERR(bitmap->storage.sb_page);
477 bitmap->storage.sb_page = NULL;
478 return err;
479 }
480 bitmap->storage.sb_page->index = 0;
481
482 sb = kmap_atomic(bitmap->storage.sb_page);
483
484 sb->magic = cpu_to_le32(BITMAP_MAGIC);
485 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
486
487 chunksize = bitmap->mddev->bitmap_info.chunksize;
488 BUG_ON(!chunksize);
489 if (!is_power_of_2(chunksize)) {
490 kunmap_atomic(sb);
491 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
492 return -EINVAL;
493 }
494 sb->chunksize = cpu_to_le32(chunksize);
495
496 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
497 if (!daemon_sleep ||
498 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
499 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
500 daemon_sleep = 5 * HZ;
501 }
502 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
503 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
504
505 /*
506 * FIXME: write_behind for RAID1. If not specified, what
507 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
508 */
509 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
510 if (write_behind > COUNTER_MAX)
511 write_behind = COUNTER_MAX / 2;
512 sb->write_behind = cpu_to_le32(write_behind);
513 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
514
515 /* keep the array size field of the bitmap superblock up to date */
516 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
517
518 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
519
520 set_bit(BITMAP_STALE, &bitmap->flags);
521 sb->state = cpu_to_le32(bitmap->flags);
522 bitmap->events_cleared = bitmap->mddev->events;
523 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
524
525 kunmap_atomic(sb);
526
527 return 0;
528 }
529
530 /* read the superblock from the bitmap file and initialize some bitmap fields */
531 static int bitmap_read_sb(struct bitmap *bitmap)
532 {
533 char *reason = NULL;
534 bitmap_super_t *sb;
535 unsigned long chunksize, daemon_sleep, write_behind;
536 unsigned long long events;
537 int err = -EINVAL;
538 struct page *sb_page;
539
540 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
541 chunksize = 128 * 1024 * 1024;
542 daemon_sleep = 5 * HZ;
543 write_behind = 0;
544 set_bit(BITMAP_STALE, &bitmap->flags);
545 err = 0;
546 goto out_no_sb;
547 }
548 /* page 0 is the superblock, read it... */
549 sb_page = alloc_page(GFP_KERNEL);
550 if (!sb_page)
551 return -ENOMEM;
552 bitmap->storage.sb_page = sb_page;
553
554 if (bitmap->storage.file) {
555 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
556 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
557
558 err = read_page(bitmap->storage.file, 0,
559 bitmap, bytes, sb_page);
560 } else {
561 err = read_sb_page(bitmap->mddev,
562 bitmap->mddev->bitmap_info.offset,
563 sb_page,
564 0, sizeof(bitmap_super_t));
565 }
566 if (err)
567 return err;
568
569 sb = kmap_atomic(sb_page);
570
571 chunksize = le32_to_cpu(sb->chunksize);
572 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
573 write_behind = le32_to_cpu(sb->write_behind);
574
575 /* verify that the bitmap-specific fields are valid */
576 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
577 reason = "bad magic";
578 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
579 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
580 reason = "unrecognized superblock version";
581 else if (chunksize < 512)
582 reason = "bitmap chunksize too small";
583 else if (!is_power_of_2(chunksize))
584 reason = "bitmap chunksize not a power of 2";
585 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
586 reason = "daemon sleep period out of range";
587 else if (write_behind > COUNTER_MAX)
588 reason = "write-behind limit out of range (0 - 16383)";
589 if (reason) {
590 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
591 bmname(bitmap), reason);
592 goto out;
593 }
594
595 /* keep the array size field of the bitmap superblock up to date */
596 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
597
598 if (bitmap->mddev->persistent) {
599 /*
600 * We have a persistent array superblock, so compare the
601 * bitmap's UUID and event counter to the mddev's
602 */
603 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
604 printk(KERN_INFO
605 "%s: bitmap superblock UUID mismatch\n",
606 bmname(bitmap));
607 goto out;
608 }
609 events = le64_to_cpu(sb->events);
610 if (events < bitmap->mddev->events) {
611 printk(KERN_INFO
612 "%s: bitmap file is out of date (%llu < %llu) "
613 "-- forcing full recovery\n",
614 bmname(bitmap), events,
615 (unsigned long long) bitmap->mddev->events);
616 set_bit(BITMAP_STALE, &bitmap->flags);
617 }
618 }
619
620 /* assign fields using values from superblock */
621 bitmap->flags |= le32_to_cpu(sb->state);
622 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
623 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
624 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
625 err = 0;
626 out:
627 kunmap_atomic(sb);
628 out_no_sb:
629 if (test_bit(BITMAP_STALE, &bitmap->flags))
630 bitmap->events_cleared = bitmap->mddev->events;
631 bitmap->mddev->bitmap_info.chunksize = chunksize;
632 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
633 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
634 if (err)
635 bitmap_print_sb(bitmap);
636 return err;
637 }
638
639 /*
640 * general bitmap file operations
641 */
642
643 /*
644 * on-disk bitmap:
645 *
646 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
647 * file a page at a time. There's a superblock at the start of the file.
648 */
649 /* calculate the index of the page that contains this bit */
650 static inline unsigned long file_page_index(struct bitmap_storage *store,
651 unsigned long chunk)
652 {
653 if (store->sb_page)
654 chunk += sizeof(bitmap_super_t) << 3;
655 return chunk >> PAGE_BIT_SHIFT;
656 }
657
658 /* calculate the (bit) offset of this bit within a page */
659 static inline unsigned long file_page_offset(struct bitmap_storage *store,
660 unsigned long chunk)
661 {
662 if (store->sb_page)
663 chunk += sizeof(bitmap_super_t) << 3;
664 return chunk & (PAGE_BITS - 1);
665 }
666
667 /*
668 * return a pointer to the page in the filemap that contains the given bit
669 *
670 * this lookup is complicated by the fact that the bitmap sb might be exactly
671 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
672 * 0 or page 1
673 */
674 static inline struct page *filemap_get_page(struct bitmap_storage *store,
675 unsigned long chunk)
676 {
677 if (file_page_index(store, chunk) >= store->file_pages)
678 return NULL;
679 return store->filemap[file_page_index(store, chunk)
680 - file_page_index(store, 0)];
681 }
682
683 static int bitmap_storage_alloc(struct bitmap_storage *store,
684 unsigned long chunks, int with_super)
685 {
686 int pnum;
687 unsigned long num_pages;
688 unsigned long bytes;
689
690 bytes = DIV_ROUND_UP(chunks, 8);
691 if (with_super)
692 bytes += sizeof(bitmap_super_t);
693
694 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
695
696 store->filemap = kmalloc(sizeof(struct page *)
697 * num_pages, GFP_KERNEL);
698 if (!store->filemap)
699 return -ENOMEM;
700
701 if (with_super && !store->sb_page) {
702 store->sb_page = alloc_page(GFP_KERNEL);
703 if (store->sb_page == NULL)
704 return -ENOMEM;
705 store->sb_page->index = 0;
706 }
707 pnum = 0;
708 if (store->sb_page) {
709 store->filemap[0] = store->sb_page;
710 pnum = 1;
711 }
712 for ( ; pnum < num_pages; pnum++) {
713 store->filemap[pnum] = alloc_page(GFP_KERNEL);
714 if (!store->filemap[pnum]) {
715 store->file_pages = pnum;
716 return -ENOMEM;
717 }
718 store->filemap[pnum]->index = pnum;
719 }
720 store->file_pages = pnum;
721
722 /* We need 4 bits per page, rounded up to a multiple
723 * of sizeof(unsigned long) */
724 store->filemap_attr = kzalloc(
725 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
726 GFP_KERNEL);
727 if (!store->filemap_attr)
728 return -ENOMEM;
729
730 store->bytes = bytes;
731
732 return 0;
733 }
734
735 static void bitmap_file_unmap(struct bitmap_storage *store)
736 {
737 struct page **map, *sb_page;
738 int pages;
739 struct file *file;
740
741 file = store->file;
742 map = store->filemap;
743 pages = store->file_pages;
744 sb_page = store->sb_page;
745
746 while (pages--)
747 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
748 free_buffers(map[pages]);
749 kfree(map);
750 kfree(store->filemap_attr);
751
752 if (sb_page)
753 free_buffers(sb_page);
754
755 if (file) {
756 struct inode *inode = file->f_path.dentry->d_inode;
757 invalidate_mapping_pages(inode->i_mapping, 0, -1);
758 fput(file);
759 }
760 }
761
762 /*
763 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
764 * then it is no longer reliable, so we stop using it and we mark the file
765 * as failed in the superblock
766 */
767 static void bitmap_file_kick(struct bitmap *bitmap)
768 {
769 char *path, *ptr = NULL;
770
771 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
772 bitmap_update_sb(bitmap);
773
774 if (bitmap->storage.file) {
775 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
776 if (path)
777 ptr = d_path(&bitmap->storage.file->f_path,
778 path, PAGE_SIZE);
779
780 printk(KERN_ALERT
781 "%s: kicking failed bitmap file %s from array!\n",
782 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
783
784 kfree(path);
785 } else
786 printk(KERN_ALERT
787 "%s: disabling internal bitmap due to errors\n",
788 bmname(bitmap));
789 }
790 }
791
792 enum bitmap_page_attr {
793 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
794 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
795 * i.e. counter is 1 or 2. */
796 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
797 };
798
799 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
800 enum bitmap_page_attr attr)
801 {
802 __set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
803 }
804
805 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
806 enum bitmap_page_attr attr)
807 {
808 __clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
809 }
810
811 static inline unsigned long test_page_attr(struct bitmap *bitmap, int pnum,
812 enum bitmap_page_attr attr)
813 {
814 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
815 }
816
817 /*
818 * bitmap_file_set_bit -- called before performing a write to the md device
819 * to set (and eventually sync) a particular bit in the bitmap file
820 *
821 * we set the bit immediately, then we record the page number so that
822 * when an unplug occurs, we can flush the dirty pages out to disk
823 */
824 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
825 {
826 unsigned long bit;
827 struct page *page;
828 void *kaddr;
829 unsigned long chunk = block >> bitmap->chunkshift;
830
831 page = filemap_get_page(&bitmap->storage, chunk);
832 if (!page)
833 return;
834 bit = file_page_offset(&bitmap->storage, chunk);
835
836 /* set the bit */
837 kaddr = kmap_atomic(page);
838 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
839 set_bit(bit, kaddr);
840 else
841 __set_bit_le(bit, kaddr);
842 kunmap_atomic(kaddr);
843 pr_debug("set file bit %lu page %lu\n", bit, page->index);
844 /* record page number so it gets flushed to disk when unplug occurs */
845 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
846 }
847
848 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
849 {
850 unsigned long bit;
851 struct page *page;
852 void *paddr;
853 unsigned long chunk = block >> bitmap->chunkshift;
854
855 page = filemap_get_page(&bitmap->storage, chunk);
856 if (!page)
857 return;
858 bit = file_page_offset(&bitmap->storage, chunk);
859 paddr = kmap_atomic(page);
860 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
861 clear_bit(bit, paddr);
862 else
863 __clear_bit_le(bit, paddr);
864 kunmap_atomic(paddr);
865 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
866 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
867 bitmap->allclean = 0;
868 }
869 }
870
871 /* this gets called when the md device is ready to unplug its underlying
872 * (slave) device queues -- before we let any writes go down, we need to
873 * sync the dirty pages of the bitmap file to disk */
874 void bitmap_unplug(struct bitmap *bitmap)
875 {
876 unsigned long i;
877 int dirty, need_write;
878 int wait = 0;
879
880 if (!bitmap || !bitmap->storage.filemap ||
881 test_bit(BITMAP_STALE, &bitmap->flags))
882 return;
883
884 /* look at each page to see if there are any set bits that need to be
885 * flushed out to disk */
886 for (i = 0; i < bitmap->storage.file_pages; i++) {
887 spin_lock_irq(&bitmap->lock);
888 if (!bitmap->storage.filemap) {
889 spin_unlock_irq(&bitmap->lock);
890 return;
891 }
892 dirty = test_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
893 need_write = test_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
894 clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
895 clear_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
896 if (dirty || need_write)
897 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
898 if (dirty)
899 wait = 1;
900 spin_unlock_irq(&bitmap->lock);
901
902 if (dirty || need_write)
903 write_page(bitmap, bitmap->storage.filemap[i], 0);
904 }
905 if (wait) { /* if any writes were performed, we need to wait on them */
906 if (bitmap->storage.file)
907 wait_event(bitmap->write_wait,
908 atomic_read(&bitmap->pending_writes)==0);
909 else
910 md_super_wait(bitmap->mddev);
911 }
912 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
913 bitmap_file_kick(bitmap);
914 }
915 EXPORT_SYMBOL(bitmap_unplug);
916
917 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
918 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
919 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
920 * memory mapping of the bitmap file
921 * Special cases:
922 * if there's no bitmap file, or if the bitmap file had been
923 * previously kicked from the array, we mark all the bits as
924 * 1's in order to cause a full resync.
925 *
926 * We ignore all bits for sectors that end earlier than 'start'.
927 * This is used when reading an out-of-date bitmap...
928 */
929 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
930 {
931 unsigned long i, chunks, index, oldindex, bit;
932 struct page *page = NULL;
933 unsigned long bit_cnt = 0;
934 struct file *file;
935 unsigned long offset;
936 int outofdate;
937 int ret = -ENOSPC;
938 void *paddr;
939 struct bitmap_storage *store = &bitmap->storage;
940
941 chunks = bitmap->chunks;
942 file = store->file;
943
944 if (!file && !bitmap->mddev->bitmap_info.offset) {
945 /* No permanent bitmap - fill with '1s'. */
946 store->filemap = NULL;
947 store->file_pages = 0;
948 for (i = 0; i < chunks ; i++) {
949 /* if the disk bit is set, set the memory bit */
950 int needed = ((sector_t)(i+1) << (bitmap->chunkshift)
951 >= start);
952 bitmap_set_memory_bits(bitmap,
953 (sector_t)i << bitmap->chunkshift,
954 needed);
955 }
956 return 0;
957 }
958
959 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
960 if (outofdate)
961 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
962 "recovery\n", bmname(bitmap));
963
964 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
965 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
966 bmname(bitmap),
967 (unsigned long) i_size_read(file->f_mapping->host),
968 store->bytes);
969 goto err;
970 }
971
972 oldindex = ~0L;
973 offset = 0;
974 if (!bitmap->mddev->bitmap_info.external)
975 offset = sizeof(bitmap_super_t);
976
977 for (i = 0; i < chunks; i++) {
978 int b;
979 index = file_page_index(&bitmap->storage, i);
980 bit = file_page_offset(&bitmap->storage, i);
981 if (index != oldindex) { /* this is a new page, read it in */
982 int count;
983 /* unmap the old page, we're done with it */
984 if (index == store->file_pages-1)
985 count = store->bytes - index * PAGE_SIZE;
986 else
987 count = PAGE_SIZE;
988 page = store->filemap[index];
989 if (file)
990 ret = read_page(file, index, bitmap,
991 count, page);
992 else
993 ret = read_sb_page(
994 bitmap->mddev,
995 bitmap->mddev->bitmap_info.offset,
996 page,
997 index, count);
998
999 if (ret)
1000 goto err;
1001
1002 oldindex = index;
1003
1004 if (outofdate) {
1005 /*
1006 * if bitmap is out of date, dirty the
1007 * whole page and write it out
1008 */
1009 paddr = kmap_atomic(page);
1010 memset(paddr + offset, 0xff,
1011 PAGE_SIZE - offset);
1012 kunmap_atomic(paddr);
1013 write_page(bitmap, page, 1);
1014
1015 ret = -EIO;
1016 if (test_bit(BITMAP_WRITE_ERROR,
1017 &bitmap->flags))
1018 goto err;
1019 }
1020 }
1021 paddr = kmap_atomic(page);
1022 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1023 b = test_bit(bit, paddr);
1024 else
1025 b = test_bit_le(bit, paddr);
1026 kunmap_atomic(paddr);
1027 if (b) {
1028 /* if the disk bit is set, set the memory bit */
1029 int needed = ((sector_t)(i+1) << bitmap->chunkshift
1030 >= start);
1031 bitmap_set_memory_bits(bitmap,
1032 (sector_t)i << bitmap->chunkshift,
1033 needed);
1034 bit_cnt++;
1035 }
1036 offset = 0;
1037 }
1038
1039 printk(KERN_INFO "%s: bitmap initialized from disk: "
1040 "read %lu pages, set %lu of %lu bits\n",
1041 bmname(bitmap), store->file_pages,
1042 bit_cnt, chunks);
1043
1044 return 0;
1045
1046 err:
1047 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1048 bmname(bitmap), ret);
1049 return ret;
1050 }
1051
1052 void bitmap_write_all(struct bitmap *bitmap)
1053 {
1054 /* We don't actually write all bitmap blocks here,
1055 * just flag them as needing to be written
1056 */
1057 int i;
1058
1059 if (!bitmap || !bitmap->storage.filemap)
1060 return;
1061 if (bitmap->storage.file)
1062 /* Only one copy, so nothing needed */
1063 return;
1064
1065 spin_lock_irq(&bitmap->lock);
1066 for (i = 0; i < bitmap->storage.file_pages; i++)
1067 set_page_attr(bitmap, i,
1068 BITMAP_PAGE_NEEDWRITE);
1069 bitmap->allclean = 0;
1070 spin_unlock_irq(&bitmap->lock);
1071 }
1072
1073 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1074 {
1075 sector_t chunk = offset >> bitmap->chunkshift;
1076 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1077 bitmap->bp[page].count += inc;
1078 bitmap_checkfree(bitmap, page);
1079 }
1080
1081 static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset)
1082 {
1083 sector_t chunk = offset >> bitmap->chunkshift;
1084 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1085 struct bitmap_page *bp = &bitmap->bp[page];
1086
1087 if (!bp->pending)
1088 bp->pending = 1;
1089 }
1090
1091 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1092 sector_t offset, sector_t *blocks,
1093 int create);
1094
1095 /*
1096 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1097 * out to disk
1098 */
1099
1100 void bitmap_daemon_work(struct mddev *mddev)
1101 {
1102 struct bitmap *bitmap;
1103 unsigned long j;
1104 unsigned long nextpage;
1105 sector_t blocks;
1106
1107 /* Use a mutex to guard daemon_work against
1108 * bitmap_destroy.
1109 */
1110 mutex_lock(&mddev->bitmap_info.mutex);
1111 bitmap = mddev->bitmap;
1112 if (bitmap == NULL) {
1113 mutex_unlock(&mddev->bitmap_info.mutex);
1114 return;
1115 }
1116 if (time_before(jiffies, bitmap->daemon_lastrun
1117 + mddev->bitmap_info.daemon_sleep))
1118 goto done;
1119
1120 bitmap->daemon_lastrun = jiffies;
1121 if (bitmap->allclean) {
1122 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1123 goto done;
1124 }
1125 bitmap->allclean = 1;
1126
1127 /* Any file-page which is PENDING now needs to be written.
1128 * So set NEEDWRITE now, then after we make any last-minute changes
1129 * we will write it.
1130 */
1131 spin_lock_irq(&bitmap->lock);
1132 for (j = 0; j < bitmap->storage.file_pages; j++)
1133 if (test_page_attr(bitmap, j,
1134 BITMAP_PAGE_PENDING)) {
1135 set_page_attr(bitmap, j,
1136 BITMAP_PAGE_NEEDWRITE);
1137 clear_page_attr(bitmap, j,
1138 BITMAP_PAGE_PENDING);
1139 }
1140
1141 if (bitmap->need_sync &&
1142 mddev->bitmap_info.external == 0) {
1143 /* Arrange for superblock update as well as
1144 * other changes */
1145 bitmap_super_t *sb;
1146 bitmap->need_sync = 0;
1147 if (bitmap->storage.filemap) {
1148 sb = kmap_atomic(bitmap->storage.sb_page);
1149 sb->events_cleared =
1150 cpu_to_le64(bitmap->events_cleared);
1151 kunmap_atomic(sb);
1152 set_page_attr(bitmap, 0,
1153 BITMAP_PAGE_NEEDWRITE);
1154 }
1155 }
1156 /* Now look at the bitmap counters and if any are '2' or '1',
1157 * decrement and handle accordingly.
1158 */
1159 nextpage = 0;
1160 for (j = 0; j < bitmap->chunks; j++) {
1161 bitmap_counter_t *bmc;
1162 sector_t block = (sector_t)j << bitmap->chunkshift;
1163
1164 if (j == nextpage) {
1165 nextpage += PAGE_COUNTER_RATIO;
1166 if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1167 j |= PAGE_COUNTER_MASK;
1168 continue;
1169 }
1170 bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1171 }
1172 bmc = bitmap_get_counter(bitmap,
1173 block,
1174 &blocks, 0);
1175
1176 if (!bmc) {
1177 j |= PAGE_COUNTER_MASK;
1178 continue;
1179 }
1180 if (*bmc == 1 && !bitmap->need_sync) {
1181 /* We can clear the bit */
1182 *bmc = 0;
1183 bitmap_count_page(bitmap, block, -1);
1184 bitmap_file_clear_bit(bitmap, block);
1185 } else if (*bmc && *bmc <= 2) {
1186 *bmc = 1;
1187 bitmap_set_pending(bitmap, block);
1188 bitmap->allclean = 0;
1189 }
1190 }
1191
1192 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1193 * DIRTY pages need to be written by bitmap_unplug so it can wait
1194 * for them.
1195 * If we find any DIRTY page we stop there and let bitmap_unplug
1196 * handle all the rest. This is important in the case where
1197 * the first blocking holds the superblock and it has been updated.
1198 * We mustn't write any other blocks before the superblock.
1199 */
1200 for (j = 0;
1201 j < bitmap->storage.file_pages
1202 && !test_bit(BITMAP_STALE, &bitmap->flags);
1203 j++) {
1204
1205 if (test_page_attr(bitmap, j,
1206 BITMAP_PAGE_DIRTY))
1207 /* bitmap_unplug will handle the rest */
1208 break;
1209 if (test_page_attr(bitmap, j,
1210 BITMAP_PAGE_NEEDWRITE)) {
1211 clear_page_attr(bitmap, j,
1212 BITMAP_PAGE_NEEDWRITE);
1213 spin_unlock_irq(&bitmap->lock);
1214 write_page(bitmap, bitmap->storage.filemap[j], 0);
1215 spin_lock_irq(&bitmap->lock);
1216 }
1217 }
1218 spin_unlock_irq(&bitmap->lock);
1219
1220 done:
1221 if (bitmap->allclean == 0)
1222 mddev->thread->timeout =
1223 mddev->bitmap_info.daemon_sleep;
1224 mutex_unlock(&mddev->bitmap_info.mutex);
1225 }
1226
1227 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1228 sector_t offset, sector_t *blocks,
1229 int create)
1230 __releases(bitmap->lock)
1231 __acquires(bitmap->lock)
1232 {
1233 /* If 'create', we might release the lock and reclaim it.
1234 * The lock must have been taken with interrupts enabled.
1235 * If !create, we don't release the lock.
1236 */
1237 sector_t chunk = offset >> bitmap->chunkshift;
1238 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1239 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1240 sector_t csize;
1241 int err;
1242
1243 err = bitmap_checkpage(bitmap, page, create);
1244
1245 if (bitmap->bp[page].hijacked ||
1246 bitmap->bp[page].map == NULL)
1247 csize = ((sector_t)1) << (bitmap->chunkshift +
1248 PAGE_COUNTER_SHIFT - 1);
1249 else
1250 csize = ((sector_t)1) << bitmap->chunkshift;
1251 *blocks = csize - (offset & (csize - 1));
1252
1253 if (err < 0)
1254 return NULL;
1255
1256 /* now locked ... */
1257
1258 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1259 /* should we use the first or second counter field
1260 * of the hijacked pointer? */
1261 int hi = (pageoff > PAGE_COUNTER_MASK);
1262 return &((bitmap_counter_t *)
1263 &bitmap->bp[page].map)[hi];
1264 } else /* page is allocated */
1265 return (bitmap_counter_t *)
1266 &(bitmap->bp[page].map[pageoff]);
1267 }
1268
1269 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1270 {
1271 if (!bitmap)
1272 return 0;
1273
1274 if (behind) {
1275 int bw;
1276 atomic_inc(&bitmap->behind_writes);
1277 bw = atomic_read(&bitmap->behind_writes);
1278 if (bw > bitmap->behind_writes_used)
1279 bitmap->behind_writes_used = bw;
1280
1281 pr_debug("inc write-behind count %d/%lu\n",
1282 bw, bitmap->mddev->bitmap_info.max_write_behind);
1283 }
1284
1285 while (sectors) {
1286 sector_t blocks;
1287 bitmap_counter_t *bmc;
1288
1289 spin_lock_irq(&bitmap->lock);
1290 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1291 if (!bmc) {
1292 spin_unlock_irq(&bitmap->lock);
1293 return 0;
1294 }
1295
1296 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1297 DEFINE_WAIT(__wait);
1298 /* note that it is safe to do the prepare_to_wait
1299 * after the test as long as we do it before dropping
1300 * the spinlock.
1301 */
1302 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1303 TASK_UNINTERRUPTIBLE);
1304 spin_unlock_irq(&bitmap->lock);
1305 io_schedule();
1306 finish_wait(&bitmap->overflow_wait, &__wait);
1307 continue;
1308 }
1309
1310 switch (*bmc) {
1311 case 0:
1312 bitmap_file_set_bit(bitmap, offset);
1313 bitmap_count_page(bitmap, offset, 1);
1314 /* fall through */
1315 case 1:
1316 *bmc = 2;
1317 }
1318
1319 (*bmc)++;
1320
1321 spin_unlock_irq(&bitmap->lock);
1322
1323 offset += blocks;
1324 if (sectors > blocks)
1325 sectors -= blocks;
1326 else
1327 sectors = 0;
1328 }
1329 return 0;
1330 }
1331 EXPORT_SYMBOL(bitmap_startwrite);
1332
1333 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1334 int success, int behind)
1335 {
1336 if (!bitmap)
1337 return;
1338 if (behind) {
1339 if (atomic_dec_and_test(&bitmap->behind_writes))
1340 wake_up(&bitmap->behind_wait);
1341 pr_debug("dec write-behind count %d/%lu\n",
1342 atomic_read(&bitmap->behind_writes),
1343 bitmap->mddev->bitmap_info.max_write_behind);
1344 }
1345
1346 while (sectors) {
1347 sector_t blocks;
1348 unsigned long flags;
1349 bitmap_counter_t *bmc;
1350
1351 spin_lock_irqsave(&bitmap->lock, flags);
1352 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1353 if (!bmc) {
1354 spin_unlock_irqrestore(&bitmap->lock, flags);
1355 return;
1356 }
1357
1358 if (success && !bitmap->mddev->degraded &&
1359 bitmap->events_cleared < bitmap->mddev->events) {
1360 bitmap->events_cleared = bitmap->mddev->events;
1361 bitmap->need_sync = 1;
1362 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1363 }
1364
1365 if (!success && !NEEDED(*bmc))
1366 *bmc |= NEEDED_MASK;
1367
1368 if (COUNTER(*bmc) == COUNTER_MAX)
1369 wake_up(&bitmap->overflow_wait);
1370
1371 (*bmc)--;
1372 if (*bmc <= 2) {
1373 bitmap_set_pending(bitmap, offset);
1374 bitmap->allclean = 0;
1375 }
1376 spin_unlock_irqrestore(&bitmap->lock, flags);
1377 offset += blocks;
1378 if (sectors > blocks)
1379 sectors -= blocks;
1380 else
1381 sectors = 0;
1382 }
1383 }
1384 EXPORT_SYMBOL(bitmap_endwrite);
1385
1386 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1387 int degraded)
1388 {
1389 bitmap_counter_t *bmc;
1390 int rv;
1391 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1392 *blocks = 1024;
1393 return 1; /* always resync if no bitmap */
1394 }
1395 spin_lock_irq(&bitmap->lock);
1396 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1397 rv = 0;
1398 if (bmc) {
1399 /* locked */
1400 if (RESYNC(*bmc))
1401 rv = 1;
1402 else if (NEEDED(*bmc)) {
1403 rv = 1;
1404 if (!degraded) { /* don't set/clear bits if degraded */
1405 *bmc |= RESYNC_MASK;
1406 *bmc &= ~NEEDED_MASK;
1407 }
1408 }
1409 }
1410 spin_unlock_irq(&bitmap->lock);
1411 return rv;
1412 }
1413
1414 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1415 int degraded)
1416 {
1417 /* bitmap_start_sync must always report on multiples of whole
1418 * pages, otherwise resync (which is very PAGE_SIZE based) will
1419 * get confused.
1420 * So call __bitmap_start_sync repeatedly (if needed) until
1421 * At least PAGE_SIZE>>9 blocks are covered.
1422 * Return the 'or' of the result.
1423 */
1424 int rv = 0;
1425 sector_t blocks1;
1426
1427 *blocks = 0;
1428 while (*blocks < (PAGE_SIZE>>9)) {
1429 rv |= __bitmap_start_sync(bitmap, offset,
1430 &blocks1, degraded);
1431 offset += blocks1;
1432 *blocks += blocks1;
1433 }
1434 return rv;
1435 }
1436 EXPORT_SYMBOL(bitmap_start_sync);
1437
1438 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1439 {
1440 bitmap_counter_t *bmc;
1441 unsigned long flags;
1442
1443 if (bitmap == NULL) {
1444 *blocks = 1024;
1445 return;
1446 }
1447 spin_lock_irqsave(&bitmap->lock, flags);
1448 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1449 if (bmc == NULL)
1450 goto unlock;
1451 /* locked */
1452 if (RESYNC(*bmc)) {
1453 *bmc &= ~RESYNC_MASK;
1454
1455 if (!NEEDED(*bmc) && aborted)
1456 *bmc |= NEEDED_MASK;
1457 else {
1458 if (*bmc <= 2) {
1459 bitmap_set_pending(bitmap, offset);
1460 bitmap->allclean = 0;
1461 }
1462 }
1463 }
1464 unlock:
1465 spin_unlock_irqrestore(&bitmap->lock, flags);
1466 }
1467 EXPORT_SYMBOL(bitmap_end_sync);
1468
1469 void bitmap_close_sync(struct bitmap *bitmap)
1470 {
1471 /* Sync has finished, and any bitmap chunks that weren't synced
1472 * properly have been aborted. It remains to us to clear the
1473 * RESYNC bit wherever it is still on
1474 */
1475 sector_t sector = 0;
1476 sector_t blocks;
1477 if (!bitmap)
1478 return;
1479 while (sector < bitmap->mddev->resync_max_sectors) {
1480 bitmap_end_sync(bitmap, sector, &blocks, 0);
1481 sector += blocks;
1482 }
1483 }
1484 EXPORT_SYMBOL(bitmap_close_sync);
1485
1486 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1487 {
1488 sector_t s = 0;
1489 sector_t blocks;
1490
1491 if (!bitmap)
1492 return;
1493 if (sector == 0) {
1494 bitmap->last_end_sync = jiffies;
1495 return;
1496 }
1497 if (time_before(jiffies, (bitmap->last_end_sync
1498 + bitmap->mddev->bitmap_info.daemon_sleep)))
1499 return;
1500 wait_event(bitmap->mddev->recovery_wait,
1501 atomic_read(&bitmap->mddev->recovery_active) == 0);
1502
1503 bitmap->mddev->curr_resync_completed = sector;
1504 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1505 sector &= ~((1ULL << bitmap->chunkshift) - 1);
1506 s = 0;
1507 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1508 bitmap_end_sync(bitmap, s, &blocks, 0);
1509 s += blocks;
1510 }
1511 bitmap->last_end_sync = jiffies;
1512 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1513 }
1514 EXPORT_SYMBOL(bitmap_cond_end_sync);
1515
1516 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1517 {
1518 /* For each chunk covered by any of these sectors, set the
1519 * counter to 2 and possibly set resync_needed. They should all
1520 * be 0 at this point
1521 */
1522
1523 sector_t secs;
1524 bitmap_counter_t *bmc;
1525 spin_lock_irq(&bitmap->lock);
1526 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1527 if (!bmc) {
1528 spin_unlock_irq(&bitmap->lock);
1529 return;
1530 }
1531 if (!*bmc) {
1532 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1533 bitmap_count_page(bitmap, offset, 1);
1534 bitmap_set_pending(bitmap, offset);
1535 bitmap->allclean = 0;
1536 }
1537 spin_unlock_irq(&bitmap->lock);
1538 }
1539
1540 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1541 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1542 {
1543 unsigned long chunk;
1544
1545 for (chunk = s; chunk <= e; chunk++) {
1546 sector_t sec = (sector_t)chunk << bitmap->chunkshift;
1547 bitmap_set_memory_bits(bitmap, sec, 1);
1548 spin_lock_irq(&bitmap->lock);
1549 bitmap_file_set_bit(bitmap, sec);
1550 spin_unlock_irq(&bitmap->lock);
1551 if (sec < bitmap->mddev->recovery_cp)
1552 /* We are asserting that the array is dirty,
1553 * so move the recovery_cp address back so
1554 * that it is obvious that it is dirty
1555 */
1556 bitmap->mddev->recovery_cp = sec;
1557 }
1558 }
1559
1560 /*
1561 * flush out any pending updates
1562 */
1563 void bitmap_flush(struct mddev *mddev)
1564 {
1565 struct bitmap *bitmap = mddev->bitmap;
1566 long sleep;
1567
1568 if (!bitmap) /* there was no bitmap */
1569 return;
1570
1571 /* run the daemon_work three time to ensure everything is flushed
1572 * that can be
1573 */
1574 sleep = mddev->bitmap_info.daemon_sleep * 2;
1575 bitmap->daemon_lastrun -= sleep;
1576 bitmap_daemon_work(mddev);
1577 bitmap->daemon_lastrun -= sleep;
1578 bitmap_daemon_work(mddev);
1579 bitmap->daemon_lastrun -= sleep;
1580 bitmap_daemon_work(mddev);
1581 bitmap_update_sb(bitmap);
1582 }
1583
1584 /*
1585 * free memory that was allocated
1586 */
1587 static void bitmap_free(struct bitmap *bitmap)
1588 {
1589 unsigned long k, pages;
1590 struct bitmap_page *bp;
1591
1592 if (!bitmap) /* there was no bitmap */
1593 return;
1594
1595 /* Shouldn't be needed - but just in case.... */
1596 wait_event(bitmap->write_wait,
1597 atomic_read(&bitmap->pending_writes) == 0);
1598
1599 /* release the bitmap file */
1600 bitmap_file_unmap(&bitmap->storage);
1601
1602 bp = bitmap->bp;
1603 pages = bitmap->pages;
1604
1605 /* free all allocated memory */
1606
1607 if (bp) /* deallocate the page memory */
1608 for (k = 0; k < pages; k++)
1609 if (bp[k].map && !bp[k].hijacked)
1610 kfree(bp[k].map);
1611 kfree(bp);
1612 kfree(bitmap);
1613 }
1614
1615 void bitmap_destroy(struct mddev *mddev)
1616 {
1617 struct bitmap *bitmap = mddev->bitmap;
1618
1619 if (!bitmap) /* there was no bitmap */
1620 return;
1621
1622 mutex_lock(&mddev->bitmap_info.mutex);
1623 mddev->bitmap = NULL; /* disconnect from the md device */
1624 mutex_unlock(&mddev->bitmap_info.mutex);
1625 if (mddev->thread)
1626 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1627
1628 if (bitmap->sysfs_can_clear)
1629 sysfs_put(bitmap->sysfs_can_clear);
1630
1631 bitmap_free(bitmap);
1632 }
1633
1634 /*
1635 * initialize the bitmap structure
1636 * if this returns an error, bitmap_destroy must be called to do clean up
1637 */
1638 int bitmap_create(struct mddev *mddev)
1639 {
1640 struct bitmap *bitmap;
1641 sector_t blocks = mddev->resync_max_sectors;
1642 unsigned long chunks;
1643 unsigned long pages;
1644 struct file *file = mddev->bitmap_info.file;
1645 int err;
1646 struct sysfs_dirent *bm = NULL;
1647
1648 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1649
1650 BUG_ON(file && mddev->bitmap_info.offset);
1651
1652 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1653 if (!bitmap)
1654 return -ENOMEM;
1655
1656 spin_lock_init(&bitmap->lock);
1657 atomic_set(&bitmap->pending_writes, 0);
1658 init_waitqueue_head(&bitmap->write_wait);
1659 init_waitqueue_head(&bitmap->overflow_wait);
1660 init_waitqueue_head(&bitmap->behind_wait);
1661
1662 bitmap->mddev = mddev;
1663
1664 if (mddev->kobj.sd)
1665 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1666 if (bm) {
1667 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1668 sysfs_put(bm);
1669 } else
1670 bitmap->sysfs_can_clear = NULL;
1671
1672 bitmap->storage.file = file;
1673 if (file) {
1674 get_file(file);
1675 /* As future accesses to this file will use bmap,
1676 * and bypass the page cache, we must sync the file
1677 * first.
1678 */
1679 vfs_fsync(file, 1);
1680 }
1681 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1682 if (!mddev->bitmap_info.external) {
1683 /*
1684 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1685 * instructing us to create a new on-disk bitmap instance.
1686 */
1687 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1688 err = bitmap_new_disk_sb(bitmap);
1689 else
1690 err = bitmap_read_sb(bitmap);
1691 } else {
1692 err = 0;
1693 if (mddev->bitmap_info.chunksize == 0 ||
1694 mddev->bitmap_info.daemon_sleep == 0)
1695 /* chunksize and time_base need to be
1696 * set first. */
1697 err = -EINVAL;
1698 }
1699 if (err)
1700 goto error;
1701
1702 bitmap->daemon_lastrun = jiffies;
1703 bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize)
1704 - BITMAP_BLOCK_SHIFT);
1705
1706 chunks = (blocks + (1 << bitmap->chunkshift) - 1) >>
1707 bitmap->chunkshift;
1708 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1709
1710 BUG_ON(!pages);
1711
1712 bitmap->chunks = chunks;
1713 bitmap->pages = pages;
1714 bitmap->missing_pages = pages;
1715
1716 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1717
1718 err = -ENOMEM;
1719 if (!bitmap->bp)
1720 goto error;
1721
1722 if (file || mddev->bitmap_info.offset) {
1723 err = bitmap_storage_alloc(&bitmap->storage, bitmap->chunks,
1724 !mddev->bitmap_info.external);
1725 if (err)
1726 goto error;
1727 }
1728 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1729 pages, bmname(bitmap));
1730
1731 mddev->bitmap = bitmap;
1732
1733
1734 return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1735
1736 error:
1737 bitmap_free(bitmap);
1738 return err;
1739 }
1740
1741 int bitmap_load(struct mddev *mddev)
1742 {
1743 int err = 0;
1744 sector_t start = 0;
1745 sector_t sector = 0;
1746 struct bitmap *bitmap = mddev->bitmap;
1747
1748 if (!bitmap)
1749 goto out;
1750
1751 /* Clear out old bitmap info first: Either there is none, or we
1752 * are resuming after someone else has possibly changed things,
1753 * so we should forget old cached info.
1754 * All chunks should be clean, but some might need_sync.
1755 */
1756 while (sector < mddev->resync_max_sectors) {
1757 sector_t blocks;
1758 bitmap_start_sync(bitmap, sector, &blocks, 0);
1759 sector += blocks;
1760 }
1761 bitmap_close_sync(bitmap);
1762
1763 if (mddev->degraded == 0
1764 || bitmap->events_cleared == mddev->events)
1765 /* no need to keep dirty bits to optimise a
1766 * re-add of a missing device */
1767 start = mddev->recovery_cp;
1768
1769 mutex_lock(&mddev->bitmap_info.mutex);
1770 err = bitmap_init_from_disk(bitmap, start);
1771 mutex_unlock(&mddev->bitmap_info.mutex);
1772
1773 if (err)
1774 goto out;
1775 clear_bit(BITMAP_STALE, &bitmap->flags);
1776
1777 /* Kick recovery in case any bits were set */
1778 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1779
1780 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1781 md_wakeup_thread(mddev->thread);
1782
1783 bitmap_update_sb(bitmap);
1784
1785 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1786 err = -EIO;
1787 out:
1788 return err;
1789 }
1790 EXPORT_SYMBOL_GPL(bitmap_load);
1791
1792 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1793 {
1794 unsigned long chunk_kb;
1795
1796 if (!bitmap)
1797 return;
1798
1799 spin_lock_irq(&bitmap->lock);
1800 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1801 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1802 "%lu%s chunk",
1803 bitmap->pages - bitmap->missing_pages,
1804 bitmap->pages,
1805 (bitmap->pages - bitmap->missing_pages)
1806 << (PAGE_SHIFT - 10),
1807 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1808 chunk_kb ? "KB" : "B");
1809 if (bitmap->storage.file) {
1810 seq_printf(seq, ", file: ");
1811 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1812 }
1813
1814 seq_printf(seq, "\n");
1815 spin_unlock_irq(&bitmap->lock);
1816 }
1817
1818 static ssize_t
1819 location_show(struct mddev *mddev, char *page)
1820 {
1821 ssize_t len;
1822 if (mddev->bitmap_info.file)
1823 len = sprintf(page, "file");
1824 else if (mddev->bitmap_info.offset)
1825 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1826 else
1827 len = sprintf(page, "none");
1828 len += sprintf(page+len, "\n");
1829 return len;
1830 }
1831
1832 static ssize_t
1833 location_store(struct mddev *mddev, const char *buf, size_t len)
1834 {
1835
1836 if (mddev->pers) {
1837 if (!mddev->pers->quiesce)
1838 return -EBUSY;
1839 if (mddev->recovery || mddev->sync_thread)
1840 return -EBUSY;
1841 }
1842
1843 if (mddev->bitmap || mddev->bitmap_info.file ||
1844 mddev->bitmap_info.offset) {
1845 /* bitmap already configured. Only option is to clear it */
1846 if (strncmp(buf, "none", 4) != 0)
1847 return -EBUSY;
1848 if (mddev->pers) {
1849 mddev->pers->quiesce(mddev, 1);
1850 bitmap_destroy(mddev);
1851 mddev->pers->quiesce(mddev, 0);
1852 }
1853 mddev->bitmap_info.offset = 0;
1854 if (mddev->bitmap_info.file) {
1855 struct file *f = mddev->bitmap_info.file;
1856 mddev->bitmap_info.file = NULL;
1857 restore_bitmap_write_access(f);
1858 fput(f);
1859 }
1860 } else {
1861 /* No bitmap, OK to set a location */
1862 long long offset;
1863 if (strncmp(buf, "none", 4) == 0)
1864 /* nothing to be done */;
1865 else if (strncmp(buf, "file:", 5) == 0) {
1866 /* Not supported yet */
1867 return -EINVAL;
1868 } else {
1869 int rv;
1870 if (buf[0] == '+')
1871 rv = strict_strtoll(buf+1, 10, &offset);
1872 else
1873 rv = strict_strtoll(buf, 10, &offset);
1874 if (rv)
1875 return rv;
1876 if (offset == 0)
1877 return -EINVAL;
1878 if (mddev->bitmap_info.external == 0 &&
1879 mddev->major_version == 0 &&
1880 offset != mddev->bitmap_info.default_offset)
1881 return -EINVAL;
1882 mddev->bitmap_info.offset = offset;
1883 if (mddev->pers) {
1884 mddev->pers->quiesce(mddev, 1);
1885 rv = bitmap_create(mddev);
1886 if (!rv)
1887 rv = bitmap_load(mddev);
1888 if (rv) {
1889 bitmap_destroy(mddev);
1890 mddev->bitmap_info.offset = 0;
1891 }
1892 mddev->pers->quiesce(mddev, 0);
1893 if (rv)
1894 return rv;
1895 }
1896 }
1897 }
1898 if (!mddev->external) {
1899 /* Ensure new bitmap info is stored in
1900 * metadata promptly.
1901 */
1902 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1903 md_wakeup_thread(mddev->thread);
1904 }
1905 return len;
1906 }
1907
1908 static struct md_sysfs_entry bitmap_location =
1909 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1910
1911 /* 'bitmap/space' is the space available at 'location' for the
1912 * bitmap. This allows the kernel to know when it is safe to
1913 * resize the bitmap to match a resized array.
1914 */
1915 static ssize_t
1916 space_show(struct mddev *mddev, char *page)
1917 {
1918 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
1919 }
1920
1921 static ssize_t
1922 space_store(struct mddev *mddev, const char *buf, size_t len)
1923 {
1924 unsigned long sectors;
1925 int rv;
1926
1927 rv = kstrtoul(buf, 10, &sectors);
1928 if (rv)
1929 return rv;
1930
1931 if (sectors == 0)
1932 return -EINVAL;
1933
1934 if (mddev->bitmap &&
1935 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
1936 return -EFBIG; /* Bitmap is too big for this small space */
1937
1938 /* could make sure it isn't too big, but that isn't really
1939 * needed - user-space should be careful.
1940 */
1941 mddev->bitmap_info.space = sectors;
1942 return len;
1943 }
1944
1945 static struct md_sysfs_entry bitmap_space =
1946 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
1947
1948 static ssize_t
1949 timeout_show(struct mddev *mddev, char *page)
1950 {
1951 ssize_t len;
1952 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1953 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1954
1955 len = sprintf(page, "%lu", secs);
1956 if (jifs)
1957 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1958 len += sprintf(page+len, "\n");
1959 return len;
1960 }
1961
1962 static ssize_t
1963 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1964 {
1965 /* timeout can be set at any time */
1966 unsigned long timeout;
1967 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1968 if (rv)
1969 return rv;
1970
1971 /* just to make sure we don't overflow... */
1972 if (timeout >= LONG_MAX / HZ)
1973 return -EINVAL;
1974
1975 timeout = timeout * HZ / 10000;
1976
1977 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1978 timeout = MAX_SCHEDULE_TIMEOUT-1;
1979 if (timeout < 1)
1980 timeout = 1;
1981 mddev->bitmap_info.daemon_sleep = timeout;
1982 if (mddev->thread) {
1983 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1984 * the bitmap is all clean and we don't need to
1985 * adjust the timeout right now
1986 */
1987 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1988 mddev->thread->timeout = timeout;
1989 md_wakeup_thread(mddev->thread);
1990 }
1991 }
1992 return len;
1993 }
1994
1995 static struct md_sysfs_entry bitmap_timeout =
1996 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1997
1998 static ssize_t
1999 backlog_show(struct mddev *mddev, char *page)
2000 {
2001 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2002 }
2003
2004 static ssize_t
2005 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2006 {
2007 unsigned long backlog;
2008 int rv = strict_strtoul(buf, 10, &backlog);
2009 if (rv)
2010 return rv;
2011 if (backlog > COUNTER_MAX)
2012 return -EINVAL;
2013 mddev->bitmap_info.max_write_behind = backlog;
2014 return len;
2015 }
2016
2017 static struct md_sysfs_entry bitmap_backlog =
2018 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2019
2020 static ssize_t
2021 chunksize_show(struct mddev *mddev, char *page)
2022 {
2023 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2024 }
2025
2026 static ssize_t
2027 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2028 {
2029 /* Can only be changed when no bitmap is active */
2030 int rv;
2031 unsigned long csize;
2032 if (mddev->bitmap)
2033 return -EBUSY;
2034 rv = strict_strtoul(buf, 10, &csize);
2035 if (rv)
2036 return rv;
2037 if (csize < 512 ||
2038 !is_power_of_2(csize))
2039 return -EINVAL;
2040 mddev->bitmap_info.chunksize = csize;
2041 return len;
2042 }
2043
2044 static struct md_sysfs_entry bitmap_chunksize =
2045 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2046
2047 static ssize_t metadata_show(struct mddev *mddev, char *page)
2048 {
2049 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2050 ? "external" : "internal"));
2051 }
2052
2053 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2054 {
2055 if (mddev->bitmap ||
2056 mddev->bitmap_info.file ||
2057 mddev->bitmap_info.offset)
2058 return -EBUSY;
2059 if (strncmp(buf, "external", 8) == 0)
2060 mddev->bitmap_info.external = 1;
2061 else if (strncmp(buf, "internal", 8) == 0)
2062 mddev->bitmap_info.external = 0;
2063 else
2064 return -EINVAL;
2065 return len;
2066 }
2067
2068 static struct md_sysfs_entry bitmap_metadata =
2069 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2070
2071 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2072 {
2073 int len;
2074 if (mddev->bitmap)
2075 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2076 "false" : "true"));
2077 else
2078 len = sprintf(page, "\n");
2079 return len;
2080 }
2081
2082 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2083 {
2084 if (mddev->bitmap == NULL)
2085 return -ENOENT;
2086 if (strncmp(buf, "false", 5) == 0)
2087 mddev->bitmap->need_sync = 1;
2088 else if (strncmp(buf, "true", 4) == 0) {
2089 if (mddev->degraded)
2090 return -EBUSY;
2091 mddev->bitmap->need_sync = 0;
2092 } else
2093 return -EINVAL;
2094 return len;
2095 }
2096
2097 static struct md_sysfs_entry bitmap_can_clear =
2098 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2099
2100 static ssize_t
2101 behind_writes_used_show(struct mddev *mddev, char *page)
2102 {
2103 if (mddev->bitmap == NULL)
2104 return sprintf(page, "0\n");
2105 return sprintf(page, "%lu\n",
2106 mddev->bitmap->behind_writes_used);
2107 }
2108
2109 static ssize_t
2110 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2111 {
2112 if (mddev->bitmap)
2113 mddev->bitmap->behind_writes_used = 0;
2114 return len;
2115 }
2116
2117 static struct md_sysfs_entry max_backlog_used =
2118 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2119 behind_writes_used_show, behind_writes_used_reset);
2120
2121 static struct attribute *md_bitmap_attrs[] = {
2122 &bitmap_location.attr,
2123 &bitmap_space.attr,
2124 &bitmap_timeout.attr,
2125 &bitmap_backlog.attr,
2126 &bitmap_chunksize.attr,
2127 &bitmap_metadata.attr,
2128 &bitmap_can_clear.attr,
2129 &max_backlog_used.attr,
2130 NULL
2131 };
2132 struct attribute_group md_bitmap_group = {
2133 .name = "bitmap",
2134 .attrs = md_bitmap_attrs,
2135 };
2136
This page took 0.072304 seconds and 6 git commands to generate.