md/bitmap: disentangle two different 'pending' flags.
[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 struct page *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 int did_alloc = 0;
142
143 if (!page) {
144 page = alloc_page(GFP_KERNEL);
145 if (!page)
146 return ERR_PTR(-ENOMEM);
147 did_alloc = 1;
148 }
149
150 rdev_for_each(rdev, mddev) {
151 if (! test_bit(In_sync, &rdev->flags)
152 || test_bit(Faulty, &rdev->flags))
153 continue;
154
155 target = offset + index * (PAGE_SIZE/512);
156
157 if (sync_page_io(rdev, target,
158 roundup(size, bdev_logical_block_size(rdev->bdev)),
159 page, READ, true)) {
160 page->index = index;
161 attach_page_buffers(page, NULL); /* so that free_buffer will
162 * quietly no-op */
163 return page;
164 }
165 }
166 if (did_alloc)
167 put_page(page);
168 return ERR_PTR(-EIO);
169
170 }
171
172 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
173 {
174 /* Iterate the disks of an mddev, using rcu to protect access to the
175 * linked list, and raising the refcount of devices we return to ensure
176 * they don't disappear while in use.
177 * As devices are only added or removed when raid_disk is < 0 and
178 * nr_pending is 0 and In_sync is clear, the entries we return will
179 * still be in the same position on the list when we re-enter
180 * list_for_each_continue_rcu.
181 */
182 struct list_head *pos;
183 rcu_read_lock();
184 if (rdev == NULL)
185 /* start at the beginning */
186 pos = &mddev->disks;
187 else {
188 /* release the previous rdev and start from there. */
189 rdev_dec_pending(rdev, mddev);
190 pos = &rdev->same_set;
191 }
192 list_for_each_continue_rcu(pos, &mddev->disks) {
193 rdev = list_entry(pos, struct md_rdev, same_set);
194 if (rdev->raid_disk >= 0 &&
195 !test_bit(Faulty, &rdev->flags)) {
196 /* this is a usable devices */
197 atomic_inc(&rdev->nr_pending);
198 rcu_read_unlock();
199 return rdev;
200 }
201 }
202 rcu_read_unlock();
203 return NULL;
204 }
205
206 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
207 {
208 struct md_rdev *rdev = NULL;
209 struct block_device *bdev;
210 struct mddev *mddev = bitmap->mddev;
211
212 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
213 int size = PAGE_SIZE;
214 loff_t offset = mddev->bitmap_info.offset;
215
216 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
217
218 if (page->index == bitmap->file_pages-1)
219 size = roundup(bitmap->last_page_size,
220 bdev_logical_block_size(bdev));
221 /* Just make sure we aren't corrupting data or
222 * metadata
223 */
224 if (mddev->external) {
225 /* Bitmap could be anywhere. */
226 if (rdev->sb_start + offset + (page->index
227 * (PAGE_SIZE/512))
228 > rdev->data_offset
229 &&
230 rdev->sb_start + offset
231 < (rdev->data_offset + mddev->dev_sectors
232 + (PAGE_SIZE/512)))
233 goto bad_alignment;
234 } else if (offset < 0) {
235 /* DATA BITMAP METADATA */
236 if (offset
237 + (long)(page->index * (PAGE_SIZE/512))
238 + size/512 > 0)
239 /* bitmap runs in to metadata */
240 goto bad_alignment;
241 if (rdev->data_offset + mddev->dev_sectors
242 > rdev->sb_start + offset)
243 /* data runs in to bitmap */
244 goto bad_alignment;
245 } else if (rdev->sb_start < rdev->data_offset) {
246 /* METADATA BITMAP DATA */
247 if (rdev->sb_start
248 + offset
249 + page->index*(PAGE_SIZE/512) + size/512
250 > rdev->data_offset)
251 /* bitmap runs in to data */
252 goto bad_alignment;
253 } else {
254 /* DATA METADATA BITMAP - no problems */
255 }
256 md_super_write(mddev, rdev,
257 rdev->sb_start + offset
258 + page->index * (PAGE_SIZE/512),
259 size,
260 page);
261 }
262
263 if (wait)
264 md_super_wait(mddev);
265 return 0;
266
267 bad_alignment:
268 return -EINVAL;
269 }
270
271 static void bitmap_file_kick(struct bitmap *bitmap);
272 /*
273 * write out a page to a file
274 */
275 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
276 {
277 struct buffer_head *bh;
278
279 if (bitmap->file == NULL) {
280 switch (write_sb_page(bitmap, page, wait)) {
281 case -EINVAL:
282 bitmap->flags |= BITMAP_WRITE_ERROR;
283 }
284 } else {
285
286 bh = page_buffers(page);
287
288 while (bh && bh->b_blocknr) {
289 atomic_inc(&bitmap->pending_writes);
290 set_buffer_locked(bh);
291 set_buffer_mapped(bh);
292 submit_bh(WRITE | REQ_SYNC, bh);
293 bh = bh->b_this_page;
294 }
295
296 if (wait)
297 wait_event(bitmap->write_wait,
298 atomic_read(&bitmap->pending_writes)==0);
299 }
300 if (bitmap->flags & BITMAP_WRITE_ERROR)
301 bitmap_file_kick(bitmap);
302 }
303
304 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
305 {
306 struct bitmap *bitmap = bh->b_private;
307 unsigned long flags;
308
309 if (!uptodate) {
310 spin_lock_irqsave(&bitmap->lock, flags);
311 bitmap->flags |= BITMAP_WRITE_ERROR;
312 spin_unlock_irqrestore(&bitmap->lock, flags);
313 }
314 if (atomic_dec_and_test(&bitmap->pending_writes))
315 wake_up(&bitmap->write_wait);
316 }
317
318 /* copied from buffer.c */
319 static void
320 __clear_page_buffers(struct page *page)
321 {
322 ClearPagePrivate(page);
323 set_page_private(page, 0);
324 page_cache_release(page);
325 }
326 static void free_buffers(struct page *page)
327 {
328 struct buffer_head *bh = page_buffers(page);
329
330 while (bh) {
331 struct buffer_head *next = bh->b_this_page;
332 free_buffer_head(bh);
333 bh = next;
334 }
335 __clear_page_buffers(page);
336 put_page(page);
337 }
338
339 /* read a page from a file.
340 * We both read the page, and attach buffers to the page to record the
341 * address of each block (using bmap). These addresses will be used
342 * to write the block later, completely bypassing the filesystem.
343 * This usage is similar to how swap files are handled, and allows us
344 * to write to a file with no concerns of memory allocation failing.
345 */
346 static struct page *read_page(struct file *file, unsigned long index,
347 struct bitmap *bitmap,
348 unsigned long count)
349 {
350 struct page *page = NULL;
351 struct inode *inode = file->f_path.dentry->d_inode;
352 struct buffer_head *bh;
353 sector_t block;
354
355 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
356 (unsigned long long)index << PAGE_SHIFT);
357
358 page = alloc_page(GFP_KERNEL);
359 if (!page)
360 page = ERR_PTR(-ENOMEM);
361 if (IS_ERR(page))
362 goto out;
363
364 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
365 if (!bh) {
366 put_page(page);
367 page = ERR_PTR(-ENOMEM);
368 goto out;
369 }
370 attach_page_buffers(page, bh);
371 block = index << (PAGE_SHIFT - inode->i_blkbits);
372 while (bh) {
373 if (count == 0)
374 bh->b_blocknr = 0;
375 else {
376 bh->b_blocknr = bmap(inode, block);
377 if (bh->b_blocknr == 0) {
378 /* Cannot use this file! */
379 free_buffers(page);
380 page = ERR_PTR(-EINVAL);
381 goto out;
382 }
383 bh->b_bdev = inode->i_sb->s_bdev;
384 if (count < (1<<inode->i_blkbits))
385 count = 0;
386 else
387 count -= (1<<inode->i_blkbits);
388
389 bh->b_end_io = end_bitmap_write;
390 bh->b_private = bitmap;
391 atomic_inc(&bitmap->pending_writes);
392 set_buffer_locked(bh);
393 set_buffer_mapped(bh);
394 submit_bh(READ, bh);
395 }
396 block++;
397 bh = bh->b_this_page;
398 }
399 page->index = index;
400
401 wait_event(bitmap->write_wait,
402 atomic_read(&bitmap->pending_writes)==0);
403 if (bitmap->flags & BITMAP_WRITE_ERROR) {
404 free_buffers(page);
405 page = ERR_PTR(-EIO);
406 }
407 out:
408 if (IS_ERR(page))
409 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n",
410 (int)PAGE_SIZE,
411 (unsigned long long)index << PAGE_SHIFT,
412 PTR_ERR(page));
413 return page;
414 }
415
416 /*
417 * bitmap file superblock operations
418 */
419
420 /* update the event counter and sync the superblock to disk */
421 void bitmap_update_sb(struct bitmap *bitmap)
422 {
423 bitmap_super_t *sb;
424
425 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
426 return;
427 if (bitmap->mddev->bitmap_info.external)
428 return;
429 if (!bitmap->sb_page) /* no superblock */
430 return;
431 sb = kmap_atomic(bitmap->sb_page);
432 sb->events = cpu_to_le64(bitmap->mddev->events);
433 if (bitmap->mddev->events < bitmap->events_cleared)
434 /* rocking back to read-only */
435 bitmap->events_cleared = bitmap->mddev->events;
436 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
437 sb->state = cpu_to_le32(bitmap->flags);
438 /* Just in case these have been changed via sysfs: */
439 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
440 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
441 kunmap_atomic(sb);
442 write_page(bitmap, bitmap->sb_page, 1);
443 }
444
445 /* print out the bitmap file superblock */
446 void bitmap_print_sb(struct bitmap *bitmap)
447 {
448 bitmap_super_t *sb;
449
450 if (!bitmap || !bitmap->sb_page)
451 return;
452 sb = kmap_atomic(bitmap->sb_page);
453 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
454 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
455 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
456 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
457 *(__u32 *)(sb->uuid+0),
458 *(__u32 *)(sb->uuid+4),
459 *(__u32 *)(sb->uuid+8),
460 *(__u32 *)(sb->uuid+12));
461 printk(KERN_DEBUG " events: %llu\n",
462 (unsigned long long) le64_to_cpu(sb->events));
463 printk(KERN_DEBUG "events cleared: %llu\n",
464 (unsigned long long) le64_to_cpu(sb->events_cleared));
465 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
466 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
467 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
468 printk(KERN_DEBUG " sync size: %llu KB\n",
469 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
470 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
471 kunmap_atomic(sb);
472 }
473
474 /*
475 * bitmap_new_disk_sb
476 * @bitmap
477 *
478 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
479 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
480 * This function verifies 'bitmap_info' and populates the on-disk bitmap
481 * structure, which is to be written to disk.
482 *
483 * Returns: 0 on success, -Exxx on error
484 */
485 static int bitmap_new_disk_sb(struct bitmap *bitmap)
486 {
487 bitmap_super_t *sb;
488 unsigned long chunksize, daemon_sleep, write_behind;
489 int err = -EINVAL;
490
491 bitmap->sb_page = alloc_page(GFP_KERNEL);
492 if (IS_ERR(bitmap->sb_page)) {
493 err = PTR_ERR(bitmap->sb_page);
494 bitmap->sb_page = NULL;
495 return err;
496 }
497 bitmap->sb_page->index = 0;
498
499 sb = kmap_atomic(bitmap->sb_page);
500
501 sb->magic = cpu_to_le32(BITMAP_MAGIC);
502 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
503
504 chunksize = bitmap->mddev->bitmap_info.chunksize;
505 BUG_ON(!chunksize);
506 if (!is_power_of_2(chunksize)) {
507 kunmap_atomic(sb);
508 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
509 return -EINVAL;
510 }
511 sb->chunksize = cpu_to_le32(chunksize);
512
513 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
514 if (!daemon_sleep ||
515 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
516 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
517 daemon_sleep = 5 * HZ;
518 }
519 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
520 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
521
522 /*
523 * FIXME: write_behind for RAID1. If not specified, what
524 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
525 */
526 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
527 if (write_behind > COUNTER_MAX)
528 write_behind = COUNTER_MAX / 2;
529 sb->write_behind = cpu_to_le32(write_behind);
530 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
531
532 /* keep the array size field of the bitmap superblock up to date */
533 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
534
535 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
536
537 bitmap->flags |= BITMAP_STALE;
538 sb->state |= cpu_to_le32(BITMAP_STALE);
539 bitmap->events_cleared = bitmap->mddev->events;
540 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
541
542 kunmap_atomic(sb);
543
544 return 0;
545 }
546
547 /* read the superblock from the bitmap file and initialize some bitmap fields */
548 static int bitmap_read_sb(struct bitmap *bitmap)
549 {
550 char *reason = NULL;
551 bitmap_super_t *sb;
552 unsigned long chunksize, daemon_sleep, write_behind;
553 unsigned long long events;
554 int err = -EINVAL;
555
556 /* page 0 is the superblock, read it... */
557 if (bitmap->file) {
558 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
559 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
560
561 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
562 } else {
563 bitmap->sb_page = read_sb_page(bitmap->mddev,
564 bitmap->mddev->bitmap_info.offset,
565 NULL,
566 0, sizeof(bitmap_super_t));
567 }
568 if (IS_ERR(bitmap->sb_page)) {
569 err = PTR_ERR(bitmap->sb_page);
570 bitmap->sb_page = NULL;
571 return err;
572 }
573
574 sb = kmap_atomic(bitmap->sb_page);
575
576 chunksize = le32_to_cpu(sb->chunksize);
577 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
578 write_behind = le32_to_cpu(sb->write_behind);
579
580 /* verify that the bitmap-specific fields are valid */
581 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
582 reason = "bad magic";
583 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
584 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
585 reason = "unrecognized superblock version";
586 else if (chunksize < 512)
587 reason = "bitmap chunksize too small";
588 else if (!is_power_of_2(chunksize))
589 reason = "bitmap chunksize not a power of 2";
590 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
591 reason = "daemon sleep period out of range";
592 else if (write_behind > COUNTER_MAX)
593 reason = "write-behind limit out of range (0 - 16383)";
594 if (reason) {
595 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
596 bmname(bitmap), reason);
597 goto out;
598 }
599
600 /* keep the array size field of the bitmap superblock up to date */
601 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
602
603 if (bitmap->mddev->persistent) {
604 /*
605 * We have a persistent array superblock, so compare the
606 * bitmap's UUID and event counter to the mddev's
607 */
608 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
609 printk(KERN_INFO
610 "%s: bitmap superblock UUID mismatch\n",
611 bmname(bitmap));
612 goto out;
613 }
614 events = le64_to_cpu(sb->events);
615 if (events < bitmap->mddev->events) {
616 printk(KERN_INFO
617 "%s: bitmap file is out of date (%llu < %llu) "
618 "-- forcing full recovery\n",
619 bmname(bitmap), events,
620 (unsigned long long) bitmap->mddev->events);
621 sb->state |= cpu_to_le32(BITMAP_STALE);
622 }
623 }
624
625 /* assign fields using values from superblock */
626 bitmap->mddev->bitmap_info.chunksize = chunksize;
627 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
628 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
629 bitmap->flags |= le32_to_cpu(sb->state);
630 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
631 bitmap->flags |= BITMAP_HOSTENDIAN;
632 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
633 if (bitmap->flags & BITMAP_STALE)
634 bitmap->events_cleared = bitmap->mddev->events;
635 err = 0;
636 out:
637 kunmap_atomic(sb);
638 if (err)
639 bitmap_print_sb(bitmap);
640 return err;
641 }
642
643 enum bitmap_mask_op {
644 MASK_SET,
645 MASK_UNSET
646 };
647
648 /* record the state of the bitmap in the superblock. Return the old value */
649 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
650 enum bitmap_mask_op op)
651 {
652 bitmap_super_t *sb;
653 int old;
654
655 if (!bitmap->sb_page) /* can't set the state */
656 return 0;
657 sb = kmap_atomic(bitmap->sb_page);
658 old = le32_to_cpu(sb->state) & bits;
659 switch (op) {
660 case MASK_SET:
661 sb->state |= cpu_to_le32(bits);
662 bitmap->flags |= bits;
663 break;
664 case MASK_UNSET:
665 sb->state &= cpu_to_le32(~bits);
666 bitmap->flags &= ~bits;
667 break;
668 default:
669 BUG();
670 }
671 kunmap_atomic(sb);
672 return old;
673 }
674
675 /*
676 * general bitmap file operations
677 */
678
679 /*
680 * on-disk bitmap:
681 *
682 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
683 * file a page at a time. There's a superblock at the start of the file.
684 */
685 /* calculate the index of the page that contains this bit */
686 static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk)
687 {
688 if (!bitmap->mddev->bitmap_info.external)
689 chunk += sizeof(bitmap_super_t) << 3;
690 return chunk >> PAGE_BIT_SHIFT;
691 }
692
693 /* calculate the (bit) offset of this bit within a page */
694 static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk)
695 {
696 if (!bitmap->mddev->bitmap_info.external)
697 chunk += sizeof(bitmap_super_t) << 3;
698 return chunk & (PAGE_BITS - 1);
699 }
700
701 /*
702 * return a pointer to the page in the filemap that contains the given bit
703 *
704 * this lookup is complicated by the fact that the bitmap sb might be exactly
705 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
706 * 0 or page 1
707 */
708 static inline struct page *filemap_get_page(struct bitmap *bitmap,
709 unsigned long chunk)
710 {
711 if (file_page_index(bitmap, chunk) >= bitmap->file_pages)
712 return NULL;
713 return bitmap->filemap[file_page_index(bitmap, chunk)
714 - file_page_index(bitmap, 0)];
715 }
716
717 static void bitmap_file_unmap(struct bitmap *bitmap)
718 {
719 struct page **map, *sb_page;
720 unsigned long *attr;
721 int pages;
722 unsigned long flags;
723
724 spin_lock_irqsave(&bitmap->lock, flags);
725 map = bitmap->filemap;
726 bitmap->filemap = NULL;
727 attr = bitmap->filemap_attr;
728 bitmap->filemap_attr = NULL;
729 pages = bitmap->file_pages;
730 bitmap->file_pages = 0;
731 sb_page = bitmap->sb_page;
732 bitmap->sb_page = NULL;
733 spin_unlock_irqrestore(&bitmap->lock, flags);
734
735 while (pages--)
736 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
737 free_buffers(map[pages]);
738 kfree(map);
739 kfree(attr);
740
741 if (sb_page)
742 free_buffers(sb_page);
743 }
744
745 static void bitmap_file_put(struct bitmap *bitmap)
746 {
747 struct file *file;
748 unsigned long flags;
749
750 spin_lock_irqsave(&bitmap->lock, flags);
751 file = bitmap->file;
752 bitmap->file = NULL;
753 spin_unlock_irqrestore(&bitmap->lock, flags);
754
755 if (file)
756 wait_event(bitmap->write_wait,
757 atomic_read(&bitmap->pending_writes)==0);
758 bitmap_file_unmap(bitmap);
759
760 if (file) {
761 struct inode *inode = file->f_path.dentry->d_inode;
762 invalidate_mapping_pages(inode->i_mapping, 0, -1);
763 fput(file);
764 }
765 }
766
767 /*
768 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
769 * then it is no longer reliable, so we stop using it and we mark the file
770 * as failed in the superblock
771 */
772 static void bitmap_file_kick(struct bitmap *bitmap)
773 {
774 char *path, *ptr = NULL;
775
776 if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
777 bitmap_update_sb(bitmap);
778
779 if (bitmap->file) {
780 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
781 if (path)
782 ptr = d_path(&bitmap->file->f_path, path,
783 PAGE_SIZE);
784
785 printk(KERN_ALERT
786 "%s: kicking failed bitmap file %s from array!\n",
787 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
788
789 kfree(path);
790 } else
791 printk(KERN_ALERT
792 "%s: disabling internal bitmap due to errors\n",
793 bmname(bitmap));
794 }
795
796 bitmap_file_put(bitmap);
797
798 return;
799 }
800
801 enum bitmap_page_attr {
802 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
803 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
804 * i.e. counter is 1 or 2. */
805 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
806 };
807
808 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
809 enum bitmap_page_attr attr)
810 {
811 __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
812 }
813
814 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
815 enum bitmap_page_attr attr)
816 {
817 __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
818 }
819
820 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
821 enum bitmap_page_attr attr)
822 {
823 return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
824 }
825
826 /*
827 * bitmap_file_set_bit -- called before performing a write to the md device
828 * to set (and eventually sync) a particular bit in the bitmap file
829 *
830 * we set the bit immediately, then we record the page number so that
831 * when an unplug occurs, we can flush the dirty pages out to disk
832 */
833 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
834 {
835 unsigned long bit;
836 struct page *page;
837 void *kaddr;
838 unsigned long chunk = block >> bitmap->chunkshift;
839
840 if (!bitmap->filemap)
841 return;
842
843 page = filemap_get_page(bitmap, chunk);
844 if (!page)
845 return;
846 bit = file_page_offset(bitmap, chunk);
847
848 /* set the bit */
849 kaddr = kmap_atomic(page);
850 if (bitmap->flags & BITMAP_HOSTENDIAN)
851 set_bit(bit, kaddr);
852 else
853 __set_bit_le(bit, kaddr);
854 kunmap_atomic(kaddr);
855 pr_debug("set file bit %lu page %lu\n", bit, page->index);
856 /* record page number so it gets flushed to disk when unplug occurs */
857 set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
858 }
859
860 /* this gets called when the md device is ready to unplug its underlying
861 * (slave) device queues -- before we let any writes go down, we need to
862 * sync the dirty pages of the bitmap file to disk */
863 void bitmap_unplug(struct bitmap *bitmap)
864 {
865 unsigned long i, flags;
866 int dirty, need_write;
867 struct page *page;
868 int wait = 0;
869
870 if (!bitmap)
871 return;
872
873 /* look at each page to see if there are any set bits that need to be
874 * flushed out to disk */
875 for (i = 0; i < bitmap->file_pages; i++) {
876 spin_lock_irqsave(&bitmap->lock, flags);
877 if (!bitmap->filemap) {
878 spin_unlock_irqrestore(&bitmap->lock, flags);
879 return;
880 }
881 page = bitmap->filemap[i];
882 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
883 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
884 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
885 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
886 if (dirty || need_write)
887 clear_page_attr(bitmap, page, BITMAP_PAGE_PENDING);
888 if (dirty)
889 wait = 1;
890 spin_unlock_irqrestore(&bitmap->lock, flags);
891
892 if (dirty || need_write)
893 write_page(bitmap, page, 0);
894 }
895 if (wait) { /* if any writes were performed, we need to wait on them */
896 if (bitmap->file)
897 wait_event(bitmap->write_wait,
898 atomic_read(&bitmap->pending_writes)==0);
899 else
900 md_super_wait(bitmap->mddev);
901 }
902 if (bitmap->flags & BITMAP_WRITE_ERROR)
903 bitmap_file_kick(bitmap);
904 }
905 EXPORT_SYMBOL(bitmap_unplug);
906
907 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
908 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
909 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
910 * memory mapping of the bitmap file
911 * Special cases:
912 * if there's no bitmap file, or if the bitmap file had been
913 * previously kicked from the array, we mark all the bits as
914 * 1's in order to cause a full resync.
915 *
916 * We ignore all bits for sectors that end earlier than 'start'.
917 * This is used when reading an out-of-date bitmap...
918 */
919 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
920 {
921 unsigned long i, chunks, index, oldindex, bit;
922 struct page *page = NULL, *oldpage = NULL;
923 unsigned long num_pages, bit_cnt = 0;
924 struct file *file;
925 unsigned long bytes, offset;
926 int outofdate;
927 int ret = -ENOSPC;
928 void *paddr;
929
930 chunks = bitmap->chunks;
931 file = bitmap->file;
932
933 BUG_ON(!file && !bitmap->mddev->bitmap_info.offset);
934
935 outofdate = bitmap->flags & BITMAP_STALE;
936 if (outofdate)
937 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
938 "recovery\n", bmname(bitmap));
939
940 bytes = DIV_ROUND_UP(bitmap->chunks, 8);
941 if (!bitmap->mddev->bitmap_info.external)
942 bytes += sizeof(bitmap_super_t);
943
944 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
945
946 if (file && i_size_read(file->f_mapping->host) < bytes) {
947 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
948 bmname(bitmap),
949 (unsigned long) i_size_read(file->f_mapping->host),
950 bytes);
951 goto err;
952 }
953
954 ret = -ENOMEM;
955
956 bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
957 if (!bitmap->filemap)
958 goto err;
959
960 /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
961 bitmap->filemap_attr = kzalloc(
962 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
963 GFP_KERNEL);
964 if (!bitmap->filemap_attr)
965 goto err;
966
967 oldindex = ~0L;
968
969 for (i = 0; i < chunks; i++) {
970 int b;
971 index = file_page_index(bitmap, i);
972 bit = file_page_offset(bitmap, i);
973 if (index != oldindex) { /* this is a new page, read it in */
974 int count;
975 /* unmap the old page, we're done with it */
976 if (index == num_pages-1)
977 count = bytes - index * PAGE_SIZE;
978 else
979 count = PAGE_SIZE;
980 if (index == 0 && bitmap->sb_page) {
981 /*
982 * if we're here then the superblock page
983 * contains some bits (PAGE_SIZE != sizeof sb)
984 * we've already read it in, so just use it
985 */
986 page = bitmap->sb_page;
987 offset = sizeof(bitmap_super_t);
988 if (!file)
989 page = read_sb_page(
990 bitmap->mddev,
991 bitmap->mddev->bitmap_info.offset,
992 page,
993 index, count);
994 } else if (file) {
995 page = read_page(file, index, bitmap, count);
996 offset = 0;
997 } else {
998 page = read_sb_page(bitmap->mddev,
999 bitmap->mddev->bitmap_info.offset,
1000 NULL,
1001 index, count);
1002 offset = 0;
1003 }
1004 if (IS_ERR(page)) { /* read error */
1005 ret = PTR_ERR(page);
1006 goto err;
1007 }
1008
1009 oldindex = index;
1010 oldpage = page;
1011
1012 bitmap->filemap[bitmap->file_pages++] = page;
1013 bitmap->last_page_size = count;
1014
1015 if (outofdate) {
1016 /*
1017 * if bitmap is out of date, dirty the
1018 * whole page and write it out
1019 */
1020 paddr = kmap_atomic(page);
1021 memset(paddr + offset, 0xff,
1022 PAGE_SIZE - offset);
1023 kunmap_atomic(paddr);
1024 write_page(bitmap, page, 1);
1025
1026 ret = -EIO;
1027 if (bitmap->flags & BITMAP_WRITE_ERROR)
1028 goto err;
1029 }
1030 }
1031 paddr = kmap_atomic(page);
1032 if (bitmap->flags & BITMAP_HOSTENDIAN)
1033 b = test_bit(bit, paddr);
1034 else
1035 b = test_bit_le(bit, paddr);
1036 kunmap_atomic(paddr);
1037 if (b) {
1038 /* if the disk bit is set, set the memory bit */
1039 int needed = ((sector_t)(i+1) << bitmap->chunkshift
1040 >= start);
1041 bitmap_set_memory_bits(bitmap,
1042 (sector_t)i << bitmap->chunkshift,
1043 needed);
1044 bit_cnt++;
1045 }
1046 }
1047
1048 /* everything went OK */
1049 ret = 0;
1050 bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1051
1052 if (bit_cnt) { /* Kick recovery if any bits were set */
1053 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1054 md_wakeup_thread(bitmap->mddev->thread);
1055 }
1056
1057 printk(KERN_INFO "%s: bitmap initialized from disk: "
1058 "read %lu/%lu pages, set %lu of %lu bits\n",
1059 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, chunks);
1060
1061 return 0;
1062
1063 err:
1064 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1065 bmname(bitmap), ret);
1066 return ret;
1067 }
1068
1069 void bitmap_write_all(struct bitmap *bitmap)
1070 {
1071 /* We don't actually write all bitmap blocks here,
1072 * just flag them as needing to be written
1073 */
1074 int i;
1075
1076 spin_lock_irq(&bitmap->lock);
1077 for (i = 0; i < bitmap->file_pages; i++)
1078 set_page_attr(bitmap, bitmap->filemap[i],
1079 BITMAP_PAGE_NEEDWRITE);
1080 bitmap->allclean = 0;
1081 spin_unlock_irq(&bitmap->lock);
1082 }
1083
1084 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1085 {
1086 sector_t chunk = offset >> bitmap->chunkshift;
1087 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1088 bitmap->bp[page].count += inc;
1089 bitmap_checkfree(bitmap, page);
1090 }
1091
1092 static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset)
1093 {
1094 sector_t chunk = offset >> bitmap->chunkshift;
1095 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1096 struct bitmap_page *bp = &bitmap->bp[page];
1097
1098 if (!bp->pending)
1099 bp->pending = 1;
1100 }
1101
1102 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1103 sector_t offset, sector_t *blocks,
1104 int create);
1105
1106 /*
1107 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1108 * out to disk
1109 */
1110
1111 void bitmap_daemon_work(struct mddev *mddev)
1112 {
1113 struct bitmap *bitmap;
1114 unsigned long j;
1115 unsigned long nextpage;
1116 unsigned long flags;
1117 sector_t blocks;
1118 void *paddr;
1119
1120 /* Use a mutex to guard daemon_work against
1121 * bitmap_destroy.
1122 */
1123 mutex_lock(&mddev->bitmap_info.mutex);
1124 bitmap = mddev->bitmap;
1125 if (bitmap == NULL) {
1126 mutex_unlock(&mddev->bitmap_info.mutex);
1127 return;
1128 }
1129 if (time_before(jiffies, bitmap->daemon_lastrun
1130 + mddev->bitmap_info.daemon_sleep))
1131 goto done;
1132
1133 bitmap->daemon_lastrun = jiffies;
1134 if (bitmap->allclean) {
1135 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1136 goto done;
1137 }
1138 bitmap->allclean = 1;
1139
1140 /* Any file-page which is PENDING now needs to be written.
1141 * So set NEEDWRITE now, then after we make any last-minute changes
1142 * we will write it.
1143 */
1144 spin_lock_irqsave(&bitmap->lock, flags);
1145 if (!bitmap->filemap)
1146 /* error or shutdown */
1147 goto out;
1148
1149 for (j = 0; j < bitmap->file_pages; j++)
1150 if (test_page_attr(bitmap, bitmap->filemap[j],
1151 BITMAP_PAGE_PENDING)) {
1152 set_page_attr(bitmap, bitmap->filemap[j],
1153 BITMAP_PAGE_NEEDWRITE);
1154 clear_page_attr(bitmap, bitmap->filemap[j],
1155 BITMAP_PAGE_PENDING);
1156 }
1157
1158 if (bitmap->need_sync &&
1159 mddev->bitmap_info.external == 0) {
1160 /* Arrange for superblock update as well as
1161 * other changes */
1162 bitmap_super_t *sb;
1163 bitmap->need_sync = 0;
1164 sb = kmap_atomic(bitmap->sb_page);
1165 sb->events_cleared =
1166 cpu_to_le64(bitmap->events_cleared);
1167 kunmap_atomic(sb);
1168 set_page_attr(bitmap, bitmap->sb_page, BITMAP_PAGE_NEEDWRITE);
1169 }
1170 /* Now look at the bitmap counters and if any are '2' or '1',
1171 * decrement and handle accordingly.
1172 */
1173 nextpage = 0;
1174 for (j = 0; j < bitmap->chunks; j++) {
1175 bitmap_counter_t *bmc;
1176
1177 if (j == nextpage) {
1178 nextpage += PAGE_COUNTER_RATIO;
1179 if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1180 j |= PAGE_COUNTER_MASK;
1181 continue;
1182 }
1183 bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1184 }
1185 bmc = bitmap_get_counter(bitmap,
1186 (sector_t)j << bitmap->chunkshift,
1187 &blocks, 0);
1188
1189 if (!bmc) {
1190 j |= PAGE_COUNTER_MASK;
1191 continue;
1192 }
1193 if (*bmc == 1 && !bitmap->need_sync) {
1194 /* We can clear the bit */
1195 struct page *page;
1196 *bmc = 0;
1197 bitmap_count_page(
1198 bitmap,
1199 (sector_t)j << bitmap->chunkshift,
1200 -1);
1201
1202 page = filemap_get_page(bitmap, j);
1203 paddr = kmap_atomic(page);
1204 if (bitmap->flags & BITMAP_HOSTENDIAN)
1205 clear_bit(file_page_offset(bitmap, j),
1206 paddr);
1207 else
1208 __clear_bit_le(file_page_offset(bitmap, j),
1209 paddr);
1210 kunmap_atomic(paddr);
1211 if (!test_page_attr(bitmap, page,
1212 BITMAP_PAGE_NEEDWRITE)) {
1213 set_page_attr(bitmap, page,
1214 BITMAP_PAGE_PENDING);
1215 bitmap->allclean = 0;
1216 }
1217 } else if (*bmc && *bmc <= 2) {
1218 *bmc = 1;
1219 bitmap_set_pending(
1220 bitmap,
1221 (sector_t)j << bitmap->chunkshift);
1222 bitmap->allclean = 0;
1223 }
1224 }
1225
1226 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1227 * DIRTY pages need to be written by bitmap_unplug so it can wait
1228 * for them.
1229 * If we find any DIRTY page we stop there and let bitmap_unplug
1230 * handle all the rest. This is important in the case where
1231 * the first blocking holds the superblock and it has been updated.
1232 * We mustn't write any other blocks before the superblock.
1233 */
1234 for (j = 0; j < bitmap->file_pages; j++) {
1235 struct page *page = bitmap->filemap[j];
1236
1237 if (test_page_attr(bitmap, page,
1238 BITMAP_PAGE_DIRTY))
1239 /* bitmap_unplug will handle the rest */
1240 break;
1241 if (test_page_attr(bitmap, page,
1242 BITMAP_PAGE_NEEDWRITE)) {
1243 clear_page_attr(bitmap, page,
1244 BITMAP_PAGE_NEEDWRITE);
1245 spin_unlock_irqrestore(&bitmap->lock, flags);
1246 write_page(bitmap, page, 0);
1247 spin_lock_irqsave(&bitmap->lock, flags);
1248 if (!bitmap->filemap)
1249 break;
1250 }
1251 }
1252 out:
1253 spin_unlock_irqrestore(&bitmap->lock, flags);
1254
1255 done:
1256 if (bitmap->allclean == 0)
1257 mddev->thread->timeout =
1258 mddev->bitmap_info.daemon_sleep;
1259 mutex_unlock(&mddev->bitmap_info.mutex);
1260 }
1261
1262 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1263 sector_t offset, sector_t *blocks,
1264 int create)
1265 __releases(bitmap->lock)
1266 __acquires(bitmap->lock)
1267 {
1268 /* If 'create', we might release the lock and reclaim it.
1269 * The lock must have been taken with interrupts enabled.
1270 * If !create, we don't release the lock.
1271 */
1272 sector_t chunk = offset >> bitmap->chunkshift;
1273 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1274 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1275 sector_t csize;
1276 int err;
1277
1278 err = bitmap_checkpage(bitmap, page, create);
1279
1280 if (bitmap->bp[page].hijacked ||
1281 bitmap->bp[page].map == NULL)
1282 csize = ((sector_t)1) << (bitmap->chunkshift +
1283 PAGE_COUNTER_SHIFT - 1);
1284 else
1285 csize = ((sector_t)1) << bitmap->chunkshift;
1286 *blocks = csize - (offset & (csize - 1));
1287
1288 if (err < 0)
1289 return NULL;
1290
1291 /* now locked ... */
1292
1293 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1294 /* should we use the first or second counter field
1295 * of the hijacked pointer? */
1296 int hi = (pageoff > PAGE_COUNTER_MASK);
1297 return &((bitmap_counter_t *)
1298 &bitmap->bp[page].map)[hi];
1299 } else /* page is allocated */
1300 return (bitmap_counter_t *)
1301 &(bitmap->bp[page].map[pageoff]);
1302 }
1303
1304 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1305 {
1306 if (!bitmap)
1307 return 0;
1308
1309 if (behind) {
1310 int bw;
1311 atomic_inc(&bitmap->behind_writes);
1312 bw = atomic_read(&bitmap->behind_writes);
1313 if (bw > bitmap->behind_writes_used)
1314 bitmap->behind_writes_used = bw;
1315
1316 pr_debug("inc write-behind count %d/%lu\n",
1317 bw, bitmap->mddev->bitmap_info.max_write_behind);
1318 }
1319
1320 while (sectors) {
1321 sector_t blocks;
1322 bitmap_counter_t *bmc;
1323
1324 spin_lock_irq(&bitmap->lock);
1325 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1326 if (!bmc) {
1327 spin_unlock_irq(&bitmap->lock);
1328 return 0;
1329 }
1330
1331 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1332 DEFINE_WAIT(__wait);
1333 /* note that it is safe to do the prepare_to_wait
1334 * after the test as long as we do it before dropping
1335 * the spinlock.
1336 */
1337 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1338 TASK_UNINTERRUPTIBLE);
1339 spin_unlock_irq(&bitmap->lock);
1340 io_schedule();
1341 finish_wait(&bitmap->overflow_wait, &__wait);
1342 continue;
1343 }
1344
1345 switch (*bmc) {
1346 case 0:
1347 bitmap_file_set_bit(bitmap, offset);
1348 bitmap_count_page(bitmap, offset, 1);
1349 /* fall through */
1350 case 1:
1351 *bmc = 2;
1352 }
1353
1354 (*bmc)++;
1355
1356 spin_unlock_irq(&bitmap->lock);
1357
1358 offset += blocks;
1359 if (sectors > blocks)
1360 sectors -= blocks;
1361 else
1362 sectors = 0;
1363 }
1364 return 0;
1365 }
1366 EXPORT_SYMBOL(bitmap_startwrite);
1367
1368 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1369 int success, int behind)
1370 {
1371 if (!bitmap)
1372 return;
1373 if (behind) {
1374 if (atomic_dec_and_test(&bitmap->behind_writes))
1375 wake_up(&bitmap->behind_wait);
1376 pr_debug("dec write-behind count %d/%lu\n",
1377 atomic_read(&bitmap->behind_writes),
1378 bitmap->mddev->bitmap_info.max_write_behind);
1379 }
1380
1381 while (sectors) {
1382 sector_t blocks;
1383 unsigned long flags;
1384 bitmap_counter_t *bmc;
1385
1386 spin_lock_irqsave(&bitmap->lock, flags);
1387 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1388 if (!bmc) {
1389 spin_unlock_irqrestore(&bitmap->lock, flags);
1390 return;
1391 }
1392
1393 if (success && !bitmap->mddev->degraded &&
1394 bitmap->events_cleared < bitmap->mddev->events) {
1395 bitmap->events_cleared = bitmap->mddev->events;
1396 bitmap->need_sync = 1;
1397 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1398 }
1399
1400 if (!success && !NEEDED(*bmc))
1401 *bmc |= NEEDED_MASK;
1402
1403 if (COUNTER(*bmc) == COUNTER_MAX)
1404 wake_up(&bitmap->overflow_wait);
1405
1406 (*bmc)--;
1407 if (*bmc <= 2) {
1408 bitmap_set_pending(bitmap, offset);
1409 bitmap->allclean = 0;
1410 }
1411 spin_unlock_irqrestore(&bitmap->lock, flags);
1412 offset += blocks;
1413 if (sectors > blocks)
1414 sectors -= blocks;
1415 else
1416 sectors = 0;
1417 }
1418 }
1419 EXPORT_SYMBOL(bitmap_endwrite);
1420
1421 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1422 int degraded)
1423 {
1424 bitmap_counter_t *bmc;
1425 int rv;
1426 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1427 *blocks = 1024;
1428 return 1; /* always resync if no bitmap */
1429 }
1430 spin_lock_irq(&bitmap->lock);
1431 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1432 rv = 0;
1433 if (bmc) {
1434 /* locked */
1435 if (RESYNC(*bmc))
1436 rv = 1;
1437 else if (NEEDED(*bmc)) {
1438 rv = 1;
1439 if (!degraded) { /* don't set/clear bits if degraded */
1440 *bmc |= RESYNC_MASK;
1441 *bmc &= ~NEEDED_MASK;
1442 }
1443 }
1444 }
1445 spin_unlock_irq(&bitmap->lock);
1446 return rv;
1447 }
1448
1449 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1450 int degraded)
1451 {
1452 /* bitmap_start_sync must always report on multiples of whole
1453 * pages, otherwise resync (which is very PAGE_SIZE based) will
1454 * get confused.
1455 * So call __bitmap_start_sync repeatedly (if needed) until
1456 * At least PAGE_SIZE>>9 blocks are covered.
1457 * Return the 'or' of the result.
1458 */
1459 int rv = 0;
1460 sector_t blocks1;
1461
1462 *blocks = 0;
1463 while (*blocks < (PAGE_SIZE>>9)) {
1464 rv |= __bitmap_start_sync(bitmap, offset,
1465 &blocks1, degraded);
1466 offset += blocks1;
1467 *blocks += blocks1;
1468 }
1469 return rv;
1470 }
1471 EXPORT_SYMBOL(bitmap_start_sync);
1472
1473 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1474 {
1475 bitmap_counter_t *bmc;
1476 unsigned long flags;
1477
1478 if (bitmap == NULL) {
1479 *blocks = 1024;
1480 return;
1481 }
1482 spin_lock_irqsave(&bitmap->lock, flags);
1483 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1484 if (bmc == NULL)
1485 goto unlock;
1486 /* locked */
1487 if (RESYNC(*bmc)) {
1488 *bmc &= ~RESYNC_MASK;
1489
1490 if (!NEEDED(*bmc) && aborted)
1491 *bmc |= NEEDED_MASK;
1492 else {
1493 if (*bmc <= 2) {
1494 bitmap_set_pending(bitmap, offset);
1495 bitmap->allclean = 0;
1496 }
1497 }
1498 }
1499 unlock:
1500 spin_unlock_irqrestore(&bitmap->lock, flags);
1501 }
1502 EXPORT_SYMBOL(bitmap_end_sync);
1503
1504 void bitmap_close_sync(struct bitmap *bitmap)
1505 {
1506 /* Sync has finished, and any bitmap chunks that weren't synced
1507 * properly have been aborted. It remains to us to clear the
1508 * RESYNC bit wherever it is still on
1509 */
1510 sector_t sector = 0;
1511 sector_t blocks;
1512 if (!bitmap)
1513 return;
1514 while (sector < bitmap->mddev->resync_max_sectors) {
1515 bitmap_end_sync(bitmap, sector, &blocks, 0);
1516 sector += blocks;
1517 }
1518 }
1519 EXPORT_SYMBOL(bitmap_close_sync);
1520
1521 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1522 {
1523 sector_t s = 0;
1524 sector_t blocks;
1525
1526 if (!bitmap)
1527 return;
1528 if (sector == 0) {
1529 bitmap->last_end_sync = jiffies;
1530 return;
1531 }
1532 if (time_before(jiffies, (bitmap->last_end_sync
1533 + bitmap->mddev->bitmap_info.daemon_sleep)))
1534 return;
1535 wait_event(bitmap->mddev->recovery_wait,
1536 atomic_read(&bitmap->mddev->recovery_active) == 0);
1537
1538 bitmap->mddev->curr_resync_completed = sector;
1539 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1540 sector &= ~((1ULL << bitmap->chunkshift) - 1);
1541 s = 0;
1542 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1543 bitmap_end_sync(bitmap, s, &blocks, 0);
1544 s += blocks;
1545 }
1546 bitmap->last_end_sync = jiffies;
1547 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1548 }
1549 EXPORT_SYMBOL(bitmap_cond_end_sync);
1550
1551 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1552 {
1553 /* For each chunk covered by any of these sectors, set the
1554 * counter to 1 and set resync_needed. They should all
1555 * be 0 at this point
1556 */
1557
1558 sector_t secs;
1559 bitmap_counter_t *bmc;
1560 spin_lock_irq(&bitmap->lock);
1561 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1562 if (!bmc) {
1563 spin_unlock_irq(&bitmap->lock);
1564 return;
1565 }
1566 if (!*bmc) {
1567 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1568 bitmap_count_page(bitmap, offset, 1);
1569 bitmap_set_pending(bitmap, offset);
1570 bitmap->allclean = 0;
1571 }
1572 spin_unlock_irq(&bitmap->lock);
1573 }
1574
1575 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1576 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1577 {
1578 unsigned long chunk;
1579
1580 for (chunk = s; chunk <= e; chunk++) {
1581 sector_t sec = (sector_t)chunk << bitmap->chunkshift;
1582 bitmap_set_memory_bits(bitmap, sec, 1);
1583 spin_lock_irq(&bitmap->lock);
1584 bitmap_file_set_bit(bitmap, sec);
1585 spin_unlock_irq(&bitmap->lock);
1586 if (sec < bitmap->mddev->recovery_cp)
1587 /* We are asserting that the array is dirty,
1588 * so move the recovery_cp address back so
1589 * that it is obvious that it is dirty
1590 */
1591 bitmap->mddev->recovery_cp = sec;
1592 }
1593 }
1594
1595 /*
1596 * flush out any pending updates
1597 */
1598 void bitmap_flush(struct mddev *mddev)
1599 {
1600 struct bitmap *bitmap = mddev->bitmap;
1601 long sleep;
1602
1603 if (!bitmap) /* there was no bitmap */
1604 return;
1605
1606 /* run the daemon_work three time to ensure everything is flushed
1607 * that can be
1608 */
1609 sleep = mddev->bitmap_info.daemon_sleep * 2;
1610 bitmap->daemon_lastrun -= sleep;
1611 bitmap_daemon_work(mddev);
1612 bitmap->daemon_lastrun -= sleep;
1613 bitmap_daemon_work(mddev);
1614 bitmap->daemon_lastrun -= sleep;
1615 bitmap_daemon_work(mddev);
1616 bitmap_update_sb(bitmap);
1617 }
1618
1619 /*
1620 * free memory that was allocated
1621 */
1622 static void bitmap_free(struct bitmap *bitmap)
1623 {
1624 unsigned long k, pages;
1625 struct bitmap_page *bp;
1626
1627 if (!bitmap) /* there was no bitmap */
1628 return;
1629
1630 /* release the bitmap file and kill the daemon */
1631 bitmap_file_put(bitmap);
1632
1633 bp = bitmap->bp;
1634 pages = bitmap->pages;
1635
1636 /* free all allocated memory */
1637
1638 if (bp) /* deallocate the page memory */
1639 for (k = 0; k < pages; k++)
1640 if (bp[k].map && !bp[k].hijacked)
1641 kfree(bp[k].map);
1642 kfree(bp);
1643 kfree(bitmap);
1644 }
1645
1646 void bitmap_destroy(struct mddev *mddev)
1647 {
1648 struct bitmap *bitmap = mddev->bitmap;
1649
1650 if (!bitmap) /* there was no bitmap */
1651 return;
1652
1653 mutex_lock(&mddev->bitmap_info.mutex);
1654 mddev->bitmap = NULL; /* disconnect from the md device */
1655 mutex_unlock(&mddev->bitmap_info.mutex);
1656 if (mddev->thread)
1657 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1658
1659 if (bitmap->sysfs_can_clear)
1660 sysfs_put(bitmap->sysfs_can_clear);
1661
1662 bitmap_free(bitmap);
1663 }
1664
1665 /*
1666 * initialize the bitmap structure
1667 * if this returns an error, bitmap_destroy must be called to do clean up
1668 */
1669 int bitmap_create(struct mddev *mddev)
1670 {
1671 struct bitmap *bitmap;
1672 sector_t blocks = mddev->resync_max_sectors;
1673 unsigned long chunks;
1674 unsigned long pages;
1675 struct file *file = mddev->bitmap_info.file;
1676 int err;
1677 struct sysfs_dirent *bm = NULL;
1678
1679 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1680
1681 if (!file
1682 && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */
1683 return 0;
1684
1685 BUG_ON(file && mddev->bitmap_info.offset);
1686
1687 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1688 if (!bitmap)
1689 return -ENOMEM;
1690
1691 spin_lock_init(&bitmap->lock);
1692 atomic_set(&bitmap->pending_writes, 0);
1693 init_waitqueue_head(&bitmap->write_wait);
1694 init_waitqueue_head(&bitmap->overflow_wait);
1695 init_waitqueue_head(&bitmap->behind_wait);
1696
1697 bitmap->mddev = mddev;
1698
1699 if (mddev->kobj.sd)
1700 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1701 if (bm) {
1702 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1703 sysfs_put(bm);
1704 } else
1705 bitmap->sysfs_can_clear = NULL;
1706
1707 bitmap->file = file;
1708 if (file) {
1709 get_file(file);
1710 /* As future accesses to this file will use bmap,
1711 * and bypass the page cache, we must sync the file
1712 * first.
1713 */
1714 vfs_fsync(file, 1);
1715 }
1716 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1717 if (!mddev->bitmap_info.external) {
1718 /*
1719 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1720 * instructing us to create a new on-disk bitmap instance.
1721 */
1722 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1723 err = bitmap_new_disk_sb(bitmap);
1724 else
1725 err = bitmap_read_sb(bitmap);
1726 } else {
1727 err = 0;
1728 if (mddev->bitmap_info.chunksize == 0 ||
1729 mddev->bitmap_info.daemon_sleep == 0)
1730 /* chunksize and time_base need to be
1731 * set first. */
1732 err = -EINVAL;
1733 }
1734 if (err)
1735 goto error;
1736
1737 bitmap->daemon_lastrun = jiffies;
1738 bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize)
1739 - BITMAP_BLOCK_SHIFT);
1740
1741 chunks = (blocks + (1 << bitmap->chunkshift) - 1) >>
1742 bitmap->chunkshift;
1743 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1744
1745 BUG_ON(!pages);
1746
1747 bitmap->chunks = chunks;
1748 bitmap->pages = pages;
1749 bitmap->missing_pages = pages;
1750
1751 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1752
1753 err = -ENOMEM;
1754 if (!bitmap->bp)
1755 goto error;
1756
1757 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1758 pages, bmname(bitmap));
1759
1760 mddev->bitmap = bitmap;
1761
1762
1763 return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1764
1765 error:
1766 bitmap_free(bitmap);
1767 return err;
1768 }
1769
1770 int bitmap_load(struct mddev *mddev)
1771 {
1772 int err = 0;
1773 sector_t start = 0;
1774 sector_t sector = 0;
1775 struct bitmap *bitmap = mddev->bitmap;
1776
1777 if (!bitmap)
1778 goto out;
1779
1780 /* Clear out old bitmap info first: Either there is none, or we
1781 * are resuming after someone else has possibly changed things,
1782 * so we should forget old cached info.
1783 * All chunks should be clean, but some might need_sync.
1784 */
1785 while (sector < mddev->resync_max_sectors) {
1786 sector_t blocks;
1787 bitmap_start_sync(bitmap, sector, &blocks, 0);
1788 sector += blocks;
1789 }
1790 bitmap_close_sync(bitmap);
1791
1792 if (mddev->degraded == 0
1793 || bitmap->events_cleared == mddev->events)
1794 /* no need to keep dirty bits to optimise a
1795 * re-add of a missing device */
1796 start = mddev->recovery_cp;
1797
1798 mutex_lock(&mddev->bitmap_info.mutex);
1799 err = bitmap_init_from_disk(bitmap, start);
1800 mutex_unlock(&mddev->bitmap_info.mutex);
1801
1802 if (err)
1803 goto out;
1804
1805 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1806 md_wakeup_thread(mddev->thread);
1807
1808 bitmap_update_sb(bitmap);
1809
1810 if (bitmap->flags & BITMAP_WRITE_ERROR)
1811 err = -EIO;
1812 out:
1813 return err;
1814 }
1815 EXPORT_SYMBOL_GPL(bitmap_load);
1816
1817 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1818 {
1819 unsigned long chunk_kb;
1820 unsigned long flags;
1821
1822 if (!bitmap)
1823 return;
1824
1825 spin_lock_irqsave(&bitmap->lock, flags);
1826 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1827 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1828 "%lu%s chunk",
1829 bitmap->pages - bitmap->missing_pages,
1830 bitmap->pages,
1831 (bitmap->pages - bitmap->missing_pages)
1832 << (PAGE_SHIFT - 10),
1833 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1834 chunk_kb ? "KB" : "B");
1835 if (bitmap->file) {
1836 seq_printf(seq, ", file: ");
1837 seq_path(seq, &bitmap->file->f_path, " \t\n");
1838 }
1839
1840 seq_printf(seq, "\n");
1841 spin_unlock_irqrestore(&bitmap->lock, flags);
1842 }
1843
1844 static ssize_t
1845 location_show(struct mddev *mddev, char *page)
1846 {
1847 ssize_t len;
1848 if (mddev->bitmap_info.file)
1849 len = sprintf(page, "file");
1850 else if (mddev->bitmap_info.offset)
1851 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1852 else
1853 len = sprintf(page, "none");
1854 len += sprintf(page+len, "\n");
1855 return len;
1856 }
1857
1858 static ssize_t
1859 location_store(struct mddev *mddev, const char *buf, size_t len)
1860 {
1861
1862 if (mddev->pers) {
1863 if (!mddev->pers->quiesce)
1864 return -EBUSY;
1865 if (mddev->recovery || mddev->sync_thread)
1866 return -EBUSY;
1867 }
1868
1869 if (mddev->bitmap || mddev->bitmap_info.file ||
1870 mddev->bitmap_info.offset) {
1871 /* bitmap already configured. Only option is to clear it */
1872 if (strncmp(buf, "none", 4) != 0)
1873 return -EBUSY;
1874 if (mddev->pers) {
1875 mddev->pers->quiesce(mddev, 1);
1876 bitmap_destroy(mddev);
1877 mddev->pers->quiesce(mddev, 0);
1878 }
1879 mddev->bitmap_info.offset = 0;
1880 if (mddev->bitmap_info.file) {
1881 struct file *f = mddev->bitmap_info.file;
1882 mddev->bitmap_info.file = NULL;
1883 restore_bitmap_write_access(f);
1884 fput(f);
1885 }
1886 } else {
1887 /* No bitmap, OK to set a location */
1888 long long offset;
1889 if (strncmp(buf, "none", 4) == 0)
1890 /* nothing to be done */;
1891 else if (strncmp(buf, "file:", 5) == 0) {
1892 /* Not supported yet */
1893 return -EINVAL;
1894 } else {
1895 int rv;
1896 if (buf[0] == '+')
1897 rv = strict_strtoll(buf+1, 10, &offset);
1898 else
1899 rv = strict_strtoll(buf, 10, &offset);
1900 if (rv)
1901 return rv;
1902 if (offset == 0)
1903 return -EINVAL;
1904 if (mddev->bitmap_info.external == 0 &&
1905 mddev->major_version == 0 &&
1906 offset != mddev->bitmap_info.default_offset)
1907 return -EINVAL;
1908 mddev->bitmap_info.offset = offset;
1909 if (mddev->pers) {
1910 mddev->pers->quiesce(mddev, 1);
1911 rv = bitmap_create(mddev);
1912 if (!rv)
1913 rv = bitmap_load(mddev);
1914 if (rv) {
1915 bitmap_destroy(mddev);
1916 mddev->bitmap_info.offset = 0;
1917 }
1918 mddev->pers->quiesce(mddev, 0);
1919 if (rv)
1920 return rv;
1921 }
1922 }
1923 }
1924 if (!mddev->external) {
1925 /* Ensure new bitmap info is stored in
1926 * metadata promptly.
1927 */
1928 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1929 md_wakeup_thread(mddev->thread);
1930 }
1931 return len;
1932 }
1933
1934 static struct md_sysfs_entry bitmap_location =
1935 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1936
1937 static ssize_t
1938 timeout_show(struct mddev *mddev, char *page)
1939 {
1940 ssize_t len;
1941 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1942 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1943
1944 len = sprintf(page, "%lu", secs);
1945 if (jifs)
1946 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1947 len += sprintf(page+len, "\n");
1948 return len;
1949 }
1950
1951 static ssize_t
1952 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1953 {
1954 /* timeout can be set at any time */
1955 unsigned long timeout;
1956 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1957 if (rv)
1958 return rv;
1959
1960 /* just to make sure we don't overflow... */
1961 if (timeout >= LONG_MAX / HZ)
1962 return -EINVAL;
1963
1964 timeout = timeout * HZ / 10000;
1965
1966 if (timeout >= MAX_SCHEDULE_TIMEOUT)
1967 timeout = MAX_SCHEDULE_TIMEOUT-1;
1968 if (timeout < 1)
1969 timeout = 1;
1970 mddev->bitmap_info.daemon_sleep = timeout;
1971 if (mddev->thread) {
1972 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
1973 * the bitmap is all clean and we don't need to
1974 * adjust the timeout right now
1975 */
1976 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
1977 mddev->thread->timeout = timeout;
1978 md_wakeup_thread(mddev->thread);
1979 }
1980 }
1981 return len;
1982 }
1983
1984 static struct md_sysfs_entry bitmap_timeout =
1985 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
1986
1987 static ssize_t
1988 backlog_show(struct mddev *mddev, char *page)
1989 {
1990 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
1991 }
1992
1993 static ssize_t
1994 backlog_store(struct mddev *mddev, const char *buf, size_t len)
1995 {
1996 unsigned long backlog;
1997 int rv = strict_strtoul(buf, 10, &backlog);
1998 if (rv)
1999 return rv;
2000 if (backlog > COUNTER_MAX)
2001 return -EINVAL;
2002 mddev->bitmap_info.max_write_behind = backlog;
2003 return len;
2004 }
2005
2006 static struct md_sysfs_entry bitmap_backlog =
2007 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2008
2009 static ssize_t
2010 chunksize_show(struct mddev *mddev, char *page)
2011 {
2012 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2013 }
2014
2015 static ssize_t
2016 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2017 {
2018 /* Can only be changed when no bitmap is active */
2019 int rv;
2020 unsigned long csize;
2021 if (mddev->bitmap)
2022 return -EBUSY;
2023 rv = strict_strtoul(buf, 10, &csize);
2024 if (rv)
2025 return rv;
2026 if (csize < 512 ||
2027 !is_power_of_2(csize))
2028 return -EINVAL;
2029 mddev->bitmap_info.chunksize = csize;
2030 return len;
2031 }
2032
2033 static struct md_sysfs_entry bitmap_chunksize =
2034 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2035
2036 static ssize_t metadata_show(struct mddev *mddev, char *page)
2037 {
2038 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2039 ? "external" : "internal"));
2040 }
2041
2042 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2043 {
2044 if (mddev->bitmap ||
2045 mddev->bitmap_info.file ||
2046 mddev->bitmap_info.offset)
2047 return -EBUSY;
2048 if (strncmp(buf, "external", 8) == 0)
2049 mddev->bitmap_info.external = 1;
2050 else if (strncmp(buf, "internal", 8) == 0)
2051 mddev->bitmap_info.external = 0;
2052 else
2053 return -EINVAL;
2054 return len;
2055 }
2056
2057 static struct md_sysfs_entry bitmap_metadata =
2058 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2059
2060 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2061 {
2062 int len;
2063 if (mddev->bitmap)
2064 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2065 "false" : "true"));
2066 else
2067 len = sprintf(page, "\n");
2068 return len;
2069 }
2070
2071 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2072 {
2073 if (mddev->bitmap == NULL)
2074 return -ENOENT;
2075 if (strncmp(buf, "false", 5) == 0)
2076 mddev->bitmap->need_sync = 1;
2077 else if (strncmp(buf, "true", 4) == 0) {
2078 if (mddev->degraded)
2079 return -EBUSY;
2080 mddev->bitmap->need_sync = 0;
2081 } else
2082 return -EINVAL;
2083 return len;
2084 }
2085
2086 static struct md_sysfs_entry bitmap_can_clear =
2087 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2088
2089 static ssize_t
2090 behind_writes_used_show(struct mddev *mddev, char *page)
2091 {
2092 if (mddev->bitmap == NULL)
2093 return sprintf(page, "0\n");
2094 return sprintf(page, "%lu\n",
2095 mddev->bitmap->behind_writes_used);
2096 }
2097
2098 static ssize_t
2099 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2100 {
2101 if (mddev->bitmap)
2102 mddev->bitmap->behind_writes_used = 0;
2103 return len;
2104 }
2105
2106 static struct md_sysfs_entry max_backlog_used =
2107 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2108 behind_writes_used_show, behind_writes_used_reset);
2109
2110 static struct attribute *md_bitmap_attrs[] = {
2111 &bitmap_location.attr,
2112 &bitmap_timeout.attr,
2113 &bitmap_backlog.attr,
2114 &bitmap_chunksize.attr,
2115 &bitmap_metadata.attr,
2116 &bitmap_can_clear.attr,
2117 &max_backlog_used.attr,
2118 NULL
2119 };
2120 struct attribute_group md_bitmap_group = {
2121 .name = "bitmap",
2122 .attrs = md_bitmap_attrs,
2123 };
2124
This page took 0.070615 seconds and 6 git commands to generate.