md/bitmap: use set_bit, test_bit, etc for operation on bitmap->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 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 *bitmap)
736 {
737 struct page **map, *sb_page;
738 unsigned long *attr;
739 int pages;
740 unsigned long flags;
741 struct bitmap_storage *store = &bitmap->storage;
742
743 spin_lock_irqsave(&bitmap->lock, flags);
744 map = store->filemap;
745 store->filemap = NULL;
746 attr = store->filemap_attr;
747 store->filemap_attr = NULL;
748 pages = store->file_pages;
749 store->file_pages = 0;
750 sb_page = store->sb_page;
751 store->sb_page = NULL;
752 spin_unlock_irqrestore(&bitmap->lock, flags);
753
754 while (pages--)
755 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
756 free_buffers(map[pages]);
757 kfree(map);
758 kfree(attr);
759
760 if (sb_page)
761 free_buffers(sb_page);
762 }
763
764 static void bitmap_file_put(struct bitmap *bitmap)
765 {
766 struct file *file;
767 unsigned long flags;
768
769 spin_lock_irqsave(&bitmap->lock, flags);
770 file = bitmap->storage.file;
771 bitmap->storage.file = NULL;
772 spin_unlock_irqrestore(&bitmap->lock, flags);
773
774 if (file)
775 wait_event(bitmap->write_wait,
776 atomic_read(&bitmap->pending_writes)==0);
777 bitmap_file_unmap(bitmap);
778
779 if (file) {
780 struct inode *inode = file->f_path.dentry->d_inode;
781 invalidate_mapping_pages(inode->i_mapping, 0, -1);
782 fput(file);
783 }
784 }
785
786 /*
787 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
788 * then it is no longer reliable, so we stop using it and we mark the file
789 * as failed in the superblock
790 */
791 static void bitmap_file_kick(struct bitmap *bitmap)
792 {
793 char *path, *ptr = NULL;
794
795 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
796 bitmap_update_sb(bitmap);
797
798 if (bitmap->storage.file) {
799 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
800 if (path)
801 ptr = d_path(&bitmap->storage.file->f_path,
802 path, PAGE_SIZE);
803
804 printk(KERN_ALERT
805 "%s: kicking failed bitmap file %s from array!\n",
806 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
807
808 kfree(path);
809 } else
810 printk(KERN_ALERT
811 "%s: disabling internal bitmap due to errors\n",
812 bmname(bitmap));
813 }
814
815 bitmap_file_put(bitmap);
816
817 return;
818 }
819
820 enum bitmap_page_attr {
821 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
822 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
823 * i.e. counter is 1 or 2. */
824 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
825 };
826
827 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
828 enum bitmap_page_attr attr)
829 {
830 __set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
831 }
832
833 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
834 enum bitmap_page_attr attr)
835 {
836 __clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
837 }
838
839 static inline unsigned long test_page_attr(struct bitmap *bitmap, int pnum,
840 enum bitmap_page_attr attr)
841 {
842 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
843 }
844
845 /*
846 * bitmap_file_set_bit -- called before performing a write to the md device
847 * to set (and eventually sync) a particular bit in the bitmap file
848 *
849 * we set the bit immediately, then we record the page number so that
850 * when an unplug occurs, we can flush the dirty pages out to disk
851 */
852 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
853 {
854 unsigned long bit;
855 struct page *page;
856 void *kaddr;
857 unsigned long chunk = block >> bitmap->chunkshift;
858
859 page = filemap_get_page(&bitmap->storage, chunk);
860 if (!page)
861 return;
862 bit = file_page_offset(&bitmap->storage, chunk);
863
864 /* set the bit */
865 kaddr = kmap_atomic(page);
866 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
867 set_bit(bit, kaddr);
868 else
869 __set_bit_le(bit, kaddr);
870 kunmap_atomic(kaddr);
871 pr_debug("set file bit %lu page %lu\n", bit, page->index);
872 /* record page number so it gets flushed to disk when unplug occurs */
873 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
874 }
875
876 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
877 {
878 unsigned long bit;
879 struct page *page;
880 void *paddr;
881 unsigned long chunk = block >> bitmap->chunkshift;
882
883 page = filemap_get_page(&bitmap->storage, chunk);
884 if (!page)
885 return;
886 bit = file_page_offset(&bitmap->storage, chunk);
887 paddr = kmap_atomic(page);
888 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
889 clear_bit(bit, paddr);
890 else
891 __clear_bit_le(bit, paddr);
892 kunmap_atomic(paddr);
893 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
894 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
895 bitmap->allclean = 0;
896 }
897 }
898
899 /* this gets called when the md device is ready to unplug its underlying
900 * (slave) device queues -- before we let any writes go down, we need to
901 * sync the dirty pages of the bitmap file to disk */
902 void bitmap_unplug(struct bitmap *bitmap)
903 {
904 unsigned long i, flags;
905 int dirty, need_write;
906 int wait = 0;
907
908 if (!bitmap || !bitmap->storage.filemap)
909 return;
910
911 /* look at each page to see if there are any set bits that need to be
912 * flushed out to disk */
913 for (i = 0; i < bitmap->storage.file_pages; i++) {
914 spin_lock_irqsave(&bitmap->lock, flags);
915 if (!bitmap->storage.filemap) {
916 spin_unlock_irqrestore(&bitmap->lock, flags);
917 return;
918 }
919 dirty = test_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
920 need_write = test_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
921 clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
922 clear_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
923 if (dirty || need_write)
924 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
925 if (dirty)
926 wait = 1;
927 spin_unlock_irqrestore(&bitmap->lock, flags);
928
929 if (dirty || need_write)
930 write_page(bitmap, bitmap->storage.filemap[i], 0);
931 }
932 if (wait) { /* if any writes were performed, we need to wait on them */
933 if (bitmap->storage.file)
934 wait_event(bitmap->write_wait,
935 atomic_read(&bitmap->pending_writes)==0);
936 else
937 md_super_wait(bitmap->mddev);
938 }
939 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
940 bitmap_file_kick(bitmap);
941 }
942 EXPORT_SYMBOL(bitmap_unplug);
943
944 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
945 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
946 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
947 * memory mapping of the bitmap file
948 * Special cases:
949 * if there's no bitmap file, or if the bitmap file had been
950 * previously kicked from the array, we mark all the bits as
951 * 1's in order to cause a full resync.
952 *
953 * We ignore all bits for sectors that end earlier than 'start'.
954 * This is used when reading an out-of-date bitmap...
955 */
956 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
957 {
958 unsigned long i, chunks, index, oldindex, bit;
959 struct page *page = NULL;
960 unsigned long bit_cnt = 0;
961 struct file *file;
962 unsigned long offset;
963 int outofdate;
964 int ret = -ENOSPC;
965 void *paddr;
966 struct bitmap_storage *store = &bitmap->storage;
967
968 chunks = bitmap->chunks;
969 file = store->file;
970
971 if (!file && !bitmap->mddev->bitmap_info.offset) {
972 /* No permanent bitmap - fill with '1s'. */
973 store->filemap = NULL;
974 store->file_pages = 0;
975 for (i = 0; i < chunks ; i++) {
976 /* if the disk bit is set, set the memory bit */
977 int needed = ((sector_t)(i+1) << (bitmap->chunkshift)
978 >= start);
979 bitmap_set_memory_bits(bitmap,
980 (sector_t)i << bitmap->chunkshift,
981 needed);
982 }
983 return 0;
984 }
985
986 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
987 if (outofdate)
988 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
989 "recovery\n", bmname(bitmap));
990
991 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
992 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
993 bmname(bitmap),
994 (unsigned long) i_size_read(file->f_mapping->host),
995 store->bytes);
996 goto err;
997 }
998
999 oldindex = ~0L;
1000 offset = 0;
1001 if (!bitmap->mddev->bitmap_info.external)
1002 offset = sizeof(bitmap_super_t);
1003
1004 for (i = 0; i < chunks; i++) {
1005 int b;
1006 index = file_page_index(&bitmap->storage, i);
1007 bit = file_page_offset(&bitmap->storage, i);
1008 if (index != oldindex) { /* this is a new page, read it in */
1009 int count;
1010 /* unmap the old page, we're done with it */
1011 if (index == store->file_pages-1)
1012 count = store->bytes - index * PAGE_SIZE;
1013 else
1014 count = PAGE_SIZE;
1015 page = store->filemap[index];
1016 if (file)
1017 ret = read_page(file, index, bitmap,
1018 count, page);
1019 else
1020 ret = read_sb_page(
1021 bitmap->mddev,
1022 bitmap->mddev->bitmap_info.offset,
1023 page,
1024 index, count);
1025
1026 if (ret)
1027 goto err;
1028
1029 oldindex = index;
1030
1031 if (outofdate) {
1032 /*
1033 * if bitmap is out of date, dirty the
1034 * whole page and write it out
1035 */
1036 paddr = kmap_atomic(page);
1037 memset(paddr + offset, 0xff,
1038 PAGE_SIZE - offset);
1039 kunmap_atomic(paddr);
1040 write_page(bitmap, page, 1);
1041
1042 ret = -EIO;
1043 if (test_bit(BITMAP_WRITE_ERROR,
1044 &bitmap->flags))
1045 goto err;
1046 }
1047 }
1048 paddr = kmap_atomic(page);
1049 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1050 b = test_bit(bit, paddr);
1051 else
1052 b = test_bit_le(bit, paddr);
1053 kunmap_atomic(paddr);
1054 if (b) {
1055 /* if the disk bit is set, set the memory bit */
1056 int needed = ((sector_t)(i+1) << bitmap->chunkshift
1057 >= start);
1058 bitmap_set_memory_bits(bitmap,
1059 (sector_t)i << bitmap->chunkshift,
1060 needed);
1061 bit_cnt++;
1062 }
1063 offset = 0;
1064 }
1065
1066 printk(KERN_INFO "%s: bitmap initialized from disk: "
1067 "read %lu pages, set %lu of %lu bits\n",
1068 bmname(bitmap), store->file_pages,
1069 bit_cnt, chunks);
1070
1071 return 0;
1072
1073 err:
1074 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1075 bmname(bitmap), ret);
1076 return ret;
1077 }
1078
1079 void bitmap_write_all(struct bitmap *bitmap)
1080 {
1081 /* We don't actually write all bitmap blocks here,
1082 * just flag them as needing to be written
1083 */
1084 int i;
1085
1086 if (!bitmap || !bitmap->storage.filemap)
1087 return;
1088 if (bitmap->storage.file)
1089 /* Only one copy, so nothing needed */
1090 return;
1091
1092 spin_lock_irq(&bitmap->lock);
1093 for (i = 0; i < bitmap->storage.file_pages; i++)
1094 set_page_attr(bitmap, i,
1095 BITMAP_PAGE_NEEDWRITE);
1096 bitmap->allclean = 0;
1097 spin_unlock_irq(&bitmap->lock);
1098 }
1099
1100 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1101 {
1102 sector_t chunk = offset >> bitmap->chunkshift;
1103 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1104 bitmap->bp[page].count += inc;
1105 bitmap_checkfree(bitmap, page);
1106 }
1107
1108 static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset)
1109 {
1110 sector_t chunk = offset >> bitmap->chunkshift;
1111 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1112 struct bitmap_page *bp = &bitmap->bp[page];
1113
1114 if (!bp->pending)
1115 bp->pending = 1;
1116 }
1117
1118 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1119 sector_t offset, sector_t *blocks,
1120 int create);
1121
1122 /*
1123 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1124 * out to disk
1125 */
1126
1127 void bitmap_daemon_work(struct mddev *mddev)
1128 {
1129 struct bitmap *bitmap;
1130 unsigned long j;
1131 unsigned long nextpage;
1132 unsigned long flags;
1133 sector_t blocks;
1134
1135 /* Use a mutex to guard daemon_work against
1136 * bitmap_destroy.
1137 */
1138 mutex_lock(&mddev->bitmap_info.mutex);
1139 bitmap = mddev->bitmap;
1140 if (bitmap == NULL) {
1141 mutex_unlock(&mddev->bitmap_info.mutex);
1142 return;
1143 }
1144 if (time_before(jiffies, bitmap->daemon_lastrun
1145 + mddev->bitmap_info.daemon_sleep))
1146 goto done;
1147
1148 bitmap->daemon_lastrun = jiffies;
1149 if (bitmap->allclean) {
1150 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1151 goto done;
1152 }
1153 bitmap->allclean = 1;
1154
1155 /* Any file-page which is PENDING now needs to be written.
1156 * So set NEEDWRITE now, then after we make any last-minute changes
1157 * we will write it.
1158 */
1159 spin_lock_irqsave(&bitmap->lock, flags);
1160 for (j = 0; j < bitmap->storage.file_pages; j++)
1161 if (test_page_attr(bitmap, j,
1162 BITMAP_PAGE_PENDING)) {
1163 set_page_attr(bitmap, j,
1164 BITMAP_PAGE_NEEDWRITE);
1165 clear_page_attr(bitmap, j,
1166 BITMAP_PAGE_PENDING);
1167 }
1168
1169 if (bitmap->need_sync &&
1170 mddev->bitmap_info.external == 0) {
1171 /* Arrange for superblock update as well as
1172 * other changes */
1173 bitmap_super_t *sb;
1174 bitmap->need_sync = 0;
1175 if (bitmap->storage.filemap) {
1176 sb = kmap_atomic(bitmap->storage.sb_page);
1177 sb->events_cleared =
1178 cpu_to_le64(bitmap->events_cleared);
1179 kunmap_atomic(sb);
1180 set_page_attr(bitmap, 0,
1181 BITMAP_PAGE_NEEDWRITE);
1182 }
1183 }
1184 /* Now look at the bitmap counters and if any are '2' or '1',
1185 * decrement and handle accordingly.
1186 */
1187 nextpage = 0;
1188 for (j = 0; j < bitmap->chunks; j++) {
1189 bitmap_counter_t *bmc;
1190 sector_t block = (sector_t)j << bitmap->chunkshift;
1191
1192 if (j == nextpage) {
1193 nextpage += PAGE_COUNTER_RATIO;
1194 if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1195 j |= PAGE_COUNTER_MASK;
1196 continue;
1197 }
1198 bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1199 }
1200 bmc = bitmap_get_counter(bitmap,
1201 block,
1202 &blocks, 0);
1203
1204 if (!bmc) {
1205 j |= PAGE_COUNTER_MASK;
1206 continue;
1207 }
1208 if (*bmc == 1 && !bitmap->need_sync) {
1209 /* We can clear the bit */
1210 *bmc = 0;
1211 bitmap_count_page(bitmap, block, -1);
1212 bitmap_file_clear_bit(bitmap, block);
1213 } else if (*bmc && *bmc <= 2) {
1214 *bmc = 1;
1215 bitmap_set_pending(bitmap, block);
1216 bitmap->allclean = 0;
1217 }
1218 }
1219
1220 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1221 * DIRTY pages need to be written by bitmap_unplug so it can wait
1222 * for them.
1223 * If we find any DIRTY page we stop there and let bitmap_unplug
1224 * handle all the rest. This is important in the case where
1225 * the first blocking holds the superblock and it has been updated.
1226 * We mustn't write any other blocks before the superblock.
1227 */
1228 for (j = 0; j < bitmap->storage.file_pages; j++) {
1229
1230 if (test_page_attr(bitmap, j,
1231 BITMAP_PAGE_DIRTY))
1232 /* bitmap_unplug will handle the rest */
1233 break;
1234 if (test_page_attr(bitmap, j,
1235 BITMAP_PAGE_NEEDWRITE)) {
1236 clear_page_attr(bitmap, j,
1237 BITMAP_PAGE_NEEDWRITE);
1238 spin_unlock_irqrestore(&bitmap->lock, flags);
1239 write_page(bitmap, bitmap->storage.filemap[j], 0);
1240 spin_lock_irqsave(&bitmap->lock, flags);
1241 if (!bitmap->storage.filemap)
1242 break;
1243 }
1244 }
1245 spin_unlock_irqrestore(&bitmap->lock, flags);
1246
1247 done:
1248 if (bitmap->allclean == 0)
1249 mddev->thread->timeout =
1250 mddev->bitmap_info.daemon_sleep;
1251 mutex_unlock(&mddev->bitmap_info.mutex);
1252 }
1253
1254 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1255 sector_t offset, sector_t *blocks,
1256 int create)
1257 __releases(bitmap->lock)
1258 __acquires(bitmap->lock)
1259 {
1260 /* If 'create', we might release the lock and reclaim it.
1261 * The lock must have been taken with interrupts enabled.
1262 * If !create, we don't release the lock.
1263 */
1264 sector_t chunk = offset >> bitmap->chunkshift;
1265 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1266 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1267 sector_t csize;
1268 int err;
1269
1270 err = bitmap_checkpage(bitmap, page, create);
1271
1272 if (bitmap->bp[page].hijacked ||
1273 bitmap->bp[page].map == NULL)
1274 csize = ((sector_t)1) << (bitmap->chunkshift +
1275 PAGE_COUNTER_SHIFT - 1);
1276 else
1277 csize = ((sector_t)1) << bitmap->chunkshift;
1278 *blocks = csize - (offset & (csize - 1));
1279
1280 if (err < 0)
1281 return NULL;
1282
1283 /* now locked ... */
1284
1285 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1286 /* should we use the first or second counter field
1287 * of the hijacked pointer? */
1288 int hi = (pageoff > PAGE_COUNTER_MASK);
1289 return &((bitmap_counter_t *)
1290 &bitmap->bp[page].map)[hi];
1291 } else /* page is allocated */
1292 return (bitmap_counter_t *)
1293 &(bitmap->bp[page].map[pageoff]);
1294 }
1295
1296 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1297 {
1298 if (!bitmap)
1299 return 0;
1300
1301 if (behind) {
1302 int bw;
1303 atomic_inc(&bitmap->behind_writes);
1304 bw = atomic_read(&bitmap->behind_writes);
1305 if (bw > bitmap->behind_writes_used)
1306 bitmap->behind_writes_used = bw;
1307
1308 pr_debug("inc write-behind count %d/%lu\n",
1309 bw, bitmap->mddev->bitmap_info.max_write_behind);
1310 }
1311
1312 while (sectors) {
1313 sector_t blocks;
1314 bitmap_counter_t *bmc;
1315
1316 spin_lock_irq(&bitmap->lock);
1317 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1318 if (!bmc) {
1319 spin_unlock_irq(&bitmap->lock);
1320 return 0;
1321 }
1322
1323 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1324 DEFINE_WAIT(__wait);
1325 /* note that it is safe to do the prepare_to_wait
1326 * after the test as long as we do it before dropping
1327 * the spinlock.
1328 */
1329 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1330 TASK_UNINTERRUPTIBLE);
1331 spin_unlock_irq(&bitmap->lock);
1332 io_schedule();
1333 finish_wait(&bitmap->overflow_wait, &__wait);
1334 continue;
1335 }
1336
1337 switch (*bmc) {
1338 case 0:
1339 bitmap_file_set_bit(bitmap, offset);
1340 bitmap_count_page(bitmap, offset, 1);
1341 /* fall through */
1342 case 1:
1343 *bmc = 2;
1344 }
1345
1346 (*bmc)++;
1347
1348 spin_unlock_irq(&bitmap->lock);
1349
1350 offset += blocks;
1351 if (sectors > blocks)
1352 sectors -= blocks;
1353 else
1354 sectors = 0;
1355 }
1356 return 0;
1357 }
1358 EXPORT_SYMBOL(bitmap_startwrite);
1359
1360 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1361 int success, int behind)
1362 {
1363 if (!bitmap)
1364 return;
1365 if (behind) {
1366 if (atomic_dec_and_test(&bitmap->behind_writes))
1367 wake_up(&bitmap->behind_wait);
1368 pr_debug("dec write-behind count %d/%lu\n",
1369 atomic_read(&bitmap->behind_writes),
1370 bitmap->mddev->bitmap_info.max_write_behind);
1371 }
1372
1373 while (sectors) {
1374 sector_t blocks;
1375 unsigned long flags;
1376 bitmap_counter_t *bmc;
1377
1378 spin_lock_irqsave(&bitmap->lock, flags);
1379 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1380 if (!bmc) {
1381 spin_unlock_irqrestore(&bitmap->lock, flags);
1382 return;
1383 }
1384
1385 if (success && !bitmap->mddev->degraded &&
1386 bitmap->events_cleared < bitmap->mddev->events) {
1387 bitmap->events_cleared = bitmap->mddev->events;
1388 bitmap->need_sync = 1;
1389 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1390 }
1391
1392 if (!success && !NEEDED(*bmc))
1393 *bmc |= NEEDED_MASK;
1394
1395 if (COUNTER(*bmc) == COUNTER_MAX)
1396 wake_up(&bitmap->overflow_wait);
1397
1398 (*bmc)--;
1399 if (*bmc <= 2) {
1400 bitmap_set_pending(bitmap, offset);
1401 bitmap->allclean = 0;
1402 }
1403 spin_unlock_irqrestore(&bitmap->lock, flags);
1404 offset += blocks;
1405 if (sectors > blocks)
1406 sectors -= blocks;
1407 else
1408 sectors = 0;
1409 }
1410 }
1411 EXPORT_SYMBOL(bitmap_endwrite);
1412
1413 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1414 int degraded)
1415 {
1416 bitmap_counter_t *bmc;
1417 int rv;
1418 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1419 *blocks = 1024;
1420 return 1; /* always resync if no bitmap */
1421 }
1422 spin_lock_irq(&bitmap->lock);
1423 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1424 rv = 0;
1425 if (bmc) {
1426 /* locked */
1427 if (RESYNC(*bmc))
1428 rv = 1;
1429 else if (NEEDED(*bmc)) {
1430 rv = 1;
1431 if (!degraded) { /* don't set/clear bits if degraded */
1432 *bmc |= RESYNC_MASK;
1433 *bmc &= ~NEEDED_MASK;
1434 }
1435 }
1436 }
1437 spin_unlock_irq(&bitmap->lock);
1438 return rv;
1439 }
1440
1441 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1442 int degraded)
1443 {
1444 /* bitmap_start_sync must always report on multiples of whole
1445 * pages, otherwise resync (which is very PAGE_SIZE based) will
1446 * get confused.
1447 * So call __bitmap_start_sync repeatedly (if needed) until
1448 * At least PAGE_SIZE>>9 blocks are covered.
1449 * Return the 'or' of the result.
1450 */
1451 int rv = 0;
1452 sector_t blocks1;
1453
1454 *blocks = 0;
1455 while (*blocks < (PAGE_SIZE>>9)) {
1456 rv |= __bitmap_start_sync(bitmap, offset,
1457 &blocks1, degraded);
1458 offset += blocks1;
1459 *blocks += blocks1;
1460 }
1461 return rv;
1462 }
1463 EXPORT_SYMBOL(bitmap_start_sync);
1464
1465 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1466 {
1467 bitmap_counter_t *bmc;
1468 unsigned long flags;
1469
1470 if (bitmap == NULL) {
1471 *blocks = 1024;
1472 return;
1473 }
1474 spin_lock_irqsave(&bitmap->lock, flags);
1475 bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1476 if (bmc == NULL)
1477 goto unlock;
1478 /* locked */
1479 if (RESYNC(*bmc)) {
1480 *bmc &= ~RESYNC_MASK;
1481
1482 if (!NEEDED(*bmc) && aborted)
1483 *bmc |= NEEDED_MASK;
1484 else {
1485 if (*bmc <= 2) {
1486 bitmap_set_pending(bitmap, offset);
1487 bitmap->allclean = 0;
1488 }
1489 }
1490 }
1491 unlock:
1492 spin_unlock_irqrestore(&bitmap->lock, flags);
1493 }
1494 EXPORT_SYMBOL(bitmap_end_sync);
1495
1496 void bitmap_close_sync(struct bitmap *bitmap)
1497 {
1498 /* Sync has finished, and any bitmap chunks that weren't synced
1499 * properly have been aborted. It remains to us to clear the
1500 * RESYNC bit wherever it is still on
1501 */
1502 sector_t sector = 0;
1503 sector_t blocks;
1504 if (!bitmap)
1505 return;
1506 while (sector < bitmap->mddev->resync_max_sectors) {
1507 bitmap_end_sync(bitmap, sector, &blocks, 0);
1508 sector += blocks;
1509 }
1510 }
1511 EXPORT_SYMBOL(bitmap_close_sync);
1512
1513 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1514 {
1515 sector_t s = 0;
1516 sector_t blocks;
1517
1518 if (!bitmap)
1519 return;
1520 if (sector == 0) {
1521 bitmap->last_end_sync = jiffies;
1522 return;
1523 }
1524 if (time_before(jiffies, (bitmap->last_end_sync
1525 + bitmap->mddev->bitmap_info.daemon_sleep)))
1526 return;
1527 wait_event(bitmap->mddev->recovery_wait,
1528 atomic_read(&bitmap->mddev->recovery_active) == 0);
1529
1530 bitmap->mddev->curr_resync_completed = sector;
1531 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1532 sector &= ~((1ULL << bitmap->chunkshift) - 1);
1533 s = 0;
1534 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1535 bitmap_end_sync(bitmap, s, &blocks, 0);
1536 s += blocks;
1537 }
1538 bitmap->last_end_sync = jiffies;
1539 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1540 }
1541 EXPORT_SYMBOL(bitmap_cond_end_sync);
1542
1543 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1544 {
1545 /* For each chunk covered by any of these sectors, set the
1546 * counter to 2 and possibly set resync_needed. They should all
1547 * be 0 at this point
1548 */
1549
1550 sector_t secs;
1551 bitmap_counter_t *bmc;
1552 spin_lock_irq(&bitmap->lock);
1553 bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1554 if (!bmc) {
1555 spin_unlock_irq(&bitmap->lock);
1556 return;
1557 }
1558 if (!*bmc) {
1559 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1560 bitmap_count_page(bitmap, offset, 1);
1561 bitmap_set_pending(bitmap, offset);
1562 bitmap->allclean = 0;
1563 }
1564 spin_unlock_irq(&bitmap->lock);
1565 }
1566
1567 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1568 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1569 {
1570 unsigned long chunk;
1571
1572 for (chunk = s; chunk <= e; chunk++) {
1573 sector_t sec = (sector_t)chunk << bitmap->chunkshift;
1574 bitmap_set_memory_bits(bitmap, sec, 1);
1575 spin_lock_irq(&bitmap->lock);
1576 bitmap_file_set_bit(bitmap, sec);
1577 spin_unlock_irq(&bitmap->lock);
1578 if (sec < bitmap->mddev->recovery_cp)
1579 /* We are asserting that the array is dirty,
1580 * so move the recovery_cp address back so
1581 * that it is obvious that it is dirty
1582 */
1583 bitmap->mddev->recovery_cp = sec;
1584 }
1585 }
1586
1587 /*
1588 * flush out any pending updates
1589 */
1590 void bitmap_flush(struct mddev *mddev)
1591 {
1592 struct bitmap *bitmap = mddev->bitmap;
1593 long sleep;
1594
1595 if (!bitmap) /* there was no bitmap */
1596 return;
1597
1598 /* run the daemon_work three time to ensure everything is flushed
1599 * that can be
1600 */
1601 sleep = mddev->bitmap_info.daemon_sleep * 2;
1602 bitmap->daemon_lastrun -= sleep;
1603 bitmap_daemon_work(mddev);
1604 bitmap->daemon_lastrun -= sleep;
1605 bitmap_daemon_work(mddev);
1606 bitmap->daemon_lastrun -= sleep;
1607 bitmap_daemon_work(mddev);
1608 bitmap_update_sb(bitmap);
1609 }
1610
1611 /*
1612 * free memory that was allocated
1613 */
1614 static void bitmap_free(struct bitmap *bitmap)
1615 {
1616 unsigned long k, pages;
1617 struct bitmap_page *bp;
1618
1619 if (!bitmap) /* there was no bitmap */
1620 return;
1621
1622 /* release the bitmap file and kill the daemon */
1623 bitmap_file_put(bitmap);
1624
1625 bp = bitmap->bp;
1626 pages = bitmap->pages;
1627
1628 /* free all allocated memory */
1629
1630 if (bp) /* deallocate the page memory */
1631 for (k = 0; k < pages; k++)
1632 if (bp[k].map && !bp[k].hijacked)
1633 kfree(bp[k].map);
1634 kfree(bp);
1635 kfree(bitmap);
1636 }
1637
1638 void bitmap_destroy(struct mddev *mddev)
1639 {
1640 struct bitmap *bitmap = mddev->bitmap;
1641
1642 if (!bitmap) /* there was no bitmap */
1643 return;
1644
1645 mutex_lock(&mddev->bitmap_info.mutex);
1646 mddev->bitmap = NULL; /* disconnect from the md device */
1647 mutex_unlock(&mddev->bitmap_info.mutex);
1648 if (mddev->thread)
1649 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1650
1651 if (bitmap->sysfs_can_clear)
1652 sysfs_put(bitmap->sysfs_can_clear);
1653
1654 bitmap_free(bitmap);
1655 }
1656
1657 /*
1658 * initialize the bitmap structure
1659 * if this returns an error, bitmap_destroy must be called to do clean up
1660 */
1661 int bitmap_create(struct mddev *mddev)
1662 {
1663 struct bitmap *bitmap;
1664 sector_t blocks = mddev->resync_max_sectors;
1665 unsigned long chunks;
1666 unsigned long pages;
1667 struct file *file = mddev->bitmap_info.file;
1668 int err;
1669 struct sysfs_dirent *bm = NULL;
1670
1671 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1672
1673 BUG_ON(file && mddev->bitmap_info.offset);
1674
1675 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1676 if (!bitmap)
1677 return -ENOMEM;
1678
1679 spin_lock_init(&bitmap->lock);
1680 atomic_set(&bitmap->pending_writes, 0);
1681 init_waitqueue_head(&bitmap->write_wait);
1682 init_waitqueue_head(&bitmap->overflow_wait);
1683 init_waitqueue_head(&bitmap->behind_wait);
1684
1685 bitmap->mddev = mddev;
1686
1687 if (mddev->kobj.sd)
1688 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1689 if (bm) {
1690 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1691 sysfs_put(bm);
1692 } else
1693 bitmap->sysfs_can_clear = NULL;
1694
1695 bitmap->storage.file = file;
1696 if (file) {
1697 get_file(file);
1698 /* As future accesses to this file will use bmap,
1699 * and bypass the page cache, we must sync the file
1700 * first.
1701 */
1702 vfs_fsync(file, 1);
1703 }
1704 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1705 if (!mddev->bitmap_info.external) {
1706 /*
1707 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1708 * instructing us to create a new on-disk bitmap instance.
1709 */
1710 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1711 err = bitmap_new_disk_sb(bitmap);
1712 else
1713 err = bitmap_read_sb(bitmap);
1714 } else {
1715 err = 0;
1716 if (mddev->bitmap_info.chunksize == 0 ||
1717 mddev->bitmap_info.daemon_sleep == 0)
1718 /* chunksize and time_base need to be
1719 * set first. */
1720 err = -EINVAL;
1721 }
1722 if (err)
1723 goto error;
1724
1725 bitmap->daemon_lastrun = jiffies;
1726 bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize)
1727 - BITMAP_BLOCK_SHIFT);
1728
1729 chunks = (blocks + (1 << bitmap->chunkshift) - 1) >>
1730 bitmap->chunkshift;
1731 pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1732
1733 BUG_ON(!pages);
1734
1735 bitmap->chunks = chunks;
1736 bitmap->pages = pages;
1737 bitmap->missing_pages = pages;
1738
1739 bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1740
1741 err = -ENOMEM;
1742 if (!bitmap->bp)
1743 goto error;
1744
1745 if (file || mddev->bitmap_info.offset) {
1746 err = bitmap_storage_alloc(&bitmap->storage, bitmap->chunks,
1747 !mddev->bitmap_info.external);
1748 if (err)
1749 goto error;
1750 }
1751 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1752 pages, bmname(bitmap));
1753
1754 mddev->bitmap = bitmap;
1755
1756
1757 return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1758
1759 error:
1760 bitmap_free(bitmap);
1761 return err;
1762 }
1763
1764 int bitmap_load(struct mddev *mddev)
1765 {
1766 int err = 0;
1767 sector_t start = 0;
1768 sector_t sector = 0;
1769 struct bitmap *bitmap = mddev->bitmap;
1770
1771 if (!bitmap)
1772 goto out;
1773
1774 /* Clear out old bitmap info first: Either there is none, or we
1775 * are resuming after someone else has possibly changed things,
1776 * so we should forget old cached info.
1777 * All chunks should be clean, but some might need_sync.
1778 */
1779 while (sector < mddev->resync_max_sectors) {
1780 sector_t blocks;
1781 bitmap_start_sync(bitmap, sector, &blocks, 0);
1782 sector += blocks;
1783 }
1784 bitmap_close_sync(bitmap);
1785
1786 if (mddev->degraded == 0
1787 || bitmap->events_cleared == mddev->events)
1788 /* no need to keep dirty bits to optimise a
1789 * re-add of a missing device */
1790 start = mddev->recovery_cp;
1791
1792 mutex_lock(&mddev->bitmap_info.mutex);
1793 err = bitmap_init_from_disk(bitmap, start);
1794 mutex_unlock(&mddev->bitmap_info.mutex);
1795
1796 if (err)
1797 goto out;
1798 clear_bit(BITMAP_STALE, &bitmap->flags);
1799
1800 /* Kick recovery in case any bits were set */
1801 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1802
1803 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1804 md_wakeup_thread(mddev->thread);
1805
1806 bitmap_update_sb(bitmap);
1807
1808 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1809 err = -EIO;
1810 out:
1811 return err;
1812 }
1813 EXPORT_SYMBOL_GPL(bitmap_load);
1814
1815 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1816 {
1817 unsigned long chunk_kb;
1818 unsigned long flags;
1819
1820 if (!bitmap)
1821 return;
1822
1823 spin_lock_irqsave(&bitmap->lock, flags);
1824 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1825 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1826 "%lu%s chunk",
1827 bitmap->pages - bitmap->missing_pages,
1828 bitmap->pages,
1829 (bitmap->pages - bitmap->missing_pages)
1830 << (PAGE_SHIFT - 10),
1831 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1832 chunk_kb ? "KB" : "B");
1833 if (bitmap->storage.file) {
1834 seq_printf(seq, ", file: ");
1835 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1836 }
1837
1838 seq_printf(seq, "\n");
1839 spin_unlock_irqrestore(&bitmap->lock, flags);
1840 }
1841
1842 static ssize_t
1843 location_show(struct mddev *mddev, char *page)
1844 {
1845 ssize_t len;
1846 if (mddev->bitmap_info.file)
1847 len = sprintf(page, "file");
1848 else if (mddev->bitmap_info.offset)
1849 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1850 else
1851 len = sprintf(page, "none");
1852 len += sprintf(page+len, "\n");
1853 return len;
1854 }
1855
1856 static ssize_t
1857 location_store(struct mddev *mddev, const char *buf, size_t len)
1858 {
1859
1860 if (mddev->pers) {
1861 if (!mddev->pers->quiesce)
1862 return -EBUSY;
1863 if (mddev->recovery || mddev->sync_thread)
1864 return -EBUSY;
1865 }
1866
1867 if (mddev->bitmap || mddev->bitmap_info.file ||
1868 mddev->bitmap_info.offset) {
1869 /* bitmap already configured. Only option is to clear it */
1870 if (strncmp(buf, "none", 4) != 0)
1871 return -EBUSY;
1872 if (mddev->pers) {
1873 mddev->pers->quiesce(mddev, 1);
1874 bitmap_destroy(mddev);
1875 mddev->pers->quiesce(mddev, 0);
1876 }
1877 mddev->bitmap_info.offset = 0;
1878 if (mddev->bitmap_info.file) {
1879 struct file *f = mddev->bitmap_info.file;
1880 mddev->bitmap_info.file = NULL;
1881 restore_bitmap_write_access(f);
1882 fput(f);
1883 }
1884 } else {
1885 /* No bitmap, OK to set a location */
1886 long long offset;
1887 if (strncmp(buf, "none", 4) == 0)
1888 /* nothing to be done */;
1889 else if (strncmp(buf, "file:", 5) == 0) {
1890 /* Not supported yet */
1891 return -EINVAL;
1892 } else {
1893 int rv;
1894 if (buf[0] == '+')
1895 rv = strict_strtoll(buf+1, 10, &offset);
1896 else
1897 rv = strict_strtoll(buf, 10, &offset);
1898 if (rv)
1899 return rv;
1900 if (offset == 0)
1901 return -EINVAL;
1902 if (mddev->bitmap_info.external == 0 &&
1903 mddev->major_version == 0 &&
1904 offset != mddev->bitmap_info.default_offset)
1905 return -EINVAL;
1906 mddev->bitmap_info.offset = offset;
1907 if (mddev->pers) {
1908 mddev->pers->quiesce(mddev, 1);
1909 rv = bitmap_create(mddev);
1910 if (!rv)
1911 rv = bitmap_load(mddev);
1912 if (rv) {
1913 bitmap_destroy(mddev);
1914 mddev->bitmap_info.offset = 0;
1915 }
1916 mddev->pers->quiesce(mddev, 0);
1917 if (rv)
1918 return rv;
1919 }
1920 }
1921 }
1922 if (!mddev->external) {
1923 /* Ensure new bitmap info is stored in
1924 * metadata promptly.
1925 */
1926 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1927 md_wakeup_thread(mddev->thread);
1928 }
1929 return len;
1930 }
1931
1932 static struct md_sysfs_entry bitmap_location =
1933 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
1934
1935 /* 'bitmap/space' is the space available at 'location' for the
1936 * bitmap. This allows the kernel to know when it is safe to
1937 * resize the bitmap to match a resized array.
1938 */
1939 static ssize_t
1940 space_show(struct mddev *mddev, char *page)
1941 {
1942 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
1943 }
1944
1945 static ssize_t
1946 space_store(struct mddev *mddev, const char *buf, size_t len)
1947 {
1948 unsigned long sectors;
1949 int rv;
1950
1951 rv = kstrtoul(buf, 10, &sectors);
1952 if (rv)
1953 return rv;
1954
1955 if (sectors == 0)
1956 return -EINVAL;
1957
1958 if (mddev->bitmap &&
1959 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
1960 return -EFBIG; /* Bitmap is too big for this small space */
1961
1962 /* could make sure it isn't too big, but that isn't really
1963 * needed - user-space should be careful.
1964 */
1965 mddev->bitmap_info.space = sectors;
1966 return len;
1967 }
1968
1969 static struct md_sysfs_entry bitmap_space =
1970 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
1971
1972 static ssize_t
1973 timeout_show(struct mddev *mddev, char *page)
1974 {
1975 ssize_t len;
1976 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
1977 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
1978
1979 len = sprintf(page, "%lu", secs);
1980 if (jifs)
1981 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
1982 len += sprintf(page+len, "\n");
1983 return len;
1984 }
1985
1986 static ssize_t
1987 timeout_store(struct mddev *mddev, const char *buf, size_t len)
1988 {
1989 /* timeout can be set at any time */
1990 unsigned long timeout;
1991 int rv = strict_strtoul_scaled(buf, &timeout, 4);
1992 if (rv)
1993 return rv;
1994
1995 /* just to make sure we don't overflow... */
1996 if (timeout >= LONG_MAX / HZ)
1997 return -EINVAL;
1998
1999 timeout = timeout * HZ / 10000;
2000
2001 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2002 timeout = MAX_SCHEDULE_TIMEOUT-1;
2003 if (timeout < 1)
2004 timeout = 1;
2005 mddev->bitmap_info.daemon_sleep = timeout;
2006 if (mddev->thread) {
2007 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2008 * the bitmap is all clean and we don't need to
2009 * adjust the timeout right now
2010 */
2011 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2012 mddev->thread->timeout = timeout;
2013 md_wakeup_thread(mddev->thread);
2014 }
2015 }
2016 return len;
2017 }
2018
2019 static struct md_sysfs_entry bitmap_timeout =
2020 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2021
2022 static ssize_t
2023 backlog_show(struct mddev *mddev, char *page)
2024 {
2025 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2026 }
2027
2028 static ssize_t
2029 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2030 {
2031 unsigned long backlog;
2032 int rv = strict_strtoul(buf, 10, &backlog);
2033 if (rv)
2034 return rv;
2035 if (backlog > COUNTER_MAX)
2036 return -EINVAL;
2037 mddev->bitmap_info.max_write_behind = backlog;
2038 return len;
2039 }
2040
2041 static struct md_sysfs_entry bitmap_backlog =
2042 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2043
2044 static ssize_t
2045 chunksize_show(struct mddev *mddev, char *page)
2046 {
2047 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2048 }
2049
2050 static ssize_t
2051 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2052 {
2053 /* Can only be changed when no bitmap is active */
2054 int rv;
2055 unsigned long csize;
2056 if (mddev->bitmap)
2057 return -EBUSY;
2058 rv = strict_strtoul(buf, 10, &csize);
2059 if (rv)
2060 return rv;
2061 if (csize < 512 ||
2062 !is_power_of_2(csize))
2063 return -EINVAL;
2064 mddev->bitmap_info.chunksize = csize;
2065 return len;
2066 }
2067
2068 static struct md_sysfs_entry bitmap_chunksize =
2069 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2070
2071 static ssize_t metadata_show(struct mddev *mddev, char *page)
2072 {
2073 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2074 ? "external" : "internal"));
2075 }
2076
2077 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2078 {
2079 if (mddev->bitmap ||
2080 mddev->bitmap_info.file ||
2081 mddev->bitmap_info.offset)
2082 return -EBUSY;
2083 if (strncmp(buf, "external", 8) == 0)
2084 mddev->bitmap_info.external = 1;
2085 else if (strncmp(buf, "internal", 8) == 0)
2086 mddev->bitmap_info.external = 0;
2087 else
2088 return -EINVAL;
2089 return len;
2090 }
2091
2092 static struct md_sysfs_entry bitmap_metadata =
2093 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2094
2095 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2096 {
2097 int len;
2098 if (mddev->bitmap)
2099 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2100 "false" : "true"));
2101 else
2102 len = sprintf(page, "\n");
2103 return len;
2104 }
2105
2106 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2107 {
2108 if (mddev->bitmap == NULL)
2109 return -ENOENT;
2110 if (strncmp(buf, "false", 5) == 0)
2111 mddev->bitmap->need_sync = 1;
2112 else if (strncmp(buf, "true", 4) == 0) {
2113 if (mddev->degraded)
2114 return -EBUSY;
2115 mddev->bitmap->need_sync = 0;
2116 } else
2117 return -EINVAL;
2118 return len;
2119 }
2120
2121 static struct md_sysfs_entry bitmap_can_clear =
2122 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2123
2124 static ssize_t
2125 behind_writes_used_show(struct mddev *mddev, char *page)
2126 {
2127 if (mddev->bitmap == NULL)
2128 return sprintf(page, "0\n");
2129 return sprintf(page, "%lu\n",
2130 mddev->bitmap->behind_writes_used);
2131 }
2132
2133 static ssize_t
2134 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2135 {
2136 if (mddev->bitmap)
2137 mddev->bitmap->behind_writes_used = 0;
2138 return len;
2139 }
2140
2141 static struct md_sysfs_entry max_backlog_used =
2142 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2143 behind_writes_used_show, behind_writes_used_reset);
2144
2145 static struct attribute *md_bitmap_attrs[] = {
2146 &bitmap_location.attr,
2147 &bitmap_space.attr,
2148 &bitmap_timeout.attr,
2149 &bitmap_backlog.attr,
2150 &bitmap_chunksize.attr,
2151 &bitmap_metadata.attr,
2152 &bitmap_can_clear.attr,
2153 &max_backlog_used.attr,
2154 NULL
2155 };
2156 struct attribute_group md_bitmap_group = {
2157 .name = "bitmap",
2158 .attrs = md_bitmap_attrs,
2159 };
2160
This page took 0.134519 seconds and 6 git commands to generate.