md/bitmap: disable bitmap_resize for file-backed bitmaps.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / bitmap.c
1 /*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
10 */
11
12 /*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
16 */
17
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
30 #include "md.h"
31 #include "bitmap.h"
32
33 static inline char *bmname(struct bitmap *bitmap)
34 {
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
36 }
37
38 /*
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
40 *
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
44 *
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
47 */
48 static int bitmap_checkpage(struct bitmap_counts *bitmap,
49 unsigned long page, int create)
50 __releases(bitmap->lock)
51 __acquires(bitmap->lock)
52 {
53 unsigned char *mappage;
54
55 if (page >= bitmap->pages) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
58 * It is harmless.
59 */
60 return -EINVAL;
61 }
62
63 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
64 return 0;
65
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
67 return 0;
68
69 if (!create)
70 return -ENOENT;
71
72 /* this page has not been allocated yet */
73
74 spin_unlock_irq(&bitmap->lock);
75 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
76 spin_lock_irq(&bitmap->lock);
77
78 if (mappage == NULL) {
79 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
80 /* failed - set the hijacked flag so that we can use the
81 * pointer as a counter */
82 if (!bitmap->bp[page].map)
83 bitmap->bp[page].hijacked = 1;
84 } else if (bitmap->bp[page].map ||
85 bitmap->bp[page].hijacked) {
86 /* somebody beat us to getting the page */
87 kfree(mappage);
88 return 0;
89 } else {
90
91 /* no page was in place and we have one, so install it */
92
93 bitmap->bp[page].map = mappage;
94 bitmap->missing_pages--;
95 }
96 return 0;
97 }
98
99 /* if page is completely empty, put it back on the free list, or dealloc it */
100 /* if page was hijacked, unmark the flag so it might get alloced next time */
101 /* Note: lock should be held when calling this */
102 static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
103 {
104 char *ptr;
105
106 if (bitmap->bp[page].count) /* page is still busy */
107 return;
108
109 /* page is no longer in use, it can be released */
110
111 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
112 bitmap->bp[page].hijacked = 0;
113 bitmap->bp[page].map = NULL;
114 } else {
115 /* normal case, free the page */
116 ptr = bitmap->bp[page].map;
117 bitmap->bp[page].map = NULL;
118 bitmap->missing_pages++;
119 kfree(ptr);
120 }
121 }
122
123 /*
124 * bitmap file handling - read and write the bitmap file and its superblock
125 */
126
127 /*
128 * basic page I/O operations
129 */
130
131 /* IO operations when bitmap is stored near all superblocks */
132 static int read_sb_page(struct mddev *mddev, loff_t offset,
133 struct page *page,
134 unsigned long index, int size)
135 {
136 /* choose a good rdev and read the page from there */
137
138 struct md_rdev *rdev;
139 sector_t target;
140
141 rdev_for_each(rdev, mddev) {
142 if (! test_bit(In_sync, &rdev->flags)
143 || test_bit(Faulty, &rdev->flags))
144 continue;
145
146 target = offset + index * (PAGE_SIZE/512);
147
148 if (sync_page_io(rdev, target,
149 roundup(size, bdev_logical_block_size(rdev->bdev)),
150 page, READ, true)) {
151 page->index = index;
152 return 0;
153 }
154 }
155 return -EIO;
156 }
157
158 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
159 {
160 /* Iterate the disks of an mddev, using rcu to protect access to the
161 * linked list, and raising the refcount of devices we return to ensure
162 * they don't disappear while in use.
163 * As devices are only added or removed when raid_disk is < 0 and
164 * nr_pending is 0 and In_sync is clear, the entries we return will
165 * still be in the same position on the list when we re-enter
166 * list_for_each_entry_continue_rcu.
167 */
168 rcu_read_lock();
169 if (rdev == NULL)
170 /* start at the beginning */
171 rdev = list_entry_rcu(&mddev->disks, struct md_rdev, same_set);
172 else {
173 /* release the previous rdev and start from there. */
174 rdev_dec_pending(rdev, mddev);
175 }
176 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
177 if (rdev->raid_disk >= 0 &&
178 !test_bit(Faulty, &rdev->flags)) {
179 /* this is a usable devices */
180 atomic_inc(&rdev->nr_pending);
181 rcu_read_unlock();
182 return rdev;
183 }
184 }
185 rcu_read_unlock();
186 return NULL;
187 }
188
189 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
190 {
191 struct md_rdev *rdev = NULL;
192 struct block_device *bdev;
193 struct mddev *mddev = bitmap->mddev;
194 struct bitmap_storage *store = &bitmap->storage;
195
196 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
197 int size = PAGE_SIZE;
198 loff_t offset = mddev->bitmap_info.offset;
199
200 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
201
202 if (page->index == store->file_pages-1) {
203 int last_page_size = store->bytes & (PAGE_SIZE-1);
204 if (last_page_size == 0)
205 last_page_size = PAGE_SIZE;
206 size = roundup(last_page_size,
207 bdev_logical_block_size(bdev));
208 }
209 /* Just make sure we aren't corrupting data or
210 * metadata
211 */
212 if (mddev->external) {
213 /* Bitmap could be anywhere. */
214 if (rdev->sb_start + offset + (page->index
215 * (PAGE_SIZE/512))
216 > rdev->data_offset
217 &&
218 rdev->sb_start + offset
219 < (rdev->data_offset + mddev->dev_sectors
220 + (PAGE_SIZE/512)))
221 goto bad_alignment;
222 } else if (offset < 0) {
223 /* DATA BITMAP METADATA */
224 if (offset
225 + (long)(page->index * (PAGE_SIZE/512))
226 + size/512 > 0)
227 /* bitmap runs in to metadata */
228 goto bad_alignment;
229 if (rdev->data_offset + mddev->dev_sectors
230 > rdev->sb_start + offset)
231 /* data runs in to bitmap */
232 goto bad_alignment;
233 } else if (rdev->sb_start < rdev->data_offset) {
234 /* METADATA BITMAP DATA */
235 if (rdev->sb_start
236 + offset
237 + page->index*(PAGE_SIZE/512) + size/512
238 > rdev->data_offset)
239 /* bitmap runs in to data */
240 goto bad_alignment;
241 } else {
242 /* DATA METADATA BITMAP - no problems */
243 }
244 md_super_write(mddev, rdev,
245 rdev->sb_start + offset
246 + page->index * (PAGE_SIZE/512),
247 size,
248 page);
249 }
250
251 if (wait)
252 md_super_wait(mddev);
253 return 0;
254
255 bad_alignment:
256 return -EINVAL;
257 }
258
259 static void bitmap_file_kick(struct bitmap *bitmap);
260 /*
261 * write out a page to a file
262 */
263 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
264 {
265 struct buffer_head *bh;
266
267 if (bitmap->storage.file == NULL) {
268 switch (write_sb_page(bitmap, page, wait)) {
269 case -EINVAL:
270 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
271 }
272 } else {
273
274 bh = page_buffers(page);
275
276 while (bh && bh->b_blocknr) {
277 atomic_inc(&bitmap->pending_writes);
278 set_buffer_locked(bh);
279 set_buffer_mapped(bh);
280 submit_bh(WRITE | REQ_SYNC, bh);
281 bh = bh->b_this_page;
282 }
283
284 if (wait)
285 wait_event(bitmap->write_wait,
286 atomic_read(&bitmap->pending_writes)==0);
287 }
288 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
289 bitmap_file_kick(bitmap);
290 }
291
292 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
293 {
294 struct bitmap *bitmap = bh->b_private;
295
296 if (!uptodate)
297 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
298 if (atomic_dec_and_test(&bitmap->pending_writes))
299 wake_up(&bitmap->write_wait);
300 }
301
302 /* copied from buffer.c */
303 static void
304 __clear_page_buffers(struct page *page)
305 {
306 ClearPagePrivate(page);
307 set_page_private(page, 0);
308 page_cache_release(page);
309 }
310 static void free_buffers(struct page *page)
311 {
312 struct buffer_head *bh;
313
314 if (!PagePrivate(page))
315 return;
316
317 bh = page_buffers(page);
318 while (bh) {
319 struct buffer_head *next = bh->b_this_page;
320 free_buffer_head(bh);
321 bh = next;
322 }
323 __clear_page_buffers(page);
324 put_page(page);
325 }
326
327 /* read a page from a file.
328 * We both read the page, and attach buffers to the page to record the
329 * address of each block (using bmap). These addresses will be used
330 * to write the block later, completely bypassing the filesystem.
331 * This usage is similar to how swap files are handled, and allows us
332 * to write to a file with no concerns of memory allocation failing.
333 */
334 static int read_page(struct file *file, unsigned long index,
335 struct bitmap *bitmap,
336 unsigned long count,
337 struct page *page)
338 {
339 int ret = 0;
340 struct inode *inode = file_inode(file);
341 struct buffer_head *bh;
342 sector_t block;
343
344 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
345 (unsigned long long)index << PAGE_SHIFT);
346
347 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
348 if (!bh) {
349 ret = -ENOMEM;
350 goto out;
351 }
352 attach_page_buffers(page, bh);
353 block = index << (PAGE_SHIFT - inode->i_blkbits);
354 while (bh) {
355 if (count == 0)
356 bh->b_blocknr = 0;
357 else {
358 bh->b_blocknr = bmap(inode, block);
359 if (bh->b_blocknr == 0) {
360 /* Cannot use this file! */
361 ret = -EINVAL;
362 goto out;
363 }
364 bh->b_bdev = inode->i_sb->s_bdev;
365 if (count < (1<<inode->i_blkbits))
366 count = 0;
367 else
368 count -= (1<<inode->i_blkbits);
369
370 bh->b_end_io = end_bitmap_write;
371 bh->b_private = bitmap;
372 atomic_inc(&bitmap->pending_writes);
373 set_buffer_locked(bh);
374 set_buffer_mapped(bh);
375 submit_bh(READ, bh);
376 }
377 block++;
378 bh = bh->b_this_page;
379 }
380 page->index = index;
381
382 wait_event(bitmap->write_wait,
383 atomic_read(&bitmap->pending_writes)==0);
384 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
385 ret = -EIO;
386 out:
387 if (ret)
388 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
389 (int)PAGE_SIZE,
390 (unsigned long long)index << PAGE_SHIFT,
391 ret);
392 return ret;
393 }
394
395 /*
396 * bitmap file superblock operations
397 */
398
399 /* update the event counter and sync the superblock to disk */
400 void bitmap_update_sb(struct bitmap *bitmap)
401 {
402 bitmap_super_t *sb;
403
404 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
405 return;
406 if (bitmap->mddev->bitmap_info.external)
407 return;
408 if (!bitmap->storage.sb_page) /* no superblock */
409 return;
410 sb = kmap_atomic(bitmap->storage.sb_page);
411 sb->events = cpu_to_le64(bitmap->mddev->events);
412 if (bitmap->mddev->events < bitmap->events_cleared)
413 /* rocking back to read-only */
414 bitmap->events_cleared = bitmap->mddev->events;
415 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
416 sb->state = cpu_to_le32(bitmap->flags);
417 /* Just in case these have been changed via sysfs: */
418 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
419 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
420 /* This might have been changed by a reshape */
421 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
422 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
423 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
424 bitmap_info.space);
425 kunmap_atomic(sb);
426 write_page(bitmap, bitmap->storage.sb_page, 1);
427 }
428
429 /* print out the bitmap file superblock */
430 void bitmap_print_sb(struct bitmap *bitmap)
431 {
432 bitmap_super_t *sb;
433
434 if (!bitmap || !bitmap->storage.sb_page)
435 return;
436 sb = kmap_atomic(bitmap->storage.sb_page);
437 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
438 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
439 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
440 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
441 *(__u32 *)(sb->uuid+0),
442 *(__u32 *)(sb->uuid+4),
443 *(__u32 *)(sb->uuid+8),
444 *(__u32 *)(sb->uuid+12));
445 printk(KERN_DEBUG " events: %llu\n",
446 (unsigned long long) le64_to_cpu(sb->events));
447 printk(KERN_DEBUG "events cleared: %llu\n",
448 (unsigned long long) le64_to_cpu(sb->events_cleared));
449 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
450 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
451 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
452 printk(KERN_DEBUG " sync size: %llu KB\n",
453 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
454 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
455 kunmap_atomic(sb);
456 }
457
458 /*
459 * bitmap_new_disk_sb
460 * @bitmap
461 *
462 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
463 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
464 * This function verifies 'bitmap_info' and populates the on-disk bitmap
465 * structure, which is to be written to disk.
466 *
467 * Returns: 0 on success, -Exxx on error
468 */
469 static int bitmap_new_disk_sb(struct bitmap *bitmap)
470 {
471 bitmap_super_t *sb;
472 unsigned long chunksize, daemon_sleep, write_behind;
473
474 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
475 if (bitmap->storage.sb_page == NULL)
476 return -ENOMEM;
477 bitmap->storage.sb_page->index = 0;
478
479 sb = kmap_atomic(bitmap->storage.sb_page);
480
481 sb->magic = cpu_to_le32(BITMAP_MAGIC);
482 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
483
484 chunksize = bitmap->mddev->bitmap_info.chunksize;
485 BUG_ON(!chunksize);
486 if (!is_power_of_2(chunksize)) {
487 kunmap_atomic(sb);
488 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
489 return -EINVAL;
490 }
491 sb->chunksize = cpu_to_le32(chunksize);
492
493 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
494 if (!daemon_sleep ||
495 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
496 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
497 daemon_sleep = 5 * HZ;
498 }
499 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
500 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
501
502 /*
503 * FIXME: write_behind for RAID1. If not specified, what
504 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
505 */
506 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
507 if (write_behind > COUNTER_MAX)
508 write_behind = COUNTER_MAX / 2;
509 sb->write_behind = cpu_to_le32(write_behind);
510 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
511
512 /* keep the array size field of the bitmap superblock up to date */
513 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
514
515 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
516
517 set_bit(BITMAP_STALE, &bitmap->flags);
518 sb->state = cpu_to_le32(bitmap->flags);
519 bitmap->events_cleared = bitmap->mddev->events;
520 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
521
522 kunmap_atomic(sb);
523
524 return 0;
525 }
526
527 /* read the superblock from the bitmap file and initialize some bitmap fields */
528 static int bitmap_read_sb(struct bitmap *bitmap)
529 {
530 char *reason = NULL;
531 bitmap_super_t *sb;
532 unsigned long chunksize, daemon_sleep, write_behind;
533 unsigned long long events;
534 unsigned long sectors_reserved = 0;
535 int err = -EINVAL;
536 struct page *sb_page;
537
538 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
539 chunksize = 128 * 1024 * 1024;
540 daemon_sleep = 5 * HZ;
541 write_behind = 0;
542 set_bit(BITMAP_STALE, &bitmap->flags);
543 err = 0;
544 goto out_no_sb;
545 }
546 /* page 0 is the superblock, read it... */
547 sb_page = alloc_page(GFP_KERNEL);
548 if (!sb_page)
549 return -ENOMEM;
550 bitmap->storage.sb_page = sb_page;
551
552 if (bitmap->storage.file) {
553 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
554 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
555
556 err = read_page(bitmap->storage.file, 0,
557 bitmap, bytes, sb_page);
558 } else {
559 err = read_sb_page(bitmap->mddev,
560 bitmap->mddev->bitmap_info.offset,
561 sb_page,
562 0, sizeof(bitmap_super_t));
563 }
564 if (err)
565 return err;
566
567 err = -EINVAL;
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 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
575
576 /* verify that the bitmap-specific fields are valid */
577 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
578 reason = "bad magic";
579 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
580 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
581 reason = "unrecognized superblock version";
582 else if (chunksize < 512)
583 reason = "bitmap chunksize too small";
584 else if (!is_power_of_2(chunksize))
585 reason = "bitmap chunksize not a power of 2";
586 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
587 reason = "daemon sleep period out of range";
588 else if (write_behind > COUNTER_MAX)
589 reason = "write-behind limit out of range (0 - 16383)";
590 if (reason) {
591 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
592 bmname(bitmap), reason);
593 goto out;
594 }
595
596 /* keep the array size field of the bitmap superblock up to date */
597 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
598
599 if (bitmap->mddev->persistent) {
600 /*
601 * We have a persistent array superblock, so compare the
602 * bitmap's UUID and event counter to the mddev's
603 */
604 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
605 printk(KERN_INFO
606 "%s: bitmap superblock UUID mismatch\n",
607 bmname(bitmap));
608 goto out;
609 }
610 events = le64_to_cpu(sb->events);
611 if (events < bitmap->mddev->events) {
612 printk(KERN_INFO
613 "%s: bitmap file is out of date (%llu < %llu) "
614 "-- forcing full recovery\n",
615 bmname(bitmap), events,
616 (unsigned long long) bitmap->mddev->events);
617 set_bit(BITMAP_STALE, &bitmap->flags);
618 }
619 }
620
621 /* assign fields using values from superblock */
622 bitmap->flags |= le32_to_cpu(sb->state);
623 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
624 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
625 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
626 err = 0;
627 out:
628 kunmap_atomic(sb);
629 out_no_sb:
630 if (test_bit(BITMAP_STALE, &bitmap->flags))
631 bitmap->events_cleared = bitmap->mddev->events;
632 bitmap->mddev->bitmap_info.chunksize = chunksize;
633 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
634 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
635 if (bitmap->mddev->bitmap_info.space == 0 ||
636 bitmap->mddev->bitmap_info.space > sectors_reserved)
637 bitmap->mddev->bitmap_info.space = sectors_reserved;
638 if (err)
639 bitmap_print_sb(bitmap);
640 return err;
641 }
642
643 /*
644 * general bitmap file operations
645 */
646
647 /*
648 * on-disk bitmap:
649 *
650 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
651 * file a page at a time. There's a superblock at the start of the file.
652 */
653 /* calculate the index of the page that contains this bit */
654 static inline unsigned long file_page_index(struct bitmap_storage *store,
655 unsigned long chunk)
656 {
657 if (store->sb_page)
658 chunk += sizeof(bitmap_super_t) << 3;
659 return chunk >> PAGE_BIT_SHIFT;
660 }
661
662 /* calculate the (bit) offset of this bit within a page */
663 static inline unsigned long file_page_offset(struct bitmap_storage *store,
664 unsigned long chunk)
665 {
666 if (store->sb_page)
667 chunk += sizeof(bitmap_super_t) << 3;
668 return chunk & (PAGE_BITS - 1);
669 }
670
671 /*
672 * return a pointer to the page in the filemap that contains the given bit
673 *
674 * this lookup is complicated by the fact that the bitmap sb might be exactly
675 * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
676 * 0 or page 1
677 */
678 static inline struct page *filemap_get_page(struct bitmap_storage *store,
679 unsigned long chunk)
680 {
681 if (file_page_index(store, chunk) >= store->file_pages)
682 return NULL;
683 return store->filemap[file_page_index(store, chunk)
684 - file_page_index(store, 0)];
685 }
686
687 static int bitmap_storage_alloc(struct bitmap_storage *store,
688 unsigned long chunks, int with_super)
689 {
690 int pnum;
691 unsigned long num_pages;
692 unsigned long bytes;
693
694 bytes = DIV_ROUND_UP(chunks, 8);
695 if (with_super)
696 bytes += sizeof(bitmap_super_t);
697
698 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
699
700 store->filemap = kmalloc(sizeof(struct page *)
701 * num_pages, GFP_KERNEL);
702 if (!store->filemap)
703 return -ENOMEM;
704
705 if (with_super && !store->sb_page) {
706 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
707 if (store->sb_page == NULL)
708 return -ENOMEM;
709 store->sb_page->index = 0;
710 }
711 pnum = 0;
712 if (store->sb_page) {
713 store->filemap[0] = store->sb_page;
714 pnum = 1;
715 }
716 for ( ; pnum < num_pages; pnum++) {
717 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
718 if (!store->filemap[pnum]) {
719 store->file_pages = pnum;
720 return -ENOMEM;
721 }
722 store->filemap[pnum]->index = pnum;
723 }
724 store->file_pages = pnum;
725
726 /* We need 4 bits per page, rounded up to a multiple
727 * of sizeof(unsigned long) */
728 store->filemap_attr = kzalloc(
729 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
730 GFP_KERNEL);
731 if (!store->filemap_attr)
732 return -ENOMEM;
733
734 store->bytes = bytes;
735
736 return 0;
737 }
738
739 static void bitmap_file_unmap(struct bitmap_storage *store)
740 {
741 struct page **map, *sb_page;
742 int pages;
743 struct file *file;
744
745 file = store->file;
746 map = store->filemap;
747 pages = store->file_pages;
748 sb_page = store->sb_page;
749
750 while (pages--)
751 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
752 free_buffers(map[pages]);
753 kfree(map);
754 kfree(store->filemap_attr);
755
756 if (sb_page)
757 free_buffers(sb_page);
758
759 if (file) {
760 struct inode *inode = file_inode(file);
761 invalidate_mapping_pages(inode->i_mapping, 0, -1);
762 fput(file);
763 }
764 }
765
766 /*
767 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
768 * then it is no longer reliable, so we stop using it and we mark the file
769 * as failed in the superblock
770 */
771 static void bitmap_file_kick(struct bitmap *bitmap)
772 {
773 char *path, *ptr = NULL;
774
775 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
776 bitmap_update_sb(bitmap);
777
778 if (bitmap->storage.file) {
779 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
780 if (path)
781 ptr = d_path(&bitmap->storage.file->f_path,
782 path, PAGE_SIZE);
783
784 printk(KERN_ALERT
785 "%s: kicking failed bitmap file %s from array!\n",
786 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
787
788 kfree(path);
789 } else
790 printk(KERN_ALERT
791 "%s: disabling internal bitmap due to errors\n",
792 bmname(bitmap));
793 }
794 }
795
796 enum bitmap_page_attr {
797 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
798 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
799 * i.e. counter is 1 or 2. */
800 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
801 };
802
803 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
804 enum bitmap_page_attr attr)
805 {
806 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
807 }
808
809 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
810 enum bitmap_page_attr attr)
811 {
812 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
813 }
814
815 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
816 enum bitmap_page_attr attr)
817 {
818 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
819 }
820
821 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
822 enum bitmap_page_attr attr)
823 {
824 return test_and_clear_bit((pnum<<2) + attr,
825 bitmap->storage.filemap_attr);
826 }
827 /*
828 * bitmap_file_set_bit -- called before performing a write to the md device
829 * to set (and eventually sync) a particular bit in the bitmap file
830 *
831 * we set the bit immediately, then we record the page number so that
832 * when an unplug occurs, we can flush the dirty pages out to disk
833 */
834 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
835 {
836 unsigned long bit;
837 struct page *page;
838 void *kaddr;
839 unsigned long chunk = block >> bitmap->counts.chunkshift;
840
841 page = filemap_get_page(&bitmap->storage, chunk);
842 if (!page)
843 return;
844 bit = file_page_offset(&bitmap->storage, chunk);
845
846 /* set the bit */
847 kaddr = kmap_atomic(page);
848 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
849 set_bit(bit, kaddr);
850 else
851 set_bit_le(bit, kaddr);
852 kunmap_atomic(kaddr);
853 pr_debug("set file bit %lu page %lu\n", bit, page->index);
854 /* record page number so it gets flushed to disk when unplug occurs */
855 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
856 }
857
858 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
859 {
860 unsigned long bit;
861 struct page *page;
862 void *paddr;
863 unsigned long chunk = block >> bitmap->counts.chunkshift;
864
865 page = filemap_get_page(&bitmap->storage, chunk);
866 if (!page)
867 return;
868 bit = file_page_offset(&bitmap->storage, chunk);
869 paddr = kmap_atomic(page);
870 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
871 clear_bit(bit, paddr);
872 else
873 clear_bit_le(bit, paddr);
874 kunmap_atomic(paddr);
875 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
876 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
877 bitmap->allclean = 0;
878 }
879 }
880
881 /* this gets called when the md device is ready to unplug its underlying
882 * (slave) device queues -- before we let any writes go down, we need to
883 * sync the dirty pages of the bitmap file to disk */
884 void bitmap_unplug(struct bitmap *bitmap)
885 {
886 unsigned long i;
887 int dirty, need_write;
888
889 if (!bitmap || !bitmap->storage.filemap ||
890 test_bit(BITMAP_STALE, &bitmap->flags))
891 return;
892
893 /* look at each page to see if there are any set bits that need to be
894 * flushed out to disk */
895 for (i = 0; i < bitmap->storage.file_pages; i++) {
896 if (!bitmap->storage.filemap)
897 return;
898 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
899 need_write = test_and_clear_page_attr(bitmap, i,
900 BITMAP_PAGE_NEEDWRITE);
901 if (dirty || need_write) {
902 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
903 write_page(bitmap, bitmap->storage.filemap[i], 0);
904 }
905 }
906 if (bitmap->storage.file)
907 wait_event(bitmap->write_wait,
908 atomic_read(&bitmap->pending_writes)==0);
909 else
910 md_super_wait(bitmap->mddev);
911
912 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
913 bitmap_file_kick(bitmap);
914 }
915 EXPORT_SYMBOL(bitmap_unplug);
916
917 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
918 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
919 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
920 * memory mapping of the bitmap file
921 * Special cases:
922 * if there's no bitmap file, or if the bitmap file had been
923 * previously kicked from the array, we mark all the bits as
924 * 1's in order to cause a full resync.
925 *
926 * We ignore all bits for sectors that end earlier than 'start'.
927 * This is used when reading an out-of-date bitmap...
928 */
929 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
930 {
931 unsigned long i, chunks, index, oldindex, bit;
932 struct page *page = NULL;
933 unsigned long bit_cnt = 0;
934 struct file *file;
935 unsigned long offset;
936 int outofdate;
937 int ret = -ENOSPC;
938 void *paddr;
939 struct bitmap_storage *store = &bitmap->storage;
940
941 chunks = bitmap->counts.chunks;
942 file = store->file;
943
944 if (!file && !bitmap->mddev->bitmap_info.offset) {
945 /* No permanent bitmap - fill with '1s'. */
946 store->filemap = NULL;
947 store->file_pages = 0;
948 for (i = 0; i < chunks ; i++) {
949 /* if the disk bit is set, set the memory bit */
950 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
951 >= start);
952 bitmap_set_memory_bits(bitmap,
953 (sector_t)i << bitmap->counts.chunkshift,
954 needed);
955 }
956 return 0;
957 }
958
959 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
960 if (outofdate)
961 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
962 "recovery\n", bmname(bitmap));
963
964 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
965 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
966 bmname(bitmap),
967 (unsigned long) i_size_read(file->f_mapping->host),
968 store->bytes);
969 goto err;
970 }
971
972 oldindex = ~0L;
973 offset = 0;
974 if (!bitmap->mddev->bitmap_info.external)
975 offset = sizeof(bitmap_super_t);
976
977 for (i = 0; i < chunks; i++) {
978 int b;
979 index = file_page_index(&bitmap->storage, i);
980 bit = file_page_offset(&bitmap->storage, i);
981 if (index != oldindex) { /* this is a new page, read it in */
982 int count;
983 /* unmap the old page, we're done with it */
984 if (index == store->file_pages-1)
985 count = store->bytes - index * PAGE_SIZE;
986 else
987 count = PAGE_SIZE;
988 page = store->filemap[index];
989 if (file)
990 ret = read_page(file, index, bitmap,
991 count, page);
992 else
993 ret = read_sb_page(
994 bitmap->mddev,
995 bitmap->mddev->bitmap_info.offset,
996 page,
997 index, count);
998
999 if (ret)
1000 goto err;
1001
1002 oldindex = index;
1003
1004 if (outofdate) {
1005 /*
1006 * if bitmap is out of date, dirty the
1007 * whole page and write it out
1008 */
1009 paddr = kmap_atomic(page);
1010 memset(paddr + offset, 0xff,
1011 PAGE_SIZE - offset);
1012 kunmap_atomic(paddr);
1013 write_page(bitmap, page, 1);
1014
1015 ret = -EIO;
1016 if (test_bit(BITMAP_WRITE_ERROR,
1017 &bitmap->flags))
1018 goto err;
1019 }
1020 }
1021 paddr = kmap_atomic(page);
1022 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1023 b = test_bit(bit, paddr);
1024 else
1025 b = test_bit_le(bit, paddr);
1026 kunmap_atomic(paddr);
1027 if (b) {
1028 /* if the disk bit is set, set the memory bit */
1029 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1030 >= start);
1031 bitmap_set_memory_bits(bitmap,
1032 (sector_t)i << bitmap->counts.chunkshift,
1033 needed);
1034 bit_cnt++;
1035 }
1036 offset = 0;
1037 }
1038
1039 printk(KERN_INFO "%s: bitmap initialized from disk: "
1040 "read %lu pages, set %lu of %lu bits\n",
1041 bmname(bitmap), store->file_pages,
1042 bit_cnt, chunks);
1043
1044 return 0;
1045
1046 err:
1047 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1048 bmname(bitmap), ret);
1049 return ret;
1050 }
1051
1052 void bitmap_write_all(struct bitmap *bitmap)
1053 {
1054 /* We don't actually write all bitmap blocks here,
1055 * just flag them as needing to be written
1056 */
1057 int i;
1058
1059 if (!bitmap || !bitmap->storage.filemap)
1060 return;
1061 if (bitmap->storage.file)
1062 /* Only one copy, so nothing needed */
1063 return;
1064
1065 for (i = 0; i < bitmap->storage.file_pages; i++)
1066 set_page_attr(bitmap, i,
1067 BITMAP_PAGE_NEEDWRITE);
1068 bitmap->allclean = 0;
1069 }
1070
1071 static void bitmap_count_page(struct bitmap_counts *bitmap,
1072 sector_t offset, int inc)
1073 {
1074 sector_t chunk = offset >> bitmap->chunkshift;
1075 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1076 bitmap->bp[page].count += inc;
1077 bitmap_checkfree(bitmap, page);
1078 }
1079
1080 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1081 {
1082 sector_t chunk = offset >> bitmap->chunkshift;
1083 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1084 struct bitmap_page *bp = &bitmap->bp[page];
1085
1086 if (!bp->pending)
1087 bp->pending = 1;
1088 }
1089
1090 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1091 sector_t offset, sector_t *blocks,
1092 int create);
1093
1094 /*
1095 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1096 * out to disk
1097 */
1098
1099 void bitmap_daemon_work(struct mddev *mddev)
1100 {
1101 struct bitmap *bitmap;
1102 unsigned long j;
1103 unsigned long nextpage;
1104 sector_t blocks;
1105 struct bitmap_counts *counts;
1106
1107 /* Use a mutex to guard daemon_work against
1108 * bitmap_destroy.
1109 */
1110 mutex_lock(&mddev->bitmap_info.mutex);
1111 bitmap = mddev->bitmap;
1112 if (bitmap == NULL) {
1113 mutex_unlock(&mddev->bitmap_info.mutex);
1114 return;
1115 }
1116 if (time_before(jiffies, bitmap->daemon_lastrun
1117 + mddev->bitmap_info.daemon_sleep))
1118 goto done;
1119
1120 bitmap->daemon_lastrun = jiffies;
1121 if (bitmap->allclean) {
1122 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1123 goto done;
1124 }
1125 bitmap->allclean = 1;
1126
1127 /* Any file-page which is PENDING now needs to be written.
1128 * So set NEEDWRITE now, then after we make any last-minute changes
1129 * we will write it.
1130 */
1131 for (j = 0; j < bitmap->storage.file_pages; j++)
1132 if (test_and_clear_page_attr(bitmap, j,
1133 BITMAP_PAGE_PENDING))
1134 set_page_attr(bitmap, j,
1135 BITMAP_PAGE_NEEDWRITE);
1136
1137 if (bitmap->need_sync &&
1138 mddev->bitmap_info.external == 0) {
1139 /* Arrange for superblock update as well as
1140 * other changes */
1141 bitmap_super_t *sb;
1142 bitmap->need_sync = 0;
1143 if (bitmap->storage.filemap) {
1144 sb = kmap_atomic(bitmap->storage.sb_page);
1145 sb->events_cleared =
1146 cpu_to_le64(bitmap->events_cleared);
1147 kunmap_atomic(sb);
1148 set_page_attr(bitmap, 0,
1149 BITMAP_PAGE_NEEDWRITE);
1150 }
1151 }
1152 /* Now look at the bitmap counters and if any are '2' or '1',
1153 * decrement and handle accordingly.
1154 */
1155 counts = &bitmap->counts;
1156 spin_lock_irq(&counts->lock);
1157 nextpage = 0;
1158 for (j = 0; j < counts->chunks; j++) {
1159 bitmap_counter_t *bmc;
1160 sector_t block = (sector_t)j << counts->chunkshift;
1161
1162 if (j == nextpage) {
1163 nextpage += PAGE_COUNTER_RATIO;
1164 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1165 j |= PAGE_COUNTER_MASK;
1166 continue;
1167 }
1168 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1169 }
1170 bmc = bitmap_get_counter(counts,
1171 block,
1172 &blocks, 0);
1173
1174 if (!bmc) {
1175 j |= PAGE_COUNTER_MASK;
1176 continue;
1177 }
1178 if (*bmc == 1 && !bitmap->need_sync) {
1179 /* We can clear the bit */
1180 *bmc = 0;
1181 bitmap_count_page(counts, block, -1);
1182 bitmap_file_clear_bit(bitmap, block);
1183 } else if (*bmc && *bmc <= 2) {
1184 *bmc = 1;
1185 bitmap_set_pending(counts, block);
1186 bitmap->allclean = 0;
1187 }
1188 }
1189 spin_unlock_irq(&counts->lock);
1190
1191 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1192 * DIRTY pages need to be written by bitmap_unplug so it can wait
1193 * for them.
1194 * If we find any DIRTY page we stop there and let bitmap_unplug
1195 * handle all the rest. This is important in the case where
1196 * the first blocking holds the superblock and it has been updated.
1197 * We mustn't write any other blocks before the superblock.
1198 */
1199 for (j = 0;
1200 j < bitmap->storage.file_pages
1201 && !test_bit(BITMAP_STALE, &bitmap->flags);
1202 j++) {
1203
1204 if (test_page_attr(bitmap, j,
1205 BITMAP_PAGE_DIRTY))
1206 /* bitmap_unplug will handle the rest */
1207 break;
1208 if (test_and_clear_page_attr(bitmap, j,
1209 BITMAP_PAGE_NEEDWRITE)) {
1210 write_page(bitmap, bitmap->storage.filemap[j], 0);
1211 }
1212 }
1213
1214 done:
1215 if (bitmap->allclean == 0)
1216 mddev->thread->timeout =
1217 mddev->bitmap_info.daemon_sleep;
1218 mutex_unlock(&mddev->bitmap_info.mutex);
1219 }
1220
1221 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1222 sector_t offset, sector_t *blocks,
1223 int create)
1224 __releases(bitmap->lock)
1225 __acquires(bitmap->lock)
1226 {
1227 /* If 'create', we might release the lock and reclaim it.
1228 * The lock must have been taken with interrupts enabled.
1229 * If !create, we don't release the lock.
1230 */
1231 sector_t chunk = offset >> bitmap->chunkshift;
1232 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1233 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1234 sector_t csize;
1235 int err;
1236
1237 err = bitmap_checkpage(bitmap, page, create);
1238
1239 if (bitmap->bp[page].hijacked ||
1240 bitmap->bp[page].map == NULL)
1241 csize = ((sector_t)1) << (bitmap->chunkshift +
1242 PAGE_COUNTER_SHIFT - 1);
1243 else
1244 csize = ((sector_t)1) << bitmap->chunkshift;
1245 *blocks = csize - (offset & (csize - 1));
1246
1247 if (err < 0)
1248 return NULL;
1249
1250 /* now locked ... */
1251
1252 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1253 /* should we use the first or second counter field
1254 * of the hijacked pointer? */
1255 int hi = (pageoff > PAGE_COUNTER_MASK);
1256 return &((bitmap_counter_t *)
1257 &bitmap->bp[page].map)[hi];
1258 } else /* page is allocated */
1259 return (bitmap_counter_t *)
1260 &(bitmap->bp[page].map[pageoff]);
1261 }
1262
1263 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1264 {
1265 if (!bitmap)
1266 return 0;
1267
1268 if (behind) {
1269 int bw;
1270 atomic_inc(&bitmap->behind_writes);
1271 bw = atomic_read(&bitmap->behind_writes);
1272 if (bw > bitmap->behind_writes_used)
1273 bitmap->behind_writes_used = bw;
1274
1275 pr_debug("inc write-behind count %d/%lu\n",
1276 bw, bitmap->mddev->bitmap_info.max_write_behind);
1277 }
1278
1279 while (sectors) {
1280 sector_t blocks;
1281 bitmap_counter_t *bmc;
1282
1283 spin_lock_irq(&bitmap->counts.lock);
1284 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1285 if (!bmc) {
1286 spin_unlock_irq(&bitmap->counts.lock);
1287 return 0;
1288 }
1289
1290 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1291 DEFINE_WAIT(__wait);
1292 /* note that it is safe to do the prepare_to_wait
1293 * after the test as long as we do it before dropping
1294 * the spinlock.
1295 */
1296 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1297 TASK_UNINTERRUPTIBLE);
1298 spin_unlock_irq(&bitmap->counts.lock);
1299 schedule();
1300 finish_wait(&bitmap->overflow_wait, &__wait);
1301 continue;
1302 }
1303
1304 switch (*bmc) {
1305 case 0:
1306 bitmap_file_set_bit(bitmap, offset);
1307 bitmap_count_page(&bitmap->counts, offset, 1);
1308 /* fall through */
1309 case 1:
1310 *bmc = 2;
1311 }
1312
1313 (*bmc)++;
1314
1315 spin_unlock_irq(&bitmap->counts.lock);
1316
1317 offset += blocks;
1318 if (sectors > blocks)
1319 sectors -= blocks;
1320 else
1321 sectors = 0;
1322 }
1323 return 0;
1324 }
1325 EXPORT_SYMBOL(bitmap_startwrite);
1326
1327 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1328 int success, int behind)
1329 {
1330 if (!bitmap)
1331 return;
1332 if (behind) {
1333 if (atomic_dec_and_test(&bitmap->behind_writes))
1334 wake_up(&bitmap->behind_wait);
1335 pr_debug("dec write-behind count %d/%lu\n",
1336 atomic_read(&bitmap->behind_writes),
1337 bitmap->mddev->bitmap_info.max_write_behind);
1338 }
1339
1340 while (sectors) {
1341 sector_t blocks;
1342 unsigned long flags;
1343 bitmap_counter_t *bmc;
1344
1345 spin_lock_irqsave(&bitmap->counts.lock, flags);
1346 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1347 if (!bmc) {
1348 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1349 return;
1350 }
1351
1352 if (success && !bitmap->mddev->degraded &&
1353 bitmap->events_cleared < bitmap->mddev->events) {
1354 bitmap->events_cleared = bitmap->mddev->events;
1355 bitmap->need_sync = 1;
1356 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1357 }
1358
1359 if (!success && !NEEDED(*bmc))
1360 *bmc |= NEEDED_MASK;
1361
1362 if (COUNTER(*bmc) == COUNTER_MAX)
1363 wake_up(&bitmap->overflow_wait);
1364
1365 (*bmc)--;
1366 if (*bmc <= 2) {
1367 bitmap_set_pending(&bitmap->counts, offset);
1368 bitmap->allclean = 0;
1369 }
1370 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1371 offset += blocks;
1372 if (sectors > blocks)
1373 sectors -= blocks;
1374 else
1375 sectors = 0;
1376 }
1377 }
1378 EXPORT_SYMBOL(bitmap_endwrite);
1379
1380 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1381 int degraded)
1382 {
1383 bitmap_counter_t *bmc;
1384 int rv;
1385 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1386 *blocks = 1024;
1387 return 1; /* always resync if no bitmap */
1388 }
1389 spin_lock_irq(&bitmap->counts.lock);
1390 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1391 rv = 0;
1392 if (bmc) {
1393 /* locked */
1394 if (RESYNC(*bmc))
1395 rv = 1;
1396 else if (NEEDED(*bmc)) {
1397 rv = 1;
1398 if (!degraded) { /* don't set/clear bits if degraded */
1399 *bmc |= RESYNC_MASK;
1400 *bmc &= ~NEEDED_MASK;
1401 }
1402 }
1403 }
1404 spin_unlock_irq(&bitmap->counts.lock);
1405 return rv;
1406 }
1407
1408 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1409 int degraded)
1410 {
1411 /* bitmap_start_sync must always report on multiples of whole
1412 * pages, otherwise resync (which is very PAGE_SIZE based) will
1413 * get confused.
1414 * So call __bitmap_start_sync repeatedly (if needed) until
1415 * At least PAGE_SIZE>>9 blocks are covered.
1416 * Return the 'or' of the result.
1417 */
1418 int rv = 0;
1419 sector_t blocks1;
1420
1421 *blocks = 0;
1422 while (*blocks < (PAGE_SIZE>>9)) {
1423 rv |= __bitmap_start_sync(bitmap, offset,
1424 &blocks1, degraded);
1425 offset += blocks1;
1426 *blocks += blocks1;
1427 }
1428 return rv;
1429 }
1430 EXPORT_SYMBOL(bitmap_start_sync);
1431
1432 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1433 {
1434 bitmap_counter_t *bmc;
1435 unsigned long flags;
1436
1437 if (bitmap == NULL) {
1438 *blocks = 1024;
1439 return;
1440 }
1441 spin_lock_irqsave(&bitmap->counts.lock, flags);
1442 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1443 if (bmc == NULL)
1444 goto unlock;
1445 /* locked */
1446 if (RESYNC(*bmc)) {
1447 *bmc &= ~RESYNC_MASK;
1448
1449 if (!NEEDED(*bmc) && aborted)
1450 *bmc |= NEEDED_MASK;
1451 else {
1452 if (*bmc <= 2) {
1453 bitmap_set_pending(&bitmap->counts, offset);
1454 bitmap->allclean = 0;
1455 }
1456 }
1457 }
1458 unlock:
1459 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1460 }
1461 EXPORT_SYMBOL(bitmap_end_sync);
1462
1463 void bitmap_close_sync(struct bitmap *bitmap)
1464 {
1465 /* Sync has finished, and any bitmap chunks that weren't synced
1466 * properly have been aborted. It remains to us to clear the
1467 * RESYNC bit wherever it is still on
1468 */
1469 sector_t sector = 0;
1470 sector_t blocks;
1471 if (!bitmap)
1472 return;
1473 while (sector < bitmap->mddev->resync_max_sectors) {
1474 bitmap_end_sync(bitmap, sector, &blocks, 0);
1475 sector += blocks;
1476 }
1477 }
1478 EXPORT_SYMBOL(bitmap_close_sync);
1479
1480 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1481 {
1482 sector_t s = 0;
1483 sector_t blocks;
1484
1485 if (!bitmap)
1486 return;
1487 if (sector == 0) {
1488 bitmap->last_end_sync = jiffies;
1489 return;
1490 }
1491 if (time_before(jiffies, (bitmap->last_end_sync
1492 + bitmap->mddev->bitmap_info.daemon_sleep)))
1493 return;
1494 wait_event(bitmap->mddev->recovery_wait,
1495 atomic_read(&bitmap->mddev->recovery_active) == 0);
1496
1497 bitmap->mddev->curr_resync_completed = sector;
1498 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1499 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1500 s = 0;
1501 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1502 bitmap_end_sync(bitmap, s, &blocks, 0);
1503 s += blocks;
1504 }
1505 bitmap->last_end_sync = jiffies;
1506 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1507 }
1508 EXPORT_SYMBOL(bitmap_cond_end_sync);
1509
1510 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1511 {
1512 /* For each chunk covered by any of these sectors, set the
1513 * counter to 2 and possibly set resync_needed. They should all
1514 * be 0 at this point
1515 */
1516
1517 sector_t secs;
1518 bitmap_counter_t *bmc;
1519 spin_lock_irq(&bitmap->counts.lock);
1520 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1521 if (!bmc) {
1522 spin_unlock_irq(&bitmap->counts.lock);
1523 return;
1524 }
1525 if (!*bmc) {
1526 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1527 bitmap_count_page(&bitmap->counts, offset, 1);
1528 bitmap_set_pending(&bitmap->counts, offset);
1529 bitmap->allclean = 0;
1530 }
1531 spin_unlock_irq(&bitmap->counts.lock);
1532 }
1533
1534 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1535 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1536 {
1537 unsigned long chunk;
1538
1539 for (chunk = s; chunk <= e; chunk++) {
1540 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1541 bitmap_set_memory_bits(bitmap, sec, 1);
1542 bitmap_file_set_bit(bitmap, sec);
1543 if (sec < bitmap->mddev->recovery_cp)
1544 /* We are asserting that the array is dirty,
1545 * so move the recovery_cp address back so
1546 * that it is obvious that it is dirty
1547 */
1548 bitmap->mddev->recovery_cp = sec;
1549 }
1550 }
1551
1552 /*
1553 * flush out any pending updates
1554 */
1555 void bitmap_flush(struct mddev *mddev)
1556 {
1557 struct bitmap *bitmap = mddev->bitmap;
1558 long sleep;
1559
1560 if (!bitmap) /* there was no bitmap */
1561 return;
1562
1563 /* run the daemon_work three time to ensure everything is flushed
1564 * that can be
1565 */
1566 sleep = mddev->bitmap_info.daemon_sleep * 2;
1567 bitmap->daemon_lastrun -= sleep;
1568 bitmap_daemon_work(mddev);
1569 bitmap->daemon_lastrun -= sleep;
1570 bitmap_daemon_work(mddev);
1571 bitmap->daemon_lastrun -= sleep;
1572 bitmap_daemon_work(mddev);
1573 bitmap_update_sb(bitmap);
1574 }
1575
1576 /*
1577 * free memory that was allocated
1578 */
1579 static void bitmap_free(struct bitmap *bitmap)
1580 {
1581 unsigned long k, pages;
1582 struct bitmap_page *bp;
1583
1584 if (!bitmap) /* there was no bitmap */
1585 return;
1586
1587 /* Shouldn't be needed - but just in case.... */
1588 wait_event(bitmap->write_wait,
1589 atomic_read(&bitmap->pending_writes) == 0);
1590
1591 /* release the bitmap file */
1592 bitmap_file_unmap(&bitmap->storage);
1593
1594 bp = bitmap->counts.bp;
1595 pages = bitmap->counts.pages;
1596
1597 /* free all allocated memory */
1598
1599 if (bp) /* deallocate the page memory */
1600 for (k = 0; k < pages; k++)
1601 if (bp[k].map && !bp[k].hijacked)
1602 kfree(bp[k].map);
1603 kfree(bp);
1604 kfree(bitmap);
1605 }
1606
1607 void bitmap_destroy(struct mddev *mddev)
1608 {
1609 struct bitmap *bitmap = mddev->bitmap;
1610
1611 if (!bitmap) /* there was no bitmap */
1612 return;
1613
1614 mutex_lock(&mddev->bitmap_info.mutex);
1615 mddev->bitmap = NULL; /* disconnect from the md device */
1616 mutex_unlock(&mddev->bitmap_info.mutex);
1617 if (mddev->thread)
1618 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1619
1620 if (bitmap->sysfs_can_clear)
1621 sysfs_put(bitmap->sysfs_can_clear);
1622
1623 bitmap_free(bitmap);
1624 }
1625
1626 /*
1627 * initialize the bitmap structure
1628 * if this returns an error, bitmap_destroy must be called to do clean up
1629 */
1630 int bitmap_create(struct mddev *mddev)
1631 {
1632 struct bitmap *bitmap;
1633 sector_t blocks = mddev->resync_max_sectors;
1634 struct file *file = mddev->bitmap_info.file;
1635 int err;
1636 struct sysfs_dirent *bm = NULL;
1637
1638 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1639
1640 BUG_ON(file && mddev->bitmap_info.offset);
1641
1642 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1643 if (!bitmap)
1644 return -ENOMEM;
1645
1646 spin_lock_init(&bitmap->counts.lock);
1647 atomic_set(&bitmap->pending_writes, 0);
1648 init_waitqueue_head(&bitmap->write_wait);
1649 init_waitqueue_head(&bitmap->overflow_wait);
1650 init_waitqueue_head(&bitmap->behind_wait);
1651
1652 bitmap->mddev = mddev;
1653
1654 if (mddev->kobj.sd)
1655 bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap");
1656 if (bm) {
1657 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear");
1658 sysfs_put(bm);
1659 } else
1660 bitmap->sysfs_can_clear = NULL;
1661
1662 bitmap->storage.file = file;
1663 if (file) {
1664 get_file(file);
1665 /* As future accesses to this file will use bmap,
1666 * and bypass the page cache, we must sync the file
1667 * first.
1668 */
1669 vfs_fsync(file, 1);
1670 }
1671 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1672 if (!mddev->bitmap_info.external) {
1673 /*
1674 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1675 * instructing us to create a new on-disk bitmap instance.
1676 */
1677 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1678 err = bitmap_new_disk_sb(bitmap);
1679 else
1680 err = bitmap_read_sb(bitmap);
1681 } else {
1682 err = 0;
1683 if (mddev->bitmap_info.chunksize == 0 ||
1684 mddev->bitmap_info.daemon_sleep == 0)
1685 /* chunksize and time_base need to be
1686 * set first. */
1687 err = -EINVAL;
1688 }
1689 if (err)
1690 goto error;
1691
1692 bitmap->daemon_lastrun = jiffies;
1693 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1694 if (err)
1695 goto error;
1696
1697 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1698 bitmap->counts.pages, bmname(bitmap));
1699
1700 mddev->bitmap = bitmap;
1701 return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1702
1703 error:
1704 bitmap_free(bitmap);
1705 return err;
1706 }
1707
1708 int bitmap_load(struct mddev *mddev)
1709 {
1710 int err = 0;
1711 sector_t start = 0;
1712 sector_t sector = 0;
1713 struct bitmap *bitmap = mddev->bitmap;
1714
1715 if (!bitmap)
1716 goto out;
1717
1718 /* Clear out old bitmap info first: Either there is none, or we
1719 * are resuming after someone else has possibly changed things,
1720 * so we should forget old cached info.
1721 * All chunks should be clean, but some might need_sync.
1722 */
1723 while (sector < mddev->resync_max_sectors) {
1724 sector_t blocks;
1725 bitmap_start_sync(bitmap, sector, &blocks, 0);
1726 sector += blocks;
1727 }
1728 bitmap_close_sync(bitmap);
1729
1730 if (mddev->degraded == 0
1731 || bitmap->events_cleared == mddev->events)
1732 /* no need to keep dirty bits to optimise a
1733 * re-add of a missing device */
1734 start = mddev->recovery_cp;
1735
1736 mutex_lock(&mddev->bitmap_info.mutex);
1737 err = bitmap_init_from_disk(bitmap, start);
1738 mutex_unlock(&mddev->bitmap_info.mutex);
1739
1740 if (err)
1741 goto out;
1742 clear_bit(BITMAP_STALE, &bitmap->flags);
1743
1744 /* Kick recovery in case any bits were set */
1745 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1746
1747 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1748 md_wakeup_thread(mddev->thread);
1749
1750 bitmap_update_sb(bitmap);
1751
1752 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1753 err = -EIO;
1754 out:
1755 return err;
1756 }
1757 EXPORT_SYMBOL_GPL(bitmap_load);
1758
1759 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1760 {
1761 unsigned long chunk_kb;
1762 struct bitmap_counts *counts;
1763
1764 if (!bitmap)
1765 return;
1766
1767 counts = &bitmap->counts;
1768
1769 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1770 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1771 "%lu%s chunk",
1772 counts->pages - counts->missing_pages,
1773 counts->pages,
1774 (counts->pages - counts->missing_pages)
1775 << (PAGE_SHIFT - 10),
1776 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1777 chunk_kb ? "KB" : "B");
1778 if (bitmap->storage.file) {
1779 seq_printf(seq, ", file: ");
1780 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1781 }
1782
1783 seq_printf(seq, "\n");
1784 }
1785
1786 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1787 int chunksize, int init)
1788 {
1789 /* If chunk_size is 0, choose an appropriate chunk size.
1790 * Then possibly allocate new storage space.
1791 * Then quiesce, copy bits, replace bitmap, and re-start
1792 *
1793 * This function is called both to set up the initial bitmap
1794 * and to resize the bitmap while the array is active.
1795 * If this happens as a result of the array being resized,
1796 * chunksize will be zero, and we need to choose a suitable
1797 * chunksize, otherwise we use what we are given.
1798 */
1799 struct bitmap_storage store;
1800 struct bitmap_counts old_counts;
1801 unsigned long chunks;
1802 sector_t block;
1803 sector_t old_blocks, new_blocks;
1804 int chunkshift;
1805 int ret = 0;
1806 long pages;
1807 struct bitmap_page *new_bp;
1808
1809 if (bitmap->storage.file && !init) {
1810 pr_info("md: cannot resize file-based bitmap\n");
1811 return -EINVAL;
1812 }
1813
1814 if (chunksize == 0) {
1815 /* If there is enough space, leave the chunk size unchanged,
1816 * else increase by factor of two until there is enough space.
1817 */
1818 long bytes;
1819 long space = bitmap->mddev->bitmap_info.space;
1820
1821 if (space == 0) {
1822 /* We don't know how much space there is, so limit
1823 * to current size - in sectors.
1824 */
1825 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1826 if (!bitmap->mddev->bitmap_info.external)
1827 bytes += sizeof(bitmap_super_t);
1828 space = DIV_ROUND_UP(bytes, 512);
1829 bitmap->mddev->bitmap_info.space = space;
1830 }
1831 chunkshift = bitmap->counts.chunkshift;
1832 chunkshift--;
1833 do {
1834 /* 'chunkshift' is shift from block size to chunk size */
1835 chunkshift++;
1836 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1837 bytes = DIV_ROUND_UP(chunks, 8);
1838 if (!bitmap->mddev->bitmap_info.external)
1839 bytes += sizeof(bitmap_super_t);
1840 } while (bytes > (space << 9));
1841 } else
1842 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1843
1844 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1845 memset(&store, 0, sizeof(store));
1846 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1847 ret = bitmap_storage_alloc(&store, chunks,
1848 !bitmap->mddev->bitmap_info.external);
1849 if (ret)
1850 goto err;
1851
1852 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
1853
1854 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
1855 ret = -ENOMEM;
1856 if (!new_bp) {
1857 bitmap_file_unmap(&store);
1858 goto err;
1859 }
1860
1861 if (!init)
1862 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
1863
1864 store.file = bitmap->storage.file;
1865 bitmap->storage.file = NULL;
1866
1867 if (store.sb_page && bitmap->storage.sb_page)
1868 memcpy(page_address(store.sb_page),
1869 page_address(bitmap->storage.sb_page),
1870 sizeof(bitmap_super_t));
1871 bitmap_file_unmap(&bitmap->storage);
1872 bitmap->storage = store;
1873
1874 old_counts = bitmap->counts;
1875 bitmap->counts.bp = new_bp;
1876 bitmap->counts.pages = pages;
1877 bitmap->counts.missing_pages = pages;
1878 bitmap->counts.chunkshift = chunkshift;
1879 bitmap->counts.chunks = chunks;
1880 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
1881 BITMAP_BLOCK_SHIFT);
1882
1883 blocks = min(old_counts.chunks << old_counts.chunkshift,
1884 chunks << chunkshift);
1885
1886 spin_lock_irq(&bitmap->counts.lock);
1887 for (block = 0; block < blocks; ) {
1888 bitmap_counter_t *bmc_old, *bmc_new;
1889 int set;
1890
1891 bmc_old = bitmap_get_counter(&old_counts, block,
1892 &old_blocks, 0);
1893 set = bmc_old && NEEDED(*bmc_old);
1894
1895 if (set) {
1896 bmc_new = bitmap_get_counter(&bitmap->counts, block,
1897 &new_blocks, 1);
1898 if (*bmc_new == 0) {
1899 /* need to set on-disk bits too. */
1900 sector_t end = block + new_blocks;
1901 sector_t start = block >> chunkshift;
1902 start <<= chunkshift;
1903 while (start < end) {
1904 bitmap_file_set_bit(bitmap, block);
1905 start += 1 << chunkshift;
1906 }
1907 *bmc_new = 2;
1908 bitmap_count_page(&bitmap->counts,
1909 block, 1);
1910 bitmap_set_pending(&bitmap->counts,
1911 block);
1912 }
1913 *bmc_new |= NEEDED_MASK;
1914 if (new_blocks < old_blocks)
1915 old_blocks = new_blocks;
1916 }
1917 block += old_blocks;
1918 }
1919
1920 if (!init) {
1921 int i;
1922 while (block < (chunks << chunkshift)) {
1923 bitmap_counter_t *bmc;
1924 bmc = bitmap_get_counter(&bitmap->counts, block,
1925 &new_blocks, 1);
1926 if (bmc) {
1927 /* new space. It needs to be resynced, so
1928 * we set NEEDED_MASK.
1929 */
1930 if (*bmc == 0) {
1931 *bmc = NEEDED_MASK | 2;
1932 bitmap_count_page(&bitmap->counts,
1933 block, 1);
1934 bitmap_set_pending(&bitmap->counts,
1935 block);
1936 }
1937 }
1938 block += new_blocks;
1939 }
1940 for (i = 0; i < bitmap->storage.file_pages; i++)
1941 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1942 }
1943 spin_unlock_irq(&bitmap->counts.lock);
1944
1945 if (!init) {
1946 bitmap_unplug(bitmap);
1947 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
1948 }
1949 ret = 0;
1950 err:
1951 return ret;
1952 }
1953 EXPORT_SYMBOL_GPL(bitmap_resize);
1954
1955 static ssize_t
1956 location_show(struct mddev *mddev, char *page)
1957 {
1958 ssize_t len;
1959 if (mddev->bitmap_info.file)
1960 len = sprintf(page, "file");
1961 else if (mddev->bitmap_info.offset)
1962 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1963 else
1964 len = sprintf(page, "none");
1965 len += sprintf(page+len, "\n");
1966 return len;
1967 }
1968
1969 static ssize_t
1970 location_store(struct mddev *mddev, const char *buf, size_t len)
1971 {
1972
1973 if (mddev->pers) {
1974 if (!mddev->pers->quiesce)
1975 return -EBUSY;
1976 if (mddev->recovery || mddev->sync_thread)
1977 return -EBUSY;
1978 }
1979
1980 if (mddev->bitmap || mddev->bitmap_info.file ||
1981 mddev->bitmap_info.offset) {
1982 /* bitmap already configured. Only option is to clear it */
1983 if (strncmp(buf, "none", 4) != 0)
1984 return -EBUSY;
1985 if (mddev->pers) {
1986 mddev->pers->quiesce(mddev, 1);
1987 bitmap_destroy(mddev);
1988 mddev->pers->quiesce(mddev, 0);
1989 }
1990 mddev->bitmap_info.offset = 0;
1991 if (mddev->bitmap_info.file) {
1992 struct file *f = mddev->bitmap_info.file;
1993 mddev->bitmap_info.file = NULL;
1994 restore_bitmap_write_access(f);
1995 fput(f);
1996 }
1997 } else {
1998 /* No bitmap, OK to set a location */
1999 long long offset;
2000 if (strncmp(buf, "none", 4) == 0)
2001 /* nothing to be done */;
2002 else if (strncmp(buf, "file:", 5) == 0) {
2003 /* Not supported yet */
2004 return -EINVAL;
2005 } else {
2006 int rv;
2007 if (buf[0] == '+')
2008 rv = strict_strtoll(buf+1, 10, &offset);
2009 else
2010 rv = strict_strtoll(buf, 10, &offset);
2011 if (rv)
2012 return rv;
2013 if (offset == 0)
2014 return -EINVAL;
2015 if (mddev->bitmap_info.external == 0 &&
2016 mddev->major_version == 0 &&
2017 offset != mddev->bitmap_info.default_offset)
2018 return -EINVAL;
2019 mddev->bitmap_info.offset = offset;
2020 if (mddev->pers) {
2021 mddev->pers->quiesce(mddev, 1);
2022 rv = bitmap_create(mddev);
2023 if (!rv)
2024 rv = bitmap_load(mddev);
2025 if (rv) {
2026 bitmap_destroy(mddev);
2027 mddev->bitmap_info.offset = 0;
2028 }
2029 mddev->pers->quiesce(mddev, 0);
2030 if (rv)
2031 return rv;
2032 }
2033 }
2034 }
2035 if (!mddev->external) {
2036 /* Ensure new bitmap info is stored in
2037 * metadata promptly.
2038 */
2039 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2040 md_wakeup_thread(mddev->thread);
2041 }
2042 return len;
2043 }
2044
2045 static struct md_sysfs_entry bitmap_location =
2046 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2047
2048 /* 'bitmap/space' is the space available at 'location' for the
2049 * bitmap. This allows the kernel to know when it is safe to
2050 * resize the bitmap to match a resized array.
2051 */
2052 static ssize_t
2053 space_show(struct mddev *mddev, char *page)
2054 {
2055 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2056 }
2057
2058 static ssize_t
2059 space_store(struct mddev *mddev, const char *buf, size_t len)
2060 {
2061 unsigned long sectors;
2062 int rv;
2063
2064 rv = kstrtoul(buf, 10, &sectors);
2065 if (rv)
2066 return rv;
2067
2068 if (sectors == 0)
2069 return -EINVAL;
2070
2071 if (mddev->bitmap &&
2072 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2073 return -EFBIG; /* Bitmap is too big for this small space */
2074
2075 /* could make sure it isn't too big, but that isn't really
2076 * needed - user-space should be careful.
2077 */
2078 mddev->bitmap_info.space = sectors;
2079 return len;
2080 }
2081
2082 static struct md_sysfs_entry bitmap_space =
2083 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2084
2085 static ssize_t
2086 timeout_show(struct mddev *mddev, char *page)
2087 {
2088 ssize_t len;
2089 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2090 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2091
2092 len = sprintf(page, "%lu", secs);
2093 if (jifs)
2094 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2095 len += sprintf(page+len, "\n");
2096 return len;
2097 }
2098
2099 static ssize_t
2100 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2101 {
2102 /* timeout can be set at any time */
2103 unsigned long timeout;
2104 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2105 if (rv)
2106 return rv;
2107
2108 /* just to make sure we don't overflow... */
2109 if (timeout >= LONG_MAX / HZ)
2110 return -EINVAL;
2111
2112 timeout = timeout * HZ / 10000;
2113
2114 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2115 timeout = MAX_SCHEDULE_TIMEOUT-1;
2116 if (timeout < 1)
2117 timeout = 1;
2118 mddev->bitmap_info.daemon_sleep = timeout;
2119 if (mddev->thread) {
2120 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2121 * the bitmap is all clean and we don't need to
2122 * adjust the timeout right now
2123 */
2124 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2125 mddev->thread->timeout = timeout;
2126 md_wakeup_thread(mddev->thread);
2127 }
2128 }
2129 return len;
2130 }
2131
2132 static struct md_sysfs_entry bitmap_timeout =
2133 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2134
2135 static ssize_t
2136 backlog_show(struct mddev *mddev, char *page)
2137 {
2138 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2139 }
2140
2141 static ssize_t
2142 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2143 {
2144 unsigned long backlog;
2145 int rv = strict_strtoul(buf, 10, &backlog);
2146 if (rv)
2147 return rv;
2148 if (backlog > COUNTER_MAX)
2149 return -EINVAL;
2150 mddev->bitmap_info.max_write_behind = backlog;
2151 return len;
2152 }
2153
2154 static struct md_sysfs_entry bitmap_backlog =
2155 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2156
2157 static ssize_t
2158 chunksize_show(struct mddev *mddev, char *page)
2159 {
2160 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2161 }
2162
2163 static ssize_t
2164 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2165 {
2166 /* Can only be changed when no bitmap is active */
2167 int rv;
2168 unsigned long csize;
2169 if (mddev->bitmap)
2170 return -EBUSY;
2171 rv = strict_strtoul(buf, 10, &csize);
2172 if (rv)
2173 return rv;
2174 if (csize < 512 ||
2175 !is_power_of_2(csize))
2176 return -EINVAL;
2177 mddev->bitmap_info.chunksize = csize;
2178 return len;
2179 }
2180
2181 static struct md_sysfs_entry bitmap_chunksize =
2182 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2183
2184 static ssize_t metadata_show(struct mddev *mddev, char *page)
2185 {
2186 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2187 ? "external" : "internal"));
2188 }
2189
2190 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2191 {
2192 if (mddev->bitmap ||
2193 mddev->bitmap_info.file ||
2194 mddev->bitmap_info.offset)
2195 return -EBUSY;
2196 if (strncmp(buf, "external", 8) == 0)
2197 mddev->bitmap_info.external = 1;
2198 else if (strncmp(buf, "internal", 8) == 0)
2199 mddev->bitmap_info.external = 0;
2200 else
2201 return -EINVAL;
2202 return len;
2203 }
2204
2205 static struct md_sysfs_entry bitmap_metadata =
2206 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2207
2208 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2209 {
2210 int len;
2211 if (mddev->bitmap)
2212 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2213 "false" : "true"));
2214 else
2215 len = sprintf(page, "\n");
2216 return len;
2217 }
2218
2219 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2220 {
2221 if (mddev->bitmap == NULL)
2222 return -ENOENT;
2223 if (strncmp(buf, "false", 5) == 0)
2224 mddev->bitmap->need_sync = 1;
2225 else if (strncmp(buf, "true", 4) == 0) {
2226 if (mddev->degraded)
2227 return -EBUSY;
2228 mddev->bitmap->need_sync = 0;
2229 } else
2230 return -EINVAL;
2231 return len;
2232 }
2233
2234 static struct md_sysfs_entry bitmap_can_clear =
2235 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2236
2237 static ssize_t
2238 behind_writes_used_show(struct mddev *mddev, char *page)
2239 {
2240 if (mddev->bitmap == NULL)
2241 return sprintf(page, "0\n");
2242 return sprintf(page, "%lu\n",
2243 mddev->bitmap->behind_writes_used);
2244 }
2245
2246 static ssize_t
2247 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2248 {
2249 if (mddev->bitmap)
2250 mddev->bitmap->behind_writes_used = 0;
2251 return len;
2252 }
2253
2254 static struct md_sysfs_entry max_backlog_used =
2255 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2256 behind_writes_used_show, behind_writes_used_reset);
2257
2258 static struct attribute *md_bitmap_attrs[] = {
2259 &bitmap_location.attr,
2260 &bitmap_space.attr,
2261 &bitmap_timeout.attr,
2262 &bitmap_backlog.attr,
2263 &bitmap_chunksize.attr,
2264 &bitmap_metadata.attr,
2265 &bitmap_can_clear.attr,
2266 &max_backlog_used.attr,
2267 NULL
2268 };
2269 struct attribute_group md_bitmap_group = {
2270 .name = "bitmap",
2271 .attrs = md_bitmap_attrs,
2272 };
2273