Merge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
[deliverable/linux.git] / drivers / md / bitmap.c
1 /*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
10 */
11
12 /*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
16 */
17
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
30 #include "md.h"
31 #include "bitmap.h"
32
33 static inline char *bmname(struct bitmap *bitmap)
34 {
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
36 }
37
38 /*
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
40 *
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
44 *
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
47 */
48 static int bitmap_checkpage(struct bitmap_counts *bitmap,
49 unsigned long page, int create)
50 __releases(bitmap->lock)
51 __acquires(bitmap->lock)
52 {
53 unsigned char *mappage;
54
55 if (page >= bitmap->pages) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
58 * It is harmless.
59 */
60 return -EINVAL;
61 }
62
63 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
64 return 0;
65
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
67 return 0;
68
69 if (!create)
70 return -ENOENT;
71
72 /* this page has not been allocated yet */
73
74 spin_unlock_irq(&bitmap->lock);
75 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
76 spin_lock_irq(&bitmap->lock);
77
78 if (mappage == NULL) {
79 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
80 /* failed - set the hijacked flag so that we can use the
81 * pointer as a counter */
82 if (!bitmap->bp[page].map)
83 bitmap->bp[page].hijacked = 1;
84 } else if (bitmap->bp[page].map ||
85 bitmap->bp[page].hijacked) {
86 /* somebody beat us to getting the page */
87 kfree(mappage);
88 return 0;
89 } else {
90
91 /* no page was in place and we have one, so install it */
92
93 bitmap->bp[page].map = mappage;
94 bitmap->missing_pages--;
95 }
96 return 0;
97 }
98
99 /* if page is completely empty, put it back on the free list, or dealloc it */
100 /* if page was hijacked, unmark the flag so it might get alloced next time */
101 /* Note: lock should be held when calling this */
102 static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
103 {
104 char *ptr;
105
106 if (bitmap->bp[page].count) /* page is still busy */
107 return;
108
109 /* page is no longer in use, it can be released */
110
111 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
112 bitmap->bp[page].hijacked = 0;
113 bitmap->bp[page].map = NULL;
114 } else {
115 /* normal case, free the page */
116 ptr = bitmap->bp[page].map;
117 bitmap->bp[page].map = NULL;
118 bitmap->missing_pages++;
119 kfree(ptr);
120 }
121 }
122
123 /*
124 * bitmap file handling - read and write the bitmap file and its superblock
125 */
126
127 /*
128 * basic page I/O operations
129 */
130
131 /* IO operations when bitmap is stored near all superblocks */
132 static int read_sb_page(struct mddev *mddev, loff_t offset,
133 struct page *page,
134 unsigned long index, int size)
135 {
136 /* choose a good rdev and read the page from there */
137
138 struct md_rdev *rdev;
139 sector_t target;
140
141 rdev_for_each(rdev, mddev) {
142 if (! test_bit(In_sync, &rdev->flags)
143 || test_bit(Faulty, &rdev->flags))
144 continue;
145
146 target = offset + index * (PAGE_SIZE/512);
147
148 if (sync_page_io(rdev, target,
149 roundup(size, bdev_logical_block_size(rdev->bdev)),
150 page, READ, true)) {
151 page->index = index;
152 return 0;
153 }
154 }
155 return -EIO;
156 }
157
158 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
159 {
160 /* Iterate the disks of an mddev, using rcu to protect access to the
161 * linked list, and raising the refcount of devices we return to ensure
162 * they don't disappear while in use.
163 * As devices are only added or removed when raid_disk is < 0 and
164 * nr_pending is 0 and In_sync is clear, the entries we return will
165 * still be in the same position on the list when we re-enter
166 * list_for_each_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 sb = kmap_atomic(sb_page);
568
569 chunksize = le32_to_cpu(sb->chunksize);
570 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
571 write_behind = le32_to_cpu(sb->write_behind);
572 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
573
574 /* verify that the bitmap-specific fields are valid */
575 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
576 reason = "bad magic";
577 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
578 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
579 reason = "unrecognized superblock version";
580 else if (chunksize < 512)
581 reason = "bitmap chunksize too small";
582 else if (!is_power_of_2(chunksize))
583 reason = "bitmap chunksize not a power of 2";
584 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
585 reason = "daemon sleep period out of range";
586 else if (write_behind > COUNTER_MAX)
587 reason = "write-behind limit out of range (0 - 16383)";
588 if (reason) {
589 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
590 bmname(bitmap), reason);
591 goto out;
592 }
593
594 /* keep the array size field of the bitmap superblock up to date */
595 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
596
597 if (bitmap->mddev->persistent) {
598 /*
599 * We have a persistent array superblock, so compare the
600 * bitmap's UUID and event counter to the mddev's
601 */
602 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
603 printk(KERN_INFO
604 "%s: bitmap superblock UUID mismatch\n",
605 bmname(bitmap));
606 goto out;
607 }
608 events = le64_to_cpu(sb->events);
609 if (events < bitmap->mddev->events) {
610 printk(KERN_INFO
611 "%s: bitmap file is out of date (%llu < %llu) "
612 "-- forcing full recovery\n",
613 bmname(bitmap), events,
614 (unsigned long long) bitmap->mddev->events);
615 set_bit(BITMAP_STALE, &bitmap->flags);
616 }
617 }
618
619 /* assign fields using values from superblock */
620 bitmap->flags |= le32_to_cpu(sb->state);
621 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
622 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
623 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
624 err = 0;
625 out:
626 kunmap_atomic(sb);
627 out_no_sb:
628 if (test_bit(BITMAP_STALE, &bitmap->flags))
629 bitmap->events_cleared = bitmap->mddev->events;
630 bitmap->mddev->bitmap_info.chunksize = chunksize;
631 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
632 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
633 if (bitmap->mddev->bitmap_info.space == 0 ||
634 bitmap->mddev->bitmap_info.space > sectors_reserved)
635 bitmap->mddev->bitmap_info.space = sectors_reserved;
636 if (err)
637 bitmap_print_sb(bitmap);
638 return err;
639 }
640
641 /*
642 * general bitmap file operations
643 */
644
645 /*
646 * on-disk bitmap:
647 *
648 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
649 * file a page at a time. There's a superblock at the start of the file.
650 */
651 /* calculate the index of the page that contains this bit */
652 static inline unsigned long file_page_index(struct bitmap_storage *store,
653 unsigned long chunk)
654 {
655 if (store->sb_page)
656 chunk += sizeof(bitmap_super_t) << 3;
657 return chunk >> PAGE_BIT_SHIFT;
658 }
659
660 /* calculate the (bit) offset of this bit within a page */
661 static inline unsigned long file_page_offset(struct bitmap_storage *store,
662 unsigned long chunk)
663 {
664 if (store->sb_page)
665 chunk += sizeof(bitmap_super_t) << 3;
666 return chunk & (PAGE_BITS - 1);
667 }
668
669 /*
670 * return a pointer to the page in the filemap that contains the given bit
671 *
672 */
673 static inline struct page *filemap_get_page(struct bitmap_storage *store,
674 unsigned long chunk)
675 {
676 if (file_page_index(store, chunk) >= store->file_pages)
677 return NULL;
678 return store->filemap[file_page_index(store, chunk)];
679 }
680
681 static int bitmap_storage_alloc(struct bitmap_storage *store,
682 unsigned long chunks, int with_super)
683 {
684 int pnum;
685 unsigned long num_pages;
686 unsigned long bytes;
687
688 bytes = DIV_ROUND_UP(chunks, 8);
689 if (with_super)
690 bytes += sizeof(bitmap_super_t);
691
692 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
693
694 store->filemap = kmalloc(sizeof(struct page *)
695 * num_pages, GFP_KERNEL);
696 if (!store->filemap)
697 return -ENOMEM;
698
699 if (with_super && !store->sb_page) {
700 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
701 if (store->sb_page == NULL)
702 return -ENOMEM;
703 store->sb_page->index = 0;
704 }
705 pnum = 0;
706 if (store->sb_page) {
707 store->filemap[0] = store->sb_page;
708 pnum = 1;
709 }
710 for ( ; pnum < num_pages; pnum++) {
711 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
712 if (!store->filemap[pnum]) {
713 store->file_pages = pnum;
714 return -ENOMEM;
715 }
716 store->filemap[pnum]->index = pnum;
717 }
718 store->file_pages = pnum;
719
720 /* We need 4 bits per page, rounded up to a multiple
721 * of sizeof(unsigned long) */
722 store->filemap_attr = kzalloc(
723 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
724 GFP_KERNEL);
725 if (!store->filemap_attr)
726 return -ENOMEM;
727
728 store->bytes = bytes;
729
730 return 0;
731 }
732
733 static void bitmap_file_unmap(struct bitmap_storage *store)
734 {
735 struct page **map, *sb_page;
736 int pages;
737 struct file *file;
738
739 file = store->file;
740 map = store->filemap;
741 pages = store->file_pages;
742 sb_page = store->sb_page;
743
744 while (pages--)
745 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
746 free_buffers(map[pages]);
747 kfree(map);
748 kfree(store->filemap_attr);
749
750 if (sb_page)
751 free_buffers(sb_page);
752
753 if (file) {
754 struct inode *inode = file_inode(file);
755 invalidate_mapping_pages(inode->i_mapping, 0, -1);
756 fput(file);
757 }
758 }
759
760 /*
761 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
762 * then it is no longer reliable, so we stop using it and we mark the file
763 * as failed in the superblock
764 */
765 static void bitmap_file_kick(struct bitmap *bitmap)
766 {
767 char *path, *ptr = NULL;
768
769 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
770 bitmap_update_sb(bitmap);
771
772 if (bitmap->storage.file) {
773 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
774 if (path)
775 ptr = d_path(&bitmap->storage.file->f_path,
776 path, PAGE_SIZE);
777
778 printk(KERN_ALERT
779 "%s: kicking failed bitmap file %s from array!\n",
780 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
781
782 kfree(path);
783 } else
784 printk(KERN_ALERT
785 "%s: disabling internal bitmap due to errors\n",
786 bmname(bitmap));
787 }
788 }
789
790 enum bitmap_page_attr {
791 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
792 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
793 * i.e. counter is 1 or 2. */
794 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
795 };
796
797 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
798 enum bitmap_page_attr attr)
799 {
800 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
801 }
802
803 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
804 enum bitmap_page_attr attr)
805 {
806 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
807 }
808
809 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
810 enum bitmap_page_attr attr)
811 {
812 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
813 }
814
815 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
816 enum bitmap_page_attr attr)
817 {
818 return test_and_clear_bit((pnum<<2) + attr,
819 bitmap->storage.filemap_attr);
820 }
821 /*
822 * bitmap_file_set_bit -- called before performing a write to the md device
823 * to set (and eventually sync) a particular bit in the bitmap file
824 *
825 * we set the bit immediately, then we record the page number so that
826 * when an unplug occurs, we can flush the dirty pages out to disk
827 */
828 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
829 {
830 unsigned long bit;
831 struct page *page;
832 void *kaddr;
833 unsigned long chunk = block >> bitmap->counts.chunkshift;
834
835 page = filemap_get_page(&bitmap->storage, chunk);
836 if (!page)
837 return;
838 bit = file_page_offset(&bitmap->storage, chunk);
839
840 /* set the bit */
841 kaddr = kmap_atomic(page);
842 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
843 set_bit(bit, kaddr);
844 else
845 set_bit_le(bit, kaddr);
846 kunmap_atomic(kaddr);
847 pr_debug("set file bit %lu page %lu\n", bit, page->index);
848 /* record page number so it gets flushed to disk when unplug occurs */
849 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
850 }
851
852 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
853 {
854 unsigned long bit;
855 struct page *page;
856 void *paddr;
857 unsigned long chunk = block >> bitmap->counts.chunkshift;
858
859 page = filemap_get_page(&bitmap->storage, chunk);
860 if (!page)
861 return;
862 bit = file_page_offset(&bitmap->storage, chunk);
863 paddr = kmap_atomic(page);
864 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
865 clear_bit(bit, paddr);
866 else
867 clear_bit_le(bit, paddr);
868 kunmap_atomic(paddr);
869 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
870 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
871 bitmap->allclean = 0;
872 }
873 }
874
875 /* this gets called when the md device is ready to unplug its underlying
876 * (slave) device queues -- before we let any writes go down, we need to
877 * sync the dirty pages of the bitmap file to disk */
878 void bitmap_unplug(struct bitmap *bitmap)
879 {
880 unsigned long i;
881 int dirty, need_write;
882 int wait = 0;
883
884 if (!bitmap || !bitmap->storage.filemap ||
885 test_bit(BITMAP_STALE, &bitmap->flags))
886 return;
887
888 /* look at each page to see if there are any set bits that need to be
889 * flushed out to disk */
890 for (i = 0; i < bitmap->storage.file_pages; i++) {
891 if (!bitmap->storage.filemap)
892 return;
893 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
894 need_write = test_and_clear_page_attr(bitmap, i,
895 BITMAP_PAGE_NEEDWRITE);
896 if (dirty || need_write) {
897 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
898 write_page(bitmap, bitmap->storage.filemap[i], 0);
899 }
900 if (dirty)
901 wait = 1;
902 }
903 if (wait) { /* if any writes were performed, we need to wait on them */
904 if (bitmap->storage.file)
905 wait_event(bitmap->write_wait,
906 atomic_read(&bitmap->pending_writes)==0);
907 else
908 md_super_wait(bitmap->mddev);
909 }
910 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
911 bitmap_file_kick(bitmap);
912 }
913 EXPORT_SYMBOL(bitmap_unplug);
914
915 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
916 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
917 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
918 * memory mapping of the bitmap file
919 * Special cases:
920 * if there's no bitmap file, or if the bitmap file had been
921 * previously kicked from the array, we mark all the bits as
922 * 1's in order to cause a full resync.
923 *
924 * We ignore all bits for sectors that end earlier than 'start'.
925 * This is used when reading an out-of-date bitmap...
926 */
927 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
928 {
929 unsigned long i, chunks, index, oldindex, bit;
930 struct page *page = NULL;
931 unsigned long bit_cnt = 0;
932 struct file *file;
933 unsigned long offset;
934 int outofdate;
935 int ret = -ENOSPC;
936 void *paddr;
937 struct bitmap_storage *store = &bitmap->storage;
938
939 chunks = bitmap->counts.chunks;
940 file = store->file;
941
942 if (!file && !bitmap->mddev->bitmap_info.offset) {
943 /* No permanent bitmap - fill with '1s'. */
944 store->filemap = NULL;
945 store->file_pages = 0;
946 for (i = 0; i < chunks ; i++) {
947 /* if the disk bit is set, set the memory bit */
948 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
949 >= start);
950 bitmap_set_memory_bits(bitmap,
951 (sector_t)i << bitmap->counts.chunkshift,
952 needed);
953 }
954 return 0;
955 }
956
957 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
958 if (outofdate)
959 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
960 "recovery\n", bmname(bitmap));
961
962 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
963 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
964 bmname(bitmap),
965 (unsigned long) i_size_read(file->f_mapping->host),
966 store->bytes);
967 goto err;
968 }
969
970 oldindex = ~0L;
971 offset = 0;
972 if (!bitmap->mddev->bitmap_info.external)
973 offset = sizeof(bitmap_super_t);
974
975 for (i = 0; i < chunks; i++) {
976 int b;
977 index = file_page_index(&bitmap->storage, i);
978 bit = file_page_offset(&bitmap->storage, i);
979 if (index != oldindex) { /* this is a new page, read it in */
980 int count;
981 /* unmap the old page, we're done with it */
982 if (index == store->file_pages-1)
983 count = store->bytes - index * PAGE_SIZE;
984 else
985 count = PAGE_SIZE;
986 page = store->filemap[index];
987 if (file)
988 ret = read_page(file, index, bitmap,
989 count, page);
990 else
991 ret = read_sb_page(
992 bitmap->mddev,
993 bitmap->mddev->bitmap_info.offset,
994 page,
995 index, count);
996
997 if (ret)
998 goto err;
999
1000 oldindex = index;
1001
1002 if (outofdate) {
1003 /*
1004 * if bitmap is out of date, dirty the
1005 * whole page and write it out
1006 */
1007 paddr = kmap_atomic(page);
1008 memset(paddr + offset, 0xff,
1009 PAGE_SIZE - offset);
1010 kunmap_atomic(paddr);
1011 write_page(bitmap, page, 1);
1012
1013 ret = -EIO;
1014 if (test_bit(BITMAP_WRITE_ERROR,
1015 &bitmap->flags))
1016 goto err;
1017 }
1018 }
1019 paddr = kmap_atomic(page);
1020 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1021 b = test_bit(bit, paddr);
1022 else
1023 b = test_bit_le(bit, paddr);
1024 kunmap_atomic(paddr);
1025 if (b) {
1026 /* if the disk bit is set, set the memory bit */
1027 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1028 >= start);
1029 bitmap_set_memory_bits(bitmap,
1030 (sector_t)i << bitmap->counts.chunkshift,
1031 needed);
1032 bit_cnt++;
1033 }
1034 offset = 0;
1035 }
1036
1037 printk(KERN_INFO "%s: bitmap initialized from disk: "
1038 "read %lu pages, set %lu of %lu bits\n",
1039 bmname(bitmap), store->file_pages,
1040 bit_cnt, chunks);
1041
1042 return 0;
1043
1044 err:
1045 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1046 bmname(bitmap), ret);
1047 return ret;
1048 }
1049
1050 void bitmap_write_all(struct bitmap *bitmap)
1051 {
1052 /* We don't actually write all bitmap blocks here,
1053 * just flag them as needing to be written
1054 */
1055 int i;
1056
1057 if (!bitmap || !bitmap->storage.filemap)
1058 return;
1059 if (bitmap->storage.file)
1060 /* Only one copy, so nothing needed */
1061 return;
1062
1063 for (i = 0; i < bitmap->storage.file_pages; i++)
1064 set_page_attr(bitmap, i,
1065 BITMAP_PAGE_NEEDWRITE);
1066 bitmap->allclean = 0;
1067 }
1068
1069 static void bitmap_count_page(struct bitmap_counts *bitmap,
1070 sector_t offset, int inc)
1071 {
1072 sector_t chunk = offset >> bitmap->chunkshift;
1073 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1074 bitmap->bp[page].count += inc;
1075 bitmap_checkfree(bitmap, page);
1076 }
1077
1078 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1079 {
1080 sector_t chunk = offset >> bitmap->chunkshift;
1081 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1082 struct bitmap_page *bp = &bitmap->bp[page];
1083
1084 if (!bp->pending)
1085 bp->pending = 1;
1086 }
1087
1088 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1089 sector_t offset, sector_t *blocks,
1090 int create);
1091
1092 /*
1093 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1094 * out to disk
1095 */
1096
1097 void bitmap_daemon_work(struct mddev *mddev)
1098 {
1099 struct bitmap *bitmap;
1100 unsigned long j;
1101 unsigned long nextpage;
1102 sector_t blocks;
1103 struct bitmap_counts *counts;
1104
1105 /* Use a mutex to guard daemon_work against
1106 * bitmap_destroy.
1107 */
1108 mutex_lock(&mddev->bitmap_info.mutex);
1109 bitmap = mddev->bitmap;
1110 if (bitmap == NULL) {
1111 mutex_unlock(&mddev->bitmap_info.mutex);
1112 return;
1113 }
1114 if (time_before(jiffies, bitmap->daemon_lastrun
1115 + mddev->bitmap_info.daemon_sleep))
1116 goto done;
1117
1118 bitmap->daemon_lastrun = jiffies;
1119 if (bitmap->allclean) {
1120 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1121 goto done;
1122 }
1123 bitmap->allclean = 1;
1124
1125 /* Any file-page which is PENDING now needs to be written.
1126 * So set NEEDWRITE now, then after we make any last-minute changes
1127 * we will write it.
1128 */
1129 for (j = 0; j < bitmap->storage.file_pages; j++)
1130 if (test_and_clear_page_attr(bitmap, j,
1131 BITMAP_PAGE_PENDING))
1132 set_page_attr(bitmap, j,
1133 BITMAP_PAGE_NEEDWRITE);
1134
1135 if (bitmap->need_sync &&
1136 mddev->bitmap_info.external == 0) {
1137 /* Arrange for superblock update as well as
1138 * other changes */
1139 bitmap_super_t *sb;
1140 bitmap->need_sync = 0;
1141 if (bitmap->storage.filemap) {
1142 sb = kmap_atomic(bitmap->storage.sb_page);
1143 sb->events_cleared =
1144 cpu_to_le64(bitmap->events_cleared);
1145 kunmap_atomic(sb);
1146 set_page_attr(bitmap, 0,
1147 BITMAP_PAGE_NEEDWRITE);
1148 }
1149 }
1150 /* Now look at the bitmap counters and if any are '2' or '1',
1151 * decrement and handle accordingly.
1152 */
1153 counts = &bitmap->counts;
1154 spin_lock_irq(&counts->lock);
1155 nextpage = 0;
1156 for (j = 0; j < counts->chunks; j++) {
1157 bitmap_counter_t *bmc;
1158 sector_t block = (sector_t)j << counts->chunkshift;
1159
1160 if (j == nextpage) {
1161 nextpage += PAGE_COUNTER_RATIO;
1162 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1163 j |= PAGE_COUNTER_MASK;
1164 continue;
1165 }
1166 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1167 }
1168 bmc = bitmap_get_counter(counts,
1169 block,
1170 &blocks, 0);
1171
1172 if (!bmc) {
1173 j |= PAGE_COUNTER_MASK;
1174 continue;
1175 }
1176 if (*bmc == 1 && !bitmap->need_sync) {
1177 /* We can clear the bit */
1178 *bmc = 0;
1179 bitmap_count_page(counts, block, -1);
1180 bitmap_file_clear_bit(bitmap, block);
1181 } else if (*bmc && *bmc <= 2) {
1182 *bmc = 1;
1183 bitmap_set_pending(counts, block);
1184 bitmap->allclean = 0;
1185 }
1186 }
1187 spin_unlock_irq(&counts->lock);
1188
1189 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1190 * DIRTY pages need to be written by bitmap_unplug so it can wait
1191 * for them.
1192 * If we find any DIRTY page we stop there and let bitmap_unplug
1193 * handle all the rest. This is important in the case where
1194 * the first blocking holds the superblock and it has been updated.
1195 * We mustn't write any other blocks before the superblock.
1196 */
1197 for (j = 0;
1198 j < bitmap->storage.file_pages
1199 && !test_bit(BITMAP_STALE, &bitmap->flags);
1200 j++) {
1201
1202 if (test_page_attr(bitmap, j,
1203 BITMAP_PAGE_DIRTY))
1204 /* bitmap_unplug will handle the rest */
1205 break;
1206 if (test_and_clear_page_attr(bitmap, j,
1207 BITMAP_PAGE_NEEDWRITE)) {
1208 write_page(bitmap, bitmap->storage.filemap[j], 0);
1209 }
1210 }
1211
1212 done:
1213 if (bitmap->allclean == 0)
1214 mddev->thread->timeout =
1215 mddev->bitmap_info.daemon_sleep;
1216 mutex_unlock(&mddev->bitmap_info.mutex);
1217 }
1218
1219 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1220 sector_t offset, sector_t *blocks,
1221 int create)
1222 __releases(bitmap->lock)
1223 __acquires(bitmap->lock)
1224 {
1225 /* If 'create', we might release the lock and reclaim it.
1226 * The lock must have been taken with interrupts enabled.
1227 * If !create, we don't release the lock.
1228 */
1229 sector_t chunk = offset >> bitmap->chunkshift;
1230 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1231 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1232 sector_t csize;
1233 int err;
1234
1235 err = bitmap_checkpage(bitmap, page, create);
1236
1237 if (bitmap->bp[page].hijacked ||
1238 bitmap->bp[page].map == NULL)
1239 csize = ((sector_t)1) << (bitmap->chunkshift +
1240 PAGE_COUNTER_SHIFT - 1);
1241 else
1242 csize = ((sector_t)1) << bitmap->chunkshift;
1243 *blocks = csize - (offset & (csize - 1));
1244
1245 if (err < 0)
1246 return NULL;
1247
1248 /* now locked ... */
1249
1250 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1251 /* should we use the first or second counter field
1252 * of the hijacked pointer? */
1253 int hi = (pageoff > PAGE_COUNTER_MASK);
1254 return &((bitmap_counter_t *)
1255 &bitmap->bp[page].map)[hi];
1256 } else /* page is allocated */
1257 return (bitmap_counter_t *)
1258 &(bitmap->bp[page].map[pageoff]);
1259 }
1260
1261 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1262 {
1263 if (!bitmap)
1264 return 0;
1265
1266 if (behind) {
1267 int bw;
1268 atomic_inc(&bitmap->behind_writes);
1269 bw = atomic_read(&bitmap->behind_writes);
1270 if (bw > bitmap->behind_writes_used)
1271 bitmap->behind_writes_used = bw;
1272
1273 pr_debug("inc write-behind count %d/%lu\n",
1274 bw, bitmap->mddev->bitmap_info.max_write_behind);
1275 }
1276
1277 while (sectors) {
1278 sector_t blocks;
1279 bitmap_counter_t *bmc;
1280
1281 spin_lock_irq(&bitmap->counts.lock);
1282 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1283 if (!bmc) {
1284 spin_unlock_irq(&bitmap->counts.lock);
1285 return 0;
1286 }
1287
1288 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1289 DEFINE_WAIT(__wait);
1290 /* note that it is safe to do the prepare_to_wait
1291 * after the test as long as we do it before dropping
1292 * the spinlock.
1293 */
1294 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1295 TASK_UNINTERRUPTIBLE);
1296 spin_unlock_irq(&bitmap->counts.lock);
1297 schedule();
1298 finish_wait(&bitmap->overflow_wait, &__wait);
1299 continue;
1300 }
1301
1302 switch (*bmc) {
1303 case 0:
1304 bitmap_file_set_bit(bitmap, offset);
1305 bitmap_count_page(&bitmap->counts, offset, 1);
1306 /* fall through */
1307 case 1:
1308 *bmc = 2;
1309 }
1310
1311 (*bmc)++;
1312
1313 spin_unlock_irq(&bitmap->counts.lock);
1314
1315 offset += blocks;
1316 if (sectors > blocks)
1317 sectors -= blocks;
1318 else
1319 sectors = 0;
1320 }
1321 return 0;
1322 }
1323 EXPORT_SYMBOL(bitmap_startwrite);
1324
1325 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1326 int success, int behind)
1327 {
1328 if (!bitmap)
1329 return;
1330 if (behind) {
1331 if (atomic_dec_and_test(&bitmap->behind_writes))
1332 wake_up(&bitmap->behind_wait);
1333 pr_debug("dec write-behind count %d/%lu\n",
1334 atomic_read(&bitmap->behind_writes),
1335 bitmap->mddev->bitmap_info.max_write_behind);
1336 }
1337
1338 while (sectors) {
1339 sector_t blocks;
1340 unsigned long flags;
1341 bitmap_counter_t *bmc;
1342
1343 spin_lock_irqsave(&bitmap->counts.lock, flags);
1344 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1345 if (!bmc) {
1346 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1347 return;
1348 }
1349
1350 if (success && !bitmap->mddev->degraded &&
1351 bitmap->events_cleared < bitmap->mddev->events) {
1352 bitmap->events_cleared = bitmap->mddev->events;
1353 bitmap->need_sync = 1;
1354 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1355 }
1356
1357 if (!success && !NEEDED(*bmc))
1358 *bmc |= NEEDED_MASK;
1359
1360 if (COUNTER(*bmc) == COUNTER_MAX)
1361 wake_up(&bitmap->overflow_wait);
1362
1363 (*bmc)--;
1364 if (*bmc <= 2) {
1365 bitmap_set_pending(&bitmap->counts, offset);
1366 bitmap->allclean = 0;
1367 }
1368 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1369 offset += blocks;
1370 if (sectors > blocks)
1371 sectors -= blocks;
1372 else
1373 sectors = 0;
1374 }
1375 }
1376 EXPORT_SYMBOL(bitmap_endwrite);
1377
1378 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1379 int degraded)
1380 {
1381 bitmap_counter_t *bmc;
1382 int rv;
1383 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1384 *blocks = 1024;
1385 return 1; /* always resync if no bitmap */
1386 }
1387 spin_lock_irq(&bitmap->counts.lock);
1388 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1389 rv = 0;
1390 if (bmc) {
1391 /* locked */
1392 if (RESYNC(*bmc))
1393 rv = 1;
1394 else if (NEEDED(*bmc)) {
1395 rv = 1;
1396 if (!degraded) { /* don't set/clear bits if degraded */
1397 *bmc |= RESYNC_MASK;
1398 *bmc &= ~NEEDED_MASK;
1399 }
1400 }
1401 }
1402 spin_unlock_irq(&bitmap->counts.lock);
1403 return rv;
1404 }
1405
1406 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1407 int degraded)
1408 {
1409 /* bitmap_start_sync must always report on multiples of whole
1410 * pages, otherwise resync (which is very PAGE_SIZE based) will
1411 * get confused.
1412 * So call __bitmap_start_sync repeatedly (if needed) until
1413 * At least PAGE_SIZE>>9 blocks are covered.
1414 * Return the 'or' of the result.
1415 */
1416 int rv = 0;
1417 sector_t blocks1;
1418
1419 *blocks = 0;
1420 while (*blocks < (PAGE_SIZE>>9)) {
1421 rv |= __bitmap_start_sync(bitmap, offset,
1422 &blocks1, degraded);
1423 offset += blocks1;
1424 *blocks += blocks1;
1425 }
1426 return rv;
1427 }
1428 EXPORT_SYMBOL(bitmap_start_sync);
1429
1430 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1431 {
1432 bitmap_counter_t *bmc;
1433 unsigned long flags;
1434
1435 if (bitmap == NULL) {
1436 *blocks = 1024;
1437 return;
1438 }
1439 spin_lock_irqsave(&bitmap->counts.lock, flags);
1440 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1441 if (bmc == NULL)
1442 goto unlock;
1443 /* locked */
1444 if (RESYNC(*bmc)) {
1445 *bmc &= ~RESYNC_MASK;
1446
1447 if (!NEEDED(*bmc) && aborted)
1448 *bmc |= NEEDED_MASK;
1449 else {
1450 if (*bmc <= 2) {
1451 bitmap_set_pending(&bitmap->counts, offset);
1452 bitmap->allclean = 0;
1453 }
1454 }
1455 }
1456 unlock:
1457 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1458 }
1459 EXPORT_SYMBOL(bitmap_end_sync);
1460
1461 void bitmap_close_sync(struct bitmap *bitmap)
1462 {
1463 /* Sync has finished, and any bitmap chunks that weren't synced
1464 * properly have been aborted. It remains to us to clear the
1465 * RESYNC bit wherever it is still on
1466 */
1467 sector_t sector = 0;
1468 sector_t blocks;
1469 if (!bitmap)
1470 return;
1471 while (sector < bitmap->mddev->resync_max_sectors) {
1472 bitmap_end_sync(bitmap, sector, &blocks, 0);
1473 sector += blocks;
1474 }
1475 }
1476 EXPORT_SYMBOL(bitmap_close_sync);
1477
1478 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1479 {
1480 sector_t s = 0;
1481 sector_t blocks;
1482
1483 if (!bitmap)
1484 return;
1485 if (sector == 0) {
1486 bitmap->last_end_sync = jiffies;
1487 return;
1488 }
1489 if (time_before(jiffies, (bitmap->last_end_sync
1490 + bitmap->mddev->bitmap_info.daemon_sleep)))
1491 return;
1492 wait_event(bitmap->mddev->recovery_wait,
1493 atomic_read(&bitmap->mddev->recovery_active) == 0);
1494
1495 bitmap->mddev->curr_resync_completed = sector;
1496 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1497 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1498 s = 0;
1499 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1500 bitmap_end_sync(bitmap, s, &blocks, 0);
1501 s += blocks;
1502 }
1503 bitmap->last_end_sync = jiffies;
1504 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1505 }
1506 EXPORT_SYMBOL(bitmap_cond_end_sync);
1507
1508 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1509 {
1510 /* For each chunk covered by any of these sectors, set the
1511 * counter to 2 and possibly set resync_needed. They should all
1512 * be 0 at this point
1513 */
1514
1515 sector_t secs;
1516 bitmap_counter_t *bmc;
1517 spin_lock_irq(&bitmap->counts.lock);
1518 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1519 if (!bmc) {
1520 spin_unlock_irq(&bitmap->counts.lock);
1521 return;
1522 }
1523 if (!*bmc) {
1524 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1525 bitmap_count_page(&bitmap->counts, offset, 1);
1526 bitmap_set_pending(&bitmap->counts, offset);
1527 bitmap->allclean = 0;
1528 }
1529 spin_unlock_irq(&bitmap->counts.lock);
1530 }
1531
1532 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1533 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1534 {
1535 unsigned long chunk;
1536
1537 for (chunk = s; chunk <= e; chunk++) {
1538 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1539 bitmap_set_memory_bits(bitmap, sec, 1);
1540 bitmap_file_set_bit(bitmap, sec);
1541 if (sec < bitmap->mddev->recovery_cp)
1542 /* We are asserting that the array is dirty,
1543 * so move the recovery_cp address back so
1544 * that it is obvious that it is dirty
1545 */
1546 bitmap->mddev->recovery_cp = sec;
1547 }
1548 }
1549
1550 /*
1551 * flush out any pending updates
1552 */
1553 void bitmap_flush(struct mddev *mddev)
1554 {
1555 struct bitmap *bitmap = mddev->bitmap;
1556 long sleep;
1557
1558 if (!bitmap) /* there was no bitmap */
1559 return;
1560
1561 /* run the daemon_work three time to ensure everything is flushed
1562 * that can be
1563 */
1564 sleep = mddev->bitmap_info.daemon_sleep * 2;
1565 bitmap->daemon_lastrun -= sleep;
1566 bitmap_daemon_work(mddev);
1567 bitmap->daemon_lastrun -= sleep;
1568 bitmap_daemon_work(mddev);
1569 bitmap->daemon_lastrun -= sleep;
1570 bitmap_daemon_work(mddev);
1571 bitmap_update_sb(bitmap);
1572 }
1573
1574 /*
1575 * free memory that was allocated
1576 */
1577 static void bitmap_free(struct bitmap *bitmap)
1578 {
1579 unsigned long k, pages;
1580 struct bitmap_page *bp;
1581
1582 if (!bitmap) /* there was no bitmap */
1583 return;
1584
1585 /* Shouldn't be needed - but just in case.... */
1586 wait_event(bitmap->write_wait,
1587 atomic_read(&bitmap->pending_writes) == 0);
1588
1589 /* release the bitmap file */
1590 bitmap_file_unmap(&bitmap->storage);
1591
1592 bp = bitmap->counts.bp;
1593 pages = bitmap->counts.pages;
1594
1595 /* free all allocated memory */
1596
1597 if (bp) /* deallocate the page memory */
1598 for (k = 0; k < pages; k++)
1599 if (bp[k].map && !bp[k].hijacked)
1600 kfree(bp[k].map);
1601 kfree(bp);
1602 kfree(bitmap);
1603 }
1604
1605 void bitmap_destroy(struct mddev *mddev)
1606 {
1607 struct bitmap *bitmap = mddev->bitmap;
1608
1609 if (!bitmap) /* there was no bitmap */
1610 return;
1611
1612 mutex_lock(&mddev->bitmap_info.mutex);
1613 mddev->bitmap = NULL; /* disconnect from the md device */
1614 mutex_unlock(&mddev->bitmap_info.mutex);
1615 if (mddev->thread)
1616 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1617
1618 if (bitmap->sysfs_can_clear)
1619 sysfs_put(bitmap->sysfs_can_clear);
1620
1621 bitmap_free(bitmap);
1622 }
1623
1624 /*
1625 * initialize the bitmap structure
1626 * if this returns an error, bitmap_destroy must be called to do clean up
1627 */
1628 int bitmap_create(struct mddev *mddev)
1629 {
1630 struct bitmap *bitmap;
1631 sector_t blocks = mddev->resync_max_sectors;
1632 struct file *file = mddev->bitmap_info.file;
1633 int err;
1634 struct kernfs_node *bm = NULL;
1635
1636 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1637
1638 BUG_ON(file && mddev->bitmap_info.offset);
1639
1640 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1641 if (!bitmap)
1642 return -ENOMEM;
1643
1644 spin_lock_init(&bitmap->counts.lock);
1645 atomic_set(&bitmap->pending_writes, 0);
1646 init_waitqueue_head(&bitmap->write_wait);
1647 init_waitqueue_head(&bitmap->overflow_wait);
1648 init_waitqueue_head(&bitmap->behind_wait);
1649
1650 bitmap->mddev = mddev;
1651
1652 if (mddev->kobj.sd)
1653 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1654 if (bm) {
1655 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1656 sysfs_put(bm);
1657 } else
1658 bitmap->sysfs_can_clear = NULL;
1659
1660 bitmap->storage.file = file;
1661 if (file) {
1662 get_file(file);
1663 /* As future accesses to this file will use bmap,
1664 * and bypass the page cache, we must sync the file
1665 * first.
1666 */
1667 vfs_fsync(file, 1);
1668 }
1669 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1670 if (!mddev->bitmap_info.external) {
1671 /*
1672 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1673 * instructing us to create a new on-disk bitmap instance.
1674 */
1675 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1676 err = bitmap_new_disk_sb(bitmap);
1677 else
1678 err = bitmap_read_sb(bitmap);
1679 } else {
1680 err = 0;
1681 if (mddev->bitmap_info.chunksize == 0 ||
1682 mddev->bitmap_info.daemon_sleep == 0)
1683 /* chunksize and time_base need to be
1684 * set first. */
1685 err = -EINVAL;
1686 }
1687 if (err)
1688 goto error;
1689
1690 bitmap->daemon_lastrun = jiffies;
1691 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1692 if (err)
1693 goto error;
1694
1695 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1696 bitmap->counts.pages, bmname(bitmap));
1697
1698 mddev->bitmap = bitmap;
1699 return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1700
1701 error:
1702 bitmap_free(bitmap);
1703 return err;
1704 }
1705
1706 int bitmap_load(struct mddev *mddev)
1707 {
1708 int err = 0;
1709 sector_t start = 0;
1710 sector_t sector = 0;
1711 struct bitmap *bitmap = mddev->bitmap;
1712
1713 if (!bitmap)
1714 goto out;
1715
1716 /* Clear out old bitmap info first: Either there is none, or we
1717 * are resuming after someone else has possibly changed things,
1718 * so we should forget old cached info.
1719 * All chunks should be clean, but some might need_sync.
1720 */
1721 while (sector < mddev->resync_max_sectors) {
1722 sector_t blocks;
1723 bitmap_start_sync(bitmap, sector, &blocks, 0);
1724 sector += blocks;
1725 }
1726 bitmap_close_sync(bitmap);
1727
1728 if (mddev->degraded == 0
1729 || bitmap->events_cleared == mddev->events)
1730 /* no need to keep dirty bits to optimise a
1731 * re-add of a missing device */
1732 start = mddev->recovery_cp;
1733
1734 mutex_lock(&mddev->bitmap_info.mutex);
1735 err = bitmap_init_from_disk(bitmap, start);
1736 mutex_unlock(&mddev->bitmap_info.mutex);
1737
1738 if (err)
1739 goto out;
1740 clear_bit(BITMAP_STALE, &bitmap->flags);
1741
1742 /* Kick recovery in case any bits were set */
1743 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1744
1745 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1746 md_wakeup_thread(mddev->thread);
1747
1748 bitmap_update_sb(bitmap);
1749
1750 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1751 err = -EIO;
1752 out:
1753 return err;
1754 }
1755 EXPORT_SYMBOL_GPL(bitmap_load);
1756
1757 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1758 {
1759 unsigned long chunk_kb;
1760 struct bitmap_counts *counts;
1761
1762 if (!bitmap)
1763 return;
1764
1765 counts = &bitmap->counts;
1766
1767 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1768 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1769 "%lu%s chunk",
1770 counts->pages - counts->missing_pages,
1771 counts->pages,
1772 (counts->pages - counts->missing_pages)
1773 << (PAGE_SHIFT - 10),
1774 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1775 chunk_kb ? "KB" : "B");
1776 if (bitmap->storage.file) {
1777 seq_printf(seq, ", file: ");
1778 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1779 }
1780
1781 seq_printf(seq, "\n");
1782 }
1783
1784 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1785 int chunksize, int init)
1786 {
1787 /* If chunk_size is 0, choose an appropriate chunk size.
1788 * Then possibly allocate new storage space.
1789 * Then quiesce, copy bits, replace bitmap, and re-start
1790 *
1791 * This function is called both to set up the initial bitmap
1792 * and to resize the bitmap while the array is active.
1793 * If this happens as a result of the array being resized,
1794 * chunksize will be zero, and we need to choose a suitable
1795 * chunksize, otherwise we use what we are given.
1796 */
1797 struct bitmap_storage store;
1798 struct bitmap_counts old_counts;
1799 unsigned long chunks;
1800 sector_t block;
1801 sector_t old_blocks, new_blocks;
1802 int chunkshift;
1803 int ret = 0;
1804 long pages;
1805 struct bitmap_page *new_bp;
1806
1807 if (chunksize == 0) {
1808 /* If there is enough space, leave the chunk size unchanged,
1809 * else increase by factor of two until there is enough space.
1810 */
1811 long bytes;
1812 long space = bitmap->mddev->bitmap_info.space;
1813
1814 if (space == 0) {
1815 /* We don't know how much space there is, so limit
1816 * to current size - in sectors.
1817 */
1818 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1819 if (!bitmap->mddev->bitmap_info.external)
1820 bytes += sizeof(bitmap_super_t);
1821 space = DIV_ROUND_UP(bytes, 512);
1822 bitmap->mddev->bitmap_info.space = space;
1823 }
1824 chunkshift = bitmap->counts.chunkshift;
1825 chunkshift--;
1826 do {
1827 /* 'chunkshift' is shift from block size to chunk size */
1828 chunkshift++;
1829 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1830 bytes = DIV_ROUND_UP(chunks, 8);
1831 if (!bitmap->mddev->bitmap_info.external)
1832 bytes += sizeof(bitmap_super_t);
1833 } while (bytes > (space << 9));
1834 } else
1835 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1836
1837 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1838 memset(&store, 0, sizeof(store));
1839 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1840 ret = bitmap_storage_alloc(&store, chunks,
1841 !bitmap->mddev->bitmap_info.external);
1842 if (ret)
1843 goto err;
1844
1845 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
1846
1847 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
1848 ret = -ENOMEM;
1849 if (!new_bp) {
1850 bitmap_file_unmap(&store);
1851 goto err;
1852 }
1853
1854 if (!init)
1855 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
1856
1857 store.file = bitmap->storage.file;
1858 bitmap->storage.file = NULL;
1859
1860 if (store.sb_page && bitmap->storage.sb_page)
1861 memcpy(page_address(store.sb_page),
1862 page_address(bitmap->storage.sb_page),
1863 sizeof(bitmap_super_t));
1864 bitmap_file_unmap(&bitmap->storage);
1865 bitmap->storage = store;
1866
1867 old_counts = bitmap->counts;
1868 bitmap->counts.bp = new_bp;
1869 bitmap->counts.pages = pages;
1870 bitmap->counts.missing_pages = pages;
1871 bitmap->counts.chunkshift = chunkshift;
1872 bitmap->counts.chunks = chunks;
1873 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
1874 BITMAP_BLOCK_SHIFT);
1875
1876 blocks = min(old_counts.chunks << old_counts.chunkshift,
1877 chunks << chunkshift);
1878
1879 spin_lock_irq(&bitmap->counts.lock);
1880 for (block = 0; block < blocks; ) {
1881 bitmap_counter_t *bmc_old, *bmc_new;
1882 int set;
1883
1884 bmc_old = bitmap_get_counter(&old_counts, block,
1885 &old_blocks, 0);
1886 set = bmc_old && NEEDED(*bmc_old);
1887
1888 if (set) {
1889 bmc_new = bitmap_get_counter(&bitmap->counts, block,
1890 &new_blocks, 1);
1891 if (*bmc_new == 0) {
1892 /* need to set on-disk bits too. */
1893 sector_t end = block + new_blocks;
1894 sector_t start = block >> chunkshift;
1895 start <<= chunkshift;
1896 while (start < end) {
1897 bitmap_file_set_bit(bitmap, block);
1898 start += 1 << chunkshift;
1899 }
1900 *bmc_new = 2;
1901 bitmap_count_page(&bitmap->counts,
1902 block, 1);
1903 bitmap_set_pending(&bitmap->counts,
1904 block);
1905 }
1906 *bmc_new |= NEEDED_MASK;
1907 if (new_blocks < old_blocks)
1908 old_blocks = new_blocks;
1909 }
1910 block += old_blocks;
1911 }
1912
1913 if (!init) {
1914 int i;
1915 while (block < (chunks << chunkshift)) {
1916 bitmap_counter_t *bmc;
1917 bmc = bitmap_get_counter(&bitmap->counts, block,
1918 &new_blocks, 1);
1919 if (bmc) {
1920 /* new space. It needs to be resynced, so
1921 * we set NEEDED_MASK.
1922 */
1923 if (*bmc == 0) {
1924 *bmc = NEEDED_MASK | 2;
1925 bitmap_count_page(&bitmap->counts,
1926 block, 1);
1927 bitmap_set_pending(&bitmap->counts,
1928 block);
1929 }
1930 }
1931 block += new_blocks;
1932 }
1933 for (i = 0; i < bitmap->storage.file_pages; i++)
1934 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1935 }
1936 spin_unlock_irq(&bitmap->counts.lock);
1937
1938 if (!init) {
1939 bitmap_unplug(bitmap);
1940 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
1941 }
1942 ret = 0;
1943 err:
1944 return ret;
1945 }
1946 EXPORT_SYMBOL_GPL(bitmap_resize);
1947
1948 static ssize_t
1949 location_show(struct mddev *mddev, char *page)
1950 {
1951 ssize_t len;
1952 if (mddev->bitmap_info.file)
1953 len = sprintf(page, "file");
1954 else if (mddev->bitmap_info.offset)
1955 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1956 else
1957 len = sprintf(page, "none");
1958 len += sprintf(page+len, "\n");
1959 return len;
1960 }
1961
1962 static ssize_t
1963 location_store(struct mddev *mddev, const char *buf, size_t len)
1964 {
1965
1966 if (mddev->pers) {
1967 if (!mddev->pers->quiesce)
1968 return -EBUSY;
1969 if (mddev->recovery || mddev->sync_thread)
1970 return -EBUSY;
1971 }
1972
1973 if (mddev->bitmap || mddev->bitmap_info.file ||
1974 mddev->bitmap_info.offset) {
1975 /* bitmap already configured. Only option is to clear it */
1976 if (strncmp(buf, "none", 4) != 0)
1977 return -EBUSY;
1978 if (mddev->pers) {
1979 mddev->pers->quiesce(mddev, 1);
1980 bitmap_destroy(mddev);
1981 mddev->pers->quiesce(mddev, 0);
1982 }
1983 mddev->bitmap_info.offset = 0;
1984 if (mddev->bitmap_info.file) {
1985 struct file *f = mddev->bitmap_info.file;
1986 mddev->bitmap_info.file = NULL;
1987 fput(f);
1988 }
1989 } else {
1990 /* No bitmap, OK to set a location */
1991 long long offset;
1992 if (strncmp(buf, "none", 4) == 0)
1993 /* nothing to be done */;
1994 else if (strncmp(buf, "file:", 5) == 0) {
1995 /* Not supported yet */
1996 return -EINVAL;
1997 } else {
1998 int rv;
1999 if (buf[0] == '+')
2000 rv = kstrtoll(buf+1, 10, &offset);
2001 else
2002 rv = kstrtoll(buf, 10, &offset);
2003 if (rv)
2004 return rv;
2005 if (offset == 0)
2006 return -EINVAL;
2007 if (mddev->bitmap_info.external == 0 &&
2008 mddev->major_version == 0 &&
2009 offset != mddev->bitmap_info.default_offset)
2010 return -EINVAL;
2011 mddev->bitmap_info.offset = offset;
2012 if (mddev->pers) {
2013 mddev->pers->quiesce(mddev, 1);
2014 rv = bitmap_create(mddev);
2015 if (!rv)
2016 rv = bitmap_load(mddev);
2017 if (rv) {
2018 bitmap_destroy(mddev);
2019 mddev->bitmap_info.offset = 0;
2020 }
2021 mddev->pers->quiesce(mddev, 0);
2022 if (rv)
2023 return rv;
2024 }
2025 }
2026 }
2027 if (!mddev->external) {
2028 /* Ensure new bitmap info is stored in
2029 * metadata promptly.
2030 */
2031 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2032 md_wakeup_thread(mddev->thread);
2033 }
2034 return len;
2035 }
2036
2037 static struct md_sysfs_entry bitmap_location =
2038 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2039
2040 /* 'bitmap/space' is the space available at 'location' for the
2041 * bitmap. This allows the kernel to know when it is safe to
2042 * resize the bitmap to match a resized array.
2043 */
2044 static ssize_t
2045 space_show(struct mddev *mddev, char *page)
2046 {
2047 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2048 }
2049
2050 static ssize_t
2051 space_store(struct mddev *mddev, const char *buf, size_t len)
2052 {
2053 unsigned long sectors;
2054 int rv;
2055
2056 rv = kstrtoul(buf, 10, &sectors);
2057 if (rv)
2058 return rv;
2059
2060 if (sectors == 0)
2061 return -EINVAL;
2062
2063 if (mddev->bitmap &&
2064 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2065 return -EFBIG; /* Bitmap is too big for this small space */
2066
2067 /* could make sure it isn't too big, but that isn't really
2068 * needed - user-space should be careful.
2069 */
2070 mddev->bitmap_info.space = sectors;
2071 return len;
2072 }
2073
2074 static struct md_sysfs_entry bitmap_space =
2075 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2076
2077 static ssize_t
2078 timeout_show(struct mddev *mddev, char *page)
2079 {
2080 ssize_t len;
2081 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2082 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2083
2084 len = sprintf(page, "%lu", secs);
2085 if (jifs)
2086 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2087 len += sprintf(page+len, "\n");
2088 return len;
2089 }
2090
2091 static ssize_t
2092 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2093 {
2094 /* timeout can be set at any time */
2095 unsigned long timeout;
2096 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2097 if (rv)
2098 return rv;
2099
2100 /* just to make sure we don't overflow... */
2101 if (timeout >= LONG_MAX / HZ)
2102 return -EINVAL;
2103
2104 timeout = timeout * HZ / 10000;
2105
2106 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2107 timeout = MAX_SCHEDULE_TIMEOUT-1;
2108 if (timeout < 1)
2109 timeout = 1;
2110 mddev->bitmap_info.daemon_sleep = timeout;
2111 if (mddev->thread) {
2112 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2113 * the bitmap is all clean and we don't need to
2114 * adjust the timeout right now
2115 */
2116 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2117 mddev->thread->timeout = timeout;
2118 md_wakeup_thread(mddev->thread);
2119 }
2120 }
2121 return len;
2122 }
2123
2124 static struct md_sysfs_entry bitmap_timeout =
2125 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2126
2127 static ssize_t
2128 backlog_show(struct mddev *mddev, char *page)
2129 {
2130 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2131 }
2132
2133 static ssize_t
2134 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2135 {
2136 unsigned long backlog;
2137 int rv = kstrtoul(buf, 10, &backlog);
2138 if (rv)
2139 return rv;
2140 if (backlog > COUNTER_MAX)
2141 return -EINVAL;
2142 mddev->bitmap_info.max_write_behind = backlog;
2143 return len;
2144 }
2145
2146 static struct md_sysfs_entry bitmap_backlog =
2147 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2148
2149 static ssize_t
2150 chunksize_show(struct mddev *mddev, char *page)
2151 {
2152 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2153 }
2154
2155 static ssize_t
2156 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2157 {
2158 /* Can only be changed when no bitmap is active */
2159 int rv;
2160 unsigned long csize;
2161 if (mddev->bitmap)
2162 return -EBUSY;
2163 rv = kstrtoul(buf, 10, &csize);
2164 if (rv)
2165 return rv;
2166 if (csize < 512 ||
2167 !is_power_of_2(csize))
2168 return -EINVAL;
2169 mddev->bitmap_info.chunksize = csize;
2170 return len;
2171 }
2172
2173 static struct md_sysfs_entry bitmap_chunksize =
2174 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2175
2176 static ssize_t metadata_show(struct mddev *mddev, char *page)
2177 {
2178 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2179 ? "external" : "internal"));
2180 }
2181
2182 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2183 {
2184 if (mddev->bitmap ||
2185 mddev->bitmap_info.file ||
2186 mddev->bitmap_info.offset)
2187 return -EBUSY;
2188 if (strncmp(buf, "external", 8) == 0)
2189 mddev->bitmap_info.external = 1;
2190 else if (strncmp(buf, "internal", 8) == 0)
2191 mddev->bitmap_info.external = 0;
2192 else
2193 return -EINVAL;
2194 return len;
2195 }
2196
2197 static struct md_sysfs_entry bitmap_metadata =
2198 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2199
2200 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2201 {
2202 int len;
2203 if (mddev->bitmap)
2204 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2205 "false" : "true"));
2206 else
2207 len = sprintf(page, "\n");
2208 return len;
2209 }
2210
2211 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2212 {
2213 if (mddev->bitmap == NULL)
2214 return -ENOENT;
2215 if (strncmp(buf, "false", 5) == 0)
2216 mddev->bitmap->need_sync = 1;
2217 else if (strncmp(buf, "true", 4) == 0) {
2218 if (mddev->degraded)
2219 return -EBUSY;
2220 mddev->bitmap->need_sync = 0;
2221 } else
2222 return -EINVAL;
2223 return len;
2224 }
2225
2226 static struct md_sysfs_entry bitmap_can_clear =
2227 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2228
2229 static ssize_t
2230 behind_writes_used_show(struct mddev *mddev, char *page)
2231 {
2232 if (mddev->bitmap == NULL)
2233 return sprintf(page, "0\n");
2234 return sprintf(page, "%lu\n",
2235 mddev->bitmap->behind_writes_used);
2236 }
2237
2238 static ssize_t
2239 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2240 {
2241 if (mddev->bitmap)
2242 mddev->bitmap->behind_writes_used = 0;
2243 return len;
2244 }
2245
2246 static struct md_sysfs_entry max_backlog_used =
2247 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2248 behind_writes_used_show, behind_writes_used_reset);
2249
2250 static struct attribute *md_bitmap_attrs[] = {
2251 &bitmap_location.attr,
2252 &bitmap_space.attr,
2253 &bitmap_timeout.attr,
2254 &bitmap_backlog.attr,
2255 &bitmap_chunksize.attr,
2256 &bitmap_metadata.attr,
2257 &bitmap_can_clear.attr,
2258 &max_backlog_used.attr,
2259 NULL
2260 };
2261 struct attribute_group md_bitmap_group = {
2262 .name = "bitmap",
2263 .attrs = md_bitmap_attrs,
2264 };
2265
This page took 0.081594 seconds and 5 git commands to generate.