Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[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 /* It is possible that this is being called inside a
76 * prepare_to_wait/finish_wait loop from raid5c:make_request().
77 * In general it is not permitted to sleep in that context as it
78 * can cause the loop to spin freely.
79 * That doesn't apply here as we can only reach this point
80 * once with any loop.
81 * When this function completes, either bp[page].map or
82 * bp[page].hijacked. In either case, this function will
83 * abort before getting to this point again. So there is
84 * no risk of a free-spin, and so it is safe to assert
85 * that sleeping here is allowed.
86 */
87 sched_annotate_sleep();
88 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
89 spin_lock_irq(&bitmap->lock);
90
91 if (mappage == NULL) {
92 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
93 /* failed - set the hijacked flag so that we can use the
94 * pointer as a counter */
95 if (!bitmap->bp[page].map)
96 bitmap->bp[page].hijacked = 1;
97 } else if (bitmap->bp[page].map ||
98 bitmap->bp[page].hijacked) {
99 /* somebody beat us to getting the page */
100 kfree(mappage);
101 return 0;
102 } else {
103
104 /* no page was in place and we have one, so install it */
105
106 bitmap->bp[page].map = mappage;
107 bitmap->missing_pages--;
108 }
109 return 0;
110 }
111
112 /* if page is completely empty, put it back on the free list, or dealloc it */
113 /* if page was hijacked, unmark the flag so it might get alloced next time */
114 /* Note: lock should be held when calling this */
115 static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
116 {
117 char *ptr;
118
119 if (bitmap->bp[page].count) /* page is still busy */
120 return;
121
122 /* page is no longer in use, it can be released */
123
124 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
125 bitmap->bp[page].hijacked = 0;
126 bitmap->bp[page].map = NULL;
127 } else {
128 /* normal case, free the page */
129 ptr = bitmap->bp[page].map;
130 bitmap->bp[page].map = NULL;
131 bitmap->missing_pages++;
132 kfree(ptr);
133 }
134 }
135
136 /*
137 * bitmap file handling - read and write the bitmap file and its superblock
138 */
139
140 /*
141 * basic page I/O operations
142 */
143
144 /* IO operations when bitmap is stored near all superblocks */
145 static int read_sb_page(struct mddev *mddev, loff_t offset,
146 struct page *page,
147 unsigned long index, int size)
148 {
149 /* choose a good rdev and read the page from there */
150
151 struct md_rdev *rdev;
152 sector_t target;
153
154 rdev_for_each(rdev, mddev) {
155 if (! test_bit(In_sync, &rdev->flags)
156 || test_bit(Faulty, &rdev->flags))
157 continue;
158
159 target = offset + index * (PAGE_SIZE/512);
160
161 if (sync_page_io(rdev, target,
162 roundup(size, bdev_logical_block_size(rdev->bdev)),
163 page, READ, true)) {
164 page->index = index;
165 return 0;
166 }
167 }
168 return -EIO;
169 }
170
171 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
172 {
173 /* Iterate the disks of an mddev, using rcu to protect access to the
174 * linked list, and raising the refcount of devices we return to ensure
175 * they don't disappear while in use.
176 * As devices are only added or removed when raid_disk is < 0 and
177 * nr_pending is 0 and In_sync is clear, the entries we return will
178 * still be in the same position on the list when we re-enter
179 * list_for_each_entry_continue_rcu.
180 */
181 rcu_read_lock();
182 if (rdev == NULL)
183 /* start at the beginning */
184 rdev = list_entry_rcu(&mddev->disks, struct md_rdev, same_set);
185 else {
186 /* release the previous rdev and start from there. */
187 rdev_dec_pending(rdev, mddev);
188 }
189 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
190 if (rdev->raid_disk >= 0 &&
191 !test_bit(Faulty, &rdev->flags)) {
192 /* this is a usable devices */
193 atomic_inc(&rdev->nr_pending);
194 rcu_read_unlock();
195 return rdev;
196 }
197 }
198 rcu_read_unlock();
199 return NULL;
200 }
201
202 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
203 {
204 struct md_rdev *rdev = NULL;
205 struct block_device *bdev;
206 struct mddev *mddev = bitmap->mddev;
207 struct bitmap_storage *store = &bitmap->storage;
208
209 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
210 int size = PAGE_SIZE;
211 loff_t offset = mddev->bitmap_info.offset;
212
213 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
214
215 if (page->index == store->file_pages-1) {
216 int last_page_size = store->bytes & (PAGE_SIZE-1);
217 if (last_page_size == 0)
218 last_page_size = PAGE_SIZE;
219 size = roundup(last_page_size,
220 bdev_logical_block_size(bdev));
221 }
222 /* Just make sure we aren't corrupting data or
223 * metadata
224 */
225 if (mddev->external) {
226 /* Bitmap could be anywhere. */
227 if (rdev->sb_start + offset + (page->index
228 * (PAGE_SIZE/512))
229 > rdev->data_offset
230 &&
231 rdev->sb_start + offset
232 < (rdev->data_offset + mddev->dev_sectors
233 + (PAGE_SIZE/512)))
234 goto bad_alignment;
235 } else if (offset < 0) {
236 /* DATA BITMAP METADATA */
237 if (offset
238 + (long)(page->index * (PAGE_SIZE/512))
239 + size/512 > 0)
240 /* bitmap runs in to metadata */
241 goto bad_alignment;
242 if (rdev->data_offset + mddev->dev_sectors
243 > rdev->sb_start + offset)
244 /* data runs in to bitmap */
245 goto bad_alignment;
246 } else if (rdev->sb_start < rdev->data_offset) {
247 /* METADATA BITMAP DATA */
248 if (rdev->sb_start
249 + offset
250 + page->index*(PAGE_SIZE/512) + size/512
251 > rdev->data_offset)
252 /* bitmap runs in to data */
253 goto bad_alignment;
254 } else {
255 /* DATA METADATA BITMAP - no problems */
256 }
257 md_super_write(mddev, rdev,
258 rdev->sb_start + offset
259 + page->index * (PAGE_SIZE/512),
260 size,
261 page);
262 }
263
264 if (wait)
265 md_super_wait(mddev);
266 return 0;
267
268 bad_alignment:
269 return -EINVAL;
270 }
271
272 static void bitmap_file_kick(struct bitmap *bitmap);
273 /*
274 * write out a page to a file
275 */
276 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
277 {
278 struct buffer_head *bh;
279
280 if (bitmap->storage.file == NULL) {
281 switch (write_sb_page(bitmap, page, wait)) {
282 case -EINVAL:
283 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
284 }
285 } else {
286
287 bh = page_buffers(page);
288
289 while (bh && bh->b_blocknr) {
290 atomic_inc(&bitmap->pending_writes);
291 set_buffer_locked(bh);
292 set_buffer_mapped(bh);
293 submit_bh(WRITE | REQ_SYNC, bh);
294 bh = bh->b_this_page;
295 }
296
297 if (wait)
298 wait_event(bitmap->write_wait,
299 atomic_read(&bitmap->pending_writes)==0);
300 }
301 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
302 bitmap_file_kick(bitmap);
303 }
304
305 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
306 {
307 struct bitmap *bitmap = bh->b_private;
308
309 if (!uptodate)
310 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
311 if (atomic_dec_and_test(&bitmap->pending_writes))
312 wake_up(&bitmap->write_wait);
313 }
314
315 /* copied from buffer.c */
316 static void
317 __clear_page_buffers(struct page *page)
318 {
319 ClearPagePrivate(page);
320 set_page_private(page, 0);
321 page_cache_release(page);
322 }
323 static void free_buffers(struct page *page)
324 {
325 struct buffer_head *bh;
326
327 if (!PagePrivate(page))
328 return;
329
330 bh = page_buffers(page);
331 while (bh) {
332 struct buffer_head *next = bh->b_this_page;
333 free_buffer_head(bh);
334 bh = next;
335 }
336 __clear_page_buffers(page);
337 put_page(page);
338 }
339
340 /* read a page from a file.
341 * We both read the page, and attach buffers to the page to record the
342 * address of each block (using bmap). These addresses will be used
343 * to write the block later, completely bypassing the filesystem.
344 * This usage is similar to how swap files are handled, and allows us
345 * to write to a file with no concerns of memory allocation failing.
346 */
347 static int read_page(struct file *file, unsigned long index,
348 struct bitmap *bitmap,
349 unsigned long count,
350 struct page *page)
351 {
352 int ret = 0;
353 struct inode *inode = file_inode(file);
354 struct buffer_head *bh;
355 sector_t block;
356
357 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
358 (unsigned long long)index << PAGE_SHIFT);
359
360 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
361 if (!bh) {
362 ret = -ENOMEM;
363 goto out;
364 }
365 attach_page_buffers(page, bh);
366 block = index << (PAGE_SHIFT - inode->i_blkbits);
367 while (bh) {
368 if (count == 0)
369 bh->b_blocknr = 0;
370 else {
371 bh->b_blocknr = bmap(inode, block);
372 if (bh->b_blocknr == 0) {
373 /* Cannot use this file! */
374 ret = -EINVAL;
375 goto out;
376 }
377 bh->b_bdev = inode->i_sb->s_bdev;
378 if (count < (1<<inode->i_blkbits))
379 count = 0;
380 else
381 count -= (1<<inode->i_blkbits);
382
383 bh->b_end_io = end_bitmap_write;
384 bh->b_private = bitmap;
385 atomic_inc(&bitmap->pending_writes);
386 set_buffer_locked(bh);
387 set_buffer_mapped(bh);
388 submit_bh(READ, bh);
389 }
390 block++;
391 bh = bh->b_this_page;
392 }
393 page->index = index;
394
395 wait_event(bitmap->write_wait,
396 atomic_read(&bitmap->pending_writes)==0);
397 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
398 ret = -EIO;
399 out:
400 if (ret)
401 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
402 (int)PAGE_SIZE,
403 (unsigned long long)index << PAGE_SHIFT,
404 ret);
405 return ret;
406 }
407
408 /*
409 * bitmap file superblock operations
410 */
411
412 /* update the event counter and sync the superblock to disk */
413 void bitmap_update_sb(struct bitmap *bitmap)
414 {
415 bitmap_super_t *sb;
416
417 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
418 return;
419 if (bitmap->mddev->bitmap_info.external)
420 return;
421 if (!bitmap->storage.sb_page) /* no superblock */
422 return;
423 sb = kmap_atomic(bitmap->storage.sb_page);
424 sb->events = cpu_to_le64(bitmap->mddev->events);
425 if (bitmap->mddev->events < bitmap->events_cleared)
426 /* rocking back to read-only */
427 bitmap->events_cleared = bitmap->mddev->events;
428 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
429 sb->state = cpu_to_le32(bitmap->flags);
430 /* Just in case these have been changed via sysfs: */
431 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
432 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
433 /* This might have been changed by a reshape */
434 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
435 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
436 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
437 bitmap_info.space);
438 kunmap_atomic(sb);
439 write_page(bitmap, bitmap->storage.sb_page, 1);
440 }
441
442 /* print out the bitmap file superblock */
443 void bitmap_print_sb(struct bitmap *bitmap)
444 {
445 bitmap_super_t *sb;
446
447 if (!bitmap || !bitmap->storage.sb_page)
448 return;
449 sb = kmap_atomic(bitmap->storage.sb_page);
450 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
451 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
452 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
453 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
454 *(__u32 *)(sb->uuid+0),
455 *(__u32 *)(sb->uuid+4),
456 *(__u32 *)(sb->uuid+8),
457 *(__u32 *)(sb->uuid+12));
458 printk(KERN_DEBUG " events: %llu\n",
459 (unsigned long long) le64_to_cpu(sb->events));
460 printk(KERN_DEBUG "events cleared: %llu\n",
461 (unsigned long long) le64_to_cpu(sb->events_cleared));
462 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
463 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
464 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
465 printk(KERN_DEBUG " sync size: %llu KB\n",
466 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
467 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
468 kunmap_atomic(sb);
469 }
470
471 /*
472 * bitmap_new_disk_sb
473 * @bitmap
474 *
475 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
476 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
477 * This function verifies 'bitmap_info' and populates the on-disk bitmap
478 * structure, which is to be written to disk.
479 *
480 * Returns: 0 on success, -Exxx on error
481 */
482 static int bitmap_new_disk_sb(struct bitmap *bitmap)
483 {
484 bitmap_super_t *sb;
485 unsigned long chunksize, daemon_sleep, write_behind;
486
487 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
488 if (bitmap->storage.sb_page == NULL)
489 return -ENOMEM;
490 bitmap->storage.sb_page->index = 0;
491
492 sb = kmap_atomic(bitmap->storage.sb_page);
493
494 sb->magic = cpu_to_le32(BITMAP_MAGIC);
495 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
496
497 chunksize = bitmap->mddev->bitmap_info.chunksize;
498 BUG_ON(!chunksize);
499 if (!is_power_of_2(chunksize)) {
500 kunmap_atomic(sb);
501 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
502 return -EINVAL;
503 }
504 sb->chunksize = cpu_to_le32(chunksize);
505
506 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
507 if (!daemon_sleep ||
508 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
509 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
510 daemon_sleep = 5 * HZ;
511 }
512 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
513 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
514
515 /*
516 * FIXME: write_behind for RAID1. If not specified, what
517 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
518 */
519 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
520 if (write_behind > COUNTER_MAX)
521 write_behind = COUNTER_MAX / 2;
522 sb->write_behind = cpu_to_le32(write_behind);
523 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
524
525 /* keep the array size field of the bitmap superblock up to date */
526 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
527
528 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
529
530 set_bit(BITMAP_STALE, &bitmap->flags);
531 sb->state = cpu_to_le32(bitmap->flags);
532 bitmap->events_cleared = bitmap->mddev->events;
533 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
534
535 kunmap_atomic(sb);
536
537 return 0;
538 }
539
540 /* read the superblock from the bitmap file and initialize some bitmap fields */
541 static int bitmap_read_sb(struct bitmap *bitmap)
542 {
543 char *reason = NULL;
544 bitmap_super_t *sb;
545 unsigned long chunksize, daemon_sleep, write_behind;
546 unsigned long long events;
547 unsigned long sectors_reserved = 0;
548 int err = -EINVAL;
549 struct page *sb_page;
550
551 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
552 chunksize = 128 * 1024 * 1024;
553 daemon_sleep = 5 * HZ;
554 write_behind = 0;
555 set_bit(BITMAP_STALE, &bitmap->flags);
556 err = 0;
557 goto out_no_sb;
558 }
559 /* page 0 is the superblock, read it... */
560 sb_page = alloc_page(GFP_KERNEL);
561 if (!sb_page)
562 return -ENOMEM;
563 bitmap->storage.sb_page = sb_page;
564
565 if (bitmap->storage.file) {
566 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
567 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
568
569 err = read_page(bitmap->storage.file, 0,
570 bitmap, bytes, sb_page);
571 } else {
572 err = read_sb_page(bitmap->mddev,
573 bitmap->mddev->bitmap_info.offset,
574 sb_page,
575 0, sizeof(bitmap_super_t));
576 }
577 if (err)
578 return err;
579
580 sb = kmap_atomic(sb_page);
581
582 chunksize = le32_to_cpu(sb->chunksize);
583 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
584 write_behind = le32_to_cpu(sb->write_behind);
585 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
586
587 /* verify that the bitmap-specific fields are valid */
588 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
589 reason = "bad magic";
590 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
591 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
592 reason = "unrecognized superblock version";
593 else if (chunksize < 512)
594 reason = "bitmap chunksize too small";
595 else if (!is_power_of_2(chunksize))
596 reason = "bitmap chunksize not a power of 2";
597 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
598 reason = "daemon sleep period out of range";
599 else if (write_behind > COUNTER_MAX)
600 reason = "write-behind limit out of range (0 - 16383)";
601 if (reason) {
602 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
603 bmname(bitmap), reason);
604 goto out;
605 }
606
607 /* keep the array size field of the bitmap superblock up to date */
608 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
609
610 if (bitmap->mddev->persistent) {
611 /*
612 * We have a persistent array superblock, so compare the
613 * bitmap's UUID and event counter to the mddev's
614 */
615 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
616 printk(KERN_INFO
617 "%s: bitmap superblock UUID mismatch\n",
618 bmname(bitmap));
619 goto out;
620 }
621 events = le64_to_cpu(sb->events);
622 if (events < bitmap->mddev->events) {
623 printk(KERN_INFO
624 "%s: bitmap file is out of date (%llu < %llu) "
625 "-- forcing full recovery\n",
626 bmname(bitmap), events,
627 (unsigned long long) bitmap->mddev->events);
628 set_bit(BITMAP_STALE, &bitmap->flags);
629 }
630 }
631
632 /* assign fields using values from superblock */
633 bitmap->flags |= le32_to_cpu(sb->state);
634 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
635 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
636 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
637 err = 0;
638 out:
639 kunmap_atomic(sb);
640 out_no_sb:
641 if (test_bit(BITMAP_STALE, &bitmap->flags))
642 bitmap->events_cleared = bitmap->mddev->events;
643 bitmap->mddev->bitmap_info.chunksize = chunksize;
644 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
645 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
646 if (bitmap->mddev->bitmap_info.space == 0 ||
647 bitmap->mddev->bitmap_info.space > sectors_reserved)
648 bitmap->mddev->bitmap_info.space = sectors_reserved;
649 if (err)
650 bitmap_print_sb(bitmap);
651 return err;
652 }
653
654 /*
655 * general bitmap file operations
656 */
657
658 /*
659 * on-disk bitmap:
660 *
661 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
662 * file a page at a time. There's a superblock at the start of the file.
663 */
664 /* calculate the index of the page that contains this bit */
665 static inline unsigned long file_page_index(struct bitmap_storage *store,
666 unsigned long chunk)
667 {
668 if (store->sb_page)
669 chunk += sizeof(bitmap_super_t) << 3;
670 return chunk >> PAGE_BIT_SHIFT;
671 }
672
673 /* calculate the (bit) offset of this bit within a page */
674 static inline unsigned long file_page_offset(struct bitmap_storage *store,
675 unsigned long chunk)
676 {
677 if (store->sb_page)
678 chunk += sizeof(bitmap_super_t) << 3;
679 return chunk & (PAGE_BITS - 1);
680 }
681
682 /*
683 * return a pointer to the page in the filemap that contains the given bit
684 *
685 */
686 static inline struct page *filemap_get_page(struct bitmap_storage *store,
687 unsigned long chunk)
688 {
689 if (file_page_index(store, chunk) >= store->file_pages)
690 return NULL;
691 return store->filemap[file_page_index(store, chunk)];
692 }
693
694 static int bitmap_storage_alloc(struct bitmap_storage *store,
695 unsigned long chunks, int with_super)
696 {
697 int pnum;
698 unsigned long num_pages;
699 unsigned long bytes;
700
701 bytes = DIV_ROUND_UP(chunks, 8);
702 if (with_super)
703 bytes += sizeof(bitmap_super_t);
704
705 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
706
707 store->filemap = kmalloc(sizeof(struct page *)
708 * num_pages, GFP_KERNEL);
709 if (!store->filemap)
710 return -ENOMEM;
711
712 if (with_super && !store->sb_page) {
713 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
714 if (store->sb_page == NULL)
715 return -ENOMEM;
716 store->sb_page->index = 0;
717 }
718 pnum = 0;
719 if (store->sb_page) {
720 store->filemap[0] = store->sb_page;
721 pnum = 1;
722 }
723 for ( ; pnum < num_pages; pnum++) {
724 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
725 if (!store->filemap[pnum]) {
726 store->file_pages = pnum;
727 return -ENOMEM;
728 }
729 store->filemap[pnum]->index = pnum;
730 }
731 store->file_pages = pnum;
732
733 /* We need 4 bits per page, rounded up to a multiple
734 * of sizeof(unsigned long) */
735 store->filemap_attr = kzalloc(
736 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
737 GFP_KERNEL);
738 if (!store->filemap_attr)
739 return -ENOMEM;
740
741 store->bytes = bytes;
742
743 return 0;
744 }
745
746 static void bitmap_file_unmap(struct bitmap_storage *store)
747 {
748 struct page **map, *sb_page;
749 int pages;
750 struct file *file;
751
752 file = store->file;
753 map = store->filemap;
754 pages = store->file_pages;
755 sb_page = store->sb_page;
756
757 while (pages--)
758 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
759 free_buffers(map[pages]);
760 kfree(map);
761 kfree(store->filemap_attr);
762
763 if (sb_page)
764 free_buffers(sb_page);
765
766 if (file) {
767 struct inode *inode = file_inode(file);
768 invalidate_mapping_pages(inode->i_mapping, 0, -1);
769 fput(file);
770 }
771 }
772
773 /*
774 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
775 * then it is no longer reliable, so we stop using it and we mark the file
776 * as failed in the superblock
777 */
778 static void bitmap_file_kick(struct bitmap *bitmap)
779 {
780 char *path, *ptr = NULL;
781
782 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
783 bitmap_update_sb(bitmap);
784
785 if (bitmap->storage.file) {
786 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
787 if (path)
788 ptr = d_path(&bitmap->storage.file->f_path,
789 path, PAGE_SIZE);
790
791 printk(KERN_ALERT
792 "%s: kicking failed bitmap file %s from array!\n",
793 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
794
795 kfree(path);
796 } else
797 printk(KERN_ALERT
798 "%s: disabling internal bitmap due to errors\n",
799 bmname(bitmap));
800 }
801 }
802
803 enum bitmap_page_attr {
804 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
805 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
806 * i.e. counter is 1 or 2. */
807 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
808 };
809
810 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
811 enum bitmap_page_attr attr)
812 {
813 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
814 }
815
816 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
817 enum bitmap_page_attr attr)
818 {
819 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
820 }
821
822 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
823 enum bitmap_page_attr attr)
824 {
825 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
826 }
827
828 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
829 enum bitmap_page_attr attr)
830 {
831 return test_and_clear_bit((pnum<<2) + attr,
832 bitmap->storage.filemap_attr);
833 }
834 /*
835 * bitmap_file_set_bit -- called before performing a write to the md device
836 * to set (and eventually sync) a particular bit in the bitmap file
837 *
838 * we set the bit immediately, then we record the page number so that
839 * when an unplug occurs, we can flush the dirty pages out to disk
840 */
841 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
842 {
843 unsigned long bit;
844 struct page *page;
845 void *kaddr;
846 unsigned long chunk = block >> bitmap->counts.chunkshift;
847
848 page = filemap_get_page(&bitmap->storage, chunk);
849 if (!page)
850 return;
851 bit = file_page_offset(&bitmap->storage, chunk);
852
853 /* set the bit */
854 kaddr = kmap_atomic(page);
855 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
856 set_bit(bit, kaddr);
857 else
858 set_bit_le(bit, kaddr);
859 kunmap_atomic(kaddr);
860 pr_debug("set file bit %lu page %lu\n", bit, page->index);
861 /* record page number so it gets flushed to disk when unplug occurs */
862 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
863 }
864
865 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
866 {
867 unsigned long bit;
868 struct page *page;
869 void *paddr;
870 unsigned long chunk = block >> bitmap->counts.chunkshift;
871
872 page = filemap_get_page(&bitmap->storage, chunk);
873 if (!page)
874 return;
875 bit = file_page_offset(&bitmap->storage, chunk);
876 paddr = kmap_atomic(page);
877 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
878 clear_bit(bit, paddr);
879 else
880 clear_bit_le(bit, paddr);
881 kunmap_atomic(paddr);
882 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
883 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
884 bitmap->allclean = 0;
885 }
886 }
887
888 /* this gets called when the md device is ready to unplug its underlying
889 * (slave) device queues -- before we let any writes go down, we need to
890 * sync the dirty pages of the bitmap file to disk */
891 void bitmap_unplug(struct bitmap *bitmap)
892 {
893 unsigned long i;
894 int dirty, need_write;
895
896 if (!bitmap || !bitmap->storage.filemap ||
897 test_bit(BITMAP_STALE, &bitmap->flags))
898 return;
899
900 /* look at each page to see if there are any set bits that need to be
901 * flushed out to disk */
902 for (i = 0; i < bitmap->storage.file_pages; i++) {
903 if (!bitmap->storage.filemap)
904 return;
905 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
906 need_write = test_and_clear_page_attr(bitmap, i,
907 BITMAP_PAGE_NEEDWRITE);
908 if (dirty || need_write) {
909 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
910 write_page(bitmap, bitmap->storage.filemap[i], 0);
911 }
912 }
913 if (bitmap->storage.file)
914 wait_event(bitmap->write_wait,
915 atomic_read(&bitmap->pending_writes)==0);
916 else
917 md_super_wait(bitmap->mddev);
918
919 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
920 bitmap_file_kick(bitmap);
921 }
922 EXPORT_SYMBOL(bitmap_unplug);
923
924 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
925 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
926 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
927 * memory mapping of the bitmap file
928 * Special cases:
929 * if there's no bitmap file, or if the bitmap file had been
930 * previously kicked from the array, we mark all the bits as
931 * 1's in order to cause a full resync.
932 *
933 * We ignore all bits for sectors that end earlier than 'start'.
934 * This is used when reading an out-of-date bitmap...
935 */
936 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
937 {
938 unsigned long i, chunks, index, oldindex, bit;
939 struct page *page = NULL;
940 unsigned long bit_cnt = 0;
941 struct file *file;
942 unsigned long offset;
943 int outofdate;
944 int ret = -ENOSPC;
945 void *paddr;
946 struct bitmap_storage *store = &bitmap->storage;
947
948 chunks = bitmap->counts.chunks;
949 file = store->file;
950
951 if (!file && !bitmap->mddev->bitmap_info.offset) {
952 /* No permanent bitmap - fill with '1s'. */
953 store->filemap = NULL;
954 store->file_pages = 0;
955 for (i = 0; i < chunks ; i++) {
956 /* if the disk bit is set, set the memory bit */
957 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
958 >= start);
959 bitmap_set_memory_bits(bitmap,
960 (sector_t)i << bitmap->counts.chunkshift,
961 needed);
962 }
963 return 0;
964 }
965
966 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
967 if (outofdate)
968 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
969 "recovery\n", bmname(bitmap));
970
971 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
972 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
973 bmname(bitmap),
974 (unsigned long) i_size_read(file->f_mapping->host),
975 store->bytes);
976 goto err;
977 }
978
979 oldindex = ~0L;
980 offset = 0;
981 if (!bitmap->mddev->bitmap_info.external)
982 offset = sizeof(bitmap_super_t);
983
984 for (i = 0; i < chunks; i++) {
985 int b;
986 index = file_page_index(&bitmap->storage, i);
987 bit = file_page_offset(&bitmap->storage, i);
988 if (index != oldindex) { /* this is a new page, read it in */
989 int count;
990 /* unmap the old page, we're done with it */
991 if (index == store->file_pages-1)
992 count = store->bytes - index * PAGE_SIZE;
993 else
994 count = PAGE_SIZE;
995 page = store->filemap[index];
996 if (file)
997 ret = read_page(file, index, bitmap,
998 count, page);
999 else
1000 ret = read_sb_page(
1001 bitmap->mddev,
1002 bitmap->mddev->bitmap_info.offset,
1003 page,
1004 index, count);
1005
1006 if (ret)
1007 goto err;
1008
1009 oldindex = index;
1010
1011 if (outofdate) {
1012 /*
1013 * if bitmap is out of date, dirty the
1014 * whole page and write it out
1015 */
1016 paddr = kmap_atomic(page);
1017 memset(paddr + offset, 0xff,
1018 PAGE_SIZE - offset);
1019 kunmap_atomic(paddr);
1020 write_page(bitmap, page, 1);
1021
1022 ret = -EIO;
1023 if (test_bit(BITMAP_WRITE_ERROR,
1024 &bitmap->flags))
1025 goto err;
1026 }
1027 }
1028 paddr = kmap_atomic(page);
1029 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1030 b = test_bit(bit, paddr);
1031 else
1032 b = test_bit_le(bit, paddr);
1033 kunmap_atomic(paddr);
1034 if (b) {
1035 /* if the disk bit is set, set the memory bit */
1036 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1037 >= start);
1038 bitmap_set_memory_bits(bitmap,
1039 (sector_t)i << bitmap->counts.chunkshift,
1040 needed);
1041 bit_cnt++;
1042 }
1043 offset = 0;
1044 }
1045
1046 printk(KERN_INFO "%s: bitmap initialized from disk: "
1047 "read %lu pages, set %lu of %lu bits\n",
1048 bmname(bitmap), store->file_pages,
1049 bit_cnt, chunks);
1050
1051 return 0;
1052
1053 err:
1054 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1055 bmname(bitmap), ret);
1056 return ret;
1057 }
1058
1059 void bitmap_write_all(struct bitmap *bitmap)
1060 {
1061 /* We don't actually write all bitmap blocks here,
1062 * just flag them as needing to be written
1063 */
1064 int i;
1065
1066 if (!bitmap || !bitmap->storage.filemap)
1067 return;
1068 if (bitmap->storage.file)
1069 /* Only one copy, so nothing needed */
1070 return;
1071
1072 for (i = 0; i < bitmap->storage.file_pages; i++)
1073 set_page_attr(bitmap, i,
1074 BITMAP_PAGE_NEEDWRITE);
1075 bitmap->allclean = 0;
1076 }
1077
1078 static void bitmap_count_page(struct bitmap_counts *bitmap,
1079 sector_t offset, int inc)
1080 {
1081 sector_t chunk = offset >> bitmap->chunkshift;
1082 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1083 bitmap->bp[page].count += inc;
1084 bitmap_checkfree(bitmap, page);
1085 }
1086
1087 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1088 {
1089 sector_t chunk = offset >> bitmap->chunkshift;
1090 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1091 struct bitmap_page *bp = &bitmap->bp[page];
1092
1093 if (!bp->pending)
1094 bp->pending = 1;
1095 }
1096
1097 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1098 sector_t offset, sector_t *blocks,
1099 int create);
1100
1101 /*
1102 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1103 * out to disk
1104 */
1105
1106 void bitmap_daemon_work(struct mddev *mddev)
1107 {
1108 struct bitmap *bitmap;
1109 unsigned long j;
1110 unsigned long nextpage;
1111 sector_t blocks;
1112 struct bitmap_counts *counts;
1113
1114 /* Use a mutex to guard daemon_work against
1115 * bitmap_destroy.
1116 */
1117 mutex_lock(&mddev->bitmap_info.mutex);
1118 bitmap = mddev->bitmap;
1119 if (bitmap == NULL) {
1120 mutex_unlock(&mddev->bitmap_info.mutex);
1121 return;
1122 }
1123 if (time_before(jiffies, bitmap->daemon_lastrun
1124 + mddev->bitmap_info.daemon_sleep))
1125 goto done;
1126
1127 bitmap->daemon_lastrun = jiffies;
1128 if (bitmap->allclean) {
1129 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1130 goto done;
1131 }
1132 bitmap->allclean = 1;
1133
1134 /* Any file-page which is PENDING now needs to be written.
1135 * So set NEEDWRITE now, then after we make any last-minute changes
1136 * we will write it.
1137 */
1138 for (j = 0; j < bitmap->storage.file_pages; j++)
1139 if (test_and_clear_page_attr(bitmap, j,
1140 BITMAP_PAGE_PENDING))
1141 set_page_attr(bitmap, j,
1142 BITMAP_PAGE_NEEDWRITE);
1143
1144 if (bitmap->need_sync &&
1145 mddev->bitmap_info.external == 0) {
1146 /* Arrange for superblock update as well as
1147 * other changes */
1148 bitmap_super_t *sb;
1149 bitmap->need_sync = 0;
1150 if (bitmap->storage.filemap) {
1151 sb = kmap_atomic(bitmap->storage.sb_page);
1152 sb->events_cleared =
1153 cpu_to_le64(bitmap->events_cleared);
1154 kunmap_atomic(sb);
1155 set_page_attr(bitmap, 0,
1156 BITMAP_PAGE_NEEDWRITE);
1157 }
1158 }
1159 /* Now look at the bitmap counters and if any are '2' or '1',
1160 * decrement and handle accordingly.
1161 */
1162 counts = &bitmap->counts;
1163 spin_lock_irq(&counts->lock);
1164 nextpage = 0;
1165 for (j = 0; j < counts->chunks; j++) {
1166 bitmap_counter_t *bmc;
1167 sector_t block = (sector_t)j << counts->chunkshift;
1168
1169 if (j == nextpage) {
1170 nextpage += PAGE_COUNTER_RATIO;
1171 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1172 j |= PAGE_COUNTER_MASK;
1173 continue;
1174 }
1175 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1176 }
1177 bmc = bitmap_get_counter(counts,
1178 block,
1179 &blocks, 0);
1180
1181 if (!bmc) {
1182 j |= PAGE_COUNTER_MASK;
1183 continue;
1184 }
1185 if (*bmc == 1 && !bitmap->need_sync) {
1186 /* We can clear the bit */
1187 *bmc = 0;
1188 bitmap_count_page(counts, block, -1);
1189 bitmap_file_clear_bit(bitmap, block);
1190 } else if (*bmc && *bmc <= 2) {
1191 *bmc = 1;
1192 bitmap_set_pending(counts, block);
1193 bitmap->allclean = 0;
1194 }
1195 }
1196 spin_unlock_irq(&counts->lock);
1197
1198 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1199 * DIRTY pages need to be written by bitmap_unplug so it can wait
1200 * for them.
1201 * If we find any DIRTY page we stop there and let bitmap_unplug
1202 * handle all the rest. This is important in the case where
1203 * the first blocking holds the superblock and it has been updated.
1204 * We mustn't write any other blocks before the superblock.
1205 */
1206 for (j = 0;
1207 j < bitmap->storage.file_pages
1208 && !test_bit(BITMAP_STALE, &bitmap->flags);
1209 j++) {
1210
1211 if (test_page_attr(bitmap, j,
1212 BITMAP_PAGE_DIRTY))
1213 /* bitmap_unplug will handle the rest */
1214 break;
1215 if (test_and_clear_page_attr(bitmap, j,
1216 BITMAP_PAGE_NEEDWRITE)) {
1217 write_page(bitmap, bitmap->storage.filemap[j], 0);
1218 }
1219 }
1220
1221 done:
1222 if (bitmap->allclean == 0)
1223 mddev->thread->timeout =
1224 mddev->bitmap_info.daemon_sleep;
1225 mutex_unlock(&mddev->bitmap_info.mutex);
1226 }
1227
1228 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1229 sector_t offset, sector_t *blocks,
1230 int create)
1231 __releases(bitmap->lock)
1232 __acquires(bitmap->lock)
1233 {
1234 /* If 'create', we might release the lock and reclaim it.
1235 * The lock must have been taken with interrupts enabled.
1236 * If !create, we don't release the lock.
1237 */
1238 sector_t chunk = offset >> bitmap->chunkshift;
1239 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1240 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1241 sector_t csize;
1242 int err;
1243
1244 err = bitmap_checkpage(bitmap, page, create);
1245
1246 if (bitmap->bp[page].hijacked ||
1247 bitmap->bp[page].map == NULL)
1248 csize = ((sector_t)1) << (bitmap->chunkshift +
1249 PAGE_COUNTER_SHIFT - 1);
1250 else
1251 csize = ((sector_t)1) << bitmap->chunkshift;
1252 *blocks = csize - (offset & (csize - 1));
1253
1254 if (err < 0)
1255 return NULL;
1256
1257 /* now locked ... */
1258
1259 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1260 /* should we use the first or second counter field
1261 * of the hijacked pointer? */
1262 int hi = (pageoff > PAGE_COUNTER_MASK);
1263 return &((bitmap_counter_t *)
1264 &bitmap->bp[page].map)[hi];
1265 } else /* page is allocated */
1266 return (bitmap_counter_t *)
1267 &(bitmap->bp[page].map[pageoff]);
1268 }
1269
1270 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1271 {
1272 if (!bitmap)
1273 return 0;
1274
1275 if (behind) {
1276 int bw;
1277 atomic_inc(&bitmap->behind_writes);
1278 bw = atomic_read(&bitmap->behind_writes);
1279 if (bw > bitmap->behind_writes_used)
1280 bitmap->behind_writes_used = bw;
1281
1282 pr_debug("inc write-behind count %d/%lu\n",
1283 bw, bitmap->mddev->bitmap_info.max_write_behind);
1284 }
1285
1286 while (sectors) {
1287 sector_t blocks;
1288 bitmap_counter_t *bmc;
1289
1290 spin_lock_irq(&bitmap->counts.lock);
1291 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1292 if (!bmc) {
1293 spin_unlock_irq(&bitmap->counts.lock);
1294 return 0;
1295 }
1296
1297 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1298 DEFINE_WAIT(__wait);
1299 /* note that it is safe to do the prepare_to_wait
1300 * after the test as long as we do it before dropping
1301 * the spinlock.
1302 */
1303 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1304 TASK_UNINTERRUPTIBLE);
1305 spin_unlock_irq(&bitmap->counts.lock);
1306 schedule();
1307 finish_wait(&bitmap->overflow_wait, &__wait);
1308 continue;
1309 }
1310
1311 switch (*bmc) {
1312 case 0:
1313 bitmap_file_set_bit(bitmap, offset);
1314 bitmap_count_page(&bitmap->counts, offset, 1);
1315 /* fall through */
1316 case 1:
1317 *bmc = 2;
1318 }
1319
1320 (*bmc)++;
1321
1322 spin_unlock_irq(&bitmap->counts.lock);
1323
1324 offset += blocks;
1325 if (sectors > blocks)
1326 sectors -= blocks;
1327 else
1328 sectors = 0;
1329 }
1330 return 0;
1331 }
1332 EXPORT_SYMBOL(bitmap_startwrite);
1333
1334 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1335 int success, int behind)
1336 {
1337 if (!bitmap)
1338 return;
1339 if (behind) {
1340 if (atomic_dec_and_test(&bitmap->behind_writes))
1341 wake_up(&bitmap->behind_wait);
1342 pr_debug("dec write-behind count %d/%lu\n",
1343 atomic_read(&bitmap->behind_writes),
1344 bitmap->mddev->bitmap_info.max_write_behind);
1345 }
1346
1347 while (sectors) {
1348 sector_t blocks;
1349 unsigned long flags;
1350 bitmap_counter_t *bmc;
1351
1352 spin_lock_irqsave(&bitmap->counts.lock, flags);
1353 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1354 if (!bmc) {
1355 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1356 return;
1357 }
1358
1359 if (success && !bitmap->mddev->degraded &&
1360 bitmap->events_cleared < bitmap->mddev->events) {
1361 bitmap->events_cleared = bitmap->mddev->events;
1362 bitmap->need_sync = 1;
1363 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1364 }
1365
1366 if (!success && !NEEDED(*bmc))
1367 *bmc |= NEEDED_MASK;
1368
1369 if (COUNTER(*bmc) == COUNTER_MAX)
1370 wake_up(&bitmap->overflow_wait);
1371
1372 (*bmc)--;
1373 if (*bmc <= 2) {
1374 bitmap_set_pending(&bitmap->counts, offset);
1375 bitmap->allclean = 0;
1376 }
1377 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1378 offset += blocks;
1379 if (sectors > blocks)
1380 sectors -= blocks;
1381 else
1382 sectors = 0;
1383 }
1384 }
1385 EXPORT_SYMBOL(bitmap_endwrite);
1386
1387 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1388 int degraded)
1389 {
1390 bitmap_counter_t *bmc;
1391 int rv;
1392 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1393 *blocks = 1024;
1394 return 1; /* always resync if no bitmap */
1395 }
1396 spin_lock_irq(&bitmap->counts.lock);
1397 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1398 rv = 0;
1399 if (bmc) {
1400 /* locked */
1401 if (RESYNC(*bmc))
1402 rv = 1;
1403 else if (NEEDED(*bmc)) {
1404 rv = 1;
1405 if (!degraded) { /* don't set/clear bits if degraded */
1406 *bmc |= RESYNC_MASK;
1407 *bmc &= ~NEEDED_MASK;
1408 }
1409 }
1410 }
1411 spin_unlock_irq(&bitmap->counts.lock);
1412 return rv;
1413 }
1414
1415 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1416 int degraded)
1417 {
1418 /* bitmap_start_sync must always report on multiples of whole
1419 * pages, otherwise resync (which is very PAGE_SIZE based) will
1420 * get confused.
1421 * So call __bitmap_start_sync repeatedly (if needed) until
1422 * At least PAGE_SIZE>>9 blocks are covered.
1423 * Return the 'or' of the result.
1424 */
1425 int rv = 0;
1426 sector_t blocks1;
1427
1428 *blocks = 0;
1429 while (*blocks < (PAGE_SIZE>>9)) {
1430 rv |= __bitmap_start_sync(bitmap, offset,
1431 &blocks1, degraded);
1432 offset += blocks1;
1433 *blocks += blocks1;
1434 }
1435 return rv;
1436 }
1437 EXPORT_SYMBOL(bitmap_start_sync);
1438
1439 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1440 {
1441 bitmap_counter_t *bmc;
1442 unsigned long flags;
1443
1444 if (bitmap == NULL) {
1445 *blocks = 1024;
1446 return;
1447 }
1448 spin_lock_irqsave(&bitmap->counts.lock, flags);
1449 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1450 if (bmc == NULL)
1451 goto unlock;
1452 /* locked */
1453 if (RESYNC(*bmc)) {
1454 *bmc &= ~RESYNC_MASK;
1455
1456 if (!NEEDED(*bmc) && aborted)
1457 *bmc |= NEEDED_MASK;
1458 else {
1459 if (*bmc <= 2) {
1460 bitmap_set_pending(&bitmap->counts, offset);
1461 bitmap->allclean = 0;
1462 }
1463 }
1464 }
1465 unlock:
1466 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1467 }
1468 EXPORT_SYMBOL(bitmap_end_sync);
1469
1470 void bitmap_close_sync(struct bitmap *bitmap)
1471 {
1472 /* Sync has finished, and any bitmap chunks that weren't synced
1473 * properly have been aborted. It remains to us to clear the
1474 * RESYNC bit wherever it is still on
1475 */
1476 sector_t sector = 0;
1477 sector_t blocks;
1478 if (!bitmap)
1479 return;
1480 while (sector < bitmap->mddev->resync_max_sectors) {
1481 bitmap_end_sync(bitmap, sector, &blocks, 0);
1482 sector += blocks;
1483 }
1484 }
1485 EXPORT_SYMBOL(bitmap_close_sync);
1486
1487 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1488 {
1489 sector_t s = 0;
1490 sector_t blocks;
1491
1492 if (!bitmap)
1493 return;
1494 if (sector == 0) {
1495 bitmap->last_end_sync = jiffies;
1496 return;
1497 }
1498 if (time_before(jiffies, (bitmap->last_end_sync
1499 + bitmap->mddev->bitmap_info.daemon_sleep)))
1500 return;
1501 wait_event(bitmap->mddev->recovery_wait,
1502 atomic_read(&bitmap->mddev->recovery_active) == 0);
1503
1504 bitmap->mddev->curr_resync_completed = sector;
1505 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1506 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1507 s = 0;
1508 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1509 bitmap_end_sync(bitmap, s, &blocks, 0);
1510 s += blocks;
1511 }
1512 bitmap->last_end_sync = jiffies;
1513 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1514 }
1515 EXPORT_SYMBOL(bitmap_cond_end_sync);
1516
1517 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1518 {
1519 /* For each chunk covered by any of these sectors, set the
1520 * counter to 2 and possibly set resync_needed. They should all
1521 * be 0 at this point
1522 */
1523
1524 sector_t secs;
1525 bitmap_counter_t *bmc;
1526 spin_lock_irq(&bitmap->counts.lock);
1527 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1528 if (!bmc) {
1529 spin_unlock_irq(&bitmap->counts.lock);
1530 return;
1531 }
1532 if (!*bmc) {
1533 *bmc = 2 | (needed ? NEEDED_MASK : 0);
1534 bitmap_count_page(&bitmap->counts, offset, 1);
1535 bitmap_set_pending(&bitmap->counts, offset);
1536 bitmap->allclean = 0;
1537 }
1538 spin_unlock_irq(&bitmap->counts.lock);
1539 }
1540
1541 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1542 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1543 {
1544 unsigned long chunk;
1545
1546 for (chunk = s; chunk <= e; chunk++) {
1547 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1548 bitmap_set_memory_bits(bitmap, sec, 1);
1549 bitmap_file_set_bit(bitmap, sec);
1550 if (sec < bitmap->mddev->recovery_cp)
1551 /* We are asserting that the array is dirty,
1552 * so move the recovery_cp address back so
1553 * that it is obvious that it is dirty
1554 */
1555 bitmap->mddev->recovery_cp = sec;
1556 }
1557 }
1558
1559 /*
1560 * flush out any pending updates
1561 */
1562 void bitmap_flush(struct mddev *mddev)
1563 {
1564 struct bitmap *bitmap = mddev->bitmap;
1565 long sleep;
1566
1567 if (!bitmap) /* there was no bitmap */
1568 return;
1569
1570 /* run the daemon_work three time to ensure everything is flushed
1571 * that can be
1572 */
1573 sleep = mddev->bitmap_info.daemon_sleep * 2;
1574 bitmap->daemon_lastrun -= sleep;
1575 bitmap_daemon_work(mddev);
1576 bitmap->daemon_lastrun -= sleep;
1577 bitmap_daemon_work(mddev);
1578 bitmap->daemon_lastrun -= sleep;
1579 bitmap_daemon_work(mddev);
1580 bitmap_update_sb(bitmap);
1581 }
1582
1583 /*
1584 * free memory that was allocated
1585 */
1586 static void bitmap_free(struct bitmap *bitmap)
1587 {
1588 unsigned long k, pages;
1589 struct bitmap_page *bp;
1590
1591 if (!bitmap) /* there was no bitmap */
1592 return;
1593
1594 /* Shouldn't be needed - but just in case.... */
1595 wait_event(bitmap->write_wait,
1596 atomic_read(&bitmap->pending_writes) == 0);
1597
1598 /* release the bitmap file */
1599 bitmap_file_unmap(&bitmap->storage);
1600
1601 bp = bitmap->counts.bp;
1602 pages = bitmap->counts.pages;
1603
1604 /* free all allocated memory */
1605
1606 if (bp) /* deallocate the page memory */
1607 for (k = 0; k < pages; k++)
1608 if (bp[k].map && !bp[k].hijacked)
1609 kfree(bp[k].map);
1610 kfree(bp);
1611 kfree(bitmap);
1612 }
1613
1614 void bitmap_destroy(struct mddev *mddev)
1615 {
1616 struct bitmap *bitmap = mddev->bitmap;
1617
1618 if (!bitmap) /* there was no bitmap */
1619 return;
1620
1621 mutex_lock(&mddev->bitmap_info.mutex);
1622 spin_lock(&mddev->lock);
1623 mddev->bitmap = NULL; /* disconnect from the md device */
1624 spin_unlock(&mddev->lock);
1625 mutex_unlock(&mddev->bitmap_info.mutex);
1626 if (mddev->thread)
1627 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1628
1629 if (bitmap->sysfs_can_clear)
1630 sysfs_put(bitmap->sysfs_can_clear);
1631
1632 bitmap_free(bitmap);
1633 }
1634
1635 /*
1636 * initialize the bitmap structure
1637 * if this returns an error, bitmap_destroy must be called to do clean up
1638 */
1639 int bitmap_create(struct mddev *mddev)
1640 {
1641 struct bitmap *bitmap;
1642 sector_t blocks = mddev->resync_max_sectors;
1643 struct file *file = mddev->bitmap_info.file;
1644 int err;
1645 struct kernfs_node *bm = NULL;
1646
1647 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1648
1649 BUG_ON(file && mddev->bitmap_info.offset);
1650
1651 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1652 if (!bitmap)
1653 return -ENOMEM;
1654
1655 spin_lock_init(&bitmap->counts.lock);
1656 atomic_set(&bitmap->pending_writes, 0);
1657 init_waitqueue_head(&bitmap->write_wait);
1658 init_waitqueue_head(&bitmap->overflow_wait);
1659 init_waitqueue_head(&bitmap->behind_wait);
1660
1661 bitmap->mddev = mddev;
1662
1663 if (mddev->kobj.sd)
1664 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1665 if (bm) {
1666 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1667 sysfs_put(bm);
1668 } else
1669 bitmap->sysfs_can_clear = NULL;
1670
1671 bitmap->storage.file = file;
1672 if (file) {
1673 get_file(file);
1674 /* As future accesses to this file will use bmap,
1675 * and bypass the page cache, we must sync the file
1676 * first.
1677 */
1678 vfs_fsync(file, 1);
1679 }
1680 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1681 if (!mddev->bitmap_info.external) {
1682 /*
1683 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1684 * instructing us to create a new on-disk bitmap instance.
1685 */
1686 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1687 err = bitmap_new_disk_sb(bitmap);
1688 else
1689 err = bitmap_read_sb(bitmap);
1690 } else {
1691 err = 0;
1692 if (mddev->bitmap_info.chunksize == 0 ||
1693 mddev->bitmap_info.daemon_sleep == 0)
1694 /* chunksize and time_base need to be
1695 * set first. */
1696 err = -EINVAL;
1697 }
1698 if (err)
1699 goto error;
1700
1701 bitmap->daemon_lastrun = jiffies;
1702 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1703 if (err)
1704 goto error;
1705
1706 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1707 bitmap->counts.pages, bmname(bitmap));
1708
1709 mddev->bitmap = bitmap;
1710 return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1711
1712 error:
1713 bitmap_free(bitmap);
1714 return err;
1715 }
1716
1717 int bitmap_load(struct mddev *mddev)
1718 {
1719 int err = 0;
1720 sector_t start = 0;
1721 sector_t sector = 0;
1722 struct bitmap *bitmap = mddev->bitmap;
1723
1724 if (!bitmap)
1725 goto out;
1726
1727 /* Clear out old bitmap info first: Either there is none, or we
1728 * are resuming after someone else has possibly changed things,
1729 * so we should forget old cached info.
1730 * All chunks should be clean, but some might need_sync.
1731 */
1732 while (sector < mddev->resync_max_sectors) {
1733 sector_t blocks;
1734 bitmap_start_sync(bitmap, sector, &blocks, 0);
1735 sector += blocks;
1736 }
1737 bitmap_close_sync(bitmap);
1738
1739 if (mddev->degraded == 0
1740 || bitmap->events_cleared == mddev->events)
1741 /* no need to keep dirty bits to optimise a
1742 * re-add of a missing device */
1743 start = mddev->recovery_cp;
1744
1745 mutex_lock(&mddev->bitmap_info.mutex);
1746 err = bitmap_init_from_disk(bitmap, start);
1747 mutex_unlock(&mddev->bitmap_info.mutex);
1748
1749 if (err)
1750 goto out;
1751 clear_bit(BITMAP_STALE, &bitmap->flags);
1752
1753 /* Kick recovery in case any bits were set */
1754 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1755
1756 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1757 md_wakeup_thread(mddev->thread);
1758
1759 bitmap_update_sb(bitmap);
1760
1761 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1762 err = -EIO;
1763 out:
1764 return err;
1765 }
1766 EXPORT_SYMBOL_GPL(bitmap_load);
1767
1768 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1769 {
1770 unsigned long chunk_kb;
1771 struct bitmap_counts *counts;
1772
1773 if (!bitmap)
1774 return;
1775
1776 counts = &bitmap->counts;
1777
1778 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1779 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1780 "%lu%s chunk",
1781 counts->pages - counts->missing_pages,
1782 counts->pages,
1783 (counts->pages - counts->missing_pages)
1784 << (PAGE_SHIFT - 10),
1785 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1786 chunk_kb ? "KB" : "B");
1787 if (bitmap->storage.file) {
1788 seq_printf(seq, ", file: ");
1789 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1790 }
1791
1792 seq_printf(seq, "\n");
1793 }
1794
1795 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1796 int chunksize, int init)
1797 {
1798 /* If chunk_size is 0, choose an appropriate chunk size.
1799 * Then possibly allocate new storage space.
1800 * Then quiesce, copy bits, replace bitmap, and re-start
1801 *
1802 * This function is called both to set up the initial bitmap
1803 * and to resize the bitmap while the array is active.
1804 * If this happens as a result of the array being resized,
1805 * chunksize will be zero, and we need to choose a suitable
1806 * chunksize, otherwise we use what we are given.
1807 */
1808 struct bitmap_storage store;
1809 struct bitmap_counts old_counts;
1810 unsigned long chunks;
1811 sector_t block;
1812 sector_t old_blocks, new_blocks;
1813 int chunkshift;
1814 int ret = 0;
1815 long pages;
1816 struct bitmap_page *new_bp;
1817
1818 if (chunksize == 0) {
1819 /* If there is enough space, leave the chunk size unchanged,
1820 * else increase by factor of two until there is enough space.
1821 */
1822 long bytes;
1823 long space = bitmap->mddev->bitmap_info.space;
1824
1825 if (space == 0) {
1826 /* We don't know how much space there is, so limit
1827 * to current size - in sectors.
1828 */
1829 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1830 if (!bitmap->mddev->bitmap_info.external)
1831 bytes += sizeof(bitmap_super_t);
1832 space = DIV_ROUND_UP(bytes, 512);
1833 bitmap->mddev->bitmap_info.space = space;
1834 }
1835 chunkshift = bitmap->counts.chunkshift;
1836 chunkshift--;
1837 do {
1838 /* 'chunkshift' is shift from block size to chunk size */
1839 chunkshift++;
1840 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1841 bytes = DIV_ROUND_UP(chunks, 8);
1842 if (!bitmap->mddev->bitmap_info.external)
1843 bytes += sizeof(bitmap_super_t);
1844 } while (bytes > (space << 9));
1845 } else
1846 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1847
1848 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1849 memset(&store, 0, sizeof(store));
1850 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1851 ret = bitmap_storage_alloc(&store, chunks,
1852 !bitmap->mddev->bitmap_info.external);
1853 if (ret)
1854 goto err;
1855
1856 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
1857
1858 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
1859 ret = -ENOMEM;
1860 if (!new_bp) {
1861 bitmap_file_unmap(&store);
1862 goto err;
1863 }
1864
1865 if (!init)
1866 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
1867
1868 store.file = bitmap->storage.file;
1869 bitmap->storage.file = NULL;
1870
1871 if (store.sb_page && bitmap->storage.sb_page)
1872 memcpy(page_address(store.sb_page),
1873 page_address(bitmap->storage.sb_page),
1874 sizeof(bitmap_super_t));
1875 bitmap_file_unmap(&bitmap->storage);
1876 bitmap->storage = store;
1877
1878 old_counts = bitmap->counts;
1879 bitmap->counts.bp = new_bp;
1880 bitmap->counts.pages = pages;
1881 bitmap->counts.missing_pages = pages;
1882 bitmap->counts.chunkshift = chunkshift;
1883 bitmap->counts.chunks = chunks;
1884 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
1885 BITMAP_BLOCK_SHIFT);
1886
1887 blocks = min(old_counts.chunks << old_counts.chunkshift,
1888 chunks << chunkshift);
1889
1890 spin_lock_irq(&bitmap->counts.lock);
1891 for (block = 0; block < blocks; ) {
1892 bitmap_counter_t *bmc_old, *bmc_new;
1893 int set;
1894
1895 bmc_old = bitmap_get_counter(&old_counts, block,
1896 &old_blocks, 0);
1897 set = bmc_old && NEEDED(*bmc_old);
1898
1899 if (set) {
1900 bmc_new = bitmap_get_counter(&bitmap->counts, block,
1901 &new_blocks, 1);
1902 if (*bmc_new == 0) {
1903 /* need to set on-disk bits too. */
1904 sector_t end = block + new_blocks;
1905 sector_t start = block >> chunkshift;
1906 start <<= chunkshift;
1907 while (start < end) {
1908 bitmap_file_set_bit(bitmap, block);
1909 start += 1 << chunkshift;
1910 }
1911 *bmc_new = 2;
1912 bitmap_count_page(&bitmap->counts,
1913 block, 1);
1914 bitmap_set_pending(&bitmap->counts,
1915 block);
1916 }
1917 *bmc_new |= NEEDED_MASK;
1918 if (new_blocks < old_blocks)
1919 old_blocks = new_blocks;
1920 }
1921 block += old_blocks;
1922 }
1923
1924 if (!init) {
1925 int i;
1926 while (block < (chunks << chunkshift)) {
1927 bitmap_counter_t *bmc;
1928 bmc = bitmap_get_counter(&bitmap->counts, block,
1929 &new_blocks, 1);
1930 if (bmc) {
1931 /* new space. It needs to be resynced, so
1932 * we set NEEDED_MASK.
1933 */
1934 if (*bmc == 0) {
1935 *bmc = NEEDED_MASK | 2;
1936 bitmap_count_page(&bitmap->counts,
1937 block, 1);
1938 bitmap_set_pending(&bitmap->counts,
1939 block);
1940 }
1941 }
1942 block += new_blocks;
1943 }
1944 for (i = 0; i < bitmap->storage.file_pages; i++)
1945 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1946 }
1947 spin_unlock_irq(&bitmap->counts.lock);
1948
1949 if (!init) {
1950 bitmap_unplug(bitmap);
1951 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
1952 }
1953 ret = 0;
1954 err:
1955 return ret;
1956 }
1957 EXPORT_SYMBOL_GPL(bitmap_resize);
1958
1959 static ssize_t
1960 location_show(struct mddev *mddev, char *page)
1961 {
1962 ssize_t len;
1963 if (mddev->bitmap_info.file)
1964 len = sprintf(page, "file");
1965 else if (mddev->bitmap_info.offset)
1966 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
1967 else
1968 len = sprintf(page, "none");
1969 len += sprintf(page+len, "\n");
1970 return len;
1971 }
1972
1973 static ssize_t
1974 location_store(struct mddev *mddev, const char *buf, size_t len)
1975 {
1976
1977 if (mddev->pers) {
1978 if (!mddev->pers->quiesce)
1979 return -EBUSY;
1980 if (mddev->recovery || mddev->sync_thread)
1981 return -EBUSY;
1982 }
1983
1984 if (mddev->bitmap || mddev->bitmap_info.file ||
1985 mddev->bitmap_info.offset) {
1986 /* bitmap already configured. Only option is to clear it */
1987 if (strncmp(buf, "none", 4) != 0)
1988 return -EBUSY;
1989 if (mddev->pers) {
1990 mddev->pers->quiesce(mddev, 1);
1991 bitmap_destroy(mddev);
1992 mddev->pers->quiesce(mddev, 0);
1993 }
1994 mddev->bitmap_info.offset = 0;
1995 if (mddev->bitmap_info.file) {
1996 struct file *f = mddev->bitmap_info.file;
1997 mddev->bitmap_info.file = NULL;
1998 fput(f);
1999 }
2000 } else {
2001 /* No bitmap, OK to set a location */
2002 long long offset;
2003 if (strncmp(buf, "none", 4) == 0)
2004 /* nothing to be done */;
2005 else if (strncmp(buf, "file:", 5) == 0) {
2006 /* Not supported yet */
2007 return -EINVAL;
2008 } else {
2009 int rv;
2010 if (buf[0] == '+')
2011 rv = kstrtoll(buf+1, 10, &offset);
2012 else
2013 rv = kstrtoll(buf, 10, &offset);
2014 if (rv)
2015 return rv;
2016 if (offset == 0)
2017 return -EINVAL;
2018 if (mddev->bitmap_info.external == 0 &&
2019 mddev->major_version == 0 &&
2020 offset != mddev->bitmap_info.default_offset)
2021 return -EINVAL;
2022 mddev->bitmap_info.offset = offset;
2023 if (mddev->pers) {
2024 mddev->pers->quiesce(mddev, 1);
2025 rv = bitmap_create(mddev);
2026 if (!rv)
2027 rv = bitmap_load(mddev);
2028 if (rv) {
2029 bitmap_destroy(mddev);
2030 mddev->bitmap_info.offset = 0;
2031 }
2032 mddev->pers->quiesce(mddev, 0);
2033 if (rv)
2034 return rv;
2035 }
2036 }
2037 }
2038 if (!mddev->external) {
2039 /* Ensure new bitmap info is stored in
2040 * metadata promptly.
2041 */
2042 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2043 md_wakeup_thread(mddev->thread);
2044 }
2045 return len;
2046 }
2047
2048 static struct md_sysfs_entry bitmap_location =
2049 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2050
2051 /* 'bitmap/space' is the space available at 'location' for the
2052 * bitmap. This allows the kernel to know when it is safe to
2053 * resize the bitmap to match a resized array.
2054 */
2055 static ssize_t
2056 space_show(struct mddev *mddev, char *page)
2057 {
2058 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2059 }
2060
2061 static ssize_t
2062 space_store(struct mddev *mddev, const char *buf, size_t len)
2063 {
2064 unsigned long sectors;
2065 int rv;
2066
2067 rv = kstrtoul(buf, 10, &sectors);
2068 if (rv)
2069 return rv;
2070
2071 if (sectors == 0)
2072 return -EINVAL;
2073
2074 if (mddev->bitmap &&
2075 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2076 return -EFBIG; /* Bitmap is too big for this small space */
2077
2078 /* could make sure it isn't too big, but that isn't really
2079 * needed - user-space should be careful.
2080 */
2081 mddev->bitmap_info.space = sectors;
2082 return len;
2083 }
2084
2085 static struct md_sysfs_entry bitmap_space =
2086 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2087
2088 static ssize_t
2089 timeout_show(struct mddev *mddev, char *page)
2090 {
2091 ssize_t len;
2092 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2093 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2094
2095 len = sprintf(page, "%lu", secs);
2096 if (jifs)
2097 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2098 len += sprintf(page+len, "\n");
2099 return len;
2100 }
2101
2102 static ssize_t
2103 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2104 {
2105 /* timeout can be set at any time */
2106 unsigned long timeout;
2107 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2108 if (rv)
2109 return rv;
2110
2111 /* just to make sure we don't overflow... */
2112 if (timeout >= LONG_MAX / HZ)
2113 return -EINVAL;
2114
2115 timeout = timeout * HZ / 10000;
2116
2117 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2118 timeout = MAX_SCHEDULE_TIMEOUT-1;
2119 if (timeout < 1)
2120 timeout = 1;
2121 mddev->bitmap_info.daemon_sleep = timeout;
2122 if (mddev->thread) {
2123 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2124 * the bitmap is all clean and we don't need to
2125 * adjust the timeout right now
2126 */
2127 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2128 mddev->thread->timeout = timeout;
2129 md_wakeup_thread(mddev->thread);
2130 }
2131 }
2132 return len;
2133 }
2134
2135 static struct md_sysfs_entry bitmap_timeout =
2136 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2137
2138 static ssize_t
2139 backlog_show(struct mddev *mddev, char *page)
2140 {
2141 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2142 }
2143
2144 static ssize_t
2145 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2146 {
2147 unsigned long backlog;
2148 int rv = kstrtoul(buf, 10, &backlog);
2149 if (rv)
2150 return rv;
2151 if (backlog > COUNTER_MAX)
2152 return -EINVAL;
2153 mddev->bitmap_info.max_write_behind = backlog;
2154 return len;
2155 }
2156
2157 static struct md_sysfs_entry bitmap_backlog =
2158 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2159
2160 static ssize_t
2161 chunksize_show(struct mddev *mddev, char *page)
2162 {
2163 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2164 }
2165
2166 static ssize_t
2167 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2168 {
2169 /* Can only be changed when no bitmap is active */
2170 int rv;
2171 unsigned long csize;
2172 if (mddev->bitmap)
2173 return -EBUSY;
2174 rv = kstrtoul(buf, 10, &csize);
2175 if (rv)
2176 return rv;
2177 if (csize < 512 ||
2178 !is_power_of_2(csize))
2179 return -EINVAL;
2180 mddev->bitmap_info.chunksize = csize;
2181 return len;
2182 }
2183
2184 static struct md_sysfs_entry bitmap_chunksize =
2185 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2186
2187 static ssize_t metadata_show(struct mddev *mddev, char *page)
2188 {
2189 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2190 ? "external" : "internal"));
2191 }
2192
2193 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2194 {
2195 if (mddev->bitmap ||
2196 mddev->bitmap_info.file ||
2197 mddev->bitmap_info.offset)
2198 return -EBUSY;
2199 if (strncmp(buf, "external", 8) == 0)
2200 mddev->bitmap_info.external = 1;
2201 else if (strncmp(buf, "internal", 8) == 0)
2202 mddev->bitmap_info.external = 0;
2203 else
2204 return -EINVAL;
2205 return len;
2206 }
2207
2208 static struct md_sysfs_entry bitmap_metadata =
2209 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2210
2211 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2212 {
2213 int len;
2214 spin_lock(&mddev->lock);
2215 if (mddev->bitmap)
2216 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2217 "false" : "true"));
2218 else
2219 len = sprintf(page, "\n");
2220 spin_unlock(&mddev->lock);
2221 return len;
2222 }
2223
2224 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2225 {
2226 if (mddev->bitmap == NULL)
2227 return -ENOENT;
2228 if (strncmp(buf, "false", 5) == 0)
2229 mddev->bitmap->need_sync = 1;
2230 else if (strncmp(buf, "true", 4) == 0) {
2231 if (mddev->degraded)
2232 return -EBUSY;
2233 mddev->bitmap->need_sync = 0;
2234 } else
2235 return -EINVAL;
2236 return len;
2237 }
2238
2239 static struct md_sysfs_entry bitmap_can_clear =
2240 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2241
2242 static ssize_t
2243 behind_writes_used_show(struct mddev *mddev, char *page)
2244 {
2245 ssize_t ret;
2246 spin_lock(&mddev->lock);
2247 if (mddev->bitmap == NULL)
2248 ret = sprintf(page, "0\n");
2249 else
2250 ret = sprintf(page, "%lu\n",
2251 mddev->bitmap->behind_writes_used);
2252 spin_unlock(&mddev->lock);
2253 return ret;
2254 }
2255
2256 static ssize_t
2257 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2258 {
2259 if (mddev->bitmap)
2260 mddev->bitmap->behind_writes_used = 0;
2261 return len;
2262 }
2263
2264 static struct md_sysfs_entry max_backlog_used =
2265 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2266 behind_writes_used_show, behind_writes_used_reset);
2267
2268 static struct attribute *md_bitmap_attrs[] = {
2269 &bitmap_location.attr,
2270 &bitmap_space.attr,
2271 &bitmap_timeout.attr,
2272 &bitmap_backlog.attr,
2273 &bitmap_chunksize.attr,
2274 &bitmap_metadata.attr,
2275 &bitmap_can_clear.attr,
2276 &max_backlog_used.attr,
2277 NULL
2278 };
2279 struct attribute_group md_bitmap_group = {
2280 .name = "bitmap",
2281 .attrs = md_bitmap_attrs,
2282 };
2283
This page took 0.077699 seconds and 6 git commands to generate.