mm, debug: replace dump_flags() with the new printk formats
[linux/fpc-iii.git] / drivers / md / bitmap.c
blobd80cce499a56e595d1f2679170124ca5fc620315
1 /*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
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
13 * Still to do:
15 * flush after percent set rather than just time based. (maybe both).
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"
33 static inline char *bmname(struct bitmap *bitmap)
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
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
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
48 static int bitmap_checkpage(struct bitmap_counts *bitmap,
49 unsigned long page, int create)
50 __releases(bitmap->lock)
51 __acquires(bitmap->lock)
53 unsigned char *mappage;
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.
60 return -EINVAL;
63 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
64 return 0;
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
67 return 0;
69 if (!create)
70 return -ENOENT;
72 /* this page has not been allocated yet */
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.
87 sched_annotate_sleep();
88 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
89 spin_lock_irq(&bitmap->lock);
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 {
104 /* no page was in place and we have one, so install it */
106 bitmap->bp[page].map = mappage;
107 bitmap->missing_pages--;
109 return 0;
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)
117 char *ptr;
119 if (bitmap->bp[page].count) /* page is still busy */
120 return;
122 /* page is no longer in use, it can be released */
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);
137 * bitmap file handling - read and write the bitmap file and its superblock
141 * basic page I/O operations
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)
149 /* choose a good rdev and read the page from there */
151 struct md_rdev *rdev;
152 sector_t target;
154 rdev_for_each(rdev, mddev) {
155 if (! test_bit(In_sync, &rdev->flags)
156 || test_bit(Faulty, &rdev->flags))
157 continue;
159 target = offset + index * (PAGE_SIZE/512);
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;
168 return -EIO;
171 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
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.
181 * Note that if entered with 'rdev == NULL' to start at the
182 * beginning, we temporarily assign 'rdev' to an address which
183 * isn't really an rdev, but which can be used by
184 * list_for_each_entry_continue_rcu() to find the first entry.
186 rcu_read_lock();
187 if (rdev == NULL)
188 /* start at the beginning */
189 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
190 else {
191 /* release the previous rdev and start from there. */
192 rdev_dec_pending(rdev, mddev);
194 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
195 if (rdev->raid_disk >= 0 &&
196 !test_bit(Faulty, &rdev->flags)) {
197 /* this is a usable devices */
198 atomic_inc(&rdev->nr_pending);
199 rcu_read_unlock();
200 return rdev;
203 rcu_read_unlock();
204 return NULL;
207 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
209 struct md_rdev *rdev = NULL;
210 struct block_device *bdev;
211 struct mddev *mddev = bitmap->mddev;
212 struct bitmap_storage *store = &bitmap->storage;
214 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
215 int size = PAGE_SIZE;
216 loff_t offset = mddev->bitmap_info.offset;
218 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
220 if (page->index == store->file_pages-1) {
221 int last_page_size = store->bytes & (PAGE_SIZE-1);
222 if (last_page_size == 0)
223 last_page_size = PAGE_SIZE;
224 size = roundup(last_page_size,
225 bdev_logical_block_size(bdev));
227 /* Just make sure we aren't corrupting data or
228 * metadata
230 if (mddev->external) {
231 /* Bitmap could be anywhere. */
232 if (rdev->sb_start + offset + (page->index
233 * (PAGE_SIZE/512))
234 > rdev->data_offset
236 rdev->sb_start + offset
237 < (rdev->data_offset + mddev->dev_sectors
238 + (PAGE_SIZE/512)))
239 goto bad_alignment;
240 } else if (offset < 0) {
241 /* DATA BITMAP METADATA */
242 if (offset
243 + (long)(page->index * (PAGE_SIZE/512))
244 + size/512 > 0)
245 /* bitmap runs in to metadata */
246 goto bad_alignment;
247 if (rdev->data_offset + mddev->dev_sectors
248 > rdev->sb_start + offset)
249 /* data runs in to bitmap */
250 goto bad_alignment;
251 } else if (rdev->sb_start < rdev->data_offset) {
252 /* METADATA BITMAP DATA */
253 if (rdev->sb_start
254 + offset
255 + page->index*(PAGE_SIZE/512) + size/512
256 > rdev->data_offset)
257 /* bitmap runs in to data */
258 goto bad_alignment;
259 } else {
260 /* DATA METADATA BITMAP - no problems */
262 md_super_write(mddev, rdev,
263 rdev->sb_start + offset
264 + page->index * (PAGE_SIZE/512),
265 size,
266 page);
269 if (wait)
270 md_super_wait(mddev);
271 return 0;
273 bad_alignment:
274 return -EINVAL;
277 static void bitmap_file_kick(struct bitmap *bitmap);
279 * write out a page to a file
281 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
283 struct buffer_head *bh;
285 if (bitmap->storage.file == NULL) {
286 switch (write_sb_page(bitmap, page, wait)) {
287 case -EINVAL:
288 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
290 } else {
292 bh = page_buffers(page);
294 while (bh && bh->b_blocknr) {
295 atomic_inc(&bitmap->pending_writes);
296 set_buffer_locked(bh);
297 set_buffer_mapped(bh);
298 submit_bh(WRITE | REQ_SYNC, bh);
299 bh = bh->b_this_page;
302 if (wait)
303 wait_event(bitmap->write_wait,
304 atomic_read(&bitmap->pending_writes)==0);
306 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
307 bitmap_file_kick(bitmap);
310 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
312 struct bitmap *bitmap = bh->b_private;
314 if (!uptodate)
315 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
316 if (atomic_dec_and_test(&bitmap->pending_writes))
317 wake_up(&bitmap->write_wait);
320 /* copied from buffer.c */
321 static void
322 __clear_page_buffers(struct page *page)
324 ClearPagePrivate(page);
325 set_page_private(page, 0);
326 page_cache_release(page);
328 static void free_buffers(struct page *page)
330 struct buffer_head *bh;
332 if (!PagePrivate(page))
333 return;
335 bh = page_buffers(page);
336 while (bh) {
337 struct buffer_head *next = bh->b_this_page;
338 free_buffer_head(bh);
339 bh = next;
341 __clear_page_buffers(page);
342 put_page(page);
345 /* read a page from a file.
346 * We both read the page, and attach buffers to the page to record the
347 * address of each block (using bmap). These addresses will be used
348 * to write the block later, completely bypassing the filesystem.
349 * This usage is similar to how swap files are handled, and allows us
350 * to write to a file with no concerns of memory allocation failing.
352 static int read_page(struct file *file, unsigned long index,
353 struct bitmap *bitmap,
354 unsigned long count,
355 struct page *page)
357 int ret = 0;
358 struct inode *inode = file_inode(file);
359 struct buffer_head *bh;
360 sector_t block;
362 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
363 (unsigned long long)index << PAGE_SHIFT);
365 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
366 if (!bh) {
367 ret = -ENOMEM;
368 goto out;
370 attach_page_buffers(page, bh);
371 block = index << (PAGE_SHIFT - inode->i_blkbits);
372 while (bh) {
373 if (count == 0)
374 bh->b_blocknr = 0;
375 else {
376 bh->b_blocknr = bmap(inode, block);
377 if (bh->b_blocknr == 0) {
378 /* Cannot use this file! */
379 ret = -EINVAL;
380 goto out;
382 bh->b_bdev = inode->i_sb->s_bdev;
383 if (count < (1<<inode->i_blkbits))
384 count = 0;
385 else
386 count -= (1<<inode->i_blkbits);
388 bh->b_end_io = end_bitmap_write;
389 bh->b_private = bitmap;
390 atomic_inc(&bitmap->pending_writes);
391 set_buffer_locked(bh);
392 set_buffer_mapped(bh);
393 submit_bh(READ, bh);
395 block++;
396 bh = bh->b_this_page;
398 page->index = index;
400 wait_event(bitmap->write_wait,
401 atomic_read(&bitmap->pending_writes)==0);
402 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
403 ret = -EIO;
404 out:
405 if (ret)
406 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
407 (int)PAGE_SIZE,
408 (unsigned long long)index << PAGE_SHIFT,
409 ret);
410 return ret;
414 * bitmap file superblock operations
417 /* update the event counter and sync the superblock to disk */
418 void bitmap_update_sb(struct bitmap *bitmap)
420 bitmap_super_t *sb;
422 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
423 return;
424 if (bitmap->mddev->bitmap_info.external)
425 return;
426 if (!bitmap->storage.sb_page) /* no superblock */
427 return;
428 sb = kmap_atomic(bitmap->storage.sb_page);
429 sb->events = cpu_to_le64(bitmap->mddev->events);
430 if (bitmap->mddev->events < bitmap->events_cleared)
431 /* rocking back to read-only */
432 bitmap->events_cleared = bitmap->mddev->events;
433 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
434 sb->state = cpu_to_le32(bitmap->flags);
435 /* Just in case these have been changed via sysfs: */
436 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
437 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
438 /* This might have been changed by a reshape */
439 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
440 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
441 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
442 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
443 bitmap_info.space);
444 kunmap_atomic(sb);
445 write_page(bitmap, bitmap->storage.sb_page, 1);
448 /* print out the bitmap file superblock */
449 void bitmap_print_sb(struct bitmap *bitmap)
451 bitmap_super_t *sb;
453 if (!bitmap || !bitmap->storage.sb_page)
454 return;
455 sb = kmap_atomic(bitmap->storage.sb_page);
456 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
457 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
458 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
459 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
460 *(__u32 *)(sb->uuid+0),
461 *(__u32 *)(sb->uuid+4),
462 *(__u32 *)(sb->uuid+8),
463 *(__u32 *)(sb->uuid+12));
464 printk(KERN_DEBUG " events: %llu\n",
465 (unsigned long long) le64_to_cpu(sb->events));
466 printk(KERN_DEBUG "events cleared: %llu\n",
467 (unsigned long long) le64_to_cpu(sb->events_cleared));
468 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
469 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
470 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
471 printk(KERN_DEBUG " sync size: %llu KB\n",
472 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
473 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
474 kunmap_atomic(sb);
478 * bitmap_new_disk_sb
479 * @bitmap
481 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
482 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
483 * This function verifies 'bitmap_info' and populates the on-disk bitmap
484 * structure, which is to be written to disk.
486 * Returns: 0 on success, -Exxx on error
488 static int bitmap_new_disk_sb(struct bitmap *bitmap)
490 bitmap_super_t *sb;
491 unsigned long chunksize, daemon_sleep, write_behind;
493 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
494 if (bitmap->storage.sb_page == NULL)
495 return -ENOMEM;
496 bitmap->storage.sb_page->index = 0;
498 sb = kmap_atomic(bitmap->storage.sb_page);
500 sb->magic = cpu_to_le32(BITMAP_MAGIC);
501 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
503 chunksize = bitmap->mddev->bitmap_info.chunksize;
504 BUG_ON(!chunksize);
505 if (!is_power_of_2(chunksize)) {
506 kunmap_atomic(sb);
507 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
508 return -EINVAL;
510 sb->chunksize = cpu_to_le32(chunksize);
512 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
513 if (!daemon_sleep ||
514 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
515 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
516 daemon_sleep = 5 * HZ;
518 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
519 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
522 * FIXME: write_behind for RAID1. If not specified, what
523 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
525 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
526 if (write_behind > COUNTER_MAX)
527 write_behind = COUNTER_MAX / 2;
528 sb->write_behind = cpu_to_le32(write_behind);
529 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
531 /* keep the array size field of the bitmap superblock up to date */
532 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
534 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
536 set_bit(BITMAP_STALE, &bitmap->flags);
537 sb->state = cpu_to_le32(bitmap->flags);
538 bitmap->events_cleared = bitmap->mddev->events;
539 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
540 bitmap->mddev->bitmap_info.nodes = 0;
542 kunmap_atomic(sb);
544 return 0;
547 /* read the superblock from the bitmap file and initialize some bitmap fields */
548 static int bitmap_read_sb(struct bitmap *bitmap)
550 char *reason = NULL;
551 bitmap_super_t *sb;
552 unsigned long chunksize, daemon_sleep, write_behind;
553 unsigned long long events;
554 int nodes = 0;
555 unsigned long sectors_reserved = 0;
556 int err = -EINVAL;
557 struct page *sb_page;
558 loff_t offset = bitmap->mddev->bitmap_info.offset;
560 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
561 chunksize = 128 * 1024 * 1024;
562 daemon_sleep = 5 * HZ;
563 write_behind = 0;
564 set_bit(BITMAP_STALE, &bitmap->flags);
565 err = 0;
566 goto out_no_sb;
568 /* page 0 is the superblock, read it... */
569 sb_page = alloc_page(GFP_KERNEL);
570 if (!sb_page)
571 return -ENOMEM;
572 bitmap->storage.sb_page = sb_page;
574 re_read:
575 /* If cluster_slot is set, the cluster is setup */
576 if (bitmap->cluster_slot >= 0) {
577 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
579 sector_div(bm_blocks,
580 bitmap->mddev->bitmap_info.chunksize >> 9);
581 /* bits to bytes */
582 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
583 /* to 4k blocks */
584 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
585 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
586 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
587 bitmap->cluster_slot, offset);
590 if (bitmap->storage.file) {
591 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
592 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
594 err = read_page(bitmap->storage.file, 0,
595 bitmap, bytes, sb_page);
596 } else {
597 err = read_sb_page(bitmap->mddev,
598 offset,
599 sb_page,
600 0, sizeof(bitmap_super_t));
602 if (err)
603 return err;
605 err = -EINVAL;
606 sb = kmap_atomic(sb_page);
608 chunksize = le32_to_cpu(sb->chunksize);
609 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
610 write_behind = le32_to_cpu(sb->write_behind);
611 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
612 /* Setup nodes/clustername only if bitmap version is
613 * cluster-compatible
615 if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
616 nodes = le32_to_cpu(sb->nodes);
617 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
618 sb->cluster_name, 64);
621 /* verify that the bitmap-specific fields are valid */
622 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
623 reason = "bad magic";
624 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
625 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
626 reason = "unrecognized superblock version";
627 else if (chunksize < 512)
628 reason = "bitmap chunksize too small";
629 else if (!is_power_of_2(chunksize))
630 reason = "bitmap chunksize not a power of 2";
631 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
632 reason = "daemon sleep period out of range";
633 else if (write_behind > COUNTER_MAX)
634 reason = "write-behind limit out of range (0 - 16383)";
635 if (reason) {
636 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
637 bmname(bitmap), reason);
638 goto out;
641 /* keep the array size field of the bitmap superblock up to date */
642 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
644 if (bitmap->mddev->persistent) {
646 * We have a persistent array superblock, so compare the
647 * bitmap's UUID and event counter to the mddev's
649 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
650 printk(KERN_INFO
651 "%s: bitmap superblock UUID mismatch\n",
652 bmname(bitmap));
653 goto out;
655 events = le64_to_cpu(sb->events);
656 if (!nodes && (events < bitmap->mddev->events)) {
657 printk(KERN_INFO
658 "%s: bitmap file is out of date (%llu < %llu) "
659 "-- forcing full recovery\n",
660 bmname(bitmap), events,
661 (unsigned long long) bitmap->mddev->events);
662 set_bit(BITMAP_STALE, &bitmap->flags);
666 /* assign fields using values from superblock */
667 bitmap->flags |= le32_to_cpu(sb->state);
668 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
669 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
670 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
671 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
672 err = 0;
674 out:
675 kunmap_atomic(sb);
676 /* Assiging chunksize is required for "re_read" */
677 bitmap->mddev->bitmap_info.chunksize = chunksize;
678 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
679 err = md_setup_cluster(bitmap->mddev, nodes);
680 if (err) {
681 pr_err("%s: Could not setup cluster service (%d)\n",
682 bmname(bitmap), err);
683 goto out_no_sb;
685 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
686 goto re_read;
690 out_no_sb:
691 if (test_bit(BITMAP_STALE, &bitmap->flags))
692 bitmap->events_cleared = bitmap->mddev->events;
693 bitmap->mddev->bitmap_info.chunksize = chunksize;
694 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
695 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
696 bitmap->mddev->bitmap_info.nodes = nodes;
697 if (bitmap->mddev->bitmap_info.space == 0 ||
698 bitmap->mddev->bitmap_info.space > sectors_reserved)
699 bitmap->mddev->bitmap_info.space = sectors_reserved;
700 if (err) {
701 bitmap_print_sb(bitmap);
702 if (bitmap->cluster_slot < 0)
703 md_cluster_stop(bitmap->mddev);
705 return err;
709 * general bitmap file operations
713 * on-disk bitmap:
715 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
716 * file a page at a time. There's a superblock at the start of the file.
718 /* calculate the index of the page that contains this bit */
719 static inline unsigned long file_page_index(struct bitmap_storage *store,
720 unsigned long chunk)
722 if (store->sb_page)
723 chunk += sizeof(bitmap_super_t) << 3;
724 return chunk >> PAGE_BIT_SHIFT;
727 /* calculate the (bit) offset of this bit within a page */
728 static inline unsigned long file_page_offset(struct bitmap_storage *store,
729 unsigned long chunk)
731 if (store->sb_page)
732 chunk += sizeof(bitmap_super_t) << 3;
733 return chunk & (PAGE_BITS - 1);
737 * return a pointer to the page in the filemap that contains the given bit
740 static inline struct page *filemap_get_page(struct bitmap_storage *store,
741 unsigned long chunk)
743 if (file_page_index(store, chunk) >= store->file_pages)
744 return NULL;
745 return store->filemap[file_page_index(store, chunk)];
748 static int bitmap_storage_alloc(struct bitmap_storage *store,
749 unsigned long chunks, int with_super,
750 int slot_number)
752 int pnum, offset = 0;
753 unsigned long num_pages;
754 unsigned long bytes;
756 bytes = DIV_ROUND_UP(chunks, 8);
757 if (with_super)
758 bytes += sizeof(bitmap_super_t);
760 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
761 offset = slot_number * (num_pages - 1);
763 store->filemap = kmalloc(sizeof(struct page *)
764 * num_pages, GFP_KERNEL);
765 if (!store->filemap)
766 return -ENOMEM;
768 if (with_super && !store->sb_page) {
769 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
770 if (store->sb_page == NULL)
771 return -ENOMEM;
774 pnum = 0;
775 if (store->sb_page) {
776 store->filemap[0] = store->sb_page;
777 pnum = 1;
778 store->sb_page->index = offset;
781 for ( ; pnum < num_pages; pnum++) {
782 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
783 if (!store->filemap[pnum]) {
784 store->file_pages = pnum;
785 return -ENOMEM;
787 store->filemap[pnum]->index = pnum + offset;
789 store->file_pages = pnum;
791 /* We need 4 bits per page, rounded up to a multiple
792 * of sizeof(unsigned long) */
793 store->filemap_attr = kzalloc(
794 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
795 GFP_KERNEL);
796 if (!store->filemap_attr)
797 return -ENOMEM;
799 store->bytes = bytes;
801 return 0;
804 static void bitmap_file_unmap(struct bitmap_storage *store)
806 struct page **map, *sb_page;
807 int pages;
808 struct file *file;
810 file = store->file;
811 map = store->filemap;
812 pages = store->file_pages;
813 sb_page = store->sb_page;
815 while (pages--)
816 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
817 free_buffers(map[pages]);
818 kfree(map);
819 kfree(store->filemap_attr);
821 if (sb_page)
822 free_buffers(sb_page);
824 if (file) {
825 struct inode *inode = file_inode(file);
826 invalidate_mapping_pages(inode->i_mapping, 0, -1);
827 fput(file);
832 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
833 * then it is no longer reliable, so we stop using it and we mark the file
834 * as failed in the superblock
836 static void bitmap_file_kick(struct bitmap *bitmap)
838 char *path, *ptr = NULL;
840 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
841 bitmap_update_sb(bitmap);
843 if (bitmap->storage.file) {
844 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
845 if (path)
846 ptr = file_path(bitmap->storage.file,
847 path, PAGE_SIZE);
849 printk(KERN_ALERT
850 "%s: kicking failed bitmap file %s from array!\n",
851 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
853 kfree(path);
854 } else
855 printk(KERN_ALERT
856 "%s: disabling internal bitmap due to errors\n",
857 bmname(bitmap));
861 enum bitmap_page_attr {
862 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
863 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
864 * i.e. counter is 1 or 2. */
865 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
868 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
869 enum bitmap_page_attr attr)
871 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
874 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
875 enum bitmap_page_attr attr)
877 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
880 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
881 enum bitmap_page_attr attr)
883 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
886 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
887 enum bitmap_page_attr attr)
889 return test_and_clear_bit((pnum<<2) + attr,
890 bitmap->storage.filemap_attr);
893 * bitmap_file_set_bit -- called before performing a write to the md device
894 * to set (and eventually sync) a particular bit in the bitmap file
896 * we set the bit immediately, then we record the page number so that
897 * when an unplug occurs, we can flush the dirty pages out to disk
899 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
901 unsigned long bit;
902 struct page *page;
903 void *kaddr;
904 unsigned long chunk = block >> bitmap->counts.chunkshift;
906 page = filemap_get_page(&bitmap->storage, chunk);
907 if (!page)
908 return;
909 bit = file_page_offset(&bitmap->storage, chunk);
911 /* set the bit */
912 kaddr = kmap_atomic(page);
913 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
914 set_bit(bit, kaddr);
915 else
916 set_bit_le(bit, kaddr);
917 kunmap_atomic(kaddr);
918 pr_debug("set file bit %lu page %lu\n", bit, page->index);
919 /* record page number so it gets flushed to disk when unplug occurs */
920 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
923 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
925 unsigned long bit;
926 struct page *page;
927 void *paddr;
928 unsigned long chunk = block >> bitmap->counts.chunkshift;
930 page = filemap_get_page(&bitmap->storage, chunk);
931 if (!page)
932 return;
933 bit = file_page_offset(&bitmap->storage, chunk);
934 paddr = kmap_atomic(page);
935 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
936 clear_bit(bit, paddr);
937 else
938 clear_bit_le(bit, paddr);
939 kunmap_atomic(paddr);
940 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
941 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
942 bitmap->allclean = 0;
946 static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
948 unsigned long bit;
949 struct page *page;
950 void *paddr;
951 unsigned long chunk = block >> bitmap->counts.chunkshift;
952 int set = 0;
954 page = filemap_get_page(&bitmap->storage, chunk);
955 if (!page)
956 return -EINVAL;
957 bit = file_page_offset(&bitmap->storage, chunk);
958 paddr = kmap_atomic(page);
959 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
960 set = test_bit(bit, paddr);
961 else
962 set = test_bit_le(bit, paddr);
963 kunmap_atomic(paddr);
964 return set;
968 /* this gets called when the md device is ready to unplug its underlying
969 * (slave) device queues -- before we let any writes go down, we need to
970 * sync the dirty pages of the bitmap file to disk */
971 void bitmap_unplug(struct bitmap *bitmap)
973 unsigned long i;
974 int dirty, need_write;
976 if (!bitmap || !bitmap->storage.filemap ||
977 test_bit(BITMAP_STALE, &bitmap->flags))
978 return;
980 /* look at each page to see if there are any set bits that need to be
981 * flushed out to disk */
982 for (i = 0; i < bitmap->storage.file_pages; i++) {
983 if (!bitmap->storage.filemap)
984 return;
985 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
986 need_write = test_and_clear_page_attr(bitmap, i,
987 BITMAP_PAGE_NEEDWRITE);
988 if (dirty || need_write) {
989 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
990 write_page(bitmap, bitmap->storage.filemap[i], 0);
993 if (bitmap->storage.file)
994 wait_event(bitmap->write_wait,
995 atomic_read(&bitmap->pending_writes)==0);
996 else
997 md_super_wait(bitmap->mddev);
999 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1000 bitmap_file_kick(bitmap);
1002 EXPORT_SYMBOL(bitmap_unplug);
1004 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1005 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1006 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1007 * memory mapping of the bitmap file
1008 * Special cases:
1009 * if there's no bitmap file, or if the bitmap file had been
1010 * previously kicked from the array, we mark all the bits as
1011 * 1's in order to cause a full resync.
1013 * We ignore all bits for sectors that end earlier than 'start'.
1014 * This is used when reading an out-of-date bitmap...
1016 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1018 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1019 struct page *page = NULL;
1020 unsigned long bit_cnt = 0;
1021 struct file *file;
1022 unsigned long offset;
1023 int outofdate;
1024 int ret = -ENOSPC;
1025 void *paddr;
1026 struct bitmap_storage *store = &bitmap->storage;
1028 chunks = bitmap->counts.chunks;
1029 file = store->file;
1031 if (!file && !bitmap->mddev->bitmap_info.offset) {
1032 /* No permanent bitmap - fill with '1s'. */
1033 store->filemap = NULL;
1034 store->file_pages = 0;
1035 for (i = 0; i < chunks ; i++) {
1036 /* if the disk bit is set, set the memory bit */
1037 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1038 >= start);
1039 bitmap_set_memory_bits(bitmap,
1040 (sector_t)i << bitmap->counts.chunkshift,
1041 needed);
1043 return 0;
1046 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1047 if (outofdate)
1048 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1049 "recovery\n", bmname(bitmap));
1051 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1052 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
1053 bmname(bitmap),
1054 (unsigned long) i_size_read(file->f_mapping->host),
1055 store->bytes);
1056 goto err;
1059 oldindex = ~0L;
1060 offset = 0;
1061 if (!bitmap->mddev->bitmap_info.external)
1062 offset = sizeof(bitmap_super_t);
1064 if (mddev_is_clustered(bitmap->mddev))
1065 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1067 for (i = 0; i < chunks; i++) {
1068 int b;
1069 index = file_page_index(&bitmap->storage, i);
1070 bit = file_page_offset(&bitmap->storage, i);
1071 if (index != oldindex) { /* this is a new page, read it in */
1072 int count;
1073 /* unmap the old page, we're done with it */
1074 if (index == store->file_pages-1)
1075 count = store->bytes - index * PAGE_SIZE;
1076 else
1077 count = PAGE_SIZE;
1078 page = store->filemap[index];
1079 if (file)
1080 ret = read_page(file, index, bitmap,
1081 count, page);
1082 else
1083 ret = read_sb_page(
1084 bitmap->mddev,
1085 bitmap->mddev->bitmap_info.offset,
1086 page,
1087 index + node_offset, count);
1089 if (ret)
1090 goto err;
1092 oldindex = index;
1094 if (outofdate) {
1096 * if bitmap is out of date, dirty the
1097 * whole page and write it out
1099 paddr = kmap_atomic(page);
1100 memset(paddr + offset, 0xff,
1101 PAGE_SIZE - offset);
1102 kunmap_atomic(paddr);
1103 write_page(bitmap, page, 1);
1105 ret = -EIO;
1106 if (test_bit(BITMAP_WRITE_ERROR,
1107 &bitmap->flags))
1108 goto err;
1111 paddr = kmap_atomic(page);
1112 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1113 b = test_bit(bit, paddr);
1114 else
1115 b = test_bit_le(bit, paddr);
1116 kunmap_atomic(paddr);
1117 if (b) {
1118 /* if the disk bit is set, set the memory bit */
1119 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1120 >= start);
1121 bitmap_set_memory_bits(bitmap,
1122 (sector_t)i << bitmap->counts.chunkshift,
1123 needed);
1124 bit_cnt++;
1126 offset = 0;
1129 printk(KERN_INFO "%s: bitmap initialized from disk: "
1130 "read %lu pages, set %lu of %lu bits\n",
1131 bmname(bitmap), store->file_pages,
1132 bit_cnt, chunks);
1134 return 0;
1136 err:
1137 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1138 bmname(bitmap), ret);
1139 return ret;
1142 void bitmap_write_all(struct bitmap *bitmap)
1144 /* We don't actually write all bitmap blocks here,
1145 * just flag them as needing to be written
1147 int i;
1149 if (!bitmap || !bitmap->storage.filemap)
1150 return;
1151 if (bitmap->storage.file)
1152 /* Only one copy, so nothing needed */
1153 return;
1155 for (i = 0; i < bitmap->storage.file_pages; i++)
1156 set_page_attr(bitmap, i,
1157 BITMAP_PAGE_NEEDWRITE);
1158 bitmap->allclean = 0;
1161 static void bitmap_count_page(struct bitmap_counts *bitmap,
1162 sector_t offset, int inc)
1164 sector_t chunk = offset >> bitmap->chunkshift;
1165 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1166 bitmap->bp[page].count += inc;
1167 bitmap_checkfree(bitmap, page);
1170 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1172 sector_t chunk = offset >> bitmap->chunkshift;
1173 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1174 struct bitmap_page *bp = &bitmap->bp[page];
1176 if (!bp->pending)
1177 bp->pending = 1;
1180 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1181 sector_t offset, sector_t *blocks,
1182 int create);
1185 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1186 * out to disk
1189 void bitmap_daemon_work(struct mddev *mddev)
1191 struct bitmap *bitmap;
1192 unsigned long j;
1193 unsigned long nextpage;
1194 sector_t blocks;
1195 struct bitmap_counts *counts;
1197 /* Use a mutex to guard daemon_work against
1198 * bitmap_destroy.
1200 mutex_lock(&mddev->bitmap_info.mutex);
1201 bitmap = mddev->bitmap;
1202 if (bitmap == NULL) {
1203 mutex_unlock(&mddev->bitmap_info.mutex);
1204 return;
1206 if (time_before(jiffies, bitmap->daemon_lastrun
1207 + mddev->bitmap_info.daemon_sleep))
1208 goto done;
1210 bitmap->daemon_lastrun = jiffies;
1211 if (bitmap->allclean) {
1212 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1213 goto done;
1215 bitmap->allclean = 1;
1217 /* Any file-page which is PENDING now needs to be written.
1218 * So set NEEDWRITE now, then after we make any last-minute changes
1219 * we will write it.
1221 for (j = 0; j < bitmap->storage.file_pages; j++)
1222 if (test_and_clear_page_attr(bitmap, j,
1223 BITMAP_PAGE_PENDING))
1224 set_page_attr(bitmap, j,
1225 BITMAP_PAGE_NEEDWRITE);
1227 if (bitmap->need_sync &&
1228 mddev->bitmap_info.external == 0) {
1229 /* Arrange for superblock update as well as
1230 * other changes */
1231 bitmap_super_t *sb;
1232 bitmap->need_sync = 0;
1233 if (bitmap->storage.filemap) {
1234 sb = kmap_atomic(bitmap->storage.sb_page);
1235 sb->events_cleared =
1236 cpu_to_le64(bitmap->events_cleared);
1237 kunmap_atomic(sb);
1238 set_page_attr(bitmap, 0,
1239 BITMAP_PAGE_NEEDWRITE);
1242 /* Now look at the bitmap counters and if any are '2' or '1',
1243 * decrement and handle accordingly.
1245 counts = &bitmap->counts;
1246 spin_lock_irq(&counts->lock);
1247 nextpage = 0;
1248 for (j = 0; j < counts->chunks; j++) {
1249 bitmap_counter_t *bmc;
1250 sector_t block = (sector_t)j << counts->chunkshift;
1252 if (j == nextpage) {
1253 nextpage += PAGE_COUNTER_RATIO;
1254 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1255 j |= PAGE_COUNTER_MASK;
1256 continue;
1258 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1260 bmc = bitmap_get_counter(counts,
1261 block,
1262 &blocks, 0);
1264 if (!bmc) {
1265 j |= PAGE_COUNTER_MASK;
1266 continue;
1268 if (*bmc == 1 && !bitmap->need_sync) {
1269 /* We can clear the bit */
1270 *bmc = 0;
1271 bitmap_count_page(counts, block, -1);
1272 bitmap_file_clear_bit(bitmap, block);
1273 } else if (*bmc && *bmc <= 2) {
1274 *bmc = 1;
1275 bitmap_set_pending(counts, block);
1276 bitmap->allclean = 0;
1279 spin_unlock_irq(&counts->lock);
1281 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1282 * DIRTY pages need to be written by bitmap_unplug so it can wait
1283 * for them.
1284 * If we find any DIRTY page we stop there and let bitmap_unplug
1285 * handle all the rest. This is important in the case where
1286 * the first blocking holds the superblock and it has been updated.
1287 * We mustn't write any other blocks before the superblock.
1289 for (j = 0;
1290 j < bitmap->storage.file_pages
1291 && !test_bit(BITMAP_STALE, &bitmap->flags);
1292 j++) {
1293 if (test_page_attr(bitmap, j,
1294 BITMAP_PAGE_DIRTY))
1295 /* bitmap_unplug will handle the rest */
1296 break;
1297 if (test_and_clear_page_attr(bitmap, j,
1298 BITMAP_PAGE_NEEDWRITE)) {
1299 write_page(bitmap, bitmap->storage.filemap[j], 0);
1303 done:
1304 if (bitmap->allclean == 0)
1305 mddev->thread->timeout =
1306 mddev->bitmap_info.daemon_sleep;
1307 mutex_unlock(&mddev->bitmap_info.mutex);
1310 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1311 sector_t offset, sector_t *blocks,
1312 int create)
1313 __releases(bitmap->lock)
1314 __acquires(bitmap->lock)
1316 /* If 'create', we might release the lock and reclaim it.
1317 * The lock must have been taken with interrupts enabled.
1318 * If !create, we don't release the lock.
1320 sector_t chunk = offset >> bitmap->chunkshift;
1321 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1322 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1323 sector_t csize;
1324 int err;
1326 err = bitmap_checkpage(bitmap, page, create);
1328 if (bitmap->bp[page].hijacked ||
1329 bitmap->bp[page].map == NULL)
1330 csize = ((sector_t)1) << (bitmap->chunkshift +
1331 PAGE_COUNTER_SHIFT - 1);
1332 else
1333 csize = ((sector_t)1) << bitmap->chunkshift;
1334 *blocks = csize - (offset & (csize - 1));
1336 if (err < 0)
1337 return NULL;
1339 /* now locked ... */
1341 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1342 /* should we use the first or second counter field
1343 * of the hijacked pointer? */
1344 int hi = (pageoff > PAGE_COUNTER_MASK);
1345 return &((bitmap_counter_t *)
1346 &bitmap->bp[page].map)[hi];
1347 } else /* page is allocated */
1348 return (bitmap_counter_t *)
1349 &(bitmap->bp[page].map[pageoff]);
1352 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1354 if (!bitmap)
1355 return 0;
1357 if (behind) {
1358 int bw;
1359 atomic_inc(&bitmap->behind_writes);
1360 bw = atomic_read(&bitmap->behind_writes);
1361 if (bw > bitmap->behind_writes_used)
1362 bitmap->behind_writes_used = bw;
1364 pr_debug("inc write-behind count %d/%lu\n",
1365 bw, bitmap->mddev->bitmap_info.max_write_behind);
1368 while (sectors) {
1369 sector_t blocks;
1370 bitmap_counter_t *bmc;
1372 spin_lock_irq(&bitmap->counts.lock);
1373 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1374 if (!bmc) {
1375 spin_unlock_irq(&bitmap->counts.lock);
1376 return 0;
1379 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1380 DEFINE_WAIT(__wait);
1381 /* note that it is safe to do the prepare_to_wait
1382 * after the test as long as we do it before dropping
1383 * the spinlock.
1385 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1386 TASK_UNINTERRUPTIBLE);
1387 spin_unlock_irq(&bitmap->counts.lock);
1388 schedule();
1389 finish_wait(&bitmap->overflow_wait, &__wait);
1390 continue;
1393 switch (*bmc) {
1394 case 0:
1395 bitmap_file_set_bit(bitmap, offset);
1396 bitmap_count_page(&bitmap->counts, offset, 1);
1397 /* fall through */
1398 case 1:
1399 *bmc = 2;
1402 (*bmc)++;
1404 spin_unlock_irq(&bitmap->counts.lock);
1406 offset += blocks;
1407 if (sectors > blocks)
1408 sectors -= blocks;
1409 else
1410 sectors = 0;
1412 return 0;
1414 EXPORT_SYMBOL(bitmap_startwrite);
1416 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1417 int success, int behind)
1419 if (!bitmap)
1420 return;
1421 if (behind) {
1422 if (atomic_dec_and_test(&bitmap->behind_writes))
1423 wake_up(&bitmap->behind_wait);
1424 pr_debug("dec write-behind count %d/%lu\n",
1425 atomic_read(&bitmap->behind_writes),
1426 bitmap->mddev->bitmap_info.max_write_behind);
1429 while (sectors) {
1430 sector_t blocks;
1431 unsigned long flags;
1432 bitmap_counter_t *bmc;
1434 spin_lock_irqsave(&bitmap->counts.lock, flags);
1435 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1436 if (!bmc) {
1437 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1438 return;
1441 if (success && !bitmap->mddev->degraded &&
1442 bitmap->events_cleared < bitmap->mddev->events) {
1443 bitmap->events_cleared = bitmap->mddev->events;
1444 bitmap->need_sync = 1;
1445 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1448 if (!success && !NEEDED(*bmc))
1449 *bmc |= NEEDED_MASK;
1451 if (COUNTER(*bmc) == COUNTER_MAX)
1452 wake_up(&bitmap->overflow_wait);
1454 (*bmc)--;
1455 if (*bmc <= 2) {
1456 bitmap_set_pending(&bitmap->counts, offset);
1457 bitmap->allclean = 0;
1459 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1460 offset += blocks;
1461 if (sectors > blocks)
1462 sectors -= blocks;
1463 else
1464 sectors = 0;
1467 EXPORT_SYMBOL(bitmap_endwrite);
1469 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1470 int degraded)
1472 bitmap_counter_t *bmc;
1473 int rv;
1474 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1475 *blocks = 1024;
1476 return 1; /* always resync if no bitmap */
1478 spin_lock_irq(&bitmap->counts.lock);
1479 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1480 rv = 0;
1481 if (bmc) {
1482 /* locked */
1483 if (RESYNC(*bmc))
1484 rv = 1;
1485 else if (NEEDED(*bmc)) {
1486 rv = 1;
1487 if (!degraded) { /* don't set/clear bits if degraded */
1488 *bmc |= RESYNC_MASK;
1489 *bmc &= ~NEEDED_MASK;
1493 spin_unlock_irq(&bitmap->counts.lock);
1494 return rv;
1497 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1498 int degraded)
1500 /* bitmap_start_sync must always report on multiples of whole
1501 * pages, otherwise resync (which is very PAGE_SIZE based) will
1502 * get confused.
1503 * So call __bitmap_start_sync repeatedly (if needed) until
1504 * At least PAGE_SIZE>>9 blocks are covered.
1505 * Return the 'or' of the result.
1507 int rv = 0;
1508 sector_t blocks1;
1510 *blocks = 0;
1511 while (*blocks < (PAGE_SIZE>>9)) {
1512 rv |= __bitmap_start_sync(bitmap, offset,
1513 &blocks1, degraded);
1514 offset += blocks1;
1515 *blocks += blocks1;
1517 return rv;
1519 EXPORT_SYMBOL(bitmap_start_sync);
1521 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1523 bitmap_counter_t *bmc;
1524 unsigned long flags;
1526 if (bitmap == NULL) {
1527 *blocks = 1024;
1528 return;
1530 spin_lock_irqsave(&bitmap->counts.lock, flags);
1531 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1532 if (bmc == NULL)
1533 goto unlock;
1534 /* locked */
1535 if (RESYNC(*bmc)) {
1536 *bmc &= ~RESYNC_MASK;
1538 if (!NEEDED(*bmc) && aborted)
1539 *bmc |= NEEDED_MASK;
1540 else {
1541 if (*bmc <= 2) {
1542 bitmap_set_pending(&bitmap->counts, offset);
1543 bitmap->allclean = 0;
1547 unlock:
1548 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1550 EXPORT_SYMBOL(bitmap_end_sync);
1552 void bitmap_close_sync(struct bitmap *bitmap)
1554 /* Sync has finished, and any bitmap chunks that weren't synced
1555 * properly have been aborted. It remains to us to clear the
1556 * RESYNC bit wherever it is still on
1558 sector_t sector = 0;
1559 sector_t blocks;
1560 if (!bitmap)
1561 return;
1562 while (sector < bitmap->mddev->resync_max_sectors) {
1563 bitmap_end_sync(bitmap, sector, &blocks, 0);
1564 sector += blocks;
1567 EXPORT_SYMBOL(bitmap_close_sync);
1569 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1571 sector_t s = 0;
1572 sector_t blocks;
1574 if (!bitmap)
1575 return;
1576 if (sector == 0) {
1577 bitmap->last_end_sync = jiffies;
1578 return;
1580 if (!force && time_before(jiffies, (bitmap->last_end_sync
1581 + bitmap->mddev->bitmap_info.daemon_sleep)))
1582 return;
1583 wait_event(bitmap->mddev->recovery_wait,
1584 atomic_read(&bitmap->mddev->recovery_active) == 0);
1586 bitmap->mddev->curr_resync_completed = sector;
1587 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1588 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1589 s = 0;
1590 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1591 bitmap_end_sync(bitmap, s, &blocks, 0);
1592 s += blocks;
1594 bitmap->last_end_sync = jiffies;
1595 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1597 EXPORT_SYMBOL(bitmap_cond_end_sync);
1599 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1601 /* For each chunk covered by any of these sectors, set the
1602 * counter to 2 and possibly set resync_needed. They should all
1603 * be 0 at this point
1606 sector_t secs;
1607 bitmap_counter_t *bmc;
1608 spin_lock_irq(&bitmap->counts.lock);
1609 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1610 if (!bmc) {
1611 spin_unlock_irq(&bitmap->counts.lock);
1612 return;
1614 if (!*bmc) {
1615 *bmc = 2;
1616 bitmap_count_page(&bitmap->counts, offset, 1);
1617 bitmap_set_pending(&bitmap->counts, offset);
1618 bitmap->allclean = 0;
1620 if (needed)
1621 *bmc |= NEEDED_MASK;
1622 spin_unlock_irq(&bitmap->counts.lock);
1625 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1626 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1628 unsigned long chunk;
1630 for (chunk = s; chunk <= e; chunk++) {
1631 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1632 bitmap_set_memory_bits(bitmap, sec, 1);
1633 bitmap_file_set_bit(bitmap, sec);
1634 if (sec < bitmap->mddev->recovery_cp)
1635 /* We are asserting that the array is dirty,
1636 * so move the recovery_cp address back so
1637 * that it is obvious that it is dirty
1639 bitmap->mddev->recovery_cp = sec;
1644 * flush out any pending updates
1646 void bitmap_flush(struct mddev *mddev)
1648 struct bitmap *bitmap = mddev->bitmap;
1649 long sleep;
1651 if (!bitmap) /* there was no bitmap */
1652 return;
1654 /* run the daemon_work three time to ensure everything is flushed
1655 * that can be
1657 sleep = mddev->bitmap_info.daemon_sleep * 2;
1658 bitmap->daemon_lastrun -= sleep;
1659 bitmap_daemon_work(mddev);
1660 bitmap->daemon_lastrun -= sleep;
1661 bitmap_daemon_work(mddev);
1662 bitmap->daemon_lastrun -= sleep;
1663 bitmap_daemon_work(mddev);
1664 bitmap_update_sb(bitmap);
1668 * free memory that was allocated
1670 static void bitmap_free(struct bitmap *bitmap)
1672 unsigned long k, pages;
1673 struct bitmap_page *bp;
1675 if (!bitmap) /* there was no bitmap */
1676 return;
1678 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1679 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1680 md_cluster_stop(bitmap->mddev);
1682 /* Shouldn't be needed - but just in case.... */
1683 wait_event(bitmap->write_wait,
1684 atomic_read(&bitmap->pending_writes) == 0);
1686 /* release the bitmap file */
1687 bitmap_file_unmap(&bitmap->storage);
1689 bp = bitmap->counts.bp;
1690 pages = bitmap->counts.pages;
1692 /* free all allocated memory */
1694 if (bp) /* deallocate the page memory */
1695 for (k = 0; k < pages; k++)
1696 if (bp[k].map && !bp[k].hijacked)
1697 kfree(bp[k].map);
1698 kfree(bp);
1699 kfree(bitmap);
1702 void bitmap_destroy(struct mddev *mddev)
1704 struct bitmap *bitmap = mddev->bitmap;
1706 if (!bitmap) /* there was no bitmap */
1707 return;
1709 mutex_lock(&mddev->bitmap_info.mutex);
1710 spin_lock(&mddev->lock);
1711 mddev->bitmap = NULL; /* disconnect from the md device */
1712 spin_unlock(&mddev->lock);
1713 mutex_unlock(&mddev->bitmap_info.mutex);
1714 if (mddev->thread)
1715 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1717 if (bitmap->sysfs_can_clear)
1718 sysfs_put(bitmap->sysfs_can_clear);
1720 bitmap_free(bitmap);
1724 * initialize the bitmap structure
1725 * if this returns an error, bitmap_destroy must be called to do clean up
1727 struct bitmap *bitmap_create(struct mddev *mddev, int slot)
1729 struct bitmap *bitmap;
1730 sector_t blocks = mddev->resync_max_sectors;
1731 struct file *file = mddev->bitmap_info.file;
1732 int err;
1733 struct kernfs_node *bm = NULL;
1735 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1737 BUG_ON(file && mddev->bitmap_info.offset);
1739 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1740 if (!bitmap)
1741 return ERR_PTR(-ENOMEM);
1743 spin_lock_init(&bitmap->counts.lock);
1744 atomic_set(&bitmap->pending_writes, 0);
1745 init_waitqueue_head(&bitmap->write_wait);
1746 init_waitqueue_head(&bitmap->overflow_wait);
1747 init_waitqueue_head(&bitmap->behind_wait);
1749 bitmap->mddev = mddev;
1750 bitmap->cluster_slot = slot;
1752 if (mddev->kobj.sd)
1753 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1754 if (bm) {
1755 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1756 sysfs_put(bm);
1757 } else
1758 bitmap->sysfs_can_clear = NULL;
1760 bitmap->storage.file = file;
1761 if (file) {
1762 get_file(file);
1763 /* As future accesses to this file will use bmap,
1764 * and bypass the page cache, we must sync the file
1765 * first.
1767 vfs_fsync(file, 1);
1769 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1770 if (!mddev->bitmap_info.external) {
1772 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1773 * instructing us to create a new on-disk bitmap instance.
1775 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1776 err = bitmap_new_disk_sb(bitmap);
1777 else
1778 err = bitmap_read_sb(bitmap);
1779 } else {
1780 err = 0;
1781 if (mddev->bitmap_info.chunksize == 0 ||
1782 mddev->bitmap_info.daemon_sleep == 0)
1783 /* chunksize and time_base need to be
1784 * set first. */
1785 err = -EINVAL;
1787 if (err)
1788 goto error;
1790 bitmap->daemon_lastrun = jiffies;
1791 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1792 if (err)
1793 goto error;
1795 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1796 bitmap->counts.pages, bmname(bitmap));
1798 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1799 if (err)
1800 goto error;
1802 return bitmap;
1803 error:
1804 bitmap_free(bitmap);
1805 return ERR_PTR(err);
1808 int bitmap_load(struct mddev *mddev)
1810 int err = 0;
1811 sector_t start = 0;
1812 sector_t sector = 0;
1813 struct bitmap *bitmap = mddev->bitmap;
1815 if (!bitmap)
1816 goto out;
1818 /* Clear out old bitmap info first: Either there is none, or we
1819 * are resuming after someone else has possibly changed things,
1820 * so we should forget old cached info.
1821 * All chunks should be clean, but some might need_sync.
1823 while (sector < mddev->resync_max_sectors) {
1824 sector_t blocks;
1825 bitmap_start_sync(bitmap, sector, &blocks, 0);
1826 sector += blocks;
1828 bitmap_close_sync(bitmap);
1830 if (mddev->degraded == 0
1831 || bitmap->events_cleared == mddev->events)
1832 /* no need to keep dirty bits to optimise a
1833 * re-add of a missing device */
1834 start = mddev->recovery_cp;
1836 mutex_lock(&mddev->bitmap_info.mutex);
1837 err = bitmap_init_from_disk(bitmap, start);
1838 mutex_unlock(&mddev->bitmap_info.mutex);
1840 if (err)
1841 goto out;
1842 clear_bit(BITMAP_STALE, &bitmap->flags);
1844 /* Kick recovery in case any bits were set */
1845 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1847 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1848 md_wakeup_thread(mddev->thread);
1850 bitmap_update_sb(bitmap);
1852 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1853 err = -EIO;
1854 out:
1855 return err;
1857 EXPORT_SYMBOL_GPL(bitmap_load);
1859 /* Loads the bitmap associated with slot and copies the resync information
1860 * to our bitmap
1862 int bitmap_copy_from_slot(struct mddev *mddev, int slot,
1863 sector_t *low, sector_t *high, bool clear_bits)
1865 int rv = 0, i, j;
1866 sector_t block, lo = 0, hi = 0;
1867 struct bitmap_counts *counts;
1868 struct bitmap *bitmap = bitmap_create(mddev, slot);
1870 if (IS_ERR(bitmap))
1871 return PTR_ERR(bitmap);
1873 rv = bitmap_init_from_disk(bitmap, 0);
1874 if (rv)
1875 goto err;
1877 counts = &bitmap->counts;
1878 for (j = 0; j < counts->chunks; j++) {
1879 block = (sector_t)j << counts->chunkshift;
1880 if (bitmap_file_test_bit(bitmap, block)) {
1881 if (!lo)
1882 lo = block;
1883 hi = block;
1884 bitmap_file_clear_bit(bitmap, block);
1885 bitmap_set_memory_bits(mddev->bitmap, block, 1);
1886 bitmap_file_set_bit(mddev->bitmap, block);
1890 if (clear_bits) {
1891 bitmap_update_sb(bitmap);
1892 /* Setting this for the ev_page should be enough.
1893 * And we do not require both write_all and PAGE_DIRT either
1895 for (i = 0; i < bitmap->storage.file_pages; i++)
1896 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1897 bitmap_write_all(bitmap);
1898 bitmap_unplug(bitmap);
1900 *low = lo;
1901 *high = hi;
1902 err:
1903 bitmap_free(bitmap);
1904 return rv;
1906 EXPORT_SYMBOL_GPL(bitmap_copy_from_slot);
1909 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1911 unsigned long chunk_kb;
1912 struct bitmap_counts *counts;
1914 if (!bitmap)
1915 return;
1917 counts = &bitmap->counts;
1919 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1920 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1921 "%lu%s chunk",
1922 counts->pages - counts->missing_pages,
1923 counts->pages,
1924 (counts->pages - counts->missing_pages)
1925 << (PAGE_SHIFT - 10),
1926 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1927 chunk_kb ? "KB" : "B");
1928 if (bitmap->storage.file) {
1929 seq_printf(seq, ", file: ");
1930 seq_file_path(seq, bitmap->storage.file, " \t\n");
1933 seq_printf(seq, "\n");
1936 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1937 int chunksize, int init)
1939 /* If chunk_size is 0, choose an appropriate chunk size.
1940 * Then possibly allocate new storage space.
1941 * Then quiesce, copy bits, replace bitmap, and re-start
1943 * This function is called both to set up the initial bitmap
1944 * and to resize the bitmap while the array is active.
1945 * If this happens as a result of the array being resized,
1946 * chunksize will be zero, and we need to choose a suitable
1947 * chunksize, otherwise we use what we are given.
1949 struct bitmap_storage store;
1950 struct bitmap_counts old_counts;
1951 unsigned long chunks;
1952 sector_t block;
1953 sector_t old_blocks, new_blocks;
1954 int chunkshift;
1955 int ret = 0;
1956 long pages;
1957 struct bitmap_page *new_bp;
1959 if (chunksize == 0) {
1960 /* If there is enough space, leave the chunk size unchanged,
1961 * else increase by factor of two until there is enough space.
1963 long bytes;
1964 long space = bitmap->mddev->bitmap_info.space;
1966 if (space == 0) {
1967 /* We don't know how much space there is, so limit
1968 * to current size - in sectors.
1970 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1971 if (!bitmap->mddev->bitmap_info.external)
1972 bytes += sizeof(bitmap_super_t);
1973 space = DIV_ROUND_UP(bytes, 512);
1974 bitmap->mddev->bitmap_info.space = space;
1976 chunkshift = bitmap->counts.chunkshift;
1977 chunkshift--;
1978 do {
1979 /* 'chunkshift' is shift from block size to chunk size */
1980 chunkshift++;
1981 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1982 bytes = DIV_ROUND_UP(chunks, 8);
1983 if (!bitmap->mddev->bitmap_info.external)
1984 bytes += sizeof(bitmap_super_t);
1985 } while (bytes > (space << 9));
1986 } else
1987 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1989 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1990 memset(&store, 0, sizeof(store));
1991 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1992 ret = bitmap_storage_alloc(&store, chunks,
1993 !bitmap->mddev->bitmap_info.external,
1994 mddev_is_clustered(bitmap->mddev)
1995 ? bitmap->cluster_slot : 0);
1996 if (ret)
1997 goto err;
1999 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2001 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
2002 ret = -ENOMEM;
2003 if (!new_bp) {
2004 bitmap_file_unmap(&store);
2005 goto err;
2008 if (!init)
2009 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2011 store.file = bitmap->storage.file;
2012 bitmap->storage.file = NULL;
2014 if (store.sb_page && bitmap->storage.sb_page)
2015 memcpy(page_address(store.sb_page),
2016 page_address(bitmap->storage.sb_page),
2017 sizeof(bitmap_super_t));
2018 bitmap_file_unmap(&bitmap->storage);
2019 bitmap->storage = store;
2021 old_counts = bitmap->counts;
2022 bitmap->counts.bp = new_bp;
2023 bitmap->counts.pages = pages;
2024 bitmap->counts.missing_pages = pages;
2025 bitmap->counts.chunkshift = chunkshift;
2026 bitmap->counts.chunks = chunks;
2027 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2028 BITMAP_BLOCK_SHIFT);
2030 blocks = min(old_counts.chunks << old_counts.chunkshift,
2031 chunks << chunkshift);
2033 spin_lock_irq(&bitmap->counts.lock);
2034 for (block = 0; block < blocks; ) {
2035 bitmap_counter_t *bmc_old, *bmc_new;
2036 int set;
2038 bmc_old = bitmap_get_counter(&old_counts, block,
2039 &old_blocks, 0);
2040 set = bmc_old && NEEDED(*bmc_old);
2042 if (set) {
2043 bmc_new = bitmap_get_counter(&bitmap->counts, block,
2044 &new_blocks, 1);
2045 if (*bmc_new == 0) {
2046 /* need to set on-disk bits too. */
2047 sector_t end = block + new_blocks;
2048 sector_t start = block >> chunkshift;
2049 start <<= chunkshift;
2050 while (start < end) {
2051 bitmap_file_set_bit(bitmap, block);
2052 start += 1 << chunkshift;
2054 *bmc_new = 2;
2055 bitmap_count_page(&bitmap->counts,
2056 block, 1);
2057 bitmap_set_pending(&bitmap->counts,
2058 block);
2060 *bmc_new |= NEEDED_MASK;
2061 if (new_blocks < old_blocks)
2062 old_blocks = new_blocks;
2064 block += old_blocks;
2067 if (!init) {
2068 int i;
2069 while (block < (chunks << chunkshift)) {
2070 bitmap_counter_t *bmc;
2071 bmc = bitmap_get_counter(&bitmap->counts, block,
2072 &new_blocks, 1);
2073 if (bmc) {
2074 /* new space. It needs to be resynced, so
2075 * we set NEEDED_MASK.
2077 if (*bmc == 0) {
2078 *bmc = NEEDED_MASK | 2;
2079 bitmap_count_page(&bitmap->counts,
2080 block, 1);
2081 bitmap_set_pending(&bitmap->counts,
2082 block);
2085 block += new_blocks;
2087 for (i = 0; i < bitmap->storage.file_pages; i++)
2088 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2090 spin_unlock_irq(&bitmap->counts.lock);
2092 if (!init) {
2093 bitmap_unplug(bitmap);
2094 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2096 ret = 0;
2097 err:
2098 return ret;
2100 EXPORT_SYMBOL_GPL(bitmap_resize);
2102 static ssize_t
2103 location_show(struct mddev *mddev, char *page)
2105 ssize_t len;
2106 if (mddev->bitmap_info.file)
2107 len = sprintf(page, "file");
2108 else if (mddev->bitmap_info.offset)
2109 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2110 else
2111 len = sprintf(page, "none");
2112 len += sprintf(page+len, "\n");
2113 return len;
2116 static ssize_t
2117 location_store(struct mddev *mddev, const char *buf, size_t len)
2120 if (mddev->pers) {
2121 if (!mddev->pers->quiesce)
2122 return -EBUSY;
2123 if (mddev->recovery || mddev->sync_thread)
2124 return -EBUSY;
2127 if (mddev->bitmap || mddev->bitmap_info.file ||
2128 mddev->bitmap_info.offset) {
2129 /* bitmap already configured. Only option is to clear it */
2130 if (strncmp(buf, "none", 4) != 0)
2131 return -EBUSY;
2132 if (mddev->pers) {
2133 mddev->pers->quiesce(mddev, 1);
2134 bitmap_destroy(mddev);
2135 mddev->pers->quiesce(mddev, 0);
2137 mddev->bitmap_info.offset = 0;
2138 if (mddev->bitmap_info.file) {
2139 struct file *f = mddev->bitmap_info.file;
2140 mddev->bitmap_info.file = NULL;
2141 fput(f);
2143 } else {
2144 /* No bitmap, OK to set a location */
2145 long long offset;
2146 if (strncmp(buf, "none", 4) == 0)
2147 /* nothing to be done */;
2148 else if (strncmp(buf, "file:", 5) == 0) {
2149 /* Not supported yet */
2150 return -EINVAL;
2151 } else {
2152 int rv;
2153 if (buf[0] == '+')
2154 rv = kstrtoll(buf+1, 10, &offset);
2155 else
2156 rv = kstrtoll(buf, 10, &offset);
2157 if (rv)
2158 return rv;
2159 if (offset == 0)
2160 return -EINVAL;
2161 if (mddev->bitmap_info.external == 0 &&
2162 mddev->major_version == 0 &&
2163 offset != mddev->bitmap_info.default_offset)
2164 return -EINVAL;
2165 mddev->bitmap_info.offset = offset;
2166 if (mddev->pers) {
2167 struct bitmap *bitmap;
2168 mddev->pers->quiesce(mddev, 1);
2169 bitmap = bitmap_create(mddev, -1);
2170 if (IS_ERR(bitmap))
2171 rv = PTR_ERR(bitmap);
2172 else {
2173 mddev->bitmap = bitmap;
2174 rv = bitmap_load(mddev);
2175 if (rv) {
2176 bitmap_destroy(mddev);
2177 mddev->bitmap_info.offset = 0;
2180 mddev->pers->quiesce(mddev, 0);
2181 if (rv)
2182 return rv;
2186 if (!mddev->external) {
2187 /* Ensure new bitmap info is stored in
2188 * metadata promptly.
2190 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2191 md_wakeup_thread(mddev->thread);
2193 return len;
2196 static struct md_sysfs_entry bitmap_location =
2197 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2199 /* 'bitmap/space' is the space available at 'location' for the
2200 * bitmap. This allows the kernel to know when it is safe to
2201 * resize the bitmap to match a resized array.
2203 static ssize_t
2204 space_show(struct mddev *mddev, char *page)
2206 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2209 static ssize_t
2210 space_store(struct mddev *mddev, const char *buf, size_t len)
2212 unsigned long sectors;
2213 int rv;
2215 rv = kstrtoul(buf, 10, &sectors);
2216 if (rv)
2217 return rv;
2219 if (sectors == 0)
2220 return -EINVAL;
2222 if (mddev->bitmap &&
2223 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2224 return -EFBIG; /* Bitmap is too big for this small space */
2226 /* could make sure it isn't too big, but that isn't really
2227 * needed - user-space should be careful.
2229 mddev->bitmap_info.space = sectors;
2230 return len;
2233 static struct md_sysfs_entry bitmap_space =
2234 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2236 static ssize_t
2237 timeout_show(struct mddev *mddev, char *page)
2239 ssize_t len;
2240 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2241 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2243 len = sprintf(page, "%lu", secs);
2244 if (jifs)
2245 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2246 len += sprintf(page+len, "\n");
2247 return len;
2250 static ssize_t
2251 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2253 /* timeout can be set at any time */
2254 unsigned long timeout;
2255 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2256 if (rv)
2257 return rv;
2259 /* just to make sure we don't overflow... */
2260 if (timeout >= LONG_MAX / HZ)
2261 return -EINVAL;
2263 timeout = timeout * HZ / 10000;
2265 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2266 timeout = MAX_SCHEDULE_TIMEOUT-1;
2267 if (timeout < 1)
2268 timeout = 1;
2269 mddev->bitmap_info.daemon_sleep = timeout;
2270 if (mddev->thread) {
2271 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2272 * the bitmap is all clean and we don't need to
2273 * adjust the timeout right now
2275 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2276 mddev->thread->timeout = timeout;
2277 md_wakeup_thread(mddev->thread);
2280 return len;
2283 static struct md_sysfs_entry bitmap_timeout =
2284 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2286 static ssize_t
2287 backlog_show(struct mddev *mddev, char *page)
2289 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2292 static ssize_t
2293 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2295 unsigned long backlog;
2296 int rv = kstrtoul(buf, 10, &backlog);
2297 if (rv)
2298 return rv;
2299 if (backlog > COUNTER_MAX)
2300 return -EINVAL;
2301 mddev->bitmap_info.max_write_behind = backlog;
2302 return len;
2305 static struct md_sysfs_entry bitmap_backlog =
2306 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2308 static ssize_t
2309 chunksize_show(struct mddev *mddev, char *page)
2311 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2314 static ssize_t
2315 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2317 /* Can only be changed when no bitmap is active */
2318 int rv;
2319 unsigned long csize;
2320 if (mddev->bitmap)
2321 return -EBUSY;
2322 rv = kstrtoul(buf, 10, &csize);
2323 if (rv)
2324 return rv;
2325 if (csize < 512 ||
2326 !is_power_of_2(csize))
2327 return -EINVAL;
2328 mddev->bitmap_info.chunksize = csize;
2329 return len;
2332 static struct md_sysfs_entry bitmap_chunksize =
2333 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2335 static ssize_t metadata_show(struct mddev *mddev, char *page)
2337 if (mddev_is_clustered(mddev))
2338 return sprintf(page, "clustered\n");
2339 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2340 ? "external" : "internal"));
2343 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2345 if (mddev->bitmap ||
2346 mddev->bitmap_info.file ||
2347 mddev->bitmap_info.offset)
2348 return -EBUSY;
2349 if (strncmp(buf, "external", 8) == 0)
2350 mddev->bitmap_info.external = 1;
2351 else if ((strncmp(buf, "internal", 8) == 0) ||
2352 (strncmp(buf, "clustered", 9) == 0))
2353 mddev->bitmap_info.external = 0;
2354 else
2355 return -EINVAL;
2356 return len;
2359 static struct md_sysfs_entry bitmap_metadata =
2360 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2362 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2364 int len;
2365 spin_lock(&mddev->lock);
2366 if (mddev->bitmap)
2367 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2368 "false" : "true"));
2369 else
2370 len = sprintf(page, "\n");
2371 spin_unlock(&mddev->lock);
2372 return len;
2375 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2377 if (mddev->bitmap == NULL)
2378 return -ENOENT;
2379 if (strncmp(buf, "false", 5) == 0)
2380 mddev->bitmap->need_sync = 1;
2381 else if (strncmp(buf, "true", 4) == 0) {
2382 if (mddev->degraded)
2383 return -EBUSY;
2384 mddev->bitmap->need_sync = 0;
2385 } else
2386 return -EINVAL;
2387 return len;
2390 static struct md_sysfs_entry bitmap_can_clear =
2391 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2393 static ssize_t
2394 behind_writes_used_show(struct mddev *mddev, char *page)
2396 ssize_t ret;
2397 spin_lock(&mddev->lock);
2398 if (mddev->bitmap == NULL)
2399 ret = sprintf(page, "0\n");
2400 else
2401 ret = sprintf(page, "%lu\n",
2402 mddev->bitmap->behind_writes_used);
2403 spin_unlock(&mddev->lock);
2404 return ret;
2407 static ssize_t
2408 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2410 if (mddev->bitmap)
2411 mddev->bitmap->behind_writes_used = 0;
2412 return len;
2415 static struct md_sysfs_entry max_backlog_used =
2416 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2417 behind_writes_used_show, behind_writes_used_reset);
2419 static struct attribute *md_bitmap_attrs[] = {
2420 &bitmap_location.attr,
2421 &bitmap_space.attr,
2422 &bitmap_timeout.attr,
2423 &bitmap_backlog.attr,
2424 &bitmap_chunksize.attr,
2425 &bitmap_metadata.attr,
2426 &bitmap_can_clear.attr,
2427 &max_backlog_used.attr,
2428 NULL
2430 struct attribute_group md_bitmap_group = {
2431 .name = "bitmap",
2432 .attrs = md_bitmap_attrs,