bna: remove oper_state_cbfn from struct bna_rxf
[linux/fpc-iii.git] / drivers / md / bitmap.c
blob135a0907e9de413d140e9fb9b793a91b638a1606
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;
213 int node_offset = 0;
215 if (mddev_is_clustered(bitmap->mddev))
216 node_offset = bitmap->cluster_slot * store->file_pages;
218 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
219 int size = PAGE_SIZE;
220 loff_t offset = mddev->bitmap_info.offset;
222 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
224 if (page->index == store->file_pages-1) {
225 int last_page_size = store->bytes & (PAGE_SIZE-1);
226 if (last_page_size == 0)
227 last_page_size = PAGE_SIZE;
228 size = roundup(last_page_size,
229 bdev_logical_block_size(bdev));
231 /* Just make sure we aren't corrupting data or
232 * metadata
234 if (mddev->external) {
235 /* Bitmap could be anywhere. */
236 if (rdev->sb_start + offset + (page->index
237 * (PAGE_SIZE/512))
238 > rdev->data_offset
240 rdev->sb_start + offset
241 < (rdev->data_offset + mddev->dev_sectors
242 + (PAGE_SIZE/512)))
243 goto bad_alignment;
244 } else if (offset < 0) {
245 /* DATA BITMAP METADATA */
246 if (offset
247 + (long)(page->index * (PAGE_SIZE/512))
248 + size/512 > 0)
249 /* bitmap runs in to metadata */
250 goto bad_alignment;
251 if (rdev->data_offset + mddev->dev_sectors
252 > rdev->sb_start + offset)
253 /* data runs in to bitmap */
254 goto bad_alignment;
255 } else if (rdev->sb_start < rdev->data_offset) {
256 /* METADATA BITMAP DATA */
257 if (rdev->sb_start
258 + offset
259 + page->index*(PAGE_SIZE/512) + size/512
260 > rdev->data_offset)
261 /* bitmap runs in to data */
262 goto bad_alignment;
263 } else {
264 /* DATA METADATA BITMAP - no problems */
266 md_super_write(mddev, rdev,
267 rdev->sb_start + offset
268 + page->index * (PAGE_SIZE/512),
269 size,
270 page);
273 if (wait)
274 md_super_wait(mddev);
275 return 0;
277 bad_alignment:
278 return -EINVAL;
281 static void bitmap_file_kick(struct bitmap *bitmap);
283 * write out a page to a file
285 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
287 struct buffer_head *bh;
289 if (bitmap->storage.file == NULL) {
290 switch (write_sb_page(bitmap, page, wait)) {
291 case -EINVAL:
292 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
294 } else {
296 bh = page_buffers(page);
298 while (bh && bh->b_blocknr) {
299 atomic_inc(&bitmap->pending_writes);
300 set_buffer_locked(bh);
301 set_buffer_mapped(bh);
302 submit_bh(WRITE | REQ_SYNC, bh);
303 bh = bh->b_this_page;
306 if (wait)
307 wait_event(bitmap->write_wait,
308 atomic_read(&bitmap->pending_writes)==0);
310 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
311 bitmap_file_kick(bitmap);
314 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
316 struct bitmap *bitmap = bh->b_private;
318 if (!uptodate)
319 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
320 if (atomic_dec_and_test(&bitmap->pending_writes))
321 wake_up(&bitmap->write_wait);
324 /* copied from buffer.c */
325 static void
326 __clear_page_buffers(struct page *page)
328 ClearPagePrivate(page);
329 set_page_private(page, 0);
330 page_cache_release(page);
332 static void free_buffers(struct page *page)
334 struct buffer_head *bh;
336 if (!PagePrivate(page))
337 return;
339 bh = page_buffers(page);
340 while (bh) {
341 struct buffer_head *next = bh->b_this_page;
342 free_buffer_head(bh);
343 bh = next;
345 __clear_page_buffers(page);
346 put_page(page);
349 /* read a page from a file.
350 * We both read the page, and attach buffers to the page to record the
351 * address of each block (using bmap). These addresses will be used
352 * to write the block later, completely bypassing the filesystem.
353 * This usage is similar to how swap files are handled, and allows us
354 * to write to a file with no concerns of memory allocation failing.
356 static int read_page(struct file *file, unsigned long index,
357 struct bitmap *bitmap,
358 unsigned long count,
359 struct page *page)
361 int ret = 0;
362 struct inode *inode = file_inode(file);
363 struct buffer_head *bh;
364 sector_t block;
366 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
367 (unsigned long long)index << PAGE_SHIFT);
369 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
370 if (!bh) {
371 ret = -ENOMEM;
372 goto out;
374 attach_page_buffers(page, bh);
375 block = index << (PAGE_SHIFT - inode->i_blkbits);
376 while (bh) {
377 if (count == 0)
378 bh->b_blocknr = 0;
379 else {
380 bh->b_blocknr = bmap(inode, block);
381 if (bh->b_blocknr == 0) {
382 /* Cannot use this file! */
383 ret = -EINVAL;
384 goto out;
386 bh->b_bdev = inode->i_sb->s_bdev;
387 if (count < (1<<inode->i_blkbits))
388 count = 0;
389 else
390 count -= (1<<inode->i_blkbits);
392 bh->b_end_io = end_bitmap_write;
393 bh->b_private = bitmap;
394 atomic_inc(&bitmap->pending_writes);
395 set_buffer_locked(bh);
396 set_buffer_mapped(bh);
397 submit_bh(READ, bh);
399 block++;
400 bh = bh->b_this_page;
402 page->index = index;
404 wait_event(bitmap->write_wait,
405 atomic_read(&bitmap->pending_writes)==0);
406 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
407 ret = -EIO;
408 out:
409 if (ret)
410 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
411 (int)PAGE_SIZE,
412 (unsigned long long)index << PAGE_SHIFT,
413 ret);
414 return ret;
418 * bitmap file superblock operations
421 /* update the event counter and sync the superblock to disk */
422 void bitmap_update_sb(struct bitmap *bitmap)
424 bitmap_super_t *sb;
426 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
427 return;
428 if (bitmap->mddev->bitmap_info.external)
429 return;
430 if (!bitmap->storage.sb_page) /* no superblock */
431 return;
432 sb = kmap_atomic(bitmap->storage.sb_page);
433 sb->events = cpu_to_le64(bitmap->mddev->events);
434 if (bitmap->mddev->events < bitmap->events_cleared)
435 /* rocking back to read-only */
436 bitmap->events_cleared = bitmap->mddev->events;
437 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
438 sb->state = cpu_to_le32(bitmap->flags);
439 /* Just in case these have been changed via sysfs: */
440 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
441 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
442 /* This might have been changed by a reshape */
443 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
444 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
445 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
446 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
447 bitmap_info.space);
448 kunmap_atomic(sb);
449 write_page(bitmap, bitmap->storage.sb_page, 1);
452 /* print out the bitmap file superblock */
453 void bitmap_print_sb(struct bitmap *bitmap)
455 bitmap_super_t *sb;
457 if (!bitmap || !bitmap->storage.sb_page)
458 return;
459 sb = kmap_atomic(bitmap->storage.sb_page);
460 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
461 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
462 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
463 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
464 *(__u32 *)(sb->uuid+0),
465 *(__u32 *)(sb->uuid+4),
466 *(__u32 *)(sb->uuid+8),
467 *(__u32 *)(sb->uuid+12));
468 printk(KERN_DEBUG " events: %llu\n",
469 (unsigned long long) le64_to_cpu(sb->events));
470 printk(KERN_DEBUG "events cleared: %llu\n",
471 (unsigned long long) le64_to_cpu(sb->events_cleared));
472 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
473 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
474 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
475 printk(KERN_DEBUG " sync size: %llu KB\n",
476 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
477 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
478 kunmap_atomic(sb);
482 * bitmap_new_disk_sb
483 * @bitmap
485 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
486 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
487 * This function verifies 'bitmap_info' and populates the on-disk bitmap
488 * structure, which is to be written to disk.
490 * Returns: 0 on success, -Exxx on error
492 static int bitmap_new_disk_sb(struct bitmap *bitmap)
494 bitmap_super_t *sb;
495 unsigned long chunksize, daemon_sleep, write_behind;
497 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
498 if (bitmap->storage.sb_page == NULL)
499 return -ENOMEM;
500 bitmap->storage.sb_page->index = 0;
502 sb = kmap_atomic(bitmap->storage.sb_page);
504 sb->magic = cpu_to_le32(BITMAP_MAGIC);
505 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
507 chunksize = bitmap->mddev->bitmap_info.chunksize;
508 BUG_ON(!chunksize);
509 if (!is_power_of_2(chunksize)) {
510 kunmap_atomic(sb);
511 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
512 return -EINVAL;
514 sb->chunksize = cpu_to_le32(chunksize);
516 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
517 if (!daemon_sleep ||
518 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
519 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
520 daemon_sleep = 5 * HZ;
522 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
523 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
526 * FIXME: write_behind for RAID1. If not specified, what
527 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
529 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
530 if (write_behind > COUNTER_MAX)
531 write_behind = COUNTER_MAX / 2;
532 sb->write_behind = cpu_to_le32(write_behind);
533 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
535 /* keep the array size field of the bitmap superblock up to date */
536 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
538 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
540 set_bit(BITMAP_STALE, &bitmap->flags);
541 sb->state = cpu_to_le32(bitmap->flags);
542 bitmap->events_cleared = bitmap->mddev->events;
543 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
545 kunmap_atomic(sb);
547 return 0;
550 /* read the superblock from the bitmap file and initialize some bitmap fields */
551 static int bitmap_read_sb(struct bitmap *bitmap)
553 char *reason = NULL;
554 bitmap_super_t *sb;
555 unsigned long chunksize, daemon_sleep, write_behind;
556 unsigned long long events;
557 int nodes = 0;
558 unsigned long sectors_reserved = 0;
559 int err = -EINVAL;
560 struct page *sb_page;
562 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
563 chunksize = 128 * 1024 * 1024;
564 daemon_sleep = 5 * HZ;
565 write_behind = 0;
566 set_bit(BITMAP_STALE, &bitmap->flags);
567 err = 0;
568 goto out_no_sb;
570 /* page 0 is the superblock, read it... */
571 sb_page = alloc_page(GFP_KERNEL);
572 if (!sb_page)
573 return -ENOMEM;
574 bitmap->storage.sb_page = sb_page;
576 re_read:
577 /* If cluster_slot is set, the cluster is setup */
578 if (bitmap->cluster_slot >= 0) {
579 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
581 sector_div(bm_blocks,
582 bitmap->mddev->bitmap_info.chunksize >> 9);
583 /* bits to bytes */
584 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
585 /* to 4k blocks */
586 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
587 bitmap->mddev->bitmap_info.offset += bitmap->cluster_slot * (bm_blocks << 3);
588 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
589 bitmap->cluster_slot, (unsigned long long)bitmap->mddev->bitmap_info.offset);
592 if (bitmap->storage.file) {
593 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
594 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
596 err = read_page(bitmap->storage.file, 0,
597 bitmap, bytes, sb_page);
598 } else {
599 err = read_sb_page(bitmap->mddev,
600 bitmap->mddev->bitmap_info.offset,
601 sb_page,
602 0, sizeof(bitmap_super_t));
604 if (err)
605 return err;
607 err = -EINVAL;
608 sb = kmap_atomic(sb_page);
610 chunksize = le32_to_cpu(sb->chunksize);
611 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
612 write_behind = le32_to_cpu(sb->write_behind);
613 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
614 nodes = le32_to_cpu(sb->nodes);
615 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
617 /* verify that the bitmap-specific fields are valid */
618 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
619 reason = "bad magic";
620 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
621 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
622 reason = "unrecognized superblock version";
623 else if (chunksize < 512)
624 reason = "bitmap chunksize too small";
625 else if (!is_power_of_2(chunksize))
626 reason = "bitmap chunksize not a power of 2";
627 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
628 reason = "daemon sleep period out of range";
629 else if (write_behind > COUNTER_MAX)
630 reason = "write-behind limit out of range (0 - 16383)";
631 if (reason) {
632 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
633 bmname(bitmap), reason);
634 goto out;
637 /* keep the array size field of the bitmap superblock up to date */
638 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
640 if (bitmap->mddev->persistent) {
642 * We have a persistent array superblock, so compare the
643 * bitmap's UUID and event counter to the mddev's
645 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
646 printk(KERN_INFO
647 "%s: bitmap superblock UUID mismatch\n",
648 bmname(bitmap));
649 goto out;
651 events = le64_to_cpu(sb->events);
652 if (!nodes && (events < bitmap->mddev->events)) {
653 printk(KERN_INFO
654 "%s: bitmap file is out of date (%llu < %llu) "
655 "-- forcing full recovery\n",
656 bmname(bitmap), events,
657 (unsigned long long) bitmap->mddev->events);
658 set_bit(BITMAP_STALE, &bitmap->flags);
662 /* assign fields using values from superblock */
663 bitmap->flags |= le32_to_cpu(sb->state);
664 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
665 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
666 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
667 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
668 err = 0;
670 out:
671 kunmap_atomic(sb);
672 /* Assiging chunksize is required for "re_read" */
673 bitmap->mddev->bitmap_info.chunksize = chunksize;
674 if (nodes && (bitmap->cluster_slot < 0)) {
675 err = md_setup_cluster(bitmap->mddev, nodes);
676 if (err) {
677 pr_err("%s: Could not setup cluster service (%d)\n",
678 bmname(bitmap), err);
679 goto out_no_sb;
681 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
682 goto re_read;
686 out_no_sb:
687 if (test_bit(BITMAP_STALE, &bitmap->flags))
688 bitmap->events_cleared = bitmap->mddev->events;
689 bitmap->mddev->bitmap_info.chunksize = chunksize;
690 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
691 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
692 bitmap->mddev->bitmap_info.nodes = nodes;
693 if (bitmap->mddev->bitmap_info.space == 0 ||
694 bitmap->mddev->bitmap_info.space > sectors_reserved)
695 bitmap->mddev->bitmap_info.space = sectors_reserved;
696 if (err) {
697 bitmap_print_sb(bitmap);
698 if (bitmap->cluster_slot < 0)
699 md_cluster_stop(bitmap->mddev);
701 return err;
705 * general bitmap file operations
709 * on-disk bitmap:
711 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
712 * file a page at a time. There's a superblock at the start of the file.
714 /* calculate the index of the page that contains this bit */
715 static inline unsigned long file_page_index(struct bitmap_storage *store,
716 unsigned long chunk)
718 if (store->sb_page)
719 chunk += sizeof(bitmap_super_t) << 3;
720 return chunk >> PAGE_BIT_SHIFT;
723 /* calculate the (bit) offset of this bit within a page */
724 static inline unsigned long file_page_offset(struct bitmap_storage *store,
725 unsigned long chunk)
727 if (store->sb_page)
728 chunk += sizeof(bitmap_super_t) << 3;
729 return chunk & (PAGE_BITS - 1);
733 * return a pointer to the page in the filemap that contains the given bit
736 static inline struct page *filemap_get_page(struct bitmap_storage *store,
737 unsigned long chunk)
739 if (file_page_index(store, chunk) >= store->file_pages)
740 return NULL;
741 return store->filemap[file_page_index(store, chunk)];
744 static int bitmap_storage_alloc(struct bitmap_storage *store,
745 unsigned long chunks, int with_super,
746 int slot_number)
748 int pnum, offset = 0;
749 unsigned long num_pages;
750 unsigned long bytes;
752 bytes = DIV_ROUND_UP(chunks, 8);
753 if (with_super)
754 bytes += sizeof(bitmap_super_t);
756 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
757 offset = slot_number * (num_pages - 1);
759 store->filemap = kmalloc(sizeof(struct page *)
760 * num_pages, GFP_KERNEL);
761 if (!store->filemap)
762 return -ENOMEM;
764 if (with_super && !store->sb_page) {
765 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
766 if (store->sb_page == NULL)
767 return -ENOMEM;
770 pnum = 0;
771 if (store->sb_page) {
772 store->filemap[0] = store->sb_page;
773 pnum = 1;
774 store->sb_page->index = offset;
777 for ( ; pnum < num_pages; pnum++) {
778 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
779 if (!store->filemap[pnum]) {
780 store->file_pages = pnum;
781 return -ENOMEM;
783 store->filemap[pnum]->index = pnum + offset;
785 store->file_pages = pnum;
787 /* We need 4 bits per page, rounded up to a multiple
788 * of sizeof(unsigned long) */
789 store->filemap_attr = kzalloc(
790 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
791 GFP_KERNEL);
792 if (!store->filemap_attr)
793 return -ENOMEM;
795 store->bytes = bytes;
797 return 0;
800 static void bitmap_file_unmap(struct bitmap_storage *store)
802 struct page **map, *sb_page;
803 int pages;
804 struct file *file;
806 file = store->file;
807 map = store->filemap;
808 pages = store->file_pages;
809 sb_page = store->sb_page;
811 while (pages--)
812 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
813 free_buffers(map[pages]);
814 kfree(map);
815 kfree(store->filemap_attr);
817 if (sb_page)
818 free_buffers(sb_page);
820 if (file) {
821 struct inode *inode = file_inode(file);
822 invalidate_mapping_pages(inode->i_mapping, 0, -1);
823 fput(file);
828 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
829 * then it is no longer reliable, so we stop using it and we mark the file
830 * as failed in the superblock
832 static void bitmap_file_kick(struct bitmap *bitmap)
834 char *path, *ptr = NULL;
836 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
837 bitmap_update_sb(bitmap);
839 if (bitmap->storage.file) {
840 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
841 if (path)
842 ptr = d_path(&bitmap->storage.file->f_path,
843 path, PAGE_SIZE);
845 printk(KERN_ALERT
846 "%s: kicking failed bitmap file %s from array!\n",
847 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
849 kfree(path);
850 } else
851 printk(KERN_ALERT
852 "%s: disabling internal bitmap due to errors\n",
853 bmname(bitmap));
857 enum bitmap_page_attr {
858 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
859 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
860 * i.e. counter is 1 or 2. */
861 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
864 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
865 enum bitmap_page_attr attr)
867 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
870 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
871 enum bitmap_page_attr attr)
873 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
876 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
877 enum bitmap_page_attr attr)
879 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
882 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
883 enum bitmap_page_attr attr)
885 return test_and_clear_bit((pnum<<2) + attr,
886 bitmap->storage.filemap_attr);
889 * bitmap_file_set_bit -- called before performing a write to the md device
890 * to set (and eventually sync) a particular bit in the bitmap file
892 * we set the bit immediately, then we record the page number so that
893 * when an unplug occurs, we can flush the dirty pages out to disk
895 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
897 unsigned long bit;
898 struct page *page;
899 void *kaddr;
900 unsigned long chunk = block >> bitmap->counts.chunkshift;
902 page = filemap_get_page(&bitmap->storage, chunk);
903 if (!page)
904 return;
905 bit = file_page_offset(&bitmap->storage, chunk);
907 /* set the bit */
908 kaddr = kmap_atomic(page);
909 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
910 set_bit(bit, kaddr);
911 else
912 set_bit_le(bit, kaddr);
913 kunmap_atomic(kaddr);
914 pr_debug("set file bit %lu page %lu\n", bit, page->index);
915 /* record page number so it gets flushed to disk when unplug occurs */
916 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
919 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
921 unsigned long bit;
922 struct page *page;
923 void *paddr;
924 unsigned long chunk = block >> bitmap->counts.chunkshift;
926 page = filemap_get_page(&bitmap->storage, chunk);
927 if (!page)
928 return;
929 bit = file_page_offset(&bitmap->storage, chunk);
930 paddr = kmap_atomic(page);
931 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
932 clear_bit(bit, paddr);
933 else
934 clear_bit_le(bit, paddr);
935 kunmap_atomic(paddr);
936 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
937 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
938 bitmap->allclean = 0;
942 static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
944 unsigned long bit;
945 struct page *page;
946 void *paddr;
947 unsigned long chunk = block >> bitmap->counts.chunkshift;
948 int set = 0;
950 page = filemap_get_page(&bitmap->storage, chunk);
951 if (!page)
952 return -EINVAL;
953 bit = file_page_offset(&bitmap->storage, chunk);
954 paddr = kmap_atomic(page);
955 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
956 set = test_bit(bit, paddr);
957 else
958 set = test_bit_le(bit, paddr);
959 kunmap_atomic(paddr);
960 return set;
964 /* this gets called when the md device is ready to unplug its underlying
965 * (slave) device queues -- before we let any writes go down, we need to
966 * sync the dirty pages of the bitmap file to disk */
967 void bitmap_unplug(struct bitmap *bitmap)
969 unsigned long i;
970 int dirty, need_write;
972 if (!bitmap || !bitmap->storage.filemap ||
973 test_bit(BITMAP_STALE, &bitmap->flags))
974 return;
976 /* look at each page to see if there are any set bits that need to be
977 * flushed out to disk */
978 for (i = 0; i < bitmap->storage.file_pages; i++) {
979 if (!bitmap->storage.filemap)
980 return;
981 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
982 need_write = test_and_clear_page_attr(bitmap, i,
983 BITMAP_PAGE_NEEDWRITE);
984 if (dirty || need_write) {
985 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
986 write_page(bitmap, bitmap->storage.filemap[i], 0);
989 if (bitmap->storage.file)
990 wait_event(bitmap->write_wait,
991 atomic_read(&bitmap->pending_writes)==0);
992 else
993 md_super_wait(bitmap->mddev);
995 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
996 bitmap_file_kick(bitmap);
998 EXPORT_SYMBOL(bitmap_unplug);
1000 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1001 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1002 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1003 * memory mapping of the bitmap file
1004 * Special cases:
1005 * if there's no bitmap file, or if the bitmap file had been
1006 * previously kicked from the array, we mark all the bits as
1007 * 1's in order to cause a full resync.
1009 * We ignore all bits for sectors that end earlier than 'start'.
1010 * This is used when reading an out-of-date bitmap...
1012 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1014 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1015 struct page *page = NULL;
1016 unsigned long bit_cnt = 0;
1017 struct file *file;
1018 unsigned long offset;
1019 int outofdate;
1020 int ret = -ENOSPC;
1021 void *paddr;
1022 struct bitmap_storage *store = &bitmap->storage;
1024 chunks = bitmap->counts.chunks;
1025 file = store->file;
1027 if (!file && !bitmap->mddev->bitmap_info.offset) {
1028 /* No permanent bitmap - fill with '1s'. */
1029 store->filemap = NULL;
1030 store->file_pages = 0;
1031 for (i = 0; i < chunks ; i++) {
1032 /* if the disk bit is set, set the memory bit */
1033 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1034 >= start);
1035 bitmap_set_memory_bits(bitmap,
1036 (sector_t)i << bitmap->counts.chunkshift,
1037 needed);
1039 return 0;
1042 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1043 if (outofdate)
1044 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1045 "recovery\n", bmname(bitmap));
1047 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1048 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
1049 bmname(bitmap),
1050 (unsigned long) i_size_read(file->f_mapping->host),
1051 store->bytes);
1052 goto err;
1055 oldindex = ~0L;
1056 offset = 0;
1057 if (!bitmap->mddev->bitmap_info.external)
1058 offset = sizeof(bitmap_super_t);
1060 if (mddev_is_clustered(bitmap->mddev))
1061 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1063 for (i = 0; i < chunks; i++) {
1064 int b;
1065 index = file_page_index(&bitmap->storage, i);
1066 bit = file_page_offset(&bitmap->storage, i);
1067 if (index != oldindex) { /* this is a new page, read it in */
1068 int count;
1069 /* unmap the old page, we're done with it */
1070 if (index == store->file_pages-1)
1071 count = store->bytes - index * PAGE_SIZE;
1072 else
1073 count = PAGE_SIZE;
1074 page = store->filemap[index];
1075 if (file)
1076 ret = read_page(file, index, bitmap,
1077 count, page);
1078 else
1079 ret = read_sb_page(
1080 bitmap->mddev,
1081 bitmap->mddev->bitmap_info.offset,
1082 page,
1083 index + node_offset, count);
1085 if (ret)
1086 goto err;
1088 oldindex = index;
1090 if (outofdate) {
1092 * if bitmap is out of date, dirty the
1093 * whole page and write it out
1095 paddr = kmap_atomic(page);
1096 memset(paddr + offset, 0xff,
1097 PAGE_SIZE - offset);
1098 kunmap_atomic(paddr);
1099 write_page(bitmap, page, 1);
1101 ret = -EIO;
1102 if (test_bit(BITMAP_WRITE_ERROR,
1103 &bitmap->flags))
1104 goto err;
1107 paddr = kmap_atomic(page);
1108 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1109 b = test_bit(bit, paddr);
1110 else
1111 b = test_bit_le(bit, paddr);
1112 kunmap_atomic(paddr);
1113 if (b) {
1114 /* if the disk bit is set, set the memory bit */
1115 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1116 >= start);
1117 bitmap_set_memory_bits(bitmap,
1118 (sector_t)i << bitmap->counts.chunkshift,
1119 needed);
1120 bit_cnt++;
1122 offset = 0;
1125 printk(KERN_INFO "%s: bitmap initialized from disk: "
1126 "read %lu pages, set %lu of %lu bits\n",
1127 bmname(bitmap), store->file_pages,
1128 bit_cnt, chunks);
1130 return 0;
1132 err:
1133 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1134 bmname(bitmap), ret);
1135 return ret;
1138 void bitmap_write_all(struct bitmap *bitmap)
1140 /* We don't actually write all bitmap blocks here,
1141 * just flag them as needing to be written
1143 int i;
1145 if (!bitmap || !bitmap->storage.filemap)
1146 return;
1147 if (bitmap->storage.file)
1148 /* Only one copy, so nothing needed */
1149 return;
1151 for (i = 0; i < bitmap->storage.file_pages; i++)
1152 set_page_attr(bitmap, i,
1153 BITMAP_PAGE_NEEDWRITE);
1154 bitmap->allclean = 0;
1157 static void bitmap_count_page(struct bitmap_counts *bitmap,
1158 sector_t offset, int inc)
1160 sector_t chunk = offset >> bitmap->chunkshift;
1161 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1162 bitmap->bp[page].count += inc;
1163 bitmap_checkfree(bitmap, page);
1166 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1168 sector_t chunk = offset >> bitmap->chunkshift;
1169 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1170 struct bitmap_page *bp = &bitmap->bp[page];
1172 if (!bp->pending)
1173 bp->pending = 1;
1176 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1177 sector_t offset, sector_t *blocks,
1178 int create);
1181 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1182 * out to disk
1185 void bitmap_daemon_work(struct mddev *mddev)
1187 struct bitmap *bitmap;
1188 unsigned long j;
1189 unsigned long nextpage;
1190 sector_t blocks;
1191 struct bitmap_counts *counts;
1193 /* Use a mutex to guard daemon_work against
1194 * bitmap_destroy.
1196 mutex_lock(&mddev->bitmap_info.mutex);
1197 bitmap = mddev->bitmap;
1198 if (bitmap == NULL) {
1199 mutex_unlock(&mddev->bitmap_info.mutex);
1200 return;
1202 if (time_before(jiffies, bitmap->daemon_lastrun
1203 + mddev->bitmap_info.daemon_sleep))
1204 goto done;
1206 bitmap->daemon_lastrun = jiffies;
1207 if (bitmap->allclean) {
1208 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1209 goto done;
1211 bitmap->allclean = 1;
1213 /* Any file-page which is PENDING now needs to be written.
1214 * So set NEEDWRITE now, then after we make any last-minute changes
1215 * we will write it.
1217 for (j = 0; j < bitmap->storage.file_pages; j++)
1218 if (test_and_clear_page_attr(bitmap, j,
1219 BITMAP_PAGE_PENDING))
1220 set_page_attr(bitmap, j,
1221 BITMAP_PAGE_NEEDWRITE);
1223 if (bitmap->need_sync &&
1224 mddev->bitmap_info.external == 0) {
1225 /* Arrange for superblock update as well as
1226 * other changes */
1227 bitmap_super_t *sb;
1228 bitmap->need_sync = 0;
1229 if (bitmap->storage.filemap) {
1230 sb = kmap_atomic(bitmap->storage.sb_page);
1231 sb->events_cleared =
1232 cpu_to_le64(bitmap->events_cleared);
1233 kunmap_atomic(sb);
1234 set_page_attr(bitmap, 0,
1235 BITMAP_PAGE_NEEDWRITE);
1238 /* Now look at the bitmap counters and if any are '2' or '1',
1239 * decrement and handle accordingly.
1241 counts = &bitmap->counts;
1242 spin_lock_irq(&counts->lock);
1243 nextpage = 0;
1244 for (j = 0; j < counts->chunks; j++) {
1245 bitmap_counter_t *bmc;
1246 sector_t block = (sector_t)j << counts->chunkshift;
1248 if (j == nextpage) {
1249 nextpage += PAGE_COUNTER_RATIO;
1250 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1251 j |= PAGE_COUNTER_MASK;
1252 continue;
1254 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1256 bmc = bitmap_get_counter(counts,
1257 block,
1258 &blocks, 0);
1260 if (!bmc) {
1261 j |= PAGE_COUNTER_MASK;
1262 continue;
1264 if (*bmc == 1 && !bitmap->need_sync) {
1265 /* We can clear the bit */
1266 *bmc = 0;
1267 bitmap_count_page(counts, block, -1);
1268 bitmap_file_clear_bit(bitmap, block);
1269 } else if (*bmc && *bmc <= 2) {
1270 *bmc = 1;
1271 bitmap_set_pending(counts, block);
1272 bitmap->allclean = 0;
1275 spin_unlock_irq(&counts->lock);
1277 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1278 * DIRTY pages need to be written by bitmap_unplug so it can wait
1279 * for them.
1280 * If we find any DIRTY page we stop there and let bitmap_unplug
1281 * handle all the rest. This is important in the case where
1282 * the first blocking holds the superblock and it has been updated.
1283 * We mustn't write any other blocks before the superblock.
1285 for (j = 0;
1286 j < bitmap->storage.file_pages
1287 && !test_bit(BITMAP_STALE, &bitmap->flags);
1288 j++) {
1289 if (test_page_attr(bitmap, j,
1290 BITMAP_PAGE_DIRTY))
1291 /* bitmap_unplug will handle the rest */
1292 break;
1293 if (test_and_clear_page_attr(bitmap, j,
1294 BITMAP_PAGE_NEEDWRITE)) {
1295 write_page(bitmap, bitmap->storage.filemap[j], 0);
1299 done:
1300 if (bitmap->allclean == 0)
1301 mddev->thread->timeout =
1302 mddev->bitmap_info.daemon_sleep;
1303 mutex_unlock(&mddev->bitmap_info.mutex);
1306 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1307 sector_t offset, sector_t *blocks,
1308 int create)
1309 __releases(bitmap->lock)
1310 __acquires(bitmap->lock)
1312 /* If 'create', we might release the lock and reclaim it.
1313 * The lock must have been taken with interrupts enabled.
1314 * If !create, we don't release the lock.
1316 sector_t chunk = offset >> bitmap->chunkshift;
1317 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1318 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1319 sector_t csize;
1320 int err;
1322 err = bitmap_checkpage(bitmap, page, create);
1324 if (bitmap->bp[page].hijacked ||
1325 bitmap->bp[page].map == NULL)
1326 csize = ((sector_t)1) << (bitmap->chunkshift +
1327 PAGE_COUNTER_SHIFT - 1);
1328 else
1329 csize = ((sector_t)1) << bitmap->chunkshift;
1330 *blocks = csize - (offset & (csize - 1));
1332 if (err < 0)
1333 return NULL;
1335 /* now locked ... */
1337 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1338 /* should we use the first or second counter field
1339 * of the hijacked pointer? */
1340 int hi = (pageoff > PAGE_COUNTER_MASK);
1341 return &((bitmap_counter_t *)
1342 &bitmap->bp[page].map)[hi];
1343 } else /* page is allocated */
1344 return (bitmap_counter_t *)
1345 &(bitmap->bp[page].map[pageoff]);
1348 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1350 if (!bitmap)
1351 return 0;
1353 if (behind) {
1354 int bw;
1355 atomic_inc(&bitmap->behind_writes);
1356 bw = atomic_read(&bitmap->behind_writes);
1357 if (bw > bitmap->behind_writes_used)
1358 bitmap->behind_writes_used = bw;
1360 pr_debug("inc write-behind count %d/%lu\n",
1361 bw, bitmap->mddev->bitmap_info.max_write_behind);
1364 while (sectors) {
1365 sector_t blocks;
1366 bitmap_counter_t *bmc;
1368 spin_lock_irq(&bitmap->counts.lock);
1369 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1370 if (!bmc) {
1371 spin_unlock_irq(&bitmap->counts.lock);
1372 return 0;
1375 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1376 DEFINE_WAIT(__wait);
1377 /* note that it is safe to do the prepare_to_wait
1378 * after the test as long as we do it before dropping
1379 * the spinlock.
1381 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1382 TASK_UNINTERRUPTIBLE);
1383 spin_unlock_irq(&bitmap->counts.lock);
1384 schedule();
1385 finish_wait(&bitmap->overflow_wait, &__wait);
1386 continue;
1389 switch (*bmc) {
1390 case 0:
1391 bitmap_file_set_bit(bitmap, offset);
1392 bitmap_count_page(&bitmap->counts, offset, 1);
1393 /* fall through */
1394 case 1:
1395 *bmc = 2;
1398 (*bmc)++;
1400 spin_unlock_irq(&bitmap->counts.lock);
1402 offset += blocks;
1403 if (sectors > blocks)
1404 sectors -= blocks;
1405 else
1406 sectors = 0;
1408 return 0;
1410 EXPORT_SYMBOL(bitmap_startwrite);
1412 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1413 int success, int behind)
1415 if (!bitmap)
1416 return;
1417 if (behind) {
1418 if (atomic_dec_and_test(&bitmap->behind_writes))
1419 wake_up(&bitmap->behind_wait);
1420 pr_debug("dec write-behind count %d/%lu\n",
1421 atomic_read(&bitmap->behind_writes),
1422 bitmap->mddev->bitmap_info.max_write_behind);
1425 while (sectors) {
1426 sector_t blocks;
1427 unsigned long flags;
1428 bitmap_counter_t *bmc;
1430 spin_lock_irqsave(&bitmap->counts.lock, flags);
1431 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1432 if (!bmc) {
1433 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1434 return;
1437 if (success && !bitmap->mddev->degraded &&
1438 bitmap->events_cleared < bitmap->mddev->events) {
1439 bitmap->events_cleared = bitmap->mddev->events;
1440 bitmap->need_sync = 1;
1441 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1444 if (!success && !NEEDED(*bmc))
1445 *bmc |= NEEDED_MASK;
1447 if (COUNTER(*bmc) == COUNTER_MAX)
1448 wake_up(&bitmap->overflow_wait);
1450 (*bmc)--;
1451 if (*bmc <= 2) {
1452 bitmap_set_pending(&bitmap->counts, offset);
1453 bitmap->allclean = 0;
1455 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1456 offset += blocks;
1457 if (sectors > blocks)
1458 sectors -= blocks;
1459 else
1460 sectors = 0;
1463 EXPORT_SYMBOL(bitmap_endwrite);
1465 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1466 int degraded)
1468 bitmap_counter_t *bmc;
1469 int rv;
1470 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1471 *blocks = 1024;
1472 return 1; /* always resync if no bitmap */
1474 spin_lock_irq(&bitmap->counts.lock);
1475 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1476 rv = 0;
1477 if (bmc) {
1478 /* locked */
1479 if (RESYNC(*bmc))
1480 rv = 1;
1481 else if (NEEDED(*bmc)) {
1482 rv = 1;
1483 if (!degraded) { /* don't set/clear bits if degraded */
1484 *bmc |= RESYNC_MASK;
1485 *bmc &= ~NEEDED_MASK;
1489 spin_unlock_irq(&bitmap->counts.lock);
1490 return rv;
1493 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1494 int degraded)
1496 /* bitmap_start_sync must always report on multiples of whole
1497 * pages, otherwise resync (which is very PAGE_SIZE based) will
1498 * get confused.
1499 * So call __bitmap_start_sync repeatedly (if needed) until
1500 * At least PAGE_SIZE>>9 blocks are covered.
1501 * Return the 'or' of the result.
1503 int rv = 0;
1504 sector_t blocks1;
1506 *blocks = 0;
1507 while (*blocks < (PAGE_SIZE>>9)) {
1508 rv |= __bitmap_start_sync(bitmap, offset,
1509 &blocks1, degraded);
1510 offset += blocks1;
1511 *blocks += blocks1;
1513 return rv;
1515 EXPORT_SYMBOL(bitmap_start_sync);
1517 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1519 bitmap_counter_t *bmc;
1520 unsigned long flags;
1522 if (bitmap == NULL) {
1523 *blocks = 1024;
1524 return;
1526 spin_lock_irqsave(&bitmap->counts.lock, flags);
1527 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1528 if (bmc == NULL)
1529 goto unlock;
1530 /* locked */
1531 if (RESYNC(*bmc)) {
1532 *bmc &= ~RESYNC_MASK;
1534 if (!NEEDED(*bmc) && aborted)
1535 *bmc |= NEEDED_MASK;
1536 else {
1537 if (*bmc <= 2) {
1538 bitmap_set_pending(&bitmap->counts, offset);
1539 bitmap->allclean = 0;
1543 unlock:
1544 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1546 EXPORT_SYMBOL(bitmap_end_sync);
1548 void bitmap_close_sync(struct bitmap *bitmap)
1550 /* Sync has finished, and any bitmap chunks that weren't synced
1551 * properly have been aborted. It remains to us to clear the
1552 * RESYNC bit wherever it is still on
1554 sector_t sector = 0;
1555 sector_t blocks;
1556 if (!bitmap)
1557 return;
1558 while (sector < bitmap->mddev->resync_max_sectors) {
1559 bitmap_end_sync(bitmap, sector, &blocks, 0);
1560 sector += blocks;
1563 EXPORT_SYMBOL(bitmap_close_sync);
1565 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1567 sector_t s = 0;
1568 sector_t blocks;
1570 if (!bitmap)
1571 return;
1572 if (sector == 0) {
1573 bitmap->last_end_sync = jiffies;
1574 return;
1576 if (time_before(jiffies, (bitmap->last_end_sync
1577 + bitmap->mddev->bitmap_info.daemon_sleep)))
1578 return;
1579 wait_event(bitmap->mddev->recovery_wait,
1580 atomic_read(&bitmap->mddev->recovery_active) == 0);
1582 bitmap->mddev->curr_resync_completed = sector;
1583 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1584 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1585 s = 0;
1586 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1587 bitmap_end_sync(bitmap, s, &blocks, 0);
1588 s += blocks;
1590 bitmap->last_end_sync = jiffies;
1591 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1593 EXPORT_SYMBOL(bitmap_cond_end_sync);
1595 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1597 /* For each chunk covered by any of these sectors, set the
1598 * counter to 2 and possibly set resync_needed. They should all
1599 * be 0 at this point
1602 sector_t secs;
1603 bitmap_counter_t *bmc;
1604 spin_lock_irq(&bitmap->counts.lock);
1605 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1606 if (!bmc) {
1607 spin_unlock_irq(&bitmap->counts.lock);
1608 return;
1610 if (!*bmc) {
1611 *bmc = 2;
1612 bitmap_count_page(&bitmap->counts, offset, 1);
1613 bitmap_set_pending(&bitmap->counts, offset);
1614 bitmap->allclean = 0;
1616 if (needed)
1617 *bmc |= NEEDED_MASK;
1618 spin_unlock_irq(&bitmap->counts.lock);
1621 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1622 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1624 unsigned long chunk;
1626 for (chunk = s; chunk <= e; chunk++) {
1627 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1628 bitmap_set_memory_bits(bitmap, sec, 1);
1629 bitmap_file_set_bit(bitmap, sec);
1630 if (sec < bitmap->mddev->recovery_cp)
1631 /* We are asserting that the array is dirty,
1632 * so move the recovery_cp address back so
1633 * that it is obvious that it is dirty
1635 bitmap->mddev->recovery_cp = sec;
1640 * flush out any pending updates
1642 void bitmap_flush(struct mddev *mddev)
1644 struct bitmap *bitmap = mddev->bitmap;
1645 long sleep;
1647 if (!bitmap) /* there was no bitmap */
1648 return;
1650 /* run the daemon_work three time to ensure everything is flushed
1651 * that can be
1653 sleep = mddev->bitmap_info.daemon_sleep * 2;
1654 bitmap->daemon_lastrun -= sleep;
1655 bitmap_daemon_work(mddev);
1656 bitmap->daemon_lastrun -= sleep;
1657 bitmap_daemon_work(mddev);
1658 bitmap->daemon_lastrun -= sleep;
1659 bitmap_daemon_work(mddev);
1660 bitmap_update_sb(bitmap);
1664 * free memory that was allocated
1666 static void bitmap_free(struct bitmap *bitmap)
1668 unsigned long k, pages;
1669 struct bitmap_page *bp;
1671 if (!bitmap) /* there was no bitmap */
1672 return;
1674 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1675 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1676 md_cluster_stop(bitmap->mddev);
1678 /* Shouldn't be needed - but just in case.... */
1679 wait_event(bitmap->write_wait,
1680 atomic_read(&bitmap->pending_writes) == 0);
1682 /* release the bitmap file */
1683 bitmap_file_unmap(&bitmap->storage);
1685 bp = bitmap->counts.bp;
1686 pages = bitmap->counts.pages;
1688 /* free all allocated memory */
1690 if (bp) /* deallocate the page memory */
1691 for (k = 0; k < pages; k++)
1692 if (bp[k].map && !bp[k].hijacked)
1693 kfree(bp[k].map);
1694 kfree(bp);
1695 kfree(bitmap);
1698 void bitmap_destroy(struct mddev *mddev)
1700 struct bitmap *bitmap = mddev->bitmap;
1702 if (!bitmap) /* there was no bitmap */
1703 return;
1705 mutex_lock(&mddev->bitmap_info.mutex);
1706 spin_lock(&mddev->lock);
1707 mddev->bitmap = NULL; /* disconnect from the md device */
1708 spin_unlock(&mddev->lock);
1709 mutex_unlock(&mddev->bitmap_info.mutex);
1710 if (mddev->thread)
1711 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1713 if (bitmap->sysfs_can_clear)
1714 sysfs_put(bitmap->sysfs_can_clear);
1716 bitmap_free(bitmap);
1720 * initialize the bitmap structure
1721 * if this returns an error, bitmap_destroy must be called to do clean up
1723 struct bitmap *bitmap_create(struct mddev *mddev, int slot)
1725 struct bitmap *bitmap;
1726 sector_t blocks = mddev->resync_max_sectors;
1727 struct file *file = mddev->bitmap_info.file;
1728 int err;
1729 struct kernfs_node *bm = NULL;
1731 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1733 BUG_ON(file && mddev->bitmap_info.offset);
1735 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1736 if (!bitmap)
1737 return ERR_PTR(-ENOMEM);
1739 spin_lock_init(&bitmap->counts.lock);
1740 atomic_set(&bitmap->pending_writes, 0);
1741 init_waitqueue_head(&bitmap->write_wait);
1742 init_waitqueue_head(&bitmap->overflow_wait);
1743 init_waitqueue_head(&bitmap->behind_wait);
1745 bitmap->mddev = mddev;
1746 bitmap->cluster_slot = slot;
1748 if (mddev->kobj.sd)
1749 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1750 if (bm) {
1751 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1752 sysfs_put(bm);
1753 } else
1754 bitmap->sysfs_can_clear = NULL;
1756 bitmap->storage.file = file;
1757 if (file) {
1758 get_file(file);
1759 /* As future accesses to this file will use bmap,
1760 * and bypass the page cache, we must sync the file
1761 * first.
1763 vfs_fsync(file, 1);
1765 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1766 if (!mddev->bitmap_info.external) {
1768 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1769 * instructing us to create a new on-disk bitmap instance.
1771 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1772 err = bitmap_new_disk_sb(bitmap);
1773 else
1774 err = bitmap_read_sb(bitmap);
1775 } else {
1776 err = 0;
1777 if (mddev->bitmap_info.chunksize == 0 ||
1778 mddev->bitmap_info.daemon_sleep == 0)
1779 /* chunksize and time_base need to be
1780 * set first. */
1781 err = -EINVAL;
1783 if (err)
1784 goto error;
1786 bitmap->daemon_lastrun = jiffies;
1787 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1788 if (err)
1789 goto error;
1791 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1792 bitmap->counts.pages, bmname(bitmap));
1794 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1795 if (err)
1796 goto error;
1798 return bitmap;
1799 error:
1800 bitmap_free(bitmap);
1801 return ERR_PTR(err);
1804 int bitmap_load(struct mddev *mddev)
1806 int err = 0;
1807 sector_t start = 0;
1808 sector_t sector = 0;
1809 struct bitmap *bitmap = mddev->bitmap;
1811 if (!bitmap)
1812 goto out;
1814 /* Clear out old bitmap info first: Either there is none, or we
1815 * are resuming after someone else has possibly changed things,
1816 * so we should forget old cached info.
1817 * All chunks should be clean, but some might need_sync.
1819 while (sector < mddev->resync_max_sectors) {
1820 sector_t blocks;
1821 bitmap_start_sync(bitmap, sector, &blocks, 0);
1822 sector += blocks;
1824 bitmap_close_sync(bitmap);
1826 if (mddev->degraded == 0
1827 || bitmap->events_cleared == mddev->events)
1828 /* no need to keep dirty bits to optimise a
1829 * re-add of a missing device */
1830 start = mddev->recovery_cp;
1832 mutex_lock(&mddev->bitmap_info.mutex);
1833 err = bitmap_init_from_disk(bitmap, start);
1834 mutex_unlock(&mddev->bitmap_info.mutex);
1836 if (err)
1837 goto out;
1838 clear_bit(BITMAP_STALE, &bitmap->flags);
1840 /* Kick recovery in case any bits were set */
1841 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1843 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1844 md_wakeup_thread(mddev->thread);
1846 bitmap_update_sb(bitmap);
1848 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1849 err = -EIO;
1850 out:
1851 return err;
1853 EXPORT_SYMBOL_GPL(bitmap_load);
1855 /* Loads the bitmap associated with slot and copies the resync information
1856 * to our bitmap
1858 int bitmap_copy_from_slot(struct mddev *mddev, int slot,
1859 sector_t *low, sector_t *high, bool clear_bits)
1861 int rv = 0, i, j;
1862 sector_t block, lo = 0, hi = 0;
1863 struct bitmap_counts *counts;
1864 struct bitmap *bitmap = bitmap_create(mddev, slot);
1866 if (IS_ERR(bitmap))
1867 return PTR_ERR(bitmap);
1869 rv = bitmap_read_sb(bitmap);
1870 if (rv)
1871 goto err;
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_path(seq, &bitmap->storage.file->f_path, " \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 bitmap->cluster_slot);
1995 if (ret)
1996 goto err;
1998 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2000 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
2001 ret = -ENOMEM;
2002 if (!new_bp) {
2003 bitmap_file_unmap(&store);
2004 goto err;
2007 if (!init)
2008 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2010 store.file = bitmap->storage.file;
2011 bitmap->storage.file = NULL;
2013 if (store.sb_page && bitmap->storage.sb_page)
2014 memcpy(page_address(store.sb_page),
2015 page_address(bitmap->storage.sb_page),
2016 sizeof(bitmap_super_t));
2017 bitmap_file_unmap(&bitmap->storage);
2018 bitmap->storage = store;
2020 old_counts = bitmap->counts;
2021 bitmap->counts.bp = new_bp;
2022 bitmap->counts.pages = pages;
2023 bitmap->counts.missing_pages = pages;
2024 bitmap->counts.chunkshift = chunkshift;
2025 bitmap->counts.chunks = chunks;
2026 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2027 BITMAP_BLOCK_SHIFT);
2029 blocks = min(old_counts.chunks << old_counts.chunkshift,
2030 chunks << chunkshift);
2032 spin_lock_irq(&bitmap->counts.lock);
2033 for (block = 0; block < blocks; ) {
2034 bitmap_counter_t *bmc_old, *bmc_new;
2035 int set;
2037 bmc_old = bitmap_get_counter(&old_counts, block,
2038 &old_blocks, 0);
2039 set = bmc_old && NEEDED(*bmc_old);
2041 if (set) {
2042 bmc_new = bitmap_get_counter(&bitmap->counts, block,
2043 &new_blocks, 1);
2044 if (*bmc_new == 0) {
2045 /* need to set on-disk bits too. */
2046 sector_t end = block + new_blocks;
2047 sector_t start = block >> chunkshift;
2048 start <<= chunkshift;
2049 while (start < end) {
2050 bitmap_file_set_bit(bitmap, block);
2051 start += 1 << chunkshift;
2053 *bmc_new = 2;
2054 bitmap_count_page(&bitmap->counts,
2055 block, 1);
2056 bitmap_set_pending(&bitmap->counts,
2057 block);
2059 *bmc_new |= NEEDED_MASK;
2060 if (new_blocks < old_blocks)
2061 old_blocks = new_blocks;
2063 block += old_blocks;
2066 if (!init) {
2067 int i;
2068 while (block < (chunks << chunkshift)) {
2069 bitmap_counter_t *bmc;
2070 bmc = bitmap_get_counter(&bitmap->counts, block,
2071 &new_blocks, 1);
2072 if (bmc) {
2073 /* new space. It needs to be resynced, so
2074 * we set NEEDED_MASK.
2076 if (*bmc == 0) {
2077 *bmc = NEEDED_MASK | 2;
2078 bitmap_count_page(&bitmap->counts,
2079 block, 1);
2080 bitmap_set_pending(&bitmap->counts,
2081 block);
2084 block += new_blocks;
2086 for (i = 0; i < bitmap->storage.file_pages; i++)
2087 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2089 spin_unlock_irq(&bitmap->counts.lock);
2091 if (!init) {
2092 bitmap_unplug(bitmap);
2093 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2095 ret = 0;
2096 err:
2097 return ret;
2099 EXPORT_SYMBOL_GPL(bitmap_resize);
2101 static ssize_t
2102 location_show(struct mddev *mddev, char *page)
2104 ssize_t len;
2105 if (mddev->bitmap_info.file)
2106 len = sprintf(page, "file");
2107 else if (mddev->bitmap_info.offset)
2108 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2109 else
2110 len = sprintf(page, "none");
2111 len += sprintf(page+len, "\n");
2112 return len;
2115 static ssize_t
2116 location_store(struct mddev *mddev, const char *buf, size_t len)
2119 if (mddev->pers) {
2120 if (!mddev->pers->quiesce)
2121 return -EBUSY;
2122 if (mddev->recovery || mddev->sync_thread)
2123 return -EBUSY;
2126 if (mddev->bitmap || mddev->bitmap_info.file ||
2127 mddev->bitmap_info.offset) {
2128 /* bitmap already configured. Only option is to clear it */
2129 if (strncmp(buf, "none", 4) != 0)
2130 return -EBUSY;
2131 if (mddev->pers) {
2132 mddev->pers->quiesce(mddev, 1);
2133 bitmap_destroy(mddev);
2134 mddev->pers->quiesce(mddev, 0);
2136 mddev->bitmap_info.offset = 0;
2137 if (mddev->bitmap_info.file) {
2138 struct file *f = mddev->bitmap_info.file;
2139 mddev->bitmap_info.file = NULL;
2140 fput(f);
2142 } else {
2143 /* No bitmap, OK to set a location */
2144 long long offset;
2145 if (strncmp(buf, "none", 4) == 0)
2146 /* nothing to be done */;
2147 else if (strncmp(buf, "file:", 5) == 0) {
2148 /* Not supported yet */
2149 return -EINVAL;
2150 } else {
2151 int rv;
2152 if (buf[0] == '+')
2153 rv = kstrtoll(buf+1, 10, &offset);
2154 else
2155 rv = kstrtoll(buf, 10, &offset);
2156 if (rv)
2157 return rv;
2158 if (offset == 0)
2159 return -EINVAL;
2160 if (mddev->bitmap_info.external == 0 &&
2161 mddev->major_version == 0 &&
2162 offset != mddev->bitmap_info.default_offset)
2163 return -EINVAL;
2164 mddev->bitmap_info.offset = offset;
2165 if (mddev->pers) {
2166 struct bitmap *bitmap;
2167 mddev->pers->quiesce(mddev, 1);
2168 bitmap = bitmap_create(mddev, -1);
2169 if (IS_ERR(bitmap))
2170 rv = PTR_ERR(bitmap);
2171 else {
2172 mddev->bitmap = bitmap;
2173 rv = bitmap_load(mddev);
2174 if (rv) {
2175 bitmap_destroy(mddev);
2176 mddev->bitmap_info.offset = 0;
2179 mddev->pers->quiesce(mddev, 0);
2180 if (rv)
2181 return rv;
2185 if (!mddev->external) {
2186 /* Ensure new bitmap info is stored in
2187 * metadata promptly.
2189 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2190 md_wakeup_thread(mddev->thread);
2192 return len;
2195 static struct md_sysfs_entry bitmap_location =
2196 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2198 /* 'bitmap/space' is the space available at 'location' for the
2199 * bitmap. This allows the kernel to know when it is safe to
2200 * resize the bitmap to match a resized array.
2202 static ssize_t
2203 space_show(struct mddev *mddev, char *page)
2205 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2208 static ssize_t
2209 space_store(struct mddev *mddev, const char *buf, size_t len)
2211 unsigned long sectors;
2212 int rv;
2214 rv = kstrtoul(buf, 10, &sectors);
2215 if (rv)
2216 return rv;
2218 if (sectors == 0)
2219 return -EINVAL;
2221 if (mddev->bitmap &&
2222 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2223 return -EFBIG; /* Bitmap is too big for this small space */
2225 /* could make sure it isn't too big, but that isn't really
2226 * needed - user-space should be careful.
2228 mddev->bitmap_info.space = sectors;
2229 return len;
2232 static struct md_sysfs_entry bitmap_space =
2233 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2235 static ssize_t
2236 timeout_show(struct mddev *mddev, char *page)
2238 ssize_t len;
2239 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2240 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2242 len = sprintf(page, "%lu", secs);
2243 if (jifs)
2244 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2245 len += sprintf(page+len, "\n");
2246 return len;
2249 static ssize_t
2250 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2252 /* timeout can be set at any time */
2253 unsigned long timeout;
2254 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2255 if (rv)
2256 return rv;
2258 /* just to make sure we don't overflow... */
2259 if (timeout >= LONG_MAX / HZ)
2260 return -EINVAL;
2262 timeout = timeout * HZ / 10000;
2264 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2265 timeout = MAX_SCHEDULE_TIMEOUT-1;
2266 if (timeout < 1)
2267 timeout = 1;
2268 mddev->bitmap_info.daemon_sleep = timeout;
2269 if (mddev->thread) {
2270 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2271 * the bitmap is all clean and we don't need to
2272 * adjust the timeout right now
2274 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2275 mddev->thread->timeout = timeout;
2276 md_wakeup_thread(mddev->thread);
2279 return len;
2282 static struct md_sysfs_entry bitmap_timeout =
2283 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2285 static ssize_t
2286 backlog_show(struct mddev *mddev, char *page)
2288 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2291 static ssize_t
2292 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2294 unsigned long backlog;
2295 int rv = kstrtoul(buf, 10, &backlog);
2296 if (rv)
2297 return rv;
2298 if (backlog > COUNTER_MAX)
2299 return -EINVAL;
2300 mddev->bitmap_info.max_write_behind = backlog;
2301 return len;
2304 static struct md_sysfs_entry bitmap_backlog =
2305 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2307 static ssize_t
2308 chunksize_show(struct mddev *mddev, char *page)
2310 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2313 static ssize_t
2314 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2316 /* Can only be changed when no bitmap is active */
2317 int rv;
2318 unsigned long csize;
2319 if (mddev->bitmap)
2320 return -EBUSY;
2321 rv = kstrtoul(buf, 10, &csize);
2322 if (rv)
2323 return rv;
2324 if (csize < 512 ||
2325 !is_power_of_2(csize))
2326 return -EINVAL;
2327 mddev->bitmap_info.chunksize = csize;
2328 return len;
2331 static struct md_sysfs_entry bitmap_chunksize =
2332 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2334 static ssize_t metadata_show(struct mddev *mddev, char *page)
2336 if (mddev_is_clustered(mddev))
2337 return sprintf(page, "clustered\n");
2338 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2339 ? "external" : "internal"));
2342 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2344 if (mddev->bitmap ||
2345 mddev->bitmap_info.file ||
2346 mddev->bitmap_info.offset)
2347 return -EBUSY;
2348 if (strncmp(buf, "external", 8) == 0)
2349 mddev->bitmap_info.external = 1;
2350 else if ((strncmp(buf, "internal", 8) == 0) ||
2351 (strncmp(buf, "clustered", 9) == 0))
2352 mddev->bitmap_info.external = 0;
2353 else
2354 return -EINVAL;
2355 return len;
2358 static struct md_sysfs_entry bitmap_metadata =
2359 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2361 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2363 int len;
2364 spin_lock(&mddev->lock);
2365 if (mddev->bitmap)
2366 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2367 "false" : "true"));
2368 else
2369 len = sprintf(page, "\n");
2370 spin_unlock(&mddev->lock);
2371 return len;
2374 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2376 if (mddev->bitmap == NULL)
2377 return -ENOENT;
2378 if (strncmp(buf, "false", 5) == 0)
2379 mddev->bitmap->need_sync = 1;
2380 else if (strncmp(buf, "true", 4) == 0) {
2381 if (mddev->degraded)
2382 return -EBUSY;
2383 mddev->bitmap->need_sync = 0;
2384 } else
2385 return -EINVAL;
2386 return len;
2389 static struct md_sysfs_entry bitmap_can_clear =
2390 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2392 static ssize_t
2393 behind_writes_used_show(struct mddev *mddev, char *page)
2395 ssize_t ret;
2396 spin_lock(&mddev->lock);
2397 if (mddev->bitmap == NULL)
2398 ret = sprintf(page, "0\n");
2399 else
2400 ret = sprintf(page, "%lu\n",
2401 mddev->bitmap->behind_writes_used);
2402 spin_unlock(&mddev->lock);
2403 return ret;
2406 static ssize_t
2407 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2409 if (mddev->bitmap)
2410 mddev->bitmap->behind_writes_used = 0;
2411 return len;
2414 static struct md_sysfs_entry max_backlog_used =
2415 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2416 behind_writes_used_show, behind_writes_used_reset);
2418 static struct attribute *md_bitmap_attrs[] = {
2419 &bitmap_location.attr,
2420 &bitmap_space.attr,
2421 &bitmap_timeout.attr,
2422 &bitmap_backlog.attr,
2423 &bitmap_chunksize.attr,
2424 &bitmap_metadata.attr,
2425 &bitmap_can_clear.attr,
2426 &max_backlog_used.attr,
2427 NULL
2429 struct attribute_group md_bitmap_group = {
2430 .name = "bitmap",
2431 .attrs = md_bitmap_attrs,