treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / drivers / md / md-bitmap.c
blobe230052c2107783c2d2410103c8d444eb1ded34f
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
5 * bitmap_create - sets up the bitmap structure
6 * bitmap_destroy - destroys the bitmap structure
8 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
9 * - added disk storage for bitmap
10 * - changes to allow various bitmap chunk sizes
14 * Still to do:
16 * flush after percent set rather than just time based. (maybe both).
19 #include <linux/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/timer.h>
25 #include <linux/sched.h>
26 #include <linux/list.h>
27 #include <linux/file.h>
28 #include <linux/mount.h>
29 #include <linux/buffer_head.h>
30 #include <linux/seq_file.h>
31 #include <trace/events/block.h>
32 #include "md.h"
33 #include "md-bitmap.h"
35 static inline char *bmname(struct bitmap *bitmap)
37 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
41 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
43 * 1) check to see if this page is allocated, if it's not then try to alloc
44 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
45 * page pointer directly as a counter
47 * if we find our page, we increment the page's refcount so that it stays
48 * allocated while we're using it
50 static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
51 unsigned long page, int create, int no_hijack)
52 __releases(bitmap->lock)
53 __acquires(bitmap->lock)
55 unsigned char *mappage;
57 if (page >= bitmap->pages) {
58 /* This can happen if bitmap_start_sync goes beyond
59 * End-of-device while looking for a whole page.
60 * It is harmless.
62 return -EINVAL;
65 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
66 return 0;
68 if (bitmap->bp[page].map) /* page is already allocated, just return */
69 return 0;
71 if (!create)
72 return -ENOENT;
74 /* this page has not been allocated yet */
76 spin_unlock_irq(&bitmap->lock);
77 /* It is possible that this is being called inside a
78 * prepare_to_wait/finish_wait loop from raid5c:make_request().
79 * In general it is not permitted to sleep in that context as it
80 * can cause the loop to spin freely.
81 * That doesn't apply here as we can only reach this point
82 * once with any loop.
83 * When this function completes, either bp[page].map or
84 * bp[page].hijacked. In either case, this function will
85 * abort before getting to this point again. So there is
86 * no risk of a free-spin, and so it is safe to assert
87 * that sleeping here is allowed.
89 sched_annotate_sleep();
90 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
91 spin_lock_irq(&bitmap->lock);
93 if (mappage == NULL) {
94 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
95 /* We don't support hijack for cluster raid */
96 if (no_hijack)
97 return -ENOMEM;
98 /* failed - set the hijacked flag so that we can use the
99 * pointer as a counter */
100 if (!bitmap->bp[page].map)
101 bitmap->bp[page].hijacked = 1;
102 } else if (bitmap->bp[page].map ||
103 bitmap->bp[page].hijacked) {
104 /* somebody beat us to getting the page */
105 kfree(mappage);
106 } else {
108 /* no page was in place and we have one, so install it */
110 bitmap->bp[page].map = mappage;
111 bitmap->missing_pages--;
113 return 0;
116 /* if page is completely empty, put it back on the free list, or dealloc it */
117 /* if page was hijacked, unmark the flag so it might get alloced next time */
118 /* Note: lock should be held when calling this */
119 static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
121 char *ptr;
123 if (bitmap->bp[page].count) /* page is still busy */
124 return;
126 /* page is no longer in use, it can be released */
128 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
129 bitmap->bp[page].hijacked = 0;
130 bitmap->bp[page].map = NULL;
131 } else {
132 /* normal case, free the page */
133 ptr = bitmap->bp[page].map;
134 bitmap->bp[page].map = NULL;
135 bitmap->missing_pages++;
136 kfree(ptr);
141 * bitmap file handling - read and write the bitmap file and its superblock
145 * basic page I/O operations
148 /* IO operations when bitmap is stored near all superblocks */
149 static int read_sb_page(struct mddev *mddev, loff_t offset,
150 struct page *page,
151 unsigned long index, int size)
153 /* choose a good rdev and read the page from there */
155 struct md_rdev *rdev;
156 sector_t target;
158 rdev_for_each(rdev, mddev) {
159 if (! test_bit(In_sync, &rdev->flags)
160 || test_bit(Faulty, &rdev->flags)
161 || test_bit(Bitmap_sync, &rdev->flags))
162 continue;
164 target = offset + index * (PAGE_SIZE/512);
166 if (sync_page_io(rdev, target,
167 roundup(size, bdev_logical_block_size(rdev->bdev)),
168 page, REQ_OP_READ, 0, true)) {
169 page->index = index;
170 return 0;
173 return -EIO;
176 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
178 /* Iterate the disks of an mddev, using rcu to protect access to the
179 * linked list, and raising the refcount of devices we return to ensure
180 * they don't disappear while in use.
181 * As devices are only added or removed when raid_disk is < 0 and
182 * nr_pending is 0 and In_sync is clear, the entries we return will
183 * still be in the same position on the list when we re-enter
184 * list_for_each_entry_continue_rcu.
186 * Note that if entered with 'rdev == NULL' to start at the
187 * beginning, we temporarily assign 'rdev' to an address which
188 * isn't really an rdev, but which can be used by
189 * list_for_each_entry_continue_rcu() to find the first entry.
191 rcu_read_lock();
192 if (rdev == NULL)
193 /* start at the beginning */
194 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
195 else {
196 /* release the previous rdev and start from there. */
197 rdev_dec_pending(rdev, mddev);
199 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
200 if (rdev->raid_disk >= 0 &&
201 !test_bit(Faulty, &rdev->flags)) {
202 /* this is a usable devices */
203 atomic_inc(&rdev->nr_pending);
204 rcu_read_unlock();
205 return rdev;
208 rcu_read_unlock();
209 return NULL;
212 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
214 struct md_rdev *rdev;
215 struct block_device *bdev;
216 struct mddev *mddev = bitmap->mddev;
217 struct bitmap_storage *store = &bitmap->storage;
219 restart:
220 rdev = NULL;
221 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
222 int size = PAGE_SIZE;
223 loff_t offset = mddev->bitmap_info.offset;
225 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
227 if (page->index == store->file_pages-1) {
228 int last_page_size = store->bytes & (PAGE_SIZE-1);
229 if (last_page_size == 0)
230 last_page_size = PAGE_SIZE;
231 size = roundup(last_page_size,
232 bdev_logical_block_size(bdev));
234 /* Just make sure we aren't corrupting data or
235 * metadata
237 if (mddev->external) {
238 /* Bitmap could be anywhere. */
239 if (rdev->sb_start + offset + (page->index
240 * (PAGE_SIZE/512))
241 > rdev->data_offset
243 rdev->sb_start + offset
244 < (rdev->data_offset + mddev->dev_sectors
245 + (PAGE_SIZE/512)))
246 goto bad_alignment;
247 } else if (offset < 0) {
248 /* DATA BITMAP METADATA */
249 if (offset
250 + (long)(page->index * (PAGE_SIZE/512))
251 + size/512 > 0)
252 /* bitmap runs in to metadata */
253 goto bad_alignment;
254 if (rdev->data_offset + mddev->dev_sectors
255 > rdev->sb_start + offset)
256 /* data runs in to bitmap */
257 goto bad_alignment;
258 } else if (rdev->sb_start < rdev->data_offset) {
259 /* METADATA BITMAP DATA */
260 if (rdev->sb_start
261 + offset
262 + page->index*(PAGE_SIZE/512) + size/512
263 > rdev->data_offset)
264 /* bitmap runs in to data */
265 goto bad_alignment;
266 } else {
267 /* DATA METADATA BITMAP - no problems */
269 md_super_write(mddev, rdev,
270 rdev->sb_start + offset
271 + page->index * (PAGE_SIZE/512),
272 size,
273 page);
276 if (wait && md_super_wait(mddev) < 0)
277 goto restart;
278 return 0;
280 bad_alignment:
281 return -EINVAL;
284 static void md_bitmap_file_kick(struct bitmap *bitmap);
286 * write out a page to a file
288 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
290 struct buffer_head *bh;
292 if (bitmap->storage.file == NULL) {
293 switch (write_sb_page(bitmap, page, wait)) {
294 case -EINVAL:
295 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
297 } else {
299 bh = page_buffers(page);
301 while (bh && bh->b_blocknr) {
302 atomic_inc(&bitmap->pending_writes);
303 set_buffer_locked(bh);
304 set_buffer_mapped(bh);
305 submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
306 bh = bh->b_this_page;
309 if (wait)
310 wait_event(bitmap->write_wait,
311 atomic_read(&bitmap->pending_writes)==0);
313 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
314 md_bitmap_file_kick(bitmap);
317 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
319 struct bitmap *bitmap = bh->b_private;
321 if (!uptodate)
322 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
323 if (atomic_dec_and_test(&bitmap->pending_writes))
324 wake_up(&bitmap->write_wait);
327 /* copied from buffer.c */
328 static void
329 __clear_page_buffers(struct page *page)
331 ClearPagePrivate(page);
332 set_page_private(page, 0);
333 put_page(page);
335 static void free_buffers(struct page *page)
337 struct buffer_head *bh;
339 if (!PagePrivate(page))
340 return;
342 bh = page_buffers(page);
343 while (bh) {
344 struct buffer_head *next = bh->b_this_page;
345 free_buffer_head(bh);
346 bh = next;
348 __clear_page_buffers(page);
349 put_page(page);
352 /* read a page from a file.
353 * We both read the page, and attach buffers to the page to record the
354 * address of each block (using bmap). These addresses will be used
355 * to write the block later, completely bypassing the filesystem.
356 * This usage is similar to how swap files are handled, and allows us
357 * to write to a file with no concerns of memory allocation failing.
359 static int read_page(struct file *file, unsigned long index,
360 struct bitmap *bitmap,
361 unsigned long count,
362 struct page *page)
364 int ret = 0;
365 struct inode *inode = file_inode(file);
366 struct buffer_head *bh;
367 sector_t block;
369 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
370 (unsigned long long)index << PAGE_SHIFT);
372 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, false);
373 if (!bh) {
374 ret = -ENOMEM;
375 goto out;
377 attach_page_buffers(page, bh);
378 block = index << (PAGE_SHIFT - inode->i_blkbits);
379 while (bh) {
380 if (count == 0)
381 bh->b_blocknr = 0;
382 else {
383 bh->b_blocknr = bmap(inode, block);
384 if (bh->b_blocknr == 0) {
385 /* Cannot use this file! */
386 ret = -EINVAL;
387 goto out;
389 bh->b_bdev = inode->i_sb->s_bdev;
390 if (count < (1<<inode->i_blkbits))
391 count = 0;
392 else
393 count -= (1<<inode->i_blkbits);
395 bh->b_end_io = end_bitmap_write;
396 bh->b_private = bitmap;
397 atomic_inc(&bitmap->pending_writes);
398 set_buffer_locked(bh);
399 set_buffer_mapped(bh);
400 submit_bh(REQ_OP_READ, 0, bh);
402 block++;
403 bh = bh->b_this_page;
405 page->index = index;
407 wait_event(bitmap->write_wait,
408 atomic_read(&bitmap->pending_writes)==0);
409 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
410 ret = -EIO;
411 out:
412 if (ret)
413 pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
414 (int)PAGE_SIZE,
415 (unsigned long long)index << PAGE_SHIFT,
416 ret);
417 return ret;
421 * bitmap file superblock operations
425 * md_bitmap_wait_writes() should be called before writing any bitmap
426 * blocks, to ensure previous writes, particularly from
427 * md_bitmap_daemon_work(), have completed.
429 static void md_bitmap_wait_writes(struct bitmap *bitmap)
431 if (bitmap->storage.file)
432 wait_event(bitmap->write_wait,
433 atomic_read(&bitmap->pending_writes)==0);
434 else
435 /* Note that we ignore the return value. The writes
436 * might have failed, but that would just mean that
437 * some bits which should be cleared haven't been,
438 * which is safe. The relevant bitmap blocks will
439 * probably get written again, but there is no great
440 * loss if they aren't.
442 md_super_wait(bitmap->mddev);
446 /* update the event counter and sync the superblock to disk */
447 void md_bitmap_update_sb(struct bitmap *bitmap)
449 bitmap_super_t *sb;
451 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
452 return;
453 if (bitmap->mddev->bitmap_info.external)
454 return;
455 if (!bitmap->storage.sb_page) /* no superblock */
456 return;
457 sb = kmap_atomic(bitmap->storage.sb_page);
458 sb->events = cpu_to_le64(bitmap->mddev->events);
459 if (bitmap->mddev->events < bitmap->events_cleared)
460 /* rocking back to read-only */
461 bitmap->events_cleared = bitmap->mddev->events;
462 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
464 * clear BITMAP_WRITE_ERROR bit to protect against the case that
465 * a bitmap write error occurred but the later writes succeeded.
467 sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
468 /* Just in case these have been changed via sysfs: */
469 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
470 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
471 /* This might have been changed by a reshape */
472 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
473 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
474 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
475 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
476 bitmap_info.space);
477 kunmap_atomic(sb);
478 write_page(bitmap, bitmap->storage.sb_page, 1);
480 EXPORT_SYMBOL(md_bitmap_update_sb);
482 /* print out the bitmap file superblock */
483 void md_bitmap_print_sb(struct bitmap *bitmap)
485 bitmap_super_t *sb;
487 if (!bitmap || !bitmap->storage.sb_page)
488 return;
489 sb = kmap_atomic(bitmap->storage.sb_page);
490 pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
491 pr_debug(" magic: %08x\n", le32_to_cpu(sb->magic));
492 pr_debug(" version: %d\n", le32_to_cpu(sb->version));
493 pr_debug(" uuid: %08x.%08x.%08x.%08x\n",
494 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
495 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
496 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
497 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
498 pr_debug(" events: %llu\n",
499 (unsigned long long) le64_to_cpu(sb->events));
500 pr_debug("events cleared: %llu\n",
501 (unsigned long long) le64_to_cpu(sb->events_cleared));
502 pr_debug(" state: %08x\n", le32_to_cpu(sb->state));
503 pr_debug(" chunksize: %d B\n", le32_to_cpu(sb->chunksize));
504 pr_debug(" daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
505 pr_debug(" sync size: %llu KB\n",
506 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
507 pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind));
508 kunmap_atomic(sb);
512 * bitmap_new_disk_sb
513 * @bitmap
515 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
516 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
517 * This function verifies 'bitmap_info' and populates the on-disk bitmap
518 * structure, which is to be written to disk.
520 * Returns: 0 on success, -Exxx on error
522 static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
524 bitmap_super_t *sb;
525 unsigned long chunksize, daemon_sleep, write_behind;
527 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
528 if (bitmap->storage.sb_page == NULL)
529 return -ENOMEM;
530 bitmap->storage.sb_page->index = 0;
532 sb = kmap_atomic(bitmap->storage.sb_page);
534 sb->magic = cpu_to_le32(BITMAP_MAGIC);
535 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
537 chunksize = bitmap->mddev->bitmap_info.chunksize;
538 BUG_ON(!chunksize);
539 if (!is_power_of_2(chunksize)) {
540 kunmap_atomic(sb);
541 pr_warn("bitmap chunksize not a power of 2\n");
542 return -EINVAL;
544 sb->chunksize = cpu_to_le32(chunksize);
546 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
547 if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
548 pr_debug("Choosing daemon_sleep default (5 sec)\n");
549 daemon_sleep = 5 * HZ;
551 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
552 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
555 * FIXME: write_behind for RAID1. If not specified, what
556 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
558 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
559 if (write_behind > COUNTER_MAX)
560 write_behind = COUNTER_MAX / 2;
561 sb->write_behind = cpu_to_le32(write_behind);
562 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
564 /* keep the array size field of the bitmap superblock up to date */
565 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
567 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
569 set_bit(BITMAP_STALE, &bitmap->flags);
570 sb->state = cpu_to_le32(bitmap->flags);
571 bitmap->events_cleared = bitmap->mddev->events;
572 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
573 bitmap->mddev->bitmap_info.nodes = 0;
575 kunmap_atomic(sb);
577 return 0;
580 /* read the superblock from the bitmap file and initialize some bitmap fields */
581 static int md_bitmap_read_sb(struct bitmap *bitmap)
583 char *reason = NULL;
584 bitmap_super_t *sb;
585 unsigned long chunksize, daemon_sleep, write_behind;
586 unsigned long long events;
587 int nodes = 0;
588 unsigned long sectors_reserved = 0;
589 int err = -EINVAL;
590 struct page *sb_page;
591 loff_t offset = bitmap->mddev->bitmap_info.offset;
593 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
594 chunksize = 128 * 1024 * 1024;
595 daemon_sleep = 5 * HZ;
596 write_behind = 0;
597 set_bit(BITMAP_STALE, &bitmap->flags);
598 err = 0;
599 goto out_no_sb;
601 /* page 0 is the superblock, read it... */
602 sb_page = alloc_page(GFP_KERNEL);
603 if (!sb_page)
604 return -ENOMEM;
605 bitmap->storage.sb_page = sb_page;
607 re_read:
608 /* If cluster_slot is set, the cluster is setup */
609 if (bitmap->cluster_slot >= 0) {
610 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
612 sector_div(bm_blocks,
613 bitmap->mddev->bitmap_info.chunksize >> 9);
614 /* bits to bytes */
615 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
616 /* to 4k blocks */
617 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
618 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
619 pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
620 bitmap->cluster_slot, offset);
623 if (bitmap->storage.file) {
624 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
625 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
627 err = read_page(bitmap->storage.file, 0,
628 bitmap, bytes, sb_page);
629 } else {
630 err = read_sb_page(bitmap->mddev,
631 offset,
632 sb_page,
633 0, sizeof(bitmap_super_t));
635 if (err)
636 return err;
638 err = -EINVAL;
639 sb = kmap_atomic(sb_page);
641 chunksize = le32_to_cpu(sb->chunksize);
642 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
643 write_behind = le32_to_cpu(sb->write_behind);
644 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
645 /* Setup nodes/clustername only if bitmap version is
646 * cluster-compatible
648 if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
649 nodes = le32_to_cpu(sb->nodes);
650 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
651 sb->cluster_name, 64);
654 /* verify that the bitmap-specific fields are valid */
655 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
656 reason = "bad magic";
657 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
658 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
659 reason = "unrecognized superblock version";
660 else if (chunksize < 512)
661 reason = "bitmap chunksize too small";
662 else if (!is_power_of_2(chunksize))
663 reason = "bitmap chunksize not a power of 2";
664 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
665 reason = "daemon sleep period out of range";
666 else if (write_behind > COUNTER_MAX)
667 reason = "write-behind limit out of range (0 - 16383)";
668 if (reason) {
669 pr_warn("%s: invalid bitmap file superblock: %s\n",
670 bmname(bitmap), reason);
671 goto out;
674 /* keep the array size field of the bitmap superblock up to date */
675 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
677 if (bitmap->mddev->persistent) {
679 * We have a persistent array superblock, so compare the
680 * bitmap's UUID and event counter to the mddev's
682 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
683 pr_warn("%s: bitmap superblock UUID mismatch\n",
684 bmname(bitmap));
685 goto out;
687 events = le64_to_cpu(sb->events);
688 if (!nodes && (events < bitmap->mddev->events)) {
689 pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
690 bmname(bitmap), events,
691 (unsigned long long) bitmap->mddev->events);
692 set_bit(BITMAP_STALE, &bitmap->flags);
696 /* assign fields using values from superblock */
697 bitmap->flags |= le32_to_cpu(sb->state);
698 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
699 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
700 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
701 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
702 err = 0;
704 out:
705 kunmap_atomic(sb);
706 /* Assigning chunksize is required for "re_read" */
707 bitmap->mddev->bitmap_info.chunksize = chunksize;
708 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
709 err = md_setup_cluster(bitmap->mddev, nodes);
710 if (err) {
711 pr_warn("%s: Could not setup cluster service (%d)\n",
712 bmname(bitmap), err);
713 goto out_no_sb;
715 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
716 goto re_read;
720 out_no_sb:
721 if (test_bit(BITMAP_STALE, &bitmap->flags))
722 bitmap->events_cleared = bitmap->mddev->events;
723 bitmap->mddev->bitmap_info.chunksize = chunksize;
724 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
725 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
726 bitmap->mddev->bitmap_info.nodes = nodes;
727 if (bitmap->mddev->bitmap_info.space == 0 ||
728 bitmap->mddev->bitmap_info.space > sectors_reserved)
729 bitmap->mddev->bitmap_info.space = sectors_reserved;
730 if (err) {
731 md_bitmap_print_sb(bitmap);
732 if (bitmap->cluster_slot < 0)
733 md_cluster_stop(bitmap->mddev);
735 return err;
739 * general bitmap file operations
743 * on-disk bitmap:
745 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
746 * file a page at a time. There's a superblock at the start of the file.
748 /* calculate the index of the page that contains this bit */
749 static inline unsigned long file_page_index(struct bitmap_storage *store,
750 unsigned long chunk)
752 if (store->sb_page)
753 chunk += sizeof(bitmap_super_t) << 3;
754 return chunk >> PAGE_BIT_SHIFT;
757 /* calculate the (bit) offset of this bit within a page */
758 static inline unsigned long file_page_offset(struct bitmap_storage *store,
759 unsigned long chunk)
761 if (store->sb_page)
762 chunk += sizeof(bitmap_super_t) << 3;
763 return chunk & (PAGE_BITS - 1);
767 * return a pointer to the page in the filemap that contains the given bit
770 static inline struct page *filemap_get_page(struct bitmap_storage *store,
771 unsigned long chunk)
773 if (file_page_index(store, chunk) >= store->file_pages)
774 return NULL;
775 return store->filemap[file_page_index(store, chunk)];
778 static int md_bitmap_storage_alloc(struct bitmap_storage *store,
779 unsigned long chunks, int with_super,
780 int slot_number)
782 int pnum, offset = 0;
783 unsigned long num_pages;
784 unsigned long bytes;
786 bytes = DIV_ROUND_UP(chunks, 8);
787 if (with_super)
788 bytes += sizeof(bitmap_super_t);
790 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
791 offset = slot_number * num_pages;
793 store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
794 GFP_KERNEL);
795 if (!store->filemap)
796 return -ENOMEM;
798 if (with_super && !store->sb_page) {
799 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
800 if (store->sb_page == NULL)
801 return -ENOMEM;
804 pnum = 0;
805 if (store->sb_page) {
806 store->filemap[0] = store->sb_page;
807 pnum = 1;
808 store->sb_page->index = offset;
811 for ( ; pnum < num_pages; pnum++) {
812 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
813 if (!store->filemap[pnum]) {
814 store->file_pages = pnum;
815 return -ENOMEM;
817 store->filemap[pnum]->index = pnum + offset;
819 store->file_pages = pnum;
821 /* We need 4 bits per page, rounded up to a multiple
822 * of sizeof(unsigned long) */
823 store->filemap_attr = kzalloc(
824 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
825 GFP_KERNEL);
826 if (!store->filemap_attr)
827 return -ENOMEM;
829 store->bytes = bytes;
831 return 0;
834 static void md_bitmap_file_unmap(struct bitmap_storage *store)
836 struct page **map, *sb_page;
837 int pages;
838 struct file *file;
840 file = store->file;
841 map = store->filemap;
842 pages = store->file_pages;
843 sb_page = store->sb_page;
845 while (pages--)
846 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
847 free_buffers(map[pages]);
848 kfree(map);
849 kfree(store->filemap_attr);
851 if (sb_page)
852 free_buffers(sb_page);
854 if (file) {
855 struct inode *inode = file_inode(file);
856 invalidate_mapping_pages(inode->i_mapping, 0, -1);
857 fput(file);
862 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
863 * then it is no longer reliable, so we stop using it and we mark the file
864 * as failed in the superblock
866 static void md_bitmap_file_kick(struct bitmap *bitmap)
868 char *path, *ptr = NULL;
870 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
871 md_bitmap_update_sb(bitmap);
873 if (bitmap->storage.file) {
874 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
875 if (path)
876 ptr = file_path(bitmap->storage.file,
877 path, PAGE_SIZE);
879 pr_warn("%s: kicking failed bitmap file %s from array!\n",
880 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
882 kfree(path);
883 } else
884 pr_warn("%s: disabling internal bitmap due to errors\n",
885 bmname(bitmap));
889 enum bitmap_page_attr {
890 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
891 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
892 * i.e. counter is 1 or 2. */
893 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
896 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
897 enum bitmap_page_attr attr)
899 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
902 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
903 enum bitmap_page_attr attr)
905 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
908 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
909 enum bitmap_page_attr attr)
911 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
914 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
915 enum bitmap_page_attr attr)
917 return test_and_clear_bit((pnum<<2) + attr,
918 bitmap->storage.filemap_attr);
921 * bitmap_file_set_bit -- called before performing a write to the md device
922 * to set (and eventually sync) a particular bit in the bitmap file
924 * we set the bit immediately, then we record the page number so that
925 * when an unplug occurs, we can flush the dirty pages out to disk
927 static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
929 unsigned long bit;
930 struct page *page;
931 void *kaddr;
932 unsigned long chunk = block >> bitmap->counts.chunkshift;
933 struct bitmap_storage *store = &bitmap->storage;
934 unsigned long node_offset = 0;
936 if (mddev_is_clustered(bitmap->mddev))
937 node_offset = bitmap->cluster_slot * store->file_pages;
939 page = filemap_get_page(&bitmap->storage, chunk);
940 if (!page)
941 return;
942 bit = file_page_offset(&bitmap->storage, chunk);
944 /* set the bit */
945 kaddr = kmap_atomic(page);
946 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
947 set_bit(bit, kaddr);
948 else
949 set_bit_le(bit, kaddr);
950 kunmap_atomic(kaddr);
951 pr_debug("set file bit %lu page %lu\n", bit, page->index);
952 /* record page number so it gets flushed to disk when unplug occurs */
953 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY);
956 static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
958 unsigned long bit;
959 struct page *page;
960 void *paddr;
961 unsigned long chunk = block >> bitmap->counts.chunkshift;
962 struct bitmap_storage *store = &bitmap->storage;
963 unsigned long node_offset = 0;
965 if (mddev_is_clustered(bitmap->mddev))
966 node_offset = bitmap->cluster_slot * store->file_pages;
968 page = filemap_get_page(&bitmap->storage, chunk);
969 if (!page)
970 return;
971 bit = file_page_offset(&bitmap->storage, chunk);
972 paddr = kmap_atomic(page);
973 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
974 clear_bit(bit, paddr);
975 else
976 clear_bit_le(bit, paddr);
977 kunmap_atomic(paddr);
978 if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
979 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
980 bitmap->allclean = 0;
984 static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
986 unsigned long bit;
987 struct page *page;
988 void *paddr;
989 unsigned long chunk = block >> bitmap->counts.chunkshift;
990 int set = 0;
992 page = filemap_get_page(&bitmap->storage, chunk);
993 if (!page)
994 return -EINVAL;
995 bit = file_page_offset(&bitmap->storage, chunk);
996 paddr = kmap_atomic(page);
997 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
998 set = test_bit(bit, paddr);
999 else
1000 set = test_bit_le(bit, paddr);
1001 kunmap_atomic(paddr);
1002 return set;
1006 /* this gets called when the md device is ready to unplug its underlying
1007 * (slave) device queues -- before we let any writes go down, we need to
1008 * sync the dirty pages of the bitmap file to disk */
1009 void md_bitmap_unplug(struct bitmap *bitmap)
1011 unsigned long i;
1012 int dirty, need_write;
1013 int writing = 0;
1015 if (!bitmap || !bitmap->storage.filemap ||
1016 test_bit(BITMAP_STALE, &bitmap->flags))
1017 return;
1019 /* look at each page to see if there are any set bits that need to be
1020 * flushed out to disk */
1021 for (i = 0; i < bitmap->storage.file_pages; i++) {
1022 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1023 need_write = test_and_clear_page_attr(bitmap, i,
1024 BITMAP_PAGE_NEEDWRITE);
1025 if (dirty || need_write) {
1026 if (!writing) {
1027 md_bitmap_wait_writes(bitmap);
1028 if (bitmap->mddev->queue)
1029 blk_add_trace_msg(bitmap->mddev->queue,
1030 "md bitmap_unplug");
1032 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1033 write_page(bitmap, bitmap->storage.filemap[i], 0);
1034 writing = 1;
1037 if (writing)
1038 md_bitmap_wait_writes(bitmap);
1040 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1041 md_bitmap_file_kick(bitmap);
1043 EXPORT_SYMBOL(md_bitmap_unplug);
1045 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1046 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1047 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1048 * memory mapping of the bitmap file
1049 * Special cases:
1050 * if there's no bitmap file, or if the bitmap file had been
1051 * previously kicked from the array, we mark all the bits as
1052 * 1's in order to cause a full resync.
1054 * We ignore all bits for sectors that end earlier than 'start'.
1055 * This is used when reading an out-of-date bitmap...
1057 static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1059 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1060 struct page *page = NULL;
1061 unsigned long bit_cnt = 0;
1062 struct file *file;
1063 unsigned long offset;
1064 int outofdate;
1065 int ret = -ENOSPC;
1066 void *paddr;
1067 struct bitmap_storage *store = &bitmap->storage;
1069 chunks = bitmap->counts.chunks;
1070 file = store->file;
1072 if (!file && !bitmap->mddev->bitmap_info.offset) {
1073 /* No permanent bitmap - fill with '1s'. */
1074 store->filemap = NULL;
1075 store->file_pages = 0;
1076 for (i = 0; i < chunks ; i++) {
1077 /* if the disk bit is set, set the memory bit */
1078 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1079 >= start);
1080 md_bitmap_set_memory_bits(bitmap,
1081 (sector_t)i << bitmap->counts.chunkshift,
1082 needed);
1084 return 0;
1087 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1088 if (outofdate)
1089 pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap));
1091 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1092 pr_warn("%s: bitmap file too short %lu < %lu\n",
1093 bmname(bitmap),
1094 (unsigned long) i_size_read(file->f_mapping->host),
1095 store->bytes);
1096 goto err;
1099 oldindex = ~0L;
1100 offset = 0;
1101 if (!bitmap->mddev->bitmap_info.external)
1102 offset = sizeof(bitmap_super_t);
1104 if (mddev_is_clustered(bitmap->mddev))
1105 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1107 for (i = 0; i < chunks; i++) {
1108 int b;
1109 index = file_page_index(&bitmap->storage, i);
1110 bit = file_page_offset(&bitmap->storage, i);
1111 if (index != oldindex) { /* this is a new page, read it in */
1112 int count;
1113 /* unmap the old page, we're done with it */
1114 if (index == store->file_pages-1)
1115 count = store->bytes - index * PAGE_SIZE;
1116 else
1117 count = PAGE_SIZE;
1118 page = store->filemap[index];
1119 if (file)
1120 ret = read_page(file, index, bitmap,
1121 count, page);
1122 else
1123 ret = read_sb_page(
1124 bitmap->mddev,
1125 bitmap->mddev->bitmap_info.offset,
1126 page,
1127 index + node_offset, count);
1129 if (ret)
1130 goto err;
1132 oldindex = index;
1134 if (outofdate) {
1136 * if bitmap is out of date, dirty the
1137 * whole page and write it out
1139 paddr = kmap_atomic(page);
1140 memset(paddr + offset, 0xff,
1141 PAGE_SIZE - offset);
1142 kunmap_atomic(paddr);
1143 write_page(bitmap, page, 1);
1145 ret = -EIO;
1146 if (test_bit(BITMAP_WRITE_ERROR,
1147 &bitmap->flags))
1148 goto err;
1151 paddr = kmap_atomic(page);
1152 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1153 b = test_bit(bit, paddr);
1154 else
1155 b = test_bit_le(bit, paddr);
1156 kunmap_atomic(paddr);
1157 if (b) {
1158 /* if the disk bit is set, set the memory bit */
1159 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1160 >= start);
1161 md_bitmap_set_memory_bits(bitmap,
1162 (sector_t)i << bitmap->counts.chunkshift,
1163 needed);
1164 bit_cnt++;
1166 offset = 0;
1169 pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1170 bmname(bitmap), store->file_pages,
1171 bit_cnt, chunks);
1173 return 0;
1175 err:
1176 pr_warn("%s: bitmap initialisation failed: %d\n",
1177 bmname(bitmap), ret);
1178 return ret;
1181 void md_bitmap_write_all(struct bitmap *bitmap)
1183 /* We don't actually write all bitmap blocks here,
1184 * just flag them as needing to be written
1186 int i;
1188 if (!bitmap || !bitmap->storage.filemap)
1189 return;
1190 if (bitmap->storage.file)
1191 /* Only one copy, so nothing needed */
1192 return;
1194 for (i = 0; i < bitmap->storage.file_pages; i++)
1195 set_page_attr(bitmap, i,
1196 BITMAP_PAGE_NEEDWRITE);
1197 bitmap->allclean = 0;
1200 static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1201 sector_t offset, int inc)
1203 sector_t chunk = offset >> bitmap->chunkshift;
1204 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1205 bitmap->bp[page].count += inc;
1206 md_bitmap_checkfree(bitmap, page);
1209 static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1211 sector_t chunk = offset >> bitmap->chunkshift;
1212 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1213 struct bitmap_page *bp = &bitmap->bp[page];
1215 if (!bp->pending)
1216 bp->pending = 1;
1219 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1220 sector_t offset, sector_t *blocks,
1221 int create);
1224 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1225 * out to disk
1228 void md_bitmap_daemon_work(struct mddev *mddev)
1230 struct bitmap *bitmap;
1231 unsigned long j;
1232 unsigned long nextpage;
1233 sector_t blocks;
1234 struct bitmap_counts *counts;
1236 /* Use a mutex to guard daemon_work against
1237 * bitmap_destroy.
1239 mutex_lock(&mddev->bitmap_info.mutex);
1240 bitmap = mddev->bitmap;
1241 if (bitmap == NULL) {
1242 mutex_unlock(&mddev->bitmap_info.mutex);
1243 return;
1245 if (time_before(jiffies, bitmap->daemon_lastrun
1246 + mddev->bitmap_info.daemon_sleep))
1247 goto done;
1249 bitmap->daemon_lastrun = jiffies;
1250 if (bitmap->allclean) {
1251 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1252 goto done;
1254 bitmap->allclean = 1;
1256 if (bitmap->mddev->queue)
1257 blk_add_trace_msg(bitmap->mddev->queue,
1258 "md bitmap_daemon_work");
1260 /* Any file-page which is PENDING now needs to be written.
1261 * So set NEEDWRITE now, then after we make any last-minute changes
1262 * we will write it.
1264 for (j = 0; j < bitmap->storage.file_pages; j++)
1265 if (test_and_clear_page_attr(bitmap, j,
1266 BITMAP_PAGE_PENDING))
1267 set_page_attr(bitmap, j,
1268 BITMAP_PAGE_NEEDWRITE);
1270 if (bitmap->need_sync &&
1271 mddev->bitmap_info.external == 0) {
1272 /* Arrange for superblock update as well as
1273 * other changes */
1274 bitmap_super_t *sb;
1275 bitmap->need_sync = 0;
1276 if (bitmap->storage.filemap) {
1277 sb = kmap_atomic(bitmap->storage.sb_page);
1278 sb->events_cleared =
1279 cpu_to_le64(bitmap->events_cleared);
1280 kunmap_atomic(sb);
1281 set_page_attr(bitmap, 0,
1282 BITMAP_PAGE_NEEDWRITE);
1285 /* Now look at the bitmap counters and if any are '2' or '1',
1286 * decrement and handle accordingly.
1288 counts = &bitmap->counts;
1289 spin_lock_irq(&counts->lock);
1290 nextpage = 0;
1291 for (j = 0; j < counts->chunks; j++) {
1292 bitmap_counter_t *bmc;
1293 sector_t block = (sector_t)j << counts->chunkshift;
1295 if (j == nextpage) {
1296 nextpage += PAGE_COUNTER_RATIO;
1297 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1298 j |= PAGE_COUNTER_MASK;
1299 continue;
1301 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1304 bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1305 if (!bmc) {
1306 j |= PAGE_COUNTER_MASK;
1307 continue;
1309 if (*bmc == 1 && !bitmap->need_sync) {
1310 /* We can clear the bit */
1311 *bmc = 0;
1312 md_bitmap_count_page(counts, block, -1);
1313 md_bitmap_file_clear_bit(bitmap, block);
1314 } else if (*bmc && *bmc <= 2) {
1315 *bmc = 1;
1316 md_bitmap_set_pending(counts, block);
1317 bitmap->allclean = 0;
1320 spin_unlock_irq(&counts->lock);
1322 md_bitmap_wait_writes(bitmap);
1323 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1324 * DIRTY pages need to be written by bitmap_unplug so it can wait
1325 * for them.
1326 * If we find any DIRTY page we stop there and let bitmap_unplug
1327 * handle all the rest. This is important in the case where
1328 * the first blocking holds the superblock and it has been updated.
1329 * We mustn't write any other blocks before the superblock.
1331 for (j = 0;
1332 j < bitmap->storage.file_pages
1333 && !test_bit(BITMAP_STALE, &bitmap->flags);
1334 j++) {
1335 if (test_page_attr(bitmap, j,
1336 BITMAP_PAGE_DIRTY))
1337 /* bitmap_unplug will handle the rest */
1338 break;
1339 if (bitmap->storage.filemap &&
1340 test_and_clear_page_attr(bitmap, j,
1341 BITMAP_PAGE_NEEDWRITE)) {
1342 write_page(bitmap, bitmap->storage.filemap[j], 0);
1346 done:
1347 if (bitmap->allclean == 0)
1348 mddev->thread->timeout =
1349 mddev->bitmap_info.daemon_sleep;
1350 mutex_unlock(&mddev->bitmap_info.mutex);
1353 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1354 sector_t offset, sector_t *blocks,
1355 int create)
1356 __releases(bitmap->lock)
1357 __acquires(bitmap->lock)
1359 /* If 'create', we might release the lock and reclaim it.
1360 * The lock must have been taken with interrupts enabled.
1361 * If !create, we don't release the lock.
1363 sector_t chunk = offset >> bitmap->chunkshift;
1364 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1365 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1366 sector_t csize;
1367 int err;
1369 err = md_bitmap_checkpage(bitmap, page, create, 0);
1371 if (bitmap->bp[page].hijacked ||
1372 bitmap->bp[page].map == NULL)
1373 csize = ((sector_t)1) << (bitmap->chunkshift +
1374 PAGE_COUNTER_SHIFT - 1);
1375 else
1376 csize = ((sector_t)1) << bitmap->chunkshift;
1377 *blocks = csize - (offset & (csize - 1));
1379 if (err < 0)
1380 return NULL;
1382 /* now locked ... */
1384 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1385 /* should we use the first or second counter field
1386 * of the hijacked pointer? */
1387 int hi = (pageoff > PAGE_COUNTER_MASK);
1388 return &((bitmap_counter_t *)
1389 &bitmap->bp[page].map)[hi];
1390 } else /* page is allocated */
1391 return (bitmap_counter_t *)
1392 &(bitmap->bp[page].map[pageoff]);
1395 int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1397 if (!bitmap)
1398 return 0;
1400 if (behind) {
1401 int bw;
1402 atomic_inc(&bitmap->behind_writes);
1403 bw = atomic_read(&bitmap->behind_writes);
1404 if (bw > bitmap->behind_writes_used)
1405 bitmap->behind_writes_used = bw;
1407 pr_debug("inc write-behind count %d/%lu\n",
1408 bw, bitmap->mddev->bitmap_info.max_write_behind);
1411 while (sectors) {
1412 sector_t blocks;
1413 bitmap_counter_t *bmc;
1415 spin_lock_irq(&bitmap->counts.lock);
1416 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1417 if (!bmc) {
1418 spin_unlock_irq(&bitmap->counts.lock);
1419 return 0;
1422 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1423 DEFINE_WAIT(__wait);
1424 /* note that it is safe to do the prepare_to_wait
1425 * after the test as long as we do it before dropping
1426 * the spinlock.
1428 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1429 TASK_UNINTERRUPTIBLE);
1430 spin_unlock_irq(&bitmap->counts.lock);
1431 schedule();
1432 finish_wait(&bitmap->overflow_wait, &__wait);
1433 continue;
1436 switch (*bmc) {
1437 case 0:
1438 md_bitmap_file_set_bit(bitmap, offset);
1439 md_bitmap_count_page(&bitmap->counts, offset, 1);
1440 /* fall through */
1441 case 1:
1442 *bmc = 2;
1445 (*bmc)++;
1447 spin_unlock_irq(&bitmap->counts.lock);
1449 offset += blocks;
1450 if (sectors > blocks)
1451 sectors -= blocks;
1452 else
1453 sectors = 0;
1455 return 0;
1457 EXPORT_SYMBOL(md_bitmap_startwrite);
1459 void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1460 unsigned long sectors, int success, int behind)
1462 if (!bitmap)
1463 return;
1464 if (behind) {
1465 if (atomic_dec_and_test(&bitmap->behind_writes))
1466 wake_up(&bitmap->behind_wait);
1467 pr_debug("dec write-behind count %d/%lu\n",
1468 atomic_read(&bitmap->behind_writes),
1469 bitmap->mddev->bitmap_info.max_write_behind);
1472 while (sectors) {
1473 sector_t blocks;
1474 unsigned long flags;
1475 bitmap_counter_t *bmc;
1477 spin_lock_irqsave(&bitmap->counts.lock, flags);
1478 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1479 if (!bmc) {
1480 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1481 return;
1484 if (success && !bitmap->mddev->degraded &&
1485 bitmap->events_cleared < bitmap->mddev->events) {
1486 bitmap->events_cleared = bitmap->mddev->events;
1487 bitmap->need_sync = 1;
1488 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1491 if (!success && !NEEDED(*bmc))
1492 *bmc |= NEEDED_MASK;
1494 if (COUNTER(*bmc) == COUNTER_MAX)
1495 wake_up(&bitmap->overflow_wait);
1497 (*bmc)--;
1498 if (*bmc <= 2) {
1499 md_bitmap_set_pending(&bitmap->counts, offset);
1500 bitmap->allclean = 0;
1502 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1503 offset += blocks;
1504 if (sectors > blocks)
1505 sectors -= blocks;
1506 else
1507 sectors = 0;
1510 EXPORT_SYMBOL(md_bitmap_endwrite);
1512 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1513 int degraded)
1515 bitmap_counter_t *bmc;
1516 int rv;
1517 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1518 *blocks = 1024;
1519 return 1; /* always resync if no bitmap */
1521 spin_lock_irq(&bitmap->counts.lock);
1522 bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1523 rv = 0;
1524 if (bmc) {
1525 /* locked */
1526 if (RESYNC(*bmc))
1527 rv = 1;
1528 else if (NEEDED(*bmc)) {
1529 rv = 1;
1530 if (!degraded) { /* don't set/clear bits if degraded */
1531 *bmc |= RESYNC_MASK;
1532 *bmc &= ~NEEDED_MASK;
1536 spin_unlock_irq(&bitmap->counts.lock);
1537 return rv;
1540 int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1541 int degraded)
1543 /* bitmap_start_sync must always report on multiples of whole
1544 * pages, otherwise resync (which is very PAGE_SIZE based) will
1545 * get confused.
1546 * So call __bitmap_start_sync repeatedly (if needed) until
1547 * At least PAGE_SIZE>>9 blocks are covered.
1548 * Return the 'or' of the result.
1550 int rv = 0;
1551 sector_t blocks1;
1553 *blocks = 0;
1554 while (*blocks < (PAGE_SIZE>>9)) {
1555 rv |= __bitmap_start_sync(bitmap, offset,
1556 &blocks1, degraded);
1557 offset += blocks1;
1558 *blocks += blocks1;
1560 return rv;
1562 EXPORT_SYMBOL(md_bitmap_start_sync);
1564 void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1566 bitmap_counter_t *bmc;
1567 unsigned long flags;
1569 if (bitmap == NULL) {
1570 *blocks = 1024;
1571 return;
1573 spin_lock_irqsave(&bitmap->counts.lock, flags);
1574 bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1575 if (bmc == NULL)
1576 goto unlock;
1577 /* locked */
1578 if (RESYNC(*bmc)) {
1579 *bmc &= ~RESYNC_MASK;
1581 if (!NEEDED(*bmc) && aborted)
1582 *bmc |= NEEDED_MASK;
1583 else {
1584 if (*bmc <= 2) {
1585 md_bitmap_set_pending(&bitmap->counts, offset);
1586 bitmap->allclean = 0;
1590 unlock:
1591 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1593 EXPORT_SYMBOL(md_bitmap_end_sync);
1595 void md_bitmap_close_sync(struct bitmap *bitmap)
1597 /* Sync has finished, and any bitmap chunks that weren't synced
1598 * properly have been aborted. It remains to us to clear the
1599 * RESYNC bit wherever it is still on
1601 sector_t sector = 0;
1602 sector_t blocks;
1603 if (!bitmap)
1604 return;
1605 while (sector < bitmap->mddev->resync_max_sectors) {
1606 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1607 sector += blocks;
1610 EXPORT_SYMBOL(md_bitmap_close_sync);
1612 void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1614 sector_t s = 0;
1615 sector_t blocks;
1617 if (!bitmap)
1618 return;
1619 if (sector == 0) {
1620 bitmap->last_end_sync = jiffies;
1621 return;
1623 if (!force && time_before(jiffies, (bitmap->last_end_sync
1624 + bitmap->mddev->bitmap_info.daemon_sleep)))
1625 return;
1626 wait_event(bitmap->mddev->recovery_wait,
1627 atomic_read(&bitmap->mddev->recovery_active) == 0);
1629 bitmap->mddev->curr_resync_completed = sector;
1630 set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1631 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1632 s = 0;
1633 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1634 md_bitmap_end_sync(bitmap, s, &blocks, 0);
1635 s += blocks;
1637 bitmap->last_end_sync = jiffies;
1638 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1640 EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1642 void md_bitmap_sync_with_cluster(struct mddev *mddev,
1643 sector_t old_lo, sector_t old_hi,
1644 sector_t new_lo, sector_t new_hi)
1646 struct bitmap *bitmap = mddev->bitmap;
1647 sector_t sector, blocks = 0;
1649 for (sector = old_lo; sector < new_lo; ) {
1650 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1651 sector += blocks;
1653 WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1655 for (sector = old_hi; sector < new_hi; ) {
1656 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1657 sector += blocks;
1659 WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1661 EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1663 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1665 /* For each chunk covered by any of these sectors, set the
1666 * counter to 2 and possibly set resync_needed. They should all
1667 * be 0 at this point
1670 sector_t secs;
1671 bitmap_counter_t *bmc;
1672 spin_lock_irq(&bitmap->counts.lock);
1673 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1674 if (!bmc) {
1675 spin_unlock_irq(&bitmap->counts.lock);
1676 return;
1678 if (!*bmc) {
1679 *bmc = 2;
1680 md_bitmap_count_page(&bitmap->counts, offset, 1);
1681 md_bitmap_set_pending(&bitmap->counts, offset);
1682 bitmap->allclean = 0;
1684 if (needed)
1685 *bmc |= NEEDED_MASK;
1686 spin_unlock_irq(&bitmap->counts.lock);
1689 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1690 void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1692 unsigned long chunk;
1694 for (chunk = s; chunk <= e; chunk++) {
1695 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1696 md_bitmap_set_memory_bits(bitmap, sec, 1);
1697 md_bitmap_file_set_bit(bitmap, sec);
1698 if (sec < bitmap->mddev->recovery_cp)
1699 /* We are asserting that the array is dirty,
1700 * so move the recovery_cp address back so
1701 * that it is obvious that it is dirty
1703 bitmap->mddev->recovery_cp = sec;
1708 * flush out any pending updates
1710 void md_bitmap_flush(struct mddev *mddev)
1712 struct bitmap *bitmap = mddev->bitmap;
1713 long sleep;
1715 if (!bitmap) /* there was no bitmap */
1716 return;
1718 /* run the daemon_work three time to ensure everything is flushed
1719 * that can be
1721 sleep = mddev->bitmap_info.daemon_sleep * 2;
1722 bitmap->daemon_lastrun -= sleep;
1723 md_bitmap_daemon_work(mddev);
1724 bitmap->daemon_lastrun -= sleep;
1725 md_bitmap_daemon_work(mddev);
1726 bitmap->daemon_lastrun -= sleep;
1727 md_bitmap_daemon_work(mddev);
1728 md_bitmap_update_sb(bitmap);
1732 * free memory that was allocated
1734 void md_bitmap_free(struct bitmap *bitmap)
1736 unsigned long k, pages;
1737 struct bitmap_page *bp;
1739 if (!bitmap) /* there was no bitmap */
1740 return;
1742 if (bitmap->sysfs_can_clear)
1743 sysfs_put(bitmap->sysfs_can_clear);
1745 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1746 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1747 md_cluster_stop(bitmap->mddev);
1749 /* Shouldn't be needed - but just in case.... */
1750 wait_event(bitmap->write_wait,
1751 atomic_read(&bitmap->pending_writes) == 0);
1753 /* release the bitmap file */
1754 md_bitmap_file_unmap(&bitmap->storage);
1756 bp = bitmap->counts.bp;
1757 pages = bitmap->counts.pages;
1759 /* free all allocated memory */
1761 if (bp) /* deallocate the page memory */
1762 for (k = 0; k < pages; k++)
1763 if (bp[k].map && !bp[k].hijacked)
1764 kfree(bp[k].map);
1765 kfree(bp);
1766 kfree(bitmap);
1768 EXPORT_SYMBOL(md_bitmap_free);
1770 void md_bitmap_wait_behind_writes(struct mddev *mddev)
1772 struct bitmap *bitmap = mddev->bitmap;
1774 /* wait for behind writes to complete */
1775 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1776 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1777 mdname(mddev));
1778 /* need to kick something here to make sure I/O goes? */
1779 wait_event(bitmap->behind_wait,
1780 atomic_read(&bitmap->behind_writes) == 0);
1784 void md_bitmap_destroy(struct mddev *mddev)
1786 struct bitmap *bitmap = mddev->bitmap;
1788 if (!bitmap) /* there was no bitmap */
1789 return;
1791 md_bitmap_wait_behind_writes(mddev);
1792 if (!mddev->serialize_policy)
1793 mddev_destroy_serial_pool(mddev, NULL, true);
1795 mutex_lock(&mddev->bitmap_info.mutex);
1796 spin_lock(&mddev->lock);
1797 mddev->bitmap = NULL; /* disconnect from the md device */
1798 spin_unlock(&mddev->lock);
1799 mutex_unlock(&mddev->bitmap_info.mutex);
1800 if (mddev->thread)
1801 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1803 md_bitmap_free(bitmap);
1807 * initialize the bitmap structure
1808 * if this returns an error, bitmap_destroy must be called to do clean up
1809 * once mddev->bitmap is set
1811 struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1813 struct bitmap *bitmap;
1814 sector_t blocks = mddev->resync_max_sectors;
1815 struct file *file = mddev->bitmap_info.file;
1816 int err;
1817 struct kernfs_node *bm = NULL;
1819 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1821 BUG_ON(file && mddev->bitmap_info.offset);
1823 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1824 pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1825 mdname(mddev));
1826 return ERR_PTR(-EBUSY);
1829 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1830 if (!bitmap)
1831 return ERR_PTR(-ENOMEM);
1833 spin_lock_init(&bitmap->counts.lock);
1834 atomic_set(&bitmap->pending_writes, 0);
1835 init_waitqueue_head(&bitmap->write_wait);
1836 init_waitqueue_head(&bitmap->overflow_wait);
1837 init_waitqueue_head(&bitmap->behind_wait);
1839 bitmap->mddev = mddev;
1840 bitmap->cluster_slot = slot;
1842 if (mddev->kobj.sd)
1843 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1844 if (bm) {
1845 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1846 sysfs_put(bm);
1847 } else
1848 bitmap->sysfs_can_clear = NULL;
1850 bitmap->storage.file = file;
1851 if (file) {
1852 get_file(file);
1853 /* As future accesses to this file will use bmap,
1854 * and bypass the page cache, we must sync the file
1855 * first.
1857 vfs_fsync(file, 1);
1859 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1860 if (!mddev->bitmap_info.external) {
1862 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1863 * instructing us to create a new on-disk bitmap instance.
1865 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1866 err = md_bitmap_new_disk_sb(bitmap);
1867 else
1868 err = md_bitmap_read_sb(bitmap);
1869 } else {
1870 err = 0;
1871 if (mddev->bitmap_info.chunksize == 0 ||
1872 mddev->bitmap_info.daemon_sleep == 0)
1873 /* chunksize and time_base need to be
1874 * set first. */
1875 err = -EINVAL;
1877 if (err)
1878 goto error;
1880 bitmap->daemon_lastrun = jiffies;
1881 err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1882 if (err)
1883 goto error;
1885 pr_debug("created bitmap (%lu pages) for device %s\n",
1886 bitmap->counts.pages, bmname(bitmap));
1888 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1889 if (err)
1890 goto error;
1892 return bitmap;
1893 error:
1894 md_bitmap_free(bitmap);
1895 return ERR_PTR(err);
1898 int md_bitmap_load(struct mddev *mddev)
1900 int err = 0;
1901 sector_t start = 0;
1902 sector_t sector = 0;
1903 struct bitmap *bitmap = mddev->bitmap;
1904 struct md_rdev *rdev;
1906 if (!bitmap)
1907 goto out;
1909 rdev_for_each(rdev, mddev)
1910 mddev_create_serial_pool(mddev, rdev, true);
1912 if (mddev_is_clustered(mddev))
1913 md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1915 /* Clear out old bitmap info first: Either there is none, or we
1916 * are resuming after someone else has possibly changed things,
1917 * so we should forget old cached info.
1918 * All chunks should be clean, but some might need_sync.
1920 while (sector < mddev->resync_max_sectors) {
1921 sector_t blocks;
1922 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1923 sector += blocks;
1925 md_bitmap_close_sync(bitmap);
1927 if (mddev->degraded == 0
1928 || bitmap->events_cleared == mddev->events)
1929 /* no need to keep dirty bits to optimise a
1930 * re-add of a missing device */
1931 start = mddev->recovery_cp;
1933 mutex_lock(&mddev->bitmap_info.mutex);
1934 err = md_bitmap_init_from_disk(bitmap, start);
1935 mutex_unlock(&mddev->bitmap_info.mutex);
1937 if (err)
1938 goto out;
1939 clear_bit(BITMAP_STALE, &bitmap->flags);
1941 /* Kick recovery in case any bits were set */
1942 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1944 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1945 md_wakeup_thread(mddev->thread);
1947 md_bitmap_update_sb(bitmap);
1949 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1950 err = -EIO;
1951 out:
1952 return err;
1954 EXPORT_SYMBOL_GPL(md_bitmap_load);
1956 struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot)
1958 int rv = 0;
1959 struct bitmap *bitmap;
1961 bitmap = md_bitmap_create(mddev, slot);
1962 if (IS_ERR(bitmap)) {
1963 rv = PTR_ERR(bitmap);
1964 return ERR_PTR(rv);
1967 rv = md_bitmap_init_from_disk(bitmap, 0);
1968 if (rv) {
1969 md_bitmap_free(bitmap);
1970 return ERR_PTR(rv);
1973 return bitmap;
1975 EXPORT_SYMBOL(get_bitmap_from_slot);
1977 /* Loads the bitmap associated with slot and copies the resync information
1978 * to our bitmap
1980 int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
1981 sector_t *low, sector_t *high, bool clear_bits)
1983 int rv = 0, i, j;
1984 sector_t block, lo = 0, hi = 0;
1985 struct bitmap_counts *counts;
1986 struct bitmap *bitmap;
1988 bitmap = get_bitmap_from_slot(mddev, slot);
1989 if (IS_ERR(bitmap)) {
1990 pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
1991 return -1;
1994 counts = &bitmap->counts;
1995 for (j = 0; j < counts->chunks; j++) {
1996 block = (sector_t)j << counts->chunkshift;
1997 if (md_bitmap_file_test_bit(bitmap, block)) {
1998 if (!lo)
1999 lo = block;
2000 hi = block;
2001 md_bitmap_file_clear_bit(bitmap, block);
2002 md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2003 md_bitmap_file_set_bit(mddev->bitmap, block);
2007 if (clear_bits) {
2008 md_bitmap_update_sb(bitmap);
2009 /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2010 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2011 for (i = 0; i < bitmap->storage.file_pages; i++)
2012 if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2013 set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2014 md_bitmap_unplug(bitmap);
2016 md_bitmap_unplug(mddev->bitmap);
2017 *low = lo;
2018 *high = hi;
2020 return rv;
2022 EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2025 void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2027 unsigned long chunk_kb;
2028 struct bitmap_counts *counts;
2030 if (!bitmap)
2031 return;
2033 counts = &bitmap->counts;
2035 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2036 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2037 "%lu%s chunk",
2038 counts->pages - counts->missing_pages,
2039 counts->pages,
2040 (counts->pages - counts->missing_pages)
2041 << (PAGE_SHIFT - 10),
2042 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2043 chunk_kb ? "KB" : "B");
2044 if (bitmap->storage.file) {
2045 seq_printf(seq, ", file: ");
2046 seq_file_path(seq, bitmap->storage.file, " \t\n");
2049 seq_printf(seq, "\n");
2052 int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2053 int chunksize, int init)
2055 /* If chunk_size is 0, choose an appropriate chunk size.
2056 * Then possibly allocate new storage space.
2057 * Then quiesce, copy bits, replace bitmap, and re-start
2059 * This function is called both to set up the initial bitmap
2060 * and to resize the bitmap while the array is active.
2061 * If this happens as a result of the array being resized,
2062 * chunksize will be zero, and we need to choose a suitable
2063 * chunksize, otherwise we use what we are given.
2065 struct bitmap_storage store;
2066 struct bitmap_counts old_counts;
2067 unsigned long chunks;
2068 sector_t block;
2069 sector_t old_blocks, new_blocks;
2070 int chunkshift;
2071 int ret = 0;
2072 long pages;
2073 struct bitmap_page *new_bp;
2075 if (bitmap->storage.file && !init) {
2076 pr_info("md: cannot resize file-based bitmap\n");
2077 return -EINVAL;
2080 if (chunksize == 0) {
2081 /* If there is enough space, leave the chunk size unchanged,
2082 * else increase by factor of two until there is enough space.
2084 long bytes;
2085 long space = bitmap->mddev->bitmap_info.space;
2087 if (space == 0) {
2088 /* We don't know how much space there is, so limit
2089 * to current size - in sectors.
2091 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2092 if (!bitmap->mddev->bitmap_info.external)
2093 bytes += sizeof(bitmap_super_t);
2094 space = DIV_ROUND_UP(bytes, 512);
2095 bitmap->mddev->bitmap_info.space = space;
2097 chunkshift = bitmap->counts.chunkshift;
2098 chunkshift--;
2099 do {
2100 /* 'chunkshift' is shift from block size to chunk size */
2101 chunkshift++;
2102 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2103 bytes = DIV_ROUND_UP(chunks, 8);
2104 if (!bitmap->mddev->bitmap_info.external)
2105 bytes += sizeof(bitmap_super_t);
2106 } while (bytes > (space << 9));
2107 } else
2108 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2110 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2111 memset(&store, 0, sizeof(store));
2112 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2113 ret = md_bitmap_storage_alloc(&store, chunks,
2114 !bitmap->mddev->bitmap_info.external,
2115 mddev_is_clustered(bitmap->mddev)
2116 ? bitmap->cluster_slot : 0);
2117 if (ret) {
2118 md_bitmap_file_unmap(&store);
2119 goto err;
2122 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2124 new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2125 ret = -ENOMEM;
2126 if (!new_bp) {
2127 md_bitmap_file_unmap(&store);
2128 goto err;
2131 if (!init)
2132 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2134 store.file = bitmap->storage.file;
2135 bitmap->storage.file = NULL;
2137 if (store.sb_page && bitmap->storage.sb_page)
2138 memcpy(page_address(store.sb_page),
2139 page_address(bitmap->storage.sb_page),
2140 sizeof(bitmap_super_t));
2141 spin_lock_irq(&bitmap->counts.lock);
2142 md_bitmap_file_unmap(&bitmap->storage);
2143 bitmap->storage = store;
2145 old_counts = bitmap->counts;
2146 bitmap->counts.bp = new_bp;
2147 bitmap->counts.pages = pages;
2148 bitmap->counts.missing_pages = pages;
2149 bitmap->counts.chunkshift = chunkshift;
2150 bitmap->counts.chunks = chunks;
2151 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2152 BITMAP_BLOCK_SHIFT);
2154 blocks = min(old_counts.chunks << old_counts.chunkshift,
2155 chunks << chunkshift);
2157 /* For cluster raid, need to pre-allocate bitmap */
2158 if (mddev_is_clustered(bitmap->mddev)) {
2159 unsigned long page;
2160 for (page = 0; page < pages; page++) {
2161 ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2162 if (ret) {
2163 unsigned long k;
2165 /* deallocate the page memory */
2166 for (k = 0; k < page; k++) {
2167 kfree(new_bp[k].map);
2169 kfree(new_bp);
2171 /* restore some fields from old_counts */
2172 bitmap->counts.bp = old_counts.bp;
2173 bitmap->counts.pages = old_counts.pages;
2174 bitmap->counts.missing_pages = old_counts.pages;
2175 bitmap->counts.chunkshift = old_counts.chunkshift;
2176 bitmap->counts.chunks = old_counts.chunks;
2177 bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift +
2178 BITMAP_BLOCK_SHIFT);
2179 blocks = old_counts.chunks << old_counts.chunkshift;
2180 pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2181 break;
2182 } else
2183 bitmap->counts.bp[page].count += 1;
2187 for (block = 0; block < blocks; ) {
2188 bitmap_counter_t *bmc_old, *bmc_new;
2189 int set;
2191 bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2192 set = bmc_old && NEEDED(*bmc_old);
2194 if (set) {
2195 bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2196 if (*bmc_new == 0) {
2197 /* need to set on-disk bits too. */
2198 sector_t end = block + new_blocks;
2199 sector_t start = block >> chunkshift;
2200 start <<= chunkshift;
2201 while (start < end) {
2202 md_bitmap_file_set_bit(bitmap, block);
2203 start += 1 << chunkshift;
2205 *bmc_new = 2;
2206 md_bitmap_count_page(&bitmap->counts, block, 1);
2207 md_bitmap_set_pending(&bitmap->counts, block);
2209 *bmc_new |= NEEDED_MASK;
2210 if (new_blocks < old_blocks)
2211 old_blocks = new_blocks;
2213 block += old_blocks;
2216 if (bitmap->counts.bp != old_counts.bp) {
2217 unsigned long k;
2218 for (k = 0; k < old_counts.pages; k++)
2219 if (!old_counts.bp[k].hijacked)
2220 kfree(old_counts.bp[k].map);
2221 kfree(old_counts.bp);
2224 if (!init) {
2225 int i;
2226 while (block < (chunks << chunkshift)) {
2227 bitmap_counter_t *bmc;
2228 bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2229 if (bmc) {
2230 /* new space. It needs to be resynced, so
2231 * we set NEEDED_MASK.
2233 if (*bmc == 0) {
2234 *bmc = NEEDED_MASK | 2;
2235 md_bitmap_count_page(&bitmap->counts, block, 1);
2236 md_bitmap_set_pending(&bitmap->counts, block);
2239 block += new_blocks;
2241 for (i = 0; i < bitmap->storage.file_pages; i++)
2242 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2244 spin_unlock_irq(&bitmap->counts.lock);
2246 if (!init) {
2247 md_bitmap_unplug(bitmap);
2248 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2250 ret = 0;
2251 err:
2252 return ret;
2254 EXPORT_SYMBOL_GPL(md_bitmap_resize);
2256 static ssize_t
2257 location_show(struct mddev *mddev, char *page)
2259 ssize_t len;
2260 if (mddev->bitmap_info.file)
2261 len = sprintf(page, "file");
2262 else if (mddev->bitmap_info.offset)
2263 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2264 else
2265 len = sprintf(page, "none");
2266 len += sprintf(page+len, "\n");
2267 return len;
2270 static ssize_t
2271 location_store(struct mddev *mddev, const char *buf, size_t len)
2273 int rv;
2275 rv = mddev_lock(mddev);
2276 if (rv)
2277 return rv;
2278 if (mddev->pers) {
2279 if (!mddev->pers->quiesce) {
2280 rv = -EBUSY;
2281 goto out;
2283 if (mddev->recovery || mddev->sync_thread) {
2284 rv = -EBUSY;
2285 goto out;
2289 if (mddev->bitmap || mddev->bitmap_info.file ||
2290 mddev->bitmap_info.offset) {
2291 /* bitmap already configured. Only option is to clear it */
2292 if (strncmp(buf, "none", 4) != 0) {
2293 rv = -EBUSY;
2294 goto out;
2296 if (mddev->pers) {
2297 mddev_suspend(mddev);
2298 md_bitmap_destroy(mddev);
2299 mddev_resume(mddev);
2301 mddev->bitmap_info.offset = 0;
2302 if (mddev->bitmap_info.file) {
2303 struct file *f = mddev->bitmap_info.file;
2304 mddev->bitmap_info.file = NULL;
2305 fput(f);
2307 } else {
2308 /* No bitmap, OK to set a location */
2309 long long offset;
2310 if (strncmp(buf, "none", 4) == 0)
2311 /* nothing to be done */;
2312 else if (strncmp(buf, "file:", 5) == 0) {
2313 /* Not supported yet */
2314 rv = -EINVAL;
2315 goto out;
2316 } else {
2317 if (buf[0] == '+')
2318 rv = kstrtoll(buf+1, 10, &offset);
2319 else
2320 rv = kstrtoll(buf, 10, &offset);
2321 if (rv)
2322 goto out;
2323 if (offset == 0) {
2324 rv = -EINVAL;
2325 goto out;
2327 if (mddev->bitmap_info.external == 0 &&
2328 mddev->major_version == 0 &&
2329 offset != mddev->bitmap_info.default_offset) {
2330 rv = -EINVAL;
2331 goto out;
2333 mddev->bitmap_info.offset = offset;
2334 if (mddev->pers) {
2335 struct bitmap *bitmap;
2336 bitmap = md_bitmap_create(mddev, -1);
2337 mddev_suspend(mddev);
2338 if (IS_ERR(bitmap))
2339 rv = PTR_ERR(bitmap);
2340 else {
2341 mddev->bitmap = bitmap;
2342 rv = md_bitmap_load(mddev);
2343 if (rv)
2344 mddev->bitmap_info.offset = 0;
2346 if (rv) {
2347 md_bitmap_destroy(mddev);
2348 mddev_resume(mddev);
2349 goto out;
2351 mddev_resume(mddev);
2355 if (!mddev->external) {
2356 /* Ensure new bitmap info is stored in
2357 * metadata promptly.
2359 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2360 md_wakeup_thread(mddev->thread);
2362 rv = 0;
2363 out:
2364 mddev_unlock(mddev);
2365 if (rv)
2366 return rv;
2367 return len;
2370 static struct md_sysfs_entry bitmap_location =
2371 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2373 /* 'bitmap/space' is the space available at 'location' for the
2374 * bitmap. This allows the kernel to know when it is safe to
2375 * resize the bitmap to match a resized array.
2377 static ssize_t
2378 space_show(struct mddev *mddev, char *page)
2380 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2383 static ssize_t
2384 space_store(struct mddev *mddev, const char *buf, size_t len)
2386 unsigned long sectors;
2387 int rv;
2389 rv = kstrtoul(buf, 10, &sectors);
2390 if (rv)
2391 return rv;
2393 if (sectors == 0)
2394 return -EINVAL;
2396 if (mddev->bitmap &&
2397 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2398 return -EFBIG; /* Bitmap is too big for this small space */
2400 /* could make sure it isn't too big, but that isn't really
2401 * needed - user-space should be careful.
2403 mddev->bitmap_info.space = sectors;
2404 return len;
2407 static struct md_sysfs_entry bitmap_space =
2408 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2410 static ssize_t
2411 timeout_show(struct mddev *mddev, char *page)
2413 ssize_t len;
2414 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2415 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2417 len = sprintf(page, "%lu", secs);
2418 if (jifs)
2419 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2420 len += sprintf(page+len, "\n");
2421 return len;
2424 static ssize_t
2425 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2427 /* timeout can be set at any time */
2428 unsigned long timeout;
2429 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2430 if (rv)
2431 return rv;
2433 /* just to make sure we don't overflow... */
2434 if (timeout >= LONG_MAX / HZ)
2435 return -EINVAL;
2437 timeout = timeout * HZ / 10000;
2439 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2440 timeout = MAX_SCHEDULE_TIMEOUT-1;
2441 if (timeout < 1)
2442 timeout = 1;
2443 mddev->bitmap_info.daemon_sleep = timeout;
2444 if (mddev->thread) {
2445 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2446 * the bitmap is all clean and we don't need to
2447 * adjust the timeout right now
2449 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2450 mddev->thread->timeout = timeout;
2451 md_wakeup_thread(mddev->thread);
2454 return len;
2457 static struct md_sysfs_entry bitmap_timeout =
2458 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2460 static ssize_t
2461 backlog_show(struct mddev *mddev, char *page)
2463 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2466 static ssize_t
2467 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2469 unsigned long backlog;
2470 unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2471 int rv = kstrtoul(buf, 10, &backlog);
2472 if (rv)
2473 return rv;
2474 if (backlog > COUNTER_MAX)
2475 return -EINVAL;
2476 mddev->bitmap_info.max_write_behind = backlog;
2477 if (!backlog && mddev->serial_info_pool) {
2478 /* serial_info_pool is not needed if backlog is zero */
2479 if (!mddev->serialize_policy)
2480 mddev_destroy_serial_pool(mddev, NULL, false);
2481 } else if (backlog && !mddev->serial_info_pool) {
2482 /* serial_info_pool is needed since backlog is not zero */
2483 struct md_rdev *rdev;
2485 rdev_for_each(rdev, mddev)
2486 mddev_create_serial_pool(mddev, rdev, false);
2488 if (old_mwb != backlog)
2489 md_bitmap_update_sb(mddev->bitmap);
2490 return len;
2493 static struct md_sysfs_entry bitmap_backlog =
2494 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2496 static ssize_t
2497 chunksize_show(struct mddev *mddev, char *page)
2499 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2502 static ssize_t
2503 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2505 /* Can only be changed when no bitmap is active */
2506 int rv;
2507 unsigned long csize;
2508 if (mddev->bitmap)
2509 return -EBUSY;
2510 rv = kstrtoul(buf, 10, &csize);
2511 if (rv)
2512 return rv;
2513 if (csize < 512 ||
2514 !is_power_of_2(csize))
2515 return -EINVAL;
2516 mddev->bitmap_info.chunksize = csize;
2517 return len;
2520 static struct md_sysfs_entry bitmap_chunksize =
2521 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2523 static ssize_t metadata_show(struct mddev *mddev, char *page)
2525 if (mddev_is_clustered(mddev))
2526 return sprintf(page, "clustered\n");
2527 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2528 ? "external" : "internal"));
2531 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2533 if (mddev->bitmap ||
2534 mddev->bitmap_info.file ||
2535 mddev->bitmap_info.offset)
2536 return -EBUSY;
2537 if (strncmp(buf, "external", 8) == 0)
2538 mddev->bitmap_info.external = 1;
2539 else if ((strncmp(buf, "internal", 8) == 0) ||
2540 (strncmp(buf, "clustered", 9) == 0))
2541 mddev->bitmap_info.external = 0;
2542 else
2543 return -EINVAL;
2544 return len;
2547 static struct md_sysfs_entry bitmap_metadata =
2548 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2550 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2552 int len;
2553 spin_lock(&mddev->lock);
2554 if (mddev->bitmap)
2555 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2556 "false" : "true"));
2557 else
2558 len = sprintf(page, "\n");
2559 spin_unlock(&mddev->lock);
2560 return len;
2563 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2565 if (mddev->bitmap == NULL)
2566 return -ENOENT;
2567 if (strncmp(buf, "false", 5) == 0)
2568 mddev->bitmap->need_sync = 1;
2569 else if (strncmp(buf, "true", 4) == 0) {
2570 if (mddev->degraded)
2571 return -EBUSY;
2572 mddev->bitmap->need_sync = 0;
2573 } else
2574 return -EINVAL;
2575 return len;
2578 static struct md_sysfs_entry bitmap_can_clear =
2579 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2581 static ssize_t
2582 behind_writes_used_show(struct mddev *mddev, char *page)
2584 ssize_t ret;
2585 spin_lock(&mddev->lock);
2586 if (mddev->bitmap == NULL)
2587 ret = sprintf(page, "0\n");
2588 else
2589 ret = sprintf(page, "%lu\n",
2590 mddev->bitmap->behind_writes_used);
2591 spin_unlock(&mddev->lock);
2592 return ret;
2595 static ssize_t
2596 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2598 if (mddev->bitmap)
2599 mddev->bitmap->behind_writes_used = 0;
2600 return len;
2603 static struct md_sysfs_entry max_backlog_used =
2604 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2605 behind_writes_used_show, behind_writes_used_reset);
2607 static struct attribute *md_bitmap_attrs[] = {
2608 &bitmap_location.attr,
2609 &bitmap_space.attr,
2610 &bitmap_timeout.attr,
2611 &bitmap_backlog.attr,
2612 &bitmap_chunksize.attr,
2613 &bitmap_metadata.attr,
2614 &bitmap_can_clear.attr,
2615 &max_backlog_used.attr,
2616 NULL
2618 struct attribute_group md_bitmap_group = {
2619 .name = "bitmap",
2620 .attrs = md_bitmap_attrs,