Merge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
[linux/fpc-iii.git] / drivers / md / md-bitmap.c
blob95a5f3757fa30854ab46227a1bc92b413dfd8e10
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 static void free_buffers(struct page *page)
329 struct buffer_head *bh;
331 if (!PagePrivate(page))
332 return;
334 bh = page_buffers(page);
335 while (bh) {
336 struct buffer_head *next = bh->b_this_page;
337 free_buffer_head(bh);
338 bh = next;
340 detach_page_private(page);
341 put_page(page);
344 /* read a page from a file.
345 * We both read the page, and attach buffers to the page to record the
346 * address of each block (using bmap). These addresses will be used
347 * to write the block later, completely bypassing the filesystem.
348 * This usage is similar to how swap files are handled, and allows us
349 * to write to a file with no concerns of memory allocation failing.
351 static int read_page(struct file *file, unsigned long index,
352 struct bitmap *bitmap,
353 unsigned long count,
354 struct page *page)
356 int ret = 0;
357 struct inode *inode = file_inode(file);
358 struct buffer_head *bh;
359 sector_t block, blk_cur;
361 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
362 (unsigned long long)index << PAGE_SHIFT);
364 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, false);
365 if (!bh) {
366 ret = -ENOMEM;
367 goto out;
369 attach_page_private(page, bh);
370 blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
371 while (bh) {
372 block = blk_cur;
374 if (count == 0)
375 bh->b_blocknr = 0;
376 else {
377 ret = bmap(inode, &block);
378 if (ret || !block) {
379 ret = -EINVAL;
380 bh->b_blocknr = 0;
381 goto out;
384 bh->b_blocknr = block;
385 bh->b_bdev = inode->i_sb->s_bdev;
386 if (count < (1<<inode->i_blkbits))
387 count = 0;
388 else
389 count -= (1<<inode->i_blkbits);
391 bh->b_end_io = end_bitmap_write;
392 bh->b_private = bitmap;
393 atomic_inc(&bitmap->pending_writes);
394 set_buffer_locked(bh);
395 set_buffer_mapped(bh);
396 submit_bh(REQ_OP_READ, 0, bh);
398 blk_cur++;
399 bh = bh->b_this_page;
401 page->index = index;
403 wait_event(bitmap->write_wait,
404 atomic_read(&bitmap->pending_writes)==0);
405 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
406 ret = -EIO;
407 out:
408 if (ret)
409 pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
410 (int)PAGE_SIZE,
411 (unsigned long long)index << PAGE_SHIFT,
412 ret);
413 return ret;
417 * bitmap file superblock operations
421 * md_bitmap_wait_writes() should be called before writing any bitmap
422 * blocks, to ensure previous writes, particularly from
423 * md_bitmap_daemon_work(), have completed.
425 static void md_bitmap_wait_writes(struct bitmap *bitmap)
427 if (bitmap->storage.file)
428 wait_event(bitmap->write_wait,
429 atomic_read(&bitmap->pending_writes)==0);
430 else
431 /* Note that we ignore the return value. The writes
432 * might have failed, but that would just mean that
433 * some bits which should be cleared haven't been,
434 * which is safe. The relevant bitmap blocks will
435 * probably get written again, but there is no great
436 * loss if they aren't.
438 md_super_wait(bitmap->mddev);
442 /* update the event counter and sync the superblock to disk */
443 void md_bitmap_update_sb(struct bitmap *bitmap)
445 bitmap_super_t *sb;
447 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
448 return;
449 if (bitmap->mddev->bitmap_info.external)
450 return;
451 if (!bitmap->storage.sb_page) /* no superblock */
452 return;
453 sb = kmap_atomic(bitmap->storage.sb_page);
454 sb->events = cpu_to_le64(bitmap->mddev->events);
455 if (bitmap->mddev->events < bitmap->events_cleared)
456 /* rocking back to read-only */
457 bitmap->events_cleared = bitmap->mddev->events;
458 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
460 * clear BITMAP_WRITE_ERROR bit to protect against the case that
461 * a bitmap write error occurred but the later writes succeeded.
463 sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
464 /* Just in case these have been changed via sysfs: */
465 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
466 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
467 /* This might have been changed by a reshape */
468 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
469 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
470 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
471 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
472 bitmap_info.space);
473 kunmap_atomic(sb);
474 write_page(bitmap, bitmap->storage.sb_page, 1);
476 EXPORT_SYMBOL(md_bitmap_update_sb);
478 /* print out the bitmap file superblock */
479 void md_bitmap_print_sb(struct bitmap *bitmap)
481 bitmap_super_t *sb;
483 if (!bitmap || !bitmap->storage.sb_page)
484 return;
485 sb = kmap_atomic(bitmap->storage.sb_page);
486 pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
487 pr_debug(" magic: %08x\n", le32_to_cpu(sb->magic));
488 pr_debug(" version: %d\n", le32_to_cpu(sb->version));
489 pr_debug(" uuid: %08x.%08x.%08x.%08x\n",
490 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
491 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
492 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
493 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
494 pr_debug(" events: %llu\n",
495 (unsigned long long) le64_to_cpu(sb->events));
496 pr_debug("events cleared: %llu\n",
497 (unsigned long long) le64_to_cpu(sb->events_cleared));
498 pr_debug(" state: %08x\n", le32_to_cpu(sb->state));
499 pr_debug(" chunksize: %d B\n", le32_to_cpu(sb->chunksize));
500 pr_debug(" daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
501 pr_debug(" sync size: %llu KB\n",
502 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
503 pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind));
504 kunmap_atomic(sb);
508 * bitmap_new_disk_sb
509 * @bitmap
511 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
512 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
513 * This function verifies 'bitmap_info' and populates the on-disk bitmap
514 * structure, which is to be written to disk.
516 * Returns: 0 on success, -Exxx on error
518 static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
520 bitmap_super_t *sb;
521 unsigned long chunksize, daemon_sleep, write_behind;
523 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
524 if (bitmap->storage.sb_page == NULL)
525 return -ENOMEM;
526 bitmap->storage.sb_page->index = 0;
528 sb = kmap_atomic(bitmap->storage.sb_page);
530 sb->magic = cpu_to_le32(BITMAP_MAGIC);
531 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
533 chunksize = bitmap->mddev->bitmap_info.chunksize;
534 BUG_ON(!chunksize);
535 if (!is_power_of_2(chunksize)) {
536 kunmap_atomic(sb);
537 pr_warn("bitmap chunksize not a power of 2\n");
538 return -EINVAL;
540 sb->chunksize = cpu_to_le32(chunksize);
542 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
543 if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
544 pr_debug("Choosing daemon_sleep default (5 sec)\n");
545 daemon_sleep = 5 * HZ;
547 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
548 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
551 * FIXME: write_behind for RAID1. If not specified, what
552 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
554 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
555 if (write_behind > COUNTER_MAX)
556 write_behind = COUNTER_MAX / 2;
557 sb->write_behind = cpu_to_le32(write_behind);
558 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
560 /* keep the array size field of the bitmap superblock up to date */
561 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
563 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
565 set_bit(BITMAP_STALE, &bitmap->flags);
566 sb->state = cpu_to_le32(bitmap->flags);
567 bitmap->events_cleared = bitmap->mddev->events;
568 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
569 bitmap->mddev->bitmap_info.nodes = 0;
571 kunmap_atomic(sb);
573 return 0;
576 /* read the superblock from the bitmap file and initialize some bitmap fields */
577 static int md_bitmap_read_sb(struct bitmap *bitmap)
579 char *reason = NULL;
580 bitmap_super_t *sb;
581 unsigned long chunksize, daemon_sleep, write_behind;
582 unsigned long long events;
583 int nodes = 0;
584 unsigned long sectors_reserved = 0;
585 int err = -EINVAL;
586 struct page *sb_page;
587 loff_t offset = bitmap->mddev->bitmap_info.offset;
589 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
590 chunksize = 128 * 1024 * 1024;
591 daemon_sleep = 5 * HZ;
592 write_behind = 0;
593 set_bit(BITMAP_STALE, &bitmap->flags);
594 err = 0;
595 goto out_no_sb;
597 /* page 0 is the superblock, read it... */
598 sb_page = alloc_page(GFP_KERNEL);
599 if (!sb_page)
600 return -ENOMEM;
601 bitmap->storage.sb_page = sb_page;
603 re_read:
604 /* If cluster_slot is set, the cluster is setup */
605 if (bitmap->cluster_slot >= 0) {
606 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
608 sector_div(bm_blocks,
609 bitmap->mddev->bitmap_info.chunksize >> 9);
610 /* bits to bytes */
611 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
612 /* to 4k blocks */
613 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
614 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
615 pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
616 bitmap->cluster_slot, offset);
619 if (bitmap->storage.file) {
620 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
621 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
623 err = read_page(bitmap->storage.file, 0,
624 bitmap, bytes, sb_page);
625 } else {
626 err = read_sb_page(bitmap->mddev,
627 offset,
628 sb_page,
629 0, sizeof(bitmap_super_t));
631 if (err)
632 return err;
634 err = -EINVAL;
635 sb = kmap_atomic(sb_page);
637 chunksize = le32_to_cpu(sb->chunksize);
638 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
639 write_behind = le32_to_cpu(sb->write_behind);
640 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
641 /* Setup nodes/clustername only if bitmap version is
642 * cluster-compatible
644 if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
645 nodes = le32_to_cpu(sb->nodes);
646 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
647 sb->cluster_name, 64);
650 /* verify that the bitmap-specific fields are valid */
651 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
652 reason = "bad magic";
653 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
654 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
655 reason = "unrecognized superblock version";
656 else if (chunksize < 512)
657 reason = "bitmap chunksize too small";
658 else if (!is_power_of_2(chunksize))
659 reason = "bitmap chunksize not a power of 2";
660 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
661 reason = "daemon sleep period out of range";
662 else if (write_behind > COUNTER_MAX)
663 reason = "write-behind limit out of range (0 - 16383)";
664 if (reason) {
665 pr_warn("%s: invalid bitmap file superblock: %s\n",
666 bmname(bitmap), reason);
667 goto out;
670 /* keep the array size field of the bitmap superblock up to date */
671 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
673 if (bitmap->mddev->persistent) {
675 * We have a persistent array superblock, so compare the
676 * bitmap's UUID and event counter to the mddev's
678 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
679 pr_warn("%s: bitmap superblock UUID mismatch\n",
680 bmname(bitmap));
681 goto out;
683 events = le64_to_cpu(sb->events);
684 if (!nodes && (events < bitmap->mddev->events)) {
685 pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
686 bmname(bitmap), events,
687 (unsigned long long) bitmap->mddev->events);
688 set_bit(BITMAP_STALE, &bitmap->flags);
692 /* assign fields using values from superblock */
693 bitmap->flags |= le32_to_cpu(sb->state);
694 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
695 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
696 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
697 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
698 err = 0;
700 out:
701 kunmap_atomic(sb);
702 /* Assigning chunksize is required for "re_read" */
703 bitmap->mddev->bitmap_info.chunksize = chunksize;
704 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
705 err = md_setup_cluster(bitmap->mddev, nodes);
706 if (err) {
707 pr_warn("%s: Could not setup cluster service (%d)\n",
708 bmname(bitmap), err);
709 goto out_no_sb;
711 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
712 goto re_read;
716 out_no_sb:
717 if (test_bit(BITMAP_STALE, &bitmap->flags))
718 bitmap->events_cleared = bitmap->mddev->events;
719 bitmap->mddev->bitmap_info.chunksize = chunksize;
720 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
721 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
722 bitmap->mddev->bitmap_info.nodes = nodes;
723 if (bitmap->mddev->bitmap_info.space == 0 ||
724 bitmap->mddev->bitmap_info.space > sectors_reserved)
725 bitmap->mddev->bitmap_info.space = sectors_reserved;
726 if (err) {
727 md_bitmap_print_sb(bitmap);
728 if (bitmap->cluster_slot < 0)
729 md_cluster_stop(bitmap->mddev);
731 return err;
735 * general bitmap file operations
739 * on-disk bitmap:
741 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
742 * file a page at a time. There's a superblock at the start of the file.
744 /* calculate the index of the page that contains this bit */
745 static inline unsigned long file_page_index(struct bitmap_storage *store,
746 unsigned long chunk)
748 if (store->sb_page)
749 chunk += sizeof(bitmap_super_t) << 3;
750 return chunk >> PAGE_BIT_SHIFT;
753 /* calculate the (bit) offset of this bit within a page */
754 static inline unsigned long file_page_offset(struct bitmap_storage *store,
755 unsigned long chunk)
757 if (store->sb_page)
758 chunk += sizeof(bitmap_super_t) << 3;
759 return chunk & (PAGE_BITS - 1);
763 * return a pointer to the page in the filemap that contains the given bit
766 static inline struct page *filemap_get_page(struct bitmap_storage *store,
767 unsigned long chunk)
769 if (file_page_index(store, chunk) >= store->file_pages)
770 return NULL;
771 return store->filemap[file_page_index(store, chunk)];
774 static int md_bitmap_storage_alloc(struct bitmap_storage *store,
775 unsigned long chunks, int with_super,
776 int slot_number)
778 int pnum, offset = 0;
779 unsigned long num_pages;
780 unsigned long bytes;
782 bytes = DIV_ROUND_UP(chunks, 8);
783 if (with_super)
784 bytes += sizeof(bitmap_super_t);
786 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
787 offset = slot_number * num_pages;
789 store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
790 GFP_KERNEL);
791 if (!store->filemap)
792 return -ENOMEM;
794 if (with_super && !store->sb_page) {
795 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
796 if (store->sb_page == NULL)
797 return -ENOMEM;
800 pnum = 0;
801 if (store->sb_page) {
802 store->filemap[0] = store->sb_page;
803 pnum = 1;
804 store->sb_page->index = offset;
807 for ( ; pnum < num_pages; pnum++) {
808 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
809 if (!store->filemap[pnum]) {
810 store->file_pages = pnum;
811 return -ENOMEM;
813 store->filemap[pnum]->index = pnum + offset;
815 store->file_pages = pnum;
817 /* We need 4 bits per page, rounded up to a multiple
818 * of sizeof(unsigned long) */
819 store->filemap_attr = kzalloc(
820 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
821 GFP_KERNEL);
822 if (!store->filemap_attr)
823 return -ENOMEM;
825 store->bytes = bytes;
827 return 0;
830 static void md_bitmap_file_unmap(struct bitmap_storage *store)
832 struct page **map, *sb_page;
833 int pages;
834 struct file *file;
836 file = store->file;
837 map = store->filemap;
838 pages = store->file_pages;
839 sb_page = store->sb_page;
841 while (pages--)
842 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
843 free_buffers(map[pages]);
844 kfree(map);
845 kfree(store->filemap_attr);
847 if (sb_page)
848 free_buffers(sb_page);
850 if (file) {
851 struct inode *inode = file_inode(file);
852 invalidate_mapping_pages(inode->i_mapping, 0, -1);
853 fput(file);
858 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
859 * then it is no longer reliable, so we stop using it and we mark the file
860 * as failed in the superblock
862 static void md_bitmap_file_kick(struct bitmap *bitmap)
864 char *path, *ptr = NULL;
866 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
867 md_bitmap_update_sb(bitmap);
869 if (bitmap->storage.file) {
870 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
871 if (path)
872 ptr = file_path(bitmap->storage.file,
873 path, PAGE_SIZE);
875 pr_warn("%s: kicking failed bitmap file %s from array!\n",
876 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
878 kfree(path);
879 } else
880 pr_warn("%s: disabling internal bitmap due to errors\n",
881 bmname(bitmap));
885 enum bitmap_page_attr {
886 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
887 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
888 * i.e. counter is 1 or 2. */
889 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
892 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
893 enum bitmap_page_attr attr)
895 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
898 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
899 enum bitmap_page_attr attr)
901 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
904 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
905 enum bitmap_page_attr attr)
907 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
910 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
911 enum bitmap_page_attr attr)
913 return test_and_clear_bit((pnum<<2) + attr,
914 bitmap->storage.filemap_attr);
917 * bitmap_file_set_bit -- called before performing a write to the md device
918 * to set (and eventually sync) a particular bit in the bitmap file
920 * we set the bit immediately, then we record the page number so that
921 * when an unplug occurs, we can flush the dirty pages out to disk
923 static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
925 unsigned long bit;
926 struct page *page;
927 void *kaddr;
928 unsigned long chunk = block >> bitmap->counts.chunkshift;
929 struct bitmap_storage *store = &bitmap->storage;
930 unsigned long node_offset = 0;
932 if (mddev_is_clustered(bitmap->mddev))
933 node_offset = bitmap->cluster_slot * store->file_pages;
935 page = filemap_get_page(&bitmap->storage, chunk);
936 if (!page)
937 return;
938 bit = file_page_offset(&bitmap->storage, chunk);
940 /* set the bit */
941 kaddr = kmap_atomic(page);
942 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
943 set_bit(bit, kaddr);
944 else
945 set_bit_le(bit, kaddr);
946 kunmap_atomic(kaddr);
947 pr_debug("set file bit %lu page %lu\n", bit, page->index);
948 /* record page number so it gets flushed to disk when unplug occurs */
949 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY);
952 static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
954 unsigned long bit;
955 struct page *page;
956 void *paddr;
957 unsigned long chunk = block >> bitmap->counts.chunkshift;
958 struct bitmap_storage *store = &bitmap->storage;
959 unsigned long node_offset = 0;
961 if (mddev_is_clustered(bitmap->mddev))
962 node_offset = bitmap->cluster_slot * store->file_pages;
964 page = filemap_get_page(&bitmap->storage, chunk);
965 if (!page)
966 return;
967 bit = file_page_offset(&bitmap->storage, chunk);
968 paddr = kmap_atomic(page);
969 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
970 clear_bit(bit, paddr);
971 else
972 clear_bit_le(bit, paddr);
973 kunmap_atomic(paddr);
974 if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
975 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
976 bitmap->allclean = 0;
980 static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
982 unsigned long bit;
983 struct page *page;
984 void *paddr;
985 unsigned long chunk = block >> bitmap->counts.chunkshift;
986 int set = 0;
988 page = filemap_get_page(&bitmap->storage, chunk);
989 if (!page)
990 return -EINVAL;
991 bit = file_page_offset(&bitmap->storage, chunk);
992 paddr = kmap_atomic(page);
993 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
994 set = test_bit(bit, paddr);
995 else
996 set = test_bit_le(bit, paddr);
997 kunmap_atomic(paddr);
998 return set;
1002 /* this gets called when the md device is ready to unplug its underlying
1003 * (slave) device queues -- before we let any writes go down, we need to
1004 * sync the dirty pages of the bitmap file to disk */
1005 void md_bitmap_unplug(struct bitmap *bitmap)
1007 unsigned long i;
1008 int dirty, need_write;
1009 int writing = 0;
1011 if (!bitmap || !bitmap->storage.filemap ||
1012 test_bit(BITMAP_STALE, &bitmap->flags))
1013 return;
1015 /* look at each page to see if there are any set bits that need to be
1016 * flushed out to disk */
1017 for (i = 0; i < bitmap->storage.file_pages; i++) {
1018 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1019 need_write = test_and_clear_page_attr(bitmap, i,
1020 BITMAP_PAGE_NEEDWRITE);
1021 if (dirty || need_write) {
1022 if (!writing) {
1023 md_bitmap_wait_writes(bitmap);
1024 if (bitmap->mddev->queue)
1025 blk_add_trace_msg(bitmap->mddev->queue,
1026 "md bitmap_unplug");
1028 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1029 write_page(bitmap, bitmap->storage.filemap[i], 0);
1030 writing = 1;
1033 if (writing)
1034 md_bitmap_wait_writes(bitmap);
1036 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1037 md_bitmap_file_kick(bitmap);
1039 EXPORT_SYMBOL(md_bitmap_unplug);
1041 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1042 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1043 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1044 * memory mapping of the bitmap file
1045 * Special cases:
1046 * if there's no bitmap file, or if the bitmap file had been
1047 * previously kicked from the array, we mark all the bits as
1048 * 1's in order to cause a full resync.
1050 * We ignore all bits for sectors that end earlier than 'start'.
1051 * This is used when reading an out-of-date bitmap...
1053 static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1055 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1056 struct page *page = NULL;
1057 unsigned long bit_cnt = 0;
1058 struct file *file;
1059 unsigned long offset;
1060 int outofdate;
1061 int ret = -ENOSPC;
1062 void *paddr;
1063 struct bitmap_storage *store = &bitmap->storage;
1065 chunks = bitmap->counts.chunks;
1066 file = store->file;
1068 if (!file && !bitmap->mddev->bitmap_info.offset) {
1069 /* No permanent bitmap - fill with '1s'. */
1070 store->filemap = NULL;
1071 store->file_pages = 0;
1072 for (i = 0; i < chunks ; i++) {
1073 /* if the disk bit is set, set the memory bit */
1074 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1075 >= start);
1076 md_bitmap_set_memory_bits(bitmap,
1077 (sector_t)i << bitmap->counts.chunkshift,
1078 needed);
1080 return 0;
1083 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1084 if (outofdate)
1085 pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap));
1087 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1088 pr_warn("%s: bitmap file too short %lu < %lu\n",
1089 bmname(bitmap),
1090 (unsigned long) i_size_read(file->f_mapping->host),
1091 store->bytes);
1092 goto err;
1095 oldindex = ~0L;
1096 offset = 0;
1097 if (!bitmap->mddev->bitmap_info.external)
1098 offset = sizeof(bitmap_super_t);
1100 if (mddev_is_clustered(bitmap->mddev))
1101 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1103 for (i = 0; i < chunks; i++) {
1104 int b;
1105 index = file_page_index(&bitmap->storage, i);
1106 bit = file_page_offset(&bitmap->storage, i);
1107 if (index != oldindex) { /* this is a new page, read it in */
1108 int count;
1109 /* unmap the old page, we're done with it */
1110 if (index == store->file_pages-1)
1111 count = store->bytes - index * PAGE_SIZE;
1112 else
1113 count = PAGE_SIZE;
1114 page = store->filemap[index];
1115 if (file)
1116 ret = read_page(file, index, bitmap,
1117 count, page);
1118 else
1119 ret = read_sb_page(
1120 bitmap->mddev,
1121 bitmap->mddev->bitmap_info.offset,
1122 page,
1123 index + node_offset, count);
1125 if (ret)
1126 goto err;
1128 oldindex = index;
1130 if (outofdate) {
1132 * if bitmap is out of date, dirty the
1133 * whole page and write it out
1135 paddr = kmap_atomic(page);
1136 memset(paddr + offset, 0xff,
1137 PAGE_SIZE - offset);
1138 kunmap_atomic(paddr);
1139 write_page(bitmap, page, 1);
1141 ret = -EIO;
1142 if (test_bit(BITMAP_WRITE_ERROR,
1143 &bitmap->flags))
1144 goto err;
1147 paddr = kmap_atomic(page);
1148 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1149 b = test_bit(bit, paddr);
1150 else
1151 b = test_bit_le(bit, paddr);
1152 kunmap_atomic(paddr);
1153 if (b) {
1154 /* if the disk bit is set, set the memory bit */
1155 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1156 >= start);
1157 md_bitmap_set_memory_bits(bitmap,
1158 (sector_t)i << bitmap->counts.chunkshift,
1159 needed);
1160 bit_cnt++;
1162 offset = 0;
1165 pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1166 bmname(bitmap), store->file_pages,
1167 bit_cnt, chunks);
1169 return 0;
1171 err:
1172 pr_warn("%s: bitmap initialisation failed: %d\n",
1173 bmname(bitmap), ret);
1174 return ret;
1177 void md_bitmap_write_all(struct bitmap *bitmap)
1179 /* We don't actually write all bitmap blocks here,
1180 * just flag them as needing to be written
1182 int i;
1184 if (!bitmap || !bitmap->storage.filemap)
1185 return;
1186 if (bitmap->storage.file)
1187 /* Only one copy, so nothing needed */
1188 return;
1190 for (i = 0; i < bitmap->storage.file_pages; i++)
1191 set_page_attr(bitmap, i,
1192 BITMAP_PAGE_NEEDWRITE);
1193 bitmap->allclean = 0;
1196 static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1197 sector_t offset, int inc)
1199 sector_t chunk = offset >> bitmap->chunkshift;
1200 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1201 bitmap->bp[page].count += inc;
1202 md_bitmap_checkfree(bitmap, page);
1205 static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1207 sector_t chunk = offset >> bitmap->chunkshift;
1208 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1209 struct bitmap_page *bp = &bitmap->bp[page];
1211 if (!bp->pending)
1212 bp->pending = 1;
1215 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1216 sector_t offset, sector_t *blocks,
1217 int create);
1220 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1221 * out to disk
1224 void md_bitmap_daemon_work(struct mddev *mddev)
1226 struct bitmap *bitmap;
1227 unsigned long j;
1228 unsigned long nextpage;
1229 sector_t blocks;
1230 struct bitmap_counts *counts;
1232 /* Use a mutex to guard daemon_work against
1233 * bitmap_destroy.
1235 mutex_lock(&mddev->bitmap_info.mutex);
1236 bitmap = mddev->bitmap;
1237 if (bitmap == NULL) {
1238 mutex_unlock(&mddev->bitmap_info.mutex);
1239 return;
1241 if (time_before(jiffies, bitmap->daemon_lastrun
1242 + mddev->bitmap_info.daemon_sleep))
1243 goto done;
1245 bitmap->daemon_lastrun = jiffies;
1246 if (bitmap->allclean) {
1247 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1248 goto done;
1250 bitmap->allclean = 1;
1252 if (bitmap->mddev->queue)
1253 blk_add_trace_msg(bitmap->mddev->queue,
1254 "md bitmap_daemon_work");
1256 /* Any file-page which is PENDING now needs to be written.
1257 * So set NEEDWRITE now, then after we make any last-minute changes
1258 * we will write it.
1260 for (j = 0; j < bitmap->storage.file_pages; j++)
1261 if (test_and_clear_page_attr(bitmap, j,
1262 BITMAP_PAGE_PENDING))
1263 set_page_attr(bitmap, j,
1264 BITMAP_PAGE_NEEDWRITE);
1266 if (bitmap->need_sync &&
1267 mddev->bitmap_info.external == 0) {
1268 /* Arrange for superblock update as well as
1269 * other changes */
1270 bitmap_super_t *sb;
1271 bitmap->need_sync = 0;
1272 if (bitmap->storage.filemap) {
1273 sb = kmap_atomic(bitmap->storage.sb_page);
1274 sb->events_cleared =
1275 cpu_to_le64(bitmap->events_cleared);
1276 kunmap_atomic(sb);
1277 set_page_attr(bitmap, 0,
1278 BITMAP_PAGE_NEEDWRITE);
1281 /* Now look at the bitmap counters and if any are '2' or '1',
1282 * decrement and handle accordingly.
1284 counts = &bitmap->counts;
1285 spin_lock_irq(&counts->lock);
1286 nextpage = 0;
1287 for (j = 0; j < counts->chunks; j++) {
1288 bitmap_counter_t *bmc;
1289 sector_t block = (sector_t)j << counts->chunkshift;
1291 if (j == nextpage) {
1292 nextpage += PAGE_COUNTER_RATIO;
1293 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1294 j |= PAGE_COUNTER_MASK;
1295 continue;
1297 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1300 bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1301 if (!bmc) {
1302 j |= PAGE_COUNTER_MASK;
1303 continue;
1305 if (*bmc == 1 && !bitmap->need_sync) {
1306 /* We can clear the bit */
1307 *bmc = 0;
1308 md_bitmap_count_page(counts, block, -1);
1309 md_bitmap_file_clear_bit(bitmap, block);
1310 } else if (*bmc && *bmc <= 2) {
1311 *bmc = 1;
1312 md_bitmap_set_pending(counts, block);
1313 bitmap->allclean = 0;
1316 spin_unlock_irq(&counts->lock);
1318 md_bitmap_wait_writes(bitmap);
1319 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1320 * DIRTY pages need to be written by bitmap_unplug so it can wait
1321 * for them.
1322 * If we find any DIRTY page we stop there and let bitmap_unplug
1323 * handle all the rest. This is important in the case where
1324 * the first blocking holds the superblock and it has been updated.
1325 * We mustn't write any other blocks before the superblock.
1327 for (j = 0;
1328 j < bitmap->storage.file_pages
1329 && !test_bit(BITMAP_STALE, &bitmap->flags);
1330 j++) {
1331 if (test_page_attr(bitmap, j,
1332 BITMAP_PAGE_DIRTY))
1333 /* bitmap_unplug will handle the rest */
1334 break;
1335 if (bitmap->storage.filemap &&
1336 test_and_clear_page_attr(bitmap, j,
1337 BITMAP_PAGE_NEEDWRITE)) {
1338 write_page(bitmap, bitmap->storage.filemap[j], 0);
1342 done:
1343 if (bitmap->allclean == 0)
1344 mddev->thread->timeout =
1345 mddev->bitmap_info.daemon_sleep;
1346 mutex_unlock(&mddev->bitmap_info.mutex);
1349 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1350 sector_t offset, sector_t *blocks,
1351 int create)
1352 __releases(bitmap->lock)
1353 __acquires(bitmap->lock)
1355 /* If 'create', we might release the lock and reclaim it.
1356 * The lock must have been taken with interrupts enabled.
1357 * If !create, we don't release the lock.
1359 sector_t chunk = offset >> bitmap->chunkshift;
1360 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1361 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1362 sector_t csize;
1363 int err;
1365 err = md_bitmap_checkpage(bitmap, page, create, 0);
1367 if (bitmap->bp[page].hijacked ||
1368 bitmap->bp[page].map == NULL)
1369 csize = ((sector_t)1) << (bitmap->chunkshift +
1370 PAGE_COUNTER_SHIFT - 1);
1371 else
1372 csize = ((sector_t)1) << bitmap->chunkshift;
1373 *blocks = csize - (offset & (csize - 1));
1375 if (err < 0)
1376 return NULL;
1378 /* now locked ... */
1380 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1381 /* should we use the first or second counter field
1382 * of the hijacked pointer? */
1383 int hi = (pageoff > PAGE_COUNTER_MASK);
1384 return &((bitmap_counter_t *)
1385 &bitmap->bp[page].map)[hi];
1386 } else /* page is allocated */
1387 return (bitmap_counter_t *)
1388 &(bitmap->bp[page].map[pageoff]);
1391 int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1393 if (!bitmap)
1394 return 0;
1396 if (behind) {
1397 int bw;
1398 atomic_inc(&bitmap->behind_writes);
1399 bw = atomic_read(&bitmap->behind_writes);
1400 if (bw > bitmap->behind_writes_used)
1401 bitmap->behind_writes_used = bw;
1403 pr_debug("inc write-behind count %d/%lu\n",
1404 bw, bitmap->mddev->bitmap_info.max_write_behind);
1407 while (sectors) {
1408 sector_t blocks;
1409 bitmap_counter_t *bmc;
1411 spin_lock_irq(&bitmap->counts.lock);
1412 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1413 if (!bmc) {
1414 spin_unlock_irq(&bitmap->counts.lock);
1415 return 0;
1418 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1419 DEFINE_WAIT(__wait);
1420 /* note that it is safe to do the prepare_to_wait
1421 * after the test as long as we do it before dropping
1422 * the spinlock.
1424 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1425 TASK_UNINTERRUPTIBLE);
1426 spin_unlock_irq(&bitmap->counts.lock);
1427 schedule();
1428 finish_wait(&bitmap->overflow_wait, &__wait);
1429 continue;
1432 switch (*bmc) {
1433 case 0:
1434 md_bitmap_file_set_bit(bitmap, offset);
1435 md_bitmap_count_page(&bitmap->counts, offset, 1);
1436 /* fall through */
1437 case 1:
1438 *bmc = 2;
1441 (*bmc)++;
1443 spin_unlock_irq(&bitmap->counts.lock);
1445 offset += blocks;
1446 if (sectors > blocks)
1447 sectors -= blocks;
1448 else
1449 sectors = 0;
1451 return 0;
1453 EXPORT_SYMBOL(md_bitmap_startwrite);
1455 void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1456 unsigned long sectors, int success, int behind)
1458 if (!bitmap)
1459 return;
1460 if (behind) {
1461 if (atomic_dec_and_test(&bitmap->behind_writes))
1462 wake_up(&bitmap->behind_wait);
1463 pr_debug("dec write-behind count %d/%lu\n",
1464 atomic_read(&bitmap->behind_writes),
1465 bitmap->mddev->bitmap_info.max_write_behind);
1468 while (sectors) {
1469 sector_t blocks;
1470 unsigned long flags;
1471 bitmap_counter_t *bmc;
1473 spin_lock_irqsave(&bitmap->counts.lock, flags);
1474 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1475 if (!bmc) {
1476 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1477 return;
1480 if (success && !bitmap->mddev->degraded &&
1481 bitmap->events_cleared < bitmap->mddev->events) {
1482 bitmap->events_cleared = bitmap->mddev->events;
1483 bitmap->need_sync = 1;
1484 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1487 if (!success && !NEEDED(*bmc))
1488 *bmc |= NEEDED_MASK;
1490 if (COUNTER(*bmc) == COUNTER_MAX)
1491 wake_up(&bitmap->overflow_wait);
1493 (*bmc)--;
1494 if (*bmc <= 2) {
1495 md_bitmap_set_pending(&bitmap->counts, offset);
1496 bitmap->allclean = 0;
1498 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1499 offset += blocks;
1500 if (sectors > blocks)
1501 sectors -= blocks;
1502 else
1503 sectors = 0;
1506 EXPORT_SYMBOL(md_bitmap_endwrite);
1508 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1509 int degraded)
1511 bitmap_counter_t *bmc;
1512 int rv;
1513 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1514 *blocks = 1024;
1515 return 1; /* always resync if no bitmap */
1517 spin_lock_irq(&bitmap->counts.lock);
1518 bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1519 rv = 0;
1520 if (bmc) {
1521 /* locked */
1522 if (RESYNC(*bmc))
1523 rv = 1;
1524 else if (NEEDED(*bmc)) {
1525 rv = 1;
1526 if (!degraded) { /* don't set/clear bits if degraded */
1527 *bmc |= RESYNC_MASK;
1528 *bmc &= ~NEEDED_MASK;
1532 spin_unlock_irq(&bitmap->counts.lock);
1533 return rv;
1536 int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1537 int degraded)
1539 /* bitmap_start_sync must always report on multiples of whole
1540 * pages, otherwise resync (which is very PAGE_SIZE based) will
1541 * get confused.
1542 * So call __bitmap_start_sync repeatedly (if needed) until
1543 * At least PAGE_SIZE>>9 blocks are covered.
1544 * Return the 'or' of the result.
1546 int rv = 0;
1547 sector_t blocks1;
1549 *blocks = 0;
1550 while (*blocks < (PAGE_SIZE>>9)) {
1551 rv |= __bitmap_start_sync(bitmap, offset,
1552 &blocks1, degraded);
1553 offset += blocks1;
1554 *blocks += blocks1;
1556 return rv;
1558 EXPORT_SYMBOL(md_bitmap_start_sync);
1560 void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1562 bitmap_counter_t *bmc;
1563 unsigned long flags;
1565 if (bitmap == NULL) {
1566 *blocks = 1024;
1567 return;
1569 spin_lock_irqsave(&bitmap->counts.lock, flags);
1570 bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1571 if (bmc == NULL)
1572 goto unlock;
1573 /* locked */
1574 if (RESYNC(*bmc)) {
1575 *bmc &= ~RESYNC_MASK;
1577 if (!NEEDED(*bmc) && aborted)
1578 *bmc |= NEEDED_MASK;
1579 else {
1580 if (*bmc <= 2) {
1581 md_bitmap_set_pending(&bitmap->counts, offset);
1582 bitmap->allclean = 0;
1586 unlock:
1587 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1589 EXPORT_SYMBOL(md_bitmap_end_sync);
1591 void md_bitmap_close_sync(struct bitmap *bitmap)
1593 /* Sync has finished, and any bitmap chunks that weren't synced
1594 * properly have been aborted. It remains to us to clear the
1595 * RESYNC bit wherever it is still on
1597 sector_t sector = 0;
1598 sector_t blocks;
1599 if (!bitmap)
1600 return;
1601 while (sector < bitmap->mddev->resync_max_sectors) {
1602 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1603 sector += blocks;
1606 EXPORT_SYMBOL(md_bitmap_close_sync);
1608 void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1610 sector_t s = 0;
1611 sector_t blocks;
1613 if (!bitmap)
1614 return;
1615 if (sector == 0) {
1616 bitmap->last_end_sync = jiffies;
1617 return;
1619 if (!force && time_before(jiffies, (bitmap->last_end_sync
1620 + bitmap->mddev->bitmap_info.daemon_sleep)))
1621 return;
1622 wait_event(bitmap->mddev->recovery_wait,
1623 atomic_read(&bitmap->mddev->recovery_active) == 0);
1625 bitmap->mddev->curr_resync_completed = sector;
1626 set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1627 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1628 s = 0;
1629 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1630 md_bitmap_end_sync(bitmap, s, &blocks, 0);
1631 s += blocks;
1633 bitmap->last_end_sync = jiffies;
1634 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1636 EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1638 void md_bitmap_sync_with_cluster(struct mddev *mddev,
1639 sector_t old_lo, sector_t old_hi,
1640 sector_t new_lo, sector_t new_hi)
1642 struct bitmap *bitmap = mddev->bitmap;
1643 sector_t sector, blocks = 0;
1645 for (sector = old_lo; sector < new_lo; ) {
1646 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1647 sector += blocks;
1649 WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1651 for (sector = old_hi; sector < new_hi; ) {
1652 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1653 sector += blocks;
1655 WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1657 EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1659 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1661 /* For each chunk covered by any of these sectors, set the
1662 * counter to 2 and possibly set resync_needed. They should all
1663 * be 0 at this point
1666 sector_t secs;
1667 bitmap_counter_t *bmc;
1668 spin_lock_irq(&bitmap->counts.lock);
1669 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1670 if (!bmc) {
1671 spin_unlock_irq(&bitmap->counts.lock);
1672 return;
1674 if (!*bmc) {
1675 *bmc = 2;
1676 md_bitmap_count_page(&bitmap->counts, offset, 1);
1677 md_bitmap_set_pending(&bitmap->counts, offset);
1678 bitmap->allclean = 0;
1680 if (needed)
1681 *bmc |= NEEDED_MASK;
1682 spin_unlock_irq(&bitmap->counts.lock);
1685 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1686 void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1688 unsigned long chunk;
1690 for (chunk = s; chunk <= e; chunk++) {
1691 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1692 md_bitmap_set_memory_bits(bitmap, sec, 1);
1693 md_bitmap_file_set_bit(bitmap, sec);
1694 if (sec < bitmap->mddev->recovery_cp)
1695 /* We are asserting that the array is dirty,
1696 * so move the recovery_cp address back so
1697 * that it is obvious that it is dirty
1699 bitmap->mddev->recovery_cp = sec;
1704 * flush out any pending updates
1706 void md_bitmap_flush(struct mddev *mddev)
1708 struct bitmap *bitmap = mddev->bitmap;
1709 long sleep;
1711 if (!bitmap) /* there was no bitmap */
1712 return;
1714 /* run the daemon_work three time to ensure everything is flushed
1715 * that can be
1717 sleep = mddev->bitmap_info.daemon_sleep * 2;
1718 bitmap->daemon_lastrun -= sleep;
1719 md_bitmap_daemon_work(mddev);
1720 bitmap->daemon_lastrun -= sleep;
1721 md_bitmap_daemon_work(mddev);
1722 bitmap->daemon_lastrun -= sleep;
1723 md_bitmap_daemon_work(mddev);
1724 md_bitmap_update_sb(bitmap);
1728 * free memory that was allocated
1730 void md_bitmap_free(struct bitmap *bitmap)
1732 unsigned long k, pages;
1733 struct bitmap_page *bp;
1735 if (!bitmap) /* there was no bitmap */
1736 return;
1738 if (bitmap->sysfs_can_clear)
1739 sysfs_put(bitmap->sysfs_can_clear);
1741 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1742 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1743 md_cluster_stop(bitmap->mddev);
1745 /* Shouldn't be needed - but just in case.... */
1746 wait_event(bitmap->write_wait,
1747 atomic_read(&bitmap->pending_writes) == 0);
1749 /* release the bitmap file */
1750 md_bitmap_file_unmap(&bitmap->storage);
1752 bp = bitmap->counts.bp;
1753 pages = bitmap->counts.pages;
1755 /* free all allocated memory */
1757 if (bp) /* deallocate the page memory */
1758 for (k = 0; k < pages; k++)
1759 if (bp[k].map && !bp[k].hijacked)
1760 kfree(bp[k].map);
1761 kfree(bp);
1762 kfree(bitmap);
1764 EXPORT_SYMBOL(md_bitmap_free);
1766 void md_bitmap_wait_behind_writes(struct mddev *mddev)
1768 struct bitmap *bitmap = mddev->bitmap;
1770 /* wait for behind writes to complete */
1771 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1772 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1773 mdname(mddev));
1774 /* need to kick something here to make sure I/O goes? */
1775 wait_event(bitmap->behind_wait,
1776 atomic_read(&bitmap->behind_writes) == 0);
1780 void md_bitmap_destroy(struct mddev *mddev)
1782 struct bitmap *bitmap = mddev->bitmap;
1784 if (!bitmap) /* there was no bitmap */
1785 return;
1787 md_bitmap_wait_behind_writes(mddev);
1788 if (!mddev->serialize_policy)
1789 mddev_destroy_serial_pool(mddev, NULL, true);
1791 mutex_lock(&mddev->bitmap_info.mutex);
1792 spin_lock(&mddev->lock);
1793 mddev->bitmap = NULL; /* disconnect from the md device */
1794 spin_unlock(&mddev->lock);
1795 mutex_unlock(&mddev->bitmap_info.mutex);
1796 if (mddev->thread)
1797 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1799 md_bitmap_free(bitmap);
1803 * initialize the bitmap structure
1804 * if this returns an error, bitmap_destroy must be called to do clean up
1805 * once mddev->bitmap is set
1807 struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1809 struct bitmap *bitmap;
1810 sector_t blocks = mddev->resync_max_sectors;
1811 struct file *file = mddev->bitmap_info.file;
1812 int err;
1813 struct kernfs_node *bm = NULL;
1815 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1817 BUG_ON(file && mddev->bitmap_info.offset);
1819 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1820 pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1821 mdname(mddev));
1822 return ERR_PTR(-EBUSY);
1825 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1826 if (!bitmap)
1827 return ERR_PTR(-ENOMEM);
1829 spin_lock_init(&bitmap->counts.lock);
1830 atomic_set(&bitmap->pending_writes, 0);
1831 init_waitqueue_head(&bitmap->write_wait);
1832 init_waitqueue_head(&bitmap->overflow_wait);
1833 init_waitqueue_head(&bitmap->behind_wait);
1835 bitmap->mddev = mddev;
1836 bitmap->cluster_slot = slot;
1838 if (mddev->kobj.sd)
1839 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1840 if (bm) {
1841 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1842 sysfs_put(bm);
1843 } else
1844 bitmap->sysfs_can_clear = NULL;
1846 bitmap->storage.file = file;
1847 if (file) {
1848 get_file(file);
1849 /* As future accesses to this file will use bmap,
1850 * and bypass the page cache, we must sync the file
1851 * first.
1853 vfs_fsync(file, 1);
1855 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1856 if (!mddev->bitmap_info.external) {
1858 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1859 * instructing us to create a new on-disk bitmap instance.
1861 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1862 err = md_bitmap_new_disk_sb(bitmap);
1863 else
1864 err = md_bitmap_read_sb(bitmap);
1865 } else {
1866 err = 0;
1867 if (mddev->bitmap_info.chunksize == 0 ||
1868 mddev->bitmap_info.daemon_sleep == 0)
1869 /* chunksize and time_base need to be
1870 * set first. */
1871 err = -EINVAL;
1873 if (err)
1874 goto error;
1876 bitmap->daemon_lastrun = jiffies;
1877 err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1878 if (err)
1879 goto error;
1881 pr_debug("created bitmap (%lu pages) for device %s\n",
1882 bitmap->counts.pages, bmname(bitmap));
1884 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1885 if (err)
1886 goto error;
1888 return bitmap;
1889 error:
1890 md_bitmap_free(bitmap);
1891 return ERR_PTR(err);
1894 int md_bitmap_load(struct mddev *mddev)
1896 int err = 0;
1897 sector_t start = 0;
1898 sector_t sector = 0;
1899 struct bitmap *bitmap = mddev->bitmap;
1900 struct md_rdev *rdev;
1902 if (!bitmap)
1903 goto out;
1905 rdev_for_each(rdev, mddev)
1906 mddev_create_serial_pool(mddev, rdev, true);
1908 if (mddev_is_clustered(mddev))
1909 md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1911 /* Clear out old bitmap info first: Either there is none, or we
1912 * are resuming after someone else has possibly changed things,
1913 * so we should forget old cached info.
1914 * All chunks should be clean, but some might need_sync.
1916 while (sector < mddev->resync_max_sectors) {
1917 sector_t blocks;
1918 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1919 sector += blocks;
1921 md_bitmap_close_sync(bitmap);
1923 if (mddev->degraded == 0
1924 || bitmap->events_cleared == mddev->events)
1925 /* no need to keep dirty bits to optimise a
1926 * re-add of a missing device */
1927 start = mddev->recovery_cp;
1929 mutex_lock(&mddev->bitmap_info.mutex);
1930 err = md_bitmap_init_from_disk(bitmap, start);
1931 mutex_unlock(&mddev->bitmap_info.mutex);
1933 if (err)
1934 goto out;
1935 clear_bit(BITMAP_STALE, &bitmap->flags);
1937 /* Kick recovery in case any bits were set */
1938 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1940 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1941 md_wakeup_thread(mddev->thread);
1943 md_bitmap_update_sb(bitmap);
1945 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1946 err = -EIO;
1947 out:
1948 return err;
1950 EXPORT_SYMBOL_GPL(md_bitmap_load);
1952 struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot)
1954 int rv = 0;
1955 struct bitmap *bitmap;
1957 bitmap = md_bitmap_create(mddev, slot);
1958 if (IS_ERR(bitmap)) {
1959 rv = PTR_ERR(bitmap);
1960 return ERR_PTR(rv);
1963 rv = md_bitmap_init_from_disk(bitmap, 0);
1964 if (rv) {
1965 md_bitmap_free(bitmap);
1966 return ERR_PTR(rv);
1969 return bitmap;
1971 EXPORT_SYMBOL(get_bitmap_from_slot);
1973 /* Loads the bitmap associated with slot and copies the resync information
1974 * to our bitmap
1976 int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
1977 sector_t *low, sector_t *high, bool clear_bits)
1979 int rv = 0, i, j;
1980 sector_t block, lo = 0, hi = 0;
1981 struct bitmap_counts *counts;
1982 struct bitmap *bitmap;
1984 bitmap = get_bitmap_from_slot(mddev, slot);
1985 if (IS_ERR(bitmap)) {
1986 pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
1987 return -1;
1990 counts = &bitmap->counts;
1991 for (j = 0; j < counts->chunks; j++) {
1992 block = (sector_t)j << counts->chunkshift;
1993 if (md_bitmap_file_test_bit(bitmap, block)) {
1994 if (!lo)
1995 lo = block;
1996 hi = block;
1997 md_bitmap_file_clear_bit(bitmap, block);
1998 md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
1999 md_bitmap_file_set_bit(mddev->bitmap, block);
2003 if (clear_bits) {
2004 md_bitmap_update_sb(bitmap);
2005 /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2006 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2007 for (i = 0; i < bitmap->storage.file_pages; i++)
2008 if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2009 set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2010 md_bitmap_unplug(bitmap);
2012 md_bitmap_unplug(mddev->bitmap);
2013 *low = lo;
2014 *high = hi;
2016 return rv;
2018 EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2021 void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2023 unsigned long chunk_kb;
2024 struct bitmap_counts *counts;
2026 if (!bitmap)
2027 return;
2029 counts = &bitmap->counts;
2031 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2032 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2033 "%lu%s chunk",
2034 counts->pages - counts->missing_pages,
2035 counts->pages,
2036 (counts->pages - counts->missing_pages)
2037 << (PAGE_SHIFT - 10),
2038 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2039 chunk_kb ? "KB" : "B");
2040 if (bitmap->storage.file) {
2041 seq_printf(seq, ", file: ");
2042 seq_file_path(seq, bitmap->storage.file, " \t\n");
2045 seq_printf(seq, "\n");
2048 int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2049 int chunksize, int init)
2051 /* If chunk_size is 0, choose an appropriate chunk size.
2052 * Then possibly allocate new storage space.
2053 * Then quiesce, copy bits, replace bitmap, and re-start
2055 * This function is called both to set up the initial bitmap
2056 * and to resize the bitmap while the array is active.
2057 * If this happens as a result of the array being resized,
2058 * chunksize will be zero, and we need to choose a suitable
2059 * chunksize, otherwise we use what we are given.
2061 struct bitmap_storage store;
2062 struct bitmap_counts old_counts;
2063 unsigned long chunks;
2064 sector_t block;
2065 sector_t old_blocks, new_blocks;
2066 int chunkshift;
2067 int ret = 0;
2068 long pages;
2069 struct bitmap_page *new_bp;
2071 if (bitmap->storage.file && !init) {
2072 pr_info("md: cannot resize file-based bitmap\n");
2073 return -EINVAL;
2076 if (chunksize == 0) {
2077 /* If there is enough space, leave the chunk size unchanged,
2078 * else increase by factor of two until there is enough space.
2080 long bytes;
2081 long space = bitmap->mddev->bitmap_info.space;
2083 if (space == 0) {
2084 /* We don't know how much space there is, so limit
2085 * to current size - in sectors.
2087 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2088 if (!bitmap->mddev->bitmap_info.external)
2089 bytes += sizeof(bitmap_super_t);
2090 space = DIV_ROUND_UP(bytes, 512);
2091 bitmap->mddev->bitmap_info.space = space;
2093 chunkshift = bitmap->counts.chunkshift;
2094 chunkshift--;
2095 do {
2096 /* 'chunkshift' is shift from block size to chunk size */
2097 chunkshift++;
2098 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2099 bytes = DIV_ROUND_UP(chunks, 8);
2100 if (!bitmap->mddev->bitmap_info.external)
2101 bytes += sizeof(bitmap_super_t);
2102 } while (bytes > (space << 9));
2103 } else
2104 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2106 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2107 memset(&store, 0, sizeof(store));
2108 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2109 ret = md_bitmap_storage_alloc(&store, chunks,
2110 !bitmap->mddev->bitmap_info.external,
2111 mddev_is_clustered(bitmap->mddev)
2112 ? bitmap->cluster_slot : 0);
2113 if (ret) {
2114 md_bitmap_file_unmap(&store);
2115 goto err;
2118 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2120 new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2121 ret = -ENOMEM;
2122 if (!new_bp) {
2123 md_bitmap_file_unmap(&store);
2124 goto err;
2127 if (!init)
2128 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2130 store.file = bitmap->storage.file;
2131 bitmap->storage.file = NULL;
2133 if (store.sb_page && bitmap->storage.sb_page)
2134 memcpy(page_address(store.sb_page),
2135 page_address(bitmap->storage.sb_page),
2136 sizeof(bitmap_super_t));
2137 spin_lock_irq(&bitmap->counts.lock);
2138 md_bitmap_file_unmap(&bitmap->storage);
2139 bitmap->storage = store;
2141 old_counts = bitmap->counts;
2142 bitmap->counts.bp = new_bp;
2143 bitmap->counts.pages = pages;
2144 bitmap->counts.missing_pages = pages;
2145 bitmap->counts.chunkshift = chunkshift;
2146 bitmap->counts.chunks = chunks;
2147 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2148 BITMAP_BLOCK_SHIFT);
2150 blocks = min(old_counts.chunks << old_counts.chunkshift,
2151 chunks << chunkshift);
2153 /* For cluster raid, need to pre-allocate bitmap */
2154 if (mddev_is_clustered(bitmap->mddev)) {
2155 unsigned long page;
2156 for (page = 0; page < pages; page++) {
2157 ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2158 if (ret) {
2159 unsigned long k;
2161 /* deallocate the page memory */
2162 for (k = 0; k < page; k++) {
2163 kfree(new_bp[k].map);
2165 kfree(new_bp);
2167 /* restore some fields from old_counts */
2168 bitmap->counts.bp = old_counts.bp;
2169 bitmap->counts.pages = old_counts.pages;
2170 bitmap->counts.missing_pages = old_counts.pages;
2171 bitmap->counts.chunkshift = old_counts.chunkshift;
2172 bitmap->counts.chunks = old_counts.chunks;
2173 bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift +
2174 BITMAP_BLOCK_SHIFT);
2175 blocks = old_counts.chunks << old_counts.chunkshift;
2176 pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2177 break;
2178 } else
2179 bitmap->counts.bp[page].count += 1;
2183 for (block = 0; block < blocks; ) {
2184 bitmap_counter_t *bmc_old, *bmc_new;
2185 int set;
2187 bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2188 set = bmc_old && NEEDED(*bmc_old);
2190 if (set) {
2191 bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2192 if (*bmc_new == 0) {
2193 /* need to set on-disk bits too. */
2194 sector_t end = block + new_blocks;
2195 sector_t start = block >> chunkshift;
2196 start <<= chunkshift;
2197 while (start < end) {
2198 md_bitmap_file_set_bit(bitmap, block);
2199 start += 1 << chunkshift;
2201 *bmc_new = 2;
2202 md_bitmap_count_page(&bitmap->counts, block, 1);
2203 md_bitmap_set_pending(&bitmap->counts, block);
2205 *bmc_new |= NEEDED_MASK;
2206 if (new_blocks < old_blocks)
2207 old_blocks = new_blocks;
2209 block += old_blocks;
2212 if (bitmap->counts.bp != old_counts.bp) {
2213 unsigned long k;
2214 for (k = 0; k < old_counts.pages; k++)
2215 if (!old_counts.bp[k].hijacked)
2216 kfree(old_counts.bp[k].map);
2217 kfree(old_counts.bp);
2220 if (!init) {
2221 int i;
2222 while (block < (chunks << chunkshift)) {
2223 bitmap_counter_t *bmc;
2224 bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2225 if (bmc) {
2226 /* new space. It needs to be resynced, so
2227 * we set NEEDED_MASK.
2229 if (*bmc == 0) {
2230 *bmc = NEEDED_MASK | 2;
2231 md_bitmap_count_page(&bitmap->counts, block, 1);
2232 md_bitmap_set_pending(&bitmap->counts, block);
2235 block += new_blocks;
2237 for (i = 0; i < bitmap->storage.file_pages; i++)
2238 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2240 spin_unlock_irq(&bitmap->counts.lock);
2242 if (!init) {
2243 md_bitmap_unplug(bitmap);
2244 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2246 ret = 0;
2247 err:
2248 return ret;
2250 EXPORT_SYMBOL_GPL(md_bitmap_resize);
2252 static ssize_t
2253 location_show(struct mddev *mddev, char *page)
2255 ssize_t len;
2256 if (mddev->bitmap_info.file)
2257 len = sprintf(page, "file");
2258 else if (mddev->bitmap_info.offset)
2259 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2260 else
2261 len = sprintf(page, "none");
2262 len += sprintf(page+len, "\n");
2263 return len;
2266 static ssize_t
2267 location_store(struct mddev *mddev, const char *buf, size_t len)
2269 int rv;
2271 rv = mddev_lock(mddev);
2272 if (rv)
2273 return rv;
2274 if (mddev->pers) {
2275 if (!mddev->pers->quiesce) {
2276 rv = -EBUSY;
2277 goto out;
2279 if (mddev->recovery || mddev->sync_thread) {
2280 rv = -EBUSY;
2281 goto out;
2285 if (mddev->bitmap || mddev->bitmap_info.file ||
2286 mddev->bitmap_info.offset) {
2287 /* bitmap already configured. Only option is to clear it */
2288 if (strncmp(buf, "none", 4) != 0) {
2289 rv = -EBUSY;
2290 goto out;
2292 if (mddev->pers) {
2293 mddev_suspend(mddev);
2294 md_bitmap_destroy(mddev);
2295 mddev_resume(mddev);
2297 mddev->bitmap_info.offset = 0;
2298 if (mddev->bitmap_info.file) {
2299 struct file *f = mddev->bitmap_info.file;
2300 mddev->bitmap_info.file = NULL;
2301 fput(f);
2303 } else {
2304 /* No bitmap, OK to set a location */
2305 long long offset;
2306 if (strncmp(buf, "none", 4) == 0)
2307 /* nothing to be done */;
2308 else if (strncmp(buf, "file:", 5) == 0) {
2309 /* Not supported yet */
2310 rv = -EINVAL;
2311 goto out;
2312 } else {
2313 if (buf[0] == '+')
2314 rv = kstrtoll(buf+1, 10, &offset);
2315 else
2316 rv = kstrtoll(buf, 10, &offset);
2317 if (rv)
2318 goto out;
2319 if (offset == 0) {
2320 rv = -EINVAL;
2321 goto out;
2323 if (mddev->bitmap_info.external == 0 &&
2324 mddev->major_version == 0 &&
2325 offset != mddev->bitmap_info.default_offset) {
2326 rv = -EINVAL;
2327 goto out;
2329 mddev->bitmap_info.offset = offset;
2330 if (mddev->pers) {
2331 struct bitmap *bitmap;
2332 bitmap = md_bitmap_create(mddev, -1);
2333 mddev_suspend(mddev);
2334 if (IS_ERR(bitmap))
2335 rv = PTR_ERR(bitmap);
2336 else {
2337 mddev->bitmap = bitmap;
2338 rv = md_bitmap_load(mddev);
2339 if (rv)
2340 mddev->bitmap_info.offset = 0;
2342 if (rv) {
2343 md_bitmap_destroy(mddev);
2344 mddev_resume(mddev);
2345 goto out;
2347 mddev_resume(mddev);
2351 if (!mddev->external) {
2352 /* Ensure new bitmap info is stored in
2353 * metadata promptly.
2355 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2356 md_wakeup_thread(mddev->thread);
2358 rv = 0;
2359 out:
2360 mddev_unlock(mddev);
2361 if (rv)
2362 return rv;
2363 return len;
2366 static struct md_sysfs_entry bitmap_location =
2367 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2369 /* 'bitmap/space' is the space available at 'location' for the
2370 * bitmap. This allows the kernel to know when it is safe to
2371 * resize the bitmap to match a resized array.
2373 static ssize_t
2374 space_show(struct mddev *mddev, char *page)
2376 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2379 static ssize_t
2380 space_store(struct mddev *mddev, const char *buf, size_t len)
2382 unsigned long sectors;
2383 int rv;
2385 rv = kstrtoul(buf, 10, &sectors);
2386 if (rv)
2387 return rv;
2389 if (sectors == 0)
2390 return -EINVAL;
2392 if (mddev->bitmap &&
2393 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2394 return -EFBIG; /* Bitmap is too big for this small space */
2396 /* could make sure it isn't too big, but that isn't really
2397 * needed - user-space should be careful.
2399 mddev->bitmap_info.space = sectors;
2400 return len;
2403 static struct md_sysfs_entry bitmap_space =
2404 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2406 static ssize_t
2407 timeout_show(struct mddev *mddev, char *page)
2409 ssize_t len;
2410 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2411 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2413 len = sprintf(page, "%lu", secs);
2414 if (jifs)
2415 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2416 len += sprintf(page+len, "\n");
2417 return len;
2420 static ssize_t
2421 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2423 /* timeout can be set at any time */
2424 unsigned long timeout;
2425 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2426 if (rv)
2427 return rv;
2429 /* just to make sure we don't overflow... */
2430 if (timeout >= LONG_MAX / HZ)
2431 return -EINVAL;
2433 timeout = timeout * HZ / 10000;
2435 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2436 timeout = MAX_SCHEDULE_TIMEOUT-1;
2437 if (timeout < 1)
2438 timeout = 1;
2439 mddev->bitmap_info.daemon_sleep = timeout;
2440 if (mddev->thread) {
2441 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2442 * the bitmap is all clean and we don't need to
2443 * adjust the timeout right now
2445 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2446 mddev->thread->timeout = timeout;
2447 md_wakeup_thread(mddev->thread);
2450 return len;
2453 static struct md_sysfs_entry bitmap_timeout =
2454 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2456 static ssize_t
2457 backlog_show(struct mddev *mddev, char *page)
2459 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2462 static ssize_t
2463 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2465 unsigned long backlog;
2466 unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2467 int rv = kstrtoul(buf, 10, &backlog);
2468 if (rv)
2469 return rv;
2470 if (backlog > COUNTER_MAX)
2471 return -EINVAL;
2472 mddev->bitmap_info.max_write_behind = backlog;
2473 if (!backlog && mddev->serial_info_pool) {
2474 /* serial_info_pool is not needed if backlog is zero */
2475 if (!mddev->serialize_policy)
2476 mddev_destroy_serial_pool(mddev, NULL, false);
2477 } else if (backlog && !mddev->serial_info_pool) {
2478 /* serial_info_pool is needed since backlog is not zero */
2479 struct md_rdev *rdev;
2481 rdev_for_each(rdev, mddev)
2482 mddev_create_serial_pool(mddev, rdev, false);
2484 if (old_mwb != backlog)
2485 md_bitmap_update_sb(mddev->bitmap);
2486 return len;
2489 static struct md_sysfs_entry bitmap_backlog =
2490 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2492 static ssize_t
2493 chunksize_show(struct mddev *mddev, char *page)
2495 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2498 static ssize_t
2499 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2501 /* Can only be changed when no bitmap is active */
2502 int rv;
2503 unsigned long csize;
2504 if (mddev->bitmap)
2505 return -EBUSY;
2506 rv = kstrtoul(buf, 10, &csize);
2507 if (rv)
2508 return rv;
2509 if (csize < 512 ||
2510 !is_power_of_2(csize))
2511 return -EINVAL;
2512 mddev->bitmap_info.chunksize = csize;
2513 return len;
2516 static struct md_sysfs_entry bitmap_chunksize =
2517 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2519 static ssize_t metadata_show(struct mddev *mddev, char *page)
2521 if (mddev_is_clustered(mddev))
2522 return sprintf(page, "clustered\n");
2523 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2524 ? "external" : "internal"));
2527 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2529 if (mddev->bitmap ||
2530 mddev->bitmap_info.file ||
2531 mddev->bitmap_info.offset)
2532 return -EBUSY;
2533 if (strncmp(buf, "external", 8) == 0)
2534 mddev->bitmap_info.external = 1;
2535 else if ((strncmp(buf, "internal", 8) == 0) ||
2536 (strncmp(buf, "clustered", 9) == 0))
2537 mddev->bitmap_info.external = 0;
2538 else
2539 return -EINVAL;
2540 return len;
2543 static struct md_sysfs_entry bitmap_metadata =
2544 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2546 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2548 int len;
2549 spin_lock(&mddev->lock);
2550 if (mddev->bitmap)
2551 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2552 "false" : "true"));
2553 else
2554 len = sprintf(page, "\n");
2555 spin_unlock(&mddev->lock);
2556 return len;
2559 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2561 if (mddev->bitmap == NULL)
2562 return -ENOENT;
2563 if (strncmp(buf, "false", 5) == 0)
2564 mddev->bitmap->need_sync = 1;
2565 else if (strncmp(buf, "true", 4) == 0) {
2566 if (mddev->degraded)
2567 return -EBUSY;
2568 mddev->bitmap->need_sync = 0;
2569 } else
2570 return -EINVAL;
2571 return len;
2574 static struct md_sysfs_entry bitmap_can_clear =
2575 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2577 static ssize_t
2578 behind_writes_used_show(struct mddev *mddev, char *page)
2580 ssize_t ret;
2581 spin_lock(&mddev->lock);
2582 if (mddev->bitmap == NULL)
2583 ret = sprintf(page, "0\n");
2584 else
2585 ret = sprintf(page, "%lu\n",
2586 mddev->bitmap->behind_writes_used);
2587 spin_unlock(&mddev->lock);
2588 return ret;
2591 static ssize_t
2592 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2594 if (mddev->bitmap)
2595 mddev->bitmap->behind_writes_used = 0;
2596 return len;
2599 static struct md_sysfs_entry max_backlog_used =
2600 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2601 behind_writes_used_show, behind_writes_used_reset);
2603 static struct attribute *md_bitmap_attrs[] = {
2604 &bitmap_location.attr,
2605 &bitmap_space.attr,
2606 &bitmap_timeout.attr,
2607 &bitmap_backlog.attr,
2608 &bitmap_chunksize.attr,
2609 &bitmap_metadata.attr,
2610 &bitmap_can_clear.attr,
2611 &max_backlog_used.attr,
2612 NULL
2614 struct attribute_group md_bitmap_group = {
2615 .name = "bitmap",
2616 .attrs = md_bitmap_attrs,