Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
4 *
5 * bitmap_create - sets up the bitmap structure
6 * bitmap_destroy - destroys the bitmap structure
7 *
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
11 */
12
13/*
14 * Still to do:
15 *
16 * flush after percent set rather than just time based. (maybe both).
17 */
18
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
33#include "md.h"
34#include "md-bitmap.h"
35#include "md-cluster.h"
36
37/*
38 * in-memory bitmap:
39 *
40 * Use 16 bit block counters to track pending writes to each "chunk".
41 * The 2 high order bits are special-purpose, the first is a flag indicating
42 * whether a resync is needed. The second is a flag indicating whether a
43 * resync is active.
44 * This means that the counter is actually 14 bits:
45 *
46 * +--------+--------+------------------------------------------------+
47 * | resync | resync | counter |
48 * | needed | active | |
49 * | (0-1) | (0-1) | (0-16383) |
50 * +--------+--------+------------------------------------------------+
51 *
52 * The "resync needed" bit is set when:
53 * a '1' bit is read from storage at startup.
54 * a write request fails on some drives
55 * a resync is aborted on a chunk with 'resync active' set
56 * It is cleared (and resync-active set) when a resync starts across all drives
57 * of the chunk.
58 *
59 *
60 * The "resync active" bit is set when:
61 * a resync is started on all drives, and resync_needed is set.
62 * resync_needed will be cleared (as long as resync_active wasn't already set).
63 * It is cleared when a resync completes.
64 *
65 * The counter counts pending write requests, plus the on-disk bit.
66 * When the counter is '1' and the resync bits are clear, the on-disk
67 * bit can be cleared as well, thus setting the counter to 0.
68 * When we set a bit, or in the counter (to start a write), if the fields is
69 * 0, we first set the disk bit and set the counter to 1.
70 *
71 * If the counter is 0, the on-disk bit is clear and the stripe is clean
72 * Anything that dirties the stripe pushes the counter to 2 (at least)
73 * and sets the on-disk bit (lazily).
74 * If a periodic sweep find the counter at 2, it is decremented to 1.
75 * If the sweep find the counter at 1, the on-disk bit is cleared and the
76 * counter goes to zero.
77 *
78 * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
79 * counters as a fallback when "page" memory cannot be allocated:
80 *
81 * Normal case (page memory allocated):
82 *
83 * page pointer (32-bit)
84 *
85 * [ ] ------+
86 * |
87 * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters)
88 * c1 c2 c2048
89 *
90 * Hijacked case (page memory allocation failed):
91 *
92 * hijacked page pointer (32-bit)
93 *
94 * [ ][ ] (no page memory allocated)
95 * counter #1 (16-bit) counter #2 (16-bit)
96 *
97 */
98
99typedef __u16 bitmap_counter_t;
100
101#define PAGE_BITS (PAGE_SIZE << 3)
102#define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
103
104#define COUNTER_BITS 16
105#define COUNTER_BIT_SHIFT 4
106#define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3)
107
108#define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1)))
109#define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2)))
110#define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1)
111
112#define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK)
113#define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK)
114#define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX)
115
116/* how many counters per page? */
117#define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS)
118/* same, except a shift value for more efficient bitops */
119#define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT)
120/* same, except a mask value for more efficient bitops */
121#define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1)
122
123#define BITMAP_BLOCK_SHIFT 9
124
125/*
126 * bitmap structures:
127 */
128
129/* the in-memory bitmap is represented by bitmap_pages */
130struct bitmap_page {
131 /*
132 * map points to the actual memory page
133 */
134 char *map;
135 /*
136 * in emergencies (when map cannot be alloced), hijack the map
137 * pointer and use it as two counters itself
138 */
139 unsigned int hijacked:1;
140 /*
141 * If any counter in this page is '1' or '2' - and so could be
142 * cleared then that page is marked as 'pending'
143 */
144 unsigned int pending:1;
145 /*
146 * count of dirty bits on the page
147 */
148 unsigned int count:30;
149};
150
151/* the main bitmap structure - one per mddev */
152struct bitmap {
153
154 struct bitmap_counts {
155 spinlock_t lock;
156 struct bitmap_page *bp;
157 /* total number of pages in the bitmap */
158 unsigned long pages;
159 /* number of pages not yet allocated */
160 unsigned long missing_pages;
161 /* chunksize = 2^chunkshift (for bitops) */
162 unsigned long chunkshift;
163 /* total number of data chunks for the array */
164 unsigned long chunks;
165 } counts;
166
167 struct mddev *mddev; /* the md device that the bitmap is for */
168
169 __u64 events_cleared;
170 int need_sync;
171
172 struct bitmap_storage {
173 /* backing disk file */
174 struct file *file;
175 /* cached copy of the bitmap file superblock */
176 struct page *sb_page;
177 unsigned long sb_index;
178 /* list of cache pages for the file */
179 struct page **filemap;
180 /* attributes associated filemap pages */
181 unsigned long *filemap_attr;
182 /* number of pages in the file */
183 unsigned long file_pages;
184 /* total bytes in the bitmap */
185 unsigned long bytes;
186 } storage;
187
188 unsigned long flags;
189
190 int allclean;
191
192 atomic_t behind_writes;
193 /* highest actual value at runtime */
194 unsigned long behind_writes_used;
195
196 /*
197 * the bitmap daemon - periodically wakes up and sweeps the bitmap
198 * file, cleaning up bits and flushing out pages to disk as necessary
199 */
200 unsigned long daemon_lastrun; /* jiffies of last run */
201 /*
202 * when we lasted called end_sync to update bitmap with resync
203 * progress.
204 */
205 unsigned long last_end_sync;
206
207 /* pending writes to the bitmap file */
208 atomic_t pending_writes;
209 wait_queue_head_t write_wait;
210 wait_queue_head_t overflow_wait;
211 wait_queue_head_t behind_wait;
212
213 struct kernfs_node *sysfs_can_clear;
214 /* slot offset for clustered env */
215 int cluster_slot;
216};
217
218static struct workqueue_struct *md_bitmap_wq;
219static struct attribute_group md_bitmap_internal_group;
220
221static int __bitmap_resize(struct bitmap *bitmap, sector_t blocks,
222 int chunksize, bool init);
223
224static inline char *bmname(struct bitmap *bitmap)
225{
226 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
227}
228
229static bool bitmap_enabled(void *data, bool flush)
230{
231 struct bitmap *bitmap = data;
232
233 if (!flush)
234 return true;
235
236 /*
237 * If caller want to flush bitmap pages to underlying disks, check if
238 * there are cached pages in filemap.
239 */
240 return !test_bit(BITMAP_STALE, &bitmap->flags) &&
241 bitmap->storage.filemap != NULL;
242}
243
244/*
245 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
246 *
247 * 1) check to see if this page is allocated, if it's not then try to alloc
248 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
249 * page pointer directly as a counter
250 *
251 * if we find our page, we increment the page's refcount so that it stays
252 * allocated while we're using it
253 */
254static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
255 unsigned long page, int create, int no_hijack)
256__releases(bitmap->lock)
257__acquires(bitmap->lock)
258{
259 unsigned char *mappage;
260
261 WARN_ON_ONCE(page >= bitmap->pages);
262 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
263 return 0;
264
265 if (bitmap->bp[page].map) /* page is already allocated, just return */
266 return 0;
267
268 if (!create)
269 return -ENOENT;
270
271 /* this page has not been allocated yet */
272
273 spin_unlock_irq(&bitmap->lock);
274 /* It is possible that this is being called inside a
275 * prepare_to_wait/finish_wait loop from raid5c:make_request().
276 * In general it is not permitted to sleep in that context as it
277 * can cause the loop to spin freely.
278 * That doesn't apply here as we can only reach this point
279 * once with any loop.
280 * When this function completes, either bp[page].map or
281 * bp[page].hijacked. In either case, this function will
282 * abort before getting to this point again. So there is
283 * no risk of a free-spin, and so it is safe to assert
284 * that sleeping here is allowed.
285 */
286 sched_annotate_sleep();
287 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
288 spin_lock_irq(&bitmap->lock);
289
290 if (mappage == NULL) {
291 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
292 /* We don't support hijack for cluster raid */
293 if (no_hijack)
294 return -ENOMEM;
295 /* failed - set the hijacked flag so that we can use the
296 * pointer as a counter */
297 if (!bitmap->bp[page].map)
298 bitmap->bp[page].hijacked = 1;
299 } else if (bitmap->bp[page].map ||
300 bitmap->bp[page].hijacked) {
301 /* somebody beat us to getting the page */
302 kfree(mappage);
303 } else {
304
305 /* no page was in place and we have one, so install it */
306
307 bitmap->bp[page].map = mappage;
308 bitmap->missing_pages--;
309 }
310 return 0;
311}
312
313/* if page is completely empty, put it back on the free list, or dealloc it */
314/* if page was hijacked, unmark the flag so it might get alloced next time */
315/* Note: lock should be held when calling this */
316static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
317{
318 char *ptr;
319
320 if (bitmap->bp[page].count) /* page is still busy */
321 return;
322
323 /* page is no longer in use, it can be released */
324
325 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
326 bitmap->bp[page].hijacked = 0;
327 bitmap->bp[page].map = NULL;
328 } else {
329 /* normal case, free the page */
330 ptr = bitmap->bp[page].map;
331 bitmap->bp[page].map = NULL;
332 bitmap->missing_pages++;
333 kfree(ptr);
334 }
335}
336
337/*
338 * bitmap file handling - read and write the bitmap file and its superblock
339 */
340
341/*
342 * basic page I/O operations
343 */
344
345/* IO operations when bitmap is stored near all superblocks */
346
347/* choose a good rdev and read the page from there */
348static int read_sb_page(struct mddev *mddev, loff_t offset,
349 struct page *page, unsigned long index, int size)
350{
351
352 sector_t sector = mddev->bitmap_info.offset + offset +
353 index * (PAGE_SIZE / SECTOR_SIZE);
354 struct md_rdev *rdev;
355
356 rdev_for_each(rdev, mddev) {
357 u32 iosize = roundup(size, bdev_logical_block_size(rdev->bdev));
358
359 if (!test_bit(In_sync, &rdev->flags) ||
360 test_bit(Faulty, &rdev->flags) ||
361 test_bit(Bitmap_sync, &rdev->flags))
362 continue;
363
364 if (sync_page_io(rdev, sector, iosize, page, REQ_OP_READ, true))
365 return 0;
366 }
367 return -EIO;
368}
369
370static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
371{
372 /* Iterate the disks of an mddev, using rcu to protect access to the
373 * linked list, and raising the refcount of devices we return to ensure
374 * they don't disappear while in use.
375 * As devices are only added or removed when raid_disk is < 0 and
376 * nr_pending is 0 and In_sync is clear, the entries we return will
377 * still be in the same position on the list when we re-enter
378 * list_for_each_entry_continue_rcu.
379 *
380 * Note that if entered with 'rdev == NULL' to start at the
381 * beginning, we temporarily assign 'rdev' to an address which
382 * isn't really an rdev, but which can be used by
383 * list_for_each_entry_continue_rcu() to find the first entry.
384 */
385 rcu_read_lock();
386 if (rdev == NULL)
387 /* start at the beginning */
388 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
389 else {
390 /* release the previous rdev and start from there. */
391 rdev_dec_pending(rdev, mddev);
392 }
393 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
394 if (rdev->raid_disk >= 0 &&
395 !test_bit(Faulty, &rdev->flags)) {
396 /* this is a usable devices */
397 atomic_inc(&rdev->nr_pending);
398 rcu_read_unlock();
399 return rdev;
400 }
401 }
402 rcu_read_unlock();
403 return NULL;
404}
405
406static unsigned int optimal_io_size(struct block_device *bdev,
407 unsigned int last_page_size,
408 unsigned int io_size)
409{
410 if (bdev_io_opt(bdev) > bdev_logical_block_size(bdev))
411 return roundup(last_page_size, bdev_io_opt(bdev));
412 return io_size;
413}
414
415static unsigned int bitmap_io_size(unsigned int io_size, unsigned int opt_size,
416 loff_t start, loff_t boundary)
417{
418 if (io_size != opt_size &&
419 start + opt_size / SECTOR_SIZE <= boundary)
420 return opt_size;
421 if (start + io_size / SECTOR_SIZE <= boundary)
422 return io_size;
423
424 /* Overflows boundary */
425 return 0;
426}
427
428static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
429 unsigned long pg_index, struct page *page)
430{
431 struct block_device *bdev;
432 struct mddev *mddev = bitmap->mddev;
433 struct bitmap_storage *store = &bitmap->storage;
434 unsigned long num_pages = bitmap->storage.file_pages;
435 unsigned int bitmap_limit = (num_pages - pg_index % num_pages) << PAGE_SHIFT;
436 loff_t sboff, offset = mddev->bitmap_info.offset;
437 sector_t ps = pg_index * PAGE_SIZE / SECTOR_SIZE;
438 unsigned int size = PAGE_SIZE;
439 unsigned int opt_size = PAGE_SIZE;
440 sector_t doff;
441
442 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
443 /* we compare length (page numbers), not page offset. */
444 if ((pg_index - store->sb_index) == num_pages - 1) {
445 unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
446
447 if (last_page_size == 0)
448 last_page_size = PAGE_SIZE;
449 size = roundup(last_page_size, bdev_logical_block_size(bdev));
450 opt_size = optimal_io_size(bdev, last_page_size, size);
451 }
452
453 sboff = rdev->sb_start + offset;
454 doff = rdev->data_offset;
455
456 /* Just make sure we aren't corrupting data or metadata */
457 if (mddev->external) {
458 /* Bitmap could be anywhere. */
459 if (sboff + ps > doff &&
460 sboff < (doff + mddev->dev_sectors + PAGE_SIZE / SECTOR_SIZE))
461 return -EINVAL;
462 } else if (offset < 0) {
463 /* DATA BITMAP METADATA */
464 size = bitmap_io_size(size, opt_size, offset + ps, 0);
465 if (size == 0)
466 /* bitmap runs in to metadata */
467 return -EINVAL;
468
469 if (doff + mddev->dev_sectors > sboff)
470 /* data runs in to bitmap */
471 return -EINVAL;
472 } else if (rdev->sb_start < rdev->data_offset) {
473 /* METADATA BITMAP DATA */
474 size = bitmap_io_size(size, opt_size, sboff + ps, doff);
475 if (size == 0)
476 /* bitmap runs in to data */
477 return -EINVAL;
478 }
479
480 md_write_metadata(mddev, rdev, sboff + ps, (int)min(size, bitmap_limit),
481 page, 0);
482 return 0;
483}
484
485static void write_sb_page(struct bitmap *bitmap, unsigned long pg_index,
486 struct page *page, bool wait)
487{
488 struct mddev *mddev = bitmap->mddev;
489
490 do {
491 struct md_rdev *rdev = NULL;
492
493 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
494 if (__write_sb_page(rdev, bitmap, pg_index, page) < 0) {
495 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
496 return;
497 }
498 }
499 } while (wait && md_super_wait(mddev) < 0);
500}
501
502static void md_bitmap_file_kick(struct bitmap *bitmap);
503
504#ifdef CONFIG_MD_BITMAP_FILE
505static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
506{
507 struct buffer_head *bh = page_buffers(page);
508
509 while (bh && bh->b_blocknr) {
510 atomic_inc(&bitmap->pending_writes);
511 set_buffer_locked(bh);
512 set_buffer_mapped(bh);
513 submit_bh(REQ_OP_WRITE | REQ_SYNC, bh);
514 bh = bh->b_this_page;
515 }
516
517 if (wait)
518 wait_event(bitmap->write_wait,
519 atomic_read(&bitmap->pending_writes) == 0);
520}
521
522static void end_bitmap_write(struct buffer_head *bh, int uptodate)
523{
524 struct bitmap *bitmap = bh->b_private;
525
526 if (!uptodate)
527 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
528 if (atomic_dec_and_test(&bitmap->pending_writes))
529 wake_up(&bitmap->write_wait);
530}
531
532static void free_buffers(struct page *page)
533{
534 struct buffer_head *bh;
535
536 if (!PagePrivate(page))
537 return;
538
539 bh = page_buffers(page);
540 while (bh) {
541 struct buffer_head *next = bh->b_this_page;
542 free_buffer_head(bh);
543 bh = next;
544 }
545 detach_page_private(page);
546 put_page(page);
547}
548
549/* read a page from a file.
550 * We both read the page, and attach buffers to the page to record the
551 * address of each block (using bmap). These addresses will be used
552 * to write the block later, completely bypassing the filesystem.
553 * This usage is similar to how swap files are handled, and allows us
554 * to write to a file with no concerns of memory allocation failing.
555 */
556static int read_file_page(struct file *file, unsigned long index,
557 struct bitmap *bitmap, unsigned long count, struct page *page)
558{
559 int ret = 0;
560 struct inode *inode = file_inode(file);
561 struct buffer_head *bh;
562 sector_t block, blk_cur;
563 unsigned long blocksize = i_blocksize(inode);
564
565 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
566 (unsigned long long)index << PAGE_SHIFT);
567
568 bh = alloc_page_buffers(page, blocksize);
569 if (!bh) {
570 ret = -ENOMEM;
571 goto out;
572 }
573 attach_page_private(page, bh);
574 blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
575 while (bh) {
576 block = blk_cur;
577
578 if (count == 0)
579 bh->b_blocknr = 0;
580 else {
581 ret = bmap(inode, &block);
582 if (ret || !block) {
583 ret = -EINVAL;
584 bh->b_blocknr = 0;
585 goto out;
586 }
587
588 bh->b_blocknr = block;
589 bh->b_bdev = inode->i_sb->s_bdev;
590 if (count < blocksize)
591 count = 0;
592 else
593 count -= blocksize;
594
595 bh->b_end_io = end_bitmap_write;
596 bh->b_private = bitmap;
597 atomic_inc(&bitmap->pending_writes);
598 set_buffer_locked(bh);
599 set_buffer_mapped(bh);
600 submit_bh(REQ_OP_READ, bh);
601 }
602 blk_cur++;
603 bh = bh->b_this_page;
604 }
605
606 wait_event(bitmap->write_wait,
607 atomic_read(&bitmap->pending_writes)==0);
608 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
609 ret = -EIO;
610out:
611 if (ret)
612 pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
613 (int)PAGE_SIZE,
614 (unsigned long long)index << PAGE_SHIFT,
615 ret);
616 return ret;
617}
618#else /* CONFIG_MD_BITMAP_FILE */
619static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
620{
621}
622static int read_file_page(struct file *file, unsigned long index,
623 struct bitmap *bitmap, unsigned long count, struct page *page)
624{
625 return -EIO;
626}
627static void free_buffers(struct page *page)
628{
629 put_page(page);
630}
631#endif /* CONFIG_MD_BITMAP_FILE */
632
633/*
634 * bitmap file superblock operations
635 */
636
637/*
638 * write out a page to a file
639 */
640static void filemap_write_page(struct bitmap *bitmap, unsigned long pg_index,
641 bool wait)
642{
643 struct bitmap_storage *store = &bitmap->storage;
644 struct page *page = store->filemap[pg_index];
645
646 if (mddev_is_clustered(bitmap->mddev)) {
647 /* go to node bitmap area starting point */
648 pg_index += store->sb_index;
649 }
650
651 if (store->file)
652 write_file_page(bitmap, page, wait);
653 else
654 write_sb_page(bitmap, pg_index, page, wait);
655}
656
657/*
658 * md_bitmap_wait_writes() should be called before writing any bitmap
659 * blocks, to ensure previous writes, particularly from
660 * md_bitmap_daemon_work(), have completed.
661 */
662static void md_bitmap_wait_writes(struct bitmap *bitmap)
663{
664 if (bitmap->storage.file)
665 wait_event(bitmap->write_wait,
666 atomic_read(&bitmap->pending_writes)==0);
667 else
668 /* Note that we ignore the return value. The writes
669 * might have failed, but that would just mean that
670 * some bits which should be cleared haven't been,
671 * which is safe. The relevant bitmap blocks will
672 * probably get written again, but there is no great
673 * loss if they aren't.
674 */
675 md_super_wait(bitmap->mddev);
676}
677
678
679/* update the event counter and sync the superblock to disk */
680static void bitmap_update_sb(void *data)
681{
682 bitmap_super_t *sb;
683 struct bitmap *bitmap = data;
684
685 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
686 return;
687 if (bitmap->mddev->bitmap_info.external)
688 return;
689 if (!bitmap->storage.sb_page) /* no superblock */
690 return;
691 sb = kmap_local_page(bitmap->storage.sb_page);
692 sb->events = cpu_to_le64(bitmap->mddev->events);
693 if (bitmap->mddev->events < bitmap->events_cleared)
694 /* rocking back to read-only */
695 bitmap->events_cleared = bitmap->mddev->events;
696 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
697 /*
698 * clear BITMAP_WRITE_ERROR bit to protect against the case that
699 * a bitmap write error occurred but the later writes succeeded.
700 */
701 sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
702 /* Just in case these have been changed via sysfs: */
703 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
704 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
705 /* This might have been changed by a reshape */
706 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
707 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
708 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
709 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
710 bitmap_info.space);
711 kunmap_local(sb);
712
713 if (bitmap->storage.file)
714 write_file_page(bitmap, bitmap->storage.sb_page, 1);
715 else
716 write_sb_page(bitmap, bitmap->storage.sb_index,
717 bitmap->storage.sb_page, 1);
718}
719
720static void bitmap_print_sb(struct bitmap *bitmap)
721{
722 bitmap_super_t *sb;
723
724 if (!bitmap || !bitmap->storage.sb_page)
725 return;
726 sb = kmap_local_page(bitmap->storage.sb_page);
727 pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
728 pr_debug(" magic: %08x\n", le32_to_cpu(sb->magic));
729 pr_debug(" version: %u\n", le32_to_cpu(sb->version));
730 pr_debug(" uuid: %08x.%08x.%08x.%08x\n",
731 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
732 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
733 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
734 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
735 pr_debug(" events: %llu\n",
736 (unsigned long long) le64_to_cpu(sb->events));
737 pr_debug("events cleared: %llu\n",
738 (unsigned long long) le64_to_cpu(sb->events_cleared));
739 pr_debug(" state: %08x\n", le32_to_cpu(sb->state));
740 pr_debug(" chunksize: %u B\n", le32_to_cpu(sb->chunksize));
741 pr_debug(" daemon sleep: %us\n", le32_to_cpu(sb->daemon_sleep));
742 pr_debug(" sync size: %llu KB\n",
743 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
744 pr_debug("max write behind: %u\n", le32_to_cpu(sb->write_behind));
745 kunmap_local(sb);
746}
747
748/*
749 * bitmap_new_disk_sb
750 * @bitmap
751 *
752 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
753 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
754 * This function verifies 'bitmap_info' and populates the on-disk bitmap
755 * structure, which is to be written to disk.
756 *
757 * Returns: 0 on success, -Exxx on error
758 */
759static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
760{
761 bitmap_super_t *sb;
762 unsigned long chunksize, daemon_sleep, write_behind;
763
764 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
765 if (bitmap->storage.sb_page == NULL)
766 return -ENOMEM;
767 bitmap->storage.sb_index = 0;
768
769 sb = kmap_local_page(bitmap->storage.sb_page);
770
771 sb->magic = cpu_to_le32(BITMAP_MAGIC);
772 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
773
774 chunksize = bitmap->mddev->bitmap_info.chunksize;
775 BUG_ON(!chunksize);
776 if (!is_power_of_2(chunksize)) {
777 kunmap_local(sb);
778 pr_warn("bitmap chunksize not a power of 2\n");
779 return -EINVAL;
780 }
781 sb->chunksize = cpu_to_le32(chunksize);
782
783 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
784 if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
785 pr_debug("Choosing daemon_sleep default (5 sec)\n");
786 daemon_sleep = 5 * HZ;
787 }
788 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
789 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
790
791 /*
792 * FIXME: write_behind for RAID1. If not specified, what
793 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
794 */
795 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
796 if (write_behind > COUNTER_MAX / 2)
797 write_behind = COUNTER_MAX / 2;
798 sb->write_behind = cpu_to_le32(write_behind);
799 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
800
801 /* keep the array size field of the bitmap superblock up to date */
802 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
803
804 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
805
806 set_bit(BITMAP_STALE, &bitmap->flags);
807 sb->state = cpu_to_le32(bitmap->flags);
808 bitmap->events_cleared = bitmap->mddev->events;
809 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
810 bitmap->mddev->bitmap_info.nodes = 0;
811
812 kunmap_local(sb);
813
814 return 0;
815}
816
817/* read the superblock from the bitmap file and initialize some bitmap fields */
818static int md_bitmap_read_sb(struct bitmap *bitmap)
819{
820 char *reason = NULL;
821 bitmap_super_t *sb;
822 unsigned long chunksize, daemon_sleep, write_behind;
823 unsigned long long events;
824 int nodes = 0;
825 unsigned long sectors_reserved = 0;
826 int err = -EINVAL;
827 struct page *sb_page;
828 loff_t offset = 0;
829
830 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
831 chunksize = 128 * 1024 * 1024;
832 daemon_sleep = 5 * HZ;
833 write_behind = 0;
834 set_bit(BITMAP_STALE, &bitmap->flags);
835 err = 0;
836 goto out_no_sb;
837 }
838 /* page 0 is the superblock, read it... */
839 sb_page = alloc_page(GFP_KERNEL);
840 if (!sb_page)
841 return -ENOMEM;
842 bitmap->storage.sb_page = sb_page;
843
844re_read:
845 /* If cluster_slot is set, the cluster is setup */
846 if (bitmap->cluster_slot >= 0) {
847 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
848
849 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
850 (bitmap->mddev->bitmap_info.chunksize >> 9));
851 /* bits to bytes */
852 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
853 /* to 4k blocks */
854 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
855 offset = bitmap->cluster_slot * (bm_blocks << 3);
856 pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
857 bitmap->cluster_slot, offset);
858 }
859
860 if (bitmap->storage.file) {
861 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
862 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
863
864 err = read_file_page(bitmap->storage.file, 0,
865 bitmap, bytes, sb_page);
866 } else {
867 err = read_sb_page(bitmap->mddev, offset, sb_page, 0,
868 sizeof(bitmap_super_t));
869 }
870 if (err)
871 return err;
872
873 err = -EINVAL;
874 sb = kmap_local_page(sb_page);
875
876 chunksize = le32_to_cpu(sb->chunksize);
877 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
878 write_behind = le32_to_cpu(sb->write_behind);
879 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
880
881 /* verify that the bitmap-specific fields are valid */
882 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
883 reason = "bad magic";
884 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
885 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
886 reason = "unrecognized superblock version";
887 else if (chunksize < 512)
888 reason = "bitmap chunksize too small";
889 else if (!is_power_of_2(chunksize))
890 reason = "bitmap chunksize not a power of 2";
891 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
892 reason = "daemon sleep period out of range";
893 else if (write_behind > COUNTER_MAX)
894 reason = "write-behind limit out of range (0 - 16383)";
895 if (reason) {
896 pr_warn("%s: invalid bitmap file superblock: %s\n",
897 bmname(bitmap), reason);
898 goto out;
899 }
900
901 /*
902 * Setup nodes/clustername only if bitmap version is
903 * cluster-compatible
904 */
905 if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
906 nodes = le32_to_cpu(sb->nodes);
907 strscpy(bitmap->mddev->bitmap_info.cluster_name,
908 sb->cluster_name, 64);
909 }
910
911 /* keep the array size field of the bitmap superblock up to date */
912 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
913
914 if (bitmap->mddev->persistent) {
915 /*
916 * We have a persistent array superblock, so compare the
917 * bitmap's UUID and event counter to the mddev's
918 */
919 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
920 pr_warn("%s: bitmap superblock UUID mismatch\n",
921 bmname(bitmap));
922 goto out;
923 }
924 events = le64_to_cpu(sb->events);
925 if (!nodes && (events < bitmap->mddev->events)) {
926 pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
927 bmname(bitmap), events,
928 (unsigned long long) bitmap->mddev->events);
929 set_bit(BITMAP_STALE, &bitmap->flags);
930 }
931 }
932
933 /* assign fields using values from superblock */
934 bitmap->flags |= le32_to_cpu(sb->state);
935 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
936 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
937 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
938 err = 0;
939
940out:
941 kunmap_local(sb);
942 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
943 /* Assigning chunksize is required for "re_read" */
944 bitmap->mddev->bitmap_info.chunksize = chunksize;
945 err = md_setup_cluster(bitmap->mddev, nodes);
946 if (err) {
947 pr_warn("%s: Could not setup cluster service (%d)\n",
948 bmname(bitmap), err);
949 goto out_no_sb;
950 }
951 bitmap->cluster_slot = bitmap->mddev->cluster_ops->slot_number(bitmap->mddev);
952 goto re_read;
953 }
954
955out_no_sb:
956 if (err == 0) {
957 if (test_bit(BITMAP_STALE, &bitmap->flags))
958 bitmap->events_cleared = bitmap->mddev->events;
959 bitmap->mddev->bitmap_info.chunksize = chunksize;
960 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
961 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
962 bitmap->mddev->bitmap_info.nodes = nodes;
963 if (bitmap->mddev->bitmap_info.space == 0 ||
964 bitmap->mddev->bitmap_info.space > sectors_reserved)
965 bitmap->mddev->bitmap_info.space = sectors_reserved;
966 } else {
967 bitmap_print_sb(bitmap);
968 if (bitmap->cluster_slot < 0)
969 md_cluster_stop(bitmap->mddev);
970 }
971 return err;
972}
973
974/*
975 * general bitmap file operations
976 */
977
978/*
979 * on-disk bitmap:
980 *
981 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
982 * file a page at a time. There's a superblock at the start of the file.
983 */
984/* calculate the index of the page that contains this bit */
985static inline unsigned long file_page_index(struct bitmap_storage *store,
986 unsigned long chunk)
987{
988 if (store->sb_page)
989 chunk += sizeof(bitmap_super_t) << 3;
990 return chunk >> PAGE_BIT_SHIFT;
991}
992
993/* calculate the (bit) offset of this bit within a page */
994static inline unsigned long file_page_offset(struct bitmap_storage *store,
995 unsigned long chunk)
996{
997 if (store->sb_page)
998 chunk += sizeof(bitmap_super_t) << 3;
999 return chunk & (PAGE_BITS - 1);
1000}
1001
1002/*
1003 * return a pointer to the page in the filemap that contains the given bit
1004 *
1005 */
1006static inline struct page *filemap_get_page(struct bitmap_storage *store,
1007 unsigned long chunk)
1008{
1009 if (file_page_index(store, chunk) >= store->file_pages)
1010 return NULL;
1011 return store->filemap[file_page_index(store, chunk)];
1012}
1013
1014static int md_bitmap_storage_alloc(struct bitmap_storage *store,
1015 unsigned long chunks, int with_super,
1016 int slot_number)
1017{
1018 int pnum, offset = 0;
1019 unsigned long num_pages;
1020 unsigned long bytes;
1021
1022 bytes = DIV_ROUND_UP(chunks, 8);
1023 if (with_super)
1024 bytes += sizeof(bitmap_super_t);
1025
1026 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
1027 offset = slot_number * num_pages;
1028
1029 store->filemap = kmalloc_objs(struct page *, num_pages);
1030 if (!store->filemap)
1031 return -ENOMEM;
1032
1033 if (with_super && !store->sb_page) {
1034 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
1035 if (store->sb_page == NULL)
1036 return -ENOMEM;
1037 }
1038
1039 pnum = 0;
1040 if (store->sb_page) {
1041 store->filemap[0] = store->sb_page;
1042 pnum = 1;
1043 store->sb_index = offset;
1044 }
1045
1046 for ( ; pnum < num_pages; pnum++) {
1047 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
1048 if (!store->filemap[pnum]) {
1049 store->file_pages = pnum;
1050 return -ENOMEM;
1051 }
1052 }
1053 store->file_pages = pnum;
1054
1055 /* We need 4 bits per page, rounded up to a multiple
1056 * of sizeof(unsigned long) */
1057 store->filemap_attr = kzalloc(
1058 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
1059 GFP_KERNEL);
1060 if (!store->filemap_attr)
1061 return -ENOMEM;
1062
1063 store->bytes = bytes;
1064
1065 return 0;
1066}
1067
1068static void md_bitmap_file_unmap(struct bitmap_storage *store)
1069{
1070 struct file *file = store->file;
1071 struct page *sb_page = store->sb_page;
1072 struct page **map = store->filemap;
1073 int pages = store->file_pages;
1074
1075 while (pages--)
1076 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
1077 free_buffers(map[pages]);
1078 kfree(map);
1079 kfree(store->filemap_attr);
1080
1081 if (sb_page)
1082 free_buffers(sb_page);
1083
1084 if (file) {
1085 struct inode *inode = file_inode(file);
1086 invalidate_mapping_pages(inode->i_mapping, 0, -1);
1087 fput(file);
1088 }
1089}
1090
1091/*
1092 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
1093 * then it is no longer reliable, so we stop using it and we mark the file
1094 * as failed in the superblock
1095 */
1096static void md_bitmap_file_kick(struct bitmap *bitmap)
1097{
1098 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
1099 bitmap_update_sb(bitmap);
1100
1101 if (bitmap->storage.file) {
1102 pr_warn("%s: kicking failed bitmap file %pD4 from array!\n",
1103 bmname(bitmap), bitmap->storage.file);
1104
1105 } else
1106 pr_warn("%s: disabling internal bitmap due to errors\n",
1107 bmname(bitmap));
1108 }
1109}
1110
1111enum bitmap_page_attr {
1112 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
1113 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
1114 * i.e. counter is 1 or 2. */
1115 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
1116};
1117
1118static inline void set_page_attr(struct bitmap *bitmap, int pnum,
1119 enum bitmap_page_attr attr)
1120{
1121 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1122}
1123
1124static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
1125 enum bitmap_page_attr attr)
1126{
1127 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1128}
1129
1130static inline int test_page_attr(struct bitmap *bitmap, int pnum,
1131 enum bitmap_page_attr attr)
1132{
1133 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1134}
1135
1136static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
1137 enum bitmap_page_attr attr)
1138{
1139 return test_and_clear_bit((pnum<<2) + attr,
1140 bitmap->storage.filemap_attr);
1141}
1142/*
1143 * bitmap_file_set_bit -- called before performing a write to the md device
1144 * to set (and eventually sync) a particular bit in the bitmap file
1145 *
1146 * we set the bit immediately, then we record the page number so that
1147 * when an unplug occurs, we can flush the dirty pages out to disk
1148 */
1149static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
1150{
1151 unsigned long bit;
1152 struct page *page;
1153 void *kaddr;
1154 unsigned long chunk = block >> bitmap->counts.chunkshift;
1155 struct bitmap_storage *store = &bitmap->storage;
1156 unsigned long index = file_page_index(store, chunk);
1157 unsigned long node_offset = 0;
1158
1159 index += store->sb_index;
1160 if (mddev_is_clustered(bitmap->mddev))
1161 node_offset = bitmap->cluster_slot * store->file_pages;
1162
1163 page = filemap_get_page(&bitmap->storage, chunk);
1164 if (!page)
1165 return;
1166 bit = file_page_offset(&bitmap->storage, chunk);
1167
1168 /* set the bit */
1169 kaddr = kmap_local_page(page);
1170 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1171 set_bit(bit, kaddr);
1172 else
1173 set_bit_le(bit, kaddr);
1174 kunmap_local(kaddr);
1175 pr_debug("set file bit %lu page %lu\n", bit, index);
1176 /* record page number so it gets flushed to disk when unplug occurs */
1177 set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_DIRTY);
1178}
1179
1180static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
1181{
1182 unsigned long bit;
1183 struct page *page;
1184 void *paddr;
1185 unsigned long chunk = block >> bitmap->counts.chunkshift;
1186 struct bitmap_storage *store = &bitmap->storage;
1187 unsigned long index = file_page_index(store, chunk);
1188 unsigned long node_offset = 0;
1189
1190 index += store->sb_index;
1191 if (mddev_is_clustered(bitmap->mddev))
1192 node_offset = bitmap->cluster_slot * store->file_pages;
1193
1194 page = filemap_get_page(&bitmap->storage, chunk);
1195 if (!page)
1196 return;
1197 bit = file_page_offset(&bitmap->storage, chunk);
1198 paddr = kmap_local_page(page);
1199 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1200 clear_bit(bit, paddr);
1201 else
1202 clear_bit_le(bit, paddr);
1203 kunmap_local(paddr);
1204 if (!test_page_attr(bitmap, index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
1205 set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_PENDING);
1206 bitmap->allclean = 0;
1207 }
1208}
1209
1210static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
1211{
1212 unsigned long bit;
1213 struct page *page;
1214 void *paddr;
1215 unsigned long chunk = block >> bitmap->counts.chunkshift;
1216 int set = 0;
1217
1218 page = filemap_get_page(&bitmap->storage, chunk);
1219 if (!page)
1220 return -EINVAL;
1221 bit = file_page_offset(&bitmap->storage, chunk);
1222 paddr = kmap_local_page(page);
1223 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1224 set = test_bit(bit, paddr);
1225 else
1226 set = test_bit_le(bit, paddr);
1227 kunmap_local(paddr);
1228 return set;
1229}
1230
1231/* this gets called when the md device is ready to unplug its underlying
1232 * (slave) device queues -- before we let any writes go down, we need to
1233 * sync the dirty pages of the bitmap file to disk */
1234static void __bitmap_unplug(struct bitmap *bitmap)
1235{
1236 unsigned long i;
1237 int dirty, need_write;
1238 int writing = 0;
1239
1240 if (!bitmap_enabled(bitmap, true))
1241 return;
1242
1243 /* look at each page to see if there are any set bits that need to be
1244 * flushed out to disk */
1245 for (i = 0; i < bitmap->storage.file_pages; i++) {
1246 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1247 need_write = test_and_clear_page_attr(bitmap, i,
1248 BITMAP_PAGE_NEEDWRITE);
1249 if (dirty || need_write) {
1250 if (!writing) {
1251 md_bitmap_wait_writes(bitmap);
1252 mddev_add_trace_msg(bitmap->mddev,
1253 "md bitmap_unplug");
1254 }
1255 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1256 filemap_write_page(bitmap, i, false);
1257 writing = 1;
1258 }
1259 }
1260 if (writing)
1261 md_bitmap_wait_writes(bitmap);
1262
1263 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1264 md_bitmap_file_kick(bitmap);
1265}
1266
1267struct bitmap_unplug_work {
1268 struct work_struct work;
1269 struct bitmap *bitmap;
1270 struct completion *done;
1271};
1272
1273static void md_bitmap_unplug_fn(struct work_struct *work)
1274{
1275 struct bitmap_unplug_work *unplug_work =
1276 container_of(work, struct bitmap_unplug_work, work);
1277
1278 __bitmap_unplug(unplug_work->bitmap);
1279 complete(unplug_work->done);
1280}
1281
1282static void bitmap_unplug_async(struct bitmap *bitmap)
1283{
1284 DECLARE_COMPLETION_ONSTACK(done);
1285 struct bitmap_unplug_work unplug_work;
1286
1287 INIT_WORK_ONSTACK(&unplug_work.work, md_bitmap_unplug_fn);
1288 unplug_work.bitmap = bitmap;
1289 unplug_work.done = &done;
1290
1291 queue_work(md_bitmap_wq, &unplug_work.work);
1292 wait_for_completion(&done);
1293 destroy_work_on_stack(&unplug_work.work);
1294}
1295
1296static void bitmap_unplug(struct mddev *mddev, bool sync)
1297{
1298 struct bitmap *bitmap = mddev->bitmap;
1299
1300 if (!bitmap)
1301 return;
1302
1303 if (sync)
1304 __bitmap_unplug(bitmap);
1305 else
1306 bitmap_unplug_async(bitmap);
1307}
1308
1309static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1310
1311/*
1312 * Initialize the in-memory bitmap from the on-disk bitmap and set up the memory
1313 * mapping of the bitmap file.
1314 *
1315 * Special case: If there's no bitmap file, or if the bitmap file had been
1316 * previously kicked from the array, we mark all the bits as 1's in order to
1317 * cause a full resync.
1318 *
1319 * We ignore all bits for sectors that end earlier than 'start'.
1320 * This is used when reading an out-of-date bitmap.
1321 */
1322static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1323{
1324 bool outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1325 struct mddev *mddev = bitmap->mddev;
1326 unsigned long chunks = bitmap->counts.chunks;
1327 struct bitmap_storage *store = &bitmap->storage;
1328 struct file *file = store->file;
1329 unsigned long node_offset = 0;
1330 unsigned long bit_cnt = 0;
1331 unsigned long i;
1332 int ret;
1333
1334 if (!file && !mddev->bitmap_info.offset) {
1335 /* No permanent bitmap - fill with '1s'. */
1336 store->filemap = NULL;
1337 store->file_pages = 0;
1338 for (i = 0; i < chunks ; i++) {
1339 /* if the disk bit is set, set the memory bit */
1340 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1341 >= start);
1342 md_bitmap_set_memory_bits(bitmap,
1343 (sector_t)i << bitmap->counts.chunkshift,
1344 needed);
1345 }
1346 return 0;
1347 }
1348
1349 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1350 pr_warn("%s: bitmap file too short %lu < %lu\n",
1351 bmname(bitmap),
1352 (unsigned long) i_size_read(file->f_mapping->host),
1353 store->bytes);
1354 ret = -ENOSPC;
1355 goto err;
1356 }
1357
1358 if (mddev_is_clustered(mddev))
1359 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1360
1361 for (i = 0; i < store->file_pages; i++) {
1362 struct page *page = store->filemap[i];
1363 int count;
1364
1365 /* unmap the old page, we're done with it */
1366 if (i == store->file_pages - 1)
1367 count = store->bytes - i * PAGE_SIZE;
1368 else
1369 count = PAGE_SIZE;
1370
1371 if (file)
1372 ret = read_file_page(file, i, bitmap, count, page);
1373 else
1374 ret = read_sb_page(mddev, 0, page, i + node_offset,
1375 count);
1376 if (ret)
1377 goto err;
1378 }
1379
1380 if (outofdate) {
1381 pr_warn("%s: bitmap file is out of date, doing full recovery\n",
1382 bmname(bitmap));
1383
1384 for (i = 0; i < store->file_pages; i++) {
1385 struct page *page = store->filemap[i];
1386 unsigned long offset = 0;
1387 void *paddr;
1388
1389 if (i == 0 && !mddev->bitmap_info.external)
1390 offset = sizeof(bitmap_super_t);
1391
1392 /*
1393 * If the bitmap is out of date, dirty the whole page
1394 * and write it out
1395 */
1396 paddr = kmap_local_page(page);
1397 memset(paddr + offset, 0xff, PAGE_SIZE - offset);
1398 kunmap_local(paddr);
1399
1400 filemap_write_page(bitmap, i, true);
1401 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) {
1402 ret = -EIO;
1403 goto err;
1404 }
1405 }
1406 }
1407
1408 for (i = 0; i < chunks; i++) {
1409 struct page *page = filemap_get_page(&bitmap->storage, i);
1410 unsigned long bit = file_page_offset(&bitmap->storage, i);
1411 void *paddr;
1412 bool was_set;
1413
1414 paddr = kmap_local_page(page);
1415 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1416 was_set = test_bit(bit, paddr);
1417 else
1418 was_set = test_bit_le(bit, paddr);
1419 kunmap_local(paddr);
1420
1421 if (was_set) {
1422 /* if the disk bit is set, set the memory bit */
1423 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1424 >= start);
1425 md_bitmap_set_memory_bits(bitmap,
1426 (sector_t)i << bitmap->counts.chunkshift,
1427 needed);
1428 bit_cnt++;
1429 }
1430 }
1431
1432 pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1433 bmname(bitmap), store->file_pages,
1434 bit_cnt, chunks);
1435
1436 return 0;
1437
1438 err:
1439 pr_warn("%s: bitmap initialisation failed: %d\n",
1440 bmname(bitmap), ret);
1441 return ret;
1442}
1443
1444/* just flag bitmap pages as needing to be written. */
1445static void bitmap_write_all(struct mddev *mddev)
1446{
1447 int i;
1448 struct bitmap *bitmap = mddev->bitmap;
1449
1450 if (!bitmap || !bitmap->storage.filemap)
1451 return;
1452
1453 /* Only one copy, so nothing needed */
1454 if (bitmap->storage.file)
1455 return;
1456
1457 for (i = 0; i < bitmap->storage.file_pages; i++)
1458 set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
1459 bitmap->allclean = 0;
1460}
1461
1462static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1463 sector_t offset, int inc)
1464{
1465 sector_t chunk = offset >> bitmap->chunkshift;
1466 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1467 bitmap->bp[page].count += inc;
1468 md_bitmap_checkfree(bitmap, page);
1469}
1470
1471static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1472{
1473 sector_t chunk = offset >> bitmap->chunkshift;
1474 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1475 struct bitmap_page *bp = &bitmap->bp[page];
1476
1477 if (!bp->pending)
1478 bp->pending = 1;
1479}
1480
1481static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1482 sector_t offset, sector_t *blocks,
1483 int create);
1484
1485static void mddev_set_timeout(struct mddev *mddev, unsigned long timeout,
1486 bool force)
1487{
1488 struct md_thread *thread;
1489
1490 rcu_read_lock();
1491 thread = rcu_dereference(mddev->thread);
1492
1493 if (!thread)
1494 goto out;
1495
1496 if (force || thread->timeout < MAX_SCHEDULE_TIMEOUT)
1497 thread->timeout = timeout;
1498
1499out:
1500 rcu_read_unlock();
1501}
1502
1503/*
1504 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1505 * out to disk
1506 */
1507static void bitmap_daemon_work(struct mddev *mddev)
1508{
1509 struct bitmap *bitmap;
1510 unsigned long j;
1511 unsigned long nextpage;
1512 sector_t blocks;
1513 struct bitmap_counts *counts;
1514
1515 /* Use a mutex to guard daemon_work against
1516 * bitmap_destroy.
1517 */
1518 mutex_lock(&mddev->bitmap_info.mutex);
1519 bitmap = mddev->bitmap;
1520 if (bitmap == NULL) {
1521 mutex_unlock(&mddev->bitmap_info.mutex);
1522 return;
1523 }
1524 if (time_before(jiffies, bitmap->daemon_lastrun
1525 + mddev->bitmap_info.daemon_sleep))
1526 goto done;
1527
1528 bitmap->daemon_lastrun = jiffies;
1529 if (bitmap->allclean) {
1530 mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
1531 goto done;
1532 }
1533 bitmap->allclean = 1;
1534
1535 mddev_add_trace_msg(bitmap->mddev, "md bitmap_daemon_work");
1536
1537 /* Any file-page which is PENDING now needs to be written.
1538 * So set NEEDWRITE now, then after we make any last-minute changes
1539 * we will write it.
1540 */
1541 for (j = 0; j < bitmap->storage.file_pages; j++)
1542 if (test_and_clear_page_attr(bitmap, j,
1543 BITMAP_PAGE_PENDING))
1544 set_page_attr(bitmap, j,
1545 BITMAP_PAGE_NEEDWRITE);
1546
1547 if (bitmap->need_sync &&
1548 mddev->bitmap_info.external == 0) {
1549 /* Arrange for superblock update as well as
1550 * other changes */
1551 bitmap_super_t *sb;
1552 bitmap->need_sync = 0;
1553 if (bitmap->storage.filemap) {
1554 sb = kmap_local_page(bitmap->storage.sb_page);
1555 sb->events_cleared =
1556 cpu_to_le64(bitmap->events_cleared);
1557 kunmap_local(sb);
1558 set_page_attr(bitmap, 0,
1559 BITMAP_PAGE_NEEDWRITE);
1560 }
1561 }
1562 /* Now look at the bitmap counters and if any are '2' or '1',
1563 * decrement and handle accordingly.
1564 */
1565 counts = &bitmap->counts;
1566 spin_lock_irq(&counts->lock);
1567 nextpage = 0;
1568 for (j = 0; j < counts->chunks; j++) {
1569 bitmap_counter_t *bmc;
1570 sector_t block = (sector_t)j << counts->chunkshift;
1571
1572 if (j == nextpage) {
1573 nextpage += PAGE_COUNTER_RATIO;
1574 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1575 j |= PAGE_COUNTER_MASK;
1576 continue;
1577 }
1578 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1579 }
1580
1581 bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1582 if (!bmc) {
1583 j |= PAGE_COUNTER_MASK;
1584 continue;
1585 }
1586 if (*bmc == 1 && !bitmap->need_sync) {
1587 /* We can clear the bit */
1588 *bmc = 0;
1589 md_bitmap_count_page(counts, block, -1);
1590 md_bitmap_file_clear_bit(bitmap, block);
1591 } else if (*bmc && *bmc <= 2) {
1592 *bmc = 1;
1593 md_bitmap_set_pending(counts, block);
1594 bitmap->allclean = 0;
1595 }
1596 }
1597 spin_unlock_irq(&counts->lock);
1598
1599 md_bitmap_wait_writes(bitmap);
1600 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1601 * DIRTY pages need to be written by bitmap_unplug so it can wait
1602 * for them.
1603 * If we find any DIRTY page we stop there and let bitmap_unplug
1604 * handle all the rest. This is important in the case where
1605 * the first blocking holds the superblock and it has been updated.
1606 * We mustn't write any other blocks before the superblock.
1607 */
1608 for (j = 0;
1609 j < bitmap->storage.file_pages
1610 && !test_bit(BITMAP_STALE, &bitmap->flags);
1611 j++) {
1612 if (test_page_attr(bitmap, j,
1613 BITMAP_PAGE_DIRTY))
1614 /* bitmap_unplug will handle the rest */
1615 break;
1616 if (bitmap->storage.filemap &&
1617 test_and_clear_page_attr(bitmap, j,
1618 BITMAP_PAGE_NEEDWRITE))
1619 filemap_write_page(bitmap, j, false);
1620 }
1621
1622 done:
1623 if (bitmap->allclean == 0)
1624 mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
1625 mutex_unlock(&mddev->bitmap_info.mutex);
1626}
1627
1628static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1629 sector_t offset, sector_t *blocks,
1630 int create)
1631__releases(bitmap->lock)
1632__acquires(bitmap->lock)
1633{
1634 /* If 'create', we might release the lock and reclaim it.
1635 * The lock must have been taken with interrupts enabled.
1636 * If !create, we don't release the lock.
1637 */
1638 sector_t chunk = offset >> bitmap->chunkshift;
1639 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1640 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1641 sector_t csize = ((sector_t)1) << bitmap->chunkshift;
1642 int err;
1643
1644 if (page >= bitmap->pages) {
1645 /*
1646 * This can happen if bitmap_start_sync goes beyond
1647 * End-of-device while looking for a whole page or
1648 * user set a huge number to sysfs bitmap_set_bits.
1649 */
1650 *blocks = csize - (offset & (csize - 1));
1651 return NULL;
1652 }
1653 err = md_bitmap_checkpage(bitmap, page, create, 0);
1654
1655 if (bitmap->bp[page].hijacked ||
1656 bitmap->bp[page].map == NULL)
1657 csize = ((sector_t)1) << (bitmap->chunkshift +
1658 PAGE_COUNTER_SHIFT);
1659
1660 *blocks = csize - (offset & (csize - 1));
1661
1662 if (err < 0)
1663 return NULL;
1664
1665 /* now locked ... */
1666
1667 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1668 /* should we use the first or second counter field
1669 * of the hijacked pointer? */
1670 int hi = (pageoff > PAGE_COUNTER_MASK);
1671 return &((bitmap_counter_t *)
1672 &bitmap->bp[page].map)[hi];
1673 } else /* page is allocated */
1674 return (bitmap_counter_t *)
1675 &(bitmap->bp[page].map[pageoff]);
1676}
1677
1678static void bitmap_start_write(struct mddev *mddev, sector_t offset,
1679 unsigned long sectors)
1680{
1681 struct bitmap *bitmap = mddev->bitmap;
1682
1683 if (!bitmap)
1684 return;
1685
1686 while (sectors) {
1687 sector_t blocks;
1688 bitmap_counter_t *bmc;
1689
1690 spin_lock_irq(&bitmap->counts.lock);
1691 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1692 if (!bmc) {
1693 spin_unlock_irq(&bitmap->counts.lock);
1694 return;
1695 }
1696
1697 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1698 DEFINE_WAIT(__wait);
1699 /* note that it is safe to do the prepare_to_wait
1700 * after the test as long as we do it before dropping
1701 * the spinlock.
1702 */
1703 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1704 TASK_UNINTERRUPTIBLE);
1705 spin_unlock_irq(&bitmap->counts.lock);
1706 schedule();
1707 finish_wait(&bitmap->overflow_wait, &__wait);
1708 continue;
1709 }
1710
1711 switch (*bmc) {
1712 case 0:
1713 md_bitmap_file_set_bit(bitmap, offset);
1714 md_bitmap_count_page(&bitmap->counts, offset, 1);
1715 fallthrough;
1716 case 1:
1717 *bmc = 2;
1718 }
1719
1720 (*bmc)++;
1721
1722 spin_unlock_irq(&bitmap->counts.lock);
1723
1724 offset += blocks;
1725 if (sectors > blocks)
1726 sectors -= blocks;
1727 else
1728 sectors = 0;
1729 }
1730}
1731
1732static void bitmap_end_write(struct mddev *mddev, sector_t offset,
1733 unsigned long sectors)
1734{
1735 struct bitmap *bitmap = mddev->bitmap;
1736
1737 if (!bitmap)
1738 return;
1739
1740 while (sectors) {
1741 sector_t blocks;
1742 unsigned long flags;
1743 bitmap_counter_t *bmc;
1744
1745 spin_lock_irqsave(&bitmap->counts.lock, flags);
1746 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1747 if (!bmc) {
1748 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1749 return;
1750 }
1751
1752 if (!bitmap->mddev->degraded) {
1753 if (bitmap->events_cleared < bitmap->mddev->events) {
1754 bitmap->events_cleared = bitmap->mddev->events;
1755 bitmap->need_sync = 1;
1756 sysfs_notify_dirent_safe(
1757 bitmap->sysfs_can_clear);
1758 }
1759 } else if (!NEEDED(*bmc)) {
1760 *bmc |= NEEDED_MASK;
1761 }
1762
1763 if (COUNTER(*bmc) == COUNTER_MAX)
1764 wake_up(&bitmap->overflow_wait);
1765
1766 (*bmc)--;
1767 if (*bmc <= 2) {
1768 md_bitmap_set_pending(&bitmap->counts, offset);
1769 bitmap->allclean = 0;
1770 }
1771 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1772 offset += blocks;
1773 if (sectors > blocks)
1774 sectors -= blocks;
1775 else
1776 sectors = 0;
1777 }
1778}
1779
1780static bool __bitmap_start_sync(struct bitmap *bitmap, sector_t offset,
1781 sector_t *blocks, bool degraded)
1782{
1783 bitmap_counter_t *bmc;
1784 bool rv = false;
1785
1786 spin_lock_irq(&bitmap->counts.lock);
1787 bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1788 if (bmc) {
1789 /* locked */
1790 if (RESYNC(*bmc)) {
1791 rv = true;
1792 } else if (NEEDED(*bmc)) {
1793 rv = true;
1794 if (!degraded) { /* don't set/clear bits if degraded */
1795 *bmc |= RESYNC_MASK;
1796 *bmc &= ~NEEDED_MASK;
1797 }
1798 }
1799 }
1800 spin_unlock_irq(&bitmap->counts.lock);
1801
1802 return rv;
1803}
1804
1805static bool bitmap_start_sync(struct mddev *mddev, sector_t offset,
1806 sector_t *blocks, bool degraded)
1807{
1808 /* bitmap_start_sync must always report on multiples of whole
1809 * pages, otherwise resync (which is very PAGE_SIZE based) will
1810 * get confused.
1811 * So call __bitmap_start_sync repeatedly (if needed) until
1812 * At least PAGE_SIZE>>9 blocks are covered.
1813 * Return the 'or' of the result.
1814 */
1815 bool rv = false;
1816 sector_t blocks1;
1817
1818 *blocks = 0;
1819 while (*blocks < (PAGE_SIZE>>9)) {
1820 rv |= __bitmap_start_sync(mddev->bitmap, offset,
1821 &blocks1, degraded);
1822 offset += blocks1;
1823 *blocks += blocks1;
1824 }
1825
1826 return rv;
1827}
1828
1829static void __bitmap_end_sync(struct bitmap *bitmap, sector_t offset,
1830 sector_t *blocks, bool aborted)
1831{
1832 bitmap_counter_t *bmc;
1833 unsigned long flags;
1834
1835 spin_lock_irqsave(&bitmap->counts.lock, flags);
1836 bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1837 if (bmc == NULL)
1838 goto unlock;
1839 /* locked */
1840 if (RESYNC(*bmc)) {
1841 *bmc &= ~RESYNC_MASK;
1842
1843 if (!NEEDED(*bmc) && aborted)
1844 *bmc |= NEEDED_MASK;
1845 else {
1846 if (*bmc <= 2) {
1847 md_bitmap_set_pending(&bitmap->counts, offset);
1848 bitmap->allclean = 0;
1849 }
1850 }
1851 }
1852 unlock:
1853 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1854}
1855
1856static void bitmap_end_sync(struct mddev *mddev, sector_t offset,
1857 sector_t *blocks)
1858{
1859 __bitmap_end_sync(mddev->bitmap, offset, blocks, true);
1860}
1861
1862static void bitmap_close_sync(struct mddev *mddev)
1863{
1864 /* Sync has finished, and any bitmap chunks that weren't synced
1865 * properly have been aborted. It remains to us to clear the
1866 * RESYNC bit wherever it is still on
1867 */
1868 sector_t sector = 0;
1869 sector_t blocks;
1870 struct bitmap *bitmap = mddev->bitmap;
1871
1872 if (!bitmap)
1873 return;
1874
1875 while (sector < bitmap->mddev->resync_max_sectors) {
1876 __bitmap_end_sync(bitmap, sector, &blocks, false);
1877 sector += blocks;
1878 }
1879}
1880
1881static void bitmap_cond_end_sync(struct mddev *mddev, sector_t sector,
1882 bool force)
1883{
1884 sector_t s = 0;
1885 sector_t blocks;
1886 struct bitmap *bitmap = mddev->bitmap;
1887
1888 if (!bitmap)
1889 return;
1890 if (sector == 0) {
1891 bitmap->last_end_sync = jiffies;
1892 return;
1893 }
1894 if (!force && time_before(jiffies, (bitmap->last_end_sync
1895 + bitmap->mddev->bitmap_info.daemon_sleep)))
1896 return;
1897 wait_event(bitmap->mddev->recovery_wait,
1898 atomic_read(&bitmap->mddev->recovery_active) == 0);
1899
1900 bitmap->mddev->curr_resync_completed = sector;
1901 set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1902 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1903 s = 0;
1904 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1905 __bitmap_end_sync(bitmap, s, &blocks, false);
1906 s += blocks;
1907 }
1908 bitmap->last_end_sync = jiffies;
1909 sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1910}
1911
1912static void bitmap_sync_with_cluster(struct mddev *mddev,
1913 sector_t old_lo, sector_t old_hi,
1914 sector_t new_lo, sector_t new_hi)
1915{
1916 struct bitmap *bitmap = mddev->bitmap;
1917 sector_t sector, blocks = 0;
1918
1919 for (sector = old_lo; sector < new_lo; ) {
1920 __bitmap_end_sync(bitmap, sector, &blocks, false);
1921 sector += blocks;
1922 }
1923 WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1924
1925 for (sector = old_hi; sector < new_hi; ) {
1926 bitmap_start_sync(mddev, sector, &blocks, false);
1927 sector += blocks;
1928 }
1929 WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1930}
1931
1932static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1933{
1934 /* For each chunk covered by any of these sectors, set the
1935 * counter to 2 and possibly set resync_needed. They should all
1936 * be 0 at this point
1937 */
1938
1939 sector_t secs;
1940 bitmap_counter_t *bmc;
1941 spin_lock_irq(&bitmap->counts.lock);
1942 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1943 if (!bmc) {
1944 spin_unlock_irq(&bitmap->counts.lock);
1945 return;
1946 }
1947 if (!*bmc) {
1948 *bmc = 2;
1949 md_bitmap_count_page(&bitmap->counts, offset, 1);
1950 md_bitmap_set_pending(&bitmap->counts, offset);
1951 bitmap->allclean = 0;
1952 }
1953 if (needed)
1954 *bmc |= NEEDED_MASK;
1955 spin_unlock_irq(&bitmap->counts.lock);
1956}
1957
1958/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1959static void bitmap_dirty_bits(struct mddev *mddev, unsigned long s,
1960 unsigned long e)
1961{
1962 unsigned long chunk;
1963 struct bitmap *bitmap = mddev->bitmap;
1964
1965 if (!bitmap)
1966 return;
1967
1968 for (chunk = s; chunk <= e; chunk++) {
1969 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1970
1971 md_bitmap_set_memory_bits(bitmap, sec, 1);
1972 md_bitmap_file_set_bit(bitmap, sec);
1973 if (sec < bitmap->mddev->resync_offset)
1974 /* We are asserting that the array is dirty,
1975 * so move the resync_offset address back so
1976 * that it is obvious that it is dirty
1977 */
1978 bitmap->mddev->resync_offset = sec;
1979 }
1980}
1981
1982static void bitmap_flush(struct mddev *mddev)
1983{
1984 struct bitmap *bitmap = mddev->bitmap;
1985 long sleep;
1986
1987 if (!bitmap) /* there was no bitmap */
1988 return;
1989
1990 /* run the daemon_work three time to ensure everything is flushed
1991 * that can be
1992 */
1993 sleep = mddev->bitmap_info.daemon_sleep * 2;
1994 bitmap->daemon_lastrun -= sleep;
1995 bitmap_daemon_work(mddev);
1996 bitmap->daemon_lastrun -= sleep;
1997 bitmap_daemon_work(mddev);
1998 bitmap->daemon_lastrun -= sleep;
1999 bitmap_daemon_work(mddev);
2000 if (mddev->bitmap_info.external)
2001 md_super_wait(mddev);
2002 bitmap_update_sb(bitmap);
2003}
2004
2005static void md_bitmap_free(void *data)
2006{
2007 unsigned long k, pages;
2008 struct bitmap_page *bp;
2009 struct bitmap *bitmap = data;
2010
2011 if (!bitmap) /* there was no bitmap */
2012 return;
2013
2014 if (bitmap->sysfs_can_clear)
2015 sysfs_put(bitmap->sysfs_can_clear);
2016
2017 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
2018 bitmap->cluster_slot == bitmap->mddev->cluster_ops->slot_number(bitmap->mddev))
2019 md_cluster_stop(bitmap->mddev);
2020
2021 /* Shouldn't be needed - but just in case.... */
2022 wait_event(bitmap->write_wait,
2023 atomic_read(&bitmap->pending_writes) == 0);
2024
2025 /* release the bitmap file */
2026 md_bitmap_file_unmap(&bitmap->storage);
2027
2028 bp = bitmap->counts.bp;
2029 pages = bitmap->counts.pages;
2030
2031 /* free all allocated memory */
2032
2033 if (bp) /* deallocate the page memory */
2034 for (k = 0; k < pages; k++)
2035 if (bp[k].map && !bp[k].hijacked)
2036 kfree(bp[k].map);
2037 kfree(bp);
2038 kfree(bitmap);
2039}
2040
2041static void bitmap_start_behind_write(struct mddev *mddev)
2042{
2043 struct bitmap *bitmap = mddev->bitmap;
2044 int bw;
2045
2046 atomic_inc(&bitmap->behind_writes);
2047 bw = atomic_read(&bitmap->behind_writes);
2048 if (bw > bitmap->behind_writes_used)
2049 bitmap->behind_writes_used = bw;
2050
2051 pr_debug("inc write-behind count %d/%lu\n",
2052 bw, bitmap->mddev->bitmap_info.max_write_behind);
2053}
2054
2055static void bitmap_end_behind_write(struct mddev *mddev)
2056{
2057 struct bitmap *bitmap = mddev->bitmap;
2058
2059 if (atomic_dec_and_test(&bitmap->behind_writes))
2060 wake_up(&bitmap->behind_wait);
2061 pr_debug("dec write-behind count %d/%lu\n",
2062 atomic_read(&bitmap->behind_writes),
2063 bitmap->mddev->bitmap_info.max_write_behind);
2064}
2065
2066static void bitmap_wait_behind_writes(struct mddev *mddev)
2067{
2068 struct bitmap *bitmap = mddev->bitmap;
2069
2070 /* wait for behind writes to complete */
2071 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
2072 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
2073 mdname(mddev));
2074 /* need to kick something here to make sure I/O goes? */
2075 wait_event(bitmap->behind_wait,
2076 atomic_read(&bitmap->behind_writes) == 0);
2077 }
2078}
2079
2080static void bitmap_destroy(struct mddev *mddev)
2081{
2082 struct bitmap *bitmap = mddev->bitmap;
2083
2084 if (!bitmap) /* there was no bitmap */
2085 return;
2086
2087 bitmap_wait_behind_writes(mddev);
2088 if (!test_bit(MD_SERIALIZE_POLICY, &mddev->flags))
2089 mddev_destroy_serial_pool(mddev, NULL);
2090
2091 mutex_lock(&mddev->bitmap_info.mutex);
2092 spin_lock(&mddev->lock);
2093 mddev->bitmap = NULL; /* disconnect from the md device */
2094 spin_unlock(&mddev->lock);
2095 mutex_unlock(&mddev->bitmap_info.mutex);
2096 mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
2097
2098 md_bitmap_free(bitmap);
2099}
2100
2101/*
2102 * initialize the bitmap structure
2103 * if this returns an error, bitmap_destroy must be called to do clean up
2104 * once mddev->bitmap is set
2105 */
2106static struct bitmap *__bitmap_create(struct mddev *mddev, int slot)
2107{
2108 struct bitmap *bitmap;
2109 sector_t blocks = mddev->resync_max_sectors;
2110 struct file *file = mddev->bitmap_info.file;
2111 int err;
2112 struct kernfs_node *bm = NULL;
2113
2114 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
2115
2116 BUG_ON(file && mddev->bitmap_info.offset);
2117
2118 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
2119 pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
2120 mdname(mddev));
2121 return ERR_PTR(-EBUSY);
2122 }
2123
2124 bitmap = kzalloc_obj(*bitmap);
2125 if (!bitmap)
2126 return ERR_PTR(-ENOMEM);
2127
2128 spin_lock_init(&bitmap->counts.lock);
2129 atomic_set(&bitmap->pending_writes, 0);
2130 init_waitqueue_head(&bitmap->write_wait);
2131 init_waitqueue_head(&bitmap->overflow_wait);
2132 init_waitqueue_head(&bitmap->behind_wait);
2133
2134 bitmap->mddev = mddev;
2135 bitmap->cluster_slot = slot;
2136
2137 if (mddev->kobj.sd)
2138 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
2139 if (bm) {
2140 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
2141 sysfs_put(bm);
2142 } else
2143 bitmap->sysfs_can_clear = NULL;
2144
2145 bitmap->storage.file = file;
2146 if (file) {
2147 get_file(file);
2148 /* As future accesses to this file will use bmap,
2149 * and bypass the page cache, we must sync the file
2150 * first.
2151 */
2152 vfs_fsync(file, 1);
2153 }
2154 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
2155 if (!mddev->bitmap_info.external) {
2156 /*
2157 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
2158 * instructing us to create a new on-disk bitmap instance.
2159 */
2160 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
2161 err = md_bitmap_new_disk_sb(bitmap);
2162 else
2163 err = md_bitmap_read_sb(bitmap);
2164 } else {
2165 err = 0;
2166 if (mddev->bitmap_info.chunksize == 0 ||
2167 mddev->bitmap_info.daemon_sleep == 0)
2168 /* chunksize and time_base need to be
2169 * set first. */
2170 err = -EINVAL;
2171 }
2172 if (err)
2173 goto error;
2174
2175 bitmap->daemon_lastrun = jiffies;
2176 err = __bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize,
2177 true);
2178 if (err)
2179 goto error;
2180
2181 pr_debug("created bitmap (%lu pages) for device %s\n",
2182 bitmap->counts.pages, bmname(bitmap));
2183
2184 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
2185 if (err)
2186 goto error;
2187
2188 return bitmap;
2189 error:
2190 md_bitmap_free(bitmap);
2191 return ERR_PTR(err);
2192}
2193
2194static int bitmap_create(struct mddev *mddev)
2195{
2196 struct bitmap *bitmap = __bitmap_create(mddev, -1);
2197
2198 if (IS_ERR(bitmap))
2199 return PTR_ERR(bitmap);
2200
2201 mddev->bitmap = bitmap;
2202 return 0;
2203}
2204
2205static int bitmap_load(struct mddev *mddev)
2206{
2207 int err = 0;
2208 sector_t start = 0;
2209 sector_t sector = 0;
2210 struct bitmap *bitmap = mddev->bitmap;
2211 struct md_rdev *rdev;
2212
2213 if (!bitmap)
2214 goto out;
2215
2216 rdev_for_each(rdev, mddev)
2217 mddev_create_serial_pool(mddev, rdev);
2218
2219 if (mddev_is_clustered(mddev))
2220 mddev->cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
2221
2222 /* Clear out old bitmap info first: Either there is none, or we
2223 * are resuming after someone else has possibly changed things,
2224 * so we should forget old cached info.
2225 * All chunks should be clean, but some might need_sync.
2226 */
2227 while (sector < mddev->resync_max_sectors) {
2228 sector_t blocks;
2229 bitmap_start_sync(mddev, sector, &blocks, false);
2230 sector += blocks;
2231 }
2232 bitmap_close_sync(mddev);
2233
2234 if (mddev->degraded == 0
2235 || bitmap->events_cleared == mddev->events)
2236 /* no need to keep dirty bits to optimise a
2237 * re-add of a missing device */
2238 start = mddev->resync_offset;
2239
2240 mutex_lock(&mddev->bitmap_info.mutex);
2241 err = md_bitmap_init_from_disk(bitmap, start);
2242 mutex_unlock(&mddev->bitmap_info.mutex);
2243
2244 if (err)
2245 goto out;
2246 clear_bit(BITMAP_STALE, &bitmap->flags);
2247
2248 /* Kick recovery in case any bits were set */
2249 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
2250
2251 mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
2252 md_wakeup_thread(mddev->thread);
2253
2254 bitmap_update_sb(bitmap);
2255
2256 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
2257 err = -EIO;
2258out:
2259 return err;
2260}
2261
2262/* caller need to free returned bitmap with md_bitmap_free() */
2263static void *bitmap_get_from_slot(struct mddev *mddev, int slot)
2264{
2265 int rv = 0;
2266 struct bitmap *bitmap;
2267
2268 bitmap = __bitmap_create(mddev, slot);
2269 if (IS_ERR(bitmap)) {
2270 rv = PTR_ERR(bitmap);
2271 return ERR_PTR(rv);
2272 }
2273
2274 rv = md_bitmap_init_from_disk(bitmap, 0);
2275 if (rv) {
2276 md_bitmap_free(bitmap);
2277 return ERR_PTR(rv);
2278 }
2279
2280 return bitmap;
2281}
2282
2283/* Loads the bitmap associated with slot and copies the resync information
2284 * to our bitmap
2285 */
2286static int bitmap_copy_from_slot(struct mddev *mddev, int slot, sector_t *low,
2287 sector_t *high, bool clear_bits)
2288{
2289 int rv = 0, i, j;
2290 sector_t block, lo = 0, hi = 0;
2291 struct bitmap_counts *counts;
2292 struct bitmap *bitmap;
2293
2294 bitmap = bitmap_get_from_slot(mddev, slot);
2295 if (IS_ERR(bitmap)) {
2296 pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
2297 return -1;
2298 }
2299
2300 counts = &bitmap->counts;
2301 for (j = 0; j < counts->chunks; j++) {
2302 block = (sector_t)j << counts->chunkshift;
2303 if (md_bitmap_file_test_bit(bitmap, block)) {
2304 if (!lo)
2305 lo = block;
2306 hi = block;
2307 md_bitmap_file_clear_bit(bitmap, block);
2308 md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2309 md_bitmap_file_set_bit(mddev->bitmap, block);
2310 }
2311 }
2312
2313 if (clear_bits) {
2314 bitmap_update_sb(bitmap);
2315 /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2316 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2317 for (i = 0; i < bitmap->storage.file_pages; i++)
2318 if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2319 set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2320 __bitmap_unplug(bitmap);
2321 }
2322 __bitmap_unplug(mddev->bitmap);
2323 *low = lo;
2324 *high = hi;
2325 md_bitmap_free(bitmap);
2326
2327 return rv;
2328}
2329
2330static void bitmap_set_pages(void *data, unsigned long pages)
2331{
2332 struct bitmap *bitmap = data;
2333
2334 bitmap->counts.pages = pages;
2335}
2336
2337static int bitmap_get_stats(void *data, struct md_bitmap_stats *stats)
2338{
2339 struct bitmap_storage *storage;
2340 struct bitmap_counts *counts;
2341 struct bitmap *bitmap = data;
2342 bitmap_super_t *sb;
2343
2344 if (!bitmap)
2345 return -ENOENT;
2346 if (!bitmap->storage.sb_page)
2347 return -EINVAL;
2348 sb = kmap_local_page(bitmap->storage.sb_page);
2349 stats->sync_size = le64_to_cpu(sb->sync_size);
2350 kunmap_local(sb);
2351
2352 counts = &bitmap->counts;
2353 stats->missing_pages = counts->missing_pages;
2354 stats->pages = counts->pages;
2355
2356 storage = &bitmap->storage;
2357 stats->file_pages = storage->file_pages;
2358 stats->file = storage->file;
2359
2360 stats->behind_writes = atomic_read(&bitmap->behind_writes);
2361 stats->behind_wait = wq_has_sleeper(&bitmap->behind_wait);
2362 stats->events_cleared = bitmap->events_cleared;
2363 return 0;
2364}
2365
2366static int __bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2367 int chunksize, bool init)
2368{
2369 /* If chunk_size is 0, choose an appropriate chunk size.
2370 * Then possibly allocate new storage space.
2371 * Then quiesce, copy bits, replace bitmap, and re-start
2372 *
2373 * This function is called both to set up the initial bitmap
2374 * and to resize the bitmap while the array is active.
2375 * If this happens as a result of the array being resized,
2376 * chunksize will be zero, and we need to choose a suitable
2377 * chunksize, otherwise we use what we are given.
2378 */
2379 struct bitmap_storage store;
2380 struct bitmap_counts old_counts;
2381 unsigned long chunks;
2382 sector_t block;
2383 sector_t old_blocks, new_blocks;
2384 int chunkshift;
2385 int ret = 0;
2386 long pages;
2387 struct bitmap_page *new_bp;
2388
2389 if (bitmap->storage.file && !init) {
2390 pr_info("md: cannot resize file-based bitmap\n");
2391 return -EINVAL;
2392 }
2393
2394 if (chunksize == 0) {
2395 /* If there is enough space, leave the chunk size unchanged,
2396 * else increase by factor of two until there is enough space.
2397 */
2398 long bytes;
2399 long space = bitmap->mddev->bitmap_info.space;
2400
2401 if (space == 0) {
2402 /* We don't know how much space there is, so limit
2403 * to current size - in sectors.
2404 */
2405 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2406 if (!bitmap->mddev->bitmap_info.external)
2407 bytes += sizeof(bitmap_super_t);
2408 space = DIV_ROUND_UP(bytes, 512);
2409 bitmap->mddev->bitmap_info.space = space;
2410 }
2411 chunkshift = bitmap->counts.chunkshift;
2412 chunkshift--;
2413 do {
2414 /* 'chunkshift' is shift from block size to chunk size */
2415 chunkshift++;
2416 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2417 bytes = DIV_ROUND_UP(chunks, 8);
2418 if (!bitmap->mddev->bitmap_info.external)
2419 bytes += sizeof(bitmap_super_t);
2420 } while (bytes > (space << 9) && (chunkshift + BITMAP_BLOCK_SHIFT) <
2421 (BITS_PER_BYTE * sizeof(((bitmap_super_t *)0)->chunksize) - 1));
2422 } else
2423 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2424
2425 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2426 memset(&store, 0, sizeof(store));
2427 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2428 ret = md_bitmap_storage_alloc(&store, chunks,
2429 !bitmap->mddev->bitmap_info.external,
2430 mddev_is_clustered(bitmap->mddev)
2431 ? bitmap->cluster_slot : 0);
2432 if (ret) {
2433 md_bitmap_file_unmap(&store);
2434 goto err;
2435 }
2436
2437 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2438
2439 new_bp = kzalloc_objs(*new_bp, pages);
2440 ret = -ENOMEM;
2441 if (!new_bp) {
2442 md_bitmap_file_unmap(&store);
2443 goto err;
2444 }
2445
2446 if (!init)
2447 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2448
2449 store.file = bitmap->storage.file;
2450 bitmap->storage.file = NULL;
2451
2452 if (store.sb_page && bitmap->storage.sb_page)
2453 memcpy(page_address(store.sb_page),
2454 page_address(bitmap->storage.sb_page),
2455 sizeof(bitmap_super_t));
2456 mutex_lock(&bitmap->mddev->bitmap_info.mutex);
2457 spin_lock_irq(&bitmap->counts.lock);
2458 md_bitmap_file_unmap(&bitmap->storage);
2459 bitmap->storage = store;
2460
2461 old_counts = bitmap->counts;
2462 bitmap->counts.bp = new_bp;
2463 bitmap->counts.pages = pages;
2464 bitmap->counts.missing_pages = pages;
2465 bitmap->counts.chunkshift = chunkshift;
2466 bitmap->counts.chunks = chunks;
2467 bitmap->mddev->bitmap_info.chunksize = 1UL << (chunkshift +
2468 BITMAP_BLOCK_SHIFT);
2469
2470 blocks = min(old_counts.chunks << old_counts.chunkshift,
2471 chunks << chunkshift);
2472
2473 /* For cluster raid, need to pre-allocate bitmap */
2474 if (mddev_is_clustered(bitmap->mddev)) {
2475 unsigned long page;
2476 for (page = 0; page < pages; page++) {
2477 ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2478 if (ret) {
2479 unsigned long k;
2480
2481 /* deallocate the page memory */
2482 for (k = 0; k < page; k++) {
2483 kfree(new_bp[k].map);
2484 }
2485 kfree(new_bp);
2486
2487 /* restore some fields from old_counts */
2488 bitmap->counts.bp = old_counts.bp;
2489 bitmap->counts.pages = old_counts.pages;
2490 bitmap->counts.missing_pages = old_counts.pages;
2491 bitmap->counts.chunkshift = old_counts.chunkshift;
2492 bitmap->counts.chunks = old_counts.chunks;
2493 bitmap->mddev->bitmap_info.chunksize =
2494 1UL << (old_counts.chunkshift + BITMAP_BLOCK_SHIFT);
2495 blocks = old_counts.chunks << old_counts.chunkshift;
2496 pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2497 break;
2498 } else
2499 bitmap->counts.bp[page].count += 1;
2500 }
2501 }
2502
2503 for (block = 0; block < blocks; ) {
2504 bitmap_counter_t *bmc_old, *bmc_new;
2505 int set;
2506
2507 bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2508 set = bmc_old && NEEDED(*bmc_old);
2509
2510 if (set) {
2511 bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2512 if (bmc_new) {
2513 if (*bmc_new == 0) {
2514 /* need to set on-disk bits too. */
2515 sector_t end = block + new_blocks;
2516 sector_t start = block >> chunkshift;
2517
2518 start <<= chunkshift;
2519 while (start < end) {
2520 md_bitmap_file_set_bit(bitmap, block);
2521 start += 1 << chunkshift;
2522 }
2523 *bmc_new = 2;
2524 md_bitmap_count_page(&bitmap->counts, block, 1);
2525 md_bitmap_set_pending(&bitmap->counts, block);
2526 }
2527 *bmc_new |= NEEDED_MASK;
2528 }
2529 if (new_blocks < old_blocks)
2530 old_blocks = new_blocks;
2531 }
2532 block += old_blocks;
2533 }
2534
2535 if (bitmap->counts.bp != old_counts.bp) {
2536 unsigned long k;
2537 for (k = 0; k < old_counts.pages; k++)
2538 if (!old_counts.bp[k].hijacked)
2539 kfree(old_counts.bp[k].map);
2540 kfree(old_counts.bp);
2541 }
2542
2543 if (!init) {
2544 int i;
2545 while (block < (chunks << chunkshift)) {
2546 bitmap_counter_t *bmc;
2547 bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2548 if (bmc) {
2549 /* new space. It needs to be resynced, so
2550 * we set NEEDED_MASK.
2551 */
2552 if (*bmc == 0) {
2553 *bmc = NEEDED_MASK | 2;
2554 md_bitmap_count_page(&bitmap->counts, block, 1);
2555 md_bitmap_set_pending(&bitmap->counts, block);
2556 }
2557 }
2558 block += new_blocks;
2559 }
2560 for (i = 0; i < bitmap->storage.file_pages; i++)
2561 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2562 }
2563 spin_unlock_irq(&bitmap->counts.lock);
2564 mutex_unlock(&bitmap->mddev->bitmap_info.mutex);
2565 if (!init) {
2566 __bitmap_unplug(bitmap);
2567 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2568 }
2569 ret = 0;
2570err:
2571 return ret;
2572}
2573
2574static int bitmap_resize(struct mddev *mddev, sector_t blocks, int chunksize)
2575{
2576 struct bitmap *bitmap = mddev->bitmap;
2577
2578 if (!bitmap)
2579 return 0;
2580
2581 return __bitmap_resize(bitmap, blocks, chunksize, false);
2582}
2583
2584static bool bitmap_none_enabled(void *data, bool flush)
2585{
2586 return false;
2587}
2588
2589static int bitmap_none_create(struct mddev *mddev)
2590{
2591 return 0;
2592}
2593
2594static int bitmap_none_load(struct mddev *mddev)
2595{
2596 return 0;
2597}
2598
2599static void bitmap_none_destroy(struct mddev *mddev)
2600{
2601}
2602
2603static int bitmap_none_get_stats(void *data, struct md_bitmap_stats *stats)
2604{
2605 return -ENOENT;
2606}
2607
2608static ssize_t
2609location_show(struct mddev *mddev, char *page)
2610{
2611 ssize_t len;
2612 if (mddev->bitmap_info.file)
2613 len = sprintf(page, "file");
2614 else if (mddev->bitmap_info.offset)
2615 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2616 else
2617 len = sprintf(page, "none");
2618 len += sprintf(page+len, "\n");
2619 return len;
2620}
2621
2622static ssize_t
2623location_store(struct mddev *mddev, const char *buf, size_t len)
2624{
2625 int rv;
2626
2627 rv = mddev_suspend_and_lock(mddev);
2628 if (rv)
2629 return rv;
2630
2631 if (mddev->pers) {
2632 if (mddev->recovery || mddev->sync_thread) {
2633 rv = -EBUSY;
2634 goto out;
2635 }
2636 }
2637
2638 if (mddev->bitmap || mddev->bitmap_info.file ||
2639 mddev->bitmap_info.offset) {
2640 /* bitmap already configured. Only option is to clear it */
2641 if (strncmp(buf, "none", 4) != 0) {
2642 rv = -EBUSY;
2643 goto out;
2644 }
2645
2646 sysfs_unmerge_group(&mddev->kobj, &md_bitmap_internal_group);
2647 md_bitmap_destroy_nosysfs(mddev);
2648 mddev->bitmap_id = ID_BITMAP_NONE;
2649 if (!mddev_set_bitmap_ops_nosysfs(mddev))
2650 goto none_err;
2651 mddev->bitmap_info.offset = 0;
2652 if (mddev->bitmap_info.file) {
2653 struct file *f = mddev->bitmap_info.file;
2654 mddev->bitmap_info.file = NULL;
2655 fput(f);
2656 }
2657 } else {
2658 /* No bitmap, OK to set a location */
2659 long long offset;
2660
2661 if (strncmp(buf, "none", 4) == 0)
2662 /* nothing to be done */;
2663 else if (strncmp(buf, "file:", 5) == 0) {
2664 /* Not supported yet */
2665 rv = -EINVAL;
2666 goto out;
2667 } else {
2668 if (buf[0] == '+')
2669 rv = kstrtoll(buf+1, 10, &offset);
2670 else
2671 rv = kstrtoll(buf, 10, &offset);
2672 if (rv)
2673 goto out;
2674 if (offset == 0) {
2675 rv = -EINVAL;
2676 goto out;
2677 }
2678 if (mddev->bitmap_info.external == 0 &&
2679 mddev->major_version == 0 &&
2680 offset != mddev->bitmap_info.default_offset) {
2681 rv = -EINVAL;
2682 goto out;
2683 }
2684
2685 mddev->bitmap_info.offset = offset;
2686 md_bitmap_destroy_nosysfs(mddev);
2687 mddev->bitmap_id = ID_BITMAP;
2688 if (!mddev_set_bitmap_ops_nosysfs(mddev))
2689 goto bitmap_err;
2690
2691 rv = md_bitmap_create_nosysfs(mddev);
2692 if (rv)
2693 goto create_err;
2694
2695 rv = mddev->bitmap_ops->load(mddev);
2696 if (rv) {
2697 mddev->bitmap_info.offset = 0;
2698 goto load_err;
2699 }
2700
2701 rv = sysfs_merge_group(&mddev->kobj,
2702 &md_bitmap_internal_group);
2703 if (rv)
2704 goto merge_err;
2705 }
2706 }
2707 if (!mddev->external) {
2708 /* Ensure new bitmap info is stored in
2709 * metadata promptly.
2710 */
2711 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2712 md_wakeup_thread(mddev->thread);
2713 }
2714 rv = 0;
2715out:
2716 mddev_unlock_and_resume(mddev);
2717 if (rv)
2718 return rv;
2719 return len;
2720
2721merge_err:
2722 mddev->bitmap_info.offset = 0;
2723load_err:
2724 md_bitmap_destroy_nosysfs(mddev);
2725create_err:
2726 mddev->bitmap_info.offset = 0;
2727 mddev->bitmap_id = ID_BITMAP_NONE;
2728 if (!mddev_set_bitmap_ops_nosysfs(mddev))
2729 rv = -ENOENT;
2730 goto out;
2731bitmap_err:
2732 rv = -ENOENT;
2733none_err:
2734 mddev->bitmap_info.offset = 0;
2735 goto out;
2736}
2737
2738static struct md_sysfs_entry bitmap_location =
2739__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2740
2741/* 'bitmap/space' is the space available at 'location' for the
2742 * bitmap. This allows the kernel to know when it is safe to
2743 * resize the bitmap to match a resized array.
2744 */
2745static ssize_t
2746space_show(struct mddev *mddev, char *page)
2747{
2748 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2749}
2750
2751static ssize_t
2752space_store(struct mddev *mddev, const char *buf, size_t len)
2753{
2754 struct bitmap *bitmap;
2755 unsigned long sectors;
2756 int rv;
2757
2758 rv = kstrtoul(buf, 10, §ors);
2759 if (rv)
2760 return rv;
2761
2762 if (sectors == 0)
2763 return -EINVAL;
2764
2765 bitmap = mddev->bitmap;
2766 if (bitmap && sectors < (bitmap->storage.bytes + 511) >> 9)
2767 return -EFBIG; /* Bitmap is too big for this small space */
2768
2769 /* could make sure it isn't too big, but that isn't really
2770 * needed - user-space should be careful.
2771 */
2772 mddev->bitmap_info.space = sectors;
2773 return len;
2774}
2775
2776static struct md_sysfs_entry bitmap_space =
2777__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2778
2779static ssize_t
2780timeout_show(struct mddev *mddev, char *page)
2781{
2782 ssize_t len;
2783 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2784 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2785
2786 len = sprintf(page, "%lu", secs);
2787 if (jifs)
2788 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2789 len += sprintf(page+len, "\n");
2790 return len;
2791}
2792
2793static ssize_t
2794timeout_store(struct mddev *mddev, const char *buf, size_t len)
2795{
2796 /* timeout can be set at any time */
2797 unsigned long timeout;
2798 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2799 if (rv)
2800 return rv;
2801
2802 /* just to make sure we don't overflow... */
2803 if (timeout >= LONG_MAX / HZ)
2804 return -EINVAL;
2805
2806 timeout = timeout * HZ / 10000;
2807
2808 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2809 timeout = MAX_SCHEDULE_TIMEOUT-1;
2810 if (timeout < 1)
2811 timeout = 1;
2812
2813 mddev->bitmap_info.daemon_sleep = timeout;
2814 mddev_set_timeout(mddev, timeout, false);
2815 md_wakeup_thread(mddev->thread);
2816
2817 return len;
2818}
2819
2820static struct md_sysfs_entry bitmap_timeout =
2821__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2822
2823static ssize_t
2824backlog_show(struct mddev *mddev, char *page)
2825{
2826 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2827}
2828
2829static ssize_t
2830backlog_store(struct mddev *mddev, const char *buf, size_t len)
2831{
2832 unsigned long backlog;
2833 unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2834 struct md_rdev *rdev;
2835 bool has_write_mostly = false;
2836 int rv = kstrtoul(buf, 10, &backlog);
2837 if (rv)
2838 return rv;
2839 if (backlog > COUNTER_MAX)
2840 return -EINVAL;
2841
2842 rv = mddev_suspend_and_lock(mddev);
2843 if (rv)
2844 return rv;
2845
2846 /*
2847 * Without write mostly device, it doesn't make sense to set
2848 * backlog for max_write_behind.
2849 */
2850 rdev_for_each(rdev, mddev) {
2851 if (test_bit(WriteMostly, &rdev->flags)) {
2852 has_write_mostly = true;
2853 break;
2854 }
2855 }
2856 if (!has_write_mostly) {
2857 pr_warn_ratelimited("%s: can't set backlog, no write mostly device available\n",
2858 mdname(mddev));
2859 mddev_unlock(mddev);
2860 return -EINVAL;
2861 }
2862
2863 mddev->bitmap_info.max_write_behind = backlog;
2864 if (!backlog && mddev->serial_info_pool) {
2865 /* serial_info_pool is not needed if backlog is zero */
2866 if (!test_bit(MD_SERIALIZE_POLICY, &mddev->flags))
2867 mddev_destroy_serial_pool(mddev, NULL);
2868 } else if (backlog && !mddev->serial_info_pool) {
2869 /* serial_info_pool is needed since backlog is not zero */
2870 rdev_for_each(rdev, mddev)
2871 mddev_create_serial_pool(mddev, rdev);
2872 }
2873 if (old_mwb != backlog)
2874 bitmap_update_sb(mddev->bitmap);
2875
2876 mddev_unlock_and_resume(mddev);
2877 return len;
2878}
2879
2880static struct md_sysfs_entry bitmap_backlog =
2881__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2882
2883static ssize_t
2884chunksize_show(struct mddev *mddev, char *page)
2885{
2886 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2887}
2888
2889static ssize_t
2890chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2891{
2892 /* Can only be changed when no bitmap is active */
2893 int rv;
2894 unsigned long csize;
2895 if (mddev->bitmap)
2896 return -EBUSY;
2897 rv = kstrtoul(buf, 10, &csize);
2898 if (rv)
2899 return rv;
2900 if (csize < 512 ||
2901 !is_power_of_2(csize))
2902 return -EINVAL;
2903 if (BITS_PER_LONG > 32 && csize >= (1ULL << (BITS_PER_BYTE *
2904 sizeof(((bitmap_super_t *)0)->chunksize))))
2905 return -EOVERFLOW;
2906 mddev->bitmap_info.chunksize = csize;
2907 return len;
2908}
2909
2910static struct md_sysfs_entry bitmap_chunksize =
2911__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2912
2913static ssize_t metadata_show(struct mddev *mddev, char *page)
2914{
2915 if (mddev_is_clustered(mddev))
2916 return sprintf(page, "clustered\n");
2917 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2918 ? "external" : "internal"));
2919}
2920
2921static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2922{
2923 if (mddev->bitmap ||
2924 mddev->bitmap_info.file ||
2925 mddev->bitmap_info.offset)
2926 return -EBUSY;
2927 if (strncmp(buf, "external", 8) == 0)
2928 mddev->bitmap_info.external = 1;
2929 else if ((strncmp(buf, "internal", 8) == 0) ||
2930 (strncmp(buf, "clustered", 9) == 0))
2931 mddev->bitmap_info.external = 0;
2932 else
2933 return -EINVAL;
2934 return len;
2935}
2936
2937static struct md_sysfs_entry bitmap_metadata =
2938__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2939
2940static ssize_t can_clear_show(struct mddev *mddev, char *page)
2941{
2942 int len;
2943 struct bitmap *bitmap;
2944
2945 spin_lock(&mddev->lock);
2946 bitmap = mddev->bitmap;
2947 if (bitmap)
2948 len = sprintf(page, "%s\n", (bitmap->need_sync ? "false" :
2949 "true"));
2950 else
2951 len = sprintf(page, "\n");
2952 spin_unlock(&mddev->lock);
2953 return len;
2954}
2955
2956static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2957{
2958 struct bitmap *bitmap = mddev->bitmap;
2959
2960 if (!bitmap)
2961 return -ENOENT;
2962
2963 if (strncmp(buf, "false", 5) == 0) {
2964 bitmap->need_sync = 1;
2965 return len;
2966 }
2967
2968 if (strncmp(buf, "true", 4) == 0) {
2969 if (mddev->degraded)
2970 return -EBUSY;
2971 bitmap->need_sync = 0;
2972 return len;
2973 }
2974
2975 return -EINVAL;
2976}
2977
2978static struct md_sysfs_entry bitmap_can_clear =
2979__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2980
2981static ssize_t
2982behind_writes_used_show(struct mddev *mddev, char *page)
2983{
2984 ssize_t ret;
2985 struct bitmap *bitmap;
2986
2987 spin_lock(&mddev->lock);
2988 bitmap = mddev->bitmap;
2989 if (!bitmap)
2990 ret = sprintf(page, "0\n");
2991 else
2992 ret = sprintf(page, "%lu\n", bitmap->behind_writes_used);
2993 spin_unlock(&mddev->lock);
2994
2995 return ret;
2996}
2997
2998static ssize_t
2999behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
3000{
3001 struct bitmap *bitmap = mddev->bitmap;
3002
3003 if (bitmap)
3004 bitmap->behind_writes_used = 0;
3005 return len;
3006}
3007
3008static struct md_sysfs_entry max_backlog_used =
3009__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
3010 behind_writes_used_show, behind_writes_used_reset);
3011
3012static struct attribute *md_bitmap_common_attrs[] = {
3013 &bitmap_location.attr,
3014 NULL
3015};
3016
3017static struct attribute *md_bitmap_internal_attrs[] = {
3018 &bitmap_space.attr,
3019 &bitmap_timeout.attr,
3020 &bitmap_backlog.attr,
3021 &bitmap_chunksize.attr,
3022 &bitmap_metadata.attr,
3023 &bitmap_can_clear.attr,
3024 &max_backlog_used.attr,
3025 NULL
3026};
3027
3028static struct attribute_group md_bitmap_common_group = {
3029 .name = "bitmap",
3030 .attrs = md_bitmap_common_attrs,
3031};
3032
3033static struct attribute_group md_bitmap_internal_group = {
3034 .name = "bitmap",
3035 .attrs = md_bitmap_internal_attrs,
3036};
3037
3038static const struct attribute_group *bitmap_groups[] = {
3039 &md_bitmap_common_group,
3040 &md_bitmap_internal_group,
3041 NULL,
3042};
3043
3044static const struct attribute_group *bitmap_none_groups[] = {
3045 &md_bitmap_common_group,
3046 NULL,
3047};
3048
3049static struct bitmap_operations bitmap_none_ops = {
3050 .head = {
3051 .type = MD_BITMAP,
3052 .id = ID_BITMAP_NONE,
3053 .name = "none",
3054 },
3055
3056 .enabled = bitmap_none_enabled,
3057 .create = bitmap_none_create,
3058 .load = bitmap_none_load,
3059 .destroy = bitmap_none_destroy,
3060 .get_stats = bitmap_none_get_stats,
3061
3062 .groups = bitmap_none_groups,
3063};
3064
3065static struct bitmap_operations bitmap_ops = {
3066 .head = {
3067 .type = MD_BITMAP,
3068 .id = ID_BITMAP,
3069 .name = "bitmap",
3070 },
3071
3072 .enabled = bitmap_enabled,
3073 .create = bitmap_create,
3074 .resize = bitmap_resize,
3075 .load = bitmap_load,
3076 .destroy = bitmap_destroy,
3077 .flush = bitmap_flush,
3078 .write_all = bitmap_write_all,
3079 .dirty_bits = bitmap_dirty_bits,
3080 .unplug = bitmap_unplug,
3081 .daemon_work = bitmap_daemon_work,
3082
3083 .start_behind_write = bitmap_start_behind_write,
3084 .end_behind_write = bitmap_end_behind_write,
3085 .wait_behind_writes = bitmap_wait_behind_writes,
3086
3087 .start_write = bitmap_start_write,
3088 .end_write = bitmap_end_write,
3089 .start_discard = bitmap_start_write,
3090 .end_discard = bitmap_end_write,
3091
3092 .start_sync = bitmap_start_sync,
3093 .end_sync = bitmap_end_sync,
3094 .cond_end_sync = bitmap_cond_end_sync,
3095 .close_sync = bitmap_close_sync,
3096
3097 .update_sb = bitmap_update_sb,
3098 .get_stats = bitmap_get_stats,
3099
3100 .sync_with_cluster = bitmap_sync_with_cluster,
3101 .get_from_slot = bitmap_get_from_slot,
3102 .copy_from_slot = bitmap_copy_from_slot,
3103 .set_pages = bitmap_set_pages,
3104 .free = md_bitmap_free,
3105
3106 .groups = bitmap_groups,
3107};
3108
3109int md_bitmap_init(void)
3110{
3111 int err;
3112
3113 md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
3114 0);
3115 if (!md_bitmap_wq)
3116 return -ENOMEM;
3117
3118 err = register_md_submodule(&bitmap_none_ops.head);
3119 if (err)
3120 goto err_wq;
3121
3122 err = register_md_submodule(&bitmap_ops.head);
3123 if (err)
3124 goto err_none;
3125
3126 return 0;
3127
3128err_none:
3129 unregister_md_submodule(&bitmap_none_ops.head);
3130err_wq:
3131 destroy_workqueue(md_bitmap_wq);
3132 return err;
3133}
3134
3135void md_bitmap_exit(void)
3136{
3137 unregister_md_submodule(&bitmap_ops.head);
3138 unregister_md_submodule(&bitmap_none_ops.head);
3139 destroy_workqueue(md_bitmap_wq);
3140}