1 /* SPDX-License-Identifier: GPL-2.0 */
3 #ifndef BTRFS_EXTENT_IO_H
4 #define BTRFS_EXTENT_IO_H
6 #include <linux/rbtree.h>
7 #include <linux/refcount.h>
10 /* bits for the extent state */
11 #define EXTENT_DIRTY (1U << 0)
12 #define EXTENT_WRITEBACK (1U << 1)
13 #define EXTENT_UPTODATE (1U << 2)
14 #define EXTENT_LOCKED (1U << 3)
15 #define EXTENT_NEW (1U << 4)
16 #define EXTENT_DELALLOC (1U << 5)
17 #define EXTENT_DEFRAG (1U << 6)
18 #define EXTENT_BOUNDARY (1U << 9)
19 #define EXTENT_NODATASUM (1U << 10)
20 #define EXTENT_CLEAR_META_RESV (1U << 11)
21 #define EXTENT_FIRST_DELALLOC (1U << 12)
22 #define EXTENT_NEED_WAIT (1U << 13)
23 #define EXTENT_DAMAGED (1U << 14)
24 #define EXTENT_NORESERVE (1U << 15)
25 #define EXTENT_QGROUP_RESERVED (1U << 16)
26 #define EXTENT_CLEAR_DATA_RESV (1U << 17)
27 #define EXTENT_DELALLOC_NEW (1U << 18)
28 #define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
29 #define EXTENT_DO_ACCOUNTING (EXTENT_CLEAR_META_RESV | \
30 EXTENT_CLEAR_DATA_RESV)
31 #define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC)
34 * flags for bio submission. The high bits indicate the compression
37 #define EXTENT_BIO_COMPRESSED 1
38 #define EXTENT_BIO_FLAG_SHIFT 16
40 /* these are bit numbers for test/set bit */
41 #define EXTENT_BUFFER_UPTODATE 0
42 #define EXTENT_BUFFER_DIRTY 2
43 #define EXTENT_BUFFER_CORRUPT 3
44 #define EXTENT_BUFFER_READAHEAD 4 /* this got triggered by readahead */
45 #define EXTENT_BUFFER_TREE_REF 5
46 #define EXTENT_BUFFER_STALE 6
47 #define EXTENT_BUFFER_WRITEBACK 7
48 #define EXTENT_BUFFER_READ_ERR 8 /* read IO error */
49 #define EXTENT_BUFFER_DUMMY 9
50 #define EXTENT_BUFFER_IN_TREE 10
51 #define EXTENT_BUFFER_WRITE_ERR 11 /* write IO error */
53 /* these are flags for __process_pages_contig */
54 #define PAGE_UNLOCK (1 << 0)
55 #define PAGE_CLEAR_DIRTY (1 << 1)
56 #define PAGE_SET_WRITEBACK (1 << 2)
57 #define PAGE_END_WRITEBACK (1 << 3)
58 #define PAGE_SET_PRIVATE2 (1 << 4)
59 #define PAGE_SET_ERROR (1 << 5)
60 #define PAGE_LOCK (1 << 6)
63 * page->private values. Every page that is controlled by the extent
64 * map has page->private set to one.
66 #define EXTENT_PAGE_PRIVATE 1
69 * The extent buffer bitmap operations are done with byte granularity instead of
70 * word granularity for two reasons:
71 * 1. The bitmaps must be little-endian on disk.
72 * 2. Bitmap items are not guaranteed to be aligned to a word and therefore a
73 * single word in a bitmap may straddle two pages in the extent buffer.
75 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
76 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
77 #define BITMAP_FIRST_BYTE_MASK(start) \
78 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
79 #define BITMAP_LAST_BYTE_MASK(nbits) \
80 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
82 static inline int le_test_bit(int nr, const u8 *addr)
84 return 1U & (addr[BIT_BYTE(nr)] >> (nr & (BITS_PER_BYTE-1)));
87 void le_bitmap_set(u8 *map, unsigned int start, int len);
88 void le_bitmap_clear(u8 *map, unsigned int start, int len);
94 struct io_failure_record;
96 typedef blk_status_t (extent_submit_bio_hook_t)(void *private_data, struct bio *bio,
97 int mirror_num, unsigned long bio_flags,
100 typedef blk_status_t (extent_submit_bio_start_t)(void *private_data,
101 struct bio *bio, u64 bio_offset);
103 typedef blk_status_t (extent_submit_bio_done_t)(void *private_data,
104 struct bio *bio, int mirror_num);
106 struct extent_io_ops {
108 * The following callbacks must be allways defined, the function
109 * pointer will be called unconditionally.
111 extent_submit_bio_hook_t *submit_bio_hook;
112 int (*readpage_end_io_hook)(struct btrfs_io_bio *io_bio, u64 phy_offset,
113 struct page *page, u64 start, u64 end,
115 int (*merge_bio_hook)(struct page *page, unsigned long offset,
116 size_t size, struct bio *bio,
117 unsigned long bio_flags);
118 int (*readpage_io_failed_hook)(struct page *page, int failed_mirror);
119 struct btrfs_fs_info *(*tree_fs_info)(void *private_data);
120 void (*set_range_writeback)(void *private_data, u64 start, u64 end);
123 * Optional hooks, called if the pointer is not NULL
125 int (*fill_delalloc)(void *private_data, struct page *locked_page,
126 u64 start, u64 end, int *page_started,
127 unsigned long *nr_written,
128 struct writeback_control *wbc);
130 int (*writepage_start_hook)(struct page *page, u64 start, u64 end);
131 void (*writepage_end_io_hook)(struct page *page, u64 start, u64 end,
132 struct extent_state *state, int uptodate);
133 void (*set_bit_hook)(void *private_data, struct extent_state *state,
135 void (*clear_bit_hook)(void *private_data,
136 struct extent_state *state,
138 void (*merge_extent_hook)(void *private_data,
139 struct extent_state *new,
140 struct extent_state *other);
141 void (*split_extent_hook)(void *private_data,
142 struct extent_state *orig, u64 split);
143 void (*check_extent_io_range)(void *private_data, const char *caller,
147 struct extent_io_tree {
148 struct rb_root state;
153 const struct extent_io_ops *ops;
156 struct extent_state {
158 u64 end; /* inclusive */
159 struct rb_node rb_node;
161 /* ADD NEW ELEMENTS AFTER THIS */
162 wait_queue_head_t wq;
166 struct io_failure_record *failrec;
168 #ifdef CONFIG_BTRFS_DEBUG
169 struct list_head leak_list;
173 #define INLINE_EXTENT_BUFFER_PAGES 16
174 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE)
175 struct extent_buffer {
178 unsigned long bflags;
179 struct btrfs_fs_info *fs_info;
180 spinlock_t refs_lock;
184 struct rcu_head rcu_head;
187 /* count of read lock holders on the extent buffer */
188 atomic_t write_locks;
190 atomic_t blocking_writers;
191 atomic_t blocking_readers;
192 atomic_t spinning_readers;
193 atomic_t spinning_writers;
195 /* >= 0 if eb belongs to a log tree, -1 otherwise */
198 /* protects write locks */
201 /* readers use lock_wq while they wait for the write
202 * lock holders to unlock
204 wait_queue_head_t write_lock_wq;
206 /* writers use read_lock_wq while they wait for readers
209 wait_queue_head_t read_lock_wq;
210 struct page *pages[INLINE_EXTENT_BUFFER_PAGES];
211 #ifdef CONFIG_BTRFS_DEBUG
212 struct list_head leak_list;
217 * Structure to record how many bytes and which ranges are set/cleared
219 struct extent_changeset {
220 /* How many bytes are set/cleared in this operation */
221 unsigned int bytes_changed;
224 struct ulist range_changed;
227 static inline void extent_changeset_init(struct extent_changeset *changeset)
229 changeset->bytes_changed = 0;
230 ulist_init(&changeset->range_changed);
233 static inline struct extent_changeset *extent_changeset_alloc(void)
235 struct extent_changeset *ret;
237 ret = kmalloc(sizeof(*ret), GFP_KERNEL);
241 extent_changeset_init(ret);
245 static inline void extent_changeset_release(struct extent_changeset *changeset)
249 changeset->bytes_changed = 0;
250 ulist_release(&changeset->range_changed);
253 static inline void extent_changeset_free(struct extent_changeset *changeset)
257 extent_changeset_release(changeset);
261 static inline void extent_set_compress_type(unsigned long *bio_flags,
264 *bio_flags |= compress_type << EXTENT_BIO_FLAG_SHIFT;
267 static inline int extent_compress_type(unsigned long bio_flags)
269 return bio_flags >> EXTENT_BIO_FLAG_SHIFT;
272 struct extent_map_tree;
274 typedef struct extent_map *(get_extent_t)(struct btrfs_inode *inode,
280 void extent_io_tree_init(struct extent_io_tree *tree, void *private_data);
281 int try_release_extent_mapping(struct extent_map_tree *map,
282 struct extent_io_tree *tree, struct page *page,
284 int try_release_extent_buffer(struct page *page);
285 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
286 struct extent_state **cached);
288 static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
290 return lock_extent_bits(tree, start, end, NULL);
293 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
294 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
295 get_extent_t *get_extent, int mirror_num);
296 int __init extent_io_init(void);
297 void __cold extent_io_exit(void);
299 u64 count_range_bits(struct extent_io_tree *tree,
300 u64 *start, u64 search_end,
301 u64 max_bytes, unsigned bits, int contig);
303 void free_extent_state(struct extent_state *state);
304 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
305 unsigned bits, int filled,
306 struct extent_state *cached_state);
307 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
308 unsigned bits, struct extent_changeset *changeset);
309 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
310 unsigned bits, int wake, int delete,
311 struct extent_state **cached);
312 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
313 unsigned bits, int wake, int delete,
314 struct extent_state **cached, gfp_t mask,
315 struct extent_changeset *changeset);
317 static inline int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
319 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL);
322 static inline int unlock_extent_cached(struct extent_io_tree *tree, u64 start,
323 u64 end, struct extent_state **cached)
325 return __clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
329 static inline int unlock_extent_cached_atomic(struct extent_io_tree *tree,
330 u64 start, u64 end, struct extent_state **cached)
332 return __clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
336 static inline int clear_extent_bits(struct extent_io_tree *tree, u64 start,
337 u64 end, unsigned bits)
341 if (bits & EXTENT_LOCKED)
344 return clear_extent_bit(tree, start, end, bits, wake, 0, NULL);
347 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
348 unsigned bits, struct extent_changeset *changeset);
349 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
350 unsigned bits, u64 *failed_start,
351 struct extent_state **cached_state, gfp_t mask);
353 static inline int set_extent_bits(struct extent_io_tree *tree, u64 start,
354 u64 end, unsigned bits)
356 return set_extent_bit(tree, start, end, bits, NULL, NULL, GFP_NOFS);
359 static inline int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
360 u64 end, struct extent_state **cached_state)
362 return __clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
363 cached_state, GFP_NOFS, NULL);
366 static inline int set_extent_dirty(struct extent_io_tree *tree, u64 start,
369 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
373 static inline int clear_extent_dirty(struct extent_io_tree *tree, u64 start,
376 return clear_extent_bit(tree, start, end,
377 EXTENT_DIRTY | EXTENT_DELALLOC |
378 EXTENT_DO_ACCOUNTING, 0, 0, NULL);
381 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
382 unsigned bits, unsigned clear_bits,
383 struct extent_state **cached_state);
385 static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start,
386 u64 end, unsigned int extra_bits,
387 struct extent_state **cached_state)
389 return set_extent_bit(tree, start, end,
390 EXTENT_DELALLOC | EXTENT_UPTODATE | extra_bits,
391 NULL, cached_state, GFP_NOFS);
394 static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start,
395 u64 end, struct extent_state **cached_state)
397 return set_extent_bit(tree, start, end,
398 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
399 NULL, cached_state, GFP_NOFS);
402 static inline int set_extent_new(struct extent_io_tree *tree, u64 start,
405 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, NULL,
409 static inline int set_extent_uptodate(struct extent_io_tree *tree, u64 start,
410 u64 end, struct extent_state **cached_state, gfp_t mask)
412 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
416 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
417 u64 *start_ret, u64 *end_ret, unsigned bits,
418 struct extent_state **cached_state);
419 int extent_invalidatepage(struct extent_io_tree *tree,
420 struct page *page, unsigned long offset);
421 int extent_write_full_page(struct page *page, struct writeback_control *wbc);
422 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
424 int extent_writepages(struct extent_io_tree *tree,
425 struct address_space *mapping,
426 struct writeback_control *wbc);
427 int btree_write_cache_pages(struct address_space *mapping,
428 struct writeback_control *wbc);
429 int extent_readpages(struct extent_io_tree *tree,
430 struct address_space *mapping,
431 struct list_head *pages, unsigned nr_pages);
432 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
433 __u64 start, __u64 len);
434 void set_page_extent_mapped(struct page *page);
436 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
438 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
439 u64 start, unsigned long len);
440 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
442 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src);
443 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
445 void free_extent_buffer(struct extent_buffer *eb);
446 void free_extent_buffer_stale(struct extent_buffer *eb);
448 #define WAIT_COMPLETE 1
449 #define WAIT_PAGE_LOCK 2
450 int read_extent_buffer_pages(struct extent_io_tree *tree,
451 struct extent_buffer *eb, int wait,
453 void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
455 static inline unsigned long num_extent_pages(u64 start, u64 len)
457 return ((start + len + PAGE_SIZE - 1) >> PAGE_SHIFT) -
458 (start >> PAGE_SHIFT);
461 static inline void extent_buffer_get(struct extent_buffer *eb)
463 atomic_inc(&eb->refs);
466 static inline int extent_buffer_uptodate(struct extent_buffer *eb)
468 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
471 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
472 unsigned long start, unsigned long len);
473 void read_extent_buffer(const struct extent_buffer *eb, void *dst,
476 int read_extent_buffer_to_user(const struct extent_buffer *eb,
477 void __user *dst, unsigned long start,
479 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *src);
480 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
482 void write_extent_buffer(struct extent_buffer *eb, const void *src,
483 unsigned long start, unsigned long len);
484 void copy_extent_buffer_full(struct extent_buffer *dst,
485 struct extent_buffer *src);
486 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
487 unsigned long dst_offset, unsigned long src_offset,
489 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
490 unsigned long src_offset, unsigned long len);
491 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
492 unsigned long src_offset, unsigned long len);
493 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
495 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
497 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
498 unsigned long pos, unsigned long len);
499 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
500 unsigned long pos, unsigned long len);
501 void clear_extent_buffer_dirty(struct extent_buffer *eb);
502 int set_extent_buffer_dirty(struct extent_buffer *eb);
503 void set_extent_buffer_uptodate(struct extent_buffer *eb);
504 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
505 int extent_buffer_under_io(struct extent_buffer *eb);
506 int map_private_extent_buffer(const struct extent_buffer *eb,
507 unsigned long offset, unsigned long min_len,
508 char **map, unsigned long *map_start,
509 unsigned long *map_len);
510 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
511 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end);
512 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
513 u64 delalloc_end, struct page *locked_page,
514 unsigned bits_to_clear,
515 unsigned long page_ops);
516 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte);
517 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs);
518 struct bio *btrfs_bio_clone(struct bio *bio);
519 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size);
521 struct btrfs_fs_info;
524 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
525 u64 length, u64 logical, struct page *page,
526 unsigned int pg_offset, int mirror_num);
527 int clean_io_failure(struct btrfs_fs_info *fs_info,
528 struct extent_io_tree *failure_tree,
529 struct extent_io_tree *io_tree, u64 start,
530 struct page *page, u64 ino, unsigned int pg_offset);
531 void end_extent_writepage(struct page *page, int err, u64 start, u64 end);
532 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
533 struct extent_buffer *eb, int mirror_num);
536 * When IO fails, either with EIO or csum verification fails, we
537 * try other mirrors that might have a good copy of the data. This
538 * io_failure_record is used to record state as we go through all the
539 * mirrors. If another mirror has good data, the page is set up to date
540 * and things continue. If a good mirror can't be found, the original
541 * bio end_io callback is called to indicate things have failed.
543 struct io_failure_record {
548 unsigned long bio_flags;
555 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start,
557 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
558 struct io_failure_record **failrec_ret);
559 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
560 struct io_failure_record *failrec, int fail_mirror);
561 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
562 struct io_failure_record *failrec,
563 struct page *page, int pg_offset, int icsum,
564 bio_end_io_t *endio_func, void *data);
565 int free_io_failure(struct extent_io_tree *failure_tree,
566 struct extent_io_tree *io_tree,
567 struct io_failure_record *rec);
568 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
569 noinline u64 find_lock_delalloc_range(struct inode *inode,
570 struct extent_io_tree *tree,
571 struct page *locked_page, u64 *start,
572 u64 *end, u64 max_bytes);
574 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,