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_UNMAPPED 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)))
86 struct io_failure_record;
88 typedef blk_status_t (extent_submit_bio_hook_t)(void *private_data, struct bio *bio,
89 int mirror_num, unsigned long bio_flags,
92 typedef blk_status_t (extent_submit_bio_start_t)(void *private_data,
93 struct bio *bio, u64 bio_offset);
95 struct extent_io_ops {
97 * The following callbacks must be allways defined, the function
98 * pointer will be called unconditionally.
100 extent_submit_bio_hook_t *submit_bio_hook;
101 int (*readpage_end_io_hook)(struct btrfs_io_bio *io_bio, u64 phy_offset,
102 struct page *page, u64 start, u64 end,
104 int (*readpage_io_failed_hook)(struct page *page, int failed_mirror);
107 * Optional hooks, called if the pointer is not NULL
109 int (*writepage_start_hook)(struct page *page, u64 start, u64 end);
110 void (*writepage_end_io_hook)(struct page *page, u64 start, u64 end,
111 struct extent_state *state, int uptodate);
112 void (*set_bit_hook)(void *private_data, struct extent_state *state,
114 void (*clear_bit_hook)(void *private_data,
115 struct extent_state *state,
117 void (*merge_extent_hook)(void *private_data,
118 struct extent_state *new,
119 struct extent_state *other);
120 void (*split_extent_hook)(void *private_data,
121 struct extent_state *orig, u64 split);
122 void (*check_extent_io_range)(void *private_data, const char *caller,
126 struct extent_io_tree {
127 struct rb_root state;
132 const struct extent_io_ops *ops;
135 struct extent_state {
137 u64 end; /* inclusive */
138 struct rb_node rb_node;
140 /* ADD NEW ELEMENTS AFTER THIS */
141 wait_queue_head_t wq;
145 struct io_failure_record *failrec;
147 #ifdef CONFIG_BTRFS_DEBUG
148 struct list_head leak_list;
152 #define INLINE_EXTENT_BUFFER_PAGES 16
153 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE)
154 struct extent_buffer {
157 unsigned long bflags;
158 struct btrfs_fs_info *fs_info;
159 spinlock_t refs_lock;
163 struct rcu_head rcu_head;
166 /* count of read lock holders on the extent buffer */
167 atomic_t write_locks;
169 atomic_t blocking_writers;
170 atomic_t blocking_readers;
171 atomic_t spinning_readers;
172 atomic_t spinning_writers;
174 /* >= 0 if eb belongs to a log tree, -1 otherwise */
177 /* protects write locks */
180 /* readers use lock_wq while they wait for the write
181 * lock holders to unlock
183 wait_queue_head_t write_lock_wq;
185 /* writers use read_lock_wq while they wait for readers
188 wait_queue_head_t read_lock_wq;
189 struct page *pages[INLINE_EXTENT_BUFFER_PAGES];
190 #ifdef CONFIG_BTRFS_DEBUG
191 struct list_head leak_list;
196 * Structure to record how many bytes and which ranges are set/cleared
198 struct extent_changeset {
199 /* How many bytes are set/cleared in this operation */
200 unsigned int bytes_changed;
203 struct ulist range_changed;
206 static inline void extent_changeset_init(struct extent_changeset *changeset)
208 changeset->bytes_changed = 0;
209 ulist_init(&changeset->range_changed);
212 static inline struct extent_changeset *extent_changeset_alloc(void)
214 struct extent_changeset *ret;
216 ret = kmalloc(sizeof(*ret), GFP_KERNEL);
220 extent_changeset_init(ret);
224 static inline void extent_changeset_release(struct extent_changeset *changeset)
228 changeset->bytes_changed = 0;
229 ulist_release(&changeset->range_changed);
232 static inline void extent_changeset_free(struct extent_changeset *changeset)
236 extent_changeset_release(changeset);
240 static inline void extent_set_compress_type(unsigned long *bio_flags,
243 *bio_flags |= compress_type << EXTENT_BIO_FLAG_SHIFT;
246 static inline int extent_compress_type(unsigned long bio_flags)
248 return bio_flags >> EXTENT_BIO_FLAG_SHIFT;
251 struct extent_map_tree;
253 typedef struct extent_map *(get_extent_t)(struct btrfs_inode *inode,
259 void extent_io_tree_init(struct extent_io_tree *tree, void *private_data);
260 int try_release_extent_mapping(struct page *page, gfp_t mask);
261 int try_release_extent_buffer(struct page *page);
262 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
263 struct extent_state **cached);
265 static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
267 return lock_extent_bits(tree, start, end, NULL);
270 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
271 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
272 get_extent_t *get_extent, int mirror_num);
273 int __init extent_io_init(void);
274 void __cold extent_io_exit(void);
276 u64 count_range_bits(struct extent_io_tree *tree,
277 u64 *start, u64 search_end,
278 u64 max_bytes, unsigned bits, int contig);
280 void free_extent_state(struct extent_state *state);
281 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
282 unsigned bits, int filled,
283 struct extent_state *cached_state);
284 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
285 unsigned bits, struct extent_changeset *changeset);
286 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
287 unsigned bits, int wake, int delete,
288 struct extent_state **cached);
289 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
290 unsigned bits, int wake, int delete,
291 struct extent_state **cached, gfp_t mask,
292 struct extent_changeset *changeset);
294 static inline int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
296 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL);
299 static inline int unlock_extent_cached(struct extent_io_tree *tree, u64 start,
300 u64 end, struct extent_state **cached)
302 return __clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
306 static inline int unlock_extent_cached_atomic(struct extent_io_tree *tree,
307 u64 start, u64 end, struct extent_state **cached)
309 return __clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
313 static inline int clear_extent_bits(struct extent_io_tree *tree, u64 start,
314 u64 end, unsigned bits)
318 if (bits & EXTENT_LOCKED)
321 return clear_extent_bit(tree, start, end, bits, wake, 0, NULL);
324 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
325 unsigned bits, struct extent_changeset *changeset);
326 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
327 unsigned bits, u64 *failed_start,
328 struct extent_state **cached_state, gfp_t mask);
330 static inline int set_extent_bits(struct extent_io_tree *tree, u64 start,
331 u64 end, unsigned bits)
333 return set_extent_bit(tree, start, end, bits, NULL, NULL, GFP_NOFS);
336 static inline int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
337 u64 end, struct extent_state **cached_state)
339 return __clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
340 cached_state, GFP_NOFS, NULL);
343 static inline int set_extent_dirty(struct extent_io_tree *tree, u64 start,
346 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
350 static inline int clear_extent_dirty(struct extent_io_tree *tree, u64 start,
351 u64 end, struct extent_state **cached)
353 return clear_extent_bit(tree, start, end,
354 EXTENT_DIRTY | EXTENT_DELALLOC |
355 EXTENT_DO_ACCOUNTING, 0, 0, cached);
358 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
359 unsigned bits, unsigned clear_bits,
360 struct extent_state **cached_state);
362 static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start,
363 u64 end, unsigned int extra_bits,
364 struct extent_state **cached_state)
366 return set_extent_bit(tree, start, end,
367 EXTENT_DELALLOC | EXTENT_UPTODATE | extra_bits,
368 NULL, cached_state, GFP_NOFS);
371 static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start,
372 u64 end, struct extent_state **cached_state)
374 return set_extent_bit(tree, start, end,
375 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
376 NULL, cached_state, GFP_NOFS);
379 static inline int set_extent_new(struct extent_io_tree *tree, u64 start,
382 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, NULL,
386 static inline int set_extent_uptodate(struct extent_io_tree *tree, u64 start,
387 u64 end, struct extent_state **cached_state, gfp_t mask)
389 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
393 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
394 u64 *start_ret, u64 *end_ret, unsigned bits,
395 struct extent_state **cached_state);
396 int extent_invalidatepage(struct extent_io_tree *tree,
397 struct page *page, unsigned long offset);
398 int extent_write_full_page(struct page *page, struct writeback_control *wbc);
399 int extent_write_locked_range(struct inode *inode, u64 start, u64 end,
401 int extent_writepages(struct address_space *mapping,
402 struct writeback_control *wbc);
403 int btree_write_cache_pages(struct address_space *mapping,
404 struct writeback_control *wbc);
405 int extent_readpages(struct address_space *mapping, struct list_head *pages,
407 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
408 __u64 start, __u64 len);
409 void set_page_extent_mapped(struct page *page);
411 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
413 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
414 u64 start, unsigned long len);
415 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
417 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src);
418 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
420 void free_extent_buffer(struct extent_buffer *eb);
421 void free_extent_buffer_stale(struct extent_buffer *eb);
423 #define WAIT_COMPLETE 1
424 #define WAIT_PAGE_LOCK 2
425 int read_extent_buffer_pages(struct extent_io_tree *tree,
426 struct extent_buffer *eb, int wait,
428 void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
430 static inline int num_extent_pages(const struct extent_buffer *eb)
432 return (round_up(eb->start + eb->len, PAGE_SIZE) >> PAGE_SHIFT) -
433 (eb->start >> PAGE_SHIFT);
436 static inline void extent_buffer_get(struct extent_buffer *eb)
438 atomic_inc(&eb->refs);
441 static inline int extent_buffer_uptodate(struct extent_buffer *eb)
443 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
446 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
447 unsigned long start, unsigned long len);
448 void read_extent_buffer(const struct extent_buffer *eb, void *dst,
451 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
452 void __user *dst, unsigned long start,
454 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *src);
455 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
457 void write_extent_buffer(struct extent_buffer *eb, const void *src,
458 unsigned long start, unsigned long len);
459 void copy_extent_buffer_full(struct extent_buffer *dst,
460 struct extent_buffer *src);
461 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
462 unsigned long dst_offset, unsigned long src_offset,
464 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
465 unsigned long src_offset, unsigned long len);
466 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
467 unsigned long src_offset, unsigned long len);
468 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
470 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
472 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
473 unsigned long pos, unsigned long len);
474 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
475 unsigned long pos, unsigned long len);
476 void clear_extent_buffer_dirty(struct extent_buffer *eb);
477 int set_extent_buffer_dirty(struct extent_buffer *eb);
478 void set_extent_buffer_uptodate(struct extent_buffer *eb);
479 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
480 int extent_buffer_under_io(struct extent_buffer *eb);
481 int map_private_extent_buffer(const struct extent_buffer *eb,
482 unsigned long offset, unsigned long min_len,
483 char **map, unsigned long *map_start,
484 unsigned long *map_len);
485 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
486 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end);
487 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
488 u64 delalloc_end, struct page *locked_page,
489 unsigned bits_to_clear,
490 unsigned long page_ops);
491 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte);
492 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs);
493 struct bio *btrfs_bio_clone(struct bio *bio);
494 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size);
496 struct btrfs_fs_info;
499 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
500 u64 length, u64 logical, struct page *page,
501 unsigned int pg_offset, int mirror_num);
502 int clean_io_failure(struct btrfs_fs_info *fs_info,
503 struct extent_io_tree *failure_tree,
504 struct extent_io_tree *io_tree, u64 start,
505 struct page *page, u64 ino, unsigned int pg_offset);
506 void end_extent_writepage(struct page *page, int err, u64 start, u64 end);
507 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
508 struct extent_buffer *eb, int mirror_num);
511 * When IO fails, either with EIO or csum verification fails, we
512 * try other mirrors that might have a good copy of the data. This
513 * io_failure_record is used to record state as we go through all the
514 * mirrors. If another mirror has good data, the page is set up to date
515 * and things continue. If a good mirror can't be found, the original
516 * bio end_io callback is called to indicate things have failed.
518 struct io_failure_record {
523 unsigned long bio_flags;
530 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start,
532 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
533 struct io_failure_record **failrec_ret);
534 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages,
535 struct io_failure_record *failrec, int fail_mirror);
536 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
537 struct io_failure_record *failrec,
538 struct page *page, int pg_offset, int icsum,
539 bio_end_io_t *endio_func, void *data);
540 int free_io_failure(struct extent_io_tree *failure_tree,
541 struct extent_io_tree *io_tree,
542 struct io_failure_record *rec);
543 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
544 noinline u64 find_lock_delalloc_range(struct inode *inode,
545 struct extent_io_tree *tree,
546 struct page *locked_page, u64 *start,
547 u64 *end, u64 max_bytes);
549 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,