xref: /openbmc/linux/fs/btrfs/extent_io.h (revision d0b5e15f)
1 #ifndef __EXTENTIO__
2 #define __EXTENTIO__
3 
4 #include <linux/rbtree.h>
5 
6 /* bits for the extent state */
7 #define EXTENT_DIRTY		(1U << 0)
8 #define EXTENT_WRITEBACK	(1U << 1)
9 #define EXTENT_UPTODATE		(1U << 2)
10 #define EXTENT_LOCKED		(1U << 3)
11 #define EXTENT_NEW		(1U << 4)
12 #define EXTENT_DELALLOC		(1U << 5)
13 #define EXTENT_DEFRAG		(1U << 6)
14 #define EXTENT_BOUNDARY		(1U << 9)
15 #define EXTENT_NODATASUM	(1U << 10)
16 #define EXTENT_DO_ACCOUNTING	(1U << 11)
17 #define EXTENT_FIRST_DELALLOC	(1U << 12)
18 #define EXTENT_NEED_WAIT	(1U << 13)
19 #define EXTENT_DAMAGED		(1U << 14)
20 #define EXTENT_NORESERVE	(1U << 15)
21 #define EXTENT_IOBITS		(EXTENT_LOCKED | EXTENT_WRITEBACK)
22 #define EXTENT_CTLBITS		(EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC)
23 
24 /*
25  * flags for bio submission. The high bits indicate the compression
26  * type for this bio
27  */
28 #define EXTENT_BIO_COMPRESSED 1
29 #define EXTENT_BIO_TREE_LOG 2
30 #define EXTENT_BIO_PARENT_LOCKED 4
31 #define EXTENT_BIO_FLAG_SHIFT 16
32 
33 /* these are bit numbers for test/set bit */
34 #define EXTENT_BUFFER_UPTODATE 0
35 #define EXTENT_BUFFER_DIRTY 2
36 #define EXTENT_BUFFER_CORRUPT 3
37 #define EXTENT_BUFFER_READAHEAD 4	/* this got triggered by readahead */
38 #define EXTENT_BUFFER_TREE_REF 5
39 #define EXTENT_BUFFER_STALE 6
40 #define EXTENT_BUFFER_WRITEBACK 7
41 #define EXTENT_BUFFER_READ_ERR 8        /* read IO error */
42 #define EXTENT_BUFFER_DUMMY 9
43 #define EXTENT_BUFFER_IN_TREE 10
44 #define EXTENT_BUFFER_WRITE_ERR 11    /* write IO error */
45 
46 /* these are flags for extent_clear_unlock_delalloc */
47 #define PAGE_UNLOCK		(1 << 0)
48 #define PAGE_CLEAR_DIRTY	(1 << 1)
49 #define PAGE_SET_WRITEBACK	(1 << 2)
50 #define PAGE_END_WRITEBACK	(1 << 3)
51 #define PAGE_SET_PRIVATE2	(1 << 4)
52 #define PAGE_SET_ERROR		(1 << 5)
53 
54 /*
55  * page->private values.  Every page that is controlled by the extent
56  * map has page->private set to one.
57  */
58 #define EXTENT_PAGE_PRIVATE 1
59 
60 struct extent_state;
61 struct btrfs_root;
62 struct btrfs_io_bio;
63 
64 typedef	int (extent_submit_bio_hook_t)(struct inode *inode, int rw,
65 				       struct bio *bio, int mirror_num,
66 				       unsigned long bio_flags, u64 bio_offset);
67 struct extent_io_ops {
68 	int (*fill_delalloc)(struct inode *inode, struct page *locked_page,
69 			     u64 start, u64 end, int *page_started,
70 			     unsigned long *nr_written);
71 	int (*writepage_start_hook)(struct page *page, u64 start, u64 end);
72 	int (*writepage_io_hook)(struct page *page, u64 start, u64 end);
73 	extent_submit_bio_hook_t *submit_bio_hook;
74 	int (*merge_bio_hook)(int rw, struct page *page, unsigned long offset,
75 			      size_t size, struct bio *bio,
76 			      unsigned long bio_flags);
77 	int (*readpage_io_failed_hook)(struct page *page, int failed_mirror);
78 	int (*readpage_end_io_hook)(struct btrfs_io_bio *io_bio, u64 phy_offset,
79 				    struct page *page, u64 start, u64 end,
80 				    int mirror);
81 	int (*writepage_end_io_hook)(struct page *page, u64 start, u64 end,
82 				      struct extent_state *state, int uptodate);
83 	void (*set_bit_hook)(struct inode *inode, struct extent_state *state,
84 			     unsigned *bits);
85 	void (*clear_bit_hook)(struct inode *inode, struct extent_state *state,
86 			       unsigned *bits);
87 	void (*merge_extent_hook)(struct inode *inode,
88 				  struct extent_state *new,
89 				  struct extent_state *other);
90 	void (*split_extent_hook)(struct inode *inode,
91 				  struct extent_state *orig, u64 split);
92 };
93 
94 struct extent_io_tree {
95 	struct rb_root state;
96 	struct address_space *mapping;
97 	u64 dirty_bytes;
98 	int track_uptodate;
99 	spinlock_t lock;
100 	struct extent_io_ops *ops;
101 };
102 
103 struct extent_state {
104 	u64 start;
105 	u64 end; /* inclusive */
106 	struct rb_node rb_node;
107 
108 	/* ADD NEW ELEMENTS AFTER THIS */
109 	wait_queue_head_t wq;
110 	atomic_t refs;
111 	unsigned state;
112 
113 	/* for use by the FS */
114 	u64 private;
115 
116 #ifdef CONFIG_BTRFS_DEBUG
117 	struct list_head leak_list;
118 #endif
119 };
120 
121 #define INLINE_EXTENT_BUFFER_PAGES 16
122 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_CACHE_SIZE)
123 struct extent_buffer {
124 	u64 start;
125 	unsigned long len;
126 	unsigned long bflags;
127 	struct btrfs_fs_info *fs_info;
128 	spinlock_t refs_lock;
129 	atomic_t refs;
130 	atomic_t io_pages;
131 	int read_mirror;
132 	struct rcu_head rcu_head;
133 	pid_t lock_owner;
134 
135 	/* count of read lock holders on the extent buffer */
136 	atomic_t write_locks;
137 	atomic_t read_locks;
138 	atomic_t blocking_writers;
139 	atomic_t blocking_readers;
140 	atomic_t spinning_readers;
141 	atomic_t spinning_writers;
142 	short lock_nested;
143 	/* >= 0 if eb belongs to a log tree, -1 otherwise */
144 	short log_index;
145 
146 	/* protects write locks */
147 	rwlock_t lock;
148 
149 	/* readers use lock_wq while they wait for the write
150 	 * lock holders to unlock
151 	 */
152 	wait_queue_head_t write_lock_wq;
153 
154 	/* writers use read_lock_wq while they wait for readers
155 	 * to unlock
156 	 */
157 	wait_queue_head_t read_lock_wq;
158 	struct page *pages[INLINE_EXTENT_BUFFER_PAGES];
159 #ifdef CONFIG_BTRFS_DEBUG
160 	struct list_head leak_list;
161 #endif
162 };
163 
164 static inline void extent_set_compress_type(unsigned long *bio_flags,
165 					    int compress_type)
166 {
167 	*bio_flags |= compress_type << EXTENT_BIO_FLAG_SHIFT;
168 }
169 
170 static inline int extent_compress_type(unsigned long bio_flags)
171 {
172 	return bio_flags >> EXTENT_BIO_FLAG_SHIFT;
173 }
174 
175 struct extent_map_tree;
176 
177 typedef struct extent_map *(get_extent_t)(struct inode *inode,
178 					  struct page *page,
179 					  size_t pg_offset,
180 					  u64 start, u64 len,
181 					  int create);
182 
183 void extent_io_tree_init(struct extent_io_tree *tree,
184 			 struct address_space *mapping);
185 int try_release_extent_mapping(struct extent_map_tree *map,
186 			       struct extent_io_tree *tree, struct page *page,
187 			       gfp_t mask);
188 int try_release_extent_buffer(struct page *page);
189 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
190 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
191 		     unsigned bits, struct extent_state **cached);
192 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end);
193 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
194 			 struct extent_state **cached, gfp_t mask);
195 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
196 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
197 			  get_extent_t *get_extent, int mirror_num);
198 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
199 				 get_extent_t *get_extent, int mirror_num);
200 int __init extent_io_init(void);
201 void extent_io_exit(void);
202 
203 u64 count_range_bits(struct extent_io_tree *tree,
204 		     u64 *start, u64 search_end,
205 		     u64 max_bytes, unsigned bits, int contig);
206 
207 void free_extent_state(struct extent_state *state);
208 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
209 		   unsigned bits, int filled,
210 		   struct extent_state *cached_state);
211 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
212 		      unsigned bits, gfp_t mask);
213 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
214 		     unsigned bits, int wake, int delete,
215 		     struct extent_state **cached, gfp_t mask);
216 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
217 		    unsigned bits, gfp_t mask);
218 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
219 		   unsigned bits, u64 *failed_start,
220 		   struct extent_state **cached_state, gfp_t mask);
221 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
222 			struct extent_state **cached_state, gfp_t mask);
223 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
224 			  struct extent_state **cached_state, gfp_t mask);
225 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
226 		   gfp_t mask);
227 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
228 		     gfp_t mask);
229 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
230 		       gfp_t mask);
231 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
232 		       unsigned bits, unsigned clear_bits,
233 		       struct extent_state **cached_state, gfp_t mask);
234 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
235 			struct extent_state **cached_state, gfp_t mask);
236 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
237 		      struct extent_state **cached_state, gfp_t mask);
238 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
239 			  u64 *start_ret, u64 *end_ret, unsigned bits,
240 			  struct extent_state **cached_state);
241 int extent_invalidatepage(struct extent_io_tree *tree,
242 			  struct page *page, unsigned long offset);
243 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
244 			  get_extent_t *get_extent,
245 			  struct writeback_control *wbc);
246 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
247 			      u64 start, u64 end, get_extent_t *get_extent,
248 			      int mode);
249 int extent_writepages(struct extent_io_tree *tree,
250 		      struct address_space *mapping,
251 		      get_extent_t *get_extent,
252 		      struct writeback_control *wbc);
253 int btree_write_cache_pages(struct address_space *mapping,
254 			    struct writeback_control *wbc);
255 int extent_readpages(struct extent_io_tree *tree,
256 		     struct address_space *mapping,
257 		     struct list_head *pages, unsigned nr_pages,
258 		     get_extent_t get_extent);
259 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
260 		__u64 start, __u64 len, get_extent_t *get_extent);
261 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private);
262 void set_page_extent_mapped(struct page *page);
263 
264 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
265 					  u64 start);
266 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
267 		u64 start);
268 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src);
269 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
270 					 u64 start);
271 void free_extent_buffer(struct extent_buffer *eb);
272 void free_extent_buffer_stale(struct extent_buffer *eb);
273 #define WAIT_NONE	0
274 #define WAIT_COMPLETE	1
275 #define WAIT_PAGE_LOCK	2
276 int read_extent_buffer_pages(struct extent_io_tree *tree,
277 			     struct extent_buffer *eb, u64 start, int wait,
278 			     get_extent_t *get_extent, int mirror_num);
279 void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
280 
281 static inline unsigned long num_extent_pages(u64 start, u64 len)
282 {
283 	return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
284 		(start >> PAGE_CACHE_SHIFT);
285 }
286 
287 static inline void extent_buffer_get(struct extent_buffer *eb)
288 {
289 	atomic_inc(&eb->refs);
290 }
291 
292 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
293 			  unsigned long start,
294 			  unsigned long len);
295 void read_extent_buffer(struct extent_buffer *eb, void *dst,
296 			unsigned long start,
297 			unsigned long len);
298 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dst,
299 			       unsigned long start,
300 			       unsigned long len);
301 void write_extent_buffer(struct extent_buffer *eb, const void *src,
302 			 unsigned long start, unsigned long len);
303 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
304 			unsigned long dst_offset, unsigned long src_offset,
305 			unsigned long len);
306 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
307 			   unsigned long src_offset, unsigned long len);
308 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
309 			   unsigned long src_offset, unsigned long len);
310 void memset_extent_buffer(struct extent_buffer *eb, char c,
311 			  unsigned long start, unsigned long len);
312 void clear_extent_buffer_dirty(struct extent_buffer *eb);
313 int set_extent_buffer_dirty(struct extent_buffer *eb);
314 int set_extent_buffer_uptodate(struct extent_buffer *eb);
315 int clear_extent_buffer_uptodate(struct extent_buffer *eb);
316 int extent_buffer_uptodate(struct extent_buffer *eb);
317 int extent_buffer_under_io(struct extent_buffer *eb);
318 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long offset,
319 		      unsigned long min_len, char **map,
320 		      unsigned long *map_start,
321 		      unsigned long *map_len);
322 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
323 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end);
324 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
325 				 struct page *locked_page,
326 				 unsigned bits_to_clear,
327 				 unsigned long page_ops);
328 struct bio *
329 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
330 		gfp_t gfp_flags);
331 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs);
332 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask);
333 
334 struct btrfs_fs_info;
335 
336 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
337 		      struct page *page, unsigned int pg_offset,
338 		      int mirror_num);
339 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
340 		     unsigned int pg_offset);
341 int end_extent_writepage(struct page *page, int err, u64 start, u64 end);
342 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
343 			 int mirror_num);
344 
345 /*
346  * When IO fails, either with EIO or csum verification fails, we
347  * try other mirrors that might have a good copy of the data.  This
348  * io_failure_record is used to record state as we go through all the
349  * mirrors.  If another mirror has good data, the page is set up to date
350  * and things continue.  If a good mirror can't be found, the original
351  * bio end_io callback is called to indicate things have failed.
352  */
353 struct io_failure_record {
354 	struct page *page;
355 	u64 start;
356 	u64 len;
357 	u64 logical;
358 	unsigned long bio_flags;
359 	int this_mirror;
360 	int failed_mirror;
361 	int in_validation;
362 };
363 
364 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end);
365 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
366 				struct io_failure_record **failrec_ret);
367 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
368 			   struct io_failure_record *failrec, int fail_mirror);
369 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
370 				    struct io_failure_record *failrec,
371 				    struct page *page, int pg_offset, int icsum,
372 				    bio_end_io_t *endio_func, void *data);
373 int free_io_failure(struct inode *inode, struct io_failure_record *rec);
374 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
375 noinline u64 find_lock_delalloc_range(struct inode *inode,
376 				      struct extent_io_tree *tree,
377 				      struct page *locked_page, u64 *start,
378 				      u64 *end, u64 max_bytes);
379 #endif
380 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
381 					       u64 start);
382 #endif
383