xref: /linux-6.15/include/linux/buffer_head.h (revision 2008d89f)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * include/linux/buffer_head.h
4  *
5  * Everything to do with buffer_heads.
6  */
7 
8 #ifndef _LINUX_BUFFER_HEAD_H
9 #define _LINUX_BUFFER_HEAD_H
10 
11 #include <linux/types.h>
12 #include <linux/blk_types.h>
13 #include <linux/fs.h>
14 #include <linux/linkage.h>
15 #include <linux/pagemap.h>
16 #include <linux/wait.h>
17 #include <linux/atomic.h>
18 
19 enum bh_state_bits {
20 	BH_Uptodate,	/* Contains valid data */
21 	BH_Dirty,	/* Is dirty */
22 	BH_Lock,	/* Is locked */
23 	BH_Req,		/* Has been submitted for I/O */
24 
25 	BH_Mapped,	/* Has a disk mapping */
26 	BH_New,		/* Disk mapping was newly created by get_block */
27 	BH_Async_Read,	/* Is under end_buffer_async_read I/O */
28 	BH_Async_Write,	/* Is under end_buffer_async_write I/O */
29 	BH_Delay,	/* Buffer is not yet allocated on disk */
30 	BH_Boundary,	/* Block is followed by a discontiguity */
31 	BH_Write_EIO,	/* I/O error on write */
32 	BH_Unwritten,	/* Buffer is allocated on disk but not written */
33 	BH_Quiet,	/* Buffer Error Prinks to be quiet */
34 	BH_Meta,	/* Buffer contains metadata */
35 	BH_Prio,	/* Buffer should be submitted with REQ_PRIO */
36 	BH_Defer_Completion, /* Defer AIO completion to workqueue */
37 
38 	BH_PrivateStart,/* not a state bit, but the first bit available
39 			 * for private allocation by other entities
40 			 */
41 };
42 
43 #define MAX_BUF_PER_PAGE (PAGE_SIZE / 512)
44 
45 struct page;
46 struct buffer_head;
47 struct address_space;
48 typedef void (bh_end_io_t)(struct buffer_head *bh, int uptodate);
49 
50 /*
51  * Historically, a buffer_head was used to map a single block
52  * within a page, and of course as the unit of I/O through the
53  * filesystem and block layers.  Nowadays the basic I/O unit
54  * is the bio, and buffer_heads are used for extracting block
55  * mappings (via a get_block_t call), for tracking state within
56  * a page (via a page_mapping) and for wrapping bio submission
57  * for backward compatibility reasons (e.g. submit_bh).
58  */
59 struct buffer_head {
60 	unsigned long b_state;		/* buffer state bitmap (see above) */
61 	struct buffer_head *b_this_page;/* circular list of page's buffers */
62 	union {
63 		struct page *b_page;	/* the page this bh is mapped to */
64 		struct folio *b_folio;	/* the folio this bh is mapped to */
65 	};
66 
67 	sector_t b_blocknr;		/* start block number */
68 	size_t b_size;			/* size of mapping */
69 	char *b_data;			/* pointer to data within the page */
70 
71 	struct block_device *b_bdev;
72 	bh_end_io_t *b_end_io;		/* I/O completion */
73  	void *b_private;		/* reserved for b_end_io */
74 	struct list_head b_assoc_buffers; /* associated with another mapping */
75 	struct address_space *b_assoc_map;	/* mapping this buffer is
76 						   associated with */
77 	atomic_t b_count;		/* users using this buffer_head */
78 	spinlock_t b_uptodate_lock;	/* Used by the first bh in a page, to
79 					 * serialise IO completion of other
80 					 * buffers in the page */
81 };
82 
83 /*
84  * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()
85  * and buffer_foo() functions.
86  * To avoid reset buffer flags that are already set, because that causes
87  * a costly cache line transition, check the flag first.
88  */
89 #define BUFFER_FNS(bit, name)						\
90 static __always_inline void set_buffer_##name(struct buffer_head *bh)	\
91 {									\
92 	if (!test_bit(BH_##bit, &(bh)->b_state))			\
93 		set_bit(BH_##bit, &(bh)->b_state);			\
94 }									\
95 static __always_inline void clear_buffer_##name(struct buffer_head *bh)	\
96 {									\
97 	clear_bit(BH_##bit, &(bh)->b_state);				\
98 }									\
99 static __always_inline int buffer_##name(const struct buffer_head *bh)	\
100 {									\
101 	return test_bit(BH_##bit, &(bh)->b_state);			\
102 }
103 
104 /*
105  * test_set_buffer_foo() and test_clear_buffer_foo()
106  */
107 #define TAS_BUFFER_FNS(bit, name)					\
108 static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \
109 {									\
110 	return test_and_set_bit(BH_##bit, &(bh)->b_state);		\
111 }									\
112 static __always_inline int test_clear_buffer_##name(struct buffer_head *bh) \
113 {									\
114 	return test_and_clear_bit(BH_##bit, &(bh)->b_state);		\
115 }									\
116 
117 /*
118  * Emit the buffer bitops functions.   Note that there are also functions
119  * of the form "mark_buffer_foo()".  These are higher-level functions which
120  * do something in addition to setting a b_state bit.
121  */
122 BUFFER_FNS(Dirty, dirty)
123 TAS_BUFFER_FNS(Dirty, dirty)
124 BUFFER_FNS(Lock, locked)
125 BUFFER_FNS(Req, req)
126 TAS_BUFFER_FNS(Req, req)
127 BUFFER_FNS(Mapped, mapped)
128 BUFFER_FNS(New, new)
129 BUFFER_FNS(Async_Read, async_read)
130 BUFFER_FNS(Async_Write, async_write)
131 BUFFER_FNS(Delay, delay)
132 BUFFER_FNS(Boundary, boundary)
133 BUFFER_FNS(Write_EIO, write_io_error)
134 BUFFER_FNS(Unwritten, unwritten)
135 BUFFER_FNS(Meta, meta)
136 BUFFER_FNS(Prio, prio)
137 BUFFER_FNS(Defer_Completion, defer_completion)
138 
139 static __always_inline void set_buffer_uptodate(struct buffer_head *bh)
140 {
141 	/*
142 	 * If somebody else already set this uptodate, they will
143 	 * have done the memory barrier, and a reader will thus
144 	 * see *some* valid buffer state.
145 	 *
146 	 * Any other serialization (with IO errors or whatever that
147 	 * might clear the bit) has to come from other state (eg BH_Lock).
148 	 */
149 	if (test_bit(BH_Uptodate, &bh->b_state))
150 		return;
151 
152 	/*
153 	 * make it consistent with folio_mark_uptodate
154 	 * pairs with smp_load_acquire in buffer_uptodate
155 	 */
156 	smp_mb__before_atomic();
157 	set_bit(BH_Uptodate, &bh->b_state);
158 }
159 
160 static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)
161 {
162 	clear_bit(BH_Uptodate, &bh->b_state);
163 }
164 
165 static __always_inline int buffer_uptodate(const struct buffer_head *bh)
166 {
167 	/*
168 	 * make it consistent with folio_test_uptodate
169 	 * pairs with smp_mb__before_atomic in set_buffer_uptodate
170 	 */
171 	return test_bit_acquire(BH_Uptodate, &bh->b_state);
172 }
173 
174 #define bh_offset(bh)		((unsigned long)(bh)->b_data & ~PAGE_MASK)
175 
176 /* If we *know* page->private refers to buffer_heads */
177 #define page_buffers(page)					\
178 	({							\
179 		BUG_ON(!PagePrivate(page));			\
180 		((struct buffer_head *)page_private(page));	\
181 	})
182 #define page_has_buffers(page)	PagePrivate(page)
183 #define folio_buffers(folio)		folio_get_private(folio)
184 
185 void buffer_check_dirty_writeback(struct folio *folio,
186 				     bool *dirty, bool *writeback);
187 
188 /*
189  * Declarations
190  */
191 
192 void mark_buffer_dirty(struct buffer_head *bh);
193 void mark_buffer_write_io_error(struct buffer_head *bh);
194 void touch_buffer(struct buffer_head *bh);
195 void set_bh_page(struct buffer_head *bh,
196 		struct page *page, unsigned long offset);
197 void folio_set_bh(struct buffer_head *bh, struct folio *folio,
198 		  unsigned long offset);
199 struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size,
200 					bool retry);
201 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
202 		bool retry);
203 void create_empty_buffers(struct page *, unsigned long,
204 			unsigned long b_state);
205 void folio_create_empty_buffers(struct folio *folio, unsigned long blocksize,
206 				unsigned long b_state);
207 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
208 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
209 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
210 
211 /* Things to do with buffers at mapping->private_list */
212 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
213 int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end,
214 				  bool datasync);
215 int generic_buffers_fsync(struct file *file, loff_t start, loff_t end,
216 			  bool datasync);
217 void clean_bdev_aliases(struct block_device *bdev, sector_t block,
218 			sector_t len);
219 static inline void clean_bdev_bh_alias(struct buffer_head *bh)
220 {
221 	clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
222 }
223 
224 void mark_buffer_async_write(struct buffer_head *bh);
225 void __wait_on_buffer(struct buffer_head *);
226 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
227 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
228 			unsigned size);
229 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
230 				  unsigned size, gfp_t gfp);
231 void __brelse(struct buffer_head *);
232 void __bforget(struct buffer_head *);
233 void __breadahead(struct block_device *, sector_t block, unsigned int size);
234 struct buffer_head *__bread_gfp(struct block_device *,
235 				sector_t block, unsigned size, gfp_t gfp);
236 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
237 void free_buffer_head(struct buffer_head * bh);
238 void unlock_buffer(struct buffer_head *bh);
239 void __lock_buffer(struct buffer_head *bh);
240 int sync_dirty_buffer(struct buffer_head *bh);
241 int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
242 void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
243 void submit_bh(blk_opf_t, struct buffer_head *);
244 void write_boundary_block(struct block_device *bdev,
245 			sector_t bblock, unsigned blocksize);
246 int bh_uptodate_or_lock(struct buffer_head *bh);
247 int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait);
248 void __bh_read_batch(int nr, struct buffer_head *bhs[],
249 		     blk_opf_t op_flags, bool force_lock);
250 
251 /*
252  * Generic address_space_operations implementations for buffer_head-backed
253  * address_spaces.
254  */
255 void block_invalidate_folio(struct folio *folio, size_t offset, size_t length);
256 int block_write_full_page(struct page *page, get_block_t *get_block,
257 				struct writeback_control *wbc);
258 int __block_write_full_folio(struct inode *inode, struct folio *folio,
259 			get_block_t *get_block, struct writeback_control *wbc,
260 			bh_end_io_t *handler);
261 int block_read_full_folio(struct folio *, get_block_t *);
262 bool block_is_partially_uptodate(struct folio *, size_t from, size_t count);
263 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
264 		struct page **pagep, get_block_t *get_block);
265 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
266 		get_block_t *get_block);
267 int block_write_end(struct file *, struct address_space *,
268 				loff_t, unsigned, unsigned,
269 				struct page *, void *);
270 int generic_write_end(struct file *, struct address_space *,
271 				loff_t, unsigned, unsigned,
272 				struct page *, void *);
273 void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to);
274 void clean_page_buffers(struct page *page);
275 int cont_write_begin(struct file *, struct address_space *, loff_t,
276 			unsigned, struct page **, void **,
277 			get_block_t *, loff_t *);
278 int generic_cont_expand_simple(struct inode *inode, loff_t size);
279 int block_commit_write(struct page *page, unsigned from, unsigned to);
280 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
281 				get_block_t get_block);
282 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
283 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
284 
285 #ifdef CONFIG_MIGRATION
286 extern int buffer_migrate_folio(struct address_space *,
287 		struct folio *dst, struct folio *src, enum migrate_mode);
288 extern int buffer_migrate_folio_norefs(struct address_space *,
289 		struct folio *dst, struct folio *src, enum migrate_mode);
290 #else
291 #define buffer_migrate_folio NULL
292 #define buffer_migrate_folio_norefs NULL
293 #endif
294 
295 /*
296  * inline definitions
297  */
298 
299 static inline void get_bh(struct buffer_head *bh)
300 {
301         atomic_inc(&bh->b_count);
302 }
303 
304 static inline void put_bh(struct buffer_head *bh)
305 {
306         smp_mb__before_atomic();
307         atomic_dec(&bh->b_count);
308 }
309 
310 static inline void brelse(struct buffer_head *bh)
311 {
312 	if (bh)
313 		__brelse(bh);
314 }
315 
316 static inline void bforget(struct buffer_head *bh)
317 {
318 	if (bh)
319 		__bforget(bh);
320 }
321 
322 static inline struct buffer_head *
323 sb_bread(struct super_block *sb, sector_t block)
324 {
325 	return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
326 }
327 
328 static inline struct buffer_head *
329 sb_bread_unmovable(struct super_block *sb, sector_t block)
330 {
331 	return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
332 }
333 
334 static inline void
335 sb_breadahead(struct super_block *sb, sector_t block)
336 {
337 	__breadahead(sb->s_bdev, block, sb->s_blocksize);
338 }
339 
340 static inline struct buffer_head *
341 sb_getblk(struct super_block *sb, sector_t block)
342 {
343 	return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
344 }
345 
346 
347 static inline struct buffer_head *
348 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
349 {
350 	return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
351 }
352 
353 static inline struct buffer_head *
354 sb_find_get_block(struct super_block *sb, sector_t block)
355 {
356 	return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
357 }
358 
359 static inline void
360 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
361 {
362 	set_buffer_mapped(bh);
363 	bh->b_bdev = sb->s_bdev;
364 	bh->b_blocknr = block;
365 	bh->b_size = sb->s_blocksize;
366 }
367 
368 static inline void wait_on_buffer(struct buffer_head *bh)
369 {
370 	might_sleep();
371 	if (buffer_locked(bh))
372 		__wait_on_buffer(bh);
373 }
374 
375 static inline int trylock_buffer(struct buffer_head *bh)
376 {
377 	return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
378 }
379 
380 static inline void lock_buffer(struct buffer_head *bh)
381 {
382 	might_sleep();
383 	if (!trylock_buffer(bh))
384 		__lock_buffer(bh);
385 }
386 
387 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
388 						   sector_t block,
389 						   unsigned size)
390 {
391 	return __getblk_gfp(bdev, block, size, 0);
392 }
393 
394 static inline struct buffer_head *__getblk(struct block_device *bdev,
395 					   sector_t block,
396 					   unsigned size)
397 {
398 	return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
399 }
400 
401 static inline void bh_readahead(struct buffer_head *bh, blk_opf_t op_flags)
402 {
403 	if (!buffer_uptodate(bh) && trylock_buffer(bh)) {
404 		if (!buffer_uptodate(bh))
405 			__bh_read(bh, op_flags, false);
406 		else
407 			unlock_buffer(bh);
408 	}
409 }
410 
411 static inline void bh_read_nowait(struct buffer_head *bh, blk_opf_t op_flags)
412 {
413 	if (!bh_uptodate_or_lock(bh))
414 		__bh_read(bh, op_flags, false);
415 }
416 
417 /* Returns 1 if buffer uptodated, 0 on success, and -EIO on error. */
418 static inline int bh_read(struct buffer_head *bh, blk_opf_t op_flags)
419 {
420 	if (bh_uptodate_or_lock(bh))
421 		return 1;
422 	return __bh_read(bh, op_flags, true);
423 }
424 
425 static inline void bh_read_batch(int nr, struct buffer_head *bhs[])
426 {
427 	__bh_read_batch(nr, bhs, 0, true);
428 }
429 
430 static inline void bh_readahead_batch(int nr, struct buffer_head *bhs[],
431 				      blk_opf_t op_flags)
432 {
433 	__bh_read_batch(nr, bhs, op_flags, false);
434 }
435 
436 /**
437  *  __bread() - reads a specified block and returns the bh
438  *  @bdev: the block_device to read from
439  *  @block: number of block
440  *  @size: size (in bytes) to read
441  *
442  *  Reads a specified block, and returns buffer head that contains it.
443  *  The page cache is allocated from movable area so that it can be migrated.
444  *  It returns NULL if the block was unreadable.
445  */
446 static inline struct buffer_head *
447 __bread(struct block_device *bdev, sector_t block, unsigned size)
448 {
449 	return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
450 }
451 
452 bool block_dirty_folio(struct address_space *mapping, struct folio *folio);
453 
454 #ifdef CONFIG_BUFFER_HEAD
455 
456 void buffer_init(void);
457 bool try_to_free_buffers(struct folio *folio);
458 int inode_has_buffers(struct inode *inode);
459 void invalidate_inode_buffers(struct inode *inode);
460 int remove_inode_buffers(struct inode *inode);
461 int sync_mapping_buffers(struct address_space *mapping);
462 void invalidate_bh_lrus(void);
463 void invalidate_bh_lrus_cpu(void);
464 bool has_bh_in_lru(int cpu, void *dummy);
465 extern int buffer_heads_over_limit;
466 
467 #else /* CONFIG_BUFFER_HEAD */
468 
469 static inline void buffer_init(void) {}
470 static inline bool try_to_free_buffers(struct folio *folio) { return true; }
471 static inline int inode_has_buffers(struct inode *inode) { return 0; }
472 static inline void invalidate_inode_buffers(struct inode *inode) {}
473 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
474 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
475 static inline void invalidate_bh_lrus(void) {}
476 static inline void invalidate_bh_lrus_cpu(void) {}
477 static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }
478 #define buffer_heads_over_limit 0
479 
480 #endif /* CONFIG_BUFFER_HEAD */
481 #endif /* _LINUX_BUFFER_HEAD_H */
482