xref: /linux-6.15/include/linux/buffer_head.h (revision 2d91ecac)
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/fs.h>
13 #include <linux/linkage.h>
14 #include <linux/pagemap.h>
15 #include <linux/wait.h>
16 #include <linux/atomic.h>
17 
18 #ifdef CONFIG_BLOCK
19 
20 enum bh_state_bits {
21 	BH_Uptodate,	/* Contains valid data */
22 	BH_Dirty,	/* Is dirty */
23 	BH_Lock,	/* Is locked */
24 	BH_Req,		/* Has been submitted for I/O */
25 
26 	BH_Mapped,	/* Has a disk mapping */
27 	BH_New,		/* Disk mapping was newly created by get_block */
28 	BH_Async_Read,	/* Is under end_buffer_async_read I/O */
29 	BH_Async_Write,	/* Is under end_buffer_async_write I/O */
30 	BH_Delay,	/* Buffer is not yet allocated on disk */
31 	BH_Boundary,	/* Block is followed by a discontiguity */
32 	BH_Write_EIO,	/* I/O error on write */
33 	BH_Unwritten,	/* Buffer is allocated on disk but not written */
34 	BH_Quiet,	/* Buffer Error Prinks to be quiet */
35 	BH_Meta,	/* Buffer contains metadata */
36 	BH_Prio,	/* Buffer should be submitted with REQ_PRIO */
37 	BH_Defer_Completion, /* Defer AIO completion to workqueue */
38 
39 	BH_PrivateStart,/* not a state bit, but the first bit available
40 			 * for private allocation by other entities
41 			 */
42 };
43 
44 #define MAX_BUF_PER_PAGE (PAGE_SIZE / 512)
45 
46 struct page;
47 struct buffer_head;
48 struct address_space;
49 typedef void (bh_end_io_t)(struct buffer_head *bh, int uptodate);
50 
51 /*
52  * Historically, a buffer_head was used to map a single block
53  * within a page, and of course as the unit of I/O through the
54  * filesystem and block layers.  Nowadays the basic I/O unit
55  * is the bio, and buffer_heads are used for extracting block
56  * mappings (via a get_block_t call), for tracking state within
57  * a page (via a page_mapping) and for wrapping bio submission
58  * for backward compatibility reasons (e.g. submit_bh).
59  */
60 struct buffer_head {
61 	unsigned long b_state;		/* buffer state bitmap (see above) */
62 	struct buffer_head *b_this_page;/* circular list of page's buffers */
63 	struct page *b_page;		/* the page this bh is mapped to */
64 
65 	sector_t b_blocknr;		/* start block number */
66 	size_t b_size;			/* size of mapping */
67 	char *b_data;			/* pointer to data within the page */
68 
69 	struct block_device *b_bdev;
70 	bh_end_io_t *b_end_io;		/* I/O completion */
71  	void *b_private;		/* reserved for b_end_io */
72 	struct list_head b_assoc_buffers; /* associated with another mapping */
73 	struct address_space *b_assoc_map;	/* mapping this buffer is
74 						   associated with */
75 	atomic_t b_count;		/* users using this buffer_head */
76 	spinlock_t b_uptodate_lock;	/* Used by the first bh in a page, to
77 					 * serialise IO completion of other
78 					 * buffers in the page */
79 };
80 
81 /*
82  * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()
83  * and buffer_foo() functions.
84  * To avoid reset buffer flags that are already set, because that causes
85  * a costly cache line transition, check the flag first.
86  */
87 #define BUFFER_FNS(bit, name)						\
88 static __always_inline void set_buffer_##name(struct buffer_head *bh)	\
89 {									\
90 	if (!test_bit(BH_##bit, &(bh)->b_state))			\
91 		set_bit(BH_##bit, &(bh)->b_state);			\
92 }									\
93 static __always_inline void clear_buffer_##name(struct buffer_head *bh)	\
94 {									\
95 	clear_bit(BH_##bit, &(bh)->b_state);				\
96 }									\
97 static __always_inline int buffer_##name(const struct buffer_head *bh)	\
98 {									\
99 	return test_bit(BH_##bit, &(bh)->b_state);			\
100 }
101 
102 /*
103  * test_set_buffer_foo() and test_clear_buffer_foo()
104  */
105 #define TAS_BUFFER_FNS(bit, name)					\
106 static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \
107 {									\
108 	return test_and_set_bit(BH_##bit, &(bh)->b_state);		\
109 }									\
110 static __always_inline int test_clear_buffer_##name(struct buffer_head *bh) \
111 {									\
112 	return test_and_clear_bit(BH_##bit, &(bh)->b_state);		\
113 }									\
114 
115 /*
116  * Emit the buffer bitops functions.   Note that there are also functions
117  * of the form "mark_buffer_foo()".  These are higher-level functions which
118  * do something in addition to setting a b_state bit.
119  */
120 BUFFER_FNS(Uptodate, uptodate)
121 BUFFER_FNS(Dirty, dirty)
122 TAS_BUFFER_FNS(Dirty, dirty)
123 BUFFER_FNS(Lock, locked)
124 BUFFER_FNS(Req, req)
125 TAS_BUFFER_FNS(Req, req)
126 BUFFER_FNS(Mapped, mapped)
127 BUFFER_FNS(New, new)
128 BUFFER_FNS(Async_Read, async_read)
129 BUFFER_FNS(Async_Write, async_write)
130 BUFFER_FNS(Delay, delay)
131 BUFFER_FNS(Boundary, boundary)
132 BUFFER_FNS(Write_EIO, write_io_error)
133 BUFFER_FNS(Unwritten, unwritten)
134 BUFFER_FNS(Meta, meta)
135 BUFFER_FNS(Prio, prio)
136 BUFFER_FNS(Defer_Completion, defer_completion)
137 
138 #define bh_offset(bh)		((unsigned long)(bh)->b_data & ~PAGE_MASK)
139 
140 /* If we *know* page->private refers to buffer_heads */
141 #define page_buffers(page)					\
142 	({							\
143 		BUG_ON(!PagePrivate(page));			\
144 		((struct buffer_head *)page_private(page));	\
145 	})
146 #define page_has_buffers(page)	PagePrivate(page)
147 #define folio_buffers(folio)		folio_get_private(folio)
148 
149 void buffer_check_dirty_writeback(struct folio *folio,
150 				     bool *dirty, bool *writeback);
151 
152 /*
153  * Declarations
154  */
155 
156 void mark_buffer_dirty(struct buffer_head *bh);
157 void mark_buffer_write_io_error(struct buffer_head *bh);
158 void touch_buffer(struct buffer_head *bh);
159 void set_bh_page(struct buffer_head *bh,
160 		struct page *page, unsigned long offset);
161 bool try_to_free_buffers(struct folio *);
162 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
163 		bool retry);
164 void create_empty_buffers(struct page *, unsigned long,
165 			unsigned long b_state);
166 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
167 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
168 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
169 
170 /* Things to do with buffers at mapping->private_list */
171 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
172 int inode_has_buffers(struct inode *);
173 void invalidate_inode_buffers(struct inode *);
174 int remove_inode_buffers(struct inode *inode);
175 int sync_mapping_buffers(struct address_space *mapping);
176 void clean_bdev_aliases(struct block_device *bdev, sector_t block,
177 			sector_t len);
178 static inline void clean_bdev_bh_alias(struct buffer_head *bh)
179 {
180 	clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
181 }
182 
183 void mark_buffer_async_write(struct buffer_head *bh);
184 void __wait_on_buffer(struct buffer_head *);
185 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
186 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
187 			unsigned size);
188 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
189 				  unsigned size, gfp_t gfp);
190 void __brelse(struct buffer_head *);
191 void __bforget(struct buffer_head *);
192 void __breadahead(struct block_device *, sector_t block, unsigned int size);
193 void __breadahead_gfp(struct block_device *, sector_t block, unsigned int size,
194 		  gfp_t gfp);
195 struct buffer_head *__bread_gfp(struct block_device *,
196 				sector_t block, unsigned size, gfp_t gfp);
197 void invalidate_bh_lrus(void);
198 void invalidate_bh_lrus_cpu(void);
199 bool has_bh_in_lru(int cpu, void *dummy);
200 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
201 void free_buffer_head(struct buffer_head * bh);
202 void unlock_buffer(struct buffer_head *bh);
203 void __lock_buffer(struct buffer_head *bh);
204 void ll_rw_block(int, int, int, struct buffer_head * bh[]);
205 int sync_dirty_buffer(struct buffer_head *bh);
206 int __sync_dirty_buffer(struct buffer_head *bh, int op_flags);
207 void write_dirty_buffer(struct buffer_head *bh, int op_flags);
208 int submit_bh(int, int, struct buffer_head *);
209 void write_boundary_block(struct block_device *bdev,
210 			sector_t bblock, unsigned blocksize);
211 int bh_uptodate_or_lock(struct buffer_head *bh);
212 int bh_submit_read(struct buffer_head *bh);
213 
214 extern int buffer_heads_over_limit;
215 
216 /*
217  * Generic address_space_operations implementations for buffer_head-backed
218  * address_spaces.
219  */
220 void block_invalidate_folio(struct folio *folio, size_t offset, size_t length);
221 int block_write_full_page(struct page *page, get_block_t *get_block,
222 				struct writeback_control *wbc);
223 int __block_write_full_page(struct inode *inode, struct page *page,
224 			get_block_t *get_block, struct writeback_control *wbc,
225 			bh_end_io_t *handler);
226 int block_read_full_folio(struct folio *, get_block_t *);
227 bool block_is_partially_uptodate(struct folio *, size_t from, size_t count);
228 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
229 		struct page **pagep, get_block_t *get_block);
230 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
231 		get_block_t *get_block);
232 int block_write_end(struct file *, struct address_space *,
233 				loff_t, unsigned, unsigned,
234 				struct page *, void *);
235 int generic_write_end(struct file *, struct address_space *,
236 				loff_t, unsigned, unsigned,
237 				struct page *, void *);
238 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to);
239 void clean_page_buffers(struct page *page);
240 int cont_write_begin(struct file *, struct address_space *, loff_t,
241 			unsigned, struct page **, void **,
242 			get_block_t *, loff_t *);
243 int generic_cont_expand_simple(struct inode *inode, loff_t size);
244 int block_commit_write(struct page *page, unsigned from, unsigned to);
245 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
246 				get_block_t get_block);
247 /* Convert errno to return value from ->page_mkwrite() call */
248 static inline vm_fault_t block_page_mkwrite_return(int err)
249 {
250 	if (err == 0)
251 		return VM_FAULT_LOCKED;
252 	if (err == -EFAULT || err == -EAGAIN)
253 		return VM_FAULT_NOPAGE;
254 	if (err == -ENOMEM)
255 		return VM_FAULT_OOM;
256 	/* -ENOSPC, -EDQUOT, -EIO ... */
257 	return VM_FAULT_SIGBUS;
258 }
259 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
260 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
261 int nobh_write_begin(struct address_space *, loff_t, unsigned len,
262 				struct page **, void **, get_block_t*);
263 int nobh_write_end(struct file *, struct address_space *,
264 				loff_t, unsigned, unsigned,
265 				struct page *, void *);
266 int nobh_truncate_page(struct address_space *, loff_t, get_block_t *);
267 int nobh_writepage(struct page *page, get_block_t *get_block,
268                         struct writeback_control *wbc);
269 
270 void buffer_init(void);
271 
272 /*
273  * inline definitions
274  */
275 
276 static inline void get_bh(struct buffer_head *bh)
277 {
278         atomic_inc(&bh->b_count);
279 }
280 
281 static inline void put_bh(struct buffer_head *bh)
282 {
283         smp_mb__before_atomic();
284         atomic_dec(&bh->b_count);
285 }
286 
287 static inline void brelse(struct buffer_head *bh)
288 {
289 	if (bh)
290 		__brelse(bh);
291 }
292 
293 static inline void bforget(struct buffer_head *bh)
294 {
295 	if (bh)
296 		__bforget(bh);
297 }
298 
299 static inline struct buffer_head *
300 sb_bread(struct super_block *sb, sector_t block)
301 {
302 	return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
303 }
304 
305 static inline struct buffer_head *
306 sb_bread_unmovable(struct super_block *sb, sector_t block)
307 {
308 	return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
309 }
310 
311 static inline void
312 sb_breadahead(struct super_block *sb, sector_t block)
313 {
314 	__breadahead(sb->s_bdev, block, sb->s_blocksize);
315 }
316 
317 static inline void
318 sb_breadahead_unmovable(struct super_block *sb, sector_t block)
319 {
320 	__breadahead_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
321 }
322 
323 static inline struct buffer_head *
324 sb_getblk(struct super_block *sb, sector_t block)
325 {
326 	return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
327 }
328 
329 
330 static inline struct buffer_head *
331 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
332 {
333 	return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
334 }
335 
336 static inline struct buffer_head *
337 sb_find_get_block(struct super_block *sb, sector_t block)
338 {
339 	return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
340 }
341 
342 static inline void
343 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
344 {
345 	set_buffer_mapped(bh);
346 	bh->b_bdev = sb->s_bdev;
347 	bh->b_blocknr = block;
348 	bh->b_size = sb->s_blocksize;
349 }
350 
351 static inline void wait_on_buffer(struct buffer_head *bh)
352 {
353 	might_sleep();
354 	if (buffer_locked(bh))
355 		__wait_on_buffer(bh);
356 }
357 
358 static inline int trylock_buffer(struct buffer_head *bh)
359 {
360 	return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
361 }
362 
363 static inline void lock_buffer(struct buffer_head *bh)
364 {
365 	might_sleep();
366 	if (!trylock_buffer(bh))
367 		__lock_buffer(bh);
368 }
369 
370 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
371 						   sector_t block,
372 						   unsigned size)
373 {
374 	return __getblk_gfp(bdev, block, size, 0);
375 }
376 
377 static inline struct buffer_head *__getblk(struct block_device *bdev,
378 					   sector_t block,
379 					   unsigned size)
380 {
381 	return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
382 }
383 
384 /**
385  *  __bread() - reads a specified block and returns the bh
386  *  @bdev: the block_device to read from
387  *  @block: number of block
388  *  @size: size (in bytes) to read
389  *
390  *  Reads a specified block, and returns buffer head that contains it.
391  *  The page cache is allocated from movable area so that it can be migrated.
392  *  It returns NULL if the block was unreadable.
393  */
394 static inline struct buffer_head *
395 __bread(struct block_device *bdev, sector_t block, unsigned size)
396 {
397 	return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
398 }
399 
400 bool block_dirty_folio(struct address_space *mapping, struct folio *folio);
401 
402 #else /* CONFIG_BLOCK */
403 
404 static inline void buffer_init(void) {}
405 static inline bool try_to_free_buffers(struct folio *folio) { return true; }
406 static inline int inode_has_buffers(struct inode *inode) { return 0; }
407 static inline void invalidate_inode_buffers(struct inode *inode) {}
408 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
409 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
410 static inline void invalidate_bh_lrus_cpu(void) {}
411 static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }
412 #define buffer_heads_over_limit 0
413 
414 #endif /* CONFIG_BLOCK */
415 #endif /* _LINUX_BUFFER_HEAD_H */
416