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
19enum 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 BH_Migrate, /* Buffer is being migrated (norefs) */
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
46struct page;
47struct buffer_head;
48struct address_space;
49typedef 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 folio (via a folio_mapping) and for wrapping bio submission
58 * for backward compatibility reasons (e.g. submit_bh).
59 */
60struct 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 union {
64 struct page *b_page; /* the page this bh is mapped to */
65 struct folio *b_folio; /* the folio this bh is mapped to */
66 };
67
68 sector_t b_blocknr; /* start block number */
69 size_t b_size; /* size of mapping */
70 char *b_data; /* pointer to data within the page */
71
72 struct block_device *b_bdev;
73 bh_end_io_t *b_end_io; /* I/O completion */
74 void *b_private; /* reserved for b_end_io */
75 struct list_head b_assoc_buffers; /* associated with another mapping */
76 struct address_space *b_assoc_map; /* mapping this buffer is
77 associated with */
78 atomic_t b_count; /* users using this buffer_head */
79 spinlock_t b_uptodate_lock; /* Used by the first bh in a page, to
80 * serialise IO completion of other
81 * buffers in the page */
82};
83
84/*
85 * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()
86 * and buffer_foo() functions.
87 * To avoid reset buffer flags that are already set, because that causes
88 * a costly cache line transition, check the flag first.
89 */
90#define BUFFER_FNS(bit, name) \
91static __always_inline void set_buffer_##name(struct buffer_head *bh) \
92{ \
93 if (!test_bit(BH_##bit, &(bh)->b_state)) \
94 set_bit(BH_##bit, &(bh)->b_state); \
95} \
96static __always_inline void clear_buffer_##name(struct buffer_head *bh) \
97{ \
98 clear_bit(BH_##bit, &(bh)->b_state); \
99} \
100static __always_inline int buffer_##name(const struct buffer_head *bh) \
101{ \
102 return test_bit(BH_##bit, &(bh)->b_state); \
103}
104
105/*
106 * test_set_buffer_foo() and test_clear_buffer_foo()
107 */
108#define TAS_BUFFER_FNS(bit, name) \
109static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \
110{ \
111 return test_and_set_bit(BH_##bit, &(bh)->b_state); \
112} \
113static __always_inline int test_clear_buffer_##name(struct buffer_head *bh) \
114{ \
115 return test_and_clear_bit(BH_##bit, &(bh)->b_state); \
116} \
117
118/*
119 * Emit the buffer bitops functions. Note that there are also functions
120 * of the form "mark_buffer_foo()". These are higher-level functions which
121 * do something in addition to setting a b_state bit.
122 */
123BUFFER_FNS(Dirty, dirty)
124TAS_BUFFER_FNS(Dirty, dirty)
125BUFFER_FNS(Lock, locked)
126BUFFER_FNS(Req, req)
127TAS_BUFFER_FNS(Req, req)
128BUFFER_FNS(Mapped, mapped)
129BUFFER_FNS(New, new)
130BUFFER_FNS(Async_Read, async_read)
131BUFFER_FNS(Async_Write, async_write)
132BUFFER_FNS(Delay, delay)
133BUFFER_FNS(Boundary, boundary)
134BUFFER_FNS(Write_EIO, write_io_error)
135BUFFER_FNS(Unwritten, unwritten)
136BUFFER_FNS(Meta, meta)
137BUFFER_FNS(Prio, prio)
138BUFFER_FNS(Defer_Completion, defer_completion)
139
140static __always_inline void set_buffer_uptodate(struct buffer_head *bh)
141{
142 /*
143 * If somebody else already set this uptodate, they will
144 * have done the memory barrier, and a reader will thus
145 * see *some* valid buffer state.
146 *
147 * Any other serialization (with IO errors or whatever that
148 * might clear the bit) has to come from other state (eg BH_Lock).
149 */
150 if (test_bit(BH_Uptodate, &bh->b_state))
151 return;
152
153 /*
154 * make it consistent with folio_mark_uptodate
155 * pairs with smp_load_acquire in buffer_uptodate
156 */
157 smp_mb__before_atomic();
158 set_bit(nr: BH_Uptodate, addr: &bh->b_state);
159}
160
161static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)
162{
163 clear_bit(nr: BH_Uptodate, addr: &bh->b_state);
164}
165
166static __always_inline int buffer_uptodate(const struct buffer_head *bh)
167{
168 /*
169 * make it consistent with folio_test_uptodate
170 * pairs with smp_mb__before_atomic in set_buffer_uptodate
171 */
172 return test_bit_acquire(BH_Uptodate, &bh->b_state);
173}
174
175static inline unsigned long bh_offset(const struct buffer_head *bh)
176{
177 return (unsigned long)(bh)->b_data & (page_size(page: bh->b_page) - 1);
178}
179
180/* If we *know* page->private refers to buffer_heads */
181#define page_buffers(page) \
182 ({ \
183 BUG_ON(!PagePrivate(page)); \
184 ((struct buffer_head *)page_private(page)); \
185 })
186#define folio_buffers(folio) folio_get_private(folio)
187
188void buffer_check_dirty_writeback(struct folio *folio,
189 bool *dirty, bool *writeback);
190
191/*
192 * Declarations
193 */
194
195void mark_buffer_dirty(struct buffer_head *bh);
196void mark_buffer_write_io_error(struct buffer_head *bh);
197void touch_buffer(struct buffer_head *bh);
198void folio_set_bh(struct buffer_head *bh, struct folio *folio,
199 unsigned long offset);
200struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size,
201 gfp_t gfp);
202struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size);
203struct buffer_head *create_empty_buffers(struct folio *folio,
204 unsigned long blocksize, unsigned long b_state);
205void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
206void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
207
208/* Things to do with buffers at mapping->private_list */
209void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
210int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end,
211 bool datasync);
212int generic_buffers_fsync(struct file *file, loff_t start, loff_t end,
213 bool datasync);
214void clean_bdev_aliases(struct block_device *bdev, sector_t block,
215 sector_t len);
216static inline void clean_bdev_bh_alias(struct buffer_head *bh)
217{
218 clean_bdev_aliases(bdev: bh->b_bdev, block: bh->b_blocknr, len: 1);
219}
220
221void mark_buffer_async_write(struct buffer_head *bh);
222void __wait_on_buffer(struct buffer_head *);
223wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
224struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
225 unsigned size);
226struct buffer_head *__find_get_block_nonatomic(struct block_device *bdev,
227 sector_t block, unsigned size);
228struct buffer_head *bdev_getblk(struct block_device *bdev, sector_t block,
229 unsigned size, gfp_t gfp);
230void __brelse(struct buffer_head *);
231void __bforget(struct buffer_head *);
232void __breadahead(struct block_device *, sector_t block, unsigned int size);
233struct buffer_head *__bread_gfp(struct block_device *,
234 sector_t block, unsigned size, gfp_t gfp);
235struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
236void free_buffer_head(struct buffer_head * bh);
237void unlock_buffer(struct buffer_head *bh);
238void __lock_buffer(struct buffer_head *bh);
239int sync_dirty_buffer(struct buffer_head *bh);
240int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
241void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
242void submit_bh(blk_opf_t, struct buffer_head *);
243void write_boundary_block(struct block_device *bdev,
244 sector_t bblock, unsigned blocksize);
245int bh_uptodate_or_lock(struct buffer_head *bh);
246int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait);
247void __bh_read_batch(int nr, struct buffer_head *bhs[],
248 blk_opf_t op_flags, bool force_lock);
249
250/*
251 * Generic address_space_operations implementations for buffer_head-backed
252 * address_spaces.
253 */
254void block_invalidate_folio(struct folio *folio, size_t offset, size_t length);
255int block_write_full_folio(struct folio *folio, struct writeback_control *wbc,
256 void *get_block);
257int __block_write_full_folio(struct inode *inode, struct folio *folio,
258 get_block_t *get_block, struct writeback_control *wbc);
259int block_read_full_folio(struct folio *, get_block_t *);
260bool block_is_partially_uptodate(struct folio *, size_t from, size_t count);
261int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
262 struct folio **foliop, get_block_t *get_block);
263int __block_write_begin(struct folio *folio, loff_t pos, unsigned len,
264 get_block_t *get_block);
265int block_write_end(loff_t pos, unsigned len, unsigned copied, struct folio *);
266int generic_write_end(const struct kiocb *, struct address_space *,
267 loff_t, unsigned len, unsigned copied,
268 struct folio *, void *);
269void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to);
270int cont_write_begin(const struct kiocb *, struct address_space *, loff_t,
271 unsigned, struct folio **, void **,
272 get_block_t *, loff_t *);
273int generic_cont_expand_simple(struct inode *inode, loff_t size);
274void block_commit_write(struct folio *folio, size_t from, size_t to);
275int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
276 get_block_t get_block);
277sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
278int block_truncate_page(struct address_space *, loff_t, get_block_t *);
279
280#ifdef CONFIG_MIGRATION
281extern int buffer_migrate_folio(struct address_space *,
282 struct folio *dst, struct folio *src, enum migrate_mode);
283extern int buffer_migrate_folio_norefs(struct address_space *,
284 struct folio *dst, struct folio *src, enum migrate_mode);
285#else
286#define buffer_migrate_folio NULL
287#define buffer_migrate_folio_norefs NULL
288#endif
289
290/*
291 * inline definitions
292 */
293
294static inline void get_bh(struct buffer_head *bh)
295{
296 atomic_inc(v: &bh->b_count);
297}
298
299static inline void put_bh(struct buffer_head *bh)
300{
301 smp_mb__before_atomic();
302 atomic_dec(v: &bh->b_count);
303}
304
305/**
306 * brelse - Release a buffer.
307 * @bh: The buffer to release.
308 *
309 * Decrement a buffer_head's reference count. If @bh is NULL, this
310 * function is a no-op.
311 *
312 * If all buffers on a folio have zero reference count, are clean
313 * and unlocked, and if the folio is unlocked and not under writeback
314 * then try_to_free_buffers() may strip the buffers from the folio in
315 * preparation for freeing it (sometimes, rarely, buffers are removed
316 * from a folio but it ends up not being freed, and buffers may later
317 * be reattached).
318 *
319 * Context: Any context.
320 */
321static inline void brelse(struct buffer_head *bh)
322{
323 if (bh)
324 __brelse(bh);
325}
326
327/**
328 * bforget - Discard any dirty data in a buffer.
329 * @bh: The buffer to forget.
330 *
331 * Call this function instead of brelse() if the data written to a buffer
332 * no longer needs to be written back. It will clear the buffer's dirty
333 * flag so writeback of this buffer will be skipped.
334 *
335 * Context: Any context.
336 */
337static inline void bforget(struct buffer_head *bh)
338{
339 if (bh)
340 __bforget(bh);
341}
342
343static inline struct buffer_head *
344sb_bread(struct super_block *sb, sector_t block)
345{
346 return __bread_gfp(sb->s_bdev, block, size: sb->s_blocksize, __GFP_MOVABLE);
347}
348
349static inline struct buffer_head *
350sb_bread_unmovable(struct super_block *sb, sector_t block)
351{
352 return __bread_gfp(sb->s_bdev, block, size: sb->s_blocksize, gfp: 0);
353}
354
355static inline void
356sb_breadahead(struct super_block *sb, sector_t block)
357{
358 __breadahead(sb->s_bdev, block, size: sb->s_blocksize);
359}
360
361static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
362 sector_t block, unsigned size)
363{
364 gfp_t gfp;
365
366 gfp = mapping_gfp_constraint(mapping: bdev->bd_mapping, gfp_mask: ~__GFP_FS);
367 gfp |= __GFP_NOFAIL;
368
369 return bdev_getblk(bdev, block, size, gfp);
370}
371
372static inline struct buffer_head *__getblk(struct block_device *bdev,
373 sector_t block, unsigned size)
374{
375 gfp_t gfp;
376
377 gfp = mapping_gfp_constraint(mapping: bdev->bd_mapping, gfp_mask: ~__GFP_FS);
378 gfp |= __GFP_MOVABLE | __GFP_NOFAIL;
379
380 return bdev_getblk(bdev, block, size, gfp);
381}
382
383static inline struct buffer_head *sb_getblk(struct super_block *sb,
384 sector_t block)
385{
386 return __getblk(bdev: sb->s_bdev, block, size: sb->s_blocksize);
387}
388
389static inline struct buffer_head *sb_getblk_gfp(struct super_block *sb,
390 sector_t block, gfp_t gfp)
391{
392 return bdev_getblk(bdev: sb->s_bdev, block, size: sb->s_blocksize, gfp);
393}
394
395static inline struct buffer_head *
396sb_find_get_block(struct super_block *sb, sector_t block)
397{
398 return __find_get_block(bdev: sb->s_bdev, block, size: sb->s_blocksize);
399}
400
401static inline struct buffer_head *
402sb_find_get_block_nonatomic(struct super_block *sb, sector_t block)
403{
404 return __find_get_block_nonatomic(bdev: sb->s_bdev, block, size: sb->s_blocksize);
405}
406
407static inline void
408map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
409{
410 set_buffer_mapped(bh);
411 bh->b_bdev = sb->s_bdev;
412 bh->b_blocknr = block;
413 bh->b_size = sb->s_blocksize;
414}
415
416static inline void wait_on_buffer(struct buffer_head *bh)
417{
418 might_sleep();
419 if (buffer_locked(bh))
420 __wait_on_buffer(bh);
421}
422
423static inline int trylock_buffer(struct buffer_head *bh)
424{
425 return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
426}
427
428static inline void lock_buffer(struct buffer_head *bh)
429{
430 might_sleep();
431 if (!trylock_buffer(bh))
432 __lock_buffer(bh);
433}
434
435static inline void bh_readahead(struct buffer_head *bh, blk_opf_t op_flags)
436{
437 if (!buffer_uptodate(bh) && trylock_buffer(bh)) {
438 if (!buffer_uptodate(bh))
439 __bh_read(bh, op_flags, wait: false);
440 else
441 unlock_buffer(bh);
442 }
443}
444
445static inline void bh_read_nowait(struct buffer_head *bh, blk_opf_t op_flags)
446{
447 if (!bh_uptodate_or_lock(bh))
448 __bh_read(bh, op_flags, wait: false);
449}
450
451/* Returns 1 if buffer uptodated, 0 on success, and -EIO on error. */
452static inline int bh_read(struct buffer_head *bh, blk_opf_t op_flags)
453{
454 if (bh_uptodate_or_lock(bh))
455 return 1;
456 return __bh_read(bh, op_flags, wait: true);
457}
458
459static inline void bh_read_batch(int nr, struct buffer_head *bhs[])
460{
461 __bh_read_batch(nr, bhs, op_flags: 0, force_lock: true);
462}
463
464static inline void bh_readahead_batch(int nr, struct buffer_head *bhs[],
465 blk_opf_t op_flags)
466{
467 __bh_read_batch(nr, bhs, op_flags, force_lock: false);
468}
469
470/**
471 * __bread() - Read a block.
472 * @bdev: The block device to read from.
473 * @block: Block number in units of block size.
474 * @size: The block size of this device in bytes.
475 *
476 * Read a specified block, and return the buffer head that refers
477 * to it. The memory is allocated from the movable area so that it can
478 * be migrated. The returned buffer head has its refcount increased.
479 * The caller should call brelse() when it has finished with the buffer.
480 *
481 * Context: May sleep waiting for I/O.
482 * Return: NULL if the block was unreadable.
483 */
484static inline struct buffer_head *__bread(struct block_device *bdev,
485 sector_t block, unsigned size)
486{
487 return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
488}
489
490/**
491 * get_nth_bh - Get a reference on the n'th buffer after this one.
492 * @bh: The buffer to start counting from.
493 * @count: How many buffers to skip.
494 *
495 * This is primarily useful for finding the nth buffer in a folio; in
496 * that case you pass the head buffer and the byte offset in the folio
497 * divided by the block size. It can be used for other purposes, but
498 * it will wrap at the end of the folio rather than returning NULL or
499 * proceeding to the next folio for you.
500 *
501 * Return: The requested buffer with an elevated refcount.
502 */
503static inline __must_check
504struct buffer_head *get_nth_bh(struct buffer_head *bh, unsigned int count)
505{
506 while (count--)
507 bh = bh->b_this_page;
508 get_bh(bh);
509 return bh;
510}
511
512bool block_dirty_folio(struct address_space *mapping, struct folio *folio);
513
514#ifdef CONFIG_BUFFER_HEAD
515
516void buffer_init(void);
517bool try_to_free_buffers(struct folio *folio);
518int inode_has_buffers(struct inode *inode);
519void invalidate_inode_buffers(struct inode *inode);
520int remove_inode_buffers(struct inode *inode);
521int sync_mapping_buffers(struct address_space *mapping);
522void invalidate_bh_lrus(void);
523void invalidate_bh_lrus_cpu(void);
524bool has_bh_in_lru(int cpu, void *dummy);
525extern int buffer_heads_over_limit;
526
527#else /* CONFIG_BUFFER_HEAD */
528
529static inline void buffer_init(void) {}
530static inline bool try_to_free_buffers(struct folio *folio) { return true; }
531static inline int inode_has_buffers(struct inode *inode) { return 0; }
532static inline void invalidate_inode_buffers(struct inode *inode) {}
533static inline int remove_inode_buffers(struct inode *inode) { return 1; }
534static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
535static inline void invalidate_bh_lrus(void) {}
536static inline void invalidate_bh_lrus_cpu(void) {}
537static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }
538#define buffer_heads_over_limit 0
539
540#endif /* CONFIG_BUFFER_HEAD */
541#endif /* _LINUX_BUFFER_HEAD_H */
542