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
20enum 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 BH_Uptodate_Lock,/* Used by the first bh in a page, to serialise
26 * IO completion of other buffers in the page
27 */
28
29 BH_Mapped, /* Has a disk mapping */
30 BH_New, /* Disk mapping was newly created by get_block */
31 BH_Async_Read, /* Is under end_buffer_async_read I/O */
32 BH_Async_Write, /* Is under end_buffer_async_write I/O */
33 BH_Delay, /* Buffer is not yet allocated on disk */
34 BH_Boundary, /* Block is followed by a discontiguity */
35 BH_Write_EIO, /* I/O error on write */
36 BH_Unwritten, /* Buffer is allocated on disk but not written */
37 BH_Quiet, /* Buffer Error Prinks to be quiet */
38 BH_Meta, /* Buffer contains metadata */
39 BH_Prio, /* Buffer should be submitted with REQ_PRIO */
40 BH_Defer_Completion, /* Defer AIO completion to workqueue */
41
42 BH_PrivateStart,/* not a state bit, but the first bit available
43 * for private allocation by other entities
44 */
45};
46
47#define MAX_BUF_PER_PAGE (PAGE_SIZE / 512)
48
49struct page;
50struct buffer_head;
51struct address_space;
52typedef void (bh_end_io_t)(struct buffer_head *bh, int uptodate);
53
54/*
55 * Historically, a buffer_head was used to map a single block
56 * within a page, and of course as the unit of I/O through the
57 * filesystem and block layers. Nowadays the basic I/O unit
58 * is the bio, and buffer_heads are used for extracting block
59 * mappings (via a get_block_t call), for tracking state within
60 * a page (via a page_mapping) and for wrapping bio submission
61 * for backward compatibility reasons (e.g. submit_bh).
62 */
63struct buffer_head {
64 unsigned long b_state; /* buffer state bitmap (see above) */
65 struct buffer_head *b_this_page;/* circular list of page's buffers */
66 struct page *b_page; /* the page this bh is mapped to */
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};
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) \
88static __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} \
93static __always_inline void clear_buffer_##name(struct buffer_head *bh) \
94{ \
95 clear_bit(BH_##bit, &(bh)->b_state); \
96} \
97static __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) \
106static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \
107{ \
108 return test_and_set_bit(BH_##bit, &(bh)->b_state); \
109} \
110static __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 */
120BUFFER_FNS(Uptodate, uptodate)
121BUFFER_FNS(Dirty, dirty)
122TAS_BUFFER_FNS(Dirty, dirty)
123BUFFER_FNS(Lock, locked)
124BUFFER_FNS(Req, req)
125TAS_BUFFER_FNS(Req, req)
126BUFFER_FNS(Mapped, mapped)
127BUFFER_FNS(New, new)
128BUFFER_FNS(Async_Read, async_read)
129BUFFER_FNS(Async_Write, async_write)
130BUFFER_FNS(Delay, delay)
131BUFFER_FNS(Boundary, boundary)
132BUFFER_FNS(Write_EIO, write_io_error)
133BUFFER_FNS(Unwritten, unwritten)
134BUFFER_FNS(Meta, meta)
135BUFFER_FNS(Prio, prio)
136BUFFER_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
148void buffer_check_dirty_writeback(struct page *page,
149 bool *dirty, bool *writeback);
150
151/*
152 * Declarations
153 */
154
155void mark_buffer_dirty(struct buffer_head *bh);
156void mark_buffer_write_io_error(struct buffer_head *bh);
157void touch_buffer(struct buffer_head *bh);
158void set_bh_page(struct buffer_head *bh,
159 struct page *page, unsigned long offset);
160int try_to_free_buffers(struct page *);
161struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
162 bool retry);
163void create_empty_buffers(struct page *, unsigned long,
164 unsigned long b_state);
165void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
166void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
167void end_buffer_async_write(struct buffer_head *bh, int uptodate);
168
169/* Things to do with buffers at mapping->private_list */
170void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
171int inode_has_buffers(struct inode *);
172void invalidate_inode_buffers(struct inode *);
173int remove_inode_buffers(struct inode *inode);
174int sync_mapping_buffers(struct address_space *mapping);
175void clean_bdev_aliases(struct block_device *bdev, sector_t block,
176 sector_t len);
177static inline void clean_bdev_bh_alias(struct buffer_head *bh)
178{
179 clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
180}
181
182void mark_buffer_async_write(struct buffer_head *bh);
183void __wait_on_buffer(struct buffer_head *);
184wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
185struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
186 unsigned size);
187struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
188 unsigned size, gfp_t gfp);
189void __brelse(struct buffer_head *);
190void __bforget(struct buffer_head *);
191void __breadahead(struct block_device *, sector_t block, unsigned int size);
192struct buffer_head *__bread_gfp(struct block_device *,
193 sector_t block, unsigned size, gfp_t gfp);
194void invalidate_bh_lrus(void);
195struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
196void free_buffer_head(struct buffer_head * bh);
197void unlock_buffer(struct buffer_head *bh);
198void __lock_buffer(struct buffer_head *bh);
199void ll_rw_block(int, int, int, struct buffer_head * bh[]);
200int sync_dirty_buffer(struct buffer_head *bh);
201int __sync_dirty_buffer(struct buffer_head *bh, int op_flags);
202void write_dirty_buffer(struct buffer_head *bh, int op_flags);
203int submit_bh(int, int, struct buffer_head *);
204void write_boundary_block(struct block_device *bdev,
205 sector_t bblock, unsigned blocksize);
206int bh_uptodate_or_lock(struct buffer_head *bh);
207int bh_submit_read(struct buffer_head *bh);
208
209extern int buffer_heads_over_limit;
210
211/*
212 * Generic address_space_operations implementations for buffer_head-backed
213 * address_spaces.
214 */
215void block_invalidatepage(struct page *page, unsigned int offset,
216 unsigned int length);
217int block_write_full_page(struct page *page, get_block_t *get_block,
218 struct writeback_control *wbc);
219int __block_write_full_page(struct inode *inode, struct page *page,
220 get_block_t *get_block, struct writeback_control *wbc,
221 bh_end_io_t *handler);
222int block_read_full_page(struct page*, get_block_t*);
223int block_is_partially_uptodate(struct page *page, unsigned long from,
224 unsigned long count);
225int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
226 unsigned flags, struct page **pagep, get_block_t *get_block);
227int __block_write_begin(struct page *page, loff_t pos, unsigned len,
228 get_block_t *get_block);
229int block_write_end(struct file *, struct address_space *,
230 loff_t, unsigned, unsigned,
231 struct page *, void *);
232int generic_write_end(struct file *, struct address_space *,
233 loff_t, unsigned, unsigned,
234 struct page *, void *);
235void page_zero_new_buffers(struct page *page, unsigned from, unsigned to);
236void clean_page_buffers(struct page *page);
237int cont_write_begin(struct file *, struct address_space *, loff_t,
238 unsigned, unsigned, struct page **, void **,
239 get_block_t *, loff_t *);
240int generic_cont_expand_simple(struct inode *inode, loff_t size);
241int block_commit_write(struct page *page, unsigned from, unsigned to);
242int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
243 get_block_t get_block);
244/* Convert errno to return value from ->page_mkwrite() call */
245static inline vm_fault_t block_page_mkwrite_return(int err)
246{
247 if (err == 0)
248 return VM_FAULT_LOCKED;
249 if (err == -EFAULT || err == -EAGAIN)
250 return VM_FAULT_NOPAGE;
251 if (err == -ENOMEM)
252 return VM_FAULT_OOM;
253 /* -ENOSPC, -EDQUOT, -EIO ... */
254 return VM_FAULT_SIGBUS;
255}
256sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
257int block_truncate_page(struct address_space *, loff_t, get_block_t *);
258int nobh_write_begin(struct address_space *, loff_t, unsigned, unsigned,
259 struct page **, void **, get_block_t*);
260int nobh_write_end(struct file *, struct address_space *,
261 loff_t, unsigned, unsigned,
262 struct page *, void *);
263int nobh_truncate_page(struct address_space *, loff_t, get_block_t *);
264int nobh_writepage(struct page *page, get_block_t *get_block,
265 struct writeback_control *wbc);
266
267void buffer_init(void);
268
269/*
270 * inline definitions
271 */
272
273static inline void attach_page_buffers(struct page *page,
274 struct buffer_head *head)
275{
276 get_page(page);
277 SetPagePrivate(page);
278 set_page_private(page, (unsigned long)head);
279}
280
281static inline void get_bh(struct buffer_head *bh)
282{
283 atomic_inc(&bh->b_count);
284}
285
286static inline void put_bh(struct buffer_head *bh)
287{
288 smp_mb__before_atomic();
289 atomic_dec(&bh->b_count);
290}
291
292static inline void brelse(struct buffer_head *bh)
293{
294 if (bh)
295 __brelse(bh);
296}
297
298static inline void bforget(struct buffer_head *bh)
299{
300 if (bh)
301 __bforget(bh);
302}
303
304static inline struct buffer_head *
305sb_bread(struct super_block *sb, sector_t block)
306{
307 return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
308}
309
310static inline struct buffer_head *
311sb_bread_unmovable(struct super_block *sb, sector_t block)
312{
313 return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
314}
315
316static inline void
317sb_breadahead(struct super_block *sb, sector_t block)
318{
319 __breadahead(sb->s_bdev, block, sb->s_blocksize);
320}
321
322static inline struct buffer_head *
323sb_getblk(struct super_block *sb, sector_t block)
324{
325 return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
326}
327
328
329static inline struct buffer_head *
330sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
331{
332 return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
333}
334
335static inline struct buffer_head *
336sb_find_get_block(struct super_block *sb, sector_t block)
337{
338 return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
339}
340
341static inline void
342map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
343{
344 set_buffer_mapped(bh);
345 bh->b_bdev = sb->s_bdev;
346 bh->b_blocknr = block;
347 bh->b_size = sb->s_blocksize;
348}
349
350static inline void wait_on_buffer(struct buffer_head *bh)
351{
352 might_sleep();
353 if (buffer_locked(bh))
354 __wait_on_buffer(bh);
355}
356
357static inline int trylock_buffer(struct buffer_head *bh)
358{
359 return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
360}
361
362static inline void lock_buffer(struct buffer_head *bh)
363{
364 might_sleep();
365 if (!trylock_buffer(bh))
366 __lock_buffer(bh);
367}
368
369static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
370 sector_t block,
371 unsigned size)
372{
373 return __getblk_gfp(bdev, block, size, 0);
374}
375
376static inline struct buffer_head *__getblk(struct block_device *bdev,
377 sector_t block,
378 unsigned size)
379{
380 return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
381}
382
383/**
384 * __bread() - reads a specified block and returns the bh
385 * @bdev: the block_device to read from
386 * @block: number of block
387 * @size: size (in bytes) to read
388 *
389 * Reads a specified block, and returns buffer head that contains it.
390 * The page cache is allocated from movable area so that it can be migrated.
391 * It returns NULL if the block was unreadable.
392 */
393static inline struct buffer_head *
394__bread(struct block_device *bdev, sector_t block, unsigned size)
395{
396 return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
397}
398
399extern int __set_page_dirty_buffers(struct page *page);
400
401#else /* CONFIG_BLOCK */
402
403static inline void buffer_init(void) {}
404static inline int try_to_free_buffers(struct page *page) { return 1; }
405static inline int inode_has_buffers(struct inode *inode) { return 0; }
406static inline void invalidate_inode_buffers(struct inode *inode) {}
407static inline int remove_inode_buffers(struct inode *inode) { return 1; }
408static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
409
410#endif /* CONFIG_BLOCK */
411#endif /* _LINUX_BUFFER_HEAD_H */
412