include/linux/buffer_head.h

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