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         union {
  65                 struct page *b_page;    /* the page this bh is mapped to */
  66                 struct folio *b_folio;  /* the folio this bh is mapped to */
  67         };
  68
  69         sector_t b_blocknr;             /* start block number */
  70         size_t b_size;                  /* size of mapping */
  71         char *b_data;                   /* pointer to data within the page */
  72
  73         struct block_device *b_bdev;
  74         bh_end_io_t *b_end_io;          /* I/O completion */
  75         void *b_private;                /* reserved for b_end_io */
  76         struct list_head b_assoc_buffers; /* associated with another mapping */
  77         struct address_space *b_assoc_map;      /* mapping this buffer is
  78                                                    associated with */
  79         atomic_t b_count;               /* users using this buffer_head */
  80         spinlock_t b_uptodate_lock;     /* Used by the first bh in a page, to
  81                                          * serialise IO completion of other
  82                                          * buffers in the page */
  83 };
  84
  85 /*
  86  * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()
  87  * and buffer_foo() functions.
  88  * To avoid reset buffer flags that are already set, because that causes
  89  * a costly cache line transition, check the flag first.
  90  */
  91 #define BUFFER_FNS(bit, name)                                           \
  92 static __always_inline void set_buffer_##name(struct buffer_head *bh)   \
  93 {                                                                       \
  94         if (!test_bit(BH_##bit, &(bh)->b_state))                        \
  95                 set_bit(BH_##bit, &(bh)->b_state);                      \
  96 }                                                                       \
  97 static __always_inline void clear_buffer_##name(struct buffer_head *bh) \
  98 {                                                                       \
  99         clear_bit(BH_##bit, &(bh)->b_state);                            \
 100 }                                                                       \
 101 static __always_inline int buffer_##name(const struct buffer_head *bh)  \
 102 {                                                                       \
 103         return test_bit(BH_##bit, &(bh)->b_state);                      \
 104 }
 105
 106 /*
 107  * test_set_buffer_foo() and test_clear_buffer_foo()
 108  */
 109 #define TAS_BUFFER_FNS(bit, name)                                       \
 110 static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \
 111 {                                                                       \
 112         return test_and_set_bit(BH_##bit, &(bh)->b_state);              \
 113 }                                                                       \
 114 static __always_inline int test_clear_buffer_##name(struct buffer_head *bh) \
 115 {                                                                       \
 116         return test_and_clear_bit(BH_##bit, &(bh)->b_state);            \
 117 }                                                                       \
 118
 119 /*
 120  * Emit the buffer bitops functions.   Note that there are also functions
 121  * of the form "mark_buffer_foo()".  These are higher-level functions which
 122  * do something in addition to setting a b_state bit.
 123  */
 124 BUFFER_FNS(Dirty, dirty)
 125 TAS_BUFFER_FNS(Dirty, dirty)
 126 BUFFER_FNS(Lock, locked)
 127 BUFFER_FNS(Req, req)
 128 TAS_BUFFER_FNS(Req, req)
 129 BUFFER_FNS(Mapped, mapped)
 130 BUFFER_FNS(New, new)
 131 BUFFER_FNS(Async_Read, async_read)
 132 BUFFER_FNS(Async_Write, async_write)
 133 BUFFER_FNS(Delay, delay)
 134 BUFFER_FNS(Boundary, boundary)
 135 BUFFER_FNS(Write_EIO, write_io_error)
 136 BUFFER_FNS(Unwritten, unwritten)
 137 BUFFER_FNS(Meta, meta)
 138 BUFFER_FNS(Prio, prio)
 139 BUFFER_FNS(Defer_Completion, defer_completion)
 140
 141 static __always_inline void set_buffer_uptodate(struct buffer_head *bh)
 142 {
 143         /*
 144          * If somebody else already set this uptodate, they will
 145          * have done the memory barrier, and a reader will thus
 146          * see *some* valid buffer state.
 147          *
 148          * Any other serialization (with IO errors or whatever that
 149          * might clear the bit) has to come from other state (eg BH_Lock).
 150          */
 151         if (test_bit(BH_Uptodate, &bh->b_state))
 152                 return;
 153
 154         /*
 155          * make it consistent with folio_mark_uptodate
 156          * pairs with smp_load_acquire in buffer_uptodate
 157          */
 158         smp_mb__before_atomic();
 159         set_bit(BH_Uptodate, &bh->b_state);
 160 }
 161
 162 static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)
 163 {
 164         clear_bit(BH_Uptodate, &bh->b_state);
 165 }
 166
 167 static __always_inline int buffer_uptodate(const struct buffer_head *bh)
 168 {
 169         /*
 170          * make it consistent with folio_test_uptodate
 171          * pairs with smp_mb__before_atomic in set_buffer_uptodate
 172          */
 173         return test_bit_acquire(BH_Uptodate, &bh->b_state);
 174 }
 175
 176 #define bh_offset(bh)           ((unsigned long)(bh)->b_data & ~PAGE_MASK)
 177
 178 /* If we *know* page->private refers to buffer_heads */
 179 #define page_buffers(page)                                      \
 180         ({                                                      \
 181                 BUG_ON(!PagePrivate(page));                     \
 182                 ((struct buffer_head *)page_private(page));     \
 183         })
 184 #define page_has_buffers(page)  PagePrivate(page)
 185 #define folio_buffers(folio)            folio_get_private(folio)
 186
 187 void buffer_check_dirty_writeback(struct folio *folio,
 188                                      bool *dirty, bool *writeback);
 189
 190 /*
 191  * Declarations
 192  */
 193
 194 void mark_buffer_dirty(struct buffer_head *bh);
 195 void mark_buffer_write_io_error(struct buffer_head *bh);
 196 void touch_buffer(struct buffer_head *bh);
 197 void set_bh_page(struct buffer_head *bh,
 198                 struct page *page, unsigned long offset);
 199 void folio_set_bh(struct buffer_head *bh, struct folio *folio,
 200                   unsigned long offset);
 201 bool try_to_free_buffers(struct folio *);
 202 struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size,
 203                                         bool retry);
 204 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
 205                 bool retry);
 206 void create_empty_buffers(struct page *, unsigned long,
 207                         unsigned long b_state);
 208 void folio_create_empty_buffers(struct folio *folio, unsigned long blocksize,
 209                                 unsigned long b_state);
 210 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
 211 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
 212 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
 213
 214 /* Things to do with buffers at mapping->private_list */
 215 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
 216 int inode_has_buffers(struct inode *);
 217 void invalidate_inode_buffers(struct inode *);
 218 int remove_inode_buffers(struct inode *inode);
 219 int sync_mapping_buffers(struct address_space *mapping);
 220 int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end,
 221                                   bool datasync);
 222 int generic_buffers_fsync(struct file *file, loff_t start, loff_t end,
 223                           bool datasync);
 224 void clean_bdev_aliases(struct block_device *bdev, sector_t block,
 225                         sector_t len);
 226 static inline void clean_bdev_bh_alias(struct buffer_head *bh)
 227 {
 228         clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
 229 }
 230
 231 void mark_buffer_async_write(struct buffer_head *bh);
 232 void __wait_on_buffer(struct buffer_head *);
 233 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
 234 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
 235                         unsigned size);
 236 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
 237                                   unsigned size, gfp_t gfp);
 238 void __brelse(struct buffer_head *);
 239 void __bforget(struct buffer_head *);
 240 void __breadahead(struct block_device *, sector_t block, unsigned int size);
 241 struct buffer_head *__bread_gfp(struct block_device *,
 242                                 sector_t block, unsigned size, gfp_t gfp);
 243 void invalidate_bh_lrus(void);
 244 void invalidate_bh_lrus_cpu(void);
 245 bool has_bh_in_lru(int cpu, void *dummy);
 246 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 247 void free_buffer_head(struct buffer_head * bh);
 248 void unlock_buffer(struct buffer_head *bh);
 249 void __lock_buffer(struct buffer_head *bh);
 250 int sync_dirty_buffer(struct buffer_head *bh);
 251 int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
 252 void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
 253 void submit_bh(blk_opf_t, struct buffer_head *);
 254 void write_boundary_block(struct block_device *bdev,
 255                         sector_t bblock, unsigned blocksize);
 256 int bh_uptodate_or_lock(struct buffer_head *bh);
 257 int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait);
 258 void __bh_read_batch(int nr, struct buffer_head *bhs[],
 259                      blk_opf_t op_flags, bool force_lock);
 260
 261 extern int buffer_heads_over_limit;
 262
 263 /*
 264  * Generic address_space_operations implementations for buffer_head-backed
 265  * address_spaces.
 266  */
 267 void block_invalidate_folio(struct folio *folio, size_t offset, size_t length);
 268 int block_write_full_page(struct page *page, get_block_t *get_block,
 269                                 struct writeback_control *wbc);
 270 int __block_write_full_folio(struct inode *inode, struct folio *folio,
 271                         get_block_t *get_block, struct writeback_control *wbc,
 272                         bh_end_io_t *handler);
 273 int block_read_full_folio(struct folio *, get_block_t *);
 274 bool block_is_partially_uptodate(struct folio *, size_t from, size_t count);
 275 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
 276                 struct page **pagep, get_block_t *get_block);
 277 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
 278                 get_block_t *get_block);
 279 int block_write_end(struct file *, struct address_space *,
 280                                 loff_t, unsigned, unsigned,
 281                                 struct page *, void *);
 282 int generic_write_end(struct file *, struct address_space *,
 283                                 loff_t, unsigned, unsigned,
 284                                 struct page *, void *);
 285 void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to);
 286 void clean_page_buffers(struct page *page);
 287 int cont_write_begin(struct file *, struct address_space *, loff_t,
 288                         unsigned, struct page **, void **,
 289                         get_block_t *, loff_t *);
 290 int generic_cont_expand_simple(struct inode *inode, loff_t size);
 291 int block_commit_write(struct page *page, unsigned from, unsigned to);
 292 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 293                                 get_block_t get_block);
 294 /* Convert errno to return value from ->page_mkwrite() call */
 295 static inline vm_fault_t block_page_mkwrite_return(int err)
 296 {
 297         if (err == 0)
 298                 return VM_FAULT_LOCKED;
 299         if (err == -EFAULT || err == -EAGAIN)
 300                 return VM_FAULT_NOPAGE;
 301         if (err == -ENOMEM)
 302                 return VM_FAULT_OOM;
 303         /* -ENOSPC, -EDQUOT, -EIO ... */
 304         return VM_FAULT_SIGBUS;
 305 }
 306 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
 307 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
 308
 309 #ifdef CONFIG_MIGRATION
 310 extern int buffer_migrate_folio(struct address_space *,
 311                 struct folio *dst, struct folio *src, enum migrate_mode);
 312 extern int buffer_migrate_folio_norefs(struct address_space *,
 313                 struct folio *dst, struct folio *src, enum migrate_mode);
 314 #else
 315 #define buffer_migrate_folio NULL
 316 #define buffer_migrate_folio_norefs NULL
 317 #endif
 318
 319 void buffer_init(void);
 320
 321 /*
 322  * inline definitions
 323  */
 324
 325 static inline void get_bh(struct buffer_head *bh)
 326 {
 327         atomic_inc(&bh->b_count);
 328 }
 329
 330 static inline void put_bh(struct buffer_head *bh)
 331 {
 332         smp_mb__before_atomic();
 333         atomic_dec(&bh->b_count);
 334 }
 335
 336 static inline void brelse(struct buffer_head *bh)
 337 {
 338         if (bh)
 339                 __brelse(bh);
 340 }
 341
 342 static inline void bforget(struct buffer_head *bh)
 343 {
 344         if (bh)
 345                 __bforget(bh);
 346 }
 347
 348 static inline struct buffer_head *
 349 sb_bread(struct super_block *sb, sector_t block)
 350 {
 351         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 352 }
 353
 354 static inline struct buffer_head *
 355 sb_bread_unmovable(struct super_block *sb, sector_t block)
 356 {
 357         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 358 }
 359
 360 static inline void
 361 sb_breadahead(struct super_block *sb, sector_t block)
 362 {
 363         __breadahead(sb->s_bdev, block, sb->s_blocksize);
 364 }
 365
 366 static inline struct buffer_head *
 367 sb_getblk(struct super_block *sb, sector_t block)
 368 {
 369         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 370 }
 371
 372
 373 static inline struct buffer_head *
 374 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
 375 {
 376         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
 377 }
 378
 379 static inline struct buffer_head *
 380 sb_find_get_block(struct super_block *sb, sector_t block)
 381 {
 382         return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
 383 }
 384
 385 static inline void
 386 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
 387 {
 388         set_buffer_mapped(bh);
 389         bh->b_bdev = sb->s_bdev;
 390         bh->b_blocknr = block;
 391         bh->b_size = sb->s_blocksize;
 392 }
 393
 394 static inline void wait_on_buffer(struct buffer_head *bh)
 395 {
 396         might_sleep();
 397         if (buffer_locked(bh))
 398                 __wait_on_buffer(bh);
 399 }
 400
 401 static inline int trylock_buffer(struct buffer_head *bh)
 402 {
 403         return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
 404 }
 405
 406 static inline void lock_buffer(struct buffer_head *bh)
 407 {
 408         might_sleep();
 409         if (!trylock_buffer(bh))
 410                 __lock_buffer(bh);
 411 }
 412
 413 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
 414                                                    sector_t block,
 415                                                    unsigned size)
 416 {
 417         return __getblk_gfp(bdev, block, size, 0);
 418 }
 419
 420 static inline struct buffer_head *__getblk(struct block_device *bdev,
 421                                            sector_t block,
 422                                            unsigned size)
 423 {
 424         return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
 425 }
 426
 427 static inline void bh_readahead(struct buffer_head *bh, blk_opf_t op_flags)
 428 {
 429         if (!buffer_uptodate(bh) && trylock_buffer(bh)) {
 430                 if (!buffer_uptodate(bh))
 431                         __bh_read(bh, op_flags, false);
 432                 else
 433                         unlock_buffer(bh);
 434         }
 435 }
 436
 437 static inline void bh_read_nowait(struct buffer_head *bh, blk_opf_t op_flags)
 438 {
 439         if (!bh_uptodate_or_lock(bh))
 440                 __bh_read(bh, op_flags, false);
 441 }
 442
 443 /* Returns 1 if buffer uptodated, 0 on success, and -EIO on error. */
 444 static inline int bh_read(struct buffer_head *bh, blk_opf_t op_flags)
 445 {
 446         if (bh_uptodate_or_lock(bh))
 447                 return 1;
 448         return __bh_read(bh, op_flags, true);
 449 }
 450
 451 static inline void bh_read_batch(int nr, struct buffer_head *bhs[])
 452 {
 453         __bh_read_batch(nr, bhs, 0, true);
 454 }
 455
 456 static inline void bh_readahead_batch(int nr, struct buffer_head *bhs[],
 457                                       blk_opf_t op_flags)
 458 {
 459         __bh_read_batch(nr, bhs, op_flags, false);
 460 }
 461
 462 /**
 463  *  __bread() - reads a specified block and returns the bh
 464  *  @bdev: the block_device to read from
 465  *  @block: number of block
 466  *  @size: size (in bytes) to read
 467  *
 468  *  Reads a specified block, and returns buffer head that contains it.
 469  *  The page cache is allocated from movable area so that it can be migrated.
 470  *  It returns NULL if the block was unreadable.
 471  */
 472 static inline struct buffer_head *
 473 __bread(struct block_device *bdev, sector_t block, unsigned size)
 474 {
 475         return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
 476 }
 477
 478 bool block_dirty_folio(struct address_space *mapping, struct folio *folio);
 479
 480 #else /* CONFIG_BLOCK */
 481
 482 static inline void buffer_init(void) {}
 483 static inline bool try_to_free_buffers(struct folio *folio) { return true; }
 484 static inline int inode_has_buffers(struct inode *inode) { return 0; }
 485 static inline void invalidate_inode_buffers(struct inode *inode) {}
 486 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 487 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
 488 static inline void invalidate_bh_lrus_cpu(void) {}
 489 static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }
 490 #define buffer_heads_over_limit 0
 491
 492 #endif /* CONFIG_BLOCK */
 493 #endif /* _LINUX_BUFFER_HEAD_H */