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 bool try_to_free_buffers(struct folio *);
 200 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
 201                 bool retry);
 202 void create_empty_buffers(struct page *, unsigned long,
 203                         unsigned long b_state);
 204 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
 205 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
 206 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
 207
 208 /* Things to do with buffers at mapping->private_list */
 209 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
 210 int inode_has_buffers(struct inode *);
 211 void invalidate_inode_buffers(struct inode *);
 212 int remove_inode_buffers(struct inode *inode);
 213 int sync_mapping_buffers(struct address_space *mapping);
 214 void clean_bdev_aliases(struct block_device *bdev, sector_t block,
 215                         sector_t len);
 216 static inline void clean_bdev_bh_alias(struct buffer_head *bh)
 217 {
 218         clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
 219 }
 220
 221 void mark_buffer_async_write(struct buffer_head *bh);
 222 void __wait_on_buffer(struct buffer_head *);
 223 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
 224 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
 225                         unsigned size);
 226 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
 227                                   unsigned size, gfp_t gfp);
 228 void __brelse(struct buffer_head *);
 229 void __bforget(struct buffer_head *);
 230 void __breadahead(struct block_device *, sector_t block, unsigned int size);
 231 struct buffer_head *__bread_gfp(struct block_device *,
 232                                 sector_t block, unsigned size, gfp_t gfp);
 233 void invalidate_bh_lrus(void);
 234 void invalidate_bh_lrus_cpu(void);
 235 bool has_bh_in_lru(int cpu, void *dummy);
 236 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 237 void free_buffer_head(struct buffer_head * bh);
 238 void unlock_buffer(struct buffer_head *bh);
 239 void __lock_buffer(struct buffer_head *bh);
 240 int sync_dirty_buffer(struct buffer_head *bh);
 241 int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
 242 void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);
 243 void submit_bh(blk_opf_t, struct buffer_head *);
 244 void write_boundary_block(struct block_device *bdev,
 245                         sector_t bblock, unsigned blocksize);
 246 int bh_uptodate_or_lock(struct buffer_head *bh);
 247 int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait);
 248 void __bh_read_batch(int nr, struct buffer_head *bhs[],
 249                      blk_opf_t op_flags, bool force_lock);
 250
 251 extern int buffer_heads_over_limit;
 252
 253 /*
 254  * Generic address_space_operations implementations for buffer_head-backed
 255  * address_spaces.
 256  */
 257 void block_invalidate_folio(struct folio *folio, size_t offset, size_t length);
 258 int block_write_full_page(struct page *page, get_block_t *get_block,
 259                                 struct writeback_control *wbc);
 260 int __block_write_full_page(struct inode *inode, struct page *page,
 261                         get_block_t *get_block, struct writeback_control *wbc,
 262                         bh_end_io_t *handler);
 263 int block_read_full_folio(struct folio *, get_block_t *);
 264 bool block_is_partially_uptodate(struct folio *, size_t from, size_t count);
 265 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
 266                 struct page **pagep, get_block_t *get_block);
 267 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
 268                 get_block_t *get_block);
 269 int block_write_end(struct file *, struct address_space *,
 270                                 loff_t, unsigned, unsigned,
 271                                 struct page *, void *);
 272 int generic_write_end(struct file *, struct address_space *,
 273                                 loff_t, unsigned, unsigned,
 274                                 struct page *, void *);
 275 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to);
 276 void clean_page_buffers(struct page *page);
 277 int cont_write_begin(struct file *, struct address_space *, loff_t,
 278                         unsigned, struct page **, void **,
 279                         get_block_t *, loff_t *);
 280 int generic_cont_expand_simple(struct inode *inode, loff_t size);
 281 int block_commit_write(struct page *page, unsigned from, unsigned to);
 282 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 283                                 get_block_t get_block);
 284 /* Convert errno to return value from ->page_mkwrite() call */
 285 static inline vm_fault_t block_page_mkwrite_return(int err)
 286 {
 287         if (err == 0)
 288                 return VM_FAULT_LOCKED;
 289         if (err == -EFAULT || err == -EAGAIN)
 290                 return VM_FAULT_NOPAGE;
 291         if (err == -ENOMEM)
 292                 return VM_FAULT_OOM;
 293         /* -ENOSPC, -EDQUOT, -EIO ... */
 294         return VM_FAULT_SIGBUS;
 295 }
 296 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
 297 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
 298
 299 #ifdef CONFIG_MIGRATION
 300 extern int buffer_migrate_folio(struct address_space *,
 301                 struct folio *dst, struct folio *src, enum migrate_mode);
 302 extern int buffer_migrate_folio_norefs(struct address_space *,
 303                 struct folio *dst, struct folio *src, enum migrate_mode);
 304 #else
 305 #define buffer_migrate_folio NULL
 306 #define buffer_migrate_folio_norefs NULL
 307 #endif
 308
 309 void buffer_init(void);
 310
 311 /*
 312  * inline definitions
 313  */
 314
 315 static inline void get_bh(struct buffer_head *bh)
 316 {
 317         atomic_inc(&bh->b_count);
 318 }
 319
 320 static inline void put_bh(struct buffer_head *bh)
 321 {
 322         smp_mb__before_atomic();
 323         atomic_dec(&bh->b_count);
 324 }
 325
 326 static inline void brelse(struct buffer_head *bh)
 327 {
 328         if (bh)
 329                 __brelse(bh);
 330 }
 331
 332 static inline void bforget(struct buffer_head *bh)
 333 {
 334         if (bh)
 335                 __bforget(bh);
 336 }
 337
 338 static inline struct buffer_head *
 339 sb_bread(struct super_block *sb, sector_t block)
 340 {
 341         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 342 }
 343
 344 static inline struct buffer_head *
 345 sb_bread_unmovable(struct super_block *sb, sector_t block)
 346 {
 347         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 348 }
 349
 350 static inline void
 351 sb_breadahead(struct super_block *sb, sector_t block)
 352 {
 353         __breadahead(sb->s_bdev, block, sb->s_blocksize);
 354 }
 355
 356 static inline struct buffer_head *
 357 sb_getblk(struct super_block *sb, sector_t block)
 358 {
 359         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 360 }
 361
 362
 363 static inline struct buffer_head *
 364 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
 365 {
 366         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
 367 }
 368
 369 static inline struct buffer_head *
 370 sb_find_get_block(struct super_block *sb, sector_t block)
 371 {
 372         return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
 373 }
 374
 375 static inline void
 376 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
 377 {
 378         set_buffer_mapped(bh);
 379         bh->b_bdev = sb->s_bdev;
 380         bh->b_blocknr = block;
 381         bh->b_size = sb->s_blocksize;
 382 }
 383
 384 static inline void wait_on_buffer(struct buffer_head *bh)
 385 {
 386         might_sleep();
 387         if (buffer_locked(bh))
 388                 __wait_on_buffer(bh);
 389 }
 390
 391 static inline int trylock_buffer(struct buffer_head *bh)
 392 {
 393         return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
 394 }
 395
 396 static inline void lock_buffer(struct buffer_head *bh)
 397 {
 398         might_sleep();
 399         if (!trylock_buffer(bh))
 400                 __lock_buffer(bh);
 401 }
 402
 403 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
 404                                                    sector_t block,
 405                                                    unsigned size)
 406 {
 407         return __getblk_gfp(bdev, block, size, 0);
 408 }
 409
 410 static inline struct buffer_head *__getblk(struct block_device *bdev,
 411                                            sector_t block,
 412                                            unsigned size)
 413 {
 414         return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
 415 }
 416
 417 static inline void bh_readahead(struct buffer_head *bh, blk_opf_t op_flags)
 418 {
 419         if (!buffer_uptodate(bh) && trylock_buffer(bh)) {
 420                 if (!buffer_uptodate(bh))
 421                         __bh_read(bh, op_flags, false);
 422                 else
 423                         unlock_buffer(bh);
 424         }
 425 }
 426
 427 static inline void bh_read_nowait(struct buffer_head *bh, blk_opf_t op_flags)
 428 {
 429         if (!bh_uptodate_or_lock(bh))
 430                 __bh_read(bh, op_flags, false);
 431 }
 432
 433 /* Returns 1 if buffer uptodated, 0 on success, and -EIO on error. */
 434 static inline int bh_read(struct buffer_head *bh, blk_opf_t op_flags)
 435 {
 436         if (bh_uptodate_or_lock(bh))
 437                 return 1;
 438         return __bh_read(bh, op_flags, true);
 439 }
 440
 441 static inline void bh_read_batch(int nr, struct buffer_head *bhs[])
 442 {
 443         __bh_read_batch(nr, bhs, 0, true);
 444 }
 445
 446 static inline void bh_readahead_batch(int nr, struct buffer_head *bhs[],
 447                                       blk_opf_t op_flags)
 448 {
 449         __bh_read_batch(nr, bhs, op_flags, false);
 450 }
 451
 452 /**
 453  *  __bread() - reads a specified block and returns the bh
 454  *  @bdev: the block_device to read from
 455  *  @block: number of block
 456  *  @size: size (in bytes) to read
 457  *
 458  *  Reads a specified block, and returns buffer head that contains it.
 459  *  The page cache is allocated from movable area so that it can be migrated.
 460  *  It returns NULL if the block was unreadable.
 461  */
 462 static inline struct buffer_head *
 463 __bread(struct block_device *bdev, sector_t block, unsigned size)
 464 {
 465         return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
 466 }
 467
 468 bool block_dirty_folio(struct address_space *mapping, struct folio *folio);
 469
 470 #else /* CONFIG_BLOCK */
 471
 472 static inline void buffer_init(void) {}
 473 static inline bool try_to_free_buffers(struct folio *folio) { return true; }
 474 static inline int inode_has_buffers(struct inode *inode) { return 0; }
 475 static inline void invalidate_inode_buffers(struct inode *inode) {}
 476 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 477 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
 478 static inline void invalidate_bh_lrus_cpu(void) {}
 479 static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }
 480 #define buffer_heads_over_limit 0
 481
 482 #endif /* CONFIG_BLOCK */
 483 #endif /* _LINUX_BUFFER_HEAD_H */