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/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
  20 enum 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
  49 struct page;
  50 struct buffer_head;
  51 struct address_space;
  52 typedef 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  */
  63 struct 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  */
  85 #define BUFFER_FNS(bit, name)                                           \
  86 static __always_inline void set_buffer_##name(struct buffer_head *bh)   \
  87 {                                                                       \
  88         set_bit(BH_##bit, &(bh)->b_state);                              \
  89 }                                                                       \
  90 static __always_inline void clear_buffer_##name(struct buffer_head *bh) \
  91 {                                                                       \
  92         clear_bit(BH_##bit, &(bh)->b_state);                            \
  93 }                                                                       \
  94 static __always_inline int buffer_##name(const struct buffer_head *bh)  \
  95 {                                                                       \
  96         return test_bit(BH_##bit, &(bh)->b_state);                      \
  97 }
  98
  99 /*
 100  * test_set_buffer_foo() and test_clear_buffer_foo()
 101  */
 102 #define TAS_BUFFER_FNS(bit, name)                                       \
 103 static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \
 104 {                                                                       \
 105         return test_and_set_bit(BH_##bit, &(bh)->b_state);              \
 106 }                                                                       \
 107 static __always_inline int test_clear_buffer_##name(struct buffer_head *bh) \
 108 {                                                                       \
 109         return test_and_clear_bit(BH_##bit, &(bh)->b_state);            \
 110 }                                                                       \
 111
 112 /*
 113  * Emit the buffer bitops functions.   Note that there are also functions
 114  * of the form "mark_buffer_foo()".  These are higher-level functions which
 115  * do something in addition to setting a b_state bit.
 116  */
 117 BUFFER_FNS(Dirty, dirty)
 118 TAS_BUFFER_FNS(Dirty, dirty)
 119 BUFFER_FNS(Lock, locked)
 120 BUFFER_FNS(Req, req)
 121 TAS_BUFFER_FNS(Req, req)
 122 BUFFER_FNS(Mapped, mapped)
 123 BUFFER_FNS(New, new)
 124 BUFFER_FNS(Async_Read, async_read)
 125 BUFFER_FNS(Async_Write, async_write)
 126 BUFFER_FNS(Delay, delay)
 127 BUFFER_FNS(Boundary, boundary)
 128 BUFFER_FNS(Write_EIO, write_io_error)
 129 BUFFER_FNS(Unwritten, unwritten)
 130 BUFFER_FNS(Meta, meta)
 131 BUFFER_FNS(Prio, prio)
 132 BUFFER_FNS(Defer_Completion, defer_completion)
 133
 134 static __always_inline void set_buffer_uptodate(struct buffer_head *bh)
 135 {
 136         /*
 137          * If somebody else already set this uptodate, they will
 138          * have done the memory barrier, and a reader will thus
 139          * see *some* valid buffer state.
 140          *
 141          * Any other serialization (with IO errors or whatever that
 142          * might clear the bit) has to come from other state (eg BH_Lock).
 143          */
 144         if (test_bit(BH_Uptodate, &bh->b_state))
 145                 return;
 146
 147         /*
 148          * make it consistent with folio_mark_uptodate
 149          * pairs with smp_load_acquire in buffer_uptodate
 150          */
 151         smp_mb__before_atomic();
 152         set_bit(BH_Uptodate, &bh->b_state);
 153 }
 154
 155 static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)
 156 {
 157         clear_bit(BH_Uptodate, &bh->b_state);
 158 }
 159
 160 static __always_inline int buffer_uptodate(const struct buffer_head *bh)
 161 {
 162         /*
 163          * make it consistent with folio_test_uptodate
 164          * pairs with smp_mb__before_atomic in set_buffer_uptodate
 165          */
 166         return (smp_load_acquire(&bh->b_state) & (1UL << BH_Uptodate)) != 0;
 167 }
 168
 169 #define bh_offset(bh)           ((unsigned long)(bh)->b_data & ~PAGE_MASK)
 170
 171 /* If we *know* page->private refers to buffer_heads */
 172 #define page_buffers(page)                                      \
 173         ({                                                      \
 174                 BUG_ON(!PagePrivate(page));                     \
 175                 ((struct buffer_head *)page_private(page));     \
 176         })
 177 #define page_has_buffers(page)  PagePrivate(page)
 178
 179 void buffer_check_dirty_writeback(struct page *page,
 180                                      bool *dirty, bool *writeback);
 181
 182 /*
 183  * Declarations
 184  */
 185
 186 void mark_buffer_dirty(struct buffer_head *bh);
 187 void mark_buffer_write_io_error(struct buffer_head *bh);
 188 void init_buffer(struct buffer_head *, bh_end_io_t *, void *);
 189 void touch_buffer(struct buffer_head *bh);
 190 void set_bh_page(struct buffer_head *bh,
 191                 struct page *page, unsigned long offset);
 192 int try_to_free_buffers(struct page *);
 193 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
 194                 int retry);
 195 void create_empty_buffers(struct page *, unsigned long,
 196                         unsigned long b_state);
 197 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
 198 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
 199 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
 200
 201 /* Things to do with buffers at mapping->private_list */
 202 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
 203 int inode_has_buffers(struct inode *);
 204 void invalidate_inode_buffers(struct inode *);
 205 int remove_inode_buffers(struct inode *inode);
 206 int sync_mapping_buffers(struct address_space *mapping);
 207 void clean_bdev_aliases(struct block_device *bdev, sector_t block,
 208                         sector_t len);
 209 static inline void clean_bdev_bh_alias(struct buffer_head *bh)
 210 {
 211         clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
 212 }
 213
 214 void mark_buffer_async_write(struct buffer_head *bh);
 215 void __wait_on_buffer(struct buffer_head *);
 216 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
 217 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
 218                         unsigned size);
 219 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
 220                                   unsigned size, gfp_t gfp);
 221 void __brelse(struct buffer_head *);
 222 void __bforget(struct buffer_head *);
 223 void __breadahead(struct block_device *, sector_t block, unsigned int size);
 224 void __breadahead_gfp(struct block_device *, sector_t block, unsigned int size,
 225                   gfp_t gfp);
 226 struct buffer_head *__bread_gfp(struct block_device *,
 227                                 sector_t block, unsigned size, gfp_t gfp);
 228 void invalidate_bh_lrus(void);
 229 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 230 void free_buffer_head(struct buffer_head * bh);
 231 void unlock_buffer(struct buffer_head *bh);
 232 void __lock_buffer(struct buffer_head *bh);
 233 void ll_rw_block(int, int, int, struct buffer_head * bh[]);
 234 int sync_dirty_buffer(struct buffer_head *bh);
 235 int __sync_dirty_buffer(struct buffer_head *bh, int op_flags);
 236 void write_dirty_buffer(struct buffer_head *bh, int op_flags);
 237 int submit_bh(int, int, struct buffer_head *);
 238 void write_boundary_block(struct block_device *bdev,
 239                         sector_t bblock, unsigned blocksize);
 240 int bh_uptodate_or_lock(struct buffer_head *bh);
 241 int bh_submit_read(struct buffer_head *bh);
 242 loff_t page_cache_seek_hole_data(struct inode *inode, loff_t offset,
 243                                  loff_t length, int whence);
 244
 245 extern int buffer_heads_over_limit;
 246
 247 /*
 248  * Generic address_space_operations implementations for buffer_head-backed
 249  * address_spaces.
 250  */
 251 void block_invalidatepage(struct page *page, unsigned int offset,
 252                           unsigned int length);
 253 int block_write_full_page(struct page *page, get_block_t *get_block,
 254                                 struct writeback_control *wbc);
 255 int __block_write_full_page(struct inode *inode, struct page *page,
 256                         get_block_t *get_block, struct writeback_control *wbc,
 257                         bh_end_io_t *handler);
 258 int block_read_full_page(struct page*, get_block_t*);
 259 int block_is_partially_uptodate(struct page *page, unsigned long from,
 260                                 unsigned long count);
 261 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
 262                 unsigned flags, struct page **pagep, get_block_t *get_block);
 263 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
 264                 get_block_t *get_block);
 265 int block_write_end(struct file *, struct address_space *,
 266                                 loff_t, unsigned, unsigned,
 267                                 struct page *, void *);
 268 int generic_write_end(struct file *, struct address_space *,
 269                                 loff_t, unsigned, unsigned,
 270                                 struct page *, void *);
 271 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to);
 272 void clean_page_buffers(struct page *page);
 273 int cont_write_begin(struct file *, struct address_space *, loff_t,
 274                         unsigned, unsigned, struct page **, void **,
 275                         get_block_t *, loff_t *);
 276 int generic_cont_expand_simple(struct inode *inode, loff_t size);
 277 int block_commit_write(struct page *page, unsigned from, unsigned to);
 278 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 279                                 get_block_t get_block);
 280 /* Convert errno to return value from ->page_mkwrite() call */
 281 static inline int block_page_mkwrite_return(int err)
 282 {
 283         if (err == 0)
 284                 return VM_FAULT_LOCKED;
 285         if (err == -EFAULT || err == -EAGAIN)
 286                 return VM_FAULT_NOPAGE;
 287         if (err == -ENOMEM)
 288                 return VM_FAULT_OOM;
 289         /* -ENOSPC, -EDQUOT, -EIO ... */
 290         return VM_FAULT_SIGBUS;
 291 }
 292 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
 293 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
 294 int nobh_write_begin(struct address_space *, loff_t, unsigned, unsigned,
 295                                 struct page **, void **, get_block_t*);
 296 int nobh_write_end(struct file *, struct address_space *,
 297                                 loff_t, unsigned, unsigned,
 298                                 struct page *, void *);
 299 int nobh_truncate_page(struct address_space *, loff_t, get_block_t *);
 300 int nobh_writepage(struct page *page, get_block_t *get_block,
 301                         struct writeback_control *wbc);
 302
 303 void buffer_init(void);
 304
 305 /*
 306  * inline definitions
 307  */
 308
 309 static inline void attach_page_buffers(struct page *page,
 310                 struct buffer_head *head)
 311 {
 312         get_page(page);
 313         SetPagePrivate(page);
 314         set_page_private(page, (unsigned long)head);
 315 }
 316
 317 static inline void get_bh(struct buffer_head *bh)
 318 {
 319         atomic_inc(&bh->b_count);
 320 }
 321
 322 static inline void put_bh(struct buffer_head *bh)
 323 {
 324         smp_mb__before_atomic();
 325         atomic_dec(&bh->b_count);
 326 }
 327
 328 static inline void brelse(struct buffer_head *bh)
 329 {
 330         if (bh)
 331                 __brelse(bh);
 332 }
 333
 334 static inline void bforget(struct buffer_head *bh)
 335 {
 336         if (bh)
 337                 __bforget(bh);
 338 }
 339
 340 static inline struct buffer_head *
 341 sb_bread(struct super_block *sb, sector_t block)
 342 {
 343         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 344 }
 345
 346 static inline struct buffer_head *
 347 sb_bread_unmovable(struct super_block *sb, sector_t block)
 348 {
 349         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 350 }
 351
 352 static inline void
 353 sb_breadahead(struct super_block *sb, sector_t block)
 354 {
 355         __breadahead(sb->s_bdev, block, sb->s_blocksize);
 356 }
 357
 358 static inline void
 359 sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 360 {
 361         __breadahead_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 362 }
 363
 364 static inline struct buffer_head *
 365 sb_getblk(struct super_block *sb, sector_t block)
 366 {
 367         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 368 }
 369
 370
 371 static inline struct buffer_head *
 372 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
 373 {
 374         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
 375 }
 376
 377 static inline struct buffer_head *
 378 sb_find_get_block(struct super_block *sb, sector_t block)
 379 {
 380         return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
 381 }
 382
 383 static inline void
 384 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
 385 {
 386         set_buffer_mapped(bh);
 387         bh->b_bdev = sb->s_bdev;
 388         bh->b_blocknr = block;
 389         bh->b_size = sb->s_blocksize;
 390 }
 391
 392 static inline void wait_on_buffer(struct buffer_head *bh)
 393 {
 394         might_sleep();
 395         if (buffer_locked(bh))
 396                 __wait_on_buffer(bh);
 397 }
 398
 399 static inline int trylock_buffer(struct buffer_head *bh)
 400 {
 401         return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
 402 }
 403
 404 static inline void lock_buffer(struct buffer_head *bh)
 405 {
 406         might_sleep();
 407         if (!trylock_buffer(bh))
 408                 __lock_buffer(bh);
 409 }
 410
 411 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
 412                                                    sector_t block,
 413                                                    unsigned size)
 414 {
 415         return __getblk_gfp(bdev, block, size, 0);
 416 }
 417
 418 static inline struct buffer_head *__getblk(struct block_device *bdev,
 419                                            sector_t block,
 420                                            unsigned size)
 421 {
 422         return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
 423 }
 424
 425 /**
 426  *  __bread() - reads a specified block and returns the bh
 427  *  @bdev: the block_device to read from
 428  *  @block: number of block
 429  *  @size: size (in bytes) to read
 430  *
 431  *  Reads a specified block, and returns buffer head that contains it.
 432  *  The page cache is allocated from movable area so that it can be migrated.
 433  *  It returns NULL if the block was unreadable.
 434  */
 435 static inline struct buffer_head *
 436 __bread(struct block_device *bdev, sector_t block, unsigned size)
 437 {
 438         return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
 439 }
 440
 441 extern int __set_page_dirty_buffers(struct page *page);
 442
 443 #else /* CONFIG_BLOCK */
 444
 445 static inline void buffer_init(void) {}
 446 static inline int try_to_free_buffers(struct page *page) { return 1; }
 447 static inline int inode_has_buffers(struct inode *inode) { return 0; }
 448 static inline void invalidate_inode_buffers(struct inode *inode) {}
 449 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 450 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
 451
 452 #endif /* CONFIG_BLOCK */
 453 #endif /* _LINUX_BUFFER_HEAD_H */