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
  26         BH_Mapped,      /* Has a disk mapping */
  27         BH_New,         /* Disk mapping was newly created by get_block */
  28         BH_Async_Read,  /* Is under end_buffer_async_read I/O */
  29         BH_Async_Write, /* Is under end_buffer_async_write I/O */
  30         BH_Delay,       /* Buffer is not yet allocated on disk */
  31         BH_Boundary,    /* Block is followed by a discontiguity */
  32         BH_Write_EIO,   /* I/O error on write */
  33         BH_Unwritten,   /* Buffer is allocated on disk but not written */
  34         BH_Quiet,       /* Buffer Error Prinks to be quiet */
  35         BH_Meta,        /* Buffer contains metadata */
  36         BH_Prio,        /* Buffer should be submitted with REQ_PRIO */
  37         BH_Defer_Completion, /* Defer AIO completion to workqueue */
  38
  39         BH_PrivateStart,/* not a state bit, but the first bit available
  40                          * for private allocation by other entities
  41                          */
  42 };
  43
  44 #define MAX_BUF_PER_PAGE (PAGE_SIZE / 512)
  45
  46 struct page;
  47 struct buffer_head;
  48 struct address_space;
  49 typedef void (bh_end_io_t)(struct buffer_head *bh, int uptodate);
  50
  51 /*
  52  * Historically, a buffer_head was used to map a single block
  53  * within a page, and of course as the unit of I/O through the
  54  * filesystem and block layers.  Nowadays the basic I/O unit
  55  * is the bio, and buffer_heads are used for extracting block
  56  * mappings (via a get_block_t call), for tracking state within
  57  * a page (via a page_mapping) and for wrapping bio submission
  58  * for backward compatibility reasons (e.g. submit_bh).
  59  */
  60 struct buffer_head {
  61         unsigned long b_state;          /* buffer state bitmap (see above) */
  62         struct buffer_head *b_this_page;/* circular list of page's buffers */
  63         struct page *b_page;            /* the page this bh is mapped to */
  64
  65         sector_t b_blocknr;             /* start block number */
  66         size_t b_size;                  /* size of mapping */
  67         char *b_data;                   /* pointer to data within the page */
  68
  69         struct block_device *b_bdev;
  70         bh_end_io_t *b_end_io;          /* I/O completion */
  71         void *b_private;                /* reserved for b_end_io */
  72         struct list_head b_assoc_buffers; /* associated with another mapping */
  73         struct address_space *b_assoc_map;      /* mapping this buffer is
  74                                                    associated with */
  75         atomic_t b_count;               /* users using this buffer_head */
  76         spinlock_t b_uptodate_lock;     /* Used by the first bh in a page, to
  77                                          * serialise IO completion of other
  78                                          * buffers in the page */
  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)                                           \
  88 static __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 }                                                                       \
  93 static __always_inline void clear_buffer_##name(struct buffer_head *bh) \
  94 {                                                                       \
  95         clear_bit(BH_##bit, &(bh)->b_state);                            \
  96 }                                                                       \
  97 static __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)                                       \
 106 static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \
 107 {                                                                       \
 108         return test_and_set_bit(BH_##bit, &(bh)->b_state);              \
 109 }                                                                       \
 110 static __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  */
 120 BUFFER_FNS(Dirty, dirty)
 121 TAS_BUFFER_FNS(Dirty, dirty)
 122 BUFFER_FNS(Lock, locked)
 123 BUFFER_FNS(Req, req)
 124 TAS_BUFFER_FNS(Req, req)
 125 BUFFER_FNS(Mapped, mapped)
 126 BUFFER_FNS(New, new)
 127 BUFFER_FNS(Async_Read, async_read)
 128 BUFFER_FNS(Async_Write, async_write)
 129 BUFFER_FNS(Delay, delay)
 130 BUFFER_FNS(Boundary, boundary)
 131 BUFFER_FNS(Write_EIO, write_io_error)
 132 BUFFER_FNS(Unwritten, unwritten)
 133 BUFFER_FNS(Meta, meta)
 134 BUFFER_FNS(Prio, prio)
 135 BUFFER_FNS(Defer_Completion, defer_completion)
 136
 137 static __always_inline void set_buffer_uptodate(struct buffer_head *bh)
 138 {
 139         /*
 140          * If somebody else already set this uptodate, they will
 141          * have done the memory barrier, and a reader will thus
 142          * see *some* valid buffer state.
 143          *
 144          * Any other serialization (with IO errors or whatever that
 145          * might clear the bit) has to come from other state (eg BH_Lock).
 146          */
 147         if (test_bit(BH_Uptodate, &bh->b_state))
 148                 return;
 149
 150         /*
 151          * make it consistent with folio_mark_uptodate
 152          * pairs with smp_load_acquire in buffer_uptodate
 153          */
 154         smp_mb__before_atomic();
 155         set_bit(BH_Uptodate, &bh->b_state);
 156 }
 157
 158 static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)
 159 {
 160         clear_bit(BH_Uptodate, &bh->b_state);
 161 }
 162
 163 static __always_inline int buffer_uptodate(const struct buffer_head *bh)
 164 {
 165         /*
 166          * make it consistent with folio_test_uptodate
 167          * pairs with smp_mb__before_atomic in set_buffer_uptodate
 168          */
 169         return (smp_load_acquire(&bh->b_state) & (1UL << BH_Uptodate)) != 0;
 170 }
 171
 172 #define bh_offset(bh)           ((unsigned long)(bh)->b_data & ~PAGE_MASK)
 173
 174 /* If we *know* page->private refers to buffer_heads */
 175 #define page_buffers(page)                                      \
 176         ({                                                      \
 177                 BUG_ON(!PagePrivate(page));                     \
 178                 ((struct buffer_head *)page_private(page));     \
 179         })
 180 #define page_has_buffers(page)  PagePrivate(page)
 181
 182 void buffer_check_dirty_writeback(struct page *page,
 183                                      bool *dirty, bool *writeback);
 184
 185 /*
 186  * Declarations
 187  */
 188
 189 void mark_buffer_dirty(struct buffer_head *bh);
 190 void mark_buffer_write_io_error(struct buffer_head *bh);
 191 void touch_buffer(struct buffer_head *bh);
 192 void set_bh_page(struct buffer_head *bh,
 193                 struct page *page, unsigned long offset);
 194 int try_to_free_buffers(struct page *);
 195 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
 196                 bool retry);
 197 void create_empty_buffers(struct page *, unsigned long,
 198                         unsigned long b_state);
 199 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
 200 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
 201 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
 202
 203 /* Things to do with buffers at mapping->private_list */
 204 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
 205 int inode_has_buffers(struct inode *);
 206 void invalidate_inode_buffers(struct inode *);
 207 int remove_inode_buffers(struct inode *inode);
 208 int sync_mapping_buffers(struct address_space *mapping);
 209 void clean_bdev_aliases(struct block_device *bdev, sector_t block,
 210                         sector_t len);
 211 static inline void clean_bdev_bh_alias(struct buffer_head *bh)
 212 {
 213         clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
 214 }
 215
 216 void mark_buffer_async_write(struct buffer_head *bh);
 217 void __wait_on_buffer(struct buffer_head *);
 218 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
 219 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
 220                         unsigned size);
 221 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
 222                                   unsigned size, gfp_t gfp);
 223 void __brelse(struct buffer_head *);
 224 void __bforget(struct buffer_head *);
 225 void __breadahead(struct block_device *, sector_t block, unsigned int size);
 226 void __breadahead_gfp(struct block_device *, sector_t block, unsigned int size,
 227                   gfp_t gfp);
 228 struct buffer_head *__bread_gfp(struct block_device *,
 229                                 sector_t block, unsigned size, gfp_t gfp);
 230 void invalidate_bh_lrus(void);
 231 void invalidate_bh_lrus_cpu(void);
 232 bool has_bh_in_lru(int cpu, void *dummy);
 233 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 234 void free_buffer_head(struct buffer_head * bh);
 235 void unlock_buffer(struct buffer_head *bh);
 236 void __lock_buffer(struct buffer_head *bh);
 237 void ll_rw_block(int, int, int, struct buffer_head * bh[]);
 238 int sync_dirty_buffer(struct buffer_head *bh);
 239 int __sync_dirty_buffer(struct buffer_head *bh, int op_flags);
 240 void write_dirty_buffer(struct buffer_head *bh, int op_flags);
 241 int submit_bh(int, int, struct buffer_head *);
 242 void write_boundary_block(struct block_device *bdev,
 243                         sector_t bblock, unsigned blocksize);
 244 int bh_uptodate_or_lock(struct buffer_head *bh);
 245 int bh_submit_read(struct buffer_head *bh);
 246
 247 extern int buffer_heads_over_limit;
 248
 249 /*
 250  * Generic address_space_operations implementations for buffer_head-backed
 251  * address_spaces.
 252  */
 253 void block_invalidatepage(struct page *page, unsigned int offset,
 254                           unsigned int length);
 255 int block_write_full_page(struct page *page, get_block_t *get_block,
 256                                 struct writeback_control *wbc);
 257 int __block_write_full_page(struct inode *inode, struct page *page,
 258                         get_block_t *get_block, struct writeback_control *wbc,
 259                         bh_end_io_t *handler);
 260 int block_read_full_page(struct page*, get_block_t*);
 261 int block_is_partially_uptodate(struct page *page, unsigned long from,
 262                                 unsigned long count);
 263 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
 264                 unsigned flags, struct page **pagep, get_block_t *get_block);
 265 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
 266                 get_block_t *get_block);
 267 int block_write_end(struct file *, struct address_space *,
 268                                 loff_t, unsigned, unsigned,
 269                                 struct page *, void *);
 270 int generic_write_end(struct file *, struct address_space *,
 271                                 loff_t, unsigned, unsigned,
 272                                 struct page *, void *);
 273 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to);
 274 void clean_page_buffers(struct page *page);
 275 int cont_write_begin(struct file *, struct address_space *, loff_t,
 276                         unsigned, unsigned, struct page **, void **,
 277                         get_block_t *, loff_t *);
 278 int generic_cont_expand_simple(struct inode *inode, loff_t size);
 279 int block_commit_write(struct page *page, unsigned from, unsigned to);
 280 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 281                                 get_block_t get_block);
 282 /* Convert errno to return value from ->page_mkwrite() call */
 283 static inline vm_fault_t block_page_mkwrite_return(int err)
 284 {
 285         if (err == 0)
 286                 return VM_FAULT_LOCKED;
 287         if (err == -EFAULT || err == -EAGAIN)
 288                 return VM_FAULT_NOPAGE;
 289         if (err == -ENOMEM)
 290                 return VM_FAULT_OOM;
 291         /* -ENOSPC, -EDQUOT, -EIO ... */
 292         return VM_FAULT_SIGBUS;
 293 }
 294 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
 295 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
 296 int nobh_write_begin(struct address_space *, loff_t, unsigned, unsigned,
 297                                 struct page **, void **, get_block_t*);
 298 int nobh_write_end(struct file *, struct address_space *,
 299                                 loff_t, unsigned, unsigned,
 300                                 struct page *, void *);
 301 int nobh_truncate_page(struct address_space *, loff_t, get_block_t *);
 302 int nobh_writepage(struct page *page, get_block_t *get_block,
 303                         struct writeback_control *wbc);
 304
 305 void buffer_init(void);
 306
 307 /*
 308  * inline definitions
 309  */
 310
 311 static inline void get_bh(struct buffer_head *bh)
 312 {
 313         atomic_inc(&bh->b_count);
 314 }
 315
 316 static inline void put_bh(struct buffer_head *bh)
 317 {
 318         smp_mb__before_atomic();
 319         atomic_dec(&bh->b_count);
 320 }
 321
 322 static inline void brelse(struct buffer_head *bh)
 323 {
 324         if (bh)
 325                 __brelse(bh);
 326 }
 327
 328 static inline void bforget(struct buffer_head *bh)
 329 {
 330         if (bh)
 331                 __bforget(bh);
 332 }
 333
 334 static inline struct buffer_head *
 335 sb_bread(struct super_block *sb, sector_t block)
 336 {
 337         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 338 }
 339
 340 static inline struct buffer_head *
 341 sb_bread_unmovable(struct super_block *sb, sector_t block)
 342 {
 343         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 344 }
 345
 346 static inline void
 347 sb_breadahead(struct super_block *sb, sector_t block)
 348 {
 349         __breadahead(sb->s_bdev, block, sb->s_blocksize);
 350 }
 351
 352 static inline void
 353 sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 354 {
 355         __breadahead_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 356 }
 357
 358 static inline struct buffer_head *
 359 sb_getblk(struct super_block *sb, sector_t block)
 360 {
 361         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 362 }
 363
 364
 365 static inline struct buffer_head *
 366 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
 367 {
 368         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
 369 }
 370
 371 static inline struct buffer_head *
 372 sb_find_get_block(struct super_block *sb, sector_t block)
 373 {
 374         return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
 375 }
 376
 377 static inline void
 378 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
 379 {
 380         set_buffer_mapped(bh);
 381         bh->b_bdev = sb->s_bdev;
 382         bh->b_blocknr = block;
 383         bh->b_size = sb->s_blocksize;
 384 }
 385
 386 static inline void wait_on_buffer(struct buffer_head *bh)
 387 {
 388         might_sleep();
 389         if (buffer_locked(bh))
 390                 __wait_on_buffer(bh);
 391 }
 392
 393 static inline int trylock_buffer(struct buffer_head *bh)
 394 {
 395         return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
 396 }
 397
 398 static inline void lock_buffer(struct buffer_head *bh)
 399 {
 400         might_sleep();
 401         if (!trylock_buffer(bh))
 402                 __lock_buffer(bh);
 403 }
 404
 405 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
 406                                                    sector_t block,
 407                                                    unsigned size)
 408 {
 409         return __getblk_gfp(bdev, block, size, 0);
 410 }
 411
 412 static inline struct buffer_head *__getblk(struct block_device *bdev,
 413                                            sector_t block,
 414                                            unsigned size)
 415 {
 416         return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
 417 }
 418
 419 /**
 420  *  __bread() - reads a specified block and returns the bh
 421  *  @bdev: the block_device to read from
 422  *  @block: number of block
 423  *  @size: size (in bytes) to read
 424  *
 425  *  Reads a specified block, and returns buffer head that contains it.
 426  *  The page cache is allocated from movable area so that it can be migrated.
 427  *  It returns NULL if the block was unreadable.
 428  */
 429 static inline struct buffer_head *
 430 __bread(struct block_device *bdev, sector_t block, unsigned size)
 431 {
 432         return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
 433 }
 434
 435 extern int __set_page_dirty_buffers(struct page *page);
 436
 437 #else /* CONFIG_BLOCK */
 438
 439 static inline void buffer_init(void) {}
 440 static inline int try_to_free_buffers(struct page *page) { return 1; }
 441 static inline int inode_has_buffers(struct inode *inode) { return 0; }
 442 static inline void invalidate_inode_buffers(struct inode *inode) {}
 443 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 444 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
 445 static inline void invalidate_bh_lrus_cpu(void) {}
 446 static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }
 447 #define buffer_heads_over_limit 0
 448
 449 #endif /* CONFIG_BLOCK */
 450 #endif /* _LINUX_BUFFER_HEAD_H */