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          * If somebody else already set this uptodate, they will
 142          * have done the memory barrier, and a reader will thus
 143          * see *some* valid buffer state.
 144          *
 145          * Any other serialization (with IO errors or whatever that
 146          * might clear the bit) has to come from other state (eg BH_Lock).
 147          */
 148         if (test_bit(BH_Uptodate, &bh->b_state))
 149                 return;
 150
 151         /*
 152          * make it consistent with folio_mark_uptodate
 153          * pairs with smp_load_acquire in buffer_uptodate
 154          */
 155         smp_mb__before_atomic();
 156         set_bit(BH_Uptodate, &bh->b_state);
 157 }
 158
 159 static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)
 160 {
 161         clear_bit(BH_Uptodate, &bh->b_state);
 162 }
 163
 164 static __always_inline int buffer_uptodate(const struct buffer_head *bh)
 165 {
 166         /*
 167          * make it consistent with folio_test_uptodate
 168          * pairs with smp_mb__before_atomic in set_buffer_uptodate
 169          */
 170         return test_bit_acquire(BH_Uptodate, &bh->b_state);
 171 }
 172
 173 #define bh_offset(bh)           ((unsigned long)(bh)->b_data & ~PAGE_MASK)
 174
 175 /* If we *know* page->private refers to buffer_heads */
 176 #define page_buffers(page)                                      \
 177         ({                                                      \
 178                 BUG_ON(!PagePrivate(page));                     \
 179                 ((struct buffer_head *)page_private(page));     \
 180         })
 181 #define page_has_buffers(page)  PagePrivate(page)
 182 #define folio_buffers(folio)            folio_get_private(folio)
 183
 184 void buffer_check_dirty_writeback(struct folio *folio,
 185                                      bool *dirty, bool *writeback);
 186
 187 /*
 188  * Declarations
 189  */
 190
 191 void mark_buffer_dirty(struct buffer_head *bh);
 192 void mark_buffer_write_io_error(struct buffer_head *bh);
 193 void touch_buffer(struct buffer_head *bh);
 194 void set_bh_page(struct buffer_head *bh,
 195                 struct page *page, unsigned long offset);
 196 bool try_to_free_buffers(struct folio *);
 197 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
 198                 bool retry);
 199 void create_empty_buffers(struct page *, unsigned long,
 200                         unsigned long b_state);
 201 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
 202 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
 203 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
 204
 205 /* Things to do with buffers at mapping->private_list */
 206 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
 207 int inode_has_buffers(struct inode *);
 208 void invalidate_inode_buffers(struct inode *);
 209 int remove_inode_buffers(struct inode *inode);
 210 int sync_mapping_buffers(struct address_space *mapping);
 211 void clean_bdev_aliases(struct block_device *bdev, sector_t block,
 212                         sector_t len);
 213 static inline void clean_bdev_bh_alias(struct buffer_head *bh)
 214 {
 215         clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);
 216 }
 217
 218 void mark_buffer_async_write(struct buffer_head *bh);
 219 void __wait_on_buffer(struct buffer_head *);
 220 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
 221 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
 222                         unsigned size);
 223 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
 224                                   unsigned size, gfp_t gfp);
 225 void __brelse(struct buffer_head *);
 226 void __bforget(struct buffer_head *);
 227 void __breadahead(struct block_device *, sector_t block, unsigned int size);
 228 void __breadahead_gfp(struct block_device *, sector_t block, unsigned int size,
 229                   gfp_t gfp);
 230 struct buffer_head *__bread_gfp(struct block_device *,
 231                                 sector_t block, unsigned size, gfp_t gfp);
 232 void invalidate_bh_lrus(void);
 233 void invalidate_bh_lrus_cpu(void);
 234 bool has_bh_in_lru(int cpu, void *dummy);
 235 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 236 void free_buffer_head(struct buffer_head * bh);
 237 void unlock_buffer(struct buffer_head *bh);
 238 void __lock_buffer(struct buffer_head *bh);
 239 void ll_rw_block(blk_opf_t, int, 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 int 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_submit_read(struct buffer_head *bh);
 248
 249 extern int buffer_heads_over_limit;
 250
 251 /*
 252  * Generic address_space_operations implementations for buffer_head-backed
 253  * address_spaces.
 254  */
 255 void block_invalidate_folio(struct folio *folio, size_t offset, size_t length);
 256 int block_write_full_page(struct page *page, get_block_t *get_block,
 257                                 struct writeback_control *wbc);
 258 int __block_write_full_page(struct inode *inode, struct page *page,
 259                         get_block_t *get_block, struct writeback_control *wbc,
 260                         bh_end_io_t *handler);
 261 int block_read_full_folio(struct folio *, get_block_t *);
 262 bool block_is_partially_uptodate(struct folio *, size_t from, size_t count);
 263 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
 264                 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, 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
 297 #ifdef CONFIG_MIGRATION
 298 extern int buffer_migrate_folio(struct address_space *,
 299                 struct folio *dst, struct folio *src, enum migrate_mode);
 300 extern int buffer_migrate_folio_norefs(struct address_space *,
 301                 struct folio *dst, struct folio *src, enum migrate_mode);
 302 #else
 303 #define buffer_migrate_folio NULL
 304 #define buffer_migrate_folio_norefs NULL
 305 #endif
 306
 307 void buffer_init(void);
 308
 309 /*
 310  * inline definitions
 311  */
 312
 313 static inline void get_bh(struct buffer_head *bh)
 314 {
 315         atomic_inc(&bh->b_count);
 316 }
 317
 318 static inline void put_bh(struct buffer_head *bh)
 319 {
 320         smp_mb__before_atomic();
 321         atomic_dec(&bh->b_count);
 322 }
 323
 324 static inline void brelse(struct buffer_head *bh)
 325 {
 326         if (bh)
 327                 __brelse(bh);
 328 }
 329
 330 static inline void bforget(struct buffer_head *bh)
 331 {
 332         if (bh)
 333                 __bforget(bh);
 334 }
 335
 336 static inline struct buffer_head *
 337 sb_bread(struct super_block *sb, sector_t block)
 338 {
 339         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 340 }
 341
 342 static inline struct buffer_head *
 343 sb_bread_unmovable(struct super_block *sb, sector_t block)
 344 {
 345         return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 346 }
 347
 348 static inline void
 349 sb_breadahead(struct super_block *sb, sector_t block)
 350 {
 351         __breadahead(sb->s_bdev, block, sb->s_blocksize);
 352 }
 353
 354 static inline void
 355 sb_breadahead_unmovable(struct super_block *sb, sector_t block)
 356 {
 357         __breadahead_gfp(sb->s_bdev, block, sb->s_blocksize, 0);
 358 }
 359
 360 static inline struct buffer_head *
 361 sb_getblk(struct super_block *sb, sector_t block)
 362 {
 363         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 364 }
 365
 366
 367 static inline struct buffer_head *
 368 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
 369 {
 370         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
 371 }
 372
 373 static inline struct buffer_head *
 374 sb_find_get_block(struct super_block *sb, sector_t block)
 375 {
 376         return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
 377 }
 378
 379 static inline void
 380 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
 381 {
 382         set_buffer_mapped(bh);
 383         bh->b_bdev = sb->s_bdev;
 384         bh->b_blocknr = block;
 385         bh->b_size = sb->s_blocksize;
 386 }
 387
 388 static inline void wait_on_buffer(struct buffer_head *bh)
 389 {
 390         might_sleep();
 391         if (buffer_locked(bh))
 392                 __wait_on_buffer(bh);
 393 }
 394
 395 static inline int trylock_buffer(struct buffer_head *bh)
 396 {
 397         return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
 398 }
 399
 400 static inline void lock_buffer(struct buffer_head *bh)
 401 {
 402         might_sleep();
 403         if (!trylock_buffer(bh))
 404                 __lock_buffer(bh);
 405 }
 406
 407 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
 408                                                    sector_t block,
 409                                                    unsigned size)
 410 {
 411         return __getblk_gfp(bdev, block, size, 0);
 412 }
 413
 414 static inline struct buffer_head *__getblk(struct block_device *bdev,
 415                                            sector_t block,
 416                                            unsigned size)
 417 {
 418         return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
 419 }
 420
 421 /**
 422  *  __bread() - reads a specified block and returns the bh
 423  *  @bdev: the block_device to read from
 424  *  @block: number of block
 425  *  @size: size (in bytes) to read
 426  *
 427  *  Reads a specified block, and returns buffer head that contains it.
 428  *  The page cache is allocated from movable area so that it can be migrated.
 429  *  It returns NULL if the block was unreadable.
 430  */
 431 static inline struct buffer_head *
 432 __bread(struct block_device *bdev, sector_t block, unsigned size)
 433 {
 434         return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
 435 }
 436
 437 bool block_dirty_folio(struct address_space *mapping, struct folio *folio);
 438
 439 #else /* CONFIG_BLOCK */
 440
 441 static inline void buffer_init(void) {}
 442 static inline bool try_to_free_buffers(struct folio *folio) { return true; }
 443 static inline int inode_has_buffers(struct inode *inode) { return 0; }
 444 static inline void invalidate_inode_buffers(struct inode *inode) {}
 445 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 446 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
 447 static inline void invalidate_bh_lrus_cpu(void) {}
 448 static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }
 449 #define buffer_heads_over_limit 0
 450
 451 #endif /* CONFIG_BLOCK */
 452 #endif /* _LINUX_BUFFER_HEAD_H */