kernel/include/linux/buffer_head.h

   1 /*
   2  * include/linux/buffer_head.h
   3  *
   4  * Everything to do with buffer_heads.
   5  */
   6
   7 #ifndef _LINUX_BUFFER_HEAD_H
   8 #define _LINUX_BUFFER_HEAD_H
   9
  10 #include <linux/types.h>
  11 #include <linux/fs.h>
  12 #include <linux/linkage.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/wait.h>
  15 #include <linux/atomic.h>
  16
  17 #ifdef CONFIG_BLOCK
  18
  19 enum bh_state_bits {
  20         BH_Uptodate,    /* Contains valid data */
  21         BH_Dirty,       /* Is dirty */
  22         BH_Lock,        /* Is locked */
  23         BH_Req,         /* Has been submitted for I/O */
  24         BH_Uptodate_Lock,/* Used by the first bh in a page, to serialise
  25                           * IO completion of other buffers in the page
  26                           */
  27
  28         BH_Mapped,      /* Has a disk mapping */
  29         BH_New,         /* Disk mapping was newly created by get_block */
  30         BH_Async_Read,  /* Is under end_buffer_async_read I/O */
  31         BH_Async_Write, /* Is under end_buffer_async_write I/O */
  32         BH_Delay,       /* Buffer is not yet allocated on disk */
  33         BH_Boundary,    /* Block is followed by a discontiguity */
  34         BH_Write_EIO,   /* I/O error on write */
  35         BH_Unwritten,   /* Buffer is allocated on disk but not written */
  36         BH_Quiet,       /* Buffer Error Prinks to be quiet */
  37         BH_Meta,        /* Buffer contains metadata */
  38         BH_Prio,        /* Buffer should be submitted with REQ_PRIO */
  39         BH_Defer_Completion, /* Defer AIO completion to workqueue */
  40
  41         BH_PrivateStart,/* not a state bit, but the first bit available
  42                          * for private allocation by other entities
  43                          */
  44 };
  45
  46 #define MAX_BUF_PER_PAGE (PAGE_CACHE_SIZE / 512)
  47
  48 struct page;
  49 struct buffer_head;
  50 struct address_space;
  51 typedef void (bh_end_io_t)(struct buffer_head *bh, int uptodate);
  52
  53 /*
  54  * Historically, a buffer_head was used to map a single block
  55  * within a page, and of course as the unit of I/O through the
  56  * filesystem and block layers.  Nowadays the basic I/O unit
  57  * is the bio, and buffer_heads are used for extracting block
  58  * mappings (via a get_block_t call), for tracking state within
  59  * a page (via a page_mapping) and for wrapping bio submission
  60  * for backward compatibility reasons (e.g. submit_bh).
  61  */
  62 struct buffer_head {
  63         unsigned long b_state;          /* buffer state bitmap (see above) */
  64         struct buffer_head *b_this_page;/* circular list of page's buffers */
  65         struct page *b_page;            /* the page this bh is mapped to */
  66
  67         sector_t b_blocknr;             /* start block number */
  68         size_t b_size;                  /* size of mapping */
  69         char *b_data;                   /* pointer to data within the page */
  70
  71         struct block_device *b_bdev;
  72         bh_end_io_t *b_end_io;          /* I/O completion */
  73         void *b_private;                /* reserved for b_end_io */
  74         struct list_head b_assoc_buffers; /* associated with another mapping */
  75         struct address_space *b_assoc_map;      /* mapping this buffer is
  76                                                    associated with */
  77         atomic_t b_count;               /* users using this buffer_head */
  78 #ifdef CONFIG_PREEMPT_RT_BASE
  79         spinlock_t b_uptodate_lock;
  80 #if IS_ENABLED(CONFIG_JBD2)
  81         spinlock_t b_state_lock;
  82         spinlock_t b_journal_head_lock;
  83 #endif
  84 #endif
  85 };
  86
  87 static inline unsigned long bh_uptodate_lock_irqsave(struct buffer_head *bh)
  88 {
  89         unsigned long flags;
  90
  91 #ifndef CONFIG_PREEMPT_RT_BASE
  92         local_irq_save(flags);
  93         bit_spin_lock(BH_Uptodate_Lock, &bh->b_state);
  94 #else
  95         spin_lock_irqsave(&bh->b_uptodate_lock, flags);
  96 #endif
  97         return flags;
  98 }
  99
 100 static inline void
 101 bh_uptodate_unlock_irqrestore(struct buffer_head *bh, unsigned long flags)
 102 {
 103 #ifndef CONFIG_PREEMPT_RT_BASE
 104         bit_spin_unlock(BH_Uptodate_Lock, &bh->b_state);
 105         local_irq_restore(flags);
 106 #else
 107         spin_unlock_irqrestore(&bh->b_uptodate_lock, flags);
 108 #endif
 109 }
 110
 111 static inline void buffer_head_init_locks(struct buffer_head *bh)
 112 {
 113 #ifdef CONFIG_PREEMPT_RT_BASE
 114         spin_lock_init(&bh->b_uptodate_lock);
 115 #if IS_ENABLED(CONFIG_JBD2)
 116         spin_lock_init(&bh->b_state_lock);
 117         spin_lock_init(&bh->b_journal_head_lock);
 118 #endif
 119 #endif
 120 }
 121
 122 /*
 123  * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()
 124  * and buffer_foo() functions.
 125  */
 126 #define BUFFER_FNS(bit, name)                                           \
 127 static inline void set_buffer_##name(struct buffer_head *bh)            \
 128 {                                                                       \
 129         set_bit(BH_##bit, &(bh)->b_state);                              \
 130 }                                                                       \
 131 static inline void clear_buffer_##name(struct buffer_head *bh)          \
 132 {                                                                       \
 133         clear_bit(BH_##bit, &(bh)->b_state);                            \
 134 }                                                                       \
 135 static inline int buffer_##name(const struct buffer_head *bh)           \
 136 {                                                                       \
 137         return test_bit(BH_##bit, &(bh)->b_state);                      \
 138 }
 139
 140 /*
 141  * test_set_buffer_foo() and test_clear_buffer_foo()
 142  */
 143 #define TAS_BUFFER_FNS(bit, name)                                       \
 144 static inline int test_set_buffer_##name(struct buffer_head *bh)        \
 145 {                                                                       \
 146         return test_and_set_bit(BH_##bit, &(bh)->b_state);              \
 147 }                                                                       \
 148 static inline int test_clear_buffer_##name(struct buffer_head *bh)      \
 149 {                                                                       \
 150         return test_and_clear_bit(BH_##bit, &(bh)->b_state);            \
 151 }                                                                       \
 152
 153 /*
 154  * Emit the buffer bitops functions.   Note that there are also functions
 155  * of the form "mark_buffer_foo()".  These are higher-level functions which
 156  * do something in addition to setting a b_state bit.
 157  */
 158 BUFFER_FNS(Uptodate, uptodate)
 159 BUFFER_FNS(Dirty, dirty)
 160 TAS_BUFFER_FNS(Dirty, dirty)
 161 BUFFER_FNS(Lock, locked)
 162 BUFFER_FNS(Req, req)
 163 TAS_BUFFER_FNS(Req, req)
 164 BUFFER_FNS(Mapped, mapped)
 165 BUFFER_FNS(New, new)
 166 BUFFER_FNS(Async_Read, async_read)
 167 BUFFER_FNS(Async_Write, async_write)
 168 BUFFER_FNS(Delay, delay)
 169 BUFFER_FNS(Boundary, boundary)
 170 BUFFER_FNS(Write_EIO, write_io_error)
 171 BUFFER_FNS(Unwritten, unwritten)
 172 BUFFER_FNS(Meta, meta)
 173 BUFFER_FNS(Prio, prio)
 174 BUFFER_FNS(Defer_Completion, defer_completion)
 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
 186 void buffer_check_dirty_writeback(struct page *page,
 187                                      bool *dirty, bool *writeback);
 188
 189 /*
 190  * Declarations
 191  */
 192
 193 void mark_buffer_dirty(struct buffer_head *bh);
 194 void init_buffer(struct buffer_head *, bh_end_io_t *, void *);
 195 void touch_buffer(struct buffer_head *bh);
 196 void set_bh_page(struct buffer_head *bh,
 197                 struct page *page, unsigned long offset);
 198 int try_to_free_buffers(struct page *);
 199 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
 200                 int retry);
 201 void create_empty_buffers(struct page *, unsigned long,
 202                         unsigned long b_state);
 203 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
 204 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
 205 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
 206
 207 /* Things to do with buffers at mapping->private_list */
 208 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
 209 int inode_has_buffers(struct inode *);
 210 void invalidate_inode_buffers(struct inode *);
 211 int remove_inode_buffers(struct inode *inode);
 212 int sync_mapping_buffers(struct address_space *mapping);
 213 void unmap_underlying_metadata(struct block_device *bdev, sector_t block);
 214
 215 void mark_buffer_async_write(struct buffer_head *bh);
 216 void __wait_on_buffer(struct buffer_head *);
 217 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
 218 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
 219                         unsigned size);
 220 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
 221                                   unsigned size, gfp_t gfp);
 222 void __brelse(struct buffer_head *);
 223 void __bforget(struct buffer_head *);
 224 void __breadahead(struct block_device *, sector_t block, unsigned int size);
 225 struct buffer_head *__bread_gfp(struct block_device *,
 226                                 sector_t block, unsigned size, gfp_t gfp);
 227 void invalidate_bh_lrus(void);
 228 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 229 void free_buffer_head(struct buffer_head * bh);
 230 void unlock_buffer(struct buffer_head *bh);
 231 void __lock_buffer(struct buffer_head *bh);
 232 void ll_rw_block(int, int, struct buffer_head * bh[]);
 233 int sync_dirty_buffer(struct buffer_head *bh);
 234 int __sync_dirty_buffer(struct buffer_head *bh, int rw);
 235 void write_dirty_buffer(struct buffer_head *bh, int rw);
 236 int _submit_bh(int rw, struct buffer_head *bh, unsigned long bio_flags);
 237 int submit_bh(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
 243 extern int buffer_heads_over_limit;
 244
 245 /*
 246  * Generic address_space_operations implementations for buffer_head-backed
 247  * address_spaces.
 248  */
 249 void block_invalidatepage(struct page *page, unsigned int offset,
 250                           unsigned int length);
 251 int block_write_full_page(struct page *page, get_block_t *get_block,
 252                                 struct writeback_control *wbc);
 253 int block_read_full_page(struct page*, get_block_t*);
 254 int block_is_partially_uptodate(struct page *page, unsigned long from,
 255                                 unsigned long count);
 256 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
 257                 unsigned flags, struct page **pagep, get_block_t *get_block);
 258 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
 259                 get_block_t *get_block);
 260 int block_write_end(struct file *, struct address_space *,
 261                                 loff_t, unsigned, unsigned,
 262                                 struct page *, void *);
 263 int generic_write_end(struct file *, struct address_space *,
 264                                 loff_t, unsigned, unsigned,
 265                                 struct page *, void *);
 266 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to);
 267 int cont_write_begin(struct file *, struct address_space *, loff_t,
 268                         unsigned, unsigned, struct page **, void **,
 269                         get_block_t *, loff_t *);
 270 int generic_cont_expand_simple(struct inode *inode, loff_t size);
 271 int block_commit_write(struct page *page, unsigned from, unsigned to);
 272 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 273                                 get_block_t get_block);
 274 /* Convert errno to return value from ->page_mkwrite() call */
 275 static inline int block_page_mkwrite_return(int err)
 276 {
 277         if (err == 0)
 278                 return VM_FAULT_LOCKED;
 279         if (err == -EFAULT)
 280                 return VM_FAULT_NOPAGE;
 281         if (err == -ENOMEM)
 282                 return VM_FAULT_OOM;
 283         if (err == -EAGAIN)
 284                 return VM_FAULT_RETRY;
 285         /* -ENOSPC, -EDQUOT, -EIO ... */
 286         return VM_FAULT_SIGBUS;
 287 }
 288 sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);
 289 int block_truncate_page(struct address_space *, loff_t, get_block_t *);
 290 int nobh_write_begin(struct address_space *, loff_t, unsigned, unsigned,
 291                                 struct page **, void **, get_block_t*);
 292 int nobh_write_end(struct file *, struct address_space *,
 293                                 loff_t, unsigned, unsigned,
 294                                 struct page *, void *);
 295 int nobh_truncate_page(struct address_space *, loff_t, get_block_t *);
 296 int nobh_writepage(struct page *page, get_block_t *get_block,
 297                         struct writeback_control *wbc);
 298
 299 void buffer_init(void);
 300
 301 /*
 302  * inline definitions
 303  */
 304
 305 static inline void attach_page_buffers(struct page *page,
 306                 struct buffer_head *head)
 307 {
 308         page_cache_get(page);
 309         SetPagePrivate(page);
 310         set_page_private(page, (unsigned long)head);
 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 struct buffer_head *
 355 sb_getblk(struct super_block *sb, sector_t block)
 356 {
 357         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);
 358 }
 359
 360
 361 static inline struct buffer_head *
 362 sb_getblk_gfp(struct super_block *sb, sector_t block, gfp_t gfp)
 363 {
 364         return __getblk_gfp(sb->s_bdev, block, sb->s_blocksize, gfp);
 365 }
 366
 367 static inline struct buffer_head *
 368 sb_find_get_block(struct super_block *sb, sector_t block)
 369 {
 370         return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
 371 }
 372
 373 static inline void
 374 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
 375 {
 376         set_buffer_mapped(bh);
 377         bh->b_bdev = sb->s_bdev;
 378         bh->b_blocknr = block;
 379         bh->b_size = sb->s_blocksize;
 380 }
 381
 382 static inline void wait_on_buffer(struct buffer_head *bh)
 383 {
 384         might_sleep();
 385         if (buffer_locked(bh))
 386                 __wait_on_buffer(bh);
 387 }
 388
 389 static inline int trylock_buffer(struct buffer_head *bh)
 390 {
 391         return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
 392 }
 393
 394 static inline void lock_buffer(struct buffer_head *bh)
 395 {
 396         might_sleep();
 397         if (!trylock_buffer(bh))
 398                 __lock_buffer(bh);
 399 }
 400
 401 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
 402                                                    sector_t block,
 403                                                    unsigned size)
 404 {
 405         return __getblk_gfp(bdev, block, size, 0);
 406 }
 407
 408 static inline struct buffer_head *__getblk(struct block_device *bdev,
 409                                            sector_t block,
 410                                            unsigned size)
 411 {
 412         return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
 413 }
 414
 415 /**
 416  *  __bread() - reads a specified block and returns the bh
 417  *  @bdev: the block_device to read from
 418  *  @block: number of block
 419  *  @size: size (in bytes) to read
 420  *
 421  *  Reads a specified block, and returns buffer head that contains it.
 422  *  The page cache is allocated from movable area so that it can be migrated.
 423  *  It returns NULL if the block was unreadable.
 424  */
 425 static inline struct buffer_head *
 426 __bread(struct block_device *bdev, sector_t block, unsigned size)
 427 {
 428         return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
 429 }
 430
 431 extern int __set_page_dirty_buffers(struct page *page);
 432
 433 #else /* CONFIG_BLOCK */
 434
 435 static inline void buffer_init(void) {}
 436 static inline int try_to_free_buffers(struct page *page) { return 1; }
 437 static inline int inode_has_buffers(struct inode *inode) { return 0; }
 438 static inline void invalidate_inode_buffers(struct inode *inode) {}
 439 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 440 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
 441
 442 #endif /* CONFIG_BLOCK */
 443 #endif /* _LINUX_BUFFER_HEAD_H */