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 defined(CONFIG_JBD) || defined(CONFIG_JBD_MODULE) || \
  81         defined(CONFIG_JBD2) || defined(CONFIG_JBD2_MODULE)
  82         spinlock_t b_state_lock;
  83         spinlock_t b_journal_head_lock;
  84 #endif
  85 #endif
  86 };
  87
  88 static inline unsigned long bh_uptodate_lock_irqsave(struct buffer_head *bh)
  89 {
  90         unsigned long flags;
  91
  92 #ifndef CONFIG_PREEMPT_RT_BASE
  93         local_irq_save(flags);
  94         bit_spin_lock(BH_Uptodate_Lock, &bh->b_state);
  95 #else
  96         spin_lock_irqsave(&bh->b_uptodate_lock, flags);
  97 #endif
  98         return flags;
  99 }
 100
 101 static inline void
 102 bh_uptodate_unlock_irqrestore(struct buffer_head *bh, unsigned long flags)
 103 {
 104 #ifndef CONFIG_PREEMPT_RT_BASE
 105         bit_spin_unlock(BH_Uptodate_Lock, &bh->b_state);
 106         local_irq_restore(flags);
 107 #else
 108         spin_unlock_irqrestore(&bh->b_uptodate_lock, flags);
 109 #endif
 110 }
 111
 112 static inline void buffer_head_init_locks(struct buffer_head *bh)
 113 {
 114 #ifdef CONFIG_PREEMPT_RT_BASE
 115         spin_lock_init(&bh->b_uptodate_lock);
 116 #if defined(CONFIG_JBD) || defined(CONFIG_JBD_MODULE) || \
 117         defined(CONFIG_JBD2) || defined(CONFIG_JBD2_MODULE)
 118         spin_lock_init(&bh->b_state_lock);
 119         spin_lock_init(&bh->b_journal_head_lock);
 120 #endif
 121 #endif
 122 }
 123
 124 /*
 125  * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()
 126  * and buffer_foo() functions.
 127  */
 128 #define BUFFER_FNS(bit, name)                                           \
 129 static inline void set_buffer_##name(struct buffer_head *bh)            \
 130 {                                                                       \
 131         set_bit(BH_##bit, &(bh)->b_state);                              \
 132 }                                                                       \
 133 static inline void clear_buffer_##name(struct buffer_head *bh)          \
 134 {                                                                       \
 135         clear_bit(BH_##bit, &(bh)->b_state);                            \
 136 }                                                                       \
 137 static inline int buffer_##name(const struct buffer_head *bh)           \
 138 {                                                                       \
 139         return test_bit(BH_##bit, &(bh)->b_state);                      \
 140 }
 141
 142 /*
 143  * test_set_buffer_foo() and test_clear_buffer_foo()
 144  */
 145 #define TAS_BUFFER_FNS(bit, name)                                       \
 146 static inline int test_set_buffer_##name(struct buffer_head *bh)        \
 147 {                                                                       \
 148         return test_and_set_bit(BH_##bit, &(bh)->b_state);              \
 149 }                                                                       \
 150 static inline int test_clear_buffer_##name(struct buffer_head *bh)      \
 151 {                                                                       \
 152         return test_and_clear_bit(BH_##bit, &(bh)->b_state);            \
 153 }                                                                       \
 154
 155 /*
 156  * Emit the buffer bitops functions.   Note that there are also functions
 157  * of the form "mark_buffer_foo()".  These are higher-level functions which
 158  * do something in addition to setting a b_state bit.
 159  */
 160 BUFFER_FNS(Uptodate, uptodate)
 161 BUFFER_FNS(Dirty, dirty)
 162 TAS_BUFFER_FNS(Dirty, dirty)
 163 BUFFER_FNS(Lock, locked)
 164 BUFFER_FNS(Req, req)
 165 TAS_BUFFER_FNS(Req, req)
 166 BUFFER_FNS(Mapped, mapped)
 167 BUFFER_FNS(New, new)
 168 BUFFER_FNS(Async_Read, async_read)
 169 BUFFER_FNS(Async_Write, async_write)
 170 BUFFER_FNS(Delay, delay)
 171 BUFFER_FNS(Boundary, boundary)
 172 BUFFER_FNS(Write_EIO, write_io_error)
 173 BUFFER_FNS(Unwritten, unwritten)
 174 BUFFER_FNS(Meta, meta)
 175 BUFFER_FNS(Prio, prio)
 176 BUFFER_FNS(Defer_Completion, defer_completion)
 177
 178 #define bh_offset(bh)           ((unsigned long)(bh)->b_data & ~PAGE_MASK)
 179
 180 /* If we *know* page->private refers to buffer_heads */
 181 #define page_buffers(page)                                      \
 182         ({                                                      \
 183                 BUG_ON(!PagePrivate(page));                     \
 184                 ((struct buffer_head *)page_private(page));     \
 185         })
 186 #define page_has_buffers(page)  PagePrivate(page)
 187
 188 void buffer_check_dirty_writeback(struct page *page,
 189                                      bool *dirty, bool *writeback);
 190
 191 /*
 192  * Declarations
 193  */
 194
 195 void mark_buffer_dirty(struct buffer_head *bh);
 196 void init_buffer(struct buffer_head *, bh_end_io_t *, void *);
 197 void touch_buffer(struct buffer_head *bh);
 198 void set_bh_page(struct buffer_head *bh,
 199                 struct page *page, unsigned long offset);
 200 int try_to_free_buffers(struct page *);
 201 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
 202                 int retry);
 203 void create_empty_buffers(struct page *, unsigned long,
 204                         unsigned long b_state);
 205 void end_buffer_read_sync(struct buffer_head *bh, int uptodate);
 206 void end_buffer_write_sync(struct buffer_head *bh, int uptodate);
 207 void end_buffer_async_write(struct buffer_head *bh, int uptodate);
 208
 209 /* Things to do with buffers at mapping->private_list */
 210 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);
 211 int inode_has_buffers(struct inode *);
 212 void invalidate_inode_buffers(struct inode *);
 213 int remove_inode_buffers(struct inode *inode);
 214 int sync_mapping_buffers(struct address_space *mapping);
 215 void unmap_underlying_metadata(struct block_device *bdev, sector_t block);
 216
 217 void mark_buffer_async_write(struct buffer_head *bh);
 218 void __wait_on_buffer(struct buffer_head *);
 219 wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);
 220 struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,
 221                         unsigned size);
 222 struct buffer_head *__getblk_gfp(struct block_device *bdev, sector_t block,
 223                                   unsigned size, gfp_t gfp);
 224 void __brelse(struct buffer_head *);
 225 void __bforget(struct buffer_head *);
 226 void __breadahead(struct block_device *, sector_t block, unsigned int size);
 227 struct buffer_head *__bread_gfp(struct block_device *,
 228                                 sector_t block, unsigned size, gfp_t gfp);
 229 void invalidate_bh_lrus(void);
 230 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 231 void free_buffer_head(struct buffer_head * bh);
 232 void unlock_buffer(struct buffer_head *bh);
 233 void __lock_buffer(struct buffer_head *bh);
 234 void ll_rw_block(int, int, struct buffer_head * bh[]);
 235 int sync_dirty_buffer(struct buffer_head *bh);
 236 int __sync_dirty_buffer(struct buffer_head *bh, int rw);
 237 void write_dirty_buffer(struct buffer_head *bh, int rw);
 238 int _submit_bh(int rw, struct buffer_head *bh, unsigned long bio_flags);
 239 int submit_bh(int, struct buffer_head *);
 240 void write_boundary_block(struct block_device *bdev,
 241                         sector_t bblock, unsigned blocksize);
 242 int bh_uptodate_or_lock(struct buffer_head *bh);
 243 int bh_submit_read(struct buffer_head *bh);
 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_read_full_page(struct page*, get_block_t*);
 256 int block_is_partially_uptodate(struct page *page, unsigned long from,
 257                                 unsigned long count);
 258 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
 259                 unsigned flags, struct page **pagep, get_block_t *get_block);
 260 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
 261                 get_block_t *get_block);
 262 int block_write_end(struct file *, struct address_space *,
 263                                 loff_t, unsigned, unsigned,
 264                                 struct page *, void *);
 265 int generic_write_end(struct file *, struct address_space *,
 266                                 loff_t, unsigned, unsigned,
 267                                 struct page *, void *);
 268 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to);
 269 int cont_write_begin(struct file *, struct address_space *, loff_t,
 270                         unsigned, unsigned, struct page **, void **,
 271                         get_block_t *, loff_t *);
 272 int generic_cont_expand_simple(struct inode *inode, loff_t size);
 273 int block_commit_write(struct page *page, unsigned from, unsigned to);
 274 int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 275                                 get_block_t get_block);
 276 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 277                                 get_block_t get_block);
 278 /* Convert errno to return value from ->page_mkwrite() call */
 279 static inline int block_page_mkwrite_return(int err)
 280 {
 281         if (err == 0)
 282                 return VM_FAULT_LOCKED;
 283         if (err == -EFAULT)
 284                 return VM_FAULT_NOPAGE;
 285         if (err == -ENOMEM)
 286                 return VM_FAULT_OOM;
 287         if (err == -EAGAIN)
 288                 return VM_FAULT_RETRY;
 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         page_cache_get(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 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 static inline struct buffer_head *
 365 sb_find_get_block(struct super_block *sb, sector_t block)
 366 {
 367         return __find_get_block(sb->s_bdev, block, sb->s_blocksize);
 368 }
 369
 370 static inline void
 371 map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)
 372 {
 373         set_buffer_mapped(bh);
 374         bh->b_bdev = sb->s_bdev;
 375         bh->b_blocknr = block;
 376         bh->b_size = sb->s_blocksize;
 377 }
 378
 379 static inline void wait_on_buffer(struct buffer_head *bh)
 380 {
 381         might_sleep();
 382         if (buffer_locked(bh))
 383                 __wait_on_buffer(bh);
 384 }
 385
 386 static inline int trylock_buffer(struct buffer_head *bh)
 387 {
 388         return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));
 389 }
 390
 391 static inline void lock_buffer(struct buffer_head *bh)
 392 {
 393         might_sleep();
 394         if (!trylock_buffer(bh))
 395                 __lock_buffer(bh);
 396 }
 397
 398 static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,
 399                                                    sector_t block,
 400                                                    unsigned size)
 401 {
 402         return __getblk_gfp(bdev, block, size, 0);
 403 }
 404
 405 static inline struct buffer_head *__getblk(struct block_device *bdev,
 406                                            sector_t block,
 407                                            unsigned size)
 408 {
 409         return __getblk_gfp(bdev, block, size, __GFP_MOVABLE);
 410 }
 411
 412 /**
 413  *  __bread() - reads a specified block and returns the bh
 414  *  @bdev: the block_device to read from
 415  *  @block: number of block
 416  *  @size: size (in bytes) to read
 417  *
 418  *  Reads a specified block, and returns buffer head that contains it.
 419  *  The page cache is allocated from movable area so that it can be migrated.
 420  *  It returns NULL if the block was unreadable.
 421  */
 422 static inline struct buffer_head *
 423 __bread(struct block_device *bdev, sector_t block, unsigned size)
 424 {
 425         return __bread_gfp(bdev, block, size, __GFP_MOVABLE);
 426 }
 427
 428 extern int __set_page_dirty_buffers(struct page *page);
 429
 430 #else /* CONFIG_BLOCK */
 431
 432 static inline void buffer_init(void) {}
 433 static inline int try_to_free_buffers(struct page *page) { return 1; }
 434 static inline int inode_has_buffers(struct inode *inode) { return 0; }
 435 static inline void invalidate_inode_buffers(struct inode *inode) {}
 436 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 437 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
 438
 439 #endif /* CONFIG_BLOCK */
 440 #endif /* _LINUX_BUFFER_HEAD_H */