536 lines · c
1/* SPDX-License-Identifier: GPL-2.0 */2/*3 * include/linux/buffer_head.h4 *5 * Everything to do with buffer_heads.6 */7 8#ifndef _LINUX_BUFFER_HEAD_H9#define _LINUX_BUFFER_HEAD_H10 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 19enum 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 25 BH_Mapped, /* Has a disk mapping */26 BH_New, /* Disk mapping was newly created by get_block */27 BH_Async_Read, /* Is under end_buffer_async_read I/O */28 BH_Async_Write, /* Is under end_buffer_async_write I/O */29 BH_Delay, /* Buffer is not yet allocated on disk */30 BH_Boundary, /* Block is followed by a discontiguity */31 BH_Write_EIO, /* I/O error on write */32 BH_Unwritten, /* Buffer is allocated on disk but not written */33 BH_Quiet, /* Buffer Error Prinks to be quiet */34 BH_Meta, /* Buffer contains metadata */35 BH_Prio, /* Buffer should be submitted with REQ_PRIO */36 BH_Defer_Completion, /* Defer AIO completion to workqueue */37 38 BH_PrivateStart,/* not a state bit, but the first bit available39 * for private allocation by other entities40 */41};42 43#define MAX_BUF_PER_PAGE (PAGE_SIZE / 512)44 45struct page;46struct buffer_head;47struct address_space;48typedef void (bh_end_io_t)(struct buffer_head *bh, int uptodate);49 50/*51 * Historically, a buffer_head was used to map a single block52 * within a page, and of course as the unit of I/O through the53 * filesystem and block layers. Nowadays the basic I/O unit54 * is the bio, and buffer_heads are used for extracting block55 * mappings (via a get_block_t call), for tracking state within56 * a folio (via a folio_mapping) and for wrapping bio submission57 * for backward compatibility reasons (e.g. submit_bh).58 */59struct buffer_head {60 unsigned long b_state; /* buffer state bitmap (see above) */61 struct buffer_head *b_this_page;/* circular list of page's buffers */62 union {63 struct page *b_page; /* the page this bh is mapped to */64 struct folio *b_folio; /* the folio this bh is mapped to */65 };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 is76 associated with */77 atomic_t b_count; /* users using this buffer_head */78 spinlock_t b_uptodate_lock; /* Used by the first bh in a page, to79 * serialise IO completion of other80 * buffers in the page */81};82 83/*84 * macro tricks to expand the set_buffer_foo(), clear_buffer_foo()85 * and buffer_foo() functions.86 * To avoid reset buffer flags that are already set, because that causes87 * a costly cache line transition, check the flag first.88 */89#define BUFFER_FNS(bit, name) \90static __always_inline void set_buffer_##name(struct buffer_head *bh) \91{ \92 if (!test_bit(BH_##bit, &(bh)->b_state)) \93 set_bit(BH_##bit, &(bh)->b_state); \94} \95static __always_inline void clear_buffer_##name(struct buffer_head *bh) \96{ \97 clear_bit(BH_##bit, &(bh)->b_state); \98} \99static __always_inline int buffer_##name(const struct buffer_head *bh) \100{ \101 return test_bit(BH_##bit, &(bh)->b_state); \102}103 104/*105 * test_set_buffer_foo() and test_clear_buffer_foo()106 */107#define TAS_BUFFER_FNS(bit, name) \108static __always_inline int test_set_buffer_##name(struct buffer_head *bh) \109{ \110 return test_and_set_bit(BH_##bit, &(bh)->b_state); \111} \112static __always_inline int test_clear_buffer_##name(struct buffer_head *bh) \113{ \114 return test_and_clear_bit(BH_##bit, &(bh)->b_state); \115} \116 117/*118 * Emit the buffer bitops functions. Note that there are also functions119 * of the form "mark_buffer_foo()". These are higher-level functions which120 * do something in addition to setting a b_state bit.121 */122BUFFER_FNS(Dirty, dirty)123TAS_BUFFER_FNS(Dirty, dirty)124BUFFER_FNS(Lock, locked)125BUFFER_FNS(Req, req)126TAS_BUFFER_FNS(Req, req)127BUFFER_FNS(Mapped, mapped)128BUFFER_FNS(New, new)129BUFFER_FNS(Async_Read, async_read)130BUFFER_FNS(Async_Write, async_write)131BUFFER_FNS(Delay, delay)132BUFFER_FNS(Boundary, boundary)133BUFFER_FNS(Write_EIO, write_io_error)134BUFFER_FNS(Unwritten, unwritten)135BUFFER_FNS(Meta, meta)136BUFFER_FNS(Prio, prio)137BUFFER_FNS(Defer_Completion, defer_completion)138 139static __always_inline void set_buffer_uptodate(struct buffer_head *bh)140{141 /*142 * If somebody else already set this uptodate, they will143 * have done the memory barrier, and a reader will thus144 * see *some* valid buffer state.145 *146 * Any other serialization (with IO errors or whatever that147 * might clear the bit) has to come from other state (eg BH_Lock).148 */149 if (test_bit(BH_Uptodate, &bh->b_state))150 return;151 152 /*153 * make it consistent with folio_mark_uptodate154 * pairs with smp_load_acquire in buffer_uptodate155 */156 smp_mb__before_atomic();157 set_bit(BH_Uptodate, &bh->b_state);158}159 160static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)161{162 clear_bit(BH_Uptodate, &bh->b_state);163}164 165static __always_inline int buffer_uptodate(const struct buffer_head *bh)166{167 /*168 * make it consistent with folio_test_uptodate169 * pairs with smp_mb__before_atomic in set_buffer_uptodate170 */171 return test_bit_acquire(BH_Uptodate, &bh->b_state);172}173 174static inline unsigned long bh_offset(const struct buffer_head *bh)175{176 return (unsigned long)(bh)->b_data & (page_size(bh->b_page) - 1);177}178 179/* If we *know* page->private refers to buffer_heads */180#define page_buffers(page) \181 ({ \182 BUG_ON(!PagePrivate(page)); \183 ((struct buffer_head *)page_private(page)); \184 })185#define page_has_buffers(page) PagePrivate(page)186#define folio_buffers(folio) folio_get_private(folio)187 188void buffer_check_dirty_writeback(struct folio *folio,189 bool *dirty, bool *writeback);190 191/*192 * Declarations193 */194 195void mark_buffer_dirty(struct buffer_head *bh);196void mark_buffer_write_io_error(struct buffer_head *bh);197void touch_buffer(struct buffer_head *bh);198void folio_set_bh(struct buffer_head *bh, struct folio *folio,199 unsigned long offset);200struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size,201 gfp_t gfp);202struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size);203struct buffer_head *create_empty_buffers(struct folio *folio,204 unsigned long blocksize, unsigned long b_state);205void end_buffer_read_sync(struct buffer_head *bh, int uptodate);206void end_buffer_write_sync(struct buffer_head *bh, int uptodate);207 208/* Things to do with buffers at mapping->private_list */209void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode);210int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end,211 bool datasync);212int generic_buffers_fsync(struct file *file, loff_t start, loff_t end,213 bool datasync);214void clean_bdev_aliases(struct block_device *bdev, sector_t block,215 sector_t len);216static inline void clean_bdev_bh_alias(struct buffer_head *bh)217{218 clean_bdev_aliases(bh->b_bdev, bh->b_blocknr, 1);219}220 221void mark_buffer_async_write(struct buffer_head *bh);222void __wait_on_buffer(struct buffer_head *);223wait_queue_head_t *bh_waitq_head(struct buffer_head *bh);224struct buffer_head *__find_get_block(struct block_device *bdev, sector_t block,225 unsigned size);226struct buffer_head *bdev_getblk(struct block_device *bdev, sector_t block,227 unsigned size, gfp_t gfp);228void __brelse(struct buffer_head *);229void __bforget(struct buffer_head *);230void __breadahead(struct block_device *, sector_t block, unsigned int size);231struct buffer_head *__bread_gfp(struct block_device *,232 sector_t block, unsigned size, gfp_t gfp);233struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);234void free_buffer_head(struct buffer_head * bh);235void unlock_buffer(struct buffer_head *bh);236void __lock_buffer(struct buffer_head *bh);237int sync_dirty_buffer(struct buffer_head *bh);238int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);239void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags);240void submit_bh(blk_opf_t, struct buffer_head *);241void write_boundary_block(struct block_device *bdev,242 sector_t bblock, unsigned blocksize);243int bh_uptodate_or_lock(struct buffer_head *bh);244int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait);245void __bh_read_batch(int nr, struct buffer_head *bhs[],246 blk_opf_t op_flags, bool force_lock);247 248/*249 * Generic address_space_operations implementations for buffer_head-backed250 * address_spaces.251 */252void block_invalidate_folio(struct folio *folio, size_t offset, size_t length);253int block_write_full_folio(struct folio *folio, struct writeback_control *wbc,254 void *get_block);255int __block_write_full_folio(struct inode *inode, struct folio *folio,256 get_block_t *get_block, struct writeback_control *wbc);257int block_read_full_folio(struct folio *, get_block_t *);258bool block_is_partially_uptodate(struct folio *, size_t from, size_t count);259int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,260 struct folio **foliop, get_block_t *get_block);261int __block_write_begin(struct folio *folio, loff_t pos, unsigned len,262 get_block_t *get_block);263int block_write_end(struct file *, struct address_space *,264 loff_t, unsigned len, unsigned copied,265 struct folio *, void *);266int generic_write_end(struct file *, struct address_space *,267 loff_t, unsigned len, unsigned copied,268 struct folio *, void *);269void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to);270int cont_write_begin(struct file *, struct address_space *, loff_t,271 unsigned, struct folio **, void **,272 get_block_t *, loff_t *);273int generic_cont_expand_simple(struct inode *inode, loff_t size);274void block_commit_write(struct page *page, unsigned int from, unsigned int to);275int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,276 get_block_t get_block);277sector_t generic_block_bmap(struct address_space *, sector_t, get_block_t *);278int block_truncate_page(struct address_space *, loff_t, get_block_t *);279 280#ifdef CONFIG_MIGRATION281extern int buffer_migrate_folio(struct address_space *,282 struct folio *dst, struct folio *src, enum migrate_mode);283extern int buffer_migrate_folio_norefs(struct address_space *,284 struct folio *dst, struct folio *src, enum migrate_mode);285#else286#define buffer_migrate_folio NULL287#define buffer_migrate_folio_norefs NULL288#endif289 290/*291 * inline definitions292 */293 294static inline void get_bh(struct buffer_head *bh)295{296 atomic_inc(&bh->b_count);297}298 299static inline void put_bh(struct buffer_head *bh)300{301 smp_mb__before_atomic();302 atomic_dec(&bh->b_count);303}304 305/**306 * brelse - Release a buffer.307 * @bh: The buffer to release.308 *309 * Decrement a buffer_head's reference count. If @bh is NULL, this310 * function is a no-op.311 *312 * If all buffers on a folio have zero reference count, are clean313 * and unlocked, and if the folio is unlocked and not under writeback314 * then try_to_free_buffers() may strip the buffers from the folio in315 * preparation for freeing it (sometimes, rarely, buffers are removed316 * from a folio but it ends up not being freed, and buffers may later317 * be reattached).318 *319 * Context: Any context.320 */321static inline void brelse(struct buffer_head *bh)322{323 if (bh)324 __brelse(bh);325}326 327/**328 * bforget - Discard any dirty data in a buffer.329 * @bh: The buffer to forget.330 *331 * Call this function instead of brelse() if the data written to a buffer332 * no longer needs to be written back. It will clear the buffer's dirty333 * flag so writeback of this buffer will be skipped.334 *335 * Context: Any context.336 */337static inline void bforget(struct buffer_head *bh)338{339 if (bh)340 __bforget(bh);341}342 343static inline struct buffer_head *344sb_bread(struct super_block *sb, sector_t block)345{346 return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, __GFP_MOVABLE);347}348 349static inline struct buffer_head *350sb_bread_unmovable(struct super_block *sb, sector_t block)351{352 return __bread_gfp(sb->s_bdev, block, sb->s_blocksize, 0);353}354 355static inline void356sb_breadahead(struct super_block *sb, sector_t block)357{358 __breadahead(sb->s_bdev, block, sb->s_blocksize);359}360 361static inline struct buffer_head *getblk_unmovable(struct block_device *bdev,362 sector_t block, unsigned size)363{364 gfp_t gfp;365 366 gfp = mapping_gfp_constraint(bdev->bd_mapping, ~__GFP_FS);367 gfp |= __GFP_NOFAIL;368 369 return bdev_getblk(bdev, block, size, gfp);370}371 372static inline struct buffer_head *__getblk(struct block_device *bdev,373 sector_t block, unsigned size)374{375 gfp_t gfp;376 377 gfp = mapping_gfp_constraint(bdev->bd_mapping, ~__GFP_FS);378 gfp |= __GFP_MOVABLE | __GFP_NOFAIL;379 380 return bdev_getblk(bdev, block, size, gfp);381}382 383static inline struct buffer_head *sb_getblk(struct super_block *sb,384 sector_t block)385{386 return __getblk(sb->s_bdev, block, sb->s_blocksize);387}388 389static inline struct buffer_head *sb_getblk_gfp(struct super_block *sb,390 sector_t block, gfp_t gfp)391{392 return bdev_getblk(sb->s_bdev, block, sb->s_blocksize, gfp);393}394 395static inline struct buffer_head *396sb_find_get_block(struct super_block *sb, sector_t block)397{398 return __find_get_block(sb->s_bdev, block, sb->s_blocksize);399}400 401static inline void402map_bh(struct buffer_head *bh, struct super_block *sb, sector_t block)403{404 set_buffer_mapped(bh);405 bh->b_bdev = sb->s_bdev;406 bh->b_blocknr = block;407 bh->b_size = sb->s_blocksize;408}409 410static inline void wait_on_buffer(struct buffer_head *bh)411{412 might_sleep();413 if (buffer_locked(bh))414 __wait_on_buffer(bh);415}416 417static inline int trylock_buffer(struct buffer_head *bh)418{419 return likely(!test_and_set_bit_lock(BH_Lock, &bh->b_state));420}421 422static inline void lock_buffer(struct buffer_head *bh)423{424 might_sleep();425 if (!trylock_buffer(bh))426 __lock_buffer(bh);427}428 429static inline void bh_readahead(struct buffer_head *bh, blk_opf_t op_flags)430{431 if (!buffer_uptodate(bh) && trylock_buffer(bh)) {432 if (!buffer_uptodate(bh))433 __bh_read(bh, op_flags, false);434 else435 unlock_buffer(bh);436 }437}438 439static inline void bh_read_nowait(struct buffer_head *bh, blk_opf_t op_flags)440{441 if (!bh_uptodate_or_lock(bh))442 __bh_read(bh, op_flags, false);443}444 445/* Returns 1 if buffer uptodated, 0 on success, and -EIO on error. */446static inline int bh_read(struct buffer_head *bh, blk_opf_t op_flags)447{448 if (bh_uptodate_or_lock(bh))449 return 1;450 return __bh_read(bh, op_flags, true);451}452 453static inline void bh_read_batch(int nr, struct buffer_head *bhs[])454{455 __bh_read_batch(nr, bhs, 0, true);456}457 458static inline void bh_readahead_batch(int nr, struct buffer_head *bhs[],459 blk_opf_t op_flags)460{461 __bh_read_batch(nr, bhs, op_flags, false);462}463 464/**465 * __bread() - Read a block.466 * @bdev: The block device to read from.467 * @block: Block number in units of block size.468 * @size: The block size of this device in bytes.469 *470 * Read a specified block, and return the buffer head that refers471 * to it. The memory is allocated from the movable area so that it can472 * be migrated. The returned buffer head has its refcount increased.473 * The caller should call brelse() when it has finished with the buffer.474 *475 * Context: May sleep waiting for I/O.476 * Return: NULL if the block was unreadable.477 */478static inline struct buffer_head *__bread(struct block_device *bdev,479 sector_t block, unsigned size)480{481 return __bread_gfp(bdev, block, size, __GFP_MOVABLE);482}483 484/**485 * get_nth_bh - Get a reference on the n'th buffer after this one.486 * @bh: The buffer to start counting from.487 * @count: How many buffers to skip.488 *489 * This is primarily useful for finding the nth buffer in a folio; in490 * that case you pass the head buffer and the byte offset in the folio491 * divided by the block size. It can be used for other purposes, but492 * it will wrap at the end of the folio rather than returning NULL or493 * proceeding to the next folio for you.494 *495 * Return: The requested buffer with an elevated refcount.496 */497static inline __must_check498struct buffer_head *get_nth_bh(struct buffer_head *bh, unsigned int count)499{500 while (count--)501 bh = bh->b_this_page;502 get_bh(bh);503 return bh;504}505 506bool block_dirty_folio(struct address_space *mapping, struct folio *folio);507 508#ifdef CONFIG_BUFFER_HEAD509 510void buffer_init(void);511bool try_to_free_buffers(struct folio *folio);512int inode_has_buffers(struct inode *inode);513void invalidate_inode_buffers(struct inode *inode);514int remove_inode_buffers(struct inode *inode);515int sync_mapping_buffers(struct address_space *mapping);516void invalidate_bh_lrus(void);517void invalidate_bh_lrus_cpu(void);518bool has_bh_in_lru(int cpu, void *dummy);519extern int buffer_heads_over_limit;520 521#else /* CONFIG_BUFFER_HEAD */522 523static inline void buffer_init(void) {}524static inline bool try_to_free_buffers(struct folio *folio) { return true; }525static inline int inode_has_buffers(struct inode *inode) { return 0; }526static inline void invalidate_inode_buffers(struct inode *inode) {}527static inline int remove_inode_buffers(struct inode *inode) { return 1; }528static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }529static inline void invalidate_bh_lrus(void) {}530static inline void invalidate_bh_lrus_cpu(void) {}531static inline bool has_bh_in_lru(int cpu, void *dummy) { return false; }532#define buffer_heads_over_limit 0533 534#endif /* CONFIG_BUFFER_HEAD */535#endif /* _LINUX_BUFFER_HEAD_H */536