1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef LINUX_IOMAP_H 3 #define LINUX_IOMAP_H 1 4 5 #include <linux/atomic.h> 6 #include <linux/bitmap.h> 7 #include <linux/blk_types.h> 8 #include <linux/mm.h> 9 #include <linux/types.h> 10 #include <linux/mm_types.h> 11 #include <linux/blkdev.h> 12 13 struct address_space; 14 struct fiemap_extent_info; 15 struct inode; 16 struct iomap_iter; 17 struct iomap_dio; 18 struct iomap_writepage_ctx; 19 struct iov_iter; 20 struct kiocb; 21 struct page; 22 struct vm_area_struct; 23 struct vm_fault; 24 25 /* 26 * Types of block ranges for iomap mappings: 27 */ 28 #define IOMAP_HOLE 0 /* no blocks allocated, need allocation */ 29 #define IOMAP_DELALLOC 1 /* delayed allocation blocks */ 30 #define IOMAP_MAPPED 2 /* blocks allocated at @addr */ 31 #define IOMAP_UNWRITTEN 3 /* blocks allocated at @addr in unwritten state */ 32 #define IOMAP_INLINE 4 /* data inline in the inode */ 33 34 /* 35 * Flags reported by the file system from iomap_begin: 36 * 37 * IOMAP_F_NEW indicates that the blocks have been newly allocated and need 38 * zeroing for areas that no data is copied to. 39 * 40 * IOMAP_F_DIRTY indicates the inode has uncommitted metadata needed to access 41 * written data and requires fdatasync to commit them to persistent storage. 42 * This needs to take into account metadata changes that *may* be made at IO 43 * completion, such as file size updates from direct IO. 44 * 45 * IOMAP_F_SHARED indicates that the blocks are shared, and will need to be 46 * unshared as part a write. 47 * 48 * IOMAP_F_MERGED indicates that the iomap contains the merge of multiple block 49 * mappings. 50 * 51 * IOMAP_F_BUFFER_HEAD indicates that the file system requires the use of 52 * buffer heads for this mapping. 53 * 54 * IOMAP_F_XATTR indicates that the iomap is for an extended attribute extent 55 * rather than a file data extent. 56 */ 57 #define IOMAP_F_NEW (1U << 0) 58 #define IOMAP_F_DIRTY (1U << 1) 59 #define IOMAP_F_SHARED (1U << 2) 60 #define IOMAP_F_MERGED (1U << 3) 61 #ifdef CONFIG_BUFFER_HEAD 62 #define IOMAP_F_BUFFER_HEAD (1U << 4) 63 #else 64 #define IOMAP_F_BUFFER_HEAD 0 65 #endif /* CONFIG_BUFFER_HEAD */ 66 #define IOMAP_F_XATTR (1U << 5) 67 68 /* 69 * Flags set by the core iomap code during operations: 70 * 71 * IOMAP_F_SIZE_CHANGED indicates to the iomap_end method that the file size 72 * has changed as the result of this write operation. 73 * 74 * IOMAP_F_STALE indicates that the iomap is not valid any longer and the file 75 * range it covers needs to be remapped by the high level before the operation 76 * can proceed. 77 */ 78 #define IOMAP_F_SIZE_CHANGED (1U << 8) 79 #define IOMAP_F_STALE (1U << 9) 80 81 /* 82 * Flags from 0x1000 up are for file system specific usage: 83 */ 84 #define IOMAP_F_PRIVATE (1U << 12) 85 86 87 /* 88 * Magic value for addr: 89 */ 90 #define IOMAP_NULL_ADDR -1ULL /* addr is not valid */ 91 92 struct iomap_folio_ops; 93 94 struct iomap { 95 u64 addr; /* disk offset of mapping, bytes */ 96 loff_t offset; /* file offset of mapping, bytes */ 97 u64 length; /* length of mapping, bytes */ 98 u16 type; /* type of mapping */ 99 u16 flags; /* flags for mapping */ 100 struct block_device *bdev; /* block device for I/O */ 101 struct dax_device *dax_dev; /* dax_dev for dax operations */ 102 void *inline_data; 103 void *private; /* filesystem private */ 104 const struct iomap_folio_ops *folio_ops; 105 u64 validity_cookie; /* used with .iomap_valid() */ 106 }; 107 iomap_sector(const struct iomap * iomap,loff_t pos)108 static inline sector_t iomap_sector(const struct iomap *iomap, loff_t pos) 109 { 110 return (iomap->addr + pos - iomap->offset) >> SECTOR_SHIFT; 111 } 112 113 /* 114 * Returns the inline data pointer for logical offset @pos. 115 */ iomap_inline_data(const struct iomap * iomap,loff_t pos)116 static inline void *iomap_inline_data(const struct iomap *iomap, loff_t pos) 117 { 118 return iomap->inline_data + pos - iomap->offset; 119 } 120 121 /* 122 * Check if the mapping's length is within the valid range for inline data. 123 * This is used to guard against accessing data beyond the page inline_data 124 * points at. 125 */ iomap_inline_data_valid(const struct iomap * iomap)126 static inline bool iomap_inline_data_valid(const struct iomap *iomap) 127 { 128 return iomap->length <= PAGE_SIZE - offset_in_page(iomap->inline_data); 129 } 130 131 /* 132 * When a filesystem sets folio_ops in an iomap mapping it returns, get_folio 133 * and put_folio will be called for each folio written to. This only applies 134 * to buffered writes as unbuffered writes will not typically have folios 135 * associated with them. 136 * 137 * When get_folio succeeds, put_folio will always be called to do any 138 * cleanup work necessary. put_folio is responsible for unlocking and putting 139 * @folio. 140 */ 141 struct iomap_folio_ops { 142 struct folio *(*get_folio)(struct iomap_iter *iter, loff_t pos, 143 unsigned len); 144 void (*put_folio)(struct inode *inode, loff_t pos, unsigned copied, 145 struct folio *folio); 146 147 /* 148 * Check that the cached iomap still maps correctly to the filesystem's 149 * internal extent map. FS internal extent maps can change while iomap 150 * is iterating a cached iomap, so this hook allows iomap to detect that 151 * the iomap needs to be refreshed during a long running write 152 * operation. 153 * 154 * The filesystem can store internal state (e.g. a sequence number) in 155 * iomap->validity_cookie when the iomap is first mapped to be able to 156 * detect changes between mapping time and whenever .iomap_valid() is 157 * called. 158 * 159 * This is called with the folio over the specified file position held 160 * locked by the iomap code. 161 */ 162 bool (*iomap_valid)(struct inode *inode, const struct iomap *iomap); 163 }; 164 165 /* 166 * Flags for iomap_begin / iomap_end. No flag implies a read. 167 */ 168 #define IOMAP_WRITE (1 << 0) /* writing, must allocate blocks */ 169 #define IOMAP_ZERO (1 << 1) /* zeroing operation, may skip holes */ 170 #define IOMAP_REPORT (1 << 2) /* report extent status, e.g. FIEMAP */ 171 #define IOMAP_FAULT (1 << 3) /* mapping for page fault */ 172 #define IOMAP_DIRECT (1 << 4) /* direct I/O */ 173 #define IOMAP_NOWAIT (1 << 5) /* do not block */ 174 #define IOMAP_OVERWRITE_ONLY (1 << 6) /* only pure overwrites allowed */ 175 #define IOMAP_UNSHARE (1 << 7) /* unshare_file_range */ 176 #ifdef CONFIG_FS_DAX 177 #define IOMAP_DAX (1 << 8) /* DAX mapping */ 178 #else 179 #define IOMAP_DAX 0 180 #endif /* CONFIG_FS_DAX */ 181 182 struct iomap_ops { 183 /* 184 * Return the existing mapping at pos, or reserve space starting at 185 * pos for up to length, as long as we can do it as a single mapping. 186 * The actual length is returned in iomap->length. 187 */ 188 int (*iomap_begin)(struct inode *inode, loff_t pos, loff_t length, 189 unsigned flags, struct iomap *iomap, 190 struct iomap *srcmap); 191 192 /* 193 * Commit and/or unreserve space previous allocated using iomap_begin. 194 * Written indicates the length of the successful write operation which 195 * needs to be commited, while the rest needs to be unreserved. 196 * Written might be zero if no data was written. 197 */ 198 int (*iomap_end)(struct inode *inode, loff_t pos, loff_t length, 199 ssize_t written, unsigned flags, struct iomap *iomap); 200 }; 201 202 /** 203 * struct iomap_iter - Iterate through a range of a file 204 * @inode: Set at the start of the iteration and should not change. 205 * @pos: The current file position we are operating on. It is updated by 206 * calls to iomap_iter(). Treat as read-only in the body. 207 * @len: The remaining length of the file segment we're operating on. 208 * It is updated at the same time as @pos. 209 * @processed: The number of bytes processed by the body in the most recent 210 * iteration, or a negative errno. 0 causes the iteration to stop. 211 * @flags: Zero or more of the iomap_begin flags above. 212 * @iomap: Map describing the I/O iteration 213 * @srcmap: Source map for COW operations 214 */ 215 struct iomap_iter { 216 struct inode *inode; 217 loff_t pos; 218 u64 len; 219 s64 processed; 220 unsigned flags; 221 struct iomap iomap; 222 struct iomap srcmap; 223 void *private; 224 }; 225 226 int iomap_iter(struct iomap_iter *iter, const struct iomap_ops *ops); 227 228 /** 229 * iomap_length - length of the current iomap iteration 230 * @iter: iteration structure 231 * 232 * Returns the length that the operation applies to for the current iteration. 233 */ iomap_length(const struct iomap_iter * iter)234 static inline u64 iomap_length(const struct iomap_iter *iter) 235 { 236 u64 end = iter->iomap.offset + iter->iomap.length; 237 238 if (iter->srcmap.type != IOMAP_HOLE) 239 end = min(end, iter->srcmap.offset + iter->srcmap.length); 240 return min(iter->len, end - iter->pos); 241 } 242 243 /** 244 * iomap_iter_srcmap - return the source map for the current iomap iteration 245 * @i: iteration structure 246 * 247 * Write operations on file systems with reflink support might require a 248 * source and a destination map. This function retourns the source map 249 * for a given operation, which may or may no be identical to the destination 250 * map in &i->iomap. 251 */ iomap_iter_srcmap(const struct iomap_iter * i)252 static inline const struct iomap *iomap_iter_srcmap(const struct iomap_iter *i) 253 { 254 if (i->srcmap.type != IOMAP_HOLE) 255 return &i->srcmap; 256 return &i->iomap; 257 } 258 259 /* 260 * Return the file offset for the first unchanged block after a short write. 261 * 262 * If nothing was written, round @pos down to point at the first block in 263 * the range, else round up to include the partially written block. 264 */ iomap_last_written_block(struct inode * inode,loff_t pos,ssize_t written)265 static inline loff_t iomap_last_written_block(struct inode *inode, loff_t pos, 266 ssize_t written) 267 { 268 if (unlikely(!written)) 269 return round_down(pos, i_blocksize(inode)); 270 return round_up(pos + written, i_blocksize(inode)); 271 } 272 273 /* 274 * Check if the range needs to be unshared for a FALLOC_FL_UNSHARE_RANGE 275 * operation. 276 * 277 * Don't bother with blocks that are not shared to start with; or mappings that 278 * cannot be shared, such as inline data, delalloc reservations, holes or 279 * unwritten extents. 280 * 281 * Note that we use srcmap directly instead of iomap_iter_srcmap as unsharing 282 * requires providing a separate source map, and the presence of one is a good 283 * indicator that unsharing is needed, unlike IOMAP_F_SHARED which can be set 284 * for any data that goes into the COW fork for XFS. 285 */ iomap_want_unshare_iter(const struct iomap_iter * iter)286 static inline bool iomap_want_unshare_iter(const struct iomap_iter *iter) 287 { 288 return (iter->iomap.flags & IOMAP_F_SHARED) && 289 iter->srcmap.type == IOMAP_MAPPED; 290 } 291 292 ssize_t iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *from, 293 const struct iomap_ops *ops, void *private); 294 int iomap_read_folio(struct folio *folio, const struct iomap_ops *ops); 295 void iomap_readahead(struct readahead_control *, const struct iomap_ops *ops); 296 bool iomap_is_partially_uptodate(struct folio *, size_t from, size_t count); 297 struct folio *iomap_get_folio(struct iomap_iter *iter, loff_t pos, size_t len); 298 bool iomap_release_folio(struct folio *folio, gfp_t gfp_flags); 299 void iomap_invalidate_folio(struct folio *folio, size_t offset, size_t len); 300 bool iomap_dirty_folio(struct address_space *mapping, struct folio *folio); 301 int iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 302 const struct iomap_ops *ops); 303 int iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, 304 bool *did_zero, const struct iomap_ops *ops); 305 int iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 306 const struct iomap_ops *ops); 307 vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, 308 const struct iomap_ops *ops); 309 310 typedef void (*iomap_punch_t)(struct inode *inode, loff_t offset, loff_t length, 311 struct iomap *iomap); 312 void iomap_write_delalloc_release(struct inode *inode, loff_t start_byte, 313 loff_t end_byte, unsigned flags, struct iomap *iomap, 314 iomap_punch_t punch); 315 316 int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 317 u64 start, u64 len, const struct iomap_ops *ops); 318 loff_t iomap_seek_hole(struct inode *inode, loff_t offset, 319 const struct iomap_ops *ops); 320 loff_t iomap_seek_data(struct inode *inode, loff_t offset, 321 const struct iomap_ops *ops); 322 sector_t iomap_bmap(struct address_space *mapping, sector_t bno, 323 const struct iomap_ops *ops); 324 325 /* 326 * Structure for writeback I/O completions. 327 */ 328 struct iomap_ioend { 329 struct list_head io_list; /* next ioend in chain */ 330 u16 io_type; 331 u16 io_flags; /* IOMAP_F_* */ 332 struct inode *io_inode; /* file being written to */ 333 size_t io_size; /* size of the extent */ 334 loff_t io_offset; /* offset in the file */ 335 sector_t io_sector; /* start sector of ioend */ 336 struct bio io_bio; /* MUST BE LAST! */ 337 }; 338 iomap_ioend_from_bio(struct bio * bio)339 static inline struct iomap_ioend *iomap_ioend_from_bio(struct bio *bio) 340 { 341 return container_of(bio, struct iomap_ioend, io_bio); 342 } 343 344 struct iomap_writeback_ops { 345 /* 346 * Required, maps the blocks so that writeback can be performed on 347 * the range starting at offset. 348 * 349 * Can return arbitrarily large regions, but we need to call into it at 350 * least once per folio to allow the file systems to synchronize with 351 * the write path that could be invalidating mappings. 352 * 353 * An existing mapping from a previous call to this method can be reused 354 * by the file system if it is still valid. 355 */ 356 int (*map_blocks)(struct iomap_writepage_ctx *wpc, struct inode *inode, 357 loff_t offset, unsigned len); 358 359 /* 360 * Optional, allows the file systems to perform actions just before 361 * submitting the bio and/or override the bio end_io handler for complex 362 * operations like copy on write extent manipulation or unwritten extent 363 * conversions. 364 */ 365 int (*prepare_ioend)(struct iomap_ioend *ioend, int status); 366 367 /* 368 * Optional, allows the file system to discard state on a page where 369 * we failed to submit any I/O. 370 */ 371 void (*discard_folio)(struct folio *folio, loff_t pos); 372 }; 373 374 struct iomap_writepage_ctx { 375 struct iomap iomap; 376 struct iomap_ioend *ioend; 377 const struct iomap_writeback_ops *ops; 378 u32 nr_folios; /* folios added to the ioend */ 379 }; 380 381 void iomap_finish_ioends(struct iomap_ioend *ioend, int error); 382 void iomap_ioend_try_merge(struct iomap_ioend *ioend, 383 struct list_head *more_ioends); 384 void iomap_sort_ioends(struct list_head *ioend_list); 385 int iomap_writepages(struct address_space *mapping, 386 struct writeback_control *wbc, struct iomap_writepage_ctx *wpc, 387 const struct iomap_writeback_ops *ops); 388 389 /* 390 * Flags for direct I/O ->end_io: 391 */ 392 #define IOMAP_DIO_UNWRITTEN (1 << 0) /* covers unwritten extent(s) */ 393 #define IOMAP_DIO_COW (1 << 1) /* covers COW extent(s) */ 394 395 struct iomap_dio_ops { 396 int (*end_io)(struct kiocb *iocb, ssize_t size, int error, 397 unsigned flags); 398 void (*submit_io)(const struct iomap_iter *iter, struct bio *bio, 399 loff_t file_offset); 400 401 /* 402 * Filesystems wishing to attach private information to a direct io bio 403 * must provide a ->submit_io method that attaches the additional 404 * information to the bio and changes the ->bi_end_io callback to a 405 * custom function. This function should, at a minimum, perform any 406 * relevant post-processing of the bio and end with a call to 407 * iomap_dio_bio_end_io. 408 */ 409 struct bio_set *bio_set; 410 }; 411 412 /* 413 * Wait for the I/O to complete in iomap_dio_rw even if the kiocb is not 414 * synchronous. 415 */ 416 #define IOMAP_DIO_FORCE_WAIT (1 << 0) 417 418 /* 419 * Do not allocate blocks or zero partial blocks, but instead fall back to 420 * the caller by returning -EAGAIN. Used to optimize direct I/O writes that 421 * are not aligned to the file system block size. 422 */ 423 #define IOMAP_DIO_OVERWRITE_ONLY (1 << 1) 424 425 /* 426 * When a page fault occurs, return a partial synchronous result and allow 427 * the caller to retry the rest of the operation after dealing with the page 428 * fault. 429 */ 430 #define IOMAP_DIO_PARTIAL (1 << 2) 431 432 ssize_t iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, 433 const struct iomap_ops *ops, const struct iomap_dio_ops *dops, 434 unsigned int dio_flags, void *private, size_t done_before); 435 struct iomap_dio *__iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, 436 const struct iomap_ops *ops, const struct iomap_dio_ops *dops, 437 unsigned int dio_flags, void *private, size_t done_before); 438 ssize_t iomap_dio_complete(struct iomap_dio *dio); 439 void iomap_dio_bio_end_io(struct bio *bio); 440 441 #ifdef CONFIG_SWAP 442 struct file; 443 struct swap_info_struct; 444 445 int iomap_swapfile_activate(struct swap_info_struct *sis, 446 struct file *swap_file, sector_t *pagespan, 447 const struct iomap_ops *ops); 448 #else 449 # define iomap_swapfile_activate(sis, swapfile, pagespan, ops) (-EIO) 450 #endif /* CONFIG_SWAP */ 451 452 #endif /* LINUX_IOMAP_H */ 453