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_dio; 17 struct iomap_writepage_ctx; 18 struct iov_iter; 19 struct kiocb; 20 struct page; 21 struct vm_area_struct; 22 struct vm_fault; 23 24 /* 25 * Types of block ranges for iomap mappings: 26 */ 27 #define IOMAP_HOLE 0 /* no blocks allocated, need allocation */ 28 #define IOMAP_DELALLOC 1 /* delayed allocation blocks */ 29 #define IOMAP_MAPPED 2 /* blocks allocated at @addr */ 30 #define IOMAP_UNWRITTEN 3 /* blocks allocated at @addr in unwritten state */ 31 #define IOMAP_INLINE 4 /* data inline in the inode */ 32 33 /* 34 * Flags reported by the file system from iomap_begin: 35 * 36 * IOMAP_F_NEW indicates that the blocks have been newly allocated and need 37 * zeroing for areas that no data is copied to. 38 * 39 * IOMAP_F_DIRTY indicates the inode has uncommitted metadata needed to access 40 * written data and requires fdatasync to commit them to persistent storage. 41 * This needs to take into account metadata changes that *may* be made at IO 42 * completion, such as file size updates from direct IO. 43 * 44 * IOMAP_F_SHARED indicates that the blocks are shared, and will need to be 45 * unshared as part a write. 46 * 47 * IOMAP_F_MERGED indicates that the iomap contains the merge of multiple block 48 * mappings. 49 * 50 * IOMAP_F_BUFFER_HEAD indicates that the file system requires the use of 51 * buffer heads for this mapping. 52 */ 53 #define IOMAP_F_NEW 0x01 54 #define IOMAP_F_DIRTY 0x02 55 #define IOMAP_F_SHARED 0x04 56 #define IOMAP_F_MERGED 0x08 57 #define IOMAP_F_BUFFER_HEAD 0x10 58 #define IOMAP_F_ZONE_APPEND 0x20 59 60 /* 61 * Flags set by the core iomap code during operations: 62 * 63 * IOMAP_F_SIZE_CHANGED indicates to the iomap_end method that the file size 64 * has changed as the result of this write operation. 65 */ 66 #define IOMAP_F_SIZE_CHANGED 0x100 67 68 /* 69 * Flags from 0x1000 up are for file system specific usage: 70 */ 71 #define IOMAP_F_PRIVATE 0x1000 72 73 74 /* 75 * Magic value for addr: 76 */ 77 #define IOMAP_NULL_ADDR -1ULL /* addr is not valid */ 78 79 struct iomap_page_ops; 80 81 struct iomap { 82 u64 addr; /* disk offset of mapping, bytes */ 83 loff_t offset; /* file offset of mapping, bytes */ 84 u64 length; /* length of mapping, bytes */ 85 u16 type; /* type of mapping */ 86 u16 flags; /* flags for mapping */ 87 struct block_device *bdev; /* block device for I/O */ 88 struct dax_device *dax_dev; /* dax_dev for dax operations */ 89 void *inline_data; 90 void *private; /* filesystem private */ 91 const struct iomap_page_ops *page_ops; 92 }; 93 94 static inline sector_t iomap_sector(const struct iomap *iomap, loff_t pos) 95 { 96 return (iomap->addr + pos - iomap->offset) >> SECTOR_SHIFT; 97 } 98 99 /* 100 * Returns the inline data pointer for logical offset @pos. 101 */ 102 static inline void *iomap_inline_data(const struct iomap *iomap, loff_t pos) 103 { 104 return iomap->inline_data + pos - iomap->offset; 105 } 106 107 /* 108 * Check if the mapping's length is within the valid range for inline data. 109 * This is used to guard against accessing data beyond the page inline_data 110 * points at. 111 */ 112 static inline bool iomap_inline_data_valid(const struct iomap *iomap) 113 { 114 return iomap->length <= PAGE_SIZE - offset_in_page(iomap->inline_data); 115 } 116 117 /* 118 * When a filesystem sets page_ops in an iomap mapping it returns, page_prepare 119 * and page_done will be called for each page written to. This only applies to 120 * buffered writes as unbuffered writes will not typically have pages 121 * associated with them. 122 * 123 * When page_prepare succeeds, page_done will always be called to do any 124 * cleanup work necessary. In that page_done call, @page will be NULL if the 125 * associated page could not be obtained. 126 */ 127 struct iomap_page_ops { 128 int (*page_prepare)(struct inode *inode, loff_t pos, unsigned len); 129 void (*page_done)(struct inode *inode, loff_t pos, unsigned copied, 130 struct page *page); 131 }; 132 133 /* 134 * Flags for iomap_begin / iomap_end. No flag implies a read. 135 */ 136 #define IOMAP_WRITE (1 << 0) /* writing, must allocate blocks */ 137 #define IOMAP_ZERO (1 << 1) /* zeroing operation, may skip holes */ 138 #define IOMAP_REPORT (1 << 2) /* report extent status, e.g. FIEMAP */ 139 #define IOMAP_FAULT (1 << 3) /* mapping for page fault */ 140 #define IOMAP_DIRECT (1 << 4) /* direct I/O */ 141 #define IOMAP_NOWAIT (1 << 5) /* do not block */ 142 #define IOMAP_OVERWRITE_ONLY (1 << 6) /* only pure overwrites allowed */ 143 #define IOMAP_UNSHARE (1 << 7) /* unshare_file_range */ 144 #ifdef CONFIG_FS_DAX 145 #define IOMAP_DAX (1 << 8) /* DAX mapping */ 146 #else 147 #define IOMAP_DAX 0 148 #endif /* CONFIG_FS_DAX */ 149 150 struct iomap_ops { 151 /* 152 * Return the existing mapping at pos, or reserve space starting at 153 * pos for up to length, as long as we can do it as a single mapping. 154 * The actual length is returned in iomap->length. 155 */ 156 int (*iomap_begin)(struct inode *inode, loff_t pos, loff_t length, 157 unsigned flags, struct iomap *iomap, 158 struct iomap *srcmap); 159 160 /* 161 * Commit and/or unreserve space previous allocated using iomap_begin. 162 * Written indicates the length of the successful write operation which 163 * needs to be commited, while the rest needs to be unreserved. 164 * Written might be zero if no data was written. 165 */ 166 int (*iomap_end)(struct inode *inode, loff_t pos, loff_t length, 167 ssize_t written, unsigned flags, struct iomap *iomap); 168 }; 169 170 /** 171 * struct iomap_iter - Iterate through a range of a file 172 * @inode: Set at the start of the iteration and should not change. 173 * @pos: The current file position we are operating on. It is updated by 174 * calls to iomap_iter(). Treat as read-only in the body. 175 * @len: The remaining length of the file segment we're operating on. 176 * It is updated at the same time as @pos. 177 * @processed: The number of bytes processed by the body in the most recent 178 * iteration, or a negative errno. 0 causes the iteration to stop. 179 * @flags: Zero or more of the iomap_begin flags above. 180 * @iomap: Map describing the I/O iteration 181 * @srcmap: Source map for COW operations 182 */ 183 struct iomap_iter { 184 struct inode *inode; 185 loff_t pos; 186 u64 len; 187 s64 processed; 188 unsigned flags; 189 struct iomap iomap; 190 struct iomap srcmap; 191 void *private; 192 }; 193 194 int iomap_iter(struct iomap_iter *iter, const struct iomap_ops *ops); 195 196 /** 197 * iomap_length - length of the current iomap iteration 198 * @iter: iteration structure 199 * 200 * Returns the length that the operation applies to for the current iteration. 201 */ 202 static inline u64 iomap_length(const struct iomap_iter *iter) 203 { 204 u64 end = iter->iomap.offset + iter->iomap.length; 205 206 if (iter->srcmap.type != IOMAP_HOLE) 207 end = min(end, iter->srcmap.offset + iter->srcmap.length); 208 return min(iter->len, end - iter->pos); 209 } 210 211 /** 212 * iomap_iter_srcmap - return the source map for the current iomap iteration 213 * @i: iteration structure 214 * 215 * Write operations on file systems with reflink support might require a 216 * source and a destination map. This function retourns the source map 217 * for a given operation, which may or may no be identical to the destination 218 * map in &i->iomap. 219 */ 220 static inline const struct iomap *iomap_iter_srcmap(const struct iomap_iter *i) 221 { 222 if (i->srcmap.type != IOMAP_HOLE) 223 return &i->srcmap; 224 return &i->iomap; 225 } 226 227 ssize_t iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *from, 228 const struct iomap_ops *ops); 229 int iomap_read_folio(struct folio *folio, const struct iomap_ops *ops); 230 void iomap_readahead(struct readahead_control *, const struct iomap_ops *ops); 231 bool iomap_is_partially_uptodate(struct folio *, size_t from, size_t count); 232 bool iomap_release_folio(struct folio *folio, gfp_t gfp_flags); 233 void iomap_invalidate_folio(struct folio *folio, size_t offset, size_t len); 234 int iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 235 const struct iomap_ops *ops); 236 int iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, 237 bool *did_zero, const struct iomap_ops *ops); 238 int iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 239 const struct iomap_ops *ops); 240 vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, 241 const struct iomap_ops *ops); 242 int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 243 u64 start, u64 len, const struct iomap_ops *ops); 244 loff_t iomap_seek_hole(struct inode *inode, loff_t offset, 245 const struct iomap_ops *ops); 246 loff_t iomap_seek_data(struct inode *inode, loff_t offset, 247 const struct iomap_ops *ops); 248 sector_t iomap_bmap(struct address_space *mapping, sector_t bno, 249 const struct iomap_ops *ops); 250 251 /* 252 * Structure for writeback I/O completions. 253 */ 254 struct iomap_ioend { 255 struct list_head io_list; /* next ioend in chain */ 256 u16 io_type; 257 u16 io_flags; /* IOMAP_F_* */ 258 u32 io_folios; /* folios added to ioend */ 259 struct inode *io_inode; /* file being written to */ 260 size_t io_size; /* size of the extent */ 261 loff_t io_offset; /* offset in the file */ 262 sector_t io_sector; /* start sector of ioend */ 263 struct bio *io_bio; /* bio being built */ 264 struct bio io_inline_bio; /* MUST BE LAST! */ 265 }; 266 267 struct iomap_writeback_ops { 268 /* 269 * Required, maps the blocks so that writeback can be performed on 270 * the range starting at offset. 271 */ 272 int (*map_blocks)(struct iomap_writepage_ctx *wpc, struct inode *inode, 273 loff_t offset); 274 275 /* 276 * Optional, allows the file systems to perform actions just before 277 * submitting the bio and/or override the bio end_io handler for complex 278 * operations like copy on write extent manipulation or unwritten extent 279 * conversions. 280 */ 281 int (*prepare_ioend)(struct iomap_ioend *ioend, int status); 282 283 /* 284 * Optional, allows the file system to discard state on a page where 285 * we failed to submit any I/O. 286 */ 287 void (*discard_folio)(struct folio *folio, loff_t pos); 288 }; 289 290 struct iomap_writepage_ctx { 291 struct iomap iomap; 292 struct iomap_ioend *ioend; 293 const struct iomap_writeback_ops *ops; 294 }; 295 296 void iomap_finish_ioends(struct iomap_ioend *ioend, int error); 297 void iomap_ioend_try_merge(struct iomap_ioend *ioend, 298 struct list_head *more_ioends); 299 void iomap_sort_ioends(struct list_head *ioend_list); 300 int iomap_writepages(struct address_space *mapping, 301 struct writeback_control *wbc, struct iomap_writepage_ctx *wpc, 302 const struct iomap_writeback_ops *ops); 303 304 /* 305 * Flags for direct I/O ->end_io: 306 */ 307 #define IOMAP_DIO_UNWRITTEN (1 << 0) /* covers unwritten extent(s) */ 308 #define IOMAP_DIO_COW (1 << 1) /* covers COW extent(s) */ 309 310 struct iomap_dio_ops { 311 int (*end_io)(struct kiocb *iocb, ssize_t size, int error, 312 unsigned flags); 313 void (*submit_io)(const struct iomap_iter *iter, struct bio *bio, 314 loff_t file_offset); 315 316 /* 317 * Filesystems wishing to attach private information to a direct io bio 318 * must provide a ->submit_io method that attaches the additional 319 * information to the bio and changes the ->bi_end_io callback to a 320 * custom function. This function should, at a minimum, perform any 321 * relevant post-processing of the bio and end with a call to 322 * iomap_dio_bio_end_io. 323 */ 324 struct bio_set *bio_set; 325 }; 326 327 /* 328 * Wait for the I/O to complete in iomap_dio_rw even if the kiocb is not 329 * synchronous. 330 */ 331 #define IOMAP_DIO_FORCE_WAIT (1 << 0) 332 333 /* 334 * Do not allocate blocks or zero partial blocks, but instead fall back to 335 * the caller by returning -EAGAIN. Used to optimize direct I/O writes that 336 * are not aligned to the file system block size. 337 */ 338 #define IOMAP_DIO_OVERWRITE_ONLY (1 << 1) 339 340 /* 341 * When a page fault occurs, return a partial synchronous result and allow 342 * the caller to retry the rest of the operation after dealing with the page 343 * fault. 344 */ 345 #define IOMAP_DIO_PARTIAL (1 << 2) 346 347 /* 348 * The caller will sync the write if needed; do not sync it within 349 * iomap_dio_rw. Overrides IOMAP_DIO_FORCE_WAIT. 350 */ 351 #define IOMAP_DIO_NOSYNC (1 << 3) 352 353 ssize_t iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, 354 const struct iomap_ops *ops, const struct iomap_dio_ops *dops, 355 unsigned int dio_flags, void *private, size_t done_before); 356 struct iomap_dio *__iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, 357 const struct iomap_ops *ops, const struct iomap_dio_ops *dops, 358 unsigned int dio_flags, void *private, size_t done_before); 359 ssize_t iomap_dio_complete(struct iomap_dio *dio); 360 void iomap_dio_bio_end_io(struct bio *bio); 361 362 #ifdef CONFIG_SWAP 363 struct file; 364 struct swap_info_struct; 365 366 int iomap_swapfile_activate(struct swap_info_struct *sis, 367 struct file *swap_file, sector_t *pagespan, 368 const struct iomap_ops *ops); 369 #else 370 # define iomap_swapfile_activate(sis, swapfile, pagespan, ops) (-EIO) 371 #endif /* CONFIG_SWAP */ 372 373 #endif /* LINUX_IOMAP_H */ 374