| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | #ifndef _LINUX_DAX_H |
| 3 | #define _LINUX_DAX_H |
| 4 | |
| 5 | #include <linux/fs.h> |
| 6 | #include <linux/mm.h> |
| 7 | #include <linux/radix-tree.h> |
| 8 | |
| 9 | typedef unsigned long dax_entry_t; |
| 10 | |
| 11 | struct dax_device; |
| 12 | struct gendisk; |
| 13 | struct iomap_ops; |
| 14 | struct iomap_iter; |
| 15 | struct iomap; |
| 16 | |
| 17 | enum dax_access_mode { |
| 18 | DAX_ACCESS, |
| 19 | DAX_RECOVERY_WRITE, |
| 20 | }; |
| 21 | |
| 22 | struct dax_operations { |
| 23 | /* |
| 24 | * direct_access: translate a device-relative |
| 25 | * logical-page-offset into an absolute physical pfn. Return the |
| 26 | * number of pages available for DAX at that pfn. |
| 27 | */ |
| 28 | long (*direct_access)(struct dax_device *, pgoff_t, long, |
| 29 | enum dax_access_mode, void **, unsigned long *); |
| 30 | /* zero_page_range: required operation. Zero page range */ |
| 31 | int (*zero_page_range)(struct dax_device *, pgoff_t, size_t); |
| 32 | /* |
| 33 | * recovery_write: recover a poisoned range by DAX device driver |
| 34 | * capable of clearing poison. |
| 35 | */ |
| 36 | size_t (*recovery_write)(struct dax_device *dax_dev, pgoff_t pgoff, |
| 37 | void *addr, size_t bytes, struct iov_iter *iter); |
| 38 | }; |
| 39 | |
| 40 | struct dax_holder_operations { |
| 41 | /* |
| 42 | * notify_failure - notify memory failure into inner holder device |
| 43 | * @dax_dev: the dax device which contains the holder |
| 44 | * @offset: offset on this dax device where memory failure occurs |
| 45 | * @len: length of this memory failure event |
| 46 | * @flags: action flags for memory failure handler |
| 47 | */ |
| 48 | int (*notify_failure)(struct dax_device *dax_dev, u64 offset, |
| 49 | u64 len, int mf_flags); |
| 50 | }; |
| 51 | |
| 52 | #if IS_ENABLED(CONFIG_DAX) |
| 53 | struct dax_device *alloc_dax(void *private, const struct dax_operations *ops); |
| 54 | void *dax_holder(struct dax_device *dax_dev); |
| 55 | void put_dax(struct dax_device *dax_dev); |
| 56 | void kill_dax(struct dax_device *dax_dev); |
| 57 | void dax_write_cache(struct dax_device *dax_dev, bool wc); |
| 58 | bool dax_write_cache_enabled(struct dax_device *dax_dev); |
| 59 | bool dax_synchronous(struct dax_device *dax_dev); |
| 60 | void set_dax_nocache(struct dax_device *dax_dev); |
| 61 | void set_dax_nomc(struct dax_device *dax_dev); |
| 62 | void set_dax_synchronous(struct dax_device *dax_dev); |
| 63 | size_t dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff, |
| 64 | void *addr, size_t bytes, struct iov_iter *i); |
| 65 | /* |
| 66 | * Check if given mapping is supported by the file / underlying device. |
| 67 | */ |
| 68 | static inline bool daxdev_mapping_supported(vm_flags_t vm_flags, |
| 69 | const struct inode *inode, |
| 70 | struct dax_device *dax_dev) |
| 71 | { |
| 72 | if (!(vm_flags & VM_SYNC)) |
| 73 | return true; |
| 74 | if (!IS_DAX(inode)) |
| 75 | return false; |
| 76 | return dax_synchronous(dax_dev); |
| 77 | } |
| 78 | #else |
| 79 | static inline void *dax_holder(struct dax_device *dax_dev) |
| 80 | { |
| 81 | return NULL; |
| 82 | } |
| 83 | static inline struct dax_device *alloc_dax(void *private, |
| 84 | const struct dax_operations *ops) |
| 85 | { |
| 86 | return ERR_PTR(-EOPNOTSUPP); |
| 87 | } |
| 88 | static inline void put_dax(struct dax_device *dax_dev) |
| 89 | { |
| 90 | } |
| 91 | static inline void kill_dax(struct dax_device *dax_dev) |
| 92 | { |
| 93 | } |
| 94 | static inline void dax_write_cache(struct dax_device *dax_dev, bool wc) |
| 95 | { |
| 96 | } |
| 97 | static inline bool dax_write_cache_enabled(struct dax_device *dax_dev) |
| 98 | { |
| 99 | return false; |
| 100 | } |
| 101 | static inline bool dax_synchronous(struct dax_device *dax_dev) |
| 102 | { |
| 103 | return true; |
| 104 | } |
| 105 | static inline void set_dax_nocache(struct dax_device *dax_dev) |
| 106 | { |
| 107 | } |
| 108 | static inline void set_dax_nomc(struct dax_device *dax_dev) |
| 109 | { |
| 110 | } |
| 111 | static inline void set_dax_synchronous(struct dax_device *dax_dev) |
| 112 | { |
| 113 | } |
| 114 | static inline bool daxdev_mapping_supported(vm_flags_t vm_flags, |
| 115 | const struct inode *inode, |
| 116 | struct dax_device *dax_dev) |
| 117 | { |
| 118 | return !(vm_flags & VM_SYNC); |
| 119 | } |
| 120 | static inline size_t dax_recovery_write(struct dax_device *dax_dev, |
| 121 | pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i) |
| 122 | { |
| 123 | return 0; |
| 124 | } |
| 125 | #endif |
| 126 | |
| 127 | struct writeback_control; |
| 128 | #if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX) |
| 129 | int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk); |
| 130 | void dax_remove_host(struct gendisk *disk); |
| 131 | struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off, |
| 132 | void *holder, const struct dax_holder_operations *ops); |
| 133 | void fs_put_dax(struct dax_device *dax_dev, void *holder); |
| 134 | #else |
| 135 | static inline int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk) |
| 136 | { |
| 137 | return 0; |
| 138 | } |
| 139 | static inline void dax_remove_host(struct gendisk *disk) |
| 140 | { |
| 141 | } |
| 142 | static inline struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, |
| 143 | u64 *start_off, void *holder, |
| 144 | const struct dax_holder_operations *ops) |
| 145 | { |
| 146 | return NULL; |
| 147 | } |
| 148 | static inline void fs_put_dax(struct dax_device *dax_dev, void *holder) |
| 149 | { |
| 150 | } |
| 151 | #endif /* CONFIG_BLOCK && CONFIG_FS_DAX */ |
| 152 | |
| 153 | #if IS_ENABLED(CONFIG_FS_DAX) |
| 154 | int dax_writeback_mapping_range(struct address_space *mapping, |
| 155 | struct dax_device *dax_dev, struct writeback_control *wbc); |
| 156 | |
| 157 | struct page *dax_layout_busy_page(struct address_space *mapping); |
| 158 | struct page *dax_layout_busy_page_range(struct address_space *mapping, loff_t start, loff_t end); |
| 159 | dax_entry_t dax_lock_folio(struct folio *folio); |
| 160 | void dax_unlock_folio(struct folio *folio, dax_entry_t cookie); |
| 161 | dax_entry_t dax_lock_mapping_entry(struct address_space *mapping, |
| 162 | unsigned long index, struct page **page); |
| 163 | void dax_unlock_mapping_entry(struct address_space *mapping, |
| 164 | unsigned long index, dax_entry_t cookie); |
| 165 | #else |
| 166 | static inline struct page *dax_layout_busy_page(struct address_space *mapping) |
| 167 | { |
| 168 | return NULL; |
| 169 | } |
| 170 | |
| 171 | static inline struct page *dax_layout_busy_page_range(struct address_space *mapping, pgoff_t start, pgoff_t nr_pages) |
| 172 | { |
| 173 | return NULL; |
| 174 | } |
| 175 | |
| 176 | static inline int dax_writeback_mapping_range(struct address_space *mapping, |
| 177 | struct dax_device *dax_dev, struct writeback_control *wbc) |
| 178 | { |
| 179 | return -EOPNOTSUPP; |
| 180 | } |
| 181 | |
| 182 | static inline dax_entry_t dax_lock_folio(struct folio *folio) |
| 183 | { |
| 184 | if (IS_DAX(folio->mapping->host)) |
| 185 | return ~0UL; |
| 186 | return 0; |
| 187 | } |
| 188 | |
| 189 | static inline void dax_unlock_folio(struct folio *folio, dax_entry_t cookie) |
| 190 | { |
| 191 | } |
| 192 | |
| 193 | static inline dax_entry_t dax_lock_mapping_entry(struct address_space *mapping, |
| 194 | unsigned long index, struct page **page) |
| 195 | { |
| 196 | return 0; |
| 197 | } |
| 198 | |
| 199 | static inline void dax_unlock_mapping_entry(struct address_space *mapping, |
| 200 | unsigned long index, dax_entry_t cookie) |
| 201 | { |
| 202 | } |
| 203 | #endif |
| 204 | |
| 205 | int dax_file_unshare(struct inode *inode, loff_t pos, loff_t len, |
| 206 | const struct iomap_ops *ops); |
| 207 | int dax_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, |
| 208 | const struct iomap_ops *ops); |
| 209 | int dax_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, |
| 210 | const struct iomap_ops *ops); |
| 211 | |
| 212 | static inline bool dax_page_is_idle(struct page *page) |
| 213 | { |
| 214 | return page && page_ref_count(page) == 0; |
| 215 | } |
| 216 | |
| 217 | #if IS_ENABLED(CONFIG_DAX) |
| 218 | int dax_read_lock(void); |
| 219 | void dax_read_unlock(int id); |
| 220 | #else |
| 221 | static inline int dax_read_lock(void) |
| 222 | { |
| 223 | return 0; |
| 224 | } |
| 225 | |
| 226 | static inline void dax_read_unlock(int id) |
| 227 | { |
| 228 | } |
| 229 | #endif /* CONFIG_DAX */ |
| 230 | |
| 231 | #if !IS_ENABLED(CONFIG_FS_DAX) |
| 232 | static inline int __must_check dax_break_layout(struct inode *inode, |
| 233 | loff_t start, loff_t end, void (cb)(struct inode *)) |
| 234 | { |
| 235 | return 0; |
| 236 | } |
| 237 | |
| 238 | static inline void dax_break_layout_final(struct inode *inode) |
| 239 | { |
| 240 | } |
| 241 | #endif |
| 242 | |
| 243 | bool dax_alive(struct dax_device *dax_dev); |
| 244 | void *dax_get_private(struct dax_device *dax_dev); |
| 245 | long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, |
| 246 | enum dax_access_mode mode, void **kaddr, unsigned long *pfn); |
| 247 | size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, |
| 248 | size_t bytes, struct iov_iter *i); |
| 249 | size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, |
| 250 | size_t bytes, struct iov_iter *i); |
| 251 | int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, |
| 252 | size_t nr_pages); |
| 253 | int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off, u64 len, |
| 254 | int mf_flags); |
| 255 | void dax_flush(struct dax_device *dax_dev, void *addr, size_t size); |
| 256 | |
| 257 | ssize_t dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter, |
| 258 | const struct iomap_ops *ops); |
| 259 | vm_fault_t dax_iomap_fault(struct vm_fault *vmf, unsigned int order, |
| 260 | unsigned long *pfnp, int *errp, |
| 261 | const struct iomap_ops *ops); |
| 262 | vm_fault_t dax_finish_sync_fault(struct vm_fault *vmf, |
| 263 | unsigned int order, unsigned long pfn); |
| 264 | int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index); |
| 265 | void dax_delete_mapping_range(struct address_space *mapping, |
| 266 | loff_t start, loff_t end); |
| 267 | int dax_invalidate_mapping_entry_sync(struct address_space *mapping, |
| 268 | pgoff_t index); |
| 269 | int __must_check dax_break_layout(struct inode *inode, loff_t start, |
| 270 | loff_t end, void (cb)(struct inode *)); |
| 271 | static inline int __must_check dax_break_layout_inode(struct inode *inode, |
| 272 | void (cb)(struct inode *)) |
| 273 | { |
| 274 | return dax_break_layout(inode, start: 0, LLONG_MAX, cb); |
| 275 | } |
| 276 | void dax_break_layout_final(struct inode *inode); |
| 277 | int dax_dedupe_file_range_compare(struct inode *src, loff_t srcoff, |
| 278 | struct inode *dest, loff_t destoff, |
| 279 | loff_t len, bool *is_same, |
| 280 | const struct iomap_ops *ops); |
| 281 | int dax_remap_file_range_prep(struct file *file_in, loff_t pos_in, |
| 282 | struct file *file_out, loff_t pos_out, |
| 283 | loff_t *len, unsigned int remap_flags, |
| 284 | const struct iomap_ops *ops); |
| 285 | static inline bool dax_mapping(struct address_space *mapping) |
| 286 | { |
| 287 | return mapping->host && IS_DAX(mapping->host); |
| 288 | } |
| 289 | |
| 290 | /* |
| 291 | * Due to dax's memory and block duo personalities, hwpoison reporting |
| 292 | * takes into consideration which personality is presently visible. |
| 293 | * When dax acts like a block device, such as in block IO, an encounter of |
| 294 | * dax hwpoison is reported as -EIO. |
| 295 | * When dax acts like memory, such as in page fault, a detection of hwpoison |
| 296 | * is reported as -EHWPOISON which leads to VM_FAULT_HWPOISON. |
| 297 | */ |
| 298 | static inline int dax_mem2blk_err(int err) |
| 299 | { |
| 300 | return (err == -EHWPOISON) ? -EIO : err; |
| 301 | } |
| 302 | |
| 303 | #ifdef CONFIG_DEV_DAX_HMEM_DEVICES |
| 304 | void hmem_register_resource(int target_nid, struct resource *r); |
| 305 | #else |
| 306 | static inline void hmem_register_resource(int target_nid, struct resource *r) |
| 307 | { |
| 308 | } |
| 309 | #endif |
| 310 | |
| 311 | typedef int (*walk_hmem_fn)(struct device *dev, int target_nid, |
| 312 | const struct resource *res); |
| 313 | int walk_hmem_resources(struct device *dev, walk_hmem_fn fn); |
| 314 | #endif |
| 315 | |