xref: /linux/include/linux/dax.h (revision c34e9ab9a612ee8b18273398ef75c207b01f516d)
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 **, pfn_t *);
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(struct vm_area_struct *vma,
69 					     struct dax_device *dax_dev)
70 {
71 	if (!(vma->vm_flags & VM_SYNC))
72 		return true;
73 	if (!IS_DAX(file_inode(vma->vm_file)))
74 		return false;
75 	return dax_synchronous(dax_dev);
76 }
77 #else
78 static inline void *dax_holder(struct dax_device *dax_dev)
79 {
80 	return NULL;
81 }
82 static inline struct dax_device *alloc_dax(void *private,
83 		const struct dax_operations *ops)
84 {
85 	return ERR_PTR(-EOPNOTSUPP);
86 }
87 static inline void put_dax(struct dax_device *dax_dev)
88 {
89 }
90 static inline void kill_dax(struct dax_device *dax_dev)
91 {
92 }
93 static inline void dax_write_cache(struct dax_device *dax_dev, bool wc)
94 {
95 }
96 static inline bool dax_write_cache_enabled(struct dax_device *dax_dev)
97 {
98 	return false;
99 }
100 static inline bool dax_synchronous(struct dax_device *dax_dev)
101 {
102 	return true;
103 }
104 static inline void set_dax_nocache(struct dax_device *dax_dev)
105 {
106 }
107 static inline void set_dax_nomc(struct dax_device *dax_dev)
108 {
109 }
110 static inline void set_dax_synchronous(struct dax_device *dax_dev)
111 {
112 }
113 static inline bool daxdev_mapping_supported(struct vm_area_struct *vma,
114 				struct dax_device *dax_dev)
115 {
116 	return !(vma->vm_flags & VM_SYNC);
117 }
118 static inline size_t dax_recovery_write(struct dax_device *dax_dev,
119 		pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i)
120 {
121 	return 0;
122 }
123 #endif
124 
125 struct writeback_control;
126 #if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX)
127 int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk);
128 void dax_remove_host(struct gendisk *disk);
129 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off,
130 		void *holder, const struct dax_holder_operations *ops);
131 void fs_put_dax(struct dax_device *dax_dev, void *holder);
132 #else
133 static inline int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk)
134 {
135 	return 0;
136 }
137 static inline void dax_remove_host(struct gendisk *disk)
138 {
139 }
140 static inline struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev,
141 		u64 *start_off, void *holder,
142 		const struct dax_holder_operations *ops)
143 {
144 	return NULL;
145 }
146 static inline void fs_put_dax(struct dax_device *dax_dev, void *holder)
147 {
148 }
149 #endif /* CONFIG_BLOCK && CONFIG_FS_DAX */
150 
151 #if IS_ENABLED(CONFIG_FS_DAX)
152 int dax_writeback_mapping_range(struct address_space *mapping,
153 		struct dax_device *dax_dev, struct writeback_control *wbc);
154 
155 struct page *dax_layout_busy_page(struct address_space *mapping);
156 struct page *dax_layout_busy_page_range(struct address_space *mapping, loff_t start, loff_t end);
157 dax_entry_t dax_lock_folio(struct folio *folio);
158 void dax_unlock_folio(struct folio *folio, dax_entry_t cookie);
159 dax_entry_t dax_lock_mapping_entry(struct address_space *mapping,
160 		unsigned long index, struct page **page);
161 void dax_unlock_mapping_entry(struct address_space *mapping,
162 		unsigned long index, dax_entry_t cookie);
163 #else
164 static inline struct page *dax_layout_busy_page(struct address_space *mapping)
165 {
166 	return NULL;
167 }
168 
169 static inline struct page *dax_layout_busy_page_range(struct address_space *mapping, pgoff_t start, pgoff_t nr_pages)
170 {
171 	return NULL;
172 }
173 
174 static inline int dax_writeback_mapping_range(struct address_space *mapping,
175 		struct dax_device *dax_dev, struct writeback_control *wbc)
176 {
177 	return -EOPNOTSUPP;
178 }
179 
180 static inline dax_entry_t dax_lock_folio(struct folio *folio)
181 {
182 	if (IS_DAX(folio->mapping->host))
183 		return ~0UL;
184 	return 0;
185 }
186 
187 static inline void dax_unlock_folio(struct folio *folio, dax_entry_t cookie)
188 {
189 }
190 
191 static inline dax_entry_t dax_lock_mapping_entry(struct address_space *mapping,
192 		unsigned long index, struct page **page)
193 {
194 	return 0;
195 }
196 
197 static inline void dax_unlock_mapping_entry(struct address_space *mapping,
198 		unsigned long index, dax_entry_t cookie)
199 {
200 }
201 #endif
202 
203 int dax_file_unshare(struct inode *inode, loff_t pos, loff_t len,
204 		const struct iomap_ops *ops);
205 int dax_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
206 		const struct iomap_ops *ops);
207 int dax_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
208 		const struct iomap_ops *ops);
209 
210 #if IS_ENABLED(CONFIG_DAX)
211 int dax_read_lock(void);
212 void dax_read_unlock(int id);
213 #else
214 static inline int dax_read_lock(void)
215 {
216 	return 0;
217 }
218 
219 static inline void dax_read_unlock(int id)
220 {
221 }
222 #endif /* CONFIG_DAX */
223 bool dax_alive(struct dax_device *dax_dev);
224 void *dax_get_private(struct dax_device *dax_dev);
225 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
226 		enum dax_access_mode mode, void **kaddr, pfn_t *pfn);
227 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
228 		size_t bytes, struct iov_iter *i);
229 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
230 		size_t bytes, struct iov_iter *i);
231 int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
232 			size_t nr_pages);
233 int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off, u64 len,
234 		int mf_flags);
235 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size);
236 
237 ssize_t dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
238 		const struct iomap_ops *ops);
239 vm_fault_t dax_iomap_fault(struct vm_fault *vmf, unsigned int order,
240 		    pfn_t *pfnp, int *errp, const struct iomap_ops *ops);
241 vm_fault_t dax_finish_sync_fault(struct vm_fault *vmf,
242 		unsigned int order, pfn_t pfn);
243 int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index);
244 int dax_invalidate_mapping_entry_sync(struct address_space *mapping,
245 				      pgoff_t index);
246 int dax_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
247 				  struct inode *dest, loff_t destoff,
248 				  loff_t len, bool *is_same,
249 				  const struct iomap_ops *ops);
250 int dax_remap_file_range_prep(struct file *file_in, loff_t pos_in,
251 			      struct file *file_out, loff_t pos_out,
252 			      loff_t *len, unsigned int remap_flags,
253 			      const struct iomap_ops *ops);
254 static inline bool dax_mapping(struct address_space *mapping)
255 {
256 	return mapping->host && IS_DAX(mapping->host);
257 }
258 
259 /*
260  * Due to dax's memory and block duo personalities, hwpoison reporting
261  * takes into consideration which personality is presently visible.
262  * When dax acts like a block device, such as in block IO, an encounter of
263  * dax hwpoison is reported as -EIO.
264  * When dax acts like memory, such as in page fault, a detection of hwpoison
265  * is reported as -EHWPOISON which leads to VM_FAULT_HWPOISON.
266  */
267 static inline int dax_mem2blk_err(int err)
268 {
269 	return (err == -EHWPOISON) ? -EIO : err;
270 }
271 
272 #ifdef CONFIG_DEV_DAX_HMEM_DEVICES
273 void hmem_register_resource(int target_nid, struct resource *r);
274 #else
275 static inline void hmem_register_resource(int target_nid, struct resource *r)
276 {
277 }
278 #endif
279 
280 typedef int (*walk_hmem_fn)(struct device *dev, int target_nid,
281 			    const struct resource *res);
282 int walk_hmem_resources(struct device *dev, walk_hmem_fn fn);
283 #endif
284