projects/linux·blob master · refs

313 lines · 9.6 KiB

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
9typedef unsigned long dax_entry_t;
10
11struct dax_device;
12struct gendisk;
13struct iomap_ops;
14struct iomap_iter;
15struct iomap;
16
17enum dax_access_mode {
18 DAX_ACCESS,
19 DAX_RECOVERY_WRITE,
20};
21
22struct 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
40struct 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)
53struct dax_device *alloc_dax(void *private, const struct dax_operations *ops);
54void *dax_holder(struct dax_device *dax_dev);
55void put_dax(struct dax_device *dax_dev);
56void kill_dax(struct dax_device *dax_dev);
57struct dax_device *dax_dev_get(dev_t devt);
58void dax_write_cache(struct dax_device *dax_dev, bool wc);
59bool dax_write_cache_enabled(struct dax_device *dax_dev);
60bool dax_synchronous(struct dax_device *dax_dev);
61void set_dax_nocache(struct dax_device *dax_dev);
62void set_dax_nomc(struct dax_device *dax_dev);
63void set_dax_synchronous(struct dax_device *dax_dev);
64size_t dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
65 void *addr, size_t bytes, struct iov_iter *i);
66/*
67 * Check if given mapping is supported by the file / underlying device.
68 */
69static inline bool daxdev_mapping_supported(const struct vm_area_desc *desc,
70 const struct inode *inode,
71 struct dax_device *dax_dev)
72{
73 if (!vma_desc_test(desc, VMA_SYNC_BIT))
74 return true;
75 if (!IS_DAX(inode))
76 return false;
77 return dax_synchronous(dax_dev);
78}
79#else
80static inline void *dax_holder(struct dax_device *dax_dev)
81{
82 return NULL;
83}
84static inline struct dax_device *alloc_dax(void *private,
85 const struct dax_operations *ops)
86{
87 return ERR_PTR(-EOPNOTSUPP);
88}
89static inline void put_dax(struct dax_device *dax_dev)
90{
91}
92static inline void kill_dax(struct dax_device *dax_dev)
93{
94}
95static inline void dax_write_cache(struct dax_device *dax_dev, bool wc)
96{
97}
98static inline bool dax_write_cache_enabled(struct dax_device *dax_dev)
99{
100 return false;
101}
102static inline bool dax_synchronous(struct dax_device *dax_dev)
103{
104 return true;
105}
106static inline void set_dax_nocache(struct dax_device *dax_dev)
107{
108}
109static inline void set_dax_nomc(struct dax_device *dax_dev)
110{
111}
112static inline void set_dax_synchronous(struct dax_device *dax_dev)
113{
114}
115static inline bool daxdev_mapping_supported(const struct vm_area_desc *desc,
116 const struct inode *inode,
117 struct dax_device *dax_dev)
118{
119 return !vma_desc_test(desc, VMA_SYNC_BIT);
120}
121static inline size_t dax_recovery_write(struct dax_device *dax_dev,
122 pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i)
123{
124 return 0;
125}
126#endif
127
128struct writeback_control;
129#if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX)
130int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk);
131void dax_remove_host(struct gendisk *disk);
132struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off,
133 void *holder, const struct dax_holder_operations *ops);
134#else
135static inline int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk)
136{
137 return 0;
138}
139static inline void dax_remove_host(struct gendisk *disk)
140{
141}
142static 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#endif /* CONFIG_BLOCK && CONFIG_FS_DAX */
149
150#if IS_ENABLED(CONFIG_FS_DAX)
151void fs_put_dax(struct dax_device *dax_dev, void *holder);
152int fs_dax_get(struct dax_device *dax_dev, void *holder,
153 const struct dax_holder_operations *hops);
154int dax_writeback_mapping_range(struct address_space *mapping,
155 struct dax_device *dax_dev, struct writeback_control *wbc);
156int dax_folio_reset_order(struct folio *folio);
157
158dax_entry_t dax_lock_folio(struct folio *folio);
159void dax_unlock_folio(struct folio *folio, dax_entry_t cookie);
160dax_entry_t dax_lock_mapping_entry(struct address_space *mapping,
161 unsigned long index, struct page **page);
162void dax_unlock_mapping_entry(struct address_space *mapping,
163 unsigned long index, dax_entry_t cookie);
164#else
165static inline void fs_put_dax(struct dax_device *dax_dev, void *holder)
166{
167}
168
169static inline int fs_dax_get(struct dax_device *dax_dev, void *holder,
170 const struct dax_holder_operations *hops)
171{
172 return -EOPNOTSUPP;
173}
174static 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
180static 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
187static inline void dax_unlock_folio(struct folio *folio, dax_entry_t cookie)
188{
189}
190
191static 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
197static inline void dax_unlock_mapping_entry(struct address_space *mapping,
198 unsigned long index, dax_entry_t cookie)
199{
200}
201#endif
202
203int dax_file_unshare(struct inode *inode, loff_t pos, loff_t len,
204 const struct iomap_ops *ops);
205int dax_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
206 const struct iomap_ops *ops);
207int dax_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
208 const struct iomap_ops *ops);
209
210static inline bool dax_page_is_idle(struct page *page)
211{
212 return page && page_ref_count(page) == 0;
213}
214
215#if IS_ENABLED(CONFIG_DAX)
216int dax_read_lock(void);
217void dax_read_unlock(int id);
218#else
219static inline int dax_read_lock(void)
220{
221 return 0;
222}
223
224static inline void dax_read_unlock(int id)
225{
226}
227#endif /* CONFIG_DAX */
228
229#if !IS_ENABLED(CONFIG_FS_DAX)
230static inline int __must_check dax_break_layout(struct inode *inode,
231 loff_t start, loff_t end, void (cb)(struct inode *))
232{
233 return 0;
234}
235
236static inline void dax_break_layout_final(struct inode *inode)
237{
238}
239#endif
240
241bool dax_alive(struct dax_device *dax_dev);
242void *dax_get_private(struct dax_device *dax_dev);
243int dax_set_ops(struct dax_device *dax_dev, const struct dax_operations *ops);
244long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
245 enum dax_access_mode mode, void **kaddr, unsigned long *pfn);
246size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
247 size_t bytes, struct iov_iter *i);
248size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
249 size_t bytes, struct iov_iter *i);
250int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
251 size_t nr_pages);
252int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off, u64 len,
253 int mf_flags);
254void dax_flush(struct dax_device *dax_dev, void *addr, size_t size);
255
256ssize_t dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
257 const struct iomap_ops *ops);
258vm_fault_t dax_iomap_fault(struct vm_fault *vmf, unsigned int order,
259 unsigned long *pfnp, int *errp,
260 const struct iomap_ops *ops);
261vm_fault_t dax_finish_sync_fault(struct vm_fault *vmf,
262 unsigned int order, unsigned long pfn);
263int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index);
264void dax_delete_mapping_range(struct address_space *mapping,
265 loff_t start, loff_t end);
266int dax_invalidate_mapping_entry_sync(struct address_space *mapping,
267 pgoff_t index);
268int __must_check dax_break_layout(struct inode *inode, loff_t start,
269 loff_t end, void (cb)(struct inode *));
270static inline int __must_check dax_break_layout_inode(struct inode *inode,
271 void (cb)(struct inode *))
272{
273 return dax_break_layout(inode, 0, LLONG_MAX, cb);
274}
275void dax_break_layout_final(struct inode *inode);
276int dax_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
277 struct inode *dest, loff_t destoff,
278 loff_t len, bool *is_same,
279 const struct iomap_ops *ops);
280int dax_remap_file_range_prep(struct file *file_in, loff_t pos_in,
281 struct file *file_out, loff_t pos_out,
282 loff_t *len, unsigned int remap_flags,
283 const struct iomap_ops *ops);
284static inline bool dax_mapping(struct address_space *mapping)
285{
286 return mapping->host && IS_DAX(mapping->host);
287}
288
289/*
290 * Due to dax's memory and block duo personalities, hwpoison reporting
291 * takes into consideration which personality is presently visible.
292 * When dax acts like a block device, such as in block IO, an encounter of
293 * dax hwpoison is reported as -EIO.
294 * When dax acts like memory, such as in page fault, a detection of hwpoison
295 * is reported as -EHWPOISON which leads to VM_FAULT_HWPOISON.
296 */
297static inline int dax_mem2blk_err(int err)
298{
299 return (err == -EHWPOISON) ? -EIO : err;
300}
301
302#ifdef CONFIG_DEV_DAX_HMEM_DEVICES
303void hmem_register_resource(int target_nid, struct resource *r);
304#else
305static inline void hmem_register_resource(int target_nid, struct resource *r)
306{
307}
308#endif
309
310typedef int (*walk_hmem_fn)(struct device *dev, int target_nid,
311 const struct resource *res);
312int walk_hmem_resources(struct device *dev, walk_hmem_fn fn);
313#endif