]>
Commit | Line | Data |
---|---|---|
7b6be844 DW |
1 | /* |
2 | * Copyright(c) 2017 Intel Corporation. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of version 2 of the GNU General Public License as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | */ | |
13 | #include <linux/pagemap.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/mount.h> | |
16 | #include <linux/magic.h> | |
ef510424 | 17 | #include <linux/genhd.h> |
7b6be844 DW |
18 | #include <linux/cdev.h> |
19 | #include <linux/hash.h> | |
20 | #include <linux/slab.h> | |
6568b08b | 21 | #include <linux/dax.h> |
7b6be844 DW |
22 | #include <linux/fs.h> |
23 | ||
7b6be844 DW |
24 | static dev_t dax_devt; |
25 | DEFINE_STATIC_SRCU(dax_srcu); | |
26 | static struct vfsmount *dax_mnt; | |
27 | static DEFINE_IDA(dax_minor_ida); | |
28 | static struct kmem_cache *dax_cache __read_mostly; | |
29 | static struct super_block *dax_superblock __read_mostly; | |
30 | ||
72058005 DW |
31 | #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head)) |
32 | static struct hlist_head dax_host_list[DAX_HASH_SIZE]; | |
33 | static DEFINE_SPINLOCK(dax_host_lock); | |
34 | ||
7b6be844 DW |
35 | int dax_read_lock(void) |
36 | { | |
37 | return srcu_read_lock(&dax_srcu); | |
38 | } | |
39 | EXPORT_SYMBOL_GPL(dax_read_lock); | |
40 | ||
41 | void dax_read_unlock(int id) | |
42 | { | |
43 | srcu_read_unlock(&dax_srcu, id); | |
44 | } | |
45 | EXPORT_SYMBOL_GPL(dax_read_unlock); | |
46 | ||
9d109081 | 47 | #ifdef CONFIG_BLOCK |
ef510424 DW |
48 | int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size, |
49 | pgoff_t *pgoff) | |
50 | { | |
51 | phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512; | |
52 | ||
53 | if (pgoff) | |
54 | *pgoff = PHYS_PFN(phys_off); | |
55 | if (phys_off % PAGE_SIZE || size % PAGE_SIZE) | |
56 | return -EINVAL; | |
57 | return 0; | |
58 | } | |
59 | EXPORT_SYMBOL(bdev_dax_pgoff); | |
60 | ||
61 | /** | |
62 | * __bdev_dax_supported() - Check if the device supports dax for filesystem | |
63 | * @sb: The superblock of the device | |
64 | * @blocksize: The block size of the device | |
65 | * | |
66 | * This is a library function for filesystems to check if the block device | |
67 | * can be mounted with dax option. | |
68 | * | |
69 | * Return: negative errno if unsupported, 0 if supported. | |
70 | */ | |
71 | int __bdev_dax_supported(struct super_block *sb, int blocksize) | |
72 | { | |
73 | struct block_device *bdev = sb->s_bdev; | |
74 | struct dax_device *dax_dev; | |
75 | pgoff_t pgoff; | |
76 | int err, id; | |
77 | void *kaddr; | |
78 | pfn_t pfn; | |
79 | long len; | |
80 | ||
81 | if (blocksize != PAGE_SIZE) { | |
82 | pr_err("VFS (%s): error: unsupported blocksize for dax\n", | |
83 | sb->s_id); | |
84 | return -EINVAL; | |
85 | } | |
86 | ||
87 | err = bdev_dax_pgoff(bdev, 0, PAGE_SIZE, &pgoff); | |
88 | if (err) { | |
89 | pr_err("VFS (%s): error: unaligned partition for dax\n", | |
90 | sb->s_id); | |
91 | return err; | |
92 | } | |
93 | ||
94 | dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); | |
95 | if (!dax_dev) { | |
96 | pr_err("VFS (%s): error: device does not support dax\n", | |
97 | sb->s_id); | |
98 | return -EOPNOTSUPP; | |
99 | } | |
100 | ||
101 | id = dax_read_lock(); | |
102 | len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn); | |
103 | dax_read_unlock(id); | |
104 | ||
105 | put_dax(dax_dev); | |
106 | ||
107 | if (len < 1) { | |
108 | pr_err("VFS (%s): error: dax access failed (%ld)", | |
109 | sb->s_id, len); | |
110 | return len < 0 ? len : -EIO; | |
111 | } | |
112 | ||
113 | return 0; | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(__bdev_dax_supported); | |
9d109081 | 116 | #endif |
ef510424 | 117 | |
7b6be844 DW |
118 | /** |
119 | * struct dax_device - anchor object for dax services | |
120 | * @inode: core vfs | |
121 | * @cdev: optional character interface for "device dax" | |
72058005 | 122 | * @host: optional name for lookups where the device path is not available |
7b6be844 DW |
123 | * @private: dax driver private data |
124 | * @alive: !alive + rcu grace period == no new operations / mappings | |
125 | */ | |
126 | struct dax_device { | |
72058005 | 127 | struct hlist_node list; |
7b6be844 DW |
128 | struct inode inode; |
129 | struct cdev cdev; | |
72058005 | 130 | const char *host; |
7b6be844 DW |
131 | void *private; |
132 | bool alive; | |
6568b08b | 133 | const struct dax_operations *ops; |
7b6be844 DW |
134 | }; |
135 | ||
b0686260 DW |
136 | /** |
137 | * dax_direct_access() - translate a device pgoff to an absolute pfn | |
138 | * @dax_dev: a dax_device instance representing the logical memory range | |
139 | * @pgoff: offset in pages from the start of the device to translate | |
140 | * @nr_pages: number of consecutive pages caller can handle relative to @pfn | |
141 | * @kaddr: output parameter that returns a virtual address mapping of pfn | |
142 | * @pfn: output parameter that returns an absolute pfn translation of @pgoff | |
143 | * | |
144 | * Return: negative errno if an error occurs, otherwise the number of | |
145 | * pages accessible at the device relative @pgoff. | |
146 | */ | |
147 | long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, | |
148 | void **kaddr, pfn_t *pfn) | |
149 | { | |
150 | long avail; | |
151 | ||
152 | /* | |
153 | * The device driver is allowed to sleep, in order to make the | |
154 | * memory directly accessible. | |
155 | */ | |
156 | might_sleep(); | |
157 | ||
158 | if (!dax_dev) | |
159 | return -EOPNOTSUPP; | |
160 | ||
161 | if (!dax_alive(dax_dev)) | |
162 | return -ENXIO; | |
163 | ||
164 | if (nr_pages < 0) | |
165 | return nr_pages; | |
166 | ||
167 | avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages, | |
168 | kaddr, pfn); | |
169 | if (!avail) | |
170 | return -ERANGE; | |
171 | return min(avail, nr_pages); | |
172 | } | |
173 | EXPORT_SYMBOL_GPL(dax_direct_access); | |
174 | ||
7b6be844 DW |
175 | bool dax_alive(struct dax_device *dax_dev) |
176 | { | |
177 | lockdep_assert_held(&dax_srcu); | |
178 | return dax_dev->alive; | |
179 | } | |
180 | EXPORT_SYMBOL_GPL(dax_alive); | |
181 | ||
72058005 DW |
182 | static int dax_host_hash(const char *host) |
183 | { | |
184 | return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE; | |
185 | } | |
186 | ||
7b6be844 DW |
187 | /* |
188 | * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring | |
189 | * that any fault handlers or operations that might have seen | |
190 | * dax_alive(), have completed. Any operations that start after | |
191 | * synchronize_srcu() has run will abort upon seeing !dax_alive(). | |
192 | */ | |
193 | void kill_dax(struct dax_device *dax_dev) | |
194 | { | |
195 | if (!dax_dev) | |
196 | return; | |
197 | ||
198 | dax_dev->alive = false; | |
72058005 | 199 | |
7b6be844 | 200 | synchronize_srcu(&dax_srcu); |
72058005 DW |
201 | |
202 | spin_lock(&dax_host_lock); | |
203 | hlist_del_init(&dax_dev->list); | |
204 | spin_unlock(&dax_host_lock); | |
205 | ||
7b6be844 DW |
206 | dax_dev->private = NULL; |
207 | } | |
208 | EXPORT_SYMBOL_GPL(kill_dax); | |
209 | ||
210 | static struct inode *dax_alloc_inode(struct super_block *sb) | |
211 | { | |
212 | struct dax_device *dax_dev; | |
213 | ||
214 | dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL); | |
215 | return &dax_dev->inode; | |
216 | } | |
217 | ||
218 | static struct dax_device *to_dax_dev(struct inode *inode) | |
219 | { | |
220 | return container_of(inode, struct dax_device, inode); | |
221 | } | |
222 | ||
223 | static void dax_i_callback(struct rcu_head *head) | |
224 | { | |
225 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
226 | struct dax_device *dax_dev = to_dax_dev(inode); | |
227 | ||
72058005 DW |
228 | kfree(dax_dev->host); |
229 | dax_dev->host = NULL; | |
7b6be844 DW |
230 | ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev)); |
231 | kmem_cache_free(dax_cache, dax_dev); | |
232 | } | |
233 | ||
234 | static void dax_destroy_inode(struct inode *inode) | |
235 | { | |
236 | struct dax_device *dax_dev = to_dax_dev(inode); | |
237 | ||
238 | WARN_ONCE(dax_dev->alive, | |
239 | "kill_dax() must be called before final iput()\n"); | |
240 | call_rcu(&inode->i_rcu, dax_i_callback); | |
241 | } | |
242 | ||
243 | static const struct super_operations dax_sops = { | |
244 | .statfs = simple_statfs, | |
245 | .alloc_inode = dax_alloc_inode, | |
246 | .destroy_inode = dax_destroy_inode, | |
247 | .drop_inode = generic_delete_inode, | |
248 | }; | |
249 | ||
250 | static struct dentry *dax_mount(struct file_system_type *fs_type, | |
251 | int flags, const char *dev_name, void *data) | |
252 | { | |
253 | return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC); | |
254 | } | |
255 | ||
256 | static struct file_system_type dax_fs_type = { | |
257 | .name = "dax", | |
258 | .mount = dax_mount, | |
259 | .kill_sb = kill_anon_super, | |
260 | }; | |
261 | ||
262 | static int dax_test(struct inode *inode, void *data) | |
263 | { | |
264 | dev_t devt = *(dev_t *) data; | |
265 | ||
266 | return inode->i_rdev == devt; | |
267 | } | |
268 | ||
269 | static int dax_set(struct inode *inode, void *data) | |
270 | { | |
271 | dev_t devt = *(dev_t *) data; | |
272 | ||
273 | inode->i_rdev = devt; | |
274 | return 0; | |
275 | } | |
276 | ||
277 | static struct dax_device *dax_dev_get(dev_t devt) | |
278 | { | |
279 | struct dax_device *dax_dev; | |
280 | struct inode *inode; | |
281 | ||
282 | inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31), | |
283 | dax_test, dax_set, &devt); | |
284 | ||
285 | if (!inode) | |
286 | return NULL; | |
287 | ||
288 | dax_dev = to_dax_dev(inode); | |
289 | if (inode->i_state & I_NEW) { | |
290 | dax_dev->alive = true; | |
291 | inode->i_cdev = &dax_dev->cdev; | |
292 | inode->i_mode = S_IFCHR; | |
293 | inode->i_flags = S_DAX; | |
294 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
295 | unlock_new_inode(inode); | |
296 | } | |
297 | ||
298 | return dax_dev; | |
299 | } | |
300 | ||
72058005 DW |
301 | static void dax_add_host(struct dax_device *dax_dev, const char *host) |
302 | { | |
303 | int hash; | |
304 | ||
305 | /* | |
306 | * Unconditionally init dax_dev since it's coming from a | |
307 | * non-zeroed slab cache | |
308 | */ | |
309 | INIT_HLIST_NODE(&dax_dev->list); | |
310 | dax_dev->host = host; | |
311 | if (!host) | |
312 | return; | |
313 | ||
314 | hash = dax_host_hash(host); | |
315 | spin_lock(&dax_host_lock); | |
316 | hlist_add_head(&dax_dev->list, &dax_host_list[hash]); | |
317 | spin_unlock(&dax_host_lock); | |
318 | } | |
319 | ||
6568b08b DW |
320 | struct dax_device *alloc_dax(void *private, const char *__host, |
321 | const struct dax_operations *ops) | |
7b6be844 DW |
322 | { |
323 | struct dax_device *dax_dev; | |
72058005 | 324 | const char *host; |
7b6be844 DW |
325 | dev_t devt; |
326 | int minor; | |
327 | ||
72058005 DW |
328 | host = kstrdup(__host, GFP_KERNEL); |
329 | if (__host && !host) | |
330 | return NULL; | |
331 | ||
cf1e2289 | 332 | minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL); |
7b6be844 | 333 | if (minor < 0) |
72058005 | 334 | goto err_minor; |
7b6be844 DW |
335 | |
336 | devt = MKDEV(MAJOR(dax_devt), minor); | |
337 | dax_dev = dax_dev_get(devt); | |
338 | if (!dax_dev) | |
72058005 | 339 | goto err_dev; |
7b6be844 | 340 | |
72058005 | 341 | dax_add_host(dax_dev, host); |
6568b08b | 342 | dax_dev->ops = ops; |
7b6be844 DW |
343 | dax_dev->private = private; |
344 | return dax_dev; | |
345 | ||
72058005 | 346 | err_dev: |
7b6be844 | 347 | ida_simple_remove(&dax_minor_ida, minor); |
72058005 DW |
348 | err_minor: |
349 | kfree(host); | |
7b6be844 DW |
350 | return NULL; |
351 | } | |
352 | EXPORT_SYMBOL_GPL(alloc_dax); | |
353 | ||
354 | void put_dax(struct dax_device *dax_dev) | |
355 | { | |
356 | if (!dax_dev) | |
357 | return; | |
358 | iput(&dax_dev->inode); | |
359 | } | |
360 | EXPORT_SYMBOL_GPL(put_dax); | |
361 | ||
72058005 DW |
362 | /** |
363 | * dax_get_by_host() - temporary lookup mechanism for filesystem-dax | |
364 | * @host: alternate name for the device registered by a dax driver | |
365 | */ | |
366 | struct dax_device *dax_get_by_host(const char *host) | |
367 | { | |
368 | struct dax_device *dax_dev, *found = NULL; | |
369 | int hash, id; | |
370 | ||
371 | if (!host) | |
372 | return NULL; | |
373 | ||
374 | hash = dax_host_hash(host); | |
375 | ||
376 | id = dax_read_lock(); | |
377 | spin_lock(&dax_host_lock); | |
378 | hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) { | |
379 | if (!dax_alive(dax_dev) | |
380 | || strcmp(host, dax_dev->host) != 0) | |
381 | continue; | |
382 | ||
383 | if (igrab(&dax_dev->inode)) | |
384 | found = dax_dev; | |
385 | break; | |
386 | } | |
387 | spin_unlock(&dax_host_lock); | |
388 | dax_read_unlock(id); | |
389 | ||
390 | return found; | |
391 | } | |
392 | EXPORT_SYMBOL_GPL(dax_get_by_host); | |
393 | ||
7b6be844 DW |
394 | /** |
395 | * inode_dax: convert a public inode into its dax_dev | |
396 | * @inode: An inode with i_cdev pointing to a dax_dev | |
397 | * | |
398 | * Note this is not equivalent to to_dax_dev() which is for private | |
399 | * internal use where we know the inode filesystem type == dax_fs_type. | |
400 | */ | |
401 | struct dax_device *inode_dax(struct inode *inode) | |
402 | { | |
403 | struct cdev *cdev = inode->i_cdev; | |
404 | ||
405 | return container_of(cdev, struct dax_device, cdev); | |
406 | } | |
407 | EXPORT_SYMBOL_GPL(inode_dax); | |
408 | ||
409 | struct inode *dax_inode(struct dax_device *dax_dev) | |
410 | { | |
411 | return &dax_dev->inode; | |
412 | } | |
413 | EXPORT_SYMBOL_GPL(dax_inode); | |
414 | ||
415 | void *dax_get_private(struct dax_device *dax_dev) | |
416 | { | |
417 | return dax_dev->private; | |
418 | } | |
419 | EXPORT_SYMBOL_GPL(dax_get_private); | |
420 | ||
421 | static void init_once(void *_dax_dev) | |
422 | { | |
423 | struct dax_device *dax_dev = _dax_dev; | |
424 | struct inode *inode = &dax_dev->inode; | |
425 | ||
426 | inode_init_once(inode); | |
427 | } | |
428 | ||
429 | static int __dax_fs_init(void) | |
430 | { | |
431 | int rc; | |
432 | ||
433 | dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0, | |
434 | (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| | |
435 | SLAB_MEM_SPREAD|SLAB_ACCOUNT), | |
436 | init_once); | |
437 | if (!dax_cache) | |
438 | return -ENOMEM; | |
439 | ||
440 | rc = register_filesystem(&dax_fs_type); | |
441 | if (rc) | |
442 | goto err_register_fs; | |
443 | ||
444 | dax_mnt = kern_mount(&dax_fs_type); | |
445 | if (IS_ERR(dax_mnt)) { | |
446 | rc = PTR_ERR(dax_mnt); | |
447 | goto err_mount; | |
448 | } | |
449 | dax_superblock = dax_mnt->mnt_sb; | |
450 | ||
451 | return 0; | |
452 | ||
453 | err_mount: | |
454 | unregister_filesystem(&dax_fs_type); | |
455 | err_register_fs: | |
456 | kmem_cache_destroy(dax_cache); | |
457 | ||
458 | return rc; | |
459 | } | |
460 | ||
461 | static void __dax_fs_exit(void) | |
462 | { | |
463 | kern_unmount(dax_mnt); | |
464 | unregister_filesystem(&dax_fs_type); | |
465 | kmem_cache_destroy(dax_cache); | |
466 | } | |
467 | ||
468 | static int __init dax_fs_init(void) | |
469 | { | |
470 | int rc; | |
471 | ||
472 | rc = __dax_fs_init(); | |
473 | if (rc) | |
474 | return rc; | |
475 | ||
cf1e2289 | 476 | rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax"); |
7b6be844 DW |
477 | if (rc) |
478 | __dax_fs_exit(); | |
479 | return rc; | |
480 | } | |
481 | ||
482 | static void __exit dax_fs_exit(void) | |
483 | { | |
cf1e2289 | 484 | unregister_chrdev_region(dax_devt, MINORMASK+1); |
7b6be844 DW |
485 | ida_destroy(&dax_minor_ida); |
486 | __dax_fs_exit(); | |
487 | } | |
488 | ||
489 | MODULE_AUTHOR("Intel Corporation"); | |
490 | MODULE_LICENSE("GPL v2"); | |
491 | subsys_initcall(dax_fs_init); | |
492 | module_exit(dax_fs_exit); |