1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * Copyright(c) 2017 Intel Corporation. All rights reserved. |
4 | */ |
5 | #include <linux/pagemap.h> |
6 | #include <linux/module.h> |
7 | #include <linux/mount.h> |
8 | #include <linux/pseudo_fs.h> |
9 | #include <linux/magic.h> |
10 | #include <linux/pfn_t.h> |
11 | #include <linux/cdev.h> |
12 | #include <linux/slab.h> |
13 | #include <linux/uio.h> |
14 | #include <linux/dax.h> |
15 | #include <linux/fs.h> |
16 | #include "dax-private.h" |
17 | |
18 | /** |
19 | * struct dax_device - anchor object for dax services |
20 | * @inode: core vfs |
21 | * @cdev: optional character interface for "device dax" |
22 | * @private: dax driver private data |
23 | * @flags: state and boolean properties |
24 | * @ops: operations for this device |
25 | * @holder_data: holder of a dax_device: could be filesystem or mapped device |
26 | * @holder_ops: operations for the inner holder |
27 | */ |
28 | struct dax_device { |
29 | struct inode inode; |
30 | struct cdev cdev; |
31 | void *private; |
32 | unsigned long flags; |
33 | const struct dax_operations *ops; |
34 | void *holder_data; |
35 | const struct dax_holder_operations *holder_ops; |
36 | }; |
37 | |
38 | static dev_t dax_devt; |
39 | DEFINE_STATIC_SRCU(dax_srcu); |
40 | static struct vfsmount *dax_mnt; |
41 | static DEFINE_IDA(dax_minor_ida); |
42 | static struct kmem_cache *dax_cache __read_mostly; |
43 | static struct super_block *dax_superblock __read_mostly; |
44 | |
45 | int dax_read_lock(void) |
46 | { |
47 | return srcu_read_lock(ssp: &dax_srcu); |
48 | } |
49 | EXPORT_SYMBOL_GPL(dax_read_lock); |
50 | |
51 | void dax_read_unlock(int id) |
52 | { |
53 | srcu_read_unlock(ssp: &dax_srcu, idx: id); |
54 | } |
55 | EXPORT_SYMBOL_GPL(dax_read_unlock); |
56 | |
57 | #if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX) |
58 | #include <linux/blkdev.h> |
59 | |
60 | static DEFINE_XARRAY(dax_hosts); |
61 | |
62 | int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk) |
63 | { |
64 | return xa_insert(xa: &dax_hosts, index: (unsigned long)disk, entry: dax_dev, GFP_KERNEL); |
65 | } |
66 | EXPORT_SYMBOL_GPL(dax_add_host); |
67 | |
68 | void dax_remove_host(struct gendisk *disk) |
69 | { |
70 | xa_erase(&dax_hosts, index: (unsigned long)disk); |
71 | } |
72 | EXPORT_SYMBOL_GPL(dax_remove_host); |
73 | |
74 | /** |
75 | * fs_dax_get_by_bdev() - temporary lookup mechanism for filesystem-dax |
76 | * @bdev: block device to find a dax_device for |
77 | * @start_off: returns the byte offset into the dax_device that @bdev starts |
78 | * @holder: filesystem or mapped device inside the dax_device |
79 | * @ops: operations for the inner holder |
80 | */ |
81 | struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off, |
82 | void *holder, const struct dax_holder_operations *ops) |
83 | { |
84 | struct dax_device *dax_dev; |
85 | u64 part_size; |
86 | int id; |
87 | |
88 | if (!blk_queue_dax(bdev->bd_disk->queue)) |
89 | return NULL; |
90 | |
91 | *start_off = get_start_sect(bdev) * SECTOR_SIZE; |
92 | part_size = bdev_nr_sectors(bdev) * SECTOR_SIZE; |
93 | if (*start_off % PAGE_SIZE || part_size % PAGE_SIZE) { |
94 | pr_info("%pg: error: unaligned partition for dax\n" , bdev); |
95 | return NULL; |
96 | } |
97 | |
98 | id = dax_read_lock(); |
99 | dax_dev = xa_load(&dax_hosts, index: (unsigned long)bdev->bd_disk); |
100 | if (!dax_dev || !dax_alive(dax_dev) || !igrab(&dax_dev->inode)) |
101 | dax_dev = NULL; |
102 | else if (holder) { |
103 | if (!cmpxchg(&dax_dev->holder_data, NULL, holder)) |
104 | dax_dev->holder_ops = ops; |
105 | else |
106 | dax_dev = NULL; |
107 | } |
108 | dax_read_unlock(id); |
109 | |
110 | return dax_dev; |
111 | } |
112 | EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev); |
113 | |
114 | void fs_put_dax(struct dax_device *dax_dev, void *holder) |
115 | { |
116 | if (dax_dev && holder && |
117 | cmpxchg(&dax_dev->holder_data, holder, NULL) == holder) |
118 | dax_dev->holder_ops = NULL; |
119 | put_dax(dax_dev); |
120 | } |
121 | EXPORT_SYMBOL_GPL(fs_put_dax); |
122 | #endif /* CONFIG_BLOCK && CONFIG_FS_DAX */ |
123 | |
124 | enum dax_device_flags { |
125 | /* !alive + rcu grace period == no new operations / mappings */ |
126 | DAXDEV_ALIVE, |
127 | /* gate whether dax_flush() calls the low level flush routine */ |
128 | DAXDEV_WRITE_CACHE, |
129 | /* flag to check if device supports synchronous flush */ |
130 | DAXDEV_SYNC, |
131 | /* do not leave the caches dirty after writes */ |
132 | DAXDEV_NOCACHE, |
133 | /* handle CPU fetch exceptions during reads */ |
134 | DAXDEV_NOMC, |
135 | }; |
136 | |
137 | /** |
138 | * dax_direct_access() - translate a device pgoff to an absolute pfn |
139 | * @dax_dev: a dax_device instance representing the logical memory range |
140 | * @pgoff: offset in pages from the start of the device to translate |
141 | * @nr_pages: number of consecutive pages caller can handle relative to @pfn |
142 | * @mode: indicator on normal access or recovery write |
143 | * @kaddr: output parameter that returns a virtual address mapping of pfn |
144 | * @pfn: output parameter that returns an absolute pfn translation of @pgoff |
145 | * |
146 | * Return: negative errno if an error occurs, otherwise the number of |
147 | * pages accessible at the device relative @pgoff. |
148 | */ |
149 | long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, |
150 | enum dax_access_mode mode, void **kaddr, pfn_t *pfn) |
151 | { |
152 | long avail; |
153 | |
154 | if (!dax_dev) |
155 | return -EOPNOTSUPP; |
156 | |
157 | if (!dax_alive(dax_dev)) |
158 | return -ENXIO; |
159 | |
160 | if (nr_pages < 0) |
161 | return -EINVAL; |
162 | |
163 | avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages, |
164 | mode, kaddr, pfn); |
165 | if (!avail) |
166 | return -ERANGE; |
167 | return min(avail, nr_pages); |
168 | } |
169 | EXPORT_SYMBOL_GPL(dax_direct_access); |
170 | |
171 | size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, |
172 | size_t bytes, struct iov_iter *i) |
173 | { |
174 | if (!dax_alive(dax_dev)) |
175 | return 0; |
176 | |
177 | /* |
178 | * The userspace address for the memory copy has already been validated |
179 | * via access_ok() in vfs_write, so use the 'no check' version to bypass |
180 | * the HARDENED_USERCOPY overhead. |
181 | */ |
182 | if (test_bit(DAXDEV_NOCACHE, &dax_dev->flags)) |
183 | return _copy_from_iter_flushcache(addr, bytes, i); |
184 | return _copy_from_iter(addr, bytes, i); |
185 | } |
186 | |
187 | size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, |
188 | size_t bytes, struct iov_iter *i) |
189 | { |
190 | if (!dax_alive(dax_dev)) |
191 | return 0; |
192 | |
193 | /* |
194 | * The userspace address for the memory copy has already been validated |
195 | * via access_ok() in vfs_red, so use the 'no check' version to bypass |
196 | * the HARDENED_USERCOPY overhead. |
197 | */ |
198 | if (test_bit(DAXDEV_NOMC, &dax_dev->flags)) |
199 | return _copy_mc_to_iter(addr, bytes, i); |
200 | return _copy_to_iter(addr, bytes, i); |
201 | } |
202 | |
203 | int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, |
204 | size_t nr_pages) |
205 | { |
206 | int ret; |
207 | |
208 | if (!dax_alive(dax_dev)) |
209 | return -ENXIO; |
210 | /* |
211 | * There are no callers that want to zero more than one page as of now. |
212 | * Once users are there, this check can be removed after the |
213 | * device mapper code has been updated to split ranges across targets. |
214 | */ |
215 | if (nr_pages != 1) |
216 | return -EIO; |
217 | |
218 | ret = dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages); |
219 | return dax_mem2blk_err(err: ret); |
220 | } |
221 | EXPORT_SYMBOL_GPL(dax_zero_page_range); |
222 | |
223 | size_t dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff, |
224 | void *addr, size_t bytes, struct iov_iter *iter) |
225 | { |
226 | if (!dax_dev->ops->recovery_write) |
227 | return 0; |
228 | return dax_dev->ops->recovery_write(dax_dev, pgoff, addr, bytes, iter); |
229 | } |
230 | EXPORT_SYMBOL_GPL(dax_recovery_write); |
231 | |
232 | int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off, |
233 | u64 len, int mf_flags) |
234 | { |
235 | int rc, id; |
236 | |
237 | id = dax_read_lock(); |
238 | if (!dax_alive(dax_dev)) { |
239 | rc = -ENXIO; |
240 | goto out; |
241 | } |
242 | |
243 | if (!dax_dev->holder_ops) { |
244 | rc = -EOPNOTSUPP; |
245 | goto out; |
246 | } |
247 | |
248 | rc = dax_dev->holder_ops->notify_failure(dax_dev, off, len, mf_flags); |
249 | out: |
250 | dax_read_unlock(id); |
251 | return rc; |
252 | } |
253 | EXPORT_SYMBOL_GPL(dax_holder_notify_failure); |
254 | |
255 | #ifdef CONFIG_ARCH_HAS_PMEM_API |
256 | void arch_wb_cache_pmem(void *addr, size_t size); |
257 | void dax_flush(struct dax_device *dax_dev, void *addr, size_t size) |
258 | { |
259 | if (unlikely(!dax_write_cache_enabled(dax_dev))) |
260 | return; |
261 | |
262 | arch_wb_cache_pmem(addr, size); |
263 | } |
264 | #else |
265 | void dax_flush(struct dax_device *dax_dev, void *addr, size_t size) |
266 | { |
267 | } |
268 | #endif |
269 | EXPORT_SYMBOL_GPL(dax_flush); |
270 | |
271 | void dax_write_cache(struct dax_device *dax_dev, bool wc) |
272 | { |
273 | if (wc) |
274 | set_bit(nr: DAXDEV_WRITE_CACHE, addr: &dax_dev->flags); |
275 | else |
276 | clear_bit(nr: DAXDEV_WRITE_CACHE, addr: &dax_dev->flags); |
277 | } |
278 | EXPORT_SYMBOL_GPL(dax_write_cache); |
279 | |
280 | bool dax_write_cache_enabled(struct dax_device *dax_dev) |
281 | { |
282 | return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); |
283 | } |
284 | EXPORT_SYMBOL_GPL(dax_write_cache_enabled); |
285 | |
286 | bool dax_synchronous(struct dax_device *dax_dev) |
287 | { |
288 | return test_bit(DAXDEV_SYNC, &dax_dev->flags); |
289 | } |
290 | EXPORT_SYMBOL_GPL(dax_synchronous); |
291 | |
292 | void set_dax_synchronous(struct dax_device *dax_dev) |
293 | { |
294 | set_bit(nr: DAXDEV_SYNC, addr: &dax_dev->flags); |
295 | } |
296 | EXPORT_SYMBOL_GPL(set_dax_synchronous); |
297 | |
298 | void set_dax_nocache(struct dax_device *dax_dev) |
299 | { |
300 | set_bit(nr: DAXDEV_NOCACHE, addr: &dax_dev->flags); |
301 | } |
302 | EXPORT_SYMBOL_GPL(set_dax_nocache); |
303 | |
304 | void set_dax_nomc(struct dax_device *dax_dev) |
305 | { |
306 | set_bit(nr: DAXDEV_NOMC, addr: &dax_dev->flags); |
307 | } |
308 | EXPORT_SYMBOL_GPL(set_dax_nomc); |
309 | |
310 | bool dax_alive(struct dax_device *dax_dev) |
311 | { |
312 | lockdep_assert_held(&dax_srcu); |
313 | return test_bit(DAXDEV_ALIVE, &dax_dev->flags); |
314 | } |
315 | EXPORT_SYMBOL_GPL(dax_alive); |
316 | |
317 | /* |
318 | * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring |
319 | * that any fault handlers or operations that might have seen |
320 | * dax_alive(), have completed. Any operations that start after |
321 | * synchronize_srcu() has run will abort upon seeing !dax_alive(). |
322 | */ |
323 | void kill_dax(struct dax_device *dax_dev) |
324 | { |
325 | if (!dax_dev) |
326 | return; |
327 | |
328 | if (dax_dev->holder_data != NULL) |
329 | dax_holder_notify_failure(dax_dev, 0, U64_MAX, 0); |
330 | |
331 | clear_bit(nr: DAXDEV_ALIVE, addr: &dax_dev->flags); |
332 | synchronize_srcu(ssp: &dax_srcu); |
333 | |
334 | /* clear holder data */ |
335 | dax_dev->holder_ops = NULL; |
336 | dax_dev->holder_data = NULL; |
337 | } |
338 | EXPORT_SYMBOL_GPL(kill_dax); |
339 | |
340 | void run_dax(struct dax_device *dax_dev) |
341 | { |
342 | set_bit(nr: DAXDEV_ALIVE, addr: &dax_dev->flags); |
343 | } |
344 | EXPORT_SYMBOL_GPL(run_dax); |
345 | |
346 | static struct inode *dax_alloc_inode(struct super_block *sb) |
347 | { |
348 | struct dax_device *dax_dev; |
349 | struct inode *inode; |
350 | |
351 | dax_dev = alloc_inode_sb(sb, cache: dax_cache, GFP_KERNEL); |
352 | if (!dax_dev) |
353 | return NULL; |
354 | |
355 | inode = &dax_dev->inode; |
356 | inode->i_rdev = 0; |
357 | return inode; |
358 | } |
359 | |
360 | static struct dax_device *to_dax_dev(struct inode *inode) |
361 | { |
362 | return container_of(inode, struct dax_device, inode); |
363 | } |
364 | |
365 | static void dax_free_inode(struct inode *inode) |
366 | { |
367 | struct dax_device *dax_dev = to_dax_dev(inode); |
368 | if (inode->i_rdev) |
369 | ida_free(&dax_minor_ida, id: iminor(inode)); |
370 | kmem_cache_free(s: dax_cache, objp: dax_dev); |
371 | } |
372 | |
373 | static void dax_destroy_inode(struct inode *inode) |
374 | { |
375 | struct dax_device *dax_dev = to_dax_dev(inode); |
376 | WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags), |
377 | "kill_dax() must be called before final iput()\n" ); |
378 | } |
379 | |
380 | static const struct super_operations dax_sops = { |
381 | .statfs = simple_statfs, |
382 | .alloc_inode = dax_alloc_inode, |
383 | .destroy_inode = dax_destroy_inode, |
384 | .free_inode = dax_free_inode, |
385 | .drop_inode = generic_delete_inode, |
386 | }; |
387 | |
388 | static int dax_init_fs_context(struct fs_context *fc) |
389 | { |
390 | struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC); |
391 | if (!ctx) |
392 | return -ENOMEM; |
393 | ctx->ops = &dax_sops; |
394 | return 0; |
395 | } |
396 | |
397 | static struct file_system_type dax_fs_type = { |
398 | .name = "dax" , |
399 | .init_fs_context = dax_init_fs_context, |
400 | .kill_sb = kill_anon_super, |
401 | }; |
402 | |
403 | static int dax_test(struct inode *inode, void *data) |
404 | { |
405 | dev_t devt = *(dev_t *) data; |
406 | |
407 | return inode->i_rdev == devt; |
408 | } |
409 | |
410 | static int dax_set(struct inode *inode, void *data) |
411 | { |
412 | dev_t devt = *(dev_t *) data; |
413 | |
414 | inode->i_rdev = devt; |
415 | return 0; |
416 | } |
417 | |
418 | static struct dax_device *dax_dev_get(dev_t devt) |
419 | { |
420 | struct dax_device *dax_dev; |
421 | struct inode *inode; |
422 | |
423 | inode = iget5_locked(dax_superblock, hash_32(val: devt + DAXFS_MAGIC, bits: 31), |
424 | test: dax_test, set: dax_set, &devt); |
425 | |
426 | if (!inode) |
427 | return NULL; |
428 | |
429 | dax_dev = to_dax_dev(inode); |
430 | if (inode->i_state & I_NEW) { |
431 | set_bit(nr: DAXDEV_ALIVE, addr: &dax_dev->flags); |
432 | inode->i_cdev = &dax_dev->cdev; |
433 | inode->i_mode = S_IFCHR; |
434 | inode->i_flags = S_DAX; |
435 | mapping_set_gfp_mask(m: &inode->i_data, GFP_USER); |
436 | unlock_new_inode(inode); |
437 | } |
438 | |
439 | return dax_dev; |
440 | } |
441 | |
442 | struct dax_device *alloc_dax(void *private, const struct dax_operations *ops) |
443 | { |
444 | struct dax_device *dax_dev; |
445 | dev_t devt; |
446 | int minor; |
447 | |
448 | if (WARN_ON_ONCE(ops && !ops->zero_page_range)) |
449 | return ERR_PTR(error: -EINVAL); |
450 | |
451 | minor = ida_alloc_max(ida: &dax_minor_ida, MINORMASK, GFP_KERNEL); |
452 | if (minor < 0) |
453 | return ERR_PTR(error: -ENOMEM); |
454 | |
455 | devt = MKDEV(MAJOR(dax_devt), minor); |
456 | dax_dev = dax_dev_get(devt); |
457 | if (!dax_dev) |
458 | goto err_dev; |
459 | |
460 | dax_dev->ops = ops; |
461 | dax_dev->private = private; |
462 | return dax_dev; |
463 | |
464 | err_dev: |
465 | ida_free(&dax_minor_ida, id: minor); |
466 | return ERR_PTR(error: -ENOMEM); |
467 | } |
468 | EXPORT_SYMBOL_GPL(alloc_dax); |
469 | |
470 | void put_dax(struct dax_device *dax_dev) |
471 | { |
472 | if (!dax_dev) |
473 | return; |
474 | iput(&dax_dev->inode); |
475 | } |
476 | EXPORT_SYMBOL_GPL(put_dax); |
477 | |
478 | /** |
479 | * dax_holder() - obtain the holder of a dax device |
480 | * @dax_dev: a dax_device instance |
481 | * |
482 | * Return: the holder's data which represents the holder if registered, |
483 | * otherwize NULL. |
484 | */ |
485 | void *dax_holder(struct dax_device *dax_dev) |
486 | { |
487 | return dax_dev->holder_data; |
488 | } |
489 | EXPORT_SYMBOL_GPL(dax_holder); |
490 | |
491 | /** |
492 | * inode_dax: convert a public inode into its dax_dev |
493 | * @inode: An inode with i_cdev pointing to a dax_dev |
494 | * |
495 | * Note this is not equivalent to to_dax_dev() which is for private |
496 | * internal use where we know the inode filesystem type == dax_fs_type. |
497 | */ |
498 | struct dax_device *inode_dax(struct inode *inode) |
499 | { |
500 | struct cdev *cdev = inode->i_cdev; |
501 | |
502 | return container_of(cdev, struct dax_device, cdev); |
503 | } |
504 | EXPORT_SYMBOL_GPL(inode_dax); |
505 | |
506 | struct inode *dax_inode(struct dax_device *dax_dev) |
507 | { |
508 | return &dax_dev->inode; |
509 | } |
510 | EXPORT_SYMBOL_GPL(dax_inode); |
511 | |
512 | void *dax_get_private(struct dax_device *dax_dev) |
513 | { |
514 | if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags)) |
515 | return NULL; |
516 | return dax_dev->private; |
517 | } |
518 | EXPORT_SYMBOL_GPL(dax_get_private); |
519 | |
520 | static void init_once(void *_dax_dev) |
521 | { |
522 | struct dax_device *dax_dev = _dax_dev; |
523 | struct inode *inode = &dax_dev->inode; |
524 | |
525 | memset(dax_dev, 0, sizeof(*dax_dev)); |
526 | inode_init_once(inode); |
527 | } |
528 | |
529 | static int dax_fs_init(void) |
530 | { |
531 | int rc; |
532 | |
533 | dax_cache = kmem_cache_create(name: "dax_cache" , size: sizeof(struct dax_device), align: 0, |
534 | flags: (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
535 | SLAB_MEM_SPREAD|SLAB_ACCOUNT), |
536 | ctor: init_once); |
537 | if (!dax_cache) |
538 | return -ENOMEM; |
539 | |
540 | dax_mnt = kern_mount(&dax_fs_type); |
541 | if (IS_ERR(ptr: dax_mnt)) { |
542 | rc = PTR_ERR(ptr: dax_mnt); |
543 | goto err_mount; |
544 | } |
545 | dax_superblock = dax_mnt->mnt_sb; |
546 | |
547 | return 0; |
548 | |
549 | err_mount: |
550 | kmem_cache_destroy(s: dax_cache); |
551 | |
552 | return rc; |
553 | } |
554 | |
555 | static void dax_fs_exit(void) |
556 | { |
557 | kern_unmount(mnt: dax_mnt); |
558 | rcu_barrier(); |
559 | kmem_cache_destroy(s: dax_cache); |
560 | } |
561 | |
562 | static int __init dax_core_init(void) |
563 | { |
564 | int rc; |
565 | |
566 | rc = dax_fs_init(); |
567 | if (rc) |
568 | return rc; |
569 | |
570 | rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax" ); |
571 | if (rc) |
572 | goto err_chrdev; |
573 | |
574 | rc = dax_bus_init(); |
575 | if (rc) |
576 | goto err_bus; |
577 | return 0; |
578 | |
579 | err_bus: |
580 | unregister_chrdev_region(dax_devt, MINORMASK+1); |
581 | err_chrdev: |
582 | dax_fs_exit(); |
583 | return 0; |
584 | } |
585 | |
586 | static void __exit dax_core_exit(void) |
587 | { |
588 | dax_bus_exit(); |
589 | unregister_chrdev_region(dax_devt, MINORMASK+1); |
590 | ida_destroy(ida: &dax_minor_ida); |
591 | dax_fs_exit(); |
592 | } |
593 | |
594 | MODULE_AUTHOR("Intel Corporation" ); |
595 | MODULE_LICENSE("GPL v2" ); |
596 | subsys_initcall(dax_core_init); |
597 | module_exit(dax_core_exit); |
598 | |