Loading...
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>
17#include <linux/genhd.h>
18#include <linux/pfn_t.h>
19#include <linux/cdev.h>
20#include <linux/hash.h>
21#include <linux/slab.h>
22#include <linux/uio.h>
23#include <linux/dax.h>
24#include <linux/fs.h>
25
26static dev_t dax_devt;
27DEFINE_STATIC_SRCU(dax_srcu);
28static struct vfsmount *dax_mnt;
29static DEFINE_IDA(dax_minor_ida);
30static struct kmem_cache *dax_cache __read_mostly;
31static struct super_block *dax_superblock __read_mostly;
32
33#define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
34static struct hlist_head dax_host_list[DAX_HASH_SIZE];
35static DEFINE_SPINLOCK(dax_host_lock);
36
37int dax_read_lock(void)
38{
39 return srcu_read_lock(&dax_srcu);
40}
41EXPORT_SYMBOL_GPL(dax_read_lock);
42
43void dax_read_unlock(int id)
44{
45 srcu_read_unlock(&dax_srcu, id);
46}
47EXPORT_SYMBOL_GPL(dax_read_unlock);
48
49#ifdef CONFIG_BLOCK
50#include <linux/blkdev.h>
51
52int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
53 pgoff_t *pgoff)
54{
55 phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512;
56
57 if (pgoff)
58 *pgoff = PHYS_PFN(phys_off);
59 if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
60 return -EINVAL;
61 return 0;
62}
63EXPORT_SYMBOL(bdev_dax_pgoff);
64
65#if IS_ENABLED(CONFIG_FS_DAX)
66struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
67{
68 if (!blk_queue_dax(bdev->bd_queue))
69 return NULL;
70 return fs_dax_get_by_host(bdev->bd_disk->disk_name);
71}
72EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
73#endif
74
75/**
76 * __bdev_dax_supported() - Check if the device supports dax for filesystem
77 * @sb: The superblock of the device
78 * @blocksize: The block size of the device
79 *
80 * This is a library function for filesystems to check if the block device
81 * can be mounted with dax option.
82 *
83 * Return: negative errno if unsupported, 0 if supported.
84 */
85int __bdev_dax_supported(struct super_block *sb, int blocksize)
86{
87 struct block_device *bdev = sb->s_bdev;
88 struct dax_device *dax_dev;
89 pgoff_t pgoff;
90 int err, id;
91 void *kaddr;
92 pfn_t pfn;
93 long len;
94
95 if (blocksize != PAGE_SIZE) {
96 pr_debug("VFS (%s): error: unsupported blocksize for dax\n",
97 sb->s_id);
98 return -EINVAL;
99 }
100
101 err = bdev_dax_pgoff(bdev, 0, PAGE_SIZE, &pgoff);
102 if (err) {
103 pr_debug("VFS (%s): error: unaligned partition for dax\n",
104 sb->s_id);
105 return err;
106 }
107
108 dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
109 if (!dax_dev) {
110 pr_debug("VFS (%s): error: device does not support dax\n",
111 sb->s_id);
112 return -EOPNOTSUPP;
113 }
114
115 id = dax_read_lock();
116 len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
117 dax_read_unlock(id);
118
119 put_dax(dax_dev);
120
121 if (len < 1) {
122 pr_debug("VFS (%s): error: dax access failed (%ld)\n",
123 sb->s_id, len);
124 return len < 0 ? len : -EIO;
125 }
126
127 if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
128 /*
129 * An arch that has enabled the pmem api should also
130 * have its drivers support pfn_t_devmap()
131 *
132 * This is a developer warning and should not trigger in
133 * production. dax_flush() will crash since it depends
134 * on being able to do (page_address(pfn_to_page())).
135 */
136 WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
137 } else if (pfn_t_devmap(pfn)) {
138 /* pass */;
139 } else {
140 pr_debug("VFS (%s): error: dax support not enabled\n",
141 sb->s_id);
142 return -EOPNOTSUPP;
143 }
144
145 return 0;
146}
147EXPORT_SYMBOL_GPL(__bdev_dax_supported);
148#endif
149
150enum dax_device_flags {
151 /* !alive + rcu grace period == no new operations / mappings */
152 DAXDEV_ALIVE,
153 /* gate whether dax_flush() calls the low level flush routine */
154 DAXDEV_WRITE_CACHE,
155};
156
157/**
158 * struct dax_device - anchor object for dax services
159 * @inode: core vfs
160 * @cdev: optional character interface for "device dax"
161 * @host: optional name for lookups where the device path is not available
162 * @private: dax driver private data
163 * @flags: state and boolean properties
164 */
165struct dax_device {
166 struct hlist_node list;
167 struct inode inode;
168 struct cdev cdev;
169 const char *host;
170 void *private;
171 unsigned long flags;
172 const struct dax_operations *ops;
173};
174
175static ssize_t write_cache_show(struct device *dev,
176 struct device_attribute *attr, char *buf)
177{
178 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
179 ssize_t rc;
180
181 WARN_ON_ONCE(!dax_dev);
182 if (!dax_dev)
183 return -ENXIO;
184
185 rc = sprintf(buf, "%d\n", !!test_bit(DAXDEV_WRITE_CACHE,
186 &dax_dev->flags));
187 put_dax(dax_dev);
188 return rc;
189}
190
191static ssize_t write_cache_store(struct device *dev,
192 struct device_attribute *attr, const char *buf, size_t len)
193{
194 bool write_cache;
195 int rc = strtobool(buf, &write_cache);
196 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
197
198 WARN_ON_ONCE(!dax_dev);
199 if (!dax_dev)
200 return -ENXIO;
201
202 if (rc)
203 len = rc;
204 else if (write_cache)
205 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
206 else
207 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
208
209 put_dax(dax_dev);
210 return len;
211}
212static DEVICE_ATTR_RW(write_cache);
213
214static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
215{
216 struct device *dev = container_of(kobj, typeof(*dev), kobj);
217 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
218
219 WARN_ON_ONCE(!dax_dev);
220 if (!dax_dev)
221 return 0;
222
223#ifndef CONFIG_ARCH_HAS_PMEM_API
224 if (a == &dev_attr_write_cache.attr)
225 return 0;
226#endif
227 return a->mode;
228}
229
230static struct attribute *dax_attributes[] = {
231 &dev_attr_write_cache.attr,
232 NULL,
233};
234
235struct attribute_group dax_attribute_group = {
236 .name = "dax",
237 .attrs = dax_attributes,
238 .is_visible = dax_visible,
239};
240EXPORT_SYMBOL_GPL(dax_attribute_group);
241
242/**
243 * dax_direct_access() - translate a device pgoff to an absolute pfn
244 * @dax_dev: a dax_device instance representing the logical memory range
245 * @pgoff: offset in pages from the start of the device to translate
246 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
247 * @kaddr: output parameter that returns a virtual address mapping of pfn
248 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
249 *
250 * Return: negative errno if an error occurs, otherwise the number of
251 * pages accessible at the device relative @pgoff.
252 */
253long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
254 void **kaddr, pfn_t *pfn)
255{
256 long avail;
257
258 if (!dax_dev)
259 return -EOPNOTSUPP;
260
261 if (!dax_alive(dax_dev))
262 return -ENXIO;
263
264 if (nr_pages < 0)
265 return nr_pages;
266
267 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
268 kaddr, pfn);
269 if (!avail)
270 return -ERANGE;
271 return min(avail, nr_pages);
272}
273EXPORT_SYMBOL_GPL(dax_direct_access);
274
275size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
276 size_t bytes, struct iov_iter *i)
277{
278 if (!dax_alive(dax_dev))
279 return 0;
280
281 return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
282}
283EXPORT_SYMBOL_GPL(dax_copy_from_iter);
284
285#ifdef CONFIG_ARCH_HAS_PMEM_API
286void arch_wb_cache_pmem(void *addr, size_t size);
287void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
288{
289 if (unlikely(!test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags)))
290 return;
291
292 arch_wb_cache_pmem(addr, size);
293}
294#else
295void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
296{
297}
298#endif
299EXPORT_SYMBOL_GPL(dax_flush);
300
301void dax_write_cache(struct dax_device *dax_dev, bool wc)
302{
303 if (wc)
304 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
305 else
306 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
307}
308EXPORT_SYMBOL_GPL(dax_write_cache);
309
310bool dax_write_cache_enabled(struct dax_device *dax_dev)
311{
312 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
313}
314EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
315
316bool dax_alive(struct dax_device *dax_dev)
317{
318 lockdep_assert_held(&dax_srcu);
319 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
320}
321EXPORT_SYMBOL_GPL(dax_alive);
322
323static int dax_host_hash(const char *host)
324{
325 return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
326}
327
328/*
329 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
330 * that any fault handlers or operations that might have seen
331 * dax_alive(), have completed. Any operations that start after
332 * synchronize_srcu() has run will abort upon seeing !dax_alive().
333 */
334void kill_dax(struct dax_device *dax_dev)
335{
336 if (!dax_dev)
337 return;
338
339 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
340
341 synchronize_srcu(&dax_srcu);
342
343 spin_lock(&dax_host_lock);
344 hlist_del_init(&dax_dev->list);
345 spin_unlock(&dax_host_lock);
346
347 dax_dev->private = NULL;
348}
349EXPORT_SYMBOL_GPL(kill_dax);
350
351static struct inode *dax_alloc_inode(struct super_block *sb)
352{
353 struct dax_device *dax_dev;
354 struct inode *inode;
355
356 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
357 if (!dax_dev)
358 return NULL;
359
360 inode = &dax_dev->inode;
361 inode->i_rdev = 0;
362 return inode;
363}
364
365static struct dax_device *to_dax_dev(struct inode *inode)
366{
367 return container_of(inode, struct dax_device, inode);
368}
369
370static void dax_i_callback(struct rcu_head *head)
371{
372 struct inode *inode = container_of(head, struct inode, i_rcu);
373 struct dax_device *dax_dev = to_dax_dev(inode);
374
375 kfree(dax_dev->host);
376 dax_dev->host = NULL;
377 if (inode->i_rdev)
378 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
379 kmem_cache_free(dax_cache, dax_dev);
380}
381
382static void dax_destroy_inode(struct inode *inode)
383{
384 struct dax_device *dax_dev = to_dax_dev(inode);
385
386 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
387 "kill_dax() must be called before final iput()\n");
388 call_rcu(&inode->i_rcu, dax_i_callback);
389}
390
391static const struct super_operations dax_sops = {
392 .statfs = simple_statfs,
393 .alloc_inode = dax_alloc_inode,
394 .destroy_inode = dax_destroy_inode,
395 .drop_inode = generic_delete_inode,
396};
397
398static struct dentry *dax_mount(struct file_system_type *fs_type,
399 int flags, const char *dev_name, void *data)
400{
401 return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
402}
403
404static struct file_system_type dax_fs_type = {
405 .name = "dax",
406 .mount = dax_mount,
407 .kill_sb = kill_anon_super,
408};
409
410static int dax_test(struct inode *inode, void *data)
411{
412 dev_t devt = *(dev_t *) data;
413
414 return inode->i_rdev == devt;
415}
416
417static int dax_set(struct inode *inode, void *data)
418{
419 dev_t devt = *(dev_t *) data;
420
421 inode->i_rdev = devt;
422 return 0;
423}
424
425static struct dax_device *dax_dev_get(dev_t devt)
426{
427 struct dax_device *dax_dev;
428 struct inode *inode;
429
430 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
431 dax_test, dax_set, &devt);
432
433 if (!inode)
434 return NULL;
435
436 dax_dev = to_dax_dev(inode);
437 if (inode->i_state & I_NEW) {
438 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
439 inode->i_cdev = &dax_dev->cdev;
440 inode->i_mode = S_IFCHR;
441 inode->i_flags = S_DAX;
442 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
443 unlock_new_inode(inode);
444 }
445
446 return dax_dev;
447}
448
449static void dax_add_host(struct dax_device *dax_dev, const char *host)
450{
451 int hash;
452
453 /*
454 * Unconditionally init dax_dev since it's coming from a
455 * non-zeroed slab cache
456 */
457 INIT_HLIST_NODE(&dax_dev->list);
458 dax_dev->host = host;
459 if (!host)
460 return;
461
462 hash = dax_host_hash(host);
463 spin_lock(&dax_host_lock);
464 hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
465 spin_unlock(&dax_host_lock);
466}
467
468struct dax_device *alloc_dax(void *private, const char *__host,
469 const struct dax_operations *ops)
470{
471 struct dax_device *dax_dev;
472 const char *host;
473 dev_t devt;
474 int minor;
475
476 host = kstrdup(__host, GFP_KERNEL);
477 if (__host && !host)
478 return NULL;
479
480 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
481 if (minor < 0)
482 goto err_minor;
483
484 devt = MKDEV(MAJOR(dax_devt), minor);
485 dax_dev = dax_dev_get(devt);
486 if (!dax_dev)
487 goto err_dev;
488
489 dax_add_host(dax_dev, host);
490 dax_dev->ops = ops;
491 dax_dev->private = private;
492 return dax_dev;
493
494 err_dev:
495 ida_simple_remove(&dax_minor_ida, minor);
496 err_minor:
497 kfree(host);
498 return NULL;
499}
500EXPORT_SYMBOL_GPL(alloc_dax);
501
502void put_dax(struct dax_device *dax_dev)
503{
504 if (!dax_dev)
505 return;
506 iput(&dax_dev->inode);
507}
508EXPORT_SYMBOL_GPL(put_dax);
509
510/**
511 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
512 * @host: alternate name for the device registered by a dax driver
513 */
514struct dax_device *dax_get_by_host(const char *host)
515{
516 struct dax_device *dax_dev, *found = NULL;
517 int hash, id;
518
519 if (!host)
520 return NULL;
521
522 hash = dax_host_hash(host);
523
524 id = dax_read_lock();
525 spin_lock(&dax_host_lock);
526 hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
527 if (!dax_alive(dax_dev)
528 || strcmp(host, dax_dev->host) != 0)
529 continue;
530
531 if (igrab(&dax_dev->inode))
532 found = dax_dev;
533 break;
534 }
535 spin_unlock(&dax_host_lock);
536 dax_read_unlock(id);
537
538 return found;
539}
540EXPORT_SYMBOL_GPL(dax_get_by_host);
541
542/**
543 * inode_dax: convert a public inode into its dax_dev
544 * @inode: An inode with i_cdev pointing to a dax_dev
545 *
546 * Note this is not equivalent to to_dax_dev() which is for private
547 * internal use where we know the inode filesystem type == dax_fs_type.
548 */
549struct dax_device *inode_dax(struct inode *inode)
550{
551 struct cdev *cdev = inode->i_cdev;
552
553 return container_of(cdev, struct dax_device, cdev);
554}
555EXPORT_SYMBOL_GPL(inode_dax);
556
557struct inode *dax_inode(struct dax_device *dax_dev)
558{
559 return &dax_dev->inode;
560}
561EXPORT_SYMBOL_GPL(dax_inode);
562
563void *dax_get_private(struct dax_device *dax_dev)
564{
565 return dax_dev->private;
566}
567EXPORT_SYMBOL_GPL(dax_get_private);
568
569static void init_once(void *_dax_dev)
570{
571 struct dax_device *dax_dev = _dax_dev;
572 struct inode *inode = &dax_dev->inode;
573
574 memset(dax_dev, 0, sizeof(*dax_dev));
575 inode_init_once(inode);
576}
577
578static int __dax_fs_init(void)
579{
580 int rc;
581
582 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
583 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
584 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
585 init_once);
586 if (!dax_cache)
587 return -ENOMEM;
588
589 rc = register_filesystem(&dax_fs_type);
590 if (rc)
591 goto err_register_fs;
592
593 dax_mnt = kern_mount(&dax_fs_type);
594 if (IS_ERR(dax_mnt)) {
595 rc = PTR_ERR(dax_mnt);
596 goto err_mount;
597 }
598 dax_superblock = dax_mnt->mnt_sb;
599
600 return 0;
601
602 err_mount:
603 unregister_filesystem(&dax_fs_type);
604 err_register_fs:
605 kmem_cache_destroy(dax_cache);
606
607 return rc;
608}
609
610static void __dax_fs_exit(void)
611{
612 kern_unmount(dax_mnt);
613 unregister_filesystem(&dax_fs_type);
614 kmem_cache_destroy(dax_cache);
615}
616
617static int __init dax_fs_init(void)
618{
619 int rc;
620
621 rc = __dax_fs_init();
622 if (rc)
623 return rc;
624
625 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
626 if (rc)
627 __dax_fs_exit();
628 return rc;
629}
630
631static void __exit dax_fs_exit(void)
632{
633 unregister_chrdev_region(dax_devt, MINORMASK+1);
634 ida_destroy(&dax_minor_ida);
635 __dax_fs_exit();
636}
637
638MODULE_AUTHOR("Intel Corporation");
639MODULE_LICENSE("GPL v2");
640subsys_initcall(dax_fs_init);
641module_exit(dax_fs_exit);
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 <linux/cacheinfo.h>
17#include "dax-private.h"
18
19/**
20 * struct dax_device - anchor object for dax services
21 * @inode: core vfs
22 * @cdev: optional character interface for "device dax"
23 * @private: dax driver private data
24 * @flags: state and boolean properties
25 * @ops: operations for this device
26 * @holder_data: holder of a dax_device: could be filesystem or mapped device
27 * @holder_ops: operations for the inner holder
28 */
29struct dax_device {
30 struct inode inode;
31 struct cdev cdev;
32 void *private;
33 unsigned long flags;
34 const struct dax_operations *ops;
35 void *holder_data;
36 const struct dax_holder_operations *holder_ops;
37};
38
39static dev_t dax_devt;
40DEFINE_STATIC_SRCU(dax_srcu);
41static struct vfsmount *dax_mnt;
42static DEFINE_IDA(dax_minor_ida);
43static struct kmem_cache *dax_cache __read_mostly;
44static struct super_block *dax_superblock __read_mostly;
45
46int dax_read_lock(void)
47{
48 return srcu_read_lock(&dax_srcu);
49}
50EXPORT_SYMBOL_GPL(dax_read_lock);
51
52void dax_read_unlock(int id)
53{
54 srcu_read_unlock(&dax_srcu, id);
55}
56EXPORT_SYMBOL_GPL(dax_read_unlock);
57
58#if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX)
59#include <linux/blkdev.h>
60
61static DEFINE_XARRAY(dax_hosts);
62
63int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk)
64{
65 return xa_insert(&dax_hosts, (unsigned long)disk, dax_dev, GFP_KERNEL);
66}
67EXPORT_SYMBOL_GPL(dax_add_host);
68
69void dax_remove_host(struct gendisk *disk)
70{
71 xa_erase(&dax_hosts, (unsigned long)disk);
72}
73EXPORT_SYMBOL_GPL(dax_remove_host);
74
75/**
76 * fs_dax_get_by_bdev() - temporary lookup mechanism for filesystem-dax
77 * @bdev: block device to find a dax_device for
78 * @start_off: returns the byte offset into the dax_device that @bdev starts
79 * @holder: filesystem or mapped device inside the dax_device
80 * @ops: operations for the inner holder
81 */
82struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off,
83 void *holder, const struct dax_holder_operations *ops)
84{
85 struct dax_device *dax_dev;
86 u64 part_size;
87 int id;
88
89 if (!blk_queue_dax(bdev->bd_disk->queue))
90 return NULL;
91
92 *start_off = get_start_sect(bdev) * SECTOR_SIZE;
93 part_size = bdev_nr_sectors(bdev) * SECTOR_SIZE;
94 if (*start_off % PAGE_SIZE || part_size % PAGE_SIZE) {
95 pr_info("%pg: error: unaligned partition for dax\n", bdev);
96 return NULL;
97 }
98
99 id = dax_read_lock();
100 dax_dev = xa_load(&dax_hosts, (unsigned long)bdev->bd_disk);
101 if (!dax_dev || !dax_alive(dax_dev) || !igrab(&dax_dev->inode))
102 dax_dev = NULL;
103 else if (holder) {
104 if (!cmpxchg(&dax_dev->holder_data, NULL, holder))
105 dax_dev->holder_ops = ops;
106 else
107 dax_dev = NULL;
108 }
109 dax_read_unlock(id);
110
111 return dax_dev;
112}
113EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
114
115void fs_put_dax(struct dax_device *dax_dev, void *holder)
116{
117 if (dax_dev && holder &&
118 cmpxchg(&dax_dev->holder_data, holder, NULL) == holder)
119 dax_dev->holder_ops = NULL;
120 put_dax(dax_dev);
121}
122EXPORT_SYMBOL_GPL(fs_put_dax);
123#endif /* CONFIG_BLOCK && CONFIG_FS_DAX */
124
125enum dax_device_flags {
126 /* !alive + rcu grace period == no new operations / mappings */
127 DAXDEV_ALIVE,
128 /* gate whether dax_flush() calls the low level flush routine */
129 DAXDEV_WRITE_CACHE,
130 /* flag to check if device supports synchronous flush */
131 DAXDEV_SYNC,
132 /* do not leave the caches dirty after writes */
133 DAXDEV_NOCACHE,
134 /* handle CPU fetch exceptions during reads */
135 DAXDEV_NOMC,
136};
137
138/**
139 * dax_direct_access() - translate a device pgoff to an absolute pfn
140 * @dax_dev: a dax_device instance representing the logical memory range
141 * @pgoff: offset in pages from the start of the device to translate
142 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
143 * @mode: indicator on normal access or recovery write
144 * @kaddr: output parameter that returns a virtual address mapping of pfn
145 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
146 *
147 * Return: negative errno if an error occurs, otherwise the number of
148 * pages accessible at the device relative @pgoff.
149 */
150long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
151 enum dax_access_mode mode, void **kaddr, pfn_t *pfn)
152{
153 long avail;
154
155 if (!dax_dev)
156 return -EOPNOTSUPP;
157
158 if (!dax_alive(dax_dev))
159 return -ENXIO;
160
161 if (nr_pages < 0)
162 return -EINVAL;
163
164 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
165 mode, kaddr, pfn);
166 if (!avail)
167 return -ERANGE;
168 return min(avail, nr_pages);
169}
170EXPORT_SYMBOL_GPL(dax_direct_access);
171
172size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
173 size_t bytes, struct iov_iter *i)
174{
175 if (!dax_alive(dax_dev))
176 return 0;
177
178 /*
179 * The userspace address for the memory copy has already been validated
180 * via access_ok() in vfs_write, so use the 'no check' version to bypass
181 * the HARDENED_USERCOPY overhead.
182 */
183 if (test_bit(DAXDEV_NOCACHE, &dax_dev->flags))
184 return _copy_from_iter_flushcache(addr, bytes, i);
185 return _copy_from_iter(addr, bytes, i);
186}
187
188size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
189 size_t bytes, struct iov_iter *i)
190{
191 if (!dax_alive(dax_dev))
192 return 0;
193
194 /*
195 * The userspace address for the memory copy has already been validated
196 * via access_ok() in vfs_red, so use the 'no check' version to bypass
197 * the HARDENED_USERCOPY overhead.
198 */
199 if (test_bit(DAXDEV_NOMC, &dax_dev->flags))
200 return _copy_mc_to_iter(addr, bytes, i);
201 return _copy_to_iter(addr, bytes, i);
202}
203
204int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
205 size_t nr_pages)
206{
207 int ret;
208
209 if (!dax_alive(dax_dev))
210 return -ENXIO;
211 /*
212 * There are no callers that want to zero more than one page as of now.
213 * Once users are there, this check can be removed after the
214 * device mapper code has been updated to split ranges across targets.
215 */
216 if (nr_pages != 1)
217 return -EIO;
218
219 ret = dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages);
220 return dax_mem2blk_err(ret);
221}
222EXPORT_SYMBOL_GPL(dax_zero_page_range);
223
224size_t dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
225 void *addr, size_t bytes, struct iov_iter *iter)
226{
227 if (!dax_dev->ops->recovery_write)
228 return 0;
229 return dax_dev->ops->recovery_write(dax_dev, pgoff, addr, bytes, iter);
230}
231EXPORT_SYMBOL_GPL(dax_recovery_write);
232
233int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off,
234 u64 len, int mf_flags)
235{
236 int rc, id;
237
238 id = dax_read_lock();
239 if (!dax_alive(dax_dev)) {
240 rc = -ENXIO;
241 goto out;
242 }
243
244 if (!dax_dev->holder_ops) {
245 rc = -EOPNOTSUPP;
246 goto out;
247 }
248
249 rc = dax_dev->holder_ops->notify_failure(dax_dev, off, len, mf_flags);
250out:
251 dax_read_unlock(id);
252 return rc;
253}
254EXPORT_SYMBOL_GPL(dax_holder_notify_failure);
255
256#ifdef CONFIG_ARCH_HAS_PMEM_API
257void arch_wb_cache_pmem(void *addr, size_t size);
258void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
259{
260 if (unlikely(!dax_write_cache_enabled(dax_dev)))
261 return;
262
263 arch_wb_cache_pmem(addr, size);
264}
265#else
266void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
267{
268}
269#endif
270EXPORT_SYMBOL_GPL(dax_flush);
271
272void dax_write_cache(struct dax_device *dax_dev, bool wc)
273{
274 if (wc)
275 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
276 else
277 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
278}
279EXPORT_SYMBOL_GPL(dax_write_cache);
280
281bool dax_write_cache_enabled(struct dax_device *dax_dev)
282{
283 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
284}
285EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
286
287bool dax_synchronous(struct dax_device *dax_dev)
288{
289 return test_bit(DAXDEV_SYNC, &dax_dev->flags);
290}
291EXPORT_SYMBOL_GPL(dax_synchronous);
292
293void set_dax_synchronous(struct dax_device *dax_dev)
294{
295 set_bit(DAXDEV_SYNC, &dax_dev->flags);
296}
297EXPORT_SYMBOL_GPL(set_dax_synchronous);
298
299void set_dax_nocache(struct dax_device *dax_dev)
300{
301 set_bit(DAXDEV_NOCACHE, &dax_dev->flags);
302}
303EXPORT_SYMBOL_GPL(set_dax_nocache);
304
305void set_dax_nomc(struct dax_device *dax_dev)
306{
307 set_bit(DAXDEV_NOMC, &dax_dev->flags);
308}
309EXPORT_SYMBOL_GPL(set_dax_nomc);
310
311bool dax_alive(struct dax_device *dax_dev)
312{
313 lockdep_assert_held(&dax_srcu);
314 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
315}
316EXPORT_SYMBOL_GPL(dax_alive);
317
318/*
319 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
320 * that any fault handlers or operations that might have seen
321 * dax_alive(), have completed. Any operations that start after
322 * synchronize_srcu() has run will abort upon seeing !dax_alive().
323 *
324 * Note, because alloc_dax() returns an ERR_PTR() on error, callers
325 * typically store its result into a local variable in order to check
326 * the result. Therefore, care must be taken to populate the struct
327 * device dax_dev field make sure the dax_dev is not leaked.
328 */
329void kill_dax(struct dax_device *dax_dev)
330{
331 if (!dax_dev)
332 return;
333
334 if (dax_dev->holder_data != NULL)
335 dax_holder_notify_failure(dax_dev, 0, U64_MAX,
336 MF_MEM_PRE_REMOVE);
337
338 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
339 synchronize_srcu(&dax_srcu);
340
341 /* clear holder data */
342 dax_dev->holder_ops = NULL;
343 dax_dev->holder_data = NULL;
344}
345EXPORT_SYMBOL_GPL(kill_dax);
346
347void run_dax(struct dax_device *dax_dev)
348{
349 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
350}
351EXPORT_SYMBOL_GPL(run_dax);
352
353static struct inode *dax_alloc_inode(struct super_block *sb)
354{
355 struct dax_device *dax_dev;
356 struct inode *inode;
357
358 dax_dev = alloc_inode_sb(sb, dax_cache, GFP_KERNEL);
359 if (!dax_dev)
360 return NULL;
361
362 inode = &dax_dev->inode;
363 inode->i_rdev = 0;
364 return inode;
365}
366
367static struct dax_device *to_dax_dev(struct inode *inode)
368{
369 return container_of(inode, struct dax_device, inode);
370}
371
372static void dax_free_inode(struct inode *inode)
373{
374 struct dax_device *dax_dev = to_dax_dev(inode);
375 if (inode->i_rdev)
376 ida_free(&dax_minor_ida, iminor(inode));
377 kmem_cache_free(dax_cache, dax_dev);
378}
379
380static void dax_destroy_inode(struct inode *inode)
381{
382 struct dax_device *dax_dev = to_dax_dev(inode);
383 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
384 "kill_dax() must be called before final iput()\n");
385}
386
387static const struct super_operations dax_sops = {
388 .statfs = simple_statfs,
389 .alloc_inode = dax_alloc_inode,
390 .destroy_inode = dax_destroy_inode,
391 .free_inode = dax_free_inode,
392 .drop_inode = generic_delete_inode,
393};
394
395static int dax_init_fs_context(struct fs_context *fc)
396{
397 struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC);
398 if (!ctx)
399 return -ENOMEM;
400 ctx->ops = &dax_sops;
401 return 0;
402}
403
404static struct file_system_type dax_fs_type = {
405 .name = "dax",
406 .init_fs_context = dax_init_fs_context,
407 .kill_sb = kill_anon_super,
408};
409
410static int dax_test(struct inode *inode, void *data)
411{
412 dev_t devt = *(dev_t *) data;
413
414 return inode->i_rdev == devt;
415}
416
417static int dax_set(struct inode *inode, void *data)
418{
419 dev_t devt = *(dev_t *) data;
420
421 inode->i_rdev = devt;
422 return 0;
423}
424
425static struct dax_device *dax_dev_get(dev_t devt)
426{
427 struct dax_device *dax_dev;
428 struct inode *inode;
429
430 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
431 dax_test, dax_set, &devt);
432
433 if (!inode)
434 return NULL;
435
436 dax_dev = to_dax_dev(inode);
437 if (inode->i_state & I_NEW) {
438 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
439 inode->i_cdev = &dax_dev->cdev;
440 inode->i_mode = S_IFCHR;
441 inode->i_flags = S_DAX;
442 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
443 unlock_new_inode(inode);
444 }
445
446 return dax_dev;
447}
448
449struct dax_device *alloc_dax(void *private, const struct dax_operations *ops)
450{
451 struct dax_device *dax_dev;
452 dev_t devt;
453 int minor;
454
455 /*
456 * Unavailable on architectures with virtually aliased data caches,
457 * except for device-dax (NULL operations pointer), which does
458 * not use aliased mappings from the kernel.
459 */
460 if (ops && cpu_dcache_is_aliasing())
461 return ERR_PTR(-EOPNOTSUPP);
462
463 if (WARN_ON_ONCE(ops && !ops->zero_page_range))
464 return ERR_PTR(-EINVAL);
465
466 minor = ida_alloc_max(&dax_minor_ida, MINORMASK, GFP_KERNEL);
467 if (minor < 0)
468 return ERR_PTR(-ENOMEM);
469
470 devt = MKDEV(MAJOR(dax_devt), minor);
471 dax_dev = dax_dev_get(devt);
472 if (!dax_dev)
473 goto err_dev;
474
475 dax_dev->ops = ops;
476 dax_dev->private = private;
477 return dax_dev;
478
479 err_dev:
480 ida_free(&dax_minor_ida, minor);
481 return ERR_PTR(-ENOMEM);
482}
483EXPORT_SYMBOL_GPL(alloc_dax);
484
485void put_dax(struct dax_device *dax_dev)
486{
487 if (!dax_dev)
488 return;
489 iput(&dax_dev->inode);
490}
491EXPORT_SYMBOL_GPL(put_dax);
492
493/**
494 * dax_holder() - obtain the holder of a dax device
495 * @dax_dev: a dax_device instance
496 *
497 * Return: the holder's data which represents the holder if registered,
498 * otherwize NULL.
499 */
500void *dax_holder(struct dax_device *dax_dev)
501{
502 return dax_dev->holder_data;
503}
504EXPORT_SYMBOL_GPL(dax_holder);
505
506/**
507 * inode_dax: convert a public inode into its dax_dev
508 * @inode: An inode with i_cdev pointing to a dax_dev
509 *
510 * Note this is not equivalent to to_dax_dev() which is for private
511 * internal use where we know the inode filesystem type == dax_fs_type.
512 */
513struct dax_device *inode_dax(struct inode *inode)
514{
515 struct cdev *cdev = inode->i_cdev;
516
517 return container_of(cdev, struct dax_device, cdev);
518}
519EXPORT_SYMBOL_GPL(inode_dax);
520
521struct inode *dax_inode(struct dax_device *dax_dev)
522{
523 return &dax_dev->inode;
524}
525EXPORT_SYMBOL_GPL(dax_inode);
526
527void *dax_get_private(struct dax_device *dax_dev)
528{
529 if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
530 return NULL;
531 return dax_dev->private;
532}
533EXPORT_SYMBOL_GPL(dax_get_private);
534
535static void init_once(void *_dax_dev)
536{
537 struct dax_device *dax_dev = _dax_dev;
538 struct inode *inode = &dax_dev->inode;
539
540 memset(dax_dev, 0, sizeof(*dax_dev));
541 inode_init_once(inode);
542}
543
544static int dax_fs_init(void)
545{
546 int rc;
547
548 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
549 SLAB_HWCACHE_ALIGN | SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
550 init_once);
551 if (!dax_cache)
552 return -ENOMEM;
553
554 dax_mnt = kern_mount(&dax_fs_type);
555 if (IS_ERR(dax_mnt)) {
556 rc = PTR_ERR(dax_mnt);
557 goto err_mount;
558 }
559 dax_superblock = dax_mnt->mnt_sb;
560
561 return 0;
562
563 err_mount:
564 kmem_cache_destroy(dax_cache);
565
566 return rc;
567}
568
569static void dax_fs_exit(void)
570{
571 kern_unmount(dax_mnt);
572 rcu_barrier();
573 kmem_cache_destroy(dax_cache);
574}
575
576static int __init dax_core_init(void)
577{
578 int rc;
579
580 rc = dax_fs_init();
581 if (rc)
582 return rc;
583
584 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
585 if (rc)
586 goto err_chrdev;
587
588 rc = dax_bus_init();
589 if (rc)
590 goto err_bus;
591 return 0;
592
593err_bus:
594 unregister_chrdev_region(dax_devt, MINORMASK+1);
595err_chrdev:
596 dax_fs_exit();
597 return 0;
598}
599
600static void __exit dax_core_exit(void)
601{
602 dax_bus_exit();
603 unregister_chrdev_region(dax_devt, MINORMASK+1);
604 ida_destroy(&dax_minor_ida);
605 dax_fs_exit();
606}
607
608MODULE_AUTHOR("Intel Corporation");
609MODULE_DESCRIPTION("DAX: direct access to differentiated memory");
610MODULE_LICENSE("GPL v2");
611subsys_initcall(dax_core_init);
612module_exit(dax_core_exit);