2 * Copyright(c) 2017 Intel Corporation. All rights reserved.
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.
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.
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>
26 static dev_t dax_devt
;
27 DEFINE_STATIC_SRCU(dax_srcu
);
28 static struct vfsmount
*dax_mnt
;
29 static DEFINE_IDA(dax_minor_ida
);
30 static struct kmem_cache
*dax_cache __read_mostly
;
31 static struct super_block
*dax_superblock __read_mostly
;
33 #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
34 static struct hlist_head dax_host_list
[DAX_HASH_SIZE
];
35 static DEFINE_SPINLOCK(dax_host_lock
);
37 int dax_read_lock(void)
39 return srcu_read_lock(&dax_srcu
);
41 EXPORT_SYMBOL_GPL(dax_read_lock
);
43 void dax_read_unlock(int id
)
45 srcu_read_unlock(&dax_srcu
, id
);
47 EXPORT_SYMBOL_GPL(dax_read_unlock
);
50 #include <linux/blkdev.h>
52 int bdev_dax_pgoff(struct block_device
*bdev
, sector_t sector
, size_t size
,
55 phys_addr_t phys_off
= (get_start_sect(bdev
) + sector
) * 512;
58 *pgoff
= PHYS_PFN(phys_off
);
59 if (phys_off
% PAGE_SIZE
|| size
% PAGE_SIZE
)
63 EXPORT_SYMBOL(bdev_dax_pgoff
);
65 #if IS_ENABLED(CONFIG_FS_DAX)
66 struct dax_device
*fs_dax_get_by_bdev(struct block_device
*bdev
)
68 if (!blk_queue_dax(bdev
->bd_queue
))
70 return fs_dax_get_by_host(bdev
->bd_disk
->disk_name
);
72 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev
);
76 * __bdev_dax_supported() - Check if the device supports dax for filesystem
77 * @bdev: block device to check
78 * @blocksize: The block size of the device
80 * This is a library function for filesystems to check if the block device
81 * can be mounted with dax option.
83 * Return: true if supported, false if unsupported
85 bool __bdev_dax_supported(struct block_device
*bdev
, int blocksize
)
87 struct dax_device
*dax_dev
;
88 struct request_queue
*q
;
94 char buf
[BDEVNAME_SIZE
];
96 if (blocksize
!= PAGE_SIZE
) {
97 pr_debug("%s: error: unsupported blocksize for dax\n",
102 q
= bdev_get_queue(bdev
);
103 if (!q
|| !blk_queue_dax(q
)) {
104 pr_debug("%s: error: request queue doesn't support dax\n",
105 bdevname(bdev
, buf
));
109 err
= bdev_dax_pgoff(bdev
, 0, PAGE_SIZE
, &pgoff
);
111 pr_debug("%s: error: unaligned partition for dax\n",
112 bdevname(bdev
, buf
));
116 dax_dev
= dax_get_by_host(bdev
->bd_disk
->disk_name
);
118 pr_debug("%s: error: device does not support dax\n",
119 bdevname(bdev
, buf
));
123 id
= dax_read_lock();
124 len
= dax_direct_access(dax_dev
, pgoff
, 1, &kaddr
, &pfn
);
130 pr_debug("%s: error: dax access failed (%ld)\n",
131 bdevname(bdev
, buf
), len
);
135 if ((IS_ENABLED(CONFIG_FS_DAX_LIMITED
) && pfn_t_special(pfn
))
136 || pfn_t_devmap(pfn
))
139 pr_debug("%s: error: dax support not enabled\n",
140 bdevname(bdev
, buf
));
146 EXPORT_SYMBOL_GPL(__bdev_dax_supported
);
149 enum dax_device_flags
{
150 /* !alive + rcu grace period == no new operations / mappings */
152 /* gate whether dax_flush() calls the low level flush routine */
157 * struct dax_device - anchor object for dax services
159 * @cdev: optional character interface for "device dax"
160 * @host: optional name for lookups where the device path is not available
161 * @private: dax driver private data
162 * @flags: state and boolean properties
165 struct hlist_node list
;
171 const struct dax_operations
*ops
;
174 static ssize_t
write_cache_show(struct device
*dev
,
175 struct device_attribute
*attr
, char *buf
)
177 struct dax_device
*dax_dev
= dax_get_by_host(dev_name(dev
));
180 WARN_ON_ONCE(!dax_dev
);
184 rc
= sprintf(buf
, "%d\n", !!test_bit(DAXDEV_WRITE_CACHE
,
190 static ssize_t
write_cache_store(struct device
*dev
,
191 struct device_attribute
*attr
, const char *buf
, size_t len
)
194 int rc
= strtobool(buf
, &write_cache
);
195 struct dax_device
*dax_dev
= dax_get_by_host(dev_name(dev
));
197 WARN_ON_ONCE(!dax_dev
);
203 else if (write_cache
)
204 set_bit(DAXDEV_WRITE_CACHE
, &dax_dev
->flags
);
206 clear_bit(DAXDEV_WRITE_CACHE
, &dax_dev
->flags
);
211 static DEVICE_ATTR_RW(write_cache
);
213 static umode_t
dax_visible(struct kobject
*kobj
, struct attribute
*a
, int n
)
215 struct device
*dev
= container_of(kobj
, typeof(*dev
), kobj
);
216 struct dax_device
*dax_dev
= dax_get_by_host(dev_name(dev
));
218 WARN_ON_ONCE(!dax_dev
);
222 #ifndef CONFIG_ARCH_HAS_PMEM_API
223 if (a
== &dev_attr_write_cache
.attr
)
229 static struct attribute
*dax_attributes
[] = {
230 &dev_attr_write_cache
.attr
,
234 struct attribute_group dax_attribute_group
= {
236 .attrs
= dax_attributes
,
237 .is_visible
= dax_visible
,
239 EXPORT_SYMBOL_GPL(dax_attribute_group
);
242 * dax_direct_access() - translate a device pgoff to an absolute pfn
243 * @dax_dev: a dax_device instance representing the logical memory range
244 * @pgoff: offset in pages from the start of the device to translate
245 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
246 * @kaddr: output parameter that returns a virtual address mapping of pfn
247 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
249 * Return: negative errno if an error occurs, otherwise the number of
250 * pages accessible at the device relative @pgoff.
252 long dax_direct_access(struct dax_device
*dax_dev
, pgoff_t pgoff
, long nr_pages
,
253 void **kaddr
, pfn_t
*pfn
)
258 * The device driver is allowed to sleep, in order to make the
259 * memory directly accessible.
266 if (!dax_alive(dax_dev
))
272 avail
= dax_dev
->ops
->direct_access(dax_dev
, pgoff
, nr_pages
,
276 return min(avail
, nr_pages
);
278 EXPORT_SYMBOL_GPL(dax_direct_access
);
280 size_t dax_copy_from_iter(struct dax_device
*dax_dev
, pgoff_t pgoff
, void *addr
,
281 size_t bytes
, struct iov_iter
*i
)
283 if (!dax_alive(dax_dev
))
286 return dax_dev
->ops
->copy_from_iter(dax_dev
, pgoff
, addr
, bytes
, i
);
288 EXPORT_SYMBOL_GPL(dax_copy_from_iter
);
290 #ifdef CONFIG_ARCH_HAS_PMEM_API
291 void arch_wb_cache_pmem(void *addr
, size_t size
);
292 void dax_flush(struct dax_device
*dax_dev
, void *addr
, size_t size
)
294 if (unlikely(!test_bit(DAXDEV_WRITE_CACHE
, &dax_dev
->flags
)))
297 arch_wb_cache_pmem(addr
, size
);
300 void dax_flush(struct dax_device
*dax_dev
, void *addr
, size_t size
)
304 EXPORT_SYMBOL_GPL(dax_flush
);
306 void dax_write_cache(struct dax_device
*dax_dev
, bool wc
)
309 set_bit(DAXDEV_WRITE_CACHE
, &dax_dev
->flags
);
311 clear_bit(DAXDEV_WRITE_CACHE
, &dax_dev
->flags
);
313 EXPORT_SYMBOL_GPL(dax_write_cache
);
315 bool dax_write_cache_enabled(struct dax_device
*dax_dev
)
317 return test_bit(DAXDEV_WRITE_CACHE
, &dax_dev
->flags
);
319 EXPORT_SYMBOL_GPL(dax_write_cache_enabled
);
321 bool dax_alive(struct dax_device
*dax_dev
)
323 lockdep_assert_held(&dax_srcu
);
324 return test_bit(DAXDEV_ALIVE
, &dax_dev
->flags
);
326 EXPORT_SYMBOL_GPL(dax_alive
);
328 static int dax_host_hash(const char *host
)
330 return hashlen_hash(hashlen_string("DAX", host
)) % DAX_HASH_SIZE
;
334 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
335 * that any fault handlers or operations that might have seen
336 * dax_alive(), have completed. Any operations that start after
337 * synchronize_srcu() has run will abort upon seeing !dax_alive().
339 void kill_dax(struct dax_device
*dax_dev
)
344 clear_bit(DAXDEV_ALIVE
, &dax_dev
->flags
);
346 synchronize_srcu(&dax_srcu
);
348 spin_lock(&dax_host_lock
);
349 hlist_del_init(&dax_dev
->list
);
350 spin_unlock(&dax_host_lock
);
352 dax_dev
->private = NULL
;
354 EXPORT_SYMBOL_GPL(kill_dax
);
356 static struct inode
*dax_alloc_inode(struct super_block
*sb
)
358 struct dax_device
*dax_dev
;
361 dax_dev
= kmem_cache_alloc(dax_cache
, GFP_KERNEL
);
365 inode
= &dax_dev
->inode
;
370 static struct dax_device
*to_dax_dev(struct inode
*inode
)
372 return container_of(inode
, struct dax_device
, inode
);
375 static void dax_i_callback(struct rcu_head
*head
)
377 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
378 struct dax_device
*dax_dev
= to_dax_dev(inode
);
380 kfree(dax_dev
->host
);
381 dax_dev
->host
= NULL
;
383 ida_simple_remove(&dax_minor_ida
, MINOR(inode
->i_rdev
));
384 kmem_cache_free(dax_cache
, dax_dev
);
387 static void dax_destroy_inode(struct inode
*inode
)
389 struct dax_device
*dax_dev
= to_dax_dev(inode
);
391 WARN_ONCE(test_bit(DAXDEV_ALIVE
, &dax_dev
->flags
),
392 "kill_dax() must be called before final iput()\n");
393 call_rcu(&inode
->i_rcu
, dax_i_callback
);
396 static const struct super_operations dax_sops
= {
397 .statfs
= simple_statfs
,
398 .alloc_inode
= dax_alloc_inode
,
399 .destroy_inode
= dax_destroy_inode
,
400 .drop_inode
= generic_delete_inode
,
403 static struct dentry
*dax_mount(struct file_system_type
*fs_type
,
404 int flags
, const char *dev_name
, void *data
)
406 return mount_pseudo(fs_type
, "dax:", &dax_sops
, NULL
, DAXFS_MAGIC
);
409 static struct file_system_type dax_fs_type
= {
412 .kill_sb
= kill_anon_super
,
415 static int dax_test(struct inode
*inode
, void *data
)
417 dev_t devt
= *(dev_t
*) data
;
419 return inode
->i_rdev
== devt
;
422 static int dax_set(struct inode
*inode
, void *data
)
424 dev_t devt
= *(dev_t
*) data
;
426 inode
->i_rdev
= devt
;
430 static struct dax_device
*dax_dev_get(dev_t devt
)
432 struct dax_device
*dax_dev
;
435 inode
= iget5_locked(dax_superblock
, hash_32(devt
+ DAXFS_MAGIC
, 31),
436 dax_test
, dax_set
, &devt
);
441 dax_dev
= to_dax_dev(inode
);
442 if (inode
->i_state
& I_NEW
) {
443 set_bit(DAXDEV_ALIVE
, &dax_dev
->flags
);
444 inode
->i_cdev
= &dax_dev
->cdev
;
445 inode
->i_mode
= S_IFCHR
;
446 inode
->i_flags
= S_DAX
;
447 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
448 unlock_new_inode(inode
);
454 static void dax_add_host(struct dax_device
*dax_dev
, const char *host
)
459 * Unconditionally init dax_dev since it's coming from a
460 * non-zeroed slab cache
462 INIT_HLIST_NODE(&dax_dev
->list
);
463 dax_dev
->host
= host
;
467 hash
= dax_host_hash(host
);
468 spin_lock(&dax_host_lock
);
469 hlist_add_head(&dax_dev
->list
, &dax_host_list
[hash
]);
470 spin_unlock(&dax_host_lock
);
473 struct dax_device
*alloc_dax(void *private, const char *__host
,
474 const struct dax_operations
*ops
)
476 struct dax_device
*dax_dev
;
481 host
= kstrdup(__host
, GFP_KERNEL
);
485 minor
= ida_simple_get(&dax_minor_ida
, 0, MINORMASK
+1, GFP_KERNEL
);
489 devt
= MKDEV(MAJOR(dax_devt
), minor
);
490 dax_dev
= dax_dev_get(devt
);
494 dax_add_host(dax_dev
, host
);
496 dax_dev
->private = private;
500 ida_simple_remove(&dax_minor_ida
, minor
);
505 EXPORT_SYMBOL_GPL(alloc_dax
);
507 void put_dax(struct dax_device
*dax_dev
)
511 iput(&dax_dev
->inode
);
513 EXPORT_SYMBOL_GPL(put_dax
);
516 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
517 * @host: alternate name for the device registered by a dax driver
519 struct dax_device
*dax_get_by_host(const char *host
)
521 struct dax_device
*dax_dev
, *found
= NULL
;
527 hash
= dax_host_hash(host
);
529 id
= dax_read_lock();
530 spin_lock(&dax_host_lock
);
531 hlist_for_each_entry(dax_dev
, &dax_host_list
[hash
], list
) {
532 if (!dax_alive(dax_dev
)
533 || strcmp(host
, dax_dev
->host
) != 0)
536 if (igrab(&dax_dev
->inode
))
540 spin_unlock(&dax_host_lock
);
545 EXPORT_SYMBOL_GPL(dax_get_by_host
);
548 * inode_dax: convert a public inode into its dax_dev
549 * @inode: An inode with i_cdev pointing to a dax_dev
551 * Note this is not equivalent to to_dax_dev() which is for private
552 * internal use where we know the inode filesystem type == dax_fs_type.
554 struct dax_device
*inode_dax(struct inode
*inode
)
556 struct cdev
*cdev
= inode
->i_cdev
;
558 return container_of(cdev
, struct dax_device
, cdev
);
560 EXPORT_SYMBOL_GPL(inode_dax
);
562 struct inode
*dax_inode(struct dax_device
*dax_dev
)
564 return &dax_dev
->inode
;
566 EXPORT_SYMBOL_GPL(dax_inode
);
568 void *dax_get_private(struct dax_device
*dax_dev
)
570 return dax_dev
->private;
572 EXPORT_SYMBOL_GPL(dax_get_private
);
574 static void init_once(void *_dax_dev
)
576 struct dax_device
*dax_dev
= _dax_dev
;
577 struct inode
*inode
= &dax_dev
->inode
;
579 memset(dax_dev
, 0, sizeof(*dax_dev
));
580 inode_init_once(inode
);
583 static int __dax_fs_init(void)
587 dax_cache
= kmem_cache_create("dax_cache", sizeof(struct dax_device
), 0,
588 (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
589 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
),
594 rc
= register_filesystem(&dax_fs_type
);
596 goto err_register_fs
;
598 dax_mnt
= kern_mount(&dax_fs_type
);
599 if (IS_ERR(dax_mnt
)) {
600 rc
= PTR_ERR(dax_mnt
);
603 dax_superblock
= dax_mnt
->mnt_sb
;
608 unregister_filesystem(&dax_fs_type
);
610 kmem_cache_destroy(dax_cache
);
615 static void __dax_fs_exit(void)
617 kern_unmount(dax_mnt
);
618 unregister_filesystem(&dax_fs_type
);
619 kmem_cache_destroy(dax_cache
);
622 static int __init
dax_fs_init(void)
626 rc
= __dax_fs_init();
630 rc
= alloc_chrdev_region(&dax_devt
, 0, MINORMASK
+1, "dax");
636 static void __exit
dax_fs_exit(void)
638 unregister_chrdev_region(dax_devt
, MINORMASK
+1);
639 ida_destroy(&dax_minor_ida
);
643 MODULE_AUTHOR("Intel Corporation");
644 MODULE_LICENSE("GPL v2");
645 subsys_initcall(dax_fs_init
);
646 module_exit(dax_fs_exit
);