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/cdev.h>
18 #include <linux/hash.h>
19 #include <linux/slab.h>
20 #include <linux/dax.h>
23 static int nr_dax = CONFIG_NR_DEV_DAX;
24 module_param(nr_dax, int, S_IRUGO);
25 MODULE_PARM_DESC(nr_dax, "max number of dax device instances");
27 static dev_t dax_devt;
28 DEFINE_STATIC_SRCU(dax_srcu);
29 static struct vfsmount *dax_mnt;
30 static DEFINE_IDA(dax_minor_ida);
31 static struct kmem_cache *dax_cache __read_mostly;
32 static struct super_block *dax_superblock __read_mostly;
34 #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
35 static struct hlist_head dax_host_list[DAX_HASH_SIZE];
36 static DEFINE_SPINLOCK(dax_host_lock);
38 int dax_read_lock(void)
40 return srcu_read_lock(&dax_srcu);
42 EXPORT_SYMBOL_GPL(dax_read_lock);
44 void dax_read_unlock(int id)
46 srcu_read_unlock(&dax_srcu, id);
48 EXPORT_SYMBOL_GPL(dax_read_unlock);
51 * struct dax_device - anchor object for dax services
53 * @cdev: optional character interface for "device dax"
54 * @host: optional name for lookups where the device path is not available
55 * @private: dax driver private data
56 * @alive: !alive + rcu grace period == no new operations / mappings
59 struct hlist_node list;
65 const struct dax_operations *ops;
69 * dax_direct_access() - translate a device pgoff to an absolute pfn
70 * @dax_dev: a dax_device instance representing the logical memory range
71 * @pgoff: offset in pages from the start of the device to translate
72 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
73 * @kaddr: output parameter that returns a virtual address mapping of pfn
74 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
76 * Return: negative errno if an error occurs, otherwise the number of
77 * pages accessible at the device relative @pgoff.
79 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
80 void **kaddr, pfn_t *pfn)
85 * The device driver is allowed to sleep, in order to make the
86 * memory directly accessible.
93 if (!dax_alive(dax_dev))
99 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
103 return min(avail, nr_pages);
105 EXPORT_SYMBOL_GPL(dax_direct_access);
107 bool dax_alive(struct dax_device *dax_dev)
109 lockdep_assert_held(&dax_srcu);
110 return dax_dev->alive;
112 EXPORT_SYMBOL_GPL(dax_alive);
114 static int dax_host_hash(const char *host)
116 return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
120 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
121 * that any fault handlers or operations that might have seen
122 * dax_alive(), have completed. Any operations that start after
123 * synchronize_srcu() has run will abort upon seeing !dax_alive().
125 void kill_dax(struct dax_device *dax_dev)
130 dax_dev->alive = false;
132 synchronize_srcu(&dax_srcu);
134 spin_lock(&dax_host_lock);
135 hlist_del_init(&dax_dev->list);
136 spin_unlock(&dax_host_lock);
138 dax_dev->private = NULL;
140 EXPORT_SYMBOL_GPL(kill_dax);
142 static struct inode *dax_alloc_inode(struct super_block *sb)
144 struct dax_device *dax_dev;
146 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
147 return &dax_dev->inode;
150 static struct dax_device *to_dax_dev(struct inode *inode)
152 return container_of(inode, struct dax_device, inode);
155 static void dax_i_callback(struct rcu_head *head)
157 struct inode *inode = container_of(head, struct inode, i_rcu);
158 struct dax_device *dax_dev = to_dax_dev(inode);
160 kfree(dax_dev->host);
161 dax_dev->host = NULL;
162 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
163 kmem_cache_free(dax_cache, dax_dev);
166 static void dax_destroy_inode(struct inode *inode)
168 struct dax_device *dax_dev = to_dax_dev(inode);
170 WARN_ONCE(dax_dev->alive,
171 "kill_dax() must be called before final iput()\n");
172 call_rcu(&inode->i_rcu, dax_i_callback);
175 static const struct super_operations dax_sops = {
176 .statfs = simple_statfs,
177 .alloc_inode = dax_alloc_inode,
178 .destroy_inode = dax_destroy_inode,
179 .drop_inode = generic_delete_inode,
182 static struct dentry *dax_mount(struct file_system_type *fs_type,
183 int flags, const char *dev_name, void *data)
185 return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
188 static struct file_system_type dax_fs_type = {
191 .kill_sb = kill_anon_super,
194 static int dax_test(struct inode *inode, void *data)
196 dev_t devt = *(dev_t *) data;
198 return inode->i_rdev == devt;
201 static int dax_set(struct inode *inode, void *data)
203 dev_t devt = *(dev_t *) data;
205 inode->i_rdev = devt;
209 static struct dax_device *dax_dev_get(dev_t devt)
211 struct dax_device *dax_dev;
214 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
215 dax_test, dax_set, &devt);
220 dax_dev = to_dax_dev(inode);
221 if (inode->i_state & I_NEW) {
222 dax_dev->alive = true;
223 inode->i_cdev = &dax_dev->cdev;
224 inode->i_mode = S_IFCHR;
225 inode->i_flags = S_DAX;
226 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
227 unlock_new_inode(inode);
233 static void dax_add_host(struct dax_device *dax_dev, const char *host)
238 * Unconditionally init dax_dev since it's coming from a
239 * non-zeroed slab cache
241 INIT_HLIST_NODE(&dax_dev->list);
242 dax_dev->host = host;
246 hash = dax_host_hash(host);
247 spin_lock(&dax_host_lock);
248 hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
249 spin_unlock(&dax_host_lock);
252 struct dax_device *alloc_dax(void *private, const char *__host,
253 const struct dax_operations *ops)
255 struct dax_device *dax_dev;
260 host = kstrdup(__host, GFP_KERNEL);
264 minor = ida_simple_get(&dax_minor_ida, 0, nr_dax, GFP_KERNEL);
268 devt = MKDEV(MAJOR(dax_devt), minor);
269 dax_dev = dax_dev_get(devt);
273 dax_add_host(dax_dev, host);
275 dax_dev->private = private;
279 ida_simple_remove(&dax_minor_ida, minor);
284 EXPORT_SYMBOL_GPL(alloc_dax);
286 void put_dax(struct dax_device *dax_dev)
290 iput(&dax_dev->inode);
292 EXPORT_SYMBOL_GPL(put_dax);
295 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
296 * @host: alternate name for the device registered by a dax driver
298 struct dax_device *dax_get_by_host(const char *host)
300 struct dax_device *dax_dev, *found = NULL;
306 hash = dax_host_hash(host);
308 id = dax_read_lock();
309 spin_lock(&dax_host_lock);
310 hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
311 if (!dax_alive(dax_dev)
312 || strcmp(host, dax_dev->host) != 0)
315 if (igrab(&dax_dev->inode))
319 spin_unlock(&dax_host_lock);
324 EXPORT_SYMBOL_GPL(dax_get_by_host);
327 * inode_dax: convert a public inode into its dax_dev
328 * @inode: An inode with i_cdev pointing to a dax_dev
330 * Note this is not equivalent to to_dax_dev() which is for private
331 * internal use where we know the inode filesystem type == dax_fs_type.
333 struct dax_device *inode_dax(struct inode *inode)
335 struct cdev *cdev = inode->i_cdev;
337 return container_of(cdev, struct dax_device, cdev);
339 EXPORT_SYMBOL_GPL(inode_dax);
341 struct inode *dax_inode(struct dax_device *dax_dev)
343 return &dax_dev->inode;
345 EXPORT_SYMBOL_GPL(dax_inode);
347 void *dax_get_private(struct dax_device *dax_dev)
349 return dax_dev->private;
351 EXPORT_SYMBOL_GPL(dax_get_private);
353 static void init_once(void *_dax_dev)
355 struct dax_device *dax_dev = _dax_dev;
356 struct inode *inode = &dax_dev->inode;
358 inode_init_once(inode);
361 static int __dax_fs_init(void)
365 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
366 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
367 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
372 rc = register_filesystem(&dax_fs_type);
374 goto err_register_fs;
376 dax_mnt = kern_mount(&dax_fs_type);
377 if (IS_ERR(dax_mnt)) {
378 rc = PTR_ERR(dax_mnt);
381 dax_superblock = dax_mnt->mnt_sb;
386 unregister_filesystem(&dax_fs_type);
388 kmem_cache_destroy(dax_cache);
393 static void __dax_fs_exit(void)
395 kern_unmount(dax_mnt);
396 unregister_filesystem(&dax_fs_type);
397 kmem_cache_destroy(dax_cache);
400 static int __init dax_fs_init(void)
404 rc = __dax_fs_init();
408 nr_dax = max(nr_dax, 256);
409 rc = alloc_chrdev_region(&dax_devt, 0, nr_dax, "dax");
415 static void __exit dax_fs_exit(void)
417 unregister_chrdev_region(dax_devt, nr_dax);
418 ida_destroy(&dax_minor_ida);
422 MODULE_AUTHOR("Intel Corporation");
423 MODULE_LICENSE("GPL v2");
424 subsys_initcall(dax_fs_init);
425 module_exit(dax_fs_exit);