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 bool dax_enabled = false;
89 pgoff_t pgoff, pgoff_end;
90 struct request_queue *q;
91 char buf[BDEVNAME_SIZE];
92 void *kaddr, *end_kaddr;
98 if (blocksize != PAGE_SIZE) {
99 pr_debug("%s: error: unsupported blocksize for dax\n",
100 bdevname(bdev, buf));
104 q = bdev_get_queue(bdev);
105 if (!q || !blk_queue_dax(q)) {
106 pr_debug("%s: error: request queue doesn't support dax\n",
107 bdevname(bdev, buf));
111 err = bdev_dax_pgoff(bdev, 0, PAGE_SIZE, &pgoff);
113 pr_debug("%s: error: unaligned partition for dax\n",
114 bdevname(bdev, buf));
118 last_page = PFN_DOWN(i_size_read(bdev->bd_inode) - 1) * 8;
119 err = bdev_dax_pgoff(bdev, last_page, PAGE_SIZE, &pgoff_end);
121 pr_debug("%s: error: unaligned partition for dax\n",
122 bdevname(bdev, buf));
126 dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
128 pr_debug("%s: error: device does not support dax\n",
129 bdevname(bdev, buf));
133 id = dax_read_lock();
134 len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
135 len2 = dax_direct_access(dax_dev, pgoff_end, 1, &end_kaddr, &end_pfn);
140 if (len < 1 || len2 < 1) {
141 pr_debug("%s: error: dax access failed (%ld)\n",
142 bdevname(bdev, buf), len < 1 ? len : len2);
146 if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
148 * An arch that has enabled the pmem api should also
149 * have its drivers support pfn_t_devmap()
151 * This is a developer warning and should not trigger in
152 * production. dax_flush() will crash since it depends
153 * on being able to do (page_address(pfn_to_page())).
155 WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
157 } else if (pfn_t_devmap(pfn) && pfn_t_devmap(end_pfn)) {
158 struct dev_pagemap *pgmap, *end_pgmap;
160 pgmap = get_dev_pagemap(pfn_t_to_pfn(pfn), NULL);
161 end_pgmap = get_dev_pagemap(pfn_t_to_pfn(end_pfn), NULL);
162 if (pgmap && pgmap == end_pgmap && pgmap->type == MEMORY_DEVICE_FS_DAX
163 && pfn_t_to_page(pfn)->pgmap == pgmap
164 && pfn_t_to_page(end_pfn)->pgmap == pgmap
165 && pfn_t_to_pfn(pfn) == PHYS_PFN(__pa(kaddr))
166 && pfn_t_to_pfn(end_pfn) == PHYS_PFN(__pa(end_kaddr)))
168 put_dev_pagemap(pgmap);
169 put_dev_pagemap(end_pgmap);
174 pr_debug("%s: error: dax support not enabled\n",
175 bdevname(bdev, buf));
180 EXPORT_SYMBOL_GPL(__bdev_dax_supported);
183 enum dax_device_flags {
184 /* !alive + rcu grace period == no new operations / mappings */
186 /* gate whether dax_flush() calls the low level flush routine */
191 * struct dax_device - anchor object for dax services
193 * @cdev: optional character interface for "device dax"
194 * @host: optional name for lookups where the device path is not available
195 * @private: dax driver private data
196 * @flags: state and boolean properties
199 struct hlist_node list;
205 const struct dax_operations *ops;
208 static ssize_t write_cache_show(struct device *dev,
209 struct device_attribute *attr, char *buf)
211 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
214 WARN_ON_ONCE(!dax_dev);
218 rc = sprintf(buf, "%d\n", !!dax_write_cache_enabled(dax_dev));
223 static ssize_t write_cache_store(struct device *dev,
224 struct device_attribute *attr, const char *buf, size_t len)
227 int rc = strtobool(buf, &write_cache);
228 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
230 WARN_ON_ONCE(!dax_dev);
237 dax_write_cache(dax_dev, write_cache);
242 static DEVICE_ATTR_RW(write_cache);
244 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
246 struct device *dev = container_of(kobj, typeof(*dev), kobj);
247 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
249 WARN_ON_ONCE(!dax_dev);
253 #ifndef CONFIG_ARCH_HAS_PMEM_API
254 if (a == &dev_attr_write_cache.attr)
260 static struct attribute *dax_attributes[] = {
261 &dev_attr_write_cache.attr,
265 struct attribute_group dax_attribute_group = {
267 .attrs = dax_attributes,
268 .is_visible = dax_visible,
270 EXPORT_SYMBOL_GPL(dax_attribute_group);
273 * dax_direct_access() - translate a device pgoff to an absolute pfn
274 * @dax_dev: a dax_device instance representing the logical memory range
275 * @pgoff: offset in pages from the start of the device to translate
276 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
277 * @kaddr: output parameter that returns a virtual address mapping of pfn
278 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
280 * Return: negative errno if an error occurs, otherwise the number of
281 * pages accessible at the device relative @pgoff.
283 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
284 void **kaddr, pfn_t *pfn)
291 if (!dax_alive(dax_dev))
297 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
301 return min(avail, nr_pages);
303 EXPORT_SYMBOL_GPL(dax_direct_access);
305 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
306 size_t bytes, struct iov_iter *i)
308 if (!dax_alive(dax_dev))
311 return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
313 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
315 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
316 size_t bytes, struct iov_iter *i)
318 if (!dax_alive(dax_dev))
321 return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i);
323 EXPORT_SYMBOL_GPL(dax_copy_to_iter);
325 #ifdef CONFIG_ARCH_HAS_PMEM_API
326 void arch_wb_cache_pmem(void *addr, size_t size);
327 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
329 if (unlikely(!dax_write_cache_enabled(dax_dev)))
332 arch_wb_cache_pmem(addr, size);
335 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
339 EXPORT_SYMBOL_GPL(dax_flush);
341 void dax_write_cache(struct dax_device *dax_dev, bool wc)
344 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
346 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
348 EXPORT_SYMBOL_GPL(dax_write_cache);
350 bool dax_write_cache_enabled(struct dax_device *dax_dev)
352 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
354 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
356 bool dax_alive(struct dax_device *dax_dev)
358 lockdep_assert_held(&dax_srcu);
359 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
361 EXPORT_SYMBOL_GPL(dax_alive);
363 static int dax_host_hash(const char *host)
365 return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
369 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
370 * that any fault handlers or operations that might have seen
371 * dax_alive(), have completed. Any operations that start after
372 * synchronize_srcu() has run will abort upon seeing !dax_alive().
374 void kill_dax(struct dax_device *dax_dev)
379 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
381 synchronize_srcu(&dax_srcu);
383 spin_lock(&dax_host_lock);
384 hlist_del_init(&dax_dev->list);
385 spin_unlock(&dax_host_lock);
387 dax_dev->private = NULL;
389 EXPORT_SYMBOL_GPL(kill_dax);
391 static struct inode *dax_alloc_inode(struct super_block *sb)
393 struct dax_device *dax_dev;
396 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
400 inode = &dax_dev->inode;
405 static struct dax_device *to_dax_dev(struct inode *inode)
407 return container_of(inode, struct dax_device, inode);
410 static void dax_i_callback(struct rcu_head *head)
412 struct inode *inode = container_of(head, struct inode, i_rcu);
413 struct dax_device *dax_dev = to_dax_dev(inode);
415 kfree(dax_dev->host);
416 dax_dev->host = NULL;
418 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
419 kmem_cache_free(dax_cache, dax_dev);
422 static void dax_destroy_inode(struct inode *inode)
424 struct dax_device *dax_dev = to_dax_dev(inode);
426 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
427 "kill_dax() must be called before final iput()\n");
428 call_rcu(&inode->i_rcu, dax_i_callback);
431 static const struct super_operations dax_sops = {
432 .statfs = simple_statfs,
433 .alloc_inode = dax_alloc_inode,
434 .destroy_inode = dax_destroy_inode,
435 .drop_inode = generic_delete_inode,
438 static struct dentry *dax_mount(struct file_system_type *fs_type,
439 int flags, const char *dev_name, void *data)
441 return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
444 static struct file_system_type dax_fs_type = {
447 .kill_sb = kill_anon_super,
450 static int dax_test(struct inode *inode, void *data)
452 dev_t devt = *(dev_t *) data;
454 return inode->i_rdev == devt;
457 static int dax_set(struct inode *inode, void *data)
459 dev_t devt = *(dev_t *) data;
461 inode->i_rdev = devt;
465 static struct dax_device *dax_dev_get(dev_t devt)
467 struct dax_device *dax_dev;
470 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
471 dax_test, dax_set, &devt);
476 dax_dev = to_dax_dev(inode);
477 if (inode->i_state & I_NEW) {
478 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
479 inode->i_cdev = &dax_dev->cdev;
480 inode->i_mode = S_IFCHR;
481 inode->i_flags = S_DAX;
482 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
483 unlock_new_inode(inode);
489 static void dax_add_host(struct dax_device *dax_dev, const char *host)
494 * Unconditionally init dax_dev since it's coming from a
495 * non-zeroed slab cache
497 INIT_HLIST_NODE(&dax_dev->list);
498 dax_dev->host = host;
502 hash = dax_host_hash(host);
503 spin_lock(&dax_host_lock);
504 hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
505 spin_unlock(&dax_host_lock);
508 struct dax_device *alloc_dax(void *private, const char *__host,
509 const struct dax_operations *ops)
511 struct dax_device *dax_dev;
516 host = kstrdup(__host, GFP_KERNEL);
520 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
524 devt = MKDEV(MAJOR(dax_devt), minor);
525 dax_dev = dax_dev_get(devt);
529 dax_add_host(dax_dev, host);
531 dax_dev->private = private;
535 ida_simple_remove(&dax_minor_ida, minor);
540 EXPORT_SYMBOL_GPL(alloc_dax);
542 void put_dax(struct dax_device *dax_dev)
546 iput(&dax_dev->inode);
548 EXPORT_SYMBOL_GPL(put_dax);
551 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
552 * @host: alternate name for the device registered by a dax driver
554 struct dax_device *dax_get_by_host(const char *host)
556 struct dax_device *dax_dev, *found = NULL;
562 hash = dax_host_hash(host);
564 id = dax_read_lock();
565 spin_lock(&dax_host_lock);
566 hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
567 if (!dax_alive(dax_dev)
568 || strcmp(host, dax_dev->host) != 0)
571 if (igrab(&dax_dev->inode))
575 spin_unlock(&dax_host_lock);
580 EXPORT_SYMBOL_GPL(dax_get_by_host);
583 * inode_dax: convert a public inode into its dax_dev
584 * @inode: An inode with i_cdev pointing to a dax_dev
586 * Note this is not equivalent to to_dax_dev() which is for private
587 * internal use where we know the inode filesystem type == dax_fs_type.
589 struct dax_device *inode_dax(struct inode *inode)
591 struct cdev *cdev = inode->i_cdev;
593 return container_of(cdev, struct dax_device, cdev);
595 EXPORT_SYMBOL_GPL(inode_dax);
597 struct inode *dax_inode(struct dax_device *dax_dev)
599 return &dax_dev->inode;
601 EXPORT_SYMBOL_GPL(dax_inode);
603 void *dax_get_private(struct dax_device *dax_dev)
605 return dax_dev->private;
607 EXPORT_SYMBOL_GPL(dax_get_private);
609 static void init_once(void *_dax_dev)
611 struct dax_device *dax_dev = _dax_dev;
612 struct inode *inode = &dax_dev->inode;
614 memset(dax_dev, 0, sizeof(*dax_dev));
615 inode_init_once(inode);
618 static int __dax_fs_init(void)
622 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
623 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
624 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
629 rc = register_filesystem(&dax_fs_type);
631 goto err_register_fs;
633 dax_mnt = kern_mount(&dax_fs_type);
634 if (IS_ERR(dax_mnt)) {
635 rc = PTR_ERR(dax_mnt);
638 dax_superblock = dax_mnt->mnt_sb;
643 unregister_filesystem(&dax_fs_type);
645 kmem_cache_destroy(dax_cache);
650 static void __dax_fs_exit(void)
652 kern_unmount(dax_mnt);
653 unregister_filesystem(&dax_fs_type);
654 kmem_cache_destroy(dax_cache);
657 static int __init dax_fs_init(void)
661 rc = __dax_fs_init();
665 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
671 static void __exit dax_fs_exit(void)
673 unregister_chrdev_region(dax_devt, MINORMASK+1);
674 ida_destroy(&dax_minor_ida);
678 MODULE_AUTHOR("Intel Corporation");
679 MODULE_LICENSE("GPL v2");
680 subsys_initcall(dax_fs_init);
681 module_exit(dax_fs_exit);