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>
25 #include "dax-private.h"
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 #include <linux/blkdev.h>
53 int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
56 phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512;
59 *pgoff = PHYS_PFN(phys_off);
60 if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
64 EXPORT_SYMBOL(bdev_dax_pgoff);
66 #if IS_ENABLED(CONFIG_FS_DAX)
67 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
69 if (!blk_queue_dax(bdev->bd_queue))
71 return fs_dax_get_by_host(bdev->bd_disk->disk_name);
73 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
76 bool __generic_fsdax_supported(struct dax_device *dax_dev,
77 struct block_device *bdev, int blocksize, sector_t start,
80 bool dax_enabled = false;
81 pgoff_t pgoff, pgoff_end;
82 char buf[BDEVNAME_SIZE];
83 void *kaddr, *end_kaddr;
89 if (blocksize != PAGE_SIZE) {
90 pr_debug("%s: error: unsupported blocksize for dax\n",
95 err = bdev_dax_pgoff(bdev, start, PAGE_SIZE, &pgoff);
97 pr_debug("%s: error: unaligned partition for dax\n",
102 last_page = PFN_DOWN((start + sectors - 1) * 512) * PAGE_SIZE / 512;
103 err = bdev_dax_pgoff(bdev, last_page, PAGE_SIZE, &pgoff_end);
105 pr_debug("%s: error: unaligned partition for dax\n",
106 bdevname(bdev, buf));
110 id = dax_read_lock();
111 len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
112 len2 = dax_direct_access(dax_dev, pgoff_end, 1, &end_kaddr, &end_pfn);
115 if (len < 1 || len2 < 1) {
116 pr_debug("%s: error: dax access failed (%ld)\n",
117 bdevname(bdev, buf), len < 1 ? len : len2);
121 if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
123 * An arch that has enabled the pmem api should also
124 * have its drivers support pfn_t_devmap()
126 * This is a developer warning and should not trigger in
127 * production. dax_flush() will crash since it depends
128 * on being able to do (page_address(pfn_to_page())).
130 WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
132 } else if (pfn_t_devmap(pfn) && pfn_t_devmap(end_pfn)) {
133 struct dev_pagemap *pgmap, *end_pgmap;
135 pgmap = get_dev_pagemap(pfn_t_to_pfn(pfn), NULL);
136 end_pgmap = get_dev_pagemap(pfn_t_to_pfn(end_pfn), NULL);
137 if (pgmap && pgmap == end_pgmap && pgmap->type == MEMORY_DEVICE_FS_DAX
138 && pfn_t_to_page(pfn)->pgmap == pgmap
139 && pfn_t_to_page(end_pfn)->pgmap == pgmap
140 && pfn_t_to_pfn(pfn) == PHYS_PFN(__pa(kaddr))
141 && pfn_t_to_pfn(end_pfn) == PHYS_PFN(__pa(end_kaddr)))
143 put_dev_pagemap(pgmap);
144 put_dev_pagemap(end_pgmap);
149 pr_debug("%s: error: dax support not enabled\n",
150 bdevname(bdev, buf));
155 EXPORT_SYMBOL_GPL(__generic_fsdax_supported);
158 * __bdev_dax_supported() - Check if the device supports dax for filesystem
159 * @bdev: block device to check
160 * @blocksize: The block size of the device
162 * This is a library function for filesystems to check if the block device
163 * can be mounted with dax option.
165 * Return: true if supported, false if unsupported
167 bool __bdev_dax_supported(struct block_device *bdev, int blocksize)
169 struct dax_device *dax_dev;
170 struct request_queue *q;
171 char buf[BDEVNAME_SIZE];
175 q = bdev_get_queue(bdev);
176 if (!q || !blk_queue_dax(q)) {
177 pr_debug("%s: error: request queue doesn't support dax\n",
178 bdevname(bdev, buf));
182 dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
184 pr_debug("%s: error: device does not support dax\n",
185 bdevname(bdev, buf));
189 id = dax_read_lock();
190 ret = dax_supported(dax_dev, bdev, blocksize, 0,
191 i_size_read(bdev->bd_inode) / 512);
198 EXPORT_SYMBOL_GPL(__bdev_dax_supported);
201 enum dax_device_flags {
202 /* !alive + rcu grace period == no new operations / mappings */
204 /* gate whether dax_flush() calls the low level flush routine */
209 * struct dax_device - anchor object for dax services
211 * @cdev: optional character interface for "device dax"
212 * @host: optional name for lookups where the device path is not available
213 * @private: dax driver private data
214 * @flags: state and boolean properties
217 struct hlist_node list;
223 const struct dax_operations *ops;
226 static ssize_t write_cache_show(struct device *dev,
227 struct device_attribute *attr, char *buf)
229 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
232 WARN_ON_ONCE(!dax_dev);
236 rc = sprintf(buf, "%d\n", !!dax_write_cache_enabled(dax_dev));
241 static ssize_t write_cache_store(struct device *dev,
242 struct device_attribute *attr, const char *buf, size_t len)
245 int rc = strtobool(buf, &write_cache);
246 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
248 WARN_ON_ONCE(!dax_dev);
255 dax_write_cache(dax_dev, write_cache);
260 static DEVICE_ATTR_RW(write_cache);
262 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
264 struct device *dev = container_of(kobj, typeof(*dev), kobj);
265 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
267 WARN_ON_ONCE(!dax_dev);
271 #ifndef CONFIG_ARCH_HAS_PMEM_API
272 if (a == &dev_attr_write_cache.attr)
278 static struct attribute *dax_attributes[] = {
279 &dev_attr_write_cache.attr,
283 struct attribute_group dax_attribute_group = {
285 .attrs = dax_attributes,
286 .is_visible = dax_visible,
288 EXPORT_SYMBOL_GPL(dax_attribute_group);
291 * dax_direct_access() - translate a device pgoff to an absolute pfn
292 * @dax_dev: a dax_device instance representing the logical memory range
293 * @pgoff: offset in pages from the start of the device to translate
294 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
295 * @kaddr: output parameter that returns a virtual address mapping of pfn
296 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
298 * Return: negative errno if an error occurs, otherwise the number of
299 * pages accessible at the device relative @pgoff.
301 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
302 void **kaddr, pfn_t *pfn)
309 if (!dax_alive(dax_dev))
315 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
319 return min(avail, nr_pages);
321 EXPORT_SYMBOL_GPL(dax_direct_access);
323 bool dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
324 int blocksize, sector_t start, sector_t len)
326 if (!dax_alive(dax_dev))
329 return dax_dev->ops->dax_supported(dax_dev, bdev, blocksize, start, len);
332 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
333 size_t bytes, struct iov_iter *i)
335 if (!dax_alive(dax_dev))
338 return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
340 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
342 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
343 size_t bytes, struct iov_iter *i)
345 if (!dax_alive(dax_dev))
348 return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i);
350 EXPORT_SYMBOL_GPL(dax_copy_to_iter);
352 #ifdef CONFIG_ARCH_HAS_PMEM_API
353 void arch_wb_cache_pmem(void *addr, size_t size);
354 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
356 if (unlikely(!dax_write_cache_enabled(dax_dev)))
359 arch_wb_cache_pmem(addr, size);
362 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
366 EXPORT_SYMBOL_GPL(dax_flush);
368 void dax_write_cache(struct dax_device *dax_dev, bool wc)
371 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
373 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
375 EXPORT_SYMBOL_GPL(dax_write_cache);
377 bool dax_write_cache_enabled(struct dax_device *dax_dev)
379 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
381 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
383 bool dax_alive(struct dax_device *dax_dev)
385 lockdep_assert_held(&dax_srcu);
386 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
388 EXPORT_SYMBOL_GPL(dax_alive);
390 static int dax_host_hash(const char *host)
392 return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
396 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
397 * that any fault handlers or operations that might have seen
398 * dax_alive(), have completed. Any operations that start after
399 * synchronize_srcu() has run will abort upon seeing !dax_alive().
401 void kill_dax(struct dax_device *dax_dev)
406 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
408 synchronize_srcu(&dax_srcu);
410 spin_lock(&dax_host_lock);
411 hlist_del_init(&dax_dev->list);
412 spin_unlock(&dax_host_lock);
414 EXPORT_SYMBOL_GPL(kill_dax);
416 void run_dax(struct dax_device *dax_dev)
418 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
420 EXPORT_SYMBOL_GPL(run_dax);
422 static struct inode *dax_alloc_inode(struct super_block *sb)
424 struct dax_device *dax_dev;
427 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
431 inode = &dax_dev->inode;
436 static struct dax_device *to_dax_dev(struct inode *inode)
438 return container_of(inode, struct dax_device, inode);
441 static void dax_free_inode(struct inode *inode)
443 struct dax_device *dax_dev = to_dax_dev(inode);
444 kfree(dax_dev->host);
445 dax_dev->host = NULL;
447 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
448 kmem_cache_free(dax_cache, dax_dev);
451 static void dax_destroy_inode(struct inode *inode)
453 struct dax_device *dax_dev = to_dax_dev(inode);
454 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
455 "kill_dax() must be called before final iput()\n");
458 static const struct super_operations dax_sops = {
459 .statfs = simple_statfs,
460 .alloc_inode = dax_alloc_inode,
461 .destroy_inode = dax_destroy_inode,
462 .free_inode = dax_free_inode,
463 .drop_inode = generic_delete_inode,
466 static struct dentry *dax_mount(struct file_system_type *fs_type,
467 int flags, const char *dev_name, void *data)
469 return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
472 static struct file_system_type dax_fs_type = {
475 .kill_sb = kill_anon_super,
478 static int dax_test(struct inode *inode, void *data)
480 dev_t devt = *(dev_t *) data;
482 return inode->i_rdev == devt;
485 static int dax_set(struct inode *inode, void *data)
487 dev_t devt = *(dev_t *) data;
489 inode->i_rdev = devt;
493 static struct dax_device *dax_dev_get(dev_t devt)
495 struct dax_device *dax_dev;
498 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
499 dax_test, dax_set, &devt);
504 dax_dev = to_dax_dev(inode);
505 if (inode->i_state & I_NEW) {
506 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
507 inode->i_cdev = &dax_dev->cdev;
508 inode->i_mode = S_IFCHR;
509 inode->i_flags = S_DAX;
510 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
511 unlock_new_inode(inode);
517 static void dax_add_host(struct dax_device *dax_dev, const char *host)
522 * Unconditionally init dax_dev since it's coming from a
523 * non-zeroed slab cache
525 INIT_HLIST_NODE(&dax_dev->list);
526 dax_dev->host = host;
530 hash = dax_host_hash(host);
531 spin_lock(&dax_host_lock);
532 hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
533 spin_unlock(&dax_host_lock);
536 struct dax_device *alloc_dax(void *private, const char *__host,
537 const struct dax_operations *ops)
539 struct dax_device *dax_dev;
544 host = kstrdup(__host, GFP_KERNEL);
548 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
552 devt = MKDEV(MAJOR(dax_devt), minor);
553 dax_dev = dax_dev_get(devt);
557 dax_add_host(dax_dev, host);
559 dax_dev->private = private;
563 ida_simple_remove(&dax_minor_ida, minor);
568 EXPORT_SYMBOL_GPL(alloc_dax);
570 void put_dax(struct dax_device *dax_dev)
574 iput(&dax_dev->inode);
576 EXPORT_SYMBOL_GPL(put_dax);
579 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
580 * @host: alternate name for the device registered by a dax driver
582 struct dax_device *dax_get_by_host(const char *host)
584 struct dax_device *dax_dev, *found = NULL;
590 hash = dax_host_hash(host);
592 id = dax_read_lock();
593 spin_lock(&dax_host_lock);
594 hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
595 if (!dax_alive(dax_dev)
596 || strcmp(host, dax_dev->host) != 0)
599 if (igrab(&dax_dev->inode))
603 spin_unlock(&dax_host_lock);
608 EXPORT_SYMBOL_GPL(dax_get_by_host);
611 * inode_dax: convert a public inode into its dax_dev
612 * @inode: An inode with i_cdev pointing to a dax_dev
614 * Note this is not equivalent to to_dax_dev() which is for private
615 * internal use where we know the inode filesystem type == dax_fs_type.
617 struct dax_device *inode_dax(struct inode *inode)
619 struct cdev *cdev = inode->i_cdev;
621 return container_of(cdev, struct dax_device, cdev);
623 EXPORT_SYMBOL_GPL(inode_dax);
625 struct inode *dax_inode(struct dax_device *dax_dev)
627 return &dax_dev->inode;
629 EXPORT_SYMBOL_GPL(dax_inode);
631 void *dax_get_private(struct dax_device *dax_dev)
633 if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
635 return dax_dev->private;
637 EXPORT_SYMBOL_GPL(dax_get_private);
639 static void init_once(void *_dax_dev)
641 struct dax_device *dax_dev = _dax_dev;
642 struct inode *inode = &dax_dev->inode;
644 memset(dax_dev, 0, sizeof(*dax_dev));
645 inode_init_once(inode);
648 static int dax_fs_init(void)
652 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
653 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
654 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
659 rc = register_filesystem(&dax_fs_type);
661 goto err_register_fs;
663 dax_mnt = kern_mount(&dax_fs_type);
664 if (IS_ERR(dax_mnt)) {
665 rc = PTR_ERR(dax_mnt);
668 dax_superblock = dax_mnt->mnt_sb;
673 unregister_filesystem(&dax_fs_type);
675 kmem_cache_destroy(dax_cache);
680 static void dax_fs_exit(void)
682 kern_unmount(dax_mnt);
683 unregister_filesystem(&dax_fs_type);
684 kmem_cache_destroy(dax_cache);
687 static int __init dax_core_init(void)
695 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
705 unregister_chrdev_region(dax_devt, MINORMASK+1);
711 static void __exit dax_core_exit(void)
713 unregister_chrdev_region(dax_devt, MINORMASK+1);
714 ida_destroy(&dax_minor_ida);
718 MODULE_AUTHOR("Intel Corporation");
719 MODULE_LICENSE("GPL v2");
720 subsys_initcall(dax_core_init);
721 module_exit(dax_core_exit);