1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright(c) 2017 Intel Corporation. All rights reserved.
5 #include <linux/pagemap.h>
6 #include <linux/module.h>
7 #include <linux/mount.h>
8 #include <linux/magic.h>
9 #include <linux/genhd.h>
10 #include <linux/pfn_t.h>
11 #include <linux/cdev.h>
12 #include <linux/hash.h>
13 #include <linux/slab.h>
14 #include <linux/uio.h>
15 #include <linux/dax.h>
17 #include "dax-private.h"
19 static dev_t dax_devt;
20 DEFINE_STATIC_SRCU(dax_srcu);
21 static struct vfsmount *dax_mnt;
22 static DEFINE_IDA(dax_minor_ida);
23 static struct kmem_cache *dax_cache __read_mostly;
24 static struct super_block *dax_superblock __read_mostly;
26 #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
27 static struct hlist_head dax_host_list[DAX_HASH_SIZE];
28 static DEFINE_SPINLOCK(dax_host_lock);
30 int dax_read_lock(void)
32 return srcu_read_lock(&dax_srcu);
34 EXPORT_SYMBOL_GPL(dax_read_lock);
36 void dax_read_unlock(int id)
38 srcu_read_unlock(&dax_srcu, id);
40 EXPORT_SYMBOL_GPL(dax_read_unlock);
43 #include <linux/blkdev.h>
45 int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
48 phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512;
51 *pgoff = PHYS_PFN(phys_off);
52 if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
56 EXPORT_SYMBOL(bdev_dax_pgoff);
58 #if IS_ENABLED(CONFIG_FS_DAX)
59 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
61 if (!blk_queue_dax(bdev->bd_queue))
63 return fs_dax_get_by_host(bdev->bd_disk->disk_name);
65 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
68 bool __generic_fsdax_supported(struct dax_device *dax_dev,
69 struct block_device *bdev, int blocksize, sector_t start,
72 bool dax_enabled = false;
73 pgoff_t pgoff, pgoff_end;
74 char buf[BDEVNAME_SIZE];
75 void *kaddr, *end_kaddr;
81 if (blocksize != PAGE_SIZE) {
82 pr_debug("%s: error: unsupported blocksize for dax\n",
87 err = bdev_dax_pgoff(bdev, start, PAGE_SIZE, &pgoff);
89 pr_debug("%s: error: unaligned partition for dax\n",
94 last_page = PFN_DOWN((start + sectors - 1) * 512) * PAGE_SIZE / 512;
95 err = bdev_dax_pgoff(bdev, last_page, PAGE_SIZE, &pgoff_end);
97 pr_debug("%s: error: unaligned partition for dax\n",
102 id = dax_read_lock();
103 len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
104 len2 = dax_direct_access(dax_dev, pgoff_end, 1, &end_kaddr, &end_pfn);
107 if (len < 1 || len2 < 1) {
108 pr_debug("%s: error: dax access failed (%ld)\n",
109 bdevname(bdev, buf), len < 1 ? len : len2);
113 if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
115 * An arch that has enabled the pmem api should also
116 * have its drivers support pfn_t_devmap()
118 * This is a developer warning and should not trigger in
119 * production. dax_flush() will crash since it depends
120 * on being able to do (page_address(pfn_to_page())).
122 WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
124 } else if (pfn_t_devmap(pfn) && pfn_t_devmap(end_pfn)) {
125 struct dev_pagemap *pgmap, *end_pgmap;
127 pgmap = get_dev_pagemap(pfn_t_to_pfn(pfn), NULL);
128 end_pgmap = get_dev_pagemap(pfn_t_to_pfn(end_pfn), NULL);
129 if (pgmap && pgmap == end_pgmap && pgmap->type == MEMORY_DEVICE_FS_DAX
130 && pfn_t_to_page(pfn)->pgmap == pgmap
131 && pfn_t_to_page(end_pfn)->pgmap == pgmap
132 && pfn_t_to_pfn(pfn) == PHYS_PFN(__pa(kaddr))
133 && pfn_t_to_pfn(end_pfn) == PHYS_PFN(__pa(end_kaddr)))
135 put_dev_pagemap(pgmap);
136 put_dev_pagemap(end_pgmap);
141 pr_debug("%s: error: dax support not enabled\n",
142 bdevname(bdev, buf));
147 EXPORT_SYMBOL_GPL(__generic_fsdax_supported);
150 * __bdev_dax_supported() - Check if the device supports dax for filesystem
151 * @bdev: block device to check
152 * @blocksize: The block size of the device
154 * This is a library function for filesystems to check if the block device
155 * can be mounted with dax option.
157 * Return: true if supported, false if unsupported
159 bool __bdev_dax_supported(struct block_device *bdev, int blocksize)
161 struct dax_device *dax_dev;
162 struct request_queue *q;
163 char buf[BDEVNAME_SIZE];
167 q = bdev_get_queue(bdev);
168 if (!q || !blk_queue_dax(q)) {
169 pr_debug("%s: error: request queue doesn't support dax\n",
170 bdevname(bdev, buf));
174 dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
176 pr_debug("%s: error: device does not support dax\n",
177 bdevname(bdev, buf));
181 id = dax_read_lock();
182 ret = dax_supported(dax_dev, bdev, blocksize, 0,
183 i_size_read(bdev->bd_inode) / 512);
190 EXPORT_SYMBOL_GPL(__bdev_dax_supported);
193 enum dax_device_flags {
194 /* !alive + rcu grace period == no new operations / mappings */
196 /* gate whether dax_flush() calls the low level flush routine */
201 * struct dax_device - anchor object for dax services
203 * @cdev: optional character interface for "device dax"
204 * @host: optional name for lookups where the device path is not available
205 * @private: dax driver private data
206 * @flags: state and boolean properties
209 struct hlist_node list;
215 const struct dax_operations *ops;
218 static ssize_t write_cache_show(struct device *dev,
219 struct device_attribute *attr, char *buf)
221 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
224 WARN_ON_ONCE(!dax_dev);
228 rc = sprintf(buf, "%d\n", !!dax_write_cache_enabled(dax_dev));
233 static ssize_t write_cache_store(struct device *dev,
234 struct device_attribute *attr, const char *buf, size_t len)
237 int rc = strtobool(buf, &write_cache);
238 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
240 WARN_ON_ONCE(!dax_dev);
247 dax_write_cache(dax_dev, write_cache);
252 static DEVICE_ATTR_RW(write_cache);
254 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
256 struct device *dev = container_of(kobj, typeof(*dev), kobj);
257 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
259 WARN_ON_ONCE(!dax_dev);
263 #ifndef CONFIG_ARCH_HAS_PMEM_API
264 if (a == &dev_attr_write_cache.attr)
270 static struct attribute *dax_attributes[] = {
271 &dev_attr_write_cache.attr,
275 struct attribute_group dax_attribute_group = {
277 .attrs = dax_attributes,
278 .is_visible = dax_visible,
280 EXPORT_SYMBOL_GPL(dax_attribute_group);
283 * dax_direct_access() - translate a device pgoff to an absolute pfn
284 * @dax_dev: a dax_device instance representing the logical memory range
285 * @pgoff: offset in pages from the start of the device to translate
286 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
287 * @kaddr: output parameter that returns a virtual address mapping of pfn
288 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
290 * Return: negative errno if an error occurs, otherwise the number of
291 * pages accessible at the device relative @pgoff.
293 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
294 void **kaddr, pfn_t *pfn)
301 if (!dax_alive(dax_dev))
307 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
311 return min(avail, nr_pages);
313 EXPORT_SYMBOL_GPL(dax_direct_access);
315 bool dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
316 int blocksize, sector_t start, sector_t len)
318 if (!dax_alive(dax_dev))
321 return dax_dev->ops->dax_supported(dax_dev, bdev, blocksize, start, len);
324 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
325 size_t bytes, struct iov_iter *i)
327 if (!dax_alive(dax_dev))
330 return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
332 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
334 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
335 size_t bytes, struct iov_iter *i)
337 if (!dax_alive(dax_dev))
340 return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i);
342 EXPORT_SYMBOL_GPL(dax_copy_to_iter);
344 #ifdef CONFIG_ARCH_HAS_PMEM_API
345 void arch_wb_cache_pmem(void *addr, size_t size);
346 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
348 if (unlikely(!dax_write_cache_enabled(dax_dev)))
351 arch_wb_cache_pmem(addr, size);
354 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
358 EXPORT_SYMBOL_GPL(dax_flush);
360 void dax_write_cache(struct dax_device *dax_dev, bool wc)
363 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
365 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
367 EXPORT_SYMBOL_GPL(dax_write_cache);
369 bool dax_write_cache_enabled(struct dax_device *dax_dev)
371 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
373 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
375 bool dax_alive(struct dax_device *dax_dev)
377 lockdep_assert_held(&dax_srcu);
378 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
380 EXPORT_SYMBOL_GPL(dax_alive);
382 static int dax_host_hash(const char *host)
384 return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
388 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
389 * that any fault handlers or operations that might have seen
390 * dax_alive(), have completed. Any operations that start after
391 * synchronize_srcu() has run will abort upon seeing !dax_alive().
393 void kill_dax(struct dax_device *dax_dev)
398 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
400 synchronize_srcu(&dax_srcu);
402 spin_lock(&dax_host_lock);
403 hlist_del_init(&dax_dev->list);
404 spin_unlock(&dax_host_lock);
406 EXPORT_SYMBOL_GPL(kill_dax);
408 void run_dax(struct dax_device *dax_dev)
410 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
412 EXPORT_SYMBOL_GPL(run_dax);
414 static struct inode *dax_alloc_inode(struct super_block *sb)
416 struct dax_device *dax_dev;
419 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
423 inode = &dax_dev->inode;
428 static struct dax_device *to_dax_dev(struct inode *inode)
430 return container_of(inode, struct dax_device, inode);
433 static void dax_free_inode(struct inode *inode)
435 struct dax_device *dax_dev = to_dax_dev(inode);
436 kfree(dax_dev->host);
437 dax_dev->host = NULL;
439 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
440 kmem_cache_free(dax_cache, dax_dev);
443 static void dax_destroy_inode(struct inode *inode)
445 struct dax_device *dax_dev = to_dax_dev(inode);
446 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
447 "kill_dax() must be called before final iput()\n");
450 static const struct super_operations dax_sops = {
451 .statfs = simple_statfs,
452 .alloc_inode = dax_alloc_inode,
453 .destroy_inode = dax_destroy_inode,
454 .free_inode = dax_free_inode,
455 .drop_inode = generic_delete_inode,
458 static struct dentry *dax_mount(struct file_system_type *fs_type,
459 int flags, const char *dev_name, void *data)
461 return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
464 static struct file_system_type dax_fs_type = {
467 .kill_sb = kill_anon_super,
470 static int dax_test(struct inode *inode, void *data)
472 dev_t devt = *(dev_t *) data;
474 return inode->i_rdev == devt;
477 static int dax_set(struct inode *inode, void *data)
479 dev_t devt = *(dev_t *) data;
481 inode->i_rdev = devt;
485 static struct dax_device *dax_dev_get(dev_t devt)
487 struct dax_device *dax_dev;
490 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
491 dax_test, dax_set, &devt);
496 dax_dev = to_dax_dev(inode);
497 if (inode->i_state & I_NEW) {
498 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
499 inode->i_cdev = &dax_dev->cdev;
500 inode->i_mode = S_IFCHR;
501 inode->i_flags = S_DAX;
502 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
503 unlock_new_inode(inode);
509 static void dax_add_host(struct dax_device *dax_dev, const char *host)
514 * Unconditionally init dax_dev since it's coming from a
515 * non-zeroed slab cache
517 INIT_HLIST_NODE(&dax_dev->list);
518 dax_dev->host = host;
522 hash = dax_host_hash(host);
523 spin_lock(&dax_host_lock);
524 hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
525 spin_unlock(&dax_host_lock);
528 struct dax_device *alloc_dax(void *private, const char *__host,
529 const struct dax_operations *ops)
531 struct dax_device *dax_dev;
536 host = kstrdup(__host, GFP_KERNEL);
540 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
544 devt = MKDEV(MAJOR(dax_devt), minor);
545 dax_dev = dax_dev_get(devt);
549 dax_add_host(dax_dev, host);
551 dax_dev->private = private;
555 ida_simple_remove(&dax_minor_ida, minor);
560 EXPORT_SYMBOL_GPL(alloc_dax);
562 void put_dax(struct dax_device *dax_dev)
566 iput(&dax_dev->inode);
568 EXPORT_SYMBOL_GPL(put_dax);
571 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
572 * @host: alternate name for the device registered by a dax driver
574 struct dax_device *dax_get_by_host(const char *host)
576 struct dax_device *dax_dev, *found = NULL;
582 hash = dax_host_hash(host);
584 id = dax_read_lock();
585 spin_lock(&dax_host_lock);
586 hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
587 if (!dax_alive(dax_dev)
588 || strcmp(host, dax_dev->host) != 0)
591 if (igrab(&dax_dev->inode))
595 spin_unlock(&dax_host_lock);
600 EXPORT_SYMBOL_GPL(dax_get_by_host);
603 * inode_dax: convert a public inode into its dax_dev
604 * @inode: An inode with i_cdev pointing to a dax_dev
606 * Note this is not equivalent to to_dax_dev() which is for private
607 * internal use where we know the inode filesystem type == dax_fs_type.
609 struct dax_device *inode_dax(struct inode *inode)
611 struct cdev *cdev = inode->i_cdev;
613 return container_of(cdev, struct dax_device, cdev);
615 EXPORT_SYMBOL_GPL(inode_dax);
617 struct inode *dax_inode(struct dax_device *dax_dev)
619 return &dax_dev->inode;
621 EXPORT_SYMBOL_GPL(dax_inode);
623 void *dax_get_private(struct dax_device *dax_dev)
625 if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
627 return dax_dev->private;
629 EXPORT_SYMBOL_GPL(dax_get_private);
631 static void init_once(void *_dax_dev)
633 struct dax_device *dax_dev = _dax_dev;
634 struct inode *inode = &dax_dev->inode;
636 memset(dax_dev, 0, sizeof(*dax_dev));
637 inode_init_once(inode);
640 static int dax_fs_init(void)
644 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
645 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
646 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
651 rc = register_filesystem(&dax_fs_type);
653 goto err_register_fs;
655 dax_mnt = kern_mount(&dax_fs_type);
656 if (IS_ERR(dax_mnt)) {
657 rc = PTR_ERR(dax_mnt);
660 dax_superblock = dax_mnt->mnt_sb;
665 unregister_filesystem(&dax_fs_type);
667 kmem_cache_destroy(dax_cache);
672 static void dax_fs_exit(void)
674 kern_unmount(dax_mnt);
675 unregister_filesystem(&dax_fs_type);
676 kmem_cache_destroy(dax_cache);
679 static int __init dax_core_init(void)
687 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
697 unregister_chrdev_region(dax_devt, MINORMASK+1);
703 static void __exit dax_core_exit(void)
705 unregister_chrdev_region(dax_devt, MINORMASK+1);
706 ida_destroy(&dax_minor_ida);
710 MODULE_AUTHOR("Intel Corporation");
711 MODULE_LICENSE("GPL v2");
712 subsys_initcall(dax_core_init);
713 module_exit(dax_core_exit);