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/pseudo_fs.h>
9 #include <linux/magic.h>
10 #include <linux/genhd.h>
11 #include <linux/pfn_t.h>
12 #include <linux/cdev.h>
13 #include <linux/hash.h>
14 #include <linux/slab.h>
15 #include <linux/uio.h>
16 #include <linux/dax.h>
18 #include "dax-private.h"
20 static dev_t dax_devt;
21 DEFINE_STATIC_SRCU(dax_srcu);
22 static struct vfsmount *dax_mnt;
23 static DEFINE_IDA(dax_minor_ida);
24 static struct kmem_cache *dax_cache __read_mostly;
25 static struct super_block *dax_superblock __read_mostly;
27 #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
28 static struct hlist_head dax_host_list[DAX_HASH_SIZE];
29 static DEFINE_SPINLOCK(dax_host_lock);
31 int dax_read_lock(void)
33 return srcu_read_lock(&dax_srcu);
35 EXPORT_SYMBOL_GPL(dax_read_lock);
37 void dax_read_unlock(int id)
39 srcu_read_unlock(&dax_srcu, id);
41 EXPORT_SYMBOL_GPL(dax_read_unlock);
44 #include <linux/blkdev.h>
46 int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
49 phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512;
52 *pgoff = PHYS_PFN(phys_off);
53 if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
57 EXPORT_SYMBOL(bdev_dax_pgoff);
59 #if IS_ENABLED(CONFIG_FS_DAX)
60 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
62 if (!blk_queue_dax(bdev->bd_queue))
64 return fs_dax_get_by_host(bdev->bd_disk->disk_name);
66 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
69 bool __generic_fsdax_supported(struct dax_device *dax_dev,
70 struct block_device *bdev, int blocksize, sector_t start,
73 bool dax_enabled = false;
74 pgoff_t pgoff, pgoff_end;
75 char buf[BDEVNAME_SIZE];
76 void *kaddr, *end_kaddr;
82 if (blocksize != PAGE_SIZE) {
83 pr_debug("%s: error: unsupported blocksize for dax\n",
88 err = bdev_dax_pgoff(bdev, start, PAGE_SIZE, &pgoff);
90 pr_debug("%s: error: unaligned partition for dax\n",
95 last_page = PFN_DOWN((start + sectors - 1) * 512) * PAGE_SIZE / 512;
96 err = bdev_dax_pgoff(bdev, last_page, PAGE_SIZE, &pgoff_end);
98 pr_debug("%s: error: unaligned partition for dax\n",
103 id = dax_read_lock();
104 len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
105 len2 = dax_direct_access(dax_dev, pgoff_end, 1, &end_kaddr, &end_pfn);
108 if (len < 1 || len2 < 1) {
109 pr_debug("%s: error: dax access failed (%ld)\n",
110 bdevname(bdev, buf), len < 1 ? len : len2);
114 if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
116 * An arch that has enabled the pmem api should also
117 * have its drivers support pfn_t_devmap()
119 * This is a developer warning and should not trigger in
120 * production. dax_flush() will crash since it depends
121 * on being able to do (page_address(pfn_to_page())).
123 WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
125 } else if (pfn_t_devmap(pfn) && pfn_t_devmap(end_pfn)) {
126 struct dev_pagemap *pgmap, *end_pgmap;
128 pgmap = get_dev_pagemap(pfn_t_to_pfn(pfn), NULL);
129 end_pgmap = get_dev_pagemap(pfn_t_to_pfn(end_pfn), NULL);
130 if (pgmap && pgmap == end_pgmap && pgmap->type == MEMORY_DEVICE_FS_DAX
131 && pfn_t_to_page(pfn)->pgmap == pgmap
132 && pfn_t_to_page(end_pfn)->pgmap == pgmap
133 && pfn_t_to_pfn(pfn) == PHYS_PFN(__pa(kaddr))
134 && pfn_t_to_pfn(end_pfn) == PHYS_PFN(__pa(end_kaddr)))
136 put_dev_pagemap(pgmap);
137 put_dev_pagemap(end_pgmap);
142 pr_debug("%s: error: dax support not enabled\n",
143 bdevname(bdev, buf));
148 EXPORT_SYMBOL_GPL(__generic_fsdax_supported);
151 * __bdev_dax_supported() - Check if the device supports dax for filesystem
152 * @bdev: block device to check
153 * @blocksize: The block size of the device
155 * This is a library function for filesystems to check if the block device
156 * can be mounted with dax option.
158 * Return: true if supported, false if unsupported
160 bool __bdev_dax_supported(struct block_device *bdev, int blocksize)
162 struct dax_device *dax_dev;
163 struct request_queue *q;
164 char buf[BDEVNAME_SIZE];
168 q = bdev_get_queue(bdev);
169 if (!q || !blk_queue_dax(q)) {
170 pr_debug("%s: error: request queue doesn't support dax\n",
171 bdevname(bdev, buf));
175 dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
177 pr_debug("%s: error: device does not support dax\n",
178 bdevname(bdev, buf));
182 id = dax_read_lock();
183 ret = dax_supported(dax_dev, bdev, blocksize, 0,
184 i_size_read(bdev->bd_inode) / 512);
191 EXPORT_SYMBOL_GPL(__bdev_dax_supported);
194 enum dax_device_flags {
195 /* !alive + rcu grace period == no new operations / mappings */
197 /* gate whether dax_flush() calls the low level flush routine */
199 /* flag to check if device supports synchronous flush */
204 * struct dax_device - anchor object for dax services
206 * @cdev: optional character interface for "device dax"
207 * @host: optional name for lookups where the device path is not available
208 * @private: dax driver private data
209 * @flags: state and boolean properties
212 struct hlist_node list;
218 const struct dax_operations *ops;
221 static ssize_t write_cache_show(struct device *dev,
222 struct device_attribute *attr, char *buf)
224 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
227 WARN_ON_ONCE(!dax_dev);
231 rc = sprintf(buf, "%d\n", !!dax_write_cache_enabled(dax_dev));
236 static ssize_t write_cache_store(struct device *dev,
237 struct device_attribute *attr, const char *buf, size_t len)
240 int rc = strtobool(buf, &write_cache);
241 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
243 WARN_ON_ONCE(!dax_dev);
250 dax_write_cache(dax_dev, write_cache);
255 static DEVICE_ATTR_RW(write_cache);
257 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
259 struct device *dev = container_of(kobj, typeof(*dev), kobj);
260 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
262 WARN_ON_ONCE(!dax_dev);
266 #ifndef CONFIG_ARCH_HAS_PMEM_API
267 if (a == &dev_attr_write_cache.attr)
273 static struct attribute *dax_attributes[] = {
274 &dev_attr_write_cache.attr,
278 struct attribute_group dax_attribute_group = {
280 .attrs = dax_attributes,
281 .is_visible = dax_visible,
283 EXPORT_SYMBOL_GPL(dax_attribute_group);
286 * dax_direct_access() - translate a device pgoff to an absolute pfn
287 * @dax_dev: a dax_device instance representing the logical memory range
288 * @pgoff: offset in pages from the start of the device to translate
289 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
290 * @kaddr: output parameter that returns a virtual address mapping of pfn
291 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
293 * Return: negative errno if an error occurs, otherwise the number of
294 * pages accessible at the device relative @pgoff.
296 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
297 void **kaddr, pfn_t *pfn)
304 if (!dax_alive(dax_dev))
310 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
314 return min(avail, nr_pages);
316 EXPORT_SYMBOL_GPL(dax_direct_access);
318 bool dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
319 int blocksize, sector_t start, sector_t len)
324 if (!dax_alive(dax_dev))
327 return dax_dev->ops->dax_supported(dax_dev, bdev, blocksize, start, len);
329 EXPORT_SYMBOL_GPL(dax_supported);
331 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
332 size_t bytes, struct iov_iter *i)
334 if (!dax_alive(dax_dev))
337 return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
339 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
341 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
342 size_t bytes, struct iov_iter *i)
344 if (!dax_alive(dax_dev))
347 return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i);
349 EXPORT_SYMBOL_GPL(dax_copy_to_iter);
351 #ifdef CONFIG_ARCH_HAS_PMEM_API
352 void arch_wb_cache_pmem(void *addr, size_t size);
353 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
355 if (unlikely(!dax_write_cache_enabled(dax_dev)))
358 arch_wb_cache_pmem(addr, size);
361 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
365 EXPORT_SYMBOL_GPL(dax_flush);
367 void dax_write_cache(struct dax_device *dax_dev, bool wc)
370 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
372 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
374 EXPORT_SYMBOL_GPL(dax_write_cache);
376 bool dax_write_cache_enabled(struct dax_device *dax_dev)
378 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
380 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
382 bool __dax_synchronous(struct dax_device *dax_dev)
384 return test_bit(DAXDEV_SYNC, &dax_dev->flags);
386 EXPORT_SYMBOL_GPL(__dax_synchronous);
388 void __set_dax_synchronous(struct dax_device *dax_dev)
390 set_bit(DAXDEV_SYNC, &dax_dev->flags);
392 EXPORT_SYMBOL_GPL(__set_dax_synchronous);
394 bool dax_alive(struct dax_device *dax_dev)
396 lockdep_assert_held(&dax_srcu);
397 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
399 EXPORT_SYMBOL_GPL(dax_alive);
401 static int dax_host_hash(const char *host)
403 return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
407 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
408 * that any fault handlers or operations that might have seen
409 * dax_alive(), have completed. Any operations that start after
410 * synchronize_srcu() has run will abort upon seeing !dax_alive().
412 void kill_dax(struct dax_device *dax_dev)
417 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
419 synchronize_srcu(&dax_srcu);
421 spin_lock(&dax_host_lock);
422 hlist_del_init(&dax_dev->list);
423 spin_unlock(&dax_host_lock);
425 EXPORT_SYMBOL_GPL(kill_dax);
427 void run_dax(struct dax_device *dax_dev)
429 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
431 EXPORT_SYMBOL_GPL(run_dax);
433 static struct inode *dax_alloc_inode(struct super_block *sb)
435 struct dax_device *dax_dev;
438 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
442 inode = &dax_dev->inode;
447 static struct dax_device *to_dax_dev(struct inode *inode)
449 return container_of(inode, struct dax_device, inode);
452 static void dax_free_inode(struct inode *inode)
454 struct dax_device *dax_dev = to_dax_dev(inode);
455 kfree(dax_dev->host);
456 dax_dev->host = NULL;
458 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
459 kmem_cache_free(dax_cache, dax_dev);
462 static void dax_destroy_inode(struct inode *inode)
464 struct dax_device *dax_dev = to_dax_dev(inode);
465 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
466 "kill_dax() must be called before final iput()\n");
469 static const struct super_operations dax_sops = {
470 .statfs = simple_statfs,
471 .alloc_inode = dax_alloc_inode,
472 .destroy_inode = dax_destroy_inode,
473 .free_inode = dax_free_inode,
474 .drop_inode = generic_delete_inode,
477 static int dax_init_fs_context(struct fs_context *fc)
479 struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC);
482 ctx->ops = &dax_sops;
486 static struct file_system_type dax_fs_type = {
488 .init_fs_context = dax_init_fs_context,
489 .kill_sb = kill_anon_super,
492 static int dax_test(struct inode *inode, void *data)
494 dev_t devt = *(dev_t *) data;
496 return inode->i_rdev == devt;
499 static int dax_set(struct inode *inode, void *data)
501 dev_t devt = *(dev_t *) data;
503 inode->i_rdev = devt;
507 static struct dax_device *dax_dev_get(dev_t devt)
509 struct dax_device *dax_dev;
512 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
513 dax_test, dax_set, &devt);
518 dax_dev = to_dax_dev(inode);
519 if (inode->i_state & I_NEW) {
520 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
521 inode->i_cdev = &dax_dev->cdev;
522 inode->i_mode = S_IFCHR;
523 inode->i_flags = S_DAX;
524 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
525 unlock_new_inode(inode);
531 static void dax_add_host(struct dax_device *dax_dev, const char *host)
536 * Unconditionally init dax_dev since it's coming from a
537 * non-zeroed slab cache
539 INIT_HLIST_NODE(&dax_dev->list);
540 dax_dev->host = host;
544 hash = dax_host_hash(host);
545 spin_lock(&dax_host_lock);
546 hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
547 spin_unlock(&dax_host_lock);
550 struct dax_device *alloc_dax(void *private, const char *__host,
551 const struct dax_operations *ops, unsigned long flags)
553 struct dax_device *dax_dev;
558 host = kstrdup(__host, GFP_KERNEL);
562 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
566 devt = MKDEV(MAJOR(dax_devt), minor);
567 dax_dev = dax_dev_get(devt);
571 dax_add_host(dax_dev, host);
573 dax_dev->private = private;
574 if (flags & DAXDEV_F_SYNC)
575 set_dax_synchronous(dax_dev);
580 ida_simple_remove(&dax_minor_ida, minor);
585 EXPORT_SYMBOL_GPL(alloc_dax);
587 void put_dax(struct dax_device *dax_dev)
591 iput(&dax_dev->inode);
593 EXPORT_SYMBOL_GPL(put_dax);
596 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
597 * @host: alternate name for the device registered by a dax driver
599 struct dax_device *dax_get_by_host(const char *host)
601 struct dax_device *dax_dev, *found = NULL;
607 hash = dax_host_hash(host);
609 id = dax_read_lock();
610 spin_lock(&dax_host_lock);
611 hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
612 if (!dax_alive(dax_dev)
613 || strcmp(host, dax_dev->host) != 0)
616 if (igrab(&dax_dev->inode))
620 spin_unlock(&dax_host_lock);
625 EXPORT_SYMBOL_GPL(dax_get_by_host);
628 * inode_dax: convert a public inode into its dax_dev
629 * @inode: An inode with i_cdev pointing to a dax_dev
631 * Note this is not equivalent to to_dax_dev() which is for private
632 * internal use where we know the inode filesystem type == dax_fs_type.
634 struct dax_device *inode_dax(struct inode *inode)
636 struct cdev *cdev = inode->i_cdev;
638 return container_of(cdev, struct dax_device, cdev);
640 EXPORT_SYMBOL_GPL(inode_dax);
642 struct inode *dax_inode(struct dax_device *dax_dev)
644 return &dax_dev->inode;
646 EXPORT_SYMBOL_GPL(dax_inode);
648 void *dax_get_private(struct dax_device *dax_dev)
650 if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
652 return dax_dev->private;
654 EXPORT_SYMBOL_GPL(dax_get_private);
656 static void init_once(void *_dax_dev)
658 struct dax_device *dax_dev = _dax_dev;
659 struct inode *inode = &dax_dev->inode;
661 memset(dax_dev, 0, sizeof(*dax_dev));
662 inode_init_once(inode);
665 static int dax_fs_init(void)
669 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
670 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
671 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
676 dax_mnt = kern_mount(&dax_fs_type);
677 if (IS_ERR(dax_mnt)) {
678 rc = PTR_ERR(dax_mnt);
681 dax_superblock = dax_mnt->mnt_sb;
686 kmem_cache_destroy(dax_cache);
691 static void dax_fs_exit(void)
693 kern_unmount(dax_mnt);
695 kmem_cache_destroy(dax_cache);
698 static int __init dax_core_init(void)
706 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
716 unregister_chrdev_region(dax_devt, MINORMASK+1);
722 static void __exit dax_core_exit(void)
725 unregister_chrdev_region(dax_devt, MINORMASK+1);
726 ida_destroy(&dax_minor_ida);
730 MODULE_AUTHOR("Intel Corporation");
731 MODULE_LICENSE("GPL v2");
732 subsys_initcall(dax_core_init);
733 module_exit(dax_core_exit);