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_disk->queue))
64 return 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_info("%s: error: unsupported blocksize for dax\n",
88 err = bdev_dax_pgoff(bdev, start, PAGE_SIZE, &pgoff);
90 pr_info("%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_info("%s: error: unaligned partition for dax\n",
103 if (!dax_dev && !bdev_dax_supported(bdev, blocksize)) {
104 pr_debug("%s: error: dax unsupported by block device\n",
105 bdevname(bdev, buf));
109 id = dax_read_lock();
110 len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
111 len2 = dax_direct_access(dax_dev, pgoff_end, 1, &end_kaddr, &end_pfn);
113 if (len < 1 || len2 < 1) {
114 pr_info("%s: error: dax access failed (%ld)\n",
115 bdevname(bdev, buf), len < 1 ? len : len2);
120 if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
122 * An arch that has enabled the pmem api should also
123 * have its drivers support pfn_t_devmap()
125 * This is a developer warning and should not trigger in
126 * production. dax_flush() will crash since it depends
127 * on being able to do (page_address(pfn_to_page())).
129 WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
131 } else if (pfn_t_devmap(pfn) && pfn_t_devmap(end_pfn)) {
132 struct dev_pagemap *pgmap, *end_pgmap;
134 pgmap = get_dev_pagemap(pfn_t_to_pfn(pfn), NULL);
135 end_pgmap = get_dev_pagemap(pfn_t_to_pfn(end_pfn), NULL);
136 if (pgmap && pgmap == end_pgmap && pgmap->type == MEMORY_DEVICE_FS_DAX
137 && pfn_t_to_page(pfn)->pgmap == pgmap
138 && pfn_t_to_page(end_pfn)->pgmap == pgmap
139 && pfn_t_to_pfn(pfn) == PHYS_PFN(__pa(kaddr))
140 && pfn_t_to_pfn(end_pfn) == PHYS_PFN(__pa(end_kaddr)))
142 put_dev_pagemap(pgmap);
143 put_dev_pagemap(end_pgmap);
149 pr_info("%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 */
206 /* flag to check if device supports synchronous flush */
211 * struct dax_device - anchor object for dax services
213 * @cdev: optional character interface for "device dax"
214 * @host: optional name for lookups where the device path is not available
215 * @private: dax driver private data
216 * @flags: state and boolean properties
219 struct hlist_node list;
225 const struct dax_operations *ops;
228 static ssize_t write_cache_show(struct device *dev,
229 struct device_attribute *attr, char *buf)
231 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
234 WARN_ON_ONCE(!dax_dev);
238 rc = sprintf(buf, "%d\n", !!dax_write_cache_enabled(dax_dev));
243 static ssize_t write_cache_store(struct device *dev,
244 struct device_attribute *attr, const char *buf, size_t len)
247 int rc = strtobool(buf, &write_cache);
248 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
250 WARN_ON_ONCE(!dax_dev);
257 dax_write_cache(dax_dev, write_cache);
262 static DEVICE_ATTR_RW(write_cache);
264 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
266 struct device *dev = container_of(kobj, typeof(*dev), kobj);
267 struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
269 WARN_ON_ONCE(!dax_dev);
273 #ifndef CONFIG_ARCH_HAS_PMEM_API
274 if (a == &dev_attr_write_cache.attr)
280 static struct attribute *dax_attributes[] = {
281 &dev_attr_write_cache.attr,
285 struct attribute_group dax_attribute_group = {
287 .attrs = dax_attributes,
288 .is_visible = dax_visible,
290 EXPORT_SYMBOL_GPL(dax_attribute_group);
293 * dax_direct_access() - translate a device pgoff to an absolute pfn
294 * @dax_dev: a dax_device instance representing the logical memory range
295 * @pgoff: offset in pages from the start of the device to translate
296 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
297 * @kaddr: output parameter that returns a virtual address mapping of pfn
298 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
300 * Return: negative errno if an error occurs, otherwise the number of
301 * pages accessible at the device relative @pgoff.
303 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
304 void **kaddr, pfn_t *pfn)
311 if (!dax_alive(dax_dev))
317 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
321 return min(avail, nr_pages);
323 EXPORT_SYMBOL_GPL(dax_direct_access);
325 bool dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
326 int blocksize, sector_t start, sector_t len)
328 if (!dax_alive(dax_dev))
331 return dax_dev->ops->dax_supported(dax_dev, bdev, blocksize, start, len);
334 size_t dax_copy_from_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_from_iter(dax_dev, pgoff, addr, bytes, i);
342 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
344 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
345 size_t bytes, struct iov_iter *i)
347 if (!dax_alive(dax_dev))
350 return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i);
352 EXPORT_SYMBOL_GPL(dax_copy_to_iter);
354 int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
357 if (!dax_alive(dax_dev))
360 * There are no callers that want to zero more than one page as of now.
361 * Once users are there, this check can be removed after the
362 * device mapper code has been updated to split ranges across targets.
367 return dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages);
369 EXPORT_SYMBOL_GPL(dax_zero_page_range);
371 #ifdef CONFIG_ARCH_HAS_PMEM_API
372 void arch_wb_cache_pmem(void *addr, size_t size);
373 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
375 if (unlikely(!dax_write_cache_enabled(dax_dev)))
378 arch_wb_cache_pmem(addr, size);
381 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
385 EXPORT_SYMBOL_GPL(dax_flush);
387 void dax_write_cache(struct dax_device *dax_dev, bool wc)
390 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
392 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
394 EXPORT_SYMBOL_GPL(dax_write_cache);
396 bool dax_write_cache_enabled(struct dax_device *dax_dev)
398 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
400 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
402 bool __dax_synchronous(struct dax_device *dax_dev)
404 return test_bit(DAXDEV_SYNC, &dax_dev->flags);
406 EXPORT_SYMBOL_GPL(__dax_synchronous);
408 void __set_dax_synchronous(struct dax_device *dax_dev)
410 set_bit(DAXDEV_SYNC, &dax_dev->flags);
412 EXPORT_SYMBOL_GPL(__set_dax_synchronous);
414 bool dax_alive(struct dax_device *dax_dev)
416 lockdep_assert_held(&dax_srcu);
417 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
419 EXPORT_SYMBOL_GPL(dax_alive);
421 static int dax_host_hash(const char *host)
423 return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
427 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
428 * that any fault handlers or operations that might have seen
429 * dax_alive(), have completed. Any operations that start after
430 * synchronize_srcu() has run will abort upon seeing !dax_alive().
432 void kill_dax(struct dax_device *dax_dev)
437 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
439 synchronize_srcu(&dax_srcu);
441 spin_lock(&dax_host_lock);
442 hlist_del_init(&dax_dev->list);
443 spin_unlock(&dax_host_lock);
445 EXPORT_SYMBOL_GPL(kill_dax);
447 void run_dax(struct dax_device *dax_dev)
449 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
451 EXPORT_SYMBOL_GPL(run_dax);
453 static struct inode *dax_alloc_inode(struct super_block *sb)
455 struct dax_device *dax_dev;
458 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
462 inode = &dax_dev->inode;
467 static struct dax_device *to_dax_dev(struct inode *inode)
469 return container_of(inode, struct dax_device, inode);
472 static void dax_free_inode(struct inode *inode)
474 struct dax_device *dax_dev = to_dax_dev(inode);
475 kfree(dax_dev->host);
476 dax_dev->host = NULL;
478 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
479 kmem_cache_free(dax_cache, dax_dev);
482 static void dax_destroy_inode(struct inode *inode)
484 struct dax_device *dax_dev = to_dax_dev(inode);
485 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
486 "kill_dax() must be called before final iput()\n");
489 static const struct super_operations dax_sops = {
490 .statfs = simple_statfs,
491 .alloc_inode = dax_alloc_inode,
492 .destroy_inode = dax_destroy_inode,
493 .free_inode = dax_free_inode,
494 .drop_inode = generic_delete_inode,
497 static int dax_init_fs_context(struct fs_context *fc)
499 struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC);
502 ctx->ops = &dax_sops;
506 static struct file_system_type dax_fs_type = {
508 .init_fs_context = dax_init_fs_context,
509 .kill_sb = kill_anon_super,
512 static int dax_test(struct inode *inode, void *data)
514 dev_t devt = *(dev_t *) data;
516 return inode->i_rdev == devt;
519 static int dax_set(struct inode *inode, void *data)
521 dev_t devt = *(dev_t *) data;
523 inode->i_rdev = devt;
527 static struct dax_device *dax_dev_get(dev_t devt)
529 struct dax_device *dax_dev;
532 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
533 dax_test, dax_set, &devt);
538 dax_dev = to_dax_dev(inode);
539 if (inode->i_state & I_NEW) {
540 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
541 inode->i_cdev = &dax_dev->cdev;
542 inode->i_mode = S_IFCHR;
543 inode->i_flags = S_DAX;
544 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
545 unlock_new_inode(inode);
551 static void dax_add_host(struct dax_device *dax_dev, const char *host)
556 * Unconditionally init dax_dev since it's coming from a
557 * non-zeroed slab cache
559 INIT_HLIST_NODE(&dax_dev->list);
560 dax_dev->host = host;
564 hash = dax_host_hash(host);
565 spin_lock(&dax_host_lock);
566 hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
567 spin_unlock(&dax_host_lock);
570 struct dax_device *alloc_dax(void *private, const char *__host,
571 const struct dax_operations *ops, unsigned long flags)
573 struct dax_device *dax_dev;
578 if (ops && !ops->zero_page_range) {
579 pr_debug("%s: error: device does not provide dax"
580 " operation zero_page_range()\n",
581 __host ? __host : "Unknown");
582 return ERR_PTR(-EINVAL);
585 host = kstrdup(__host, GFP_KERNEL);
587 return ERR_PTR(-ENOMEM);
589 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
593 devt = MKDEV(MAJOR(dax_devt), minor);
594 dax_dev = dax_dev_get(devt);
598 dax_add_host(dax_dev, host);
600 dax_dev->private = private;
601 if (flags & DAXDEV_F_SYNC)
602 set_dax_synchronous(dax_dev);
607 ida_simple_remove(&dax_minor_ida, minor);
610 return ERR_PTR(-ENOMEM);
612 EXPORT_SYMBOL_GPL(alloc_dax);
614 void put_dax(struct dax_device *dax_dev)
618 iput(&dax_dev->inode);
620 EXPORT_SYMBOL_GPL(put_dax);
623 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
624 * @host: alternate name for the device registered by a dax driver
626 struct dax_device *dax_get_by_host(const char *host)
628 struct dax_device *dax_dev, *found = NULL;
634 hash = dax_host_hash(host);
636 id = dax_read_lock();
637 spin_lock(&dax_host_lock);
638 hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
639 if (!dax_alive(dax_dev)
640 || strcmp(host, dax_dev->host) != 0)
643 if (igrab(&dax_dev->inode))
647 spin_unlock(&dax_host_lock);
652 EXPORT_SYMBOL_GPL(dax_get_by_host);
655 * inode_dax: convert a public inode into its dax_dev
656 * @inode: An inode with i_cdev pointing to a dax_dev
658 * Note this is not equivalent to to_dax_dev() which is for private
659 * internal use where we know the inode filesystem type == dax_fs_type.
661 struct dax_device *inode_dax(struct inode *inode)
663 struct cdev *cdev = inode->i_cdev;
665 return container_of(cdev, struct dax_device, cdev);
667 EXPORT_SYMBOL_GPL(inode_dax);
669 struct inode *dax_inode(struct dax_device *dax_dev)
671 return &dax_dev->inode;
673 EXPORT_SYMBOL_GPL(dax_inode);
675 void *dax_get_private(struct dax_device *dax_dev)
677 if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
679 return dax_dev->private;
681 EXPORT_SYMBOL_GPL(dax_get_private);
683 static void init_once(void *_dax_dev)
685 struct dax_device *dax_dev = _dax_dev;
686 struct inode *inode = &dax_dev->inode;
688 memset(dax_dev, 0, sizeof(*dax_dev));
689 inode_init_once(inode);
692 static int dax_fs_init(void)
696 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
697 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
698 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
703 dax_mnt = kern_mount(&dax_fs_type);
704 if (IS_ERR(dax_mnt)) {
705 rc = PTR_ERR(dax_mnt);
708 dax_superblock = dax_mnt->mnt_sb;
713 kmem_cache_destroy(dax_cache);
718 static void dax_fs_exit(void)
720 kern_unmount(dax_mnt);
721 kmem_cache_destroy(dax_cache);
724 static int __init dax_core_init(void)
732 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
742 unregister_chrdev_region(dax_devt, MINORMASK+1);
748 static void __exit dax_core_exit(void)
750 unregister_chrdev_region(dax_devt, MINORMASK+1);
751 ida_destroy(&dax_minor_ida);
755 MODULE_AUTHOR("Intel Corporation");
756 MODULE_LICENSE("GPL v2");
757 subsys_initcall(dax_core_init);
758 module_exit(dax_core_exit);