Linux Audio

Check our new training course

Loading...
v4.17
 
  1/*
  2 * Copyright(c) 2017 Intel Corporation. All rights reserved.
  3 *
  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.
  7 *
  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.
 12 */
 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>
 24#include <linux/fs.h>
 
 25
 26static dev_t dax_devt;
 27DEFINE_STATIC_SRCU(dax_srcu);
 28static struct vfsmount *dax_mnt;
 29static DEFINE_IDA(dax_minor_ida);
 30static struct kmem_cache *dax_cache __read_mostly;
 31static struct super_block *dax_superblock __read_mostly;
 32
 33#define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
 34static struct hlist_head dax_host_list[DAX_HASH_SIZE];
 35static DEFINE_SPINLOCK(dax_host_lock);
 36
 37int dax_read_lock(void)
 38{
 39	return srcu_read_lock(&dax_srcu);
 40}
 41EXPORT_SYMBOL_GPL(dax_read_lock);
 42
 43void dax_read_unlock(int id)
 44{
 45	srcu_read_unlock(&dax_srcu, id);
 46}
 47EXPORT_SYMBOL_GPL(dax_read_unlock);
 48
 49#ifdef CONFIG_BLOCK
 50#include <linux/blkdev.h>
 51
 52int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
 53		pgoff_t *pgoff)
 54{
 55	phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512;
 
 56
 57	if (pgoff)
 58		*pgoff = PHYS_PFN(phys_off);
 59	if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
 60		return -EINVAL;
 61	return 0;
 62}
 63EXPORT_SYMBOL(bdev_dax_pgoff);
 64
 65#if IS_ENABLED(CONFIG_FS_DAX)
 66struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
 67{
 68	if (!blk_queue_dax(bdev->bd_queue))
 69		return NULL;
 70	return fs_dax_get_by_host(bdev->bd_disk->disk_name);
 71}
 72EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
 73#endif
 74
 75/**
 76 * __bdev_dax_supported() - Check if the device supports dax for filesystem
 77 * @sb: The superblock of the device
 78 * @blocksize: The block size of the device
 79 *
 80 * This is a library function for filesystems to check if the block device
 81 * can be mounted with dax option.
 82 *
 83 * Return: negative errno if unsupported, 0 if supported.
 84 */
 85int __bdev_dax_supported(struct super_block *sb, int blocksize)
 86{
 87	struct block_device *bdev = sb->s_bdev;
 88	struct dax_device *dax_dev;
 89	pgoff_t pgoff;
 90	int err, id;
 91	void *kaddr;
 92	pfn_t pfn;
 93	long len;
 94
 95	if (blocksize != PAGE_SIZE) {
 96		pr_debug("VFS (%s): error: unsupported blocksize for dax\n",
 97				sb->s_id);
 98		return -EINVAL;
 99	}
100
101	err = bdev_dax_pgoff(bdev, 0, PAGE_SIZE, &pgoff);
 
 
 
 
 
 
102	if (err) {
103		pr_debug("VFS (%s): error: unaligned partition for dax\n",
104				sb->s_id);
105		return err;
106	}
107
108	dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
109	if (!dax_dev) {
110		pr_debug("VFS (%s): error: device does not support dax\n",
111				sb->s_id);
112		return -EOPNOTSUPP;
 
113	}
114
115	id = dax_read_lock();
116	len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
117	dax_read_unlock(id);
118
119	put_dax(dax_dev);
120
121	if (len < 1) {
122		pr_debug("VFS (%s): error: dax access failed (%ld)\n",
123				sb->s_id, len);
124		return len < 0 ? len : -EIO;
125	}
126
127	if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
128		/*
129		 * An arch that has enabled the pmem api should also
130		 * have its drivers support pfn_t_devmap()
131		 *
132		 * This is a developer warning and should not trigger in
133		 * production. dax_flush() will crash since it depends
134		 * on being able to do (page_address(pfn_to_page())).
135		 */
136		WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
137	} else if (pfn_t_devmap(pfn)) {
138		/* pass */;
139	} else {
140		pr_debug("VFS (%s): error: dax support not enabled\n",
141				sb->s_id);
142		return -EOPNOTSUPP;
 
 
 
 
 
 
 
 
 
143	}
 
144
145	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
146}
147EXPORT_SYMBOL_GPL(__bdev_dax_supported);
148#endif
149
150enum dax_device_flags {
151	/* !alive + rcu grace period == no new operations / mappings */
152	DAXDEV_ALIVE,
153	/* gate whether dax_flush() calls the low level flush routine */
154	DAXDEV_WRITE_CACHE,
 
 
155};
156
157/**
158 * struct dax_device - anchor object for dax services
159 * @inode: core vfs
160 * @cdev: optional character interface for "device dax"
161 * @host: optional name for lookups where the device path is not available
162 * @private: dax driver private data
163 * @flags: state and boolean properties
164 */
165struct dax_device {
166	struct hlist_node list;
167	struct inode inode;
168	struct cdev cdev;
169	const char *host;
170	void *private;
171	unsigned long flags;
172	const struct dax_operations *ops;
173};
174
175static ssize_t write_cache_show(struct device *dev,
176		struct device_attribute *attr, char *buf)
177{
178	struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
179	ssize_t rc;
180
181	WARN_ON_ONCE(!dax_dev);
182	if (!dax_dev)
183		return -ENXIO;
184
185	rc = sprintf(buf, "%d\n", !!test_bit(DAXDEV_WRITE_CACHE,
186				&dax_dev->flags));
187	put_dax(dax_dev);
188	return rc;
189}
190
191static ssize_t write_cache_store(struct device *dev,
192		struct device_attribute *attr, const char *buf, size_t len)
193{
194	bool write_cache;
195	int rc = strtobool(buf, &write_cache);
196	struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
197
198	WARN_ON_ONCE(!dax_dev);
199	if (!dax_dev)
200		return -ENXIO;
201
202	if (rc)
203		len = rc;
204	else if (write_cache)
205		set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
206	else
207		clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
208
209	put_dax(dax_dev);
210	return len;
211}
212static DEVICE_ATTR_RW(write_cache);
213
214static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
215{
216	struct device *dev = container_of(kobj, typeof(*dev), kobj);
217	struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
218
219	WARN_ON_ONCE(!dax_dev);
220	if (!dax_dev)
221		return 0;
222
223#ifndef CONFIG_ARCH_HAS_PMEM_API
224	if (a == &dev_attr_write_cache.attr)
225		return 0;
226#endif
227	return a->mode;
228}
229
230static struct attribute *dax_attributes[] = {
231	&dev_attr_write_cache.attr,
232	NULL,
233};
234
235struct attribute_group dax_attribute_group = {
236	.name = "dax",
237	.attrs = dax_attributes,
238	.is_visible = dax_visible,
239};
240EXPORT_SYMBOL_GPL(dax_attribute_group);
241
242/**
243 * dax_direct_access() - translate a device pgoff to an absolute pfn
244 * @dax_dev: a dax_device instance representing the logical memory range
245 * @pgoff: offset in pages from the start of the device to translate
246 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
247 * @kaddr: output parameter that returns a virtual address mapping of pfn
248 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
249 *
250 * Return: negative errno if an error occurs, otherwise the number of
251 * pages accessible at the device relative @pgoff.
252 */
253long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
254		void **kaddr, pfn_t *pfn)
255{
256	long avail;
257
258	if (!dax_dev)
259		return -EOPNOTSUPP;
260
261	if (!dax_alive(dax_dev))
262		return -ENXIO;
263
264	if (nr_pages < 0)
265		return nr_pages;
266
267	avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
268			kaddr, pfn);
269	if (!avail)
270		return -ERANGE;
271	return min(avail, nr_pages);
272}
273EXPORT_SYMBOL_GPL(dax_direct_access);
274
 
 
 
 
 
 
 
 
 
 
 
 
 
275size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
276		size_t bytes, struct iov_iter *i)
277{
278	if (!dax_alive(dax_dev))
279		return 0;
280
281	return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
282}
283EXPORT_SYMBOL_GPL(dax_copy_from_iter);
284
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
285#ifdef CONFIG_ARCH_HAS_PMEM_API
286void arch_wb_cache_pmem(void *addr, size_t size);
287void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
288{
289	if (unlikely(!test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags)))
290		return;
291
292	arch_wb_cache_pmem(addr, size);
293}
294#else
295void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
296{
297}
298#endif
299EXPORT_SYMBOL_GPL(dax_flush);
300
301void dax_write_cache(struct dax_device *dax_dev, bool wc)
302{
303	if (wc)
304		set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
305	else
306		clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
307}
308EXPORT_SYMBOL_GPL(dax_write_cache);
309
310bool dax_write_cache_enabled(struct dax_device *dax_dev)
311{
312	return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
313}
314EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
315
 
 
 
 
 
 
 
 
 
 
 
 
316bool dax_alive(struct dax_device *dax_dev)
317{
318	lockdep_assert_held(&dax_srcu);
319	return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
320}
321EXPORT_SYMBOL_GPL(dax_alive);
322
323static int dax_host_hash(const char *host)
324{
325	return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
326}
327
328/*
329 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
330 * that any fault handlers or operations that might have seen
331 * dax_alive(), have completed.  Any operations that start after
332 * synchronize_srcu() has run will abort upon seeing !dax_alive().
333 */
334void kill_dax(struct dax_device *dax_dev)
335{
336	if (!dax_dev)
337		return;
338
339	clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
340
341	synchronize_srcu(&dax_srcu);
342
343	spin_lock(&dax_host_lock);
344	hlist_del_init(&dax_dev->list);
345	spin_unlock(&dax_host_lock);
346
347	dax_dev->private = NULL;
348}
349EXPORT_SYMBOL_GPL(kill_dax);
350
 
 
 
 
 
 
351static struct inode *dax_alloc_inode(struct super_block *sb)
352{
353	struct dax_device *dax_dev;
354	struct inode *inode;
355
356	dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
357	if (!dax_dev)
358		return NULL;
359
360	inode = &dax_dev->inode;
361	inode->i_rdev = 0;
362	return inode;
363}
364
365static struct dax_device *to_dax_dev(struct inode *inode)
366{
367	return container_of(inode, struct dax_device, inode);
368}
369
370static void dax_i_callback(struct rcu_head *head)
371{
372	struct inode *inode = container_of(head, struct inode, i_rcu);
373	struct dax_device *dax_dev = to_dax_dev(inode);
374
375	kfree(dax_dev->host);
376	dax_dev->host = NULL;
377	if (inode->i_rdev)
378		ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
379	kmem_cache_free(dax_cache, dax_dev);
380}
381
382static void dax_destroy_inode(struct inode *inode)
383{
384	struct dax_device *dax_dev = to_dax_dev(inode);
385
386	WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
387			"kill_dax() must be called before final iput()\n");
388	call_rcu(&inode->i_rcu, dax_i_callback);
389}
390
391static const struct super_operations dax_sops = {
392	.statfs = simple_statfs,
393	.alloc_inode = dax_alloc_inode,
394	.destroy_inode = dax_destroy_inode,
 
395	.drop_inode = generic_delete_inode,
396};
397
398static struct dentry *dax_mount(struct file_system_type *fs_type,
399		int flags, const char *dev_name, void *data)
400{
401	return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
 
 
 
 
402}
403
404static struct file_system_type dax_fs_type = {
405	.name = "dax",
406	.mount = dax_mount,
407	.kill_sb = kill_anon_super,
408};
409
410static int dax_test(struct inode *inode, void *data)
411{
412	dev_t devt = *(dev_t *) data;
413
414	return inode->i_rdev == devt;
415}
416
417static int dax_set(struct inode *inode, void *data)
418{
419	dev_t devt = *(dev_t *) data;
420
421	inode->i_rdev = devt;
422	return 0;
423}
424
425static struct dax_device *dax_dev_get(dev_t devt)
426{
427	struct dax_device *dax_dev;
428	struct inode *inode;
429
430	inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
431			dax_test, dax_set, &devt);
432
433	if (!inode)
434		return NULL;
435
436	dax_dev = to_dax_dev(inode);
437	if (inode->i_state & I_NEW) {
438		set_bit(DAXDEV_ALIVE, &dax_dev->flags);
439		inode->i_cdev = &dax_dev->cdev;
440		inode->i_mode = S_IFCHR;
441		inode->i_flags = S_DAX;
442		mapping_set_gfp_mask(&inode->i_data, GFP_USER);
443		unlock_new_inode(inode);
444	}
445
446	return dax_dev;
447}
448
449static void dax_add_host(struct dax_device *dax_dev, const char *host)
450{
451	int hash;
452
453	/*
454	 * Unconditionally init dax_dev since it's coming from a
455	 * non-zeroed slab cache
456	 */
457	INIT_HLIST_NODE(&dax_dev->list);
458	dax_dev->host = host;
459	if (!host)
460		return;
461
462	hash = dax_host_hash(host);
463	spin_lock(&dax_host_lock);
464	hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
465	spin_unlock(&dax_host_lock);
466}
467
468struct dax_device *alloc_dax(void *private, const char *__host,
469		const struct dax_operations *ops)
470{
471	struct dax_device *dax_dev;
472	const char *host;
473	dev_t devt;
474	int minor;
475
 
 
 
 
 
 
 
476	host = kstrdup(__host, GFP_KERNEL);
477	if (__host && !host)
478		return NULL;
479
480	minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
481	if (minor < 0)
482		goto err_minor;
483
484	devt = MKDEV(MAJOR(dax_devt), minor);
485	dax_dev = dax_dev_get(devt);
486	if (!dax_dev)
487		goto err_dev;
488
489	dax_add_host(dax_dev, host);
490	dax_dev->ops = ops;
491	dax_dev->private = private;
 
 
 
492	return dax_dev;
493
494 err_dev:
495	ida_simple_remove(&dax_minor_ida, minor);
496 err_minor:
497	kfree(host);
498	return NULL;
499}
500EXPORT_SYMBOL_GPL(alloc_dax);
501
502void put_dax(struct dax_device *dax_dev)
503{
504	if (!dax_dev)
505		return;
506	iput(&dax_dev->inode);
507}
508EXPORT_SYMBOL_GPL(put_dax);
509
510/**
511 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
512 * @host: alternate name for the device registered by a dax driver
513 */
514struct dax_device *dax_get_by_host(const char *host)
515{
516	struct dax_device *dax_dev, *found = NULL;
517	int hash, id;
518
519	if (!host)
520		return NULL;
521
522	hash = dax_host_hash(host);
523
524	id = dax_read_lock();
525	spin_lock(&dax_host_lock);
526	hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
527		if (!dax_alive(dax_dev)
528				|| strcmp(host, dax_dev->host) != 0)
529			continue;
530
531		if (igrab(&dax_dev->inode))
532			found = dax_dev;
533		break;
534	}
535	spin_unlock(&dax_host_lock);
536	dax_read_unlock(id);
537
538	return found;
539}
540EXPORT_SYMBOL_GPL(dax_get_by_host);
541
542/**
543 * inode_dax: convert a public inode into its dax_dev
544 * @inode: An inode with i_cdev pointing to a dax_dev
545 *
546 * Note this is not equivalent to to_dax_dev() which is for private
547 * internal use where we know the inode filesystem type == dax_fs_type.
548 */
549struct dax_device *inode_dax(struct inode *inode)
550{
551	struct cdev *cdev = inode->i_cdev;
552
553	return container_of(cdev, struct dax_device, cdev);
554}
555EXPORT_SYMBOL_GPL(inode_dax);
556
557struct inode *dax_inode(struct dax_device *dax_dev)
558{
559	return &dax_dev->inode;
560}
561EXPORT_SYMBOL_GPL(dax_inode);
562
563void *dax_get_private(struct dax_device *dax_dev)
564{
 
 
565	return dax_dev->private;
566}
567EXPORT_SYMBOL_GPL(dax_get_private);
568
569static void init_once(void *_dax_dev)
570{
571	struct dax_device *dax_dev = _dax_dev;
572	struct inode *inode = &dax_dev->inode;
573
574	memset(dax_dev, 0, sizeof(*dax_dev));
575	inode_init_once(inode);
576}
577
578static int __dax_fs_init(void)
579{
580	int rc;
581
582	dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
583			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
584			 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
585			init_once);
586	if (!dax_cache)
587		return -ENOMEM;
588
589	rc = register_filesystem(&dax_fs_type);
590	if (rc)
591		goto err_register_fs;
592
593	dax_mnt = kern_mount(&dax_fs_type);
594	if (IS_ERR(dax_mnt)) {
595		rc = PTR_ERR(dax_mnt);
596		goto err_mount;
597	}
598	dax_superblock = dax_mnt->mnt_sb;
599
600	return 0;
601
602 err_mount:
603	unregister_filesystem(&dax_fs_type);
604 err_register_fs:
605	kmem_cache_destroy(dax_cache);
606
607	return rc;
608}
609
610static void __dax_fs_exit(void)
611{
612	kern_unmount(dax_mnt);
613	unregister_filesystem(&dax_fs_type);
614	kmem_cache_destroy(dax_cache);
615}
616
617static int __init dax_fs_init(void)
618{
619	int rc;
620
621	rc = __dax_fs_init();
622	if (rc)
623		return rc;
624
625	rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
626	if (rc)
627		__dax_fs_exit();
628	return rc;
 
 
 
 
 
 
 
 
 
 
629}
630
631static void __exit dax_fs_exit(void)
632{
 
633	unregister_chrdev_region(dax_devt, MINORMASK+1);
634	ida_destroy(&dax_minor_ida);
635	__dax_fs_exit();
636}
637
638MODULE_AUTHOR("Intel Corporation");
639MODULE_LICENSE("GPL v2");
640subsys_initcall(dax_fs_init);
641module_exit(dax_fs_exit);
v5.14.15
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright(c) 2017 Intel Corporation. All rights reserved.
 
 
 
 
 
 
 
 
 
  4 */
  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>
 17#include <linux/fs.h>
 18#include "dax-private.h"
 19
 20static dev_t dax_devt;
 21DEFINE_STATIC_SRCU(dax_srcu);
 22static struct vfsmount *dax_mnt;
 23static DEFINE_IDA(dax_minor_ida);
 24static struct kmem_cache *dax_cache __read_mostly;
 25static struct super_block *dax_superblock __read_mostly;
 26
 27#define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
 28static struct hlist_head dax_host_list[DAX_HASH_SIZE];
 29static DEFINE_SPINLOCK(dax_host_lock);
 30
 31int dax_read_lock(void)
 32{
 33	return srcu_read_lock(&dax_srcu);
 34}
 35EXPORT_SYMBOL_GPL(dax_read_lock);
 36
 37void dax_read_unlock(int id)
 38{
 39	srcu_read_unlock(&dax_srcu, id);
 40}
 41EXPORT_SYMBOL_GPL(dax_read_unlock);
 42
 43#ifdef CONFIG_BLOCK
 44#include <linux/blkdev.h>
 45
 46int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
 47		pgoff_t *pgoff)
 48{
 49	sector_t start_sect = bdev ? get_start_sect(bdev) : 0;
 50	phys_addr_t phys_off = (start_sect + sector) * 512;
 51
 52	if (pgoff)
 53		*pgoff = PHYS_PFN(phys_off);
 54	if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
 55		return -EINVAL;
 56	return 0;
 57}
 58EXPORT_SYMBOL(bdev_dax_pgoff);
 59
 60#if IS_ENABLED(CONFIG_FS_DAX)
 61struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
 62{
 63	if (!blk_queue_dax(bdev->bd_disk->queue))
 64		return NULL;
 65	return dax_get_by_host(bdev->bd_disk->disk_name);
 66}
 67EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
 68#endif
 69
 70bool __generic_fsdax_supported(struct dax_device *dax_dev,
 71		struct block_device *bdev, int blocksize, sector_t start,
 72		sector_t sectors)
 73{
 74	bool dax_enabled = false;
 75	pgoff_t pgoff, pgoff_end;
 76	char buf[BDEVNAME_SIZE];
 77	void *kaddr, *end_kaddr;
 78	pfn_t pfn, end_pfn;
 79	sector_t last_page;
 80	long len, len2;
 
 
 
 
 81	int err, id;
 
 
 
 82
 83	if (blocksize != PAGE_SIZE) {
 84		pr_info("%s: error: unsupported blocksize for dax\n",
 85				bdevname(bdev, buf));
 86		return false;
 87	}
 88
 89	if (!dax_dev) {
 90		pr_debug("%s: error: dax unsupported by block device\n",
 91				bdevname(bdev, buf));
 92		return false;
 93	}
 94
 95	err = bdev_dax_pgoff(bdev, start, PAGE_SIZE, &pgoff);
 96	if (err) {
 97		pr_info("%s: error: unaligned partition for dax\n",
 98				bdevname(bdev, buf));
 99		return false;
100	}
101
102	last_page = PFN_DOWN((start + sectors - 1) * 512) * PAGE_SIZE / 512;
103	err = bdev_dax_pgoff(bdev, last_page, PAGE_SIZE, &pgoff_end);
104	if (err) {
105		pr_info("%s: error: unaligned partition for dax\n",
106				bdevname(bdev, buf));
107		return false;
108	}
109
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);
113
114	if (len < 1 || len2 < 1) {
115		pr_info("%s: error: dax access failed (%ld)\n",
116				bdevname(bdev, buf), len < 1 ? len : len2);
117		dax_read_unlock(id);
118		return false;
 
119	}
120
121	if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) {
122		/*
123		 * An arch that has enabled the pmem api should also
124		 * have its drivers support pfn_t_devmap()
125		 *
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())).
129		 */
130		WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API));
131		dax_enabled = true;
132	} else if (pfn_t_devmap(pfn) && pfn_t_devmap(end_pfn)) {
133		struct dev_pagemap *pgmap, *end_pgmap;
134
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)))
142			dax_enabled = true;
143		put_dev_pagemap(pgmap);
144		put_dev_pagemap(end_pgmap);
145
146	}
147	dax_read_unlock(id);
148
149	if (!dax_enabled) {
150		pr_info("%s: error: dax support not enabled\n",
151				bdevname(bdev, buf));
152		return false;
153	}
154	return true;
155}
156EXPORT_SYMBOL_GPL(__generic_fsdax_supported);
157
158/**
159 * __bdev_dax_supported() - Check if the device supports dax for filesystem
160 * @bdev: block device to check
161 * @blocksize: The block size of the device
162 *
163 * This is a library function for filesystems to check if the block device
164 * can be mounted with dax option.
165 *
166 * Return: true if supported, false if unsupported
167 */
168bool __bdev_dax_supported(struct block_device *bdev, int blocksize)
169{
170	struct dax_device *dax_dev;
171	struct request_queue *q;
172	char buf[BDEVNAME_SIZE];
173	bool ret;
174	int id;
175
176	q = bdev_get_queue(bdev);
177	if (!q || !blk_queue_dax(q)) {
178		pr_debug("%s: error: request queue doesn't support dax\n",
179				bdevname(bdev, buf));
180		return false;
181	}
182
183	dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
184	if (!dax_dev) {
185		pr_debug("%s: error: device does not support dax\n",
186				bdevname(bdev, buf));
187		return false;
188	}
189
190	id = dax_read_lock();
191	ret = dax_supported(dax_dev, bdev, blocksize, 0,
192			i_size_read(bdev->bd_inode) / 512);
193	dax_read_unlock(id);
194
195	put_dax(dax_dev);
196
197	return ret;
198}
199EXPORT_SYMBOL_GPL(__bdev_dax_supported);
200#endif
201
202enum dax_device_flags {
203	/* !alive + rcu grace period == no new operations / mappings */
204	DAXDEV_ALIVE,
205	/* gate whether dax_flush() calls the low level flush routine */
206	DAXDEV_WRITE_CACHE,
207	/* flag to check if device supports synchronous flush */
208	DAXDEV_SYNC,
209};
210
211/**
212 * struct dax_device - anchor object for dax services
213 * @inode: core vfs
214 * @cdev: optional character interface for "device dax"
215 * @host: optional name for lookups where the device path is not available
216 * @private: dax driver private data
217 * @flags: state and boolean properties
218 */
219struct dax_device {
220	struct hlist_node list;
221	struct inode inode;
222	struct cdev cdev;
223	const char *host;
224	void *private;
225	unsigned long flags;
226	const struct dax_operations *ops;
227};
228
229static ssize_t write_cache_show(struct device *dev,
230		struct device_attribute *attr, char *buf)
231{
232	struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
233	ssize_t rc;
234
235	WARN_ON_ONCE(!dax_dev);
236	if (!dax_dev)
237		return -ENXIO;
238
239	rc = sprintf(buf, "%d\n", !!dax_write_cache_enabled(dax_dev));
 
240	put_dax(dax_dev);
241	return rc;
242}
243
244static ssize_t write_cache_store(struct device *dev,
245		struct device_attribute *attr, const char *buf, size_t len)
246{
247	bool write_cache;
248	int rc = strtobool(buf, &write_cache);
249	struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
250
251	WARN_ON_ONCE(!dax_dev);
252	if (!dax_dev)
253		return -ENXIO;
254
255	if (rc)
256		len = rc;
 
 
257	else
258		dax_write_cache(dax_dev, write_cache);
259
260	put_dax(dax_dev);
261	return len;
262}
263static DEVICE_ATTR_RW(write_cache);
264
265static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
266{
267	struct device *dev = container_of(kobj, typeof(*dev), kobj);
268	struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
269
270	WARN_ON_ONCE(!dax_dev);
271	if (!dax_dev)
272		return 0;
273
274#ifndef CONFIG_ARCH_HAS_PMEM_API
275	if (a == &dev_attr_write_cache.attr)
276		return 0;
277#endif
278	return a->mode;
279}
280
281static struct attribute *dax_attributes[] = {
282	&dev_attr_write_cache.attr,
283	NULL,
284};
285
286struct attribute_group dax_attribute_group = {
287	.name = "dax",
288	.attrs = dax_attributes,
289	.is_visible = dax_visible,
290};
291EXPORT_SYMBOL_GPL(dax_attribute_group);
292
293/**
294 * dax_direct_access() - translate a device pgoff to an absolute pfn
295 * @dax_dev: a dax_device instance representing the logical memory range
296 * @pgoff: offset in pages from the start of the device to translate
297 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
298 * @kaddr: output parameter that returns a virtual address mapping of pfn
299 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
300 *
301 * Return: negative errno if an error occurs, otherwise the number of
302 * pages accessible at the device relative @pgoff.
303 */
304long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
305		void **kaddr, pfn_t *pfn)
306{
307	long avail;
308
309	if (!dax_dev)
310		return -EOPNOTSUPP;
311
312	if (!dax_alive(dax_dev))
313		return -ENXIO;
314
315	if (nr_pages < 0)
316		return -EINVAL;
317
318	avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
319			kaddr, pfn);
320	if (!avail)
321		return -ERANGE;
322	return min(avail, nr_pages);
323}
324EXPORT_SYMBOL_GPL(dax_direct_access);
325
326bool dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
327		int blocksize, sector_t start, sector_t len)
328{
329	if (!dax_dev)
330		return false;
331
332	if (!dax_alive(dax_dev))
333		return false;
334
335	return dax_dev->ops->dax_supported(dax_dev, bdev, blocksize, start, len);
336}
337EXPORT_SYMBOL_GPL(dax_supported);
338
339size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
340		size_t bytes, struct iov_iter *i)
341{
342	if (!dax_alive(dax_dev))
343		return 0;
344
345	return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
346}
347EXPORT_SYMBOL_GPL(dax_copy_from_iter);
348
349size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
350		size_t bytes, struct iov_iter *i)
351{
352	if (!dax_alive(dax_dev))
353		return 0;
354
355	return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i);
356}
357EXPORT_SYMBOL_GPL(dax_copy_to_iter);
358
359int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
360			size_t nr_pages)
361{
362	if (!dax_alive(dax_dev))
363		return -ENXIO;
364	/*
365	 * There are no callers that want to zero more than one page as of now.
366	 * Once users are there, this check can be removed after the
367	 * device mapper code has been updated to split ranges across targets.
368	 */
369	if (nr_pages != 1)
370		return -EIO;
371
372	return dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages);
373}
374EXPORT_SYMBOL_GPL(dax_zero_page_range);
375
376#ifdef CONFIG_ARCH_HAS_PMEM_API
377void arch_wb_cache_pmem(void *addr, size_t size);
378void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
379{
380	if (unlikely(!dax_write_cache_enabled(dax_dev)))
381		return;
382
383	arch_wb_cache_pmem(addr, size);
384}
385#else
386void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
387{
388}
389#endif
390EXPORT_SYMBOL_GPL(dax_flush);
391
392void dax_write_cache(struct dax_device *dax_dev, bool wc)
393{
394	if (wc)
395		set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
396	else
397		clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
398}
399EXPORT_SYMBOL_GPL(dax_write_cache);
400
401bool dax_write_cache_enabled(struct dax_device *dax_dev)
402{
403	return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
404}
405EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
406
407bool __dax_synchronous(struct dax_device *dax_dev)
408{
409	return test_bit(DAXDEV_SYNC, &dax_dev->flags);
410}
411EXPORT_SYMBOL_GPL(__dax_synchronous);
412
413void __set_dax_synchronous(struct dax_device *dax_dev)
414{
415	set_bit(DAXDEV_SYNC, &dax_dev->flags);
416}
417EXPORT_SYMBOL_GPL(__set_dax_synchronous);
418
419bool dax_alive(struct dax_device *dax_dev)
420{
421	lockdep_assert_held(&dax_srcu);
422	return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
423}
424EXPORT_SYMBOL_GPL(dax_alive);
425
426static int dax_host_hash(const char *host)
427{
428	return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
429}
430
431/*
432 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
433 * that any fault handlers or operations that might have seen
434 * dax_alive(), have completed.  Any operations that start after
435 * synchronize_srcu() has run will abort upon seeing !dax_alive().
436 */
437void kill_dax(struct dax_device *dax_dev)
438{
439	if (!dax_dev)
440		return;
441
442	clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
443
444	synchronize_srcu(&dax_srcu);
445
446	spin_lock(&dax_host_lock);
447	hlist_del_init(&dax_dev->list);
448	spin_unlock(&dax_host_lock);
 
 
449}
450EXPORT_SYMBOL_GPL(kill_dax);
451
452void run_dax(struct dax_device *dax_dev)
453{
454	set_bit(DAXDEV_ALIVE, &dax_dev->flags);
455}
456EXPORT_SYMBOL_GPL(run_dax);
457
458static struct inode *dax_alloc_inode(struct super_block *sb)
459{
460	struct dax_device *dax_dev;
461	struct inode *inode;
462
463	dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
464	if (!dax_dev)
465		return NULL;
466
467	inode = &dax_dev->inode;
468	inode->i_rdev = 0;
469	return inode;
470}
471
472static struct dax_device *to_dax_dev(struct inode *inode)
473{
474	return container_of(inode, struct dax_device, inode);
475}
476
477static void dax_free_inode(struct inode *inode)
478{
 
479	struct dax_device *dax_dev = to_dax_dev(inode);
 
480	kfree(dax_dev->host);
481	dax_dev->host = NULL;
482	if (inode->i_rdev)
483		ida_simple_remove(&dax_minor_ida, iminor(inode));
484	kmem_cache_free(dax_cache, dax_dev);
485}
486
487static void dax_destroy_inode(struct inode *inode)
488{
489	struct dax_device *dax_dev = to_dax_dev(inode);
 
490	WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
491			"kill_dax() must be called before final iput()\n");
 
492}
493
494static const struct super_operations dax_sops = {
495	.statfs = simple_statfs,
496	.alloc_inode = dax_alloc_inode,
497	.destroy_inode = dax_destroy_inode,
498	.free_inode = dax_free_inode,
499	.drop_inode = generic_delete_inode,
500};
501
502static int dax_init_fs_context(struct fs_context *fc)
 
503{
504	struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC);
505	if (!ctx)
506		return -ENOMEM;
507	ctx->ops = &dax_sops;
508	return 0;
509}
510
511static struct file_system_type dax_fs_type = {
512	.name		= "dax",
513	.init_fs_context = dax_init_fs_context,
514	.kill_sb	= kill_anon_super,
515};
516
517static int dax_test(struct inode *inode, void *data)
518{
519	dev_t devt = *(dev_t *) data;
520
521	return inode->i_rdev == devt;
522}
523
524static int dax_set(struct inode *inode, void *data)
525{
526	dev_t devt = *(dev_t *) data;
527
528	inode->i_rdev = devt;
529	return 0;
530}
531
532static struct dax_device *dax_dev_get(dev_t devt)
533{
534	struct dax_device *dax_dev;
535	struct inode *inode;
536
537	inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
538			dax_test, dax_set, &devt);
539
540	if (!inode)
541		return NULL;
542
543	dax_dev = to_dax_dev(inode);
544	if (inode->i_state & I_NEW) {
545		set_bit(DAXDEV_ALIVE, &dax_dev->flags);
546		inode->i_cdev = &dax_dev->cdev;
547		inode->i_mode = S_IFCHR;
548		inode->i_flags = S_DAX;
549		mapping_set_gfp_mask(&inode->i_data, GFP_USER);
550		unlock_new_inode(inode);
551	}
552
553	return dax_dev;
554}
555
556static void dax_add_host(struct dax_device *dax_dev, const char *host)
557{
558	int hash;
559
560	/*
561	 * Unconditionally init dax_dev since it's coming from a
562	 * non-zeroed slab cache
563	 */
564	INIT_HLIST_NODE(&dax_dev->list);
565	dax_dev->host = host;
566	if (!host)
567		return;
568
569	hash = dax_host_hash(host);
570	spin_lock(&dax_host_lock);
571	hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
572	spin_unlock(&dax_host_lock);
573}
574
575struct dax_device *alloc_dax(void *private, const char *__host,
576		const struct dax_operations *ops, unsigned long flags)
577{
578	struct dax_device *dax_dev;
579	const char *host;
580	dev_t devt;
581	int minor;
582
583	if (ops && !ops->zero_page_range) {
584		pr_debug("%s: error: device does not provide dax"
585			 " operation zero_page_range()\n",
586			 __host ? __host : "Unknown");
587		return ERR_PTR(-EINVAL);
588	}
589
590	host = kstrdup(__host, GFP_KERNEL);
591	if (__host && !host)
592		return ERR_PTR(-ENOMEM);
593
594	minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
595	if (minor < 0)
596		goto err_minor;
597
598	devt = MKDEV(MAJOR(dax_devt), minor);
599	dax_dev = dax_dev_get(devt);
600	if (!dax_dev)
601		goto err_dev;
602
603	dax_add_host(dax_dev, host);
604	dax_dev->ops = ops;
605	dax_dev->private = private;
606	if (flags & DAXDEV_F_SYNC)
607		set_dax_synchronous(dax_dev);
608
609	return dax_dev;
610
611 err_dev:
612	ida_simple_remove(&dax_minor_ida, minor);
613 err_minor:
614	kfree(host);
615	return ERR_PTR(-ENOMEM);
616}
617EXPORT_SYMBOL_GPL(alloc_dax);
618
619void put_dax(struct dax_device *dax_dev)
620{
621	if (!dax_dev)
622		return;
623	iput(&dax_dev->inode);
624}
625EXPORT_SYMBOL_GPL(put_dax);
626
627/**
628 * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
629 * @host: alternate name for the device registered by a dax driver
630 */
631struct dax_device *dax_get_by_host(const char *host)
632{
633	struct dax_device *dax_dev, *found = NULL;
634	int hash, id;
635
636	if (!host)
637		return NULL;
638
639	hash = dax_host_hash(host);
640
641	id = dax_read_lock();
642	spin_lock(&dax_host_lock);
643	hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
644		if (!dax_alive(dax_dev)
645				|| strcmp(host, dax_dev->host) != 0)
646			continue;
647
648		if (igrab(&dax_dev->inode))
649			found = dax_dev;
650		break;
651	}
652	spin_unlock(&dax_host_lock);
653	dax_read_unlock(id);
654
655	return found;
656}
657EXPORT_SYMBOL_GPL(dax_get_by_host);
658
659/**
660 * inode_dax: convert a public inode into its dax_dev
661 * @inode: An inode with i_cdev pointing to a dax_dev
662 *
663 * Note this is not equivalent to to_dax_dev() which is for private
664 * internal use where we know the inode filesystem type == dax_fs_type.
665 */
666struct dax_device *inode_dax(struct inode *inode)
667{
668	struct cdev *cdev = inode->i_cdev;
669
670	return container_of(cdev, struct dax_device, cdev);
671}
672EXPORT_SYMBOL_GPL(inode_dax);
673
674struct inode *dax_inode(struct dax_device *dax_dev)
675{
676	return &dax_dev->inode;
677}
678EXPORT_SYMBOL_GPL(dax_inode);
679
680void *dax_get_private(struct dax_device *dax_dev)
681{
682	if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
683		return NULL;
684	return dax_dev->private;
685}
686EXPORT_SYMBOL_GPL(dax_get_private);
687
688static void init_once(void *_dax_dev)
689{
690	struct dax_device *dax_dev = _dax_dev;
691	struct inode *inode = &dax_dev->inode;
692
693	memset(dax_dev, 0, sizeof(*dax_dev));
694	inode_init_once(inode);
695}
696
697static int dax_fs_init(void)
698{
699	int rc;
700
701	dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
702			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
703			 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
704			init_once);
705	if (!dax_cache)
706		return -ENOMEM;
707
 
 
 
 
708	dax_mnt = kern_mount(&dax_fs_type);
709	if (IS_ERR(dax_mnt)) {
710		rc = PTR_ERR(dax_mnt);
711		goto err_mount;
712	}
713	dax_superblock = dax_mnt->mnt_sb;
714
715	return 0;
716
717 err_mount:
 
 
718	kmem_cache_destroy(dax_cache);
719
720	return rc;
721}
722
723static void dax_fs_exit(void)
724{
725	kern_unmount(dax_mnt);
 
726	kmem_cache_destroy(dax_cache);
727}
728
729static int __init dax_core_init(void)
730{
731	int rc;
732
733	rc = dax_fs_init();
734	if (rc)
735		return rc;
736
737	rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
738	if (rc)
739		goto err_chrdev;
740
741	rc = dax_bus_init();
742	if (rc)
743		goto err_bus;
744	return 0;
745
746err_bus:
747	unregister_chrdev_region(dax_devt, MINORMASK+1);
748err_chrdev:
749	dax_fs_exit();
750	return 0;
751}
752
753static void __exit dax_core_exit(void)
754{
755	dax_bus_exit();
756	unregister_chrdev_region(dax_devt, MINORMASK+1);
757	ida_destroy(&dax_minor_ida);
758	dax_fs_exit();
759}
760
761MODULE_AUTHOR("Intel Corporation");
762MODULE_LICENSE("GPL v2");
763subsys_initcall(dax_core_init);
764module_exit(dax_core_exit);