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);
v6.8
  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/pfn_t.h>
 11#include <linux/cdev.h>
 
 12#include <linux/slab.h>
 13#include <linux/uio.h>
 14#include <linux/dax.h>
 15#include <linux/fs.h>
 16#include "dax-private.h"
 17
 18/**
 19 * struct dax_device - anchor object for dax services
 20 * @inode: core vfs
 21 * @cdev: optional character interface for "device dax"
 22 * @private: dax driver private data
 23 * @flags: state and boolean properties
 24 * @ops: operations for this device
 25 * @holder_data: holder of a dax_device: could be filesystem or mapped device
 26 * @holder_ops: operations for the inner holder
 27 */
 28struct dax_device {
 29	struct inode inode;
 30	struct cdev cdev;
 31	void *private;
 32	unsigned long flags;
 33	const struct dax_operations *ops;
 34	void *holder_data;
 35	const struct dax_holder_operations *holder_ops;
 36};
 37
 38static dev_t dax_devt;
 39DEFINE_STATIC_SRCU(dax_srcu);
 40static struct vfsmount *dax_mnt;
 41static DEFINE_IDA(dax_minor_ida);
 42static struct kmem_cache *dax_cache __read_mostly;
 43static struct super_block *dax_superblock __read_mostly;
 44
 
 
 
 
 45int dax_read_lock(void)
 46{
 47	return srcu_read_lock(&dax_srcu);
 48}
 49EXPORT_SYMBOL_GPL(dax_read_lock);
 50
 51void dax_read_unlock(int id)
 52{
 53	srcu_read_unlock(&dax_srcu, id);
 54}
 55EXPORT_SYMBOL_GPL(dax_read_unlock);
 56
 57#if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX)
 58#include <linux/blkdev.h>
 59
 60static DEFINE_XARRAY(dax_hosts);
 
 
 
 61
 62int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk)
 63{
 64	return xa_insert(&dax_hosts, (unsigned long)disk, dax_dev, GFP_KERNEL);
 
 
 65}
 66EXPORT_SYMBOL_GPL(dax_add_host);
 67
 68void dax_remove_host(struct gendisk *disk)
 
 69{
 70	xa_erase(&dax_hosts, (unsigned long)disk);
 
 
 71}
 72EXPORT_SYMBOL_GPL(dax_remove_host);
 
 73
 74/**
 75 * fs_dax_get_by_bdev() - temporary lookup mechanism for filesystem-dax
 76 * @bdev: block device to find a dax_device for
 77 * @start_off: returns the byte offset into the dax_device that @bdev starts
 78 * @holder: filesystem or mapped device inside the dax_device
 79 * @ops: operations for the inner holder
 
 
 
 80 */
 81struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off,
 82		void *holder, const struct dax_holder_operations *ops)
 83{
 
 84	struct dax_device *dax_dev;
 85	u64 part_size;
 86	int id;
 
 
 
 
 
 
 
 
 
 87
 88	if (!blk_queue_dax(bdev->bd_disk->queue))
 89		return NULL;
 
 
 
 
 90
 91	*start_off = get_start_sect(bdev) * SECTOR_SIZE;
 92	part_size = bdev_nr_sectors(bdev) * SECTOR_SIZE;
 93	if (*start_off % PAGE_SIZE || part_size % PAGE_SIZE) {
 94		pr_info("%pg: error: unaligned partition for dax\n", bdev);
 95		return NULL;
 96	}
 97
 98	id = dax_read_lock();
 99	dax_dev = xa_load(&dax_hosts, (unsigned long)bdev->bd_disk);
100	if (!dax_dev || !dax_alive(dax_dev) || !igrab(&dax_dev->inode))
101		dax_dev = NULL;
102	else if (holder) {
103		if (!cmpxchg(&dax_dev->holder_data, NULL, holder))
104			dax_dev->holder_ops = ops;
105		else
106			dax_dev = NULL;
 
107	}
108	dax_read_unlock(id);
109
110	return dax_dev;
111}
112EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
113
114void fs_put_dax(struct dax_device *dax_dev, void *holder)
115{
116	if (dax_dev && holder &&
117	    cmpxchg(&dax_dev->holder_data, holder, NULL) == holder)
118		dax_dev->holder_ops = NULL;
119	put_dax(dax_dev);
120}
121EXPORT_SYMBOL_GPL(fs_put_dax);
122#endif /* CONFIG_BLOCK && CONFIG_FS_DAX */
123
124enum dax_device_flags {
125	/* !alive + rcu grace period == no new operations / mappings */
126	DAXDEV_ALIVE,
127	/* gate whether dax_flush() calls the low level flush routine */
128	DAXDEV_WRITE_CACHE,
129	/* flag to check if device supports synchronous flush */
130	DAXDEV_SYNC,
131	/* do not leave the caches dirty after writes */
132	DAXDEV_NOCACHE,
133	/* handle CPU fetch exceptions during reads */
134	DAXDEV_NOMC,
135};
136
137/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
138 * dax_direct_access() - translate a device pgoff to an absolute pfn
139 * @dax_dev: a dax_device instance representing the logical memory range
140 * @pgoff: offset in pages from the start of the device to translate
141 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
142 * @mode: indicator on normal access or recovery write
143 * @kaddr: output parameter that returns a virtual address mapping of pfn
144 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
145 *
146 * Return: negative errno if an error occurs, otherwise the number of
147 * pages accessible at the device relative @pgoff.
148 */
149long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
150		enum dax_access_mode mode, void **kaddr, pfn_t *pfn)
151{
152	long avail;
153
154	if (!dax_dev)
155		return -EOPNOTSUPP;
156
157	if (!dax_alive(dax_dev))
158		return -ENXIO;
159
160	if (nr_pages < 0)
161		return -EINVAL;
162
163	avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
164			mode, kaddr, pfn);
165	if (!avail)
166		return -ERANGE;
167	return min(avail, nr_pages);
168}
169EXPORT_SYMBOL_GPL(dax_direct_access);
170
171size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
172		size_t bytes, struct iov_iter *i)
173{
174	if (!dax_alive(dax_dev))
175		return 0;
176
177	/*
178	 * The userspace address for the memory copy has already been validated
179	 * via access_ok() in vfs_write, so use the 'no check' version to bypass
180	 * the HARDENED_USERCOPY overhead.
181	 */
182	if (test_bit(DAXDEV_NOCACHE, &dax_dev->flags))
183		return _copy_from_iter_flushcache(addr, bytes, i);
184	return _copy_from_iter(addr, bytes, i);
185}
186
187size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
188		size_t bytes, struct iov_iter *i)
189{
190	if (!dax_alive(dax_dev))
191		return 0;
192
193	/*
194	 * The userspace address for the memory copy has already been validated
195	 * via access_ok() in vfs_red, so use the 'no check' version to bypass
196	 * the HARDENED_USERCOPY overhead.
197	 */
198	if (test_bit(DAXDEV_NOMC, &dax_dev->flags))
199		return _copy_mc_to_iter(addr, bytes, i);
200	return _copy_to_iter(addr, bytes, i);
201}
202
203int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
204			size_t nr_pages)
205{
206	int ret;
207
208	if (!dax_alive(dax_dev))
209		return -ENXIO;
210	/*
211	 * There are no callers that want to zero more than one page as of now.
212	 * Once users are there, this check can be removed after the
213	 * device mapper code has been updated to split ranges across targets.
214	 */
215	if (nr_pages != 1)
216		return -EIO;
217
218	ret = dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages);
219	return dax_mem2blk_err(ret);
220}
221EXPORT_SYMBOL_GPL(dax_zero_page_range);
222
223size_t dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
224		void *addr, size_t bytes, struct iov_iter *iter)
225{
226	if (!dax_dev->ops->recovery_write)
227		return 0;
228	return dax_dev->ops->recovery_write(dax_dev, pgoff, addr, bytes, iter);
229}
230EXPORT_SYMBOL_GPL(dax_recovery_write);
231
232int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off,
233			      u64 len, int mf_flags)
234{
235	int rc, id;
236
237	id = dax_read_lock();
238	if (!dax_alive(dax_dev)) {
239		rc = -ENXIO;
240		goto out;
241	}
242
243	if (!dax_dev->holder_ops) {
244		rc = -EOPNOTSUPP;
245		goto out;
246	}
247
248	rc = dax_dev->holder_ops->notify_failure(dax_dev, off, len, mf_flags);
249out:
250	dax_read_unlock(id);
251	return rc;
252}
253EXPORT_SYMBOL_GPL(dax_holder_notify_failure);
254
255#ifdef CONFIG_ARCH_HAS_PMEM_API
256void arch_wb_cache_pmem(void *addr, size_t size);
257void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
258{
259	if (unlikely(!dax_write_cache_enabled(dax_dev)))
260		return;
261
262	arch_wb_cache_pmem(addr, size);
263}
264#else
265void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
266{
267}
268#endif
269EXPORT_SYMBOL_GPL(dax_flush);
270
271void dax_write_cache(struct dax_device *dax_dev, bool wc)
272{
273	if (wc)
274		set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
275	else
276		clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
277}
278EXPORT_SYMBOL_GPL(dax_write_cache);
279
280bool dax_write_cache_enabled(struct dax_device *dax_dev)
281{
282	return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
283}
284EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
285
286bool dax_synchronous(struct dax_device *dax_dev)
287{
288	return test_bit(DAXDEV_SYNC, &dax_dev->flags);
289}
290EXPORT_SYMBOL_GPL(dax_synchronous);
291
292void set_dax_synchronous(struct dax_device *dax_dev)
293{
294	set_bit(DAXDEV_SYNC, &dax_dev->flags);
295}
296EXPORT_SYMBOL_GPL(set_dax_synchronous);
297
298void set_dax_nocache(struct dax_device *dax_dev)
299{
300	set_bit(DAXDEV_NOCACHE, &dax_dev->flags);
301}
302EXPORT_SYMBOL_GPL(set_dax_nocache);
303
304void set_dax_nomc(struct dax_device *dax_dev)
305{
306	set_bit(DAXDEV_NOMC, &dax_dev->flags);
307}
308EXPORT_SYMBOL_GPL(set_dax_nomc);
309
310bool dax_alive(struct dax_device *dax_dev)
311{
312	lockdep_assert_held(&dax_srcu);
313	return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
314}
315EXPORT_SYMBOL_GPL(dax_alive);
316
 
 
 
 
 
317/*
318 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
319 * that any fault handlers or operations that might have seen
320 * dax_alive(), have completed.  Any operations that start after
321 * synchronize_srcu() has run will abort upon seeing !dax_alive().
322 */
323void kill_dax(struct dax_device *dax_dev)
324{
325	if (!dax_dev)
326		return;
327
328	if (dax_dev->holder_data != NULL)
329		dax_holder_notify_failure(dax_dev, 0, U64_MAX,
330				MF_MEM_PRE_REMOVE);
331
332	clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
333	synchronize_srcu(&dax_srcu);
334
335	/* clear holder data */
336	dax_dev->holder_ops = NULL;
337	dax_dev->holder_data = NULL;
 
 
338}
339EXPORT_SYMBOL_GPL(kill_dax);
340
341void run_dax(struct dax_device *dax_dev)
342{
343	set_bit(DAXDEV_ALIVE, &dax_dev->flags);
344}
345EXPORT_SYMBOL_GPL(run_dax);
346
347static struct inode *dax_alloc_inode(struct super_block *sb)
348{
349	struct dax_device *dax_dev;
350	struct inode *inode;
351
352	dax_dev = alloc_inode_sb(sb, dax_cache, GFP_KERNEL);
353	if (!dax_dev)
354		return NULL;
355
356	inode = &dax_dev->inode;
357	inode->i_rdev = 0;
358	return inode;
359}
360
361static struct dax_device *to_dax_dev(struct inode *inode)
362{
363	return container_of(inode, struct dax_device, inode);
364}
365
366static void dax_free_inode(struct inode *inode)
367{
 
368	struct dax_device *dax_dev = to_dax_dev(inode);
 
 
 
369	if (inode->i_rdev)
370		ida_free(&dax_minor_ida, iminor(inode));
371	kmem_cache_free(dax_cache, dax_dev);
372}
373
374static void dax_destroy_inode(struct inode *inode)
375{
376	struct dax_device *dax_dev = to_dax_dev(inode);
 
377	WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
378			"kill_dax() must be called before final iput()\n");
 
379}
380
381static const struct super_operations dax_sops = {
382	.statfs = simple_statfs,
383	.alloc_inode = dax_alloc_inode,
384	.destroy_inode = dax_destroy_inode,
385	.free_inode = dax_free_inode,
386	.drop_inode = generic_delete_inode,
387};
388
389static int dax_init_fs_context(struct fs_context *fc)
 
390{
391	struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC);
392	if (!ctx)
393		return -ENOMEM;
394	ctx->ops = &dax_sops;
395	return 0;
396}
397
398static struct file_system_type dax_fs_type = {
399	.name		= "dax",
400	.init_fs_context = dax_init_fs_context,
401	.kill_sb	= kill_anon_super,
402};
403
404static int dax_test(struct inode *inode, void *data)
405{
406	dev_t devt = *(dev_t *) data;
407
408	return inode->i_rdev == devt;
409}
410
411static int dax_set(struct inode *inode, void *data)
412{
413	dev_t devt = *(dev_t *) data;
414
415	inode->i_rdev = devt;
416	return 0;
417}
418
419static struct dax_device *dax_dev_get(dev_t devt)
420{
421	struct dax_device *dax_dev;
422	struct inode *inode;
423
424	inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
425			dax_test, dax_set, &devt);
426
427	if (!inode)
428		return NULL;
429
430	dax_dev = to_dax_dev(inode);
431	if (inode->i_state & I_NEW) {
432		set_bit(DAXDEV_ALIVE, &dax_dev->flags);
433		inode->i_cdev = &dax_dev->cdev;
434		inode->i_mode = S_IFCHR;
435		inode->i_flags = S_DAX;
436		mapping_set_gfp_mask(&inode->i_data, GFP_USER);
437		unlock_new_inode(inode);
438	}
439
440	return dax_dev;
441}
442
443struct dax_device *alloc_dax(void *private, const struct dax_operations *ops)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
444{
445	struct dax_device *dax_dev;
 
446	dev_t devt;
447	int minor;
448
449	if (WARN_ON_ONCE(ops && !ops->zero_page_range))
450		return ERR_PTR(-EINVAL);
 
451
452	minor = ida_alloc_max(&dax_minor_ida, MINORMASK, GFP_KERNEL);
453	if (minor < 0)
454		return ERR_PTR(-ENOMEM);
455
456	devt = MKDEV(MAJOR(dax_devt), minor);
457	dax_dev = dax_dev_get(devt);
458	if (!dax_dev)
459		goto err_dev;
460
 
461	dax_dev->ops = ops;
462	dax_dev->private = private;
463	return dax_dev;
464
465 err_dev:
466	ida_free(&dax_minor_ida, minor);
467	return ERR_PTR(-ENOMEM);
 
 
468}
469EXPORT_SYMBOL_GPL(alloc_dax);
470
471void put_dax(struct dax_device *dax_dev)
472{
473	if (!dax_dev)
474		return;
475	iput(&dax_dev->inode);
476}
477EXPORT_SYMBOL_GPL(put_dax);
478
479/**
480 * dax_holder() - obtain the holder of a dax device
481 * @dax_dev: a dax_device instance
482 *
483 * Return: the holder's data which represents the holder if registered,
484 * otherwize NULL.
485 */
486void *dax_holder(struct dax_device *dax_dev)
487{
488	return dax_dev->holder_data;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
489}
490EXPORT_SYMBOL_GPL(dax_holder);
491
492/**
493 * inode_dax: convert a public inode into its dax_dev
494 * @inode: An inode with i_cdev pointing to a dax_dev
495 *
496 * Note this is not equivalent to to_dax_dev() which is for private
497 * internal use where we know the inode filesystem type == dax_fs_type.
498 */
499struct dax_device *inode_dax(struct inode *inode)
500{
501	struct cdev *cdev = inode->i_cdev;
502
503	return container_of(cdev, struct dax_device, cdev);
504}
505EXPORT_SYMBOL_GPL(inode_dax);
506
507struct inode *dax_inode(struct dax_device *dax_dev)
508{
509	return &dax_dev->inode;
510}
511EXPORT_SYMBOL_GPL(dax_inode);
512
513void *dax_get_private(struct dax_device *dax_dev)
514{
515	if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
516		return NULL;
517	return dax_dev->private;
518}
519EXPORT_SYMBOL_GPL(dax_get_private);
520
521static void init_once(void *_dax_dev)
522{
523	struct dax_device *dax_dev = _dax_dev;
524	struct inode *inode = &dax_dev->inode;
525
526	memset(dax_dev, 0, sizeof(*dax_dev));
527	inode_init_once(inode);
528}
529
530static int dax_fs_init(void)
531{
532	int rc;
533
534	dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
535			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
536			 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
537			init_once);
538	if (!dax_cache)
539		return -ENOMEM;
540
 
 
 
 
541	dax_mnt = kern_mount(&dax_fs_type);
542	if (IS_ERR(dax_mnt)) {
543		rc = PTR_ERR(dax_mnt);
544		goto err_mount;
545	}
546	dax_superblock = dax_mnt->mnt_sb;
547
548	return 0;
549
550 err_mount:
 
 
551	kmem_cache_destroy(dax_cache);
552
553	return rc;
554}
555
556static void dax_fs_exit(void)
557{
558	kern_unmount(dax_mnt);
559	rcu_barrier();
560	kmem_cache_destroy(dax_cache);
561}
562
563static int __init dax_core_init(void)
564{
565	int rc;
566
567	rc = dax_fs_init();
568	if (rc)
569		return rc;
570
571	rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
572	if (rc)
573		goto err_chrdev;
574
575	rc = dax_bus_init();
576	if (rc)
577		goto err_bus;
578	return 0;
579
580err_bus:
581	unregister_chrdev_region(dax_devt, MINORMASK+1);
582err_chrdev:
583	dax_fs_exit();
584	return 0;
585}
586
587static void __exit dax_core_exit(void)
588{
589	dax_bus_exit();
590	unregister_chrdev_region(dax_devt, MINORMASK+1);
591	ida_destroy(&dax_minor_ida);
592	dax_fs_exit();
593}
594
595MODULE_AUTHOR("Intel Corporation");
596MODULE_LICENSE("GPL v2");
597subsys_initcall(dax_core_init);
598module_exit(dax_core_exit);