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 <linux/cacheinfo.h>
17#include "dax-private.h"
18
19/**
20 * struct dax_device - anchor object for dax services
21 * @inode: core vfs
22 * @cdev: optional character interface for "device dax"
23 * @private: dax driver private data
24 * @flags: state and boolean properties
25 * @ops: operations for this device
26 * @holder_data: holder of a dax_device: could be filesystem or mapped device
27 * @holder_ops: operations for the inner holder
28 */
29struct dax_device {
30	struct inode inode;
31	struct cdev cdev;
32	void *private;
33	unsigned long flags;
34	const struct dax_operations *ops;
35	void *holder_data;
36	const struct dax_holder_operations *holder_ops;
37};
38
39static dev_t dax_devt;
40DEFINE_STATIC_SRCU(dax_srcu);
41static struct vfsmount *dax_mnt;
42static DEFINE_IDA(dax_minor_ida);
43static struct kmem_cache *dax_cache __read_mostly;
44static struct super_block *dax_superblock __read_mostly;
45
46int dax_read_lock(void)
47{
48	return srcu_read_lock(&dax_srcu);
49}
50EXPORT_SYMBOL_GPL(dax_read_lock);
51
52void dax_read_unlock(int id)
53{
54	srcu_read_unlock(&dax_srcu, id);
55}
56EXPORT_SYMBOL_GPL(dax_read_unlock);
57
58#if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX)
59#include <linux/blkdev.h>
60
61static DEFINE_XARRAY(dax_hosts);
62
63int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk)
64{
65	return xa_insert(&dax_hosts, (unsigned long)disk, dax_dev, GFP_KERNEL);
66}
67EXPORT_SYMBOL_GPL(dax_add_host);
68
69void dax_remove_host(struct gendisk *disk)
70{
71	xa_erase(&dax_hosts, (unsigned long)disk);
72}
73EXPORT_SYMBOL_GPL(dax_remove_host);
74
75/**
76 * fs_dax_get_by_bdev() - temporary lookup mechanism for filesystem-dax
77 * @bdev: block device to find a dax_device for
78 * @start_off: returns the byte offset into the dax_device that @bdev starts
79 * @holder: filesystem or mapped device inside the dax_device
80 * @ops: operations for the inner holder
81 */
82struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off,
83		void *holder, const struct dax_holder_operations *ops)
84{
85	struct dax_device *dax_dev;
86	u64 part_size;
87	int id;
88
89	if (!blk_queue_dax(bdev->bd_disk->queue))
90		return NULL;
91
92	*start_off = get_start_sect(bdev) * SECTOR_SIZE;
93	part_size = bdev_nr_sectors(bdev) * SECTOR_SIZE;
94	if (*start_off % PAGE_SIZE || part_size % PAGE_SIZE) {
95		pr_info("%pg: error: unaligned partition for dax\n", bdev);
96		return NULL;
97	}
98
99	id = dax_read_lock();
100	dax_dev = xa_load(&dax_hosts, (unsigned long)bdev->bd_disk);
101	if (!dax_dev || !dax_alive(dax_dev) || !igrab(&dax_dev->inode))
102		dax_dev = NULL;
103	else if (holder) {
104		if (!cmpxchg(&dax_dev->holder_data, NULL, holder))
105			dax_dev->holder_ops = ops;
106		else
107			dax_dev = NULL;
108	}
109	dax_read_unlock(id);
110
111	return dax_dev;
112}
113EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
114
115void fs_put_dax(struct dax_device *dax_dev, void *holder)
116{
117	if (dax_dev && holder &&
118	    cmpxchg(&dax_dev->holder_data, holder, NULL) == holder)
119		dax_dev->holder_ops = NULL;
120	put_dax(dax_dev);
121}
122EXPORT_SYMBOL_GPL(fs_put_dax);
123#endif /* CONFIG_BLOCK && CONFIG_FS_DAX */
124
125enum dax_device_flags {
126	/* !alive + rcu grace period == no new operations / mappings */
127	DAXDEV_ALIVE,
128	/* gate whether dax_flush() calls the low level flush routine */
129	DAXDEV_WRITE_CACHE,
130	/* flag to check if device supports synchronous flush */
131	DAXDEV_SYNC,
132	/* do not leave the caches dirty after writes */
133	DAXDEV_NOCACHE,
134	/* handle CPU fetch exceptions during reads */
135	DAXDEV_NOMC,
136};
137
138/**
139 * dax_direct_access() - translate a device pgoff to an absolute pfn
140 * @dax_dev: a dax_device instance representing the logical memory range
141 * @pgoff: offset in pages from the start of the device to translate
142 * @nr_pages: number of consecutive pages caller can handle relative to @pfn
143 * @mode: indicator on normal access or recovery write
144 * @kaddr: output parameter that returns a virtual address mapping of pfn
145 * @pfn: output parameter that returns an absolute pfn translation of @pgoff
146 *
147 * Return: negative errno if an error occurs, otherwise the number of
148 * pages accessible at the device relative @pgoff.
149 */
150long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
151		enum dax_access_mode mode, void **kaddr, pfn_t *pfn)
152{
153	long avail;
154
155	if (!dax_dev)
156		return -EOPNOTSUPP;
157
158	if (!dax_alive(dax_dev))
159		return -ENXIO;
160
161	if (nr_pages < 0)
162		return -EINVAL;
163
164	avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
165			mode, kaddr, pfn);
166	if (!avail)
167		return -ERANGE;
168	return min(avail, nr_pages);
169}
170EXPORT_SYMBOL_GPL(dax_direct_access);
171
172size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
173		size_t bytes, struct iov_iter *i)
174{
175	if (!dax_alive(dax_dev))
176		return 0;
177
178	/*
179	 * The userspace address for the memory copy has already been validated
180	 * via access_ok() in vfs_write, so use the 'no check' version to bypass
181	 * the HARDENED_USERCOPY overhead.
182	 */
183	if (test_bit(DAXDEV_NOCACHE, &dax_dev->flags))
184		return _copy_from_iter_flushcache(addr, bytes, i);
185	return _copy_from_iter(addr, bytes, i);
186}
187
188size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
189		size_t bytes, struct iov_iter *i)
190{
191	if (!dax_alive(dax_dev))
192		return 0;
193
194	/*
195	 * The userspace address for the memory copy has already been validated
196	 * via access_ok() in vfs_red, so use the 'no check' version to bypass
197	 * the HARDENED_USERCOPY overhead.
198	 */
199	if (test_bit(DAXDEV_NOMC, &dax_dev->flags))
200		return _copy_mc_to_iter(addr, bytes, i);
201	return _copy_to_iter(addr, bytes, i);
202}
203
204int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
205			size_t nr_pages)
206{
207	int ret;
208
209	if (!dax_alive(dax_dev))
210		return -ENXIO;
211	/*
212	 * There are no callers that want to zero more than one page as of now.
213	 * Once users are there, this check can be removed after the
214	 * device mapper code has been updated to split ranges across targets.
215	 */
216	if (nr_pages != 1)
217		return -EIO;
218
219	ret = dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages);
220	return dax_mem2blk_err(ret);
221}
222EXPORT_SYMBOL_GPL(dax_zero_page_range);
223
224size_t dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
225		void *addr, size_t bytes, struct iov_iter *iter)
226{
227	if (!dax_dev->ops->recovery_write)
228		return 0;
229	return dax_dev->ops->recovery_write(dax_dev, pgoff, addr, bytes, iter);
230}
231EXPORT_SYMBOL_GPL(dax_recovery_write);
232
233int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off,
234			      u64 len, int mf_flags)
235{
236	int rc, id;
237
238	id = dax_read_lock();
239	if (!dax_alive(dax_dev)) {
240		rc = -ENXIO;
241		goto out;
242	}
243
244	if (!dax_dev->holder_ops) {
245		rc = -EOPNOTSUPP;
246		goto out;
247	}
248
249	rc = dax_dev->holder_ops->notify_failure(dax_dev, off, len, mf_flags);
250out:
251	dax_read_unlock(id);
252	return rc;
253}
254EXPORT_SYMBOL_GPL(dax_holder_notify_failure);
255
256#ifdef CONFIG_ARCH_HAS_PMEM_API
257void arch_wb_cache_pmem(void *addr, size_t size);
258void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
259{
260	if (unlikely(!dax_write_cache_enabled(dax_dev)))
261		return;
262
263	arch_wb_cache_pmem(addr, size);
264}
265#else
266void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
267{
268}
269#endif
270EXPORT_SYMBOL_GPL(dax_flush);
271
272void dax_write_cache(struct dax_device *dax_dev, bool wc)
273{
274	if (wc)
275		set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
276	else
277		clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
278}
279EXPORT_SYMBOL_GPL(dax_write_cache);
280
281bool dax_write_cache_enabled(struct dax_device *dax_dev)
282{
283	return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
284}
285EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
286
287bool dax_synchronous(struct dax_device *dax_dev)
288{
289	return test_bit(DAXDEV_SYNC, &dax_dev->flags);
290}
291EXPORT_SYMBOL_GPL(dax_synchronous);
292
293void set_dax_synchronous(struct dax_device *dax_dev)
294{
295	set_bit(DAXDEV_SYNC, &dax_dev->flags);
296}
297EXPORT_SYMBOL_GPL(set_dax_synchronous);
298
299void set_dax_nocache(struct dax_device *dax_dev)
300{
301	set_bit(DAXDEV_NOCACHE, &dax_dev->flags);
302}
303EXPORT_SYMBOL_GPL(set_dax_nocache);
304
305void set_dax_nomc(struct dax_device *dax_dev)
306{
307	set_bit(DAXDEV_NOMC, &dax_dev->flags);
308}
309EXPORT_SYMBOL_GPL(set_dax_nomc);
310
311bool dax_alive(struct dax_device *dax_dev)
312{
313	lockdep_assert_held(&dax_srcu);
314	return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
315}
316EXPORT_SYMBOL_GPL(dax_alive);
317
318/*
319 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
320 * that any fault handlers or operations that might have seen
321 * dax_alive(), have completed.  Any operations that start after
322 * synchronize_srcu() has run will abort upon seeing !dax_alive().
323 *
324 * Note, because alloc_dax() returns an ERR_PTR() on error, callers
325 * typically store its result into a local variable in order to check
326 * the result. Therefore, care must be taken to populate the struct
327 * device dax_dev field make sure the dax_dev is not leaked.
328 */
329void kill_dax(struct dax_device *dax_dev)
330{
331	if (!dax_dev)
332		return;
333
334	if (dax_dev->holder_data != NULL)
335		dax_holder_notify_failure(dax_dev, 0, U64_MAX,
336				MF_MEM_PRE_REMOVE);
337
338	clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
339	synchronize_srcu(&dax_srcu);
340
341	/* clear holder data */
342	dax_dev->holder_ops = NULL;
343	dax_dev->holder_data = NULL;
344}
345EXPORT_SYMBOL_GPL(kill_dax);
346
347void run_dax(struct dax_device *dax_dev)
348{
349	set_bit(DAXDEV_ALIVE, &dax_dev->flags);
350}
351EXPORT_SYMBOL_GPL(run_dax);
352
353static struct inode *dax_alloc_inode(struct super_block *sb)
354{
355	struct dax_device *dax_dev;
356	struct inode *inode;
357
358	dax_dev = alloc_inode_sb(sb, dax_cache, GFP_KERNEL);
359	if (!dax_dev)
360		return NULL;
361
362	inode = &dax_dev->inode;
363	inode->i_rdev = 0;
364	return inode;
365}
366
367static struct dax_device *to_dax_dev(struct inode *inode)
368{
369	return container_of(inode, struct dax_device, inode);
370}
371
372static void dax_free_inode(struct inode *inode)
373{
374	struct dax_device *dax_dev = to_dax_dev(inode);
375	if (inode->i_rdev)
376		ida_free(&dax_minor_ida, iminor(inode));
377	kmem_cache_free(dax_cache, dax_dev);
378}
379
380static void dax_destroy_inode(struct inode *inode)
381{
382	struct dax_device *dax_dev = to_dax_dev(inode);
383	WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
384			"kill_dax() must be called before final iput()\n");
385}
386
387static const struct super_operations dax_sops = {
388	.statfs = simple_statfs,
389	.alloc_inode = dax_alloc_inode,
390	.destroy_inode = dax_destroy_inode,
391	.free_inode = dax_free_inode,
392	.drop_inode = generic_delete_inode,
393};
394
395static int dax_init_fs_context(struct fs_context *fc)
396{
397	struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC);
398	if (!ctx)
399		return -ENOMEM;
400	ctx->ops = &dax_sops;
401	return 0;
402}
403
404static struct file_system_type dax_fs_type = {
405	.name		= "dax",
406	.init_fs_context = dax_init_fs_context,
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
449struct dax_device *alloc_dax(void *private, const struct dax_operations *ops)
450{
451	struct dax_device *dax_dev;
452	dev_t devt;
453	int minor;
454
455	/*
456	 * Unavailable on architectures with virtually aliased data caches,
457	 * except for device-dax (NULL operations pointer), which does
458	 * not use aliased mappings from the kernel.
459	 */
460	if (ops && cpu_dcache_is_aliasing())
461		return ERR_PTR(-EOPNOTSUPP);
462
463	if (WARN_ON_ONCE(ops && !ops->zero_page_range))
464		return ERR_PTR(-EINVAL);
465
466	minor = ida_alloc_max(&dax_minor_ida, MINORMASK, GFP_KERNEL);
467	if (minor < 0)
468		return ERR_PTR(-ENOMEM);
469
470	devt = MKDEV(MAJOR(dax_devt), minor);
471	dax_dev = dax_dev_get(devt);
472	if (!dax_dev)
473		goto err_dev;
474
475	dax_dev->ops = ops;
476	dax_dev->private = private;
477	return dax_dev;
478
479 err_dev:
480	ida_free(&dax_minor_ida, minor);
481	return ERR_PTR(-ENOMEM);
482}
483EXPORT_SYMBOL_GPL(alloc_dax);
484
485void put_dax(struct dax_device *dax_dev)
486{
487	if (!dax_dev)
488		return;
489	iput(&dax_dev->inode);
490}
491EXPORT_SYMBOL_GPL(put_dax);
492
493/**
494 * dax_holder() - obtain the holder of a dax device
495 * @dax_dev: a dax_device instance
496 *
497 * Return: the holder's data which represents the holder if registered,
498 * otherwize NULL.
499 */
500void *dax_holder(struct dax_device *dax_dev)
501{
502	return dax_dev->holder_data;
503}
504EXPORT_SYMBOL_GPL(dax_holder);
505
506/**
507 * inode_dax: convert a public inode into its dax_dev
508 * @inode: An inode with i_cdev pointing to a dax_dev
509 *
510 * Note this is not equivalent to to_dax_dev() which is for private
511 * internal use where we know the inode filesystem type == dax_fs_type.
512 */
513struct dax_device *inode_dax(struct inode *inode)
514{
515	struct cdev *cdev = inode->i_cdev;
516
517	return container_of(cdev, struct dax_device, cdev);
518}
519EXPORT_SYMBOL_GPL(inode_dax);
520
521struct inode *dax_inode(struct dax_device *dax_dev)
522{
523	return &dax_dev->inode;
524}
525EXPORT_SYMBOL_GPL(dax_inode);
526
527void *dax_get_private(struct dax_device *dax_dev)
528{
529	if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
530		return NULL;
531	return dax_dev->private;
532}
533EXPORT_SYMBOL_GPL(dax_get_private);
534
535static void init_once(void *_dax_dev)
536{
537	struct dax_device *dax_dev = _dax_dev;
538	struct inode *inode = &dax_dev->inode;
539
540	memset(dax_dev, 0, sizeof(*dax_dev));
541	inode_init_once(inode);
542}
543
544static int dax_fs_init(void)
545{
546	int rc;
547
548	dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
549			SLAB_HWCACHE_ALIGN | SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
550			init_once);
551	if (!dax_cache)
552		return -ENOMEM;
553
554	dax_mnt = kern_mount(&dax_fs_type);
555	if (IS_ERR(dax_mnt)) {
556		rc = PTR_ERR(dax_mnt);
557		goto err_mount;
558	}
559	dax_superblock = dax_mnt->mnt_sb;
560
561	return 0;
562
563 err_mount:
564	kmem_cache_destroy(dax_cache);
565
566	return rc;
567}
568
569static void dax_fs_exit(void)
570{
571	kern_unmount(dax_mnt);
572	rcu_barrier();
573	kmem_cache_destroy(dax_cache);
574}
575
576static int __init dax_core_init(void)
577{
578	int rc;
579
580	rc = dax_fs_init();
581	if (rc)
582		return rc;
583
584	rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
585	if (rc)
586		goto err_chrdev;
587
588	rc = dax_bus_init();
589	if (rc)
590		goto err_bus;
591	return 0;
592
593err_bus:
594	unregister_chrdev_region(dax_devt, MINORMASK+1);
595err_chrdev:
596	dax_fs_exit();
597	return 0;
598}
599
600static void __exit dax_core_exit(void)
601{
602	dax_bus_exit();
603	unregister_chrdev_region(dax_devt, MINORMASK+1);
604	ida_destroy(&dax_minor_ida);
605	dax_fs_exit();
606}
607
608MODULE_AUTHOR("Intel Corporation");
609MODULE_LICENSE("GPL v2");
610subsys_initcall(dax_core_init);
611module_exit(dax_core_exit);
612