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/pfn_t.h>
11 #include <linux/cdev.h>
12 #include <linux/slab.h>
13 #include <linux/uio.h>
14 #include <linux/dax.h>
16 #include "dax-private.h"
19 * struct dax_device - anchor object for dax services
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
33 const struct dax_operations *ops;
35 const struct dax_holder_operations *holder_ops;
38 static dev_t dax_devt;
39 DEFINE_STATIC_SRCU(dax_srcu);
40 static struct vfsmount *dax_mnt;
41 static DEFINE_IDA(dax_minor_ida);
42 static struct kmem_cache *dax_cache __read_mostly;
43 static struct super_block *dax_superblock __read_mostly;
45 int dax_read_lock(void)
47 return srcu_read_lock(&dax_srcu);
49 EXPORT_SYMBOL_GPL(dax_read_lock);
51 void dax_read_unlock(int id)
53 srcu_read_unlock(&dax_srcu, id);
55 EXPORT_SYMBOL_GPL(dax_read_unlock);
57 #if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX)
58 #include <linux/blkdev.h>
60 static DEFINE_XARRAY(dax_hosts);
62 int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk)
64 return xa_insert(&dax_hosts, (unsigned long)disk, dax_dev, GFP_KERNEL);
66 EXPORT_SYMBOL_GPL(dax_add_host);
68 void dax_remove_host(struct gendisk *disk)
70 xa_erase(&dax_hosts, (unsigned long)disk);
72 EXPORT_SYMBOL_GPL(dax_remove_host);
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
81 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off,
82 void *holder, const struct dax_holder_operations *ops)
84 struct dax_device *dax_dev;
88 if (!blk_queue_dax(bdev->bd_disk->queue))
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);
99 dax_dev = xa_load(&dax_hosts, (unsigned long)bdev->bd_disk);
100 if (!dax_dev || !dax_alive(dax_dev) || !igrab(&dax_dev->inode))
103 if (!cmpxchg(&dax_dev->holder_data, NULL, holder))
104 dax_dev->holder_ops = ops;
112 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
114 void fs_put_dax(struct dax_device *dax_dev, void *holder)
116 if (dax_dev && holder &&
117 cmpxchg(&dax_dev->holder_data, holder, NULL) == holder)
118 dax_dev->holder_ops = NULL;
121 EXPORT_SYMBOL_GPL(fs_put_dax);
122 #endif /* CONFIG_BLOCK && CONFIG_FS_DAX */
124 enum dax_device_flags {
125 /* !alive + rcu grace period == no new operations / mappings */
127 /* gate whether dax_flush() calls the low level flush routine */
129 /* flag to check if device supports synchronous flush */
131 /* do not leave the caches dirty after writes */
133 /* handle CPU fetch exceptions during reads */
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
146 * Return: negative errno if an error occurs, otherwise the number of
147 * pages accessible at the device relative @pgoff.
149 long 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)
157 if (!dax_alive(dax_dev))
163 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
167 return min(avail, nr_pages);
169 EXPORT_SYMBOL_GPL(dax_direct_access);
171 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
172 size_t bytes, struct iov_iter *i)
174 if (!dax_alive(dax_dev))
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.
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);
187 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
188 size_t bytes, struct iov_iter *i)
190 if (!dax_alive(dax_dev))
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.
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);
203 int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
206 if (!dax_alive(dax_dev))
209 * There are no callers that want to zero more than one page as of now.
210 * Once users are there, this check can be removed after the
211 * device mapper code has been updated to split ranges across targets.
216 return dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages);
218 EXPORT_SYMBOL_GPL(dax_zero_page_range);
220 size_t dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
221 void *addr, size_t bytes, struct iov_iter *iter)
223 if (!dax_dev->ops->recovery_write)
225 return dax_dev->ops->recovery_write(dax_dev, pgoff, addr, bytes, iter);
227 EXPORT_SYMBOL_GPL(dax_recovery_write);
229 int dax_holder_notify_failure(struct dax_device *dax_dev, u64 off,
230 u64 len, int mf_flags)
234 id = dax_read_lock();
235 if (!dax_alive(dax_dev)) {
240 if (!dax_dev->holder_ops) {
245 rc = dax_dev->holder_ops->notify_failure(dax_dev, off, len, mf_flags);
250 EXPORT_SYMBOL_GPL(dax_holder_notify_failure);
252 #ifdef CONFIG_ARCH_HAS_PMEM_API
253 void arch_wb_cache_pmem(void *addr, size_t size);
254 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
256 if (unlikely(!dax_write_cache_enabled(dax_dev)))
259 arch_wb_cache_pmem(addr, size);
262 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
266 EXPORT_SYMBOL_GPL(dax_flush);
268 void dax_write_cache(struct dax_device *dax_dev, bool wc)
271 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
273 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
275 EXPORT_SYMBOL_GPL(dax_write_cache);
277 bool dax_write_cache_enabled(struct dax_device *dax_dev)
279 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
281 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
283 bool dax_synchronous(struct dax_device *dax_dev)
285 return test_bit(DAXDEV_SYNC, &dax_dev->flags);
287 EXPORT_SYMBOL_GPL(dax_synchronous);
289 void set_dax_synchronous(struct dax_device *dax_dev)
291 set_bit(DAXDEV_SYNC, &dax_dev->flags);
293 EXPORT_SYMBOL_GPL(set_dax_synchronous);
295 void set_dax_nocache(struct dax_device *dax_dev)
297 set_bit(DAXDEV_NOCACHE, &dax_dev->flags);
299 EXPORT_SYMBOL_GPL(set_dax_nocache);
301 void set_dax_nomc(struct dax_device *dax_dev)
303 set_bit(DAXDEV_NOMC, &dax_dev->flags);
305 EXPORT_SYMBOL_GPL(set_dax_nomc);
307 bool dax_alive(struct dax_device *dax_dev)
309 lockdep_assert_held(&dax_srcu);
310 return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
312 EXPORT_SYMBOL_GPL(dax_alive);
315 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
316 * that any fault handlers or operations that might have seen
317 * dax_alive(), have completed. Any operations that start after
318 * synchronize_srcu() has run will abort upon seeing !dax_alive().
320 void kill_dax(struct dax_device *dax_dev)
325 if (dax_dev->holder_data != NULL)
326 dax_holder_notify_failure(dax_dev, 0, U64_MAX, 0);
328 clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
329 synchronize_srcu(&dax_srcu);
331 /* clear holder data */
332 dax_dev->holder_ops = NULL;
333 dax_dev->holder_data = NULL;
335 EXPORT_SYMBOL_GPL(kill_dax);
337 void run_dax(struct dax_device *dax_dev)
339 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
341 EXPORT_SYMBOL_GPL(run_dax);
343 static struct inode *dax_alloc_inode(struct super_block *sb)
345 struct dax_device *dax_dev;
348 dax_dev = alloc_inode_sb(sb, dax_cache, GFP_KERNEL);
352 inode = &dax_dev->inode;
357 static struct dax_device *to_dax_dev(struct inode *inode)
359 return container_of(inode, struct dax_device, inode);
362 static void dax_free_inode(struct inode *inode)
364 struct dax_device *dax_dev = to_dax_dev(inode);
366 ida_simple_remove(&dax_minor_ida, iminor(inode));
367 kmem_cache_free(dax_cache, dax_dev);
370 static void dax_destroy_inode(struct inode *inode)
372 struct dax_device *dax_dev = to_dax_dev(inode);
373 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
374 "kill_dax() must be called before final iput()\n");
377 static const struct super_operations dax_sops = {
378 .statfs = simple_statfs,
379 .alloc_inode = dax_alloc_inode,
380 .destroy_inode = dax_destroy_inode,
381 .free_inode = dax_free_inode,
382 .drop_inode = generic_delete_inode,
385 static int dax_init_fs_context(struct fs_context *fc)
387 struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC);
390 ctx->ops = &dax_sops;
394 static struct file_system_type dax_fs_type = {
396 .init_fs_context = dax_init_fs_context,
397 .kill_sb = kill_anon_super,
400 static int dax_test(struct inode *inode, void *data)
402 dev_t devt = *(dev_t *) data;
404 return inode->i_rdev == devt;
407 static int dax_set(struct inode *inode, void *data)
409 dev_t devt = *(dev_t *) data;
411 inode->i_rdev = devt;
415 static struct dax_device *dax_dev_get(dev_t devt)
417 struct dax_device *dax_dev;
420 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
421 dax_test, dax_set, &devt);
426 dax_dev = to_dax_dev(inode);
427 if (inode->i_state & I_NEW) {
428 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
429 inode->i_cdev = &dax_dev->cdev;
430 inode->i_mode = S_IFCHR;
431 inode->i_flags = S_DAX;
432 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
433 unlock_new_inode(inode);
439 struct dax_device *alloc_dax(void *private, const struct dax_operations *ops)
441 struct dax_device *dax_dev;
445 if (WARN_ON_ONCE(ops && !ops->zero_page_range))
446 return ERR_PTR(-EINVAL);
448 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
450 return ERR_PTR(-ENOMEM);
452 devt = MKDEV(MAJOR(dax_devt), minor);
453 dax_dev = dax_dev_get(devt);
458 dax_dev->private = private;
462 ida_simple_remove(&dax_minor_ida, minor);
463 return ERR_PTR(-ENOMEM);
465 EXPORT_SYMBOL_GPL(alloc_dax);
467 void put_dax(struct dax_device *dax_dev)
471 iput(&dax_dev->inode);
473 EXPORT_SYMBOL_GPL(put_dax);
476 * dax_holder() - obtain the holder of a dax device
477 * @dax_dev: a dax_device instance
479 * Return: the holder's data which represents the holder if registered,
482 void *dax_holder(struct dax_device *dax_dev)
484 return dax_dev->holder_data;
486 EXPORT_SYMBOL_GPL(dax_holder);
489 * inode_dax: convert a public inode into its dax_dev
490 * @inode: An inode with i_cdev pointing to a dax_dev
492 * Note this is not equivalent to to_dax_dev() which is for private
493 * internal use where we know the inode filesystem type == dax_fs_type.
495 struct dax_device *inode_dax(struct inode *inode)
497 struct cdev *cdev = inode->i_cdev;
499 return container_of(cdev, struct dax_device, cdev);
501 EXPORT_SYMBOL_GPL(inode_dax);
503 struct inode *dax_inode(struct dax_device *dax_dev)
505 return &dax_dev->inode;
507 EXPORT_SYMBOL_GPL(dax_inode);
509 void *dax_get_private(struct dax_device *dax_dev)
511 if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags))
513 return dax_dev->private;
515 EXPORT_SYMBOL_GPL(dax_get_private);
517 static void init_once(void *_dax_dev)
519 struct dax_device *dax_dev = _dax_dev;
520 struct inode *inode = &dax_dev->inode;
522 memset(dax_dev, 0, sizeof(*dax_dev));
523 inode_init_once(inode);
526 static int dax_fs_init(void)
530 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
531 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
532 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
537 dax_mnt = kern_mount(&dax_fs_type);
538 if (IS_ERR(dax_mnt)) {
539 rc = PTR_ERR(dax_mnt);
542 dax_superblock = dax_mnt->mnt_sb;
547 kmem_cache_destroy(dax_cache);
552 static void dax_fs_exit(void)
554 kern_unmount(dax_mnt);
556 kmem_cache_destroy(dax_cache);
559 static int __init dax_core_init(void)
567 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
577 unregister_chrdev_region(dax_devt, MINORMASK+1);
583 static void __exit dax_core_exit(void)
586 unregister_chrdev_region(dax_devt, MINORMASK+1);
587 ida_destroy(&dax_minor_ida);
591 MODULE_AUTHOR("Intel Corporation");
592 MODULE_LICENSE("GPL v2");
593 subsys_initcall(dax_core_init);
594 module_exit(dax_core_exit);