xref: /openbmc/linux/fs/overlayfs/file.c (revision f4fc91af)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017 Red Hat, Inc.
4  */
5 
6 #include <linux/cred.h>
7 #include <linux/file.h>
8 #include <linux/mount.h>
9 #include <linux/xattr.h>
10 #include <linux/uio.h>
11 #include <linux/uaccess.h>
12 #include <linux/splice.h>
13 #include <linux/security.h>
14 #include <linux/mm.h>
15 #include <linux/fs.h>
16 #include "overlayfs.h"
17 
18 struct ovl_aio_req {
19 	struct kiocb iocb;
20 	struct kiocb *orig_iocb;
21 	struct fd fd;
22 };
23 
24 static struct kmem_cache *ovl_aio_request_cachep;
25 
26 static char ovl_whatisit(struct inode *inode, struct inode *realinode)
27 {
28 	if (realinode != ovl_inode_upper(inode))
29 		return 'l';
30 	if (ovl_has_upperdata(inode))
31 		return 'u';
32 	else
33 		return 'm';
34 }
35 
36 /* No atime modificaton nor notify on underlying */
37 #define OVL_OPEN_FLAGS (O_NOATIME | FMODE_NONOTIFY)
38 
39 static struct file *ovl_open_realfile(const struct file *file,
40 				      struct inode *realinode)
41 {
42 	struct inode *inode = file_inode(file);
43 	struct file *realfile;
44 	const struct cred *old_cred;
45 	int flags = file->f_flags | OVL_OPEN_FLAGS;
46 	int acc_mode = ACC_MODE(flags);
47 	int err;
48 
49 	if (flags & O_APPEND)
50 		acc_mode |= MAY_APPEND;
51 
52 	old_cred = ovl_override_creds(inode->i_sb);
53 	err = inode_permission(realinode, MAY_OPEN | acc_mode);
54 	if (err) {
55 		realfile = ERR_PTR(err);
56 	} else if (!inode_owner_or_capable(realinode)) {
57 		realfile = ERR_PTR(-EPERM);
58 	} else {
59 		realfile = open_with_fake_path(&file->f_path, flags, realinode,
60 					       current_cred());
61 	}
62 	revert_creds(old_cred);
63 
64 	pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
65 		 file, file, ovl_whatisit(inode, realinode), file->f_flags,
66 		 realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
67 
68 	return realfile;
69 }
70 
71 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
72 
73 static int ovl_change_flags(struct file *file, unsigned int flags)
74 {
75 	struct inode *inode = file_inode(file);
76 	int err;
77 
78 	flags |= OVL_OPEN_FLAGS;
79 
80 	/* If some flag changed that cannot be changed then something's amiss */
81 	if (WARN_ON((file->f_flags ^ flags) & ~OVL_SETFL_MASK))
82 		return -EIO;
83 
84 	flags &= OVL_SETFL_MASK;
85 
86 	if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
87 		return -EPERM;
88 
89 	if (flags & O_DIRECT) {
90 		if (!file->f_mapping->a_ops ||
91 		    !file->f_mapping->a_ops->direct_IO)
92 			return -EINVAL;
93 	}
94 
95 	if (file->f_op->check_flags) {
96 		err = file->f_op->check_flags(flags);
97 		if (err)
98 			return err;
99 	}
100 
101 	spin_lock(&file->f_lock);
102 	file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
103 	spin_unlock(&file->f_lock);
104 
105 	return 0;
106 }
107 
108 static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
109 			       bool allow_meta)
110 {
111 	struct inode *inode = file_inode(file);
112 	struct inode *realinode;
113 
114 	real->flags = 0;
115 	real->file = file->private_data;
116 
117 	if (allow_meta)
118 		realinode = ovl_inode_real(inode);
119 	else
120 		realinode = ovl_inode_realdata(inode);
121 
122 	/* Has it been copied up since we'd opened it? */
123 	if (unlikely(file_inode(real->file) != realinode)) {
124 		real->flags = FDPUT_FPUT;
125 		real->file = ovl_open_realfile(file, realinode);
126 
127 		return PTR_ERR_OR_ZERO(real->file);
128 	}
129 
130 	/* Did the flags change since open? */
131 	if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
132 		return ovl_change_flags(real->file, file->f_flags);
133 
134 	return 0;
135 }
136 
137 static int ovl_real_fdget(const struct file *file, struct fd *real)
138 {
139 	if (d_is_dir(file_dentry(file))) {
140 		real->flags = 0;
141 		real->file = ovl_dir_real_file(file, false);
142 
143 		return PTR_ERR_OR_ZERO(real->file);
144 	}
145 
146 	return ovl_real_fdget_meta(file, real, false);
147 }
148 
149 static int ovl_open(struct inode *inode, struct file *file)
150 {
151 	struct file *realfile;
152 	int err;
153 
154 	err = ovl_maybe_copy_up(file_dentry(file), file->f_flags);
155 	if (err)
156 		return err;
157 
158 	/* No longer need these flags, so don't pass them on to underlying fs */
159 	file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
160 
161 	realfile = ovl_open_realfile(file, ovl_inode_realdata(inode));
162 	if (IS_ERR(realfile))
163 		return PTR_ERR(realfile);
164 
165 	file->private_data = realfile;
166 
167 	return 0;
168 }
169 
170 static int ovl_release(struct inode *inode, struct file *file)
171 {
172 	fput(file->private_data);
173 
174 	return 0;
175 }
176 
177 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
178 {
179 	struct inode *inode = file_inode(file);
180 	struct fd real;
181 	const struct cred *old_cred;
182 	loff_t ret;
183 
184 	/*
185 	 * The two special cases below do not need to involve real fs,
186 	 * so we can optimizing concurrent callers.
187 	 */
188 	if (offset == 0) {
189 		if (whence == SEEK_CUR)
190 			return file->f_pos;
191 
192 		if (whence == SEEK_SET)
193 			return vfs_setpos(file, 0, 0);
194 	}
195 
196 	ret = ovl_real_fdget(file, &real);
197 	if (ret)
198 		return ret;
199 
200 	/*
201 	 * Overlay file f_pos is the master copy that is preserved
202 	 * through copy up and modified on read/write, but only real
203 	 * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
204 	 * limitations that are more strict than ->s_maxbytes for specific
205 	 * files, so we use the real file to perform seeks.
206 	 */
207 	ovl_inode_lock(inode);
208 	real.file->f_pos = file->f_pos;
209 
210 	old_cred = ovl_override_creds(inode->i_sb);
211 	ret = vfs_llseek(real.file, offset, whence);
212 	revert_creds(old_cred);
213 
214 	file->f_pos = real.file->f_pos;
215 	ovl_inode_unlock(inode);
216 
217 	fdput(real);
218 
219 	return ret;
220 }
221 
222 static void ovl_file_accessed(struct file *file)
223 {
224 	struct inode *inode, *upperinode;
225 
226 	if (file->f_flags & O_NOATIME)
227 		return;
228 
229 	inode = file_inode(file);
230 	upperinode = ovl_inode_upper(inode);
231 
232 	if (!upperinode)
233 		return;
234 
235 	if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
236 	     !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) {
237 		inode->i_mtime = upperinode->i_mtime;
238 		inode->i_ctime = upperinode->i_ctime;
239 	}
240 
241 	touch_atime(&file->f_path);
242 }
243 
244 static rwf_t ovl_iocb_to_rwf(int ifl)
245 {
246 	rwf_t flags = 0;
247 
248 	if (ifl & IOCB_NOWAIT)
249 		flags |= RWF_NOWAIT;
250 	if (ifl & IOCB_HIPRI)
251 		flags |= RWF_HIPRI;
252 	if (ifl & IOCB_DSYNC)
253 		flags |= RWF_DSYNC;
254 	if (ifl & IOCB_SYNC)
255 		flags |= RWF_SYNC;
256 
257 	return flags;
258 }
259 
260 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
261 {
262 	struct kiocb *iocb = &aio_req->iocb;
263 	struct kiocb *orig_iocb = aio_req->orig_iocb;
264 
265 	if (iocb->ki_flags & IOCB_WRITE) {
266 		struct inode *inode = file_inode(orig_iocb->ki_filp);
267 
268 		/* Actually acquired in ovl_write_iter() */
269 		__sb_writers_acquired(file_inode(iocb->ki_filp)->i_sb,
270 				      SB_FREEZE_WRITE);
271 		file_end_write(iocb->ki_filp);
272 		ovl_copyattr(ovl_inode_real(inode), inode);
273 	}
274 
275 	orig_iocb->ki_pos = iocb->ki_pos;
276 	fdput(aio_req->fd);
277 	kmem_cache_free(ovl_aio_request_cachep, aio_req);
278 }
279 
280 static void ovl_aio_rw_complete(struct kiocb *iocb, long res, long res2)
281 {
282 	struct ovl_aio_req *aio_req = container_of(iocb,
283 						   struct ovl_aio_req, iocb);
284 	struct kiocb *orig_iocb = aio_req->orig_iocb;
285 
286 	ovl_aio_cleanup_handler(aio_req);
287 	orig_iocb->ki_complete(orig_iocb, res, res2);
288 }
289 
290 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
291 {
292 	struct file *file = iocb->ki_filp;
293 	struct fd real;
294 	const struct cred *old_cred;
295 	ssize_t ret;
296 
297 	if (!iov_iter_count(iter))
298 		return 0;
299 
300 	ret = ovl_real_fdget(file, &real);
301 	if (ret)
302 		return ret;
303 
304 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
305 	if (is_sync_kiocb(iocb)) {
306 		ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
307 				    ovl_iocb_to_rwf(iocb->ki_flags));
308 	} else {
309 		struct ovl_aio_req *aio_req;
310 
311 		ret = -ENOMEM;
312 		aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
313 		if (!aio_req)
314 			goto out;
315 
316 		aio_req->fd = real;
317 		real.flags = 0;
318 		aio_req->orig_iocb = iocb;
319 		kiocb_clone(&aio_req->iocb, iocb, real.file);
320 		aio_req->iocb.ki_complete = ovl_aio_rw_complete;
321 		ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
322 		if (ret != -EIOCBQUEUED)
323 			ovl_aio_cleanup_handler(aio_req);
324 	}
325 out:
326 	revert_creds(old_cred);
327 	ovl_file_accessed(file);
328 
329 	fdput(real);
330 
331 	return ret;
332 }
333 
334 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
335 {
336 	struct file *file = iocb->ki_filp;
337 	struct inode *inode = file_inode(file);
338 	struct fd real;
339 	const struct cred *old_cred;
340 	ssize_t ret;
341 	int ifl = iocb->ki_flags;
342 
343 	if (!iov_iter_count(iter))
344 		return 0;
345 
346 	inode_lock(inode);
347 	/* Update mode */
348 	ovl_copyattr(ovl_inode_real(inode), inode);
349 	ret = file_remove_privs(file);
350 	if (ret)
351 		goto out_unlock;
352 
353 	ret = ovl_real_fdget(file, &real);
354 	if (ret)
355 		goto out_unlock;
356 
357 	if (!ovl_should_sync(OVL_FS(inode->i_sb)))
358 		ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
359 
360 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
361 	if (is_sync_kiocb(iocb)) {
362 		file_start_write(real.file);
363 		ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
364 				     ovl_iocb_to_rwf(ifl));
365 		file_end_write(real.file);
366 		/* Update size */
367 		ovl_copyattr(ovl_inode_real(inode), inode);
368 	} else {
369 		struct ovl_aio_req *aio_req;
370 
371 		ret = -ENOMEM;
372 		aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
373 		if (!aio_req)
374 			goto out;
375 
376 		file_start_write(real.file);
377 		/* Pacify lockdep, same trick as done in aio_write() */
378 		__sb_writers_release(file_inode(real.file)->i_sb,
379 				     SB_FREEZE_WRITE);
380 		aio_req->fd = real;
381 		real.flags = 0;
382 		aio_req->orig_iocb = iocb;
383 		kiocb_clone(&aio_req->iocb, iocb, real.file);
384 		aio_req->iocb.ki_flags = ifl;
385 		aio_req->iocb.ki_complete = ovl_aio_rw_complete;
386 		ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
387 		if (ret != -EIOCBQUEUED)
388 			ovl_aio_cleanup_handler(aio_req);
389 	}
390 out:
391 	revert_creds(old_cred);
392 	fdput(real);
393 
394 out_unlock:
395 	inode_unlock(inode);
396 
397 	return ret;
398 }
399 
400 static ssize_t ovl_splice_read(struct file *in, loff_t *ppos,
401 			 struct pipe_inode_info *pipe, size_t len,
402 			 unsigned int flags)
403 {
404 	ssize_t ret;
405 	struct fd real;
406 	const struct cred *old_cred;
407 
408 	ret = ovl_real_fdget(in, &real);
409 	if (ret)
410 		return ret;
411 
412 	old_cred = ovl_override_creds(file_inode(in)->i_sb);
413 	ret = generic_file_splice_read(real.file, ppos, pipe, len, flags);
414 	revert_creds(old_cred);
415 
416 	ovl_file_accessed(in);
417 	fdput(real);
418 	return ret;
419 }
420 
421 static ssize_t
422 ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
423 			  loff_t *ppos, size_t len, unsigned int flags)
424 {
425 	struct fd real;
426 	const struct cred *old_cred;
427 	ssize_t ret;
428 
429 	ret = ovl_real_fdget(out, &real);
430 	if (ret)
431 		return ret;
432 
433 	old_cred = ovl_override_creds(file_inode(out)->i_sb);
434 	ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
435 	revert_creds(old_cred);
436 
437 	ovl_file_accessed(out);
438 	fdput(real);
439 	return ret;
440 }
441 
442 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
443 {
444 	struct fd real;
445 	const struct cred *old_cred;
446 	int ret;
447 
448 	if (!ovl_should_sync(OVL_FS(file_inode(file)->i_sb)))
449 		return 0;
450 
451 	ret = ovl_real_fdget_meta(file, &real, !datasync);
452 	if (ret)
453 		return ret;
454 
455 	/* Don't sync lower file for fear of receiving EROFS error */
456 	if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
457 		old_cred = ovl_override_creds(file_inode(file)->i_sb);
458 		ret = vfs_fsync_range(real.file, start, end, datasync);
459 		revert_creds(old_cred);
460 	}
461 
462 	fdput(real);
463 
464 	return ret;
465 }
466 
467 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
468 {
469 	struct file *realfile = file->private_data;
470 	const struct cred *old_cred;
471 	int ret;
472 
473 	if (!realfile->f_op->mmap)
474 		return -ENODEV;
475 
476 	if (WARN_ON(file != vma->vm_file))
477 		return -EIO;
478 
479 	vma->vm_file = get_file(realfile);
480 
481 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
482 	ret = call_mmap(vma->vm_file, vma);
483 	revert_creds(old_cred);
484 
485 	if (ret) {
486 		/* Drop reference count from new vm_file value */
487 		fput(realfile);
488 	} else {
489 		/* Drop reference count from previous vm_file value */
490 		fput(file);
491 	}
492 
493 	ovl_file_accessed(file);
494 
495 	return ret;
496 }
497 
498 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
499 {
500 	struct inode *inode = file_inode(file);
501 	struct fd real;
502 	const struct cred *old_cred;
503 	int ret;
504 
505 	ret = ovl_real_fdget(file, &real);
506 	if (ret)
507 		return ret;
508 
509 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
510 	ret = vfs_fallocate(real.file, mode, offset, len);
511 	revert_creds(old_cred);
512 
513 	/* Update size */
514 	ovl_copyattr(ovl_inode_real(inode), inode);
515 
516 	fdput(real);
517 
518 	return ret;
519 }
520 
521 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
522 {
523 	struct fd real;
524 	const struct cred *old_cred;
525 	int ret;
526 
527 	ret = ovl_real_fdget(file, &real);
528 	if (ret)
529 		return ret;
530 
531 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
532 	ret = vfs_fadvise(real.file, offset, len, advice);
533 	revert_creds(old_cred);
534 
535 	fdput(real);
536 
537 	return ret;
538 }
539 
540 static long ovl_real_ioctl(struct file *file, unsigned int cmd,
541 			   unsigned long arg)
542 {
543 	struct fd real;
544 	const struct cred *old_cred;
545 	long ret;
546 
547 	ret = ovl_real_fdget(file, &real);
548 	if (ret)
549 		return ret;
550 
551 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
552 	ret = security_file_ioctl(real.file, cmd, arg);
553 	if (!ret)
554 		ret = vfs_ioctl(real.file, cmd, arg);
555 	revert_creds(old_cred);
556 
557 	fdput(real);
558 
559 	return ret;
560 }
561 
562 static unsigned int ovl_iflags_to_fsflags(unsigned int iflags)
563 {
564 	unsigned int flags = 0;
565 
566 	if (iflags & S_SYNC)
567 		flags |= FS_SYNC_FL;
568 	if (iflags & S_APPEND)
569 		flags |= FS_APPEND_FL;
570 	if (iflags & S_IMMUTABLE)
571 		flags |= FS_IMMUTABLE_FL;
572 	if (iflags & S_NOATIME)
573 		flags |= FS_NOATIME_FL;
574 
575 	return flags;
576 }
577 
578 static long ovl_ioctl_set_flags(struct file *file, unsigned int cmd,
579 				unsigned long arg, unsigned int flags)
580 {
581 	long ret;
582 	struct inode *inode = file_inode(file);
583 	unsigned int oldflags;
584 
585 	if (!inode_owner_or_capable(inode))
586 		return -EACCES;
587 
588 	ret = mnt_want_write_file(file);
589 	if (ret)
590 		return ret;
591 
592 	inode_lock(inode);
593 
594 	/* Check the capability before cred override */
595 	oldflags = ovl_iflags_to_fsflags(READ_ONCE(inode->i_flags));
596 	ret = vfs_ioc_setflags_prepare(inode, oldflags, flags);
597 	if (ret)
598 		goto unlock;
599 
600 	ret = ovl_maybe_copy_up(file_dentry(file), O_WRONLY);
601 	if (ret)
602 		goto unlock;
603 
604 	ret = ovl_real_ioctl(file, cmd, arg);
605 
606 	ovl_copyflags(ovl_inode_real(inode), inode);
607 unlock:
608 	inode_unlock(inode);
609 
610 	mnt_drop_write_file(file);
611 
612 	return ret;
613 
614 }
615 
616 static long ovl_ioctl_set_fsflags(struct file *file, unsigned int cmd,
617 				  unsigned long arg)
618 {
619 	unsigned int flags;
620 
621 	if (get_user(flags, (int __user *) arg))
622 		return -EFAULT;
623 
624 	return ovl_ioctl_set_flags(file, cmd, arg, flags);
625 }
626 
627 static unsigned int ovl_fsxflags_to_fsflags(unsigned int xflags)
628 {
629 	unsigned int flags = 0;
630 
631 	if (xflags & FS_XFLAG_SYNC)
632 		flags |= FS_SYNC_FL;
633 	if (xflags & FS_XFLAG_APPEND)
634 		flags |= FS_APPEND_FL;
635 	if (xflags & FS_XFLAG_IMMUTABLE)
636 		flags |= FS_IMMUTABLE_FL;
637 	if (xflags & FS_XFLAG_NOATIME)
638 		flags |= FS_NOATIME_FL;
639 
640 	return flags;
641 }
642 
643 static long ovl_ioctl_set_fsxflags(struct file *file, unsigned int cmd,
644 				   unsigned long arg)
645 {
646 	struct fsxattr fa;
647 
648 	memset(&fa, 0, sizeof(fa));
649 	if (copy_from_user(&fa, (void __user *) arg, sizeof(fa)))
650 		return -EFAULT;
651 
652 	return ovl_ioctl_set_flags(file, cmd, arg,
653 				   ovl_fsxflags_to_fsflags(fa.fsx_xflags));
654 }
655 
656 long ovl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
657 {
658 	long ret;
659 
660 	switch (cmd) {
661 	case FS_IOC_GETFLAGS:
662 	case FS_IOC_FSGETXATTR:
663 		ret = ovl_real_ioctl(file, cmd, arg);
664 		break;
665 
666 	case FS_IOC_SETFLAGS:
667 		ret = ovl_ioctl_set_fsflags(file, cmd, arg);
668 		break;
669 
670 	case FS_IOC_FSSETXATTR:
671 		ret = ovl_ioctl_set_fsxflags(file, cmd, arg);
672 		break;
673 
674 	default:
675 		ret = -ENOTTY;
676 	}
677 
678 	return ret;
679 }
680 
681 #ifdef CONFIG_COMPAT
682 long ovl_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
683 {
684 	switch (cmd) {
685 	case FS_IOC32_GETFLAGS:
686 		cmd = FS_IOC_GETFLAGS;
687 		break;
688 
689 	case FS_IOC32_SETFLAGS:
690 		cmd = FS_IOC_SETFLAGS;
691 		break;
692 
693 	default:
694 		return -ENOIOCTLCMD;
695 	}
696 
697 	return ovl_ioctl(file, cmd, arg);
698 }
699 #endif
700 
701 enum ovl_copyop {
702 	OVL_COPY,
703 	OVL_CLONE,
704 	OVL_DEDUPE,
705 };
706 
707 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
708 			    struct file *file_out, loff_t pos_out,
709 			    loff_t len, unsigned int flags, enum ovl_copyop op)
710 {
711 	struct inode *inode_out = file_inode(file_out);
712 	struct fd real_in, real_out;
713 	const struct cred *old_cred;
714 	loff_t ret;
715 
716 	ret = ovl_real_fdget(file_out, &real_out);
717 	if (ret)
718 		return ret;
719 
720 	ret = ovl_real_fdget(file_in, &real_in);
721 	if (ret) {
722 		fdput(real_out);
723 		return ret;
724 	}
725 
726 	old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
727 	switch (op) {
728 	case OVL_COPY:
729 		ret = vfs_copy_file_range(real_in.file, pos_in,
730 					  real_out.file, pos_out, len, flags);
731 		break;
732 
733 	case OVL_CLONE:
734 		ret = vfs_clone_file_range(real_in.file, pos_in,
735 					   real_out.file, pos_out, len, flags);
736 		break;
737 
738 	case OVL_DEDUPE:
739 		ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
740 						real_out.file, pos_out, len,
741 						flags);
742 		break;
743 	}
744 	revert_creds(old_cred);
745 
746 	/* Update size */
747 	ovl_copyattr(ovl_inode_real(inode_out), inode_out);
748 
749 	fdput(real_in);
750 	fdput(real_out);
751 
752 	return ret;
753 }
754 
755 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
756 				   struct file *file_out, loff_t pos_out,
757 				   size_t len, unsigned int flags)
758 {
759 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
760 			    OVL_COPY);
761 }
762 
763 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
764 				   struct file *file_out, loff_t pos_out,
765 				   loff_t len, unsigned int remap_flags)
766 {
767 	enum ovl_copyop op;
768 
769 	if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
770 		return -EINVAL;
771 
772 	if (remap_flags & REMAP_FILE_DEDUP)
773 		op = OVL_DEDUPE;
774 	else
775 		op = OVL_CLONE;
776 
777 	/*
778 	 * Don't copy up because of a dedupe request, this wouldn't make sense
779 	 * most of the time (data would be duplicated instead of deduplicated).
780 	 */
781 	if (op == OVL_DEDUPE &&
782 	    (!ovl_inode_upper(file_inode(file_in)) ||
783 	     !ovl_inode_upper(file_inode(file_out))))
784 		return -EPERM;
785 
786 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
787 			    remap_flags, op);
788 }
789 
790 const struct file_operations ovl_file_operations = {
791 	.open		= ovl_open,
792 	.release	= ovl_release,
793 	.llseek		= ovl_llseek,
794 	.read_iter	= ovl_read_iter,
795 	.write_iter	= ovl_write_iter,
796 	.fsync		= ovl_fsync,
797 	.mmap		= ovl_mmap,
798 	.fallocate	= ovl_fallocate,
799 	.fadvise	= ovl_fadvise,
800 	.unlocked_ioctl	= ovl_ioctl,
801 #ifdef CONFIG_COMPAT
802 	.compat_ioctl	= ovl_compat_ioctl,
803 #endif
804 	.splice_read    = ovl_splice_read,
805 	.splice_write   = ovl_splice_write,
806 
807 	.copy_file_range	= ovl_copy_file_range,
808 	.remap_file_range	= ovl_remap_file_range,
809 };
810 
811 int __init ovl_aio_request_cache_init(void)
812 {
813 	ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
814 						   sizeof(struct ovl_aio_req),
815 						   0, SLAB_HWCACHE_ALIGN, NULL);
816 	if (!ovl_aio_request_cachep)
817 		return -ENOMEM;
818 
819 	return 0;
820 }
821 
822 void ovl_aio_request_cache_destroy(void)
823 {
824 	kmem_cache_destroy(ovl_aio_request_cachep);
825 }
826