xref: /openbmc/linux/fs/overlayfs/file.c (revision 6f4eaea2)
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(&init_user_ns, realinode, MAY_OPEN | acc_mode);
54 	if (err) {
55 		realfile = ERR_PTR(err);
56 	} else {
57 		if (!inode_owner_or_capable(&init_user_ns, realinode))
58 			flags &= ~O_NOATIME;
59 
60 		realfile = open_with_fake_path(&file->f_path, flags, realinode,
61 					       current_cred());
62 	}
63 	revert_creds(old_cred);
64 
65 	pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
66 		 file, file, ovl_whatisit(inode, realinode), file->f_flags,
67 		 realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
68 
69 	return realfile;
70 }
71 
72 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
73 
74 static int ovl_change_flags(struct file *file, unsigned int flags)
75 {
76 	struct inode *inode = file_inode(file);
77 	int err;
78 
79 	flags &= OVL_SETFL_MASK;
80 
81 	if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
82 		return -EPERM;
83 
84 	if (flags & O_DIRECT) {
85 		if (!file->f_mapping->a_ops ||
86 		    !file->f_mapping->a_ops->direct_IO)
87 			return -EINVAL;
88 	}
89 
90 	if (file->f_op->check_flags) {
91 		err = file->f_op->check_flags(flags);
92 		if (err)
93 			return err;
94 	}
95 
96 	spin_lock(&file->f_lock);
97 	file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
98 	spin_unlock(&file->f_lock);
99 
100 	return 0;
101 }
102 
103 static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
104 			       bool allow_meta)
105 {
106 	struct inode *inode = file_inode(file);
107 	struct inode *realinode;
108 
109 	real->flags = 0;
110 	real->file = file->private_data;
111 
112 	if (allow_meta)
113 		realinode = ovl_inode_real(inode);
114 	else
115 		realinode = ovl_inode_realdata(inode);
116 
117 	/* Has it been copied up since we'd opened it? */
118 	if (unlikely(file_inode(real->file) != realinode)) {
119 		real->flags = FDPUT_FPUT;
120 		real->file = ovl_open_realfile(file, realinode);
121 
122 		return PTR_ERR_OR_ZERO(real->file);
123 	}
124 
125 	/* Did the flags change since open? */
126 	if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
127 		return ovl_change_flags(real->file, file->f_flags);
128 
129 	return 0;
130 }
131 
132 static int ovl_real_fdget(const struct file *file, struct fd *real)
133 {
134 	if (d_is_dir(file_dentry(file))) {
135 		real->flags = 0;
136 		real->file = ovl_dir_real_file(file, false);
137 
138 		return PTR_ERR_OR_ZERO(real->file);
139 	}
140 
141 	return ovl_real_fdget_meta(file, real, false);
142 }
143 
144 static int ovl_open(struct inode *inode, struct file *file)
145 {
146 	struct file *realfile;
147 	int err;
148 
149 	err = ovl_maybe_copy_up(file_dentry(file), file->f_flags);
150 	if (err)
151 		return err;
152 
153 	/* No longer need these flags, so don't pass them on to underlying fs */
154 	file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
155 
156 	realfile = ovl_open_realfile(file, ovl_inode_realdata(inode));
157 	if (IS_ERR(realfile))
158 		return PTR_ERR(realfile);
159 
160 	file->private_data = realfile;
161 
162 	return 0;
163 }
164 
165 static int ovl_release(struct inode *inode, struct file *file)
166 {
167 	fput(file->private_data);
168 
169 	return 0;
170 }
171 
172 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
173 {
174 	struct inode *inode = file_inode(file);
175 	struct fd real;
176 	const struct cred *old_cred;
177 	loff_t ret;
178 
179 	/*
180 	 * The two special cases below do not need to involve real fs,
181 	 * so we can optimizing concurrent callers.
182 	 */
183 	if (offset == 0) {
184 		if (whence == SEEK_CUR)
185 			return file->f_pos;
186 
187 		if (whence == SEEK_SET)
188 			return vfs_setpos(file, 0, 0);
189 	}
190 
191 	ret = ovl_real_fdget(file, &real);
192 	if (ret)
193 		return ret;
194 
195 	/*
196 	 * Overlay file f_pos is the master copy that is preserved
197 	 * through copy up and modified on read/write, but only real
198 	 * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
199 	 * limitations that are more strict than ->s_maxbytes for specific
200 	 * files, so we use the real file to perform seeks.
201 	 */
202 	ovl_inode_lock(inode);
203 	real.file->f_pos = file->f_pos;
204 
205 	old_cred = ovl_override_creds(inode->i_sb);
206 	ret = vfs_llseek(real.file, offset, whence);
207 	revert_creds(old_cred);
208 
209 	file->f_pos = real.file->f_pos;
210 	ovl_inode_unlock(inode);
211 
212 	fdput(real);
213 
214 	return ret;
215 }
216 
217 static void ovl_file_accessed(struct file *file)
218 {
219 	struct inode *inode, *upperinode;
220 
221 	if (file->f_flags & O_NOATIME)
222 		return;
223 
224 	inode = file_inode(file);
225 	upperinode = ovl_inode_upper(inode);
226 
227 	if (!upperinode)
228 		return;
229 
230 	if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
231 	     !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) {
232 		inode->i_mtime = upperinode->i_mtime;
233 		inode->i_ctime = upperinode->i_ctime;
234 	}
235 
236 	touch_atime(&file->f_path);
237 }
238 
239 static rwf_t ovl_iocb_to_rwf(int ifl)
240 {
241 	rwf_t flags = 0;
242 
243 	if (ifl & IOCB_NOWAIT)
244 		flags |= RWF_NOWAIT;
245 	if (ifl & IOCB_HIPRI)
246 		flags |= RWF_HIPRI;
247 	if (ifl & IOCB_DSYNC)
248 		flags |= RWF_DSYNC;
249 	if (ifl & IOCB_SYNC)
250 		flags |= RWF_SYNC;
251 
252 	return flags;
253 }
254 
255 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
256 {
257 	struct kiocb *iocb = &aio_req->iocb;
258 	struct kiocb *orig_iocb = aio_req->orig_iocb;
259 
260 	if (iocb->ki_flags & IOCB_WRITE) {
261 		struct inode *inode = file_inode(orig_iocb->ki_filp);
262 
263 		/* Actually acquired in ovl_write_iter() */
264 		__sb_writers_acquired(file_inode(iocb->ki_filp)->i_sb,
265 				      SB_FREEZE_WRITE);
266 		file_end_write(iocb->ki_filp);
267 		ovl_copyattr(ovl_inode_real(inode), inode);
268 	}
269 
270 	orig_iocb->ki_pos = iocb->ki_pos;
271 	fdput(aio_req->fd);
272 	kmem_cache_free(ovl_aio_request_cachep, aio_req);
273 }
274 
275 static void ovl_aio_rw_complete(struct kiocb *iocb, long res, long res2)
276 {
277 	struct ovl_aio_req *aio_req = container_of(iocb,
278 						   struct ovl_aio_req, iocb);
279 	struct kiocb *orig_iocb = aio_req->orig_iocb;
280 
281 	ovl_aio_cleanup_handler(aio_req);
282 	orig_iocb->ki_complete(orig_iocb, res, res2);
283 }
284 
285 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
286 {
287 	struct file *file = iocb->ki_filp;
288 	struct fd real;
289 	const struct cred *old_cred;
290 	ssize_t ret;
291 
292 	if (!iov_iter_count(iter))
293 		return 0;
294 
295 	ret = ovl_real_fdget(file, &real);
296 	if (ret)
297 		return ret;
298 
299 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
300 	if (is_sync_kiocb(iocb)) {
301 		ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
302 				    ovl_iocb_to_rwf(iocb->ki_flags));
303 	} else {
304 		struct ovl_aio_req *aio_req;
305 
306 		ret = -ENOMEM;
307 		aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
308 		if (!aio_req)
309 			goto out;
310 
311 		aio_req->fd = real;
312 		real.flags = 0;
313 		aio_req->orig_iocb = iocb;
314 		kiocb_clone(&aio_req->iocb, iocb, real.file);
315 		aio_req->iocb.ki_complete = ovl_aio_rw_complete;
316 		ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
317 		if (ret != -EIOCBQUEUED)
318 			ovl_aio_cleanup_handler(aio_req);
319 	}
320 out:
321 	revert_creds(old_cred);
322 	ovl_file_accessed(file);
323 
324 	fdput(real);
325 
326 	return ret;
327 }
328 
329 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
330 {
331 	struct file *file = iocb->ki_filp;
332 	struct inode *inode = file_inode(file);
333 	struct fd real;
334 	const struct cred *old_cred;
335 	ssize_t ret;
336 	int ifl = iocb->ki_flags;
337 
338 	if (!iov_iter_count(iter))
339 		return 0;
340 
341 	inode_lock(inode);
342 	/* Update mode */
343 	ovl_copyattr(ovl_inode_real(inode), inode);
344 	ret = file_remove_privs(file);
345 	if (ret)
346 		goto out_unlock;
347 
348 	ret = ovl_real_fdget(file, &real);
349 	if (ret)
350 		goto out_unlock;
351 
352 	if (!ovl_should_sync(OVL_FS(inode->i_sb)))
353 		ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
354 
355 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
356 	if (is_sync_kiocb(iocb)) {
357 		file_start_write(real.file);
358 		ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
359 				     ovl_iocb_to_rwf(ifl));
360 		file_end_write(real.file);
361 		/* Update size */
362 		ovl_copyattr(ovl_inode_real(inode), inode);
363 	} else {
364 		struct ovl_aio_req *aio_req;
365 
366 		ret = -ENOMEM;
367 		aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
368 		if (!aio_req)
369 			goto out;
370 
371 		file_start_write(real.file);
372 		/* Pacify lockdep, same trick as done in aio_write() */
373 		__sb_writers_release(file_inode(real.file)->i_sb,
374 				     SB_FREEZE_WRITE);
375 		aio_req->fd = real;
376 		real.flags = 0;
377 		aio_req->orig_iocb = iocb;
378 		kiocb_clone(&aio_req->iocb, iocb, real.file);
379 		aio_req->iocb.ki_flags = ifl;
380 		aio_req->iocb.ki_complete = ovl_aio_rw_complete;
381 		ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
382 		if (ret != -EIOCBQUEUED)
383 			ovl_aio_cleanup_handler(aio_req);
384 	}
385 out:
386 	revert_creds(old_cred);
387 	fdput(real);
388 
389 out_unlock:
390 	inode_unlock(inode);
391 
392 	return ret;
393 }
394 
395 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
396 {
397 	struct fd real;
398 	const struct cred *old_cred;
399 	int ret;
400 
401 	ret = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
402 	if (ret <= 0)
403 		return ret;
404 
405 	ret = ovl_real_fdget_meta(file, &real, !datasync);
406 	if (ret)
407 		return ret;
408 
409 	/* Don't sync lower file for fear of receiving EROFS error */
410 	if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
411 		old_cred = ovl_override_creds(file_inode(file)->i_sb);
412 		ret = vfs_fsync_range(real.file, start, end, datasync);
413 		revert_creds(old_cred);
414 	}
415 
416 	fdput(real);
417 
418 	return ret;
419 }
420 
421 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
422 {
423 	struct file *realfile = file->private_data;
424 	const struct cred *old_cred;
425 	int ret;
426 
427 	if (!realfile->f_op->mmap)
428 		return -ENODEV;
429 
430 	if (WARN_ON(file != vma->vm_file))
431 		return -EIO;
432 
433 	vma->vm_file = get_file(realfile);
434 
435 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
436 	ret = call_mmap(vma->vm_file, vma);
437 	revert_creds(old_cred);
438 
439 	if (ret) {
440 		/* Drop reference count from new vm_file value */
441 		fput(realfile);
442 	} else {
443 		/* Drop reference count from previous vm_file value */
444 		fput(file);
445 	}
446 
447 	ovl_file_accessed(file);
448 
449 	return ret;
450 }
451 
452 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
453 {
454 	struct inode *inode = file_inode(file);
455 	struct fd real;
456 	const struct cred *old_cred;
457 	int ret;
458 
459 	ret = ovl_real_fdget(file, &real);
460 	if (ret)
461 		return ret;
462 
463 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
464 	ret = vfs_fallocate(real.file, mode, offset, len);
465 	revert_creds(old_cred);
466 
467 	/* Update size */
468 	ovl_copyattr(ovl_inode_real(inode), inode);
469 
470 	fdput(real);
471 
472 	return ret;
473 }
474 
475 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
476 {
477 	struct fd real;
478 	const struct cred *old_cred;
479 	int ret;
480 
481 	ret = ovl_real_fdget(file, &real);
482 	if (ret)
483 		return ret;
484 
485 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
486 	ret = vfs_fadvise(real.file, offset, len, advice);
487 	revert_creds(old_cred);
488 
489 	fdput(real);
490 
491 	return ret;
492 }
493 
494 static long ovl_real_ioctl(struct file *file, unsigned int cmd,
495 			   unsigned long arg)
496 {
497 	struct fd real;
498 	long ret;
499 
500 	ret = ovl_real_fdget(file, &real);
501 	if (ret)
502 		return ret;
503 
504 	ret = security_file_ioctl(real.file, cmd, arg);
505 	if (!ret) {
506 		/*
507 		 * Don't override creds, since we currently can't safely check
508 		 * permissions before doing so.
509 		 */
510 		ret = vfs_ioctl(real.file, cmd, arg);
511 	}
512 
513 	fdput(real);
514 
515 	return ret;
516 }
517 
518 static long ovl_ioctl_set_flags(struct file *file, unsigned int cmd,
519 				unsigned long arg)
520 {
521 	long ret;
522 	struct inode *inode = file_inode(file);
523 
524 	if (!inode_owner_or_capable(&init_user_ns, inode))
525 		return -EACCES;
526 
527 	ret = mnt_want_write_file(file);
528 	if (ret)
529 		return ret;
530 
531 	inode_lock(inode);
532 
533 	/*
534 	 * Prevent copy up if immutable and has no CAP_LINUX_IMMUTABLE
535 	 * capability.
536 	 */
537 	ret = -EPERM;
538 	if (!ovl_has_upperdata(inode) && IS_IMMUTABLE(inode) &&
539 	    !capable(CAP_LINUX_IMMUTABLE))
540 		goto unlock;
541 
542 	ret = ovl_maybe_copy_up(file_dentry(file), O_WRONLY);
543 	if (ret)
544 		goto unlock;
545 
546 	ret = ovl_real_ioctl(file, cmd, arg);
547 
548 	ovl_copyflags(ovl_inode_real(inode), inode);
549 unlock:
550 	inode_unlock(inode);
551 
552 	mnt_drop_write_file(file);
553 
554 	return ret;
555 
556 }
557 
558 long ovl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
559 {
560 	long ret;
561 
562 	switch (cmd) {
563 	case FS_IOC_GETFLAGS:
564 	case FS_IOC_FSGETXATTR:
565 		ret = ovl_real_ioctl(file, cmd, arg);
566 		break;
567 
568 	case FS_IOC_FSSETXATTR:
569 	case FS_IOC_SETFLAGS:
570 		ret = ovl_ioctl_set_flags(file, cmd, arg);
571 		break;
572 
573 	default:
574 		ret = -ENOTTY;
575 	}
576 
577 	return ret;
578 }
579 
580 #ifdef CONFIG_COMPAT
581 long ovl_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
582 {
583 	switch (cmd) {
584 	case FS_IOC32_GETFLAGS:
585 		cmd = FS_IOC_GETFLAGS;
586 		break;
587 
588 	case FS_IOC32_SETFLAGS:
589 		cmd = FS_IOC_SETFLAGS;
590 		break;
591 
592 	default:
593 		return -ENOIOCTLCMD;
594 	}
595 
596 	return ovl_ioctl(file, cmd, arg);
597 }
598 #endif
599 
600 enum ovl_copyop {
601 	OVL_COPY,
602 	OVL_CLONE,
603 	OVL_DEDUPE,
604 };
605 
606 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
607 			    struct file *file_out, loff_t pos_out,
608 			    loff_t len, unsigned int flags, enum ovl_copyop op)
609 {
610 	struct inode *inode_out = file_inode(file_out);
611 	struct fd real_in, real_out;
612 	const struct cred *old_cred;
613 	loff_t ret;
614 
615 	ret = ovl_real_fdget(file_out, &real_out);
616 	if (ret)
617 		return ret;
618 
619 	ret = ovl_real_fdget(file_in, &real_in);
620 	if (ret) {
621 		fdput(real_out);
622 		return ret;
623 	}
624 
625 	old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
626 	switch (op) {
627 	case OVL_COPY:
628 		ret = vfs_copy_file_range(real_in.file, pos_in,
629 					  real_out.file, pos_out, len, flags);
630 		break;
631 
632 	case OVL_CLONE:
633 		ret = vfs_clone_file_range(real_in.file, pos_in,
634 					   real_out.file, pos_out, len, flags);
635 		break;
636 
637 	case OVL_DEDUPE:
638 		ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
639 						real_out.file, pos_out, len,
640 						flags);
641 		break;
642 	}
643 	revert_creds(old_cred);
644 
645 	/* Update size */
646 	ovl_copyattr(ovl_inode_real(inode_out), inode_out);
647 
648 	fdput(real_in);
649 	fdput(real_out);
650 
651 	return ret;
652 }
653 
654 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
655 				   struct file *file_out, loff_t pos_out,
656 				   size_t len, unsigned int flags)
657 {
658 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
659 			    OVL_COPY);
660 }
661 
662 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
663 				   struct file *file_out, loff_t pos_out,
664 				   loff_t len, unsigned int remap_flags)
665 {
666 	enum ovl_copyop op;
667 
668 	if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
669 		return -EINVAL;
670 
671 	if (remap_flags & REMAP_FILE_DEDUP)
672 		op = OVL_DEDUPE;
673 	else
674 		op = OVL_CLONE;
675 
676 	/*
677 	 * Don't copy up because of a dedupe request, this wouldn't make sense
678 	 * most of the time (data would be duplicated instead of deduplicated).
679 	 */
680 	if (op == OVL_DEDUPE &&
681 	    (!ovl_inode_upper(file_inode(file_in)) ||
682 	     !ovl_inode_upper(file_inode(file_out))))
683 		return -EPERM;
684 
685 	return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
686 			    remap_flags, op);
687 }
688 
689 const struct file_operations ovl_file_operations = {
690 	.open		= ovl_open,
691 	.release	= ovl_release,
692 	.llseek		= ovl_llseek,
693 	.read_iter	= ovl_read_iter,
694 	.write_iter	= ovl_write_iter,
695 	.fsync		= ovl_fsync,
696 	.mmap		= ovl_mmap,
697 	.fallocate	= ovl_fallocate,
698 	.fadvise	= ovl_fadvise,
699 	.unlocked_ioctl	= ovl_ioctl,
700 #ifdef CONFIG_COMPAT
701 	.compat_ioctl	= ovl_compat_ioctl,
702 #endif
703 	.splice_read    = generic_file_splice_read,
704 	.splice_write   = iter_file_splice_write,
705 
706 	.copy_file_range	= ovl_copy_file_range,
707 	.remap_file_range	= ovl_remap_file_range,
708 };
709 
710 int __init ovl_aio_request_cache_init(void)
711 {
712 	ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
713 						   sizeof(struct ovl_aio_req),
714 						   0, SLAB_HWCACHE_ALIGN, NULL);
715 	if (!ovl_aio_request_cachep)
716 		return -ENOMEM;
717 
718 	return 0;
719 }
720 
721 void ovl_aio_request_cache_destroy(void)
722 {
723 	kmem_cache_destroy(ovl_aio_request_cachep);
724 }
725