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