Loading...
Note: File does not exist in v4.6.
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
18struct ovl_aio_req {
19 struct kiocb iocb;
20 refcount_t ref;
21 struct kiocb *orig_iocb;
22 struct fd fd;
23};
24
25static struct kmem_cache *ovl_aio_request_cachep;
26
27static char ovl_whatisit(struct inode *inode, struct inode *realinode)
28{
29 if (realinode != ovl_inode_upper(inode))
30 return 'l';
31 if (ovl_has_upperdata(inode))
32 return 'u';
33 else
34 return 'm';
35}
36
37/* No atime modification nor notify on underlying */
38#define OVL_OPEN_FLAGS (O_NOATIME | FMODE_NONOTIFY)
39
40static struct file *ovl_open_realfile(const struct file *file,
41 const struct path *realpath)
42{
43 struct inode *realinode = d_inode(realpath->dentry);
44 struct inode *inode = file_inode(file);
45 struct user_namespace *real_mnt_userns;
46 struct file *realfile;
47 const struct cred *old_cred;
48 int flags = file->f_flags | OVL_OPEN_FLAGS;
49 int acc_mode = ACC_MODE(flags);
50 int err;
51
52 if (flags & O_APPEND)
53 acc_mode |= MAY_APPEND;
54
55 old_cred = ovl_override_creds(inode->i_sb);
56 real_mnt_userns = mnt_user_ns(realpath->mnt);
57 err = inode_permission(real_mnt_userns, realinode, MAY_OPEN | acc_mode);
58 if (err) {
59 realfile = ERR_PTR(err);
60 } else {
61 if (!inode_owner_or_capable(real_mnt_userns, realinode))
62 flags &= ~O_NOATIME;
63
64 realfile = open_with_fake_path(&file->f_path, flags, realinode,
65 current_cred());
66 }
67 revert_creds(old_cred);
68
69 pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
70 file, file, ovl_whatisit(inode, realinode), file->f_flags,
71 realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
72
73 return realfile;
74}
75
76#define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
77
78static int ovl_change_flags(struct file *file, unsigned int flags)
79{
80 struct inode *inode = file_inode(file);
81 int err;
82
83 flags &= OVL_SETFL_MASK;
84
85 if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
86 return -EPERM;
87
88 if ((flags & O_DIRECT) && !(file->f_mode & FMODE_CAN_ODIRECT))
89 return -EINVAL;
90
91 if (file->f_op->check_flags) {
92 err = file->f_op->check_flags(flags);
93 if (err)
94 return err;
95 }
96
97 spin_lock(&file->f_lock);
98 file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
99 file->f_iocb_flags = iocb_flags(file);
100 spin_unlock(&file->f_lock);
101
102 return 0;
103}
104
105static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
106 bool allow_meta)
107{
108 struct dentry *dentry = file_dentry(file);
109 struct path realpath;
110
111 real->flags = 0;
112 real->file = file->private_data;
113
114 if (allow_meta)
115 ovl_path_real(dentry, &realpath);
116 else
117 ovl_path_realdata(dentry, &realpath);
118
119 /* Has it been copied up since we'd opened it? */
120 if (unlikely(file_inode(real->file) != d_inode(realpath.dentry))) {
121 real->flags = FDPUT_FPUT;
122 real->file = ovl_open_realfile(file, &realpath);
123
124 return PTR_ERR_OR_ZERO(real->file);
125 }
126
127 /* Did the flags change since open? */
128 if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
129 return ovl_change_flags(real->file, file->f_flags);
130
131 return 0;
132}
133
134static int ovl_real_fdget(const struct file *file, struct fd *real)
135{
136 if (d_is_dir(file_dentry(file))) {
137 real->flags = 0;
138 real->file = ovl_dir_real_file(file, false);
139
140 return PTR_ERR_OR_ZERO(real->file);
141 }
142
143 return ovl_real_fdget_meta(file, real, false);
144}
145
146static int ovl_open(struct inode *inode, struct file *file)
147{
148 struct dentry *dentry = file_dentry(file);
149 struct file *realfile;
150 struct path realpath;
151 int err;
152
153 err = ovl_maybe_copy_up(dentry, file->f_flags);
154 if (err)
155 return err;
156
157 /* No longer need these flags, so don't pass them on to underlying fs */
158 file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
159
160 ovl_path_realdata(dentry, &realpath);
161 realfile = ovl_open_realfile(file, &realpath);
162 if (IS_ERR(realfile))
163 return PTR_ERR(realfile);
164
165 file->private_data = realfile;
166
167 return 0;
168}
169
170static int ovl_release(struct inode *inode, struct file *file)
171{
172 fput(file->private_data);
173
174 return 0;
175}
176
177static 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
222static 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
244static 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
260static inline void ovl_aio_put(struct ovl_aio_req *aio_req)
261{
262 if (refcount_dec_and_test(&aio_req->ref)) {
263 fdput(aio_req->fd);
264 kmem_cache_free(ovl_aio_request_cachep, aio_req);
265 }
266}
267
268static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
269{
270 struct kiocb *iocb = &aio_req->iocb;
271 struct kiocb *orig_iocb = aio_req->orig_iocb;
272
273 if (iocb->ki_flags & IOCB_WRITE) {
274 struct inode *inode = file_inode(orig_iocb->ki_filp);
275
276 /* Actually acquired in ovl_write_iter() */
277 __sb_writers_acquired(file_inode(iocb->ki_filp)->i_sb,
278 SB_FREEZE_WRITE);
279 file_end_write(iocb->ki_filp);
280 ovl_copyattr(inode);
281 }
282
283 orig_iocb->ki_pos = iocb->ki_pos;
284 ovl_aio_put(aio_req);
285}
286
287static void ovl_aio_rw_complete(struct kiocb *iocb, long res)
288{
289 struct ovl_aio_req *aio_req = container_of(iocb,
290 struct ovl_aio_req, iocb);
291 struct kiocb *orig_iocb = aio_req->orig_iocb;
292
293 ovl_aio_cleanup_handler(aio_req);
294 orig_iocb->ki_complete(orig_iocb, res);
295}
296
297static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
298{
299 struct file *file = iocb->ki_filp;
300 struct fd real;
301 const struct cred *old_cred;
302 ssize_t ret;
303
304 if (!iov_iter_count(iter))
305 return 0;
306
307 ret = ovl_real_fdget(file, &real);
308 if (ret)
309 return ret;
310
311 ret = -EINVAL;
312 if (iocb->ki_flags & IOCB_DIRECT &&
313 !(real.file->f_mode & FMODE_CAN_ODIRECT))
314 goto out_fdput;
315
316 old_cred = ovl_override_creds(file_inode(file)->i_sb);
317 if (is_sync_kiocb(iocb)) {
318 ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
319 ovl_iocb_to_rwf(iocb->ki_flags));
320 } else {
321 struct ovl_aio_req *aio_req;
322
323 ret = -ENOMEM;
324 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
325 if (!aio_req)
326 goto out;
327
328 aio_req->fd = real;
329 real.flags = 0;
330 aio_req->orig_iocb = iocb;
331 kiocb_clone(&aio_req->iocb, iocb, real.file);
332 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
333 refcount_set(&aio_req->ref, 2);
334 ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
335 ovl_aio_put(aio_req);
336 if (ret != -EIOCBQUEUED)
337 ovl_aio_cleanup_handler(aio_req);
338 }
339out:
340 revert_creds(old_cred);
341 ovl_file_accessed(file);
342out_fdput:
343 fdput(real);
344
345 return ret;
346}
347
348static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
349{
350 struct file *file = iocb->ki_filp;
351 struct inode *inode = file_inode(file);
352 struct fd real;
353 const struct cred *old_cred;
354 ssize_t ret;
355 int ifl = iocb->ki_flags;
356
357 if (!iov_iter_count(iter))
358 return 0;
359
360 inode_lock(inode);
361 /* Update mode */
362 ovl_copyattr(inode);
363 ret = file_remove_privs(file);
364 if (ret)
365 goto out_unlock;
366
367 ret = ovl_real_fdget(file, &real);
368 if (ret)
369 goto out_unlock;
370
371 ret = -EINVAL;
372 if (iocb->ki_flags & IOCB_DIRECT &&
373 !(real.file->f_mode & FMODE_CAN_ODIRECT))
374 goto out_fdput;
375
376 if (!ovl_should_sync(OVL_FS(inode->i_sb)))
377 ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
378
379 old_cred = ovl_override_creds(file_inode(file)->i_sb);
380 if (is_sync_kiocb(iocb)) {
381 file_start_write(real.file);
382 ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
383 ovl_iocb_to_rwf(ifl));
384 file_end_write(real.file);
385 /* Update size */
386 ovl_copyattr(inode);
387 } else {
388 struct ovl_aio_req *aio_req;
389
390 ret = -ENOMEM;
391 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
392 if (!aio_req)
393 goto out;
394
395 file_start_write(real.file);
396 /* Pacify lockdep, same trick as done in aio_write() */
397 __sb_writers_release(file_inode(real.file)->i_sb,
398 SB_FREEZE_WRITE);
399 aio_req->fd = real;
400 real.flags = 0;
401 aio_req->orig_iocb = iocb;
402 kiocb_clone(&aio_req->iocb, iocb, real.file);
403 aio_req->iocb.ki_flags = ifl;
404 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
405 refcount_set(&aio_req->ref, 2);
406 ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
407 ovl_aio_put(aio_req);
408 if (ret != -EIOCBQUEUED)
409 ovl_aio_cleanup_handler(aio_req);
410 }
411out:
412 revert_creds(old_cred);
413out_fdput:
414 fdput(real);
415
416out_unlock:
417 inode_unlock(inode);
418
419 return ret;
420}
421
422/*
423 * Calling iter_file_splice_write() directly from overlay's f_op may deadlock
424 * due to lock order inversion between pipe->mutex in iter_file_splice_write()
425 * and file_start_write(real.file) in ovl_write_iter().
426 *
427 * So do everything ovl_write_iter() does and call iter_file_splice_write() on
428 * the real file.
429 */
430static ssize_t ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
431 loff_t *ppos, size_t len, unsigned int flags)
432{
433 struct fd real;
434 const struct cred *old_cred;
435 struct inode *inode = file_inode(out);
436 ssize_t ret;
437
438 inode_lock(inode);
439 /* Update mode */
440 ovl_copyattr(inode);
441 ret = file_remove_privs(out);
442 if (ret)
443 goto out_unlock;
444
445 ret = ovl_real_fdget(out, &real);
446 if (ret)
447 goto out_unlock;
448
449 old_cred = ovl_override_creds(inode->i_sb);
450 file_start_write(real.file);
451
452 ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
453
454 file_end_write(real.file);
455 /* Update size */
456 ovl_copyattr(inode);
457 revert_creds(old_cred);
458 fdput(real);
459
460out_unlock:
461 inode_unlock(inode);
462
463 return ret;
464}
465
466static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
467{
468 struct fd real;
469 const struct cred *old_cred;
470 int ret;
471
472 ret = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
473 if (ret <= 0)
474 return ret;
475
476 ret = ovl_real_fdget_meta(file, &real, !datasync);
477 if (ret)
478 return ret;
479
480 /* Don't sync lower file for fear of receiving EROFS error */
481 if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
482 old_cred = ovl_override_creds(file_inode(file)->i_sb);
483 ret = vfs_fsync_range(real.file, start, end, datasync);
484 revert_creds(old_cred);
485 }
486
487 fdput(real);
488
489 return ret;
490}
491
492static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
493{
494 struct file *realfile = file->private_data;
495 const struct cred *old_cred;
496 int ret;
497
498 if (!realfile->f_op->mmap)
499 return -ENODEV;
500
501 if (WARN_ON(file != vma->vm_file))
502 return -EIO;
503
504 vma_set_file(vma, realfile);
505
506 old_cred = ovl_override_creds(file_inode(file)->i_sb);
507 ret = call_mmap(vma->vm_file, vma);
508 revert_creds(old_cred);
509 ovl_file_accessed(file);
510
511 return ret;
512}
513
514static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
515{
516 struct inode *inode = file_inode(file);
517 struct fd real;
518 const struct cred *old_cred;
519 int ret;
520
521 inode_lock(inode);
522 /* Update mode */
523 ovl_copyattr(inode);
524 ret = file_remove_privs(file);
525 if (ret)
526 goto out_unlock;
527
528 ret = ovl_real_fdget(file, &real);
529 if (ret)
530 goto out_unlock;
531
532 old_cred = ovl_override_creds(file_inode(file)->i_sb);
533 ret = vfs_fallocate(real.file, mode, offset, len);
534 revert_creds(old_cred);
535
536 /* Update size */
537 ovl_copyattr(inode);
538
539 fdput(real);
540
541out_unlock:
542 inode_unlock(inode);
543
544 return ret;
545}
546
547static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
548{
549 struct fd real;
550 const struct cred *old_cred;
551 int ret;
552
553 ret = ovl_real_fdget(file, &real);
554 if (ret)
555 return ret;
556
557 old_cred = ovl_override_creds(file_inode(file)->i_sb);
558 ret = vfs_fadvise(real.file, offset, len, advice);
559 revert_creds(old_cred);
560
561 fdput(real);
562
563 return ret;
564}
565
566enum ovl_copyop {
567 OVL_COPY,
568 OVL_CLONE,
569 OVL_DEDUPE,
570};
571
572static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
573 struct file *file_out, loff_t pos_out,
574 loff_t len, unsigned int flags, enum ovl_copyop op)
575{
576 struct inode *inode_out = file_inode(file_out);
577 struct fd real_in, real_out;
578 const struct cred *old_cred;
579 loff_t ret;
580
581 inode_lock(inode_out);
582 if (op != OVL_DEDUPE) {
583 /* Update mode */
584 ovl_copyattr(inode_out);
585 ret = file_remove_privs(file_out);
586 if (ret)
587 goto out_unlock;
588 }
589
590 ret = ovl_real_fdget(file_out, &real_out);
591 if (ret)
592 goto out_unlock;
593
594 ret = ovl_real_fdget(file_in, &real_in);
595 if (ret) {
596 fdput(real_out);
597 goto out_unlock;
598 }
599
600 old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
601 switch (op) {
602 case OVL_COPY:
603 ret = vfs_copy_file_range(real_in.file, pos_in,
604 real_out.file, pos_out, len, flags);
605 break;
606
607 case OVL_CLONE:
608 ret = vfs_clone_file_range(real_in.file, pos_in,
609 real_out.file, pos_out, len, flags);
610 break;
611
612 case OVL_DEDUPE:
613 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
614 real_out.file, pos_out, len,
615 flags);
616 break;
617 }
618 revert_creds(old_cred);
619
620 /* Update size */
621 ovl_copyattr(inode_out);
622
623 fdput(real_in);
624 fdput(real_out);
625
626out_unlock:
627 inode_unlock(inode_out);
628
629 return ret;
630}
631
632static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
633 struct file *file_out, loff_t pos_out,
634 size_t len, unsigned int flags)
635{
636 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
637 OVL_COPY);
638}
639
640static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
641 struct file *file_out, loff_t pos_out,
642 loff_t len, unsigned int remap_flags)
643{
644 enum ovl_copyop op;
645
646 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
647 return -EINVAL;
648
649 if (remap_flags & REMAP_FILE_DEDUP)
650 op = OVL_DEDUPE;
651 else
652 op = OVL_CLONE;
653
654 /*
655 * Don't copy up because of a dedupe request, this wouldn't make sense
656 * most of the time (data would be duplicated instead of deduplicated).
657 */
658 if (op == OVL_DEDUPE &&
659 (!ovl_inode_upper(file_inode(file_in)) ||
660 !ovl_inode_upper(file_inode(file_out))))
661 return -EPERM;
662
663 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
664 remap_flags, op);
665}
666
667static int ovl_flush(struct file *file, fl_owner_t id)
668{
669 struct fd real;
670 const struct cred *old_cred;
671 int err;
672
673 err = ovl_real_fdget(file, &real);
674 if (err)
675 return err;
676
677 if (real.file->f_op->flush) {
678 old_cred = ovl_override_creds(file_inode(file)->i_sb);
679 err = real.file->f_op->flush(real.file, id);
680 revert_creds(old_cred);
681 }
682 fdput(real);
683
684 return err;
685}
686
687const struct file_operations ovl_file_operations = {
688 .open = ovl_open,
689 .release = ovl_release,
690 .llseek = ovl_llseek,
691 .read_iter = ovl_read_iter,
692 .write_iter = ovl_write_iter,
693 .fsync = ovl_fsync,
694 .mmap = ovl_mmap,
695 .fallocate = ovl_fallocate,
696 .fadvise = ovl_fadvise,
697 .flush = ovl_flush,
698 .splice_read = generic_file_splice_read,
699 .splice_write = ovl_splice_write,
700
701 .copy_file_range = ovl_copy_file_range,
702 .remap_file_range = ovl_remap_file_range,
703};
704
705int __init ovl_aio_request_cache_init(void)
706{
707 ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
708 sizeof(struct ovl_aio_req),
709 0, SLAB_HWCACHE_ALIGN, NULL);
710 if (!ovl_aio_request_cachep)
711 return -ENOMEM;
712
713 return 0;
714}
715
716void ovl_aio_request_cache_destroy(void)
717{
718 kmem_cache_destroy(ovl_aio_request_cachep);
719}