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/security.h>
13#include <linux/fs.h>
14#include <linux/backing-file.h>
15#include "overlayfs.h"
16
17static 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
27static struct file *ovl_open_realfile(const struct file *file,
28 const struct path *realpath)
29{
30 struct inode *realinode = d_inode(dentry: realpath->dentry);
31 struct inode *inode = file_inode(f: file);
32 struct mnt_idmap *real_idmap;
33 struct file *realfile;
34 int flags = file->f_flags | OVL_OPEN_FLAGS;
35 int acc_mode = ACC_MODE(flags);
36 int err;
37
38 if (flags & O_APPEND)
39 acc_mode |= MAY_APPEND;
40
41 with_ovl_creds(inode->i_sb) {
42 real_idmap = mnt_idmap(mnt: realpath->mnt);
43 err = inode_permission(real_idmap, realinode, MAY_OPEN | acc_mode);
44 if (err) {
45 realfile = ERR_PTR(error: err);
46 } else {
47 if (!inode_owner_or_capable(idmap: real_idmap, inode: realinode))
48 flags &= ~O_NOATIME;
49
50 realfile = backing_file_open(user_path: file_user_path(f: file),
51 flags, real_path: realpath, current_cred());
52 }
53 }
54
55 pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
56 file, file, ovl_whatisit(inode, realinode), file->f_flags,
57 realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
58
59 return realfile;
60}
61
62#define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
63
64static int ovl_change_flags(struct file *file, unsigned int flags)
65{
66 struct inode *inode = file_inode(f: file);
67 int err;
68
69 flags &= OVL_SETFL_MASK;
70
71 if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
72 return -EPERM;
73
74 if ((flags & O_DIRECT) && !(file->f_mode & FMODE_CAN_ODIRECT))
75 return -EINVAL;
76
77 if (file->f_op->check_flags) {
78 err = file->f_op->check_flags(flags);
79 if (err)
80 return err;
81 }
82
83 spin_lock(lock: &file->f_lock);
84 file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
85 file->f_iocb_flags = iocb_flags(file);
86 spin_unlock(lock: &file->f_lock);
87
88 return 0;
89}
90
91struct ovl_file {
92 struct file *realfile;
93 struct file *upperfile;
94};
95
96struct ovl_file *ovl_file_alloc(struct file *realfile)
97{
98 struct ovl_file *of = kzalloc(sizeof(struct ovl_file), GFP_KERNEL);
99
100 if (unlikely(!of))
101 return NULL;
102
103 of->realfile = realfile;
104 return of;
105}
106
107void ovl_file_free(struct ovl_file *of)
108{
109 fput(of->realfile);
110 if (of->upperfile)
111 fput(of->upperfile);
112 kfree(objp: of);
113}
114
115static bool ovl_is_real_file(const struct file *realfile,
116 const struct path *realpath)
117{
118 return file_inode(f: realfile) == d_inode(dentry: realpath->dentry);
119}
120
121static struct file *ovl_real_file_path(const struct file *file,
122 const struct path *realpath)
123{
124 struct ovl_file *of = file->private_data;
125 struct file *realfile = of->realfile;
126
127 if (WARN_ON_ONCE(!realpath->dentry))
128 return ERR_PTR(error: -EIO);
129
130 /*
131 * If the realfile that we want is not where the data used to be at
132 * open time, either we'd been copied up, or it's an fsync of a
133 * metacopied file. We need the upperfile either way, so see if it
134 * is already opened and if it is not then open and store it.
135 */
136 if (unlikely(!ovl_is_real_file(realfile, realpath))) {
137 struct file *upperfile = READ_ONCE(of->upperfile);
138 struct file *old;
139
140 if (!upperfile) { /* Nobody opened upperfile yet */
141 upperfile = ovl_open_realfile(file, realpath);
142 if (IS_ERR(ptr: upperfile))
143 return upperfile;
144
145 /* Store the upperfile for later */
146 old = cmpxchg_release(&of->upperfile, NULL, upperfile);
147 if (old) { /* Someone opened upperfile before us */
148 fput(upperfile);
149 upperfile = old;
150 }
151 }
152 /*
153 * Stored file must be from the right inode, unless someone's
154 * been corrupting the upper layer.
155 */
156 if (WARN_ON_ONCE(!ovl_is_real_file(upperfile, realpath)))
157 return ERR_PTR(error: -EIO);
158
159 realfile = upperfile;
160 }
161
162 /* Did the flags change since open? */
163 if (unlikely((file->f_flags ^ realfile->f_flags) & ~OVL_OPEN_FLAGS)) {
164 int err = ovl_change_flags(file: realfile, flags: file->f_flags);
165
166 if (err)
167 return ERR_PTR(error: err);
168 }
169
170 return realfile;
171}
172
173static struct file *ovl_real_file(const struct file *file)
174{
175 struct dentry *dentry = file_dentry(file);
176 struct path realpath;
177 int err;
178
179 if (d_is_dir(dentry)) {
180 struct file *f = ovl_dir_real_file(file, want_upper: false);
181
182 if (WARN_ON_ONCE(!f))
183 return ERR_PTR(error: -EIO);
184 return f;
185 }
186
187 /* lazy lookup and verify of lowerdata */
188 err = ovl_verify_lowerdata(dentry);
189 if (err)
190 return ERR_PTR(error: err);
191
192 ovl_path_realdata(dentry, path: &realpath);
193
194 return ovl_real_file_path(file, realpath: &realpath);
195}
196
197static int ovl_open(struct inode *inode, struct file *file)
198{
199 struct dentry *dentry = file_dentry(file);
200 struct file *realfile;
201 struct path realpath;
202 struct ovl_file *of;
203 int err;
204
205 /* lazy lookup and verify lowerdata */
206 err = ovl_verify_lowerdata(dentry);
207 if (err)
208 return err;
209
210 err = ovl_maybe_copy_up(dentry, flags: file->f_flags);
211 if (err)
212 return err;
213
214 /* No longer need these flags, so don't pass them on to underlying fs */
215 file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
216
217 ovl_path_realdata(dentry, path: &realpath);
218 if (!realpath.dentry)
219 return -EIO;
220
221 realfile = ovl_open_realfile(file, realpath: &realpath);
222 if (IS_ERR(ptr: realfile))
223 return PTR_ERR(ptr: realfile);
224
225 of = ovl_file_alloc(realfile);
226 if (!of) {
227 fput(realfile);
228 return -ENOMEM;
229 }
230
231 file->private_data = of;
232
233 return 0;
234}
235
236static int ovl_release(struct inode *inode, struct file *file)
237{
238 ovl_file_free(of: file->private_data);
239 return 0;
240}
241
242static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
243{
244 struct inode *inode = file_inode(f: file);
245 struct file *realfile;
246 loff_t ret;
247
248 /*
249 * The two special cases below do not need to involve real fs,
250 * so we can optimizing concurrent callers.
251 */
252 if (offset == 0) {
253 if (whence == SEEK_CUR)
254 return file->f_pos;
255
256 if (whence == SEEK_SET)
257 return vfs_setpos(file, offset: 0, maxsize: 0);
258 }
259
260 realfile = ovl_real_file(file);
261 if (IS_ERR(ptr: realfile))
262 return PTR_ERR(ptr: realfile);
263
264 /*
265 * Overlay file f_pos is the master copy that is preserved
266 * through copy up and modified on read/write, but only real
267 * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
268 * limitations that are more strict than ->s_maxbytes for specific
269 * files, so we use the real file to perform seeks.
270 */
271 ovl_inode_lock(inode);
272 realfile->f_pos = file->f_pos;
273
274 with_ovl_creds(inode->i_sb)
275 ret = vfs_llseek(file: realfile, offset, whence);
276
277 file->f_pos = realfile->f_pos;
278 ovl_inode_unlock(inode);
279
280 return ret;
281}
282
283static void ovl_file_modified(struct file *file)
284{
285 /* Update size/mtime */
286 ovl_copyattr(to: file_inode(f: file));
287}
288
289static void ovl_file_end_write(struct kiocb *iocb, ssize_t ret)
290{
291 ovl_file_modified(file: iocb->ki_filp);
292}
293
294static void ovl_file_accessed(struct file *file)
295{
296 struct inode *inode, *upperinode;
297 struct timespec64 ctime, uctime;
298 struct timespec64 mtime, umtime;
299
300 if (file->f_flags & O_NOATIME)
301 return;
302
303 inode = file_inode(f: file);
304 upperinode = ovl_inode_upper(inode);
305
306 if (!upperinode)
307 return;
308
309 ctime = inode_get_ctime(inode);
310 uctime = inode_get_ctime(inode: upperinode);
311 mtime = inode_get_mtime(inode);
312 umtime = inode_get_mtime(inode: upperinode);
313 if ((!timespec64_equal(a: &mtime, b: &umtime)) ||
314 !timespec64_equal(a: &ctime, b: &uctime)) {
315 inode_set_mtime_to_ts(inode, ts: inode_get_mtime(inode: upperinode));
316 inode_set_ctime_to_ts(inode, ts: uctime);
317 }
318
319 touch_atime(&file->f_path);
320}
321
322static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
323{
324 struct file *file = iocb->ki_filp;
325 struct file *realfile;
326 struct backing_file_ctx ctx = {
327 .cred = ovl_creds(sb: file_inode(f: file)->i_sb),
328 .accessed = ovl_file_accessed,
329 };
330
331 if (!iov_iter_count(i: iter))
332 return 0;
333
334 realfile = ovl_real_file(file);
335 if (IS_ERR(ptr: realfile))
336 return PTR_ERR(ptr: realfile);
337
338 return backing_file_read_iter(file: realfile, iter, iocb, flags: iocb->ki_flags,
339 ctx: &ctx);
340}
341
342static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
343{
344 struct file *file = iocb->ki_filp;
345 struct inode *inode = file_inode(f: file);
346 struct file *realfile;
347 ssize_t ret;
348 int ifl = iocb->ki_flags;
349 struct backing_file_ctx ctx = {
350 .cred = ovl_creds(sb: inode->i_sb),
351 .end_write = ovl_file_end_write,
352 };
353
354 if (!iov_iter_count(i: iter))
355 return 0;
356
357 inode_lock(inode);
358 /* Update mode */
359 ovl_copyattr(to: inode);
360
361 realfile = ovl_real_file(file);
362 ret = PTR_ERR(ptr: realfile);
363 if (IS_ERR(ptr: realfile))
364 goto out_unlock;
365
366 if (!ovl_should_sync(ofs: OVL_FS(sb: inode->i_sb)))
367 ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
368
369 ret = backing_file_write_iter(file: realfile, iter, iocb, flags: ifl, ctx: &ctx);
370
371out_unlock:
372 inode_unlock(inode);
373
374 return ret;
375}
376
377static ssize_t ovl_splice_read(struct file *in, loff_t *ppos,
378 struct pipe_inode_info *pipe, size_t len,
379 unsigned int flags)
380{
381 struct file *realfile;
382 ssize_t ret;
383 struct backing_file_ctx ctx = {
384 .cred = ovl_creds(sb: file_inode(f: in)->i_sb),
385 .accessed = ovl_file_accessed,
386 };
387 struct kiocb iocb;
388
389 realfile = ovl_real_file(file: in);
390 if (IS_ERR(ptr: realfile))
391 return PTR_ERR(ptr: realfile);
392
393 init_sync_kiocb(kiocb: &iocb, filp: in);
394 iocb.ki_pos = *ppos;
395 ret = backing_file_splice_read(in: realfile, iocb: &iocb, pipe, len, flags, ctx: &ctx);
396 *ppos = iocb.ki_pos;
397
398 return ret;
399}
400
401/*
402 * Calling iter_file_splice_write() directly from overlay's f_op may deadlock
403 * due to lock order inversion between pipe->mutex in iter_file_splice_write()
404 * and file_start_write(realfile) in ovl_write_iter().
405 *
406 * So do everything ovl_write_iter() does and call iter_file_splice_write() on
407 * the real file.
408 */
409static ssize_t ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
410 loff_t *ppos, size_t len, unsigned int flags)
411{
412 struct file *realfile;
413 struct inode *inode = file_inode(f: out);
414 ssize_t ret;
415 struct backing_file_ctx ctx = {
416 .cred = ovl_creds(sb: inode->i_sb),
417 .end_write = ovl_file_end_write,
418 };
419 struct kiocb iocb;
420
421 inode_lock(inode);
422 /* Update mode */
423 ovl_copyattr(to: inode);
424
425 realfile = ovl_real_file(file: out);
426 ret = PTR_ERR(ptr: realfile);
427 if (IS_ERR(ptr: realfile))
428 goto out_unlock;
429
430 init_sync_kiocb(kiocb: &iocb, filp: out);
431 iocb.ki_pos = *ppos;
432 ret = backing_file_splice_write(pipe, out: realfile, iocb: &iocb, len, flags, ctx: &ctx);
433 *ppos = iocb.ki_pos;
434
435out_unlock:
436 inode_unlock(inode);
437
438 return ret;
439}
440
441static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
442{
443 struct dentry *dentry = file_dentry(file);
444 enum ovl_path_type type;
445 struct path upperpath;
446 struct file *upperfile;
447 int ret;
448
449 ret = ovl_sync_status(ofs: OVL_FS(sb: file_inode(f: file)->i_sb));
450 if (ret <= 0)
451 return ret;
452
453 /* Don't sync lower file for fear of receiving EROFS error */
454 type = ovl_path_type(dentry);
455 if (!OVL_TYPE_UPPER(type) || (datasync && OVL_TYPE_MERGE(type)))
456 return 0;
457
458 ovl_path_upper(dentry, path: &upperpath);
459 upperfile = ovl_real_file_path(file, realpath: &upperpath);
460 if (IS_ERR(ptr: upperfile))
461 return PTR_ERR(ptr: upperfile);
462
463 with_ovl_creds(file_inode(file)->i_sb)
464 return vfs_fsync_range(file: upperfile, start, end, datasync);
465}
466
467static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
468{
469 struct ovl_file *of = file->private_data;
470 struct backing_file_ctx ctx = {
471 .cred = ovl_creds(sb: file_inode(f: file)->i_sb),
472 .accessed = ovl_file_accessed,
473 };
474
475 return backing_file_mmap(file: of->realfile, vma, ctx: &ctx);
476}
477
478static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
479{
480 struct inode *inode = file_inode(f: file);
481 struct file *realfile;
482 int ret;
483
484 inode_lock(inode);
485 /* Update mode */
486 ovl_copyattr(to: inode);
487 ret = file_remove_privs(file);
488 if (ret)
489 goto out_unlock;
490
491 realfile = ovl_real_file(file);
492 ret = PTR_ERR(ptr: realfile);
493 if (IS_ERR(ptr: realfile))
494 goto out_unlock;
495
496 with_ovl_creds(inode->i_sb)
497 ret = vfs_fallocate(file: realfile, mode, offset, len);
498
499 /* Update size */
500 ovl_file_modified(file);
501
502out_unlock:
503 inode_unlock(inode);
504
505 return ret;
506}
507
508static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
509{
510 struct file *realfile;
511
512 realfile = ovl_real_file(file);
513 if (IS_ERR(ptr: realfile))
514 return PTR_ERR(ptr: realfile);
515
516 with_ovl_creds(file_inode(file)->i_sb)
517 return vfs_fadvise(file: realfile, offset, len, advice);
518}
519
520enum ovl_copyop {
521 OVL_COPY,
522 OVL_CLONE,
523 OVL_DEDUPE,
524};
525
526static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
527 struct file *file_out, loff_t pos_out,
528 loff_t len, unsigned int flags, enum ovl_copyop op)
529{
530 struct inode *inode_out = file_inode(f: file_out);
531 struct file *realfile_in, *realfile_out;
532 loff_t ret;
533
534 inode_lock(inode: inode_out);
535 if (op != OVL_DEDUPE) {
536 /* Update mode */
537 ovl_copyattr(to: inode_out);
538 ret = file_remove_privs(file_out);
539 if (ret)
540 goto out_unlock;
541 }
542
543 realfile_out = ovl_real_file(file: file_out);
544 ret = PTR_ERR(ptr: realfile_out);
545 if (IS_ERR(ptr: realfile_out))
546 goto out_unlock;
547
548 realfile_in = ovl_real_file(file: file_in);
549 ret = PTR_ERR(ptr: realfile_in);
550 if (IS_ERR(ptr: realfile_in))
551 goto out_unlock;
552
553 with_ovl_creds(file_inode(file_out)->i_sb) {
554 switch (op) {
555 case OVL_COPY:
556 ret = vfs_copy_file_range(realfile_in, pos_in,
557 realfile_out, pos_out, len, flags);
558 break;
559
560 case OVL_CLONE:
561 ret = vfs_clone_file_range(file_in: realfile_in, pos_in,
562 file_out: realfile_out, pos_out, len, remap_flags: flags);
563 break;
564
565 case OVL_DEDUPE:
566 ret = vfs_dedupe_file_range_one(src_file: realfile_in, src_pos: pos_in,
567 dst_file: realfile_out, dst_pos: pos_out, len,
568 remap_flags: flags);
569 break;
570 }
571 }
572
573 /* Update size */
574 ovl_file_modified(file: file_out);
575
576out_unlock:
577 inode_unlock(inode: inode_out);
578
579 return ret;
580}
581
582static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
583 struct file *file_out, loff_t pos_out,
584 size_t len, unsigned int flags)
585{
586 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
587 op: OVL_COPY);
588}
589
590static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
591 struct file *file_out, loff_t pos_out,
592 loff_t len, unsigned int remap_flags)
593{
594 enum ovl_copyop op;
595
596 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
597 return -EINVAL;
598
599 if (remap_flags & REMAP_FILE_DEDUP)
600 op = OVL_DEDUPE;
601 else
602 op = OVL_CLONE;
603
604 /*
605 * Don't copy up because of a dedupe request, this wouldn't make sense
606 * most of the time (data would be duplicated instead of deduplicated).
607 */
608 if (op == OVL_DEDUPE &&
609 (!ovl_inode_upper(inode: file_inode(f: file_in)) ||
610 !ovl_inode_upper(inode: file_inode(f: file_out))))
611 return -EPERM;
612
613 return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
614 flags: remap_flags, op);
615}
616
617static int ovl_flush(struct file *file, fl_owner_t id)
618{
619 struct file *realfile;
620 int err = 0;
621
622 realfile = ovl_real_file(file);
623 if (IS_ERR(ptr: realfile))
624 return PTR_ERR(ptr: realfile);
625
626 if (realfile->f_op->flush) {
627 with_ovl_creds(file_inode(file)->i_sb)
628 err = realfile->f_op->flush(realfile, id);
629 }
630
631 return err;
632}
633
634const struct file_operations ovl_file_operations = {
635 .open = ovl_open,
636 .release = ovl_release,
637 .llseek = ovl_llseek,
638 .read_iter = ovl_read_iter,
639 .write_iter = ovl_write_iter,
640 .fsync = ovl_fsync,
641 .mmap = ovl_mmap,
642 .fallocate = ovl_fallocate,
643 .fadvise = ovl_fadvise,
644 .flush = ovl_flush,
645 .splice_read = ovl_splice_read,
646 .splice_write = ovl_splice_write,
647
648 .copy_file_range = ovl_copy_file_range,
649 .remap_file_range = ovl_remap_file_range,
650};
651

source code of linux/fs/overlayfs/file.c