Merge tag 'optee-for-for-v6.6' of https://git.linaro.org/people/jens.wiklander/linux...
[platform/kernel/linux-starfive.git] / fs / overlayfs / file.c
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         refcount_t ref;
21         struct kiocb *orig_iocb;
22         struct fd fd;
23 };
24
25 static struct kmem_cache *ovl_aio_request_cachep;
26
27 static 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 on underlying */
38 #define OVL_OPEN_FLAGS (O_NOATIME)
39
40 static 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 mnt_idmap *real_idmap;
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_idmap = mnt_idmap(realpath->mnt);
57         err = inode_permission(real_idmap, realinode, MAY_OPEN | acc_mode);
58         if (err) {
59                 realfile = ERR_PTR(err);
60         } else {
61                 if (!inode_owner_or_capable(real_idmap, realinode))
62                         flags &= ~O_NOATIME;
63
64                 realfile = backing_file_open(&file->f_path, flags, realpath,
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
78 static 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
105 static 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         int err;
111
112         real->flags = 0;
113         real->file = file->private_data;
114
115         if (allow_meta) {
116                 ovl_path_real(dentry, &realpath);
117         } else {
118                 /* lazy lookup and verify of lowerdata */
119                 err = ovl_verify_lowerdata(dentry);
120                 if (err)
121                         return err;
122
123                 ovl_path_realdata(dentry, &realpath);
124         }
125         if (!realpath.dentry)
126                 return -EIO;
127
128         /* Has it been copied up since we'd opened it? */
129         if (unlikely(file_inode(real->file) != d_inode(realpath.dentry))) {
130                 real->flags = FDPUT_FPUT;
131                 real->file = ovl_open_realfile(file, &realpath);
132
133                 return PTR_ERR_OR_ZERO(real->file);
134         }
135
136         /* Did the flags change since open? */
137         if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
138                 return ovl_change_flags(real->file, file->f_flags);
139
140         return 0;
141 }
142
143 static int ovl_real_fdget(const struct file *file, struct fd *real)
144 {
145         if (d_is_dir(file_dentry(file))) {
146                 real->flags = 0;
147                 real->file = ovl_dir_real_file(file, false);
148
149                 return PTR_ERR_OR_ZERO(real->file);
150         }
151
152         return ovl_real_fdget_meta(file, real, false);
153 }
154
155 static int ovl_open(struct inode *inode, struct file *file)
156 {
157         struct dentry *dentry = file_dentry(file);
158         struct file *realfile;
159         struct path realpath;
160         int err;
161
162         /* lazy lookup and verify lowerdata */
163         err = ovl_verify_lowerdata(dentry);
164         if (err)
165                 return err;
166
167         err = ovl_maybe_copy_up(dentry, file->f_flags);
168         if (err)
169                 return err;
170
171         /* No longer need these flags, so don't pass them on to underlying fs */
172         file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
173
174         ovl_path_realdata(dentry, &realpath);
175         if (!realpath.dentry)
176                 return -EIO;
177
178         realfile = ovl_open_realfile(file, &realpath);
179         if (IS_ERR(realfile))
180                 return PTR_ERR(realfile);
181
182         file->private_data = realfile;
183
184         return 0;
185 }
186
187 static int ovl_release(struct inode *inode, struct file *file)
188 {
189         fput(file->private_data);
190
191         return 0;
192 }
193
194 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
195 {
196         struct inode *inode = file_inode(file);
197         struct fd real;
198         const struct cred *old_cred;
199         loff_t ret;
200
201         /*
202          * The two special cases below do not need to involve real fs,
203          * so we can optimizing concurrent callers.
204          */
205         if (offset == 0) {
206                 if (whence == SEEK_CUR)
207                         return file->f_pos;
208
209                 if (whence == SEEK_SET)
210                         return vfs_setpos(file, 0, 0);
211         }
212
213         ret = ovl_real_fdget(file, &real);
214         if (ret)
215                 return ret;
216
217         /*
218          * Overlay file f_pos is the master copy that is preserved
219          * through copy up and modified on read/write, but only real
220          * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
221          * limitations that are more strict than ->s_maxbytes for specific
222          * files, so we use the real file to perform seeks.
223          */
224         ovl_inode_lock(inode);
225         real.file->f_pos = file->f_pos;
226
227         old_cred = ovl_override_creds(inode->i_sb);
228         ret = vfs_llseek(real.file, offset, whence);
229         revert_creds(old_cred);
230
231         file->f_pos = real.file->f_pos;
232         ovl_inode_unlock(inode);
233
234         fdput(real);
235
236         return ret;
237 }
238
239 static void ovl_file_accessed(struct file *file)
240 {
241         struct inode *inode, *upperinode;
242         struct timespec64 ctime, uctime;
243
244         if (file->f_flags & O_NOATIME)
245                 return;
246
247         inode = file_inode(file);
248         upperinode = ovl_inode_upper(inode);
249
250         if (!upperinode)
251                 return;
252
253         ctime = inode_get_ctime(inode);
254         uctime = inode_get_ctime(upperinode);
255         if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
256              !timespec64_equal(&ctime, &uctime))) {
257                 inode->i_mtime = upperinode->i_mtime;
258                 inode_set_ctime_to_ts(inode, uctime);
259         }
260
261         touch_atime(&file->f_path);
262 }
263
264 static rwf_t ovl_iocb_to_rwf(int ifl)
265 {
266         rwf_t flags = 0;
267
268         if (ifl & IOCB_NOWAIT)
269                 flags |= RWF_NOWAIT;
270         if (ifl & IOCB_HIPRI)
271                 flags |= RWF_HIPRI;
272         if (ifl & IOCB_DSYNC)
273                 flags |= RWF_DSYNC;
274         if (ifl & IOCB_SYNC)
275                 flags |= RWF_SYNC;
276
277         return flags;
278 }
279
280 static inline void ovl_aio_put(struct ovl_aio_req *aio_req)
281 {
282         if (refcount_dec_and_test(&aio_req->ref)) {
283                 fdput(aio_req->fd);
284                 kmem_cache_free(ovl_aio_request_cachep, aio_req);
285         }
286 }
287
288 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
289 {
290         struct kiocb *iocb = &aio_req->iocb;
291         struct kiocb *orig_iocb = aio_req->orig_iocb;
292
293         if (iocb->ki_flags & IOCB_WRITE) {
294                 struct inode *inode = file_inode(orig_iocb->ki_filp);
295
296                 kiocb_end_write(iocb);
297                 ovl_copyattr(inode);
298         }
299
300         orig_iocb->ki_pos = iocb->ki_pos;
301         ovl_aio_put(aio_req);
302 }
303
304 static void ovl_aio_rw_complete(struct kiocb *iocb, long res)
305 {
306         struct ovl_aio_req *aio_req = container_of(iocb,
307                                                    struct ovl_aio_req, iocb);
308         struct kiocb *orig_iocb = aio_req->orig_iocb;
309
310         ovl_aio_cleanup_handler(aio_req);
311         orig_iocb->ki_complete(orig_iocb, res);
312 }
313
314 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
315 {
316         struct file *file = iocb->ki_filp;
317         struct fd real;
318         const struct cred *old_cred;
319         ssize_t ret;
320
321         if (!iov_iter_count(iter))
322                 return 0;
323
324         ret = ovl_real_fdget(file, &real);
325         if (ret)
326                 return ret;
327
328         ret = -EINVAL;
329         if (iocb->ki_flags & IOCB_DIRECT &&
330             !(real.file->f_mode & FMODE_CAN_ODIRECT))
331                 goto out_fdput;
332
333         old_cred = ovl_override_creds(file_inode(file)->i_sb);
334         if (is_sync_kiocb(iocb)) {
335                 ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
336                                     ovl_iocb_to_rwf(iocb->ki_flags));
337         } else {
338                 struct ovl_aio_req *aio_req;
339
340                 ret = -ENOMEM;
341                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
342                 if (!aio_req)
343                         goto out;
344
345                 aio_req->fd = real;
346                 real.flags = 0;
347                 aio_req->orig_iocb = iocb;
348                 kiocb_clone(&aio_req->iocb, iocb, real.file);
349                 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
350                 refcount_set(&aio_req->ref, 2);
351                 ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
352                 ovl_aio_put(aio_req);
353                 if (ret != -EIOCBQUEUED)
354                         ovl_aio_cleanup_handler(aio_req);
355         }
356 out:
357         revert_creds(old_cred);
358         ovl_file_accessed(file);
359 out_fdput:
360         fdput(real);
361
362         return ret;
363 }
364
365 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
366 {
367         struct file *file = iocb->ki_filp;
368         struct inode *inode = file_inode(file);
369         struct fd real;
370         const struct cred *old_cred;
371         ssize_t ret;
372         int ifl = iocb->ki_flags;
373
374         if (!iov_iter_count(iter))
375                 return 0;
376
377         inode_lock(inode);
378         /* Update mode */
379         ovl_copyattr(inode);
380         ret = file_remove_privs(file);
381         if (ret)
382                 goto out_unlock;
383
384         ret = ovl_real_fdget(file, &real);
385         if (ret)
386                 goto out_unlock;
387
388         ret = -EINVAL;
389         if (iocb->ki_flags & IOCB_DIRECT &&
390             !(real.file->f_mode & FMODE_CAN_ODIRECT))
391                 goto out_fdput;
392
393         if (!ovl_should_sync(OVL_FS(inode->i_sb)))
394                 ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
395
396         old_cred = ovl_override_creds(file_inode(file)->i_sb);
397         if (is_sync_kiocb(iocb)) {
398                 file_start_write(real.file);
399                 ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
400                                      ovl_iocb_to_rwf(ifl));
401                 file_end_write(real.file);
402                 /* Update size */
403                 ovl_copyattr(inode);
404         } else {
405                 struct ovl_aio_req *aio_req;
406
407                 ret = -ENOMEM;
408                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
409                 if (!aio_req)
410                         goto out;
411
412                 aio_req->fd = real;
413                 real.flags = 0;
414                 aio_req->orig_iocb = iocb;
415                 kiocb_clone(&aio_req->iocb, iocb, real.file);
416                 aio_req->iocb.ki_flags = ifl;
417                 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
418                 refcount_set(&aio_req->ref, 2);
419                 kiocb_start_write(&aio_req->iocb);
420                 ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
421                 ovl_aio_put(aio_req);
422                 if (ret != -EIOCBQUEUED)
423                         ovl_aio_cleanup_handler(aio_req);
424         }
425 out:
426         revert_creds(old_cred);
427 out_fdput:
428         fdput(real);
429
430 out_unlock:
431         inode_unlock(inode);
432
433         return ret;
434 }
435
436 static ssize_t ovl_splice_read(struct file *in, loff_t *ppos,
437                                struct pipe_inode_info *pipe, size_t len,
438                                unsigned int flags)
439 {
440         const struct cred *old_cred;
441         struct fd real;
442         ssize_t ret;
443
444         ret = ovl_real_fdget(in, &real);
445         if (ret)
446                 return ret;
447
448         old_cred = ovl_override_creds(file_inode(in)->i_sb);
449         ret = vfs_splice_read(real.file, ppos, pipe, len, flags);
450         revert_creds(old_cred);
451         ovl_file_accessed(in);
452
453         fdput(real);
454         return ret;
455 }
456
457 /*
458  * Calling iter_file_splice_write() directly from overlay's f_op may deadlock
459  * due to lock order inversion between pipe->mutex in iter_file_splice_write()
460  * and file_start_write(real.file) in ovl_write_iter().
461  *
462  * So do everything ovl_write_iter() does and call iter_file_splice_write() on
463  * the real file.
464  */
465 static ssize_t ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
466                                 loff_t *ppos, size_t len, unsigned int flags)
467 {
468         struct fd real;
469         const struct cred *old_cred;
470         struct inode *inode = file_inode(out);
471         ssize_t ret;
472
473         inode_lock(inode);
474         /* Update mode */
475         ovl_copyattr(inode);
476         ret = file_remove_privs(out);
477         if (ret)
478                 goto out_unlock;
479
480         ret = ovl_real_fdget(out, &real);
481         if (ret)
482                 goto out_unlock;
483
484         old_cred = ovl_override_creds(inode->i_sb);
485         file_start_write(real.file);
486
487         ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
488
489         file_end_write(real.file);
490         /* Update size */
491         ovl_copyattr(inode);
492         revert_creds(old_cred);
493         fdput(real);
494
495 out_unlock:
496         inode_unlock(inode);
497
498         return ret;
499 }
500
501 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
502 {
503         struct fd real;
504         const struct cred *old_cred;
505         int ret;
506
507         ret = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
508         if (ret <= 0)
509                 return ret;
510
511         ret = ovl_real_fdget_meta(file, &real, !datasync);
512         if (ret)
513                 return ret;
514
515         /* Don't sync lower file for fear of receiving EROFS error */
516         if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
517                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
518                 ret = vfs_fsync_range(real.file, start, end, datasync);
519                 revert_creds(old_cred);
520         }
521
522         fdput(real);
523
524         return ret;
525 }
526
527 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
528 {
529         struct file *realfile = file->private_data;
530         const struct cred *old_cred;
531         int ret;
532
533         if (!realfile->f_op->mmap)
534                 return -ENODEV;
535
536         if (WARN_ON(file != vma->vm_file))
537                 return -EIO;
538
539         vma_set_file(vma, realfile);
540
541         old_cred = ovl_override_creds(file_inode(file)->i_sb);
542         ret = call_mmap(vma->vm_file, vma);
543         revert_creds(old_cred);
544         ovl_file_accessed(file);
545
546         return ret;
547 }
548
549 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
550 {
551         struct inode *inode = file_inode(file);
552         struct fd real;
553         const struct cred *old_cred;
554         int ret;
555
556         inode_lock(inode);
557         /* Update mode */
558         ovl_copyattr(inode);
559         ret = file_remove_privs(file);
560         if (ret)
561                 goto out_unlock;
562
563         ret = ovl_real_fdget(file, &real);
564         if (ret)
565                 goto out_unlock;
566
567         old_cred = ovl_override_creds(file_inode(file)->i_sb);
568         ret = vfs_fallocate(real.file, mode, offset, len);
569         revert_creds(old_cred);
570
571         /* Update size */
572         ovl_copyattr(inode);
573
574         fdput(real);
575
576 out_unlock:
577         inode_unlock(inode);
578
579         return ret;
580 }
581
582 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
583 {
584         struct fd real;
585         const struct cred *old_cred;
586         int ret;
587
588         ret = ovl_real_fdget(file, &real);
589         if (ret)
590                 return ret;
591
592         old_cred = ovl_override_creds(file_inode(file)->i_sb);
593         ret = vfs_fadvise(real.file, offset, len, advice);
594         revert_creds(old_cred);
595
596         fdput(real);
597
598         return ret;
599 }
600
601 enum ovl_copyop {
602         OVL_COPY,
603         OVL_CLONE,
604         OVL_DEDUPE,
605 };
606
607 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
608                             struct file *file_out, loff_t pos_out,
609                             loff_t len, unsigned int flags, enum ovl_copyop op)
610 {
611         struct inode *inode_out = file_inode(file_out);
612         struct fd real_in, real_out;
613         const struct cred *old_cred;
614         loff_t ret;
615
616         inode_lock(inode_out);
617         if (op != OVL_DEDUPE) {
618                 /* Update mode */
619                 ovl_copyattr(inode_out);
620                 ret = file_remove_privs(file_out);
621                 if (ret)
622                         goto out_unlock;
623         }
624
625         ret = ovl_real_fdget(file_out, &real_out);
626         if (ret)
627                 goto out_unlock;
628
629         ret = ovl_real_fdget(file_in, &real_in);
630         if (ret) {
631                 fdput(real_out);
632                 goto out_unlock;
633         }
634
635         old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
636         switch (op) {
637         case OVL_COPY:
638                 ret = vfs_copy_file_range(real_in.file, pos_in,
639                                           real_out.file, pos_out, len, flags);
640                 break;
641
642         case OVL_CLONE:
643                 ret = vfs_clone_file_range(real_in.file, pos_in,
644                                            real_out.file, pos_out, len, flags);
645                 break;
646
647         case OVL_DEDUPE:
648                 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
649                                                 real_out.file, pos_out, len,
650                                                 flags);
651                 break;
652         }
653         revert_creds(old_cred);
654
655         /* Update size */
656         ovl_copyattr(inode_out);
657
658         fdput(real_in);
659         fdput(real_out);
660
661 out_unlock:
662         inode_unlock(inode_out);
663
664         return ret;
665 }
666
667 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
668                                    struct file *file_out, loff_t pos_out,
669                                    size_t len, unsigned int flags)
670 {
671         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
672                             OVL_COPY);
673 }
674
675 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
676                                    struct file *file_out, loff_t pos_out,
677                                    loff_t len, unsigned int remap_flags)
678 {
679         enum ovl_copyop op;
680
681         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
682                 return -EINVAL;
683
684         if (remap_flags & REMAP_FILE_DEDUP)
685                 op = OVL_DEDUPE;
686         else
687                 op = OVL_CLONE;
688
689         /*
690          * Don't copy up because of a dedupe request, this wouldn't make sense
691          * most of the time (data would be duplicated instead of deduplicated).
692          */
693         if (op == OVL_DEDUPE &&
694             (!ovl_inode_upper(file_inode(file_in)) ||
695              !ovl_inode_upper(file_inode(file_out))))
696                 return -EPERM;
697
698         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
699                             remap_flags, op);
700 }
701
702 static int ovl_flush(struct file *file, fl_owner_t id)
703 {
704         struct fd real;
705         const struct cred *old_cred;
706         int err;
707
708         err = ovl_real_fdget(file, &real);
709         if (err)
710                 return err;
711
712         if (real.file->f_op->flush) {
713                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
714                 err = real.file->f_op->flush(real.file, id);
715                 revert_creds(old_cred);
716         }
717         fdput(real);
718
719         return err;
720 }
721
722 const struct file_operations ovl_file_operations = {
723         .open           = ovl_open,
724         .release        = ovl_release,
725         .llseek         = ovl_llseek,
726         .read_iter      = ovl_read_iter,
727         .write_iter     = ovl_write_iter,
728         .fsync          = ovl_fsync,
729         .mmap           = ovl_mmap,
730         .fallocate      = ovl_fallocate,
731         .fadvise        = ovl_fadvise,
732         .flush          = ovl_flush,
733         .splice_read    = ovl_splice_read,
734         .splice_write   = ovl_splice_write,
735
736         .copy_file_range        = ovl_copy_file_range,
737         .remap_file_range       = ovl_remap_file_range,
738 };
739
740 int __init ovl_aio_request_cache_init(void)
741 {
742         ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
743                                                    sizeof(struct ovl_aio_req),
744                                                    0, SLAB_HWCACHE_ALIGN, NULL);
745         if (!ovl_aio_request_cachep)
746                 return -ENOMEM;
747
748         return 0;
749 }
750
751 void ovl_aio_request_cache_destroy(void)
752 {
753         kmem_cache_destroy(ovl_aio_request_cachep);
754 }