exit: Use the correct exit_code in /proc/<pid>/stat
[linux-2.6-block.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 modificaton nor notify on underlying */
38 #define OVL_OPEN_FLAGS (O_NOATIME | FMODE_NONOTIFY)
39
40 static struct file *ovl_open_realfile(const struct file *file,
41                                       struct inode *realinode)
42 {
43         struct inode *inode = file_inode(file);
44         struct file *realfile;
45         const struct cred *old_cred;
46         int flags = file->f_flags | OVL_OPEN_FLAGS;
47         int acc_mode = ACC_MODE(flags);
48         int err;
49
50         if (flags & O_APPEND)
51                 acc_mode |= MAY_APPEND;
52
53         old_cred = ovl_override_creds(inode->i_sb);
54         err = inode_permission(&init_user_ns, realinode, MAY_OPEN | acc_mode);
55         if (err) {
56                 realfile = ERR_PTR(err);
57         } else {
58                 if (!inode_owner_or_capable(&init_user_ns, realinode))
59                         flags &= ~O_NOATIME;
60
61                 realfile = open_with_fake_path(&file->f_path, flags, realinode,
62                                                current_cred());
63         }
64         revert_creds(old_cred);
65
66         pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
67                  file, file, ovl_whatisit(inode, realinode), file->f_flags,
68                  realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
69
70         return realfile;
71 }
72
73 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
74
75 static int ovl_change_flags(struct file *file, unsigned int flags)
76 {
77         struct inode *inode = file_inode(file);
78         int err;
79
80         flags &= OVL_SETFL_MASK;
81
82         if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
83                 return -EPERM;
84
85         if (flags & O_DIRECT) {
86                 if (!file->f_mapping->a_ops ||
87                     !file->f_mapping->a_ops->direct_IO)
88                         return -EINVAL;
89         }
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         spin_unlock(&file->f_lock);
100
101         return 0;
102 }
103
104 static int ovl_real_fdget_meta(const struct file *file, struct fd *real,
105                                bool allow_meta)
106 {
107         struct inode *inode = file_inode(file);
108         struct inode *realinode;
109
110         real->flags = 0;
111         real->file = file->private_data;
112
113         if (allow_meta)
114                 realinode = ovl_inode_real(inode);
115         else
116                 realinode = ovl_inode_realdata(inode);
117
118         /* Has it been copied up since we'd opened it? */
119         if (unlikely(file_inode(real->file) != realinode)) {
120                 real->flags = FDPUT_FPUT;
121                 real->file = ovl_open_realfile(file, realinode);
122
123                 return PTR_ERR_OR_ZERO(real->file);
124         }
125
126         /* Did the flags change since open? */
127         if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
128                 return ovl_change_flags(real->file, file->f_flags);
129
130         return 0;
131 }
132
133 static int ovl_real_fdget(const struct file *file, struct fd *real)
134 {
135         if (d_is_dir(file_dentry(file))) {
136                 real->flags = 0;
137                 real->file = ovl_dir_real_file(file, false);
138
139                 return PTR_ERR_OR_ZERO(real->file);
140         }
141
142         return ovl_real_fdget_meta(file, real, false);
143 }
144
145 static int ovl_open(struct inode *inode, struct file *file)
146 {
147         struct file *realfile;
148         int err;
149
150         err = ovl_maybe_copy_up(file_dentry(file), file->f_flags);
151         if (err)
152                 return err;
153
154         /* No longer need these flags, so don't pass them on to underlying fs */
155         file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
156
157         realfile = ovl_open_realfile(file, ovl_inode_realdata(inode));
158         if (IS_ERR(realfile))
159                 return PTR_ERR(realfile);
160
161         file->private_data = realfile;
162
163         return 0;
164 }
165
166 static int ovl_release(struct inode *inode, struct file *file)
167 {
168         fput(file->private_data);
169
170         return 0;
171 }
172
173 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
174 {
175         struct inode *inode = file_inode(file);
176         struct fd real;
177         const struct cred *old_cred;
178         loff_t ret;
179
180         /*
181          * The two special cases below do not need to involve real fs,
182          * so we can optimizing concurrent callers.
183          */
184         if (offset == 0) {
185                 if (whence == SEEK_CUR)
186                         return file->f_pos;
187
188                 if (whence == SEEK_SET)
189                         return vfs_setpos(file, 0, 0);
190         }
191
192         ret = ovl_real_fdget(file, &real);
193         if (ret)
194                 return ret;
195
196         /*
197          * Overlay file f_pos is the master copy that is preserved
198          * through copy up and modified on read/write, but only real
199          * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
200          * limitations that are more strict than ->s_maxbytes for specific
201          * files, so we use the real file to perform seeks.
202          */
203         ovl_inode_lock(inode);
204         real.file->f_pos = file->f_pos;
205
206         old_cred = ovl_override_creds(inode->i_sb);
207         ret = vfs_llseek(real.file, offset, whence);
208         revert_creds(old_cred);
209
210         file->f_pos = real.file->f_pos;
211         ovl_inode_unlock(inode);
212
213         fdput(real);
214
215         return ret;
216 }
217
218 static void ovl_file_accessed(struct file *file)
219 {
220         struct inode *inode, *upperinode;
221
222         if (file->f_flags & O_NOATIME)
223                 return;
224
225         inode = file_inode(file);
226         upperinode = ovl_inode_upper(inode);
227
228         if (!upperinode)
229                 return;
230
231         if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
232              !timespec64_equal(&inode->i_ctime, &upperinode->i_ctime))) {
233                 inode->i_mtime = upperinode->i_mtime;
234                 inode->i_ctime = upperinode->i_ctime;
235         }
236
237         touch_atime(&file->f_path);
238 }
239
240 static rwf_t ovl_iocb_to_rwf(int ifl)
241 {
242         rwf_t flags = 0;
243
244         if (ifl & IOCB_NOWAIT)
245                 flags |= RWF_NOWAIT;
246         if (ifl & IOCB_HIPRI)
247                 flags |= RWF_HIPRI;
248         if (ifl & IOCB_DSYNC)
249                 flags |= RWF_DSYNC;
250         if (ifl & IOCB_SYNC)
251                 flags |= RWF_SYNC;
252
253         return flags;
254 }
255
256 static inline void ovl_aio_put(struct ovl_aio_req *aio_req)
257 {
258         if (refcount_dec_and_test(&aio_req->ref)) {
259                 fdput(aio_req->fd);
260                 kmem_cache_free(ovl_aio_request_cachep, aio_req);
261         }
262 }
263
264 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
265 {
266         struct kiocb *iocb = &aio_req->iocb;
267         struct kiocb *orig_iocb = aio_req->orig_iocb;
268
269         if (iocb->ki_flags & IOCB_WRITE) {
270                 struct inode *inode = file_inode(orig_iocb->ki_filp);
271
272                 /* Actually acquired in ovl_write_iter() */
273                 __sb_writers_acquired(file_inode(iocb->ki_filp)->i_sb,
274                                       SB_FREEZE_WRITE);
275                 file_end_write(iocb->ki_filp);
276                 ovl_copyattr(ovl_inode_real(inode), inode);
277         }
278
279         orig_iocb->ki_pos = iocb->ki_pos;
280         ovl_aio_put(aio_req);
281 }
282
283 static void ovl_aio_rw_complete(struct kiocb *iocb, long res)
284 {
285         struct ovl_aio_req *aio_req = container_of(iocb,
286                                                    struct ovl_aio_req, iocb);
287         struct kiocb *orig_iocb = aio_req->orig_iocb;
288
289         ovl_aio_cleanup_handler(aio_req);
290         orig_iocb->ki_complete(orig_iocb, res);
291 }
292
293 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
294 {
295         struct file *file = iocb->ki_filp;
296         struct fd real;
297         const struct cred *old_cred;
298         ssize_t ret;
299
300         if (!iov_iter_count(iter))
301                 return 0;
302
303         ret = ovl_real_fdget(file, &real);
304         if (ret)
305                 return ret;
306
307         ret = -EINVAL;
308         if (iocb->ki_flags & IOCB_DIRECT &&
309             (!real.file->f_mapping->a_ops ||
310              !real.file->f_mapping->a_ops->direct_IO))
311                 goto out_fdput;
312
313         old_cred = ovl_override_creds(file_inode(file)->i_sb);
314         if (is_sync_kiocb(iocb)) {
315                 ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
316                                     ovl_iocb_to_rwf(iocb->ki_flags));
317         } else {
318                 struct ovl_aio_req *aio_req;
319
320                 ret = -ENOMEM;
321                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
322                 if (!aio_req)
323                         goto out;
324
325                 aio_req->fd = real;
326                 real.flags = 0;
327                 aio_req->orig_iocb = iocb;
328                 kiocb_clone(&aio_req->iocb, iocb, real.file);
329                 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
330                 refcount_set(&aio_req->ref, 2);
331                 ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
332                 ovl_aio_put(aio_req);
333                 if (ret != -EIOCBQUEUED)
334                         ovl_aio_cleanup_handler(aio_req);
335         }
336 out:
337         revert_creds(old_cred);
338         ovl_file_accessed(file);
339 out_fdput:
340         fdput(real);
341
342         return ret;
343 }
344
345 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
346 {
347         struct file *file = iocb->ki_filp;
348         struct inode *inode = file_inode(file);
349         struct fd real;
350         const struct cred *old_cred;
351         ssize_t ret;
352         int ifl = iocb->ki_flags;
353
354         if (!iov_iter_count(iter))
355                 return 0;
356
357         inode_lock(inode);
358         /* Update mode */
359         ovl_copyattr(ovl_inode_real(inode), inode);
360         ret = file_remove_privs(file);
361         if (ret)
362                 goto out_unlock;
363
364         ret = ovl_real_fdget(file, &real);
365         if (ret)
366                 goto out_unlock;
367
368         ret = -EINVAL;
369         if (iocb->ki_flags & IOCB_DIRECT &&
370             (!real.file->f_mapping->a_ops ||
371              !real.file->f_mapping->a_ops->direct_IO))
372                 goto out_fdput;
373
374         if (!ovl_should_sync(OVL_FS(inode->i_sb)))
375                 ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
376
377         old_cred = ovl_override_creds(file_inode(file)->i_sb);
378         if (is_sync_kiocb(iocb)) {
379                 file_start_write(real.file);
380                 ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
381                                      ovl_iocb_to_rwf(ifl));
382                 file_end_write(real.file);
383                 /* Update size */
384                 ovl_copyattr(ovl_inode_real(inode), inode);
385         } else {
386                 struct ovl_aio_req *aio_req;
387
388                 ret = -ENOMEM;
389                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
390                 if (!aio_req)
391                         goto out;
392
393                 file_start_write(real.file);
394                 /* Pacify lockdep, same trick as done in aio_write() */
395                 __sb_writers_release(file_inode(real.file)->i_sb,
396                                      SB_FREEZE_WRITE);
397                 aio_req->fd = real;
398                 real.flags = 0;
399                 aio_req->orig_iocb = iocb;
400                 kiocb_clone(&aio_req->iocb, iocb, real.file);
401                 aio_req->iocb.ki_flags = ifl;
402                 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
403                 refcount_set(&aio_req->ref, 2);
404                 ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
405                 ovl_aio_put(aio_req);
406                 if (ret != -EIOCBQUEUED)
407                         ovl_aio_cleanup_handler(aio_req);
408         }
409 out:
410         revert_creds(old_cred);
411 out_fdput:
412         fdput(real);
413
414 out_unlock:
415         inode_unlock(inode);
416
417         return ret;
418 }
419
420 /*
421  * Calling iter_file_splice_write() directly from overlay's f_op may deadlock
422  * due to lock order inversion between pipe->mutex in iter_file_splice_write()
423  * and file_start_write(real.file) in ovl_write_iter().
424  *
425  * So do everything ovl_write_iter() does and call iter_file_splice_write() on
426  * the real file.
427  */
428 static ssize_t ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
429                                 loff_t *ppos, size_t len, unsigned int flags)
430 {
431         struct fd real;
432         const struct cred *old_cred;
433         struct inode *inode = file_inode(out);
434         struct inode *realinode = ovl_inode_real(inode);
435         ssize_t ret;
436
437         inode_lock(inode);
438         /* Update mode */
439         ovl_copyattr(realinode, inode);
440         ret = file_remove_privs(out);
441         if (ret)
442                 goto out_unlock;
443
444         ret = ovl_real_fdget(out, &real);
445         if (ret)
446                 goto out_unlock;
447
448         old_cred = ovl_override_creds(inode->i_sb);
449         file_start_write(real.file);
450
451         ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
452
453         file_end_write(real.file);
454         /* Update size */
455         ovl_copyattr(realinode, inode);
456         revert_creds(old_cred);
457         fdput(real);
458
459 out_unlock:
460         inode_unlock(inode);
461
462         return ret;
463 }
464
465 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
466 {
467         struct fd real;
468         const struct cred *old_cred;
469         int ret;
470
471         ret = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
472         if (ret <= 0)
473                 return ret;
474
475         ret = ovl_real_fdget_meta(file, &real, !datasync);
476         if (ret)
477                 return ret;
478
479         /* Don't sync lower file for fear of receiving EROFS error */
480         if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
481                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
482                 ret = vfs_fsync_range(real.file, start, end, datasync);
483                 revert_creds(old_cred);
484         }
485
486         fdput(real);
487
488         return ret;
489 }
490
491 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
492 {
493         struct file *realfile = file->private_data;
494         const struct cred *old_cred;
495         int ret;
496
497         if (!realfile->f_op->mmap)
498                 return -ENODEV;
499
500         if (WARN_ON(file != vma->vm_file))
501                 return -EIO;
502
503         vma_set_file(vma, realfile);
504
505         old_cred = ovl_override_creds(file_inode(file)->i_sb);
506         ret = call_mmap(vma->vm_file, vma);
507         revert_creds(old_cred);
508         ovl_file_accessed(file);
509
510         return ret;
511 }
512
513 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
514 {
515         struct inode *inode = file_inode(file);
516         struct fd real;
517         const struct cred *old_cred;
518         int ret;
519
520         ret = ovl_real_fdget(file, &real);
521         if (ret)
522                 return ret;
523
524         old_cred = ovl_override_creds(file_inode(file)->i_sb);
525         ret = vfs_fallocate(real.file, mode, offset, len);
526         revert_creds(old_cred);
527
528         /* Update size */
529         ovl_copyattr(ovl_inode_real(inode), inode);
530
531         fdput(real);
532
533         return ret;
534 }
535
536 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
537 {
538         struct fd real;
539         const struct cred *old_cred;
540         int ret;
541
542         ret = ovl_real_fdget(file, &real);
543         if (ret)
544                 return ret;
545
546         old_cred = ovl_override_creds(file_inode(file)->i_sb);
547         ret = vfs_fadvise(real.file, offset, len, advice);
548         revert_creds(old_cred);
549
550         fdput(real);
551
552         return ret;
553 }
554
555 enum ovl_copyop {
556         OVL_COPY,
557         OVL_CLONE,
558         OVL_DEDUPE,
559 };
560
561 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
562                             struct file *file_out, loff_t pos_out,
563                             loff_t len, unsigned int flags, enum ovl_copyop op)
564 {
565         struct inode *inode_out = file_inode(file_out);
566         struct fd real_in, real_out;
567         const struct cred *old_cred;
568         loff_t ret;
569
570         ret = ovl_real_fdget(file_out, &real_out);
571         if (ret)
572                 return ret;
573
574         ret = ovl_real_fdget(file_in, &real_in);
575         if (ret) {
576                 fdput(real_out);
577                 return ret;
578         }
579
580         old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
581         switch (op) {
582         case OVL_COPY:
583                 ret = vfs_copy_file_range(real_in.file, pos_in,
584                                           real_out.file, pos_out, len, flags);
585                 break;
586
587         case OVL_CLONE:
588                 ret = vfs_clone_file_range(real_in.file, pos_in,
589                                            real_out.file, pos_out, len, flags);
590                 break;
591
592         case OVL_DEDUPE:
593                 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
594                                                 real_out.file, pos_out, len,
595                                                 flags);
596                 break;
597         }
598         revert_creds(old_cred);
599
600         /* Update size */
601         ovl_copyattr(ovl_inode_real(inode_out), inode_out);
602
603         fdput(real_in);
604         fdput(real_out);
605
606         return ret;
607 }
608
609 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
610                                    struct file *file_out, loff_t pos_out,
611                                    size_t len, unsigned int flags)
612 {
613         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
614                             OVL_COPY);
615 }
616
617 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
618                                    struct file *file_out, loff_t pos_out,
619                                    loff_t len, unsigned int remap_flags)
620 {
621         enum ovl_copyop op;
622
623         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
624                 return -EINVAL;
625
626         if (remap_flags & REMAP_FILE_DEDUP)
627                 op = OVL_DEDUPE;
628         else
629                 op = OVL_CLONE;
630
631         /*
632          * Don't copy up because of a dedupe request, this wouldn't make sense
633          * most of the time (data would be duplicated instead of deduplicated).
634          */
635         if (op == OVL_DEDUPE &&
636             (!ovl_inode_upper(file_inode(file_in)) ||
637              !ovl_inode_upper(file_inode(file_out))))
638                 return -EPERM;
639
640         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
641                             remap_flags, op);
642 }
643
644 static int ovl_flush(struct file *file, fl_owner_t id)
645 {
646         struct fd real;
647         const struct cred *old_cred;
648         int err;
649
650         err = ovl_real_fdget(file, &real);
651         if (err)
652                 return err;
653
654         if (real.file->f_op->flush) {
655                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
656                 err = real.file->f_op->flush(real.file, id);
657                 revert_creds(old_cred);
658         }
659         fdput(real);
660
661         return err;
662 }
663
664 const struct file_operations ovl_file_operations = {
665         .open           = ovl_open,
666         .release        = ovl_release,
667         .llseek         = ovl_llseek,
668         .read_iter      = ovl_read_iter,
669         .write_iter     = ovl_write_iter,
670         .fsync          = ovl_fsync,
671         .mmap           = ovl_mmap,
672         .fallocate      = ovl_fallocate,
673         .fadvise        = ovl_fadvise,
674         .flush          = ovl_flush,
675         .splice_read    = generic_file_splice_read,
676         .splice_write   = ovl_splice_write,
677
678         .copy_file_range        = ovl_copy_file_range,
679         .remap_file_range       = ovl_remap_file_range,
680 };
681
682 int __init ovl_aio_request_cache_init(void)
683 {
684         ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
685                                                    sizeof(struct ovl_aio_req),
686                                                    0, SLAB_HWCACHE_ALIGN, NULL);
687         if (!ovl_aio_request_cachep)
688                 return -ENOMEM;
689
690         return 0;
691 }
692
693 void ovl_aio_request_cache_destroy(void)
694 {
695         kmem_cache_destroy(ovl_aio_request_cachep);
696 }