Merge tag 'locking-core-2023-10-28' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6-microblaze.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 };
23
24 static struct kmem_cache *ovl_aio_request_cachep;
25
26 static char ovl_whatisit(struct inode *inode, struct inode *realinode)
27 {
28         if (realinode != ovl_inode_upper(inode))
29                 return 'l';
30         if (ovl_has_upperdata(inode))
31                 return 'u';
32         else
33                 return 'm';
34 }
35
36 /* No atime modification on underlying */
37 #define OVL_OPEN_FLAGS (O_NOATIME)
38
39 static struct file *ovl_open_realfile(const struct file *file,
40                                       const struct path *realpath)
41 {
42         struct inode *realinode = d_inode(realpath->dentry);
43         struct inode *inode = file_inode(file);
44         struct mnt_idmap *real_idmap;
45         struct file *realfile;
46         const struct cred *old_cred;
47         int flags = file->f_flags | OVL_OPEN_FLAGS;
48         int acc_mode = ACC_MODE(flags);
49         int err;
50
51         if (flags & O_APPEND)
52                 acc_mode |= MAY_APPEND;
53
54         old_cred = ovl_override_creds(inode->i_sb);
55         real_idmap = mnt_idmap(realpath->mnt);
56         err = inode_permission(real_idmap, realinode, MAY_OPEN | acc_mode);
57         if (err) {
58                 realfile = ERR_PTR(err);
59         } else {
60                 if (!inode_owner_or_capable(real_idmap, realinode))
61                         flags &= ~O_NOATIME;
62
63                 realfile = backing_file_open(&file->f_path, flags, realpath,
64                                              current_cred());
65         }
66         revert_creds(old_cred);
67
68         pr_debug("open(%p[%pD2/%c], 0%o) -> (%p, 0%o)\n",
69                  file, file, ovl_whatisit(inode, realinode), file->f_flags,
70                  realfile, IS_ERR(realfile) ? 0 : realfile->f_flags);
71
72         return realfile;
73 }
74
75 #define OVL_SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT)
76
77 static int ovl_change_flags(struct file *file, unsigned int flags)
78 {
79         struct inode *inode = file_inode(file);
80         int err;
81
82         flags &= OVL_SETFL_MASK;
83
84         if (((flags ^ file->f_flags) & O_APPEND) && IS_APPEND(inode))
85                 return -EPERM;
86
87         if ((flags & O_DIRECT) && !(file->f_mode & FMODE_CAN_ODIRECT))
88                 return -EINVAL;
89
90         if (file->f_op->check_flags) {
91                 err = file->f_op->check_flags(flags);
92                 if (err)
93                         return err;
94         }
95
96         spin_lock(&file->f_lock);
97         file->f_flags = (file->f_flags & ~OVL_SETFL_MASK) | flags;
98         file->f_iocb_flags = iocb_flags(file);
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 dentry *dentry = file_dentry(file);
108         struct path realpath;
109         int err;
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                 /* lazy lookup and verify of lowerdata */
118                 err = ovl_verify_lowerdata(dentry);
119                 if (err)
120                         return err;
121
122                 ovl_path_realdata(dentry, &realpath);
123         }
124         if (!realpath.dentry)
125                 return -EIO;
126
127         /* Has it been copied up since we'd opened it? */
128         if (unlikely(file_inode(real->file) != d_inode(realpath.dentry))) {
129                 real->flags = FDPUT_FPUT;
130                 real->file = ovl_open_realfile(file, &realpath);
131
132                 return PTR_ERR_OR_ZERO(real->file);
133         }
134
135         /* Did the flags change since open? */
136         if (unlikely((file->f_flags ^ real->file->f_flags) & ~OVL_OPEN_FLAGS))
137                 return ovl_change_flags(real->file, file->f_flags);
138
139         return 0;
140 }
141
142 static int ovl_real_fdget(const struct file *file, struct fd *real)
143 {
144         if (d_is_dir(file_dentry(file))) {
145                 real->flags = 0;
146                 real->file = ovl_dir_real_file(file, false);
147
148                 return PTR_ERR_OR_ZERO(real->file);
149         }
150
151         return ovl_real_fdget_meta(file, real, false);
152 }
153
154 static int ovl_open(struct inode *inode, struct file *file)
155 {
156         struct dentry *dentry = file_dentry(file);
157         struct file *realfile;
158         struct path realpath;
159         int err;
160
161         /* lazy lookup and verify lowerdata */
162         err = ovl_verify_lowerdata(dentry);
163         if (err)
164                 return err;
165
166         err = ovl_maybe_copy_up(dentry, file->f_flags);
167         if (err)
168                 return err;
169
170         /* No longer need these flags, so don't pass them on to underlying fs */
171         file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
172
173         ovl_path_realdata(dentry, &realpath);
174         if (!realpath.dentry)
175                 return -EIO;
176
177         realfile = ovl_open_realfile(file, &realpath);
178         if (IS_ERR(realfile))
179                 return PTR_ERR(realfile);
180
181         file->private_data = realfile;
182
183         return 0;
184 }
185
186 static int ovl_release(struct inode *inode, struct file *file)
187 {
188         fput(file->private_data);
189
190         return 0;
191 }
192
193 static loff_t ovl_llseek(struct file *file, loff_t offset, int whence)
194 {
195         struct inode *inode = file_inode(file);
196         struct fd real;
197         const struct cred *old_cred;
198         loff_t ret;
199
200         /*
201          * The two special cases below do not need to involve real fs,
202          * so we can optimizing concurrent callers.
203          */
204         if (offset == 0) {
205                 if (whence == SEEK_CUR)
206                         return file->f_pos;
207
208                 if (whence == SEEK_SET)
209                         return vfs_setpos(file, 0, 0);
210         }
211
212         ret = ovl_real_fdget(file, &real);
213         if (ret)
214                 return ret;
215
216         /*
217          * Overlay file f_pos is the master copy that is preserved
218          * through copy up and modified on read/write, but only real
219          * fs knows how to SEEK_HOLE/SEEK_DATA and real fs may impose
220          * limitations that are more strict than ->s_maxbytes for specific
221          * files, so we use the real file to perform seeks.
222          */
223         ovl_inode_lock(inode);
224         real.file->f_pos = file->f_pos;
225
226         old_cred = ovl_override_creds(inode->i_sb);
227         ret = vfs_llseek(real.file, offset, whence);
228         revert_creds(old_cred);
229
230         file->f_pos = real.file->f_pos;
231         ovl_inode_unlock(inode);
232
233         fdput(real);
234
235         return ret;
236 }
237
238 static void ovl_file_accessed(struct file *file)
239 {
240         struct inode *inode, *upperinode;
241         struct timespec64 ctime, uctime;
242         struct timespec64 mtime, umtime;
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         mtime = inode_get_mtime(inode);
256         umtime = inode_get_mtime(upperinode);
257         if ((!timespec64_equal(&mtime, &umtime)) ||
258              !timespec64_equal(&ctime, &uctime)) {
259                 inode_set_mtime_to_ts(inode, inode_get_mtime(upperinode));
260                 inode_set_ctime_to_ts(inode, uctime);
261         }
262
263         touch_atime(&file->f_path);
264 }
265
266 static rwf_t ovl_iocb_to_rwf(int ifl)
267 {
268         rwf_t flags = 0;
269
270         if (ifl & IOCB_NOWAIT)
271                 flags |= RWF_NOWAIT;
272         if (ifl & IOCB_HIPRI)
273                 flags |= RWF_HIPRI;
274         if (ifl & IOCB_DSYNC)
275                 flags |= RWF_DSYNC;
276         if (ifl & IOCB_SYNC)
277                 flags |= RWF_SYNC;
278
279         return flags;
280 }
281
282 static inline void ovl_aio_put(struct ovl_aio_req *aio_req)
283 {
284         if (refcount_dec_and_test(&aio_req->ref)) {
285                 fput(aio_req->iocb.ki_filp);
286                 kmem_cache_free(ovl_aio_request_cachep, aio_req);
287         }
288 }
289
290 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
291 {
292         struct kiocb *iocb = &aio_req->iocb;
293         struct kiocb *orig_iocb = aio_req->orig_iocb;
294
295         if (iocb->ki_flags & IOCB_WRITE) {
296                 struct inode *inode = file_inode(orig_iocb->ki_filp);
297
298                 kiocb_end_write(iocb);
299                 ovl_copyattr(inode);
300         }
301
302         orig_iocb->ki_pos = iocb->ki_pos;
303         ovl_aio_put(aio_req);
304 }
305
306 static void ovl_aio_rw_complete(struct kiocb *iocb, long res)
307 {
308         struct ovl_aio_req *aio_req = container_of(iocb,
309                                                    struct ovl_aio_req, iocb);
310         struct kiocb *orig_iocb = aio_req->orig_iocb;
311
312         ovl_aio_cleanup_handler(aio_req);
313         orig_iocb->ki_complete(orig_iocb, res);
314 }
315
316 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
317 {
318         struct file *file = iocb->ki_filp;
319         struct fd real;
320         const struct cred *old_cred;
321         ssize_t ret;
322
323         if (!iov_iter_count(iter))
324                 return 0;
325
326         ret = ovl_real_fdget(file, &real);
327         if (ret)
328                 return ret;
329
330         ret = -EINVAL;
331         if (iocb->ki_flags & IOCB_DIRECT &&
332             !(real.file->f_mode & FMODE_CAN_ODIRECT))
333                 goto out_fdput;
334
335         old_cred = ovl_override_creds(file_inode(file)->i_sb);
336         if (is_sync_kiocb(iocb)) {
337                 ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
338                                     ovl_iocb_to_rwf(iocb->ki_flags));
339         } else {
340                 struct ovl_aio_req *aio_req;
341
342                 ret = -ENOMEM;
343                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
344                 if (!aio_req)
345                         goto out;
346
347                 aio_req->orig_iocb = iocb;
348                 kiocb_clone(&aio_req->iocb, iocb, get_file(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         /*
397          * Overlayfs doesn't support deferred completions, don't copy
398          * this property in case it is set by the issuer.
399          */
400         ifl &= ~IOCB_DIO_CALLER_COMP;
401
402         old_cred = ovl_override_creds(file_inode(file)->i_sb);
403         if (is_sync_kiocb(iocb)) {
404                 file_start_write(real.file);
405                 ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
406                                      ovl_iocb_to_rwf(ifl));
407                 file_end_write(real.file);
408                 /* Update size */
409                 ovl_copyattr(inode);
410         } else {
411                 struct ovl_aio_req *aio_req;
412
413                 ret = -ENOMEM;
414                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
415                 if (!aio_req)
416                         goto out;
417
418                 aio_req->orig_iocb = iocb;
419                 kiocb_clone(&aio_req->iocb, iocb, get_file(real.file));
420                 aio_req->iocb.ki_flags = ifl;
421                 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
422                 refcount_set(&aio_req->ref, 2);
423                 kiocb_start_write(&aio_req->iocb);
424                 ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
425                 ovl_aio_put(aio_req);
426                 if (ret != -EIOCBQUEUED)
427                         ovl_aio_cleanup_handler(aio_req);
428         }
429 out:
430         revert_creds(old_cred);
431 out_fdput:
432         fdput(real);
433
434 out_unlock:
435         inode_unlock(inode);
436
437         return ret;
438 }
439
440 static ssize_t ovl_splice_read(struct file *in, loff_t *ppos,
441                                struct pipe_inode_info *pipe, size_t len,
442                                unsigned int flags)
443 {
444         const struct cred *old_cred;
445         struct fd real;
446         ssize_t ret;
447
448         ret = ovl_real_fdget(in, &real);
449         if (ret)
450                 return ret;
451
452         old_cred = ovl_override_creds(file_inode(in)->i_sb);
453         ret = vfs_splice_read(real.file, ppos, pipe, len, flags);
454         revert_creds(old_cred);
455         ovl_file_accessed(in);
456
457         fdput(real);
458         return ret;
459 }
460
461 /*
462  * Calling iter_file_splice_write() directly from overlay's f_op may deadlock
463  * due to lock order inversion between pipe->mutex in iter_file_splice_write()
464  * and file_start_write(real.file) in ovl_write_iter().
465  *
466  * So do everything ovl_write_iter() does and call iter_file_splice_write() on
467  * the real file.
468  */
469 static ssize_t ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
470                                 loff_t *ppos, size_t len, unsigned int flags)
471 {
472         struct fd real;
473         const struct cred *old_cred;
474         struct inode *inode = file_inode(out);
475         ssize_t ret;
476
477         inode_lock(inode);
478         /* Update mode */
479         ovl_copyattr(inode);
480         ret = file_remove_privs(out);
481         if (ret)
482                 goto out_unlock;
483
484         ret = ovl_real_fdget(out, &real);
485         if (ret)
486                 goto out_unlock;
487
488         old_cred = ovl_override_creds(inode->i_sb);
489         file_start_write(real.file);
490
491         ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
492
493         file_end_write(real.file);
494         /* Update size */
495         ovl_copyattr(inode);
496         revert_creds(old_cred);
497         fdput(real);
498
499 out_unlock:
500         inode_unlock(inode);
501
502         return ret;
503 }
504
505 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
506 {
507         struct fd real;
508         const struct cred *old_cred;
509         int ret;
510
511         ret = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
512         if (ret <= 0)
513                 return ret;
514
515         ret = ovl_real_fdget_meta(file, &real, !datasync);
516         if (ret)
517                 return ret;
518
519         /* Don't sync lower file for fear of receiving EROFS error */
520         if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
521                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
522                 ret = vfs_fsync_range(real.file, start, end, datasync);
523                 revert_creds(old_cred);
524         }
525
526         fdput(real);
527
528         return ret;
529 }
530
531 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
532 {
533         struct file *realfile = file->private_data;
534         const struct cred *old_cred;
535         int ret;
536
537         if (!realfile->f_op->mmap)
538                 return -ENODEV;
539
540         if (WARN_ON(file != vma->vm_file))
541                 return -EIO;
542
543         vma_set_file(vma, realfile);
544
545         old_cred = ovl_override_creds(file_inode(file)->i_sb);
546         ret = call_mmap(vma->vm_file, vma);
547         revert_creds(old_cred);
548         ovl_file_accessed(file);
549
550         return ret;
551 }
552
553 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
554 {
555         struct inode *inode = file_inode(file);
556         struct fd real;
557         const struct cred *old_cred;
558         int ret;
559
560         inode_lock(inode);
561         /* Update mode */
562         ovl_copyattr(inode);
563         ret = file_remove_privs(file);
564         if (ret)
565                 goto out_unlock;
566
567         ret = ovl_real_fdget(file, &real);
568         if (ret)
569                 goto out_unlock;
570
571         old_cred = ovl_override_creds(file_inode(file)->i_sb);
572         ret = vfs_fallocate(real.file, mode, offset, len);
573         revert_creds(old_cred);
574
575         /* Update size */
576         ovl_copyattr(inode);
577
578         fdput(real);
579
580 out_unlock:
581         inode_unlock(inode);
582
583         return ret;
584 }
585
586 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
587 {
588         struct fd real;
589         const struct cred *old_cred;
590         int ret;
591
592         ret = ovl_real_fdget(file, &real);
593         if (ret)
594                 return ret;
595
596         old_cred = ovl_override_creds(file_inode(file)->i_sb);
597         ret = vfs_fadvise(real.file, offset, len, advice);
598         revert_creds(old_cred);
599
600         fdput(real);
601
602         return ret;
603 }
604
605 enum ovl_copyop {
606         OVL_COPY,
607         OVL_CLONE,
608         OVL_DEDUPE,
609 };
610
611 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
612                             struct file *file_out, loff_t pos_out,
613                             loff_t len, unsigned int flags, enum ovl_copyop op)
614 {
615         struct inode *inode_out = file_inode(file_out);
616         struct fd real_in, real_out;
617         const struct cred *old_cred;
618         loff_t ret;
619
620         inode_lock(inode_out);
621         if (op != OVL_DEDUPE) {
622                 /* Update mode */
623                 ovl_copyattr(inode_out);
624                 ret = file_remove_privs(file_out);
625                 if (ret)
626                         goto out_unlock;
627         }
628
629         ret = ovl_real_fdget(file_out, &real_out);
630         if (ret)
631                 goto out_unlock;
632
633         ret = ovl_real_fdget(file_in, &real_in);
634         if (ret) {
635                 fdput(real_out);
636                 goto out_unlock;
637         }
638
639         old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
640         switch (op) {
641         case OVL_COPY:
642                 ret = vfs_copy_file_range(real_in.file, pos_in,
643                                           real_out.file, pos_out, len, flags);
644                 break;
645
646         case OVL_CLONE:
647                 ret = vfs_clone_file_range(real_in.file, pos_in,
648                                            real_out.file, pos_out, len, flags);
649                 break;
650
651         case OVL_DEDUPE:
652                 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
653                                                 real_out.file, pos_out, len,
654                                                 flags);
655                 break;
656         }
657         revert_creds(old_cred);
658
659         /* Update size */
660         ovl_copyattr(inode_out);
661
662         fdput(real_in);
663         fdput(real_out);
664
665 out_unlock:
666         inode_unlock(inode_out);
667
668         return ret;
669 }
670
671 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
672                                    struct file *file_out, loff_t pos_out,
673                                    size_t len, unsigned int flags)
674 {
675         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
676                             OVL_COPY);
677 }
678
679 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
680                                    struct file *file_out, loff_t pos_out,
681                                    loff_t len, unsigned int remap_flags)
682 {
683         enum ovl_copyop op;
684
685         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
686                 return -EINVAL;
687
688         if (remap_flags & REMAP_FILE_DEDUP)
689                 op = OVL_DEDUPE;
690         else
691                 op = OVL_CLONE;
692
693         /*
694          * Don't copy up because of a dedupe request, this wouldn't make sense
695          * most of the time (data would be duplicated instead of deduplicated).
696          */
697         if (op == OVL_DEDUPE &&
698             (!ovl_inode_upper(file_inode(file_in)) ||
699              !ovl_inode_upper(file_inode(file_out))))
700                 return -EPERM;
701
702         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
703                             remap_flags, op);
704 }
705
706 static int ovl_flush(struct file *file, fl_owner_t id)
707 {
708         struct fd real;
709         const struct cred *old_cred;
710         int err;
711
712         err = ovl_real_fdget(file, &real);
713         if (err)
714                 return err;
715
716         if (real.file->f_op->flush) {
717                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
718                 err = real.file->f_op->flush(real.file, id);
719                 revert_creds(old_cred);
720         }
721         fdput(real);
722
723         return err;
724 }
725
726 const struct file_operations ovl_file_operations = {
727         .open           = ovl_open,
728         .release        = ovl_release,
729         .llseek         = ovl_llseek,
730         .read_iter      = ovl_read_iter,
731         .write_iter     = ovl_write_iter,
732         .fsync          = ovl_fsync,
733         .mmap           = ovl_mmap,
734         .fallocate      = ovl_fallocate,
735         .fadvise        = ovl_fadvise,
736         .flush          = ovl_flush,
737         .splice_read    = ovl_splice_read,
738         .splice_write   = ovl_splice_write,
739
740         .copy_file_range        = ovl_copy_file_range,
741         .remap_file_range       = ovl_remap_file_range,
742 };
743
744 int __init ovl_aio_request_cache_init(void)
745 {
746         ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
747                                                    sizeof(struct ovl_aio_req),
748                                                    0, SLAB_HWCACHE_ALIGN, NULL);
749         if (!ovl_aio_request_cachep)
750                 return -ENOMEM;
751
752         return 0;
753 }
754
755 void ovl_aio_request_cache_destroy(void)
756 {
757         kmem_cache_destroy(ovl_aio_request_cachep);
758 }