Merge tag 'io_uring-6.6-2023-09-28' of git://git.kernel.dk/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 };
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
243         if (file->f_flags & O_NOATIME)
244                 return;
245
246         inode = file_inode(file);
247         upperinode = ovl_inode_upper(inode);
248
249         if (!upperinode)
250                 return;
251
252         ctime = inode_get_ctime(inode);
253         uctime = inode_get_ctime(upperinode);
254         if ((!timespec64_equal(&inode->i_mtime, &upperinode->i_mtime) ||
255              !timespec64_equal(&ctime, &uctime))) {
256                 inode->i_mtime = upperinode->i_mtime;
257                 inode_set_ctime_to_ts(inode, uctime);
258         }
259
260         touch_atime(&file->f_path);
261 }
262
263 static rwf_t ovl_iocb_to_rwf(int ifl)
264 {
265         rwf_t flags = 0;
266
267         if (ifl & IOCB_NOWAIT)
268                 flags |= RWF_NOWAIT;
269         if (ifl & IOCB_HIPRI)
270                 flags |= RWF_HIPRI;
271         if (ifl & IOCB_DSYNC)
272                 flags |= RWF_DSYNC;
273         if (ifl & IOCB_SYNC)
274                 flags |= RWF_SYNC;
275
276         return flags;
277 }
278
279 static inline void ovl_aio_put(struct ovl_aio_req *aio_req)
280 {
281         if (refcount_dec_and_test(&aio_req->ref)) {
282                 fput(aio_req->iocb.ki_filp);
283                 kmem_cache_free(ovl_aio_request_cachep, aio_req);
284         }
285 }
286
287 static void ovl_aio_cleanup_handler(struct ovl_aio_req *aio_req)
288 {
289         struct kiocb *iocb = &aio_req->iocb;
290         struct kiocb *orig_iocb = aio_req->orig_iocb;
291
292         if (iocb->ki_flags & IOCB_WRITE) {
293                 struct inode *inode = file_inode(orig_iocb->ki_filp);
294
295                 kiocb_end_write(iocb);
296                 ovl_copyattr(inode);
297         }
298
299         orig_iocb->ki_pos = iocb->ki_pos;
300         ovl_aio_put(aio_req);
301 }
302
303 static void ovl_aio_rw_complete(struct kiocb *iocb, long res)
304 {
305         struct ovl_aio_req *aio_req = container_of(iocb,
306                                                    struct ovl_aio_req, iocb);
307         struct kiocb *orig_iocb = aio_req->orig_iocb;
308
309         ovl_aio_cleanup_handler(aio_req);
310         orig_iocb->ki_complete(orig_iocb, res);
311 }
312
313 static ssize_t ovl_read_iter(struct kiocb *iocb, struct iov_iter *iter)
314 {
315         struct file *file = iocb->ki_filp;
316         struct fd real;
317         const struct cred *old_cred;
318         ssize_t ret;
319
320         if (!iov_iter_count(iter))
321                 return 0;
322
323         ret = ovl_real_fdget(file, &real);
324         if (ret)
325                 return ret;
326
327         ret = -EINVAL;
328         if (iocb->ki_flags & IOCB_DIRECT &&
329             !(real.file->f_mode & FMODE_CAN_ODIRECT))
330                 goto out_fdput;
331
332         old_cred = ovl_override_creds(file_inode(file)->i_sb);
333         if (is_sync_kiocb(iocb)) {
334                 ret = vfs_iter_read(real.file, iter, &iocb->ki_pos,
335                                     ovl_iocb_to_rwf(iocb->ki_flags));
336         } else {
337                 struct ovl_aio_req *aio_req;
338
339                 ret = -ENOMEM;
340                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
341                 if (!aio_req)
342                         goto out;
343
344                 real.flags = 0;
345                 aio_req->orig_iocb = iocb;
346                 kiocb_clone(&aio_req->iocb, iocb, get_file(real.file));
347                 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
348                 refcount_set(&aio_req->ref, 2);
349                 ret = vfs_iocb_iter_read(real.file, &aio_req->iocb, iter);
350                 ovl_aio_put(aio_req);
351                 if (ret != -EIOCBQUEUED)
352                         ovl_aio_cleanup_handler(aio_req);
353         }
354 out:
355         revert_creds(old_cred);
356         ovl_file_accessed(file);
357 out_fdput:
358         fdput(real);
359
360         return ret;
361 }
362
363 static ssize_t ovl_write_iter(struct kiocb *iocb, struct iov_iter *iter)
364 {
365         struct file *file = iocb->ki_filp;
366         struct inode *inode = file_inode(file);
367         struct fd real;
368         const struct cred *old_cred;
369         ssize_t ret;
370         int ifl = iocb->ki_flags;
371
372         if (!iov_iter_count(iter))
373                 return 0;
374
375         inode_lock(inode);
376         /* Update mode */
377         ovl_copyattr(inode);
378         ret = file_remove_privs(file);
379         if (ret)
380                 goto out_unlock;
381
382         ret = ovl_real_fdget(file, &real);
383         if (ret)
384                 goto out_unlock;
385
386         ret = -EINVAL;
387         if (iocb->ki_flags & IOCB_DIRECT &&
388             !(real.file->f_mode & FMODE_CAN_ODIRECT))
389                 goto out_fdput;
390
391         if (!ovl_should_sync(OVL_FS(inode->i_sb)))
392                 ifl &= ~(IOCB_DSYNC | IOCB_SYNC);
393
394         /*
395          * Overlayfs doesn't support deferred completions, don't copy
396          * this property in case it is set by the issuer.
397          */
398         ifl &= ~IOCB_DIO_CALLER_COMP;
399
400         old_cred = ovl_override_creds(file_inode(file)->i_sb);
401         if (is_sync_kiocb(iocb)) {
402                 file_start_write(real.file);
403                 ret = vfs_iter_write(real.file, iter, &iocb->ki_pos,
404                                      ovl_iocb_to_rwf(ifl));
405                 file_end_write(real.file);
406                 /* Update size */
407                 ovl_copyattr(inode);
408         } else {
409                 struct ovl_aio_req *aio_req;
410
411                 ret = -ENOMEM;
412                 aio_req = kmem_cache_zalloc(ovl_aio_request_cachep, GFP_KERNEL);
413                 if (!aio_req)
414                         goto out;
415
416                 real.flags = 0;
417                 aio_req->orig_iocb = iocb;
418                 kiocb_clone(&aio_req->iocb, iocb, get_file(real.file));
419                 aio_req->iocb.ki_flags = ifl;
420                 aio_req->iocb.ki_complete = ovl_aio_rw_complete;
421                 refcount_set(&aio_req->ref, 2);
422                 kiocb_start_write(&aio_req->iocb);
423                 ret = vfs_iocb_iter_write(real.file, &aio_req->iocb, iter);
424                 ovl_aio_put(aio_req);
425                 if (ret != -EIOCBQUEUED)
426                         ovl_aio_cleanup_handler(aio_req);
427         }
428 out:
429         revert_creds(old_cred);
430 out_fdput:
431         fdput(real);
432
433 out_unlock:
434         inode_unlock(inode);
435
436         return ret;
437 }
438
439 static ssize_t ovl_splice_read(struct file *in, loff_t *ppos,
440                                struct pipe_inode_info *pipe, size_t len,
441                                unsigned int flags)
442 {
443         const struct cred *old_cred;
444         struct fd real;
445         ssize_t ret;
446
447         ret = ovl_real_fdget(in, &real);
448         if (ret)
449                 return ret;
450
451         old_cred = ovl_override_creds(file_inode(in)->i_sb);
452         ret = vfs_splice_read(real.file, ppos, pipe, len, flags);
453         revert_creds(old_cred);
454         ovl_file_accessed(in);
455
456         fdput(real);
457         return ret;
458 }
459
460 /*
461  * Calling iter_file_splice_write() directly from overlay's f_op may deadlock
462  * due to lock order inversion between pipe->mutex in iter_file_splice_write()
463  * and file_start_write(real.file) in ovl_write_iter().
464  *
465  * So do everything ovl_write_iter() does and call iter_file_splice_write() on
466  * the real file.
467  */
468 static ssize_t ovl_splice_write(struct pipe_inode_info *pipe, struct file *out,
469                                 loff_t *ppos, size_t len, unsigned int flags)
470 {
471         struct fd real;
472         const struct cred *old_cred;
473         struct inode *inode = file_inode(out);
474         ssize_t ret;
475
476         inode_lock(inode);
477         /* Update mode */
478         ovl_copyattr(inode);
479         ret = file_remove_privs(out);
480         if (ret)
481                 goto out_unlock;
482
483         ret = ovl_real_fdget(out, &real);
484         if (ret)
485                 goto out_unlock;
486
487         old_cred = ovl_override_creds(inode->i_sb);
488         file_start_write(real.file);
489
490         ret = iter_file_splice_write(pipe, real.file, ppos, len, flags);
491
492         file_end_write(real.file);
493         /* Update size */
494         ovl_copyattr(inode);
495         revert_creds(old_cred);
496         fdput(real);
497
498 out_unlock:
499         inode_unlock(inode);
500
501         return ret;
502 }
503
504 static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync)
505 {
506         struct fd real;
507         const struct cred *old_cred;
508         int ret;
509
510         ret = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
511         if (ret <= 0)
512                 return ret;
513
514         ret = ovl_real_fdget_meta(file, &real, !datasync);
515         if (ret)
516                 return ret;
517
518         /* Don't sync lower file for fear of receiving EROFS error */
519         if (file_inode(real.file) == ovl_inode_upper(file_inode(file))) {
520                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
521                 ret = vfs_fsync_range(real.file, start, end, datasync);
522                 revert_creds(old_cred);
523         }
524
525         fdput(real);
526
527         return ret;
528 }
529
530 static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
531 {
532         struct file *realfile = file->private_data;
533         const struct cred *old_cred;
534         int ret;
535
536         if (!realfile->f_op->mmap)
537                 return -ENODEV;
538
539         if (WARN_ON(file != vma->vm_file))
540                 return -EIO;
541
542         vma_set_file(vma, realfile);
543
544         old_cred = ovl_override_creds(file_inode(file)->i_sb);
545         ret = call_mmap(vma->vm_file, vma);
546         revert_creds(old_cred);
547         ovl_file_accessed(file);
548
549         return ret;
550 }
551
552 static long ovl_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
553 {
554         struct inode *inode = file_inode(file);
555         struct fd real;
556         const struct cred *old_cred;
557         int ret;
558
559         inode_lock(inode);
560         /* Update mode */
561         ovl_copyattr(inode);
562         ret = file_remove_privs(file);
563         if (ret)
564                 goto out_unlock;
565
566         ret = ovl_real_fdget(file, &real);
567         if (ret)
568                 goto out_unlock;
569
570         old_cred = ovl_override_creds(file_inode(file)->i_sb);
571         ret = vfs_fallocate(real.file, mode, offset, len);
572         revert_creds(old_cred);
573
574         /* Update size */
575         ovl_copyattr(inode);
576
577         fdput(real);
578
579 out_unlock:
580         inode_unlock(inode);
581
582         return ret;
583 }
584
585 static int ovl_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
586 {
587         struct fd real;
588         const struct cred *old_cred;
589         int ret;
590
591         ret = ovl_real_fdget(file, &real);
592         if (ret)
593                 return ret;
594
595         old_cred = ovl_override_creds(file_inode(file)->i_sb);
596         ret = vfs_fadvise(real.file, offset, len, advice);
597         revert_creds(old_cred);
598
599         fdput(real);
600
601         return ret;
602 }
603
604 enum ovl_copyop {
605         OVL_COPY,
606         OVL_CLONE,
607         OVL_DEDUPE,
608 };
609
610 static loff_t ovl_copyfile(struct file *file_in, loff_t pos_in,
611                             struct file *file_out, loff_t pos_out,
612                             loff_t len, unsigned int flags, enum ovl_copyop op)
613 {
614         struct inode *inode_out = file_inode(file_out);
615         struct fd real_in, real_out;
616         const struct cred *old_cred;
617         loff_t ret;
618
619         inode_lock(inode_out);
620         if (op != OVL_DEDUPE) {
621                 /* Update mode */
622                 ovl_copyattr(inode_out);
623                 ret = file_remove_privs(file_out);
624                 if (ret)
625                         goto out_unlock;
626         }
627
628         ret = ovl_real_fdget(file_out, &real_out);
629         if (ret)
630                 goto out_unlock;
631
632         ret = ovl_real_fdget(file_in, &real_in);
633         if (ret) {
634                 fdput(real_out);
635                 goto out_unlock;
636         }
637
638         old_cred = ovl_override_creds(file_inode(file_out)->i_sb);
639         switch (op) {
640         case OVL_COPY:
641                 ret = vfs_copy_file_range(real_in.file, pos_in,
642                                           real_out.file, pos_out, len, flags);
643                 break;
644
645         case OVL_CLONE:
646                 ret = vfs_clone_file_range(real_in.file, pos_in,
647                                            real_out.file, pos_out, len, flags);
648                 break;
649
650         case OVL_DEDUPE:
651                 ret = vfs_dedupe_file_range_one(real_in.file, pos_in,
652                                                 real_out.file, pos_out, len,
653                                                 flags);
654                 break;
655         }
656         revert_creds(old_cred);
657
658         /* Update size */
659         ovl_copyattr(inode_out);
660
661         fdput(real_in);
662         fdput(real_out);
663
664 out_unlock:
665         inode_unlock(inode_out);
666
667         return ret;
668 }
669
670 static ssize_t ovl_copy_file_range(struct file *file_in, loff_t pos_in,
671                                    struct file *file_out, loff_t pos_out,
672                                    size_t len, unsigned int flags)
673 {
674         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len, flags,
675                             OVL_COPY);
676 }
677
678 static loff_t ovl_remap_file_range(struct file *file_in, loff_t pos_in,
679                                    struct file *file_out, loff_t pos_out,
680                                    loff_t len, unsigned int remap_flags)
681 {
682         enum ovl_copyop op;
683
684         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
685                 return -EINVAL;
686
687         if (remap_flags & REMAP_FILE_DEDUP)
688                 op = OVL_DEDUPE;
689         else
690                 op = OVL_CLONE;
691
692         /*
693          * Don't copy up because of a dedupe request, this wouldn't make sense
694          * most of the time (data would be duplicated instead of deduplicated).
695          */
696         if (op == OVL_DEDUPE &&
697             (!ovl_inode_upper(file_inode(file_in)) ||
698              !ovl_inode_upper(file_inode(file_out))))
699                 return -EPERM;
700
701         return ovl_copyfile(file_in, pos_in, file_out, pos_out, len,
702                             remap_flags, op);
703 }
704
705 static int ovl_flush(struct file *file, fl_owner_t id)
706 {
707         struct fd real;
708         const struct cred *old_cred;
709         int err;
710
711         err = ovl_real_fdget(file, &real);
712         if (err)
713                 return err;
714
715         if (real.file->f_op->flush) {
716                 old_cred = ovl_override_creds(file_inode(file)->i_sb);
717                 err = real.file->f_op->flush(real.file, id);
718                 revert_creds(old_cred);
719         }
720         fdput(real);
721
722         return err;
723 }
724
725 const struct file_operations ovl_file_operations = {
726         .open           = ovl_open,
727         .release        = ovl_release,
728         .llseek         = ovl_llseek,
729         .read_iter      = ovl_read_iter,
730         .write_iter     = ovl_write_iter,
731         .fsync          = ovl_fsync,
732         .mmap           = ovl_mmap,
733         .fallocate      = ovl_fallocate,
734         .fadvise        = ovl_fadvise,
735         .flush          = ovl_flush,
736         .splice_read    = ovl_splice_read,
737         .splice_write   = ovl_splice_write,
738
739         .copy_file_range        = ovl_copy_file_range,
740         .remap_file_range       = ovl_remap_file_range,
741 };
742
743 int __init ovl_aio_request_cache_init(void)
744 {
745         ovl_aio_request_cachep = kmem_cache_create("ovl_aio_req",
746                                                    sizeof(struct ovl_aio_req),
747                                                    0, SLAB_HWCACHE_ALIGN, NULL);
748         if (!ovl_aio_request_cachep)
749                 return -ENOMEM;
750
751         return 0;
752 }
753
754 void ovl_aio_request_cache_destroy(void)
755 {
756         kmem_cache_destroy(ovl_aio_request_cachep);
757 }