Loading...
1/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/file.h>
10#include <linux/fdtable.h>
11#include <linux/fsnotify.h>
12#include <linux/module.h>
13#include <linux/tty.h>
14#include <linux/namei.h>
15#include <linux/backing-dev.h>
16#include <linux/capability.h>
17#include <linux/securebits.h>
18#include <linux/security.h>
19#include <linux/mount.h>
20#include <linux/fcntl.h>
21#include <linux/slab.h>
22#include <linux/uaccess.h>
23#include <linux/fs.h>
24#include <linux/personality.h>
25#include <linux/pagemap.h>
26#include <linux/syscalls.h>
27#include <linux/rcupdate.h>
28#include <linux/audit.h>
29#include <linux/falloc.h>
30#include <linux/fs_struct.h>
31#include <linux/ima.h>
32#include <linux/dnotify.h>
33#include <linux/compat.h>
34
35#include "internal.h"
36
37int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
38 struct file *filp)
39{
40 int ret;
41 struct iattr newattrs;
42
43 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
44 if (length < 0)
45 return -EINVAL;
46
47 newattrs.ia_size = length;
48 newattrs.ia_valid = ATTR_SIZE | time_attrs;
49 if (filp) {
50 newattrs.ia_file = filp;
51 newattrs.ia_valid |= ATTR_FILE;
52 }
53
54 /* Remove suid, sgid, and file capabilities on truncate too */
55 ret = dentry_needs_remove_privs(dentry);
56 if (ret < 0)
57 return ret;
58 if (ret)
59 newattrs.ia_valid |= ret | ATTR_FORCE;
60
61 inode_lock(dentry->d_inode);
62 /* Note any delegations or leases have already been broken: */
63 ret = notify_change(dentry, &newattrs, NULL);
64 inode_unlock(dentry->d_inode);
65 return ret;
66}
67
68long vfs_truncate(const struct path *path, loff_t length)
69{
70 struct inode *inode;
71 struct dentry *upperdentry;
72 long error;
73
74 inode = path->dentry->d_inode;
75
76 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
77 if (S_ISDIR(inode->i_mode))
78 return -EISDIR;
79 if (!S_ISREG(inode->i_mode))
80 return -EINVAL;
81
82 error = mnt_want_write(path->mnt);
83 if (error)
84 goto out;
85
86 error = inode_permission(inode, MAY_WRITE);
87 if (error)
88 goto mnt_drop_write_and_out;
89
90 error = -EPERM;
91 if (IS_APPEND(inode))
92 goto mnt_drop_write_and_out;
93
94 /*
95 * If this is an overlayfs then do as if opening the file so we get
96 * write access on the upper inode, not on the overlay inode. For
97 * non-overlay filesystems d_real() is an identity function.
98 */
99 upperdentry = d_real(path->dentry, NULL, O_WRONLY);
100 error = PTR_ERR(upperdentry);
101 if (IS_ERR(upperdentry))
102 goto mnt_drop_write_and_out;
103
104 error = get_write_access(upperdentry->d_inode);
105 if (error)
106 goto mnt_drop_write_and_out;
107
108 /*
109 * Make sure that there are no leases. get_write_access() protects
110 * against the truncate racing with a lease-granting setlease().
111 */
112 error = break_lease(inode, O_WRONLY);
113 if (error)
114 goto put_write_and_out;
115
116 error = locks_verify_truncate(inode, NULL, length);
117 if (!error)
118 error = security_path_truncate(path);
119 if (!error)
120 error = do_truncate(path->dentry, length, 0, NULL);
121
122put_write_and_out:
123 put_write_access(upperdentry->d_inode);
124mnt_drop_write_and_out:
125 mnt_drop_write(path->mnt);
126out:
127 return error;
128}
129EXPORT_SYMBOL_GPL(vfs_truncate);
130
131static long do_sys_truncate(const char __user *pathname, loff_t length)
132{
133 unsigned int lookup_flags = LOOKUP_FOLLOW;
134 struct path path;
135 int error;
136
137 if (length < 0) /* sorry, but loff_t says... */
138 return -EINVAL;
139
140retry:
141 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
142 if (!error) {
143 error = vfs_truncate(&path, length);
144 path_put(&path);
145 }
146 if (retry_estale(error, lookup_flags)) {
147 lookup_flags |= LOOKUP_REVAL;
148 goto retry;
149 }
150 return error;
151}
152
153SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
154{
155 return do_sys_truncate(path, length);
156}
157
158#ifdef CONFIG_COMPAT
159COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
160{
161 return do_sys_truncate(path, length);
162}
163#endif
164
165static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
166{
167 struct inode *inode;
168 struct dentry *dentry;
169 struct fd f;
170 int error;
171
172 error = -EINVAL;
173 if (length < 0)
174 goto out;
175 error = -EBADF;
176 f = fdget(fd);
177 if (!f.file)
178 goto out;
179
180 /* explicitly opened as large or we are on 64-bit box */
181 if (f.file->f_flags & O_LARGEFILE)
182 small = 0;
183
184 dentry = f.file->f_path.dentry;
185 inode = dentry->d_inode;
186 error = -EINVAL;
187 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
188 goto out_putf;
189
190 error = -EINVAL;
191 /* Cannot ftruncate over 2^31 bytes without large file support */
192 if (small && length > MAX_NON_LFS)
193 goto out_putf;
194
195 error = -EPERM;
196 if (IS_APPEND(inode))
197 goto out_putf;
198
199 sb_start_write(inode->i_sb);
200 error = locks_verify_truncate(inode, f.file, length);
201 if (!error)
202 error = security_path_truncate(&f.file->f_path);
203 if (!error)
204 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
205 sb_end_write(inode->i_sb);
206out_putf:
207 fdput(f);
208out:
209 return error;
210}
211
212SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
213{
214 return do_sys_ftruncate(fd, length, 1);
215}
216
217#ifdef CONFIG_COMPAT
218COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
219{
220 return do_sys_ftruncate(fd, length, 1);
221}
222#endif
223
224/* LFS versions of truncate are only needed on 32 bit machines */
225#if BITS_PER_LONG == 32
226SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
227{
228 return do_sys_truncate(path, length);
229}
230
231SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
232{
233 return do_sys_ftruncate(fd, length, 0);
234}
235#endif /* BITS_PER_LONG == 32 */
236
237
238int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
239{
240 struct inode *inode = file_inode(file);
241 long ret;
242
243 if (offset < 0 || len <= 0)
244 return -EINVAL;
245
246 /* Return error if mode is not supported */
247 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
248 return -EOPNOTSUPP;
249
250 /* Punch hole and zero range are mutually exclusive */
251 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
252 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
253 return -EOPNOTSUPP;
254
255 /* Punch hole must have keep size set */
256 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
257 !(mode & FALLOC_FL_KEEP_SIZE))
258 return -EOPNOTSUPP;
259
260 /* Collapse range should only be used exclusively. */
261 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
262 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
263 return -EINVAL;
264
265 /* Insert range should only be used exclusively. */
266 if ((mode & FALLOC_FL_INSERT_RANGE) &&
267 (mode & ~FALLOC_FL_INSERT_RANGE))
268 return -EINVAL;
269
270 /* Unshare range should only be used with allocate mode. */
271 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
272 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
273 return -EINVAL;
274
275 if (!(file->f_mode & FMODE_WRITE))
276 return -EBADF;
277
278 /*
279 * We can only allow pure fallocate on append only files
280 */
281 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
282 return -EPERM;
283
284 if (IS_IMMUTABLE(inode))
285 return -EPERM;
286
287 /*
288 * We cannot allow any fallocate operation on an active swapfile
289 */
290 if (IS_SWAPFILE(inode))
291 return -ETXTBSY;
292
293 /*
294 * Revalidate the write permissions, in case security policy has
295 * changed since the files were opened.
296 */
297 ret = security_file_permission(file, MAY_WRITE);
298 if (ret)
299 return ret;
300
301 if (S_ISFIFO(inode->i_mode))
302 return -ESPIPE;
303
304 /*
305 * Let individual file system decide if it supports preallocation
306 * for directories or not.
307 */
308 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
309 !S_ISBLK(inode->i_mode))
310 return -ENODEV;
311
312 /* Check for wrap through zero too */
313 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
314 return -EFBIG;
315
316 if (!file->f_op->fallocate)
317 return -EOPNOTSUPP;
318
319 sb_start_write(inode->i_sb);
320 ret = file->f_op->fallocate(file, mode, offset, len);
321
322 /*
323 * Create inotify and fanotify events.
324 *
325 * To keep the logic simple always create events if fallocate succeeds.
326 * This implies that events are even created if the file size remains
327 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
328 */
329 if (ret == 0)
330 fsnotify_modify(file);
331
332 sb_end_write(inode->i_sb);
333 return ret;
334}
335EXPORT_SYMBOL_GPL(vfs_fallocate);
336
337SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
338{
339 struct fd f = fdget(fd);
340 int error = -EBADF;
341
342 if (f.file) {
343 error = vfs_fallocate(f.file, mode, offset, len);
344 fdput(f);
345 }
346 return error;
347}
348
349/*
350 * access() needs to use the real uid/gid, not the effective uid/gid.
351 * We do this by temporarily clearing all FS-related capabilities and
352 * switching the fsuid/fsgid around to the real ones.
353 */
354SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
355{
356 const struct cred *old_cred;
357 struct cred *override_cred;
358 struct path path;
359 struct inode *inode;
360 int res;
361 unsigned int lookup_flags = LOOKUP_FOLLOW;
362
363 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
364 return -EINVAL;
365
366 override_cred = prepare_creds();
367 if (!override_cred)
368 return -ENOMEM;
369
370 override_cred->fsuid = override_cred->uid;
371 override_cred->fsgid = override_cred->gid;
372
373 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
374 /* Clear the capabilities if we switch to a non-root user */
375 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
376 if (!uid_eq(override_cred->uid, root_uid))
377 cap_clear(override_cred->cap_effective);
378 else
379 override_cred->cap_effective =
380 override_cred->cap_permitted;
381 }
382
383 old_cred = override_creds(override_cred);
384retry:
385 res = user_path_at(dfd, filename, lookup_flags, &path);
386 if (res)
387 goto out;
388
389 inode = d_backing_inode(path.dentry);
390
391 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
392 /*
393 * MAY_EXEC on regular files is denied if the fs is mounted
394 * with the "noexec" flag.
395 */
396 res = -EACCES;
397 if (path_noexec(&path))
398 goto out_path_release;
399 }
400
401 res = inode_permission(inode, mode | MAY_ACCESS);
402 /* SuS v2 requires we report a read only fs too */
403 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
404 goto out_path_release;
405 /*
406 * This is a rare case where using __mnt_is_readonly()
407 * is OK without a mnt_want/drop_write() pair. Since
408 * no actual write to the fs is performed here, we do
409 * not need to telegraph to that to anyone.
410 *
411 * By doing this, we accept that this access is
412 * inherently racy and know that the fs may change
413 * state before we even see this result.
414 */
415 if (__mnt_is_readonly(path.mnt))
416 res = -EROFS;
417
418out_path_release:
419 path_put(&path);
420 if (retry_estale(res, lookup_flags)) {
421 lookup_flags |= LOOKUP_REVAL;
422 goto retry;
423 }
424out:
425 revert_creds(old_cred);
426 put_cred(override_cred);
427 return res;
428}
429
430SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
431{
432 return sys_faccessat(AT_FDCWD, filename, mode);
433}
434
435SYSCALL_DEFINE1(chdir, const char __user *, filename)
436{
437 struct path path;
438 int error;
439 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
440retry:
441 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
442 if (error)
443 goto out;
444
445 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
446 if (error)
447 goto dput_and_out;
448
449 set_fs_pwd(current->fs, &path);
450
451dput_and_out:
452 path_put(&path);
453 if (retry_estale(error, lookup_flags)) {
454 lookup_flags |= LOOKUP_REVAL;
455 goto retry;
456 }
457out:
458 return error;
459}
460
461SYSCALL_DEFINE1(fchdir, unsigned int, fd)
462{
463 struct fd f = fdget_raw(fd);
464 struct inode *inode;
465 int error = -EBADF;
466
467 error = -EBADF;
468 if (!f.file)
469 goto out;
470
471 inode = file_inode(f.file);
472
473 error = -ENOTDIR;
474 if (!S_ISDIR(inode->i_mode))
475 goto out_putf;
476
477 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
478 if (!error)
479 set_fs_pwd(current->fs, &f.file->f_path);
480out_putf:
481 fdput(f);
482out:
483 return error;
484}
485
486SYSCALL_DEFINE1(chroot, const char __user *, filename)
487{
488 struct path path;
489 int error;
490 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
491retry:
492 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
493 if (error)
494 goto out;
495
496 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
497 if (error)
498 goto dput_and_out;
499
500 error = -EPERM;
501 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
502 goto dput_and_out;
503 error = security_path_chroot(&path);
504 if (error)
505 goto dput_and_out;
506
507 set_fs_root(current->fs, &path);
508 error = 0;
509dput_and_out:
510 path_put(&path);
511 if (retry_estale(error, lookup_flags)) {
512 lookup_flags |= LOOKUP_REVAL;
513 goto retry;
514 }
515out:
516 return error;
517}
518
519static int chmod_common(const struct path *path, umode_t mode)
520{
521 struct inode *inode = path->dentry->d_inode;
522 struct inode *delegated_inode = NULL;
523 struct iattr newattrs;
524 int error;
525
526 error = mnt_want_write(path->mnt);
527 if (error)
528 return error;
529retry_deleg:
530 inode_lock(inode);
531 error = security_path_chmod(path, mode);
532 if (error)
533 goto out_unlock;
534 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
535 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
536 error = notify_change(path->dentry, &newattrs, &delegated_inode);
537out_unlock:
538 inode_unlock(inode);
539 if (delegated_inode) {
540 error = break_deleg_wait(&delegated_inode);
541 if (!error)
542 goto retry_deleg;
543 }
544 mnt_drop_write(path->mnt);
545 return error;
546}
547
548SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
549{
550 struct fd f = fdget(fd);
551 int err = -EBADF;
552
553 if (f.file) {
554 audit_file(f.file);
555 err = chmod_common(&f.file->f_path, mode);
556 fdput(f);
557 }
558 return err;
559}
560
561SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
562{
563 struct path path;
564 int error;
565 unsigned int lookup_flags = LOOKUP_FOLLOW;
566retry:
567 error = user_path_at(dfd, filename, lookup_flags, &path);
568 if (!error) {
569 error = chmod_common(&path, mode);
570 path_put(&path);
571 if (retry_estale(error, lookup_flags)) {
572 lookup_flags |= LOOKUP_REVAL;
573 goto retry;
574 }
575 }
576 return error;
577}
578
579SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
580{
581 return sys_fchmodat(AT_FDCWD, filename, mode);
582}
583
584static int chown_common(const struct path *path, uid_t user, gid_t group)
585{
586 struct inode *inode = path->dentry->d_inode;
587 struct inode *delegated_inode = NULL;
588 int error;
589 struct iattr newattrs;
590 kuid_t uid;
591 kgid_t gid;
592
593 uid = make_kuid(current_user_ns(), user);
594 gid = make_kgid(current_user_ns(), group);
595
596retry_deleg:
597 newattrs.ia_valid = ATTR_CTIME;
598 if (user != (uid_t) -1) {
599 if (!uid_valid(uid))
600 return -EINVAL;
601 newattrs.ia_valid |= ATTR_UID;
602 newattrs.ia_uid = uid;
603 }
604 if (group != (gid_t) -1) {
605 if (!gid_valid(gid))
606 return -EINVAL;
607 newattrs.ia_valid |= ATTR_GID;
608 newattrs.ia_gid = gid;
609 }
610 if (!S_ISDIR(inode->i_mode))
611 newattrs.ia_valid |=
612 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
613 inode_lock(inode);
614 error = security_path_chown(path, uid, gid);
615 if (!error)
616 error = notify_change(path->dentry, &newattrs, &delegated_inode);
617 inode_unlock(inode);
618 if (delegated_inode) {
619 error = break_deleg_wait(&delegated_inode);
620 if (!error)
621 goto retry_deleg;
622 }
623 return error;
624}
625
626SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
627 gid_t, group, int, flag)
628{
629 struct path path;
630 int error = -EINVAL;
631 int lookup_flags;
632
633 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
634 goto out;
635
636 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
637 if (flag & AT_EMPTY_PATH)
638 lookup_flags |= LOOKUP_EMPTY;
639retry:
640 error = user_path_at(dfd, filename, lookup_flags, &path);
641 if (error)
642 goto out;
643 error = mnt_want_write(path.mnt);
644 if (error)
645 goto out_release;
646 error = chown_common(&path, user, group);
647 mnt_drop_write(path.mnt);
648out_release:
649 path_put(&path);
650 if (retry_estale(error, lookup_flags)) {
651 lookup_flags |= LOOKUP_REVAL;
652 goto retry;
653 }
654out:
655 return error;
656}
657
658SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
659{
660 return sys_fchownat(AT_FDCWD, filename, user, group, 0);
661}
662
663SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
664{
665 return sys_fchownat(AT_FDCWD, filename, user, group,
666 AT_SYMLINK_NOFOLLOW);
667}
668
669SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
670{
671 struct fd f = fdget(fd);
672 int error = -EBADF;
673
674 if (!f.file)
675 goto out;
676
677 error = mnt_want_write_file(f.file);
678 if (error)
679 goto out_fput;
680 audit_file(f.file);
681 error = chown_common(&f.file->f_path, user, group);
682 mnt_drop_write_file(f.file);
683out_fput:
684 fdput(f);
685out:
686 return error;
687}
688
689int open_check_o_direct(struct file *f)
690{
691 /* NB: we're sure to have correct a_ops only after f_op->open */
692 if (f->f_flags & O_DIRECT) {
693 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
694 return -EINVAL;
695 }
696 return 0;
697}
698
699static int do_dentry_open(struct file *f,
700 struct inode *inode,
701 int (*open)(struct inode *, struct file *),
702 const struct cred *cred)
703{
704 static const struct file_operations empty_fops = {};
705 int error;
706
707 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
708 FMODE_PREAD | FMODE_PWRITE;
709
710 path_get(&f->f_path);
711 f->f_inode = inode;
712 f->f_mapping = inode->i_mapping;
713
714 if (unlikely(f->f_flags & O_PATH)) {
715 f->f_mode = FMODE_PATH;
716 f->f_op = &empty_fops;
717 return 0;
718 }
719
720 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
721 error = get_write_access(inode);
722 if (unlikely(error))
723 goto cleanup_file;
724 error = __mnt_want_write(f->f_path.mnt);
725 if (unlikely(error)) {
726 put_write_access(inode);
727 goto cleanup_file;
728 }
729 f->f_mode |= FMODE_WRITER;
730 }
731
732 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
733 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
734 f->f_mode |= FMODE_ATOMIC_POS;
735
736 f->f_op = fops_get(inode->i_fop);
737 if (unlikely(WARN_ON(!f->f_op))) {
738 error = -ENODEV;
739 goto cleanup_all;
740 }
741
742 error = security_file_open(f, cred);
743 if (error)
744 goto cleanup_all;
745
746 error = break_lease(locks_inode(f), f->f_flags);
747 if (error)
748 goto cleanup_all;
749
750 if (!open)
751 open = f->f_op->open;
752 if (open) {
753 error = open(inode, f);
754 if (error)
755 goto cleanup_all;
756 }
757 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
758 i_readcount_inc(inode);
759 if ((f->f_mode & FMODE_READ) &&
760 likely(f->f_op->read || f->f_op->read_iter))
761 f->f_mode |= FMODE_CAN_READ;
762 if ((f->f_mode & FMODE_WRITE) &&
763 likely(f->f_op->write || f->f_op->write_iter))
764 f->f_mode |= FMODE_CAN_WRITE;
765
766 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
767
768 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
769
770 return 0;
771
772cleanup_all:
773 fops_put(f->f_op);
774 if (f->f_mode & FMODE_WRITER) {
775 put_write_access(inode);
776 __mnt_drop_write(f->f_path.mnt);
777 }
778cleanup_file:
779 path_put(&f->f_path);
780 f->f_path.mnt = NULL;
781 f->f_path.dentry = NULL;
782 f->f_inode = NULL;
783 return error;
784}
785
786/**
787 * finish_open - finish opening a file
788 * @file: file pointer
789 * @dentry: pointer to dentry
790 * @open: open callback
791 * @opened: state of open
792 *
793 * This can be used to finish opening a file passed to i_op->atomic_open().
794 *
795 * If the open callback is set to NULL, then the standard f_op->open()
796 * filesystem callback is substituted.
797 *
798 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
799 * the return value of d_splice_alias(), then the caller needs to perform dput()
800 * on it after finish_open().
801 *
802 * On successful return @file is a fully instantiated open file. After this, if
803 * an error occurs in ->atomic_open(), it needs to clean up with fput().
804 *
805 * Returns zero on success or -errno if the open failed.
806 */
807int finish_open(struct file *file, struct dentry *dentry,
808 int (*open)(struct inode *, struct file *),
809 int *opened)
810{
811 int error;
812 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
813
814 file->f_path.dentry = dentry;
815 error = do_dentry_open(file, d_backing_inode(dentry), open,
816 current_cred());
817 if (!error)
818 *opened |= FILE_OPENED;
819
820 return error;
821}
822EXPORT_SYMBOL(finish_open);
823
824/**
825 * finish_no_open - finish ->atomic_open() without opening the file
826 *
827 * @file: file pointer
828 * @dentry: dentry or NULL (as returned from ->lookup())
829 *
830 * This can be used to set the result of a successful lookup in ->atomic_open().
831 *
832 * NB: unlike finish_open() this function does consume the dentry reference and
833 * the caller need not dput() it.
834 *
835 * Returns "1" which must be the return value of ->atomic_open() after having
836 * called this function.
837 */
838int finish_no_open(struct file *file, struct dentry *dentry)
839{
840 file->f_path.dentry = dentry;
841 return 1;
842}
843EXPORT_SYMBOL(finish_no_open);
844
845char *file_path(struct file *filp, char *buf, int buflen)
846{
847 return d_path(&filp->f_path, buf, buflen);
848}
849EXPORT_SYMBOL(file_path);
850
851/**
852 * vfs_open - open the file at the given path
853 * @path: path to open
854 * @file: newly allocated file with f_flag initialized
855 * @cred: credentials to use
856 */
857int vfs_open(const struct path *path, struct file *file,
858 const struct cred *cred)
859{
860 struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
861
862 if (IS_ERR(dentry))
863 return PTR_ERR(dentry);
864
865 file->f_path = *path;
866 return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
867}
868
869struct file *dentry_open(const struct path *path, int flags,
870 const struct cred *cred)
871{
872 int error;
873 struct file *f;
874
875 validate_creds(cred);
876
877 /* We must always pass in a valid mount pointer. */
878 BUG_ON(!path->mnt);
879
880 f = get_empty_filp();
881 if (!IS_ERR(f)) {
882 f->f_flags = flags;
883 error = vfs_open(path, f, cred);
884 if (!error) {
885 /* from now on we need fput() to dispose of f */
886 error = open_check_o_direct(f);
887 if (error) {
888 fput(f);
889 f = ERR_PTR(error);
890 }
891 } else {
892 put_filp(f);
893 f = ERR_PTR(error);
894 }
895 }
896 return f;
897}
898EXPORT_SYMBOL(dentry_open);
899
900static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
901{
902 int lookup_flags = 0;
903 int acc_mode = ACC_MODE(flags);
904
905 if (flags & (O_CREAT | __O_TMPFILE))
906 op->mode = (mode & S_IALLUGO) | S_IFREG;
907 else
908 op->mode = 0;
909
910 /* Must never be set by userspace */
911 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
912
913 /*
914 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
915 * check for O_DSYNC if the need any syncing at all we enforce it's
916 * always set instead of having to deal with possibly weird behaviour
917 * for malicious applications setting only __O_SYNC.
918 */
919 if (flags & __O_SYNC)
920 flags |= O_DSYNC;
921
922 if (flags & __O_TMPFILE) {
923 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
924 return -EINVAL;
925 if (!(acc_mode & MAY_WRITE))
926 return -EINVAL;
927 } else if (flags & O_PATH) {
928 /*
929 * If we have O_PATH in the open flag. Then we
930 * cannot have anything other than the below set of flags
931 */
932 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
933 acc_mode = 0;
934 }
935
936 op->open_flag = flags;
937
938 /* O_TRUNC implies we need access checks for write permissions */
939 if (flags & O_TRUNC)
940 acc_mode |= MAY_WRITE;
941
942 /* Allow the LSM permission hook to distinguish append
943 access from general write access. */
944 if (flags & O_APPEND)
945 acc_mode |= MAY_APPEND;
946
947 op->acc_mode = acc_mode;
948
949 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
950
951 if (flags & O_CREAT) {
952 op->intent |= LOOKUP_CREATE;
953 if (flags & O_EXCL)
954 op->intent |= LOOKUP_EXCL;
955 }
956
957 if (flags & O_DIRECTORY)
958 lookup_flags |= LOOKUP_DIRECTORY;
959 if (!(flags & O_NOFOLLOW))
960 lookup_flags |= LOOKUP_FOLLOW;
961 op->lookup_flags = lookup_flags;
962 return 0;
963}
964
965/**
966 * file_open_name - open file and return file pointer
967 *
968 * @name: struct filename containing path to open
969 * @flags: open flags as per the open(2) second argument
970 * @mode: mode for the new file if O_CREAT is set, else ignored
971 *
972 * This is the helper to open a file from kernelspace if you really
973 * have to. But in generally you should not do this, so please move
974 * along, nothing to see here..
975 */
976struct file *file_open_name(struct filename *name, int flags, umode_t mode)
977{
978 struct open_flags op;
979 int err = build_open_flags(flags, mode, &op);
980 return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
981}
982
983/**
984 * filp_open - open file and return file pointer
985 *
986 * @filename: path to open
987 * @flags: open flags as per the open(2) second argument
988 * @mode: mode for the new file if O_CREAT is set, else ignored
989 *
990 * This is the helper to open a file from kernelspace if you really
991 * have to. But in generally you should not do this, so please move
992 * along, nothing to see here..
993 */
994struct file *filp_open(const char *filename, int flags, umode_t mode)
995{
996 struct filename *name = getname_kernel(filename);
997 struct file *file = ERR_CAST(name);
998
999 if (!IS_ERR(name)) {
1000 file = file_open_name(name, flags, mode);
1001 putname(name);
1002 }
1003 return file;
1004}
1005EXPORT_SYMBOL(filp_open);
1006
1007struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1008 const char *filename, int flags, umode_t mode)
1009{
1010 struct open_flags op;
1011 int err = build_open_flags(flags, mode, &op);
1012 if (err)
1013 return ERR_PTR(err);
1014 return do_file_open_root(dentry, mnt, filename, &op);
1015}
1016EXPORT_SYMBOL(file_open_root);
1017
1018struct file *filp_clone_open(struct file *oldfile)
1019{
1020 struct file *file;
1021 int retval;
1022
1023 file = get_empty_filp();
1024 if (IS_ERR(file))
1025 return file;
1026
1027 file->f_flags = oldfile->f_flags;
1028 retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1029 if (retval) {
1030 put_filp(file);
1031 return ERR_PTR(retval);
1032 }
1033
1034 return file;
1035}
1036EXPORT_SYMBOL(filp_clone_open);
1037
1038long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1039{
1040 struct open_flags op;
1041 int fd = build_open_flags(flags, mode, &op);
1042 struct filename *tmp;
1043
1044 if (fd)
1045 return fd;
1046
1047 tmp = getname(filename);
1048 if (IS_ERR(tmp))
1049 return PTR_ERR(tmp);
1050
1051 fd = get_unused_fd_flags(flags);
1052 if (fd >= 0) {
1053 struct file *f = do_filp_open(dfd, tmp, &op);
1054 if (IS_ERR(f)) {
1055 put_unused_fd(fd);
1056 fd = PTR_ERR(f);
1057 } else {
1058 fsnotify_open(f);
1059 fd_install(fd, f);
1060 }
1061 }
1062 putname(tmp);
1063 return fd;
1064}
1065
1066SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1067{
1068 if (force_o_largefile())
1069 flags |= O_LARGEFILE;
1070
1071 return do_sys_open(AT_FDCWD, filename, flags, mode);
1072}
1073
1074SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1075 umode_t, mode)
1076{
1077 if (force_o_largefile())
1078 flags |= O_LARGEFILE;
1079
1080 return do_sys_open(dfd, filename, flags, mode);
1081}
1082
1083#ifndef __alpha__
1084
1085/*
1086 * For backward compatibility? Maybe this should be moved
1087 * into arch/i386 instead?
1088 */
1089SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1090{
1091 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1092}
1093
1094#endif
1095
1096/*
1097 * "id" is the POSIX thread ID. We use the
1098 * files pointer for this..
1099 */
1100int filp_close(struct file *filp, fl_owner_t id)
1101{
1102 int retval = 0;
1103
1104 if (!file_count(filp)) {
1105 printk(KERN_ERR "VFS: Close: file count is 0\n");
1106 return 0;
1107 }
1108
1109 if (filp->f_op->flush)
1110 retval = filp->f_op->flush(filp, id);
1111
1112 if (likely(!(filp->f_mode & FMODE_PATH))) {
1113 dnotify_flush(filp, id);
1114 locks_remove_posix(filp, id);
1115 }
1116 fput(filp);
1117 return retval;
1118}
1119
1120EXPORT_SYMBOL(filp_close);
1121
1122/*
1123 * Careful here! We test whether the file pointer is NULL before
1124 * releasing the fd. This ensures that one clone task can't release
1125 * an fd while another clone is opening it.
1126 */
1127SYSCALL_DEFINE1(close, unsigned int, fd)
1128{
1129 int retval = __close_fd(current->files, fd);
1130
1131 /* can't restart close syscall because file table entry was cleared */
1132 if (unlikely(retval == -ERESTARTSYS ||
1133 retval == -ERESTARTNOINTR ||
1134 retval == -ERESTARTNOHAND ||
1135 retval == -ERESTART_RESTARTBLOCK))
1136 retval = -EINTR;
1137
1138 return retval;
1139}
1140EXPORT_SYMBOL(sys_close);
1141
1142/*
1143 * This routine simulates a hangup on the tty, to arrange that users
1144 * are given clean terminals at login time.
1145 */
1146SYSCALL_DEFINE0(vhangup)
1147{
1148 if (capable(CAP_SYS_TTY_CONFIG)) {
1149 tty_vhangup_self();
1150 return 0;
1151 }
1152 return -EPERM;
1153}
1154
1155/*
1156 * Called when an inode is about to be open.
1157 * We use this to disallow opening large files on 32bit systems if
1158 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1159 * on this flag in sys_open.
1160 */
1161int generic_file_open(struct inode * inode, struct file * filp)
1162{
1163 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1164 return -EOVERFLOW;
1165 return 0;
1166}
1167
1168EXPORT_SYMBOL(generic_file_open);
1169
1170/*
1171 * This is used by subsystems that don't want seekable
1172 * file descriptors. The function is not supposed to ever fail, the only
1173 * reason it returns an 'int' and not 'void' is so that it can be plugged
1174 * directly into file_operations structure.
1175 */
1176int nonseekable_open(struct inode *inode, struct file *filp)
1177{
1178 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1179 return 0;
1180}
1181
1182EXPORT_SYMBOL(nonseekable_open);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/open.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/string.h>
9#include <linux/mm.h>
10#include <linux/file.h>
11#include <linux/fdtable.h>
12#include <linux/fsnotify.h>
13#include <linux/module.h>
14#include <linux/tty.h>
15#include <linux/namei.h>
16#include <linux/backing-dev.h>
17#include <linux/capability.h>
18#include <linux/securebits.h>
19#include <linux/security.h>
20#include <linux/mount.h>
21#include <linux/fcntl.h>
22#include <linux/slab.h>
23#include <linux/uaccess.h>
24#include <linux/fs.h>
25#include <linux/personality.h>
26#include <linux/pagemap.h>
27#include <linux/syscalls.h>
28#include <linux/rcupdate.h>
29#include <linux/audit.h>
30#include <linux/falloc.h>
31#include <linux/fs_struct.h>
32#include <linux/ima.h>
33#include <linux/dnotify.h>
34#include <linux/compat.h>
35
36#include "internal.h"
37
38int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
39 struct file *filp)
40{
41 int ret;
42 struct iattr newattrs;
43
44 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
45 if (length < 0)
46 return -EINVAL;
47
48 newattrs.ia_size = length;
49 newattrs.ia_valid = ATTR_SIZE | time_attrs;
50 if (filp) {
51 newattrs.ia_file = filp;
52 newattrs.ia_valid |= ATTR_FILE;
53 }
54
55 /* Remove suid, sgid, and file capabilities on truncate too */
56 ret = dentry_needs_remove_privs(dentry);
57 if (ret < 0)
58 return ret;
59 if (ret)
60 newattrs.ia_valid |= ret | ATTR_FORCE;
61
62 inode_lock(dentry->d_inode);
63 /* Note any delegations or leases have already been broken: */
64 ret = notify_change(dentry, &newattrs, NULL);
65 inode_unlock(dentry->d_inode);
66 return ret;
67}
68
69long vfs_truncate(const struct path *path, loff_t length)
70{
71 struct inode *inode;
72 long error;
73
74 inode = path->dentry->d_inode;
75
76 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
77 if (S_ISDIR(inode->i_mode))
78 return -EISDIR;
79 if (!S_ISREG(inode->i_mode))
80 return -EINVAL;
81
82 error = mnt_want_write(path->mnt);
83 if (error)
84 goto out;
85
86 error = inode_permission(inode, MAY_WRITE);
87 if (error)
88 goto mnt_drop_write_and_out;
89
90 error = -EPERM;
91 if (IS_APPEND(inode))
92 goto mnt_drop_write_and_out;
93
94 error = get_write_access(inode);
95 if (error)
96 goto mnt_drop_write_and_out;
97
98 /*
99 * Make sure that there are no leases. get_write_access() protects
100 * against the truncate racing with a lease-granting setlease().
101 */
102 error = break_lease(inode, O_WRONLY);
103 if (error)
104 goto put_write_and_out;
105
106 error = locks_verify_truncate(inode, NULL, length);
107 if (!error)
108 error = security_path_truncate(path);
109 if (!error)
110 error = do_truncate(path->dentry, length, 0, NULL);
111
112put_write_and_out:
113 put_write_access(inode);
114mnt_drop_write_and_out:
115 mnt_drop_write(path->mnt);
116out:
117 return error;
118}
119EXPORT_SYMBOL_GPL(vfs_truncate);
120
121long do_sys_truncate(const char __user *pathname, loff_t length)
122{
123 unsigned int lookup_flags = LOOKUP_FOLLOW;
124 struct path path;
125 int error;
126
127 if (length < 0) /* sorry, but loff_t says... */
128 return -EINVAL;
129
130retry:
131 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
132 if (!error) {
133 error = vfs_truncate(&path, length);
134 path_put(&path);
135 }
136 if (retry_estale(error, lookup_flags)) {
137 lookup_flags |= LOOKUP_REVAL;
138 goto retry;
139 }
140 return error;
141}
142
143SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
144{
145 return do_sys_truncate(path, length);
146}
147
148#ifdef CONFIG_COMPAT
149COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
150{
151 return do_sys_truncate(path, length);
152}
153#endif
154
155long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
156{
157 struct inode *inode;
158 struct dentry *dentry;
159 struct fd f;
160 int error;
161
162 error = -EINVAL;
163 if (length < 0)
164 goto out;
165 error = -EBADF;
166 f = fdget(fd);
167 if (!f.file)
168 goto out;
169
170 /* explicitly opened as large or we are on 64-bit box */
171 if (f.file->f_flags & O_LARGEFILE)
172 small = 0;
173
174 dentry = f.file->f_path.dentry;
175 inode = dentry->d_inode;
176 error = -EINVAL;
177 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
178 goto out_putf;
179
180 error = -EINVAL;
181 /* Cannot ftruncate over 2^31 bytes without large file support */
182 if (small && length > MAX_NON_LFS)
183 goto out_putf;
184
185 error = -EPERM;
186 /* Check IS_APPEND on real upper inode */
187 if (IS_APPEND(file_inode(f.file)))
188 goto out_putf;
189
190 sb_start_write(inode->i_sb);
191 error = locks_verify_truncate(inode, f.file, length);
192 if (!error)
193 error = security_path_truncate(&f.file->f_path);
194 if (!error)
195 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
196 sb_end_write(inode->i_sb);
197out_putf:
198 fdput(f);
199out:
200 return error;
201}
202
203SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
204{
205 return do_sys_ftruncate(fd, length, 1);
206}
207
208#ifdef CONFIG_COMPAT
209COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
210{
211 return do_sys_ftruncate(fd, length, 1);
212}
213#endif
214
215/* LFS versions of truncate are only needed on 32 bit machines */
216#if BITS_PER_LONG == 32
217SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
218{
219 return do_sys_truncate(path, length);
220}
221
222SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
223{
224 return do_sys_ftruncate(fd, length, 0);
225}
226#endif /* BITS_PER_LONG == 32 */
227
228
229int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
230{
231 struct inode *inode = file_inode(file);
232 long ret;
233
234 if (offset < 0 || len <= 0)
235 return -EINVAL;
236
237 /* Return error if mode is not supported */
238 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
239 return -EOPNOTSUPP;
240
241 /* Punch hole and zero range are mutually exclusive */
242 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
243 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
244 return -EOPNOTSUPP;
245
246 /* Punch hole must have keep size set */
247 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
248 !(mode & FALLOC_FL_KEEP_SIZE))
249 return -EOPNOTSUPP;
250
251 /* Collapse range should only be used exclusively. */
252 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
253 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
254 return -EINVAL;
255
256 /* Insert range should only be used exclusively. */
257 if ((mode & FALLOC_FL_INSERT_RANGE) &&
258 (mode & ~FALLOC_FL_INSERT_RANGE))
259 return -EINVAL;
260
261 /* Unshare range should only be used with allocate mode. */
262 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
263 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
264 return -EINVAL;
265
266 if (!(file->f_mode & FMODE_WRITE))
267 return -EBADF;
268
269 /*
270 * We can only allow pure fallocate on append only files
271 */
272 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
273 return -EPERM;
274
275 if (IS_IMMUTABLE(inode))
276 return -EPERM;
277
278 /*
279 * We cannot allow any fallocate operation on an active swapfile
280 */
281 if (IS_SWAPFILE(inode))
282 return -ETXTBSY;
283
284 /*
285 * Revalidate the write permissions, in case security policy has
286 * changed since the files were opened.
287 */
288 ret = security_file_permission(file, MAY_WRITE);
289 if (ret)
290 return ret;
291
292 if (S_ISFIFO(inode->i_mode))
293 return -ESPIPE;
294
295 if (S_ISDIR(inode->i_mode))
296 return -EISDIR;
297
298 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
299 return -ENODEV;
300
301 /* Check for wrap through zero too */
302 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
303 return -EFBIG;
304
305 if (!file->f_op->fallocate)
306 return -EOPNOTSUPP;
307
308 file_start_write(file);
309 ret = file->f_op->fallocate(file, mode, offset, len);
310
311 /*
312 * Create inotify and fanotify events.
313 *
314 * To keep the logic simple always create events if fallocate succeeds.
315 * This implies that events are even created if the file size remains
316 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
317 */
318 if (ret == 0)
319 fsnotify_modify(file);
320
321 file_end_write(file);
322 return ret;
323}
324EXPORT_SYMBOL_GPL(vfs_fallocate);
325
326int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
327{
328 struct fd f = fdget(fd);
329 int error = -EBADF;
330
331 if (f.file) {
332 error = vfs_fallocate(f.file, mode, offset, len);
333 fdput(f);
334 }
335 return error;
336}
337
338SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
339{
340 return ksys_fallocate(fd, mode, offset, len);
341}
342
343/*
344 * access() needs to use the real uid/gid, not the effective uid/gid.
345 * We do this by temporarily clearing all FS-related capabilities and
346 * switching the fsuid/fsgid around to the real ones.
347 */
348static const struct cred *access_override_creds(void)
349{
350 const struct cred *old_cred;
351 struct cred *override_cred;
352
353 override_cred = prepare_creds();
354 if (!override_cred)
355 return NULL;
356
357 override_cred->fsuid = override_cred->uid;
358 override_cred->fsgid = override_cred->gid;
359
360 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
361 /* Clear the capabilities if we switch to a non-root user */
362 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
363 if (!uid_eq(override_cred->uid, root_uid))
364 cap_clear(override_cred->cap_effective);
365 else
366 override_cred->cap_effective =
367 override_cred->cap_permitted;
368 }
369
370 /*
371 * The new set of credentials can *only* be used in
372 * task-synchronous circumstances, and does not need
373 * RCU freeing, unless somebody then takes a separate
374 * reference to it.
375 *
376 * NOTE! This is _only_ true because this credential
377 * is used purely for override_creds() that installs
378 * it as the subjective cred. Other threads will be
379 * accessing ->real_cred, not the subjective cred.
380 *
381 * If somebody _does_ make a copy of this (using the
382 * 'get_current_cred()' function), that will clear the
383 * non_rcu field, because now that other user may be
384 * expecting RCU freeing. But normal thread-synchronous
385 * cred accesses will keep things non-RCY.
386 */
387 override_cred->non_rcu = 1;
388
389 old_cred = override_creds(override_cred);
390
391 /* override_cred() gets its own ref */
392 put_cred(override_cred);
393
394 return old_cred;
395}
396
397static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
398{
399 struct path path;
400 struct inode *inode;
401 int res;
402 unsigned int lookup_flags = LOOKUP_FOLLOW;
403 const struct cred *old_cred = NULL;
404
405 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
406 return -EINVAL;
407
408 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
409 return -EINVAL;
410
411 if (flags & AT_SYMLINK_NOFOLLOW)
412 lookup_flags &= ~LOOKUP_FOLLOW;
413 if (flags & AT_EMPTY_PATH)
414 lookup_flags |= LOOKUP_EMPTY;
415
416 if (!(flags & AT_EACCESS)) {
417 old_cred = access_override_creds();
418 if (!old_cred)
419 return -ENOMEM;
420 }
421
422retry:
423 res = user_path_at(dfd, filename, lookup_flags, &path);
424 if (res)
425 goto out;
426
427 inode = d_backing_inode(path.dentry);
428
429 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
430 /*
431 * MAY_EXEC on regular files is denied if the fs is mounted
432 * with the "noexec" flag.
433 */
434 res = -EACCES;
435 if (path_noexec(&path))
436 goto out_path_release;
437 }
438
439 res = inode_permission(inode, mode | MAY_ACCESS);
440 /* SuS v2 requires we report a read only fs too */
441 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
442 goto out_path_release;
443 /*
444 * This is a rare case where using __mnt_is_readonly()
445 * is OK without a mnt_want/drop_write() pair. Since
446 * no actual write to the fs is performed here, we do
447 * not need to telegraph to that to anyone.
448 *
449 * By doing this, we accept that this access is
450 * inherently racy and know that the fs may change
451 * state before we even see this result.
452 */
453 if (__mnt_is_readonly(path.mnt))
454 res = -EROFS;
455
456out_path_release:
457 path_put(&path);
458 if (retry_estale(res, lookup_flags)) {
459 lookup_flags |= LOOKUP_REVAL;
460 goto retry;
461 }
462out:
463 if (old_cred)
464 revert_creds(old_cred);
465
466 return res;
467}
468
469SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
470{
471 return do_faccessat(dfd, filename, mode, 0);
472}
473
474SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
475 int, flags)
476{
477 return do_faccessat(dfd, filename, mode, flags);
478}
479
480SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
481{
482 return do_faccessat(AT_FDCWD, filename, mode, 0);
483}
484
485SYSCALL_DEFINE1(chdir, const char __user *, filename)
486{
487 struct path path;
488 int error;
489 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
490retry:
491 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
492 if (error)
493 goto out;
494
495 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
496 if (error)
497 goto dput_and_out;
498
499 set_fs_pwd(current->fs, &path);
500
501dput_and_out:
502 path_put(&path);
503 if (retry_estale(error, lookup_flags)) {
504 lookup_flags |= LOOKUP_REVAL;
505 goto retry;
506 }
507out:
508 return error;
509}
510
511SYSCALL_DEFINE1(fchdir, unsigned int, fd)
512{
513 struct fd f = fdget_raw(fd);
514 int error;
515
516 error = -EBADF;
517 if (!f.file)
518 goto out;
519
520 error = -ENOTDIR;
521 if (!d_can_lookup(f.file->f_path.dentry))
522 goto out_putf;
523
524 error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
525 if (!error)
526 set_fs_pwd(current->fs, &f.file->f_path);
527out_putf:
528 fdput(f);
529out:
530 return error;
531}
532
533SYSCALL_DEFINE1(chroot, const char __user *, filename)
534{
535 struct path path;
536 int error;
537 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
538retry:
539 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
540 if (error)
541 goto out;
542
543 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
544 if (error)
545 goto dput_and_out;
546
547 error = -EPERM;
548 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
549 goto dput_and_out;
550 error = security_path_chroot(&path);
551 if (error)
552 goto dput_and_out;
553
554 set_fs_root(current->fs, &path);
555 error = 0;
556dput_and_out:
557 path_put(&path);
558 if (retry_estale(error, lookup_flags)) {
559 lookup_flags |= LOOKUP_REVAL;
560 goto retry;
561 }
562out:
563 return error;
564}
565
566int chmod_common(const struct path *path, umode_t mode)
567{
568 struct inode *inode = path->dentry->d_inode;
569 struct inode *delegated_inode = NULL;
570 struct iattr newattrs;
571 int error;
572
573 error = mnt_want_write(path->mnt);
574 if (error)
575 return error;
576retry_deleg:
577 inode_lock(inode);
578 error = security_path_chmod(path, mode);
579 if (error)
580 goto out_unlock;
581 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
582 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
583 error = notify_change(path->dentry, &newattrs, &delegated_inode);
584out_unlock:
585 inode_unlock(inode);
586 if (delegated_inode) {
587 error = break_deleg_wait(&delegated_inode);
588 if (!error)
589 goto retry_deleg;
590 }
591 mnt_drop_write(path->mnt);
592 return error;
593}
594
595int vfs_fchmod(struct file *file, umode_t mode)
596{
597 audit_file(file);
598 return chmod_common(&file->f_path, mode);
599}
600
601SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
602{
603 struct fd f = fdget(fd);
604 int err = -EBADF;
605
606 if (f.file) {
607 err = vfs_fchmod(f.file, mode);
608 fdput(f);
609 }
610 return err;
611}
612
613static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
614{
615 struct path path;
616 int error;
617 unsigned int lookup_flags = LOOKUP_FOLLOW;
618retry:
619 error = user_path_at(dfd, filename, lookup_flags, &path);
620 if (!error) {
621 error = chmod_common(&path, mode);
622 path_put(&path);
623 if (retry_estale(error, lookup_flags)) {
624 lookup_flags |= LOOKUP_REVAL;
625 goto retry;
626 }
627 }
628 return error;
629}
630
631SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
632 umode_t, mode)
633{
634 return do_fchmodat(dfd, filename, mode);
635}
636
637SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
638{
639 return do_fchmodat(AT_FDCWD, filename, mode);
640}
641
642int chown_common(const struct path *path, uid_t user, gid_t group)
643{
644 struct inode *inode = path->dentry->d_inode;
645 struct inode *delegated_inode = NULL;
646 int error;
647 struct iattr newattrs;
648 kuid_t uid;
649 kgid_t gid;
650
651 uid = make_kuid(current_user_ns(), user);
652 gid = make_kgid(current_user_ns(), group);
653
654retry_deleg:
655 newattrs.ia_valid = ATTR_CTIME;
656 if (user != (uid_t) -1) {
657 if (!uid_valid(uid))
658 return -EINVAL;
659 newattrs.ia_valid |= ATTR_UID;
660 newattrs.ia_uid = uid;
661 }
662 if (group != (gid_t) -1) {
663 if (!gid_valid(gid))
664 return -EINVAL;
665 newattrs.ia_valid |= ATTR_GID;
666 newattrs.ia_gid = gid;
667 }
668 if (!S_ISDIR(inode->i_mode))
669 newattrs.ia_valid |=
670 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
671 inode_lock(inode);
672 error = security_path_chown(path, uid, gid);
673 if (!error)
674 error = notify_change(path->dentry, &newattrs, &delegated_inode);
675 inode_unlock(inode);
676 if (delegated_inode) {
677 error = break_deleg_wait(&delegated_inode);
678 if (!error)
679 goto retry_deleg;
680 }
681 return error;
682}
683
684int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
685 int flag)
686{
687 struct path path;
688 int error = -EINVAL;
689 int lookup_flags;
690
691 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
692 goto out;
693
694 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
695 if (flag & AT_EMPTY_PATH)
696 lookup_flags |= LOOKUP_EMPTY;
697retry:
698 error = user_path_at(dfd, filename, lookup_flags, &path);
699 if (error)
700 goto out;
701 error = mnt_want_write(path.mnt);
702 if (error)
703 goto out_release;
704 error = chown_common(&path, user, group);
705 mnt_drop_write(path.mnt);
706out_release:
707 path_put(&path);
708 if (retry_estale(error, lookup_flags)) {
709 lookup_flags |= LOOKUP_REVAL;
710 goto retry;
711 }
712out:
713 return error;
714}
715
716SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
717 gid_t, group, int, flag)
718{
719 return do_fchownat(dfd, filename, user, group, flag);
720}
721
722SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
723{
724 return do_fchownat(AT_FDCWD, filename, user, group, 0);
725}
726
727SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
728{
729 return do_fchownat(AT_FDCWD, filename, user, group,
730 AT_SYMLINK_NOFOLLOW);
731}
732
733int vfs_fchown(struct file *file, uid_t user, gid_t group)
734{
735 int error;
736
737 error = mnt_want_write_file(file);
738 if (error)
739 return error;
740 audit_file(file);
741 error = chown_common(&file->f_path, user, group);
742 mnt_drop_write_file(file);
743 return error;
744}
745
746int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
747{
748 struct fd f = fdget(fd);
749 int error = -EBADF;
750
751 if (f.file) {
752 error = vfs_fchown(f.file, user, group);
753 fdput(f);
754 }
755 return error;
756}
757
758SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
759{
760 return ksys_fchown(fd, user, group);
761}
762
763static int do_dentry_open(struct file *f,
764 struct inode *inode,
765 int (*open)(struct inode *, struct file *))
766{
767 static const struct file_operations empty_fops = {};
768 int error;
769
770 path_get(&f->f_path);
771 f->f_inode = inode;
772 f->f_mapping = inode->i_mapping;
773 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
774 f->f_sb_err = file_sample_sb_err(f);
775
776 if (unlikely(f->f_flags & O_PATH)) {
777 f->f_mode = FMODE_PATH | FMODE_OPENED;
778 f->f_op = &empty_fops;
779 return 0;
780 }
781
782 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
783 error = get_write_access(inode);
784 if (unlikely(error))
785 goto cleanup_file;
786 error = __mnt_want_write(f->f_path.mnt);
787 if (unlikely(error)) {
788 put_write_access(inode);
789 goto cleanup_file;
790 }
791 f->f_mode |= FMODE_WRITER;
792 }
793
794 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
795 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
796 f->f_mode |= FMODE_ATOMIC_POS;
797
798 f->f_op = fops_get(inode->i_fop);
799 if (WARN_ON(!f->f_op)) {
800 error = -ENODEV;
801 goto cleanup_all;
802 }
803
804 error = security_file_open(f);
805 if (error)
806 goto cleanup_all;
807
808 error = break_lease(locks_inode(f), f->f_flags);
809 if (error)
810 goto cleanup_all;
811
812 /* normally all 3 are set; ->open() can clear them if needed */
813 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
814 if (!open)
815 open = f->f_op->open;
816 if (open) {
817 error = open(inode, f);
818 if (error)
819 goto cleanup_all;
820 }
821 f->f_mode |= FMODE_OPENED;
822 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
823 i_readcount_inc(inode);
824 if ((f->f_mode & FMODE_READ) &&
825 likely(f->f_op->read || f->f_op->read_iter))
826 f->f_mode |= FMODE_CAN_READ;
827 if ((f->f_mode & FMODE_WRITE) &&
828 likely(f->f_op->write || f->f_op->write_iter))
829 f->f_mode |= FMODE_CAN_WRITE;
830
831 f->f_write_hint = WRITE_LIFE_NOT_SET;
832 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
833
834 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
835
836 /* NB: we're sure to have correct a_ops only after f_op->open */
837 if (f->f_flags & O_DIRECT) {
838 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
839 return -EINVAL;
840 }
841
842 /*
843 * XXX: Huge page cache doesn't support writing yet. Drop all page
844 * cache for this file before processing writes.
845 */
846 if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
847 truncate_pagecache(inode, 0);
848
849 return 0;
850
851cleanup_all:
852 if (WARN_ON_ONCE(error > 0))
853 error = -EINVAL;
854 fops_put(f->f_op);
855 if (f->f_mode & FMODE_WRITER) {
856 put_write_access(inode);
857 __mnt_drop_write(f->f_path.mnt);
858 }
859cleanup_file:
860 path_put(&f->f_path);
861 f->f_path.mnt = NULL;
862 f->f_path.dentry = NULL;
863 f->f_inode = NULL;
864 return error;
865}
866
867/**
868 * finish_open - finish opening a file
869 * @file: file pointer
870 * @dentry: pointer to dentry
871 * @open: open callback
872 * @opened: state of open
873 *
874 * This can be used to finish opening a file passed to i_op->atomic_open().
875 *
876 * If the open callback is set to NULL, then the standard f_op->open()
877 * filesystem callback is substituted.
878 *
879 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
880 * the return value of d_splice_alias(), then the caller needs to perform dput()
881 * on it after finish_open().
882 *
883 * Returns zero on success or -errno if the open failed.
884 */
885int finish_open(struct file *file, struct dentry *dentry,
886 int (*open)(struct inode *, struct file *))
887{
888 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
889
890 file->f_path.dentry = dentry;
891 return do_dentry_open(file, d_backing_inode(dentry), open);
892}
893EXPORT_SYMBOL(finish_open);
894
895/**
896 * finish_no_open - finish ->atomic_open() without opening the file
897 *
898 * @file: file pointer
899 * @dentry: dentry or NULL (as returned from ->lookup())
900 *
901 * This can be used to set the result of a successful lookup in ->atomic_open().
902 *
903 * NB: unlike finish_open() this function does consume the dentry reference and
904 * the caller need not dput() it.
905 *
906 * Returns "0" which must be the return value of ->atomic_open() after having
907 * called this function.
908 */
909int finish_no_open(struct file *file, struct dentry *dentry)
910{
911 file->f_path.dentry = dentry;
912 return 0;
913}
914EXPORT_SYMBOL(finish_no_open);
915
916char *file_path(struct file *filp, char *buf, int buflen)
917{
918 return d_path(&filp->f_path, buf, buflen);
919}
920EXPORT_SYMBOL(file_path);
921
922/**
923 * vfs_open - open the file at the given path
924 * @path: path to open
925 * @file: newly allocated file with f_flag initialized
926 * @cred: credentials to use
927 */
928int vfs_open(const struct path *path, struct file *file)
929{
930 file->f_path = *path;
931 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
932}
933
934struct file *dentry_open(const struct path *path, int flags,
935 const struct cred *cred)
936{
937 int error;
938 struct file *f;
939
940 validate_creds(cred);
941
942 /* We must always pass in a valid mount pointer. */
943 BUG_ON(!path->mnt);
944
945 f = alloc_empty_file(flags, cred);
946 if (!IS_ERR(f)) {
947 error = vfs_open(path, f);
948 if (error) {
949 fput(f);
950 f = ERR_PTR(error);
951 }
952 }
953 return f;
954}
955EXPORT_SYMBOL(dentry_open);
956
957struct file *open_with_fake_path(const struct path *path, int flags,
958 struct inode *inode, const struct cred *cred)
959{
960 struct file *f = alloc_empty_file_noaccount(flags, cred);
961 if (!IS_ERR(f)) {
962 int error;
963
964 f->f_path = *path;
965 error = do_dentry_open(f, inode, NULL);
966 if (error) {
967 fput(f);
968 f = ERR_PTR(error);
969 }
970 }
971 return f;
972}
973EXPORT_SYMBOL(open_with_fake_path);
974
975#define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
976#define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
977
978inline struct open_how build_open_how(int flags, umode_t mode)
979{
980 struct open_how how = {
981 .flags = flags & VALID_OPEN_FLAGS,
982 .mode = mode & S_IALLUGO,
983 };
984
985 /* O_PATH beats everything else. */
986 if (how.flags & O_PATH)
987 how.flags &= O_PATH_FLAGS;
988 /* Modes should only be set for create-like flags. */
989 if (!WILL_CREATE(how.flags))
990 how.mode = 0;
991 return how;
992}
993
994inline int build_open_flags(const struct open_how *how, struct open_flags *op)
995{
996 int flags = how->flags;
997 int lookup_flags = 0;
998 int acc_mode = ACC_MODE(flags);
999
1000 /* Must never be set by userspace */
1001 flags &= ~(FMODE_NONOTIFY | O_CLOEXEC);
1002
1003 /*
1004 * Older syscalls implicitly clear all of the invalid flags or argument
1005 * values before calling build_open_flags(), but openat2(2) checks all
1006 * of its arguments.
1007 */
1008 if (flags & ~VALID_OPEN_FLAGS)
1009 return -EINVAL;
1010 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1011 return -EINVAL;
1012
1013 /* Deal with the mode. */
1014 if (WILL_CREATE(flags)) {
1015 if (how->mode & ~S_IALLUGO)
1016 return -EINVAL;
1017 op->mode = how->mode | S_IFREG;
1018 } else {
1019 if (how->mode != 0)
1020 return -EINVAL;
1021 op->mode = 0;
1022 }
1023
1024 /*
1025 * In order to ensure programs get explicit errors when trying to use
1026 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1027 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1028 * have to require userspace to explicitly set it.
1029 */
1030 if (flags & __O_TMPFILE) {
1031 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1032 return -EINVAL;
1033 if (!(acc_mode & MAY_WRITE))
1034 return -EINVAL;
1035 }
1036 if (flags & O_PATH) {
1037 /* O_PATH only permits certain other flags to be set. */
1038 if (flags & ~O_PATH_FLAGS)
1039 return -EINVAL;
1040 acc_mode = 0;
1041 }
1042
1043 /*
1044 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1045 * check for O_DSYNC if the need any syncing at all we enforce it's
1046 * always set instead of having to deal with possibly weird behaviour
1047 * for malicious applications setting only __O_SYNC.
1048 */
1049 if (flags & __O_SYNC)
1050 flags |= O_DSYNC;
1051
1052 op->open_flag = flags;
1053
1054 /* O_TRUNC implies we need access checks for write permissions */
1055 if (flags & O_TRUNC)
1056 acc_mode |= MAY_WRITE;
1057
1058 /* Allow the LSM permission hook to distinguish append
1059 access from general write access. */
1060 if (flags & O_APPEND)
1061 acc_mode |= MAY_APPEND;
1062
1063 op->acc_mode = acc_mode;
1064
1065 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1066
1067 if (flags & O_CREAT) {
1068 op->intent |= LOOKUP_CREATE;
1069 if (flags & O_EXCL) {
1070 op->intent |= LOOKUP_EXCL;
1071 flags |= O_NOFOLLOW;
1072 }
1073 }
1074
1075 if (flags & O_DIRECTORY)
1076 lookup_flags |= LOOKUP_DIRECTORY;
1077 if (!(flags & O_NOFOLLOW))
1078 lookup_flags |= LOOKUP_FOLLOW;
1079
1080 if (how->resolve & RESOLVE_NO_XDEV)
1081 lookup_flags |= LOOKUP_NO_XDEV;
1082 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1083 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1084 if (how->resolve & RESOLVE_NO_SYMLINKS)
1085 lookup_flags |= LOOKUP_NO_SYMLINKS;
1086 if (how->resolve & RESOLVE_BENEATH)
1087 lookup_flags |= LOOKUP_BENEATH;
1088 if (how->resolve & RESOLVE_IN_ROOT)
1089 lookup_flags |= LOOKUP_IN_ROOT;
1090
1091 op->lookup_flags = lookup_flags;
1092 return 0;
1093}
1094
1095/**
1096 * file_open_name - open file and return file pointer
1097 *
1098 * @name: struct filename containing path to open
1099 * @flags: open flags as per the open(2) second argument
1100 * @mode: mode for the new file if O_CREAT is set, else ignored
1101 *
1102 * This is the helper to open a file from kernelspace if you really
1103 * have to. But in generally you should not do this, so please move
1104 * along, nothing to see here..
1105 */
1106struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1107{
1108 struct open_flags op;
1109 struct open_how how = build_open_how(flags, mode);
1110 int err = build_open_flags(&how, &op);
1111 if (err)
1112 return ERR_PTR(err);
1113 return do_filp_open(AT_FDCWD, name, &op);
1114}
1115
1116/**
1117 * filp_open - open file and return file pointer
1118 *
1119 * @filename: path to open
1120 * @flags: open flags as per the open(2) second argument
1121 * @mode: mode for the new file if O_CREAT is set, else ignored
1122 *
1123 * This is the helper to open a file from kernelspace if you really
1124 * have to. But in generally you should not do this, so please move
1125 * along, nothing to see here..
1126 */
1127struct file *filp_open(const char *filename, int flags, umode_t mode)
1128{
1129 struct filename *name = getname_kernel(filename);
1130 struct file *file = ERR_CAST(name);
1131
1132 if (!IS_ERR(name)) {
1133 file = file_open_name(name, flags, mode);
1134 putname(name);
1135 }
1136 return file;
1137}
1138EXPORT_SYMBOL(filp_open);
1139
1140struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1141 const char *filename, int flags, umode_t mode)
1142{
1143 struct open_flags op;
1144 struct open_how how = build_open_how(flags, mode);
1145 int err = build_open_flags(&how, &op);
1146 if (err)
1147 return ERR_PTR(err);
1148 return do_file_open_root(dentry, mnt, filename, &op);
1149}
1150EXPORT_SYMBOL(file_open_root);
1151
1152static long do_sys_openat2(int dfd, const char __user *filename,
1153 struct open_how *how)
1154{
1155 struct open_flags op;
1156 int fd = build_open_flags(how, &op);
1157 struct filename *tmp;
1158
1159 if (fd)
1160 return fd;
1161
1162 tmp = getname(filename);
1163 if (IS_ERR(tmp))
1164 return PTR_ERR(tmp);
1165
1166 fd = get_unused_fd_flags(how->flags);
1167 if (fd >= 0) {
1168 struct file *f = do_filp_open(dfd, tmp, &op);
1169 if (IS_ERR(f)) {
1170 put_unused_fd(fd);
1171 fd = PTR_ERR(f);
1172 } else {
1173 fsnotify_open(f);
1174 fd_install(fd, f);
1175 }
1176 }
1177 putname(tmp);
1178 return fd;
1179}
1180
1181long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1182{
1183 struct open_how how = build_open_how(flags, mode);
1184 return do_sys_openat2(dfd, filename, &how);
1185}
1186
1187
1188SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1189{
1190 if (force_o_largefile())
1191 flags |= O_LARGEFILE;
1192 return do_sys_open(AT_FDCWD, filename, flags, mode);
1193}
1194
1195SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1196 umode_t, mode)
1197{
1198 if (force_o_largefile())
1199 flags |= O_LARGEFILE;
1200 return do_sys_open(dfd, filename, flags, mode);
1201}
1202
1203SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1204 struct open_how __user *, how, size_t, usize)
1205{
1206 int err;
1207 struct open_how tmp;
1208
1209 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1210 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1211
1212 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1213 return -EINVAL;
1214
1215 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1216 if (err)
1217 return err;
1218
1219 /* O_LARGEFILE is only allowed for non-O_PATH. */
1220 if (!(tmp.flags & O_PATH) && force_o_largefile())
1221 tmp.flags |= O_LARGEFILE;
1222
1223 return do_sys_openat2(dfd, filename, &tmp);
1224}
1225
1226#ifdef CONFIG_COMPAT
1227/*
1228 * Exactly like sys_open(), except that it doesn't set the
1229 * O_LARGEFILE flag.
1230 */
1231COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1232{
1233 return do_sys_open(AT_FDCWD, filename, flags, mode);
1234}
1235
1236/*
1237 * Exactly like sys_openat(), except that it doesn't set the
1238 * O_LARGEFILE flag.
1239 */
1240COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1241{
1242 return do_sys_open(dfd, filename, flags, mode);
1243}
1244#endif
1245
1246#ifndef __alpha__
1247
1248/*
1249 * For backward compatibility? Maybe this should be moved
1250 * into arch/i386 instead?
1251 */
1252SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1253{
1254 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1255
1256 if (force_o_largefile())
1257 flags |= O_LARGEFILE;
1258 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1259}
1260#endif
1261
1262/*
1263 * "id" is the POSIX thread ID. We use the
1264 * files pointer for this..
1265 */
1266int filp_close(struct file *filp, fl_owner_t id)
1267{
1268 int retval = 0;
1269
1270 if (!file_count(filp)) {
1271 printk(KERN_ERR "VFS: Close: file count is 0\n");
1272 return 0;
1273 }
1274
1275 if (filp->f_op->flush)
1276 retval = filp->f_op->flush(filp, id);
1277
1278 if (likely(!(filp->f_mode & FMODE_PATH))) {
1279 dnotify_flush(filp, id);
1280 locks_remove_posix(filp, id);
1281 }
1282 fput(filp);
1283 return retval;
1284}
1285
1286EXPORT_SYMBOL(filp_close);
1287
1288/*
1289 * Careful here! We test whether the file pointer is NULL before
1290 * releasing the fd. This ensures that one clone task can't release
1291 * an fd while another clone is opening it.
1292 */
1293SYSCALL_DEFINE1(close, unsigned int, fd)
1294{
1295 int retval = __close_fd(current->files, fd);
1296
1297 /* can't restart close syscall because file table entry was cleared */
1298 if (unlikely(retval == -ERESTARTSYS ||
1299 retval == -ERESTARTNOINTR ||
1300 retval == -ERESTARTNOHAND ||
1301 retval == -ERESTART_RESTARTBLOCK))
1302 retval = -EINTR;
1303
1304 return retval;
1305}
1306
1307/**
1308 * close_range() - Close all file descriptors in a given range.
1309 *
1310 * @fd: starting file descriptor to close
1311 * @max_fd: last file descriptor to close
1312 * @flags: reserved for future extensions
1313 *
1314 * This closes a range of file descriptors. All file descriptors
1315 * from @fd up to and including @max_fd are closed.
1316 * Currently, errors to close a given file descriptor are ignored.
1317 */
1318SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1319 unsigned int, flags)
1320{
1321 return __close_range(fd, max_fd, flags);
1322}
1323
1324/*
1325 * This routine simulates a hangup on the tty, to arrange that users
1326 * are given clean terminals at login time.
1327 */
1328SYSCALL_DEFINE0(vhangup)
1329{
1330 if (capable(CAP_SYS_TTY_CONFIG)) {
1331 tty_vhangup_self();
1332 return 0;
1333 }
1334 return -EPERM;
1335}
1336
1337/*
1338 * Called when an inode is about to be open.
1339 * We use this to disallow opening large files on 32bit systems if
1340 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1341 * on this flag in sys_open.
1342 */
1343int generic_file_open(struct inode * inode, struct file * filp)
1344{
1345 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1346 return -EOVERFLOW;
1347 return 0;
1348}
1349
1350EXPORT_SYMBOL(generic_file_open);
1351
1352/*
1353 * This is used by subsystems that don't want seekable
1354 * file descriptors. The function is not supposed to ever fail, the only
1355 * reason it returns an 'int' and not 'void' is so that it can be plugged
1356 * directly into file_operations structure.
1357 */
1358int nonseekable_open(struct inode *inode, struct file *filp)
1359{
1360 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1361 return 0;
1362}
1363
1364EXPORT_SYMBOL(nonseekable_open);
1365
1366/*
1367 * stream_open is used by subsystems that want stream-like file descriptors.
1368 * Such file descriptors are not seekable and don't have notion of position
1369 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1370 * Contrary to file descriptors of other regular files, .read() and .write()
1371 * can run simultaneously.
1372 *
1373 * stream_open never fails and is marked to return int so that it could be
1374 * directly used as file_operations.open .
1375 */
1376int stream_open(struct inode *inode, struct file *filp)
1377{
1378 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1379 filp->f_mode |= FMODE_STREAM;
1380 return 0;
1381}
1382
1383EXPORT_SYMBOL(stream_open);