xref: /linux/fs/open.c (revision 37744feebc086908fd89760650f458ab19071750)
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 
38 int 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 
69 long 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 
112 put_write_and_out:
113 	put_write_access(inode);
114 mnt_drop_write_and_out:
115 	mnt_drop_write(path->mnt);
116 out:
117 	return error;
118 }
119 EXPORT_SYMBOL_GPL(vfs_truncate);
120 
121 long 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 
130 retry:
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 
143 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
144 {
145 	return do_sys_truncate(path, length);
146 }
147 
148 #ifdef CONFIG_COMPAT
149 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
150 {
151 	return do_sys_truncate(path, length);
152 }
153 #endif
154 
155 long 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);
197 out_putf:
198 	fdput(f);
199 out:
200 	return error;
201 }
202 
203 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
204 {
205 	return do_sys_ftruncate(fd, length, 1);
206 }
207 
208 #ifdef CONFIG_COMPAT
209 COMPAT_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
217 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
218 {
219 	return do_sys_truncate(path, length);
220 }
221 
222 SYSCALL_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 
229 int 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 }
324 EXPORT_SYMBOL_GPL(vfs_fallocate);
325 
326 int 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 
338 SYSCALL_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  */
348 long do_faccessat(int dfd, const char __user *filename, int mode)
349 {
350 	const struct cred *old_cred;
351 	struct cred *override_cred;
352 	struct path path;
353 	struct inode *inode;
354 	int res;
355 	unsigned int lookup_flags = LOOKUP_FOLLOW;
356 
357 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
358 		return -EINVAL;
359 
360 	override_cred = prepare_creds();
361 	if (!override_cred)
362 		return -ENOMEM;
363 
364 	override_cred->fsuid = override_cred->uid;
365 	override_cred->fsgid = override_cred->gid;
366 
367 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
368 		/* Clear the capabilities if we switch to a non-root user */
369 		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
370 		if (!uid_eq(override_cred->uid, root_uid))
371 			cap_clear(override_cred->cap_effective);
372 		else
373 			override_cred->cap_effective =
374 				override_cred->cap_permitted;
375 	}
376 
377 	/*
378 	 * The new set of credentials can *only* be used in
379 	 * task-synchronous circumstances, and does not need
380 	 * RCU freeing, unless somebody then takes a separate
381 	 * reference to it.
382 	 *
383 	 * NOTE! This is _only_ true because this credential
384 	 * is used purely for override_creds() that installs
385 	 * it as the subjective cred. Other threads will be
386 	 * accessing ->real_cred, not the subjective cred.
387 	 *
388 	 * If somebody _does_ make a copy of this (using the
389 	 * 'get_current_cred()' function), that will clear the
390 	 * non_rcu field, because now that other user may be
391 	 * expecting RCU freeing. But normal thread-synchronous
392 	 * cred accesses will keep things non-RCY.
393 	 */
394 	override_cred->non_rcu = 1;
395 
396 	old_cred = override_creds(override_cred);
397 retry:
398 	res = user_path_at(dfd, filename, lookup_flags, &path);
399 	if (res)
400 		goto out;
401 
402 	inode = d_backing_inode(path.dentry);
403 
404 	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
405 		/*
406 		 * MAY_EXEC on regular files is denied if the fs is mounted
407 		 * with the "noexec" flag.
408 		 */
409 		res = -EACCES;
410 		if (path_noexec(&path))
411 			goto out_path_release;
412 	}
413 
414 	res = inode_permission(inode, mode | MAY_ACCESS);
415 	/* SuS v2 requires we report a read only fs too */
416 	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
417 		goto out_path_release;
418 	/*
419 	 * This is a rare case where using __mnt_is_readonly()
420 	 * is OK without a mnt_want/drop_write() pair.  Since
421 	 * no actual write to the fs is performed here, we do
422 	 * not need to telegraph to that to anyone.
423 	 *
424 	 * By doing this, we accept that this access is
425 	 * inherently racy and know that the fs may change
426 	 * state before we even see this result.
427 	 */
428 	if (__mnt_is_readonly(path.mnt))
429 		res = -EROFS;
430 
431 out_path_release:
432 	path_put(&path);
433 	if (retry_estale(res, lookup_flags)) {
434 		lookup_flags |= LOOKUP_REVAL;
435 		goto retry;
436 	}
437 out:
438 	revert_creds(old_cred);
439 	put_cred(override_cred);
440 	return res;
441 }
442 
443 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
444 {
445 	return do_faccessat(dfd, filename, mode);
446 }
447 
448 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
449 {
450 	return do_faccessat(AT_FDCWD, filename, mode);
451 }
452 
453 int ksys_chdir(const char __user *filename)
454 {
455 	struct path path;
456 	int error;
457 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
458 retry:
459 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
460 	if (error)
461 		goto out;
462 
463 	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
464 	if (error)
465 		goto dput_and_out;
466 
467 	set_fs_pwd(current->fs, &path);
468 
469 dput_and_out:
470 	path_put(&path);
471 	if (retry_estale(error, lookup_flags)) {
472 		lookup_flags |= LOOKUP_REVAL;
473 		goto retry;
474 	}
475 out:
476 	return error;
477 }
478 
479 SYSCALL_DEFINE1(chdir, const char __user *, filename)
480 {
481 	return ksys_chdir(filename);
482 }
483 
484 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
485 {
486 	struct fd f = fdget_raw(fd);
487 	int error;
488 
489 	error = -EBADF;
490 	if (!f.file)
491 		goto out;
492 
493 	error = -ENOTDIR;
494 	if (!d_can_lookup(f.file->f_path.dentry))
495 		goto out_putf;
496 
497 	error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
498 	if (!error)
499 		set_fs_pwd(current->fs, &f.file->f_path);
500 out_putf:
501 	fdput(f);
502 out:
503 	return error;
504 }
505 
506 int ksys_chroot(const char __user *filename)
507 {
508 	struct path path;
509 	int error;
510 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
511 retry:
512 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
513 	if (error)
514 		goto out;
515 
516 	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
517 	if (error)
518 		goto dput_and_out;
519 
520 	error = -EPERM;
521 	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
522 		goto dput_and_out;
523 	error = security_path_chroot(&path);
524 	if (error)
525 		goto dput_and_out;
526 
527 	set_fs_root(current->fs, &path);
528 	error = 0;
529 dput_and_out:
530 	path_put(&path);
531 	if (retry_estale(error, lookup_flags)) {
532 		lookup_flags |= LOOKUP_REVAL;
533 		goto retry;
534 	}
535 out:
536 	return error;
537 }
538 
539 SYSCALL_DEFINE1(chroot, const char __user *, filename)
540 {
541 	return ksys_chroot(filename);
542 }
543 
544 static int chmod_common(const struct path *path, umode_t mode)
545 {
546 	struct inode *inode = path->dentry->d_inode;
547 	struct inode *delegated_inode = NULL;
548 	struct iattr newattrs;
549 	int error;
550 
551 	error = mnt_want_write(path->mnt);
552 	if (error)
553 		return error;
554 retry_deleg:
555 	inode_lock(inode);
556 	error = security_path_chmod(path, mode);
557 	if (error)
558 		goto out_unlock;
559 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
560 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
561 	error = notify_change(path->dentry, &newattrs, &delegated_inode);
562 out_unlock:
563 	inode_unlock(inode);
564 	if (delegated_inode) {
565 		error = break_deleg_wait(&delegated_inode);
566 		if (!error)
567 			goto retry_deleg;
568 	}
569 	mnt_drop_write(path->mnt);
570 	return error;
571 }
572 
573 int ksys_fchmod(unsigned int fd, umode_t mode)
574 {
575 	struct fd f = fdget(fd);
576 	int err = -EBADF;
577 
578 	if (f.file) {
579 		audit_file(f.file);
580 		err = chmod_common(&f.file->f_path, mode);
581 		fdput(f);
582 	}
583 	return err;
584 }
585 
586 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
587 {
588 	return ksys_fchmod(fd, mode);
589 }
590 
591 int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
592 {
593 	struct path path;
594 	int error;
595 	unsigned int lookup_flags = LOOKUP_FOLLOW;
596 retry:
597 	error = user_path_at(dfd, filename, lookup_flags, &path);
598 	if (!error) {
599 		error = chmod_common(&path, mode);
600 		path_put(&path);
601 		if (retry_estale(error, lookup_flags)) {
602 			lookup_flags |= LOOKUP_REVAL;
603 			goto retry;
604 		}
605 	}
606 	return error;
607 }
608 
609 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
610 		umode_t, mode)
611 {
612 	return do_fchmodat(dfd, filename, mode);
613 }
614 
615 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
616 {
617 	return do_fchmodat(AT_FDCWD, filename, mode);
618 }
619 
620 static int chown_common(const struct path *path, uid_t user, gid_t group)
621 {
622 	struct inode *inode = path->dentry->d_inode;
623 	struct inode *delegated_inode = NULL;
624 	int error;
625 	struct iattr newattrs;
626 	kuid_t uid;
627 	kgid_t gid;
628 
629 	uid = make_kuid(current_user_ns(), user);
630 	gid = make_kgid(current_user_ns(), group);
631 
632 retry_deleg:
633 	newattrs.ia_valid =  ATTR_CTIME;
634 	if (user != (uid_t) -1) {
635 		if (!uid_valid(uid))
636 			return -EINVAL;
637 		newattrs.ia_valid |= ATTR_UID;
638 		newattrs.ia_uid = uid;
639 	}
640 	if (group != (gid_t) -1) {
641 		if (!gid_valid(gid))
642 			return -EINVAL;
643 		newattrs.ia_valid |= ATTR_GID;
644 		newattrs.ia_gid = gid;
645 	}
646 	if (!S_ISDIR(inode->i_mode))
647 		newattrs.ia_valid |=
648 			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
649 	inode_lock(inode);
650 	error = security_path_chown(path, uid, gid);
651 	if (!error)
652 		error = notify_change(path->dentry, &newattrs, &delegated_inode);
653 	inode_unlock(inode);
654 	if (delegated_inode) {
655 		error = break_deleg_wait(&delegated_inode);
656 		if (!error)
657 			goto retry_deleg;
658 	}
659 	return error;
660 }
661 
662 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
663 		int flag)
664 {
665 	struct path path;
666 	int error = -EINVAL;
667 	int lookup_flags;
668 
669 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
670 		goto out;
671 
672 	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
673 	if (flag & AT_EMPTY_PATH)
674 		lookup_flags |= LOOKUP_EMPTY;
675 retry:
676 	error = user_path_at(dfd, filename, lookup_flags, &path);
677 	if (error)
678 		goto out;
679 	error = mnt_want_write(path.mnt);
680 	if (error)
681 		goto out_release;
682 	error = chown_common(&path, user, group);
683 	mnt_drop_write(path.mnt);
684 out_release:
685 	path_put(&path);
686 	if (retry_estale(error, lookup_flags)) {
687 		lookup_flags |= LOOKUP_REVAL;
688 		goto retry;
689 	}
690 out:
691 	return error;
692 }
693 
694 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
695 		gid_t, group, int, flag)
696 {
697 	return do_fchownat(dfd, filename, user, group, flag);
698 }
699 
700 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
701 {
702 	return do_fchownat(AT_FDCWD, filename, user, group, 0);
703 }
704 
705 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
706 {
707 	return do_fchownat(AT_FDCWD, filename, user, group,
708 			   AT_SYMLINK_NOFOLLOW);
709 }
710 
711 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
712 {
713 	struct fd f = fdget(fd);
714 	int error = -EBADF;
715 
716 	if (!f.file)
717 		goto out;
718 
719 	error = mnt_want_write_file(f.file);
720 	if (error)
721 		goto out_fput;
722 	audit_file(f.file);
723 	error = chown_common(&f.file->f_path, user, group);
724 	mnt_drop_write_file(f.file);
725 out_fput:
726 	fdput(f);
727 out:
728 	return error;
729 }
730 
731 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
732 {
733 	return ksys_fchown(fd, user, group);
734 }
735 
736 static int do_dentry_open(struct file *f,
737 			  struct inode *inode,
738 			  int (*open)(struct inode *, struct file *))
739 {
740 	static const struct file_operations empty_fops = {};
741 	int error;
742 
743 	path_get(&f->f_path);
744 	f->f_inode = inode;
745 	f->f_mapping = inode->i_mapping;
746 
747 	/* Ensure that we skip any errors that predate opening of the file */
748 	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
749 
750 	if (unlikely(f->f_flags & O_PATH)) {
751 		f->f_mode = FMODE_PATH | FMODE_OPENED;
752 		f->f_op = &empty_fops;
753 		return 0;
754 	}
755 
756 	/* Any file opened for execve()/uselib() has to be a regular file. */
757 	if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
758 		error = -EACCES;
759 		goto cleanup_file;
760 	}
761 
762 	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
763 		error = get_write_access(inode);
764 		if (unlikely(error))
765 			goto cleanup_file;
766 		error = __mnt_want_write(f->f_path.mnt);
767 		if (unlikely(error)) {
768 			put_write_access(inode);
769 			goto cleanup_file;
770 		}
771 		f->f_mode |= FMODE_WRITER;
772 	}
773 
774 	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
775 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
776 		f->f_mode |= FMODE_ATOMIC_POS;
777 
778 	f->f_op = fops_get(inode->i_fop);
779 	if (WARN_ON(!f->f_op)) {
780 		error = -ENODEV;
781 		goto cleanup_all;
782 	}
783 
784 	error = security_file_open(f);
785 	if (error)
786 		goto cleanup_all;
787 
788 	error = break_lease(locks_inode(f), f->f_flags);
789 	if (error)
790 		goto cleanup_all;
791 
792 	/* normally all 3 are set; ->open() can clear them if needed */
793 	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
794 	if (!open)
795 		open = f->f_op->open;
796 	if (open) {
797 		error = open(inode, f);
798 		if (error)
799 			goto cleanup_all;
800 	}
801 	f->f_mode |= FMODE_OPENED;
802 	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
803 		i_readcount_inc(inode);
804 	if ((f->f_mode & FMODE_READ) &&
805 	     likely(f->f_op->read || f->f_op->read_iter))
806 		f->f_mode |= FMODE_CAN_READ;
807 	if ((f->f_mode & FMODE_WRITE) &&
808 	     likely(f->f_op->write || f->f_op->write_iter))
809 		f->f_mode |= FMODE_CAN_WRITE;
810 
811 	f->f_write_hint = WRITE_LIFE_NOT_SET;
812 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
813 
814 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
815 
816 	/* NB: we're sure to have correct a_ops only after f_op->open */
817 	if (f->f_flags & O_DIRECT) {
818 		if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
819 			return -EINVAL;
820 	}
821 
822 	/*
823 	 * XXX: Huge page cache doesn't support writing yet. Drop all page
824 	 * cache for this file before processing writes.
825 	 */
826 	if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
827 		truncate_pagecache(inode, 0);
828 
829 	return 0;
830 
831 cleanup_all:
832 	if (WARN_ON_ONCE(error > 0))
833 		error = -EINVAL;
834 	fops_put(f->f_op);
835 	if (f->f_mode & FMODE_WRITER) {
836 		put_write_access(inode);
837 		__mnt_drop_write(f->f_path.mnt);
838 	}
839 cleanup_file:
840 	path_put(&f->f_path);
841 	f->f_path.mnt = NULL;
842 	f->f_path.dentry = NULL;
843 	f->f_inode = NULL;
844 	return error;
845 }
846 
847 /**
848  * finish_open - finish opening a file
849  * @file: file pointer
850  * @dentry: pointer to dentry
851  * @open: open callback
852  * @opened: state of open
853  *
854  * This can be used to finish opening a file passed to i_op->atomic_open().
855  *
856  * If the open callback is set to NULL, then the standard f_op->open()
857  * filesystem callback is substituted.
858  *
859  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
860  * the return value of d_splice_alias(), then the caller needs to perform dput()
861  * on it after finish_open().
862  *
863  * Returns zero on success or -errno if the open failed.
864  */
865 int finish_open(struct file *file, struct dentry *dentry,
866 		int (*open)(struct inode *, struct file *))
867 {
868 	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
869 
870 	file->f_path.dentry = dentry;
871 	return do_dentry_open(file, d_backing_inode(dentry), open);
872 }
873 EXPORT_SYMBOL(finish_open);
874 
875 /**
876  * finish_no_open - finish ->atomic_open() without opening the file
877  *
878  * @file: file pointer
879  * @dentry: dentry or NULL (as returned from ->lookup())
880  *
881  * This can be used to set the result of a successful lookup in ->atomic_open().
882  *
883  * NB: unlike finish_open() this function does consume the dentry reference and
884  * the caller need not dput() it.
885  *
886  * Returns "0" which must be the return value of ->atomic_open() after having
887  * called this function.
888  */
889 int finish_no_open(struct file *file, struct dentry *dentry)
890 {
891 	file->f_path.dentry = dentry;
892 	return 0;
893 }
894 EXPORT_SYMBOL(finish_no_open);
895 
896 char *file_path(struct file *filp, char *buf, int buflen)
897 {
898 	return d_path(&filp->f_path, buf, buflen);
899 }
900 EXPORT_SYMBOL(file_path);
901 
902 /**
903  * vfs_open - open the file at the given path
904  * @path: path to open
905  * @file: newly allocated file with f_flag initialized
906  * @cred: credentials to use
907  */
908 int vfs_open(const struct path *path, struct file *file)
909 {
910 	file->f_path = *path;
911 	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
912 }
913 
914 struct file *dentry_open(const struct path *path, int flags,
915 			 const struct cred *cred)
916 {
917 	int error;
918 	struct file *f;
919 
920 	validate_creds(cred);
921 
922 	/* We must always pass in a valid mount pointer. */
923 	BUG_ON(!path->mnt);
924 
925 	f = alloc_empty_file(flags, cred);
926 	if (!IS_ERR(f)) {
927 		error = vfs_open(path, f);
928 		if (error) {
929 			fput(f);
930 			f = ERR_PTR(error);
931 		}
932 	}
933 	return f;
934 }
935 EXPORT_SYMBOL(dentry_open);
936 
937 struct file *open_with_fake_path(const struct path *path, int flags,
938 				struct inode *inode, const struct cred *cred)
939 {
940 	struct file *f = alloc_empty_file_noaccount(flags, cred);
941 	if (!IS_ERR(f)) {
942 		int error;
943 
944 		f->f_path = *path;
945 		error = do_dentry_open(f, inode, NULL);
946 		if (error) {
947 			fput(f);
948 			f = ERR_PTR(error);
949 		}
950 	}
951 	return f;
952 }
953 EXPORT_SYMBOL(open_with_fake_path);
954 
955 #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
956 #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
957 
958 inline struct open_how build_open_how(int flags, umode_t mode)
959 {
960 	struct open_how how = {
961 		.flags = flags & VALID_OPEN_FLAGS,
962 		.mode = mode & S_IALLUGO,
963 	};
964 
965 	/* O_PATH beats everything else. */
966 	if (how.flags & O_PATH)
967 		how.flags &= O_PATH_FLAGS;
968 	/* Modes should only be set for create-like flags. */
969 	if (!WILL_CREATE(how.flags))
970 		how.mode = 0;
971 	return how;
972 }
973 
974 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
975 {
976 	int flags = how->flags;
977 	int lookup_flags = 0;
978 	int acc_mode = ACC_MODE(flags);
979 
980 	/* Must never be set by userspace */
981 	flags &= ~(FMODE_NONOTIFY | O_CLOEXEC);
982 
983 	/*
984 	 * Older syscalls implicitly clear all of the invalid flags or argument
985 	 * values before calling build_open_flags(), but openat2(2) checks all
986 	 * of its arguments.
987 	 */
988 	if (flags & ~VALID_OPEN_FLAGS)
989 		return -EINVAL;
990 	if (how->resolve & ~VALID_RESOLVE_FLAGS)
991 		return -EINVAL;
992 
993 	/* Deal with the mode. */
994 	if (WILL_CREATE(flags)) {
995 		if (how->mode & ~S_IALLUGO)
996 			return -EINVAL;
997 		op->mode = how->mode | S_IFREG;
998 	} else {
999 		if (how->mode != 0)
1000 			return -EINVAL;
1001 		op->mode = 0;
1002 	}
1003 
1004 	/*
1005 	 * In order to ensure programs get explicit errors when trying to use
1006 	 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1007 	 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1008 	 * have to require userspace to explicitly set it.
1009 	 */
1010 	if (flags & __O_TMPFILE) {
1011 		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1012 			return -EINVAL;
1013 		if (!(acc_mode & MAY_WRITE))
1014 			return -EINVAL;
1015 	}
1016 	if (flags & O_PATH) {
1017 		/* O_PATH only permits certain other flags to be set. */
1018 		if (flags & ~O_PATH_FLAGS)
1019 			return -EINVAL;
1020 		acc_mode = 0;
1021 	}
1022 
1023 	/*
1024 	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1025 	 * check for O_DSYNC if the need any syncing at all we enforce it's
1026 	 * always set instead of having to deal with possibly weird behaviour
1027 	 * for malicious applications setting only __O_SYNC.
1028 	 */
1029 	if (flags & __O_SYNC)
1030 		flags |= O_DSYNC;
1031 
1032 	op->open_flag = flags;
1033 
1034 	/* O_TRUNC implies we need access checks for write permissions */
1035 	if (flags & O_TRUNC)
1036 		acc_mode |= MAY_WRITE;
1037 
1038 	/* Allow the LSM permission hook to distinguish append
1039 	   access from general write access. */
1040 	if (flags & O_APPEND)
1041 		acc_mode |= MAY_APPEND;
1042 
1043 	op->acc_mode = acc_mode;
1044 
1045 	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1046 
1047 	if (flags & O_CREAT) {
1048 		op->intent |= LOOKUP_CREATE;
1049 		if (flags & O_EXCL) {
1050 			op->intent |= LOOKUP_EXCL;
1051 			flags |= O_NOFOLLOW;
1052 		}
1053 	}
1054 
1055 	if (flags & O_DIRECTORY)
1056 		lookup_flags |= LOOKUP_DIRECTORY;
1057 	if (!(flags & O_NOFOLLOW))
1058 		lookup_flags |= LOOKUP_FOLLOW;
1059 
1060 	if (how->resolve & RESOLVE_NO_XDEV)
1061 		lookup_flags |= LOOKUP_NO_XDEV;
1062 	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1063 		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1064 	if (how->resolve & RESOLVE_NO_SYMLINKS)
1065 		lookup_flags |= LOOKUP_NO_SYMLINKS;
1066 	if (how->resolve & RESOLVE_BENEATH)
1067 		lookup_flags |= LOOKUP_BENEATH;
1068 	if (how->resolve & RESOLVE_IN_ROOT)
1069 		lookup_flags |= LOOKUP_IN_ROOT;
1070 
1071 	op->lookup_flags = lookup_flags;
1072 	return 0;
1073 }
1074 
1075 /**
1076  * file_open_name - open file and return file pointer
1077  *
1078  * @name:	struct filename containing path to open
1079  * @flags:	open flags as per the open(2) second argument
1080  * @mode:	mode for the new file if O_CREAT is set, else ignored
1081  *
1082  * This is the helper to open a file from kernelspace if you really
1083  * have to.  But in generally you should not do this, so please move
1084  * along, nothing to see here..
1085  */
1086 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1087 {
1088 	struct open_flags op;
1089 	struct open_how how = build_open_how(flags, mode);
1090 	int err = build_open_flags(&how, &op);
1091 	if (err)
1092 		return ERR_PTR(err);
1093 	return do_filp_open(AT_FDCWD, name, &op);
1094 }
1095 
1096 /**
1097  * filp_open - open file and return file pointer
1098  *
1099  * @filename:	path to open
1100  * @flags:	open flags as per the open(2) second argument
1101  * @mode:	mode for the new file if O_CREAT is set, else ignored
1102  *
1103  * This is the helper to open a file from kernelspace if you really
1104  * have to.  But in generally you should not do this, so please move
1105  * along, nothing to see here..
1106  */
1107 struct file *filp_open(const char *filename, int flags, umode_t mode)
1108 {
1109 	struct filename *name = getname_kernel(filename);
1110 	struct file *file = ERR_CAST(name);
1111 
1112 	if (!IS_ERR(name)) {
1113 		file = file_open_name(name, flags, mode);
1114 		putname(name);
1115 	}
1116 	return file;
1117 }
1118 EXPORT_SYMBOL(filp_open);
1119 
1120 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1121 			    const char *filename, int flags, umode_t mode)
1122 {
1123 	struct open_flags op;
1124 	struct open_how how = build_open_how(flags, mode);
1125 	int err = build_open_flags(&how, &op);
1126 	if (err)
1127 		return ERR_PTR(err);
1128 	return do_file_open_root(dentry, mnt, filename, &op);
1129 }
1130 EXPORT_SYMBOL(file_open_root);
1131 
1132 static long do_sys_openat2(int dfd, const char __user *filename,
1133 			   struct open_how *how)
1134 {
1135 	struct open_flags op;
1136 	int fd = build_open_flags(how, &op);
1137 	struct filename *tmp;
1138 
1139 	if (fd)
1140 		return fd;
1141 
1142 	tmp = getname(filename);
1143 	if (IS_ERR(tmp))
1144 		return PTR_ERR(tmp);
1145 
1146 	fd = get_unused_fd_flags(how->flags);
1147 	if (fd >= 0) {
1148 		struct file *f = do_filp_open(dfd, tmp, &op);
1149 		if (IS_ERR(f)) {
1150 			put_unused_fd(fd);
1151 			fd = PTR_ERR(f);
1152 		} else {
1153 			fsnotify_open(f);
1154 			fd_install(fd, f);
1155 		}
1156 	}
1157 	putname(tmp);
1158 	return fd;
1159 }
1160 
1161 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1162 {
1163 	struct open_how how = build_open_how(flags, mode);
1164 	return do_sys_openat2(dfd, filename, &how);
1165 }
1166 
1167 
1168 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1169 {
1170 	return ksys_open(filename, flags, mode);
1171 }
1172 
1173 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1174 		umode_t, mode)
1175 {
1176 	if (force_o_largefile())
1177 		flags |= O_LARGEFILE;
1178 	return do_sys_open(dfd, filename, flags, mode);
1179 }
1180 
1181 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1182 		struct open_how __user *, how, size_t, usize)
1183 {
1184 	int err;
1185 	struct open_how tmp;
1186 
1187 	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1188 	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1189 
1190 	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1191 		return -EINVAL;
1192 
1193 	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1194 	if (err)
1195 		return err;
1196 
1197 	/* O_LARGEFILE is only allowed for non-O_PATH. */
1198 	if (!(tmp.flags & O_PATH) && force_o_largefile())
1199 		tmp.flags |= O_LARGEFILE;
1200 
1201 	return do_sys_openat2(dfd, filename, &tmp);
1202 }
1203 
1204 #ifdef CONFIG_COMPAT
1205 /*
1206  * Exactly like sys_open(), except that it doesn't set the
1207  * O_LARGEFILE flag.
1208  */
1209 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1210 {
1211 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1212 }
1213 
1214 /*
1215  * Exactly like sys_openat(), except that it doesn't set the
1216  * O_LARGEFILE flag.
1217  */
1218 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1219 {
1220 	return do_sys_open(dfd, filename, flags, mode);
1221 }
1222 #endif
1223 
1224 #ifndef __alpha__
1225 
1226 /*
1227  * For backward compatibility?  Maybe this should be moved
1228  * into arch/i386 instead?
1229  */
1230 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1231 {
1232 	return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1233 }
1234 
1235 #endif
1236 
1237 /*
1238  * "id" is the POSIX thread ID. We use the
1239  * files pointer for this..
1240  */
1241 int filp_close(struct file *filp, fl_owner_t id)
1242 {
1243 	int retval = 0;
1244 
1245 	if (!file_count(filp)) {
1246 		printk(KERN_ERR "VFS: Close: file count is 0\n");
1247 		return 0;
1248 	}
1249 
1250 	if (filp->f_op->flush)
1251 		retval = filp->f_op->flush(filp, id);
1252 
1253 	if (likely(!(filp->f_mode & FMODE_PATH))) {
1254 		dnotify_flush(filp, id);
1255 		locks_remove_posix(filp, id);
1256 	}
1257 	fput(filp);
1258 	return retval;
1259 }
1260 
1261 EXPORT_SYMBOL(filp_close);
1262 
1263 /*
1264  * Careful here! We test whether the file pointer is NULL before
1265  * releasing the fd. This ensures that one clone task can't release
1266  * an fd while another clone is opening it.
1267  */
1268 SYSCALL_DEFINE1(close, unsigned int, fd)
1269 {
1270 	int retval = __close_fd(current->files, fd);
1271 
1272 	/* can't restart close syscall because file table entry was cleared */
1273 	if (unlikely(retval == -ERESTARTSYS ||
1274 		     retval == -ERESTARTNOINTR ||
1275 		     retval == -ERESTARTNOHAND ||
1276 		     retval == -ERESTART_RESTARTBLOCK))
1277 		retval = -EINTR;
1278 
1279 	return retval;
1280 }
1281 
1282 /*
1283  * This routine simulates a hangup on the tty, to arrange that users
1284  * are given clean terminals at login time.
1285  */
1286 SYSCALL_DEFINE0(vhangup)
1287 {
1288 	if (capable(CAP_SYS_TTY_CONFIG)) {
1289 		tty_vhangup_self();
1290 		return 0;
1291 	}
1292 	return -EPERM;
1293 }
1294 
1295 /*
1296  * Called when an inode is about to be open.
1297  * We use this to disallow opening large files on 32bit systems if
1298  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1299  * on this flag in sys_open.
1300  */
1301 int generic_file_open(struct inode * inode, struct file * filp)
1302 {
1303 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1304 		return -EOVERFLOW;
1305 	return 0;
1306 }
1307 
1308 EXPORT_SYMBOL(generic_file_open);
1309 
1310 /*
1311  * This is used by subsystems that don't want seekable
1312  * file descriptors. The function is not supposed to ever fail, the only
1313  * reason it returns an 'int' and not 'void' is so that it can be plugged
1314  * directly into file_operations structure.
1315  */
1316 int nonseekable_open(struct inode *inode, struct file *filp)
1317 {
1318 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1319 	return 0;
1320 }
1321 
1322 EXPORT_SYMBOL(nonseekable_open);
1323 
1324 /*
1325  * stream_open is used by subsystems that want stream-like file descriptors.
1326  * Such file descriptors are not seekable and don't have notion of position
1327  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1328  * Contrary to file descriptors of other regular files, .read() and .write()
1329  * can run simultaneously.
1330  *
1331  * stream_open never fails and is marked to return int so that it could be
1332  * directly used as file_operations.open .
1333  */
1334 int stream_open(struct inode *inode, struct file *filp)
1335 {
1336 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1337 	filp->f_mode |= FMODE_STREAM;
1338 	return 0;
1339 }
1340 
1341 EXPORT_SYMBOL(stream_open);
1342