xref: /linux/fs/open.c (revision 7fc2cd2e4b398c57c9cf961cfea05eadbf34c05c)
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/dnotify.h>
33 #include <linux/compat.h>
34 #include <linux/mnt_idmapping.h>
35 #include <linux/filelock.h>
36 
37 #include "internal.h"
38 
39 int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
40 		loff_t length, unsigned int time_attrs, struct file *filp)
41 {
42 	int ret;
43 	struct iattr newattrs;
44 
45 	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
46 	if (length < 0)
47 		return -EINVAL;
48 
49 	newattrs.ia_size = length;
50 	newattrs.ia_valid = ATTR_SIZE | time_attrs;
51 	if (filp) {
52 		newattrs.ia_file = filp;
53 		newattrs.ia_valid |= ATTR_FILE;
54 	}
55 
56 	/* Remove suid, sgid, and file capabilities on truncate too */
57 	ret = dentry_needs_remove_privs(idmap, dentry);
58 	if (ret < 0)
59 		return ret;
60 	if (ret)
61 		newattrs.ia_valid |= ret | ATTR_FORCE;
62 
63 	ret = inode_lock_killable(dentry->d_inode);
64 	if (ret)
65 		return ret;
66 
67 	/* Note any delegations or leases have already been broken: */
68 	ret = notify_change(idmap, dentry, &newattrs, NULL);
69 	inode_unlock(dentry->d_inode);
70 	return ret;
71 }
72 
73 int vfs_truncate(const struct path *path, loff_t length)
74 {
75 	struct mnt_idmap *idmap;
76 	struct inode *inode;
77 	int error;
78 
79 	inode = path->dentry->d_inode;
80 
81 	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
82 	if (S_ISDIR(inode->i_mode))
83 		return -EISDIR;
84 	if (!S_ISREG(inode->i_mode))
85 		return -EINVAL;
86 
87 	idmap = mnt_idmap(path->mnt);
88 	error = inode_permission(idmap, inode, MAY_WRITE);
89 	if (error)
90 		return error;
91 
92 	error = fsnotify_truncate_perm(path, length);
93 	if (error)
94 		return error;
95 
96 	error = mnt_want_write(path->mnt);
97 	if (error)
98 		return error;
99 
100 	error = -EPERM;
101 	if (IS_APPEND(inode))
102 		goto mnt_drop_write_and_out;
103 
104 	error = get_write_access(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 = security_path_truncate(path);
117 	if (!error)
118 		error = do_truncate(idmap, path->dentry, length, 0, NULL);
119 
120 put_write_and_out:
121 	put_write_access(inode);
122 mnt_drop_write_and_out:
123 	mnt_drop_write(path->mnt);
124 
125 	return error;
126 }
127 EXPORT_SYMBOL_GPL(vfs_truncate);
128 
129 int do_sys_truncate(const char __user *pathname, loff_t length)
130 {
131 	unsigned int lookup_flags = LOOKUP_FOLLOW;
132 	struct path path;
133 	int error;
134 
135 	if (length < 0)	/* sorry, but loff_t says... */
136 		return -EINVAL;
137 
138 retry:
139 	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
140 	if (!error) {
141 		error = vfs_truncate(&path, length);
142 		path_put(&path);
143 	}
144 	if (retry_estale(error, lookup_flags)) {
145 		lookup_flags |= LOOKUP_REVAL;
146 		goto retry;
147 	}
148 	return error;
149 }
150 
151 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
152 {
153 	return do_sys_truncate(path, length);
154 }
155 
156 #ifdef CONFIG_COMPAT
157 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
158 {
159 	return do_sys_truncate(path, length);
160 }
161 #endif
162 
163 int do_ftruncate(struct file *file, loff_t length, int small)
164 {
165 	struct inode *inode;
166 	struct dentry *dentry;
167 	int error;
168 
169 	/* explicitly opened as large or we are on 64-bit box */
170 	if (file->f_flags & O_LARGEFILE)
171 		small = 0;
172 
173 	dentry = file->f_path.dentry;
174 	inode = dentry->d_inode;
175 	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
176 		return -EINVAL;
177 
178 	/* Cannot ftruncate over 2^31 bytes without large file support */
179 	if (small && length > MAX_NON_LFS)
180 		return -EINVAL;
181 
182 	/* Check IS_APPEND on real upper inode */
183 	if (IS_APPEND(file_inode(file)))
184 		return -EPERM;
185 
186 	error = security_file_truncate(file);
187 	if (error)
188 		return error;
189 
190 	error = fsnotify_truncate_perm(&file->f_path, length);
191 	if (error)
192 		return error;
193 
194 	scoped_guard(super_write, inode->i_sb)
195 		return do_truncate(file_mnt_idmap(file), dentry, length,
196 				   ATTR_MTIME | ATTR_CTIME, file);
197 }
198 
199 int do_sys_ftruncate(unsigned int fd, loff_t length, int small)
200 {
201 	if (length < 0)
202 		return -EINVAL;
203 	CLASS(fd, f)(fd);
204 	if (fd_empty(f))
205 		return -EBADF;
206 
207 	return do_ftruncate(fd_file(f), length, small);
208 }
209 
210 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
211 {
212 	return do_sys_ftruncate(fd, length, 1);
213 }
214 
215 #ifdef CONFIG_COMPAT
216 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
217 {
218 	return do_sys_ftruncate(fd, length, 1);
219 }
220 #endif
221 
222 /* LFS versions of truncate are only needed on 32 bit machines */
223 #if BITS_PER_LONG == 32
224 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
225 {
226 	return do_sys_truncate(path, length);
227 }
228 
229 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
230 {
231 	return do_sys_ftruncate(fd, length, 0);
232 }
233 #endif /* BITS_PER_LONG == 32 */
234 
235 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
236 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
237 		       compat_arg_u64_dual(length))
238 {
239 	return ksys_truncate(pathname, compat_arg_u64_glue(length));
240 }
241 #endif
242 
243 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
244 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
245 		       compat_arg_u64_dual(length))
246 {
247 	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
248 }
249 #endif
250 
251 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
252 {
253 	struct inode *inode = file_inode(file);
254 	int ret;
255 	loff_t sum;
256 
257 	if (offset < 0 || len <= 0)
258 		return -EINVAL;
259 
260 	if (mode & ~(FALLOC_FL_MODE_MASK | FALLOC_FL_KEEP_SIZE))
261 		return -EOPNOTSUPP;
262 
263 	/*
264 	 * Modes are exclusive, even if that is not obvious from the encoding
265 	 * as bit masks and the mix with the flag in the same namespace.
266 	 *
267 	 * To make things even more complicated, FALLOC_FL_ALLOCATE_RANGE is
268 	 * encoded as no bit set.
269 	 */
270 	switch (mode & FALLOC_FL_MODE_MASK) {
271 	case FALLOC_FL_ALLOCATE_RANGE:
272 	case FALLOC_FL_UNSHARE_RANGE:
273 	case FALLOC_FL_ZERO_RANGE:
274 		break;
275 	case FALLOC_FL_PUNCH_HOLE:
276 		if (!(mode & FALLOC_FL_KEEP_SIZE))
277 			return -EOPNOTSUPP;
278 		break;
279 	case FALLOC_FL_COLLAPSE_RANGE:
280 	case FALLOC_FL_INSERT_RANGE:
281 	case FALLOC_FL_WRITE_ZEROES:
282 		if (mode & FALLOC_FL_KEEP_SIZE)
283 			return -EOPNOTSUPP;
284 		break;
285 	default:
286 		return -EOPNOTSUPP;
287 	}
288 
289 	if (!(file->f_mode & FMODE_WRITE))
290 		return -EBADF;
291 
292 	/*
293 	 * On append-only files only space preallocation is supported.
294 	 */
295 	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
296 		return -EPERM;
297 
298 	if (IS_IMMUTABLE(inode))
299 		return -EPERM;
300 
301 	/*
302 	 * We cannot allow any fallocate operation on an active swapfile
303 	 */
304 	if (IS_SWAPFILE(inode))
305 		return -ETXTBSY;
306 
307 	/*
308 	 * Revalidate the write permissions, in case security policy has
309 	 * changed since the files were opened.
310 	 */
311 	ret = security_file_permission(file, MAY_WRITE);
312 	if (ret)
313 		return ret;
314 
315 	ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
316 	if (ret)
317 		return ret;
318 
319 	if (S_ISFIFO(inode->i_mode))
320 		return -ESPIPE;
321 
322 	if (S_ISDIR(inode->i_mode))
323 		return -EISDIR;
324 
325 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
326 		return -ENODEV;
327 
328 	/* Check for wraparound */
329 	if (check_add_overflow(offset, len, &sum))
330 		return -EFBIG;
331 
332 	if (sum > inode->i_sb->s_maxbytes)
333 		return -EFBIG;
334 
335 	if (!file->f_op->fallocate)
336 		return -EOPNOTSUPP;
337 
338 	file_start_write(file);
339 	ret = file->f_op->fallocate(file, mode, offset, len);
340 
341 	/*
342 	 * Create inotify and fanotify events.
343 	 *
344 	 * To keep the logic simple always create events if fallocate succeeds.
345 	 * This implies that events are even created if the file size remains
346 	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
347 	 */
348 	if (ret == 0)
349 		fsnotify_modify(file);
350 
351 	file_end_write(file);
352 	return ret;
353 }
354 EXPORT_SYMBOL_GPL(vfs_fallocate);
355 
356 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
357 {
358 	CLASS(fd, f)(fd);
359 
360 	if (fd_empty(f))
361 		return -EBADF;
362 
363 	return vfs_fallocate(fd_file(f), mode, offset, len);
364 }
365 
366 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
367 {
368 	return ksys_fallocate(fd, mode, offset, len);
369 }
370 
371 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
372 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
373 		       compat_arg_u64_dual(len))
374 {
375 	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
376 			      compat_arg_u64_glue(len));
377 }
378 #endif
379 
380 /*
381  * access() needs to use the real uid/gid, not the effective uid/gid.
382  * We do this by temporarily clearing all FS-related capabilities and
383  * switching the fsuid/fsgid around to the real ones.
384  *
385  * Creating new credentials is expensive, so we try to skip doing it,
386  * which we can if the result would match what we already got.
387  */
388 static bool access_need_override_creds(int flags)
389 {
390 	const struct cred *cred;
391 
392 	if (flags & AT_EACCESS)
393 		return false;
394 
395 	cred = current_cred();
396 	if (!uid_eq(cred->fsuid, cred->uid) ||
397 	    !gid_eq(cred->fsgid, cred->gid))
398 		return true;
399 
400 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
401 		kuid_t root_uid = make_kuid(cred->user_ns, 0);
402 		if (!uid_eq(cred->uid, root_uid)) {
403 			if (!cap_isclear(cred->cap_effective))
404 				return true;
405 		} else {
406 			if (!cap_isidentical(cred->cap_effective,
407 			    cred->cap_permitted))
408 				return true;
409 		}
410 	}
411 
412 	return false;
413 }
414 
415 static const struct cred *access_override_creds(void)
416 {
417 	struct cred *override_cred;
418 
419 	override_cred = prepare_creds();
420 	if (!override_cred)
421 		return NULL;
422 
423 	/*
424 	 * XXX access_need_override_creds performs checks in hopes of skipping
425 	 * this work. Make sure it stays in sync if making any changes in this
426 	 * routine.
427 	 */
428 
429 	override_cred->fsuid = override_cred->uid;
430 	override_cred->fsgid = override_cred->gid;
431 
432 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
433 		/* Clear the capabilities if we switch to a non-root user */
434 		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
435 		if (!uid_eq(override_cred->uid, root_uid))
436 			cap_clear(override_cred->cap_effective);
437 		else
438 			override_cred->cap_effective =
439 				override_cred->cap_permitted;
440 	}
441 
442 	/*
443 	 * The new set of credentials can *only* be used in
444 	 * task-synchronous circumstances, and does not need
445 	 * RCU freeing, unless somebody then takes a separate
446 	 * reference to it.
447 	 *
448 	 * NOTE! This is _only_ true because this credential
449 	 * is used purely for override_creds() that installs
450 	 * it as the subjective cred. Other threads will be
451 	 * accessing ->real_cred, not the subjective cred.
452 	 *
453 	 * If somebody _does_ make a copy of this (using the
454 	 * 'get_current_cred()' function), that will clear the
455 	 * non_rcu field, because now that other user may be
456 	 * expecting RCU freeing. But normal thread-synchronous
457 	 * cred accesses will keep things non-racy to avoid RCU
458 	 * freeing.
459 	 */
460 	override_cred->non_rcu = 1;
461 	return override_creds(override_cred);
462 }
463 
464 static int do_faccessat(int dfd, const char __user *filename, int mode, int flags)
465 {
466 	struct path path;
467 	struct inode *inode;
468 	int res;
469 	unsigned int lookup_flags = LOOKUP_FOLLOW;
470 	const struct cred *old_cred = NULL;
471 
472 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
473 		return -EINVAL;
474 
475 	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
476 		return -EINVAL;
477 
478 	if (flags & AT_SYMLINK_NOFOLLOW)
479 		lookup_flags &= ~LOOKUP_FOLLOW;
480 	if (flags & AT_EMPTY_PATH)
481 		lookup_flags |= LOOKUP_EMPTY;
482 
483 	if (access_need_override_creds(flags)) {
484 		old_cred = access_override_creds();
485 		if (!old_cred)
486 			return -ENOMEM;
487 	}
488 
489 retry:
490 	res = user_path_at(dfd, filename, lookup_flags, &path);
491 	if (res)
492 		goto out;
493 
494 	inode = d_backing_inode(path.dentry);
495 
496 	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
497 		/*
498 		 * MAY_EXEC on regular files is denied if the fs is mounted
499 		 * with the "noexec" flag.
500 		 */
501 		res = -EACCES;
502 		if (path_noexec(&path))
503 			goto out_path_release;
504 	}
505 
506 	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
507 	/* SuS v2 requires we report a read only fs too */
508 	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
509 		goto out_path_release;
510 	/*
511 	 * This is a rare case where using __mnt_is_readonly()
512 	 * is OK without a mnt_want/drop_write() pair.  Since
513 	 * no actual write to the fs is performed here, we do
514 	 * not need to telegraph to that to anyone.
515 	 *
516 	 * By doing this, we accept that this access is
517 	 * inherently racy and know that the fs may change
518 	 * state before we even see this result.
519 	 */
520 	if (__mnt_is_readonly(path.mnt))
521 		res = -EROFS;
522 
523 out_path_release:
524 	path_put(&path);
525 	if (retry_estale(res, lookup_flags)) {
526 		lookup_flags |= LOOKUP_REVAL;
527 		goto retry;
528 	}
529 out:
530 	if (old_cred)
531 		put_cred(revert_creds(old_cred));
532 
533 	return res;
534 }
535 
536 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
537 {
538 	return do_faccessat(dfd, filename, mode, 0);
539 }
540 
541 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
542 		int, flags)
543 {
544 	return do_faccessat(dfd, filename, mode, flags);
545 }
546 
547 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
548 {
549 	return do_faccessat(AT_FDCWD, filename, mode, 0);
550 }
551 
552 SYSCALL_DEFINE1(chdir, const char __user *, filename)
553 {
554 	struct path path;
555 	int error;
556 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
557 retry:
558 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
559 	if (error)
560 		goto out;
561 
562 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
563 	if (error)
564 		goto dput_and_out;
565 
566 	set_fs_pwd(current->fs, &path);
567 
568 dput_and_out:
569 	path_put(&path);
570 	if (retry_estale(error, lookup_flags)) {
571 		lookup_flags |= LOOKUP_REVAL;
572 		goto retry;
573 	}
574 out:
575 	return error;
576 }
577 
578 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
579 {
580 	CLASS(fd_raw, f)(fd);
581 	int error;
582 
583 	if (fd_empty(f))
584 		return -EBADF;
585 
586 	if (!d_can_lookup(fd_file(f)->f_path.dentry))
587 		return -ENOTDIR;
588 
589 	error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
590 	if (!error)
591 		set_fs_pwd(current->fs, &fd_file(f)->f_path);
592 	return error;
593 }
594 
595 SYSCALL_DEFINE1(chroot, const char __user *, filename)
596 {
597 	struct path path;
598 	int error;
599 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
600 retry:
601 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
602 	if (error)
603 		goto out;
604 
605 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
606 	if (error)
607 		goto dput_and_out;
608 
609 	error = -EPERM;
610 	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
611 		goto dput_and_out;
612 	error = security_path_chroot(&path);
613 	if (error)
614 		goto dput_and_out;
615 
616 	set_fs_root(current->fs, &path);
617 	error = 0;
618 dput_and_out:
619 	path_put(&path);
620 	if (retry_estale(error, lookup_flags)) {
621 		lookup_flags |= LOOKUP_REVAL;
622 		goto retry;
623 	}
624 out:
625 	return error;
626 }
627 
628 int chmod_common(const struct path *path, umode_t mode)
629 {
630 	struct inode *inode = path->dentry->d_inode;
631 	struct delegated_inode delegated_inode = { };
632 	struct iattr newattrs;
633 	int error;
634 
635 	error = mnt_want_write(path->mnt);
636 	if (error)
637 		return error;
638 retry_deleg:
639 	error = inode_lock_killable(inode);
640 	if (error)
641 		goto out_mnt_unlock;
642 	error = security_path_chmod(path, mode);
643 	if (error)
644 		goto out_unlock;
645 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
646 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
647 	error = notify_change(mnt_idmap(path->mnt), path->dentry,
648 			      &newattrs, &delegated_inode);
649 out_unlock:
650 	inode_unlock(inode);
651 	if (is_delegated(&delegated_inode)) {
652 		error = break_deleg_wait(&delegated_inode);
653 		if (!error)
654 			goto retry_deleg;
655 	}
656 out_mnt_unlock:
657 	mnt_drop_write(path->mnt);
658 	return error;
659 }
660 
661 int vfs_fchmod(struct file *file, umode_t mode)
662 {
663 	audit_file(file);
664 	return chmod_common(&file->f_path, mode);
665 }
666 
667 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
668 {
669 	CLASS(fd, f)(fd);
670 
671 	if (fd_empty(f))
672 		return -EBADF;
673 
674 	return vfs_fchmod(fd_file(f), mode);
675 }
676 
677 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
678 		       unsigned int flags)
679 {
680 	struct path path;
681 	int error;
682 	unsigned int lookup_flags;
683 
684 	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
685 		return -EINVAL;
686 
687 	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
688 	if (flags & AT_EMPTY_PATH)
689 		lookup_flags |= LOOKUP_EMPTY;
690 
691 retry:
692 	error = user_path_at(dfd, filename, lookup_flags, &path);
693 	if (!error) {
694 		error = chmod_common(&path, mode);
695 		path_put(&path);
696 		if (retry_estale(error, lookup_flags)) {
697 			lookup_flags |= LOOKUP_REVAL;
698 			goto retry;
699 		}
700 	}
701 	return error;
702 }
703 
704 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
705 		umode_t, mode, unsigned int, flags)
706 {
707 	return do_fchmodat(dfd, filename, mode, flags);
708 }
709 
710 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
711 		umode_t, mode)
712 {
713 	return do_fchmodat(dfd, filename, mode, 0);
714 }
715 
716 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
717 {
718 	return do_fchmodat(AT_FDCWD, filename, mode, 0);
719 }
720 
721 /*
722  * Check whether @kuid is valid and if so generate and set vfsuid_t in
723  * ia_vfsuid.
724  *
725  * Return: true if @kuid is valid, false if not.
726  */
727 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
728 {
729 	if (!uid_valid(kuid))
730 		return false;
731 	attr->ia_valid |= ATTR_UID;
732 	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
733 	return true;
734 }
735 
736 /*
737  * Check whether @kgid is valid and if so generate and set vfsgid_t in
738  * ia_vfsgid.
739  *
740  * Return: true if @kgid is valid, false if not.
741  */
742 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
743 {
744 	if (!gid_valid(kgid))
745 		return false;
746 	attr->ia_valid |= ATTR_GID;
747 	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
748 	return true;
749 }
750 
751 int chown_common(const struct path *path, uid_t user, gid_t group)
752 {
753 	struct mnt_idmap *idmap;
754 	struct user_namespace *fs_userns;
755 	struct inode *inode = path->dentry->d_inode;
756 	struct delegated_inode delegated_inode = { };
757 	int error;
758 	struct iattr newattrs;
759 	kuid_t uid;
760 	kgid_t gid;
761 
762 	uid = make_kuid(current_user_ns(), user);
763 	gid = make_kgid(current_user_ns(), group);
764 
765 	idmap = mnt_idmap(path->mnt);
766 	fs_userns = i_user_ns(inode);
767 
768 retry_deleg:
769 	newattrs.ia_vfsuid = INVALID_VFSUID;
770 	newattrs.ia_vfsgid = INVALID_VFSGID;
771 	newattrs.ia_valid =  ATTR_CTIME;
772 	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
773 		return -EINVAL;
774 	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
775 		return -EINVAL;
776 	error = inode_lock_killable(inode);
777 	if (error)
778 		return error;
779 	if (!S_ISDIR(inode->i_mode))
780 		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
781 				     setattr_should_drop_sgid(idmap, inode);
782 	/* Continue to send actual fs values, not the mount values. */
783 	error = security_path_chown(
784 		path,
785 		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
786 		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
787 	if (!error)
788 		error = notify_change(idmap, path->dentry, &newattrs,
789 				      &delegated_inode);
790 	inode_unlock(inode);
791 	if (is_delegated(&delegated_inode)) {
792 		error = break_deleg_wait(&delegated_inode);
793 		if (!error)
794 			goto retry_deleg;
795 	}
796 	return error;
797 }
798 
799 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
800 		int flag)
801 {
802 	struct path path;
803 	int error = -EINVAL;
804 	int lookup_flags;
805 
806 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
807 		goto out;
808 
809 	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
810 	if (flag & AT_EMPTY_PATH)
811 		lookup_flags |= LOOKUP_EMPTY;
812 retry:
813 	error = user_path_at(dfd, filename, lookup_flags, &path);
814 	if (error)
815 		goto out;
816 	error = mnt_want_write(path.mnt);
817 	if (error)
818 		goto out_release;
819 	error = chown_common(&path, user, group);
820 	mnt_drop_write(path.mnt);
821 out_release:
822 	path_put(&path);
823 	if (retry_estale(error, lookup_flags)) {
824 		lookup_flags |= LOOKUP_REVAL;
825 		goto retry;
826 	}
827 out:
828 	return error;
829 }
830 
831 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
832 		gid_t, group, int, flag)
833 {
834 	return do_fchownat(dfd, filename, user, group, flag);
835 }
836 
837 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
838 {
839 	return do_fchownat(AT_FDCWD, filename, user, group, 0);
840 }
841 
842 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
843 {
844 	return do_fchownat(AT_FDCWD, filename, user, group,
845 			   AT_SYMLINK_NOFOLLOW);
846 }
847 
848 int vfs_fchown(struct file *file, uid_t user, gid_t group)
849 {
850 	int error;
851 
852 	error = mnt_want_write_file(file);
853 	if (error)
854 		return error;
855 	audit_file(file);
856 	error = chown_common(&file->f_path, user, group);
857 	mnt_drop_write_file(file);
858 	return error;
859 }
860 
861 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
862 {
863 	CLASS(fd, f)(fd);
864 
865 	if (fd_empty(f))
866 		return -EBADF;
867 
868 	return vfs_fchown(fd_file(f), user, group);
869 }
870 
871 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
872 {
873 	return ksys_fchown(fd, user, group);
874 }
875 
876 static inline int file_get_write_access(struct file *f)
877 {
878 	int error;
879 
880 	error = get_write_access(f->f_inode);
881 	if (unlikely(error))
882 		return error;
883 	error = mnt_get_write_access(f->f_path.mnt);
884 	if (unlikely(error))
885 		goto cleanup_inode;
886 	if (unlikely(f->f_mode & FMODE_BACKING)) {
887 		error = mnt_get_write_access(backing_file_user_path(f)->mnt);
888 		if (unlikely(error))
889 			goto cleanup_mnt;
890 	}
891 	return 0;
892 
893 cleanup_mnt:
894 	mnt_put_write_access(f->f_path.mnt);
895 cleanup_inode:
896 	put_write_access(f->f_inode);
897 	return error;
898 }
899 
900 static int do_dentry_open(struct file *f,
901 			  int (*open)(struct inode *, struct file *))
902 {
903 	static const struct file_operations empty_fops = {};
904 	struct inode *inode = f->f_path.dentry->d_inode;
905 	int error;
906 
907 	path_get(&f->f_path);
908 	f->f_inode = inode;
909 	f->f_mapping = inode->i_mapping;
910 	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
911 	f->f_sb_err = file_sample_sb_err(f);
912 
913 	if (unlikely(f->f_flags & O_PATH)) {
914 		f->f_mode = FMODE_PATH | FMODE_OPENED;
915 		file_set_fsnotify_mode(f, FMODE_NONOTIFY);
916 		f->f_op = &empty_fops;
917 		return 0;
918 	}
919 
920 	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
921 		i_readcount_inc(inode);
922 	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
923 		error = file_get_write_access(f);
924 		if (unlikely(error))
925 			goto cleanup_file;
926 		f->f_mode |= FMODE_WRITER;
927 	}
928 
929 	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
930 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
931 		f->f_mode |= FMODE_ATOMIC_POS;
932 
933 	f->f_op = fops_get(inode->i_fop);
934 	if (WARN_ON(!f->f_op)) {
935 		error = -ENODEV;
936 		goto cleanup_all;
937 	}
938 
939 	error = security_file_open(f);
940 	if (unlikely(error))
941 		goto cleanup_all;
942 
943 	/*
944 	 * Call fsnotify open permission hook and set FMODE_NONOTIFY_* bits
945 	 * according to existing permission watches.
946 	 * If FMODE_NONOTIFY mode was already set for an fanotify fd or for a
947 	 * pseudo file, this call will not change the mode.
948 	 */
949 	error = fsnotify_open_perm_and_set_mode(f);
950 	if (unlikely(error))
951 		goto cleanup_all;
952 
953 	error = break_lease(file_inode(f), f->f_flags);
954 	if (unlikely(error))
955 		goto cleanup_all;
956 
957 	/* normally all 3 are set; ->open() can clear them if needed */
958 	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
959 	if (!open)
960 		open = f->f_op->open;
961 	if (open) {
962 		error = open(inode, f);
963 		if (error)
964 			goto cleanup_all;
965 	}
966 	f->f_mode |= FMODE_OPENED;
967 	if ((f->f_mode & FMODE_READ) &&
968 	     likely(f->f_op->read || f->f_op->read_iter))
969 		f->f_mode |= FMODE_CAN_READ;
970 	if ((f->f_mode & FMODE_WRITE) &&
971 	     likely(f->f_op->write || f->f_op->write_iter))
972 		f->f_mode |= FMODE_CAN_WRITE;
973 	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
974 		f->f_mode &= ~FMODE_LSEEK;
975 	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
976 		f->f_mode |= FMODE_CAN_ODIRECT;
977 
978 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
979 	f->f_iocb_flags = iocb_flags(f);
980 
981 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
982 
983 	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
984 		return -EINVAL;
985 
986 	/*
987 	 * XXX: Huge page cache doesn't support writing yet. Drop all page
988 	 * cache for this file before processing writes.
989 	 */
990 	if (f->f_mode & FMODE_WRITE) {
991 		/*
992 		 * Depends on full fence from get_write_access() to synchronize
993 		 * against collapse_file() regarding i_writecount and nr_thps
994 		 * updates. Ensures subsequent insertion of THPs into the page
995 		 * cache will fail.
996 		 */
997 		if (filemap_nr_thps(inode->i_mapping)) {
998 			struct address_space *mapping = inode->i_mapping;
999 
1000 			filemap_invalidate_lock(inode->i_mapping);
1001 			/*
1002 			 * unmap_mapping_range just need to be called once
1003 			 * here, because the private pages is not need to be
1004 			 * unmapped mapping (e.g. data segment of dynamic
1005 			 * shared libraries here).
1006 			 */
1007 			unmap_mapping_range(mapping, 0, 0, 0);
1008 			truncate_inode_pages(mapping, 0);
1009 			filemap_invalidate_unlock(inode->i_mapping);
1010 		}
1011 	}
1012 
1013 	return 0;
1014 
1015 cleanup_all:
1016 	if (WARN_ON_ONCE(error > 0))
1017 		error = -EINVAL;
1018 	fops_put(f->f_op);
1019 	put_file_access(f);
1020 cleanup_file:
1021 	path_put(&f->f_path);
1022 	f->__f_path.mnt = NULL;
1023 	f->__f_path.dentry = NULL;
1024 	f->f_inode = NULL;
1025 	return error;
1026 }
1027 
1028 /**
1029  * finish_open - finish opening a file
1030  * @file: file pointer
1031  * @dentry: pointer to dentry
1032  * @open: open callback
1033  *
1034  * This can be used to finish opening a file passed to i_op->atomic_open().
1035  *
1036  * If the open callback is set to NULL, then the standard f_op->open()
1037  * filesystem callback is substituted.
1038  *
1039  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1040  * the return value of d_splice_alias(), then the caller needs to perform dput()
1041  * on it after finish_open().
1042  *
1043  * Returns zero on success or -errno if the open failed.
1044  */
1045 int finish_open(struct file *file, struct dentry *dentry,
1046 		int (*open)(struct inode *, struct file *))
1047 {
1048 	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1049 
1050 	file->__f_path.dentry = dentry;
1051 	return do_dentry_open(file, open);
1052 }
1053 EXPORT_SYMBOL(finish_open);
1054 
1055 /**
1056  * finish_no_open - finish ->atomic_open() without opening the file
1057  *
1058  * @file: file pointer
1059  * @dentry: dentry, ERR_PTR(-E...) or NULL (as returned from ->lookup())
1060  *
1061  * This can be used to set the result of a lookup in ->atomic_open().
1062  *
1063  * NB: unlike finish_open() this function does consume the dentry reference and
1064  * the caller need not dput() it.
1065  *
1066  * Returns 0 or -E..., which must be the return value of ->atomic_open() after
1067  * having called this function.
1068  */
1069 int finish_no_open(struct file *file, struct dentry *dentry)
1070 {
1071 	if (IS_ERR(dentry))
1072 		return PTR_ERR(dentry);
1073 	file->__f_path.dentry = dentry;
1074 	return 0;
1075 }
1076 EXPORT_SYMBOL(finish_no_open);
1077 
1078 char *file_path(struct file *filp, char *buf, int buflen)
1079 {
1080 	return d_path(&filp->f_path, buf, buflen);
1081 }
1082 EXPORT_SYMBOL(file_path);
1083 
1084 /**
1085  * vfs_open - open the file at the given path
1086  * @path: path to open
1087  * @file: newly allocated file with f_flag initialized
1088  */
1089 int vfs_open(const struct path *path, struct file *file)
1090 {
1091 	int ret;
1092 
1093 	file->__f_path = *path;
1094 	ret = do_dentry_open(file, NULL);
1095 	if (!ret) {
1096 		/*
1097 		 * Once we return a file with FMODE_OPENED, __fput() will call
1098 		 * fsnotify_close(), so we need fsnotify_open() here for
1099 		 * symmetry.
1100 		 */
1101 		fsnotify_open(file);
1102 	}
1103 	return ret;
1104 }
1105 
1106 struct file *dentry_open(const struct path *path, int flags,
1107 			 const struct cred *cred)
1108 {
1109 	int error;
1110 	struct file *f;
1111 
1112 	/* We must always pass in a valid mount pointer. */
1113 	BUG_ON(!path->mnt);
1114 
1115 	f = alloc_empty_file(flags, cred);
1116 	if (!IS_ERR(f)) {
1117 		error = vfs_open(path, f);
1118 		if (error) {
1119 			fput(f);
1120 			f = ERR_PTR(error);
1121 		}
1122 	}
1123 	return f;
1124 }
1125 EXPORT_SYMBOL(dentry_open);
1126 
1127 struct file *dentry_open_nonotify(const struct path *path, int flags,
1128 				  const struct cred *cred)
1129 {
1130 	struct file *f = alloc_empty_file(flags, cred);
1131 	if (!IS_ERR(f)) {
1132 		int error;
1133 
1134 		file_set_fsnotify_mode(f, FMODE_NONOTIFY);
1135 		error = vfs_open(path, f);
1136 		if (error) {
1137 			fput(f);
1138 			f = ERR_PTR(error);
1139 		}
1140 	}
1141 	return f;
1142 }
1143 
1144 /**
1145  * dentry_create - Create and open a file
1146  * @path: path to create
1147  * @flags: O_ flags
1148  * @mode: mode bits for new file
1149  * @cred: credentials to use
1150  *
1151  * Caller must hold the parent directory's lock, and have prepared
1152  * a negative dentry, placed in @path->dentry, for the new file.
1153  *
1154  * Caller sets @path->mnt to the vfsmount of the filesystem where
1155  * the new file is to be created. The parent directory and the
1156  * negative dentry must reside on the same filesystem instance.
1157  *
1158  * On success, returns a "struct file *". Otherwise a ERR_PTR
1159  * is returned.
1160  */
1161 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1162 			   const struct cred *cred)
1163 {
1164 	struct file *f;
1165 	int error;
1166 
1167 	f = alloc_empty_file(flags, cred);
1168 	if (IS_ERR(f))
1169 		return f;
1170 
1171 	error = vfs_create(mnt_idmap(path->mnt), path->dentry, mode, NULL);
1172 	if (!error)
1173 		error = vfs_open(path, f);
1174 
1175 	if (unlikely(error)) {
1176 		fput(f);
1177 		return ERR_PTR(error);
1178 	}
1179 	return f;
1180 }
1181 EXPORT_SYMBOL(dentry_create);
1182 
1183 /**
1184  * kernel_file_open - open a file for kernel internal use
1185  * @path:	path of the file to open
1186  * @flags:	open flags
1187  * @cred:	credentials for open
1188  *
1189  * Open a file for use by in-kernel consumers. The file is not accounted
1190  * against nr_files and must not be installed into the file descriptor
1191  * table.
1192  *
1193  * Return: Opened file on success, an error pointer on failure.
1194  */
1195 struct file *kernel_file_open(const struct path *path, int flags,
1196 				const struct cred *cred)
1197 {
1198 	struct file *f;
1199 	int error;
1200 
1201 	f = alloc_empty_file_noaccount(flags, cred);
1202 	if (IS_ERR(f))
1203 		return f;
1204 
1205 	error = vfs_open(path, f);
1206 	if (error) {
1207 		fput(f);
1208 		return ERR_PTR(error);
1209 	}
1210 	return f;
1211 }
1212 EXPORT_SYMBOL_GPL(kernel_file_open);
1213 
1214 #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1215 #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1216 
1217 inline struct open_how build_open_how(int flags, umode_t mode)
1218 {
1219 	struct open_how how = {
1220 		.flags = flags & VALID_OPEN_FLAGS,
1221 		.mode = mode & S_IALLUGO,
1222 	};
1223 
1224 	/* O_PATH beats everything else. */
1225 	if (how.flags & O_PATH)
1226 		how.flags &= O_PATH_FLAGS;
1227 	/* Modes should only be set for create-like flags. */
1228 	if (!WILL_CREATE(how.flags))
1229 		how.mode = 0;
1230 	return how;
1231 }
1232 
1233 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1234 {
1235 	u64 flags = how->flags;
1236 	u64 strip = O_CLOEXEC;
1237 	int lookup_flags = 0;
1238 	int acc_mode = ACC_MODE(flags);
1239 
1240 	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1241 			 "struct open_flags doesn't yet handle flags > 32 bits");
1242 
1243 	/*
1244 	 * Strip flags that aren't relevant in determining struct open_flags.
1245 	 */
1246 	flags &= ~strip;
1247 
1248 	/*
1249 	 * Older syscalls implicitly clear all of the invalid flags or argument
1250 	 * values before calling build_open_flags(), but openat2(2) checks all
1251 	 * of its arguments.
1252 	 */
1253 	if (flags & ~VALID_OPEN_FLAGS)
1254 		return -EINVAL;
1255 	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1256 		return -EINVAL;
1257 
1258 	/* Scoping flags are mutually exclusive. */
1259 	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1260 		return -EINVAL;
1261 
1262 	/* Deal with the mode. */
1263 	if (WILL_CREATE(flags)) {
1264 		if (how->mode & ~S_IALLUGO)
1265 			return -EINVAL;
1266 		op->mode = how->mode | S_IFREG;
1267 	} else {
1268 		if (how->mode != 0)
1269 			return -EINVAL;
1270 		op->mode = 0;
1271 	}
1272 
1273 	/*
1274 	 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1275 	 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1276 	 * O_TMPFILE below which requires O_DIRECTORY being raised.
1277 	 */
1278 	if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1279 		return -EINVAL;
1280 
1281 	/* Now handle the creative implementation of O_TMPFILE. */
1282 	if (flags & __O_TMPFILE) {
1283 		/*
1284 		 * In order to ensure programs get explicit errors when trying
1285 		 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1286 		 * is raised alongside __O_TMPFILE.
1287 		 */
1288 		if (!(flags & O_DIRECTORY))
1289 			return -EINVAL;
1290 		if (!(acc_mode & MAY_WRITE))
1291 			return -EINVAL;
1292 	}
1293 	if (flags & O_PATH) {
1294 		/* O_PATH only permits certain other flags to be set. */
1295 		if (flags & ~O_PATH_FLAGS)
1296 			return -EINVAL;
1297 		acc_mode = 0;
1298 	}
1299 
1300 	/*
1301 	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1302 	 * check for O_DSYNC if the need any syncing at all we enforce it's
1303 	 * always set instead of having to deal with possibly weird behaviour
1304 	 * for malicious applications setting only __O_SYNC.
1305 	 */
1306 	if (flags & __O_SYNC)
1307 		flags |= O_DSYNC;
1308 
1309 	op->open_flag = flags;
1310 
1311 	/* O_TRUNC implies we need access checks for write permissions */
1312 	if (flags & O_TRUNC)
1313 		acc_mode |= MAY_WRITE;
1314 
1315 	/* Allow the LSM permission hook to distinguish append
1316 	   access from general write access. */
1317 	if (flags & O_APPEND)
1318 		acc_mode |= MAY_APPEND;
1319 
1320 	op->acc_mode = acc_mode;
1321 
1322 	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1323 
1324 	if (flags & O_CREAT) {
1325 		op->intent |= LOOKUP_CREATE;
1326 		if (flags & O_EXCL) {
1327 			op->intent |= LOOKUP_EXCL;
1328 			flags |= O_NOFOLLOW;
1329 		}
1330 	}
1331 
1332 	if (flags & O_DIRECTORY)
1333 		lookup_flags |= LOOKUP_DIRECTORY;
1334 	if (!(flags & O_NOFOLLOW))
1335 		lookup_flags |= LOOKUP_FOLLOW;
1336 
1337 	if (how->resolve & RESOLVE_NO_XDEV)
1338 		lookup_flags |= LOOKUP_NO_XDEV;
1339 	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1340 		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1341 	if (how->resolve & RESOLVE_NO_SYMLINKS)
1342 		lookup_flags |= LOOKUP_NO_SYMLINKS;
1343 	if (how->resolve & RESOLVE_BENEATH)
1344 		lookup_flags |= LOOKUP_BENEATH;
1345 	if (how->resolve & RESOLVE_IN_ROOT)
1346 		lookup_flags |= LOOKUP_IN_ROOT;
1347 	if (how->resolve & RESOLVE_CACHED) {
1348 		/* Don't bother even trying for create/truncate/tmpfile open */
1349 		if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1350 			return -EAGAIN;
1351 		lookup_flags |= LOOKUP_CACHED;
1352 	}
1353 
1354 	op->lookup_flags = lookup_flags;
1355 	return 0;
1356 }
1357 
1358 /**
1359  * file_open_name - open file and return file pointer
1360  *
1361  * @name:	struct filename containing path to open
1362  * @flags:	open flags as per the open(2) second argument
1363  * @mode:	mode for the new file if O_CREAT is set, else ignored
1364  *
1365  * This is the helper to open a file from kernelspace if you really
1366  * have to.  But in generally you should not do this, so please move
1367  * along, nothing to see here..
1368  */
1369 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1370 {
1371 	struct open_flags op;
1372 	struct open_how how = build_open_how(flags, mode);
1373 	int err = build_open_flags(&how, &op);
1374 	if (err)
1375 		return ERR_PTR(err);
1376 	return do_filp_open(AT_FDCWD, name, &op);
1377 }
1378 
1379 /**
1380  * filp_open - open file and return file pointer
1381  *
1382  * @filename:	path to open
1383  * @flags:	open flags as per the open(2) second argument
1384  * @mode:	mode for the new file if O_CREAT is set, else ignored
1385  *
1386  * This is the helper to open a file from kernelspace if you really
1387  * have to.  But in generally you should not do this, so please move
1388  * along, nothing to see here..
1389  */
1390 struct file *filp_open(const char *filename, int flags, umode_t mode)
1391 {
1392 	struct filename *name = getname_kernel(filename);
1393 	struct file *file = ERR_CAST(name);
1394 
1395 	if (!IS_ERR(name)) {
1396 		file = file_open_name(name, flags, mode);
1397 		putname(name);
1398 	}
1399 	return file;
1400 }
1401 EXPORT_SYMBOL(filp_open);
1402 
1403 struct file *file_open_root(const struct path *root,
1404 			    const char *filename, int flags, umode_t mode)
1405 {
1406 	struct open_flags op;
1407 	struct open_how how = build_open_how(flags, mode);
1408 	int err = build_open_flags(&how, &op);
1409 	if (err)
1410 		return ERR_PTR(err);
1411 	return do_file_open_root(root, filename, &op);
1412 }
1413 EXPORT_SYMBOL(file_open_root);
1414 
1415 static int do_sys_openat2(int dfd, const char __user *filename,
1416 			  struct open_how *how)
1417 {
1418 	struct open_flags op;
1419 	struct filename *tmp __free(putname) = NULL;
1420 	int err;
1421 
1422 	err = build_open_flags(how, &op);
1423 	if (unlikely(err))
1424 		return err;
1425 
1426 	tmp = getname(filename);
1427 	if (IS_ERR(tmp))
1428 		return PTR_ERR(tmp);
1429 
1430 	return FD_ADD(how->flags, do_filp_open(dfd, tmp, &op));
1431 }
1432 
1433 int do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1434 {
1435 	struct open_how how = build_open_how(flags, mode);
1436 	return do_sys_openat2(dfd, filename, &how);
1437 }
1438 
1439 
1440 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1441 {
1442 	if (force_o_largefile())
1443 		flags |= O_LARGEFILE;
1444 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1445 }
1446 
1447 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1448 		umode_t, mode)
1449 {
1450 	if (force_o_largefile())
1451 		flags |= O_LARGEFILE;
1452 	return do_sys_open(dfd, filename, flags, mode);
1453 }
1454 
1455 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1456 		struct open_how __user *, how, size_t, usize)
1457 {
1458 	int err;
1459 	struct open_how tmp;
1460 
1461 	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1462 	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1463 
1464 	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1465 		return -EINVAL;
1466 	if (unlikely(usize > PAGE_SIZE))
1467 		return -E2BIG;
1468 
1469 	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1470 	if (err)
1471 		return err;
1472 
1473 	audit_openat2_how(&tmp);
1474 
1475 	/* O_LARGEFILE is only allowed for non-O_PATH. */
1476 	if (!(tmp.flags & O_PATH) && force_o_largefile())
1477 		tmp.flags |= O_LARGEFILE;
1478 
1479 	return do_sys_openat2(dfd, filename, &tmp);
1480 }
1481 
1482 #ifdef CONFIG_COMPAT
1483 /*
1484  * Exactly like sys_open(), except that it doesn't set the
1485  * O_LARGEFILE flag.
1486  */
1487 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1488 {
1489 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1490 }
1491 
1492 /*
1493  * Exactly like sys_openat(), except that it doesn't set the
1494  * O_LARGEFILE flag.
1495  */
1496 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1497 {
1498 	return do_sys_open(dfd, filename, flags, mode);
1499 }
1500 #endif
1501 
1502 #ifndef __alpha__
1503 
1504 /*
1505  * For backward compatibility?  Maybe this should be moved
1506  * into arch/i386 instead?
1507  */
1508 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1509 {
1510 	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1511 
1512 	if (force_o_largefile())
1513 		flags |= O_LARGEFILE;
1514 	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1515 }
1516 #endif
1517 
1518 /*
1519  * "id" is the POSIX thread ID. We use the
1520  * files pointer for this..
1521  */
1522 static int filp_flush(struct file *filp, fl_owner_t id)
1523 {
1524 	int retval = 0;
1525 
1526 	if (CHECK_DATA_CORRUPTION(file_count(filp) == 0, filp,
1527 			"VFS: Close: file count is 0 (f_op=%ps)",
1528 			filp->f_op)) {
1529 		return 0;
1530 	}
1531 
1532 	if (filp->f_op->flush)
1533 		retval = filp->f_op->flush(filp, id);
1534 
1535 	if (likely(!(filp->f_mode & FMODE_PATH))) {
1536 		dnotify_flush(filp, id);
1537 		locks_remove_posix(filp, id);
1538 	}
1539 	return retval;
1540 }
1541 
1542 int filp_close(struct file *filp, fl_owner_t id)
1543 {
1544 	int retval;
1545 
1546 	retval = filp_flush(filp, id);
1547 	fput_close(filp);
1548 
1549 	return retval;
1550 }
1551 EXPORT_SYMBOL(filp_close);
1552 
1553 /*
1554  * Careful here! We test whether the file pointer is NULL before
1555  * releasing the fd. This ensures that one clone task can't release
1556  * an fd while another clone is opening it.
1557  */
1558 SYSCALL_DEFINE1(close, unsigned int, fd)
1559 {
1560 	int retval;
1561 	struct file *file;
1562 
1563 	file = file_close_fd(fd);
1564 	if (!file)
1565 		return -EBADF;
1566 
1567 	retval = filp_flush(file, current->files);
1568 
1569 	/*
1570 	 * We're returning to user space. Don't bother
1571 	 * with any delayed fput() cases.
1572 	 */
1573 	fput_close_sync(file);
1574 
1575 	if (likely(retval == 0))
1576 		return 0;
1577 
1578 	/* can't restart close syscall because file table entry was cleared */
1579 	if (retval == -ERESTARTSYS ||
1580 	    retval == -ERESTARTNOINTR ||
1581 	    retval == -ERESTARTNOHAND ||
1582 	    retval == -ERESTART_RESTARTBLOCK)
1583 		retval = -EINTR;
1584 
1585 	return retval;
1586 }
1587 
1588 /*
1589  * This routine simulates a hangup on the tty, to arrange that users
1590  * are given clean terminals at login time.
1591  */
1592 SYSCALL_DEFINE0(vhangup)
1593 {
1594 	if (capable(CAP_SYS_TTY_CONFIG)) {
1595 		tty_vhangup_self();
1596 		return 0;
1597 	}
1598 	return -EPERM;
1599 }
1600 
1601 /*
1602  * Called when an inode is about to be open.
1603  * We use this to disallow opening large files on 32bit systems if
1604  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1605  * on this flag in sys_open.
1606  */
1607 int generic_file_open(struct inode * inode, struct file * filp)
1608 {
1609 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1610 		return -EOVERFLOW;
1611 	return 0;
1612 }
1613 
1614 EXPORT_SYMBOL(generic_file_open);
1615 
1616 /*
1617  * This is used by subsystems that don't want seekable
1618  * file descriptors. The function is not supposed to ever fail, the only
1619  * reason it returns an 'int' and not 'void' is so that it can be plugged
1620  * directly into file_operations structure.
1621  */
1622 int nonseekable_open(struct inode *inode, struct file *filp)
1623 {
1624 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1625 	return 0;
1626 }
1627 
1628 EXPORT_SYMBOL(nonseekable_open);
1629 
1630 /*
1631  * stream_open is used by subsystems that want stream-like file descriptors.
1632  * Such file descriptors are not seekable and don't have notion of position
1633  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1634  * Contrary to file descriptors of other regular files, .read() and .write()
1635  * can run simultaneously.
1636  *
1637  * stream_open never fails and is marked to return int so that it could be
1638  * directly used as file_operations.open .
1639  */
1640 int stream_open(struct inode *inode, struct file *filp)
1641 {
1642 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1643 	filp->f_mode |= FMODE_STREAM;
1644 	return 0;
1645 }
1646 
1647 EXPORT_SYMBOL(stream_open);
1648