xref: /linux/fs/open.c (revision 5cd2340cb6a383d04fd88e48fabc2a21a909d6a1)
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 	inode_lock(dentry->d_inode);
64 	/* Note any delegations or leases have already been broken: */
65 	ret = notify_change(idmap, dentry, &newattrs, NULL);
66 	inode_unlock(dentry->d_inode);
67 	return ret;
68 }
69 
70 long vfs_truncate(const struct path *path, loff_t length)
71 {
72 	struct mnt_idmap *idmap;
73 	struct inode *inode;
74 	long error;
75 
76 	inode = path->dentry->d_inode;
77 
78 	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79 	if (S_ISDIR(inode->i_mode))
80 		return -EISDIR;
81 	if (!S_ISREG(inode->i_mode))
82 		return -EINVAL;
83 
84 	error = mnt_want_write(path->mnt);
85 	if (error)
86 		goto out;
87 
88 	idmap = mnt_idmap(path->mnt);
89 	error = inode_permission(idmap, inode, MAY_WRITE);
90 	if (error)
91 		goto mnt_drop_write_and_out;
92 
93 	error = -EPERM;
94 	if (IS_APPEND(inode))
95 		goto mnt_drop_write_and_out;
96 
97 	error = get_write_access(inode);
98 	if (error)
99 		goto mnt_drop_write_and_out;
100 
101 	/*
102 	 * Make sure that there are no leases.  get_write_access() protects
103 	 * against the truncate racing with a lease-granting setlease().
104 	 */
105 	error = break_lease(inode, O_WRONLY);
106 	if (error)
107 		goto put_write_and_out;
108 
109 	error = security_path_truncate(path);
110 	if (!error)
111 		error = do_truncate(idmap, path->dentry, length, 0, NULL);
112 
113 put_write_and_out:
114 	put_write_access(inode);
115 mnt_drop_write_and_out:
116 	mnt_drop_write(path->mnt);
117 out:
118 	return error;
119 }
120 EXPORT_SYMBOL_GPL(vfs_truncate);
121 
122 long do_sys_truncate(const char __user *pathname, loff_t length)
123 {
124 	unsigned int lookup_flags = LOOKUP_FOLLOW;
125 	struct path path;
126 	int error;
127 
128 	if (length < 0)	/* sorry, but loff_t says... */
129 		return -EINVAL;
130 
131 retry:
132 	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
133 	if (!error) {
134 		error = vfs_truncate(&path, length);
135 		path_put(&path);
136 	}
137 	if (retry_estale(error, lookup_flags)) {
138 		lookup_flags |= LOOKUP_REVAL;
139 		goto retry;
140 	}
141 	return error;
142 }
143 
144 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
145 {
146 	return do_sys_truncate(path, length);
147 }
148 
149 #ifdef CONFIG_COMPAT
150 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
151 {
152 	return do_sys_truncate(path, length);
153 }
154 #endif
155 
156 long do_ftruncate(struct file *file, loff_t length, int small)
157 {
158 	struct inode *inode;
159 	struct dentry *dentry;
160 	int error;
161 
162 	/* explicitly opened as large or we are on 64-bit box */
163 	if (file->f_flags & O_LARGEFILE)
164 		small = 0;
165 
166 	dentry = file->f_path.dentry;
167 	inode = dentry->d_inode;
168 	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
169 		return -EINVAL;
170 
171 	/* Cannot ftruncate over 2^31 bytes without large file support */
172 	if (small && length > MAX_NON_LFS)
173 		return -EINVAL;
174 
175 	/* Check IS_APPEND on real upper inode */
176 	if (IS_APPEND(file_inode(file)))
177 		return -EPERM;
178 	sb_start_write(inode->i_sb);
179 	error = security_file_truncate(file);
180 	if (!error)
181 		error = do_truncate(file_mnt_idmap(file), dentry, length,
182 				    ATTR_MTIME | ATTR_CTIME, file);
183 	sb_end_write(inode->i_sb);
184 
185 	return error;
186 }
187 
188 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
189 {
190 	struct fd f;
191 	int error;
192 
193 	if (length < 0)
194 		return -EINVAL;
195 	f = fdget(fd);
196 	if (!f.file)
197 		return -EBADF;
198 
199 	error = do_ftruncate(f.file, length, small);
200 
201 	fdput(f);
202 	return error;
203 }
204 
205 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
206 {
207 	return do_sys_ftruncate(fd, length, 1);
208 }
209 
210 #ifdef CONFIG_COMPAT
211 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
212 {
213 	return do_sys_ftruncate(fd, length, 1);
214 }
215 #endif
216 
217 /* LFS versions of truncate are only needed on 32 bit machines */
218 #if BITS_PER_LONG == 32
219 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
220 {
221 	return do_sys_truncate(path, length);
222 }
223 
224 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
225 {
226 	return do_sys_ftruncate(fd, length, 0);
227 }
228 #endif /* BITS_PER_LONG == 32 */
229 
230 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
231 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
232 		       compat_arg_u64_dual(length))
233 {
234 	return ksys_truncate(pathname, compat_arg_u64_glue(length));
235 }
236 #endif
237 
238 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
239 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
240 		       compat_arg_u64_dual(length))
241 {
242 	return ksys_ftruncate(fd, compat_arg_u64_glue(length));
243 }
244 #endif
245 
246 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
247 {
248 	struct inode *inode = file_inode(file);
249 	long ret;
250 	loff_t sum;
251 
252 	if (offset < 0 || len <= 0)
253 		return -EINVAL;
254 
255 	/* Return error if mode is not supported */
256 	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
257 		return -EOPNOTSUPP;
258 
259 	/* Punch hole and zero range are mutually exclusive */
260 	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
261 	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
262 		return -EOPNOTSUPP;
263 
264 	/* Punch hole must have keep size set */
265 	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
266 	    !(mode & FALLOC_FL_KEEP_SIZE))
267 		return -EOPNOTSUPP;
268 
269 	/* Collapse range should only be used exclusively. */
270 	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
271 	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
272 		return -EINVAL;
273 
274 	/* Insert range should only be used exclusively. */
275 	if ((mode & FALLOC_FL_INSERT_RANGE) &&
276 	    (mode & ~FALLOC_FL_INSERT_RANGE))
277 		return -EINVAL;
278 
279 	/* Unshare range should only be used with allocate mode. */
280 	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
281 	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
282 		return -EINVAL;
283 
284 	if (!(file->f_mode & FMODE_WRITE))
285 		return -EBADF;
286 
287 	/*
288 	 * We can only allow pure fallocate on append only files
289 	 */
290 	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
291 		return -EPERM;
292 
293 	if (IS_IMMUTABLE(inode))
294 		return -EPERM;
295 
296 	/*
297 	 * We cannot allow any fallocate operation on an active swapfile
298 	 */
299 	if (IS_SWAPFILE(inode))
300 		return -ETXTBSY;
301 
302 	/*
303 	 * Revalidate the write permissions, in case security policy has
304 	 * changed since the files were opened.
305 	 */
306 	ret = security_file_permission(file, MAY_WRITE);
307 	if (ret)
308 		return ret;
309 
310 	ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
311 	if (ret)
312 		return ret;
313 
314 	if (S_ISFIFO(inode->i_mode))
315 		return -ESPIPE;
316 
317 	if (S_ISDIR(inode->i_mode))
318 		return -EISDIR;
319 
320 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
321 		return -ENODEV;
322 
323 	/* Check for wraparound */
324 	if (check_add_overflow(offset, len, &sum))
325 		return -EFBIG;
326 
327 	if (sum > inode->i_sb->s_maxbytes)
328 		return -EFBIG;
329 
330 	if (!file->f_op->fallocate)
331 		return -EOPNOTSUPP;
332 
333 	file_start_write(file);
334 	ret = file->f_op->fallocate(file, mode, offset, len);
335 
336 	/*
337 	 * Create inotify and fanotify events.
338 	 *
339 	 * To keep the logic simple always create events if fallocate succeeds.
340 	 * This implies that events are even created if the file size remains
341 	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
342 	 */
343 	if (ret == 0)
344 		fsnotify_modify(file);
345 
346 	file_end_write(file);
347 	return ret;
348 }
349 EXPORT_SYMBOL_GPL(vfs_fallocate);
350 
351 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
352 {
353 	struct fd f = fdget(fd);
354 	int error = -EBADF;
355 
356 	if (f.file) {
357 		error = vfs_fallocate(f.file, mode, offset, len);
358 		fdput(f);
359 	}
360 	return error;
361 }
362 
363 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
364 {
365 	return ksys_fallocate(fd, mode, offset, len);
366 }
367 
368 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
369 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
370 		       compat_arg_u64_dual(len))
371 {
372 	return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
373 			      compat_arg_u64_glue(len));
374 }
375 #endif
376 
377 /*
378  * access() needs to use the real uid/gid, not the effective uid/gid.
379  * We do this by temporarily clearing all FS-related capabilities and
380  * switching the fsuid/fsgid around to the real ones.
381  *
382  * Creating new credentials is expensive, so we try to skip doing it,
383  * which we can if the result would match what we already got.
384  */
385 static bool access_need_override_creds(int flags)
386 {
387 	const struct cred *cred;
388 
389 	if (flags & AT_EACCESS)
390 		return false;
391 
392 	cred = current_cred();
393 	if (!uid_eq(cred->fsuid, cred->uid) ||
394 	    !gid_eq(cred->fsgid, cred->gid))
395 		return true;
396 
397 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
398 		kuid_t root_uid = make_kuid(cred->user_ns, 0);
399 		if (!uid_eq(cred->uid, root_uid)) {
400 			if (!cap_isclear(cred->cap_effective))
401 				return true;
402 		} else {
403 			if (!cap_isidentical(cred->cap_effective,
404 			    cred->cap_permitted))
405 				return true;
406 		}
407 	}
408 
409 	return false;
410 }
411 
412 static const struct cred *access_override_creds(void)
413 {
414 	const struct cred *old_cred;
415 	struct cred *override_cred;
416 
417 	override_cred = prepare_creds();
418 	if (!override_cred)
419 		return NULL;
420 
421 	/*
422 	 * XXX access_need_override_creds performs checks in hopes of skipping
423 	 * this work. Make sure it stays in sync if making any changes in this
424 	 * routine.
425 	 */
426 
427 	override_cred->fsuid = override_cred->uid;
428 	override_cred->fsgid = override_cred->gid;
429 
430 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
431 		/* Clear the capabilities if we switch to a non-root user */
432 		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
433 		if (!uid_eq(override_cred->uid, root_uid))
434 			cap_clear(override_cred->cap_effective);
435 		else
436 			override_cred->cap_effective =
437 				override_cred->cap_permitted;
438 	}
439 
440 	/*
441 	 * The new set of credentials can *only* be used in
442 	 * task-synchronous circumstances, and does not need
443 	 * RCU freeing, unless somebody then takes a separate
444 	 * reference to it.
445 	 *
446 	 * NOTE! This is _only_ true because this credential
447 	 * is used purely for override_creds() that installs
448 	 * it as the subjective cred. Other threads will be
449 	 * accessing ->real_cred, not the subjective cred.
450 	 *
451 	 * If somebody _does_ make a copy of this (using the
452 	 * 'get_current_cred()' function), that will clear the
453 	 * non_rcu field, because now that other user may be
454 	 * expecting RCU freeing. But normal thread-synchronous
455 	 * cred accesses will keep things non-racy to avoid RCU
456 	 * freeing.
457 	 */
458 	override_cred->non_rcu = 1;
459 
460 	old_cred = override_creds(override_cred);
461 
462 	/* override_cred() gets its own ref */
463 	put_cred(override_cred);
464 
465 	return old_cred;
466 }
467 
468 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
469 {
470 	struct path path;
471 	struct inode *inode;
472 	int res;
473 	unsigned int lookup_flags = LOOKUP_FOLLOW;
474 	const struct cred *old_cred = NULL;
475 
476 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
477 		return -EINVAL;
478 
479 	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
480 		return -EINVAL;
481 
482 	if (flags & AT_SYMLINK_NOFOLLOW)
483 		lookup_flags &= ~LOOKUP_FOLLOW;
484 	if (flags & AT_EMPTY_PATH)
485 		lookup_flags |= LOOKUP_EMPTY;
486 
487 	if (access_need_override_creds(flags)) {
488 		old_cred = access_override_creds();
489 		if (!old_cred)
490 			return -ENOMEM;
491 	}
492 
493 retry:
494 	res = user_path_at(dfd, filename, lookup_flags, &path);
495 	if (res)
496 		goto out;
497 
498 	inode = d_backing_inode(path.dentry);
499 
500 	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
501 		/*
502 		 * MAY_EXEC on regular files is denied if the fs is mounted
503 		 * with the "noexec" flag.
504 		 */
505 		res = -EACCES;
506 		if (path_noexec(&path))
507 			goto out_path_release;
508 	}
509 
510 	res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
511 	/* SuS v2 requires we report a read only fs too */
512 	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
513 		goto out_path_release;
514 	/*
515 	 * This is a rare case where using __mnt_is_readonly()
516 	 * is OK without a mnt_want/drop_write() pair.  Since
517 	 * no actual write to the fs is performed here, we do
518 	 * not need to telegraph to that to anyone.
519 	 *
520 	 * By doing this, we accept that this access is
521 	 * inherently racy and know that the fs may change
522 	 * state before we even see this result.
523 	 */
524 	if (__mnt_is_readonly(path.mnt))
525 		res = -EROFS;
526 
527 out_path_release:
528 	path_put(&path);
529 	if (retry_estale(res, lookup_flags)) {
530 		lookup_flags |= LOOKUP_REVAL;
531 		goto retry;
532 	}
533 out:
534 	if (old_cred)
535 		revert_creds(old_cred);
536 
537 	return res;
538 }
539 
540 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
541 {
542 	return do_faccessat(dfd, filename, mode, 0);
543 }
544 
545 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
546 		int, flags)
547 {
548 	return do_faccessat(dfd, filename, mode, flags);
549 }
550 
551 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
552 {
553 	return do_faccessat(AT_FDCWD, filename, mode, 0);
554 }
555 
556 SYSCALL_DEFINE1(chdir, const char __user *, filename)
557 {
558 	struct path path;
559 	int error;
560 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
561 retry:
562 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
563 	if (error)
564 		goto out;
565 
566 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
567 	if (error)
568 		goto dput_and_out;
569 
570 	set_fs_pwd(current->fs, &path);
571 
572 dput_and_out:
573 	path_put(&path);
574 	if (retry_estale(error, lookup_flags)) {
575 		lookup_flags |= LOOKUP_REVAL;
576 		goto retry;
577 	}
578 out:
579 	return error;
580 }
581 
582 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
583 {
584 	struct fd f = fdget_raw(fd);
585 	int error;
586 
587 	error = -EBADF;
588 	if (!f.file)
589 		goto out;
590 
591 	error = -ENOTDIR;
592 	if (!d_can_lookup(f.file->f_path.dentry))
593 		goto out_putf;
594 
595 	error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
596 	if (!error)
597 		set_fs_pwd(current->fs, &f.file->f_path);
598 out_putf:
599 	fdput(f);
600 out:
601 	return error;
602 }
603 
604 SYSCALL_DEFINE1(chroot, const char __user *, filename)
605 {
606 	struct path path;
607 	int error;
608 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
609 retry:
610 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
611 	if (error)
612 		goto out;
613 
614 	error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
615 	if (error)
616 		goto dput_and_out;
617 
618 	error = -EPERM;
619 	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
620 		goto dput_and_out;
621 	error = security_path_chroot(&path);
622 	if (error)
623 		goto dput_and_out;
624 
625 	set_fs_root(current->fs, &path);
626 	error = 0;
627 dput_and_out:
628 	path_put(&path);
629 	if (retry_estale(error, lookup_flags)) {
630 		lookup_flags |= LOOKUP_REVAL;
631 		goto retry;
632 	}
633 out:
634 	return error;
635 }
636 
637 int chmod_common(const struct path *path, umode_t mode)
638 {
639 	struct inode *inode = path->dentry->d_inode;
640 	struct inode *delegated_inode = NULL;
641 	struct iattr newattrs;
642 	int error;
643 
644 	error = mnt_want_write(path->mnt);
645 	if (error)
646 		return error;
647 retry_deleg:
648 	inode_lock(inode);
649 	error = security_path_chmod(path, mode);
650 	if (error)
651 		goto out_unlock;
652 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
653 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
654 	error = notify_change(mnt_idmap(path->mnt), path->dentry,
655 			      &newattrs, &delegated_inode);
656 out_unlock:
657 	inode_unlock(inode);
658 	if (delegated_inode) {
659 		error = break_deleg_wait(&delegated_inode);
660 		if (!error)
661 			goto retry_deleg;
662 	}
663 	mnt_drop_write(path->mnt);
664 	return error;
665 }
666 
667 int vfs_fchmod(struct file *file, umode_t mode)
668 {
669 	audit_file(file);
670 	return chmod_common(&file->f_path, mode);
671 }
672 
673 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
674 {
675 	struct fd f = fdget(fd);
676 	int err = -EBADF;
677 
678 	if (f.file) {
679 		err = vfs_fchmod(f.file, mode);
680 		fdput(f);
681 	}
682 	return err;
683 }
684 
685 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
686 		       unsigned int flags)
687 {
688 	struct path path;
689 	int error;
690 	unsigned int lookup_flags;
691 
692 	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
693 		return -EINVAL;
694 
695 	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
696 	if (flags & AT_EMPTY_PATH)
697 		lookup_flags |= LOOKUP_EMPTY;
698 
699 retry:
700 	error = user_path_at(dfd, filename, lookup_flags, &path);
701 	if (!error) {
702 		error = chmod_common(&path, mode);
703 		path_put(&path);
704 		if (retry_estale(error, lookup_flags)) {
705 			lookup_flags |= LOOKUP_REVAL;
706 			goto retry;
707 		}
708 	}
709 	return error;
710 }
711 
712 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
713 		umode_t, mode, unsigned int, flags)
714 {
715 	return do_fchmodat(dfd, filename, mode, flags);
716 }
717 
718 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
719 		umode_t, mode)
720 {
721 	return do_fchmodat(dfd, filename, mode, 0);
722 }
723 
724 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
725 {
726 	return do_fchmodat(AT_FDCWD, filename, mode, 0);
727 }
728 
729 /*
730  * Check whether @kuid is valid and if so generate and set vfsuid_t in
731  * ia_vfsuid.
732  *
733  * Return: true if @kuid is valid, false if not.
734  */
735 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
736 {
737 	if (!uid_valid(kuid))
738 		return false;
739 	attr->ia_valid |= ATTR_UID;
740 	attr->ia_vfsuid = VFSUIDT_INIT(kuid);
741 	return true;
742 }
743 
744 /*
745  * Check whether @kgid is valid and if so generate and set vfsgid_t in
746  * ia_vfsgid.
747  *
748  * Return: true if @kgid is valid, false if not.
749  */
750 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
751 {
752 	if (!gid_valid(kgid))
753 		return false;
754 	attr->ia_valid |= ATTR_GID;
755 	attr->ia_vfsgid = VFSGIDT_INIT(kgid);
756 	return true;
757 }
758 
759 int chown_common(const struct path *path, uid_t user, gid_t group)
760 {
761 	struct mnt_idmap *idmap;
762 	struct user_namespace *fs_userns;
763 	struct inode *inode = path->dentry->d_inode;
764 	struct inode *delegated_inode = NULL;
765 	int error;
766 	struct iattr newattrs;
767 	kuid_t uid;
768 	kgid_t gid;
769 
770 	uid = make_kuid(current_user_ns(), user);
771 	gid = make_kgid(current_user_ns(), group);
772 
773 	idmap = mnt_idmap(path->mnt);
774 	fs_userns = i_user_ns(inode);
775 
776 retry_deleg:
777 	newattrs.ia_vfsuid = INVALID_VFSUID;
778 	newattrs.ia_vfsgid = INVALID_VFSGID;
779 	newattrs.ia_valid =  ATTR_CTIME;
780 	if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
781 		return -EINVAL;
782 	if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
783 		return -EINVAL;
784 	inode_lock(inode);
785 	if (!S_ISDIR(inode->i_mode))
786 		newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
787 				     setattr_should_drop_sgid(idmap, inode);
788 	/* Continue to send actual fs values, not the mount values. */
789 	error = security_path_chown(
790 		path,
791 		from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
792 		from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
793 	if (!error)
794 		error = notify_change(idmap, path->dentry, &newattrs,
795 				      &delegated_inode);
796 	inode_unlock(inode);
797 	if (delegated_inode) {
798 		error = break_deleg_wait(&delegated_inode);
799 		if (!error)
800 			goto retry_deleg;
801 	}
802 	return error;
803 }
804 
805 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
806 		int flag)
807 {
808 	struct path path;
809 	int error = -EINVAL;
810 	int lookup_flags;
811 
812 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
813 		goto out;
814 
815 	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
816 	if (flag & AT_EMPTY_PATH)
817 		lookup_flags |= LOOKUP_EMPTY;
818 retry:
819 	error = user_path_at(dfd, filename, lookup_flags, &path);
820 	if (error)
821 		goto out;
822 	error = mnt_want_write(path.mnt);
823 	if (error)
824 		goto out_release;
825 	error = chown_common(&path, user, group);
826 	mnt_drop_write(path.mnt);
827 out_release:
828 	path_put(&path);
829 	if (retry_estale(error, lookup_flags)) {
830 		lookup_flags |= LOOKUP_REVAL;
831 		goto retry;
832 	}
833 out:
834 	return error;
835 }
836 
837 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
838 		gid_t, group, int, flag)
839 {
840 	return do_fchownat(dfd, filename, user, group, flag);
841 }
842 
843 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
844 {
845 	return do_fchownat(AT_FDCWD, filename, user, group, 0);
846 }
847 
848 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
849 {
850 	return do_fchownat(AT_FDCWD, filename, user, group,
851 			   AT_SYMLINK_NOFOLLOW);
852 }
853 
854 int vfs_fchown(struct file *file, uid_t user, gid_t group)
855 {
856 	int error;
857 
858 	error = mnt_want_write_file(file);
859 	if (error)
860 		return error;
861 	audit_file(file);
862 	error = chown_common(&file->f_path, user, group);
863 	mnt_drop_write_file(file);
864 	return error;
865 }
866 
867 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
868 {
869 	struct fd f = fdget(fd);
870 	int error = -EBADF;
871 
872 	if (f.file) {
873 		error = vfs_fchown(f.file, user, group);
874 		fdput(f);
875 	}
876 	return error;
877 }
878 
879 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
880 {
881 	return ksys_fchown(fd, user, group);
882 }
883 
884 static inline int file_get_write_access(struct file *f)
885 {
886 	int error;
887 
888 	error = get_write_access(f->f_inode);
889 	if (unlikely(error))
890 		return error;
891 	error = mnt_get_write_access(f->f_path.mnt);
892 	if (unlikely(error))
893 		goto cleanup_inode;
894 	if (unlikely(f->f_mode & FMODE_BACKING)) {
895 		error = mnt_get_write_access(backing_file_user_path(f)->mnt);
896 		if (unlikely(error))
897 			goto cleanup_mnt;
898 	}
899 	return 0;
900 
901 cleanup_mnt:
902 	mnt_put_write_access(f->f_path.mnt);
903 cleanup_inode:
904 	put_write_access(f->f_inode);
905 	return error;
906 }
907 
908 static int do_dentry_open(struct file *f,
909 			  int (*open)(struct inode *, struct file *))
910 {
911 	static const struct file_operations empty_fops = {};
912 	struct inode *inode = f->f_path.dentry->d_inode;
913 	int error;
914 
915 	path_get(&f->f_path);
916 	f->f_inode = inode;
917 	f->f_mapping = inode->i_mapping;
918 	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
919 	f->f_sb_err = file_sample_sb_err(f);
920 
921 	if (unlikely(f->f_flags & O_PATH)) {
922 		f->f_mode = FMODE_PATH | FMODE_OPENED;
923 		f->f_op = &empty_fops;
924 		return 0;
925 	}
926 
927 	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
928 		i_readcount_inc(inode);
929 	} else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
930 		error = file_get_write_access(f);
931 		if (unlikely(error))
932 			goto cleanup_file;
933 		f->f_mode |= FMODE_WRITER;
934 	}
935 
936 	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
937 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
938 		f->f_mode |= FMODE_ATOMIC_POS;
939 
940 	f->f_op = fops_get(inode->i_fop);
941 	if (WARN_ON(!f->f_op)) {
942 		error = -ENODEV;
943 		goto cleanup_all;
944 	}
945 
946 	error = security_file_open(f);
947 	if (error)
948 		goto cleanup_all;
949 
950 	error = break_lease(file_inode(f), f->f_flags);
951 	if (error)
952 		goto cleanup_all;
953 
954 	/* normally all 3 are set; ->open() can clear them if needed */
955 	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
956 	if (!open)
957 		open = f->f_op->open;
958 	if (open) {
959 		error = open(inode, f);
960 		if (error)
961 			goto cleanup_all;
962 	}
963 	f->f_mode |= FMODE_OPENED;
964 	if ((f->f_mode & FMODE_READ) &&
965 	     likely(f->f_op->read || f->f_op->read_iter))
966 		f->f_mode |= FMODE_CAN_READ;
967 	if ((f->f_mode & FMODE_WRITE) &&
968 	     likely(f->f_op->write || f->f_op->write_iter))
969 		f->f_mode |= FMODE_CAN_WRITE;
970 	if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
971 		f->f_mode &= ~FMODE_LSEEK;
972 	if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
973 		f->f_mode |= FMODE_CAN_ODIRECT;
974 
975 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
976 	f->f_iocb_flags = iocb_flags(f);
977 
978 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
979 
980 	if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
981 		return -EINVAL;
982 
983 	/*
984 	 * XXX: Huge page cache doesn't support writing yet. Drop all page
985 	 * cache for this file before processing writes.
986 	 */
987 	if (f->f_mode & FMODE_WRITE) {
988 		/*
989 		 * Depends on full fence from get_write_access() to synchronize
990 		 * against collapse_file() regarding i_writecount and nr_thps
991 		 * updates. Ensures subsequent insertion of THPs into the page
992 		 * cache will fail.
993 		 */
994 		if (filemap_nr_thps(inode->i_mapping)) {
995 			struct address_space *mapping = inode->i_mapping;
996 
997 			filemap_invalidate_lock(inode->i_mapping);
998 			/*
999 			 * unmap_mapping_range just need to be called once
1000 			 * here, because the private pages is not need to be
1001 			 * unmapped mapping (e.g. data segment of dynamic
1002 			 * shared libraries here).
1003 			 */
1004 			unmap_mapping_range(mapping, 0, 0, 0);
1005 			truncate_inode_pages(mapping, 0);
1006 			filemap_invalidate_unlock(inode->i_mapping);
1007 		}
1008 	}
1009 
1010 	return 0;
1011 
1012 cleanup_all:
1013 	if (WARN_ON_ONCE(error > 0))
1014 		error = -EINVAL;
1015 	fops_put(f->f_op);
1016 	put_file_access(f);
1017 cleanup_file:
1018 	path_put(&f->f_path);
1019 	f->f_path.mnt = NULL;
1020 	f->f_path.dentry = NULL;
1021 	f->f_inode = NULL;
1022 	return error;
1023 }
1024 
1025 /**
1026  * finish_open - finish opening a file
1027  * @file: file pointer
1028  * @dentry: pointer to dentry
1029  * @open: open callback
1030  *
1031  * This can be used to finish opening a file passed to i_op->atomic_open().
1032  *
1033  * If the open callback is set to NULL, then the standard f_op->open()
1034  * filesystem callback is substituted.
1035  *
1036  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
1037  * the return value of d_splice_alias(), then the caller needs to perform dput()
1038  * on it after finish_open().
1039  *
1040  * Returns zero on success or -errno if the open failed.
1041  */
1042 int finish_open(struct file *file, struct dentry *dentry,
1043 		int (*open)(struct inode *, struct file *))
1044 {
1045 	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1046 
1047 	file->f_path.dentry = dentry;
1048 	return do_dentry_open(file, open);
1049 }
1050 EXPORT_SYMBOL(finish_open);
1051 
1052 /**
1053  * finish_no_open - finish ->atomic_open() without opening the file
1054  *
1055  * @file: file pointer
1056  * @dentry: dentry or NULL (as returned from ->lookup())
1057  *
1058  * This can be used to set the result of a successful lookup in ->atomic_open().
1059  *
1060  * NB: unlike finish_open() this function does consume the dentry reference and
1061  * the caller need not dput() it.
1062  *
1063  * Returns "0" which must be the return value of ->atomic_open() after having
1064  * called this function.
1065  */
1066 int finish_no_open(struct file *file, struct dentry *dentry)
1067 {
1068 	file->f_path.dentry = dentry;
1069 	return 0;
1070 }
1071 EXPORT_SYMBOL(finish_no_open);
1072 
1073 char *file_path(struct file *filp, char *buf, int buflen)
1074 {
1075 	return d_path(&filp->f_path, buf, buflen);
1076 }
1077 EXPORT_SYMBOL(file_path);
1078 
1079 /**
1080  * vfs_open - open the file at the given path
1081  * @path: path to open
1082  * @file: newly allocated file with f_flag initialized
1083  */
1084 int vfs_open(const struct path *path, struct file *file)
1085 {
1086 	int ret;
1087 
1088 	file->f_path = *path;
1089 	ret = do_dentry_open(file, NULL);
1090 	if (!ret) {
1091 		/*
1092 		 * Once we return a file with FMODE_OPENED, __fput() will call
1093 		 * fsnotify_close(), so we need fsnotify_open() here for
1094 		 * symmetry.
1095 		 */
1096 		fsnotify_open(file);
1097 	}
1098 	return ret;
1099 }
1100 
1101 struct file *dentry_open(const struct path *path, int flags,
1102 			 const struct cred *cred)
1103 {
1104 	int error;
1105 	struct file *f;
1106 
1107 	/* We must always pass in a valid mount pointer. */
1108 	BUG_ON(!path->mnt);
1109 
1110 	f = alloc_empty_file(flags, cred);
1111 	if (!IS_ERR(f)) {
1112 		error = vfs_open(path, f);
1113 		if (error) {
1114 			fput(f);
1115 			f = ERR_PTR(error);
1116 		}
1117 	}
1118 	return f;
1119 }
1120 EXPORT_SYMBOL(dentry_open);
1121 
1122 /**
1123  * dentry_create - Create and open a file
1124  * @path: path to create
1125  * @flags: O_ flags
1126  * @mode: mode bits for new file
1127  * @cred: credentials to use
1128  *
1129  * Caller must hold the parent directory's lock, and have prepared
1130  * a negative dentry, placed in @path->dentry, for the new file.
1131  *
1132  * Caller sets @path->mnt to the vfsmount of the filesystem where
1133  * the new file is to be created. The parent directory and the
1134  * negative dentry must reside on the same filesystem instance.
1135  *
1136  * On success, returns a "struct file *". Otherwise a ERR_PTR
1137  * is returned.
1138  */
1139 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1140 			   const struct cred *cred)
1141 {
1142 	struct file *f;
1143 	int error;
1144 
1145 	f = alloc_empty_file(flags, cred);
1146 	if (IS_ERR(f))
1147 		return f;
1148 
1149 	error = vfs_create(mnt_idmap(path->mnt),
1150 			   d_inode(path->dentry->d_parent),
1151 			   path->dentry, mode, true);
1152 	if (!error)
1153 		error = vfs_open(path, f);
1154 
1155 	if (unlikely(error)) {
1156 		fput(f);
1157 		return ERR_PTR(error);
1158 	}
1159 	return f;
1160 }
1161 EXPORT_SYMBOL(dentry_create);
1162 
1163 /**
1164  * kernel_file_open - open a file for kernel internal use
1165  * @path:	path of the file to open
1166  * @flags:	open flags
1167  * @cred:	credentials for open
1168  *
1169  * Open a file for use by in-kernel consumers. The file is not accounted
1170  * against nr_files and must not be installed into the file descriptor
1171  * table.
1172  *
1173  * Return: Opened file on success, an error pointer on failure.
1174  */
1175 struct file *kernel_file_open(const struct path *path, int flags,
1176 				const struct cred *cred)
1177 {
1178 	struct file *f;
1179 	int error;
1180 
1181 	f = alloc_empty_file_noaccount(flags, cred);
1182 	if (IS_ERR(f))
1183 		return f;
1184 
1185 	f->f_path = *path;
1186 	error = do_dentry_open(f, NULL);
1187 	if (error) {
1188 		fput(f);
1189 		return ERR_PTR(error);
1190 	}
1191 
1192 	fsnotify_open(f);
1193 	return f;
1194 }
1195 EXPORT_SYMBOL_GPL(kernel_file_open);
1196 
1197 #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
1198 #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1199 
1200 inline struct open_how build_open_how(int flags, umode_t mode)
1201 {
1202 	struct open_how how = {
1203 		.flags = flags & VALID_OPEN_FLAGS,
1204 		.mode = mode & S_IALLUGO,
1205 	};
1206 
1207 	/* O_PATH beats everything else. */
1208 	if (how.flags & O_PATH)
1209 		how.flags &= O_PATH_FLAGS;
1210 	/* Modes should only be set for create-like flags. */
1211 	if (!WILL_CREATE(how.flags))
1212 		how.mode = 0;
1213 	return how;
1214 }
1215 
1216 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1217 {
1218 	u64 flags = how->flags;
1219 	u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1220 	int lookup_flags = 0;
1221 	int acc_mode = ACC_MODE(flags);
1222 
1223 	BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1224 			 "struct open_flags doesn't yet handle flags > 32 bits");
1225 
1226 	/*
1227 	 * Strip flags that either shouldn't be set by userspace like
1228 	 * FMODE_NONOTIFY or that aren't relevant in determining struct
1229 	 * open_flags like O_CLOEXEC.
1230 	 */
1231 	flags &= ~strip;
1232 
1233 	/*
1234 	 * Older syscalls implicitly clear all of the invalid flags or argument
1235 	 * values before calling build_open_flags(), but openat2(2) checks all
1236 	 * of its arguments.
1237 	 */
1238 	if (flags & ~VALID_OPEN_FLAGS)
1239 		return -EINVAL;
1240 	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1241 		return -EINVAL;
1242 
1243 	/* Scoping flags are mutually exclusive. */
1244 	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1245 		return -EINVAL;
1246 
1247 	/* Deal with the mode. */
1248 	if (WILL_CREATE(flags)) {
1249 		if (how->mode & ~S_IALLUGO)
1250 			return -EINVAL;
1251 		op->mode = how->mode | S_IFREG;
1252 	} else {
1253 		if (how->mode != 0)
1254 			return -EINVAL;
1255 		op->mode = 0;
1256 	}
1257 
1258 	/*
1259 	 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1260 	 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1261 	 * O_TMPFILE below which requires O_DIRECTORY being raised.
1262 	 */
1263 	if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1264 		return -EINVAL;
1265 
1266 	/* Now handle the creative implementation of O_TMPFILE. */
1267 	if (flags & __O_TMPFILE) {
1268 		/*
1269 		 * In order to ensure programs get explicit errors when trying
1270 		 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1271 		 * is raised alongside __O_TMPFILE.
1272 		 */
1273 		if (!(flags & O_DIRECTORY))
1274 			return -EINVAL;
1275 		if (!(acc_mode & MAY_WRITE))
1276 			return -EINVAL;
1277 	}
1278 	if (flags & O_PATH) {
1279 		/* O_PATH only permits certain other flags to be set. */
1280 		if (flags & ~O_PATH_FLAGS)
1281 			return -EINVAL;
1282 		acc_mode = 0;
1283 	}
1284 
1285 	/*
1286 	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1287 	 * check for O_DSYNC if the need any syncing at all we enforce it's
1288 	 * always set instead of having to deal with possibly weird behaviour
1289 	 * for malicious applications setting only __O_SYNC.
1290 	 */
1291 	if (flags & __O_SYNC)
1292 		flags |= O_DSYNC;
1293 
1294 	op->open_flag = flags;
1295 
1296 	/* O_TRUNC implies we need access checks for write permissions */
1297 	if (flags & O_TRUNC)
1298 		acc_mode |= MAY_WRITE;
1299 
1300 	/* Allow the LSM permission hook to distinguish append
1301 	   access from general write access. */
1302 	if (flags & O_APPEND)
1303 		acc_mode |= MAY_APPEND;
1304 
1305 	op->acc_mode = acc_mode;
1306 
1307 	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1308 
1309 	if (flags & O_CREAT) {
1310 		op->intent |= LOOKUP_CREATE;
1311 		if (flags & O_EXCL) {
1312 			op->intent |= LOOKUP_EXCL;
1313 			flags |= O_NOFOLLOW;
1314 		}
1315 	}
1316 
1317 	if (flags & O_DIRECTORY)
1318 		lookup_flags |= LOOKUP_DIRECTORY;
1319 	if (!(flags & O_NOFOLLOW))
1320 		lookup_flags |= LOOKUP_FOLLOW;
1321 
1322 	if (how->resolve & RESOLVE_NO_XDEV)
1323 		lookup_flags |= LOOKUP_NO_XDEV;
1324 	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1325 		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1326 	if (how->resolve & RESOLVE_NO_SYMLINKS)
1327 		lookup_flags |= LOOKUP_NO_SYMLINKS;
1328 	if (how->resolve & RESOLVE_BENEATH)
1329 		lookup_flags |= LOOKUP_BENEATH;
1330 	if (how->resolve & RESOLVE_IN_ROOT)
1331 		lookup_flags |= LOOKUP_IN_ROOT;
1332 	if (how->resolve & RESOLVE_CACHED) {
1333 		/* Don't bother even trying for create/truncate/tmpfile open */
1334 		if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1335 			return -EAGAIN;
1336 		lookup_flags |= LOOKUP_CACHED;
1337 	}
1338 
1339 	op->lookup_flags = lookup_flags;
1340 	return 0;
1341 }
1342 
1343 /**
1344  * file_open_name - open file and return file pointer
1345  *
1346  * @name:	struct filename containing path to open
1347  * @flags:	open flags as per the open(2) second argument
1348  * @mode:	mode for the new file if O_CREAT is set, else ignored
1349  *
1350  * This is the helper to open a file from kernelspace if you really
1351  * have to.  But in generally you should not do this, so please move
1352  * along, nothing to see here..
1353  */
1354 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1355 {
1356 	struct open_flags op;
1357 	struct open_how how = build_open_how(flags, mode);
1358 	int err = build_open_flags(&how, &op);
1359 	if (err)
1360 		return ERR_PTR(err);
1361 	return do_filp_open(AT_FDCWD, name, &op);
1362 }
1363 
1364 /**
1365  * filp_open - open file and return file pointer
1366  *
1367  * @filename:	path to open
1368  * @flags:	open flags as per the open(2) second argument
1369  * @mode:	mode for the new file if O_CREAT is set, else ignored
1370  *
1371  * This is the helper to open a file from kernelspace if you really
1372  * have to.  But in generally you should not do this, so please move
1373  * along, nothing to see here..
1374  */
1375 struct file *filp_open(const char *filename, int flags, umode_t mode)
1376 {
1377 	struct filename *name = getname_kernel(filename);
1378 	struct file *file = ERR_CAST(name);
1379 
1380 	if (!IS_ERR(name)) {
1381 		file = file_open_name(name, flags, mode);
1382 		putname(name);
1383 	}
1384 	return file;
1385 }
1386 EXPORT_SYMBOL(filp_open);
1387 
1388 struct file *file_open_root(const struct path *root,
1389 			    const char *filename, int flags, umode_t mode)
1390 {
1391 	struct open_flags op;
1392 	struct open_how how = build_open_how(flags, mode);
1393 	int err = build_open_flags(&how, &op);
1394 	if (err)
1395 		return ERR_PTR(err);
1396 	return do_file_open_root(root, filename, &op);
1397 }
1398 EXPORT_SYMBOL(file_open_root);
1399 
1400 static long do_sys_openat2(int dfd, const char __user *filename,
1401 			   struct open_how *how)
1402 {
1403 	struct open_flags op;
1404 	int fd = build_open_flags(how, &op);
1405 	struct filename *tmp;
1406 
1407 	if (fd)
1408 		return fd;
1409 
1410 	tmp = getname(filename);
1411 	if (IS_ERR(tmp))
1412 		return PTR_ERR(tmp);
1413 
1414 	fd = get_unused_fd_flags(how->flags);
1415 	if (fd >= 0) {
1416 		struct file *f = do_filp_open(dfd, tmp, &op);
1417 		if (IS_ERR(f)) {
1418 			put_unused_fd(fd);
1419 			fd = PTR_ERR(f);
1420 		} else {
1421 			fd_install(fd, f);
1422 		}
1423 	}
1424 	putname(tmp);
1425 	return fd;
1426 }
1427 
1428 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1429 {
1430 	struct open_how how = build_open_how(flags, mode);
1431 	return do_sys_openat2(dfd, filename, &how);
1432 }
1433 
1434 
1435 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1436 {
1437 	if (force_o_largefile())
1438 		flags |= O_LARGEFILE;
1439 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1440 }
1441 
1442 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1443 		umode_t, mode)
1444 {
1445 	if (force_o_largefile())
1446 		flags |= O_LARGEFILE;
1447 	return do_sys_open(dfd, filename, flags, mode);
1448 }
1449 
1450 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1451 		struct open_how __user *, how, size_t, usize)
1452 {
1453 	int err;
1454 	struct open_how tmp;
1455 
1456 	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1457 	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1458 
1459 	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1460 		return -EINVAL;
1461 
1462 	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1463 	if (err)
1464 		return err;
1465 
1466 	audit_openat2_how(&tmp);
1467 
1468 	/* O_LARGEFILE is only allowed for non-O_PATH. */
1469 	if (!(tmp.flags & O_PATH) && force_o_largefile())
1470 		tmp.flags |= O_LARGEFILE;
1471 
1472 	return do_sys_openat2(dfd, filename, &tmp);
1473 }
1474 
1475 #ifdef CONFIG_COMPAT
1476 /*
1477  * Exactly like sys_open(), except that it doesn't set the
1478  * O_LARGEFILE flag.
1479  */
1480 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1481 {
1482 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1483 }
1484 
1485 /*
1486  * Exactly like sys_openat(), except that it doesn't set the
1487  * O_LARGEFILE flag.
1488  */
1489 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1490 {
1491 	return do_sys_open(dfd, filename, flags, mode);
1492 }
1493 #endif
1494 
1495 #ifndef __alpha__
1496 
1497 /*
1498  * For backward compatibility?  Maybe this should be moved
1499  * into arch/i386 instead?
1500  */
1501 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1502 {
1503 	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1504 
1505 	if (force_o_largefile())
1506 		flags |= O_LARGEFILE;
1507 	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1508 }
1509 #endif
1510 
1511 /*
1512  * "id" is the POSIX thread ID. We use the
1513  * files pointer for this..
1514  */
1515 static int filp_flush(struct file *filp, fl_owner_t id)
1516 {
1517 	int retval = 0;
1518 
1519 	if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1520 			"VFS: Close: file count is 0 (f_op=%ps)",
1521 			filp->f_op)) {
1522 		return 0;
1523 	}
1524 
1525 	if (filp->f_op->flush)
1526 		retval = filp->f_op->flush(filp, id);
1527 
1528 	if (likely(!(filp->f_mode & FMODE_PATH))) {
1529 		dnotify_flush(filp, id);
1530 		locks_remove_posix(filp, id);
1531 	}
1532 	return retval;
1533 }
1534 
1535 int filp_close(struct file *filp, fl_owner_t id)
1536 {
1537 	int retval;
1538 
1539 	retval = filp_flush(filp, id);
1540 	fput(filp);
1541 
1542 	return retval;
1543 }
1544 EXPORT_SYMBOL(filp_close);
1545 
1546 /*
1547  * Careful here! We test whether the file pointer is NULL before
1548  * releasing the fd. This ensures that one clone task can't release
1549  * an fd while another clone is opening it.
1550  */
1551 SYSCALL_DEFINE1(close, unsigned int, fd)
1552 {
1553 	int retval;
1554 	struct file *file;
1555 
1556 	file = file_close_fd(fd);
1557 	if (!file)
1558 		return -EBADF;
1559 
1560 	retval = filp_flush(file, current->files);
1561 
1562 	/*
1563 	 * We're returning to user space. Don't bother
1564 	 * with any delayed fput() cases.
1565 	 */
1566 	__fput_sync(file);
1567 
1568 	/* can't restart close syscall because file table entry was cleared */
1569 	if (unlikely(retval == -ERESTARTSYS ||
1570 		     retval == -ERESTARTNOINTR ||
1571 		     retval == -ERESTARTNOHAND ||
1572 		     retval == -ERESTART_RESTARTBLOCK))
1573 		retval = -EINTR;
1574 
1575 	return retval;
1576 }
1577 
1578 /**
1579  * sys_close_range() - Close all file descriptors in a given range.
1580  *
1581  * @fd:     starting file descriptor to close
1582  * @max_fd: last file descriptor to close
1583  * @flags:  reserved for future extensions
1584  *
1585  * This closes a range of file descriptors. All file descriptors
1586  * from @fd up to and including @max_fd are closed.
1587  * Currently, errors to close a given file descriptor are ignored.
1588  */
1589 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1590 		unsigned int, flags)
1591 {
1592 	return __close_range(fd, max_fd, flags);
1593 }
1594 
1595 /*
1596  * This routine simulates a hangup on the tty, to arrange that users
1597  * are given clean terminals at login time.
1598  */
1599 SYSCALL_DEFINE0(vhangup)
1600 {
1601 	if (capable(CAP_SYS_TTY_CONFIG)) {
1602 		tty_vhangup_self();
1603 		return 0;
1604 	}
1605 	return -EPERM;
1606 }
1607 
1608 /*
1609  * Called when an inode is about to be open.
1610  * We use this to disallow opening large files on 32bit systems if
1611  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1612  * on this flag in sys_open.
1613  */
1614 int generic_file_open(struct inode * inode, struct file * filp)
1615 {
1616 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1617 		return -EOVERFLOW;
1618 	return 0;
1619 }
1620 
1621 EXPORT_SYMBOL(generic_file_open);
1622 
1623 /*
1624  * This is used by subsystems that don't want seekable
1625  * file descriptors. The function is not supposed to ever fail, the only
1626  * reason it returns an 'int' and not 'void' is so that it can be plugged
1627  * directly into file_operations structure.
1628  */
1629 int nonseekable_open(struct inode *inode, struct file *filp)
1630 {
1631 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1632 	return 0;
1633 }
1634 
1635 EXPORT_SYMBOL(nonseekable_open);
1636 
1637 /*
1638  * stream_open is used by subsystems that want stream-like file descriptors.
1639  * Such file descriptors are not seekable and don't have notion of position
1640  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1641  * Contrary to file descriptors of other regular files, .read() and .write()
1642  * can run simultaneously.
1643  *
1644  * stream_open never fails and is marked to return int so that it could be
1645  * directly used as file_operations.open .
1646  */
1647 int stream_open(struct inode *inode, struct file *filp)
1648 {
1649 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1650 	filp->f_mode |= FMODE_STREAM;
1651 	return 0;
1652 }
1653 
1654 EXPORT_SYMBOL(stream_open);
1655