xref: /openbmc/linux/fs/open.c (revision 6197e5b7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/open.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/string.h>
9 #include <linux/mm.h>
10 #include <linux/file.h>
11 #include <linux/fdtable.h>
12 #include <linux/fsnotify.h>
13 #include <linux/module.h>
14 #include <linux/tty.h>
15 #include <linux/namei.h>
16 #include <linux/backing-dev.h>
17 #include <linux/capability.h>
18 #include <linux/securebits.h>
19 #include <linux/security.h>
20 #include <linux/mount.h>
21 #include <linux/fcntl.h>
22 #include <linux/slab.h>
23 #include <linux/uaccess.h>
24 #include <linux/fs.h>
25 #include <linux/personality.h>
26 #include <linux/pagemap.h>
27 #include <linux/syscalls.h>
28 #include <linux/rcupdate.h>
29 #include <linux/audit.h>
30 #include <linux/falloc.h>
31 #include <linux/fs_struct.h>
32 #include <linux/ima.h>
33 #include <linux/dnotify.h>
34 #include <linux/compat.h>
35 
36 #include "internal.h"
37 
38 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
39 	struct file *filp)
40 {
41 	int ret;
42 	struct iattr newattrs;
43 
44 	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
45 	if (length < 0)
46 		return -EINVAL;
47 
48 	newattrs.ia_size = length;
49 	newattrs.ia_valid = ATTR_SIZE | time_attrs;
50 	if (filp) {
51 		newattrs.ia_file = filp;
52 		newattrs.ia_valid |= ATTR_FILE;
53 	}
54 
55 	/* Remove suid, sgid, and file capabilities on truncate too */
56 	ret = dentry_needs_remove_privs(dentry);
57 	if (ret < 0)
58 		return ret;
59 	if (ret)
60 		newattrs.ia_valid |= ret | ATTR_FORCE;
61 
62 	inode_lock(dentry->d_inode);
63 	/* Note any delegations or leases have already been broken: */
64 	ret = notify_change(dentry, &newattrs, NULL);
65 	inode_unlock(dentry->d_inode);
66 	return ret;
67 }
68 
69 long vfs_truncate(const struct path *path, loff_t length)
70 {
71 	struct inode *inode;
72 	long error;
73 
74 	inode = path->dentry->d_inode;
75 
76 	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
77 	if (S_ISDIR(inode->i_mode))
78 		return -EISDIR;
79 	if (!S_ISREG(inode->i_mode))
80 		return -EINVAL;
81 
82 	error = mnt_want_write(path->mnt);
83 	if (error)
84 		goto out;
85 
86 	error = inode_permission(inode, MAY_WRITE);
87 	if (error)
88 		goto mnt_drop_write_and_out;
89 
90 	error = -EPERM;
91 	if (IS_APPEND(inode))
92 		goto mnt_drop_write_and_out;
93 
94 	error = get_write_access(inode);
95 	if (error)
96 		goto mnt_drop_write_and_out;
97 
98 	/*
99 	 * Make sure that there are no leases.  get_write_access() protects
100 	 * against the truncate racing with a lease-granting setlease().
101 	 */
102 	error = break_lease(inode, O_WRONLY);
103 	if (error)
104 		goto put_write_and_out;
105 
106 	error = locks_verify_truncate(inode, NULL, length);
107 	if (!error)
108 		error = security_path_truncate(path);
109 	if (!error)
110 		error = do_truncate(path->dentry, length, 0, NULL);
111 
112 put_write_and_out:
113 	put_write_access(inode);
114 mnt_drop_write_and_out:
115 	mnt_drop_write(path->mnt);
116 out:
117 	return error;
118 }
119 EXPORT_SYMBOL_GPL(vfs_truncate);
120 
121 long do_sys_truncate(const char __user *pathname, loff_t length)
122 {
123 	unsigned int lookup_flags = LOOKUP_FOLLOW;
124 	struct path path;
125 	int error;
126 
127 	if (length < 0)	/* sorry, but loff_t says... */
128 		return -EINVAL;
129 
130 retry:
131 	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
132 	if (!error) {
133 		error = vfs_truncate(&path, length);
134 		path_put(&path);
135 	}
136 	if (retry_estale(error, lookup_flags)) {
137 		lookup_flags |= LOOKUP_REVAL;
138 		goto retry;
139 	}
140 	return error;
141 }
142 
143 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
144 {
145 	return do_sys_truncate(path, length);
146 }
147 
148 #ifdef CONFIG_COMPAT
149 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
150 {
151 	return do_sys_truncate(path, length);
152 }
153 #endif
154 
155 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
156 {
157 	struct inode *inode;
158 	struct dentry *dentry;
159 	struct fd f;
160 	int error;
161 
162 	error = -EINVAL;
163 	if (length < 0)
164 		goto out;
165 	error = -EBADF;
166 	f = fdget(fd);
167 	if (!f.file)
168 		goto out;
169 
170 	/* explicitly opened as large or we are on 64-bit box */
171 	if (f.file->f_flags & O_LARGEFILE)
172 		small = 0;
173 
174 	dentry = f.file->f_path.dentry;
175 	inode = dentry->d_inode;
176 	error = -EINVAL;
177 	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
178 		goto out_putf;
179 
180 	error = -EINVAL;
181 	/* Cannot ftruncate over 2^31 bytes without large file support */
182 	if (small && length > MAX_NON_LFS)
183 		goto out_putf;
184 
185 	error = -EPERM;
186 	/* Check IS_APPEND on real upper inode */
187 	if (IS_APPEND(file_inode(f.file)))
188 		goto out_putf;
189 
190 	sb_start_write(inode->i_sb);
191 	error = locks_verify_truncate(inode, f.file, length);
192 	if (!error)
193 		error = security_path_truncate(&f.file->f_path);
194 	if (!error)
195 		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
196 	sb_end_write(inode->i_sb);
197 out_putf:
198 	fdput(f);
199 out:
200 	return error;
201 }
202 
203 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
204 {
205 	return do_sys_ftruncate(fd, length, 1);
206 }
207 
208 #ifdef CONFIG_COMPAT
209 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
210 {
211 	return do_sys_ftruncate(fd, length, 1);
212 }
213 #endif
214 
215 /* LFS versions of truncate are only needed on 32 bit machines */
216 #if BITS_PER_LONG == 32
217 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
218 {
219 	return do_sys_truncate(path, length);
220 }
221 
222 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
223 {
224 	return do_sys_ftruncate(fd, length, 0);
225 }
226 #endif /* BITS_PER_LONG == 32 */
227 
228 
229 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
230 {
231 	struct inode *inode = file_inode(file);
232 	long ret;
233 
234 	if (offset < 0 || len <= 0)
235 		return -EINVAL;
236 
237 	/* Return error if mode is not supported */
238 	if (mode & ~FALLOC_FL_SUPPORTED_MASK)
239 		return -EOPNOTSUPP;
240 
241 	/* Punch hole and zero range are mutually exclusive */
242 	if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
243 	    (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
244 		return -EOPNOTSUPP;
245 
246 	/* Punch hole must have keep size set */
247 	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
248 	    !(mode & FALLOC_FL_KEEP_SIZE))
249 		return -EOPNOTSUPP;
250 
251 	/* Collapse range should only be used exclusively. */
252 	if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
253 	    (mode & ~FALLOC_FL_COLLAPSE_RANGE))
254 		return -EINVAL;
255 
256 	/* Insert range should only be used exclusively. */
257 	if ((mode & FALLOC_FL_INSERT_RANGE) &&
258 	    (mode & ~FALLOC_FL_INSERT_RANGE))
259 		return -EINVAL;
260 
261 	/* Unshare range should only be used with allocate mode. */
262 	if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
263 	    (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
264 		return -EINVAL;
265 
266 	if (!(file->f_mode & FMODE_WRITE))
267 		return -EBADF;
268 
269 	/*
270 	 * We can only allow pure fallocate on append only files
271 	 */
272 	if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
273 		return -EPERM;
274 
275 	if (IS_IMMUTABLE(inode))
276 		return -EPERM;
277 
278 	/*
279 	 * We cannot allow any fallocate operation on an active swapfile
280 	 */
281 	if (IS_SWAPFILE(inode))
282 		return -ETXTBSY;
283 
284 	/*
285 	 * Revalidate the write permissions, in case security policy has
286 	 * changed since the files were opened.
287 	 */
288 	ret = security_file_permission(file, MAY_WRITE);
289 	if (ret)
290 		return ret;
291 
292 	if (S_ISFIFO(inode->i_mode))
293 		return -ESPIPE;
294 
295 	if (S_ISDIR(inode->i_mode))
296 		return -EISDIR;
297 
298 	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
299 		return -ENODEV;
300 
301 	/* Check for wrap through zero too */
302 	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
303 		return -EFBIG;
304 
305 	if (!file->f_op->fallocate)
306 		return -EOPNOTSUPP;
307 
308 	file_start_write(file);
309 	ret = file->f_op->fallocate(file, mode, offset, len);
310 
311 	/*
312 	 * Create inotify and fanotify events.
313 	 *
314 	 * To keep the logic simple always create events if fallocate succeeds.
315 	 * This implies that events are even created if the file size remains
316 	 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
317 	 */
318 	if (ret == 0)
319 		fsnotify_modify(file);
320 
321 	file_end_write(file);
322 	return ret;
323 }
324 EXPORT_SYMBOL_GPL(vfs_fallocate);
325 
326 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
327 {
328 	struct fd f = fdget(fd);
329 	int error = -EBADF;
330 
331 	if (f.file) {
332 		error = vfs_fallocate(f.file, mode, offset, len);
333 		fdput(f);
334 	}
335 	return error;
336 }
337 
338 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
339 {
340 	return ksys_fallocate(fd, mode, offset, len);
341 }
342 
343 /*
344  * access() needs to use the real uid/gid, not the effective uid/gid.
345  * We do this by temporarily clearing all FS-related capabilities and
346  * switching the fsuid/fsgid around to the real ones.
347  */
348 static const struct cred *access_override_creds(void)
349 {
350 	const struct cred *old_cred;
351 	struct cred *override_cred;
352 
353 	override_cred = prepare_creds();
354 	if (!override_cred)
355 		return NULL;
356 
357 	override_cred->fsuid = override_cred->uid;
358 	override_cred->fsgid = override_cred->gid;
359 
360 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
361 		/* Clear the capabilities if we switch to a non-root user */
362 		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
363 		if (!uid_eq(override_cred->uid, root_uid))
364 			cap_clear(override_cred->cap_effective);
365 		else
366 			override_cred->cap_effective =
367 				override_cred->cap_permitted;
368 	}
369 
370 	/*
371 	 * The new set of credentials can *only* be used in
372 	 * task-synchronous circumstances, and does not need
373 	 * RCU freeing, unless somebody then takes a separate
374 	 * reference to it.
375 	 *
376 	 * NOTE! This is _only_ true because this credential
377 	 * is used purely for override_creds() that installs
378 	 * it as the subjective cred. Other threads will be
379 	 * accessing ->real_cred, not the subjective cred.
380 	 *
381 	 * If somebody _does_ make a copy of this (using the
382 	 * 'get_current_cred()' function), that will clear the
383 	 * non_rcu field, because now that other user may be
384 	 * expecting RCU freeing. But normal thread-synchronous
385 	 * cred accesses will keep things non-RCY.
386 	 */
387 	override_cred->non_rcu = 1;
388 
389 	old_cred = override_creds(override_cred);
390 
391 	/* override_cred() gets its own ref */
392 	put_cred(override_cred);
393 
394 	return old_cred;
395 }
396 
397 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
398 {
399 	struct path path;
400 	struct inode *inode;
401 	int res;
402 	unsigned int lookup_flags = LOOKUP_FOLLOW;
403 	const struct cred *old_cred = NULL;
404 
405 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
406 		return -EINVAL;
407 
408 	if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
409 		return -EINVAL;
410 
411 	if (flags & AT_SYMLINK_NOFOLLOW)
412 		lookup_flags &= ~LOOKUP_FOLLOW;
413 	if (flags & AT_EMPTY_PATH)
414 		lookup_flags |= LOOKUP_EMPTY;
415 
416 	if (!(flags & AT_EACCESS)) {
417 		old_cred = access_override_creds();
418 		if (!old_cred)
419 			return -ENOMEM;
420 	}
421 
422 retry:
423 	res = user_path_at(dfd, filename, lookup_flags, &path);
424 	if (res)
425 		goto out;
426 
427 	inode = d_backing_inode(path.dentry);
428 
429 	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
430 		/*
431 		 * MAY_EXEC on regular files is denied if the fs is mounted
432 		 * with the "noexec" flag.
433 		 */
434 		res = -EACCES;
435 		if (path_noexec(&path))
436 			goto out_path_release;
437 	}
438 
439 	res = inode_permission(inode, mode | MAY_ACCESS);
440 	/* SuS v2 requires we report a read only fs too */
441 	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
442 		goto out_path_release;
443 	/*
444 	 * This is a rare case where using __mnt_is_readonly()
445 	 * is OK without a mnt_want/drop_write() pair.  Since
446 	 * no actual write to the fs is performed here, we do
447 	 * not need to telegraph to that to anyone.
448 	 *
449 	 * By doing this, we accept that this access is
450 	 * inherently racy and know that the fs may change
451 	 * state before we even see this result.
452 	 */
453 	if (__mnt_is_readonly(path.mnt))
454 		res = -EROFS;
455 
456 out_path_release:
457 	path_put(&path);
458 	if (retry_estale(res, lookup_flags)) {
459 		lookup_flags |= LOOKUP_REVAL;
460 		goto retry;
461 	}
462 out:
463 	if (old_cred)
464 		revert_creds(old_cred);
465 
466 	return res;
467 }
468 
469 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
470 {
471 	return do_faccessat(dfd, filename, mode, 0);
472 }
473 
474 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
475 		int, flags)
476 {
477 	return do_faccessat(dfd, filename, mode, flags);
478 }
479 
480 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
481 {
482 	return do_faccessat(AT_FDCWD, filename, mode, 0);
483 }
484 
485 SYSCALL_DEFINE1(chdir, const char __user *, filename)
486 {
487 	struct path path;
488 	int error;
489 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
490 retry:
491 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
492 	if (error)
493 		goto out;
494 
495 	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
496 	if (error)
497 		goto dput_and_out;
498 
499 	set_fs_pwd(current->fs, &path);
500 
501 dput_and_out:
502 	path_put(&path);
503 	if (retry_estale(error, lookup_flags)) {
504 		lookup_flags |= LOOKUP_REVAL;
505 		goto retry;
506 	}
507 out:
508 	return error;
509 }
510 
511 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
512 {
513 	struct fd f = fdget_raw(fd);
514 	int error;
515 
516 	error = -EBADF;
517 	if (!f.file)
518 		goto out;
519 
520 	error = -ENOTDIR;
521 	if (!d_can_lookup(f.file->f_path.dentry))
522 		goto out_putf;
523 
524 	error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
525 	if (!error)
526 		set_fs_pwd(current->fs, &f.file->f_path);
527 out_putf:
528 	fdput(f);
529 out:
530 	return error;
531 }
532 
533 SYSCALL_DEFINE1(chroot, const char __user *, filename)
534 {
535 	struct path path;
536 	int error;
537 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
538 retry:
539 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
540 	if (error)
541 		goto out;
542 
543 	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
544 	if (error)
545 		goto dput_and_out;
546 
547 	error = -EPERM;
548 	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
549 		goto dput_and_out;
550 	error = security_path_chroot(&path);
551 	if (error)
552 		goto dput_and_out;
553 
554 	set_fs_root(current->fs, &path);
555 	error = 0;
556 dput_and_out:
557 	path_put(&path);
558 	if (retry_estale(error, lookup_flags)) {
559 		lookup_flags |= LOOKUP_REVAL;
560 		goto retry;
561 	}
562 out:
563 	return error;
564 }
565 
566 int chmod_common(const struct path *path, umode_t mode)
567 {
568 	struct inode *inode = path->dentry->d_inode;
569 	struct inode *delegated_inode = NULL;
570 	struct iattr newattrs;
571 	int error;
572 
573 	error = mnt_want_write(path->mnt);
574 	if (error)
575 		return error;
576 retry_deleg:
577 	inode_lock(inode);
578 	error = security_path_chmod(path, mode);
579 	if (error)
580 		goto out_unlock;
581 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
582 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
583 	error = notify_change(path->dentry, &newattrs, &delegated_inode);
584 out_unlock:
585 	inode_unlock(inode);
586 	if (delegated_inode) {
587 		error = break_deleg_wait(&delegated_inode);
588 		if (!error)
589 			goto retry_deleg;
590 	}
591 	mnt_drop_write(path->mnt);
592 	return error;
593 }
594 
595 int vfs_fchmod(struct file *file, umode_t mode)
596 {
597 	audit_file(file);
598 	return chmod_common(&file->f_path, mode);
599 }
600 
601 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
602 {
603 	struct fd f = fdget(fd);
604 	int err = -EBADF;
605 
606 	if (f.file) {
607 		err = vfs_fchmod(f.file, mode);
608 		fdput(f);
609 	}
610 	return err;
611 }
612 
613 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
614 {
615 	struct path path;
616 	int error;
617 	unsigned int lookup_flags = LOOKUP_FOLLOW;
618 retry:
619 	error = user_path_at(dfd, filename, lookup_flags, &path);
620 	if (!error) {
621 		error = chmod_common(&path, mode);
622 		path_put(&path);
623 		if (retry_estale(error, lookup_flags)) {
624 			lookup_flags |= LOOKUP_REVAL;
625 			goto retry;
626 		}
627 	}
628 	return error;
629 }
630 
631 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
632 		umode_t, mode)
633 {
634 	return do_fchmodat(dfd, filename, mode);
635 }
636 
637 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
638 {
639 	return do_fchmodat(AT_FDCWD, filename, mode);
640 }
641 
642 int chown_common(const struct path *path, uid_t user, gid_t group)
643 {
644 	struct inode *inode = path->dentry->d_inode;
645 	struct inode *delegated_inode = NULL;
646 	int error;
647 	struct iattr newattrs;
648 	kuid_t uid;
649 	kgid_t gid;
650 
651 	uid = make_kuid(current_user_ns(), user);
652 	gid = make_kgid(current_user_ns(), group);
653 
654 retry_deleg:
655 	newattrs.ia_valid =  ATTR_CTIME;
656 	if (user != (uid_t) -1) {
657 		if (!uid_valid(uid))
658 			return -EINVAL;
659 		newattrs.ia_valid |= ATTR_UID;
660 		newattrs.ia_uid = uid;
661 	}
662 	if (group != (gid_t) -1) {
663 		if (!gid_valid(gid))
664 			return -EINVAL;
665 		newattrs.ia_valid |= ATTR_GID;
666 		newattrs.ia_gid = gid;
667 	}
668 	if (!S_ISDIR(inode->i_mode))
669 		newattrs.ia_valid |=
670 			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
671 	inode_lock(inode);
672 	error = security_path_chown(path, uid, gid);
673 	if (!error)
674 		error = notify_change(path->dentry, &newattrs, &delegated_inode);
675 	inode_unlock(inode);
676 	if (delegated_inode) {
677 		error = break_deleg_wait(&delegated_inode);
678 		if (!error)
679 			goto retry_deleg;
680 	}
681 	return error;
682 }
683 
684 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
685 		int flag)
686 {
687 	struct path path;
688 	int error = -EINVAL;
689 	int lookup_flags;
690 
691 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
692 		goto out;
693 
694 	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
695 	if (flag & AT_EMPTY_PATH)
696 		lookup_flags |= LOOKUP_EMPTY;
697 retry:
698 	error = user_path_at(dfd, filename, lookup_flags, &path);
699 	if (error)
700 		goto out;
701 	error = mnt_want_write(path.mnt);
702 	if (error)
703 		goto out_release;
704 	error = chown_common(&path, user, group);
705 	mnt_drop_write(path.mnt);
706 out_release:
707 	path_put(&path);
708 	if (retry_estale(error, lookup_flags)) {
709 		lookup_flags |= LOOKUP_REVAL;
710 		goto retry;
711 	}
712 out:
713 	return error;
714 }
715 
716 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
717 		gid_t, group, int, flag)
718 {
719 	return do_fchownat(dfd, filename, user, group, flag);
720 }
721 
722 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
723 {
724 	return do_fchownat(AT_FDCWD, filename, user, group, 0);
725 }
726 
727 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
728 {
729 	return do_fchownat(AT_FDCWD, filename, user, group,
730 			   AT_SYMLINK_NOFOLLOW);
731 }
732 
733 int vfs_fchown(struct file *file, uid_t user, gid_t group)
734 {
735 	int error;
736 
737 	error = mnt_want_write_file(file);
738 	if (error)
739 		return error;
740 	audit_file(file);
741 	error = chown_common(&file->f_path, user, group);
742 	mnt_drop_write_file(file);
743 	return error;
744 }
745 
746 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
747 {
748 	struct fd f = fdget(fd);
749 	int error = -EBADF;
750 
751 	if (f.file) {
752 		error = vfs_fchown(f.file, user, group);
753 		fdput(f);
754 	}
755 	return error;
756 }
757 
758 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
759 {
760 	return ksys_fchown(fd, user, group);
761 }
762 
763 static int do_dentry_open(struct file *f,
764 			  struct inode *inode,
765 			  int (*open)(struct inode *, struct file *))
766 {
767 	static const struct file_operations empty_fops = {};
768 	int error;
769 
770 	path_get(&f->f_path);
771 	f->f_inode = inode;
772 	f->f_mapping = inode->i_mapping;
773 	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
774 	f->f_sb_err = file_sample_sb_err(f);
775 
776 	if (unlikely(f->f_flags & O_PATH)) {
777 		f->f_mode = FMODE_PATH | FMODE_OPENED;
778 		f->f_op = &empty_fops;
779 		return 0;
780 	}
781 
782 	if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
783 		error = get_write_access(inode);
784 		if (unlikely(error))
785 			goto cleanup_file;
786 		error = __mnt_want_write(f->f_path.mnt);
787 		if (unlikely(error)) {
788 			put_write_access(inode);
789 			goto cleanup_file;
790 		}
791 		f->f_mode |= FMODE_WRITER;
792 	}
793 
794 	/* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
795 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
796 		f->f_mode |= FMODE_ATOMIC_POS;
797 
798 	f->f_op = fops_get(inode->i_fop);
799 	if (WARN_ON(!f->f_op)) {
800 		error = -ENODEV;
801 		goto cleanup_all;
802 	}
803 
804 	error = security_file_open(f);
805 	if (error)
806 		goto cleanup_all;
807 
808 	error = break_lease(locks_inode(f), f->f_flags);
809 	if (error)
810 		goto cleanup_all;
811 
812 	/* normally all 3 are set; ->open() can clear them if needed */
813 	f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
814 	if (!open)
815 		open = f->f_op->open;
816 	if (open) {
817 		error = open(inode, f);
818 		if (error)
819 			goto cleanup_all;
820 	}
821 	f->f_mode |= FMODE_OPENED;
822 	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
823 		i_readcount_inc(inode);
824 	if ((f->f_mode & FMODE_READ) &&
825 	     likely(f->f_op->read || f->f_op->read_iter))
826 		f->f_mode |= FMODE_CAN_READ;
827 	if ((f->f_mode & FMODE_WRITE) &&
828 	     likely(f->f_op->write || f->f_op->write_iter))
829 		f->f_mode |= FMODE_CAN_WRITE;
830 
831 	f->f_write_hint = WRITE_LIFE_NOT_SET;
832 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
833 
834 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
835 
836 	/* NB: we're sure to have correct a_ops only after f_op->open */
837 	if (f->f_flags & O_DIRECT) {
838 		if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
839 			return -EINVAL;
840 	}
841 
842 	/*
843 	 * XXX: Huge page cache doesn't support writing yet. Drop all page
844 	 * cache for this file before processing writes.
845 	 */
846 	if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
847 		truncate_pagecache(inode, 0);
848 
849 	return 0;
850 
851 cleanup_all:
852 	if (WARN_ON_ONCE(error > 0))
853 		error = -EINVAL;
854 	fops_put(f->f_op);
855 	if (f->f_mode & FMODE_WRITER) {
856 		put_write_access(inode);
857 		__mnt_drop_write(f->f_path.mnt);
858 	}
859 cleanup_file:
860 	path_put(&f->f_path);
861 	f->f_path.mnt = NULL;
862 	f->f_path.dentry = NULL;
863 	f->f_inode = NULL;
864 	return error;
865 }
866 
867 /**
868  * finish_open - finish opening a file
869  * @file: file pointer
870  * @dentry: pointer to dentry
871  * @open: open callback
872  * @opened: state of open
873  *
874  * This can be used to finish opening a file passed to i_op->atomic_open().
875  *
876  * If the open callback is set to NULL, then the standard f_op->open()
877  * filesystem callback is substituted.
878  *
879  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
880  * the return value of d_splice_alias(), then the caller needs to perform dput()
881  * on it after finish_open().
882  *
883  * Returns zero on success or -errno if the open failed.
884  */
885 int finish_open(struct file *file, struct dentry *dentry,
886 		int (*open)(struct inode *, struct file *))
887 {
888 	BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
889 
890 	file->f_path.dentry = dentry;
891 	return do_dentry_open(file, d_backing_inode(dentry), open);
892 }
893 EXPORT_SYMBOL(finish_open);
894 
895 /**
896  * finish_no_open - finish ->atomic_open() without opening the file
897  *
898  * @file: file pointer
899  * @dentry: dentry or NULL (as returned from ->lookup())
900  *
901  * This can be used to set the result of a successful lookup in ->atomic_open().
902  *
903  * NB: unlike finish_open() this function does consume the dentry reference and
904  * the caller need not dput() it.
905  *
906  * Returns "0" which must be the return value of ->atomic_open() after having
907  * called this function.
908  */
909 int finish_no_open(struct file *file, struct dentry *dentry)
910 {
911 	file->f_path.dentry = dentry;
912 	return 0;
913 }
914 EXPORT_SYMBOL(finish_no_open);
915 
916 char *file_path(struct file *filp, char *buf, int buflen)
917 {
918 	return d_path(&filp->f_path, buf, buflen);
919 }
920 EXPORT_SYMBOL(file_path);
921 
922 /**
923  * vfs_open - open the file at the given path
924  * @path: path to open
925  * @file: newly allocated file with f_flag initialized
926  * @cred: credentials to use
927  */
928 int vfs_open(const struct path *path, struct file *file)
929 {
930 	file->f_path = *path;
931 	return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
932 }
933 
934 struct file *dentry_open(const struct path *path, int flags,
935 			 const struct cred *cred)
936 {
937 	int error;
938 	struct file *f;
939 
940 	validate_creds(cred);
941 
942 	/* We must always pass in a valid mount pointer. */
943 	BUG_ON(!path->mnt);
944 
945 	f = alloc_empty_file(flags, cred);
946 	if (!IS_ERR(f)) {
947 		error = vfs_open(path, f);
948 		if (error) {
949 			fput(f);
950 			f = ERR_PTR(error);
951 		}
952 	}
953 	return f;
954 }
955 EXPORT_SYMBOL(dentry_open);
956 
957 struct file *open_with_fake_path(const struct path *path, int flags,
958 				struct inode *inode, const struct cred *cred)
959 {
960 	struct file *f = alloc_empty_file_noaccount(flags, cred);
961 	if (!IS_ERR(f)) {
962 		int error;
963 
964 		f->f_path = *path;
965 		error = do_dentry_open(f, inode, NULL);
966 		if (error) {
967 			fput(f);
968 			f = ERR_PTR(error);
969 		}
970 	}
971 	return f;
972 }
973 EXPORT_SYMBOL(open_with_fake_path);
974 
975 #define WILL_CREATE(flags)	(flags & (O_CREAT | __O_TMPFILE))
976 #define O_PATH_FLAGS		(O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
977 
978 inline struct open_how build_open_how(int flags, umode_t mode)
979 {
980 	struct open_how how = {
981 		.flags = flags & VALID_OPEN_FLAGS,
982 		.mode = mode & S_IALLUGO,
983 	};
984 
985 	/* O_PATH beats everything else. */
986 	if (how.flags & O_PATH)
987 		how.flags &= O_PATH_FLAGS;
988 	/* Modes should only be set for create-like flags. */
989 	if (!WILL_CREATE(how.flags))
990 		how.mode = 0;
991 	return how;
992 }
993 
994 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
995 {
996 	int flags = how->flags;
997 	int lookup_flags = 0;
998 	int acc_mode = ACC_MODE(flags);
999 
1000 	/* Must never be set by userspace */
1001 	flags &= ~(FMODE_NONOTIFY | O_CLOEXEC);
1002 
1003 	/*
1004 	 * Older syscalls implicitly clear all of the invalid flags or argument
1005 	 * values before calling build_open_flags(), but openat2(2) checks all
1006 	 * of its arguments.
1007 	 */
1008 	if (flags & ~VALID_OPEN_FLAGS)
1009 		return -EINVAL;
1010 	if (how->resolve & ~VALID_RESOLVE_FLAGS)
1011 		return -EINVAL;
1012 
1013 	/* Scoping flags are mutually exclusive. */
1014 	if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1015 		return -EINVAL;
1016 
1017 	/* Deal with the mode. */
1018 	if (WILL_CREATE(flags)) {
1019 		if (how->mode & ~S_IALLUGO)
1020 			return -EINVAL;
1021 		op->mode = how->mode | S_IFREG;
1022 	} else {
1023 		if (how->mode != 0)
1024 			return -EINVAL;
1025 		op->mode = 0;
1026 	}
1027 
1028 	/*
1029 	 * In order to ensure programs get explicit errors when trying to use
1030 	 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1031 	 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1032 	 * have to require userspace to explicitly set it.
1033 	 */
1034 	if (flags & __O_TMPFILE) {
1035 		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1036 			return -EINVAL;
1037 		if (!(acc_mode & MAY_WRITE))
1038 			return -EINVAL;
1039 	}
1040 	if (flags & O_PATH) {
1041 		/* O_PATH only permits certain other flags to be set. */
1042 		if (flags & ~O_PATH_FLAGS)
1043 			return -EINVAL;
1044 		acc_mode = 0;
1045 	}
1046 
1047 	/*
1048 	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1049 	 * check for O_DSYNC if the need any syncing at all we enforce it's
1050 	 * always set instead of having to deal with possibly weird behaviour
1051 	 * for malicious applications setting only __O_SYNC.
1052 	 */
1053 	if (flags & __O_SYNC)
1054 		flags |= O_DSYNC;
1055 
1056 	op->open_flag = flags;
1057 
1058 	/* O_TRUNC implies we need access checks for write permissions */
1059 	if (flags & O_TRUNC)
1060 		acc_mode |= MAY_WRITE;
1061 
1062 	/* Allow the LSM permission hook to distinguish append
1063 	   access from general write access. */
1064 	if (flags & O_APPEND)
1065 		acc_mode |= MAY_APPEND;
1066 
1067 	op->acc_mode = acc_mode;
1068 
1069 	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1070 
1071 	if (flags & O_CREAT) {
1072 		op->intent |= LOOKUP_CREATE;
1073 		if (flags & O_EXCL) {
1074 			op->intent |= LOOKUP_EXCL;
1075 			flags |= O_NOFOLLOW;
1076 		}
1077 	}
1078 
1079 	if (flags & O_DIRECTORY)
1080 		lookup_flags |= LOOKUP_DIRECTORY;
1081 	if (!(flags & O_NOFOLLOW))
1082 		lookup_flags |= LOOKUP_FOLLOW;
1083 
1084 	if (how->resolve & RESOLVE_NO_XDEV)
1085 		lookup_flags |= LOOKUP_NO_XDEV;
1086 	if (how->resolve & RESOLVE_NO_MAGICLINKS)
1087 		lookup_flags |= LOOKUP_NO_MAGICLINKS;
1088 	if (how->resolve & RESOLVE_NO_SYMLINKS)
1089 		lookup_flags |= LOOKUP_NO_SYMLINKS;
1090 	if (how->resolve & RESOLVE_BENEATH)
1091 		lookup_flags |= LOOKUP_BENEATH;
1092 	if (how->resolve & RESOLVE_IN_ROOT)
1093 		lookup_flags |= LOOKUP_IN_ROOT;
1094 	if (how->resolve & RESOLVE_CACHED) {
1095 		/* Don't bother even trying for create/truncate/tmpfile open */
1096 		if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1097 			return -EAGAIN;
1098 		lookup_flags |= LOOKUP_CACHED;
1099 	}
1100 
1101 	op->lookup_flags = lookup_flags;
1102 	return 0;
1103 }
1104 
1105 /**
1106  * file_open_name - open file and return file pointer
1107  *
1108  * @name:	struct filename containing path to open
1109  * @flags:	open flags as per the open(2) second argument
1110  * @mode:	mode for the new file if O_CREAT is set, else ignored
1111  *
1112  * This is the helper to open a file from kernelspace if you really
1113  * have to.  But in generally you should not do this, so please move
1114  * along, nothing to see here..
1115  */
1116 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1117 {
1118 	struct open_flags op;
1119 	struct open_how how = build_open_how(flags, mode);
1120 	int err = build_open_flags(&how, &op);
1121 	if (err)
1122 		return ERR_PTR(err);
1123 	return do_filp_open(AT_FDCWD, name, &op);
1124 }
1125 
1126 /**
1127  * filp_open - open file and return file pointer
1128  *
1129  * @filename:	path to open
1130  * @flags:	open flags as per the open(2) second argument
1131  * @mode:	mode for the new file if O_CREAT is set, else ignored
1132  *
1133  * This is the helper to open a file from kernelspace if you really
1134  * have to.  But in generally you should not do this, so please move
1135  * along, nothing to see here..
1136  */
1137 struct file *filp_open(const char *filename, int flags, umode_t mode)
1138 {
1139 	struct filename *name = getname_kernel(filename);
1140 	struct file *file = ERR_CAST(name);
1141 
1142 	if (!IS_ERR(name)) {
1143 		file = file_open_name(name, flags, mode);
1144 		putname(name);
1145 	}
1146 	return file;
1147 }
1148 EXPORT_SYMBOL(filp_open);
1149 
1150 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1151 			    const char *filename, int flags, umode_t mode)
1152 {
1153 	struct open_flags op;
1154 	struct open_how how = build_open_how(flags, mode);
1155 	int err = build_open_flags(&how, &op);
1156 	if (err)
1157 		return ERR_PTR(err);
1158 	return do_file_open_root(dentry, mnt, filename, &op);
1159 }
1160 EXPORT_SYMBOL(file_open_root);
1161 
1162 static long do_sys_openat2(int dfd, const char __user *filename,
1163 			   struct open_how *how)
1164 {
1165 	struct open_flags op;
1166 	int fd = build_open_flags(how, &op);
1167 	struct filename *tmp;
1168 
1169 	if (fd)
1170 		return fd;
1171 
1172 	tmp = getname(filename);
1173 	if (IS_ERR(tmp))
1174 		return PTR_ERR(tmp);
1175 
1176 	fd = get_unused_fd_flags(how->flags);
1177 	if (fd >= 0) {
1178 		struct file *f = do_filp_open(dfd, tmp, &op);
1179 		if (IS_ERR(f)) {
1180 			put_unused_fd(fd);
1181 			fd = PTR_ERR(f);
1182 		} else {
1183 			fsnotify_open(f);
1184 			fd_install(fd, f);
1185 		}
1186 	}
1187 	putname(tmp);
1188 	return fd;
1189 }
1190 
1191 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1192 {
1193 	struct open_how how = build_open_how(flags, mode);
1194 	return do_sys_openat2(dfd, filename, &how);
1195 }
1196 
1197 
1198 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1199 {
1200 	if (force_o_largefile())
1201 		flags |= O_LARGEFILE;
1202 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1203 }
1204 
1205 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1206 		umode_t, mode)
1207 {
1208 	if (force_o_largefile())
1209 		flags |= O_LARGEFILE;
1210 	return do_sys_open(dfd, filename, flags, mode);
1211 }
1212 
1213 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1214 		struct open_how __user *, how, size_t, usize)
1215 {
1216 	int err;
1217 	struct open_how tmp;
1218 
1219 	BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1220 	BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1221 
1222 	if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1223 		return -EINVAL;
1224 
1225 	err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1226 	if (err)
1227 		return err;
1228 
1229 	/* O_LARGEFILE is only allowed for non-O_PATH. */
1230 	if (!(tmp.flags & O_PATH) && force_o_largefile())
1231 		tmp.flags |= O_LARGEFILE;
1232 
1233 	return do_sys_openat2(dfd, filename, &tmp);
1234 }
1235 
1236 #ifdef CONFIG_COMPAT
1237 /*
1238  * Exactly like sys_open(), except that it doesn't set the
1239  * O_LARGEFILE flag.
1240  */
1241 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1242 {
1243 	return do_sys_open(AT_FDCWD, filename, flags, mode);
1244 }
1245 
1246 /*
1247  * Exactly like sys_openat(), except that it doesn't set the
1248  * O_LARGEFILE flag.
1249  */
1250 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1251 {
1252 	return do_sys_open(dfd, filename, flags, mode);
1253 }
1254 #endif
1255 
1256 #ifndef __alpha__
1257 
1258 /*
1259  * For backward compatibility?  Maybe this should be moved
1260  * into arch/i386 instead?
1261  */
1262 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1263 {
1264 	int flags = O_CREAT | O_WRONLY | O_TRUNC;
1265 
1266 	if (force_o_largefile())
1267 		flags |= O_LARGEFILE;
1268 	return do_sys_open(AT_FDCWD, pathname, flags, mode);
1269 }
1270 #endif
1271 
1272 /*
1273  * "id" is the POSIX thread ID. We use the
1274  * files pointer for this..
1275  */
1276 int filp_close(struct file *filp, fl_owner_t id)
1277 {
1278 	int retval = 0;
1279 
1280 	if (!file_count(filp)) {
1281 		printk(KERN_ERR "VFS: Close: file count is 0\n");
1282 		return 0;
1283 	}
1284 
1285 	if (filp->f_op->flush)
1286 		retval = filp->f_op->flush(filp, id);
1287 
1288 	if (likely(!(filp->f_mode & FMODE_PATH))) {
1289 		dnotify_flush(filp, id);
1290 		locks_remove_posix(filp, id);
1291 	}
1292 	fput(filp);
1293 	return retval;
1294 }
1295 
1296 EXPORT_SYMBOL(filp_close);
1297 
1298 /*
1299  * Careful here! We test whether the file pointer is NULL before
1300  * releasing the fd. This ensures that one clone task can't release
1301  * an fd while another clone is opening it.
1302  */
1303 SYSCALL_DEFINE1(close, unsigned int, fd)
1304 {
1305 	int retval = close_fd(fd);
1306 
1307 	/* can't restart close syscall because file table entry was cleared */
1308 	if (unlikely(retval == -ERESTARTSYS ||
1309 		     retval == -ERESTARTNOINTR ||
1310 		     retval == -ERESTARTNOHAND ||
1311 		     retval == -ERESTART_RESTARTBLOCK))
1312 		retval = -EINTR;
1313 
1314 	return retval;
1315 }
1316 
1317 /**
1318  * close_range() - Close all file descriptors in a given range.
1319  *
1320  * @fd:     starting file descriptor to close
1321  * @max_fd: last file descriptor to close
1322  * @flags:  reserved for future extensions
1323  *
1324  * This closes a range of file descriptors. All file descriptors
1325  * from @fd up to and including @max_fd are closed.
1326  * Currently, errors to close a given file descriptor are ignored.
1327  */
1328 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1329 		unsigned int, flags)
1330 {
1331 	return __close_range(fd, max_fd, flags);
1332 }
1333 
1334 /*
1335  * This routine simulates a hangup on the tty, to arrange that users
1336  * are given clean terminals at login time.
1337  */
1338 SYSCALL_DEFINE0(vhangup)
1339 {
1340 	if (capable(CAP_SYS_TTY_CONFIG)) {
1341 		tty_vhangup_self();
1342 		return 0;
1343 	}
1344 	return -EPERM;
1345 }
1346 
1347 /*
1348  * Called when an inode is about to be open.
1349  * We use this to disallow opening large files on 32bit systems if
1350  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1351  * on this flag in sys_open.
1352  */
1353 int generic_file_open(struct inode * inode, struct file * filp)
1354 {
1355 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1356 		return -EOVERFLOW;
1357 	return 0;
1358 }
1359 
1360 EXPORT_SYMBOL(generic_file_open);
1361 
1362 /*
1363  * This is used by subsystems that don't want seekable
1364  * file descriptors. The function is not supposed to ever fail, the only
1365  * reason it returns an 'int' and not 'void' is so that it can be plugged
1366  * directly into file_operations structure.
1367  */
1368 int nonseekable_open(struct inode *inode, struct file *filp)
1369 {
1370 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1371 	return 0;
1372 }
1373 
1374 EXPORT_SYMBOL(nonseekable_open);
1375 
1376 /*
1377  * stream_open is used by subsystems that want stream-like file descriptors.
1378  * Such file descriptors are not seekable and don't have notion of position
1379  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1380  * Contrary to file descriptors of other regular files, .read() and .write()
1381  * can run simultaneously.
1382  *
1383  * stream_open never fails and is marked to return int so that it could be
1384  * directly used as file_operations.open .
1385  */
1386 int stream_open(struct inode *inode, struct file *filp)
1387 {
1388 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1389 	filp->f_mode |= FMODE_STREAM;
1390 	return 0;
1391 }
1392 
1393 EXPORT_SYMBOL(stream_open);
1394