xref: /openbmc/linux/fs/open.c (revision b34081f1)
1 /*
2  *  linux/fs/open.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/string.h>
8 #include <linux/mm.h>
9 #include <linux/file.h>
10 #include <linux/fdtable.h>
11 #include <linux/fsnotify.h>
12 #include <linux/module.h>
13 #include <linux/tty.h>
14 #include <linux/namei.h>
15 #include <linux/backing-dev.h>
16 #include <linux/capability.h>
17 #include <linux/securebits.h>
18 #include <linux/security.h>
19 #include <linux/mount.h>
20 #include <linux/fcntl.h>
21 #include <linux/slab.h>
22 #include <asm/uaccess.h>
23 #include <linux/fs.h>
24 #include <linux/personality.h>
25 #include <linux/pagemap.h>
26 #include <linux/syscalls.h>
27 #include <linux/rcupdate.h>
28 #include <linux/audit.h>
29 #include <linux/falloc.h>
30 #include <linux/fs_struct.h>
31 #include <linux/ima.h>
32 #include <linux/dnotify.h>
33 #include <linux/compat.h>
34 
35 #include "internal.h"
36 
37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
38 	struct file *filp)
39 {
40 	int ret;
41 	struct iattr newattrs;
42 
43 	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
44 	if (length < 0)
45 		return -EINVAL;
46 
47 	newattrs.ia_size = length;
48 	newattrs.ia_valid = ATTR_SIZE | time_attrs;
49 	if (filp) {
50 		newattrs.ia_file = filp;
51 		newattrs.ia_valid |= ATTR_FILE;
52 	}
53 
54 	/* Remove suid/sgid on truncate too */
55 	ret = should_remove_suid(dentry);
56 	if (ret)
57 		newattrs.ia_valid |= ret | ATTR_FORCE;
58 
59 	mutex_lock(&dentry->d_inode->i_mutex);
60 	ret = notify_change(dentry, &newattrs);
61 	mutex_unlock(&dentry->d_inode->i_mutex);
62 	return ret;
63 }
64 
65 long vfs_truncate(struct path *path, loff_t length)
66 {
67 	struct inode *inode;
68 	long error;
69 
70 	inode = path->dentry->d_inode;
71 
72 	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
73 	if (S_ISDIR(inode->i_mode))
74 		return -EISDIR;
75 	if (!S_ISREG(inode->i_mode))
76 		return -EINVAL;
77 
78 	error = mnt_want_write(path->mnt);
79 	if (error)
80 		goto out;
81 
82 	error = inode_permission(inode, MAY_WRITE);
83 	if (error)
84 		goto mnt_drop_write_and_out;
85 
86 	error = -EPERM;
87 	if (IS_APPEND(inode))
88 		goto mnt_drop_write_and_out;
89 
90 	error = get_write_access(inode);
91 	if (error)
92 		goto mnt_drop_write_and_out;
93 
94 	/*
95 	 * Make sure that there are no leases.  get_write_access() protects
96 	 * against the truncate racing with a lease-granting setlease().
97 	 */
98 	error = break_lease(inode, O_WRONLY);
99 	if (error)
100 		goto put_write_and_out;
101 
102 	error = locks_verify_truncate(inode, NULL, length);
103 	if (!error)
104 		error = security_path_truncate(path);
105 	if (!error)
106 		error = do_truncate(path->dentry, length, 0, NULL);
107 
108 put_write_and_out:
109 	put_write_access(inode);
110 mnt_drop_write_and_out:
111 	mnt_drop_write(path->mnt);
112 out:
113 	return error;
114 }
115 EXPORT_SYMBOL_GPL(vfs_truncate);
116 
117 static long do_sys_truncate(const char __user *pathname, loff_t length)
118 {
119 	unsigned int lookup_flags = LOOKUP_FOLLOW;
120 	struct path path;
121 	int error;
122 
123 	if (length < 0)	/* sorry, but loff_t says... */
124 		return -EINVAL;
125 
126 retry:
127 	error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
128 	if (!error) {
129 		error = vfs_truncate(&path, length);
130 		path_put(&path);
131 	}
132 	if (retry_estale(error, lookup_flags)) {
133 		lookup_flags |= LOOKUP_REVAL;
134 		goto retry;
135 	}
136 	return error;
137 }
138 
139 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
140 {
141 	return do_sys_truncate(path, length);
142 }
143 
144 #ifdef CONFIG_COMPAT
145 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
146 {
147 	return do_sys_truncate(path, length);
148 }
149 #endif
150 
151 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
152 {
153 	struct inode *inode;
154 	struct dentry *dentry;
155 	struct fd f;
156 	int error;
157 
158 	error = -EINVAL;
159 	if (length < 0)
160 		goto out;
161 	error = -EBADF;
162 	f = fdget(fd);
163 	if (!f.file)
164 		goto out;
165 
166 	/* explicitly opened as large or we are on 64-bit box */
167 	if (f.file->f_flags & O_LARGEFILE)
168 		small = 0;
169 
170 	dentry = f.file->f_path.dentry;
171 	inode = dentry->d_inode;
172 	error = -EINVAL;
173 	if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
174 		goto out_putf;
175 
176 	error = -EINVAL;
177 	/* Cannot ftruncate over 2^31 bytes without large file support */
178 	if (small && length > MAX_NON_LFS)
179 		goto out_putf;
180 
181 	error = -EPERM;
182 	if (IS_APPEND(inode))
183 		goto out_putf;
184 
185 	sb_start_write(inode->i_sb);
186 	error = locks_verify_truncate(inode, f.file, length);
187 	if (!error)
188 		error = security_path_truncate(&f.file->f_path);
189 	if (!error)
190 		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
191 	sb_end_write(inode->i_sb);
192 out_putf:
193 	fdput(f);
194 out:
195 	return error;
196 }
197 
198 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
199 {
200 	return do_sys_ftruncate(fd, length, 1);
201 }
202 
203 #ifdef CONFIG_COMPAT
204 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
205 {
206 	return do_sys_ftruncate(fd, length, 1);
207 }
208 #endif
209 
210 /* LFS versions of truncate are only needed on 32 bit machines */
211 #if BITS_PER_LONG == 32
212 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
213 {
214 	return do_sys_truncate(path, length);
215 }
216 
217 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
218 {
219 	return do_sys_ftruncate(fd, length, 0);
220 }
221 #endif /* BITS_PER_LONG == 32 */
222 
223 
224 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
225 {
226 	struct inode *inode = file_inode(file);
227 	long ret;
228 
229 	if (offset < 0 || len <= 0)
230 		return -EINVAL;
231 
232 	/* Return error if mode is not supported */
233 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
234 		return -EOPNOTSUPP;
235 
236 	/* Punch hole must have keep size set */
237 	if ((mode & FALLOC_FL_PUNCH_HOLE) &&
238 	    !(mode & FALLOC_FL_KEEP_SIZE))
239 		return -EOPNOTSUPP;
240 
241 	if (!(file->f_mode & FMODE_WRITE))
242 		return -EBADF;
243 
244 	/* It's not possible punch hole on append only file */
245 	if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
246 		return -EPERM;
247 
248 	if (IS_IMMUTABLE(inode))
249 		return -EPERM;
250 
251 	/*
252 	 * Revalidate the write permissions, in case security policy has
253 	 * changed since the files were opened.
254 	 */
255 	ret = security_file_permission(file, MAY_WRITE);
256 	if (ret)
257 		return ret;
258 
259 	if (S_ISFIFO(inode->i_mode))
260 		return -ESPIPE;
261 
262 	/*
263 	 * Let individual file system decide if it supports preallocation
264 	 * for directories or not.
265 	 */
266 	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
267 		return -ENODEV;
268 
269 	/* Check for wrap through zero too */
270 	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
271 		return -EFBIG;
272 
273 	if (!file->f_op->fallocate)
274 		return -EOPNOTSUPP;
275 
276 	sb_start_write(inode->i_sb);
277 	ret = file->f_op->fallocate(file, mode, offset, len);
278 	sb_end_write(inode->i_sb);
279 	return ret;
280 }
281 
282 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
283 {
284 	struct fd f = fdget(fd);
285 	int error = -EBADF;
286 
287 	if (f.file) {
288 		error = do_fallocate(f.file, mode, offset, len);
289 		fdput(f);
290 	}
291 	return error;
292 }
293 
294 /*
295  * access() needs to use the real uid/gid, not the effective uid/gid.
296  * We do this by temporarily clearing all FS-related capabilities and
297  * switching the fsuid/fsgid around to the real ones.
298  */
299 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
300 {
301 	const struct cred *old_cred;
302 	struct cred *override_cred;
303 	struct path path;
304 	struct inode *inode;
305 	int res;
306 	unsigned int lookup_flags = LOOKUP_FOLLOW;
307 
308 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
309 		return -EINVAL;
310 
311 	override_cred = prepare_creds();
312 	if (!override_cred)
313 		return -ENOMEM;
314 
315 	override_cred->fsuid = override_cred->uid;
316 	override_cred->fsgid = override_cred->gid;
317 
318 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
319 		/* Clear the capabilities if we switch to a non-root user */
320 		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
321 		if (!uid_eq(override_cred->uid, root_uid))
322 			cap_clear(override_cred->cap_effective);
323 		else
324 			override_cred->cap_effective =
325 				override_cred->cap_permitted;
326 	}
327 
328 	old_cred = override_creds(override_cred);
329 retry:
330 	res = user_path_at(dfd, filename, lookup_flags, &path);
331 	if (res)
332 		goto out;
333 
334 	inode = path.dentry->d_inode;
335 
336 	if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
337 		/*
338 		 * MAY_EXEC on regular files is denied if the fs is mounted
339 		 * with the "noexec" flag.
340 		 */
341 		res = -EACCES;
342 		if (path.mnt->mnt_flags & MNT_NOEXEC)
343 			goto out_path_release;
344 	}
345 
346 	res = inode_permission(inode, mode | MAY_ACCESS);
347 	/* SuS v2 requires we report a read only fs too */
348 	if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
349 		goto out_path_release;
350 	/*
351 	 * This is a rare case where using __mnt_is_readonly()
352 	 * is OK without a mnt_want/drop_write() pair.  Since
353 	 * no actual write to the fs is performed here, we do
354 	 * not need to telegraph to that to anyone.
355 	 *
356 	 * By doing this, we accept that this access is
357 	 * inherently racy and know that the fs may change
358 	 * state before we even see this result.
359 	 */
360 	if (__mnt_is_readonly(path.mnt))
361 		res = -EROFS;
362 
363 out_path_release:
364 	path_put(&path);
365 	if (retry_estale(res, lookup_flags)) {
366 		lookup_flags |= LOOKUP_REVAL;
367 		goto retry;
368 	}
369 out:
370 	revert_creds(old_cred);
371 	put_cred(override_cred);
372 	return res;
373 }
374 
375 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
376 {
377 	return sys_faccessat(AT_FDCWD, filename, mode);
378 }
379 
380 SYSCALL_DEFINE1(chdir, const char __user *, filename)
381 {
382 	struct path path;
383 	int error;
384 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
385 retry:
386 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
387 	if (error)
388 		goto out;
389 
390 	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
391 	if (error)
392 		goto dput_and_out;
393 
394 	set_fs_pwd(current->fs, &path);
395 
396 dput_and_out:
397 	path_put(&path);
398 	if (retry_estale(error, lookup_flags)) {
399 		lookup_flags |= LOOKUP_REVAL;
400 		goto retry;
401 	}
402 out:
403 	return error;
404 }
405 
406 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
407 {
408 	struct fd f = fdget_raw(fd);
409 	struct inode *inode;
410 	int error = -EBADF;
411 
412 	error = -EBADF;
413 	if (!f.file)
414 		goto out;
415 
416 	inode = file_inode(f.file);
417 
418 	error = -ENOTDIR;
419 	if (!S_ISDIR(inode->i_mode))
420 		goto out_putf;
421 
422 	error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
423 	if (!error)
424 		set_fs_pwd(current->fs, &f.file->f_path);
425 out_putf:
426 	fdput(f);
427 out:
428 	return error;
429 }
430 
431 SYSCALL_DEFINE1(chroot, const char __user *, filename)
432 {
433 	struct path path;
434 	int error;
435 	unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
436 retry:
437 	error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
438 	if (error)
439 		goto out;
440 
441 	error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
442 	if (error)
443 		goto dput_and_out;
444 
445 	error = -EPERM;
446 	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
447 		goto dput_and_out;
448 	error = security_path_chroot(&path);
449 	if (error)
450 		goto dput_and_out;
451 
452 	set_fs_root(current->fs, &path);
453 	error = 0;
454 dput_and_out:
455 	path_put(&path);
456 	if (retry_estale(error, lookup_flags)) {
457 		lookup_flags |= LOOKUP_REVAL;
458 		goto retry;
459 	}
460 out:
461 	return error;
462 }
463 
464 static int chmod_common(struct path *path, umode_t mode)
465 {
466 	struct inode *inode = path->dentry->d_inode;
467 	struct iattr newattrs;
468 	int error;
469 
470 	error = mnt_want_write(path->mnt);
471 	if (error)
472 		return error;
473 	mutex_lock(&inode->i_mutex);
474 	error = security_path_chmod(path, mode);
475 	if (error)
476 		goto out_unlock;
477 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
478 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
479 	error = notify_change(path->dentry, &newattrs);
480 out_unlock:
481 	mutex_unlock(&inode->i_mutex);
482 	mnt_drop_write(path->mnt);
483 	return error;
484 }
485 
486 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
487 {
488 	struct fd f = fdget(fd);
489 	int err = -EBADF;
490 
491 	if (f.file) {
492 		audit_inode(NULL, f.file->f_path.dentry, 0);
493 		err = chmod_common(&f.file->f_path, mode);
494 		fdput(f);
495 	}
496 	return err;
497 }
498 
499 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
500 {
501 	struct path path;
502 	int error;
503 	unsigned int lookup_flags = LOOKUP_FOLLOW;
504 retry:
505 	error = user_path_at(dfd, filename, lookup_flags, &path);
506 	if (!error) {
507 		error = chmod_common(&path, mode);
508 		path_put(&path);
509 		if (retry_estale(error, lookup_flags)) {
510 			lookup_flags |= LOOKUP_REVAL;
511 			goto retry;
512 		}
513 	}
514 	return error;
515 }
516 
517 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
518 {
519 	return sys_fchmodat(AT_FDCWD, filename, mode);
520 }
521 
522 static int chown_common(struct path *path, uid_t user, gid_t group)
523 {
524 	struct inode *inode = path->dentry->d_inode;
525 	int error;
526 	struct iattr newattrs;
527 	kuid_t uid;
528 	kgid_t gid;
529 
530 	uid = make_kuid(current_user_ns(), user);
531 	gid = make_kgid(current_user_ns(), group);
532 
533 	newattrs.ia_valid =  ATTR_CTIME;
534 	if (user != (uid_t) -1) {
535 		if (!uid_valid(uid))
536 			return -EINVAL;
537 		newattrs.ia_valid |= ATTR_UID;
538 		newattrs.ia_uid = uid;
539 	}
540 	if (group != (gid_t) -1) {
541 		if (!gid_valid(gid))
542 			return -EINVAL;
543 		newattrs.ia_valid |= ATTR_GID;
544 		newattrs.ia_gid = gid;
545 	}
546 	if (!S_ISDIR(inode->i_mode))
547 		newattrs.ia_valid |=
548 			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
549 	mutex_lock(&inode->i_mutex);
550 	error = security_path_chown(path, uid, gid);
551 	if (!error)
552 		error = notify_change(path->dentry, &newattrs);
553 	mutex_unlock(&inode->i_mutex);
554 
555 	return error;
556 }
557 
558 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
559 		gid_t, group, int, flag)
560 {
561 	struct path path;
562 	int error = -EINVAL;
563 	int lookup_flags;
564 
565 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
566 		goto out;
567 
568 	lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
569 	if (flag & AT_EMPTY_PATH)
570 		lookup_flags |= LOOKUP_EMPTY;
571 retry:
572 	error = user_path_at(dfd, filename, lookup_flags, &path);
573 	if (error)
574 		goto out;
575 	error = mnt_want_write(path.mnt);
576 	if (error)
577 		goto out_release;
578 	error = chown_common(&path, user, group);
579 	mnt_drop_write(path.mnt);
580 out_release:
581 	path_put(&path);
582 	if (retry_estale(error, lookup_flags)) {
583 		lookup_flags |= LOOKUP_REVAL;
584 		goto retry;
585 	}
586 out:
587 	return error;
588 }
589 
590 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
591 {
592 	return sys_fchownat(AT_FDCWD, filename, user, group, 0);
593 }
594 
595 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
596 {
597 	return sys_fchownat(AT_FDCWD, filename, user, group,
598 			    AT_SYMLINK_NOFOLLOW);
599 }
600 
601 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
602 {
603 	struct fd f = fdget(fd);
604 	int error = -EBADF;
605 
606 	if (!f.file)
607 		goto out;
608 
609 	error = mnt_want_write_file(f.file);
610 	if (error)
611 		goto out_fput;
612 	audit_inode(NULL, f.file->f_path.dentry, 0);
613 	error = chown_common(&f.file->f_path, user, group);
614 	mnt_drop_write_file(f.file);
615 out_fput:
616 	fdput(f);
617 out:
618 	return error;
619 }
620 
621 /*
622  * You have to be very careful that these write
623  * counts get cleaned up in error cases and
624  * upon __fput().  This should probably never
625  * be called outside of __dentry_open().
626  */
627 static inline int __get_file_write_access(struct inode *inode,
628 					  struct vfsmount *mnt)
629 {
630 	int error;
631 	error = get_write_access(inode);
632 	if (error)
633 		return error;
634 	/*
635 	 * Do not take mount writer counts on
636 	 * special files since no writes to
637 	 * the mount itself will occur.
638 	 */
639 	if (!special_file(inode->i_mode)) {
640 		/*
641 		 * Balanced in __fput()
642 		 */
643 		error = __mnt_want_write(mnt);
644 		if (error)
645 			put_write_access(inode);
646 	}
647 	return error;
648 }
649 
650 int open_check_o_direct(struct file *f)
651 {
652 	/* NB: we're sure to have correct a_ops only after f_op->open */
653 	if (f->f_flags & O_DIRECT) {
654 		if (!f->f_mapping->a_ops ||
655 		    ((!f->f_mapping->a_ops->direct_IO) &&
656 		    (!f->f_mapping->a_ops->get_xip_mem))) {
657 			return -EINVAL;
658 		}
659 	}
660 	return 0;
661 }
662 
663 static int do_dentry_open(struct file *f,
664 			  int (*open)(struct inode *, struct file *),
665 			  const struct cred *cred)
666 {
667 	static const struct file_operations empty_fops = {};
668 	struct inode *inode;
669 	int error;
670 
671 	f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
672 				FMODE_PREAD | FMODE_PWRITE;
673 
674 	if (unlikely(f->f_flags & O_PATH))
675 		f->f_mode = FMODE_PATH;
676 
677 	path_get(&f->f_path);
678 	inode = f->f_inode = f->f_path.dentry->d_inode;
679 	if (f->f_mode & FMODE_WRITE) {
680 		error = __get_file_write_access(inode, f->f_path.mnt);
681 		if (error)
682 			goto cleanup_file;
683 		if (!special_file(inode->i_mode))
684 			file_take_write(f);
685 	}
686 
687 	f->f_mapping = inode->i_mapping;
688 	file_sb_list_add(f, inode->i_sb);
689 
690 	if (unlikely(f->f_mode & FMODE_PATH)) {
691 		f->f_op = &empty_fops;
692 		return 0;
693 	}
694 
695 	f->f_op = fops_get(inode->i_fop);
696 
697 	error = security_file_open(f, cred);
698 	if (error)
699 		goto cleanup_all;
700 
701 	error = break_lease(inode, f->f_flags);
702 	if (error)
703 		goto cleanup_all;
704 
705 	if (!open && f->f_op)
706 		open = f->f_op->open;
707 	if (open) {
708 		error = open(inode, f);
709 		if (error)
710 			goto cleanup_all;
711 	}
712 	if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
713 		i_readcount_inc(inode);
714 
715 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
716 
717 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
718 
719 	return 0;
720 
721 cleanup_all:
722 	fops_put(f->f_op);
723 	file_sb_list_del(f);
724 	if (f->f_mode & FMODE_WRITE) {
725 		put_write_access(inode);
726 		if (!special_file(inode->i_mode)) {
727 			/*
728 			 * We don't consider this a real
729 			 * mnt_want/drop_write() pair
730 			 * because it all happenend right
731 			 * here, so just reset the state.
732 			 */
733 			file_reset_write(f);
734 			__mnt_drop_write(f->f_path.mnt);
735 		}
736 	}
737 cleanup_file:
738 	path_put(&f->f_path);
739 	f->f_path.mnt = NULL;
740 	f->f_path.dentry = NULL;
741 	f->f_inode = NULL;
742 	return error;
743 }
744 
745 /**
746  * finish_open - finish opening a file
747  * @od: opaque open data
748  * @dentry: pointer to dentry
749  * @open: open callback
750  *
751  * This can be used to finish opening a file passed to i_op->atomic_open().
752  *
753  * If the open callback is set to NULL, then the standard f_op->open()
754  * filesystem callback is substituted.
755  */
756 int finish_open(struct file *file, struct dentry *dentry,
757 		int (*open)(struct inode *, struct file *),
758 		int *opened)
759 {
760 	int error;
761 	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
762 
763 	file->f_path.dentry = dentry;
764 	error = do_dentry_open(file, open, current_cred());
765 	if (!error)
766 		*opened |= FILE_OPENED;
767 
768 	return error;
769 }
770 EXPORT_SYMBOL(finish_open);
771 
772 /**
773  * finish_no_open - finish ->atomic_open() without opening the file
774  *
775  * @od: opaque open data
776  * @dentry: dentry or NULL (as returned from ->lookup())
777  *
778  * This can be used to set the result of a successful lookup in ->atomic_open().
779  * The filesystem's atomic_open() method shall return NULL after calling this.
780  */
781 int finish_no_open(struct file *file, struct dentry *dentry)
782 {
783 	file->f_path.dentry = dentry;
784 	return 1;
785 }
786 EXPORT_SYMBOL(finish_no_open);
787 
788 struct file *dentry_open(const struct path *path, int flags,
789 			 const struct cred *cred)
790 {
791 	int error;
792 	struct file *f;
793 
794 	validate_creds(cred);
795 
796 	/* We must always pass in a valid mount pointer. */
797 	BUG_ON(!path->mnt);
798 
799 	f = get_empty_filp();
800 	if (!IS_ERR(f)) {
801 		f->f_flags = flags;
802 		f->f_path = *path;
803 		error = do_dentry_open(f, NULL, cred);
804 		if (!error) {
805 			/* from now on we need fput() to dispose of f */
806 			error = open_check_o_direct(f);
807 			if (error) {
808 				fput(f);
809 				f = ERR_PTR(error);
810 			}
811 		} else {
812 			put_filp(f);
813 			f = ERR_PTR(error);
814 		}
815 	}
816 	return f;
817 }
818 EXPORT_SYMBOL(dentry_open);
819 
820 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
821 {
822 	int lookup_flags = 0;
823 	int acc_mode;
824 
825 	if (flags & (O_CREAT | __O_TMPFILE))
826 		op->mode = (mode & S_IALLUGO) | S_IFREG;
827 	else
828 		op->mode = 0;
829 
830 	/* Must never be set by userspace */
831 	flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
832 
833 	/*
834 	 * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
835 	 * check for O_DSYNC if the need any syncing at all we enforce it's
836 	 * always set instead of having to deal with possibly weird behaviour
837 	 * for malicious applications setting only __O_SYNC.
838 	 */
839 	if (flags & __O_SYNC)
840 		flags |= O_DSYNC;
841 
842 	if (flags & __O_TMPFILE) {
843 		if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
844 			return -EINVAL;
845 		acc_mode = MAY_OPEN | ACC_MODE(flags);
846 		if (!(acc_mode & MAY_WRITE))
847 			return -EINVAL;
848 	} else if (flags & O_PATH) {
849 		/*
850 		 * If we have O_PATH in the open flag. Then we
851 		 * cannot have anything other than the below set of flags
852 		 */
853 		flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
854 		acc_mode = 0;
855 	} else {
856 		acc_mode = MAY_OPEN | ACC_MODE(flags);
857 	}
858 
859 	op->open_flag = flags;
860 
861 	/* O_TRUNC implies we need access checks for write permissions */
862 	if (flags & O_TRUNC)
863 		acc_mode |= MAY_WRITE;
864 
865 	/* Allow the LSM permission hook to distinguish append
866 	   access from general write access. */
867 	if (flags & O_APPEND)
868 		acc_mode |= MAY_APPEND;
869 
870 	op->acc_mode = acc_mode;
871 
872 	op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
873 
874 	if (flags & O_CREAT) {
875 		op->intent |= LOOKUP_CREATE;
876 		if (flags & O_EXCL)
877 			op->intent |= LOOKUP_EXCL;
878 	}
879 
880 	if (flags & O_DIRECTORY)
881 		lookup_flags |= LOOKUP_DIRECTORY;
882 	if (!(flags & O_NOFOLLOW))
883 		lookup_flags |= LOOKUP_FOLLOW;
884 	op->lookup_flags = lookup_flags;
885 	return 0;
886 }
887 
888 /**
889  * file_open_name - open file and return file pointer
890  *
891  * @name:	struct filename containing path to open
892  * @flags:	open flags as per the open(2) second argument
893  * @mode:	mode for the new file if O_CREAT is set, else ignored
894  *
895  * This is the helper to open a file from kernelspace if you really
896  * have to.  But in generally you should not do this, so please move
897  * along, nothing to see here..
898  */
899 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
900 {
901 	struct open_flags op;
902 	int err = build_open_flags(flags, mode, &op);
903 	return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
904 }
905 
906 /**
907  * filp_open - open file and return file pointer
908  *
909  * @filename:	path to open
910  * @flags:	open flags as per the open(2) second argument
911  * @mode:	mode for the new file if O_CREAT is set, else ignored
912  *
913  * This is the helper to open a file from kernelspace if you really
914  * have to.  But in generally you should not do this, so please move
915  * along, nothing to see here..
916  */
917 struct file *filp_open(const char *filename, int flags, umode_t mode)
918 {
919 	struct filename name = {.name = filename};
920 	return file_open_name(&name, flags, mode);
921 }
922 EXPORT_SYMBOL(filp_open);
923 
924 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
925 			    const char *filename, int flags)
926 {
927 	struct open_flags op;
928 	int err = build_open_flags(flags, 0, &op);
929 	if (err)
930 		return ERR_PTR(err);
931 	if (flags & O_CREAT)
932 		return ERR_PTR(-EINVAL);
933 	if (!filename && (flags & O_DIRECTORY))
934 		if (!dentry->d_inode->i_op->lookup)
935 			return ERR_PTR(-ENOTDIR);
936 	return do_file_open_root(dentry, mnt, filename, &op);
937 }
938 EXPORT_SYMBOL(file_open_root);
939 
940 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
941 {
942 	struct open_flags op;
943 	int fd = build_open_flags(flags, mode, &op);
944 	struct filename *tmp;
945 
946 	if (fd)
947 		return fd;
948 
949 	tmp = getname(filename);
950 	if (IS_ERR(tmp))
951 		return PTR_ERR(tmp);
952 
953 	fd = get_unused_fd_flags(flags);
954 	if (fd >= 0) {
955 		struct file *f = do_filp_open(dfd, tmp, &op);
956 		if (IS_ERR(f)) {
957 			put_unused_fd(fd);
958 			fd = PTR_ERR(f);
959 		} else {
960 			fsnotify_open(f);
961 			fd_install(fd, f);
962 		}
963 	}
964 	putname(tmp);
965 	return fd;
966 }
967 
968 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
969 {
970 	if (force_o_largefile())
971 		flags |= O_LARGEFILE;
972 
973 	return do_sys_open(AT_FDCWD, filename, flags, mode);
974 }
975 
976 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
977 		umode_t, mode)
978 {
979 	if (force_o_largefile())
980 		flags |= O_LARGEFILE;
981 
982 	return do_sys_open(dfd, filename, flags, mode);
983 }
984 
985 #ifndef __alpha__
986 
987 /*
988  * For backward compatibility?  Maybe this should be moved
989  * into arch/i386 instead?
990  */
991 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
992 {
993 	return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
994 }
995 
996 #endif
997 
998 /*
999  * "id" is the POSIX thread ID. We use the
1000  * files pointer for this..
1001  */
1002 int filp_close(struct file *filp, fl_owner_t id)
1003 {
1004 	int retval = 0;
1005 
1006 	if (!file_count(filp)) {
1007 		printk(KERN_ERR "VFS: Close: file count is 0\n");
1008 		return 0;
1009 	}
1010 
1011 	if (filp->f_op && filp->f_op->flush)
1012 		retval = filp->f_op->flush(filp, id);
1013 
1014 	if (likely(!(filp->f_mode & FMODE_PATH))) {
1015 		dnotify_flush(filp, id);
1016 		locks_remove_posix(filp, id);
1017 	}
1018 	fput(filp);
1019 	return retval;
1020 }
1021 
1022 EXPORT_SYMBOL(filp_close);
1023 
1024 /*
1025  * Careful here! We test whether the file pointer is NULL before
1026  * releasing the fd. This ensures that one clone task can't release
1027  * an fd while another clone is opening it.
1028  */
1029 SYSCALL_DEFINE1(close, unsigned int, fd)
1030 {
1031 	int retval = __close_fd(current->files, fd);
1032 
1033 	/* can't restart close syscall because file table entry was cleared */
1034 	if (unlikely(retval == -ERESTARTSYS ||
1035 		     retval == -ERESTARTNOINTR ||
1036 		     retval == -ERESTARTNOHAND ||
1037 		     retval == -ERESTART_RESTARTBLOCK))
1038 		retval = -EINTR;
1039 
1040 	return retval;
1041 }
1042 EXPORT_SYMBOL(sys_close);
1043 
1044 /*
1045  * This routine simulates a hangup on the tty, to arrange that users
1046  * are given clean terminals at login time.
1047  */
1048 SYSCALL_DEFINE0(vhangup)
1049 {
1050 	if (capable(CAP_SYS_TTY_CONFIG)) {
1051 		tty_vhangup_self();
1052 		return 0;
1053 	}
1054 	return -EPERM;
1055 }
1056 
1057 /*
1058  * Called when an inode is about to be open.
1059  * We use this to disallow opening large files on 32bit systems if
1060  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1061  * on this flag in sys_open.
1062  */
1063 int generic_file_open(struct inode * inode, struct file * filp)
1064 {
1065 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1066 		return -EOVERFLOW;
1067 	return 0;
1068 }
1069 
1070 EXPORT_SYMBOL(generic_file_open);
1071 
1072 /*
1073  * This is used by subsystems that don't want seekable
1074  * file descriptors. The function is not supposed to ever fail, the only
1075  * reason it returns an 'int' and not 'void' is so that it can be plugged
1076  * directly into file_operations structure.
1077  */
1078 int nonseekable_open(struct inode *inode, struct file *filp)
1079 {
1080 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1081 	return 0;
1082 }
1083 
1084 EXPORT_SYMBOL(nonseekable_open);
1085