xref: /linux/fs/open.c (revision d91958815d214ea365b98cbff6215383897edcb6)
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/quotaops.h>
12 #include <linux/fsnotify.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/tty.h>
16 #include <linux/namei.h>
17 #include <linux/backing-dev.h>
18 #include <linux/capability.h>
19 #include <linux/securebits.h>
20 #include <linux/security.h>
21 #include <linux/mount.h>
22 #include <linux/vfs.h>
23 #include <linux/fcntl.h>
24 #include <asm/uaccess.h>
25 #include <linux/fs.h>
26 #include <linux/personality.h>
27 #include <linux/pagemap.h>
28 #include <linux/syscalls.h>
29 #include <linux/rcupdate.h>
30 #include <linux/audit.h>
31 #include <linux/falloc.h>
32 
33 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
34 {
35 	int retval = -ENODEV;
36 
37 	if (dentry) {
38 		retval = -ENOSYS;
39 		if (dentry->d_sb->s_op->statfs) {
40 			memset(buf, 0, sizeof(*buf));
41 			retval = security_sb_statfs(dentry);
42 			if (retval)
43 				return retval;
44 			retval = dentry->d_sb->s_op->statfs(dentry, buf);
45 			if (retval == 0 && buf->f_frsize == 0)
46 				buf->f_frsize = buf->f_bsize;
47 		}
48 	}
49 	return retval;
50 }
51 
52 EXPORT_SYMBOL(vfs_statfs);
53 
54 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
55 {
56 	struct kstatfs st;
57 	int retval;
58 
59 	retval = vfs_statfs(dentry, &st);
60 	if (retval)
61 		return retval;
62 
63 	if (sizeof(*buf) == sizeof(st))
64 		memcpy(buf, &st, sizeof(st));
65 	else {
66 		if (sizeof buf->f_blocks == 4) {
67 			if ((st.f_blocks | st.f_bfree | st.f_bavail |
68 			     st.f_bsize | st.f_frsize) &
69 			    0xffffffff00000000ULL)
70 				return -EOVERFLOW;
71 			/*
72 			 * f_files and f_ffree may be -1; it's okay to stuff
73 			 * that into 32 bits
74 			 */
75 			if (st.f_files != -1 &&
76 			    (st.f_files & 0xffffffff00000000ULL))
77 				return -EOVERFLOW;
78 			if (st.f_ffree != -1 &&
79 			    (st.f_ffree & 0xffffffff00000000ULL))
80 				return -EOVERFLOW;
81 		}
82 
83 		buf->f_type = st.f_type;
84 		buf->f_bsize = st.f_bsize;
85 		buf->f_blocks = st.f_blocks;
86 		buf->f_bfree = st.f_bfree;
87 		buf->f_bavail = st.f_bavail;
88 		buf->f_files = st.f_files;
89 		buf->f_ffree = st.f_ffree;
90 		buf->f_fsid = st.f_fsid;
91 		buf->f_namelen = st.f_namelen;
92 		buf->f_frsize = st.f_frsize;
93 		memset(buf->f_spare, 0, sizeof(buf->f_spare));
94 	}
95 	return 0;
96 }
97 
98 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
99 {
100 	struct kstatfs st;
101 	int retval;
102 
103 	retval = vfs_statfs(dentry, &st);
104 	if (retval)
105 		return retval;
106 
107 	if (sizeof(*buf) == sizeof(st))
108 		memcpy(buf, &st, sizeof(st));
109 	else {
110 		buf->f_type = st.f_type;
111 		buf->f_bsize = st.f_bsize;
112 		buf->f_blocks = st.f_blocks;
113 		buf->f_bfree = st.f_bfree;
114 		buf->f_bavail = st.f_bavail;
115 		buf->f_files = st.f_files;
116 		buf->f_ffree = st.f_ffree;
117 		buf->f_fsid = st.f_fsid;
118 		buf->f_namelen = st.f_namelen;
119 		buf->f_frsize = st.f_frsize;
120 		memset(buf->f_spare, 0, sizeof(buf->f_spare));
121 	}
122 	return 0;
123 }
124 
125 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
126 {
127 	struct nameidata nd;
128 	int error;
129 
130 	error = user_path_walk(path, &nd);
131 	if (!error) {
132 		struct statfs tmp;
133 		error = vfs_statfs_native(nd.path.dentry, &tmp);
134 		if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
135 			error = -EFAULT;
136 		path_put(&nd.path);
137 	}
138 	return error;
139 }
140 
141 
142 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
143 {
144 	struct nameidata nd;
145 	long error;
146 
147 	if (sz != sizeof(*buf))
148 		return -EINVAL;
149 	error = user_path_walk(path, &nd);
150 	if (!error) {
151 		struct statfs64 tmp;
152 		error = vfs_statfs64(nd.path.dentry, &tmp);
153 		if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
154 			error = -EFAULT;
155 		path_put(&nd.path);
156 	}
157 	return error;
158 }
159 
160 
161 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
162 {
163 	struct file * file;
164 	struct statfs tmp;
165 	int error;
166 
167 	error = -EBADF;
168 	file = fget(fd);
169 	if (!file)
170 		goto out;
171 	error = vfs_statfs_native(file->f_path.dentry, &tmp);
172 	if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
173 		error = -EFAULT;
174 	fput(file);
175 out:
176 	return error;
177 }
178 
179 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
180 {
181 	struct file * file;
182 	struct statfs64 tmp;
183 	int error;
184 
185 	if (sz != sizeof(*buf))
186 		return -EINVAL;
187 
188 	error = -EBADF;
189 	file = fget(fd);
190 	if (!file)
191 		goto out;
192 	error = vfs_statfs64(file->f_path.dentry, &tmp);
193 	if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
194 		error = -EFAULT;
195 	fput(file);
196 out:
197 	return error;
198 }
199 
200 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
201 	struct file *filp)
202 {
203 	int err;
204 	struct iattr newattrs;
205 
206 	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
207 	if (length < 0)
208 		return -EINVAL;
209 
210 	newattrs.ia_size = length;
211 	newattrs.ia_valid = ATTR_SIZE | time_attrs;
212 	if (filp) {
213 		newattrs.ia_file = filp;
214 		newattrs.ia_valid |= ATTR_FILE;
215 	}
216 
217 	/* Remove suid/sgid on truncate too */
218 	newattrs.ia_valid |= should_remove_suid(dentry);
219 
220 	mutex_lock(&dentry->d_inode->i_mutex);
221 	err = notify_change(dentry, &newattrs);
222 	mutex_unlock(&dentry->d_inode->i_mutex);
223 	return err;
224 }
225 
226 static long do_sys_truncate(const char __user * path, loff_t length)
227 {
228 	struct nameidata nd;
229 	struct inode * inode;
230 	int error;
231 
232 	error = -EINVAL;
233 	if (length < 0)	/* sorry, but loff_t says... */
234 		goto out;
235 
236 	error = user_path_walk(path, &nd);
237 	if (error)
238 		goto out;
239 	inode = nd.path.dentry->d_inode;
240 
241 	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
242 	error = -EISDIR;
243 	if (S_ISDIR(inode->i_mode))
244 		goto dput_and_out;
245 
246 	error = -EINVAL;
247 	if (!S_ISREG(inode->i_mode))
248 		goto dput_and_out;
249 
250 	error = mnt_want_write(nd.path.mnt);
251 	if (error)
252 		goto dput_and_out;
253 
254 	error = vfs_permission(&nd, MAY_WRITE);
255 	if (error)
256 		goto mnt_drop_write_and_out;
257 
258 	error = -EPERM;
259 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
260 		goto mnt_drop_write_and_out;
261 
262 	error = get_write_access(inode);
263 	if (error)
264 		goto mnt_drop_write_and_out;
265 
266 	/*
267 	 * Make sure that there are no leases.  get_write_access() protects
268 	 * against the truncate racing with a lease-granting setlease().
269 	 */
270 	error = break_lease(inode, FMODE_WRITE);
271 	if (error)
272 		goto put_write_and_out;
273 
274 	error = locks_verify_truncate(inode, NULL, length);
275 	if (!error) {
276 		DQUOT_INIT(inode);
277 		error = do_truncate(nd.path.dentry, length, 0, NULL);
278 	}
279 
280 put_write_and_out:
281 	put_write_access(inode);
282 mnt_drop_write_and_out:
283 	mnt_drop_write(nd.path.mnt);
284 dput_and_out:
285 	path_put(&nd.path);
286 out:
287 	return error;
288 }
289 
290 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
291 {
292 	/* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
293 	return do_sys_truncate(path, (long)length);
294 }
295 
296 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
297 {
298 	struct inode * inode;
299 	struct dentry *dentry;
300 	struct file * file;
301 	int error;
302 
303 	error = -EINVAL;
304 	if (length < 0)
305 		goto out;
306 	error = -EBADF;
307 	file = fget(fd);
308 	if (!file)
309 		goto out;
310 
311 	/* explicitly opened as large or we are on 64-bit box */
312 	if (file->f_flags & O_LARGEFILE)
313 		small = 0;
314 
315 	dentry = file->f_path.dentry;
316 	inode = dentry->d_inode;
317 	error = -EINVAL;
318 	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
319 		goto out_putf;
320 
321 	error = -EINVAL;
322 	/* Cannot ftruncate over 2^31 bytes without large file support */
323 	if (small && length > MAX_NON_LFS)
324 		goto out_putf;
325 
326 	error = -EPERM;
327 	if (IS_APPEND(inode))
328 		goto out_putf;
329 
330 	error = locks_verify_truncate(inode, file, length);
331 	if (!error)
332 		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
333 out_putf:
334 	fput(file);
335 out:
336 	return error;
337 }
338 
339 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
340 {
341 	long ret = do_sys_ftruncate(fd, length, 1);
342 	/* avoid REGPARM breakage on x86: */
343 	asmlinkage_protect(2, ret, fd, length);
344 	return ret;
345 }
346 
347 /* LFS versions of truncate are only needed on 32 bit machines */
348 #if BITS_PER_LONG == 32
349 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
350 {
351 	return do_sys_truncate(path, length);
352 }
353 
354 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
355 {
356 	long ret = do_sys_ftruncate(fd, length, 0);
357 	/* avoid REGPARM breakage on x86: */
358 	asmlinkage_protect(2, ret, fd, length);
359 	return ret;
360 }
361 #endif
362 
363 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
364 {
365 	struct file *file;
366 	struct inode *inode;
367 	long ret = -EINVAL;
368 
369 	if (offset < 0 || len <= 0)
370 		goto out;
371 
372 	/* Return error if mode is not supported */
373 	ret = -EOPNOTSUPP;
374 	if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
375 		goto out;
376 
377 	ret = -EBADF;
378 	file = fget(fd);
379 	if (!file)
380 		goto out;
381 	if (!(file->f_mode & FMODE_WRITE))
382 		goto out_fput;
383 	/*
384 	 * Revalidate the write permissions, in case security policy has
385 	 * changed since the files were opened.
386 	 */
387 	ret = security_file_permission(file, MAY_WRITE);
388 	if (ret)
389 		goto out_fput;
390 
391 	inode = file->f_path.dentry->d_inode;
392 
393 	ret = -ESPIPE;
394 	if (S_ISFIFO(inode->i_mode))
395 		goto out_fput;
396 
397 	ret = -ENODEV;
398 	/*
399 	 * Let individual file system decide if it supports preallocation
400 	 * for directories or not.
401 	 */
402 	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
403 		goto out_fput;
404 
405 	ret = -EFBIG;
406 	/* Check for wrap through zero too */
407 	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
408 		goto out_fput;
409 
410 	if (inode->i_op && inode->i_op->fallocate)
411 		ret = inode->i_op->fallocate(inode, mode, offset, len);
412 	else
413 		ret = -EOPNOTSUPP;
414 
415 out_fput:
416 	fput(file);
417 out:
418 	return ret;
419 }
420 
421 /*
422  * access() needs to use the real uid/gid, not the effective uid/gid.
423  * We do this by temporarily clearing all FS-related capabilities and
424  * switching the fsuid/fsgid around to the real ones.
425  */
426 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
427 {
428 	struct nameidata nd;
429 	int old_fsuid, old_fsgid;
430 	kernel_cap_t uninitialized_var(old_cap);  /* !SECURE_NO_SETUID_FIXUP */
431 	int res;
432 
433 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
434 		return -EINVAL;
435 
436 	old_fsuid = current->fsuid;
437 	old_fsgid = current->fsgid;
438 
439 	current->fsuid = current->uid;
440 	current->fsgid = current->gid;
441 
442 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
443 		/*
444 		 * Clear the capabilities if we switch to a non-root user
445 		 */
446 #ifndef CONFIG_SECURITY_FILE_CAPABILITIES
447 		/*
448 		 * FIXME: There is a race here against sys_capset.  The
449 		 * capabilities can change yet we will restore the old
450 		 * value below.  We should hold task_capabilities_lock,
451 		 * but we cannot because user_path_walk can sleep.
452 		 */
453 #endif /* ndef CONFIG_SECURITY_FILE_CAPABILITIES */
454 		if (current->uid)
455 			old_cap = cap_set_effective(__cap_empty_set);
456 		else
457 			old_cap = cap_set_effective(current->cap_permitted);
458 	}
459 
460 	res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
461 	if (res)
462 		goto out;
463 
464 	res = vfs_permission(&nd, mode);
465 	/* SuS v2 requires we report a read only fs too */
466 	if(res || !(mode & S_IWOTH) ||
467 	   special_file(nd.path.dentry->d_inode->i_mode))
468 		goto out_path_release;
469 	/*
470 	 * This is a rare case where using __mnt_is_readonly()
471 	 * is OK without a mnt_want/drop_write() pair.  Since
472 	 * no actual write to the fs is performed here, we do
473 	 * not need to telegraph to that to anyone.
474 	 *
475 	 * By doing this, we accept that this access is
476 	 * inherently racy and know that the fs may change
477 	 * state before we even see this result.
478 	 */
479 	if (__mnt_is_readonly(nd.path.mnt))
480 		res = -EROFS;
481 
482 out_path_release:
483 	path_put(&nd.path);
484 out:
485 	current->fsuid = old_fsuid;
486 	current->fsgid = old_fsgid;
487 
488 	if (!issecure(SECURE_NO_SETUID_FIXUP))
489 		cap_set_effective(old_cap);
490 
491 	return res;
492 }
493 
494 asmlinkage long sys_access(const char __user *filename, int mode)
495 {
496 	return sys_faccessat(AT_FDCWD, filename, mode);
497 }
498 
499 asmlinkage long sys_chdir(const char __user * filename)
500 {
501 	struct nameidata nd;
502 	int error;
503 
504 	error = __user_walk(filename,
505 			    LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
506 	if (error)
507 		goto out;
508 
509 	error = vfs_permission(&nd, MAY_EXEC);
510 	if (error)
511 		goto dput_and_out;
512 
513 	set_fs_pwd(current->fs, &nd.path);
514 
515 dput_and_out:
516 	path_put(&nd.path);
517 out:
518 	return error;
519 }
520 
521 asmlinkage long sys_fchdir(unsigned int fd)
522 {
523 	struct file *file;
524 	struct inode *inode;
525 	int error;
526 
527 	error = -EBADF;
528 	file = fget(fd);
529 	if (!file)
530 		goto out;
531 
532 	inode = file->f_path.dentry->d_inode;
533 
534 	error = -ENOTDIR;
535 	if (!S_ISDIR(inode->i_mode))
536 		goto out_putf;
537 
538 	error = file_permission(file, MAY_EXEC);
539 	if (!error)
540 		set_fs_pwd(current->fs, &file->f_path);
541 out_putf:
542 	fput(file);
543 out:
544 	return error;
545 }
546 
547 asmlinkage long sys_chroot(const char __user * filename)
548 {
549 	struct nameidata nd;
550 	int error;
551 
552 	error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
553 	if (error)
554 		goto out;
555 
556 	error = vfs_permission(&nd, MAY_EXEC);
557 	if (error)
558 		goto dput_and_out;
559 
560 	error = -EPERM;
561 	if (!capable(CAP_SYS_CHROOT))
562 		goto dput_and_out;
563 
564 	set_fs_root(current->fs, &nd.path);
565 	set_fs_altroot();
566 	error = 0;
567 dput_and_out:
568 	path_put(&nd.path);
569 out:
570 	return error;
571 }
572 
573 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
574 {
575 	struct inode * inode;
576 	struct dentry * dentry;
577 	struct file * file;
578 	int err = -EBADF;
579 	struct iattr newattrs;
580 
581 	file = fget(fd);
582 	if (!file)
583 		goto out;
584 
585 	dentry = file->f_path.dentry;
586 	inode = dentry->d_inode;
587 
588 	audit_inode(NULL, dentry);
589 
590 	err = mnt_want_write(file->f_path.mnt);
591 	if (err)
592 		goto out_putf;
593 	err = -EPERM;
594 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
595 		goto out_drop_write;
596 	mutex_lock(&inode->i_mutex);
597 	if (mode == (mode_t) -1)
598 		mode = inode->i_mode;
599 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
600 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
601 	err = notify_change(dentry, &newattrs);
602 	mutex_unlock(&inode->i_mutex);
603 
604 out_drop_write:
605 	mnt_drop_write(file->f_path.mnt);
606 out_putf:
607 	fput(file);
608 out:
609 	return err;
610 }
611 
612 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
613 			     mode_t mode)
614 {
615 	struct nameidata nd;
616 	struct inode * inode;
617 	int error;
618 	struct iattr newattrs;
619 
620 	error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
621 	if (error)
622 		goto out;
623 	inode = nd.path.dentry->d_inode;
624 
625 	error = mnt_want_write(nd.path.mnt);
626 	if (error)
627 		goto dput_and_out;
628 
629 	error = -EPERM;
630 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
631 		goto out_drop_write;
632 
633 	mutex_lock(&inode->i_mutex);
634 	if (mode == (mode_t) -1)
635 		mode = inode->i_mode;
636 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
637 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
638 	error = notify_change(nd.path.dentry, &newattrs);
639 	mutex_unlock(&inode->i_mutex);
640 
641 out_drop_write:
642 	mnt_drop_write(nd.path.mnt);
643 dput_and_out:
644 	path_put(&nd.path);
645 out:
646 	return error;
647 }
648 
649 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
650 {
651 	return sys_fchmodat(AT_FDCWD, filename, mode);
652 }
653 
654 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
655 {
656 	struct inode * inode;
657 	int error;
658 	struct iattr newattrs;
659 
660 	error = -ENOENT;
661 	if (!(inode = dentry->d_inode)) {
662 		printk(KERN_ERR "chown_common: NULL inode\n");
663 		goto out;
664 	}
665 	error = -EPERM;
666 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
667 		goto out;
668 	newattrs.ia_valid =  ATTR_CTIME;
669 	if (user != (uid_t) -1) {
670 		newattrs.ia_valid |= ATTR_UID;
671 		newattrs.ia_uid = user;
672 	}
673 	if (group != (gid_t) -1) {
674 		newattrs.ia_valid |= ATTR_GID;
675 		newattrs.ia_gid = group;
676 	}
677 	if (!S_ISDIR(inode->i_mode))
678 		newattrs.ia_valid |=
679 			ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
680 	mutex_lock(&inode->i_mutex);
681 	error = notify_change(dentry, &newattrs);
682 	mutex_unlock(&inode->i_mutex);
683 out:
684 	return error;
685 }
686 
687 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
688 {
689 	struct nameidata nd;
690 	int error;
691 
692 	error = user_path_walk(filename, &nd);
693 	if (error)
694 		goto out;
695 	error = mnt_want_write(nd.path.mnt);
696 	if (error)
697 		goto out_release;
698 	error = chown_common(nd.path.dentry, user, group);
699 	mnt_drop_write(nd.path.mnt);
700 out_release:
701 	path_put(&nd.path);
702 out:
703 	return error;
704 }
705 
706 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
707 			     gid_t group, int flag)
708 {
709 	struct nameidata nd;
710 	int error = -EINVAL;
711 	int follow;
712 
713 	if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
714 		goto out;
715 
716 	follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
717 	error = __user_walk_fd(dfd, filename, follow, &nd);
718 	if (error)
719 		goto out;
720 	error = mnt_want_write(nd.path.mnt);
721 	if (error)
722 		goto out_release;
723 	error = chown_common(nd.path.dentry, user, group);
724 	mnt_drop_write(nd.path.mnt);
725 out_release:
726 	path_put(&nd.path);
727 out:
728 	return error;
729 }
730 
731 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
732 {
733 	struct nameidata nd;
734 	int error;
735 
736 	error = user_path_walk_link(filename, &nd);
737 	if (error)
738 		goto out;
739 	error = mnt_want_write(nd.path.mnt);
740 	if (error)
741 		goto out_release;
742 	error = chown_common(nd.path.dentry, user, group);
743 	mnt_drop_write(nd.path.mnt);
744 out_release:
745 	path_put(&nd.path);
746 out:
747 	return error;
748 }
749 
750 
751 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
752 {
753 	struct file * file;
754 	int error = -EBADF;
755 	struct dentry * dentry;
756 
757 	file = fget(fd);
758 	if (!file)
759 		goto out;
760 
761 	error = mnt_want_write(file->f_path.mnt);
762 	if (error)
763 		goto out_fput;
764 	dentry = file->f_path.dentry;
765 	audit_inode(NULL, dentry);
766 	error = chown_common(dentry, user, group);
767 	mnt_drop_write(file->f_path.mnt);
768 out_fput:
769 	fput(file);
770 out:
771 	return error;
772 }
773 
774 /*
775  * You have to be very careful that these write
776  * counts get cleaned up in error cases and
777  * upon __fput().  This should probably never
778  * be called outside of __dentry_open().
779  */
780 static inline int __get_file_write_access(struct inode *inode,
781 					  struct vfsmount *mnt)
782 {
783 	int error;
784 	error = get_write_access(inode);
785 	if (error)
786 		return error;
787 	/*
788 	 * Do not take mount writer counts on
789 	 * special files since no writes to
790 	 * the mount itself will occur.
791 	 */
792 	if (!special_file(inode->i_mode)) {
793 		/*
794 		 * Balanced in __fput()
795 		 */
796 		error = mnt_want_write(mnt);
797 		if (error)
798 			put_write_access(inode);
799 	}
800 	return error;
801 }
802 
803 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
804 					int flags, struct file *f,
805 					int (*open)(struct inode *, struct file *))
806 {
807 	struct inode *inode;
808 	int error;
809 
810 	f->f_flags = flags;
811 	f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
812 				FMODE_PREAD | FMODE_PWRITE;
813 	inode = dentry->d_inode;
814 	if (f->f_mode & FMODE_WRITE) {
815 		error = __get_file_write_access(inode, mnt);
816 		if (error)
817 			goto cleanup_file;
818 		if (!special_file(inode->i_mode))
819 			file_take_write(f);
820 	}
821 
822 	f->f_mapping = inode->i_mapping;
823 	f->f_path.dentry = dentry;
824 	f->f_path.mnt = mnt;
825 	f->f_pos = 0;
826 	f->f_op = fops_get(inode->i_fop);
827 	file_move(f, &inode->i_sb->s_files);
828 
829 	error = security_dentry_open(f);
830 	if (error)
831 		goto cleanup_all;
832 
833 	if (!open && f->f_op)
834 		open = f->f_op->open;
835 	if (open) {
836 		error = open(inode, f);
837 		if (error)
838 			goto cleanup_all;
839 	}
840 
841 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
842 
843 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
844 
845 	/* NB: we're sure to have correct a_ops only after f_op->open */
846 	if (f->f_flags & O_DIRECT) {
847 		if (!f->f_mapping->a_ops ||
848 		    ((!f->f_mapping->a_ops->direct_IO) &&
849 		    (!f->f_mapping->a_ops->get_xip_mem))) {
850 			fput(f);
851 			f = ERR_PTR(-EINVAL);
852 		}
853 	}
854 
855 	return f;
856 
857 cleanup_all:
858 	fops_put(f->f_op);
859 	if (f->f_mode & FMODE_WRITE) {
860 		put_write_access(inode);
861 		if (!special_file(inode->i_mode)) {
862 			/*
863 			 * We don't consider this a real
864 			 * mnt_want/drop_write() pair
865 			 * because it all happenend right
866 			 * here, so just reset the state.
867 			 */
868 			file_reset_write(f);
869 			mnt_drop_write(mnt);
870 		}
871 	}
872 	file_kill(f);
873 	f->f_path.dentry = NULL;
874 	f->f_path.mnt = NULL;
875 cleanup_file:
876 	put_filp(f);
877 	dput(dentry);
878 	mntput(mnt);
879 	return ERR_PTR(error);
880 }
881 
882 /**
883  * lookup_instantiate_filp - instantiates the open intent filp
884  * @nd: pointer to nameidata
885  * @dentry: pointer to dentry
886  * @open: open callback
887  *
888  * Helper for filesystems that want to use lookup open intents and pass back
889  * a fully instantiated struct file to the caller.
890  * This function is meant to be called from within a filesystem's
891  * lookup method.
892  * Beware of calling it for non-regular files! Those ->open methods might block
893  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
894  * leading to a deadlock, as nobody can open that fifo anymore, because
895  * another process to open fifo will block on locked parent when doing lookup).
896  * Note that in case of error, nd->intent.open.file is destroyed, but the
897  * path information remains valid.
898  * If the open callback is set to NULL, then the standard f_op->open()
899  * filesystem callback is substituted.
900  */
901 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
902 		int (*open)(struct inode *, struct file *))
903 {
904 	if (IS_ERR(nd->intent.open.file))
905 		goto out;
906 	if (IS_ERR(dentry))
907 		goto out_err;
908 	nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
909 					     nd->intent.open.flags - 1,
910 					     nd->intent.open.file,
911 					     open);
912 out:
913 	return nd->intent.open.file;
914 out_err:
915 	release_open_intent(nd);
916 	nd->intent.open.file = (struct file *)dentry;
917 	goto out;
918 }
919 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
920 
921 /**
922  * nameidata_to_filp - convert a nameidata to an open filp.
923  * @nd: pointer to nameidata
924  * @flags: open flags
925  *
926  * Note that this function destroys the original nameidata
927  */
928 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
929 {
930 	struct file *filp;
931 
932 	/* Pick up the filp from the open intent */
933 	filp = nd->intent.open.file;
934 	/* Has the filesystem initialised the file for us? */
935 	if (filp->f_path.dentry == NULL)
936 		filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
937 				     NULL);
938 	else
939 		path_put(&nd->path);
940 	return filp;
941 }
942 
943 /*
944  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
945  * error.
946  */
947 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
948 {
949 	int error;
950 	struct file *f;
951 
952 	/*
953 	 * We must always pass in a valid mount pointer.   Historically
954 	 * callers got away with not passing it, but we must enforce this at
955 	 * the earliest possible point now to avoid strange problems deep in the
956 	 * filesystem stack.
957 	 */
958 	if (!mnt) {
959 		printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
960 		dump_stack();
961 		return ERR_PTR(-EINVAL);
962 	}
963 
964 	error = -ENFILE;
965 	f = get_empty_filp();
966 	if (f == NULL) {
967 		dput(dentry);
968 		mntput(mnt);
969 		return ERR_PTR(error);
970 	}
971 
972 	return __dentry_open(dentry, mnt, flags, f, NULL);
973 }
974 EXPORT_SYMBOL(dentry_open);
975 
976 /*
977  * Find an empty file descriptor entry, and mark it busy.
978  */
979 int get_unused_fd_flags(int flags)
980 {
981 	struct files_struct * files = current->files;
982 	int fd, error;
983 	struct fdtable *fdt;
984 
985   	error = -EMFILE;
986 	spin_lock(&files->file_lock);
987 
988 repeat:
989 	fdt = files_fdtable(files);
990 	fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
991 				files->next_fd);
992 
993 	/*
994 	 * N.B. For clone tasks sharing a files structure, this test
995 	 * will limit the total number of files that can be opened.
996 	 */
997 	if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
998 		goto out;
999 
1000 	/* Do we need to expand the fd array or fd set?  */
1001 	error = expand_files(files, fd);
1002 	if (error < 0)
1003 		goto out;
1004 
1005 	if (error) {
1006 		/*
1007 	 	 * If we needed to expand the fs array we
1008 		 * might have blocked - try again.
1009 		 */
1010 		error = -EMFILE;
1011 		goto repeat;
1012 	}
1013 
1014 	FD_SET(fd, fdt->open_fds);
1015 	if (flags & O_CLOEXEC)
1016 		FD_SET(fd, fdt->close_on_exec);
1017 	else
1018 		FD_CLR(fd, fdt->close_on_exec);
1019 	files->next_fd = fd + 1;
1020 #if 1
1021 	/* Sanity check */
1022 	if (fdt->fd[fd] != NULL) {
1023 		printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
1024 		fdt->fd[fd] = NULL;
1025 	}
1026 #endif
1027 	error = fd;
1028 
1029 out:
1030 	spin_unlock(&files->file_lock);
1031 	return error;
1032 }
1033 
1034 int get_unused_fd(void)
1035 {
1036 	return get_unused_fd_flags(0);
1037 }
1038 
1039 EXPORT_SYMBOL(get_unused_fd);
1040 
1041 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1042 {
1043 	struct fdtable *fdt = files_fdtable(files);
1044 	__FD_CLR(fd, fdt->open_fds);
1045 	if (fd < files->next_fd)
1046 		files->next_fd = fd;
1047 }
1048 
1049 void put_unused_fd(unsigned int fd)
1050 {
1051 	struct files_struct *files = current->files;
1052 	spin_lock(&files->file_lock);
1053 	__put_unused_fd(files, fd);
1054 	spin_unlock(&files->file_lock);
1055 }
1056 
1057 EXPORT_SYMBOL(put_unused_fd);
1058 
1059 /*
1060  * Install a file pointer in the fd array.
1061  *
1062  * The VFS is full of places where we drop the files lock between
1063  * setting the open_fds bitmap and installing the file in the file
1064  * array.  At any such point, we are vulnerable to a dup2() race
1065  * installing a file in the array before us.  We need to detect this and
1066  * fput() the struct file we are about to overwrite in this case.
1067  *
1068  * It should never happen - if we allow dup2() do it, _really_ bad things
1069  * will follow.
1070  */
1071 
1072 void fd_install(unsigned int fd, struct file *file)
1073 {
1074 	struct files_struct *files = current->files;
1075 	struct fdtable *fdt;
1076 	spin_lock(&files->file_lock);
1077 	fdt = files_fdtable(files);
1078 	BUG_ON(fdt->fd[fd] != NULL);
1079 	rcu_assign_pointer(fdt->fd[fd], file);
1080 	spin_unlock(&files->file_lock);
1081 }
1082 
1083 EXPORT_SYMBOL(fd_install);
1084 
1085 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1086 {
1087 	char *tmp = getname(filename);
1088 	int fd = PTR_ERR(tmp);
1089 
1090 	if (!IS_ERR(tmp)) {
1091 		fd = get_unused_fd_flags(flags);
1092 		if (fd >= 0) {
1093 			struct file *f = do_filp_open(dfd, tmp, flags, mode);
1094 			if (IS_ERR(f)) {
1095 				put_unused_fd(fd);
1096 				fd = PTR_ERR(f);
1097 			} else {
1098 				fsnotify_open(f->f_path.dentry);
1099 				fd_install(fd, f);
1100 			}
1101 		}
1102 		putname(tmp);
1103 	}
1104 	return fd;
1105 }
1106 
1107 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1108 {
1109 	long ret;
1110 
1111 	if (force_o_largefile())
1112 		flags |= O_LARGEFILE;
1113 
1114 	ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1115 	/* avoid REGPARM breakage on x86: */
1116 	asmlinkage_protect(3, ret, filename, flags, mode);
1117 	return ret;
1118 }
1119 
1120 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1121 			   int mode)
1122 {
1123 	long ret;
1124 
1125 	if (force_o_largefile())
1126 		flags |= O_LARGEFILE;
1127 
1128 	ret = do_sys_open(dfd, filename, flags, mode);
1129 	/* avoid REGPARM breakage on x86: */
1130 	asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1131 	return ret;
1132 }
1133 
1134 #ifndef __alpha__
1135 
1136 /*
1137  * For backward compatibility?  Maybe this should be moved
1138  * into arch/i386 instead?
1139  */
1140 asmlinkage long sys_creat(const char __user * pathname, int mode)
1141 {
1142 	return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1143 }
1144 
1145 #endif
1146 
1147 /*
1148  * "id" is the POSIX thread ID. We use the
1149  * files pointer for this..
1150  */
1151 int filp_close(struct file *filp, fl_owner_t id)
1152 {
1153 	int retval = 0;
1154 
1155 	if (!file_count(filp)) {
1156 		printk(KERN_ERR "VFS: Close: file count is 0\n");
1157 		return 0;
1158 	}
1159 
1160 	if (filp->f_op && filp->f_op->flush)
1161 		retval = filp->f_op->flush(filp, id);
1162 
1163 	dnotify_flush(filp, id);
1164 	locks_remove_posix(filp, id);
1165 	fput(filp);
1166 	return retval;
1167 }
1168 
1169 EXPORT_SYMBOL(filp_close);
1170 
1171 /*
1172  * Careful here! We test whether the file pointer is NULL before
1173  * releasing the fd. This ensures that one clone task can't release
1174  * an fd while another clone is opening it.
1175  */
1176 asmlinkage long sys_close(unsigned int fd)
1177 {
1178 	struct file * filp;
1179 	struct files_struct *files = current->files;
1180 	struct fdtable *fdt;
1181 	int retval;
1182 
1183 	spin_lock(&files->file_lock);
1184 	fdt = files_fdtable(files);
1185 	if (fd >= fdt->max_fds)
1186 		goto out_unlock;
1187 	filp = fdt->fd[fd];
1188 	if (!filp)
1189 		goto out_unlock;
1190 	rcu_assign_pointer(fdt->fd[fd], NULL);
1191 	FD_CLR(fd, fdt->close_on_exec);
1192 	__put_unused_fd(files, fd);
1193 	spin_unlock(&files->file_lock);
1194 	retval = filp_close(filp, files);
1195 
1196 	/* can't restart close syscall because file table entry was cleared */
1197 	if (unlikely(retval == -ERESTARTSYS ||
1198 		     retval == -ERESTARTNOINTR ||
1199 		     retval == -ERESTARTNOHAND ||
1200 		     retval == -ERESTART_RESTARTBLOCK))
1201 		retval = -EINTR;
1202 
1203 	return retval;
1204 
1205 out_unlock:
1206 	spin_unlock(&files->file_lock);
1207 	return -EBADF;
1208 }
1209 
1210 EXPORT_SYMBOL(sys_close);
1211 
1212 /*
1213  * This routine simulates a hangup on the tty, to arrange that users
1214  * are given clean terminals at login time.
1215  */
1216 asmlinkage long sys_vhangup(void)
1217 {
1218 	if (capable(CAP_SYS_TTY_CONFIG)) {
1219 		/* XXX: this needs locking */
1220 		tty_vhangup(current->signal->tty);
1221 		return 0;
1222 	}
1223 	return -EPERM;
1224 }
1225 
1226 /*
1227  * Called when an inode is about to be open.
1228  * We use this to disallow opening large files on 32bit systems if
1229  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1230  * on this flag in sys_open.
1231  */
1232 int generic_file_open(struct inode * inode, struct file * filp)
1233 {
1234 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1235 		return -EOVERFLOW;
1236 	return 0;
1237 }
1238 
1239 EXPORT_SYMBOL(generic_file_open);
1240 
1241 /*
1242  * This is used by subsystems that don't want seekable
1243  * file descriptors
1244  */
1245 int nonseekable_open(struct inode *inode, struct file *filp)
1246 {
1247 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1248 	return 0;
1249 }
1250 
1251 EXPORT_SYMBOL(nonseekable_open);
1252