xref: /linux/fs/open.c (revision 04c71976500352d02f60616d2b960267d8c5fe24)
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/quotaops.h>
11 #include <linux/fsnotify.h>
12 #include <linux/module.h>
13 #include <linux/slab.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/security.h>
19 #include <linux/mount.h>
20 #include <linux/vfs.h>
21 #include <linux/fcntl.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 
31 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
32 {
33 	int retval = -ENODEV;
34 
35 	if (dentry) {
36 		retval = -ENOSYS;
37 		if (dentry->d_sb->s_op->statfs) {
38 			memset(buf, 0, sizeof(*buf));
39 			retval = security_sb_statfs(dentry);
40 			if (retval)
41 				return retval;
42 			retval = dentry->d_sb->s_op->statfs(dentry, buf);
43 			if (retval == 0 && buf->f_frsize == 0)
44 				buf->f_frsize = buf->f_bsize;
45 		}
46 	}
47 	return retval;
48 }
49 
50 EXPORT_SYMBOL(vfs_statfs);
51 
52 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
53 {
54 	struct kstatfs st;
55 	int retval;
56 
57 	retval = vfs_statfs(dentry, &st);
58 	if (retval)
59 		return retval;
60 
61 	if (sizeof(*buf) == sizeof(st))
62 		memcpy(buf, &st, sizeof(st));
63 	else {
64 		if (sizeof buf->f_blocks == 4) {
65 			if ((st.f_blocks | st.f_bfree | st.f_bavail) &
66 			    0xffffffff00000000ULL)
67 				return -EOVERFLOW;
68 			/*
69 			 * f_files and f_ffree may be -1; it's okay to stuff
70 			 * that into 32 bits
71 			 */
72 			if (st.f_files != -1 &&
73 			    (st.f_files & 0xffffffff00000000ULL))
74 				return -EOVERFLOW;
75 			if (st.f_ffree != -1 &&
76 			    (st.f_ffree & 0xffffffff00000000ULL))
77 				return -EOVERFLOW;
78 		}
79 
80 		buf->f_type = st.f_type;
81 		buf->f_bsize = st.f_bsize;
82 		buf->f_blocks = st.f_blocks;
83 		buf->f_bfree = st.f_bfree;
84 		buf->f_bavail = st.f_bavail;
85 		buf->f_files = st.f_files;
86 		buf->f_ffree = st.f_ffree;
87 		buf->f_fsid = st.f_fsid;
88 		buf->f_namelen = st.f_namelen;
89 		buf->f_frsize = st.f_frsize;
90 		memset(buf->f_spare, 0, sizeof(buf->f_spare));
91 	}
92 	return 0;
93 }
94 
95 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
96 {
97 	struct kstatfs st;
98 	int retval;
99 
100 	retval = vfs_statfs(dentry, &st);
101 	if (retval)
102 		return retval;
103 
104 	if (sizeof(*buf) == sizeof(st))
105 		memcpy(buf, &st, sizeof(st));
106 	else {
107 		buf->f_type = st.f_type;
108 		buf->f_bsize = st.f_bsize;
109 		buf->f_blocks = st.f_blocks;
110 		buf->f_bfree = st.f_bfree;
111 		buf->f_bavail = st.f_bavail;
112 		buf->f_files = st.f_files;
113 		buf->f_ffree = st.f_ffree;
114 		buf->f_fsid = st.f_fsid;
115 		buf->f_namelen = st.f_namelen;
116 		buf->f_frsize = st.f_frsize;
117 		memset(buf->f_spare, 0, sizeof(buf->f_spare));
118 	}
119 	return 0;
120 }
121 
122 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
123 {
124 	struct nameidata nd;
125 	int error;
126 
127 	error = user_path_walk(path, &nd);
128 	if (!error) {
129 		struct statfs tmp;
130 		error = vfs_statfs_native(nd.dentry, &tmp);
131 		if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
132 			error = -EFAULT;
133 		path_release(&nd);
134 	}
135 	return error;
136 }
137 
138 
139 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
140 {
141 	struct nameidata nd;
142 	long error;
143 
144 	if (sz != sizeof(*buf))
145 		return -EINVAL;
146 	error = user_path_walk(path, &nd);
147 	if (!error) {
148 		struct statfs64 tmp;
149 		error = vfs_statfs64(nd.dentry, &tmp);
150 		if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
151 			error = -EFAULT;
152 		path_release(&nd);
153 	}
154 	return error;
155 }
156 
157 
158 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
159 {
160 	struct file * file;
161 	struct statfs tmp;
162 	int error;
163 
164 	error = -EBADF;
165 	file = fget(fd);
166 	if (!file)
167 		goto out;
168 	error = vfs_statfs_native(file->f_path.dentry, &tmp);
169 	if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
170 		error = -EFAULT;
171 	fput(file);
172 out:
173 	return error;
174 }
175 
176 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
177 {
178 	struct file * file;
179 	struct statfs64 tmp;
180 	int error;
181 
182 	if (sz != sizeof(*buf))
183 		return -EINVAL;
184 
185 	error = -EBADF;
186 	file = fget(fd);
187 	if (!file)
188 		goto out;
189 	error = vfs_statfs64(file->f_path.dentry, &tmp);
190 	if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
191 		error = -EFAULT;
192 	fput(file);
193 out:
194 	return error;
195 }
196 
197 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
198 	struct file *filp)
199 {
200 	int err;
201 	struct iattr newattrs;
202 
203 	/* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
204 	if (length < 0)
205 		return -EINVAL;
206 
207 	newattrs.ia_size = length;
208 	newattrs.ia_valid = ATTR_SIZE | time_attrs;
209 	if (filp) {
210 		newattrs.ia_file = filp;
211 		newattrs.ia_valid |= ATTR_FILE;
212 	}
213 
214 	/* Remove suid/sgid on truncate too */
215 	newattrs.ia_valid |= should_remove_suid(dentry);
216 
217 	mutex_lock(&dentry->d_inode->i_mutex);
218 	err = notify_change(dentry, &newattrs);
219 	mutex_unlock(&dentry->d_inode->i_mutex);
220 	return err;
221 }
222 
223 static long do_sys_truncate(const char __user * path, loff_t length)
224 {
225 	struct nameidata nd;
226 	struct inode * inode;
227 	int error;
228 
229 	error = -EINVAL;
230 	if (length < 0)	/* sorry, but loff_t says... */
231 		goto out;
232 
233 	error = user_path_walk(path, &nd);
234 	if (error)
235 		goto out;
236 	inode = nd.dentry->d_inode;
237 
238 	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
239 	error = -EISDIR;
240 	if (S_ISDIR(inode->i_mode))
241 		goto dput_and_out;
242 
243 	error = -EINVAL;
244 	if (!S_ISREG(inode->i_mode))
245 		goto dput_and_out;
246 
247 	error = vfs_permission(&nd, MAY_WRITE);
248 	if (error)
249 		goto dput_and_out;
250 
251 	error = -EROFS;
252 	if (IS_RDONLY(inode))
253 		goto dput_and_out;
254 
255 	error = -EPERM;
256 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
257 		goto dput_and_out;
258 
259 	error = get_write_access(inode);
260 	if (error)
261 		goto dput_and_out;
262 
263 	/*
264 	 * Make sure that there are no leases.  get_write_access() protects
265 	 * against the truncate racing with a lease-granting setlease().
266 	 */
267 	error = break_lease(inode, FMODE_WRITE);
268 	if (error)
269 		goto put_write_and_out;
270 
271 	error = locks_verify_truncate(inode, NULL, length);
272 	if (!error) {
273 		DQUOT_INIT(inode);
274 		error = do_truncate(nd.dentry, length, 0, NULL);
275 	}
276 
277 put_write_and_out:
278 	put_write_access(inode);
279 dput_and_out:
280 	path_release(&nd);
281 out:
282 	return error;
283 }
284 
285 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
286 {
287 	/* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
288 	return do_sys_truncate(path, (long)length);
289 }
290 
291 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
292 {
293 	struct inode * inode;
294 	struct dentry *dentry;
295 	struct file * file;
296 	int error;
297 
298 	error = -EINVAL;
299 	if (length < 0)
300 		goto out;
301 	error = -EBADF;
302 	file = fget(fd);
303 	if (!file)
304 		goto out;
305 
306 	/* explicitly opened as large or we are on 64-bit box */
307 	if (file->f_flags & O_LARGEFILE)
308 		small = 0;
309 
310 	dentry = file->f_path.dentry;
311 	inode = dentry->d_inode;
312 	error = -EINVAL;
313 	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
314 		goto out_putf;
315 
316 	error = -EINVAL;
317 	/* Cannot ftruncate over 2^31 bytes without large file support */
318 	if (small && length > MAX_NON_LFS)
319 		goto out_putf;
320 
321 	error = -EPERM;
322 	if (IS_APPEND(inode))
323 		goto out_putf;
324 
325 	error = locks_verify_truncate(inode, file, length);
326 	if (!error)
327 		error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
328 out_putf:
329 	fput(file);
330 out:
331 	return error;
332 }
333 
334 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
335 {
336 	long ret = do_sys_ftruncate(fd, length, 1);
337 	/* avoid REGPARM breakage on x86: */
338 	prevent_tail_call(ret);
339 	return ret;
340 }
341 
342 /* LFS versions of truncate are only needed on 32 bit machines */
343 #if BITS_PER_LONG == 32
344 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
345 {
346 	return do_sys_truncate(path, length);
347 }
348 
349 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
350 {
351 	long ret = do_sys_ftruncate(fd, length, 0);
352 	/* avoid REGPARM breakage on x86: */
353 	prevent_tail_call(ret);
354 	return ret;
355 }
356 #endif
357 
358 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
359 {
360 	struct file *file;
361 	struct inode *inode;
362 	long ret = -EINVAL;
363 
364 	if (offset < 0 || len <= 0)
365 		goto out;
366 
367 	/* Return error if mode is not supported */
368 	ret = -EOPNOTSUPP;
369 	if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
370 		goto out;
371 
372 	ret = -EBADF;
373 	file = fget(fd);
374 	if (!file)
375 		goto out;
376 	if (!(file->f_mode & FMODE_WRITE))
377 		goto out_fput;
378 	/*
379 	 * Revalidate the write permissions, in case security policy has
380 	 * changed since the files were opened.
381 	 */
382 	ret = security_file_permission(file, MAY_WRITE);
383 	if (ret)
384 		goto out_fput;
385 
386 	inode = file->f_path.dentry->d_inode;
387 
388 	ret = -ESPIPE;
389 	if (S_ISFIFO(inode->i_mode))
390 		goto out_fput;
391 
392 	ret = -ENODEV;
393 	/*
394 	 * Let individual file system decide if it supports preallocation
395 	 * for directories or not.
396 	 */
397 	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
398 		goto out_fput;
399 
400 	ret = -EFBIG;
401 	/* Check for wrap through zero too */
402 	if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
403 		goto out_fput;
404 
405 	if (inode->i_op && inode->i_op->fallocate)
406 		ret = inode->i_op->fallocate(inode, mode, offset, len);
407 	else
408 		ret = -EOPNOTSUPP;
409 
410 out_fput:
411 	fput(file);
412 out:
413 	return ret;
414 }
415 
416 /*
417  * access() needs to use the real uid/gid, not the effective uid/gid.
418  * We do this by temporarily clearing all FS-related capabilities and
419  * switching the fsuid/fsgid around to the real ones.
420  */
421 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
422 {
423 	struct nameidata nd;
424 	int old_fsuid, old_fsgid;
425 	kernel_cap_t old_cap;
426 	int res;
427 
428 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
429 		return -EINVAL;
430 
431 	old_fsuid = current->fsuid;
432 	old_fsgid = current->fsgid;
433 	old_cap = current->cap_effective;
434 
435 	current->fsuid = current->uid;
436 	current->fsgid = current->gid;
437 
438 	/*
439 	 * Clear the capabilities if we switch to a non-root user
440 	 *
441 	 * FIXME: There is a race here against sys_capset.  The
442 	 * capabilities can change yet we will restore the old
443 	 * value below.  We should hold task_capabilities_lock,
444 	 * but we cannot because user_path_walk can sleep.
445 	 */
446 	if (current->uid)
447 		cap_clear(current->cap_effective);
448 	else
449 		current->cap_effective = current->cap_permitted;
450 
451 	res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
452 	if (res)
453 		goto out;
454 
455 	res = vfs_permission(&nd, mode);
456 	/* SuS v2 requires we report a read only fs too */
457 	if(res || !(mode & S_IWOTH) ||
458 	   special_file(nd.dentry->d_inode->i_mode))
459 		goto out_path_release;
460 
461 	if(IS_RDONLY(nd.dentry->d_inode))
462 		res = -EROFS;
463 
464 out_path_release:
465 	path_release(&nd);
466 out:
467 	current->fsuid = old_fsuid;
468 	current->fsgid = old_fsgid;
469 	current->cap_effective = old_cap;
470 
471 	return res;
472 }
473 
474 asmlinkage long sys_access(const char __user *filename, int mode)
475 {
476 	return sys_faccessat(AT_FDCWD, filename, mode);
477 }
478 
479 asmlinkage long sys_chdir(const char __user * filename)
480 {
481 	struct nameidata nd;
482 	int error;
483 
484 	error = __user_walk(filename,
485 			    LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
486 	if (error)
487 		goto out;
488 
489 	error = vfs_permission(&nd, MAY_EXEC);
490 	if (error)
491 		goto dput_and_out;
492 
493 	set_fs_pwd(current->fs, nd.mnt, nd.dentry);
494 
495 dput_and_out:
496 	path_release(&nd);
497 out:
498 	return error;
499 }
500 
501 asmlinkage long sys_fchdir(unsigned int fd)
502 {
503 	struct file *file;
504 	struct dentry *dentry;
505 	struct inode *inode;
506 	struct vfsmount *mnt;
507 	int error;
508 
509 	error = -EBADF;
510 	file = fget(fd);
511 	if (!file)
512 		goto out;
513 
514 	dentry = file->f_path.dentry;
515 	mnt = file->f_path.mnt;
516 	inode = dentry->d_inode;
517 
518 	error = -ENOTDIR;
519 	if (!S_ISDIR(inode->i_mode))
520 		goto out_putf;
521 
522 	error = file_permission(file, MAY_EXEC);
523 	if (!error)
524 		set_fs_pwd(current->fs, mnt, dentry);
525 out_putf:
526 	fput(file);
527 out:
528 	return error;
529 }
530 
531 asmlinkage long sys_chroot(const char __user * filename)
532 {
533 	struct nameidata nd;
534 	int error;
535 
536 	error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
537 	if (error)
538 		goto out;
539 
540 	error = vfs_permission(&nd, MAY_EXEC);
541 	if (error)
542 		goto dput_and_out;
543 
544 	error = -EPERM;
545 	if (!capable(CAP_SYS_CHROOT))
546 		goto dput_and_out;
547 
548 	set_fs_root(current->fs, nd.mnt, nd.dentry);
549 	set_fs_altroot();
550 	error = 0;
551 dput_and_out:
552 	path_release(&nd);
553 out:
554 	return error;
555 }
556 
557 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
558 {
559 	struct inode * inode;
560 	struct dentry * dentry;
561 	struct file * file;
562 	int err = -EBADF;
563 	struct iattr newattrs;
564 
565 	file = fget(fd);
566 	if (!file)
567 		goto out;
568 
569 	dentry = file->f_path.dentry;
570 	inode = dentry->d_inode;
571 
572 	audit_inode(NULL, inode);
573 
574 	err = -EROFS;
575 	if (IS_RDONLY(inode))
576 		goto out_putf;
577 	err = -EPERM;
578 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
579 		goto out_putf;
580 	mutex_lock(&inode->i_mutex);
581 	if (mode == (mode_t) -1)
582 		mode = inode->i_mode;
583 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
584 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
585 	err = notify_change(dentry, &newattrs);
586 	mutex_unlock(&inode->i_mutex);
587 
588 out_putf:
589 	fput(file);
590 out:
591 	return err;
592 }
593 
594 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
595 			     mode_t mode)
596 {
597 	struct nameidata nd;
598 	struct inode * inode;
599 	int error;
600 	struct iattr newattrs;
601 
602 	error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
603 	if (error)
604 		goto out;
605 	inode = nd.dentry->d_inode;
606 
607 	error = -EROFS;
608 	if (IS_RDONLY(inode))
609 		goto dput_and_out;
610 
611 	error = -EPERM;
612 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
613 		goto dput_and_out;
614 
615 	mutex_lock(&inode->i_mutex);
616 	if (mode == (mode_t) -1)
617 		mode = inode->i_mode;
618 	newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
619 	newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
620 	error = notify_change(nd.dentry, &newattrs);
621 	mutex_unlock(&inode->i_mutex);
622 
623 dput_and_out:
624 	path_release(&nd);
625 out:
626 	return error;
627 }
628 
629 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
630 {
631 	return sys_fchmodat(AT_FDCWD, filename, mode);
632 }
633 
634 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
635 {
636 	struct inode * inode;
637 	int error;
638 	struct iattr newattrs;
639 
640 	error = -ENOENT;
641 	if (!(inode = dentry->d_inode)) {
642 		printk(KERN_ERR "chown_common: NULL inode\n");
643 		goto out;
644 	}
645 	error = -EROFS;
646 	if (IS_RDONLY(inode))
647 		goto out;
648 	error = -EPERM;
649 	if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
650 		goto out;
651 	newattrs.ia_valid =  ATTR_CTIME;
652 	if (user != (uid_t) -1) {
653 		newattrs.ia_valid |= ATTR_UID;
654 		newattrs.ia_uid = user;
655 	}
656 	if (group != (gid_t) -1) {
657 		newattrs.ia_valid |= ATTR_GID;
658 		newattrs.ia_gid = group;
659 	}
660 	if (!S_ISDIR(inode->i_mode))
661 		newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
662 	mutex_lock(&inode->i_mutex);
663 	error = notify_change(dentry, &newattrs);
664 	mutex_unlock(&inode->i_mutex);
665 out:
666 	return error;
667 }
668 
669 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
670 {
671 	struct nameidata nd;
672 	int error;
673 
674 	error = user_path_walk(filename, &nd);
675 	if (error)
676 		goto out;
677 	error = chown_common(nd.dentry, user, group);
678 	path_release(&nd);
679 out:
680 	return error;
681 }
682 
683 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
684 			     gid_t group, int flag)
685 {
686 	struct nameidata nd;
687 	int error = -EINVAL;
688 	int follow;
689 
690 	if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
691 		goto out;
692 
693 	follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
694 	error = __user_walk_fd(dfd, filename, follow, &nd);
695 	if (error)
696 		goto out;
697 	error = chown_common(nd.dentry, user, group);
698 	path_release(&nd);
699 out:
700 	return error;
701 }
702 
703 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
704 {
705 	struct nameidata nd;
706 	int error;
707 
708 	error = user_path_walk_link(filename, &nd);
709 	if (error)
710 		goto out;
711 	error = chown_common(nd.dentry, user, group);
712 	path_release(&nd);
713 out:
714 	return error;
715 }
716 
717 
718 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
719 {
720 	struct file * file;
721 	int error = -EBADF;
722 	struct dentry * dentry;
723 
724 	file = fget(fd);
725 	if (!file)
726 		goto out;
727 
728 	dentry = file->f_path.dentry;
729 	audit_inode(NULL, dentry->d_inode);
730 	error = chown_common(dentry, user, group);
731 	fput(file);
732 out:
733 	return error;
734 }
735 
736 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
737 					int flags, struct file *f,
738 					int (*open)(struct inode *, struct file *))
739 {
740 	struct inode *inode;
741 	int error;
742 
743 	f->f_flags = flags;
744 	f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
745 				FMODE_PREAD | FMODE_PWRITE;
746 	inode = dentry->d_inode;
747 	if (f->f_mode & FMODE_WRITE) {
748 		error = get_write_access(inode);
749 		if (error)
750 			goto cleanup_file;
751 	}
752 
753 	f->f_mapping = inode->i_mapping;
754 	f->f_path.dentry = dentry;
755 	f->f_path.mnt = mnt;
756 	f->f_pos = 0;
757 	f->f_op = fops_get(inode->i_fop);
758 	file_move(f, &inode->i_sb->s_files);
759 
760 	error = security_dentry_open(f);
761 	if (error)
762 		goto cleanup_all;
763 
764 	if (!open && f->f_op)
765 		open = f->f_op->open;
766 	if (open) {
767 		error = open(inode, f);
768 		if (error)
769 			goto cleanup_all;
770 	}
771 
772 	f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
773 
774 	file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
775 
776 	/* NB: we're sure to have correct a_ops only after f_op->open */
777 	if (f->f_flags & O_DIRECT) {
778 		if (!f->f_mapping->a_ops ||
779 		    ((!f->f_mapping->a_ops->direct_IO) &&
780 		    (!f->f_mapping->a_ops->get_xip_page))) {
781 			fput(f);
782 			f = ERR_PTR(-EINVAL);
783 		}
784 	}
785 
786 	return f;
787 
788 cleanup_all:
789 	fops_put(f->f_op);
790 	if (f->f_mode & FMODE_WRITE)
791 		put_write_access(inode);
792 	file_kill(f);
793 	f->f_path.dentry = NULL;
794 	f->f_path.mnt = NULL;
795 cleanup_file:
796 	put_filp(f);
797 	dput(dentry);
798 	mntput(mnt);
799 	return ERR_PTR(error);
800 }
801 
802 /*
803  * Note that while the flag value (low two bits) for sys_open means:
804  *	00 - read-only
805  *	01 - write-only
806  *	10 - read-write
807  *	11 - special
808  * it is changed into
809  *	00 - no permissions needed
810  *	01 - read-permission
811  *	10 - write-permission
812  *	11 - read-write
813  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
814  * used by symlinks.
815  */
816 static struct file *do_filp_open(int dfd, const char *filename, int flags,
817 				 int mode)
818 {
819 	int namei_flags, error;
820 	struct nameidata nd;
821 
822 	namei_flags = flags;
823 	if ((namei_flags+1) & O_ACCMODE)
824 		namei_flags++;
825 
826 	error = open_namei(dfd, filename, namei_flags, mode, &nd);
827 	if (!error)
828 		return nameidata_to_filp(&nd, flags);
829 
830 	return ERR_PTR(error);
831 }
832 
833 struct file *filp_open(const char *filename, int flags, int mode)
834 {
835 	return do_filp_open(AT_FDCWD, filename, flags, mode);
836 }
837 EXPORT_SYMBOL(filp_open);
838 
839 /**
840  * lookup_instantiate_filp - instantiates the open intent filp
841  * @nd: pointer to nameidata
842  * @dentry: pointer to dentry
843  * @open: open callback
844  *
845  * Helper for filesystems that want to use lookup open intents and pass back
846  * a fully instantiated struct file to the caller.
847  * This function is meant to be called from within a filesystem's
848  * lookup method.
849  * Beware of calling it for non-regular files! Those ->open methods might block
850  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
851  * leading to a deadlock, as nobody can open that fifo anymore, because
852  * another process to open fifo will block on locked parent when doing lookup).
853  * Note that in case of error, nd->intent.open.file is destroyed, but the
854  * path information remains valid.
855  * If the open callback is set to NULL, then the standard f_op->open()
856  * filesystem callback is substituted.
857  */
858 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
859 		int (*open)(struct inode *, struct file *))
860 {
861 	if (IS_ERR(nd->intent.open.file))
862 		goto out;
863 	if (IS_ERR(dentry))
864 		goto out_err;
865 	nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->mnt),
866 					     nd->intent.open.flags - 1,
867 					     nd->intent.open.file,
868 					     open);
869 out:
870 	return nd->intent.open.file;
871 out_err:
872 	release_open_intent(nd);
873 	nd->intent.open.file = (struct file *)dentry;
874 	goto out;
875 }
876 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
877 
878 /**
879  * nameidata_to_filp - convert a nameidata to an open filp.
880  * @nd: pointer to nameidata
881  * @flags: open flags
882  *
883  * Note that this function destroys the original nameidata
884  */
885 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
886 {
887 	struct file *filp;
888 
889 	/* Pick up the filp from the open intent */
890 	filp = nd->intent.open.file;
891 	/* Has the filesystem initialised the file for us? */
892 	if (filp->f_path.dentry == NULL)
893 		filp = __dentry_open(nd->dentry, nd->mnt, flags, filp, NULL);
894 	else
895 		path_release(nd);
896 	return filp;
897 }
898 
899 /*
900  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
901  * error.
902  */
903 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
904 {
905 	int error;
906 	struct file *f;
907 
908 	error = -ENFILE;
909 	f = get_empty_filp();
910 	if (f == NULL) {
911 		dput(dentry);
912 		mntput(mnt);
913 		return ERR_PTR(error);
914 	}
915 
916 	return __dentry_open(dentry, mnt, flags, f, NULL);
917 }
918 EXPORT_SYMBOL(dentry_open);
919 
920 /*
921  * Find an empty file descriptor entry, and mark it busy.
922  */
923 int get_unused_fd_flags(int flags)
924 {
925 	struct files_struct * files = current->files;
926 	int fd, error;
927 	struct fdtable *fdt;
928 
929   	error = -EMFILE;
930 	spin_lock(&files->file_lock);
931 
932 repeat:
933 	fdt = files_fdtable(files);
934 	fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
935 				files->next_fd);
936 
937 	/*
938 	 * N.B. For clone tasks sharing a files structure, this test
939 	 * will limit the total number of files that can be opened.
940 	 */
941 	if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
942 		goto out;
943 
944 	/* Do we need to expand the fd array or fd set?  */
945 	error = expand_files(files, fd);
946 	if (error < 0)
947 		goto out;
948 
949 	if (error) {
950 		/*
951 	 	 * If we needed to expand the fs array we
952 		 * might have blocked - try again.
953 		 */
954 		error = -EMFILE;
955 		goto repeat;
956 	}
957 
958 	FD_SET(fd, fdt->open_fds);
959 	if (flags & O_CLOEXEC)
960 		FD_SET(fd, fdt->close_on_exec);
961 	else
962 		FD_CLR(fd, fdt->close_on_exec);
963 	files->next_fd = fd + 1;
964 #if 1
965 	/* Sanity check */
966 	if (fdt->fd[fd] != NULL) {
967 		printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
968 		fdt->fd[fd] = NULL;
969 	}
970 #endif
971 	error = fd;
972 
973 out:
974 	spin_unlock(&files->file_lock);
975 	return error;
976 }
977 
978 int get_unused_fd(void)
979 {
980 	return get_unused_fd_flags(0);
981 }
982 
983 EXPORT_SYMBOL(get_unused_fd);
984 
985 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
986 {
987 	struct fdtable *fdt = files_fdtable(files);
988 	__FD_CLR(fd, fdt->open_fds);
989 	if (fd < files->next_fd)
990 		files->next_fd = fd;
991 }
992 
993 void fastcall put_unused_fd(unsigned int fd)
994 {
995 	struct files_struct *files = current->files;
996 	spin_lock(&files->file_lock);
997 	__put_unused_fd(files, fd);
998 	spin_unlock(&files->file_lock);
999 }
1000 
1001 EXPORT_SYMBOL(put_unused_fd);
1002 
1003 /*
1004  * Install a file pointer in the fd array.
1005  *
1006  * The VFS is full of places where we drop the files lock between
1007  * setting the open_fds bitmap and installing the file in the file
1008  * array.  At any such point, we are vulnerable to a dup2() race
1009  * installing a file in the array before us.  We need to detect this and
1010  * fput() the struct file we are about to overwrite in this case.
1011  *
1012  * It should never happen - if we allow dup2() do it, _really_ bad things
1013  * will follow.
1014  */
1015 
1016 void fastcall fd_install(unsigned int fd, struct file * file)
1017 {
1018 	struct files_struct *files = current->files;
1019 	struct fdtable *fdt;
1020 	spin_lock(&files->file_lock);
1021 	fdt = files_fdtable(files);
1022 	BUG_ON(fdt->fd[fd] != NULL);
1023 	rcu_assign_pointer(fdt->fd[fd], file);
1024 	spin_unlock(&files->file_lock);
1025 }
1026 
1027 EXPORT_SYMBOL(fd_install);
1028 
1029 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1030 {
1031 	char *tmp = getname(filename);
1032 	int fd = PTR_ERR(tmp);
1033 
1034 	if (!IS_ERR(tmp)) {
1035 		fd = get_unused_fd_flags(flags);
1036 		if (fd >= 0) {
1037 			struct file *f = do_filp_open(dfd, tmp, flags, mode);
1038 			if (IS_ERR(f)) {
1039 				put_unused_fd(fd);
1040 				fd = PTR_ERR(f);
1041 			} else {
1042 				fsnotify_open(f->f_path.dentry);
1043 				fd_install(fd, f);
1044 			}
1045 		}
1046 		putname(tmp);
1047 	}
1048 	return fd;
1049 }
1050 
1051 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1052 {
1053 	long ret;
1054 
1055 	if (force_o_largefile())
1056 		flags |= O_LARGEFILE;
1057 
1058 	ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1059 	/* avoid REGPARM breakage on x86: */
1060 	prevent_tail_call(ret);
1061 	return ret;
1062 }
1063 EXPORT_SYMBOL_GPL(sys_open);
1064 
1065 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1066 			   int mode)
1067 {
1068 	long ret;
1069 
1070 	if (force_o_largefile())
1071 		flags |= O_LARGEFILE;
1072 
1073 	ret = do_sys_open(dfd, filename, flags, mode);
1074 	/* avoid REGPARM breakage on x86: */
1075 	prevent_tail_call(ret);
1076 	return ret;
1077 }
1078 
1079 #ifndef __alpha__
1080 
1081 /*
1082  * For backward compatibility?  Maybe this should be moved
1083  * into arch/i386 instead?
1084  */
1085 asmlinkage long sys_creat(const char __user * pathname, int mode)
1086 {
1087 	return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1088 }
1089 
1090 #endif
1091 
1092 /*
1093  * "id" is the POSIX thread ID. We use the
1094  * files pointer for this..
1095  */
1096 int filp_close(struct file *filp, fl_owner_t id)
1097 {
1098 	int retval = 0;
1099 
1100 	if (!file_count(filp)) {
1101 		printk(KERN_ERR "VFS: Close: file count is 0\n");
1102 		return 0;
1103 	}
1104 
1105 	if (filp->f_op && filp->f_op->flush)
1106 		retval = filp->f_op->flush(filp, id);
1107 
1108 	dnotify_flush(filp, id);
1109 	locks_remove_posix(filp, id);
1110 	fput(filp);
1111 	return retval;
1112 }
1113 
1114 EXPORT_SYMBOL(filp_close);
1115 
1116 /*
1117  * Careful here! We test whether the file pointer is NULL before
1118  * releasing the fd. This ensures that one clone task can't release
1119  * an fd while another clone is opening it.
1120  */
1121 asmlinkage long sys_close(unsigned int fd)
1122 {
1123 	struct file * filp;
1124 	struct files_struct *files = current->files;
1125 	struct fdtable *fdt;
1126 	int retval;
1127 
1128 	spin_lock(&files->file_lock);
1129 	fdt = files_fdtable(files);
1130 	if (fd >= fdt->max_fds)
1131 		goto out_unlock;
1132 	filp = fdt->fd[fd];
1133 	if (!filp)
1134 		goto out_unlock;
1135 	rcu_assign_pointer(fdt->fd[fd], NULL);
1136 	FD_CLR(fd, fdt->close_on_exec);
1137 	__put_unused_fd(files, fd);
1138 	spin_unlock(&files->file_lock);
1139 	retval = filp_close(filp, files);
1140 
1141 	/* can't restart close syscall because file table entry was cleared */
1142 	if (unlikely(retval == -ERESTARTSYS ||
1143 		     retval == -ERESTARTNOINTR ||
1144 		     retval == -ERESTARTNOHAND ||
1145 		     retval == -ERESTART_RESTARTBLOCK))
1146 		retval = -EINTR;
1147 
1148 	return retval;
1149 
1150 out_unlock:
1151 	spin_unlock(&files->file_lock);
1152 	return -EBADF;
1153 }
1154 
1155 EXPORT_SYMBOL(sys_close);
1156 
1157 /*
1158  * This routine simulates a hangup on the tty, to arrange that users
1159  * are given clean terminals at login time.
1160  */
1161 asmlinkage long sys_vhangup(void)
1162 {
1163 	if (capable(CAP_SYS_TTY_CONFIG)) {
1164 		/* XXX: this needs locking */
1165 		tty_vhangup(current->signal->tty);
1166 		return 0;
1167 	}
1168 	return -EPERM;
1169 }
1170 
1171 /*
1172  * Called when an inode is about to be open.
1173  * We use this to disallow opening large files on 32bit systems if
1174  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1175  * on this flag in sys_open.
1176  */
1177 int generic_file_open(struct inode * inode, struct file * filp)
1178 {
1179 	if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1180 		return -EFBIG;
1181 	return 0;
1182 }
1183 
1184 EXPORT_SYMBOL(generic_file_open);
1185 
1186 /*
1187  * This is used by subsystems that don't want seekable
1188  * file descriptors
1189  */
1190 int nonseekable_open(struct inode *inode, struct file *filp)
1191 {
1192 	filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1193 	return 0;
1194 }
1195 
1196 EXPORT_SYMBOL(nonseekable_open);
1197