Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cjb/mmc
[cris-mirror.git] / fs / open.c
blobb52cf013ffa145de32e5c641aba7a3625de8cc29
1 /*
2 * linux/fs/open.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
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>
34 #include "internal.h"
36 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
37 struct file *filp)
39 int ret;
40 struct iattr newattrs;
42 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
43 if (length < 0)
44 return -EINVAL;
46 newattrs.ia_size = length;
47 newattrs.ia_valid = ATTR_SIZE | time_attrs;
48 if (filp) {
49 newattrs.ia_file = filp;
50 newattrs.ia_valid |= ATTR_FILE;
53 /* Remove suid/sgid on truncate too */
54 ret = should_remove_suid(dentry);
55 if (ret)
56 newattrs.ia_valid |= ret | ATTR_FORCE;
58 mutex_lock(&dentry->d_inode->i_mutex);
59 ret = notify_change(dentry, &newattrs);
60 mutex_unlock(&dentry->d_inode->i_mutex);
61 return ret;
64 static long do_sys_truncate(const char __user *pathname, loff_t length)
66 struct path path;
67 struct inode *inode;
68 int error;
70 error = -EINVAL;
71 if (length < 0) /* sorry, but loff_t says... */
72 goto out;
74 error = user_path(pathname, &path);
75 if (error)
76 goto out;
77 inode = path.dentry->d_inode;
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 error = -EISDIR;
81 if (S_ISDIR(inode->i_mode))
82 goto dput_and_out;
84 error = -EINVAL;
85 if (!S_ISREG(inode->i_mode))
86 goto dput_and_out;
88 error = mnt_want_write(path.mnt);
89 if (error)
90 goto dput_and_out;
92 error = inode_permission(inode, MAY_WRITE);
93 if (error)
94 goto mnt_drop_write_and_out;
96 error = -EPERM;
97 if (IS_APPEND(inode))
98 goto mnt_drop_write_and_out;
100 error = get_write_access(inode);
101 if (error)
102 goto mnt_drop_write_and_out;
105 * Make sure that there are no leases. get_write_access() protects
106 * against the truncate racing with a lease-granting setlease().
108 error = break_lease(inode, O_WRONLY);
109 if (error)
110 goto put_write_and_out;
112 error = locks_verify_truncate(inode, NULL, length);
113 if (!error)
114 error = security_path_truncate(&path);
115 if (!error)
116 error = do_truncate(path.dentry, length, 0, NULL);
118 put_write_and_out:
119 put_write_access(inode);
120 mnt_drop_write_and_out:
121 mnt_drop_write(path.mnt);
122 dput_and_out:
123 path_put(&path);
124 out:
125 return error;
128 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
130 return do_sys_truncate(path, length);
133 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
135 struct inode * inode;
136 struct dentry *dentry;
137 struct file * file;
138 int error;
140 error = -EINVAL;
141 if (length < 0)
142 goto out;
143 error = -EBADF;
144 file = fget(fd);
145 if (!file)
146 goto out;
148 /* explicitly opened as large or we are on 64-bit box */
149 if (file->f_flags & O_LARGEFILE)
150 small = 0;
152 dentry = file->f_path.dentry;
153 inode = dentry->d_inode;
154 error = -EINVAL;
155 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
156 goto out_putf;
158 error = -EINVAL;
159 /* Cannot ftruncate over 2^31 bytes without large file support */
160 if (small && length > MAX_NON_LFS)
161 goto out_putf;
163 error = -EPERM;
164 if (IS_APPEND(inode))
165 goto out_putf;
167 error = locks_verify_truncate(inode, file, length);
168 if (!error)
169 error = security_path_truncate(&file->f_path);
170 if (!error)
171 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
172 out_putf:
173 fput(file);
174 out:
175 return error;
178 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
180 long ret = do_sys_ftruncate(fd, length, 1);
181 /* avoid REGPARM breakage on x86: */
182 asmlinkage_protect(2, ret, fd, length);
183 return ret;
186 /* LFS versions of truncate are only needed on 32 bit machines */
187 #if BITS_PER_LONG == 32
188 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
190 return do_sys_truncate(path, length);
192 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
193 asmlinkage long SyS_truncate64(long path, loff_t length)
195 return SYSC_truncate64((const char __user *) path, length);
197 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
198 #endif
200 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
202 long ret = do_sys_ftruncate(fd, length, 0);
203 /* avoid REGPARM breakage on x86: */
204 asmlinkage_protect(2, ret, fd, length);
205 return ret;
207 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
208 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
210 return SYSC_ftruncate64((unsigned int) fd, length);
212 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
213 #endif
214 #endif /* BITS_PER_LONG == 32 */
217 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
219 struct inode *inode = file->f_path.dentry->d_inode;
220 long ret;
222 if (offset < 0 || len <= 0)
223 return -EINVAL;
225 /* Return error if mode is not supported */
226 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
227 return -EOPNOTSUPP;
229 /* Punch hole must have keep size set */
230 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
231 !(mode & FALLOC_FL_KEEP_SIZE))
232 return -EOPNOTSUPP;
234 if (!(file->f_mode & FMODE_WRITE))
235 return -EBADF;
237 /* It's not possible punch hole on append only file */
238 if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
239 return -EPERM;
241 if (IS_IMMUTABLE(inode))
242 return -EPERM;
245 * Revalidate the write permissions, in case security policy has
246 * changed since the files were opened.
248 ret = security_file_permission(file, MAY_WRITE);
249 if (ret)
250 return ret;
252 if (S_ISFIFO(inode->i_mode))
253 return -ESPIPE;
256 * Let individual file system decide if it supports preallocation
257 * for directories or not.
259 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
260 return -ENODEV;
262 /* Check for wrap through zero too */
263 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
264 return -EFBIG;
266 if (!file->f_op->fallocate)
267 return -EOPNOTSUPP;
269 return file->f_op->fallocate(file, mode, offset, len);
272 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
274 struct file *file;
275 int error = -EBADF;
277 file = fget(fd);
278 if (file) {
279 error = do_fallocate(file, mode, offset, len);
280 fput(file);
283 return error;
286 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
287 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
289 return SYSC_fallocate((int)fd, (int)mode, offset, len);
291 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
292 #endif
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.
299 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
301 const struct cred *old_cred;
302 struct cred *override_cred;
303 struct path path;
304 struct inode *inode;
305 int res;
307 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
308 return -EINVAL;
310 override_cred = prepare_creds();
311 if (!override_cred)
312 return -ENOMEM;
314 override_cred->fsuid = override_cred->uid;
315 override_cred->fsgid = override_cred->gid;
317 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
318 /* Clear the capabilities if we switch to a non-root user */
319 if (override_cred->uid)
320 cap_clear(override_cred->cap_effective);
321 else
322 override_cred->cap_effective =
323 override_cred->cap_permitted;
326 old_cred = override_creds(override_cred);
328 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
329 if (res)
330 goto out;
332 inode = path.dentry->d_inode;
334 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
336 * MAY_EXEC on regular files is denied if the fs is mounted
337 * with the "noexec" flag.
339 res = -EACCES;
340 if (path.mnt->mnt_flags & MNT_NOEXEC)
341 goto out_path_release;
344 res = inode_permission(inode, mode | MAY_ACCESS);
345 /* SuS v2 requires we report a read only fs too */
346 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
347 goto out_path_release;
349 * This is a rare case where using __mnt_is_readonly()
350 * is OK without a mnt_want/drop_write() pair. Since
351 * no actual write to the fs is performed here, we do
352 * not need to telegraph to that to anyone.
354 * By doing this, we accept that this access is
355 * inherently racy and know that the fs may change
356 * state before we even see this result.
358 if (__mnt_is_readonly(path.mnt))
359 res = -EROFS;
361 out_path_release:
362 path_put(&path);
363 out:
364 revert_creds(old_cred);
365 put_cred(override_cred);
366 return res;
369 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
371 return sys_faccessat(AT_FDCWD, filename, mode);
374 SYSCALL_DEFINE1(chdir, const char __user *, filename)
376 struct path path;
377 int error;
379 error = user_path_dir(filename, &path);
380 if (error)
381 goto out;
383 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
384 if (error)
385 goto dput_and_out;
387 set_fs_pwd(current->fs, &path);
389 dput_and_out:
390 path_put(&path);
391 out:
392 return error;
395 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
397 struct file *file;
398 struct inode *inode;
399 int error;
401 error = -EBADF;
402 file = fget(fd);
403 if (!file)
404 goto out;
406 inode = file->f_path.dentry->d_inode;
408 error = -ENOTDIR;
409 if (!S_ISDIR(inode->i_mode))
410 goto out_putf;
412 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
413 if (!error)
414 set_fs_pwd(current->fs, &file->f_path);
415 out_putf:
416 fput(file);
417 out:
418 return error;
421 SYSCALL_DEFINE1(chroot, const char __user *, filename)
423 struct path path;
424 int error;
426 error = user_path_dir(filename, &path);
427 if (error)
428 goto out;
430 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
431 if (error)
432 goto dput_and_out;
434 error = -EPERM;
435 if (!capable(CAP_SYS_CHROOT))
436 goto dput_and_out;
437 error = security_path_chroot(&path);
438 if (error)
439 goto dput_and_out;
441 set_fs_root(current->fs, &path);
442 error = 0;
443 dput_and_out:
444 path_put(&path);
445 out:
446 return error;
449 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
451 struct inode * inode;
452 struct dentry * dentry;
453 struct file * file;
454 int err = -EBADF;
455 struct iattr newattrs;
457 file = fget(fd);
458 if (!file)
459 goto out;
461 dentry = file->f_path.dentry;
462 inode = dentry->d_inode;
464 audit_inode(NULL, dentry);
466 err = mnt_want_write_file(file);
467 if (err)
468 goto out_putf;
469 mutex_lock(&inode->i_mutex);
470 err = security_path_chmod(dentry, file->f_vfsmnt, mode);
471 if (err)
472 goto out_unlock;
473 if (mode == (mode_t) -1)
474 mode = inode->i_mode;
475 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
476 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
477 err = notify_change(dentry, &newattrs);
478 out_unlock:
479 mutex_unlock(&inode->i_mutex);
480 mnt_drop_write(file->f_path.mnt);
481 out_putf:
482 fput(file);
483 out:
484 return err;
487 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
489 struct path path;
490 struct inode *inode;
491 int error;
492 struct iattr newattrs;
494 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
495 if (error)
496 goto out;
497 inode = path.dentry->d_inode;
499 error = mnt_want_write(path.mnt);
500 if (error)
501 goto dput_and_out;
502 mutex_lock(&inode->i_mutex);
503 error = security_path_chmod(path.dentry, path.mnt, mode);
504 if (error)
505 goto out_unlock;
506 if (mode == (mode_t) -1)
507 mode = inode->i_mode;
508 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
509 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
510 error = notify_change(path.dentry, &newattrs);
511 out_unlock:
512 mutex_unlock(&inode->i_mutex);
513 mnt_drop_write(path.mnt);
514 dput_and_out:
515 path_put(&path);
516 out:
517 return error;
520 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
522 return sys_fchmodat(AT_FDCWD, filename, mode);
525 static int chown_common(struct path *path, uid_t user, gid_t group)
527 struct inode *inode = path->dentry->d_inode;
528 int error;
529 struct iattr newattrs;
531 newattrs.ia_valid = ATTR_CTIME;
532 if (user != (uid_t) -1) {
533 newattrs.ia_valid |= ATTR_UID;
534 newattrs.ia_uid = user;
536 if (group != (gid_t) -1) {
537 newattrs.ia_valid |= ATTR_GID;
538 newattrs.ia_gid = group;
540 if (!S_ISDIR(inode->i_mode))
541 newattrs.ia_valid |=
542 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
543 mutex_lock(&inode->i_mutex);
544 error = security_path_chown(path, user, group);
545 if (!error)
546 error = notify_change(path->dentry, &newattrs);
547 mutex_unlock(&inode->i_mutex);
549 return error;
552 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
554 struct path path;
555 int error;
557 error = user_path(filename, &path);
558 if (error)
559 goto out;
560 error = mnt_want_write(path.mnt);
561 if (error)
562 goto out_release;
563 error = chown_common(&path, user, group);
564 mnt_drop_write(path.mnt);
565 out_release:
566 path_put(&path);
567 out:
568 return error;
571 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
572 gid_t, group, int, flag)
574 struct path path;
575 int error = -EINVAL;
576 int lookup_flags;
578 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
579 goto out;
581 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
582 if (flag & AT_EMPTY_PATH)
583 lookup_flags |= LOOKUP_EMPTY;
584 error = user_path_at(dfd, filename, lookup_flags, &path);
585 if (error)
586 goto out;
587 error = mnt_want_write(path.mnt);
588 if (error)
589 goto out_release;
590 error = chown_common(&path, user, group);
591 mnt_drop_write(path.mnt);
592 out_release:
593 path_put(&path);
594 out:
595 return error;
598 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
600 struct path path;
601 int error;
603 error = user_lpath(filename, &path);
604 if (error)
605 goto out;
606 error = mnt_want_write(path.mnt);
607 if (error)
608 goto out_release;
609 error = chown_common(&path, user, group);
610 mnt_drop_write(path.mnt);
611 out_release:
612 path_put(&path);
613 out:
614 return error;
617 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
619 struct file * file;
620 int error = -EBADF;
621 struct dentry * dentry;
623 file = fget(fd);
624 if (!file)
625 goto out;
627 error = mnt_want_write_file(file);
628 if (error)
629 goto out_fput;
630 dentry = file->f_path.dentry;
631 audit_inode(NULL, dentry);
632 error = chown_common(&file->f_path, user, group);
633 mnt_drop_write(file->f_path.mnt);
634 out_fput:
635 fput(file);
636 out:
637 return error;
641 * You have to be very careful that these write
642 * counts get cleaned up in error cases and
643 * upon __fput(). This should probably never
644 * be called outside of __dentry_open().
646 static inline int __get_file_write_access(struct inode *inode,
647 struct vfsmount *mnt)
649 int error;
650 error = get_write_access(inode);
651 if (error)
652 return error;
654 * Do not take mount writer counts on
655 * special files since no writes to
656 * the mount itself will occur.
658 if (!special_file(inode->i_mode)) {
660 * Balanced in __fput()
662 error = mnt_want_write(mnt);
663 if (error)
664 put_write_access(inode);
666 return error;
669 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
670 struct file *f,
671 int (*open)(struct inode *, struct file *),
672 const struct cred *cred)
674 static const struct file_operations empty_fops = {};
675 struct inode *inode;
676 int error;
678 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
679 FMODE_PREAD | FMODE_PWRITE;
681 if (unlikely(f->f_flags & O_PATH))
682 f->f_mode = FMODE_PATH;
684 inode = dentry->d_inode;
685 if (f->f_mode & FMODE_WRITE) {
686 error = __get_file_write_access(inode, mnt);
687 if (error)
688 goto cleanup_file;
689 if (!special_file(inode->i_mode))
690 file_take_write(f);
693 f->f_mapping = inode->i_mapping;
694 f->f_path.dentry = dentry;
695 f->f_path.mnt = mnt;
696 f->f_pos = 0;
697 file_sb_list_add(f, inode->i_sb);
699 if (unlikely(f->f_mode & FMODE_PATH)) {
700 f->f_op = &empty_fops;
701 return f;
704 f->f_op = fops_get(inode->i_fop);
706 error = security_dentry_open(f, cred);
707 if (error)
708 goto cleanup_all;
710 if (!open && f->f_op)
711 open = f->f_op->open;
712 if (open) {
713 error = open(inode, f);
714 if (error)
715 goto cleanup_all;
717 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
718 i_readcount_inc(inode);
720 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
722 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
724 /* NB: we're sure to have correct a_ops only after f_op->open */
725 if (f->f_flags & O_DIRECT) {
726 if (!f->f_mapping->a_ops ||
727 ((!f->f_mapping->a_ops->direct_IO) &&
728 (!f->f_mapping->a_ops->get_xip_mem))) {
729 fput(f);
730 f = ERR_PTR(-EINVAL);
734 return f;
736 cleanup_all:
737 fops_put(f->f_op);
738 if (f->f_mode & FMODE_WRITE) {
739 put_write_access(inode);
740 if (!special_file(inode->i_mode)) {
742 * We don't consider this a real
743 * mnt_want/drop_write() pair
744 * because it all happenend right
745 * here, so just reset the state.
747 file_reset_write(f);
748 mnt_drop_write(mnt);
751 file_sb_list_del(f);
752 f->f_path.dentry = NULL;
753 f->f_path.mnt = NULL;
754 cleanup_file:
755 put_filp(f);
756 dput(dentry);
757 mntput(mnt);
758 return ERR_PTR(error);
762 * lookup_instantiate_filp - instantiates the open intent filp
763 * @nd: pointer to nameidata
764 * @dentry: pointer to dentry
765 * @open: open callback
767 * Helper for filesystems that want to use lookup open intents and pass back
768 * a fully instantiated struct file to the caller.
769 * This function is meant to be called from within a filesystem's
770 * lookup method.
771 * Beware of calling it for non-regular files! Those ->open methods might block
772 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
773 * leading to a deadlock, as nobody can open that fifo anymore, because
774 * another process to open fifo will block on locked parent when doing lookup).
775 * Note that in case of error, nd->intent.open.file is destroyed, but the
776 * path information remains valid.
777 * If the open callback is set to NULL, then the standard f_op->open()
778 * filesystem callback is substituted.
780 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
781 int (*open)(struct inode *, struct file *))
783 const struct cred *cred = current_cred();
785 if (IS_ERR(nd->intent.open.file))
786 goto out;
787 if (IS_ERR(dentry))
788 goto out_err;
789 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
790 nd->intent.open.file,
791 open, cred);
792 out:
793 return nd->intent.open.file;
794 out_err:
795 release_open_intent(nd);
796 nd->intent.open.file = (struct file *)dentry;
797 goto out;
799 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
802 * nameidata_to_filp - convert a nameidata to an open filp.
803 * @nd: pointer to nameidata
804 * @flags: open flags
806 * Note that this function destroys the original nameidata
808 struct file *nameidata_to_filp(struct nameidata *nd)
810 const struct cred *cred = current_cred();
811 struct file *filp;
813 /* Pick up the filp from the open intent */
814 filp = nd->intent.open.file;
815 nd->intent.open.file = NULL;
817 /* Has the filesystem initialised the file for us? */
818 if (filp->f_path.dentry == NULL) {
819 path_get(&nd->path);
820 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
821 NULL, cred);
823 return filp;
827 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
828 * error.
830 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
831 const struct cred *cred)
833 int error;
834 struct file *f;
836 validate_creds(cred);
838 /* We must always pass in a valid mount pointer. */
839 BUG_ON(!mnt);
841 error = -ENFILE;
842 f = get_empty_filp();
843 if (f == NULL) {
844 dput(dentry);
845 mntput(mnt);
846 return ERR_PTR(error);
849 f->f_flags = flags;
850 return __dentry_open(dentry, mnt, f, NULL, cred);
852 EXPORT_SYMBOL(dentry_open);
854 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
856 struct fdtable *fdt = files_fdtable(files);
857 __FD_CLR(fd, fdt->open_fds);
858 if (fd < files->next_fd)
859 files->next_fd = fd;
862 void put_unused_fd(unsigned int fd)
864 struct files_struct *files = current->files;
865 spin_lock(&files->file_lock);
866 __put_unused_fd(files, fd);
867 spin_unlock(&files->file_lock);
870 EXPORT_SYMBOL(put_unused_fd);
873 * Install a file pointer in the fd array.
875 * The VFS is full of places where we drop the files lock between
876 * setting the open_fds bitmap and installing the file in the file
877 * array. At any such point, we are vulnerable to a dup2() race
878 * installing a file in the array before us. We need to detect this and
879 * fput() the struct file we are about to overwrite in this case.
881 * It should never happen - if we allow dup2() do it, _really_ bad things
882 * will follow.
885 void fd_install(unsigned int fd, struct file *file)
887 struct files_struct *files = current->files;
888 struct fdtable *fdt;
889 spin_lock(&files->file_lock);
890 fdt = files_fdtable(files);
891 BUG_ON(fdt->fd[fd] != NULL);
892 rcu_assign_pointer(fdt->fd[fd], file);
893 spin_unlock(&files->file_lock);
896 EXPORT_SYMBOL(fd_install);
898 static inline int build_open_flags(int flags, int mode, struct open_flags *op)
900 int lookup_flags = 0;
901 int acc_mode;
903 if (!(flags & O_CREAT))
904 mode = 0;
905 op->mode = mode;
907 /* Must never be set by userspace */
908 flags &= ~FMODE_NONOTIFY;
911 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
912 * check for O_DSYNC if the need any syncing at all we enforce it's
913 * always set instead of having to deal with possibly weird behaviour
914 * for malicious applications setting only __O_SYNC.
916 if (flags & __O_SYNC)
917 flags |= O_DSYNC;
920 * If we have O_PATH in the open flag. Then we
921 * cannot have anything other than the below set of flags
923 if (flags & O_PATH) {
924 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
925 acc_mode = 0;
926 } else {
927 acc_mode = MAY_OPEN | ACC_MODE(flags);
930 op->open_flag = flags;
932 /* O_TRUNC implies we need access checks for write permissions */
933 if (flags & O_TRUNC)
934 acc_mode |= MAY_WRITE;
936 /* Allow the LSM permission hook to distinguish append
937 access from general write access. */
938 if (flags & O_APPEND)
939 acc_mode |= MAY_APPEND;
941 op->acc_mode = acc_mode;
943 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
945 if (flags & O_CREAT) {
946 op->intent |= LOOKUP_CREATE;
947 if (flags & O_EXCL)
948 op->intent |= LOOKUP_EXCL;
951 if (flags & O_DIRECTORY)
952 lookup_flags |= LOOKUP_DIRECTORY;
953 if (!(flags & O_NOFOLLOW))
954 lookup_flags |= LOOKUP_FOLLOW;
955 return lookup_flags;
959 * filp_open - open file and return file pointer
961 * @filename: path to open
962 * @flags: open flags as per the open(2) second argument
963 * @mode: mode for the new file if O_CREAT is set, else ignored
965 * This is the helper to open a file from kernelspace if you really
966 * have to. But in generally you should not do this, so please move
967 * along, nothing to see here..
969 struct file *filp_open(const char *filename, int flags, int mode)
971 struct open_flags op;
972 int lookup = build_open_flags(flags, mode, &op);
973 return do_filp_open(AT_FDCWD, filename, &op, lookup);
975 EXPORT_SYMBOL(filp_open);
977 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
978 const char *filename, int flags)
980 struct open_flags op;
981 int lookup = build_open_flags(flags, 0, &op);
982 if (flags & O_CREAT)
983 return ERR_PTR(-EINVAL);
984 if (!filename && (flags & O_DIRECTORY))
985 if (!dentry->d_inode->i_op->lookup)
986 return ERR_PTR(-ENOTDIR);
987 return do_file_open_root(dentry, mnt, filename, &op, lookup);
989 EXPORT_SYMBOL(file_open_root);
991 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
993 struct open_flags op;
994 int lookup = build_open_flags(flags, mode, &op);
995 char *tmp = getname(filename);
996 int fd = PTR_ERR(tmp);
998 if (!IS_ERR(tmp)) {
999 fd = get_unused_fd_flags(flags);
1000 if (fd >= 0) {
1001 struct file *f = do_filp_open(dfd, tmp, &op, lookup);
1002 if (IS_ERR(f)) {
1003 put_unused_fd(fd);
1004 fd = PTR_ERR(f);
1005 } else {
1006 fsnotify_open(f);
1007 fd_install(fd, f);
1010 putname(tmp);
1012 return fd;
1015 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1017 long ret;
1019 if (force_o_largefile())
1020 flags |= O_LARGEFILE;
1022 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1023 /* avoid REGPARM breakage on x86: */
1024 asmlinkage_protect(3, ret, filename, flags, mode);
1025 return ret;
1028 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1029 int, mode)
1031 long ret;
1033 if (force_o_largefile())
1034 flags |= O_LARGEFILE;
1036 ret = do_sys_open(dfd, filename, flags, mode);
1037 /* avoid REGPARM breakage on x86: */
1038 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1039 return ret;
1042 #ifndef __alpha__
1045 * For backward compatibility? Maybe this should be moved
1046 * into arch/i386 instead?
1048 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1050 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1053 #endif
1056 * "id" is the POSIX thread ID. We use the
1057 * files pointer for this..
1059 int filp_close(struct file *filp, fl_owner_t id)
1061 int retval = 0;
1063 if (!file_count(filp)) {
1064 printk(KERN_ERR "VFS: Close: file count is 0\n");
1065 return 0;
1068 if (filp->f_op && filp->f_op->flush)
1069 retval = filp->f_op->flush(filp, id);
1071 if (likely(!(filp->f_mode & FMODE_PATH))) {
1072 dnotify_flush(filp, id);
1073 locks_remove_posix(filp, id);
1075 fput(filp);
1076 return retval;
1079 EXPORT_SYMBOL(filp_close);
1082 * Careful here! We test whether the file pointer is NULL before
1083 * releasing the fd. This ensures that one clone task can't release
1084 * an fd while another clone is opening it.
1086 SYSCALL_DEFINE1(close, unsigned int, fd)
1088 struct file * filp;
1089 struct files_struct *files = current->files;
1090 struct fdtable *fdt;
1091 int retval;
1093 spin_lock(&files->file_lock);
1094 fdt = files_fdtable(files);
1095 if (fd >= fdt->max_fds)
1096 goto out_unlock;
1097 filp = fdt->fd[fd];
1098 if (!filp)
1099 goto out_unlock;
1100 rcu_assign_pointer(fdt->fd[fd], NULL);
1101 FD_CLR(fd, fdt->close_on_exec);
1102 __put_unused_fd(files, fd);
1103 spin_unlock(&files->file_lock);
1104 retval = filp_close(filp, files);
1106 /* can't restart close syscall because file table entry was cleared */
1107 if (unlikely(retval == -ERESTARTSYS ||
1108 retval == -ERESTARTNOINTR ||
1109 retval == -ERESTARTNOHAND ||
1110 retval == -ERESTART_RESTARTBLOCK))
1111 retval = -EINTR;
1113 return retval;
1115 out_unlock:
1116 spin_unlock(&files->file_lock);
1117 return -EBADF;
1119 EXPORT_SYMBOL(sys_close);
1122 * This routine simulates a hangup on the tty, to arrange that users
1123 * are given clean terminals at login time.
1125 SYSCALL_DEFINE0(vhangup)
1127 if (capable(CAP_SYS_TTY_CONFIG)) {
1128 tty_vhangup_self();
1129 return 0;
1131 return -EPERM;
1135 * Called when an inode is about to be open.
1136 * We use this to disallow opening large files on 32bit systems if
1137 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1138 * on this flag in sys_open.
1140 int generic_file_open(struct inode * inode, struct file * filp)
1142 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1143 return -EOVERFLOW;
1144 return 0;
1147 EXPORT_SYMBOL(generic_file_open);
1150 * This is used by subsystems that don't want seekable
1151 * file descriptors. The function is not supposed to ever fail, the only
1152 * reason it returns an 'int' and not 'void' is so that it can be plugged
1153 * directly into file_operations structure.
1155 int nonseekable_open(struct inode *inode, struct file *filp)
1157 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1158 return 0;
1161 EXPORT_SYMBOL(nonseekable_open);