Merge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
[linux-2.6.git] / fs / open.c
blob4b3e1edf2fe4d917e69e56b3e268e32f2ae7301a
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>
33 #include <linux/compat.h>
35 #include "internal.h"
37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
38 struct file *filp)
40 int ret;
41 struct iattr newattrs;
43 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
44 if (length < 0)
45 return -EINVAL;
47 newattrs.ia_size = length;
48 newattrs.ia_valid = ATTR_SIZE | time_attrs;
49 if (filp) {
50 newattrs.ia_file = filp;
51 newattrs.ia_valid |= ATTR_FILE;
54 /* Remove suid/sgid on truncate too */
55 ret = should_remove_suid(dentry);
56 if (ret)
57 newattrs.ia_valid |= ret | ATTR_FORCE;
59 mutex_lock(&dentry->d_inode->i_mutex);
60 /* Note any delegations or leases have already been broken: */
61 ret = notify_change(dentry, &newattrs, NULL);
62 mutex_unlock(&dentry->d_inode->i_mutex);
63 return ret;
66 long vfs_truncate(struct path *path, loff_t length)
68 struct inode *inode;
69 long error;
71 inode = path->dentry->d_inode;
73 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
74 if (S_ISDIR(inode->i_mode))
75 return -EISDIR;
76 if (!S_ISREG(inode->i_mode))
77 return -EINVAL;
79 error = mnt_want_write(path->mnt);
80 if (error)
81 goto out;
83 error = inode_permission(inode, MAY_WRITE);
84 if (error)
85 goto mnt_drop_write_and_out;
87 error = -EPERM;
88 if (IS_APPEND(inode))
89 goto mnt_drop_write_and_out;
91 error = get_write_access(inode);
92 if (error)
93 goto mnt_drop_write_and_out;
96 * Make sure that there are no leases. get_write_access() protects
97 * against the truncate racing with a lease-granting setlease().
99 error = break_lease(inode, O_WRONLY);
100 if (error)
101 goto put_write_and_out;
103 error = locks_verify_truncate(inode, NULL, length);
104 if (!error)
105 error = security_path_truncate(path);
106 if (!error)
107 error = do_truncate(path->dentry, length, 0, NULL);
109 put_write_and_out:
110 put_write_access(inode);
111 mnt_drop_write_and_out:
112 mnt_drop_write(path->mnt);
113 out:
114 return error;
116 EXPORT_SYMBOL_GPL(vfs_truncate);
118 static long do_sys_truncate(const char __user *pathname, loff_t length)
120 unsigned int lookup_flags = LOOKUP_FOLLOW;
121 struct path path;
122 int error;
124 if (length < 0) /* sorry, but loff_t says... */
125 return -EINVAL;
127 retry:
128 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
129 if (!error) {
130 error = vfs_truncate(&path, length);
131 path_put(&path);
133 if (retry_estale(error, lookup_flags)) {
134 lookup_flags |= LOOKUP_REVAL;
135 goto retry;
137 return error;
140 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
142 return do_sys_truncate(path, length);
145 #ifdef CONFIG_COMPAT
146 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
148 return do_sys_truncate(path, length);
150 #endif
152 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
154 struct inode *inode;
155 struct dentry *dentry;
156 struct fd f;
157 int error;
159 error = -EINVAL;
160 if (length < 0)
161 goto out;
162 error = -EBADF;
163 f = fdget(fd);
164 if (!f.file)
165 goto out;
167 /* explicitly opened as large or we are on 64-bit box */
168 if (f.file->f_flags & O_LARGEFILE)
169 small = 0;
171 dentry = f.file->f_path.dentry;
172 inode = dentry->d_inode;
173 error = -EINVAL;
174 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
175 goto out_putf;
177 error = -EINVAL;
178 /* Cannot ftruncate over 2^31 bytes without large file support */
179 if (small && length > MAX_NON_LFS)
180 goto out_putf;
182 error = -EPERM;
183 if (IS_APPEND(inode))
184 goto out_putf;
186 sb_start_write(inode->i_sb);
187 error = locks_verify_truncate(inode, f.file, length);
188 if (!error)
189 error = security_path_truncate(&f.file->f_path);
190 if (!error)
191 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
192 sb_end_write(inode->i_sb);
193 out_putf:
194 fdput(f);
195 out:
196 return error;
199 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
201 return do_sys_ftruncate(fd, length, 1);
204 #ifdef CONFIG_COMPAT
205 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
207 return do_sys_ftruncate(fd, length, 1);
209 #endif
211 /* LFS versions of truncate are only needed on 32 bit machines */
212 #if BITS_PER_LONG == 32
213 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
215 return do_sys_truncate(path, length);
218 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
220 return do_sys_ftruncate(fd, length, 0);
222 #endif /* BITS_PER_LONG == 32 */
225 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
227 struct inode *inode = file_inode(file);
228 long ret;
230 if (offset < 0 || len <= 0)
231 return -EINVAL;
233 /* Return error if mode is not supported */
234 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
235 return -EOPNOTSUPP;
237 /* Punch hole must have keep size set */
238 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
239 !(mode & FALLOC_FL_KEEP_SIZE))
240 return -EOPNOTSUPP;
242 if (!(file->f_mode & FMODE_WRITE))
243 return -EBADF;
245 /* It's not possible punch hole on append only file */
246 if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
247 return -EPERM;
249 if (IS_IMMUTABLE(inode))
250 return -EPERM;
253 * Revalidate the write permissions, in case security policy has
254 * changed since the files were opened.
256 ret = security_file_permission(file, MAY_WRITE);
257 if (ret)
258 return ret;
260 if (S_ISFIFO(inode->i_mode))
261 return -ESPIPE;
264 * Let individual file system decide if it supports preallocation
265 * for directories or not.
267 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
268 return -ENODEV;
270 /* Check for wrap through zero too */
271 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
272 return -EFBIG;
274 if (!file->f_op->fallocate)
275 return -EOPNOTSUPP;
277 sb_start_write(inode->i_sb);
278 ret = file->f_op->fallocate(file, mode, offset, len);
279 sb_end_write(inode->i_sb);
280 return ret;
283 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
285 struct fd f = fdget(fd);
286 int error = -EBADF;
288 if (f.file) {
289 error = do_fallocate(f.file, mode, offset, len);
290 fdput(f);
292 return error;
296 * access() needs to use the real uid/gid, not the effective uid/gid.
297 * We do this by temporarily clearing all FS-related capabilities and
298 * switching the fsuid/fsgid around to the real ones.
300 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
302 const struct cred *old_cred;
303 struct cred *override_cred;
304 struct path path;
305 struct inode *inode;
306 int res;
307 unsigned int lookup_flags = LOOKUP_FOLLOW;
309 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
310 return -EINVAL;
312 override_cred = prepare_creds();
313 if (!override_cred)
314 return -ENOMEM;
316 override_cred->fsuid = override_cred->uid;
317 override_cred->fsgid = override_cred->gid;
319 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
320 /* Clear the capabilities if we switch to a non-root user */
321 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
322 if (!uid_eq(override_cred->uid, root_uid))
323 cap_clear(override_cred->cap_effective);
324 else
325 override_cred->cap_effective =
326 override_cred->cap_permitted;
329 old_cred = override_creds(override_cred);
330 retry:
331 res = user_path_at(dfd, filename, lookup_flags, &path);
332 if (res)
333 goto out;
335 inode = path.dentry->d_inode;
337 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
339 * MAY_EXEC on regular files is denied if the fs is mounted
340 * with the "noexec" flag.
342 res = -EACCES;
343 if (path.mnt->mnt_flags & MNT_NOEXEC)
344 goto out_path_release;
347 res = inode_permission(inode, mode | MAY_ACCESS);
348 /* SuS v2 requires we report a read only fs too */
349 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
350 goto out_path_release;
352 * This is a rare case where using __mnt_is_readonly()
353 * is OK without a mnt_want/drop_write() pair. Since
354 * no actual write to the fs is performed here, we do
355 * not need to telegraph to that to anyone.
357 * By doing this, we accept that this access is
358 * inherently racy and know that the fs may change
359 * state before we even see this result.
361 if (__mnt_is_readonly(path.mnt))
362 res = -EROFS;
364 out_path_release:
365 path_put(&path);
366 if (retry_estale(res, lookup_flags)) {
367 lookup_flags |= LOOKUP_REVAL;
368 goto retry;
370 out:
371 revert_creds(old_cred);
372 put_cred(override_cred);
373 return res;
376 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
378 return sys_faccessat(AT_FDCWD, filename, mode);
381 SYSCALL_DEFINE1(chdir, const char __user *, filename)
383 struct path path;
384 int error;
385 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
386 retry:
387 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
388 if (error)
389 goto out;
391 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
392 if (error)
393 goto dput_and_out;
395 set_fs_pwd(current->fs, &path);
397 dput_and_out:
398 path_put(&path);
399 if (retry_estale(error, lookup_flags)) {
400 lookup_flags |= LOOKUP_REVAL;
401 goto retry;
403 out:
404 return error;
407 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
409 struct fd f = fdget_raw(fd);
410 struct inode *inode;
411 int error = -EBADF;
413 error = -EBADF;
414 if (!f.file)
415 goto out;
417 inode = file_inode(f.file);
419 error = -ENOTDIR;
420 if (!S_ISDIR(inode->i_mode))
421 goto out_putf;
423 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
424 if (!error)
425 set_fs_pwd(current->fs, &f.file->f_path);
426 out_putf:
427 fdput(f);
428 out:
429 return error;
432 SYSCALL_DEFINE1(chroot, const char __user *, filename)
434 struct path path;
435 int error;
436 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
437 retry:
438 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
439 if (error)
440 goto out;
442 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
443 if (error)
444 goto dput_and_out;
446 error = -EPERM;
447 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
448 goto dput_and_out;
449 error = security_path_chroot(&path);
450 if (error)
451 goto dput_and_out;
453 set_fs_root(current->fs, &path);
454 error = 0;
455 dput_and_out:
456 path_put(&path);
457 if (retry_estale(error, lookup_flags)) {
458 lookup_flags |= LOOKUP_REVAL;
459 goto retry;
461 out:
462 return error;
465 static int chmod_common(struct path *path, umode_t mode)
467 struct inode *inode = path->dentry->d_inode;
468 struct inode *delegated_inode = NULL;
469 struct iattr newattrs;
470 int error;
472 error = mnt_want_write(path->mnt);
473 if (error)
474 return error;
475 retry_deleg:
476 mutex_lock(&inode->i_mutex);
477 error = security_path_chmod(path, mode);
478 if (error)
479 goto out_unlock;
480 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
481 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
482 error = notify_change(path->dentry, &newattrs, &delegated_inode);
483 out_unlock:
484 mutex_unlock(&inode->i_mutex);
485 if (delegated_inode) {
486 error = break_deleg_wait(&delegated_inode);
487 if (!error)
488 goto retry_deleg;
490 mnt_drop_write(path->mnt);
491 return error;
494 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
496 struct fd f = fdget(fd);
497 int err = -EBADF;
499 if (f.file) {
500 audit_inode(NULL, f.file->f_path.dentry, 0);
501 err = chmod_common(&f.file->f_path, mode);
502 fdput(f);
504 return err;
507 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
509 struct path path;
510 int error;
511 unsigned int lookup_flags = LOOKUP_FOLLOW;
512 retry:
513 error = user_path_at(dfd, filename, lookup_flags, &path);
514 if (!error) {
515 error = chmod_common(&path, mode);
516 path_put(&path);
517 if (retry_estale(error, lookup_flags)) {
518 lookup_flags |= LOOKUP_REVAL;
519 goto retry;
522 return error;
525 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
527 return sys_fchmodat(AT_FDCWD, filename, mode);
530 static int chown_common(struct path *path, uid_t user, gid_t group)
532 struct inode *inode = path->dentry->d_inode;
533 struct inode *delegated_inode = NULL;
534 int error;
535 struct iattr newattrs;
536 kuid_t uid;
537 kgid_t gid;
539 uid = make_kuid(current_user_ns(), user);
540 gid = make_kgid(current_user_ns(), group);
542 newattrs.ia_valid = ATTR_CTIME;
543 if (user != (uid_t) -1) {
544 if (!uid_valid(uid))
545 return -EINVAL;
546 newattrs.ia_valid |= ATTR_UID;
547 newattrs.ia_uid = uid;
549 if (group != (gid_t) -1) {
550 if (!gid_valid(gid))
551 return -EINVAL;
552 newattrs.ia_valid |= ATTR_GID;
553 newattrs.ia_gid = gid;
555 if (!S_ISDIR(inode->i_mode))
556 newattrs.ia_valid |=
557 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
558 retry_deleg:
559 mutex_lock(&inode->i_mutex);
560 error = security_path_chown(path, uid, gid);
561 if (!error)
562 error = notify_change(path->dentry, &newattrs, &delegated_inode);
563 mutex_unlock(&inode->i_mutex);
564 if (delegated_inode) {
565 error = break_deleg_wait(&delegated_inode);
566 if (!error)
567 goto retry_deleg;
569 return error;
572 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
573 gid_t, group, int, flag)
575 struct path path;
576 int error = -EINVAL;
577 int lookup_flags;
579 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
580 goto out;
582 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
583 if (flag & AT_EMPTY_PATH)
584 lookup_flags |= LOOKUP_EMPTY;
585 retry:
586 error = user_path_at(dfd, filename, lookup_flags, &path);
587 if (error)
588 goto out;
589 error = mnt_want_write(path.mnt);
590 if (error)
591 goto out_release;
592 error = chown_common(&path, user, group);
593 mnt_drop_write(path.mnt);
594 out_release:
595 path_put(&path);
596 if (retry_estale(error, lookup_flags)) {
597 lookup_flags |= LOOKUP_REVAL;
598 goto retry;
600 out:
601 return error;
604 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
606 return sys_fchownat(AT_FDCWD, filename, user, group, 0);
609 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
611 return sys_fchownat(AT_FDCWD, filename, user, group,
612 AT_SYMLINK_NOFOLLOW);
615 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
617 struct fd f = fdget(fd);
618 int error = -EBADF;
620 if (!f.file)
621 goto out;
623 error = mnt_want_write_file(f.file);
624 if (error)
625 goto out_fput;
626 audit_inode(NULL, f.file->f_path.dentry, 0);
627 error = chown_common(&f.file->f_path, user, group);
628 mnt_drop_write_file(f.file);
629 out_fput:
630 fdput(f);
631 out:
632 return error;
636 * You have to be very careful that these write
637 * counts get cleaned up in error cases and
638 * upon __fput(). This should probably never
639 * be called outside of __dentry_open().
641 static inline int __get_file_write_access(struct inode *inode,
642 struct vfsmount *mnt)
644 int error;
645 error = get_write_access(inode);
646 if (error)
647 return error;
649 * Do not take mount writer counts on
650 * special files since no writes to
651 * the mount itself will occur.
653 if (!special_file(inode->i_mode)) {
655 * Balanced in __fput()
657 error = __mnt_want_write(mnt);
658 if (error)
659 put_write_access(inode);
661 return error;
664 int open_check_o_direct(struct file *f)
666 /* NB: we're sure to have correct a_ops only after f_op->open */
667 if (f->f_flags & O_DIRECT) {
668 if (!f->f_mapping->a_ops ||
669 ((!f->f_mapping->a_ops->direct_IO) &&
670 (!f->f_mapping->a_ops->get_xip_mem))) {
671 return -EINVAL;
674 return 0;
677 static int do_dentry_open(struct file *f,
678 int (*open)(struct inode *, struct file *),
679 const struct cred *cred)
681 static const struct file_operations empty_fops = {};
682 struct inode *inode;
683 int error;
685 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
686 FMODE_PREAD | FMODE_PWRITE;
688 if (unlikely(f->f_flags & O_PATH))
689 f->f_mode = FMODE_PATH;
691 path_get(&f->f_path);
692 inode = f->f_inode = f->f_path.dentry->d_inode;
693 if (f->f_mode & FMODE_WRITE) {
694 error = __get_file_write_access(inode, f->f_path.mnt);
695 if (error)
696 goto cleanup_file;
697 if (!special_file(inode->i_mode))
698 file_take_write(f);
701 f->f_mapping = inode->i_mapping;
703 if (unlikely(f->f_mode & FMODE_PATH)) {
704 f->f_op = &empty_fops;
705 return 0;
708 f->f_op = fops_get(inode->i_fop);
709 if (unlikely(WARN_ON(!f->f_op))) {
710 error = -ENODEV;
711 goto cleanup_all;
714 error = security_file_open(f, cred);
715 if (error)
716 goto cleanup_all;
718 error = break_lease(inode, f->f_flags);
719 if (error)
720 goto cleanup_all;
722 if (!open)
723 open = f->f_op->open;
724 if (open) {
725 error = open(inode, f);
726 if (error)
727 goto cleanup_all;
729 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
730 i_readcount_inc(inode);
732 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
734 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
736 return 0;
738 cleanup_all:
739 fops_put(f->f_op);
740 if (f->f_mode & FMODE_WRITE) {
741 put_write_access(inode);
742 if (!special_file(inode->i_mode)) {
744 * We don't consider this a real
745 * mnt_want/drop_write() pair
746 * because it all happenend right
747 * here, so just reset the state.
749 file_reset_write(f);
750 __mnt_drop_write(f->f_path.mnt);
753 cleanup_file:
754 path_put(&f->f_path);
755 f->f_path.mnt = NULL;
756 f->f_path.dentry = NULL;
757 f->f_inode = NULL;
758 return error;
762 * finish_open - finish opening a file
763 * @file: file pointer
764 * @dentry: pointer to dentry
765 * @open: open callback
766 * @opened: state of open
768 * This can be used to finish opening a file passed to i_op->atomic_open().
770 * If the open callback is set to NULL, then the standard f_op->open()
771 * filesystem callback is substituted.
773 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
774 * the return value of d_splice_alias(), then the caller needs to perform dput()
775 * on it after finish_open().
777 * On successful return @file is a fully instantiated open file. After this, if
778 * an error occurs in ->atomic_open(), it needs to clean up with fput().
780 * Returns zero on success or -errno if the open failed.
782 int finish_open(struct file *file, struct dentry *dentry,
783 int (*open)(struct inode *, struct file *),
784 int *opened)
786 int error;
787 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
789 file->f_path.dentry = dentry;
790 error = do_dentry_open(file, open, current_cred());
791 if (!error)
792 *opened |= FILE_OPENED;
794 return error;
796 EXPORT_SYMBOL(finish_open);
799 * finish_no_open - finish ->atomic_open() without opening the file
801 * @file: file pointer
802 * @dentry: dentry or NULL (as returned from ->lookup())
804 * This can be used to set the result of a successful lookup in ->atomic_open().
806 * NB: unlike finish_open() this function does consume the dentry reference and
807 * the caller need not dput() it.
809 * Returns "1" which must be the return value of ->atomic_open() after having
810 * called this function.
812 int finish_no_open(struct file *file, struct dentry *dentry)
814 file->f_path.dentry = dentry;
815 return 1;
817 EXPORT_SYMBOL(finish_no_open);
819 struct file *dentry_open(const struct path *path, int flags,
820 const struct cred *cred)
822 int error;
823 struct file *f;
825 validate_creds(cred);
827 /* We must always pass in a valid mount pointer. */
828 BUG_ON(!path->mnt);
830 f = get_empty_filp();
831 if (!IS_ERR(f)) {
832 f->f_flags = flags;
833 f->f_path = *path;
834 error = do_dentry_open(f, NULL, cred);
835 if (!error) {
836 /* from now on we need fput() to dispose of f */
837 error = open_check_o_direct(f);
838 if (error) {
839 fput(f);
840 f = ERR_PTR(error);
842 } else {
843 put_filp(f);
844 f = ERR_PTR(error);
847 return f;
849 EXPORT_SYMBOL(dentry_open);
851 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
853 int lookup_flags = 0;
854 int acc_mode;
856 if (flags & (O_CREAT | __O_TMPFILE))
857 op->mode = (mode & S_IALLUGO) | S_IFREG;
858 else
859 op->mode = 0;
861 /* Must never be set by userspace */
862 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
865 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
866 * check for O_DSYNC if the need any syncing at all we enforce it's
867 * always set instead of having to deal with possibly weird behaviour
868 * for malicious applications setting only __O_SYNC.
870 if (flags & __O_SYNC)
871 flags |= O_DSYNC;
873 if (flags & __O_TMPFILE) {
874 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
875 return -EINVAL;
876 acc_mode = MAY_OPEN | ACC_MODE(flags);
877 if (!(acc_mode & MAY_WRITE))
878 return -EINVAL;
879 } else if (flags & O_PATH) {
881 * If we have O_PATH in the open flag. Then we
882 * cannot have anything other than the below set of flags
884 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
885 acc_mode = 0;
886 } else {
887 acc_mode = MAY_OPEN | ACC_MODE(flags);
890 op->open_flag = flags;
892 /* O_TRUNC implies we need access checks for write permissions */
893 if (flags & O_TRUNC)
894 acc_mode |= MAY_WRITE;
896 /* Allow the LSM permission hook to distinguish append
897 access from general write access. */
898 if (flags & O_APPEND)
899 acc_mode |= MAY_APPEND;
901 op->acc_mode = acc_mode;
903 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
905 if (flags & O_CREAT) {
906 op->intent |= LOOKUP_CREATE;
907 if (flags & O_EXCL)
908 op->intent |= LOOKUP_EXCL;
911 if (flags & O_DIRECTORY)
912 lookup_flags |= LOOKUP_DIRECTORY;
913 if (!(flags & O_NOFOLLOW))
914 lookup_flags |= LOOKUP_FOLLOW;
915 op->lookup_flags = lookup_flags;
916 return 0;
920 * file_open_name - open file and return file pointer
922 * @name: struct filename containing path to open
923 * @flags: open flags as per the open(2) second argument
924 * @mode: mode for the new file if O_CREAT is set, else ignored
926 * This is the helper to open a file from kernelspace if you really
927 * have to. But in generally you should not do this, so please move
928 * along, nothing to see here..
930 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
932 struct open_flags op;
933 int err = build_open_flags(flags, mode, &op);
934 return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
938 * filp_open - open file and return file pointer
940 * @filename: path to open
941 * @flags: open flags as per the open(2) second argument
942 * @mode: mode for the new file if O_CREAT is set, else ignored
944 * This is the helper to open a file from kernelspace if you really
945 * have to. But in generally you should not do this, so please move
946 * along, nothing to see here..
948 struct file *filp_open(const char *filename, int flags, umode_t mode)
950 struct filename name = {.name = filename};
951 return file_open_name(&name, flags, mode);
953 EXPORT_SYMBOL(filp_open);
955 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
956 const char *filename, int flags)
958 struct open_flags op;
959 int err = build_open_flags(flags, 0, &op);
960 if (err)
961 return ERR_PTR(err);
962 if (flags & O_CREAT)
963 return ERR_PTR(-EINVAL);
964 if (!filename && (flags & O_DIRECTORY))
965 if (!dentry->d_inode->i_op->lookup)
966 return ERR_PTR(-ENOTDIR);
967 return do_file_open_root(dentry, mnt, filename, &op);
969 EXPORT_SYMBOL(file_open_root);
971 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
973 struct open_flags op;
974 int fd = build_open_flags(flags, mode, &op);
975 struct filename *tmp;
977 if (fd)
978 return fd;
980 tmp = getname(filename);
981 if (IS_ERR(tmp))
982 return PTR_ERR(tmp);
984 fd = get_unused_fd_flags(flags);
985 if (fd >= 0) {
986 struct file *f = do_filp_open(dfd, tmp, &op);
987 if (IS_ERR(f)) {
988 put_unused_fd(fd);
989 fd = PTR_ERR(f);
990 } else {
991 fsnotify_open(f);
992 fd_install(fd, f);
995 putname(tmp);
996 return fd;
999 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1001 if (force_o_largefile())
1002 flags |= O_LARGEFILE;
1004 return do_sys_open(AT_FDCWD, filename, flags, mode);
1007 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1008 umode_t, mode)
1010 if (force_o_largefile())
1011 flags |= O_LARGEFILE;
1013 return do_sys_open(dfd, filename, flags, mode);
1016 #ifndef __alpha__
1019 * For backward compatibility? Maybe this should be moved
1020 * into arch/i386 instead?
1022 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1024 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1027 #endif
1030 * "id" is the POSIX thread ID. We use the
1031 * files pointer for this..
1033 int filp_close(struct file *filp, fl_owner_t id)
1035 int retval = 0;
1037 if (!file_count(filp)) {
1038 printk(KERN_ERR "VFS: Close: file count is 0\n");
1039 return 0;
1042 if (filp->f_op->flush)
1043 retval = filp->f_op->flush(filp, id);
1045 if (likely(!(filp->f_mode & FMODE_PATH))) {
1046 dnotify_flush(filp, id);
1047 locks_remove_posix(filp, id);
1049 fput(filp);
1050 return retval;
1053 EXPORT_SYMBOL(filp_close);
1056 * Careful here! We test whether the file pointer is NULL before
1057 * releasing the fd. This ensures that one clone task can't release
1058 * an fd while another clone is opening it.
1060 SYSCALL_DEFINE1(close, unsigned int, fd)
1062 int retval = __close_fd(current->files, fd);
1064 /* can't restart close syscall because file table entry was cleared */
1065 if (unlikely(retval == -ERESTARTSYS ||
1066 retval == -ERESTARTNOINTR ||
1067 retval == -ERESTARTNOHAND ||
1068 retval == -ERESTART_RESTARTBLOCK))
1069 retval = -EINTR;
1071 return retval;
1073 EXPORT_SYMBOL(sys_close);
1076 * This routine simulates a hangup on the tty, to arrange that users
1077 * are given clean terminals at login time.
1079 SYSCALL_DEFINE0(vhangup)
1081 if (capable(CAP_SYS_TTY_CONFIG)) {
1082 tty_vhangup_self();
1083 return 0;
1085 return -EPERM;
1089 * Called when an inode is about to be open.
1090 * We use this to disallow opening large files on 32bit systems if
1091 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1092 * on this flag in sys_open.
1094 int generic_file_open(struct inode * inode, struct file * filp)
1096 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1097 return -EOVERFLOW;
1098 return 0;
1101 EXPORT_SYMBOL(generic_file_open);
1104 * This is used by subsystems that don't want seekable
1105 * file descriptors. The function is not supposed to ever fail, the only
1106 * reason it returns an 'int' and not 'void' is so that it can be plugged
1107 * directly into file_operations structure.
1109 int nonseekable_open(struct inode *inode, struct file *filp)
1111 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1112 return 0;
1115 EXPORT_SYMBOL(nonseekable_open);