4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/string.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>
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>
35 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
39 struct iattr newattrs;
41 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
45 newattrs.ia_size = length;
46 newattrs.ia_valid = ATTR_SIZE | time_attrs;
48 newattrs.ia_file = filp;
49 newattrs.ia_valid |= ATTR_FILE;
52 /* Remove suid/sgid on truncate too */
53 ret = should_remove_suid(dentry);
55 newattrs.ia_valid |= ret | ATTR_FORCE;
57 mutex_lock(&dentry->d_inode->i_mutex);
58 ret = notify_change(dentry, &newattrs);
59 mutex_unlock(&dentry->d_inode->i_mutex);
63 static long do_sys_truncate(const char __user *pathname, loff_t length)
70 if (length < 0) /* sorry, but loff_t says... */
73 error = user_path(pathname, &path);
76 inode = path.dentry->d_inode;
78 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 if (S_ISDIR(inode->i_mode))
84 if (!S_ISREG(inode->i_mode))
87 error = mnt_want_write(path.mnt);
91 error = inode_permission(inode, MAY_WRITE);
93 goto mnt_drop_write_and_out;
97 goto mnt_drop_write_and_out;
99 error = get_write_access(inode);
101 goto mnt_drop_write_and_out;
104 * Make sure that there are no leases. get_write_access() protects
105 * against the truncate racing with a lease-granting setlease().
107 error = break_lease(inode, O_WRONLY);
109 goto put_write_and_out;
111 error = locks_verify_truncate(inode, NULL, length);
113 error = security_path_truncate(&path);
115 error = do_truncate(path.dentry, length, 0, NULL);
118 put_write_access(inode);
119 mnt_drop_write_and_out:
120 mnt_drop_write(path.mnt);
127 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
129 return do_sys_truncate(path, length);
132 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
134 struct inode * inode;
135 struct dentry *dentry;
147 /* explicitly opened as large or we are on 64-bit box */
148 if (file->f_flags & O_LARGEFILE)
151 dentry = file->f_path.dentry;
152 inode = dentry->d_inode;
154 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
158 /* Cannot ftruncate over 2^31 bytes without large file support */
159 if (small && length > MAX_NON_LFS)
163 if (IS_APPEND(inode))
166 error = locks_verify_truncate(inode, file, length);
168 error = security_path_truncate(&file->f_path);
170 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
177 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
179 long ret = do_sys_ftruncate(fd, length, 1);
180 /* avoid REGPARM breakage on x86: */
181 asmlinkage_protect(2, ret, fd, length);
185 /* LFS versions of truncate are only needed on 32 bit machines */
186 #if BITS_PER_LONG == 32
187 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
189 return do_sys_truncate(path, length);
191 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
192 asmlinkage long SyS_truncate64(long path, loff_t length)
194 return SYSC_truncate64((const char __user *) path, length);
196 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
199 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
201 long ret = do_sys_ftruncate(fd, length, 0);
202 /* avoid REGPARM breakage on x86: */
203 asmlinkage_protect(2, ret, fd, length);
206 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
207 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
209 return SYSC_ftruncate64((unsigned int) fd, length);
211 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
213 #endif /* BITS_PER_LONG == 32 */
216 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
218 struct inode *inode = file->f_path.dentry->d_inode;
221 if (offset < 0 || len <= 0)
224 /* Return error if mode is not supported */
225 if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
228 if (!(file->f_mode & FMODE_WRITE))
231 * Revalidate the write permissions, in case security policy has
232 * changed since the files were opened.
234 ret = security_file_permission(file, MAY_WRITE);
238 if (S_ISFIFO(inode->i_mode))
242 * Let individual file system decide if it supports preallocation
243 * for directories or not.
245 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
248 /* Check for wrap through zero too */
249 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
252 if (!inode->i_op->fallocate)
255 return inode->i_op->fallocate(inode, mode, offset, len);
258 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
265 error = do_fallocate(file, mode, offset, len);
272 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
273 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
275 return SYSC_fallocate((int)fd, (int)mode, offset, len);
277 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
281 * access() needs to use the real uid/gid, not the effective uid/gid.
282 * We do this by temporarily clearing all FS-related capabilities and
283 * switching the fsuid/fsgid around to the real ones.
285 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
287 const struct cred *old_cred;
288 struct cred *override_cred;
293 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
296 override_cred = prepare_creds();
300 override_cred->fsuid = override_cred->uid;
301 override_cred->fsgid = override_cred->gid;
303 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
304 /* Clear the capabilities if we switch to a non-root user */
305 if (override_cred->uid)
306 cap_clear(override_cred->cap_effective);
308 override_cred->cap_effective =
309 override_cred->cap_permitted;
312 old_cred = override_creds(override_cred);
314 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
318 inode = path.dentry->d_inode;
320 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
322 * MAY_EXEC on regular files is denied if the fs is mounted
323 * with the "noexec" flag.
326 if (path.mnt->mnt_flags & MNT_NOEXEC)
327 goto out_path_release;
330 res = inode_permission(inode, mode | MAY_ACCESS);
331 /* SuS v2 requires we report a read only fs too */
332 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
333 goto out_path_release;
335 * This is a rare case where using __mnt_is_readonly()
336 * is OK without a mnt_want/drop_write() pair. Since
337 * no actual write to the fs is performed here, we do
338 * not need to telegraph to that to anyone.
340 * By doing this, we accept that this access is
341 * inherently racy and know that the fs may change
342 * state before we even see this result.
344 if (__mnt_is_readonly(path.mnt))
350 revert_creds(old_cred);
351 put_cred(override_cred);
355 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
357 return sys_faccessat(AT_FDCWD, filename, mode);
360 SYSCALL_DEFINE1(chdir, const char __user *, filename)
365 error = user_path_dir(filename, &path);
369 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
373 set_fs_pwd(current->fs, &path);
381 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
392 inode = file->f_path.dentry->d_inode;
395 if (!S_ISDIR(inode->i_mode))
398 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
400 set_fs_pwd(current->fs, &file->f_path);
407 SYSCALL_DEFINE1(chroot, const char __user *, filename)
412 error = user_path_dir(filename, &path);
416 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
421 if (!capable(CAP_SYS_CHROOT))
423 error = security_path_chroot(&path);
427 set_fs_root(current->fs, &path);
435 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
437 struct inode * inode;
438 struct dentry * dentry;
441 struct iattr newattrs;
447 dentry = file->f_path.dentry;
448 inode = dentry->d_inode;
450 audit_inode(NULL, dentry);
452 err = mnt_want_write_file(file);
455 mutex_lock(&inode->i_mutex);
456 err = security_path_chmod(dentry, file->f_vfsmnt, mode);
459 if (mode == (mode_t) -1)
460 mode = inode->i_mode;
461 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
462 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
463 err = notify_change(dentry, &newattrs);
465 mutex_unlock(&inode->i_mutex);
466 mnt_drop_write(file->f_path.mnt);
473 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
478 struct iattr newattrs;
480 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
483 inode = path.dentry->d_inode;
485 error = mnt_want_write(path.mnt);
488 mutex_lock(&inode->i_mutex);
489 error = security_path_chmod(path.dentry, path.mnt, mode);
492 if (mode == (mode_t) -1)
493 mode = inode->i_mode;
494 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
495 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
496 error = notify_change(path.dentry, &newattrs);
498 mutex_unlock(&inode->i_mutex);
499 mnt_drop_write(path.mnt);
506 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
508 return sys_fchmodat(AT_FDCWD, filename, mode);
511 static int chown_common(struct path *path, uid_t user, gid_t group)
513 struct inode *inode = path->dentry->d_inode;
515 struct iattr newattrs;
517 newattrs.ia_valid = ATTR_CTIME;
518 if (user != (uid_t) -1) {
519 newattrs.ia_valid |= ATTR_UID;
520 newattrs.ia_uid = user;
522 if (group != (gid_t) -1) {
523 newattrs.ia_valid |= ATTR_GID;
524 newattrs.ia_gid = group;
526 if (!S_ISDIR(inode->i_mode))
528 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
529 mutex_lock(&inode->i_mutex);
530 error = security_path_chown(path, user, group);
532 error = notify_change(path->dentry, &newattrs);
533 mutex_unlock(&inode->i_mutex);
538 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
543 error = user_path(filename, &path);
546 error = mnt_want_write(path.mnt);
549 error = chown_common(&path, user, group);
550 mnt_drop_write(path.mnt);
557 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
558 gid_t, group, int, flag)
564 if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
567 follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
568 error = user_path_at(dfd, filename, follow, &path);
571 error = mnt_want_write(path.mnt);
574 error = chown_common(&path, user, group);
575 mnt_drop_write(path.mnt);
582 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
587 error = user_lpath(filename, &path);
590 error = mnt_want_write(path.mnt);
593 error = chown_common(&path, user, group);
594 mnt_drop_write(path.mnt);
601 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
605 struct dentry * dentry;
611 error = mnt_want_write_file(file);
614 dentry = file->f_path.dentry;
615 audit_inode(NULL, dentry);
616 error = chown_common(&file->f_path, user, group);
617 mnt_drop_write(file->f_path.mnt);
625 * You have to be very careful that these write
626 * counts get cleaned up in error cases and
627 * upon __fput(). This should probably never
628 * be called outside of __dentry_open().
630 static inline int __get_file_write_access(struct inode *inode,
631 struct vfsmount *mnt)
634 error = get_write_access(inode);
638 * Do not take mount writer counts on
639 * special files since no writes to
640 * the mount itself will occur.
642 if (!special_file(inode->i_mode)) {
644 * Balanced in __fput()
646 error = mnt_want_write(mnt);
648 put_write_access(inode);
653 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
655 int (*open)(struct inode *, struct file *),
656 const struct cred *cred)
661 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
662 FMODE_PREAD | FMODE_PWRITE;
663 inode = dentry->d_inode;
664 if (f->f_mode & FMODE_WRITE) {
665 error = __get_file_write_access(inode, mnt);
668 if (!special_file(inode->i_mode))
672 f->f_mapping = inode->i_mapping;
673 f->f_path.dentry = dentry;
676 f->f_op = fops_get(inode->i_fop);
677 file_move(f, &inode->i_sb->s_files);
679 error = security_dentry_open(f, cred);
683 if (!open && f->f_op)
684 open = f->f_op->open;
686 error = open(inode, f);
692 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
694 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
696 /* NB: we're sure to have correct a_ops only after f_op->open */
697 if (f->f_flags & O_DIRECT) {
698 if (!f->f_mapping->a_ops ||
699 ((!f->f_mapping->a_ops->direct_IO) &&
700 (!f->f_mapping->a_ops->get_xip_mem))) {
702 f = ERR_PTR(-EINVAL);
710 if (f->f_mode & FMODE_WRITE) {
711 put_write_access(inode);
712 if (!special_file(inode->i_mode)) {
714 * We don't consider this a real
715 * mnt_want/drop_write() pair
716 * because it all happenend right
717 * here, so just reset the state.
724 f->f_path.dentry = NULL;
725 f->f_path.mnt = NULL;
730 return ERR_PTR(error);
734 * lookup_instantiate_filp - instantiates the open intent filp
735 * @nd: pointer to nameidata
736 * @dentry: pointer to dentry
737 * @open: open callback
739 * Helper for filesystems that want to use lookup open intents and pass back
740 * a fully instantiated struct file to the caller.
741 * This function is meant to be called from within a filesystem's
743 * Beware of calling it for non-regular files! Those ->open methods might block
744 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
745 * leading to a deadlock, as nobody can open that fifo anymore, because
746 * another process to open fifo will block on locked parent when doing lookup).
747 * Note that in case of error, nd->intent.open.file is destroyed, but the
748 * path information remains valid.
749 * If the open callback is set to NULL, then the standard f_op->open()
750 * filesystem callback is substituted.
752 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
753 int (*open)(struct inode *, struct file *))
755 const struct cred *cred = current_cred();
757 if (IS_ERR(nd->intent.open.file))
761 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
762 nd->intent.open.file,
765 return nd->intent.open.file;
767 release_open_intent(nd);
768 nd->intent.open.file = (struct file *)dentry;
771 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
774 * nameidata_to_filp - convert a nameidata to an open filp.
775 * @nd: pointer to nameidata
778 * Note that this function destroys the original nameidata
780 struct file *nameidata_to_filp(struct nameidata *nd)
782 const struct cred *cred = current_cred();
785 /* Pick up the filp from the open intent */
786 filp = nd->intent.open.file;
787 /* Has the filesystem initialised the file for us? */
788 if (filp->f_path.dentry == NULL)
789 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
797 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
800 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
801 const struct cred *cred)
806 validate_creds(cred);
809 * We must always pass in a valid mount pointer. Historically
810 * callers got away with not passing it, but we must enforce this at
811 * the earliest possible point now to avoid strange problems deep in the
815 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
817 return ERR_PTR(-EINVAL);
821 f = get_empty_filp();
825 return ERR_PTR(error);
829 return __dentry_open(dentry, mnt, f, NULL, cred);
831 EXPORT_SYMBOL(dentry_open);
833 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
835 struct fdtable *fdt = files_fdtable(files);
836 __FD_CLR(fd, fdt->open_fds);
837 if (fd < files->next_fd)
841 void put_unused_fd(unsigned int fd)
843 struct files_struct *files = current->files;
844 spin_lock(&files->file_lock);
845 __put_unused_fd(files, fd);
846 spin_unlock(&files->file_lock);
849 EXPORT_SYMBOL(put_unused_fd);
852 * Install a file pointer in the fd array.
854 * The VFS is full of places where we drop the files lock between
855 * setting the open_fds bitmap and installing the file in the file
856 * array. At any such point, we are vulnerable to a dup2() race
857 * installing a file in the array before us. We need to detect this and
858 * fput() the struct file we are about to overwrite in this case.
860 * It should never happen - if we allow dup2() do it, _really_ bad things
864 void fd_install(unsigned int fd, struct file *file)
866 struct files_struct *files = current->files;
868 spin_lock(&files->file_lock);
869 fdt = files_fdtable(files);
870 BUG_ON(fdt->fd[fd] != NULL);
871 rcu_assign_pointer(fdt->fd[fd], file);
872 spin_unlock(&files->file_lock);
875 EXPORT_SYMBOL(fd_install);
877 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
879 char *tmp = getname(filename);
880 int fd = PTR_ERR(tmp);
883 fd = get_unused_fd_flags(flags);
885 struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
890 fsnotify_open(f->f_path.dentry);
899 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
903 if (force_o_largefile())
904 flags |= O_LARGEFILE;
906 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
907 /* avoid REGPARM breakage on x86: */
908 asmlinkage_protect(3, ret, filename, flags, mode);
912 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
917 if (force_o_largefile())
918 flags |= O_LARGEFILE;
920 ret = do_sys_open(dfd, filename, flags, mode);
921 /* avoid REGPARM breakage on x86: */
922 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
929 * For backward compatibility? Maybe this should be moved
930 * into arch/i386 instead?
932 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
934 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
940 * "id" is the POSIX thread ID. We use the
941 * files pointer for this..
943 int filp_close(struct file *filp, fl_owner_t id)
947 if (!file_count(filp)) {
948 printk(KERN_ERR "VFS: Close: file count is 0\n");
952 if (filp->f_op && filp->f_op->flush)
953 retval = filp->f_op->flush(filp, id);
955 dnotify_flush(filp, id);
956 locks_remove_posix(filp, id);
961 EXPORT_SYMBOL(filp_close);
964 * Careful here! We test whether the file pointer is NULL before
965 * releasing the fd. This ensures that one clone task can't release
966 * an fd while another clone is opening it.
968 SYSCALL_DEFINE1(close, unsigned int, fd)
971 struct files_struct *files = current->files;
975 spin_lock(&files->file_lock);
976 fdt = files_fdtable(files);
977 if (fd >= fdt->max_fds)
982 rcu_assign_pointer(fdt->fd[fd], NULL);
983 FD_CLR(fd, fdt->close_on_exec);
984 __put_unused_fd(files, fd);
985 spin_unlock(&files->file_lock);
986 retval = filp_close(filp, files);
988 /* can't restart close syscall because file table entry was cleared */
989 if (unlikely(retval == -ERESTARTSYS ||
990 retval == -ERESTARTNOINTR ||
991 retval == -ERESTARTNOHAND ||
992 retval == -ERESTART_RESTARTBLOCK))
998 spin_unlock(&files->file_lock);
1001 EXPORT_SYMBOL(sys_close);
1004 * This routine simulates a hangup on the tty, to arrange that users
1005 * are given clean terminals at login time.
1007 SYSCALL_DEFINE0(vhangup)
1009 if (capable(CAP_SYS_TTY_CONFIG)) {
1017 * Called when an inode is about to be open.
1018 * We use this to disallow opening large files on 32bit systems if
1019 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1020 * on this flag in sys_open.
1022 int generic_file_open(struct inode * inode, struct file * filp)
1024 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1029 EXPORT_SYMBOL(generic_file_open);
1032 * This is used by subsystems that don't want seekable
1035 int nonseekable_open(struct inode *inode, struct file *filp)
1037 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1041 EXPORT_SYMBOL(nonseekable_open);