fchmodat2: add support for AT_EMPTY_PATH

This allows userspace to avoid going through /proc/self/fd when dealing
with all types of file descriptors for chmod(), and makes fchmodat2() a
proper superset of all other chmod syscalls.

The primary difference between fchmodat2(AT_EMPTY_PATH) and fchmod() is
that fchmod() doesn't operate on O_PATH file descriptors by design. To
quote open(2):

> O_PATH (since Linux 2.6.39)
> [...]
> The file itself is not opened, and other file operations (e.g.,
> read(2), write(2), fchmod(2), fchown(2), fgetxattr(2), ioctl(2),
> mmap(2)) fail with the error EBADF.

However, procfs has allowed userspace to do this operation ever since
the introduction of O_PATH through magic-links, so adding this feature
is only an improvement for programs that have to mess around with
/proc/self/fd/$n today to get this behaviour. In addition,
fchownat(AT_EMPTY_PATH) has existed since the introduction of O_PATH and
allows chown() operations directly on O_PATH descriptors.

Signed-off-by: Aleksa Sarai <[email protected]>
Acked-by: Alexey Gladkov <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>

Author: cyphar

fs/open.c

@@ -678,10 +678,12 @@ static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
 	int error;
 	unsigned int lookup_flags;
 
-	if (unlikely(flags & ~AT_SYMLINK_NOFOLLOW))
+	if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
 		return -EINVAL;
 
 	lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
+	if (flags & AT_EMPTY_PATH)
+		lookup_flags |= LOOKUP_EMPTY;
 
 retry:
 	error = user_path_at(dfd, filename, lookup_flags, &path);