fs: support FD_FAILFS_ROOT in fchroot()

Allow a process to move its root directory into failfs via
fchroot(FD_FAILFS_ROOT). From that point on every absolute path lookup
and every absolute symlink fails with EOPNOTSUPP. Combined with
fchdir(FD_FAILFS_ROOT) this leaves the process with lookups anchored
at explicit directory file descriptors only. It is the fs_struct
equivalent of RESOLVE_BENEATH. This allows taks to drop their filesystem
state completely.

Callers with CAP_SYS_CHROOT in their user namespace may always do
this, mirroring chroot(2). Unprivileged callers are subject to three
requirements (which may be loosened later):

(1) no_new_privs must be set

    After entering failfs suid binaries on regular mounts remain
    reachable via inherited directory file descriptors or the working
    directory. A setuid program executing with an unusable root
    directory might be tricked by this. I'm not 100% convinced that this
    is needed but it feels more secure initially and it also forces more
    no_new_privs on userspace. So win-win imo.

(2) The caller must not already be chrooted.

    The root directory is what confines .. resolution. The failfs root
    can never be reached by walking up a real mount tree. A task whose
    root is failfs has no .. barrier left below the top of its mount
    tree. A .. walk from any real directory fd it still holds climbs
    to the mount-namespace root. Which is kinda the point if you want to
    do fd-based lookup only. If failfs prevented you from doing that
    then it doesn't make a lot of sense.

    A task that a privileged manager chrooted into a subtree could use
    chroot()ing into failfs as a way to allow for an inherited fd to
    resolve it again.

    So reject already-chrooted callers closing that issue without losing
    anything for the intended self-sandboxing use case.

(3) The caller must not share its fs_struct.

    Requirement (1) is checked on the calling thread, but the root
    lives in the fs_struct which may be shared via CLONE_FS. A sibling
    thread without no_new_privs could then execute a setuid binary with
    the failfs root and defeat (1). setns() to a mount or user namespace
    refuses a shared fs_struct for the same kind of reason, so do the
    same here and require fs->users == 1. no_new_privs is inherited
    across clone() and can never be cleared, so any CLONE_FS child
    created afterwards carries it too and the guarantee holds.

Privileged callers (CAP_SYS_CHROOT) are not subject to these
requirements and may share the fs_struct. They can already chroot and
exec a setuid binary today, so failfs hands them nothing new.

Backing out is currently hard, but that is a property of the current
implementation and not a promise. current_chrooted() treats a failfs
root as chrooted so for now the task cannot create user namespaces to
regain CAP_SYS_CHROOT and chroot()/fchroot() back out require
CAP_SYS_CHROOT. This is not guaranteed though. current_chrooted() may
change, or an unprivileged no_new_privs task could be allowed to chroot
to a real directory, either of which would loosen this. So don't treat
it as a permanent one-way door.

The remaining way out today is setns() to a mount namespace file
descriptor which requires CAP_SYS_ADMIN over the target namespace plus
CAP_SYS_CHROOT and CAP_SYS_ADMIN in the caller's user namespace and
resets both root and working directory. A task that closes or never had
such file descriptors and restricts *chdir()/*chroot()/setns() via
seccomp currently cannot get back out.

Link: https://patch.msgid.link/20260724-work-failfs-v2-4-485dabbae185@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>

fs/open.c

@@ -620,31 +620,48 @@ dput_and_out:
 
 SYSCALL_DEFINE2(fchroot, int, fd, unsigned int, flags)
 {
+	struct path path;
 	int error;
 
 	if (flags)
 		return -EINVAL;
 
-	CLASS(fd_raw, f)(fd);
-	if (fd_empty(f))
-		return -EBADF;
+	if (fd == FD_FAILFS_ROOT) {
+		if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT)) {
+			if (!task_no_new_privs(current))
+				return -EPERM;
+			/* A shared fs_struct lets a sibling exec setuid past the check above. */
+			if (current->fs->users != 1)
+				return -EINVAL;
+			/* Moving the root to failfs lifts the old root's ".." barrier. */
+			if (current_chrooted())
+				return -EPERM;
+		}
+		failfs_get_root(&path);
+	} else {
+		CLASS(fd_raw, f)(fd);
+		if (fd_empty(f))
+			return -EBADF;
 
-	if (!d_can_lookup(fd_file(f)->f_path.dentry))
-		return -ENOTDIR;
+		if (!d_can_lookup(fd_file(f)->f_path.dentry))
+			return -ENOTDIR;
 
-	error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
-	if (error)
-		return error;
+		error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
+		if (error)
+			return error;
 
-	if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
-		return -EPERM;
+		if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
+			return -EPERM;
 
-	error = security_path_chroot(&fd_file(f)->f_path);
-	if (error)
-		return error;
+		path = fd_file(f)->f_path;
+		path_get(&path);
+	}
 
-	set_fs_root(current->fs, &fd_file(f)->f_path);
-	return 0;
+	error = security_path_chroot(&path);
+	if (!error)
+		set_fs_root(current->fs, &path);
+	path_put(&path);
+	return error;
 }
 
 int chmod_common(const struct path *path, umode_t mode)