container: optionally isolate host abstract UNIX domain sockets via landlock

The socket is typically owned by the priv-user, and inaccessible by the target user, so just allowing access to the directory is not enough. This change fixes this oversight and add checks that will also be useful for merging #1.

Signed-off-by: Ophestra <cat@gensokyo.uk>

# Conflicts:
#	test/sandbox/case/device.nix
#	test/sandbox/case/tty.nix

container/container.go

@@ -92,8 +92,6 @@ type (
 		RetainSession bool
 		// Do not [syscall.CLONE_NEWNET].
 		HostNet bool
-		// Scope abstract UNIX domain sockets using LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET.
-		ScopeAbstract bool
 		// Retain CAP_SYS_ADMIN.
 		Privileged bool
 	}

container/init.go

@@ -13,7 +13,6 @@ import (
 	. "syscall"
 	"time"
 
-	"hakurei.app/container/landlock"
 	"hakurei.app/container/seccomp"
 )
 
@@ -261,12 +260,6 @@ func Init(prepare func(prefix string), setVerbose func(verbose bool)) {
 		msg.Verbose("syscall filter not configured")
 	}
 
-	if params.ScopeAbstract {
-		if err := landlock.ScopeAbstract(); err != nil {
-			log.Fatalf("could not scope abstract unix sockets: %v", err)
-		}
-	}
-
 	extraFiles := make([]*os.File, params.Count)
 	for i := range extraFiles {
 		// setup fd is placed before all extra files

container/landlock/landlock-helper.c

@@ -1,14 +0,0 @@
-#include <errno.h>
-#include <linux/landlock.h>
-#include <sys/psx_syscall.h>
-#include <sys/syscall.h>
-
-#include "landlock-helper.h"
-
-int hakurei_scope_abstract_unix_sockets(int* p_errno, int fd) {
-  int res = psx_syscall3(SYS_landlock_restrict_self, fd, 0, 0);
-
-  *p_errno = errno;
-
-  return res;
-}

container/landlock/landlock-helper.h

@@ -1,3 +0,0 @@
-#pragma once
-
-int hakurei_scope_abstract_unix_sockets(int* p_errno, int fd);

container/landlock/landlock.go

@@ -1,12 +1,8 @@
 package landlock
 
 /*
-#cgo linux pkg-config: --static libpsx
-
 #include <linux/landlock.h>
 #include <sys/syscall.h>
-
-#include "landlock-helper.h"
 */
 import "C"
 
@@ -21,10 +17,13 @@ const (
 	LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET = C.LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET
 
 	SYS_LANDLOCK_CREATE_RULESET = C.SYS_landlock_create_ruleset
+	SYS_LANDLOCK_RESTRICT_SELF  = C.SYS_landlock_restrict_self
 )
 
 type LandlockRulesetAttr = C.struct_landlock_ruleset_attr
 
+// ScopeAbstract calls landlock_restrict_self and must be called from a goroutine wired to an m
+// with the process starting from the same goroutine.
 func ScopeAbstract() error {
 	abi, _, err := syscall.Syscall(SYS_LANDLOCK_CREATE_RULESET, 0, 0, LANDLOCK_CREATE_RULESET_VERSION)
 
@@ -48,10 +47,9 @@ func ScopeAbstract() error {
 
 	defer syscall.Close(int(fd))
 
-	var errno C.int
+	r, _, err := syscall.Syscall(SYS_LANDLOCK_RESTRICT_SELF, fd, 0, 0)
-	if rv := C.hakurei_scope_abstract_unix_sockets(&errno, C.int(fd)); rv != 0 {
+	if r != 0 {
-		return fmt.Errorf("could not restrict self via landlock: errno %v", errno)
+		return fmt.Errorf("could not restrict self via landlock: errno %v", err)
 	}
-
 	return nil
 }

internal/app/seal_linux.go

@@ -416,6 +416,22 @@ func (seal *outcome) finalise(ctx context.Context, sys sys.State, config *hst.Co
 			seal.sys.ChangeHosts("#" + seal.user.uid.String())
 			seal.env[display] = d
 			seal.container.Bind(socketDir, socketDir, 0)
+
+			// the socket file at `/tmp/.X11-unix/X%d` is typically owned by the priv user
+			// and not accessible by the target user
+			var socketPath *container.Absolute
+			if len(d) > 1 && d[0] == ':' { // `:%d`
+				if n, err := strconv.Atoi(d[1:]); err == nil && n >= 0 {
+					socketPath = socketDir.Append("X" + strconv.Itoa(n))
+				}
+			} else if len(d) > 5 && strings.HasPrefix(d, "unix:") { // `unix:%s`
+				if a, err := container.NewAbs(d[5:]); err == nil {
+					socketPath = a
+				}
+			}
+			if socketPath != nil {
+				seal.sys.UpdatePermTypeOptional(system.EX11, socketPath.String(), acl.Read, acl.Write, acl.Execute)
+			}
 		}
 	}
 

package.nix

@@ -5,7 +5,6 @@
   makeBinaryWrapper,
   xdg-dbus-proxy,
   pkg-config,
-  libcap,
   libffi,
   libseccomp,
   acl,
@@ -81,16 +80,10 @@ buildGoModule rec {
         hsu = "/run/wrappers/bin/hsu";
       };
 
-  env = {
-    # required by libpsx
-    CGO_LDFLAGS_ALLOW = "-Wl,(--no-whole-archive|--whole-archive)";
-
   # nix build environment does not allow acls
-    GO_TEST_SKIP_ACL = 1;
+  env.GO_TEST_SKIP_ACL = 1;
-  };
 
   buildInputs = [
-    libcap
     libffi
     libseccomp
     acl

system/acl.go

@@ -21,7 +21,17 @@ func (sys *I) UpdatePermType(et Enablement, path string, perms ...acl.Perm) *I {
 	sys.lock.Lock()
 	defer sys.lock.Unlock()
 
-	sys.ops = append(sys.ops, &ACL{et, path, perms})
+	sys.ops = append(sys.ops, &ACL{et, path, perms, false})
+
+	return sys
+}
+
+// UpdatePermTypeOptional appends an acl update Op that silently continues if the target does not exist.
+func (sys *I) UpdatePermTypeOptional(et Enablement, path string, perms ...acl.Perm) *I {
+	sys.lock.Lock()
+	defer sys.lock.Unlock()
+
+	sys.ops = append(sys.ops, &ACL{et, path, perms, true})
 
 	return sys
 }
@@ -30,14 +40,24 @@ type ACL struct {
 	et    Enablement
 	path  string
 	perms acl.Perms
+
+	// since revert operations are cross-process, the success of apply must not affect the outcome of revert
+	skipNotExist bool
 }
 
 func (a *ACL) Type() Enablement { return a.et }
 
 func (a *ACL) apply(sys *I) error {
 	msg.Verbose("applying ACL", a)
-	return wrapErrSuffix(acl.Update(a.path, sys.uid, a.perms...),
+	if err := acl.Update(a.path, sys.uid, a.perms...); err != nil {
+		if !a.skipNotExist || !os.IsNotExist(err) {
+			return wrapErrSuffix(err,
 				fmt.Sprintf("cannot apply ACL entry to %q:", a.path))
+		}
+		msg.Verbosef("path %q does not exist", a.path)
+		return nil
+	}
+	return nil
 }
 
 func (a *ACL) revert(sys *I, ec *Criteria) error {

system/acl_test.go

@@ -20,7 +20,7 @@ func TestUpdatePerm(t *testing.T) {
 		t.Run(tc.path+permSubTestSuffix(tc.perms), func(t *testing.T) {
 			sys := New(150)
 			sys.UpdatePerm(tc.path, tc.perms...)
-			(&tcOp{Process, tc.path}).test(t, sys.ops, []Op{&ACL{Process, tc.path, tc.perms}}, "UpdatePerm")
+			(&tcOp{Process, tc.path}).test(t, sys.ops, []Op{&ACL{Process, tc.path, tc.perms, false}}, "UpdatePerm")
 		})
 	}
 }
@@ -42,7 +42,7 @@ func TestUpdatePermType(t *testing.T) {
 		t.Run(tc.path+"_"+TypeString(tc.et)+permSubTestSuffix(tc.perms), func(t *testing.T) {
 			sys := New(150)
 			sys.UpdatePermType(tc.et, tc.path, tc.perms...)
-			tc.test(t, sys.ops, []Op{&ACL{tc.et, tc.path, tc.perms}}, "UpdatePermType")
+			tc.test(t, sys.ops, []Op{&ACL{tc.et, tc.path, tc.perms, false}}, "UpdatePermType")
 		})
 	}
 }

test/sandbox/case/device.nix

@@ -243,7 +243,7 @@ in
     seccomp = true;
 
     try_socket = "/tmp/.X11-unix/X0";
-    socket_abstract = true;
+    socket_abstract = false;
     socket_pathname = true;
   };
 }

test/sandbox/case/mapuid.nix

@@ -269,7 +269,7 @@ in
     seccomp = true;
 
     try_socket = "/tmp/.X11-unix/X0";
-    socket_abstract = true;
+    socket_abstract = false;
     socket_pathname = false;
   };
 }

test/sandbox/case/pdlike.nix

@@ -264,7 +264,7 @@ in
     seccomp = true;
 
     try_socket = "/tmp/.X11-unix/X0";
-    socket_abstract = true;
+    socket_abstract = false;
     socket_pathname = false;
   };
 }

test/sandbox/case/preset.nix

@@ -262,7 +262,7 @@ in
     seccomp = true;
 
     try_socket = "/tmp/.X11-unix/X0";
-    socket_abstract = true;
+    socket_abstract = false;
     socket_pathname = false;
   };
 }

test/sandbox/case/tty.nix

@@ -275,7 +275,7 @@ in
     seccomp = true;
 
     try_socket = "/tmp/.X11-unix/X0";
-    socket_abstract = true;
+    socket_abstract = false;
     socket_pathname = true;
   };
 }