Merge pull request #12692 from dany74q:exec-seccomp-filters

PiperOrigin-RevId: 889399059

Author: gvisor-bot

pkg/sentry/control/BUILD

@@ -36,6 +36,7 @@ go_library(
     deps = [
         ":control_api_go_proto",
         "//pkg/abi/linux",
+        "//pkg/bpf",
         "//pkg/cleanup",
         "//pkg/context",
         "//pkg/eventchannel",

pkg/sentry/control/proc.go

@@ -25,6 +25,7 @@ import (
 	"time"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/bpf"
 	"gvisor.dev/gvisor/pkg/cleanup"
 	"gvisor.dev/gvisor/pkg/fd"
 	"gvisor.dev/gvisor/pkg/log"
@@ -152,6 +153,10 @@ type ExecArgs struct {
 
 	// Limits is the limit set for the process being executed.
 	Limits *limits.LimitSet
+
+	// SeccompProgram is an optional seccomp BPF program to install on the
+	// new process.
+	SeccompProgram *bpf.Program
 }
 
 // String prints the arguments as a string.
@@ -319,6 +324,13 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
 		return nil, 0, nil, err
 	}
 
+	if args.SeccompProgram != nil {
+		task := tg.Leader()
+		if err := task.AppendSyscallFilter(*args.SeccompProgram, true); err != nil {
+			return nil, 0, nil, fmt.Errorf("appending seccomp filters: %w", err)
+		}
+	}
+
 	// Start the newly created process.
 	proc.Kernel.StartProcess(tg)
 

runsc/boot/BUILD

@@ -26,6 +26,7 @@ go_library(
         "restore.go",
         "restore_impl.go",
         "seccheck.go",
+        "seccomp.go",
         "strace.go",
         "vfs.go",
     ],

runsc/boot/loader.go

@@ -28,7 +28,6 @@ import (
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/abi/linux"
-	"gvisor.dev/gvisor/pkg/bpf"
 	"gvisor.dev/gvisor/pkg/cleanup"
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/coverage"
@@ -84,7 +83,6 @@ import (
 	"gvisor.dev/gvisor/runsc/config"
 	"gvisor.dev/gvisor/runsc/profile"
 	"gvisor.dev/gvisor/runsc/specutils"
-	"gvisor.dev/gvisor/runsc/specutils/seccomp"
 
 	// Top-level inet providers.
 	"gvisor.dev/gvisor/pkg/sentry/socket/hostinet"
@@ -1361,27 +1359,15 @@ func (l *Loader) createContainerProcess(info *containerInfo) (*kernel.ThreadGrou
 	info.procArgs.FDTable.DecRef(ctx)
 
 	// Install seccomp filters with the new task if there are any.
-	if info.conf.OCISeccomp {
-		if info.spec.Linux != nil && info.spec.Linux.Seccomp != nil {
-			program, err := seccomp.BuildProgram(info.spec.Linux.Seccomp)
-			if err != nil {
-				return nil, nil, fmt.Errorf("building seccomp program: %w", err)
-			}
-
-			if log.IsLogging(log.Debug) {
-				out, _ := bpf.DecodeProgram(program)
-				log.Debugf("Installing OCI seccomp filters\nProgram:\n%s", out)
-			}
-
-			task := tg.Leader()
-			// NOTE: It seems Flags are ignored by runc so we ignore them too.
-			if err := task.AppendSyscallFilter(program, true); err != nil {
-				return nil, nil, fmt.Errorf("appending seccomp filters: %w", err)
-			}
-		}
-	} else {
-		if info.spec.Linux != nil && info.spec.Linux.Seccomp != nil {
-			log.Warningf("Seccomp spec is being ignored")
+	program, err := buildOCISeccompProgram(info.conf, info.spec)
+	if err != nil {
+		return nil, nil, err
+	}
+	if program != nil {
+		task := tg.Leader()
+		// NOTE: It seems Flags are ignored by runc so we ignore them too.
+		if err := task.AppendSyscallFilter(*program, true); err != nil {
+			return nil, nil, fmt.Errorf("appending seccomp filters: %w", err)
 		}
 	}
 
@@ -1524,6 +1510,16 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) {
 		return 0, fmt.Errorf("creating limits: %w", err)
 	}
 
+	containerName := l.k.ContainerName(args.ContainerID)
+	spec := l.containerSpecs[containerName]
+	if spec != nil {
+		seccompProgram, err := buildOCISeccompProgram(l.root.conf, spec)
+		if err != nil {
+			return 0, err
+		}
+		args.SeccompProgram = seccompProgram
+	}
+
 	// Start the process.
 	proc := control.Proc{Kernel: l.k}
 	newTG, tgid, ttyFile, err := control.ExecAsync(&proc, args)

runsc/boot/seccomp.go

@@ -0,0 +1,51 @@
+// Copyright 2026 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package boot
+
+import (
+	"fmt"
+
+	specs "github.com/opencontainers/runtime-spec/specs-go"
+
+	"gvisor.dev/gvisor/pkg/bpf"
+	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/runsc/config"
+	"gvisor.dev/gvisor/runsc/specutils/seccomp"
+)
+
+func buildOCISeccompProgram(conf *config.Config, spec *specs.Spec) (*bpf.Program, error) {
+	if !conf.OCISeccomp {
+		if spec.Linux != nil && spec.Linux.Seccomp != nil {
+			log.Warningf("Seccomp spec is being ignored because oci-seccomp is disabled")
+		}
+		return nil, nil
+	}
+
+	if spec.Linux == nil || spec.Linux.Seccomp == nil {
+		return nil, nil
+	}
+
+	program, err := seccomp.BuildProgram(spec.Linux.Seccomp)
+	if err != nil {
+		return nil, fmt.Errorf("building seccomp program: %w", err)
+	}
+
+	if log.IsLogging(log.Debug) {
+		out, _ := bpf.DecodeProgram(program)
+		log.Debugf("Installing OCI seccomp filters\nProgram:\n%s", out)
+	}
+
+	return &program, nil
+}

runsc/container/multi_container_test.go

@@ -3309,3 +3309,73 @@ func TestFSCheckpointCommand(t *testing.T) {
 		t.Errorf("Error waiting for FS restore: %v", err)
 	}
 }
+
+func TestMultiContainerExecSeccomp(t *testing.T) {
+	rootDir, cleanup, err := testutil.SetupRootDir()
+	if err != nil {
+		t.Fatalf("error creating root dir: %v", err)
+	}
+	defer cleanup()
+
+	conf := testutil.TestConfig(t)
+	conf.RootDir = rootDir
+	conf.OCISeccomp = true
+
+	testSpecs, ids := createSpecs(sleepCmd, sleepCmd)
+
+	// Container 0 (sandbox root): Block "uname" syscall.
+	testSpecs[0].Linux = &specs.Linux{
+		Seccomp: &specs.LinuxSeccomp{
+			DefaultAction: specs.ActAllow,
+			Syscalls: []specs.LinuxSyscall{
+				{
+					Names:  []string{"uname"},
+					Action: specs.ActErrno,
+				},
+			},
+		},
+	}
+
+	// Container 1 (sub-container): Block "getdents64" syscall.
+	testSpecs[1].Linux = &specs.Linux{
+		Seccomp: &specs.LinuxSeccomp{
+			DefaultAction: specs.ActAllow,
+			Syscalls: []specs.LinuxSyscall{
+				{
+					Names:  []string{"getdents", "getdents64"},
+					Action: specs.ActErrno,
+				},
+			},
+		},
+	}
+
+	containers, cleanup, err := startContainers(conf, testSpecs, ids)
+	if err != nil {
+		t.Fatalf("error starting containers: %v", err)
+	}
+	defer cleanup()
+
+	// Container 0: "uname" should be blocked.
+	_, err = executeCombinedOutput(conf, containers[0], nil, "/bin/uname")
+	if err == nil {
+		t.Errorf("uname in container 0 should have failed, but succeeded")
+	}
+
+	// Container 0: "getdents64" should not be blocked.
+	_, err = executeCombinedOutput(conf, containers[0], nil, "/bin/ls", "/")
+	if err != nil {
+		t.Errorf("ls in container 0 should have succeeded: %v", err)
+	}
+
+	// Container 1: "getdents64" should be blocked.
+	_, err = executeCombinedOutput(conf, containers[1], nil, "/bin/ls", "/")
+	if err == nil {
+		t.Errorf("ls in container 1 should have failed, but succeeded")
+	}
+
+	// Container 1: "uname" should not be blocked.
+	_, err = executeCombinedOutput(conf, containers[1], nil, "/bin/uname")
+	if err != nil {
+		t.Errorf("uname in container 1 should have succeeded: %v", err)
+	}
+}