seccomp: add support for flags

List of seccomp flags defined in runtime-spec:
* SECCOMP_FILTER_FLAG_TSYNC
* SECCOMP_FILTER_FLAG_LOG
* SECCOMP_FILTER_FLAG_SPEC_ALLOW

Signed-off-by: Alban Crequy <albancrequy@microsoft.com>

Author: alban

libcontainer/configs/config.go

@@ -33,6 +33,7 @@ type IDMap struct {
 type Seccomp struct {
 	DefaultAction    Action     `json:"default_action"`
 	Architectures    []string   `json:"architectures"`
+	Flags            []string   `json:"flags"`
 	Syscalls         []*Syscall `json:"syscalls"`
 	DefaultErrnoRet  *uint      `json:"default_errno_ret"`
 	ListenerPath     string     `json:"listener_path,omitempty"`

libcontainer/seccomp/seccomp_linux.go

@@ -86,6 +86,29 @@ func InitSeccomp(config *configs.Seccomp) (int, error) {
 		}
 	}
 
+	// Add extra flags
+	for _, flag := range config.Flags {
+		switch flag {
+		case "SECCOMP_FILTER_FLAG_TSYNC":
+			// libseccomp-golang always use filterAttrTsync when
+			// possible so all goroutines will receive the same
+			// rules, so there is nothing to do. It does not make
+			// sense to apply the seccomp filter on only one
+			// thread; other threads will be terminated after exec
+			// anyway.
+		case "SECCOMP_FILTER_FLAG_LOG":
+			if err := filter.SetLogBit(true); err != nil {
+				return -1, fmt.Errorf("error adding log flag to seccomp filter: %w", err)
+			}
+		case "SECCOMP_FILTER_FLAG_SPEC_ALLOW":
+			if err := filter.SetSSB(true); err != nil {
+				return -1, fmt.Errorf("error adding SSB flag to seccomp filter: %w", err)
+			}
+		default:
+			return -1, fmt.Errorf("seccomp flags %q not yet supported by runc", flag)
+		}
+	}
+
 	// Unset no new privs bit
 	if err := filter.SetNoNewPrivsBit(false); err != nil {
 		return -1, fmt.Errorf("error setting no new privileges: %w", err)

libcontainer/specconv/spec_linux.go

@@ -1016,14 +1016,20 @@ func SetupSeccomp(config *specs.LinuxSeccomp) (*configs.Seccomp, error) {
 		return nil, nil
 	}
 
-	// We don't currently support seccomp flags.
-	if len(config.Flags) != 0 {
-		return nil, errors.New("seccomp flags are not yet supported by runc")
-	}
-
 	newConfig := new(configs.Seccomp)
 	newConfig.Syscalls = []*configs.Syscall{}
 
+	// The list of flags defined in runtime-spec is a subset of the flags
+	// in the seccomp() syscall
+	for _, flag := range config.Flags {
+		switch flag {
+		case "SECCOMP_FILTER_FLAG_TSYNC", "SECCOMP_FILTER_FLAG_LOG", "SECCOMP_FILTER_FLAG_SPEC_ALLOW":
+			newConfig.Flags = append(newConfig.Flags, string(flag))
+		default:
+			return nil, fmt.Errorf("seccomp flags %q not yet supported by runc", flag)
+		}
+	}
+
 	if len(config.Architectures) > 0 {
 		newConfig.Architectures = []string{}
 		for _, arch := range config.Architectures {

tests/integration/seccomp.bats

@@ -66,6 +66,58 @@ function teardown() {
 	[[ "$output" == *"Network is down"* ]]
 }
 
+@test "runc run [seccomp] (SECCOMP_FILTER_FLAG_LOG)" {
+	requires_kernel 4.14 # SECCOMP_FILTER_FLAG_LOG appeared in Linux 4.14
+	update_config '   .process.args = ["/bin/sh", "-c", "mkdir /dev/shm/foo"]
+			| .process.noNewPrivileges = false
+			| .linux.seccomp = {
+				"defaultAction":"SCMP_ACT_ALLOW",
+				"architectures":["SCMP_ARCH_X86","SCMP_ARCH_X32"],
+				"flags":["SECCOMP_FILTER_FLAG_LOG"],
+				"syscalls":[{"names":["mkdir"], "action":"SCMP_ACT_ERRNO"}]
+			}'
+
+	# This test checks that the log flag is accepted but does not check the
+	# audit log
+	runc run test_busybox
+	[ "$status" -ne 0 ]
+	[[ "$output" == *"mkdir:"*"/dev/shm/foo"*"Operation not permitted"* ]]
+}
+
+@test "runc run [seccomp] (SECCOMP_FILTER_FLAG_SPEC_ALLOW)" {
+	requires_kernel 4.17 # SECCOMP_FILTER_FLAG_SPEC_ALLOW appeared in Linux 4.17
+	update_config '   .process.args = ["/bin/sh", "-c", "mkdir /dev/shm/foo"]
+			| .process.noNewPrivileges = false
+			| .linux.seccomp = {
+				"defaultAction":"SCMP_ACT_ALLOW",
+				"architectures":["SCMP_ARCH_X86","SCMP_ARCH_X32"],
+				"flags":["SECCOMP_FILTER_FLAG_SPEC_ALLOW"],
+				"syscalls":[{"names":["mkdir"], "action":"SCMP_ACT_ERRNO"}]
+			}'
+
+	# This test checks that the SSB flag is accepted but does not check the
+	# result
+	runc run test_busybox
+	[ "$status" -ne 0 ]
+	[[ "$output" == *"mkdir:"*"/dev/shm/foo"*"Operation not permitted"* ]]
+}
+
+@test "runc run [seccomp] (SECCOMP_FILTER_FLAG_TSYNC)" {
+	update_config '   .process.args = ["/bin/sh", "-c", "mkdir /dev/shm/foo"]
+			| .process.noNewPrivileges = false
+			| .linux.seccomp = {
+				"defaultAction":"SCMP_ACT_ALLOW",
+				"architectures":["SCMP_ARCH_X86","SCMP_ARCH_X32"],
+				"flags":["SECCOMP_FILTER_FLAG_TSYNC"],
+				"syscalls":[{"names":["mkdir"], "action":"SCMP_ACT_ERRNO"}]
+			}'
+
+	# This test checks that the tsync flag is accepted
+	runc run test_busybox
+	[ "$status" -ne 0 ]
+	[[ "$output" == *"mkdir:"*"/dev/shm/foo"*"Operation not permitted"* ]]
+}
+
 @test "runc run [seccomp] (SCMP_ACT_KILL)" {
 	update_config '  .process.args = ["/bin/sh", "-c", "mkdir /dev/shm/foo"]
 			| .process.noNewPrivileges = false