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

Note that runc does not apply SECCOMP_FILTER_FLAG_TSYNC. It does not
make sense to apply the seccomp filter on only one thread; other threads
will be terminated after exec anyway.

See similar commit in crun:
https://github.com/containers/crun/commit/fefabffa2816ea343068ed036a86944393db189a

Note that SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV (introduced by
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?id=c2aa2dfef243
in Linux 5.19-rc1) is not added yet because Linux 5.19 is not released
yet.

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

libcontainer/configs/config.go

@@ -31,12 +31,13 @@ type IDMap struct {
 // for syscalls. Additional architectures can be added by specifying them in
 // Architectures.
 type Seccomp struct {
-	DefaultAction    Action     `json:"default_action"`
-	Architectures    []string   `json:"architectures"`
-	Syscalls         []*Syscall `json:"syscalls"`
-	DefaultErrnoRet  *uint      `json:"default_errno_ret"`
-	ListenerPath     string     `json:"listener_path,omitempty"`
-	ListenerMetadata string     `json:"listener_metadata,omitempty"`
+	DefaultAction    Action                   `json:"default_action"`
+	Architectures    []string                 `json:"architectures"`
+	Flags            []specs.LinuxSeccompFlag `json:"flags"`
+	Syscalls         []*Syscall               `json:"syscalls"`
+	DefaultErrnoRet  *uint                    `json:"default_errno_ret"`
+	ListenerPath     string                   `json:"listener_path,omitempty"`
+	ListenerMetadata string                   `json:"listener_metadata,omitempty"`
 }
 
 // Action is taken upon rule match in Seccomp

libcontainer/seccomp/seccomp_linux.go

@@ -13,6 +13,7 @@ import (
 
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"github.com/opencontainers/runc/libcontainer/seccomp/patchbpf"
+	"github.com/opencontainers/runtime-spec/specs-go"
 )
 
 var (
@@ -86,6 +87,29 @@ 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 specs.LinuxSeccompFlagLog:
+			if err := filter.SetLogBit(true); err != nil {
+				return -1, fmt.Errorf("error adding log flag to seccomp filter: %w", err)
+			}
+		case specs.LinuxSeccompFlagSpecAllow:
+			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 +1016,22 @@ 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":
+			// Tsync can be silently ignored
+		case specs.LinuxSeccompFlagLog, specs.LinuxSeccompFlagSpecAllow:
+			newConfig.Flags = append(newConfig.Flags, flag)
+		default:
+			return nil, fmt.Errorf("seccomp flag %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 +66,38 @@ function teardown() {
 	[[ "$output" == *"Network is down"* ]]
 }
 
+@test "runc run [seccomp] (SECCOMP_FILTER_FLAG_*)" {
+	# Linux 4.14: SECCOMP_FILTER_FLAG_LOG
+	# Linux 4.17: SECCOMP_FILTER_FLAG_SPEC_ALLOW
+	requires_kernel 4.17
+	SECCOMP_FILTER_FLAGS=(
+		'' # no flag
+		'"SECCOMP_FILTER_FLAG_LOG"'
+		'"SECCOMP_FILTER_FLAG_SPEC_ALLOW"'
+		'"SECCOMP_FILTER_FLAG_TSYNC"'
+		'"SECCOMP_FILTER_FLAG_LOG","SECCOMP_FILTER_FLAG_SPEC_ALLOW"'
+		'"SECCOMP_FILTER_FLAG_LOG","SECCOMP_FILTER_FLAG_TSYNC"'
+		'"SECCOMP_FILTER_FLAG_SPEC_ALLOW","SECCOMP_FILTER_FLAG_TSYNC"'
+		'"SECCOMP_FILTER_FLAG_LOG","SECCOMP_FILTER_FLAG_SPEC_ALLOW","SECCOMP_FILTER_FLAG_TSYNC"'
+	)
+	for flags in "${SECCOMP_FILTER_FLAGS[@]}"; do
+		update_config '   .process.args = ["/bin/sh", "-c", "mkdir /dev/shm/foo"]
+				| .process.noNewPrivileges = false
+				| .linux.seccomp = {
+					"defaultAction":"SCMP_ACT_ALLOW",
+					"architectures":["SCMP_ARCH_X86","SCMP_ARCH_X32","SCMP_ARCH_X86_64","SCMP_ARCH_AARCH64","SCMP_ARCH_ARM"],
+					"flags":['"${flags}"'],
+					"syscalls":[{"names":["mkdir"], "action":"SCMP_ACT_ERRNO"}]
+				}'
+
+		# This test checks that the flags are accepted without errors but does
+		# not check they are effectively applied
+		runc run test_busybox
+		[ "$status" -ne 0 ]
+		[[ "$output" == *"mkdir:"*"/dev/shm/foo"*"Operation not permitted"* ]]
+	done
+}
+
 @test "runc run [seccomp] (SCMP_ACT_KILL)" {
 	update_config '  .process.args = ["/bin/sh", "-c", "mkdir /dev/shm/foo"]
 			| .process.noNewPrivileges = false