Add support for OCI seccomp filters in the sandbox.

OCI configuration includes support for specifying seccomp filters. In runc,
these filter configurations are converted into seccomp BPF programs and loaded
into the kernel via libseccomp. runsc needs to be a static binary so, for
runsc, we cannot rely on a C library and need to implement the functionality
in Go.

The generator added here implements basic support for taking OCI seccomp
configuration and converting it into a seccomp BPF program with the same
behavior as a program generated by libseccomp.

- New conditional operations were added to pkg/seccomp to support operations
  available in OCI.
- AllowAny and AllowValue were renamed to MatchAny and EqualTo to better reflect
  that syscalls matching the conditionals result in the provided action not
  simply SCMP_RET_ALLOW.
- BuildProgram in pkg/seccomp no longer panics if provided an empty list of
  rules. It now builds a program with the architecture sanity check only.
- ProgramBuilder now allows adding labels that are unused. However, backwards
  jumps are still not permitted.

Fixes #510

PiperOrigin-RevId: 331938697

1 files changed, 229 insertions, 0 deletions

diff --git a/runsc/specutils/seccomp/seccomp.go b/runsc/specutils/seccomp/seccomp.go
new file mode 100644
index 000000000..5932f7a41
--- /dev/null
+++ b/runsc/specutils/seccomp/seccomp.go
@@ -0,0 +1,229 @@
+// Copyright 2020 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 seccomp implements some features of libseccomp in order to support
+// OCI.
+package seccomp
+
+import (
+	"fmt"
+	"syscall"
+
+	specs "github.com/opencontainers/runtime-spec/specs-go"
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/bpf"
+	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/pkg/seccomp"
+	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux"
+)
+
+var (
+	killThreadAction = linux.SECCOMP_RET_KILL_THREAD
+	trapAction       = linux.SECCOMP_RET_TRAP
+	// runc always returns EPERM as the errorcode for SECCOMP_RET_ERRNO
+	errnoAction = linux.SECCOMP_RET_ERRNO.WithReturnCode(uint16(syscall.EPERM))
+	// runc always returns EPERM as the errorcode for SECCOMP_RET_TRACE
+	traceAction = linux.SECCOMP_RET_TRACE.WithReturnCode(uint16(syscall.EPERM))
+	allowAction = linux.SECCOMP_RET_ALLOW
+)
+
+// BuildProgram generates a bpf program based on the given OCI seccomp
+// config.
+func BuildProgram(s *specs.LinuxSeccomp) (bpf.Program, error) {
+	defaultAction, err := convertAction(s.DefaultAction)
+	if err != nil {
+		return bpf.Program{}, fmt.Errorf("secomp default action: %w", err)
+	}
+	ruleset, err := convertRules(s)
+	if err != nil {
+		return bpf.Program{}, fmt.Errorf("invalid seccomp rules: %w", err)
+	}
+
+	instrs, err := seccomp.BuildProgram(ruleset, defaultAction, killThreadAction)
+	if err != nil {
+		return bpf.Program{}, fmt.Errorf("building seccomp program: %w", err)
+	}
+
+	program, err := bpf.Compile(instrs)
+	if err != nil {
+		return bpf.Program{}, fmt.Errorf("compiling seccomp program: %w", err)
+	}
+
+	return program, nil
+}
+
+// lookupSyscallNo gets the syscall number for the syscall with the given name
+// for the given architecture.
+func lookupSyscallNo(arch uint32, name string) (uint32, error) {
+	var table *kernel.SyscallTable
+	switch arch {
+	case linux.AUDIT_ARCH_X86_64:
+		table = slinux.AMD64
+	case linux.AUDIT_ARCH_AARCH64:
+		table = slinux.ARM64
+	}
+	if table == nil {
+		return 0, fmt.Errorf("unsupported architecture: %d", arch)
+	}
+	n, err := table.LookupNo(name)
+	if err != nil {
+		return 0, err
+	}
+	return uint32(n), nil
+}
+
+// convertAction converts a LinuxSeccompAction to BPFAction
+func convertAction(act specs.LinuxSeccompAction) (linux.BPFAction, error) {
+	// TODO(gvisor.dev/issue/3124): Update specs package to include ActLog and ActKillProcess.
+	switch act {
+	case specs.ActKill:
+		return killThreadAction, nil
+	case specs.ActTrap:
+		return trapAction, nil
+	case specs.ActErrno:
+		return errnoAction, nil
+	case specs.ActTrace:
+		return traceAction, nil
+	case specs.ActAllow:
+		return allowAction, nil
+	default:
+		return 0, fmt.Errorf("invalid action: %v", act)
+	}
+}
+
+// convertRules converts OCI linux seccomp rules into RuleSets that can be used by
+// the seccomp package to build a seccomp program.
+func convertRules(s *specs.LinuxSeccomp) ([]seccomp.RuleSet, error) {
+	// NOTE: Architectures are only really relevant when calling 32bit syscalls
+	// on a 64bit system. Since we don't support that in gVisor anyway, we
+	// ignore Architectures and only test against the native architecture.
+
+	ruleset := []seccomp.RuleSet{}
+
+	for _, syscall := range s.Syscalls {
+		sysRules := seccomp.NewSyscallRules()
+
+		action, err := convertAction(syscall.Action)
+		if err != nil {
+			return nil, err
+		}
+
+		// Args
+		rules, err := convertArgs(syscall.Args)
+		if err != nil {
+			return nil, err
+		}
+
+		for _, name := range syscall.Names {
+			syscallNo, err := lookupSyscallNo(nativeArchAuditNo, name)
+			if err != nil {
+				// If there is an error looking up the syscall number, assume it is
+				// not supported on this architecture and ignore it. This is, for
+				// better or worse, what runc does.
+				log.Warningf("OCI seccomp: ignoring syscall %q", name)
+				continue
+			}
+
+			for _, rule := range rules {
+				sysRules.AddRule(uintptr(syscallNo), rule)
+			}
+		}
+
+		ruleset = append(ruleset, seccomp.RuleSet{
+			Rules:  sysRules,
+			Action: action,
+		})
+	}
+
+	return ruleset, nil
+}
+
+// convertArgs converts an OCI seccomp argument rule to a list of seccomp.Rule.
+func convertArgs(args []specs.LinuxSeccompArg) ([]seccomp.Rule, error) {
+	argCounts := make([]uint, 6)
+
+	for _, arg := range args {
+		if arg.Index > 6 {
+			return nil, fmt.Errorf("invalid index: %d", arg.Index)
+		}
+
+		argCounts[arg.Index]++
+	}
+
+	// NOTE: If multiple rules apply to the same argument (same index) the
+	// action is triggered if any one of the rules matches (OR). If not, then
+	// all rules much match in order to trigger the action (AND). This appears to
+	// be some kind of legacy behavior of runc that nevertheless needs to be
+	// supported to maintain compatibility.
+
+	hasMultipleArgs := false
+	for _, count := range argCounts {
+		if count > 1 {
+			hasMultipleArgs = true
+			break
+		}
+	}
+
+	if hasMultipleArgs {
+		rules := []seccomp.Rule{}
+
+		// Old runc behavior - do this for compatibility.
+		// Add rules as ORs by adding separate Rules.
+		for _, arg := range args {
+			rule := seccomp.Rule{nil, nil, nil, nil, nil, nil}
+
+			if err := convertRule(arg, &rule); err != nil {
+				return nil, err
+			}
+
+			rules = append(rules, rule)
+		}
+
+		return rules, nil
+	}
+
+	// Add rules as ANDs by adding to the same Rule.
+	rule := seccomp.Rule{nil, nil, nil, nil, nil, nil}
+	for _, arg := range args {
+		if err := convertRule(arg, &rule); err != nil {
+			return nil, err
+		}
+	}
+
+	return []seccomp.Rule{rule}, nil
+}
+
+// convertRule converts and adds the arg to a rule.
+func convertRule(arg specs.LinuxSeccompArg, rule *seccomp.Rule) error {
+	switch arg.Op {
+	case specs.OpEqualTo:
+		rule[arg.Index] = seccomp.EqualTo(arg.Value)
+	case specs.OpNotEqual:
+		rule[arg.Index] = seccomp.NotEqual(arg.Value)
+	case specs.OpGreaterThan:
+		rule[arg.Index] = seccomp.GreaterThan(arg.Value)
+	case specs.OpGreaterEqual:
+		rule[arg.Index] = seccomp.GreaterThanOrEqual(arg.Value)
+	case specs.OpLessThan:
+		rule[arg.Index] = seccomp.LessThan(arg.Value)
+	case specs.OpLessEqual:
+		rule[arg.Index] = seccomp.LessThanOrEqual(arg.Value)
+	case specs.OpMaskedEqual:
+		rule[arg.Index] = seccomp.MaskedEqual(uintptr(arg.Value), uintptr(arg.ValueTwo))
+	default:
+		return fmt.Errorf("unsupported operand: %q", arg.Op)
+	}
+	return nil
+}