Add SyscallRules that supports argument filtering

PiperOrigin-RevId: 198919043
Change-Id: I7f1f0a3b3430cd0936a4ee4fc6859aab71820bdf

pkg/bpf/program_builder.go

@@ -21,7 +21,10 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/abi/linux"
 )
 
-const labelTarget = math.MaxUint8
+const (
+	labelTarget       = math.MaxUint8
+	labelDirectTarget = math.MaxUint32
+)
 
 // ProgramBuilder assists with building a BPF program with jump
 // labels that are resolved to their proper offsets.
@@ -47,6 +50,14 @@ type label struct {
 	target int
 }
 
+type jmpType int
+
+const (
+	jDirect jmpType = iota
+	jTrue
+	jFalse
+)
+
 // source contains information about a single reference to a label.
 type source struct {
 	// Program line where the label reference is present.
@@ -54,7 +65,7 @@ type source struct {
 
 	// True if label reference is in the 'jump if true' part of the jump.
 	// False if label reference is in the 'jump if false' part of the jump.
-	jt bool
+	jt jmpType
 }
 
 // AddStmt adds a new statement to the program.
@@ -67,23 +78,29 @@ func (b *ProgramBuilder) AddJump(code uint16, k uint32, jt, jf uint8) {
 	b.instructions = append(b.instructions, Jump(code, k, jt, jf))
 }
 
+// AddDirectJumpLabel adds a new jump to the program where is labelled.
+func (b *ProgramBuilder) AddDirectJumpLabel(labelName string) {
+	b.addLabelSource(labelName, jDirect)
+	b.AddJump(Jmp|Ja, labelDirectTarget, 0, 0)
+}
+
 // AddJumpTrueLabel adds a new jump to the program where 'jump if true' is a label.
 func (b *ProgramBuilder) AddJumpTrueLabel(code uint16, k uint32, jtLabel string, jf uint8) {
-	b.addLabelSource(jtLabel, true)
+	b.addLabelSource(jtLabel, jTrue)
 	b.AddJump(code, k, labelTarget, jf)
 }
 
 // AddJumpFalseLabel adds a new jump to the program where 'jump if false' is a label.
 func (b *ProgramBuilder) AddJumpFalseLabel(code uint16, k uint32, jt uint8, jfLabel string) {
-	b.addLabelSource(jfLabel, false)
-	b.AddJump(code, k, jt, math.MaxUint8)
+	b.addLabelSource(jfLabel, jFalse)
+	b.AddJump(code, k, jt, labelTarget)
 }
 
 // AddJumpLabels adds a new jump to the program where both jump targets are labels.
 func (b *ProgramBuilder) AddJumpLabels(code uint16, k uint32, jtLabel, jfLabel string) {
-	b.addLabelSource(jtLabel, true)
-	b.addLabelSource(jfLabel, false)
-	b.AddJump(code, k, math.MaxUint8, math.MaxUint8)
+	b.addLabelSource(jtLabel, jTrue)
+	b.addLabelSource(jfLabel, jFalse)
+	b.AddJump(code, k, labelTarget, labelTarget)
 }
 
 // AddLabel sets the given label name at the current location. The next instruction is executed
@@ -104,20 +121,22 @@ func (b *ProgramBuilder) AddLabel(name string) error {
 
 // Instructions returns an array of BPF instructions representing the program with all labels
 // resolved. Return error in case label resolution failed due to an invalid program.
+//
+// N.B. Partial results will be returned in the error case, which is useful for debugging.
 func (b *ProgramBuilder) Instructions() ([]linux.BPFInstruction, error) {
 	if err := b.resolveLabels(); err != nil {
-		return nil, err
+		return b.instructions, err
 	}
 	return b.instructions, nil
 }
 
-func (b *ProgramBuilder) addLabelSource(labelName string, jt bool) {
+func (b *ProgramBuilder) addLabelSource(labelName string, t jmpType) {
 	l, ok := b.labels[labelName]
 	if !ok {
 		l = &label{sources: make([]source, 0), target: -1}
 		b.labels[labelName] = l
 	}
-	l.sources = append(l.sources, source{line: len(b.instructions), jt: jt})
+	l.sources = append(l.sources, source{line: len(b.instructions), jt: t})
 }
 
 func (b *ProgramBuilder) resolveLabels() error {
@@ -136,21 +155,34 @@ func (b *ProgramBuilder) resolveLabels() error {
 			}
 			// Calculates the jump offset from current line.
 			offset := v.target - s.line - 1
-			if offset > math.MaxUint8 {
-				return fmt.Errorf("jump offset to label '%v' is too large: %v", key, offset)
-			}
 			// Sets offset into jump instruction.
-			if s.jt {
+			switch s.jt {
+			case jDirect:
+				if offset > labelDirectTarget {
+					return fmt.Errorf("jump offset to label '%v' is too large: %v, inst: %v, lineno: %v", key, offset, inst, s.line)
+				}
+				if inst.K != labelDirectTarget {
+					return fmt.Errorf("jump target is not a label")
+				}
+				inst.K = uint32(offset)
+			case jTrue:
+				if offset > labelTarget {
+					return fmt.Errorf("jump offset to label '%v' is too large: %v, inst: %v, lineno: %v", key, offset, inst, s.line)
+				}
 				if inst.JumpIfTrue != labelTarget {
 					return fmt.Errorf("jump target is not a label")
 				}
 				inst.JumpIfTrue = uint8(offset)
-			} else {
+			case jFalse:
+				if offset > labelTarget {
+					return fmt.Errorf("jump offset to label '%v' is too large: %v, inst: %v, lineno: %v", key, offset, inst, s.line)
+				}
 				if inst.JumpIfFalse != labelTarget {
 					return fmt.Errorf("jump target is not a label")
 				}
 				inst.JumpIfFalse = uint8(offset)
 			}
+
 			b.instructions[s.line] = inst
 		}
 	}

pkg/seccomp/BUILD

@@ -21,14 +21,18 @@ go_library(
     name = "seccomp",
     srcs = [
         "seccomp.go",
+        "seccomp_rules.go",
         "seccomp_unsafe.go",
     ],
     importpath = "gvisor.googlesource.com/gvisor/pkg/seccomp",
     visibility = ["//visibility:public"],
     deps = [
+        "//pkg/abi",
         "//pkg/abi/linux",
         "//pkg/bpf",
         "//pkg/log",
+        "//pkg/sentry/arch",
+        "//pkg/sentry/strace",
     ],
 )
 

pkg/seccomp/seccomp.go

@@ -12,24 +12,28 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-// Package seccomp provides basic seccomp filters.
+// Package seccomp provides basic seccomp filters for x86_64 (little endian).
 package seccomp
 
 import (
 	"fmt"
+	"reflect"
 	"sort"
 
+	"gvisor.googlesource.com/gvisor/pkg/abi"
 	"gvisor.googlesource.com/gvisor/pkg/abi/linux"
 	"gvisor.googlesource.com/gvisor/pkg/bpf"
 	"gvisor.googlesource.com/gvisor/pkg/log"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/arch"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/strace"
 )
 
 const (
 	// violationLabel is added to the program to take action on a violation.
 	violationLabel = "violation"
 
-	// allowLabel is added to the program to allow the syscall to take place.
-	allowLabel = "allow"
+	// skipOneInst is the offset to take for skipping one instruction.
+	skipOneInst = 1
 )
 
 // Install generates BPF code based on the set of syscalls provided. It only
@@ -38,27 +42,19 @@ const (
 //
 // (*) The current implementation only checks the syscall number. It does NOT
 // validate any of the arguments.
-func Install(syscalls []uintptr, kill bool) error {
-	// Sort syscalls and remove duplicates to build the BST.
-	sort.Slice(syscalls, func(i, j int) bool { return syscalls[i] < syscalls[j] })
-	syscalls = filterUnique(syscalls)
-
-	log.Infof("Installing seccomp filters for %d syscalls (kill=%t)", len(syscalls), kill)
-	for _, s := range syscalls {
-		log.Infof("syscall filter: %v", s)
-	}
-
-	instrs, err := buildProgram(syscalls, kill)
-	if err != nil {
-		return err
-	}
+func Install(rules SyscallRules, kill bool) error {
+	log.Infof("Installing seccomp filters for %d syscalls (kill=%t)", len(rules), kill)
+	instrs, err := buildProgram(rules, kill)
 	if log.IsLogging(log.Debug) {
-		programStr, err := bpf.DecodeProgram(instrs)
-		if err != nil {
-			programStr = fmt.Sprintf("Error: %v\n%s", err, programStr)
+		programStr, errDecode := bpf.DecodeProgram(instrs)
+		if errDecode != nil {
+			programStr = fmt.Sprintf("Error: %v\n%s", errDecode, programStr)
 		}
 		log.Debugf("Seccomp program dump:\n%s", programStr)
 	}
+	if err != nil {
+		return err
+	}
 
 	if err := seccomp(instrs); err != nil {
 		return err
@@ -68,11 +64,8 @@ func Install(syscalls []uintptr, kill bool) error {
 	return nil
 }
 
-// buildProgram builds a BPF program that whitelists all given syscalls.
-//
-// Precondition: syscalls must be sorted and unique.
-func buildProgram(syscalls []uintptr, kill bool) ([]linux.BPFInstruction, error) {
-	const archOffset = 4 // offsetof(seccomp_data, arch)
+// buildProgram builds a BPF program that whitelists all given syscall rules.
+func buildProgram(rules SyscallRules, kill bool) ([]linux.BPFInstruction, error) {
 	program := bpf.NewProgramBuilder()
 	violationAction := uint32(linux.SECCOMP_RET_KILL)
 	if !kill {
@@ -83,10 +76,13 @@ func buildProgram(syscalls []uintptr, kill bool) ([]linux.BPFInstruction, error)
 	//
 	// A = seccomp_data.arch
 	// if (A != AUDIT_ARCH_X86_64) goto violation
-	program.AddStmt(bpf.Ld|bpf.Abs|bpf.W, archOffset)
-	program.AddJumpFalseLabel(bpf.Jmp|bpf.Jeq|bpf.K, linux.AUDIT_ARCH_X86_64, 0, violationLabel)
+	program.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetArch)
+	// violationLabel is at the bottom of the program. The size of program
+	// may exceeds 255 lines, which is the limit of a condition jump.
+	program.AddJump(bpf.Jmp|bpf.Jeq|bpf.K, linux.AUDIT_ARCH_X86_64, skipOneInst, 0)
+	program.AddDirectJumpLabel(violationLabel)
 
-	if err := buildIndex(syscalls, program); err != nil {
+	if err := buildIndex(rules, program); err != nil {
 		return nil, err
 	}
 
@@ -96,41 +92,34 @@ func buildProgram(syscalls []uintptr, kill bool) ([]linux.BPFInstruction, error)
 	}
 	program.AddStmt(bpf.Ret|bpf.K, violationAction)
 
-	// allow: return SECCOMP_RET_ALLOW
-	if err := program.AddLabel(allowLabel); err != nil {
-		return nil, err
-	}
-	program.AddStmt(bpf.Ret|bpf.K, linux.SECCOMP_RET_ALLOW)
-
 	return program.Instructions()
 }
 
-// filterUnique filters unique system calls.
-//
-// Precondition: syscalls must be sorted.
-func filterUnique(syscalls []uintptr) []uintptr {
-	filtered := make([]uintptr, 0, len(syscalls))
-	for i := 0; i < len(syscalls); i++ {
-		if len(filtered) > 0 && syscalls[i] == filtered[len(filtered)-1] {
-			// This call has already been inserted, skip.
-			continue
-		}
-		filtered = append(filtered, syscalls[i])
-	}
-	return filtered
-}
-
 // buildIndex builds a BST to quickly search through all syscalls that are whitelisted.
-//
-// Precondition: syscalls must be sorted and unique.
-func buildIndex(syscalls []uintptr, program *bpf.ProgramBuilder) error {
+func buildIndex(rules SyscallRules, program *bpf.ProgramBuilder) error {
+	syscalls := []uintptr{}
+	for sysno, _ := range rules {
+		syscalls = append(syscalls, sysno)
+	}
+
+	t, ok := strace.Lookup(abi.Linux, arch.AMD64)
+	if !ok {
+		panic("Can't find amd64 Linux syscall table")
+	}
+
+	sort.Slice(syscalls, func(i, j int) bool { return syscalls[i] < syscalls[j] })
+	for _, s := range syscalls {
+		log.Infof("syscall filter: %v (%v): %s", s, t.Name(s), rules[s])
+	}
+
 	root := createBST(syscalls)
+	root.root = true
 
 	// Load syscall number into A and run through BST.
 	//
 	// A = seccomp_data.nr
-	program.AddStmt(bpf.Ld|bpf.Abs|bpf.W, 0)
-	return root.buildBSTProgram(program, true)
+	program.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetNR)
+	return root.traverse(buildBSTProgram, program, rules)
 }
 
 // createBST converts sorted syscall slice into a balanced BST.
@@ -147,11 +136,106 @@ func createBST(syscalls []uintptr) *node {
 	return &parent
 }
 
+func ruleViolationLabel(sysno uintptr, idx int) string {
+	return fmt.Sprintf("ruleViolation_%v_%v", sysno, idx)
+}
+
+func checkArgsLabel(sysno uintptr) string {
+	return fmt.Sprintf("checkArgs_%v", sysno)
+}
+
+func addSyscallArgsCheck(p *bpf.ProgramBuilder, rules []Rule, sysno uintptr) error {
+	for ruleidx, rule := range rules {
+		labelled := false
+		for i, arg := range rule {
+			if arg != nil {
+				switch a := arg.(type) {
+				case AllowAny:
+				case AllowValue:
+					high, low := uint32(a>>32), uint32(a)
+					// assert arg_low == low
+					p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetArgLow(i))
+					p.AddJumpFalseLabel(bpf.Jmp|bpf.Jeq|bpf.K, low, 0, ruleViolationLabel(sysno, ruleidx))
+					// assert arg_high == high
+					p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetArgHigh(i))
+					p.AddJumpFalseLabel(bpf.Jmp|bpf.Jeq|bpf.K, high, 0, ruleViolationLabel(sysno, ruleidx))
+					labelled = true
+
+				default:
+					return fmt.Errorf("unknown syscall rule type: %v", reflect.TypeOf(a))
+				}
+			}
+		}
+		// Matched, allow the syscall.
+		p.AddStmt(bpf.Ret|bpf.K, linux.SECCOMP_RET_ALLOW)
+		// Label the end of the rule if necessary.
+		if labelled {
+			if err := p.AddLabel(ruleViolationLabel(sysno, ruleidx)); err != nil {
+				return err
+			}
+		}
+	}
+	// Not matched?
+	p.AddDirectJumpLabel(violationLabel)
+	return nil
+}
+
+// buildBSTProgram converts a binary tree started in 'root' into BPF code. The ouline of the code
+// is as follows:
+//
+// // SYS_PIPE(22), root
+//   (A == 22) ? goto argument check : continue
+//   (A > 22) ? goto index_35 : goto index_9
+//
+// index_9:  // SYS_MMAP(9), leaf
+//   A == 9) ? goto argument check : violation
+//
+// index_35:  // SYS_NANOSLEEP(35), single child
+//   (A == 35) ? goto argument check : continue
+//   (A > 35) ? goto index_50 : goto violation
+//
+// index_50:  // SYS_LISTEN(50), leaf
+//   (A == 50) ? goto argument check : goto violation
+//
+func buildBSTProgram(program *bpf.ProgramBuilder, rules SyscallRules, n *node) error {
+	// Root node is never referenced by label, skip it.
+	if !n.root {
+		if err := program.AddLabel(n.label()); err != nil {
+			return err
+		}
+	}
+
+	sysno := n.value
+	program.AddJumpTrueLabel(bpf.Jmp|bpf.Jeq|bpf.K, uint32(sysno), checkArgsLabel(sysno), 0)
+	if n.left == nil && n.right == nil {
+		// Leaf nodes don't require extra check.
+		program.AddDirectJumpLabel(violationLabel)
+	} else {
+		// Non-leaf node. Check which turn to take otherwise. Using direct jumps
+		// in case that the offset may exceed the limit of a conditional jump (255)
+		// Note that 'violationLabel' is returned for nil children.
+		program.AddJump(bpf.Jmp|bpf.Jgt|bpf.K, uint32(sysno), 0, skipOneInst)
+		program.AddDirectJumpLabel(n.right.label())
+		program.AddDirectJumpLabel(n.left.label())
+	}
+
+	if err := program.AddLabel(checkArgsLabel(sysno)); err != nil {
+		return err
+	}
+	// No rules, just allow it and save one jmp.
+	if len(rules[sysno]) == 0 {
+		program.AddStmt(bpf.Ret|bpf.K, linux.SECCOMP_RET_ALLOW)
+		return nil
+	}
+	return addSyscallArgsCheck(program, rules[sysno], sysno)
+}
+
 // node represents a tree node.
 type node struct {
 	value uintptr
 	left  *node
 	right *node
+	root  bool
 }
 
 // label returns the label corresponding to this node. If node is nil (syscall not present),
@@ -163,48 +247,17 @@ func (n *node) label() string {
 	return fmt.Sprintf("index_%v", n.value)
 }
 
-// buildBSTProgram converts a binary tree started in 'root' into BPF code. The ouline of the code
-// is as follows:
-//
-// // SYS_PIPE(22), root
-//   (A == 22) ? goto allow : continue
-//   (A > 22) ? goto index_35 : goto index_9
-//
-// index_9:  // SYS_MMAP(9), leaf
-//   (A == 9) ? goto allow : goto violation
-//
-// index_35:  // SYS_NANOSLEEP(35), single child
-//   (A == 35) ? goto allow : continue
-//   (A > 35) ? goto index_50 : goto violation
-//
-// index_50:  // SYS_LISTEN(50), leaf
-//   (A == 50) ? goto allow : goto violation
-//
-func (n *node) buildBSTProgram(program *bpf.ProgramBuilder, root bool) error {
+type traverseFunc func(*bpf.ProgramBuilder, SyscallRules, *node) error
+
+func (n *node) traverse(fn traverseFunc, p *bpf.ProgramBuilder, rules SyscallRules) error {
 	if n == nil {
 		return nil
 	}
-
-	// Root node is never referenced by label, skip it.
-	if !root {
-		if err := program.AddLabel(n.label()); err != nil {
-			return err
-		}
-	}
-
-	// Leaf nodes don't require extra check, they either allow or violate!
-	if n.left == nil && n.right == nil {
-		program.AddJumpLabels(bpf.Jmp|bpf.Jeq|bpf.K, uint32(n.value), allowLabel, violationLabel)
-		return nil
-	}
-
-	// Non-leaf node. Allows syscall if it matches, check which turn to take otherwise. Note
-	// that 'violationLabel' is returned for nil children.
-	program.AddJumpTrueLabel(bpf.Jmp|bpf.Jeq|bpf.K, uint32(n.value), allowLabel, 0)
-	program.AddJumpLabels(bpf.Jmp|bpf.Jgt|bpf.K, uint32(n.value), n.right.label(), n.left.label())
-
-	if err := n.left.buildBSTProgram(program, false); err != nil {
+	if err := fn(p, rules, n); err != nil {
 		return err
 	}
-	return n.right.buildBSTProgram(program, false)
+	if err := n.left.traverse(fn, p, rules); err != nil {
+		return err
+	}
+	return n.right.traverse(fn, p, rules)
 }

pkg/seccomp/seccomp_rules.go

@@ -0,0 +1,119 @@
+// Copyright 2018 Google Inc.
+//
+// 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
+
+import "fmt"
+
+// The offsets are based on the following struct in include/linux/seccomp.h.
+// struct seccomp_data {
+//	int nr;
+//	__u32 arch;
+//	__u64 instruction_pointer;
+//	__u64 args[6];
+// };
+const (
+	seccompDataOffsetNR   = 0
+	seccompDataOffsetArch = 4
+	seccompDataOffsetArgs = 16
+)
+
+func seccompDataOffsetArgLow(i int) uint32 {
+	return uint32(seccompDataOffsetArgs + i*8)
+}
+
+func seccompDataOffsetArgHigh(i int) uint32 {
+	return uint32(seccompDataOffsetArgs + i*8 + 4)
+}
+
+// AllowAny is marker to indicate any value will be accepted.
+type AllowAny struct{}
+
+func (a AllowAny) String() (s string) {
+	return "*"
+}
+
+// AllowValue specifies a value that needs to be strictly matched.
+type AllowValue uintptr
+
+func (a AllowValue) String() (s string) {
+	return fmt.Sprintf("%#x ", uintptr(a))
+}
+
+// Rule stores the whitelist of syscall arguments.
+//
+// For example:
+// rule := Rule {
+//       AllowValue(linux.ARCH_GET_FS | linux.ARCH_SET_FS), // arg0
+// }
+type Rule [6]interface{}
+
+func (r Rule) String() (s string) {
+	if len(r) == 0 {
+		return
+	}
+	s += "( "
+	for _, arg := range r {
+		if arg != nil {
+			s += fmt.Sprintf("%v ", arg)
+		}
+	}
+	s += ")"
+	return
+}
+
+// SyscallRules stores a map of OR'ed whitelist rules indexed by the syscall number.
+// If the 'Rules' is empty, we treat it as any argument is allowed.
+//
+// For example:
+//  rules := SyscallRules{
+//         syscall.SYS_FUTEX: []Rule{
+//                 {
+//                         AllowAny{},
+//                         AllowValue(linux.FUTEX_WAIT | linux.FUTEX_PRIVATE_FLAG),
+//                 }, // OR
+//                 {
+//                         AllowAny{},
+//                         AllowValue(linux.FUTEX_WAKE | linux.FUTEX_PRIVATE_FLAG),
+//                 },
+//         },
+//         syscall.SYS_GETPID: []Rule{},
+// }
+type SyscallRules map[uintptr][]Rule
+
+// NewSyscallRules returns a new SyscallRules.
+func NewSyscallRules() SyscallRules {
+	return make(map[uintptr][]Rule)
+}
+
+// AddRule adds the given rule. It will create a new entry for a new syscall, otherwise
+// it will append to the existing rules.
+func (sr SyscallRules) AddRule(sysno uintptr, r Rule) {
+	if _, ok := sr[sysno]; ok {
+		sr[sysno] = append(sr[sysno], r)
+	} else {
+		sr[sysno] = []Rule{r}
+	}
+}
+
+// Merge merges the given SyscallRules.
+func (sr SyscallRules) Merge(rules SyscallRules) {
+	for sysno, rs := range rules {
+		if _, ok := sr[sysno]; ok {
+			sr[sysno] = append(sr[sysno], rs...)
+		} else {
+			sr[sysno] = rs
+		}
+	}
+}

pkg/seccomp/seccomp_test.go

@@ -19,10 +19,10 @@ import (
 	"fmt"
 	"io"
 	"io/ioutil"
+	"math"
 	"math/rand"
 	"os"
 	"os/exec"
-	"sort"
 	"strings"
 	"testing"
 	"time"
@@ -77,12 +77,12 @@ func TestBasic(t *testing.T) {
 
 	for _, test := range []struct {
 		// filters are the set of syscall that are allowed.
-		filters []uintptr
+		filters SyscallRules
 		kill    bool
 		specs   []spec
 	}{
 		{
-			filters: []uintptr{1},
+			filters: SyscallRules{1: {}},
 			kill:    false,
 			specs: []spec{
 				{
@@ -98,8 +98,12 @@ func TestBasic(t *testing.T) {
 			},
 		},
 		{
-			filters: []uintptr{1, 3, 5},
-			kill:    false,
+			filters: SyscallRules{
+				1: {},
+				3: {},
+				5: {},
+			},
+			kill: false,
 			specs: []spec{
 				{
 					desc: "Multiple syscalls allowed (1)",
@@ -144,7 +148,7 @@ func TestBasic(t *testing.T) {
 			},
 		},
 		{
-			filters: []uintptr{1},
+			filters: SyscallRules{1: {}},
 			kill:    false,
 			specs: []spec{
 				{
@@ -155,7 +159,7 @@ func TestBasic(t *testing.T) {
 			},
 		},
 		{
-			filters: []uintptr{1},
+			filters: SyscallRules{1: {}},
 			kill:    true,
 			specs: []spec{
 				{
@@ -165,8 +169,96 @@ func TestBasic(t *testing.T) {
 				},
 			},
 		},
+		{
+			filters: SyscallRules{
+				1: []Rule{
+					{
+						AllowAny{},
+						AllowValue(0xf),
+					},
+				},
+			},
+			kill: false,
+			specs: []spec{
+				{
+					desc: "Syscall argument allowed",
+					data: seccompData{nr: 1, arch: linux.AUDIT_ARCH_X86_64, args: [6]uint64{0xf, 0xf}},
+					want: linux.SECCOMP_RET_ALLOW,
+				},
+				{
+					desc: "Syscall argument disallowed",
+					data: seccompData{nr: 1, arch: linux.AUDIT_ARCH_X86_64, args: [6]uint64{0xf, 0xe}},
+					want: linux.SECCOMP_RET_TRAP,
+				},
+			},
+		},
+		{
+			filters: SyscallRules{
+				1: []Rule{
+					{
+						AllowValue(0xf),
+					},
+					{
+						AllowValue(0xe),
+					},
+				},
+			},
+			kill: false,
+			specs: []spec{
+				{
+					desc: "Syscall argument allowed, two rules",
+					data: seccompData{nr: 1, arch: linux.AUDIT_ARCH_X86_64, args: [6]uint64{0xf}},
+					want: linux.SECCOMP_RET_ALLOW,
+				},
+				{
+					desc: "Syscall argument allowed, two rules",
+					data: seccompData{nr: 1, arch: linux.AUDIT_ARCH_X86_64, args: [6]uint64{0xe}},
+					want: linux.SECCOMP_RET_ALLOW,
+				},
+			},
+		},
+		{
+			filters: SyscallRules{
+				1: []Rule{
+					{
+						AllowValue(0),
+						AllowValue(math.MaxUint64 - 1),
+						AllowValue(math.MaxUint32),
+					},
+				},
+			},
+			kill: false,
+			specs: []spec{
+				{
+					desc: "64bit syscall argument allowed",
+					data: seccompData{
+						nr:   1,
+						arch: linux.AUDIT_ARCH_X86_64,
+						args: [6]uint64{0, math.MaxUint64 - 1, math.MaxUint32},
+					},
+					want: linux.SECCOMP_RET_ALLOW,
+				},
+				{
+					desc: "64bit syscall argument disallowed",
+					data: seccompData{
+						nr:   1,
+						arch: linux.AUDIT_ARCH_X86_64,
+						args: [6]uint64{0, math.MaxUint64, math.MaxUint32},
+					},
+					want: linux.SECCOMP_RET_TRAP,
+				},
+				{
+					desc: "64bit syscall argument disallowed",
+					data: seccompData{
+						nr:   1,
+						arch: linux.AUDIT_ARCH_X86_64,
+						args: [6]uint64{0, math.MaxUint64, math.MaxUint32 - 1},
+					},
+					want: linux.SECCOMP_RET_TRAP,
+				},
+			},
+		},
 	} {
-		sort.Slice(test.filters, func(i, j int) bool { return test.filters[i] < test.filters[j] })
 		instrs, err := buildProgram(test.filters, test.kill)
 		if err != nil {
 			t.Errorf("%s: buildProgram() got error: %v", test.specs[0].desc, err)
@@ -193,19 +285,16 @@ func TestBasic(t *testing.T) {
 func TestRandom(t *testing.T) {
 	rand.Seed(time.Now().UnixNano())
 	size := rand.Intn(50) + 1
-	syscalls := make([]uintptr, 0, size)
-	syscallMap := make(map[uintptr]struct{})
-	for len(syscalls) < size {
+	syscallRules := make(map[uintptr][]Rule)
+	for len(syscallRules) < size {
 		n := uintptr(rand.Intn(200))
-		if _, ok := syscallMap[n]; !ok {
-			syscalls = append(syscalls, n)
-			syscallMap[n] = struct{}{}
+		if _, ok := syscallRules[n]; !ok {
+			syscallRules[n] = []Rule{}
 		}
 	}
 
-	sort.Slice(syscalls, func(i, j int) bool { return syscalls[i] < syscalls[j] })
-	fmt.Printf("Testing filters: %v", syscalls)
-	instrs, err := buildProgram(syscalls, false)
+	fmt.Printf("Testing filters: %v", syscallRules)
+	instrs, err := buildProgram(syscallRules, false)
 	if err != nil {
 		t.Fatalf("buildProgram() got error: %v", err)
 	}
@@ -221,7 +310,7 @@ func TestRandom(t *testing.T) {
 			continue
 		}
 		want := uint32(linux.SECCOMP_RET_TRAP)
-		if _, ok := syscallMap[uintptr(i)]; ok {
+		if _, ok := syscallRules[uintptr(i)]; ok {
 			want = linux.SECCOMP_RET_ALLOW
 		}
 		if got != want {

pkg/seccomp/seccomp_test_victim.go

@@ -29,76 +29,81 @@ func main() {
 	dieFlag := flag.Bool("die", false, "trips over the filter if true")
 	flag.Parse()
 
-	syscalls := []uintptr{
-		syscall.SYS_ACCEPT,
-		syscall.SYS_ARCH_PRCTL,
-		syscall.SYS_BIND,
-		syscall.SYS_BRK,
-		syscall.SYS_CLOCK_GETTIME,
-		syscall.SYS_CLONE,
-		syscall.SYS_CLOSE,
-		syscall.SYS_DUP,
-		syscall.SYS_DUP2,
-		syscall.SYS_EPOLL_CREATE1,
-		syscall.SYS_EPOLL_CTL,
-		syscall.SYS_EPOLL_WAIT,
-		syscall.SYS_EXIT,
-		syscall.SYS_EXIT_GROUP,
-		syscall.SYS_FALLOCATE,
-		syscall.SYS_FCHMOD,
-		syscall.SYS_FCNTL,
-		syscall.SYS_FSTAT,
-		syscall.SYS_FSYNC,
-		syscall.SYS_FTRUNCATE,
-		syscall.SYS_FUTEX,
-		syscall.SYS_GETDENTS64,
-		syscall.SYS_GETPEERNAME,
-		syscall.SYS_GETPID,
-		syscall.SYS_GETSOCKNAME,
-		syscall.SYS_GETSOCKOPT,
-		syscall.SYS_GETTID,
-		syscall.SYS_GETTIMEOFDAY,
-		syscall.SYS_LISTEN,
-		syscall.SYS_LSEEK,
-		syscall.SYS_MADVISE,
-		syscall.SYS_MINCORE,
-		syscall.SYS_MMAP,
-		syscall.SYS_MPROTECT,
-		syscall.SYS_MUNLOCK,
-		syscall.SYS_MUNMAP,
-		syscall.SYS_NANOSLEEP,
-		syscall.SYS_NEWFSTATAT,
-		syscall.SYS_OPEN,
-		syscall.SYS_POLL,
-		syscall.SYS_PREAD64,
-		syscall.SYS_PSELECT6,
-		syscall.SYS_PWRITE64,
-		syscall.SYS_READ,
-		syscall.SYS_READLINKAT,
-		syscall.SYS_READV,
-		syscall.SYS_RECVMSG,
-		syscall.SYS_RENAMEAT,
-		syscall.SYS_RESTART_SYSCALL,
-		syscall.SYS_RT_SIGACTION,
-		syscall.SYS_RT_SIGPROCMASK,
-		syscall.SYS_RT_SIGRETURN,
-		syscall.SYS_SCHED_YIELD,
-		syscall.SYS_SENDMSG,
-		syscall.SYS_SETITIMER,
-		syscall.SYS_SET_ROBUST_LIST,
-		syscall.SYS_SETSOCKOPT,
-		syscall.SYS_SHUTDOWN,
-		syscall.SYS_SIGALTSTACK,
-		syscall.SYS_SOCKET,
-		syscall.SYS_SYNC_FILE_RANGE,
-		syscall.SYS_TGKILL,
-		syscall.SYS_UTIMENSAT,
-		syscall.SYS_WRITE,
-		syscall.SYS_WRITEV,
+	syscalls := seccomp.SyscallRules{
+		syscall.SYS_ACCEPT:          {},
+		syscall.SYS_ARCH_PRCTL:      {},
+		syscall.SYS_BIND:            {},
+		syscall.SYS_BRK:             {},
+		syscall.SYS_CLOCK_GETTIME:   {},
+		syscall.SYS_CLONE:           {},
+		syscall.SYS_CLOSE:           {},
+		syscall.SYS_DUP:             {},
+		syscall.SYS_DUP2:            {},
+		syscall.SYS_EPOLL_CREATE1:   {},
+		syscall.SYS_EPOLL_CTL:       {},
+		syscall.SYS_EPOLL_WAIT:      {},
+		syscall.SYS_EPOLL_PWAIT:     {},
+		syscall.SYS_EXIT:            {},
+		syscall.SYS_EXIT_GROUP:      {},
+		syscall.SYS_FALLOCATE:       {},
+		syscall.SYS_FCHMOD:          {},
+		syscall.SYS_FCNTL:           {},
+		syscall.SYS_FSTAT:           {},
+		syscall.SYS_FSYNC:           {},
+		syscall.SYS_FTRUNCATE:       {},
+		syscall.SYS_FUTEX:           {},
+		syscall.SYS_GETDENTS64:      {},
+		syscall.SYS_GETPEERNAME:     {},
+		syscall.SYS_GETPID:          {},
+		syscall.SYS_GETSOCKNAME:     {},
+		syscall.SYS_GETSOCKOPT:      {},
+		syscall.SYS_GETTID:          {},
+		syscall.SYS_GETTIMEOFDAY:    {},
+		syscall.SYS_LISTEN:          {},
+		syscall.SYS_LSEEK:           {},
+		syscall.SYS_MADVISE:         {},
+		syscall.SYS_MINCORE:         {},
+		syscall.SYS_MMAP:            {},
+		syscall.SYS_MPROTECT:        {},
+		syscall.SYS_MUNLOCK:         {},
+		syscall.SYS_MUNMAP:          {},
+		syscall.SYS_NANOSLEEP:       {},
+		syscall.SYS_NEWFSTATAT:      {},
+		syscall.SYS_OPEN:            {},
+		syscall.SYS_POLL:            {},
+		syscall.SYS_PREAD64:         {},
+		syscall.SYS_PSELECT6:        {},
+		syscall.SYS_PWRITE64:        {},
+		syscall.SYS_READ:            {},
+		syscall.SYS_READLINKAT:      {},
+		syscall.SYS_READV:           {},
+		syscall.SYS_RECVMSG:         {},
+		syscall.SYS_RENAMEAT:        {},
+		syscall.SYS_RESTART_SYSCALL: {},
+		syscall.SYS_RT_SIGACTION:    {},
+		syscall.SYS_RT_SIGPROCMASK:  {},
+		syscall.SYS_RT_SIGRETURN:    {},
+		syscall.SYS_SCHED_YIELD:     {},
+		syscall.SYS_SENDMSG:         {},
+		syscall.SYS_SETITIMER:       {},
+		syscall.SYS_SET_ROBUST_LIST: {},
+		syscall.SYS_SETSOCKOPT:      {},
+		syscall.SYS_SHUTDOWN:        {},
+		syscall.SYS_SIGALTSTACK:     {},
+		syscall.SYS_SOCKET:          {},
+		syscall.SYS_SYNC_FILE_RANGE: {},
+		syscall.SYS_TGKILL:          {},
+		syscall.SYS_UTIMENSAT:       {},
+		syscall.SYS_WRITE:           {},
+		syscall.SYS_WRITEV:          {},
 	}
 	die := *dieFlag
 	if !die {
-		syscalls = append(syscalls, syscall.SYS_OPENAT)
+		syscalls[syscall.SYS_OPENAT] = []seccomp.Rule{
+			{
+				seccomp.AllowValue(10),
+			},
+		}
 	}
 
 	if err := seccomp.Install(syscalls, false); err != nil {
@@ -107,6 +112,6 @@ func main() {
 	}
 	fmt.Printf("Filters installed\n")
 
-	syscall.RawSyscall(syscall.SYS_OPENAT, 0, 0, 0)
+	syscall.RawSyscall(syscall.SYS_OPENAT, 10, 0, 0)
 	fmt.Printf("Syscall was allowed!!!\n")
 }

runsc/boot/filter/config.go

@@ -18,77 +18,78 @@ import (
 	"syscall"
 
 	"golang.org/x/sys/unix"
+	"gvisor.googlesource.com/gvisor/pkg/seccomp"
 )
 
 // allowedSyscalls is the set of syscalls executed by the Sentry
 // to the host OS.
-var allowedSyscalls = []uintptr{
-	syscall.SYS_ACCEPT,
-	syscall.SYS_ARCH_PRCTL,
-	syscall.SYS_CLOCK_GETTIME,
-	syscall.SYS_CLONE,
-	syscall.SYS_CLOSE,
-	syscall.SYS_DUP,
-	syscall.SYS_DUP2,
-	syscall.SYS_EPOLL_CREATE1,
-	syscall.SYS_EPOLL_CTL,
-	syscall.SYS_EPOLL_PWAIT,
-	syscall.SYS_EPOLL_WAIT,
-	syscall.SYS_EVENTFD2,
-	syscall.SYS_EXIT,
-	syscall.SYS_EXIT_GROUP,
-	syscall.SYS_FALLOCATE,
-	syscall.SYS_FCHMOD,
-	syscall.SYS_FCNTL,
-	syscall.SYS_FSTAT,
-	syscall.SYS_FSYNC,
-	syscall.SYS_FTRUNCATE,
-	syscall.SYS_FUTEX,
-	syscall.SYS_GETDENTS64,
-	syscall.SYS_GETPID,
-	unix.SYS_GETRANDOM,
-	syscall.SYS_GETSOCKOPT,
-	syscall.SYS_GETTID,
-	syscall.SYS_GETTIMEOFDAY,
-	syscall.SYS_LISTEN,
-	syscall.SYS_LSEEK,
-	syscall.SYS_MADVISE,
-	syscall.SYS_MINCORE,
-	syscall.SYS_MMAP,
-	syscall.SYS_MPROTECT,
-	syscall.SYS_MUNMAP,
-	syscall.SYS_NEWFSTATAT,
-	syscall.SYS_POLL,
-	syscall.SYS_PREAD64,
-	syscall.SYS_PSELECT6,
-	syscall.SYS_PWRITE64,
-	syscall.SYS_READ,
-	syscall.SYS_READLINKAT,
-	syscall.SYS_READV,
-	syscall.SYS_RECVMSG,
-	syscall.SYS_RENAMEAT,
-	syscall.SYS_RESTART_SYSCALL,
-	syscall.SYS_RT_SIGACTION,
-	syscall.SYS_RT_SIGPROCMASK,
-	syscall.SYS_RT_SIGRETURN,
-	syscall.SYS_SCHED_YIELD,
-	syscall.SYS_SENDMSG,
-	syscall.SYS_SETITIMER,
-	syscall.SYS_SHUTDOWN,
-	syscall.SYS_SIGALTSTACK,
-	syscall.SYS_SYNC_FILE_RANGE,
-	syscall.SYS_TGKILL,
-	syscall.SYS_UTIMENSAT,
-	syscall.SYS_WRITE,
-	syscall.SYS_WRITEV,
+var allowedSyscalls = seccomp.SyscallRules{
+	syscall.SYS_ACCEPT:          {},
+	syscall.SYS_ARCH_PRCTL:      {},
+	syscall.SYS_CLOCK_GETTIME:   {},
+	syscall.SYS_CLONE:           {},
+	syscall.SYS_CLOSE:           {},
+	syscall.SYS_DUP:             {},
+	syscall.SYS_DUP2:            {},
+	syscall.SYS_EPOLL_CREATE1:   {},
+	syscall.SYS_EPOLL_CTL:       {},
+	syscall.SYS_EPOLL_PWAIT:     {},
+	syscall.SYS_EPOLL_WAIT:      {},
+	syscall.SYS_EVENTFD2:        {},
+	syscall.SYS_EXIT:            {},
+	syscall.SYS_EXIT_GROUP:      {},
+	syscall.SYS_FALLOCATE:       {},
+	syscall.SYS_FCHMOD:          {},
+	syscall.SYS_FCNTL:           {},
+	syscall.SYS_FSTAT:           {},
+	syscall.SYS_FSYNC:           {},
+	syscall.SYS_FTRUNCATE:       {},
+	syscall.SYS_FUTEX:           {},
+	syscall.SYS_GETDENTS64:      {},
+	syscall.SYS_GETPID:          {},
+	unix.SYS_GETRANDOM:          {},
+	syscall.SYS_GETSOCKOPT:      {},
+	syscall.SYS_GETTID:          {},
+	syscall.SYS_GETTIMEOFDAY:    {},
+	syscall.SYS_LISTEN:          {},
+	syscall.SYS_LSEEK:           {},
+	syscall.SYS_MADVISE:         {},
+	syscall.SYS_MINCORE:         {},
+	syscall.SYS_MMAP:            {},
+	syscall.SYS_MPROTECT:        {},
+	syscall.SYS_MUNMAP:          {},
+	syscall.SYS_NEWFSTATAT:      {},
+	syscall.SYS_POLL:            {},
+	syscall.SYS_PREAD64:         {},
+	syscall.SYS_PSELECT6:        {},
+	syscall.SYS_PWRITE64:        {},
+	syscall.SYS_READ:            {},
+	syscall.SYS_READLINKAT:      {},
+	syscall.SYS_READV:           {},
+	syscall.SYS_RECVMSG:         {},
+	syscall.SYS_RENAMEAT:        {},
+	syscall.SYS_RESTART_SYSCALL: {},
+	syscall.SYS_RT_SIGACTION:    {},
+	syscall.SYS_RT_SIGPROCMASK:  {},
+	syscall.SYS_RT_SIGRETURN:    {},
+	syscall.SYS_SCHED_YIELD:     {},
+	syscall.SYS_SENDMSG:         {},
+	syscall.SYS_SETITIMER:       {},
+	syscall.SYS_SHUTDOWN:        {},
+	syscall.SYS_SIGALTSTACK:     {},
+	syscall.SYS_SYNC_FILE_RANGE: {},
+	syscall.SYS_TGKILL:          {},
+	syscall.SYS_UTIMENSAT:       {},
+	syscall.SYS_WRITE:           {},
+	syscall.SYS_WRITEV:          {},
 }
 
 // TODO: Ioctl is needed in order to support tty consoles.
 // Once filters support argument-checking, we should only allow ioctl
 // with tty-related arguments.
-func consoleFilters() []uintptr {
-	return []uintptr{
-		syscall.SYS_IOCTL,
+func consoleFilters() seccomp.SyscallRules {
+	return seccomp.SyscallRules{
+		syscall.SYS_IOCTL: {},
 	}
 }
 
@@ -97,79 +98,79 @@ func consoleFilters() []uintptr {
 // file operations that would otherwise be disabled by seccomp when a Gofer is
 // used. When whitelistFS is not used, openning new FD in the Sentry is
 // disallowed.
-func whitelistFSFilters() []uintptr {
-	return []uintptr{
-		syscall.SYS_ACCESS,
-		syscall.SYS_FCHMOD,
-		syscall.SYS_FSTAT,
-		syscall.SYS_FSYNC,
-		syscall.SYS_FTRUNCATE,
-		syscall.SYS_GETCWD,
-		syscall.SYS_GETDENTS,
-		syscall.SYS_GETDENTS64,
-		syscall.SYS_LSEEK,
-		syscall.SYS_LSTAT,
-		syscall.SYS_MKDIR,
-		syscall.SYS_MKDIRAT,
-		syscall.SYS_NEWFSTATAT,
-		syscall.SYS_OPEN,
-		syscall.SYS_OPENAT,
-		syscall.SYS_PREAD64,
-		syscall.SYS_PWRITE64,
-		syscall.SYS_READ,
-		syscall.SYS_READLINK,
-		syscall.SYS_READLINKAT,
-		syscall.SYS_RENAMEAT,
-		syscall.SYS_STAT,
-		syscall.SYS_SYMLINK,
-		syscall.SYS_SYMLINKAT,
-		syscall.SYS_SYNC_FILE_RANGE,
-		syscall.SYS_UNLINK,
-		syscall.SYS_UNLINKAT,
-		syscall.SYS_UTIMENSAT,
-		syscall.SYS_WRITE,
+func whitelistFSFilters() seccomp.SyscallRules {
+	return seccomp.SyscallRules{
+		syscall.SYS_ACCESS:          {},
+		syscall.SYS_FCHMOD:          {},
+		syscall.SYS_FSTAT:           {},
+		syscall.SYS_FSYNC:           {},
+		syscall.SYS_FTRUNCATE:       {},
+		syscall.SYS_GETCWD:          {},
+		syscall.SYS_GETDENTS:        {},
+		syscall.SYS_GETDENTS64:      {},
+		syscall.SYS_LSEEK:           {},
+		syscall.SYS_LSTAT:           {},
+		syscall.SYS_MKDIR:           {},
+		syscall.SYS_MKDIRAT:         {},
+		syscall.SYS_NEWFSTATAT:      {},
+		syscall.SYS_OPEN:            {},
+		syscall.SYS_OPENAT:          {},
+		syscall.SYS_PREAD64:         {},
+		syscall.SYS_PWRITE64:        {},
+		syscall.SYS_READ:            {},
+		syscall.SYS_READLINK:        {},
+		syscall.SYS_READLINKAT:      {},
+		syscall.SYS_RENAMEAT:        {},
+		syscall.SYS_STAT:            {},
+		syscall.SYS_SYMLINK:         {},
+		syscall.SYS_SYMLINKAT:       {},
+		syscall.SYS_SYNC_FILE_RANGE: {},
+		syscall.SYS_UNLINK:          {},
+		syscall.SYS_UNLINKAT:        {},
+		syscall.SYS_UTIMENSAT:       {},
+		syscall.SYS_WRITE:           {},
 	}
 }
 
 // hostInetFilters contains syscalls that are needed by sentry/socket/hostinet.
-func hostInetFilters() []uintptr {
-	return []uintptr{
-		syscall.SYS_ACCEPT4,
-		syscall.SYS_BIND,
-		syscall.SYS_CONNECT,
-		syscall.SYS_GETPEERNAME,
-		syscall.SYS_GETSOCKNAME,
-		syscall.SYS_GETSOCKOPT,
-		syscall.SYS_IOCTL,
-		syscall.SYS_LISTEN,
-		syscall.SYS_READV,
-		syscall.SYS_RECVFROM,
-		syscall.SYS_RECVMSG,
-		syscall.SYS_SENDMSG,
-		syscall.SYS_SENDTO,
-		syscall.SYS_SETSOCKOPT,
-		syscall.SYS_SHUTDOWN,
-		syscall.SYS_SOCKET,
-		syscall.SYS_WRITEV,
+func hostInetFilters() seccomp.SyscallRules {
+	return seccomp.SyscallRules{
+		syscall.SYS_ACCEPT4:     {},
+		syscall.SYS_BIND:        {},
+		syscall.SYS_CONNECT:     {},
+		syscall.SYS_GETPEERNAME: {},
+		syscall.SYS_GETSOCKNAME: {},
+		syscall.SYS_GETSOCKOPT:  {},
+		syscall.SYS_IOCTL:       {},
+		syscall.SYS_LISTEN:      {},
+		syscall.SYS_READV:       {},
+		syscall.SYS_RECVFROM:    {},
+		syscall.SYS_RECVMSG:     {},
+		syscall.SYS_SENDMSG:     {},
+		syscall.SYS_SENDTO:      {},
+		syscall.SYS_SETSOCKOPT:  {},
+		syscall.SYS_SHUTDOWN:    {},
+		syscall.SYS_SOCKET:      {},
+		syscall.SYS_WRITEV:      {},
 	}
 }
 
 // ptraceFilters returns syscalls made exclusively by the ptrace platform.
-func ptraceFilters() []uintptr {
-	return []uintptr{
-		syscall.SYS_PTRACE,
-		syscall.SYS_WAIT4,
-		unix.SYS_GETCPU,
-		unix.SYS_SCHED_SETAFFINITY,
+func ptraceFilters() seccomp.SyscallRules {
+	return seccomp.SyscallRules{
+		syscall.SYS_PTRACE:         {},
+		syscall.SYS_WAIT4:          {},
+		unix.SYS_GETCPU:            {},
+		unix.SYS_SCHED_SETAFFINITY: {},
 	}
 }
 
 // kvmFilters returns syscalls made exclusively by the KVM platform.
-func kvmFilters() []uintptr {
-	return []uintptr{
-		syscall.SYS_IOCTL,
-		syscall.SYS_RT_SIGSUSPEND,
-		syscall.SYS_RT_SIGTIMEDWAIT,
-		0xffffffffffffffff, // KVM uses syscall -1 to transition to host.
+func kvmFilters() seccomp.SyscallRules {
+	return seccomp.SyscallRules{
+		syscall.SYS_IOCTL:           {},
+		syscall.SYS_RT_SIGSUSPEND:   {},
+		syscall.SYS_RT_SIGTIMEDWAIT: {},
+		0xffffffffffffffff:          {}, // KVM uses syscall -1 to transition to host.
 	}
 }

runsc/boot/filter/extra_filters.go

@@ -16,9 +16,13 @@
 
 package filter
 
+import (
+	"gvisor.googlesource.com/gvisor/pkg/seccomp"
+)
+
 // instrumentationFilters returns additional filters for syscalls used by
 // Go intrumentation tools, e.g. -race, -msan.
 // Returns empty when disabled.
-func instrumentationFilters() []uintptr {
+func instrumentationFilters() seccomp.SyscallRules {
 	return nil
 }

runsc/boot/filter/extra_filters_msan.go

@@ -18,13 +18,15 @@ package filter
 
 import (
 	"syscall"
+
+	"gvisor.googlesource.com/gvisor/pkg/seccomp"
 )
 
 // instrumentationFilters returns additional filters for syscalls used by MSAN.
-func instrumentationFilters() []uintptr {
+func instrumentationFilters() seccomp.SyscallRules {
 	Report("MSAN is enabled: syscall filters less restrictive!")
-	return []uintptr{
-		syscall.SYS_SCHED_GETAFFINITY,
-		syscall.SYS_SET_ROBUST_LIST,
+	return seccomp.SyscallRules{
+		syscall.SYS_SCHED_GETAFFINITY: {},
+		syscall.SYS_SET_ROBUST_LIST:   {},
 	}
 }

runsc/boot/filter/extra_filters_race.go

@@ -18,16 +18,21 @@ package filter
 
 import (
 	"syscall"
+
+	"gvisor.googlesource.com/gvisor/pkg/seccomp"
 )
 
 // instrumentationFilters returns additional filters for syscalls used by TSAN.
-func instrumentationFilters() []uintptr {
+func instrumentationFilters() seccomp.SyscallRules {
 	Report("TSAN is enabled: syscall filters less restrictive!")
-	return []uintptr{
-		syscall.SYS_BRK,
-		syscall.SYS_MUNLOCK,
-		syscall.SYS_NANOSLEEP,
-		syscall.SYS_OPEN,
-		syscall.SYS_SET_ROBUST_LIST,
+	return seccomp.SyscallRules{
+		syscall.SYS_BRK:             {},
+		syscall.SYS_CLONE:           {},
+		syscall.SYS_FUTEX:           {},
+		syscall.SYS_MMAP:            {},
+		syscall.SYS_MUNLOCK:         {},
+		syscall.SYS_NANOSLEEP:       {},
+		syscall.SYS_OPEN:            {},
+		syscall.SYS_SET_ROBUST_LIST: {},
 	}
 }

runsc/boot/filter/filter.go

@@ -33,26 +33,26 @@ func Install(p platform.Platform, whitelistFS, console, hostNetwork bool) error
 
 	// Set of additional filters used by -race and -msan. Returns empty
 	// when not enabled.
-	s = append(s, instrumentationFilters()...)
+	s.Merge(instrumentationFilters())
 
 	if whitelistFS {
 		Report("direct file access allows unrestricted file access!")
-		s = append(s, whitelistFSFilters()...)
+		s.Merge(whitelistFSFilters())
 	}
 	if console {
 		Report("console is enabled: syscall filters less restrictive!")
-		s = append(s, consoleFilters()...)
+		s.Merge(consoleFilters())
 	}
 	if hostNetwork {
 		Report("host networking enabled: syscall filters less restrictive!")
-		s = append(s, hostInetFilters()...)
+		s.Merge(hostInetFilters())
 	}
 
 	switch p := p.(type) {
 	case *ptrace.PTrace:
-		s = append(s, ptraceFilters()...)
+		s.Merge(ptraceFilters())
 	case *kvm.KVM:
-		s = append(s, kvmFilters()...)
+		s.Merge(kvmFilters())
 	default:
 		return fmt.Errorf("unknown platform type %T", p)
 	}