Allow Sentry to kill itself even when it is the init process

PiperOrigin-RevId: 733399678

pkg/sentry/platform/ptrace/BUILD

@@ -41,6 +41,7 @@ go_library(
         "//pkg/sentry/memmap",
         "//pkg/sentry/platform",
         "//pkg/sentry/platform/interrupt",
+        "//pkg/sighandling",
         "//pkg/sync",
         "@org_golang_x_sys//unix:go_default_library",
     ],

pkg/sentry/platform/ptrace/filters.go

@@ -15,6 +15,8 @@
 package ptrace
 
 import (
+	"os"
+
 	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/seccomp"
 	"gvisor.dev/gvisor/pkg/sentry/platform"
@@ -26,6 +28,11 @@ func (*PTrace) SeccompInfo() platform.SeccompInfo {
 		PlatformName: "ptrace",
 		Filters: seccomp.MakeSyscallRules(map[uintptr]seccomp.SyscallRule{
 			unix.SYS_PTRACE: seccomp.MatchAll{},
+			unix.SYS_RT_TGSIGQUEUEINFO: seccomp.PerArg{
+				seccomp.EqualTo(os.Getpid()),
+				seccomp.AnyValue{}, // tid
+				seccomp.EqualTo(unix.SIGKILL),
+			},
 			unix.SYS_TGKILL: seccomp.MatchAll{},
 			unix.SYS_WAIT4:  seccomp.MatchAll{},
 		}),

pkg/sentry/platform/ptrace/subprocess.go

@@ -27,6 +27,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
 	"gvisor.dev/gvisor/pkg/sentry/platform"
+	"gvisor.dev/gvisor/pkg/sighandling"
 	"gvisor.dev/gvisor/pkg/sync"
 )
 
@@ -357,8 +358,7 @@ func (t *thread) unexpectedStubExit() {
 		// these cases, we don't need to panic. There is no reasons to
 		// think that something wrong in gVisor.
 		log.Warningf("The ptrace stub process %v has been killed by SIGKILL.", t.tgid)
-		pid := os.Getpid()
-		unix.Tgkill(pid, pid, unix.Signal(unix.SIGKILL))
+		sighandling.KillItself()
 	}
 	t.dumpAndPanic(fmt.Sprintf("wait failed: the process %d:%d exited: %x (err %v)", t.tgid, t.tid, msg, err))
 }

pkg/sentry/platform/systrap/BUILD

@@ -99,6 +99,7 @@ go_library(
         "//pkg/sentry/platform/systrap/sysmsg",
         "//pkg/sentry/platform/systrap/usertrap",
         "//pkg/sentry/usage",
+        "//pkg/sighandling",
         "//pkg/sync",
         "//pkg/syncevent",
         "@org_golang_x_sys//unix:go_default_library",

pkg/sentry/platform/systrap/filters.go

@@ -15,6 +15,8 @@
 package systrap
 
 import (
+	"os"
+
 	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/seccomp"
@@ -93,7 +95,12 @@ func (systrapSeccomp) SyscallFilters(vars precompiledseccomp.Values) seccomp.Sys
 			},
 		},
 		unix.SYS_TGKILL: seccomp.MatchAll{},
-		unix.SYS_WAIT4:  seccomp.MatchAll{},
+		unix.SYS_RT_TGSIGQUEUEINFO: seccomp.PerArg{
+			seccomp.EqualTo(os.Getpid()),
+			seccomp.AnyValue{}, // tid
+			seccomp.EqualTo(unix.SIGKILL),
+		},
+		unix.SYS_WAIT4: seccomp.MatchAll{},
 		unix.SYS_IOCTL: seccomp.Or{
 			seccomp.PerArg{
 				seccomp.NonNegativeFD{},

pkg/sentry/platform/systrap/subprocess.go

@@ -16,7 +16,6 @@ package systrap
 
 import (
 	"fmt"
-	"os"
 	"runtime"
 	"sync"
 	"sync/atomic"
@@ -36,6 +35,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/platform/systrap/sysmsg"
 	"gvisor.dev/gvisor/pkg/sentry/platform/systrap/usertrap"
 	"gvisor.dev/gvisor/pkg/sentry/usage"
+	"gvisor.dev/gvisor/pkg/sighandling"
 )
 
 var (
@@ -609,8 +609,7 @@ func (t *thread) unexpectedStubExit() {
 		// these cases, we don't need to panic. There is no reasons to
 		// think that something wrong in gVisor.
 		log.Warningf("The ptrace stub process %v has been killed by SIGKILL.", t.tgid)
-		pid := os.Getpid()
-		unix.Tgkill(pid, pid, unix.Signal(unix.SIGKILL))
+		sighandling.KillItself()
 	}
 	t.dumpAndPanic(fmt.Sprintf("wait failed: the process %d:%d exited: %x (err %v)", t.tgid, t.tid, msg, err))
 }

pkg/sighandling/sighandling_linux_unsafe.go

@@ -19,6 +19,7 @@ package sighandling
 
 import (
 	"fmt"
+	"os"
 	"unsafe"
 
 	"golang.org/x/sys/unix"
@@ -74,3 +75,29 @@ func ReplaceSignalHandler(sig unix.Signal, handler uintptr, previous *uintptr) e
 
 	return nil
 }
+
+// KillItself sends SIGKILL to the current process, bypassing the init process
+// restriction.
+//
+// The standard `kill(getpid(), SIGKILL)` syscall doesn't work when the current
+// process is the init process within its PID namespace. This is a "known"
+// Linux feature.
+//
+// This function uses the rt_tgqueueinfo syscall to send a "kernel-generated"
+// SIGKILL.
+func KillItself() error {
+	pid := os.Getpid()
+	tid, _, _ := unix.RawSyscall(unix.SYS_GETTID, 0, 0, 0)
+	info := linux.SignalInfo{Code: linux.SI_KERNEL}
+	// The current thread can send a fake kernel siginfo to itself.
+	if _, _, e := unix.RawSyscall6(
+		unix.SYS_RT_TGSIGQUEUEINFO,
+		uintptr(pid), uintptr(tid),
+		uintptr(linux.SIGKILL),
+		uintptr(unsafe.Pointer(&info)),
+		0, 0,
+	); e != 0 {
+		return e
+	}
+	panic("unreachable")
+}