Implement membarrier(2) commands other than *_SYNC_CORE.

Updates #267

PiperOrigin-RevId: 335713923

7 files changed, 78 insertions, 3 deletions

diff --git a/pkg/sentry/platform/BUILD b/pkg/sentry/platform/BUILD
index 209b28053..db7d55ef2 100644
--- a/pkg/sentry/platform/BUILD
+++ b/pkg/sentry/platform/BUILD
@@ -15,6 +15,7 @@ go_library(
         "//pkg/context",
         "//pkg/seccomp",
         "//pkg/sentry/arch",
+        "//pkg/sentry/hostmm",
         "//pkg/sentry/memmap",
         "//pkg/usermem",
     ],
diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD
index 323837fb1..9fe23c417 100644
--- a/pkg/sentry/platform/kvm/BUILD
+++ b/pkg/sentry/platform/kvm/BUILD
@@ -56,6 +56,7 @@ go_library(
         "//pkg/sentry/time",
         "//pkg/sync",
         "//pkg/usermem",
+        "@org_golang_x_sys//unix:go_default_library",
     ],
 )
 
diff --git a/pkg/sentry/platform/kvm/filters_amd64.go b/pkg/sentry/platform/kvm/filters_amd64.go
index 7d949f1dd..d3d216aa5 100644
--- a/pkg/sentry/platform/kvm/filters_amd64.go
+++ b/pkg/sentry/platform/kvm/filters_amd64.go
@@ -17,14 +17,23 @@ package kvm
 import (
 	"syscall"
 
+	"golang.org/x/sys/unix"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/seccomp"
 )
 
 // SyscallFilters returns syscalls made exclusively by the KVM platform.
 func (*KVM) SyscallFilters() seccomp.SyscallRules {
 	return seccomp.SyscallRules{
-		syscall.SYS_ARCH_PRCTL:      {},
-		syscall.SYS_IOCTL:           {},
+		syscall.SYS_ARCH_PRCTL: {},
+		syscall.SYS_IOCTL:      {},
+		unix.SYS_MEMBARRIER: []seccomp.Rule{
+			{
+				seccomp.EqualTo(linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED),
+				seccomp.EqualTo(0),
+			},
+		},
 		syscall.SYS_MMAP:            {},
 		syscall.SYS_RT_SIGSUSPEND:   {},
 		syscall.SYS_RT_SIGTIMEDWAIT: {},
diff --git a/pkg/sentry/platform/kvm/filters_arm64.go b/pkg/sentry/platform/kvm/filters_arm64.go
index 9245d07c2..21abc2a3d 100644
--- a/pkg/sentry/platform/kvm/filters_arm64.go
+++ b/pkg/sentry/platform/kvm/filters_arm64.go
@@ -17,13 +17,22 @@ package kvm
 import (
 	"syscall"
 
+	"golang.org/x/sys/unix"
+
+	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/seccomp"
 )
 
 // SyscallFilters returns syscalls made exclusively by the KVM platform.
 func (*KVM) SyscallFilters() seccomp.SyscallRules {
 	return seccomp.SyscallRules{
-		syscall.SYS_IOCTL:           {},
+		syscall.SYS_IOCTL: {},
+		unix.SYS_MEMBARRIER: []seccomp.Rule{
+			{
+				seccomp.EqualTo(linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED),
+				seccomp.EqualTo(0),
+			},
+		},
 		syscall.SYS_MMAP:            {},
 		syscall.SYS_RT_SIGSUSPEND:   {},
 		syscall.SYS_RT_SIGTIMEDWAIT: {},
diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go
index d46946402..dd45ad10b 100644
--- a/pkg/sentry/platform/kvm/kvm.go
+++ b/pkg/sentry/platform/kvm/kvm.go
@@ -63,6 +63,9 @@ type runData struct {
 type KVM struct {
 	platform.NoCPUPreemptionDetection
 
+	// KVM never changes mm_structs.
+	platform.UseHostProcessMemoryBarrier
+
 	// machine is the backing VM.
 	machine *machine
 }
diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go
index 530e779b0..dcfe839a7 100644
--- a/pkg/sentry/platform/platform.go
+++ b/pkg/sentry/platform/platform.go
@@ -25,6 +25,7 @@ import (
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/seccomp"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
+	"gvisor.dev/gvisor/pkg/sentry/hostmm"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
 	"gvisor.dev/gvisor/pkg/usermem"
 )
@@ -52,6 +53,10 @@ type Platform interface {
 	// can reliably return ErrContextCPUPreempted.
 	DetectsCPUPreemption() bool
 
+	// HaveGlobalMemoryBarrier returns true if the GlobalMemoryBarrier method
+	// is supported.
+	HaveGlobalMemoryBarrier() bool
+
 	// MapUnit returns the alignment used for optional mappings into this
 	// platform's AddressSpaces. Higher values indicate lower per-page costs
 	// for AddressSpace.MapFile. As a special case, a MapUnit of 0 indicates
@@ -97,6 +102,15 @@ type Platform interface {
 	// called.
 	PreemptAllCPUs() error
 
+	// GlobalMemoryBarrier blocks until all threads running application code
+	// (via Context.Switch) and all task goroutines "have passed through a
+	// state where all memory accesses to user-space addresses match program
+	// order between entry to and return from [GlobalMemoryBarrier]", as for
+	// membarrier(2).
+	//
+	// Preconditions: HaveGlobalMemoryBarrier() == true.
+	GlobalMemoryBarrier() error
+
 	// SyscallFilters returns syscalls made exclusively by this platform.
 	SyscallFilters() seccomp.SyscallRules
 }
@@ -115,6 +129,43 @@ func (NoCPUPreemptionDetection) PreemptAllCPUs() error {
 	panic("This platform does not support CPU preemption detection")
 }
 
+// UseHostGlobalMemoryBarrier implements Platform.HaveGlobalMemoryBarrier and
+// Platform.GlobalMemoryBarrier by invoking equivalent functionality on the
+// host.
+type UseHostGlobalMemoryBarrier struct{}
+
+// HaveGlobalMemoryBarrier implements Platform.HaveGlobalMemoryBarrier.
+func (UseHostGlobalMemoryBarrier) HaveGlobalMemoryBarrier() bool {
+	return hostmm.HaveGlobalMemoryBarrier()
+}
+
+// GlobalMemoryBarrier implements Platform.GlobalMemoryBarrier.
+func (UseHostGlobalMemoryBarrier) GlobalMemoryBarrier() error {
+	return hostmm.GlobalMemoryBarrier()
+}
+
+// UseHostProcessMemoryBarrier implements Platform.HaveGlobalMemoryBarrier and
+// Platform.GlobalMemoryBarrier by invoking a process-local memory barrier.
+// This is faster than UseHostGlobalMemoryBarrier, but is only appropriate for
+// platforms for which application code executes while using the sentry's
+// mm_struct.
+type UseHostProcessMemoryBarrier struct{}
+
+// HaveGlobalMemoryBarrier implements Platform.HaveGlobalMemoryBarrier.
+func (UseHostProcessMemoryBarrier) HaveGlobalMemoryBarrier() bool {
+	// Fall back to a global memory barrier if a process-local one isn't
+	// available.
+	return hostmm.HaveProcessMemoryBarrier() || hostmm.HaveGlobalMemoryBarrier()
+}
+
+// GlobalMemoryBarrier implements Platform.GlobalMemoryBarrier.
+func (UseHostProcessMemoryBarrier) GlobalMemoryBarrier() error {
+	if hostmm.HaveProcessMemoryBarrier() {
+		return hostmm.ProcessMemoryBarrier()
+	}
+	return hostmm.GlobalMemoryBarrier()
+}
+
 // MemoryManager represents an abstraction above the platform address space
 // which manages memory mappings and their contents.
 type MemoryManager interface {
diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go
index b52d0fbd8..f56aa3b79 100644
--- a/pkg/sentry/platform/ptrace/ptrace.go
+++ b/pkg/sentry/platform/ptrace/ptrace.go
@@ -192,6 +192,7 @@ func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {}
 type PTrace struct {
 	platform.MMapMinAddr
 	platform.NoCPUPreemptionDetection
+	platform.UseHostGlobalMemoryBarrier
 }
 
 // New returns a new ptrace-based implementation of the platform interface.