Merge branch 'iptables-write-input-drop' into iptables-write-filter-proto

18 files changed, 1175 insertions, 0 deletions

diff --git a/pkg/sync/BUILD b/pkg/sync/BUILD
new file mode 100644
index 000000000..e8cd16b8f
--- /dev/null
+++ b/pkg/sync/BUILD
@@ -0,0 +1,53 @@
+load("//tools/go_stateify:defs.bzl", "go_library")
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+load("//tools/go_generics:defs.bzl", "go_template")
+
+package(
+    default_visibility = ["//:sandbox"],
+    licenses = ["notice"],
+)
+
+exports_files(["LICENSE"])
+
+go_template(
+    name = "generic_atomicptr",
+    srcs = ["atomicptr_unsafe.go"],
+    types = [
+        "Value",
+    ],
+)
+
+go_template(
+    name = "generic_seqatomic",
+    srcs = ["seqatomic_unsafe.go"],
+    types = [
+        "Value",
+    ],
+    deps = [
+        ":sync",
+    ],
+)
+
+go_library(
+    name = "sync",
+    srcs = [
+        "aliases.go",
+        "downgradable_rwmutex_unsafe.go",
+        "memmove_unsafe.go",
+        "norace_unsafe.go",
+        "race_unsafe.go",
+        "seqcount.go",
+        "syncutil.go",
+    ],
+    importpath = "gvisor.dev/gvisor/pkg/sync",
+)
+
+go_test(
+    name = "sync_test",
+    size = "small",
+    srcs = [
+        "downgradable_rwmutex_test.go",
+        "seqcount_test.go",
+    ],
+    embed = [":sync"],
+)
diff --git a/pkg/sync/LICENSE b/pkg/sync/LICENSE
new file mode 100644
index 000000000..6a66aea5e
--- /dev/null
+++ b/pkg/sync/LICENSE
@@ -0,0 +1,27 @@
+Copyright (c) 2009 The Go Authors. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   * Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+   * Redistributions in binary form must reproduce the above
+copyright notice, this list of conditions and the following disclaimer
+in the documentation and/or other materials provided with the
+distribution.
+   * Neither the name of Google Inc. nor the names of its
+contributors may be used to endorse or promote products derived from
+this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/pkg/sync/README.md b/pkg/sync/README.md
new file mode 100644
index 000000000..2183c4e20
--- /dev/null
+++ b/pkg/sync/README.md
@@ -0,0 +1,5 @@
+# Syncutil
+
+This package provides additional synchronization primitives not provided by the
+Go stdlib 'sync' package. It is partially derived from the upstream 'sync'
+package from go1.10.
diff --git a/pkg/sync/aliases.go b/pkg/sync/aliases.go
new file mode 100644
index 000000000..20c7ca041
--- /dev/null
+++ b/pkg/sync/aliases.go
@@ -0,0 +1,37 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package sync
+
+import (
+	"sync"
+)
+
+// Aliases of standard library types.
+type (
+	// Mutex is an alias of sync.Mutex.
+	Mutex = sync.Mutex
+
+	// RWMutex is an alias of sync.RWMutex.
+	RWMutex = sync.RWMutex
+
+	// Cond is an alias of sync.Cond.
+	Cond = sync.Cond
+
+	// Locker is an alias of sync.Locker.
+	Locker = sync.Locker
+
+	// Once is an alias of sync.Once.
+	Once = sync.Once
+
+	// Pool is an alias of sync.Pool.
+	Pool = sync.Pool
+
+	// WaitGroup is an alias of sync.WaitGroup.
+	WaitGroup = sync.WaitGroup
+
+	// Map is an alias of sync.Map.
+	Map = sync.Map
+)
diff --git a/pkg/sync/atomicptr_unsafe.go b/pkg/sync/atomicptr_unsafe.go
new file mode 100644
index 000000000..525c4beed
--- /dev/null
+++ b/pkg/sync/atomicptr_unsafe.go
@@ -0,0 +1,47 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package template doesn't exist. This file must be instantiated using the
+// go_template_instance rule in tools/go_generics/defs.bzl.
+package template
+
+import (
+	"sync/atomic"
+	"unsafe"
+)
+
+// Value is a required type parameter.
+type Value struct{}
+
+// An AtomicPtr is a pointer to a value of type Value that can be atomically
+// loaded and stored. The zero value of an AtomicPtr represents nil.
+//
+// Note that copying AtomicPtr by value performs a non-atomic read of the
+// stored pointer, which is unsafe if Store() can be called concurrently; in
+// this case, do `dst.Store(src.Load())` instead.
+//
+// +stateify savable
+type AtomicPtr struct {
+	ptr unsafe.Pointer `state:".(*Value)"`
+}
+
+func (p *AtomicPtr) savePtr() *Value {
+	return p.Load()
+}
+
+func (p *AtomicPtr) loadPtr(v *Value) {
+	p.Store(v)
+}
+
+// Load returns the value set by the most recent Store. It returns nil if there
+// has been no previous call to Store.
+func (p *AtomicPtr) Load() *Value {
+	return (*Value)(atomic.LoadPointer(&p.ptr))
+}
+
+// Store sets the value returned by Load to x.
+func (p *AtomicPtr) Store(x *Value) {
+	atomic.StorePointer(&p.ptr, (unsafe.Pointer)(x))
+}
diff --git a/pkg/sync/atomicptrtest/BUILD b/pkg/sync/atomicptrtest/BUILD
new file mode 100644
index 000000000..418eda29c
--- /dev/null
+++ b/pkg/sync/atomicptrtest/BUILD
@@ -0,0 +1,29 @@
+load("//tools/go_stateify:defs.bzl", "go_library")
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+load("//tools/go_generics:defs.bzl", "go_template_instance")
+
+package(licenses = ["notice"])
+
+go_template_instance(
+    name = "atomicptr_int",
+    out = "atomicptr_int_unsafe.go",
+    package = "atomicptr",
+    suffix = "Int",
+    template = "//pkg/sync:generic_atomicptr",
+    types = {
+        "Value": "int",
+    },
+)
+
+go_library(
+    name = "atomicptr",
+    srcs = ["atomicptr_int_unsafe.go"],
+    importpath = "gvisor.dev/gvisor/pkg/sync/atomicptr",
+)
+
+go_test(
+    name = "atomicptr_test",
+    size = "small",
+    srcs = ["atomicptr_test.go"],
+    embed = [":atomicptr"],
+)
diff --git a/pkg/sync/atomicptrtest/atomicptr_test.go b/pkg/sync/atomicptrtest/atomicptr_test.go
new file mode 100644
index 000000000..8fdc5112e
--- /dev/null
+++ b/pkg/sync/atomicptrtest/atomicptr_test.go
@@ -0,0 +1,31 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package atomicptr
+
+import (
+	"testing"
+)
+
+func newInt(val int) *int {
+	return &val
+}
+
+func TestAtomicPtr(t *testing.T) {
+	var p AtomicPtrInt
+	if got := p.Load(); got != nil {
+		t.Errorf("initial value is %p (%v), wanted nil", got, got)
+	}
+	want := newInt(42)
+	p.Store(want)
+	if got := p.Load(); got != want {
+		t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want)
+	}
+	want = newInt(100)
+	p.Store(want)
+	if got := p.Load(); got != want {
+		t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want)
+	}
+}
diff --git a/pkg/sync/downgradable_rwmutex_test.go b/pkg/sync/downgradable_rwmutex_test.go
new file mode 100644
index 000000000..f04496bc5
--- /dev/null
+++ b/pkg/sync/downgradable_rwmutex_test.go
@@ -0,0 +1,150 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Copyright 2019 The gVisor Authors.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// GOMAXPROCS=10 go test
+
+// Copy/pasted from the standard library's sync/rwmutex_test.go, except for the
+// addition of downgradingWriter and the renaming of num_iterations to
+// numIterations to shut up Golint.
+
+package sync
+
+import (
+	"fmt"
+	"runtime"
+	"sync/atomic"
+	"testing"
+)
+
+func parallelReader(m *DowngradableRWMutex, clocked, cunlock, cdone chan bool) {
+	m.RLock()
+	clocked <- true
+	<-cunlock
+	m.RUnlock()
+	cdone <- true
+}
+
+func doTestParallelReaders(numReaders, gomaxprocs int) {
+	runtime.GOMAXPROCS(gomaxprocs)
+	var m DowngradableRWMutex
+	clocked := make(chan bool)
+	cunlock := make(chan bool)
+	cdone := make(chan bool)
+	for i := 0; i < numReaders; i++ {
+		go parallelReader(&m, clocked, cunlock, cdone)
+	}
+	// Wait for all parallel RLock()s to succeed.
+	for i := 0; i < numReaders; i++ {
+		<-clocked
+	}
+	for i := 0; i < numReaders; i++ {
+		cunlock <- true
+	}
+	// Wait for the goroutines to finish.
+	for i := 0; i < numReaders; i++ {
+		<-cdone
+	}
+}
+
+func TestParallelReaders(t *testing.T) {
+	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1))
+	doTestParallelReaders(1, 4)
+	doTestParallelReaders(3, 4)
+	doTestParallelReaders(4, 2)
+}
+
+func reader(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) {
+	for i := 0; i < numIterations; i++ {
+		rwm.RLock()
+		n := atomic.AddInt32(activity, 1)
+		if n < 1 || n >= 10000 {
+			panic(fmt.Sprintf("wlock(%d)\n", n))
+		}
+		for i := 0; i < 100; i++ {
+		}
+		atomic.AddInt32(activity, -1)
+		rwm.RUnlock()
+	}
+	cdone <- true
+}
+
+func writer(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) {
+	for i := 0; i < numIterations; i++ {
+		rwm.Lock()
+		n := atomic.AddInt32(activity, 10000)
+		if n != 10000 {
+			panic(fmt.Sprintf("wlock(%d)\n", n))
+		}
+		for i := 0; i < 100; i++ {
+		}
+		atomic.AddInt32(activity, -10000)
+		rwm.Unlock()
+	}
+	cdone <- true
+}
+
+func downgradingWriter(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) {
+	for i := 0; i < numIterations; i++ {
+		rwm.Lock()
+		n := atomic.AddInt32(activity, 10000)
+		if n != 10000 {
+			panic(fmt.Sprintf("wlock(%d)\n", n))
+		}
+		for i := 0; i < 100; i++ {
+		}
+		atomic.AddInt32(activity, -10000)
+		rwm.DowngradeLock()
+		n = atomic.AddInt32(activity, 1)
+		if n < 1 || n >= 10000 {
+			panic(fmt.Sprintf("wlock(%d)\n", n))
+		}
+		for i := 0; i < 100; i++ {
+		}
+		n = atomic.AddInt32(activity, -1)
+		rwm.RUnlock()
+	}
+	cdone <- true
+}
+
+func HammerDowngradableRWMutex(gomaxprocs, numReaders, numIterations int) {
+	runtime.GOMAXPROCS(gomaxprocs)
+	// Number of active readers + 10000 * number of active writers.
+	var activity int32
+	var rwm DowngradableRWMutex
+	cdone := make(chan bool)
+	go writer(&rwm, numIterations, &activity, cdone)
+	go downgradingWriter(&rwm, numIterations, &activity, cdone)
+	var i int
+	for i = 0; i < numReaders/2; i++ {
+		go reader(&rwm, numIterations, &activity, cdone)
+	}
+	go writer(&rwm, numIterations, &activity, cdone)
+	go downgradingWriter(&rwm, numIterations, &activity, cdone)
+	for ; i < numReaders; i++ {
+		go reader(&rwm, numIterations, &activity, cdone)
+	}
+	// Wait for the 4 writers and all readers to finish.
+	for i := 0; i < 4+numReaders; i++ {
+		<-cdone
+	}
+}
+
+func TestDowngradableRWMutex(t *testing.T) {
+	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1))
+	n := 1000
+	if testing.Short() {
+		n = 5
+	}
+	HammerDowngradableRWMutex(1, 1, n)
+	HammerDowngradableRWMutex(1, 3, n)
+	HammerDowngradableRWMutex(1, 10, n)
+	HammerDowngradableRWMutex(4, 1, n)
+	HammerDowngradableRWMutex(4, 3, n)
+	HammerDowngradableRWMutex(4, 10, n)
+	HammerDowngradableRWMutex(10, 1, n)
+	HammerDowngradableRWMutex(10, 3, n)
+	HammerDowngradableRWMutex(10, 10, n)
+	HammerDowngradableRWMutex(10, 5, n)
+}
diff --git a/pkg/sync/downgradable_rwmutex_unsafe.go b/pkg/sync/downgradable_rwmutex_unsafe.go
new file mode 100644
index 000000000..9bb55cd3a
--- /dev/null
+++ b/pkg/sync/downgradable_rwmutex_unsafe.go
@@ -0,0 +1,146 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Copyright 2019 The gVisor Authors.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build go1.13
+// +build !go1.15
+
+// Check go:linkname function signatures when updating Go version.
+
+// This is mostly copied from the standard library's sync/rwmutex.go.
+//
+// Happens-before relationships indicated to the race detector:
+// - Unlock -> Lock (via writerSem)
+// - Unlock -> RLock (via readerSem)
+// - RUnlock -> Lock (via writerSem)
+// - DowngradeLock -> RLock (via readerSem)
+
+package sync
+
+import (
+	"sync"
+	"sync/atomic"
+	"unsafe"
+)
+
+//go:linkname runtimeSemacquire sync.runtime_Semacquire
+func runtimeSemacquire(s *uint32)
+
+//go:linkname runtimeSemrelease sync.runtime_Semrelease
+func runtimeSemrelease(s *uint32, handoff bool, skipframes int)
+
+// DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock
+// method.
+type DowngradableRWMutex struct {
+	w           sync.Mutex // held if there are pending writers
+	writerSem   uint32     // semaphore for writers to wait for completing readers
+	readerSem   uint32     // semaphore for readers to wait for completing writers
+	readerCount int32      // number of pending readers
+	readerWait  int32      // number of departing readers
+}
+
+const rwmutexMaxReaders = 1 << 30
+
+// RLock locks rw for reading.
+func (rw *DowngradableRWMutex) RLock() {
+	if RaceEnabled {
+		RaceDisable()
+	}
+	if atomic.AddInt32(&rw.readerCount, 1) < 0 {
+		// A writer is pending, wait for it.
+		runtimeSemacquire(&rw.readerSem)
+	}
+	if RaceEnabled {
+		RaceEnable()
+		RaceAcquire(unsafe.Pointer(&rw.readerSem))
+	}
+}
+
+// RUnlock undoes a single RLock call.
+func (rw *DowngradableRWMutex) RUnlock() {
+	if RaceEnabled {
+		RaceReleaseMerge(unsafe.Pointer(&rw.writerSem))
+		RaceDisable()
+	}
+	if r := atomic.AddInt32(&rw.readerCount, -1); r < 0 {
+		if r+1 == 0 || r+1 == -rwmutexMaxReaders {
+			panic("RUnlock of unlocked DowngradableRWMutex")
+		}
+		// A writer is pending.
+		if atomic.AddInt32(&rw.readerWait, -1) == 0 {
+			// The last reader unblocks the writer.
+			runtimeSemrelease(&rw.writerSem, false, 0)
+		}
+	}
+	if RaceEnabled {
+		RaceEnable()
+	}
+}
+
+// Lock locks rw for writing.
+func (rw *DowngradableRWMutex) Lock() {
+	if RaceEnabled {
+		RaceDisable()
+	}
+	// First, resolve competition with other writers.
+	rw.w.Lock()
+	// Announce to readers there is a pending writer.
+	r := atomic.AddInt32(&rw.readerCount, -rwmutexMaxReaders) + rwmutexMaxReaders
+	// Wait for active readers.
+	if r != 0 && atomic.AddInt32(&rw.readerWait, r) != 0 {
+		runtimeSemacquire(&rw.writerSem)
+	}
+	if RaceEnabled {
+		RaceEnable()
+		RaceAcquire(unsafe.Pointer(&rw.writerSem))
+	}
+}
+
+// Unlock unlocks rw for writing.
+func (rw *DowngradableRWMutex) Unlock() {
+	if RaceEnabled {
+		RaceRelease(unsafe.Pointer(&rw.writerSem))
+		RaceRelease(unsafe.Pointer(&rw.readerSem))
+		RaceDisable()
+	}
+	// Announce to readers there is no active writer.
+	r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders)
+	if r >= rwmutexMaxReaders {
+		panic("Unlock of unlocked DowngradableRWMutex")
+	}
+	// Unblock blocked readers, if any.
+	for i := 0; i < int(r); i++ {
+		runtimeSemrelease(&rw.readerSem, false, 0)
+	}
+	// Allow other writers to proceed.
+	rw.w.Unlock()
+	if RaceEnabled {
+		RaceEnable()
+	}
+}
+
+// DowngradeLock atomically unlocks rw for writing and locks it for reading.
+func (rw *DowngradableRWMutex) DowngradeLock() {
+	if RaceEnabled {
+		RaceRelease(unsafe.Pointer(&rw.readerSem))
+		RaceDisable()
+	}
+	// Announce to readers there is no active writer and one additional reader.
+	r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders+1)
+	if r >= rwmutexMaxReaders+1 {
+		panic("DowngradeLock of unlocked DowngradableRWMutex")
+	}
+	// Unblock blocked readers, if any. Note that this loop starts as 1 since r
+	// includes this goroutine.
+	for i := 1; i < int(r); i++ {
+		runtimeSemrelease(&rw.readerSem, false, 0)
+	}
+	// Allow other writers to proceed to rw.w.Lock(). Note that they will still
+	// block on rw.writerSem since at least this reader exists, such that
+	// DowngradeLock() is atomic with the previous write lock.
+	rw.w.Unlock()
+	if RaceEnabled {
+		RaceEnable()
+	}
+}
diff --git a/pkg/sync/memmove_unsafe.go b/pkg/sync/memmove_unsafe.go
new file mode 100644
index 000000000..ad4a3a37e
--- /dev/null
+++ b/pkg/sync/memmove_unsafe.go
@@ -0,0 +1,28 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build go1.12
+// +build !go1.15
+
+// Check go:linkname function signatures when updating Go version.
+
+package sync
+
+import (
+	"unsafe"
+)
+
+//go:linkname memmove runtime.memmove
+//go:noescape
+func memmove(to, from unsafe.Pointer, n uintptr)
+
+// Memmove is exported for SeqAtomicLoad/SeqAtomicTryLoad<T>, which can't
+// define it because go_generics can't update the go:linkname annotation.
+// Furthermore, go:linkname silently doesn't work if the local name is exported
+// (this is of course undocumented), which is why this indirection is
+// necessary.
+func Memmove(to, from unsafe.Pointer, n uintptr) {
+	memmove(to, from, n)
+}
diff --git a/pkg/sync/norace_unsafe.go b/pkg/sync/norace_unsafe.go
new file mode 100644
index 000000000..006055dd6
--- /dev/null
+++ b/pkg/sync/norace_unsafe.go
@@ -0,0 +1,35 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build !race
+
+package sync
+
+import (
+	"unsafe"
+)
+
+// RaceEnabled is true if the Go data race detector is enabled.
+const RaceEnabled = false
+
+// RaceDisable has the same semantics as runtime.RaceDisable.
+func RaceDisable() {
+}
+
+// RaceEnable has the same semantics as runtime.RaceEnable.
+func RaceEnable() {
+}
+
+// RaceAcquire has the same semantics as runtime.RaceAcquire.
+func RaceAcquire(addr unsafe.Pointer) {
+}
+
+// RaceRelease has the same semantics as runtime.RaceRelease.
+func RaceRelease(addr unsafe.Pointer) {
+}
+
+// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge.
+func RaceReleaseMerge(addr unsafe.Pointer) {
+}
diff --git a/pkg/sync/race_unsafe.go b/pkg/sync/race_unsafe.go
new file mode 100644
index 000000000..31d8fa9a6
--- /dev/null
+++ b/pkg/sync/race_unsafe.go
@@ -0,0 +1,41 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build race
+
+package sync
+
+import (
+	"runtime"
+	"unsafe"
+)
+
+// RaceEnabled is true if the Go data race detector is enabled.
+const RaceEnabled = true
+
+// RaceDisable has the same semantics as runtime.RaceDisable.
+func RaceDisable() {
+	runtime.RaceDisable()
+}
+
+// RaceEnable has the same semantics as runtime.RaceEnable.
+func RaceEnable() {
+	runtime.RaceEnable()
+}
+
+// RaceAcquire has the same semantics as runtime.RaceAcquire.
+func RaceAcquire(addr unsafe.Pointer) {
+	runtime.RaceAcquire(addr)
+}
+
+// RaceRelease has the same semantics as runtime.RaceRelease.
+func RaceRelease(addr unsafe.Pointer) {
+	runtime.RaceRelease(addr)
+}
+
+// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge.
+func RaceReleaseMerge(addr unsafe.Pointer) {
+	runtime.RaceReleaseMerge(addr)
+}
diff --git a/pkg/sync/seqatomic_unsafe.go b/pkg/sync/seqatomic_unsafe.go
new file mode 100644
index 000000000..eda6fb131
--- /dev/null
+++ b/pkg/sync/seqatomic_unsafe.go
@@ -0,0 +1,72 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package template doesn't exist. This file must be instantiated using the
+// go_template_instance rule in tools/go_generics/defs.bzl.
+package template
+
+import (
+	"fmt"
+	"reflect"
+	"strings"
+	"unsafe"
+
+	"gvisor.dev/gvisor/pkg/sync"
+)
+
+// Value is a required type parameter.
+//
+// Value must not contain any pointers, including interface objects, function
+// objects, slices, maps, channels, unsafe.Pointer, and arrays or structs
+// containing any of the above. An init() function will panic if this property
+// does not hold.
+type Value struct{}
+
+// SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race
+// with any writer critical sections in sc.
+func SeqAtomicLoad(sc *sync.SeqCount, ptr *Value) Value {
+	// This function doesn't use SeqAtomicTryLoad because doing so is
+	// measurably, significantly (~20%) slower; Go is awful at inlining.
+	var val Value
+	for {
+		epoch := sc.BeginRead()
+		if sync.RaceEnabled {
+			// runtime.RaceDisable() doesn't actually stop the race detector,
+			// so it can't help us here. Instead, call runtime.memmove
+			// directly, which is not instrumented by the race detector.
+			sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val))
+		} else {
+			// This is ~40% faster for short reads than going through memmove.
+			val = *ptr
+		}
+		if sc.ReadOk(epoch) {
+			break
+		}
+	}
+	return val
+}
+
+// SeqAtomicTryLoad returns a copy of *ptr while in a reader critical section
+// in sc initiated by a call to sc.BeginRead() that returned epoch. If the read
+// would race with a writer critical section, SeqAtomicTryLoad returns
+// (unspecified, false).
+func SeqAtomicTryLoad(sc *sync.SeqCount, epoch sync.SeqCountEpoch, ptr *Value) (Value, bool) {
+	var val Value
+	if sync.RaceEnabled {
+		sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val))
+	} else {
+		val = *ptr
+	}
+	return val, sc.ReadOk(epoch)
+}
+
+func init() {
+	var val Value
+	typ := reflect.TypeOf(val)
+	name := typ.Name()
+	if ptrs := sync.PointersInType(typ, name); len(ptrs) != 0 {
+		panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n")))
+	}
+}
diff --git a/pkg/sync/seqatomictest/BUILD b/pkg/sync/seqatomictest/BUILD
new file mode 100644
index 000000000..eba21518d
--- /dev/null
+++ b/pkg/sync/seqatomictest/BUILD
@@ -0,0 +1,33 @@
+load("//tools/go_stateify:defs.bzl", "go_library")
+load("@io_bazel_rules_go//go:def.bzl", "go_test")
+load("//tools/go_generics:defs.bzl", "go_template_instance")
+
+package(licenses = ["notice"])
+
+go_template_instance(
+    name = "seqatomic_int",
+    out = "seqatomic_int_unsafe.go",
+    package = "seqatomic",
+    suffix = "Int",
+    template = "//pkg/sync:generic_seqatomic",
+    types = {
+        "Value": "int",
+    },
+)
+
+go_library(
+    name = "seqatomic",
+    srcs = ["seqatomic_int_unsafe.go"],
+    importpath = "gvisor.dev/gvisor/pkg/sync/seqatomic",
+    deps = [
+        "//pkg/sync",
+    ],
+)
+
+go_test(
+    name = "seqatomic_test",
+    size = "small",
+    srcs = ["seqatomic_test.go"],
+    embed = [":seqatomic"],
+    deps = ["//pkg/sync"],
+)
diff --git a/pkg/sync/seqatomictest/seqatomic_test.go b/pkg/sync/seqatomictest/seqatomic_test.go
new file mode 100644
index 000000000..2c4568b07
--- /dev/null
+++ b/pkg/sync/seqatomictest/seqatomic_test.go
@@ -0,0 +1,132 @@
+// Copyright 2018 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 seqatomic
+
+import (
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"gvisor.dev/gvisor/pkg/sync"
+)
+
+func TestSeqAtomicLoadUncontended(t *testing.T) {
+	var seq sync.SeqCount
+	const want = 1
+	data := want
+	if got := SeqAtomicLoadInt(&seq, &data); got != want {
+		t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want)
+	}
+}
+
+func TestSeqAtomicLoadAfterWrite(t *testing.T) {
+	var seq sync.SeqCount
+	var data int
+	const want = 1
+	seq.BeginWrite()
+	data = want
+	seq.EndWrite()
+	if got := SeqAtomicLoadInt(&seq, &data); got != want {
+		t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want)
+	}
+}
+
+func TestSeqAtomicLoadDuringWrite(t *testing.T) {
+	var seq sync.SeqCount
+	var data int
+	const want = 1
+	seq.BeginWrite()
+	go func() {
+		time.Sleep(time.Second)
+		data = want
+		seq.EndWrite()
+	}()
+	if got := SeqAtomicLoadInt(&seq, &data); got != want {
+		t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want)
+	}
+}
+
+func TestSeqAtomicTryLoadUncontended(t *testing.T) {
+	var seq sync.SeqCount
+	const want = 1
+	data := want
+	epoch := seq.BeginRead()
+	if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want {
+		t.Errorf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want)
+	}
+}
+
+func TestSeqAtomicTryLoadDuringWrite(t *testing.T) {
+	var seq sync.SeqCount
+	var data int
+	epoch := seq.BeginRead()
+	seq.BeginWrite()
+	if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok {
+		t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got)
+	}
+	seq.EndWrite()
+}
+
+func TestSeqAtomicTryLoadAfterWrite(t *testing.T) {
+	var seq sync.SeqCount
+	var data int
+	epoch := seq.BeginRead()
+	seq.BeginWrite()
+	seq.EndWrite()
+	if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok {
+		t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got)
+	}
+}
+
+func BenchmarkSeqAtomicLoadIntUncontended(b *testing.B) {
+	var seq sync.SeqCount
+	const want = 42
+	data := want
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			if got := SeqAtomicLoadInt(&seq, &data); got != want {
+				b.Fatalf("SeqAtomicLoadInt: got %v, wanted %v", got, want)
+			}
+		}
+	})
+}
+
+func BenchmarkSeqAtomicTryLoadIntUncontended(b *testing.B) {
+	var seq sync.SeqCount
+	const want = 42
+	data := want
+	b.RunParallel(func(pb *testing.PB) {
+		epoch := seq.BeginRead()
+		for pb.Next() {
+			if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want {
+				b.Fatalf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want)
+			}
+		}
+	})
+}
+
+// For comparison:
+func BenchmarkAtomicValueLoadIntUncontended(b *testing.B) {
+	var a atomic.Value
+	const want = 42
+	a.Store(int(want))
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			if got := a.Load().(int); got != want {
+				b.Fatalf("atomic.Value.Load: got %v, wanted %v", got, want)
+			}
+		}
+	})
+}
diff --git a/pkg/sync/seqcount.go b/pkg/sync/seqcount.go
new file mode 100644
index 000000000..a1e895352
--- /dev/null
+++ b/pkg/sync/seqcount.go
@@ -0,0 +1,149 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package sync
+
+import (
+	"fmt"
+	"reflect"
+	"runtime"
+	"sync/atomic"
+)
+
+// SeqCount is a synchronization primitive for optimistic reader/writer
+// synchronization in cases where readers can work with stale data and
+// therefore do not need to block writers.
+//
+// Compared to sync/atomic.Value:
+//
+// - Mutation of SeqCount-protected data does not require memory allocation,
+// whereas atomic.Value generally does. This is a significant advantage when
+// writes are common.
+//
+// - Atomic reads of SeqCount-protected data require copying. This is a
+// disadvantage when atomic reads are common.
+//
+// - SeqCount may be more flexible: correct use of SeqCount.ReadOk allows other
+// operations to be made atomic with reads of SeqCount-protected data.
+//
+// - SeqCount may be less flexible: as of this writing, SeqCount-protected data
+// cannot include pointers.
+//
+// - SeqCount is more cumbersome to use; atomic reads of SeqCount-protected
+// data require instantiating function templates using go_generics (see
+// seqatomic.go).
+type SeqCount struct {
+	// epoch is incremented by BeginWrite and EndWrite, such that epoch is odd
+	// if a writer critical section is active, and a read from data protected
+	// by this SeqCount is atomic iff epoch is the same even value before and
+	// after the read.
+	epoch uint32
+}
+
+// SeqCountEpoch tracks writer critical sections in a SeqCount.
+type SeqCountEpoch struct {
+	val uint32
+}
+
+// We assume that:
+//
+// - All functions in sync/atomic that perform a memory read are at least a
+// read fence: memory reads before calls to such functions cannot be reordered
+// after the call, and memory reads after calls to such functions cannot be
+// reordered before the call, even if those reads do not use sync/atomic.
+//
+// - All functions in sync/atomic that perform a memory write are at least a
+// write fence: memory writes before calls to such functions cannot be
+// reordered after the call, and memory writes after calls to such functions
+// cannot be reordered before the call, even if those writes do not use
+// sync/atomic.
+//
+// As of this writing, the Go memory model completely fails to describe
+// sync/atomic, but these properties are implied by
+// https://groups.google.com/forum/#!topic/golang-nuts/7EnEhM3U7B8.
+
+// BeginRead indicates the beginning of a reader critical section. Reader
+// critical sections DO NOT BLOCK writer critical sections, so operations in a
+// reader critical section MAY RACE with writer critical sections. Races are
+// detected by ReadOk at the end of the reader critical section. Thus, the
+// low-level structure of readers is generally:
+//
+//     for {
+//         epoch := seq.BeginRead()
+//         // do something idempotent with seq-protected data
+//         if seq.ReadOk(epoch) {
+//             break
+//         }
+//     }
+//
+// However, since reader critical sections may race with writer critical
+// sections, the Go race detector will (accurately) flag data races in readers
+// using this pattern. Most users of SeqCount will need to use the
+// SeqAtomicLoad function template in seqatomic.go.
+func (s *SeqCount) BeginRead() SeqCountEpoch {
+	epoch := atomic.LoadUint32(&s.epoch)
+	for epoch&1 != 0 {
+		runtime.Gosched()
+		epoch = atomic.LoadUint32(&s.epoch)
+	}
+	return SeqCountEpoch{epoch}
+}
+
+// ReadOk returns true if the reader critical section initiated by a previous
+// call to BeginRead() that returned epoch did not race with any writer critical
+// sections.
+//
+// ReadOk may be called any number of times during a reader critical section.
+// Reader critical sections do not need to be explicitly terminated; the last
+// call to ReadOk is implicitly the end of the reader critical section.
+func (s *SeqCount) ReadOk(epoch SeqCountEpoch) bool {
+	return atomic.LoadUint32(&s.epoch) == epoch.val
+}
+
+// BeginWrite indicates the beginning of a writer critical section.
+//
+// SeqCount does not support concurrent writer critical sections; clients with
+// concurrent writers must synchronize them using e.g. sync.Mutex.
+func (s *SeqCount) BeginWrite() {
+	if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 == 0 {
+		panic("SeqCount.BeginWrite during writer critical section")
+	}
+}
+
+// EndWrite ends the effect of a preceding BeginWrite.
+func (s *SeqCount) EndWrite() {
+	if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 != 0 {
+		panic("SeqCount.EndWrite outside writer critical section")
+	}
+}
+
+// PointersInType returns a list of pointers reachable from values named
+// valName of the given type.
+//
+// PointersInType is not exhaustive, but it is guaranteed that if typ contains
+// at least one pointer, then PointersInTypeOf returns a non-empty list.
+func PointersInType(typ reflect.Type, valName string) []string {
+	switch kind := typ.Kind(); kind {
+	case reflect.Bool, reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64, reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr, reflect.Float32, reflect.Float64, reflect.Complex64, reflect.Complex128:
+		return nil
+
+	case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Ptr, reflect.Slice, reflect.String, reflect.UnsafePointer:
+		return []string{valName}
+
+	case reflect.Array:
+		return PointersInType(typ.Elem(), valName+"[]")
+
+	case reflect.Struct:
+		var ptrs []string
+		for i, n := 0, typ.NumField(); i < n; i++ {
+			field := typ.Field(i)
+			ptrs = append(ptrs, PointersInType(field.Type, fmt.Sprintf("%s.%s", valName, field.Name))...)
+		}
+		return ptrs
+
+	default:
+		return []string{fmt.Sprintf("%s (of type %s with unknown kind %s)", valName, typ, kind)}
+	}
+}
diff --git a/pkg/sync/seqcount_test.go b/pkg/sync/seqcount_test.go
new file mode 100644
index 000000000..6eb7b4b59
--- /dev/null
+++ b/pkg/sync/seqcount_test.go
@@ -0,0 +1,153 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package sync
+
+import (
+	"reflect"
+	"testing"
+	"time"
+)
+
+func TestSeqCountWriteUncontended(t *testing.T) {
+	var seq SeqCount
+	seq.BeginWrite()
+	seq.EndWrite()
+}
+
+func TestSeqCountReadUncontended(t *testing.T) {
+	var seq SeqCount
+	epoch := seq.BeginRead()
+	if !seq.ReadOk(epoch) {
+		t.Errorf("ReadOk: got false, wanted true")
+	}
+}
+
+func TestSeqCountBeginReadAfterWrite(t *testing.T) {
+	var seq SeqCount
+	var data int32
+	const want = 1
+	seq.BeginWrite()
+	data = want
+	seq.EndWrite()
+	epoch := seq.BeginRead()
+	if data != want {
+		t.Errorf("Reader: got %v, wanted %v", data, want)
+	}
+	if !seq.ReadOk(epoch) {
+		t.Errorf("ReadOk: got false, wanted true")
+	}
+}
+
+func TestSeqCountBeginReadDuringWrite(t *testing.T) {
+	var seq SeqCount
+	var data int
+	const want = 1
+	seq.BeginWrite()
+	go func() {
+		time.Sleep(time.Second)
+		data = want
+		seq.EndWrite()
+	}()
+	epoch := seq.BeginRead()
+	if data != want {
+		t.Errorf("Reader: got %v, wanted %v", data, want)
+	}
+	if !seq.ReadOk(epoch) {
+		t.Errorf("ReadOk: got false, wanted true")
+	}
+}
+
+func TestSeqCountReadOkAfterWrite(t *testing.T) {
+	var seq SeqCount
+	epoch := seq.BeginRead()
+	seq.BeginWrite()
+	seq.EndWrite()
+	if seq.ReadOk(epoch) {
+		t.Errorf("ReadOk: got true, wanted false")
+	}
+}
+
+func TestSeqCountReadOkDuringWrite(t *testing.T) {
+	var seq SeqCount
+	epoch := seq.BeginRead()
+	seq.BeginWrite()
+	if seq.ReadOk(epoch) {
+		t.Errorf("ReadOk: got true, wanted false")
+	}
+	seq.EndWrite()
+}
+
+func BenchmarkSeqCountWriteUncontended(b *testing.B) {
+	var seq SeqCount
+	for i := 0; i < b.N; i++ {
+		seq.BeginWrite()
+		seq.EndWrite()
+	}
+}
+
+func BenchmarkSeqCountReadUncontended(b *testing.B) {
+	var seq SeqCount
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			epoch := seq.BeginRead()
+			if !seq.ReadOk(epoch) {
+				b.Fatalf("ReadOk: got false, wanted true")
+			}
+		}
+	})
+}
+
+func TestPointersInType(t *testing.T) {
+	for _, test := range []struct {
+		name string // used for both test and value name
+		val  interface{}
+		ptrs []string
+	}{
+		{
+			name: "EmptyStruct",
+			val:  struct{}{},
+		},
+		{
+			name: "Int",
+			val:  int(0),
+		},
+		{
+			name: "MixedStruct",
+			val: struct {
+				b             bool
+				I             int
+				ExportedPtr   *struct{}
+				unexportedPtr *struct{}
+				arr           [2]int
+				ptrArr        [2]*int
+				nestedStruct  struct {
+					nestedNonptr int
+					nestedPtr    *int
+				}
+				structArr [1]struct {
+					nonptr int
+					ptr    *int
+				}
+			}{},
+			ptrs: []string{
+				"MixedStruct.ExportedPtr",
+				"MixedStruct.unexportedPtr",
+				"MixedStruct.ptrArr[]",
+				"MixedStruct.nestedStruct.nestedPtr",
+				"MixedStruct.structArr[].ptr",
+			},
+		},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			typ := reflect.TypeOf(test.val)
+			ptrs := PointersInType(typ, test.name)
+			t.Logf("Found pointers: %v", ptrs)
+			if (len(ptrs) != 0 || len(test.ptrs) != 0) && !reflect.DeepEqual(ptrs, test.ptrs) {
+				t.Errorf("Got %v, wanted %v", ptrs, test.ptrs)
+			}
+		})
+	}
+}
diff --git a/pkg/sync/syncutil.go b/pkg/sync/syncutil.go
new file mode 100644
index 000000000..b16cf5333
--- /dev/null
+++ b/pkg/sync/syncutil.go
@@ -0,0 +1,7 @@
+// Copyright 2019 The gVisor Authors.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package sync provides synchronization primitives.
+package sync