Merge branch 'master' into ip-forwarding

- Merges aleksej-paschenko's with HEAD
- Adds vfs2 support for ip_forward

3 files changed, 74 insertions, 63 deletions

diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD
index 34286c7a8..daa2dae76 100644
--- a/pkg/sentry/kernel/futex/BUILD
+++ b/pkg/sentry/kernel/futex/BUILD
@@ -1,6 +1,5 @@
-load("@io_bazel_rules_go//go:def.bzl", "go_test")
+load("//tools:defs.bzl", "go_library", "go_test")
 load("//tools/go_generics:defs.bzl", "go_template_instance")
-load("//tools/go_stateify:defs.bzl", "go_library")
 
 package(licenses = ["notice"])
 
@@ -9,7 +8,7 @@ go_template_instance(
     out = "atomicptr_bucket_unsafe.go",
     package = "futex",
     suffix = "Bucket",
-    template = "//third_party/gvsync:generic_atomicptr",
+    template = "//pkg/sync:generic_atomicptr",
     types = {
         "Value": "bucket",
     },
@@ -34,15 +33,15 @@ go_library(
         "futex.go",
         "waiter_list.go",
     ],
-    importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/futex",
     visibility = ["//pkg/sentry:internal"],
     deps = [
         "//pkg/abi/linux",
+        "//pkg/context",
         "//pkg/log",
-        "//pkg/sentry/context",
         "//pkg/sentry/memmap",
-        "//pkg/sentry/usermem",
+        "//pkg/sync",
         "//pkg/syserror",
+        "//pkg/usermem",
     ],
 )
 
@@ -50,6 +49,10 @@ go_test(
     name = "futex_test",
     size = "small",
     srcs = ["futex_test.go"],
-    embed = [":futex"],
-    deps = ["//pkg/sentry/usermem"],
+    library = ":futex",
+    deps = [
+        "//pkg/context",
+        "//pkg/sync",
+        "//pkg/usermem",
+    ],
 )
diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go
index 278cc8143..e4dcc4d40 100644
--- a/pkg/sentry/kernel/futex/futex.go
+++ b/pkg/sentry/kernel/futex/futex.go
@@ -18,12 +18,12 @@
 package futex
 
 import (
-	"sync"
-
 	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
-	"gvisor.dev/gvisor/pkg/sentry/usermem"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserror"
+	"gvisor.dev/gvisor/pkg/usermem"
 )
 
 // KeyKind indicates the type of a Key.
@@ -67,9 +67,9 @@ type Key struct {
 	Offset uint64
 }
 
-func (k *Key) release() {
+func (k *Key) release(t Target) {
 	if k.MappingIdentity != nil {
-		k.MappingIdentity.DecRef()
+		k.MappingIdentity.DecRef(t)
 	}
 	k.Mappable = nil
 	k.MappingIdentity = nil
@@ -95,6 +95,8 @@ func (k *Key) matches(k2 *Key) bool {
 
 // Target abstracts memory accesses and keys.
 type Target interface {
+	context.Context
+
 	// SwapUint32 gives access to usermem.IO.SwapUint32.
 	SwapUint32(addr usermem.Addr, new uint32) (uint32, error)
 
@@ -297,7 +299,7 @@ func (b *bucket) wakeWaiterLocked(w *Waiter) {
 // bucket "to".
 //
 // Preconditions: b and to must be locked.
-func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int {
+func (b *bucket) requeueLocked(t Target, to *bucket, key, nkey *Key, n int) int {
 	done := 0
 	for w := b.waiters.Front(); done < n && w != nil; {
 		if !w.key.matches(key) {
@@ -309,7 +311,7 @@ func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int {
 		requeued := w
 		w = w.Next() // Next iteration.
 		b.waiters.Remove(requeued)
-		requeued.key.release()
+		requeued.key.release(t)
 		requeued.key = nkey.clone()
 		to.waiters.PushBack(requeued)
 		requeued.bucket.Store(to)
@@ -457,7 +459,7 @@ func (m *Manager) Wake(t Target, addr usermem.Addr, private bool, bitmask uint32
 	r := b.wakeLocked(&k, bitmask, n)
 
 	b.mu.Unlock()
-	k.release()
+	k.release(t)
 	return r, nil
 }
 
@@ -466,12 +468,12 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch
 	if err != nil {
 		return 0, err
 	}
-	defer k1.release()
+	defer k1.release(t)
 	k2, err := getKey(t, naddr, private)
 	if err != nil {
 		return 0, err
 	}
-	defer k2.release()
+	defer k2.release(t)
 
 	b1, b2 := m.lockBuckets(&k1, &k2)
 	defer b1.mu.Unlock()
@@ -489,7 +491,7 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch
 	done := b1.wakeLocked(&k1, ^uint32(0), nwake)
 
 	// Requeue the number required.
-	b1.requeueLocked(b2, &k1, &k2, nreq)
+	b1.requeueLocked(t, b2, &k1, &k2, nreq)
 
 	return done, nil
 }
@@ -516,12 +518,12 @@ func (m *Manager) WakeOp(t Target, addr1, addr2 usermem.Addr, private bool, nwak
 	if err != nil {
 		return 0, err
 	}
-	defer k1.release()
+	defer k1.release(t)
 	k2, err := getKey(t, addr2, private)
 	if err != nil {
 		return 0, err
 	}
-	defer k2.release()
+	defer k2.release(t)
 
 	b1, b2 := m.lockBuckets(&k1, &k2)
 	defer b1.mu.Unlock()
@@ -572,7 +574,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo
 	// Perform our atomic check.
 	if err := check(t, addr, val); err != nil {
 		b.mu.Unlock()
-		w.key.release()
+		w.key.release(t)
 		return err
 	}
 
@@ -586,7 +588,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo
 
 // WaitComplete must be called when a Waiter previously added by WaitPrepare is
 // no longer eligible to be woken.
-func (m *Manager) WaitComplete(w *Waiter) {
+func (m *Manager) WaitComplete(w *Waiter, t Target) {
 	// Remove w from the bucket it's in.
 	for {
 		b := w.bucket.Load()
@@ -618,7 +620,7 @@ func (m *Manager) WaitComplete(w *Waiter) {
 	}
 
 	// Release references held by the waiter.
-	w.key.release()
+	w.key.release(t)
 }
 
 // LockPI attempts to lock the futex following the Priority-inheritance futex
@@ -649,13 +651,13 @@ func (m *Manager) LockPI(w *Waiter, t Target, addr usermem.Addr, tid uint32, pri
 
 	success, err := m.lockPILocked(w, t, addr, tid, b, try)
 	if err != nil {
-		w.key.release()
+		w.key.release(t)
 		b.mu.Unlock()
 		return false, err
 	}
 	if success || try {
 		// Release waiter if it's not going to be a wait.
-		w.key.release()
+		w.key.release(t)
 	}
 	b.mu.Unlock()
 	return success, nil
@@ -718,10 +720,10 @@ func (m *Manager) lockPILocked(w *Waiter, t Target, addr usermem.Addr, tid uint3
 	}
 }
 
-// UnlockPI unlock the futex following the Priority-inheritance futex
-// rules. The address provided must contain the caller's TID. If there are
-// waiters, TID of the next waiter (FIFO) is set to the given address, and the
-// waiter woken up. If there are no waiters, 0 is set to the address.
+// UnlockPI unlocks the futex following the Priority-inheritance futex rules.
+// The address provided must contain the caller's TID. If there are waiters,
+// TID of the next waiter (FIFO) is set to the given address, and the waiter
+// woken up. If there are no waiters, 0 is set to the address.
 func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool) error {
 	k, err := getKey(t, addr, private)
 	if err != nil {
@@ -731,7 +733,7 @@ func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool
 
 	err = m.unlockPILocked(t, addr, tid, b, &k)
 
-	k.release()
+	k.release(t)
 	b.mu.Unlock()
 	return err
 }
diff --git a/pkg/sentry/kernel/futex/futex_test.go b/pkg/sentry/kernel/futex/futex_test.go
index 65e5d1428..d0128c548 100644
--- a/pkg/sentry/kernel/futex/futex_test.go
+++ b/pkg/sentry/kernel/futex/futex_test.go
@@ -17,40 +17,46 @@ package futex
 import (
 	"math"
 	"runtime"
-	"sync"
 	"sync/atomic"
 	"syscall"
 	"testing"
 	"unsafe"
 
-	"gvisor.dev/gvisor/pkg/sentry/usermem"
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/usermem"
 )
 
 // testData implements the Target interface, and allows us to
 // treat the address passed for futex operations as an index in
 // a byte slice for testing simplicity.
-type testData []byte
+type testData struct {
+	context.Context
+	data []byte
+}
 
 const sizeofInt32 = 4
 
 func newTestData(size uint) testData {
-	return make([]byte, size)
+	return testData{
+		data: make([]byte, size),
+	}
 }
 
 func (t testData) SwapUint32(addr usermem.Addr, new uint32) (uint32, error) {
-	val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t[addr])), new)
+	val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), new)
 	return val, nil
 }
 
 func (t testData) CompareAndSwapUint32(addr usermem.Addr, old, new uint32) (uint32, error) {
-	if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t[addr])), old, new) {
+	if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), old, new) {
 		return old, nil
 	}
-	return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil
+	return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil
 }
 
 func (t testData) LoadUint32(addr usermem.Addr) (uint32, error) {
-	return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil
+	return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil
 }
 
 func (t testData) GetSharedKey(addr usermem.Addr) (Key, error) {
@@ -83,7 +89,7 @@ func TestFutexWake(t *testing.T) {
 
 			// Start waiting for wakeup.
 			w := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-			defer m.WaitComplete(w)
+			defer m.WaitComplete(w, d)
 
 			// Perform a wakeup.
 			if n, err := m.Wake(d, 0, private, ^uint32(0), 1); err != nil || n != 1 {
@@ -106,7 +112,7 @@ func TestFutexWakeBitmask(t *testing.T) {
 
 			// Start waiting for wakeup.
 			w := newPreparedTestWaiter(t, m, d, 0, private, 0, 0x0000ffff)
-			defer m.WaitComplete(w)
+			defer m.WaitComplete(w, d)
 
 			// Perform a wakeup using the wrong bitmask.
 			if n, err := m.Wake(d, 0, private, 0xffff0000, 1); err != nil || n != 0 {
@@ -141,7 +147,7 @@ func TestFutexWakeTwo(t *testing.T) {
 			var ws [3]*Waiter
 			for i := range ws {
 				ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-				defer m.WaitComplete(ws[i])
+				defer m.WaitComplete(ws[i], d)
 			}
 
 			// Perform two wakeups.
@@ -174,9 +180,9 @@ func TestFutexWakeUnrelated(t *testing.T) {
 
 			// Start two waiters waiting for wakeup on different addresses.
 			w1 := newPreparedTestWaiter(t, m, d, 0*sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w1)
+			defer m.WaitComplete(w1, d)
 			w2 := newPreparedTestWaiter(t, m, d, 1*sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w2)
+			defer m.WaitComplete(w2, d)
 
 			// Perform two wakeups on the second address.
 			if n, err := m.Wake(d, 1*sizeofInt32, private, ^uint32(0), 2); err != nil || n != 1 {
@@ -216,9 +222,9 @@ func TestWakeOpFirstNonEmpty(t *testing.T) {
 
 			// Add two waiters on address 0.
 			w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-			defer m.WaitComplete(w1)
+			defer m.WaitComplete(w1, d)
 			w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-			defer m.WaitComplete(w2)
+			defer m.WaitComplete(w2, d)
 
 			// Perform 10 wakeups on address 0.
 			if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 0, 0); err != nil || n != 2 {
@@ -244,9 +250,9 @@ func TestWakeOpSecondNonEmpty(t *testing.T) {
 
 			// Add two waiters on address sizeofInt32.
 			w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w1)
+			defer m.WaitComplete(w1, d)
 			w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w2)
+			defer m.WaitComplete(w2, d)
 
 			// Perform 10 wakeups on address sizeofInt32 (contingent on
 			// d.Op(0), which should succeed).
@@ -273,9 +279,9 @@ func TestWakeOpSecondNonEmptyFailingOp(t *testing.T) {
 
 			// Add two waiters on address sizeofInt32.
 			w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w1)
+			defer m.WaitComplete(w1, d)
 			w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w2)
+			defer m.WaitComplete(w2, d)
 
 			// Perform 10 wakeups on address sizeofInt32 (contingent on
 			// d.Op(1), which should fail).
@@ -302,15 +308,15 @@ func TestWakeOpAllNonEmpty(t *testing.T) {
 
 			// Add two waiters on address 0.
 			w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-			defer m.WaitComplete(w1)
+			defer m.WaitComplete(w1, d)
 			w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-			defer m.WaitComplete(w2)
+			defer m.WaitComplete(w2, d)
 
 			// Add two waiters on address sizeofInt32.
 			w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w3)
+			defer m.WaitComplete(w3, d)
 			w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w4)
+			defer m.WaitComplete(w4, d)
 
 			// Perform 10 wakeups on address 0 (unconditionally), and 10
 			// wakeups on address sizeofInt32 (contingent on d.Op(0), which
@@ -344,15 +350,15 @@ func TestWakeOpAllNonEmptyFailingOp(t *testing.T) {
 
 			// Add two waiters on address 0.
 			w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-			defer m.WaitComplete(w1)
+			defer m.WaitComplete(w1, d)
 			w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-			defer m.WaitComplete(w2)
+			defer m.WaitComplete(w2, d)
 
 			// Add two waiters on address sizeofInt32.
 			w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w3)
+			defer m.WaitComplete(w3, d)
 			w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0))
-			defer m.WaitComplete(w4)
+			defer m.WaitComplete(w4, d)
 
 			// Perform 10 wakeups on address 0 (unconditionally), and 10
 			// wakeups on address sizeofInt32 (contingent on d.Op(1), which
@@ -388,7 +394,7 @@ func TestWakeOpSameAddress(t *testing.T) {
 			var ws [4]*Waiter
 			for i := range ws {
 				ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-				defer m.WaitComplete(ws[i])
+				defer m.WaitComplete(ws[i], d)
 			}
 
 			// Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup
@@ -422,7 +428,7 @@ func TestWakeOpSameAddressFailingOp(t *testing.T) {
 			var ws [4]*Waiter
 			for i := range ws {
 				ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0))
-				defer m.WaitComplete(ws[i])
+				defer m.WaitComplete(ws[i], d)
 			}
 
 			// Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup
@@ -472,7 +478,7 @@ func (t *testMutex) Lock() {
 	for {
 		// Attempt to grab the lock.
 		if atomic.CompareAndSwapUint32(
-			(*uint32)(unsafe.Pointer(&t.d[t.a])),
+			(*uint32)(unsafe.Pointer(&t.d.data[t.a])),
 			testMutexUnlocked,
 			testMutexLocked) {
 			// Lock held.
@@ -490,7 +496,7 @@ func (t *testMutex) Lock() {
 			panic("WaitPrepare returned unexpected error: " + err.Error())
 		}
 		<-w.C
-		t.m.WaitComplete(w)
+		t.m.WaitComplete(w, t.d)
 	}
 }
 
@@ -498,7 +504,7 @@ func (t *testMutex) Lock() {
 // This will notify any waiters via the futex manager.
 func (t *testMutex) Unlock() {
 	// Unlock.
-	atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d[t.a])), testMutexUnlocked)
+	atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d.data[t.a])), testMutexUnlocked)
 
 	// Notify all waiters.
 	t.m.Wake(t.d, t.a, true, ^uint32(0), math.MaxInt32)