Require sync.Mutex to lock and unlock from the same goroutine

We would like to track locks ordering to detect ordering violations. Detecting
violations is much simpler if mutexes must be unlocked by the same goroutine
that locked them.

Thus, as a first step to tracking lock ordering, add this lock/unlock
requirement to gVisor's sync.Mutex. This is more strict than the Go standard
library's sync.Mutex, but initial testing indicates only a single lock that is
used across goroutines. The new sync.CrossGoroutineMutex relaxes the
requirement (but will not provide lock order checking).

Due to the additional overhead, enforcement is only enabled with the
"checklocks" build tag. Build with this tag using:

bazel build --define=gotags=checklocks ...

From my spot-checking, this has no changed inlining properties when disabled.

Updates #4804

PiperOrigin-RevId: 343370200

6 files changed, 185 insertions, 8 deletions

diff --git a/pkg/sync/BUILD b/pkg/sync/BUILD
index 68535c3b1..410e9a3ca 100644
--- a/pkg/sync/BUILD
+++ b/pkg/sync/BUILD
@@ -31,6 +31,8 @@ go_library(
     name = "sync",
     srcs = [
         "aliases.go",
+        "checklocks_off_unsafe.go",
+        "checklocks_on_unsafe.go",
         "memmove_unsafe.go",
         "mutex_unsafe.go",
         "nocopy.go",
@@ -43,6 +45,9 @@ go_library(
     ],
     marshal = False,
     stateify = False,
+    deps = [
+        "//pkg/goid",
+    ],
 )
 
 go_test(
diff --git a/pkg/sync/checklocks_off_unsafe.go b/pkg/sync/checklocks_off_unsafe.go
new file mode 100644
index 000000000..62c81b149
--- /dev/null
+++ b/pkg/sync/checklocks_off_unsafe.go
@@ -0,0 +1,18 @@
+// 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.
+
+// +build !checklocks
+
+package sync
+
+import (
+	"unsafe"
+)
+
+func noteLock(l unsafe.Pointer) {
+}
+
+func noteUnlock(l unsafe.Pointer) {
+}
diff --git a/pkg/sync/checklocks_on_unsafe.go b/pkg/sync/checklocks_on_unsafe.go
new file mode 100644
index 000000000..24f933ed1
--- /dev/null
+++ b/pkg/sync/checklocks_on_unsafe.go
@@ -0,0 +1,108 @@
+// 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.
+
+// +build checklocks
+
+package sync
+
+import (
+	"fmt"
+	"strings"
+	"sync"
+	"unsafe"
+
+	"gvisor.dev/gvisor/pkg/goid"
+)
+
+// gLocks contains metadata about the locks held by a goroutine.
+type gLocks struct {
+	locksHeld []unsafe.Pointer
+}
+
+// map[goid int]*gLocks
+//
+// Each key may only be written by the G with the goid it refers to.
+//
+// Note that entries are not evicted when a G exit, causing unbounded growth
+// with new G creation / destruction. If this proves problematic, entries could
+// be evicted when no locks are held at the expense of more allocations when
+// taking top-level locks.
+var locksHeld sync.Map
+
+func getGLocks() *gLocks {
+	id := goid.Get()
+
+	var locks *gLocks
+	if l, ok := locksHeld.Load(id); ok {
+		locks = l.(*gLocks)
+	} else {
+		locks = &gLocks{
+			// Initialize space for a few locks.
+			locksHeld: make([]unsafe.Pointer, 0, 8),
+		}
+		locksHeld.Store(id, locks)
+	}
+
+	return locks
+}
+
+func noteLock(l unsafe.Pointer) {
+	locks := getGLocks()
+
+	for _, lock := range locks.locksHeld {
+		if lock == l {
+			panic(fmt.Sprintf("Deadlock on goroutine %d! Double lock of %p: %+v", goid.Get(), l, locks))
+		}
+	}
+
+	// Commit only after checking for panic conditions so that this lock
+	// isn't on the list if the above panic is recovered.
+	locks.locksHeld = append(locks.locksHeld, l)
+}
+
+func noteUnlock(l unsafe.Pointer) {
+	locks := getGLocks()
+
+	if len(locks.locksHeld) == 0 {
+		panic(fmt.Sprintf("Unlock of %p on goroutine %d without any locks held! All locks:\n%s", l, goid.Get(), dumpLocks()))
+	}
+
+	// Search backwards since callers are most likely to unlock in LIFO order.
+	length := len(locks.locksHeld)
+	for i := length - 1; i >= 0; i-- {
+		if l == locks.locksHeld[i] {
+			copy(locks.locksHeld[i:length-1], locks.locksHeld[i+1:length])
+			// Clear last entry to ensure addr can be GC'd.
+			locks.locksHeld[length-1] = nil
+			locks.locksHeld = locks.locksHeld[:length-1]
+			return
+		}
+	}
+
+	panic(fmt.Sprintf("Unlock of %p on goroutine %d without matching lock! All locks:\n%s", l, goid.Get(), dumpLocks()))
+}
+
+func dumpLocks() string {
+	var s strings.Builder
+	locksHeld.Range(func(key, value interface{}) bool {
+		goid := key.(int64)
+		locks := value.(*gLocks)
+
+		// N.B. accessing gLocks of another G is fundamentally racy.
+
+		fmt.Fprintf(&s, "goroutine %d:\n", goid)
+		if len(locks.locksHeld) == 0 {
+			fmt.Fprintf(&s, "\t<none>\n")
+		}
+		for _, lock := range locks.locksHeld {
+			fmt.Fprintf(&s, "\t%p\n", lock)
+		}
+		fmt.Fprintf(&s, "\n")
+
+		return true
+	})
+
+	return s.String()
+}
diff --git a/pkg/sync/mutex_test.go b/pkg/sync/mutex_test.go
index 0838248b4..4fb51a8ab 100644
--- a/pkg/sync/mutex_test.go
+++ b/pkg/sync/mutex_test.go
@@ -32,11 +32,11 @@ func TestStructSize(t *testing.T) {
 func TestFieldValues(t *testing.T) {
 	var m Mutex
 	m.Lock()
-	if got := *m.state(); got != mutexLocked {
+	if got := *m.m.state(); got != mutexLocked {
 		t.Errorf("got locked sync.Mutex.state = %d, want = %d", got, mutexLocked)
 	}
 	m.Unlock()
-	if got := *m.state(); got != mutexUnlocked {
+	if got := *m.m.state(); got != mutexUnlocked {
 		t.Errorf("got unlocked sync.Mutex.state = %d, want = %d", got, mutexUnlocked)
 	}
 }
diff --git a/pkg/sync/mutex_unsafe.go b/pkg/sync/mutex_unsafe.go
index f4c2e9642..21084b857 100644
--- a/pkg/sync/mutex_unsafe.go
+++ b/pkg/sync/mutex_unsafe.go
@@ -17,8 +17,9 @@ import (
 	"unsafe"
 )
 
-// Mutex is a try lock.
-type Mutex struct {
+// CrossGoroutineMutex is equivalent to Mutex, but it need not be unlocked by a
+// the same goroutine that locked the mutex.
+type CrossGoroutineMutex struct {
 	sync.Mutex
 }
 
@@ -27,7 +28,7 @@ type syncMutex struct {
 	sema  uint32
 }
 
-func (m *Mutex) state() *int32 {
+func (m *CrossGoroutineMutex) state() *int32 {
 	return &(*syncMutex)(unsafe.Pointer(&m.Mutex)).state
 }
 
@@ -36,9 +37,9 @@ const (
 	mutexLocked   = 1
 )
 
-// TryLock tries to aquire the mutex. It returns true if it succeeds and false
+// TryLock tries to acquire the mutex. It returns true if it succeeds and false
 // otherwise. TryLock does not block.
-func (m *Mutex) TryLock() bool {
+func (m *CrossGoroutineMutex) TryLock() bool {
 	if atomic.CompareAndSwapInt32(m.state(), mutexUnlocked, mutexLocked) {
 		if RaceEnabled {
 			RaceAcquire(unsafe.Pointer(&m.Mutex))
@@ -47,3 +48,43 @@ func (m *Mutex) TryLock() bool {
 	}
 	return false
 }
+
+// Mutex is a mutual exclusion lock. The zero value for a Mutex is an unlocked
+// mutex.
+//
+// A Mutex must not be copied after first use.
+//
+// A Mutex must be unlocked by the same goroutine that locked it. This
+// invariant is enforced with the 'checklocks' build tag.
+type Mutex struct {
+	m CrossGoroutineMutex
+}
+
+// Lock locks m. If the lock is already in use, the calling goroutine blocks
+// until the mutex is available.
+func (m *Mutex) Lock() {
+	noteLock(unsafe.Pointer(m))
+	m.m.Lock()
+}
+
+// Unlock unlocks m.
+//
+// Preconditions:
+// * m is locked.
+// * m was locked by this goroutine.
+func (m *Mutex) Unlock() {
+	noteUnlock(unsafe.Pointer(m))
+	m.m.Unlock()
+}
+
+// TryLock tries to acquire the mutex. It returns true if it succeeds and false
+// otherwise. TryLock does not block.
+func (m *Mutex) TryLock() bool {
+	// Note lock first to enforce proper locking even if unsuccessful.
+	noteLock(unsafe.Pointer(m))
+	locked := m.m.TryLock()
+	if !locked {
+		noteUnlock(unsafe.Pointer(m))
+	}
+	return locked
+}
diff --git a/pkg/sync/rwmutex_unsafe.go b/pkg/sync/rwmutex_unsafe.go
index b3b4dee78..fa023f5bb 100644
--- a/pkg/sync/rwmutex_unsafe.go
+++ b/pkg/sync/rwmutex_unsafe.go
@@ -32,7 +32,12 @@ func runtimeSemrelease(s *uint32, handoff bool, skipframes int)
 // RWMutex is identical to sync.RWMutex, but adds the DowngradeLock,
 // TryLock and TryRLock methods.
 type RWMutex struct {
-	w           Mutex  // held if there are pending writers
+	// w is held if there are pending writers
+	//
+	// We use CrossGoroutineMutex rather than Mutex because the lock
+	// annotation instrumentation in Mutex will trigger false positives in
+	// the race detector when called inside of RaceDisable.
+	w           CrossGoroutineMutex
 	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