Improve stopped container handling

Getting state of a stopped container would fail and could lead containerd
to not detecting that the container had actually stopped. Now stopped and
deleted containers return `stopped` state.

Also makes other messages more consistent when container is stopped. Some
where still sending messages to runsc and failing in different ways. Now
they go through `initState` state machine like the other messages.

There are a few changes to improve debugability with it as well.

Fixes #5861

PiperOrigin-RevId: 380698513

1 files changed, 48 insertions, 0 deletions

diff --git a/pkg/shim/debug.go b/pkg/shim/debug.go
new file mode 100644
index 000000000..49f01990e
--- /dev/null
+++ b/pkg/shim/debug.go
@@ -0,0 +1,48 @@
+// Copyright 2021 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
+//
+//     https://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 shim
+
+import (
+	"os"
+	"os/signal"
+	"runtime"
+	"sync"
+	"syscall"
+
+	"github.com/containerd/containerd/log"
+)
+
+var once sync.Once
+
+func setDebugSigHandler() {
+	once.Do(func() {
+		dumpCh := make(chan os.Signal, 1)
+		signal.Notify(dumpCh, syscall.SIGUSR2)
+		go func() {
+			buf := make([]byte, 10240)
+			for range dumpCh {
+				for {
+					n := runtime.Stack(buf, true)
+					if n >= len(buf) {
+						buf = make([]byte, 2*len(buf))
+						continue
+					}
+					log.L.Debugf("User requested stack trace:\n%s", buf[:n])
+				}
+			}
+		}()
+		log.L.Debugf("For full process dump run: kill -%d %d", syscall.SIGUSR2, os.Getpid())
+	})
+}