Use different pidns among different containers

The different containers in a sandbox used only one pid
namespace before. This results in that a container can see
the processes in another container in the same sandbox.

This patch use different pid namespace for different containers.

Signed-off-by: chris.zn <chris.zn@antfin.com>

2 files changed, 16 insertions, 6 deletions

diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go
index 60e6c9285..3f9772b87 100644
--- a/pkg/sentry/control/proc.go
+++ b/pkg/sentry/control/proc.go
@@ -92,6 +92,9 @@ type ExecArgs struct {
 
 	// ContainerID is the container for the process being executed.
 	ContainerID string
+
+	// PIDNamespace is the pid namespace for the process being executed.
+	PIDNamespace *kernel.PIDNamespace
 }
 
 // String prints the arguments as a string.
@@ -162,6 +165,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
 		IPCNamespace:            proc.Kernel.RootIPCNamespace(),
 		AbstractSocketNamespace: proc.Kernel.RootAbstractSocketNamespace(),
 		ContainerID:             args.ContainerID,
+		PIDNamespace:            args.PIDNamespace,
 	}
 	if initArgs.Root != nil {
 		// initArgs must hold a reference on Root, which will be
@@ -341,7 +345,7 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error {
 	ts := k.TaskSet()
 	now := k.RealtimeClock().Now()
 	for _, tg := range ts.Root.ThreadGroups() {
-		pid := ts.Root.IDOfThreadGroup(tg)
+		pid := tg.PIDNamespace().IDOfThreadGroup(tg)
 		// If tg has already been reaped ignore it.
 		if pid == 0 {
 			continue
@@ -352,7 +356,7 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error {
 
 		ppid := kernel.ThreadID(0)
 		if p := tg.Leader().Parent(); p != nil {
-			ppid = ts.Root.IDOfThreadGroup(p.ThreadGroup())
+			ppid = p.PIDNamespace().IDOfThreadGroup(p.ThreadGroup())
 		}
 		*out = append(*out, &Process{
 			UID:   tg.Leader().Credentials().EffectiveKUID,
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 38b49cba2..70f5a3f0b 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -622,6 +622,9 @@ type CreateProcessArgs struct {
 	// IPCNamespace is the initial IPC namespace.
 	IPCNamespace *IPCNamespace
 
+	// PIDNamespace is the initial PID Namespace.
+	PIDNamespace *PIDNamespace
+
 	// AbstractSocketNamespace is the initial Abstract Socket namespace.
 	AbstractSocketNamespace *AbstractSocketNamespace
 
@@ -668,9 +671,7 @@ func (ctx *createProcessContext) Value(key interface{}) interface{} {
 	case CtxKernel:
 		return ctx.k
 	case CtxPIDNamespace:
-		// "The new task ... is in the root PID namespace." -
-		// Kernel.CreateProcess
-		return ctx.k.tasks.Root
+		return ctx.args.PIDNamespace
 	case CtxUTSNamespace:
 		return ctx.args.UTSNamespace
 	case CtxIPCNamespace:
@@ -745,7 +746,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
 		mounts.IncRef()
 	}
 
-	tg := k.newThreadGroup(mounts, k.tasks.Root, NewSignalHandlers(), linux.SIGCHLD, args.Limits, k.monotonicClock)
+	tg := k.newThreadGroup(mounts, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits, k.monotonicClock)
 	ctx := args.NewContext(k)
 
 	// Grab the root directory.
@@ -1018,6 +1019,11 @@ func (k *Kernel) RootIPCNamespace() *IPCNamespace {
 	return k.rootIPCNamespace
 }
 
+// RootPIDNamespace returns the root PIDNamespace.
+func (k *Kernel) RootPIDNamespace() *PIDNamespace {
+	return k.tasks.Root
+}
+
 // RootAbstractSocketNamespace returns the root AbstractSocketNamespace.
 func (k *Kernel) RootAbstractSocketNamespace() *AbstractSocketNamespace {
 	return k.rootAbstractSocketNamespace