Initial network namespace support.

TCP/IP will work with netstack networking. hostinet doesn't work, and sockets
will have the same behavior as it is now.

Before the userspace is able to create device, the default loopback device can
be used to test.

/proc/net and /sys/net will still be connected to the root network stack; this
is the same behavior now.

Issue #1833

PiperOrigin-RevId: 296309389

1 files changed, 16 insertions, 10 deletions

diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 7da0368f1..c62fd6eb1 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -111,7 +111,7 @@ type Kernel struct {
 	timekeeper                  *Timekeeper
 	tasks                       *TaskSet
 	rootUserNamespace           *auth.UserNamespace
-	networkStack                inet.Stack `state:"nosave"`
+	rootNetworkNamespace        *inet.Namespace
 	applicationCores            uint
 	useHostCores                bool
 	extraAuxv                   []arch.AuxEntry
@@ -260,8 +260,9 @@ type InitKernelArgs struct {
 	// RootUserNamespace is the root user namespace.
 	RootUserNamespace *auth.UserNamespace
 
-	// NetworkStack is the TCP/IP network stack. NetworkStack may be nil.
-	NetworkStack inet.Stack
+	// RootNetworkNamespace is the root network namespace. If nil, no networking
+	// will be available.
+	RootNetworkNamespace *inet.Namespace
 
 	// ApplicationCores is the number of logical CPUs visible to sandboxed
 	// applications. The set of logical CPU IDs is [0, ApplicationCores); thus
@@ -320,7 +321,10 @@ func (k *Kernel) Init(args InitKernelArgs) error {
 	k.rootUTSNamespace = args.RootUTSNamespace
 	k.rootIPCNamespace = args.RootIPCNamespace
 	k.rootAbstractSocketNamespace = args.RootAbstractSocketNamespace
-	k.networkStack = args.NetworkStack
+	k.rootNetworkNamespace = args.RootNetworkNamespace
+	if k.rootNetworkNamespace == nil {
+		k.rootNetworkNamespace = inet.NewRootNamespace(nil, nil)
+	}
 	k.applicationCores = args.ApplicationCores
 	if args.UseHostCores {
 		k.useHostCores = true
@@ -543,8 +547,6 @@ func (ts *TaskSet) unregisterEpollWaiters() {
 func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) error {
 	loadStart := time.Now()
 
-	k.networkStack = net
-
 	initAppCores := k.applicationCores
 
 	// Load the pre-saved CPUID FeatureSet.
@@ -575,6 +577,10 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks)
 	log.Infof("Kernel load stats: %s", &stats)
 	log.Infof("Kernel load took [%s].", time.Since(kernelStart))
 
+	// rootNetworkNamespace should be populated after loading the state file.
+	// Restore the root network stack.
+	k.rootNetworkNamespace.RestoreRootStack(net)
+
 	// Load the memory file's state.
 	memoryStart := time.Now()
 	if err := k.mf.LoadFrom(k.SupervisorContext(), r); err != nil {
@@ -905,6 +911,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
 		FSContext:               fsContext,
 		FDTable:                 args.FDTable,
 		Credentials:             args.Credentials,
+		NetworkNamespace:        k.RootNetworkNamespace(),
 		AllowedCPUMask:          sched.NewFullCPUSet(k.applicationCores),
 		UTSNamespace:            args.UTSNamespace,
 		IPCNamespace:            args.IPCNamespace,
@@ -1255,10 +1262,9 @@ func (k *Kernel) RootAbstractSocketNamespace() *AbstractSocketNamespace {
 	return k.rootAbstractSocketNamespace
 }
 
-// NetworkStack returns the network stack. NetworkStack may return nil if no
-// network stack is available.
-func (k *Kernel) NetworkStack() inet.Stack {
-	return k.networkStack
+// RootNetworkNamespace returns the root network namespace, always non-nil.
+func (k *Kernel) RootNetworkNamespace() *inet.Namespace {
+	return k.rootNetworkNamespace
 }
 
 // GlobalInit returns the thread group with ID 1 in the root PID namespace, or