2 files changed, 55 insertions, 18 deletions

diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 4c2d48e65..55a9d3d29 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -197,6 +197,11 @@ type Kernel struct {
 	// caches. Not all caches use it, only the caches that use host resources use
 	// the limiter. It may be nil if disabled.
 	DirentCacheLimiter *fs.DirentCacheLimiter
+
+	// unimplementedSyscallEmitter is used to emit unimplemented syscall
+	// events. This is initialized lazily on the first unimplemented
+	// syscall.
+	unimplementedSyscallEmitter eventchannel.Emitter `state:"nosave"`
 }
 
 // InitKernelArgs holds arguments to Init.
@@ -290,7 +295,6 @@ func (k *Kernel) Init(args InitKernelArgs) error {
 	k.monotonicClock = &timekeeperClock{tk: args.Timekeeper, c: sentrytime.Monotonic}
 	k.futexes = futex.NewManager()
 	k.netlinkPorts = port.New()
-
 	return nil
 }
 
@@ -586,11 +590,17 @@ func (k *Kernel) UniqueID() uint64 {
 
 // CreateProcessArgs holds arguments to kernel.CreateProcess.
 type CreateProcessArgs struct {
-	// Filename is the filename to load.
+	// Filename is the filename to load as the init binary.
 	//
-	// If this is provided as "", then the file will be guessed via Argv[0].
+	// If this is provided as "", File will be checked, then the file will be
+	// guessed via Argv[0].
 	Filename string
 
+	// File is a passed host FD pointing to a file to load as the init binary.
+	//
+	// This is checked if and only if Filename is "".
+	File *fs.File
+
 	// Argvv is a list of arguments.
 	Argv []string
 
@@ -775,8 +785,16 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
 		defer wd.DecRef()
 	}
 
-	if args.Filename == "" {
-		// Was anything provided?
+	// Check which file to start from.
+	switch {
+	case args.Filename != "":
+		// If a filename is given, take that.
+		// Set File to nil so we resolve the path in LoadTaskImage.
+		args.File = nil
+	case args.File != nil:
+		// If File is set, take the File provided directly.
+	default:
+		// Otherwise look at Argv and see if the first argument is a valid path.
 		if len(args.Argv) == 0 {
 			return nil, 0, fmt.Errorf("no filename or command provided")
 		}
@@ -788,7 +806,9 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID,
 
 	// Create a fresh task context.
 	remainingTraversals = uint(args.MaxSymlinkTraversals)
-	tc, se := k.LoadTaskImage(ctx, k.mounts, root, wd, &remainingTraversals, args.Filename, args.Argv, args.Envv, k.featureSet)
+
+	tc, se := k.LoadTaskImage(ctx, k.mounts, root, wd, &remainingTraversals, args.Filename, args.File, args.Argv, args.Envv, k.featureSet)
+
 	if se != nil {
 		return nil, 0, errors.New(se.String())
 	}
@@ -1168,16 +1188,6 @@ func (k *Kernel) SupervisorContext() context.Context {
 	}
 }
 
-// EmitUnimplementedEvent emits an UnimplementedSyscall event via the event
-// channel.
-func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) {
-	t := TaskFromContext(ctx)
-	eventchannel.Emit(&uspb.UnimplementedSyscall{
-		Tid:       int32(t.ThreadID()),
-		Registers: t.Arch().StateData().Proto(),
-	})
-}
-
 // SocketEntry represents a socket recorded in Kernel.sockets. It implements
 // refs.WeakRefUser for sockets stored in the socket table.
 //
@@ -1272,3 +1282,23 @@ func (ctx supervisorContext) Value(key interface{}) interface{} {
 		return nil
 	}
 }
+
+// Rate limits for the number of unimplemented syscall evants.
+const (
+	unimplementedSyscallsMaxRate = 100  // events per second
+	unimplementedSyscallBurst    = 1000 // events
+)
+
+// EmitUnimplementedEvent emits an UnimplementedSyscall event via the event
+// channel.
+func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) {
+	if k.unimplementedSyscallEmitter == nil {
+		k.unimplementedSyscallEmitter = eventchannel.RateLimitedEmitterFrom(eventchannel.DefaultEmitter, unimplementedSyscallsMaxRate, unimplementedSyscallBurst)
+	}
+
+	t := TaskFromContext(ctx)
+	k.unimplementedSyscallEmitter.Emit(&uspb.UnimplementedSyscall{
+		Tid:       int32(t.ThreadID()),
+		Registers: t.Arch().StateData().Proto(),
+	})
+}
diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go
index 54b1676b0..8639d379f 100644
--- a/pkg/sentry/kernel/task_context.go
+++ b/pkg/sentry/kernel/task_context.go
@@ -140,15 +140,22 @@ func (t *Task) Stack() *arch.Stack {
 //  * wd: Working directory to lookup filename under
 //  * maxTraversals: maximum number of symlinks to follow
 //  * filename: path to binary to load
+//  * file: an open fs.File object of the binary to load. If set,
+//  file will be loaded and not filename.
 //  * argv: Binary argv
 //  * envv: Binary envv
 //  * fs: Binary FeatureSet
-func (k *Kernel) LoadTaskImage(ctx context.Context, mounts *fs.MountNamespace, root, wd *fs.Dirent, maxTraversals *uint, filename string, argv, envv []string, fs *cpuid.FeatureSet) (*TaskContext, *syserr.Error) {
+func (k *Kernel) LoadTaskImage(ctx context.Context, mounts *fs.MountNamespace, root, wd *fs.Dirent, maxTraversals *uint, filename string, file *fs.File, argv, envv []string, fs *cpuid.FeatureSet) (*TaskContext, *syserr.Error) {
+	// If File is not nil, we should load that instead of resolving filename.
+	if file != nil {
+		filename = file.MappedName(ctx)
+	}
+
 	// Prepare a new user address space to load into.
 	m := mm.NewMemoryManager(k, k)
 	defer m.DecUsers(ctx)
 
-	os, ac, name, err := loader.Load(ctx, m, mounts, root, wd, maxTraversals, fs, filename, argv, envv, k.extraAuxv, k.vdso)
+	os, ac, name, err := loader.Load(ctx, m, mounts, root, wd, maxTraversals, fs, filename, file, argv, envv, k.extraAuxv, k.vdso)
 	if err != nil {
 		return nil, err
 	}