Merge branch 'master' into ip-forwarding

- Merges aleksej-paschenko's with HEAD
- Adds vfs2 support for ip_forward

5 files changed, 167 insertions, 101 deletions

diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD
index 5522cecd0..2c5d14be5 100644
--- a/pkg/sentry/control/BUILD
+++ b/pkg/sentry/control/BUILD
@@ -1,5 +1,4 @@
-load("//tools/go_stateify:defs.bzl", "go_library")
-load("@io_bazel_rules_go//go:def.bzl", "go_test")
+load("//tools:defs.bzl", "go_library", "go_test")
 
 package(licenses = ["notice"])
 
@@ -12,16 +11,18 @@ go_library(
         "proc.go",
         "state.go",
     ],
-    importpath = "gvisor.dev/gvisor/pkg/sentry/control",
     visibility = [
-        "//pkg/sentry:internal",
+        "//:sandbox",
     ],
     deps = [
         "//pkg/abi/linux",
         "//pkg/fd",
         "//pkg/log",
+        "//pkg/sentry/fdimport",
         "//pkg/sentry/fs",
         "//pkg/sentry/fs/host",
+        "//pkg/sentry/fs/user",
+        "//pkg/sentry/fsimpl/host",
         "//pkg/sentry/kernel",
         "//pkg/sentry/kernel/auth",
         "//pkg/sentry/kernel/time",
@@ -29,9 +30,12 @@ go_library(
         "//pkg/sentry/state",
         "//pkg/sentry/strace",
         "//pkg/sentry/usage",
+        "//pkg/sentry/vfs",
         "//pkg/sentry/watchdog",
+        "//pkg/sync",
         "//pkg/tcpip/link/sniffer",
         "//pkg/urpc",
+        "@org_golang_x_sys//unix:go_default_library",
     ],
 )
 
@@ -39,7 +43,7 @@ go_test(
     name = "control_test",
     size = "small",
     srcs = ["proc_test.go"],
-    embed = [":control"],
+    library = ":control",
     deps = [
         "//pkg/log",
         "//pkg/sentry/kernel/time",
diff --git a/pkg/sentry/control/logging.go b/pkg/sentry/control/logging.go
index 811f24324..8a500a515 100644
--- a/pkg/sentry/control/logging.go
+++ b/pkg/sentry/control/logging.go
@@ -70,8 +70,8 @@ type LoggingArgs struct {
 type Logging struct{}
 
 // Change will change the log level and strace arguments. Although
-// this functions signature requires an error it never acctually
-// return san error. It's required by the URPC interface.
+// this functions signature requires an error it never actually
+// returns an error. It's required by the URPC interface.
 // Additionally, it may look odd that this is the only method
 // attached to an empty struct but this is also part of how
 // URPC dispatches.
diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go
index 1f78d54a2..2bf3c45e1 100644
--- a/pkg/sentry/control/pprof.go
+++ b/pkg/sentry/control/pprof.go
@@ -19,9 +19,10 @@ import (
 	"runtime"
 	"runtime/pprof"
 	"runtime/trace"
-	"sync"
 
 	"gvisor.dev/gvisor/pkg/fd"
+	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/urpc"
 )
 
@@ -48,6 +49,9 @@ type ProfileOpts struct {
 // - dump out the stack trace of current go routines.
 //   sentryctl -pid <pid> pprof-goroutine
 type Profile struct {
+	// Kernel is the kernel under profile. It's immutable.
+	Kernel *kernel.Kernel
+
 	// mu protects the fields below.
 	mu sync.Mutex
 
@@ -113,9 +117,9 @@ func (p *Profile) HeapProfile(o *ProfileOpts, _ *struct{}) error {
 	return nil
 }
 
-// Goroutine is an RPC stub which dumps out the stack trace for all running
-// goroutines.
-func (p *Profile) Goroutine(o *ProfileOpts, _ *struct{}) error {
+// GoroutineProfile is an RPC stub which dumps out the stack trace for all
+// running goroutines.
+func (p *Profile) GoroutineProfile(o *ProfileOpts, _ *struct{}) error {
 	if len(o.FilePayload.Files) < 1 {
 		return errNoOutput
 	}
@@ -127,6 +131,34 @@ func (p *Profile) Goroutine(o *ProfileOpts, _ *struct{}) error {
 	return nil
 }
 
+// BlockProfile is an RPC stub which dumps out the stack trace that led to
+// blocking on synchronization primitives.
+func (p *Profile) BlockProfile(o *ProfileOpts, _ *struct{}) error {
+	if len(o.FilePayload.Files) < 1 {
+		return errNoOutput
+	}
+	output := o.FilePayload.Files[0]
+	defer output.Close()
+	if err := pprof.Lookup("block").WriteTo(output, 0); err != nil {
+		return err
+	}
+	return nil
+}
+
+// MutexProfile is an RPC stub which dumps out the stack trace of holders of
+// contended mutexes.
+func (p *Profile) MutexProfile(o *ProfileOpts, _ *struct{}) error {
+	if len(o.FilePayload.Files) < 1 {
+		return errNoOutput
+	}
+	output := o.FilePayload.Files[0]
+	defer output.Close()
+	if err := pprof.Lookup("mutex").WriteTo(output, 0); err != nil {
+		return err
+	}
+	return nil
+}
+
 // StartTrace is an RPC stub which starts collection of an execution trace.
 func (p *Profile) StartTrace(o *ProfileOpts, _ *struct{}) error {
 	if len(o.FilePayload.Files) < 1 {
@@ -147,6 +179,9 @@ func (p *Profile) StartTrace(o *ProfileOpts, _ *struct{}) error {
 		return err
 	}
 
+	// Ensure all trace contexts are registered.
+	p.Kernel.RebuildTraceContexts()
+
 	p.traceFile = output
 	return nil
 }
@@ -158,9 +193,15 @@ func (p *Profile) StopTrace(_, _ *struct{}) error {
 	defer p.mu.Unlock()
 
 	if p.traceFile == nil {
-		return errors.New("Execution tracing not start")
+		return errors.New("Execution tracing not started")
 	}
 
+	// Similarly to the case above, if tasks have not ended traces, we will
+	// lose information. Thus we need to rebuild the tasks in order to have
+	// complete information. This will not lose information if multiple
+	// traces are overlapping.
+	p.Kernel.RebuildTraceContexts()
+
 	trace.Stop()
 	p.traceFile.Close()
 	p.traceFile = nil
diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go
index c35faeb4c..dfa936563 100644
--- a/pkg/sentry/control/proc.go
+++ b/pkg/sentry/control/proc.go
@@ -23,14 +23,19 @@ import (
 	"text/tabwriter"
 	"time"
 
+	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/sentry/fdimport"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/fs/host"
+	"gvisor.dev/gvisor/pkg/sentry/fs/user"
+	hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
 	ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
 	"gvisor.dev/gvisor/pkg/sentry/limits"
 	"gvisor.dev/gvisor/pkg/sentry/usage"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/urpc"
 )
 
@@ -60,6 +65,12 @@ type ExecArgs struct {
 	// process's MountNamespace.
 	MountNamespace *fs.MountNamespace
 
+	// MountNamespaceVFS2 is the mount namespace to execute the new process in.
+	// A reference on MountNamespace must be held for the lifetime of the
+	// ExecArgs. If MountNamespace is nil, it will default to the init
+	// process's MountNamespace.
+	MountNamespaceVFS2 *vfs.MountNamespace
+
 	// WorkingDirectory defines the working directory for the new process.
 	WorkingDirectory string `json:"wd"`
 
@@ -71,15 +82,13 @@ type ExecArgs struct {
 	// the root group if not set explicitly.
 	KGID auth.KGID
 
-	// ExtraKGIDs is the list of additional groups to which the user
-	// belongs.
+	// ExtraKGIDs is the list of additional groups to which the user belongs.
 	ExtraKGIDs []auth.KGID
 
 	// Capabilities is the list of capabilities to give to the process.
 	Capabilities *auth.TaskCapabilities
 
-	// StdioIsPty indicates that FDs 0, 1, and 2 are connected to a host
-	// pty FD.
+	// StdioIsPty indicates that FDs 0, 1, and 2 are connected to a host pty FD.
 	StdioIsPty bool
 
 	// FilePayload determines the files to give to the new process.
@@ -94,6 +103,9 @@ type ExecArgs struct {
 
 // String prints the arguments as a string.
 func (args ExecArgs) String() string {
+	if len(args.Argv) == 0 {
+		return args.Filename
+	}
 	a := make([]string, len(args.Argv))
 	copy(a, args.Argv)
 	if args.Filename != "" {
@@ -104,7 +116,7 @@ func (args ExecArgs) String() string {
 
 // Exec runs a new task.
 func (proc *Proc) Exec(args *ExecArgs, waitStatus *uint32) error {
-	newTG, _, _, err := proc.execAsync(args)
+	newTG, _, _, _, err := proc.execAsync(args)
 	if err != nil {
 		return err
 	}
@@ -117,25 +129,16 @@ func (proc *Proc) Exec(args *ExecArgs, waitStatus *uint32) error {
 
 // ExecAsync runs a new task, but doesn't wait for it to finish. It is defined
 // as a function rather than a method to avoid exposing execAsync as an RPC.
-func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, error) {
+func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
 	return proc.execAsync(args)
 }
 
 // execAsync runs a new task, but doesn't wait for it to finish. It returns the
 // newly created thread group and its PID. If the stdio FDs are TTYs, then a
 // TTYFileOperations that wraps the TTY is also returned.
-func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, error) {
+func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) {
 	// Import file descriptors.
 	fdTable := proc.Kernel.NewFDTable()
-	defer fdTable.DecRef()
-
-	// No matter what happens, we should close all files in the FilePayload
-	// before returning. Any files that are imported will be duped.
-	defer func() {
-		for _, f := range args.FilePayload.Files {
-			f.Close()
-		}
-	}()
 
 	creds := auth.NewUserCredentials(
 		args.KUID,
@@ -150,6 +153,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
 		Envv:                    args.Envv,
 		WorkingDirectory:        args.WorkingDirectory,
 		MountNamespace:          args.MountNamespace,
+		MountNamespaceVFS2:      args.MountNamespaceVFS2,
 		Credentials:             creds,
 		FDTable:                 fdTable,
 		Umask:                   0022,
@@ -166,81 +170,80 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI
 		// be donated to the new process in CreateProcess.
 		initArgs.MountNamespace.IncRef()
 	}
+	if initArgs.MountNamespaceVFS2 != nil {
+		// initArgs must hold a reference on MountNamespaceVFS2, which will
+		// be donated to the new process in CreateProcess.
+		initArgs.MountNamespaceVFS2.IncRef()
+	}
 	ctx := initArgs.NewContext(proc.Kernel)
+	defer fdTable.DecRef(ctx)
 
-	if initArgs.Filename == "" {
+	if kernel.VFS2Enabled {
 		// Get the full path to the filename from the PATH env variable.
-		paths := fs.GetPath(initArgs.Envv)
-		mns := initArgs.MountNamespace
-		if mns == nil {
-			mns = proc.Kernel.GlobalInit().Leader().MountNamespace()
+		if initArgs.MountNamespaceVFS2 == nil {
+			// Set initArgs so that 'ctx' returns the namespace.
+			//
+			// MountNamespaceVFS2 adds a reference to the namespace, which is
+			// transferred to the new process.
+			initArgs.MountNamespaceVFS2 = proc.Kernel.GlobalInit().Leader().MountNamespaceVFS2()
 		}
-		f, err := mns.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths)
-		if err != nil {
-			return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err)
+	} else {
+		if initArgs.MountNamespace == nil {
+			// Set initArgs so that 'ctx' returns the namespace.
+			initArgs.MountNamespace = proc.Kernel.GlobalInit().Leader().MountNamespace()
+
+			// initArgs must hold a reference on MountNamespace, which will
+			// be donated to the new process in CreateProcess.
+			initArgs.MountNamespace.IncRef()
 		}
-		initArgs.Filename = f
 	}
+	resolved, err := user.ResolveExecutablePath(ctx, &initArgs)
+	if err != nil {
+		return nil, 0, nil, nil, err
+	}
+	initArgs.Filename = resolved
 
-	mounter := fs.FileOwnerFromContext(ctx)
-
-	var ttyFile *fs.File
-	for appFD, hostFile := range args.FilePayload.Files {
-		var appFile *fs.File
-
-		if args.StdioIsPty && appFD < 3 {
-			// Import the file as a host TTY file.
-			if ttyFile == nil {
-				var err error
-				appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, true /* isTTY */)
-				if err != nil {
-					return nil, 0, nil, err
-				}
-				defer appFile.DecRef()
-
-				// Remember this in the TTY file, as we will
-				// use it for the other stdio FDs.
-				ttyFile = appFile
-			} else {
-				// Re-use the existing TTY file, as all three
-				// stdio FDs must point to the same fs.File in
-				// order to share TTY state, specifically the
-				// foreground process group id.
-				appFile = ttyFile
-			}
-		} else {
-			// Import the file as a regular host file.
-			var err error
-			appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, false /* isTTY */)
+	fds := make([]int, len(args.FilePayload.Files))
+	for i, file := range args.FilePayload.Files {
+		if kernel.VFS2Enabled {
+			// Need to dup to remove ownership from os.File.
+			dup, err := unix.Dup(int(file.Fd()))
 			if err != nil {
-				return nil, 0, nil, err
+				return nil, 0, nil, nil, fmt.Errorf("duplicating payload files: %w", err)
 			}
-			defer appFile.DecRef()
+			fds[i] = dup
+		} else {
+			// VFS1 dups the file on import.
+			fds[i] = int(file.Fd())
 		}
-
-		// Add the file to the FD map.
-		if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil {
-			return nil, 0, nil, err
+	}
+	ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, args.StdioIsPty, fds)
+	if err != nil {
+		if kernel.VFS2Enabled {
+			for _, fd := range fds {
+				unix.Close(fd)
+			}
 		}
+		return nil, 0, nil, nil, err
 	}
 
 	tg, tid, err := proc.Kernel.CreateProcess(initArgs)
 	if err != nil {
-		return nil, 0, nil, err
+		return nil, 0, nil, nil, err
 	}
 
-	var ttyFileOps *host.TTYFileOperations
-	if ttyFile != nil {
-		// Set the foreground process group on the TTY before starting
-		// the process.
-		ttyFileOps = ttyFile.FileOperations.(*host.TTYFileOperations)
-		ttyFileOps.InitForegroundProcessGroup(tg.ProcessGroup())
+	// Set the foreground process group on the TTY before starting the process.
+	switch {
+	case ttyFile != nil:
+		ttyFile.InitForegroundProcessGroup(tg.ProcessGroup())
+	case ttyFileVFS2 != nil:
+		ttyFileVFS2.InitForegroundProcessGroup(tg.ProcessGroup())
 	}
 
 	// Start the newly created process.
 	proc.Kernel.StartProcess(tg)
 
-	return tg, tid, ttyFileOps, nil
+	return tg, tid, ttyFile, ttyFileVFS2, nil
 }
 
 // PsArgs is the set of arguments to ps.
@@ -268,14 +271,17 @@ func (proc *Proc) Ps(args *PsArgs, out *string) error {
 }
 
 // Process contains information about a single process in a Sandbox.
-// TODO(b/117881927): Implement TTY field.
 type Process struct {
 	UID auth.KUID       `json:"uid"`
 	PID kernel.ThreadID `json:"pid"`
 	// Parent PID
-	PPID kernel.ThreadID `json:"ppid"`
+	PPID    kernel.ThreadID   `json:"ppid"`
+	Threads []kernel.ThreadID `json:"threads"`
 	// Processor utilization
 	C int32 `json:"c"`
+	// TTY name of the process. Will be of the form "pts/N" if there is a
+	// TTY, or "?" if there is not.
+	TTY string `json:"tty"`
 	// Start time
 	STime string `json:"stime"`
 	// CPU time
@@ -285,18 +291,19 @@ type Process struct {
 }
 
 // ProcessListToTable prints a table with the following format:
-// UID       PID       PPID      C         STIME     TIME       CMD
-// 0         1         0         0         14:04     505262ns   tail
+// UID       PID       PPID      C         TTY		STIME     TIME       CMD
+// 0         1         0         0         pty/4	14:04     505262ns   tail
 func ProcessListToTable(pl []*Process) string {
 	var buf bytes.Buffer
 	tw := tabwriter.NewWriter(&buf, 10, 1, 3, ' ', 0)
-	fmt.Fprint(tw, "UID\tPID\tPPID\tC\tSTIME\tTIME\tCMD")
+	fmt.Fprint(tw, "UID\tPID\tPPID\tC\tTTY\tSTIME\tTIME\tCMD")
 	for _, d := range pl {
-		fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s",
+		fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s\t%s",
 			d.UID,
 			d.PID,
 			d.PPID,
 			d.C,
+			d.TTY,
 			d.STime,
 			d.Time,
 			d.Cmd)
@@ -307,7 +314,7 @@ func ProcessListToTable(pl []*Process) string {
 
 // ProcessListToJSON will return the JSON representation of ps.
 func ProcessListToJSON(pl []*Process) (string, error) {
-	b, err := json.Marshal(pl)
+	b, err := json.MarshalIndent(pl, "", "  ")
 	if err != nil {
 		return "", fmt.Errorf("couldn't marshal process list %v: %v", pl, err)
 	}
@@ -334,7 +341,9 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error {
 	ts := k.TaskSet()
 	now := k.RealtimeClock().Now()
 	for _, tg := range ts.Root.ThreadGroups() {
-		pid := tg.PIDNamespace().IDOfThreadGroup(tg)
+		pidns := tg.PIDNamespace()
+		pid := pidns.IDOfThreadGroup(tg)
+
 		// If tg has already been reaped ignore it.
 		if pid == 0 {
 			continue
@@ -345,16 +354,19 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error {
 
 		ppid := kernel.ThreadID(0)
 		if p := tg.Leader().Parent(); p != nil {
-			ppid = p.PIDNamespace().IDOfThreadGroup(p.ThreadGroup())
+			ppid = pidns.IDOfThreadGroup(p.ThreadGroup())
 		}
+		threads := tg.MemberIDs(pidns)
 		*out = append(*out, &Process{
-			UID:   tg.Leader().Credentials().EffectiveKUID,
-			PID:   pid,
-			PPID:  ppid,
-			STime: formatStartTime(now, tg.Leader().StartTime()),
-			C:     percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now),
-			Time:  tg.CPUStats().SysTime.String(),
-			Cmd:   tg.Leader().Name(),
+			UID:     tg.Leader().Credentials().EffectiveKUID,
+			PID:     pid,
+			PPID:    ppid,
+			Threads: threads,
+			STime:   formatStartTime(now, tg.Leader().StartTime()),
+			C:       percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now),
+			Time:    tg.CPUStats().SysTime.String(),
+			Cmd:     tg.Leader().Name(),
+			TTY:     ttyName(tg.TTY()),
 		})
 	}
 	sort.Slice(*out, func(i, j int) bool { return (*out)[i].PID < (*out)[j].PID })
@@ -395,3 +407,10 @@ func percentCPU(stats usage.CPUStats, startTime, now ktime.Time) int32 {
 	}
 	return int32(percentCPU)
 }
+
+func ttyName(tty *kernel.TTY) string {
+	if tty == nil {
+		return "?"
+	}
+	return fmt.Sprintf("pts/%d", tty.Index)
+}
diff --git a/pkg/sentry/control/proc_test.go b/pkg/sentry/control/proc_test.go
index d8ada2694..0a88459b2 100644
--- a/pkg/sentry/control/proc_test.go
+++ b/pkg/sentry/control/proc_test.go
@@ -34,7 +34,7 @@ func TestProcessListTable(t *testing.T) {
 	}{
 		{
 			pl:       []*Process{},
-			expected: "UID       PID       PPID      C         STIME     TIME      CMD",
+			expected: "UID       PID       PPID      C         TTY       STIME     TIME      CMD",
 		},
 		{
 			pl: []*Process{
@@ -43,6 +43,7 @@ func TestProcessListTable(t *testing.T) {
 					PID:   0,
 					PPID:  0,
 					C:     0,
+					TTY:   "?",
 					STime: "0",
 					Time:  "0",
 					Cmd:   "zero",
@@ -52,14 +53,15 @@ func TestProcessListTable(t *testing.T) {
 					PID:   1,
 					PPID:  1,
 					C:     1,
+					TTY:   "pts/4",
 					STime: "1",
 					Time:  "1",
 					Cmd:   "one",
 				},
 			},
-			expected: `UID       PID       PPID      C         STIME     TIME      CMD
-0         0         0         0         0         0         zero
-1         1         1         1         1         1         one`,
+			expected: `UID       PID       PPID      C         TTY       STIME     TIME      CMD
+0         0         0         0         ?         0         0         zero
+1         1         1         1         pts/4     1         1         one`,
 		},
 	}