Add profiling commands to runsc

Example:
  runsc debug --root=<dir> \
      --profile-heap=/tmp/heap.prof \
      --profile-cpu=/tmp/cpu.prod --profile-delay=30 \
      <container ID>
PiperOrigin-RevId: 237848456
Change-Id: Icff3f20c1b157a84d0922599eaea327320dad773

8 files changed, 135 insertions, 9 deletions

diff --git a/runsc/boot/config.go b/runsc/boot/config.go
index 400203c99..626fcabdd 100644
--- a/runsc/boot/config.go
+++ b/runsc/boot/config.go
@@ -202,6 +202,9 @@ type Config struct {
 	// SIGUSR2(12) to troubleshoot hangs. -1 disables it.
 	PanicSignal int
 
+	// ProfileEnable is set to prepare the sandbox to be profiled.
+	ProfileEnable bool
+
 	// TestOnlyAllowRunAsCurrentUserWithoutChroot should only be used in
 	// tests. It allows runsc to start the sandbox process as the current
 	// user, and without chrooting the sandbox process. This can be
@@ -228,6 +231,7 @@ func (c *Config) ToFlags() []string {
 		"--strace-log-size=" + strconv.Itoa(int(c.StraceLogSize)),
 		"--watchdog-action=" + c.WatchdogAction.String(),
 		"--panic-signal=" + strconv.Itoa(c.PanicSignal),
+		"--profile=" + strconv.FormatBool(c.ProfileEnable),
 	}
 	if c.TestOnlyAllowRunAsCurrentUserWithoutChroot {
 		// Only include if set since it is never to be used by users.
diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go
index 23d476f7f..a864be720 100644
--- a/runsc/boot/controller.go
+++ b/runsc/boot/controller.go
@@ -95,6 +95,11 @@ const (
 
 	// SandboxStacks collects sandbox stacks for debugging.
 	SandboxStacks = "debug.Stacks"
+
+	// Profiling related commands (see pprof.go for more details).
+	StartCPUProfile = "Profile.StartCPUProfile"
+	StopCPUProfile  = "Profile.StopCPUProfile"
+	HeapProfile     = "Profile.HeapProfile"
 )
 
 // ControlSocketAddr generates an abstract unix socket name for the given ID.
@@ -135,6 +140,9 @@ func newController(fd int, l *Loader) (*controller, error) {
 	}
 
 	srv.Register(&debug{})
+	if l.conf.ProfileEnable {
+		srv.Register(&control.Profile{})
+	}
 
 	return &controller{
 		srv:     srv,
diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go
index bde749861..1ba5b7257 100644
--- a/runsc/boot/filter/config.go
+++ b/runsc/boot/filter/config.go
@@ -470,3 +470,16 @@ func controlServerFilters(fd int) seccomp.SyscallRules {
 		},
 	}
 }
+
+// profileFilters returns extra syscalls made by runtime/pprof package.
+func profileFilters() seccomp.SyscallRules {
+	return seccomp.SyscallRules{
+		syscall.SYS_OPENAT: []seccomp.Rule{
+			{
+				seccomp.AllowAny{},
+				seccomp.AllowAny{},
+				seccomp.AllowValue(syscall.O_RDONLY | syscall.O_LARGEFILE | syscall.O_CLOEXEC),
+			},
+		},
+	}
+}
diff --git a/runsc/boot/filter/filter.go b/runsc/boot/filter/filter.go
index d69a6a2cc..fb197f9b1 100644
--- a/runsc/boot/filter/filter.go
+++ b/runsc/boot/filter/filter.go
@@ -29,9 +29,10 @@ import (
 
 // Options are seccomp filter related options.
 type Options struct {
-	Platform     platform.Platform
-	HostNetwork  bool
-	ControllerFD int
+	Platform      platform.Platform
+	HostNetwork   bool
+	ProfileEnable bool
+	ControllerFD  int
 }
 
 // Install installs seccomp filters for based on the given platform.
@@ -47,6 +48,10 @@ func Install(opt Options) error {
 		Report("host networking enabled: syscall filters less restrictive!")
 		s.Merge(hostInetFilters())
 	}
+	if opt.ProfileEnable {
+		Report("profile enabled: syscall filters less restrictive!")
+		s.Merge(profileFilters())
+	}
 
 	switch p := opt.Platform.(type) {
 	case *ptrace.PTrace:
diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go
index 42fe6f312..4c7e6abfc 100644
--- a/runsc/boot/loader.go
+++ b/runsc/boot/loader.go
@@ -445,9 +445,10 @@ func (l *Loader) run() error {
 		filter.Report("syscall filter is DISABLED. Running in less secure mode.")
 	} else {
 		opts := filter.Options{
-			Platform:     l.k.Platform,
-			HostNetwork:  l.conf.Network == NetworkHost,
-			ControllerFD: l.ctrl.srv.FD(),
+			Platform:      l.k.Platform,
+			HostNetwork:   l.conf.Network == NetworkHost,
+			ProfileEnable: l.conf.ProfileEnable,
+			ControllerFD:  l.ctrl.srv.FD(),
 		}
 		if err := filter.Install(opts); err != nil {
 			return fmt.Errorf("installing seccomp filters: %v", err)
diff --git a/runsc/cmd/debug.go b/runsc/cmd/debug.go
index e10326754..3ee9a9b49 100644
--- a/runsc/cmd/debug.go
+++ b/runsc/cmd/debug.go
@@ -16,7 +16,9 @@ package cmd
 
 import (
 	"context"
+	"os"
 	"syscall"
+	"time"
 
 	"flag"
 	"github.com/google/subcommands"
@@ -27,9 +29,12 @@ import (
 
 // Debug implements subcommands.Command for the "debug" command.
 type Debug struct {
-	pid    int
-	stacks bool
-	signal int
+	pid          int
+	stacks       bool
+	signal       int
+	profileHeap  string
+	profileCPU   string
+	profileDelay int
 }
 
 // Name implements subcommands.Command.
@@ -51,6 +56,9 @@ func (*Debug) Usage() string {
 func (d *Debug) SetFlags(f *flag.FlagSet) {
 	f.IntVar(&d.pid, "pid", 0, "sandbox process ID. Container ID is not necessary if this is set")
 	f.BoolVar(&d.stacks, "stacks", false, "if true, dumps all sandbox stacks to the log")
+	f.StringVar(&d.profileHeap, "profile-heap", "", "writes heap profile to the given file.")
+	f.StringVar(&d.profileCPU, "profile-cpu", "", "writes CPU profile to the given file.")
+	f.IntVar(&d.profileDelay, "profile-delay", 5, "amount of time to wait before stoping CPU profile")
 	f.IntVar(&d.signal, "signal", -1, "sends signal to the sandbox")
 }
 
@@ -114,5 +122,35 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
 		}
 		log.Infof("     *** Stack dump ***\n%s", stacks)
 	}
+	if d.profileCPU != "" {
+		f, err := os.Create(d.profileCPU)
+		if err != nil {
+			Fatalf(err.Error())
+		}
+		defer f.Close()
+
+		if err := c.Sandbox.StartCPUProfile(f); err != nil {
+			Fatalf(err.Error())
+		}
+		log.Infof("CPU profile started for %d sec, writing to %q", d.profileDelay, d.profileCPU)
+		time.Sleep(time.Duration(d.profileDelay) * time.Second)
+
+		if err := c.Sandbox.StopCPUProfile(); err != nil {
+			Fatalf(err.Error())
+		}
+		log.Infof("CPU profile written to %q", d.profileCPU)
+	}
+	if d.profileHeap != "" {
+		f, err := os.Create(d.profileHeap)
+		if err != nil {
+			Fatalf(err.Error())
+		}
+		defer f.Close()
+
+		if err := c.Sandbox.HeapProfile(f); err != nil {
+			Fatalf(err.Error())
+		}
+		log.Infof("Heap profile written to %q", d.profileHeap)
+	}
 	return subcommands.ExitSuccess
 }
diff --git a/runsc/main.go b/runsc/main.go
index 4f89312b3..82c37ec11 100644
--- a/runsc/main.go
+++ b/runsc/main.go
@@ -63,6 +63,7 @@ var (
 	overlay        = flag.Bool("overlay", false, "wrap filesystem mounts with writable overlay. All modifications are stored in memory inside the sandbox.")
 	watchdogAction = flag.String("watchdog-action", "log", "sets what action the watchdog takes when triggered: log (default), panic.")
 	panicSignal    = flag.Int("panic-signal", -1, "register signal handling that panics. Usually set to SIGUSR2(12) to troubleshoot hangs. -1 disables it.")
+	profile        = flag.Bool("profile", false, "prepares the sandbox to use Golang profiler. Note that enabling profiler loosens the seccomp protection added to the sandbox (DO NOT USE IN PRODUCTION).")
 
 	testOnlyAllowRunAsCurrentUserWithoutChroot = flag.Bool("TESTONLY-unsafe-nonroot", false, "TEST ONLY; do not ever use! This skips many security measures that isolate the host from the sandbox.")
 )
@@ -146,6 +147,7 @@ func main() {
 		StraceLogSize:  *straceLogSize,
 		WatchdogAction: wa,
 		PanicSignal:    *panicSignal,
+		ProfileEnable:  *profile,
 		TestOnlyAllowRunAsCurrentUserWithoutChroot: *testOnlyAllowRunAsCurrentUserWithoutChroot,
 	}
 	if len(*straceSyscalls) != 0 {
diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go
index ce8c21681..2698e3f86 100644
--- a/runsc/sandbox/sandbox.go
+++ b/runsc/sandbox/sandbox.go
@@ -825,6 +825,61 @@ func (s *Sandbox) Stacks() (string, error) {
 	return stacks, nil
 }
 
+// HeapProfile writes a heap profile to the given file.
+func (s *Sandbox) HeapProfile(f *os.File) error {
+	log.Debugf("Heap profile %q", s.ID)
+	conn, err := s.sandboxConnect()
+	if err != nil {
+		return err
+	}
+	defer conn.Close()
+
+	opts := control.ProfileOpts{
+		FilePayload: urpc.FilePayload{
+			Files: []*os.File{f},
+		},
+	}
+	if err := conn.Call(boot.HeapProfile, &opts, nil); err != nil {
+		return fmt.Errorf("getting sandbox %q heap profile: %v", s.ID, err)
+	}
+	return nil
+}
+
+// StartCPUProfile start CPU profile writing to the given file.
+func (s *Sandbox) StartCPUProfile(f *os.File) error {
+	log.Debugf("CPU profile start %q", s.ID)
+	conn, err := s.sandboxConnect()
+	if err != nil {
+		return err
+	}
+	defer conn.Close()
+
+	opts := control.ProfileOpts{
+		FilePayload: urpc.FilePayload{
+			Files: []*os.File{f},
+		},
+	}
+	if err := conn.Call(boot.StartCPUProfile, &opts, nil); err != nil {
+		return fmt.Errorf("starting sandbox %q CPU profile: %v", s.ID, err)
+	}
+	return nil
+}
+
+// StopCPUProfile stops a previously started CPU profile.
+func (s *Sandbox) StopCPUProfile() error {
+	log.Debugf("CPU profile stop %q", s.ID)
+	conn, err := s.sandboxConnect()
+	if err != nil {
+		return err
+	}
+	defer conn.Close()
+
+	if err := conn.Call(boot.StopCPUProfile, nil, nil); err != nil {
+		return fmt.Errorf("stopping sandbox %q CPU profile: %v", s.ID, err)
+	}
+	return nil
+}
+
 // DestroyContainer destroys the given container. If it is the root container,
 // then the entire sandbox is destroyed.
 func (s *Sandbox) DestroyContainer(cid string) error {