Drop capabilities not needed by Gofer

PiperOrigin-RevId: 199808391
Change-Id: Ib37a4fb6193dc85c1f93bc16769d6aa41854b9d4

7 files changed, 124 insertions, 60 deletions

diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD
index 1746df988..73893d699 100644
--- a/runsc/boot/BUILD
+++ b/runsc/boot/BUILD
@@ -5,7 +5,6 @@ load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test")
 go_library(
     name = "boot",
     srcs = [
-        "capability.go",
         "config.go",
         "controller.go",
         "events.go",
@@ -72,7 +71,6 @@ go_library(
         "//runsc/boot/filter",
         "//runsc/specutils",
         "@com_github_opencontainers_runtime-spec//specs-go:go_default_library",
-        "@com_github_syndtr_gocapability//capability:go_default_library",
     ],
 )
 
diff --git a/runsc/cmd/BUILD b/runsc/cmd/BUILD
index a8c84a6a3..63d8036bd 100644
--- a/runsc/cmd/BUILD
+++ b/runsc/cmd/BUILD
@@ -6,6 +6,7 @@ go_library(
     name = "cmd",
     srcs = [
         "boot.go",
+        "capability.go",
         "checkpoint.go",
         "cmd.go",
         "create.go",
@@ -39,6 +40,7 @@ go_library(
         "//runsc/specutils",
         "@com_github_google_subcommands//:go_default_library",
         "@com_github_opencontainers_runtime-spec//specs-go:go_default_library",
+        "@com_github_syndtr_gocapability//capability:go_default_library",
         "@org_golang_x_sys//unix:go_default_library",
     ],
 )
diff --git a/runsc/cmd/boot.go b/runsc/cmd/boot.go
index 3bdc2ced0..34dd8b3c0 100644
--- a/runsc/cmd/boot.go
+++ b/runsc/cmd/boot.go
@@ -16,7 +16,6 @@ package cmd
 
 import (
 	"os"
-	"runtime"
 	"runtime/debug"
 	"strings"
 	"syscall"
@@ -24,7 +23,6 @@ import (
 	"context"
 	"flag"
 	"github.com/google/subcommands"
-	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"gvisor.googlesource.com/gvisor/pkg/log"
 	"gvisor.googlesource.com/gvisor/runsc/boot"
 	"gvisor.googlesource.com/gvisor/runsc/specutils"
@@ -106,8 +104,26 @@ func (b *Boot) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
 	waitStatus := args[1].(*syscall.WaitStatus)
 
 	if b.applyCaps {
-		setCapsAndCallSelf(conf, spec)
-		Fatalf("setCapsAndCallSelf must never return")
+		caps := spec.Process.Capabilities
+		if conf.Platform == boot.PlatformPtrace {
+			// Ptrace platform requires extra capabilities.
+			const c = "CAP_SYS_PTRACE"
+			caps.Bounding = append(caps.Bounding, c)
+			caps.Effective = append(caps.Effective, c)
+			caps.Permitted = append(caps.Permitted, c)
+		}
+
+		// Remove --apply-caps arg to call myself.
+		var args []string
+		for _, arg := range os.Args {
+			if !strings.Contains(arg, "apply-caps") {
+				args = append(args, arg)
+			}
+		}
+		if err := setCapsAndCallSelf(spec, args, caps); err != nil {
+			Fatalf("%v", err)
+		}
+		panic("setCapsAndCallSelf must never return success")
 	}
 
 	// Create the loader.
@@ -130,32 +146,3 @@ func (b *Boot) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
 	*waitStatus = syscall.WaitStatus(ws.Status())
 	return subcommands.ExitSuccess
 }
-
-// setCapsAndCallSelf sets capabilities to the current thread and then execve's
-// itself again with the same arguments except '--apply-caps' to restart the
-// whole process with the desired capabilities.
-func setCapsAndCallSelf(conf *boot.Config, spec *specs.Spec) {
-	// Keep thread locked while capabilities are changed.
-	runtime.LockOSThread()
-	defer runtime.UnlockOSThread()
-
-	if err := boot.ApplyCaps(conf, spec.Process.Capabilities); err != nil {
-		Fatalf("ApplyCaps, err: %v", err)
-	}
-	binPath, err := specutils.BinPath()
-	if err != nil {
-		Fatalf("%v", err)
-	}
-
-	// Remove --apply-caps arg to call myself.
-	var args []string
-	for _, arg := range os.Args {
-		if !strings.Contains(arg, "apply-caps") {
-			args = append(args, arg)
-		}
-	}
-
-	log.Infof("Execve 'boot' again, bye!")
-	log.Infof("%s %v", binPath, args)
-	syscall.Exec(binPath, args, []string{})
-}
diff --git a/runsc/boot/capability.go b/runsc/cmd/capability.go
index efa28fb97..0209feb1b 100644
--- a/runsc/boot/capability.go
+++ b/runsc/cmd/capability.go
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package boot
+package cmd
 
 import (
 	"fmt"
@@ -20,51 +20,72 @@ import (
 
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/syndtr/gocapability/capability"
+	"gvisor.googlesource.com/gvisor/pkg/log"
 )
 
-// ApplyCaps applies the capabilities in the spec to the current thread.
+// applyCaps applies the capabilities in the spec to the current thread.
 //
 // Note that it must be called with current thread locked.
-func ApplyCaps(conf *Config, caps *specs.LinuxCapabilities) error {
+func applyCaps(caps *specs.LinuxCapabilities) error {
 	setter, err := capability.NewPid2(os.Getpid())
 	if err != nil {
 		return err
 	}
+	if err := setter.Load(); err != nil {
+		return err
+	}
 
-	bounding, err := capsFromNames(caps.Bounding)
+	bounding, err := trimCaps(caps.Bounding, setter)
 	if err != nil {
 		return err
 	}
-	effective, err := capsFromNames(caps.Effective)
+	setter.Set(capability.BOUNDS, bounding...)
+
+	effective, err := trimCaps(caps.Effective, setter)
 	if err != nil {
 		return err
 	}
-	permitted, err := capsFromNames(caps.Permitted)
+	setter.Set(capability.EFFECTIVE, effective...)
+
+	permitted, err := trimCaps(caps.Permitted, setter)
 	if err != nil {
 		return err
 	}
-	inheritable, err := capsFromNames(caps.Inheritable)
+	setter.Set(capability.PERMITTED, permitted...)
+
+	inheritable, err := trimCaps(caps.Inheritable, setter)
 	if err != nil {
 		return err
 	}
-	ambient, err := capsFromNames(caps.Ambient)
+	setter.Set(capability.INHERITABLE, inheritable...)
+
+	ambient, err := trimCaps(caps.Ambient, setter)
 	if err != nil {
 		return err
 	}
+	setter.Set(capability.AMBIENT, ambient...)
 
-	// Ptrace platform requires extra capabilities.
-	if conf.Platform == PlatformPtrace {
-		bounding = append(bounding, capability.CAP_SYS_PTRACE)
-		effective = append(effective, capability.CAP_SYS_PTRACE)
-		permitted = append(permitted, capability.CAP_SYS_PTRACE)
+	return setter.Apply(capability.CAPS | capability.BOUNDS | capability.AMBS)
+}
+
+func trimCaps(names []string, setter capability.Capabilities) ([]capability.Cap, error) {
+	wantedCaps, err := capsFromNames(names)
+	if err != nil {
+		return nil, err
 	}
 
-	setter.Set(capability.BOUNDS, bounding...)
-	setter.Set(capability.PERMITTED, permitted...)
-	setter.Set(capability.INHERITABLE, inheritable...)
-	setter.Set(capability.EFFECTIVE, effective...)
-	setter.Set(capability.AMBIENT, ambient...)
-	return setter.Apply(capability.CAPS | capability.BOUNDS | capability.AMBS)
+	// Trim down capabilities that aren't possible to acquire.
+	var caps []capability.Cap
+	for _, c := range wantedCaps {
+		// Capability rules are more complicated than this, but this catches most
+		// problems with tests running with non-priviledged user.
+		if setter.Get(capability.PERMITTED, c) {
+			caps = append(caps, c)
+		} else {
+			log.Warningf("Capability %q is not permitted, dropping it.", c)
+		}
+	}
+	return caps, nil
 }
 
 func capsFromNames(names []string) ([]capability.Cap, error) {
diff --git a/runsc/cmd/cmd.go b/runsc/cmd/cmd.go
index 9f7fd6e25..940c8cd14 100644
--- a/runsc/cmd/cmd.go
+++ b/runsc/cmd/cmd.go
@@ -18,9 +18,13 @@ package cmd
 import (
 	"fmt"
 	"os"
+	"runtime"
 	"strconv"
+	"syscall"
 
+	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"gvisor.googlesource.com/gvisor/pkg/log"
+	"gvisor.googlesource.com/gvisor/runsc/specutils"
 )
 
 // Fatalf logs to stderr and exits with a failure status code.
@@ -64,3 +68,25 @@ func (i *intFlags) Set(s string) error {
 	*i = append(*i, fd)
 	return nil
 }
+
+// setCapsAndCallSelf sets capabilities to the current thread and then execve's
+// itself again with the arguments specified in 'args' to restart the process
+// with the desired capabilities.
+func setCapsAndCallSelf(spec *specs.Spec, args []string, caps *specs.LinuxCapabilities) error {
+	// Keep thread locked while capabilities are changed.
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	if err := applyCaps(caps); err != nil {
+		return fmt.Errorf("applyCaps() failed: %v", err)
+	}
+	binPath, err := specutils.BinPath()
+	if err != nil {
+		return err
+	}
+
+	log.Infof("Capabilities applied: %+v", caps)
+	log.Infof("Execve %q again, bye!", binPath)
+	syscall.Exec(binPath, args, []string{})
+	panic("unreachable")
+}
diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go
index 844e16dbf..39803bde5 100644
--- a/runsc/cmd/gofer.go
+++ b/runsc/cmd/gofer.go
@@ -15,11 +15,13 @@
 package cmd
 
 import (
+	"os"
 	"sync"
 
 	"context"
 	"flag"
 	"github.com/google/subcommands"
+	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"gvisor.googlesource.com/gvisor/pkg/log"
 	"gvisor.googlesource.com/gvisor/pkg/p9"
 	"gvisor.googlesource.com/gvisor/pkg/unet"
@@ -32,6 +34,7 @@ import (
 type Gofer struct {
 	bundleDir string
 	ioFDs     intFlags
+	applyCaps bool
 }
 
 // Name implements subcommands.Command.
@@ -53,6 +56,7 @@ func (*Gofer) Usage() string {
 func (g *Gofer) SetFlags(f *flag.FlagSet) {
 	f.StringVar(&g.bundleDir, "bundle", "", "path to the root of the bundle directory, defaults to the current directory")
 	f.Var(&g.ioFDs, "io-fds", "list of FDs to connect 9P servers. They must follow this order: root first, then mounts as defined in the spec")
+	f.BoolVar(&g.applyCaps, "apply-caps", true, "if true, apply capabilities to restrict what the Gofer process can do")
 }
 
 // Execute implements subcommands.Command.
@@ -66,6 +70,32 @@ func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
 	if err != nil {
 		Fatalf("error reading spec: %v", err)
 	}
+
+	if g.applyCaps {
+		// Minimal set of capabilities needed by the Gofer to operate on files.
+		caps := []string{
+			"CAP_CHOWN",
+			"CAP_DAC_OVERRIDE",
+			"CAP_DAC_READ_SEARCH",
+			"CAP_FOWNER",
+			"CAP_FSETID",
+		}
+		lc := &specs.LinuxCapabilities{
+			Bounding:  caps,
+			Effective: caps,
+			Permitted: caps,
+		}
+
+		// Disable caps when calling myself again.
+		// Note: minimal argument handling for the default case to keep it simple.
+		args := os.Args
+		args = append(args, "--apply-caps=false")
+		if err := setCapsAndCallSelf(spec, args, lc); err != nil {
+			Fatalf("Unable to apply caps: %v", err)
+		}
+		panic("unreachable")
+	}
+
 	specutils.LogSpec(spec)
 
 	// Start with root mount, then add any other addition mount as needed.
diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go
index 2a434cfb7..48388aa7f 100644
--- a/runsc/sandbox/sandbox.go
+++ b/runsc/sandbox/sandbox.go
@@ -295,23 +295,23 @@ func (s *Sandbox) createSandboxProcess(spec *specs.Spec, conf *boot.Config, bund
 	// process. IPC and UTS namespaces from the host are not used as they
 	// are virtualized inside the sandbox. Be paranoid and run inside an empty
 	// namespace for these.
-	log.Infof("Sandbox will be started in empty IPC and UTS namespaces")
+	log.Infof("Sandbox will be started in new IPC and UTS namespaces")
 	nss := []specs.LinuxNamespace{
 		{Type: specs.IPCNamespace},
 		{Type: specs.UTSNamespace},
 	}
 
 	if conf.Platform == boot.PlatformPtrace {
-		// TODO: Also set an empty PID namespace so that we limit
+		// TODO: Also set a new PID namespace so that we limit
 		// access to other host processes.
 		log.Infof("Sandbox will be started in the current PID namespace")
 	} else {
-		log.Infof("Sandbox will be started in empty PID namespace")
+		log.Infof("Sandbox will be started in a new PID namespace")
 		nss = append(nss, specs.LinuxNamespace{Type: specs.PIDNamespace})
 	}
 
 	if conf.FileAccess == boot.FileAccessProxy {
-		log.Infof("Sandbox will be started in empty mount namespace")
+		log.Infof("Sandbox will be started in new mount namespace")
 		nss = append(nss, specs.LinuxNamespace{Type: specs.MountNamespace})
 	} else {
 		log.Infof("Sandbox will be started in the current mount namespace")
@@ -324,7 +324,7 @@ func (s *Sandbox) createSandboxProcess(spec *specs.Spec, conf *boot.Config, bund
 		log.Infof("Sandbox will be started in the container's network namespace: %+v", ns)
 		nss = append(nss, ns)
 	} else {
-		log.Infof("Sandbox will be started in empty network namespace")
+		log.Infof("Sandbox will be started in new network namespace")
 		nss = append(nss, specs.LinuxNamespace{Type: specs.NetworkNamespace})
 	}
 
@@ -347,7 +347,7 @@ func (s *Sandbox) createSandboxProcess(spec *specs.Spec, conf *boot.Config, bund
 		cmd.Args = append(cmd.Args, "--apply-caps=true")
 
 	} else {
-		log.Infof("Sandbox will be started in empty user namespace")
+		log.Infof("Sandbox will be started in new user namespace")
 		nss = append(nss, specs.LinuxNamespace{Type: specs.UserNamespace})
 	}