summaryrefslogtreecommitdiffhomepage
path: root/runsc
diff options
context:
space:
mode:
Diffstat (limited to 'runsc')
-rw-r--r--runsc/BUILD7
-rw-r--r--runsc/boot/BUILD2
-rw-r--r--runsc/boot/config.go4
-rw-r--r--runsc/boot/loader.go49
-rw-r--r--runsc/boot/user.go28
-rw-r--r--runsc/boot/user_test.go3
-rw-r--r--runsc/cmd/exec.go53
-rw-r--r--runsc/cmd/exec_test.go4
-rw-r--r--runsc/cmd/gofer.go5
-rw-r--r--runsc/container/BUILD1
-rw-r--r--runsc/container/container_test.go25
-rw-r--r--runsc/container/test_app/test_app.go65
-rw-r--r--runsc/dockerutil/dockerutil.go17
-rw-r--r--runsc/fsgofer/filter/config.go13
-rw-r--r--runsc/fsgofer/filter/filter.go13
-rw-r--r--runsc/fsgofer/fsgofer.go70
-rw-r--r--runsc/fsgofer/fsgofer_test.go2
-rw-r--r--runsc/main.go4
-rw-r--r--runsc/sandbox/sandbox.go2
-rw-r--r--runsc/specutils/BUILD1
-rw-r--r--runsc/specutils/namespace.go14
-rw-r--r--runsc/specutils/specutils.go10
22 files changed, 302 insertions, 90 deletions
diff --git a/runsc/BUILD b/runsc/BUILD
index 5e7dacb87..e4e8e64a3 100644
--- a/runsc/BUILD
+++ b/runsc/BUILD
@@ -1,7 +1,7 @@
package(licenses = ["notice"]) # Apache 2.0
load("@io_bazel_rules_go//go:def.bzl", "go_binary")
-load("@bazel_tools//tools/build_defs/pkg:pkg.bzl", "pkg_deb", "pkg_tar")
+load("@rules_pkg//:pkg.bzl", "pkg_deb", "pkg_tar")
go_binary(
name = "runsc",
@@ -91,11 +91,6 @@ pkg_deb(
maintainer = "The gVisor Authors <gvisor-dev@googlegroups.com>",
package = "runsc",
postinst = "debian/postinst.sh",
- tags = [
- # TODO(b/135475885): pkg_deb requires python2:
- # https://github.com/bazelbuild/bazel/issues/8443
- "manual",
- ],
version_file = ":version.txt",
visibility = [
"//visibility:public",
diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD
index 588bb8851..d90381c0f 100644
--- a/runsc/boot/BUILD
+++ b/runsc/boot/BUILD
@@ -80,6 +80,7 @@ go_library(
"//pkg/tcpip/network/ipv6",
"//pkg/tcpip/stack",
"//pkg/tcpip/transport/icmp",
+ "//pkg/tcpip/transport/raw",
"//pkg/tcpip/transport/tcp",
"//pkg/tcpip/transport/udp",
"//pkg/urpc",
@@ -109,6 +110,7 @@ go_test(
"//pkg/sentry/arch:registers_go_proto",
"//pkg/sentry/context/contexttest",
"//pkg/sentry/fs",
+ "//pkg/sentry/kernel/auth",
"//pkg/unet",
"//runsc/fsgofer",
"@com_github_opencontainers_runtime-spec//specs-go:go_default_library",
diff --git a/runsc/boot/config.go b/runsc/boot/config.go
index 31103367d..38278d0a2 100644
--- a/runsc/boot/config.go
+++ b/runsc/boot/config.go
@@ -167,6 +167,9 @@ type Config struct {
// Overlay is whether to wrap the root filesystem in an overlay.
Overlay bool
+ // FSGoferHostUDS enables the gofer to mount a host UDS.
+ FSGoferHostUDS bool
+
// Network indicates what type of network to use.
Network NetworkType
@@ -253,6 +256,7 @@ func (c *Config) ToFlags() []string {
"--debug-log-format=" + c.DebugLogFormat,
"--file-access=" + c.FileAccess.String(),
"--overlay=" + strconv.FormatBool(c.Overlay),
+ "--fsgofer-host-uds=" + strconv.FormatBool(c.FSGoferHostUDS),
"--network=" + c.Network.String(),
"--log-packets=" + strconv.FormatBool(c.LogPackets),
"--platform=" + c.Platform,
diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go
index 823a34619..adf345490 100644
--- a/runsc/boot/loader.go
+++ b/runsc/boot/loader.go
@@ -20,7 +20,6 @@ import (
mrand "math/rand"
"os"
"runtime"
- "strings"
"sync"
"sync/atomic"
"syscall"
@@ -55,6 +54,7 @@ import (
"gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
"gvisor.dev/gvisor/pkg/tcpip/stack"
"gvisor.dev/gvisor/pkg/tcpip/transport/icmp"
+ "gvisor.dev/gvisor/pkg/tcpip/transport/raw"
"gvisor.dev/gvisor/pkg/tcpip/transport/tcp"
"gvisor.dev/gvisor/pkg/tcpip/transport/udp"
"gvisor.dev/gvisor/runsc/boot/filter"
@@ -535,23 +535,12 @@ func (l *Loader) run() error {
return err
}
- // Read /etc/passwd for the user's HOME directory and set the HOME
- // environment variable as required by POSIX if it is not overridden by
- // the user.
- hasHomeEnvv := false
- for _, envv := range l.rootProcArgs.Envv {
- if strings.HasPrefix(envv, "HOME=") {
- hasHomeEnvv = true
- }
- }
- if !hasHomeEnvv {
- homeDir, err := getExecUserHome(ctx, l.rootProcArgs.MountNamespace, uint32(l.rootProcArgs.Credentials.RealKUID))
- if err != nil {
- return fmt.Errorf("error reading exec user: %v", err)
- }
-
- l.rootProcArgs.Envv = append(l.rootProcArgs.Envv, "HOME="+homeDir)
+ // Add the HOME enviroment variable if it is not already set.
+ envv, err := maybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv)
+ if err != nil {
+ return err
}
+ l.rootProcArgs.Envv = envv
// Create the root container init task. It will begin running
// when the kernel is started.
@@ -815,6 +804,16 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) {
})
defer args.MountNamespace.DecRef()
+ // Add the HOME enviroment varible if it is not already set.
+ root := args.MountNamespace.Root()
+ defer root.DecRef()
+ ctx := fs.WithRoot(l.k.SupervisorContext(), root)
+ envv, err := maybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv)
+ if err != nil {
+ return 0, err
+ }
+ args.Envv = envv
+
// Start the process.
proc := control.Proc{Kernel: l.k}
args.PIDNamespace = tg.PIDNamespace()
@@ -913,15 +912,17 @@ func newEmptyNetworkStack(conf *Config, clock tcpip.Clock) (inet.Stack, error) {
case NetworkNone, NetworkSandbox:
// NetworkNone sets up loopback using netstack.
- netProtos := []string{ipv4.ProtocolName, ipv6.ProtocolName, arp.ProtocolName}
- protoNames := []string{tcp.ProtocolName, udp.ProtocolName, icmp.ProtocolName4}
- s := epsocket.Stack{stack.New(netProtos, protoNames, stack.Options{
- Clock: clock,
- Stats: epsocket.Metrics,
- HandleLocal: true,
+ netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()}
+ transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol(), icmp.NewProtocol4()}
+ s := epsocket.Stack{stack.New(stack.Options{
+ NetworkProtocols: netProtos,
+ TransportProtocols: transProtos,
+ Clock: clock,
+ Stats: epsocket.Metrics,
+ HandleLocal: true,
// Enable raw sockets for users with sufficient
// privileges.
- Raw: true,
+ UnassociatedFactory: raw.EndpointFactory{},
})}
// Enable SACK Recovery.
diff --git a/runsc/boot/user.go b/runsc/boot/user.go
index d1d423a5c..56cc12ee0 100644
--- a/runsc/boot/user.go
+++ b/runsc/boot/user.go
@@ -16,6 +16,7 @@ package boot
import (
"bufio"
+ "fmt"
"io"
"strconv"
"strings"
@@ -23,6 +24,7 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/sentry/context"
"gvisor.dev/gvisor/pkg/sentry/fs"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/usermem"
)
@@ -42,7 +44,7 @@ func (r *fileReader) Read(buf []byte) (int, error) {
// getExecUserHome returns the home directory of the executing user read from
// /etc/passwd as read from the container filesystem.
-func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid uint32) (string, error) {
+func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.KUID) (string, error) {
// The default user home directory to return if no user matching the user
// if found in the /etc/passwd found in the image.
const defaultHome = "/"
@@ -82,7 +84,7 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid uint32
File: f,
}
- homeDir, err := findHomeInPasswd(uid, r, defaultHome)
+ homeDir, err := findHomeInPasswd(uint32(uid), r, defaultHome)
if err != nil {
return "", err
}
@@ -90,6 +92,28 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid uint32
return homeDir, nil
}
+// maybeAddExecUserHome returns a new slice with the HOME enviroment variable
+// set if the slice does not already contain it, otherwise it returns the
+// original slice unmodified.
+func maybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) {
+ // Check if the envv already contains HOME.
+ for _, env := range envv {
+ if strings.HasPrefix(env, "HOME=") {
+ // We have it. Return the original slice unmodified.
+ return envv, nil
+ }
+ }
+
+ // Read /etc/passwd for the user's HOME directory and set the HOME
+ // environment variable as required by POSIX if it is not overridden by
+ // the user.
+ homeDir, err := getExecUserHome(ctx, mns, uid)
+ if err != nil {
+ return nil, fmt.Errorf("error reading exec user: %v", err)
+ }
+ return append(envv, "HOME="+homeDir), nil
+}
+
// findHomeInPasswd parses a passwd file and returns the given user's home
// directory. This function does it's best to replicate the runc's behavior.
func findHomeInPasswd(uid uint32, passwd io.Reader, defaultHome string) (string, error) {
diff --git a/runsc/boot/user_test.go b/runsc/boot/user_test.go
index 906baf3e5..9aee2ad07 100644
--- a/runsc/boot/user_test.go
+++ b/runsc/boot/user_test.go
@@ -25,6 +25,7 @@ import (
specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/pkg/sentry/context/contexttest"
"gvisor.dev/gvisor/pkg/sentry/fs"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
)
func setupTempDir() (string, error) {
@@ -68,7 +69,7 @@ func setupPasswd(contents string, perms os.FileMode) func() (string, error) {
// TestGetExecUserHome tests the getExecUserHome function.
func TestGetExecUserHome(t *testing.T) {
tests := map[string]struct {
- uid uint32
+ uid auth.KUID
createRoot func() (string, error)
expected string
}{
diff --git a/runsc/cmd/exec.go b/runsc/cmd/exec.go
index e817eff77..d1e99243b 100644
--- a/runsc/cmd/exec.go
+++ b/runsc/cmd/exec.go
@@ -105,11 +105,11 @@ func (ex *Exec) SetFlags(f *flag.FlagSet) {
// Execute implements subcommands.Command.Execute. It starts a process in an
// already created container.
func (ex *Exec) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus {
- e, id, err := ex.parseArgs(f)
+ conf := args[0].(*boot.Config)
+ e, id, err := ex.parseArgs(f, conf.EnableRaw)
if err != nil {
Fatalf("parsing process spec: %v", err)
}
- conf := args[0].(*boot.Config)
waitStatus := args[1].(*syscall.WaitStatus)
c, err := container.Load(conf.RootDir, id)
@@ -117,6 +117,9 @@ func (ex *Exec) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
Fatalf("loading sandbox: %v", err)
}
+ log.Debugf("Exec arguments: %+v", e)
+ log.Debugf("Exec capablities: %+v", e.Capabilities)
+
// Replace empty settings with defaults from container.
if e.WorkingDirectory == "" {
e.WorkingDirectory = c.Spec.Process.Cwd
@@ -127,15 +130,13 @@ func (ex *Exec) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
Fatalf("getting environment variables: %v", err)
}
}
+
if e.Capabilities == nil {
- // enableRaw is set to true to prevent the filtering out of
- // CAP_NET_RAW. This is the opposite of Create() because exec
- // requires the capability to be set explicitly, while 'docker
- // run' sets it by default.
- e.Capabilities, err = specutils.Capabilities(true /* enableRaw */, c.Spec.Process.Capabilities)
+ e.Capabilities, err = specutils.Capabilities(conf.EnableRaw, c.Spec.Process.Capabilities)
if err != nil {
Fatalf("creating capabilities: %v", err)
}
+ log.Infof("Using exec capabilities from container: %+v", e.Capabilities)
}
// containerd expects an actual process to represent the container being
@@ -282,14 +283,14 @@ func (ex *Exec) execChildAndWait(waitStatus *syscall.WaitStatus) subcommands.Exi
// parseArgs parses exec information from the command line or a JSON file
// depending on whether the --process flag was used. Returns an ExecArgs and
// the ID of the container to be used.
-func (ex *Exec) parseArgs(f *flag.FlagSet) (*control.ExecArgs, string, error) {
+func (ex *Exec) parseArgs(f *flag.FlagSet, enableRaw bool) (*control.ExecArgs, string, error) {
if ex.processPath == "" {
// Requires at least a container ID and command.
if f.NArg() < 2 {
f.Usage()
return nil, "", fmt.Errorf("both a container-id and command are required")
}
- e, err := ex.argsFromCLI(f.Args()[1:])
+ e, err := ex.argsFromCLI(f.Args()[1:], enableRaw)
return e, f.Arg(0), err
}
// Requires only the container ID.
@@ -297,11 +298,11 @@ func (ex *Exec) parseArgs(f *flag.FlagSet) (*control.ExecArgs, string, error) {
f.Usage()
return nil, "", fmt.Errorf("a container-id is required")
}
- e, err := ex.argsFromProcessFile()
+ e, err := ex.argsFromProcessFile(enableRaw)
return e, f.Arg(0), err
}
-func (ex *Exec) argsFromCLI(argv []string) (*control.ExecArgs, error) {
+func (ex *Exec) argsFromCLI(argv []string, enableRaw bool) (*control.ExecArgs, error) {
extraKGIDs := make([]auth.KGID, 0, len(ex.extraKGIDs))
for _, s := range ex.extraKGIDs {
kgid, err := strconv.Atoi(s)
@@ -314,7 +315,7 @@ func (ex *Exec) argsFromCLI(argv []string) (*control.ExecArgs, error) {
var caps *auth.TaskCapabilities
if len(ex.caps) > 0 {
var err error
- caps, err = capabilities(ex.caps)
+ caps, err = capabilities(ex.caps, enableRaw)
if err != nil {
return nil, fmt.Errorf("capabilities error: %v", err)
}
@@ -332,7 +333,7 @@ func (ex *Exec) argsFromCLI(argv []string) (*control.ExecArgs, error) {
}, nil
}
-func (ex *Exec) argsFromProcessFile() (*control.ExecArgs, error) {
+func (ex *Exec) argsFromProcessFile(enableRaw bool) (*control.ExecArgs, error) {
f, err := os.Open(ex.processPath)
if err != nil {
return nil, fmt.Errorf("error opening process file: %s, %v", ex.processPath, err)
@@ -342,21 +343,21 @@ func (ex *Exec) argsFromProcessFile() (*control.ExecArgs, error) {
if err := json.NewDecoder(f).Decode(&p); err != nil {
return nil, fmt.Errorf("error parsing process file: %s, %v", ex.processPath, err)
}
- return argsFromProcess(&p)
+ return argsFromProcess(&p, enableRaw)
}
// argsFromProcess performs all the non-IO conversion from the Process struct
// to ExecArgs.
-func argsFromProcess(p *specs.Process) (*control.ExecArgs, error) {
+func argsFromProcess(p *specs.Process, enableRaw bool) (*control.ExecArgs, error) {
// Create capabilities.
var caps *auth.TaskCapabilities
if p.Capabilities != nil {
var err error
- // enableRaw is set to true to prevent the filtering out of
- // CAP_NET_RAW. This is the opposite of Create() because exec
- // requires the capability to be set explicitly, while 'docker
- // run' sets it by default.
- caps, err = specutils.Capabilities(true /* enableRaw */, p.Capabilities)
+ // Starting from Docker 19, capabilities are explicitly set for exec (instead
+ // of nil like before). So we can't distinguish 'exec' from
+ // 'exec --privileged', as both specify CAP_NET_RAW. Therefore, filter
+ // CAP_NET_RAW in the same way as container start.
+ caps, err = specutils.Capabilities(enableRaw, p.Capabilities)
if err != nil {
return nil, fmt.Errorf("error creating capabilities: %v", err)
}
@@ -409,7 +410,7 @@ func resolveEnvs(envs ...[]string) ([]string, error) {
// capabilities takes a list of capabilities as strings and returns an
// auth.TaskCapabilities struct with those capabilities in every capability set.
// This mimics runc's behavior.
-func capabilities(cs []string) (*auth.TaskCapabilities, error) {
+func capabilities(cs []string, enableRaw bool) (*auth.TaskCapabilities, error) {
var specCaps specs.LinuxCapabilities
for _, cap := range cs {
specCaps.Ambient = append(specCaps.Ambient, cap)
@@ -418,11 +419,11 @@ func capabilities(cs []string) (*auth.TaskCapabilities, error) {
specCaps.Inheritable = append(specCaps.Inheritable, cap)
specCaps.Permitted = append(specCaps.Permitted, cap)
}
- // enableRaw is set to true to prevent the filtering out of
- // CAP_NET_RAW. This is the opposite of Create() because exec requires
- // the capability to be set explicitly, while 'docker run' sets it by
- // default.
- return specutils.Capabilities(true /* enableRaw */, &specCaps)
+ // Starting from Docker 19, capabilities are explicitly set for exec (instead
+ // of nil like before). So we can't distinguish 'exec' from
+ // 'exec --privileged', as both specify CAP_NET_RAW. Therefore, filter
+ // CAP_NET_RAW in the same way as container start.
+ return specutils.Capabilities(enableRaw, &specCaps)
}
// stringSlice allows a flag to be used multiple times, where each occurrence
diff --git a/runsc/cmd/exec_test.go b/runsc/cmd/exec_test.go
index eb38a431f..a1e980d08 100644
--- a/runsc/cmd/exec_test.go
+++ b/runsc/cmd/exec_test.go
@@ -91,7 +91,7 @@ func TestCLIArgs(t *testing.T) {
}
for _, tc := range testCases {
- e, err := tc.ex.argsFromCLI(tc.argv)
+ e, err := tc.ex.argsFromCLI(tc.argv, true)
if err != nil {
t.Errorf("argsFromCLI(%+v): got error: %+v", tc.ex, err)
} else if !cmp.Equal(*e, tc.expected, cmpopts.IgnoreUnexported(os.File{})) {
@@ -144,7 +144,7 @@ func TestJSONArgs(t *testing.T) {
}
for _, tc := range testCases {
- e, err := argsFromProcess(&tc.p)
+ e, err := argsFromProcess(&tc.p, true)
if err != nil {
t.Errorf("argsFromProcess(%+v): got error: %+v", tc.p, err)
} else if !cmp.Equal(*e, tc.expected, cmpopts.IgnoreUnexported(os.File{})) {
diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go
index 9faabf494..fbd579fb8 100644
--- a/runsc/cmd/gofer.go
+++ b/runsc/cmd/gofer.go
@@ -182,6 +182,7 @@ func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
cfg := fsgofer.Config{
ROMount: isReadonlyMount(m.Options),
PanicOnWrite: g.panicOnWrite,
+ HostUDS: conf.FSGoferHostUDS,
}
ap, err := fsgofer.NewAttachPoint(m.Destination, cfg)
if err != nil {
@@ -200,6 +201,10 @@ func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
Fatalf("too many FDs passed for mounts. mounts: %d, FDs: %d", mountIdx, len(g.ioFDs))
}
+ if conf.FSGoferHostUDS {
+ filter.InstallUDSFilters()
+ }
+
if err := filter.Install(); err != nil {
Fatalf("installing seccomp filters: %v", err)
}
diff --git a/runsc/container/BUILD b/runsc/container/BUILD
index bc1fa25e3..26d1cd5ab 100644
--- a/runsc/container/BUILD
+++ b/runsc/container/BUILD
@@ -47,6 +47,7 @@ go_test(
],
deps = [
"//pkg/abi/linux",
+ "//pkg/bits",
"//pkg/log",
"//pkg/sentry/control",
"//pkg/sentry/kernel",
diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go
index 2ac12e5b6..519f5ed9b 100644
--- a/runsc/container/container_test.go
+++ b/runsc/container/container_test.go
@@ -34,6 +34,7 @@ import (
"github.com/cenkalti/backoff"
specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/bits"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/sentry/control"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
@@ -2049,6 +2050,30 @@ func TestMountSymlink(t *testing.T) {
}
}
+// Check that --net-raw disables the CAP_NET_RAW capability.
+func TestNetRaw(t *testing.T) {
+ capNetRaw := strconv.FormatUint(bits.MaskOf64(int(linux.CAP_NET_RAW)), 10)
+ app, err := testutil.FindFile("runsc/container/test_app/test_app")
+ if err != nil {
+ t.Fatal("error finding test_app:", err)
+ }
+
+ for _, enableRaw := range []bool{true, false} {
+ conf := testutil.TestConfig()
+ conf.EnableRaw = enableRaw
+
+ test := "--enabled"
+ if !enableRaw {
+ test = "--disabled"
+ }
+
+ spec := testutil.NewSpecWithArgs(app, "capability", test, capNetRaw)
+ if err := run(spec, conf); err != nil {
+ t.Fatalf("Error running container: %v", err)
+ }
+ }
+}
+
// executeSync synchronously executes a new process.
func (cont *Container) executeSync(args *control.ExecArgs) (syscall.WaitStatus, error) {
pid, err := cont.Execute(args)
diff --git a/runsc/container/test_app/test_app.go b/runsc/container/test_app/test_app.go
index 7f735c254..913d781c6 100644
--- a/runsc/container/test_app/test_app.go
+++ b/runsc/container/test_app/test_app.go
@@ -19,10 +19,12 @@ package main
import (
"context"
"fmt"
+ "io/ioutil"
"log"
"net"
"os"
"os/exec"
+ "regexp"
"strconv"
sys "syscall"
"time"
@@ -35,6 +37,7 @@ import (
func main() {
subcommands.Register(subcommands.HelpCommand(), "")
subcommands.Register(subcommands.FlagsCommand(), "")
+ subcommands.Register(new(capability), "")
subcommands.Register(new(fdReceiver), "")
subcommands.Register(new(fdSender), "")
subcommands.Register(new(forkBomb), "")
@@ -287,3 +290,65 @@ func (s *syscall) Execute(ctx context.Context, f *flag.FlagSet, args ...interfac
}
return subcommands.ExitSuccess
}
+
+type capability struct {
+ enabled uint64
+ disabled uint64
+}
+
+// Name implements subcommands.Command.
+func (*capability) Name() string {
+ return "capability"
+}
+
+// Synopsis implements subcommands.Command.
+func (*capability) Synopsis() string {
+ return "checks if effective capabilities are set/unset"
+}
+
+// Usage implements subcommands.Command.
+func (*capability) Usage() string {
+ return "capability [--enabled=number] [--disabled=number]"
+}
+
+// SetFlags implements subcommands.Command.
+func (c *capability) SetFlags(f *flag.FlagSet) {
+ f.Uint64Var(&c.enabled, "enabled", 0, "")
+ f.Uint64Var(&c.disabled, "disabled", 0, "")
+}
+
+// Execute implements subcommands.Command.
+func (c *capability) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus {
+ if c.enabled == 0 && c.disabled == 0 {
+ fmt.Println("One of the flags must be set")
+ return subcommands.ExitUsageError
+ }
+
+ status, err := ioutil.ReadFile("/proc/self/status")
+ if err != nil {
+ fmt.Printf("Error reading %q: %v\n", "proc/self/status", err)
+ return subcommands.ExitFailure
+ }
+ re := regexp.MustCompile("CapEff:\t([0-9a-f]+)\n")
+ matches := re.FindStringSubmatch(string(status))
+ if matches == nil || len(matches) != 2 {
+ fmt.Printf("Effective capabilities not found in\n%s\n", status)
+ return subcommands.ExitFailure
+ }
+ caps, err := strconv.ParseUint(matches[1], 16, 64)
+ if err != nil {
+ fmt.Printf("failed to convert capabilities %q: %v\n", matches[1], err)
+ return subcommands.ExitFailure
+ }
+
+ if c.enabled != 0 && (caps&c.enabled) != c.enabled {
+ fmt.Printf("Missing capabilities, want: %#x: got: %#x\n", c.enabled, caps)
+ return subcommands.ExitFailure
+ }
+ if c.disabled != 0 && (caps&c.disabled) != 0 {
+ fmt.Printf("Extra capabilities found, dont_want: %#x: got: %#x\n", c.disabled, caps)
+ return subcommands.ExitFailure
+ }
+
+ return subcommands.ExitSuccess
+}
diff --git a/runsc/dockerutil/dockerutil.go b/runsc/dockerutil/dockerutil.go
index c073d8f75..57f6ae8de 100644
--- a/runsc/dockerutil/dockerutil.go
+++ b/runsc/dockerutil/dockerutil.go
@@ -282,7 +282,22 @@ func (d *Docker) Logs() (string, error) {
// Exec calls 'docker exec' with the arguments provided.
func (d *Docker) Exec(args ...string) (string, error) {
- a := []string{"exec", d.Name}
+ return d.ExecWithFlags(nil, args...)
+}
+
+// ExecWithFlags calls 'docker exec <flags> name <args>'.
+func (d *Docker) ExecWithFlags(flags []string, args ...string) (string, error) {
+ a := []string{"exec"}
+ a = append(a, flags...)
+ a = append(a, d.Name)
+ a = append(a, args...)
+ return do(a...)
+}
+
+// ExecAsUser calls 'docker exec' as the given user with the arguments
+// provided.
+func (d *Docker) ExecAsUser(user string, args ...string) (string, error) {
+ a := []string{"exec", "--user", user, d.Name}
a = append(a, args...)
return do(a...)
}
diff --git a/runsc/fsgofer/filter/config.go b/runsc/fsgofer/filter/config.go
index 2f3f2039a..c7922b54f 100644
--- a/runsc/fsgofer/filter/config.go
+++ b/runsc/fsgofer/filter/config.go
@@ -214,3 +214,16 @@ var allowedSyscalls = seccomp.SyscallRules{
syscall.SYS_UTIMENSAT: {},
syscall.SYS_WRITE: {},
}
+
+var udsSyscalls = seccomp.SyscallRules{
+ syscall.SYS_SOCKET: []seccomp.Rule{
+ {
+ seccomp.AllowValue(syscall.AF_UNIX),
+ },
+ },
+ syscall.SYS_CONNECT: []seccomp.Rule{
+ {
+ seccomp.AllowAny{},
+ },
+ },
+}
diff --git a/runsc/fsgofer/filter/filter.go b/runsc/fsgofer/filter/filter.go
index 65053415f..289886720 100644
--- a/runsc/fsgofer/filter/filter.go
+++ b/runsc/fsgofer/filter/filter.go
@@ -23,11 +23,16 @@ import (
// Install installs seccomp filters.
func Install() error {
- s := allowedSyscalls
-
// Set of additional filters used by -race and -msan. Returns empty
// when not enabled.
- s.Merge(instrumentationFilters())
+ allowedSyscalls.Merge(instrumentationFilters())
+
+ return seccomp.Install(allowedSyscalls)
+}
- return seccomp.Install(s)
+// InstallUDSFilters extends the allowed syscalls to include those necessary for
+// connecting to a host UDS.
+func InstallUDSFilters() {
+ // Add additional filters required for connecting to the host's sockets.
+ allowedSyscalls.Merge(udsSyscalls)
}
diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go
index 7c4d2b94e..29a82138e 100644
--- a/runsc/fsgofer/fsgofer.go
+++ b/runsc/fsgofer/fsgofer.go
@@ -21,6 +21,7 @@
package fsgofer
import (
+ "errors"
"fmt"
"io"
"math"
@@ -54,6 +55,7 @@ const (
regular fileType = iota
directory
symlink
+ socket
unknown
)
@@ -66,6 +68,8 @@ func (f fileType) String() string {
return "directory"
case symlink:
return "symlink"
+ case socket:
+ return "socket"
}
return "unknown"
}
@@ -82,6 +86,9 @@ type Config struct {
// PanicOnWrite panics on attempts to write to RO mounts.
PanicOnWrite bool
+
+ // HostUDS signals whether the gofer can mount a host's UDS.
+ HostUDS bool
}
type attachPoint struct {
@@ -124,24 +131,50 @@ func (a *attachPoint) Attach() (p9.File, error) {
if err != nil {
return nil, fmt.Errorf("stat file %q, err: %v", a.prefix, err)
}
- mode := syscall.O_RDWR
- if a.conf.ROMount || (stat.Mode&syscall.S_IFMT) == syscall.S_IFDIR {
- mode = syscall.O_RDONLY
- }
-
- // Open the root directory.
- f, err := fd.Open(a.prefix, openFlags|mode, 0)
- if err != nil {
- return nil, fmt.Errorf("unable to open file %q, err: %v", a.prefix, err)
- }
+ // Acquire the attach point lock.
a.attachedMu.Lock()
defer a.attachedMu.Unlock()
+
if a.attached {
- f.Close()
return nil, fmt.Errorf("attach point already attached, prefix: %s", a.prefix)
}
+ // Hold the file descriptor we are converting into a p9.File.
+ var f *fd.FD
+
+ // Apply the S_IFMT bitmask so we can detect file type appropriately.
+ switch fmtStat := stat.Mode & syscall.S_IFMT; fmtStat {
+ case syscall.S_IFSOCK:
+ // Check to see if the CLI option has been set to allow the UDS mount.
+ if !a.conf.HostUDS {
+ return nil, errors.New("host UDS support is disabled")
+ }
+
+ // Attempt to open a connection. Bubble up the failures.
+ f, err = fd.DialUnix(a.prefix)
+ if err != nil {
+ return nil, err
+ }
+
+ default:
+ // Default to Read/Write permissions.
+ mode := syscall.O_RDWR
+
+ // If the configuration is Read Only or the mount point is a directory,
+ // set the mode to Read Only.
+ if a.conf.ROMount || fmtStat == syscall.S_IFDIR {
+ mode = syscall.O_RDONLY
+ }
+
+ // Open the mount point & capture the FD.
+ f, err = fd.Open(a.prefix, openFlags|mode, 0)
+ if err != nil {
+ return nil, fmt.Errorf("unable to open file %q, err: %v", a.prefix, err)
+ }
+ }
+
+ // Return a localFile object to the caller with the UDS FD included.
rv, err := newLocalFile(a, f, a.prefix, stat)
if err != nil {
return nil, err
@@ -295,7 +328,7 @@ func openAnyFile(path string, fn func(mode int) (*fd.FD, error)) (*fd.FD, error)
return file, nil
}
-func getSupportedFileType(stat syscall.Stat_t) (fileType, error) {
+func getSupportedFileType(stat syscall.Stat_t, permitSocket bool) (fileType, error) {
var ft fileType
switch stat.Mode & syscall.S_IFMT {
case syscall.S_IFREG:
@@ -304,6 +337,11 @@ func getSupportedFileType(stat syscall.Stat_t) (fileType, error) {
ft = directory
case syscall.S_IFLNK:
ft = symlink
+ case syscall.S_IFSOCK:
+ if !permitSocket {
+ return unknown, syscall.EPERM
+ }
+ ft = socket
default:
return unknown, syscall.EPERM
}
@@ -311,7 +349,7 @@ func getSupportedFileType(stat syscall.Stat_t) (fileType, error) {
}
func newLocalFile(a *attachPoint, file *fd.FD, path string, stat syscall.Stat_t) (*localFile, error) {
- ft, err := getSupportedFileType(stat)
+ ft, err := getSupportedFileType(stat, a.conf.HostUDS)
if err != nil {
return nil, err
}
@@ -1026,7 +1064,11 @@ func (l *localFile) Flush() error {
// Connect implements p9.File.
func (l *localFile) Connect(p9.ConnectFlags) (*fd.FD, error) {
- return nil, syscall.ECONNREFUSED
+ // Check to see if the CLI option has been set to allow the UDS mount.
+ if !l.attachPoint.conf.HostUDS {
+ return nil, syscall.ECONNREFUSED
+ }
+ return fd.DialUnix(l.hostPath)
}
// Close implements p9.File.
diff --git a/runsc/fsgofer/fsgofer_test.go b/runsc/fsgofer/fsgofer_test.go
index cbbe71019..05af7e397 100644
--- a/runsc/fsgofer/fsgofer_test.go
+++ b/runsc/fsgofer/fsgofer_test.go
@@ -665,7 +665,7 @@ func TestAttachInvalidType(t *testing.T) {
}
f, err := a.Attach()
if f != nil || err == nil {
- t.Fatalf("Attach should have failed, got (%v, nil)", f)
+ t.Fatalf("Attach should have failed, got (%v, %v)", f, err)
}
})
}
diff --git a/runsc/main.go b/runsc/main.go
index ff74c0a3d..7dce9dc00 100644
--- a/runsc/main.go
+++ b/runsc/main.go
@@ -68,6 +68,7 @@ var (
network = flag.String("network", "sandbox", "specifies which network to use: sandbox (default), host, none. Using network inside the sandbox is more secure because it's isolated from the host network.")
gso = flag.Bool("gso", true, "enable generic segmenation offload")
fileAccess = flag.String("file-access", "exclusive", "specifies which filesystem to use for the root mount: exclusive (default), shared. Volume mounts are always shared.")
+ fsGoferHostUDS = flag.Bool("fsgofer-host-uds", false, "Allow the gofer to mount Unix Domain Sockets.")
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.")
@@ -195,6 +196,7 @@ func main() {
DebugLog: *debugLog,
DebugLogFormat: *debugLogFormat,
FileAccess: fsAccess,
+ FSGoferHostUDS: *fsGoferHostUDS,
Overlay: *overlay,
Network: netType,
GSO: *gso,
@@ -239,7 +241,7 @@ func main() {
// want with them. Since Docker and Containerd both eat boot's stderr, we
// dup our stderr to the provided log FD so that panics will appear in the
// logs, rather than just disappear.
- if err := syscall.Dup2(int(f.Fd()), int(os.Stderr.Fd())); err != nil {
+ if err := syscall.Dup3(int(f.Fd()), int(os.Stderr.Fd()), 0); err != nil {
cmd.Fatalf("error dup'ing fd %d to stderr: %v", f.Fd(), err)
}
diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go
index 4c6c83fbd..ee9327fc8 100644
--- a/runsc/sandbox/sandbox.go
+++ b/runsc/sandbox/sandbox.go
@@ -352,7 +352,7 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF
}
if conf.DebugLog != "" {
test := ""
- if len(conf.TestOnlyTestNameEnv) == 0 {
+ if len(conf.TestOnlyTestNameEnv) != 0 {
// Fetch test name if one is provided and the test only flag was set.
if t, ok := specutils.EnvVar(args.Spec.Process.Env, conf.TestOnlyTestNameEnv); ok {
test = t
diff --git a/runsc/specutils/BUILD b/runsc/specutils/BUILD
index fbfb8e2f8..fa58313a0 100644
--- a/runsc/specutils/BUILD
+++ b/runsc/specutils/BUILD
@@ -13,6 +13,7 @@ go_library(
visibility = ["//:sandbox"],
deps = [
"//pkg/abi/linux",
+ "//pkg/bits",
"//pkg/log",
"//pkg/sentry/kernel/auth",
"@com_github_cenkalti_backoff//:go_default_library",
diff --git a/runsc/specutils/namespace.go b/runsc/specutils/namespace.go
index d441419cb..c7dd3051c 100644
--- a/runsc/specutils/namespace.go
+++ b/runsc/specutils/namespace.go
@@ -33,19 +33,19 @@ import (
func nsCloneFlag(nst specs.LinuxNamespaceType) uintptr {
switch nst {
case specs.IPCNamespace:
- return syscall.CLONE_NEWIPC
+ return unix.CLONE_NEWIPC
case specs.MountNamespace:
- return syscall.CLONE_NEWNS
+ return unix.CLONE_NEWNS
case specs.NetworkNamespace:
- return syscall.CLONE_NEWNET
+ return unix.CLONE_NEWNET
case specs.PIDNamespace:
- return syscall.CLONE_NEWPID
+ return unix.CLONE_NEWPID
case specs.UTSNamespace:
- return syscall.CLONE_NEWUTS
+ return unix.CLONE_NEWUTS
case specs.UserNamespace:
- return syscall.CLONE_NEWUSER
+ return unix.CLONE_NEWUSER
case specs.CgroupNamespace:
- panic("cgroup namespace has no associated clone flag")
+ return unix.CLONE_NEWCGROUP
default:
panic(fmt.Sprintf("unknown namespace %v", nst))
}
diff --git a/runsc/specutils/specutils.go b/runsc/specutils/specutils.go
index cb9e58dfb..591abe458 100644
--- a/runsc/specutils/specutils.go
+++ b/runsc/specutils/specutils.go
@@ -31,6 +31,7 @@ import (
"github.com/cenkalti/backoff"
specs "github.com/opencontainers/runtime-spec/specs-go"
"gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/bits"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
)
@@ -241,6 +242,15 @@ func AllCapabilities() *specs.LinuxCapabilities {
}
}
+// AllCapabilitiesUint64 returns a bitmask containing all capabilities set.
+func AllCapabilitiesUint64() uint64 {
+ var rv uint64
+ for _, cap := range capFromName {
+ rv |= bits.MaskOf64(int(cap))
+ }
+ return rv
+}
+
var capFromName = map[string]linux.Capability{
"CAP_CHOWN": linux.CAP_CHOWN,
"CAP_DAC_OVERRIDE": linux.CAP_DAC_OVERRIDE,
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-26 09:40:51 +0000