summaryrefslogtreecommitdiffhomepage
path: root/runsc
diff options
context:
space:
mode:
Diffstat (limited to 'runsc')
-rw-r--r--runsc/BUILD2
-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.go1
-rw-r--r--runsc/cmd/gofer.go5
-rw-r--r--runsc/dockerutil/dockerutil.go8
-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
15 files changed, 157 insertions, 49 deletions
diff --git a/runsc/BUILD b/runsc/BUILD
index 5e7dacb87..a3a0d6730 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",
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..bf1225e1c 100644
--- a/runsc/cmd/exec.go
+++ b/runsc/cmd/exec.go
@@ -127,6 +127,7 @@ 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
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/dockerutil/dockerutil.go b/runsc/dockerutil/dockerutil.go
index c073d8f75..e37ec0ffd 100644
--- a/runsc/dockerutil/dockerutil.go
+++ b/runsc/dockerutil/dockerutil.go
@@ -287,6 +287,14 @@ func (d *Docker) Exec(args ...string) (string, error) {
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...)
+}
+
// ExecWithTerminal calls 'docker exec -it' with the arguments provided and
// attaches a pty to stdio.
func (d *Docker) ExecWithTerminal(args ...string) (*exec.Cmd, *os.File, error) {
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
generated by cgit v1.2.3 (git 2.39.1) at 2025-01-08 04:34:58 +0000