Place the host UDS mounting behind --fsgofer-host-uds-allowed.

This commit allows the use of the `--fsgofer-host-uds-allowed` flag to 
enable mounting sockets and add the appropriate seccomp filters.

7 files changed, 98 insertions, 54 deletions

diff --git a/runsc/boot/config.go b/runsc/boot/config.go
index 7ae0dd05d..954ad2c2a 100644
--- a/runsc/boot/config.go
+++ b/runsc/boot/config.go
@@ -138,6 +138,9 @@ type Config struct {
 	// Overlay is whether to wrap the root filesystem in an overlay.
 	Overlay bool
 
+	// fsGoferHostUDSAllowed enables the gofer to mount a host UDS
+	FSGoferHostUDSAllowed bool
+
 	// Network indicates what type of network to use.
 	Network NetworkType
 
diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go
index 9faabf494..8e63c80e0 100644
--- a/runsc/cmd/gofer.go
+++ b/runsc/cmd/gofer.go
@@ -56,10 +56,11 @@ var goferCaps = &specs.LinuxCapabilities{
 // Gofer implements subcommands.Command for the "gofer" command, which starts a
 // filesystem gofer.  This command should not be called directly.
 type Gofer struct {
-	bundleDir string
-	ioFDs     intFlags
-	applyCaps bool
-	setUpRoot bool
+	bundleDir      string
+	ioFDs          intFlags
+	applyCaps      bool
+	hostUDSAllowed bool
+	setUpRoot      bool
 
 	panicOnWrite bool
 	specFD       int
@@ -86,6 +87,7 @@ 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")
+	f.BoolVar(&g.hostUDSAllowed, "host-uds-allowed", false, "if true, allow the Gofer to mount a host UDS")
 	f.BoolVar(&g.panicOnWrite, "panic-on-write", false, "if true, panics on attempts to write to RO mounts. RW mounts are unnaffected")
 	f.BoolVar(&g.setUpRoot, "setup-root", true, "if true, set up an empty root for the process")
 	f.IntVar(&g.specFD, "spec-fd", -1, "required fd with the container spec")
@@ -180,8 +182,9 @@ func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
 	for _, m := range spec.Mounts {
 		if specutils.Is9PMount(m) {
 			cfg := fsgofer.Config{
-				ROMount:      isReadonlyMount(m.Options),
-				PanicOnWrite: g.panicOnWrite,
+				ROMount:        isReadonlyMount(m.Options),
+				PanicOnWrite:   g.panicOnWrite,
+				HostUDSAllowed: g.hostUDSAllowed,
 			}
 			ap, err := fsgofer.NewAttachPoint(m.Destination, cfg)
 			if err != nil {
@@ -200,8 +203,14 @@ 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 err := filter.Install(); err != nil {
-		Fatalf("installing seccomp filters: %v", err)
+	if g.hostUDSAllowed {
+		if err := filter.InstallUDS(); err != nil {
+			Fatalf("installing UDS seccomp filters: %v", err)
+		}
+	} else {
+		if err := filter.Install(); err != nil {
+			Fatalf("installing seccomp filters: %v", err)
+		}
 	}
 
 	runServers(ats, g.ioFDs)
diff --git a/runsc/container/container.go b/runsc/container/container.go
index bbb364214..ceadb38aa 100644
--- a/runsc/container/container.go
+++ b/runsc/container/container.go
@@ -941,6 +941,11 @@ func (c *Container) createGoferProcess(spec *specs.Spec, conf *boot.Config, bund
 		args = append(args, "--panic-on-write=true")
 	}
 
+	// Add support for mounting host UDS in the gofer
+	if conf.FSGoferHostUDSAllowed {
+		args = append(args, "--host-uds-allowed=true")
+	}
+
 	// Open the spec file to donate to the sandbox.
 	specFile, err := specutils.OpenSpec(bundleDir)
 	if err != nil {
diff --git a/runsc/fsgofer/filter/config.go b/runsc/fsgofer/filter/config.go
index 73407383d..8989cdb2f 100644
--- a/runsc/fsgofer/filter/config.go
+++ b/runsc/fsgofer/filter/config.go
@@ -26,16 +26,6 @@ import (
 // allowedSyscalls is the set of syscalls executed by the gofer.
 var allowedSyscalls = seccomp.SyscallRules{
 	syscall.SYS_ACCEPT: {},
-	syscall.SYS_SOCKET: []seccomp.Rule{
-		{
-			seccomp.AllowValue(syscall.AF_UNIX),
-		},
-	},
-	syscall.SYS_CONNECT: []seccomp.Rule{
-		{
-			seccomp.AllowAny{},
-		},
-	},
 	syscall.SYS_ARCH_PRCTL: []seccomp.Rule{
 		{seccomp.AllowValue(linux.ARCH_GET_FS)},
 		{seccomp.AllowValue(linux.ARCH_SET_FS)},
@@ -194,3 +184,16 @@ 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..12ef19d18 100644
--- a/runsc/fsgofer/filter/filter.go
+++ b/runsc/fsgofer/filter/filter.go
@@ -31,3 +31,15 @@ func Install() error {
 
 	return seccomp.Install(s)
 }
+
+// InstallUDS installs the standard Gofer seccomp filters along with filters
+// allowing the gofer to connect to a host UDS.
+func InstallUDS() error {
+	// Use the base syscall
+	s := allowedSyscalls
+
+	// Add additional filters required for connecting to the host's sockets.
+	s.Merge(udsSyscalls)
+
+	return seccomp.Install(s)
+}
diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go
index 89171c811..d9f3ba8d6 100644
--- a/runsc/fsgofer/fsgofer.go
+++ b/runsc/fsgofer/fsgofer.go
@@ -85,6 +85,9 @@ type Config struct {
 
 	// PanicOnWrite panics on attempts to write to RO mounts.
 	PanicOnWrite bool
+
+	// HostUDS prevents
+	HostUDSAllowed bool
 }
 
 type attachPoint struct {
@@ -128,12 +131,21 @@ func (a *attachPoint) Attach() (p9.File, error) {
 		return nil, fmt.Errorf("stat file %q, err: %v", a.prefix, err)
 	}
 
+	// Acquire the attach point lock
+	a.attachedMu.Lock()
+	defer a.attachedMu.Unlock()
+
 	// 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; {
 	case fmtStat == syscall.S_IFSOCK:
+		// Check to see if the CLI option has been set to allow the UDS mount
+		if !a.conf.HostUDSAllowed {
+			return nil, fmt.Errorf("host UDS support is disabled")
+		}
+
 		// Attempt to open a connection. Bubble up the failures.
 		f, err = fd.OpenUnix(a.prefix)
 		if err != nil {
@@ -144,7 +156,7 @@ func (a *attachPoint) Attach() (p9.File, error) {
 		// Default to Read/Write permissions.
 		mode := syscall.O_RDWR
 
-		// If the configuration is Read Only & the mount point is a directory,
+		// 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
@@ -157,9 +169,7 @@ func (a *attachPoint) Attach() (p9.File, error) {
 		}
 	}
 
-	// Close the connection if the UDS is already attached.
-	a.attachedMu.Lock()
-	defer a.attachedMu.Unlock()
+	// Close the connection if already attached.
 	if a.attached {
 		f.Close()
 		return nil, fmt.Errorf("attach point already attached, prefix: %s", a.prefix)
diff --git a/runsc/main.go b/runsc/main.go
index c61583441..5eba949f6 100644
--- a/runsc/main.go
+++ b/runsc/main.go
@@ -63,17 +63,18 @@ var (
 	straceLogSize  = flag.Uint("strace-log-size", 1024, "default size (in bytes) to log data argument blobs")
 
 	// Flags that control sandbox runtime behavior.
-	platformName       = flag.String("platform", "ptrace", "specifies which platform to use: ptrace (default), kvm")
-	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.")
-	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).")
-	netRaw             = flag.Bool("net-raw", false, "enable raw sockets. When false, raw sockets are disabled by removing CAP_NET_RAW from containers (`runsc exec` will still be able to utilize raw sockets). Raw sockets allow malicious containers to craft packets and potentially attack the network.")
-	numNetworkChannels = flag.Int("num-network-channels", 1, "number of underlying channels(FDs) to use for network link endpoints.")
-	rootless           = flag.Bool("rootless", false, "it allows the sandbox to be started with a user that is not root. Sandbox and Gofer processes may run with same privileges as current user.")
+	platformName          = flag.String("platform", "ptrace", "specifies which platform to use: ptrace (default), kvm")
+	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.")
+	fsGoferHostUDSAllowed = flag.Bool("fsgofer-host-uds-allowed", 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.")
+	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).")
+	netRaw                = flag.Bool("net-raw", false, "enable raw sockets. When false, raw sockets are disabled by removing CAP_NET_RAW from containers (`runsc exec` will still be able to utilize raw sockets). Raw sockets allow malicious containers to craft packets and potentially attack the network.")
+	numNetworkChannels    = flag.Int("num-network-channels", 1, "number of underlying channels(FDs) to use for network link endpoints.")
+	rootless              = flag.Bool("rootless", false, "it allows the sandbox to be started with a user that is not root. Sandbox and Gofer processes may run with same privileges as current user.")
 
 	// Test flags, not to be used outside tests, ever.
 	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.")
@@ -171,27 +172,28 @@ func main() {
 
 	// Create a new Config from the flags.
 	conf := &boot.Config{
-		RootDir:            *rootDir,
-		Debug:              *debug,
-		LogFilename:        *logFilename,
-		LogFormat:          *logFormat,
-		DebugLog:           *debugLog,
-		DebugLogFormat:     *debugLogFormat,
-		FileAccess:         fsAccess,
-		Overlay:            *overlay,
-		Network:            netType,
-		GSO:                *gso,
-		LogPackets:         *logPackets,
-		Platform:           platformType,
-		Strace:             *strace,
-		StraceLogSize:      *straceLogSize,
-		WatchdogAction:     wa,
-		PanicSignal:        *panicSignal,
-		ProfileEnable:      *profile,
-		EnableRaw:          *netRaw,
-		NumNetworkChannels: *numNetworkChannels,
-		Rootless:           *rootless,
-		AlsoLogToStderr:    *alsoLogToStderr,
+		RootDir:               *rootDir,
+		Debug:                 *debug,
+		LogFilename:           *logFilename,
+		LogFormat:             *logFormat,
+		DebugLog:              *debugLog,
+		DebugLogFormat:        *debugLogFormat,
+		FileAccess:            fsAccess,
+		FSGoferHostUDSAllowed: *fsGoferHostUDSAllowed,
+		Overlay:               *overlay,
+		Network:               netType,
+		GSO:                   *gso,
+		LogPackets:            *logPackets,
+		Platform:              platformType,
+		Strace:                *strace,
+		StraceLogSize:         *straceLogSize,
+		WatchdogAction:        wa,
+		PanicSignal:           *panicSignal,
+		ProfileEnable:         *profile,
+		EnableRaw:             *netRaw,
+		NumNetworkChannels:    *numNetworkChannels,
+		Rootless:              *rootless,
+		AlsoLogToStderr:       *alsoLogToStderr,
 
 		TestOnlyAllowRunAsCurrentUserWithoutChroot: *testOnlyAllowRunAsCurrentUserWithoutChroot,
 	}