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.

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
 

runsc/cmd/gofer.go

@@ -59,6 +59,7 @@ type Gofer struct {
 	bundleDir      string
 	ioFDs          intFlags
 	applyCaps      bool
+	hostUDSAllowed bool
 	setUpRoot      bool
 
 	panicOnWrite bool
@@ -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")
@@ -182,6 +184,7 @@ func (g *Gofer) Execute(_ context.Context, f *flag.FlagSet, args ...interface{})
 			cfg := fsgofer.Config{
 				ROMount:        isReadonlyMount(m.Options),
 				PanicOnWrite:   g.panicOnWrite,
+				HostUDSAllowed: g.hostUDSAllowed,
 			}
 			ap, err := fsgofer.NewAttachPoint(m.Destination, cfg)
 			if err != nil {
@@ -200,9 +203,15 @@ 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 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)
 	return subcommands.ExitSuccess

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 {

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{},
+		},
+	},
+}

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)
+}

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)

runsc/main.go

@@ -67,6 +67,7 @@ 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.")
+	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.")
@@ -178,6 +179,7 @@ func main() {
 		DebugLog:              *debugLog,
 		DebugLogFormat:        *debugLogFormat,
 		FileAccess:            fsAccess,
+		FSGoferHostUDSAllowed: *fsGoferHostUDSAllowed,
 		Overlay:               *overlay,
 		Network:               netType,
 		GSO:                   *gso,