diff options
Diffstat (limited to 'runsc/boot')
-rw-r--r-- | runsc/boot/BUILD | 14 | ||||
-rw-r--r-- | runsc/boot/compat.go | 2 | ||||
-rw-r--r-- | runsc/boot/config.go | 9 | ||||
-rw-r--r-- | runsc/boot/controller.go | 24 | ||||
-rw-r--r-- | runsc/boot/fds.go | 38 | ||||
-rw-r--r-- | runsc/boot/filter/config.go | 13 | ||||
-rw-r--r-- | runsc/boot/fs.go | 24 | ||||
-rw-r--r-- | runsc/boot/loader.go | 216 | ||||
-rw-r--r-- | runsc/boot/loader_amd64.go | 5 | ||||
-rw-r--r-- | runsc/boot/loader_arm64.go | 5 | ||||
-rw-r--r-- | runsc/boot/loader_test.go | 49 | ||||
-rw-r--r-- | runsc/boot/network.go | 27 | ||||
-rw-r--r-- | runsc/boot/pprof/BUILD | 11 | ||||
-rw-r--r-- | runsc/boot/pprof/pprof.go (renamed from runsc/boot/pprof.go) | 6 | ||||
-rw-r--r-- | runsc/boot/user.go | 64 | ||||
-rw-r--r-- | runsc/boot/vfs.go | 310 |
16 files changed, 725 insertions, 92 deletions
diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index ae4dd102a..5451f1eba 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -19,9 +19,9 @@ go_library( "loader_amd64.go", "loader_arm64.go", "network.go", - "pprof.go", "strace.go", "user.go", + "vfs.go", ], visibility = [ "//runsc:__subpackages__", @@ -34,6 +34,7 @@ go_library( "//pkg/control/server", "//pkg/cpuid", "//pkg/eventchannel", + "//pkg/fspath", "//pkg/log", "//pkg/memutil", "//pkg/rand", @@ -41,6 +42,7 @@ go_library( "//pkg/sentry/arch", "//pkg/sentry/arch:registers_go_proto", "//pkg/sentry/control", + "//pkg/sentry/devices/memdev", "//pkg/sentry/fs", "//pkg/sentry/fs/dev", "//pkg/sentry/fs/gofer", @@ -50,6 +52,12 @@ go_library( "//pkg/sentry/fs/sys", "//pkg/sentry/fs/tmpfs", "//pkg/sentry/fs/tty", + "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/fsimpl/gofer", + "//pkg/sentry/fsimpl/host", + "//pkg/sentry/fsimpl/proc", + "//pkg/sentry/fsimpl/sys", + "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel:uncaught_signal_go_proto", @@ -72,6 +80,7 @@ go_library( "//pkg/sentry/time", "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", "//pkg/sentry/usage", + "//pkg/sentry/vfs", "//pkg/sentry/watchdog", "//pkg/sync", "//pkg/syserror", @@ -91,6 +100,7 @@ go_library( "//pkg/usermem", "//runsc/boot/filter", "//runsc/boot/platforms", + "//runsc/boot/pprof", "//runsc/specutils", "@com_github_golang_protobuf//proto:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", @@ -114,10 +124,12 @@ go_test( "//pkg/p9", "//pkg/sentry/contexttest", "//pkg/sentry/fs", + "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sync", "//pkg/unet", "//runsc/fsgofer", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/runsc/boot/compat.go b/runsc/boot/compat.go index 8995d678e..b7cfb35bf 100644 --- a/runsc/boot/compat.go +++ b/runsc/boot/compat.go @@ -65,7 +65,7 @@ func newCompatEmitter(logFD int) (*compatEmitter, error) { if logFD > 0 { f := os.NewFile(uintptr(logFD), "user log file") - target := &log.MultiEmitter{c.sink, &log.K8sJSONEmitter{log.Writer{Next: f}}} + target := &log.MultiEmitter{c.sink, log.K8sJSONEmitter{&log.Writer{Next: f}}} c.sink = &log.BasicLogger{Level: log.Info, Emitter: target} } return c, nil diff --git a/runsc/boot/config.go b/runsc/boot/config.go index 35391030f..715a19112 100644 --- a/runsc/boot/config.go +++ b/runsc/boot/config.go @@ -158,6 +158,9 @@ type Config struct { // DebugLog is the path to log debug information to, if not empty. DebugLog string + // PanicLog is the path to log GO's runtime messages, if not empty. + PanicLog string + // DebugLogFormat is the log format for debug. DebugLogFormat string @@ -269,6 +272,7 @@ func (c *Config) ToFlags() []string { "--log=" + c.LogFilename, "--log-format=" + c.LogFormat, "--debug-log=" + c.DebugLog, + "--panic-log=" + c.PanicLog, "--debug-log-format=" + c.DebugLogFormat, "--file-access=" + c.FileAccess.String(), "--overlay=" + strconv.FormatBool(c.Overlay), @@ -301,5 +305,10 @@ func (c *Config) ToFlags() []string { if len(c.TestOnlyTestNameEnv) != 0 { f = append(f, "--TESTONLY-test-name-env="+c.TestOnlyTestNameEnv) } + + if c.VFS2 { + f = append(f, "--vfs2=true") + } + return f } diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go index 9c9e94864..8125d5061 100644 --- a/runsc/boot/controller.go +++ b/runsc/boot/controller.go @@ -32,6 +32,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/watchdog" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/urpc" + "gvisor.dev/gvisor/runsc/boot/pprof" "gvisor.dev/gvisor/runsc/specutils" ) @@ -100,11 +101,14 @@ const ( // Profiling related commands (see pprof.go for more details). const ( - StartCPUProfile = "Profile.StartCPUProfile" - StopCPUProfile = "Profile.StopCPUProfile" - HeapProfile = "Profile.HeapProfile" - StartTrace = "Profile.StartTrace" - StopTrace = "Profile.StopTrace" + StartCPUProfile = "Profile.StartCPUProfile" + StopCPUProfile = "Profile.StopCPUProfile" + HeapProfile = "Profile.HeapProfile" + GoroutineProfile = "Profile.GoroutineProfile" + BlockProfile = "Profile.BlockProfile" + MutexProfile = "Profile.MutexProfile" + StartTrace = "Profile.StartTrace" + StopTrace = "Profile.StopTrace" ) // Logging related commands (see logging.go for more details). @@ -142,7 +146,7 @@ func newController(fd int, l *Loader) (*controller, error) { } srv.Register(manager) - if eps, ok := l.k.NetworkStack().(*netstack.Stack); ok { + if eps, ok := l.k.RootNetworkNamespace().Stack().(*netstack.Stack); ok { net := &Network{ Stack: eps.Stack, } @@ -341,7 +345,7 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error { return fmt.Errorf("creating memory file: %v", err) } k.SetMemoryFile(mf) - networkStack := cm.l.k.NetworkStack() + networkStack := cm.l.k.RootNetworkNamespace().Stack() cm.l.k = k // Set up the restore environment. @@ -365,9 +369,9 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error { } if cm.l.conf.ProfileEnable { - // initializePProf opens /proc/self/maps, so has to be - // called before installing seccomp filters. - initializePProf() + // pprof.Initialize opens /proc/self/maps, so has to be called before + // installing seccomp filters. + pprof.Initialize() } // Seccomp filters have to be applied before parsing the state file. diff --git a/runsc/boot/fds.go b/runsc/boot/fds.go index 417d2d5fb..7e49f6f9f 100644 --- a/runsc/boot/fds.go +++ b/runsc/boot/fds.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" + vfshost "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" "gvisor.dev/gvisor/pkg/sentry/kernel" ) @@ -31,10 +32,13 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F return nil, fmt.Errorf("stdioFDs should contain exactly 3 FDs (stdin, stdout, and stderr), but %d FDs received", len(stdioFDs)) } + if kernel.VFS2Enabled { + return createFDTableVFS2(ctx, console, stdioFDs) + } + k := kernel.KernelFromContext(ctx) fdTable := k.NewFDTable() defer fdTable.DecRef() - mounter := fs.FileOwnerFromContext(ctx) var ttyFile *fs.File for appFD, hostFD := range stdioFDs { @@ -44,7 +48,7 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F // Import the file as a host TTY file. if ttyFile == nil { var err error - appFile, err = host.ImportFile(ctx, hostFD, mounter, true /* isTTY */) + appFile, err = host.ImportFile(ctx, hostFD, true /* isTTY */) if err != nil { return nil, err } @@ -63,7 +67,7 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F } else { // Import the file as a regular host file. var err error - appFile, err = host.ImportFile(ctx, hostFD, mounter, false /* isTTY */) + appFile, err = host.ImportFile(ctx, hostFD, false /* isTTY */) if err != nil { return nil, err } @@ -79,3 +83,31 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F fdTable.IncRef() return fdTable, nil } + +func createFDTableVFS2(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, error) { + k := kernel.KernelFromContext(ctx) + fdTable := k.NewFDTable() + defer fdTable.DecRef() + + hostMount, err := vfshost.NewMount(k.VFS()) + if err != nil { + return nil, fmt.Errorf("creating host mount: %w", err) + } + + for appFD, hostFD := range stdioFDs { + // TODO(gvisor.dev/issue/1482): Add TTY support. + appFile, err := vfshost.ImportFD(hostMount, hostFD, false) + if err != nil { + return nil, err + } + + if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { + appFile.DecRef() + return nil, err + } + appFile.DecRef() + } + + fdTable.IncRef() + return fdTable, nil +} diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index c69f4c602..1828d116a 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -44,7 +44,7 @@ var allowedSyscalls = seccomp.SyscallRules{ { seccomp.AllowAny{}, seccomp.AllowAny{}, - seccomp.AllowValue(0), + seccomp.AllowValue(syscall.O_CLOEXEC), }, }, syscall.SYS_EPOLL_CREATE1: {}, @@ -229,7 +229,9 @@ var allowedSyscalls = seccomp.SyscallRules{ syscall.SYS_NANOSLEEP: {}, syscall.SYS_PPOLL: {}, syscall.SYS_PREAD64: {}, + syscall.SYS_PREADV: {}, syscall.SYS_PWRITE64: {}, + syscall.SYS_PWRITEV: {}, syscall.SYS_READ: {}, syscall.SYS_RECVMSG: []seccomp.Rule{ { @@ -282,12 +284,21 @@ var allowedSyscalls = seccomp.SyscallRules{ {seccomp.AllowAny{}, seccomp.AllowValue(syscall.SHUT_RDWR)}, }, syscall.SYS_SIGALTSTACK: {}, + unix.SYS_STATX: {}, syscall.SYS_SYNC_FILE_RANGE: {}, syscall.SYS_TGKILL: []seccomp.Rule{ { seccomp.AllowValue(uint64(os.Getpid())), }, }, + syscall.SYS_UTIMENSAT: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowValue(0), /* null pathname */ + seccomp.AllowAny{}, + seccomp.AllowValue(0), /* flags */ + }, + }, syscall.SYS_WRITE: {}, // The only user in rawfile.NonBlockingWrite3 always passes iovcnt with // values 2 or 3. Three iovec-s are passed, when the PACKET_VNET_HDR diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index 0f62842ea..98cce60af 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -278,6 +278,9 @@ func subtargets(root string, mnts []specs.Mount) []string { } func setupContainerFS(ctx context.Context, conf *Config, mntr *containerMounter, procArgs *kernel.CreateProcessArgs) error { + if conf.VFS2 { + return setupContainerVFS2(ctx, conf, mntr, procArgs) + } mns, err := mntr.setupFS(conf, procArgs) if err != nil { return err @@ -573,6 +576,9 @@ func newContainerMounter(spec *specs.Spec, goferFDs []int, k *kernel.Kernel, hin // should be mounted (e.g. a volume shared between containers). It must be // called for the root container only. func (c *containerMounter) processHints(conf *Config) error { + if conf.VFS2 { + return nil + } ctx := c.k.SupervisorContext() for _, hint := range c.hints.mounts { // TODO(b/142076984): Only support tmpfs for now. Bind mounts require a @@ -781,9 +787,6 @@ func (c *containerMounter) getMountNameAndOptions(conf *Config, m specs.Mount) ( useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly default: - // TODO(nlacasse): Support all the mount types and make this a fatal error. - // Most applications will "just work" without them, so this is a warning - // for now. log.Warningf("ignoring unknown filesystem type %q", m.Type) } return fsName, opts, useOverlay, nil @@ -824,7 +827,20 @@ func (c *containerMounter) mountSubmount(ctx context.Context, conf *Config, mns inode, err := filesystem.Mount(ctx, mountDevice(m), mf, strings.Join(opts, ","), nil) if err != nil { - return fmt.Errorf("creating mount with source %q: %v", m.Source, err) + err := fmt.Errorf("creating mount with source %q: %v", m.Source, err) + // Check to see if this is a common error due to a Linux bug. + // This error is generated here in order to cause it to be + // printed to the user using Docker via 'runsc create' etc. rather + // than simply printed to the logs for the 'runsc boot' command. + // + // We check the error message string rather than type because the + // actual error types (syscall.EIO, syscall.EPIPE) are lost by file system + // implementation (e.g. p9). + // TODO(gvisor.dev/issue/1765): Remove message when bug is resolved. + if strings.Contains(err.Error(), syscall.EIO.Error()) || strings.Contains(err.Error(), syscall.EPIPE.Error()) { + return fmt.Errorf("%v: %s", err, specutils.FaqErrorMsg("memlock", "you may be encountering a Linux kernel bug")) + } + return err } // If there are submounts, we need to overlay the mount on top of a ramfs diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 9f0d5d7af..cf1f47bc7 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -26,7 +26,6 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" @@ -49,6 +48,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/watchdog" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/arp" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" @@ -60,6 +60,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/runsc/boot/filter" _ "gvisor.dev/gvisor/runsc/boot/platforms" // register all platforms. + "gvisor.dev/gvisor/runsc/boot/pprof" "gvisor.dev/gvisor/runsc/specutils" // Include supported socket providers. @@ -71,6 +72,8 @@ import ( _ "gvisor.dev/gvisor/pkg/sentry/socket/unix" ) +var syscallTable *kernel.SyscallTable + // Loader keeps state needed to start the kernel and run the container.. type Loader struct { // k is the kernel. @@ -154,13 +157,17 @@ type Args struct { Spec *specs.Spec // Conf is the system configuration. Conf *Config - // ControllerFD is the FD to the URPC controller. + // ControllerFD is the FD to the URPC controller. The Loader takes ownership + // of this FD and may close it at any time. ControllerFD int - // Device is an optional argument that is passed to the platform. + // Device is an optional argument that is passed to the platform. The Loader + // takes ownership of this file and may close it at any time. Device *os.File - // GoferFDs is an array of FDs used to connect with the Gofer. + // GoferFDs is an array of FDs used to connect with the Gofer. The Loader + // takes ownership of these FDs and may close them at any time. GoferFDs []int - // StdioFDs is the stdio for the application. + // StdioFDs is the stdio for the application. The Loader takes ownership of + // these FDs and may close them at any time. StdioFDs []int // Console is set to true if using TTY. Console bool @@ -173,6 +180,9 @@ type Args struct { UserLogFD int } +// make sure stdioFDs are always the same on initial start and on restore +const startingStdioFD = 64 + // New initializes a new kernel loader configured by spec. // New also handles setting up a kernel for restoring a container. func New(args Args) (*Loader, error) { @@ -186,13 +196,14 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("setting up memory usage: %v", err) } - if args.Conf.VFS2 { - st, ok := kernel.LookupSyscallTable(abi.Linux, arch.Host) - if ok { - vfs2.Override(st.Table) - } + // Patch the syscall table. + kernel.VFS2Enabled = args.Conf.VFS2 + if kernel.VFS2Enabled { + vfs2.Override(syscallTable.Table) } + kernel.RegisterSyscallTable(syscallTable) + // Create kernel and platform. p, err := createPlatform(args.Conf, args.Device) if err != nil { @@ -230,11 +241,8 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("enabling strace: %v", err) } - // Create an empty network stack because the network namespace may be empty at - // this point. Netns is configured before Run() is called. Netstack is - // configured using a control uRPC message. Host network is configured inside - // Run(). - networkStack, err := newEmptyNetworkStack(args.Conf, k, k) + // Create root network namespace/stack. + netns, err := newRootNetworkNamespace(args.Conf, k, k) if err != nil { return nil, fmt.Errorf("creating network: %v", err) } @@ -277,7 +285,7 @@ func New(args Args) (*Loader, error) { FeatureSet: cpuid.HostFeatureSet(), Timekeeper: tk, RootUserNamespace: creds.UserNamespace, - NetworkStack: networkStack, + RootNetworkNamespace: netns, ApplicationCores: uint(args.NumCPU), Vdso: vdso, RootUTSNamespace: kernel.NewUTSNamespace(args.Spec.Hostname, args.Spec.Hostname, creds.UserNamespace), @@ -320,6 +328,24 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("creating pod mount hints: %v", err) } + // Make host FDs stable between invocations. Host FDs must map to the exact + // same number when the sandbox is restored. Otherwise the wrong FD will be + // used. + var stdioFDs []int + newfd := startingStdioFD + for _, fd := range args.StdioFDs { + err := syscall.Dup3(fd, newfd, syscall.O_CLOEXEC) + if err != nil { + return nil, fmt.Errorf("dup3 of stdioFDs failed: %v", err) + } + stdioFDs = append(stdioFDs, newfd) + err = syscall.Close(fd) + if err != nil { + return nil, fmt.Errorf("close original stdioFDs failed: %v", err) + } + newfd++ + } + eid := execID{cid: args.ID} l := &Loader{ k: k, @@ -328,7 +354,7 @@ func New(args Args) (*Loader, error) { watchdog: dog, spec: args.Spec, goferFDs: args.GoferFDs, - stdioFDs: args.StdioFDs, + stdioFDs: stdioFDs, rootProcArgs: procArgs, sandboxID: args.ID, processes: map[execID]*execProcess{eid: {}}, @@ -368,11 +394,16 @@ func newProcess(id string, spec *specs.Spec, creds *auth.Credentials, k *kernel. return kernel.CreateProcessArgs{}, fmt.Errorf("creating limits: %v", err) } + wd := spec.Process.Cwd + if wd == "" { + wd = "/" + } + // Create the process arguments. procArgs := kernel.CreateProcessArgs{ Argv: spec.Process.Args, Envv: spec.Process.Env, - WorkingDirectory: spec.Process.Cwd, // Defaults to '/' if empty. + WorkingDirectory: wd, Credentials: creds, Umask: 0022, Limits: ls, @@ -466,7 +497,7 @@ func (l *Loader) run() error { // Delay host network configuration to this point because network namespace // is configured after the loader is created and before Run() is called. log.Debugf("Configuring host network") - stack := l.k.NetworkStack().(*hostinet.Stack) + stack := l.k.RootNetworkNamespace().Stack().(*hostinet.Stack) if err := stack.Configure(); err != nil { return err } @@ -485,7 +516,7 @@ func (l *Loader) run() error { // l.restore is set by the container manager when a restore call is made. if !l.restore { if l.conf.ProfileEnable { - initializePProf() + pprof.Initialize() } // Finally done with all configuration. Setup filters before user code @@ -517,7 +548,15 @@ func (l *Loader) run() error { } // 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) + var envv []string + if kernel.VFS2Enabled { + envv, err = maybeAddExecUserHomeVFS2(ctx, l.rootProcArgs.MountNamespaceVFS2, + l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) + + } else { + envv, err = maybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, + l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) + } if err != nil { return err } @@ -570,6 +609,16 @@ func (l *Loader) run() error { } }) + // l.stdioFDs are derived from dup() in boot.New() and they are now dup()ed again + // either in createFDTable() during initial start or in descriptor.initAfterLoad() + // during restore, we can release l.stdioFDs now. + for _, fd := range l.stdioFDs { + err := syscall.Close(fd) + if err != nil { + return fmt.Errorf("close dup()ed stdioFDs: %v", err) + } + } + log.Infof("Process should have started...") l.watchdog.Start() return l.k.Start() @@ -795,16 +844,19 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) { return 0, fmt.Errorf("container %q not started", args.ContainerID) } + // TODO(gvisor.dev/issue/1623): Add VFS2 support + // Get the container MountNamespace from the Task. tg.Leader().WithMuLocked(func(t *kernel.Task) { - // task.MountNamespace() does not take a ref, so we must do so - // ourselves. + // task.MountNamespace() does not take a ref, so we must do so ourselves. args.MountNamespace = t.MountNamespace() args.MountNamespace.IncRef() }) - defer args.MountNamespace.DecRef() + if args.MountNamespace != nil { + defer args.MountNamespace.DecRef() + } - // Add the HOME enviroment varible if it is not already set. + // Add the HOME environment variable if it is not already set. root := args.MountNamespace.Root() defer root.DecRef() ctx := fs.WithRoot(l.k.SupervisorContext(), root) @@ -905,48 +957,92 @@ func (l *Loader) WaitExit() kernel.ExitStatus { return l.k.GlobalInit().ExitStatus() } -func newEmptyNetworkStack(conf *Config, clock tcpip.Clock, uniqueID stack.UniqueID) (inet.Stack, error) { +func newRootNetworkNamespace(conf *Config, clock tcpip.Clock, uniqueID stack.UniqueID) (*inet.Namespace, error) { + // Create an empty network stack because the network namespace may be empty at + // this point. Netns is configured before Run() is called. Netstack is + // configured using a control uRPC message. Host network is configured inside + // Run(). switch conf.Network { case NetworkHost: - return hostinet.NewStack(), nil + // No network namespacing support for hostinet yet, hence creator is nil. + return inet.NewRootNamespace(hostinet.NewStack(), nil), nil case NetworkNone, NetworkSandbox: - // NetworkNone sets up loopback using netstack. - netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()} - transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol(), icmp.NewProtocol4()} - s := netstack.Stack{stack.New(stack.Options{ - NetworkProtocols: netProtos, - TransportProtocols: transProtos, - Clock: clock, - Stats: netstack.Metrics, - HandleLocal: true, - // Enable raw sockets for users with sufficient - // privileges. - RawFactory: raw.EndpointFactory{}, - UniqueID: uniqueID, - })} - - // Enable SACK Recovery. - if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(true)); err != nil { - return nil, fmt.Errorf("failed to enable SACK: %v", err) + s, err := newEmptySandboxNetworkStack(clock, uniqueID) + if err != nil { + return nil, err + } + creator := &sandboxNetstackCreator{ + clock: clock, + uniqueID: uniqueID, } + return inet.NewRootNamespace(s, creator), nil + + default: + panic(fmt.Sprintf("invalid network configuration: %v", conf.Network)) + } - // Set default TTLs as required by socket/netstack. - s.Stack.SetNetworkProtocolOption(ipv4.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) - s.Stack.SetNetworkProtocolOption(ipv6.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) +} - // Enable Receive Buffer Auto-Tuning. - if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { - return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) - } +func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (inet.Stack, error) { + netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()} + transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol(), icmp.NewProtocol4()} + s := netstack.Stack{stack.New(stack.Options{ + NetworkProtocols: netProtos, + TransportProtocols: transProtos, + Clock: clock, + Stats: netstack.Metrics, + HandleLocal: true, + // Enable raw sockets for users with sufficient + // privileges. + RawFactory: raw.EndpointFactory{}, + UniqueID: uniqueID, + })} - s.FillDefaultIPTables() + // Enable SACK Recovery. + if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(true)); err != nil { + return nil, fmt.Errorf("failed to enable SACK: %v", err) + } - return &s, nil + // Set default TTLs as required by socket/netstack. + s.Stack.SetNetworkProtocolOption(ipv4.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) + s.Stack.SetNetworkProtocolOption(ipv6.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) - default: - panic(fmt.Sprintf("invalid network configuration: %v", conf.Network)) + // Enable Receive Buffer Auto-Tuning. + if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { + return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) } + + s.FillDefaultIPTables() + + return &s, nil +} + +// sandboxNetstackCreator implements kernel.NetworkStackCreator. +// +// +stateify savable +type sandboxNetstackCreator struct { + clock tcpip.Clock + uniqueID stack.UniqueID +} + +// CreateStack implements kernel.NetworkStackCreator.CreateStack. +func (f *sandboxNetstackCreator) CreateStack() (inet.Stack, error) { + s, err := newEmptySandboxNetworkStack(f.clock, f.uniqueID) + if err != nil { + return nil, err + } + + // Setup loopback. + n := &Network{Stack: s.(*netstack.Stack).Stack} + nicID := tcpip.NICID(f.uniqueID.UniqueID()) + link := DefaultLoopbackLink + linkEP := loopback.New() + if err := n.createNICWithAddrs(nicID, link.Name, linkEP, link.Addresses); err != nil { + return nil, err + } + + return s, nil } // signal sends a signal to one or more processes in a container. If PID is 0, @@ -994,7 +1090,7 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er execTG, _, err := l.threadGroupFromID(execID{cid: cid, pid: tgid}) if err == nil { // Send signal directly to the identified process. - return execTG.SendSignal(&arch.SignalInfo{Signo: signo}) + return l.k.SendExternalSignalThreadGroup(execTG, &arch.SignalInfo{Signo: signo}) } // The caller may be signaling a process not started directly via exec. @@ -1011,7 +1107,7 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er if tg.Leader().ContainerID() != cid { return fmt.Errorf("process %d is part of a different container: %q", tgid, tg.Leader().ContainerID()) } - return tg.SendSignal(&arch.SignalInfo{Signo: signo}) + return l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo}) } func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, signo int32) error { @@ -1029,7 +1125,7 @@ func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, s // No foreground process group has been set. Signal the // original thread group. log.Warningf("No foreground process group for container %q and PID %d. Sending signal directly to PID %d.", cid, tgid, tgid) - return tg.SendSignal(&arch.SignalInfo{Signo: signo}) + return l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo}) } // Send the signal to all processes in the process group. var lastErr error @@ -1037,7 +1133,7 @@ func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, s if tg.ProcessGroup() != pg { continue } - if err := tg.SendSignal(&arch.SignalInfo{Signo: signo}); err != nil { + if err := l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo}); err != nil { lastErr = err } } diff --git a/runsc/boot/loader_amd64.go b/runsc/boot/loader_amd64.go index b9669f2ac..78df86611 100644 --- a/runsc/boot/loader_amd64.go +++ b/runsc/boot/loader_amd64.go @@ -17,11 +17,10 @@ package boot import ( - "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" ) func init() { - // Register the global syscall table. - kernel.RegisterSyscallTable(linux.AMD64) + // Set the global syscall table. + syscallTable = linux.AMD64 } diff --git a/runsc/boot/loader_arm64.go b/runsc/boot/loader_arm64.go index cf64d28c8..250785010 100644 --- a/runsc/boot/loader_arm64.go +++ b/runsc/boot/loader_arm64.go @@ -17,11 +17,10 @@ package boot import ( - "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" ) func init() { - // Register the global syscall table. - kernel.RegisterSyscallTable(linux.ARM64) + // Set the global syscall table. + syscallTable = linux.ARM64 } diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index 44aa63196..e7c71734f 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -24,11 +24,13 @@ import ( "time" specs "github.com/opencontainers/runtime-spec/specs-go" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/control/server" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/runsc/fsgofer" @@ -65,6 +67,11 @@ func testSpec() *specs.Spec { } } +func resetSyscallTable() { + kernel.VFS2Enabled = false + kernel.FlushSyscallTablesTestOnly() +} + // startGofer starts a new gofer routine serving 'root' path. It returns the // sandbox side of the connection, and a function that when called will stop the // gofer. @@ -100,7 +107,7 @@ func startGofer(root string) (int, func(), error) { return sandboxEnd, cleanup, nil } -func createLoader() (*Loader, func(), error) { +func createLoader(vfsEnabled bool) (*Loader, func(), error) { fd, err := server.CreateSocket(ControlSocketAddr(fmt.Sprintf("%010d", rand.Int())[:10])) if err != nil { return nil, nil, err @@ -108,12 +115,23 @@ func createLoader() (*Loader, func(), error) { conf := testConfig() spec := testSpec() + conf.VFS2 = vfsEnabled + sandEnd, cleanup, err := startGofer(spec.Root.Path) if err != nil { return nil, nil, err } - stdio := []int{int(os.Stdin.Fd()), int(os.Stdout.Fd()), int(os.Stderr.Fd())} + // Loader takes ownership of stdio. + var stdio []int + for _, f := range []*os.File{os.Stdin, os.Stdout, os.Stderr} { + newFd, err := unix.Dup(int(f.Fd())) + if err != nil { + return nil, nil, err + } + stdio = append(stdio, newFd) + } + args := Args{ ID: "foo", Spec: spec, @@ -132,10 +150,22 @@ func createLoader() (*Loader, func(), error) { // TestRun runs a simple application in a sandbox and checks that it succeeds. func TestRun(t *testing.T) { - l, cleanup, err := createLoader() + defer resetSyscallTable() + doRun(t, false) +} + +// TestRunVFS2 runs TestRun in VFSv2. +func TestRunVFS2(t *testing.T) { + defer resetSyscallTable() + doRun(t, true) +} + +func doRun(t *testing.T, vfsEnabled bool) { + l, cleanup, err := createLoader(vfsEnabled) if err != nil { t.Fatalf("error creating loader: %v", err) } + defer l.Destroy() defer cleanup() @@ -169,7 +199,18 @@ func TestRun(t *testing.T) { // TestStartSignal tests that the controller Start message will cause // WaitForStartSignal to return. func TestStartSignal(t *testing.T) { - l, cleanup, err := createLoader() + defer resetSyscallTable() + doStartSignal(t, false) +} + +// TestStartSignalVFS2 does TestStartSignal with VFS2. +func TestStartSignalVFS2(t *testing.T) { + defer resetSyscallTable() + doStartSignal(t, true) +} + +func doStartSignal(t *testing.T, vfsEnabled bool) { + l, cleanup, err := createLoader(vfsEnabled) if err != nil { t.Fatalf("error creating loader: %v", err) } diff --git a/runsc/boot/network.go b/runsc/boot/network.go index 6a8765ec8..bee6ee336 100644 --- a/runsc/boot/network.go +++ b/runsc/boot/network.go @@ -17,6 +17,7 @@ package boot import ( "fmt" "net" + "strings" "syscall" "gvisor.dev/gvisor/pkg/log" @@ -31,6 +32,32 @@ import ( "gvisor.dev/gvisor/pkg/urpc" ) +var ( + // DefaultLoopbackLink contains IP addresses and routes of "127.0.0.1/8" and + // "::1/8" on "lo" interface. + DefaultLoopbackLink = LoopbackLink{ + Name: "lo", + Addresses: []net.IP{ + net.IP("\x7f\x00\x00\x01"), + net.IPv6loopback, + }, + Routes: []Route{ + { + Destination: net.IPNet{ + IP: net.IPv4(0x7f, 0, 0, 0), + Mask: net.IPv4Mask(0xff, 0, 0, 0), + }, + }, + { + Destination: net.IPNet{ + IP: net.IPv6loopback, + Mask: net.IPMask(strings.Repeat("\xff", net.IPv6len)), + }, + }, + }, + } +) + // Network exposes methods that can be used to configure a network stack. type Network struct { Stack *stack.Stack diff --git a/runsc/boot/pprof/BUILD b/runsc/boot/pprof/BUILD new file mode 100644 index 000000000..29cb42b2f --- /dev/null +++ b/runsc/boot/pprof/BUILD @@ -0,0 +1,11 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "pprof", + srcs = ["pprof.go"], + visibility = [ + "//runsc:__subpackages__", + ], +) diff --git a/runsc/boot/pprof.go b/runsc/boot/pprof/pprof.go index 463362f02..1ded20dee 100644 --- a/runsc/boot/pprof.go +++ b/runsc/boot/pprof/pprof.go @@ -12,7 +12,9 @@ // See the License for the specific language governing permissions and // limitations under the License. -package boot +// Package pprof provides a stub to initialize custom profilers. +package pprof -func initializePProf() { +// Initialize will be called at boot for initializing custom profilers. +func Initialize() { } diff --git a/runsc/boot/user.go b/runsc/boot/user.go index f0aa52135..332e4fce5 100644 --- a/runsc/boot/user.go +++ b/runsc/boot/user.go @@ -23,8 +23,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/usermem" ) @@ -84,6 +86,48 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.K File: f, } + return findHomeInPasswd(uint32(uid), r, defaultHome) +} + +type fileReaderVFS2 struct { + ctx context.Context + fd *vfs.FileDescription +} + +func (r *fileReaderVFS2) Read(buf []byte) (int, error) { + n, err := r.fd.Read(r.ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) + return int(n), err +} + +func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth.KUID) (string, error) { + const defaultHome = "/" + + root := mns.Root() + defer root.DecRef() + + creds := auth.CredentialsFromContext(ctx) + + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse("/etc/passwd"), + } + + opts := &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + } + + fd, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, target, opts) + if err != nil { + return defaultHome, nil + } + defer fd.DecRef() + + r := &fileReaderVFS2{ + ctx: ctx, + fd: fd, + } + homeDir, err := findHomeInPasswd(uint32(uid), r, defaultHome) if err != nil { return "", err @@ -111,6 +155,26 @@ func maybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth. if err != nil { return nil, fmt.Errorf("error reading exec user: %v", err) } + + return append(envv, "HOME="+homeDir), nil +} + +func maybeAddExecUserHomeVFS2(ctx context.Context, vmns *vfs.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 := getExecUserHomeVFS2(ctx, vmns, uid) + if err != nil { + return nil, fmt.Errorf("error reading exec user: %v", err) + } return append(envv, "HOME="+homeDir), nil } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go new file mode 100644 index 000000000..82083c57d --- /dev/null +++ b/runsc/boot/vfs.go @@ -0,0 +1,310 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package boot + +import ( + "fmt" + "path" + "strconv" + "strings" + + specs "github.com/opencontainers/runtime-spec/specs-go" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/devices/memdev" + "gvisor.dev/gvisor/pkg/sentry/fs" + devtmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + goferimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer" + procimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc" + sysimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" + tmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/syserror" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error { + + vfsObj.MustRegisterFilesystemType(rootFsName, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserList: true, + }) + + vfsObj.MustRegisterFilesystemType(bind, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserList: true, + }) + + vfsObj.MustRegisterFilesystemType(devpts, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + + vfsObj.MustRegisterFilesystemType(devtmpfs, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(proc, &procimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(sysfs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(tmpfs, &tmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(nonefs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + + // Setup files in devtmpfs. + if err := memdev.Register(vfsObj); err != nil { + return fmt.Errorf("registering memdev: %w", err) + } + a, err := devtmpfsimpl.NewAccessor(ctx, vfsObj, creds, devtmpfsimpl.Name) + if err != nil { + return fmt.Errorf("creating devtmpfs accessor: %w", err) + } + defer a.Release() + + if err := a.UserspaceInit(ctx); err != nil { + return fmt.Errorf("initializing userspace: %w", err) + } + if err := memdev.CreateDevtmpfsFiles(ctx, a); err != nil { + return fmt.Errorf("creating devtmpfs files: %w", err) + } + return nil +} + +func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounter, procArgs *kernel.CreateProcessArgs) error { + if err := mntr.k.VFS().Init(); err != nil { + return fmt.Errorf("failed to initialize VFS: %w", err) + } + mns, err := mntr.setupVFS2(ctx, conf, procArgs) + if err != nil { + return fmt.Errorf("failed to setupFS: %w", err) + } + procArgs.MountNamespaceVFS2 = mns + return setExecutablePathVFS2(ctx, procArgs) +} + +func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessArgs) error { + + exe := procArgs.Argv[0] + + // Absolute paths can be used directly. + if path.IsAbs(exe) { + procArgs.Filename = exe + return nil + } + + // Paths with '/' in them should be joined to the working directory, or + // to the root if working directory is not set. + if strings.IndexByte(exe, '/') > 0 { + + if !path.IsAbs(procArgs.WorkingDirectory) { + return fmt.Errorf("working directory %q must be absolute", procArgs.WorkingDirectory) + } + + procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe) + return nil + } + + // Paths with a '/' are relative to the CWD. + if strings.IndexByte(exe, '/') > 0 { + procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe) + return nil + } + + // Otherwise, We must lookup the name in the paths, starting from the + // root directory. + root := procArgs.MountNamespaceVFS2.Root() + defer root.DecRef() + + paths := fs.GetPath(procArgs.Envv) + creds := procArgs.Credentials + + for _, p := range paths { + + binPath := path.Join(p, exe) + + pop := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(binPath), + FollowFinalSymlink: true, + } + + opts := &vfs.OpenOptions{ + FileExec: true, + Flags: linux.O_RDONLY, + } + + dentry, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, pop, opts) + if err == syserror.ENOENT || err == syserror.EACCES { + // Didn't find it here. + continue + } + if err != nil { + return err + } + dentry.DecRef() + + procArgs.Filename = binPath + return nil + } + + return fmt.Errorf("executable %q not found in $PATH=%q", exe, strings.Join(paths, ":")) +} + +func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs *kernel.CreateProcessArgs) (*vfs.MountNamespace, error) { + log.Infof("Configuring container's file system with VFS2") + + // Create context with root credentials to mount the filesystem (the current + // user may not be privileged enough). + rootProcArgs := *procArgs + rootProcArgs.WorkingDirectory = "/" + rootProcArgs.Credentials = auth.NewRootCredentials(procArgs.Credentials.UserNamespace) + rootProcArgs.Umask = 0022 + rootProcArgs.MaxSymlinkTraversals = linux.MaxSymlinkTraversals + rootCtx := procArgs.NewContext(c.k) + + creds := procArgs.Credentials + if err := registerFilesystems(rootCtx, c.k.VFS(), creds); err != nil { + return nil, fmt.Errorf("register filesystems: %w", err) + } + + fd := c.fds.remove() + + opts := strings.Join(p9MountOptionsVFS2(fd, conf.FileAccess), ",") + + log.Infof("Mounting root over 9P, ioFD: %d", fd) + mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", rootFsName, &vfs.GetFilesystemOptions{Data: opts}) + if err != nil { + return nil, fmt.Errorf("setting up mountnamespace: %w", err) + } + + rootProcArgs.MountNamespaceVFS2 = mns + + // Mount submounts. + if err := c.mountSubmountsVFS2(rootCtx, conf, mns, creds); err != nil { + return nil, fmt.Errorf("mounting submounts vfs2: %w", err) + } + + return mns, nil +} + +func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials) error { + + for _, submount := range c.mounts { + log.Debugf("Mounting %q to %q, type: %s, options: %s", submount.Source, submount.Destination, submount.Type, submount.Options) + if err := c.mountSubmountVFS2(ctx, conf, mns, creds, &submount); err != nil { + return err + } + } + + // TODO(gvisor.dev/issue/1487): implement mountTmp from fs.go. + + return c.checkDispenser() +} + +// TODO(gvisor.dev/issue/1487): Implement submount options similar to the VFS1 version. +func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *specs.Mount) error { + root := mns.Root() + defer root.DecRef() + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(submount.Destination), + } + + _, options, useOverlay, err := c.getMountNameAndOptionsVFS2(conf, *submount) + if err != nil { + return fmt.Errorf("mountOptions failed: %w", err) + } + + opts := &vfs.MountOptions{ + GetFilesystemOptions: vfs.GetFilesystemOptions{ + Data: strings.Join(options, ","), + }, + InternalMount: true, + } + + // All writes go to upper, be paranoid and make lower readonly. + opts.ReadOnly = useOverlay + + if err := c.k.VFS().MountAt(ctx, creds, "", target, submount.Type, opts); err != nil { + return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts) + } + log.Infof("Mounted %q to %q type: %s, internal-options: %q", submount.Source, submount.Destination, submount.Type, opts) + return nil +} + +// getMountNameAndOptionsVFS2 retrieves the fsName, opts, and useOverlay values +// used for mounts. +func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m specs.Mount) (string, []string, bool, error) { + var ( + fsName string + opts []string + useOverlay bool + ) + + switch m.Type { + case devpts, devtmpfs, proc, sysfs: + fsName = m.Type + case nonefs: + fsName = sysfs + case tmpfs: + fsName = m.Type + + var err error + opts, err = parseAndFilterOptions(m.Options, tmpfsAllowedOptions...) + if err != nil { + return "", nil, false, err + } + + case bind: + fd := c.fds.remove() + fsName = "9p" + opts = p9MountOptionsVFS2(fd, c.getMountAccessType(m)) + // If configured, add overlay to all writable mounts. + useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly + + default: + log.Warningf("ignoring unknown filesystem type %q", m.Type) + } + return fsName, opts, useOverlay, nil +} + +// p9MountOptions creates a slice of options for a p9 mount. +// TODO(gvisor.dev/issue/1200): Remove this version in favor of the one in +// fs.go when privateunixsocket lands. +func p9MountOptionsVFS2(fd int, fa FileAccessType) []string { + opts := []string{ + "trans=fd", + "rfdno=" + strconv.Itoa(fd), + "wfdno=" + strconv.Itoa(fd), + } + if fa == FileAccessShared { + opts = append(opts, "cache=remote_revalidating") + } + return opts +} |