From baf4d8aaca520255d52627779ada2be56be7f861 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Mon, 26 Aug 2019 14:04:24 -0700 Subject: Internal change. PiperOrigin-RevId: 265535438 --- test/syscalls/linux/BUILD | 23 ++++++++++++++++------- 1 file changed, 16 insertions(+), 7 deletions(-) (limited to 'test/syscalls/linux/BUILD') diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index ca4344139..88f3bfcb3 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1,3 +1,5 @@ +load("//test/syscalls:build_defs.bzl", "select_for_linux") + package( default_visibility = ["//:sandbox"], licenses = ["notice"], @@ -108,20 +110,27 @@ cc_library( cc_library( name = "socket_test_util", testonly = 1, - srcs = ["socket_test_util.cc"], + srcs = [ + "socket_test_util.cc", + ] + select_for_linux( + [ + "socket_test_util_impl.cc", + ], + ), hdrs = ["socket_test_util.h"], deps = [ + "@com_google_googletest//:gtest", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/strings:str_format", + "@com_google_absl//absl/time", "//test/util:file_descriptor", "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/memory", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/strings:str_format", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", - ], + ] + select_for_linux([ + ]), ) cc_library( -- cgit v1.2.3 From 863e11ac4d6a49787cd5e5f6fe1cd771d0ceb100 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Thu, 29 Aug 2019 14:29:43 -0700 Subject: Implement /proc/net/udp. PiperOrigin-RevId: 266229756 --- pkg/sentry/fs/proc/BUILD | 1 - pkg/sentry/fs/proc/net.go | 201 +++++++++++++++++-- pkg/sentry/socket/epsocket/epsocket.go | 25 ++- pkg/tcpip/transport/udp/endpoint.go | 61 +++--- pkg/tcpip/transport/udp/endpoint_state.go | 4 +- pkg/tcpip/transport/udp/forwarder.go | 2 +- test/syscalls/BUILD | 4 + test/syscalls/linux/BUILD | 15 ++ test/syscalls/linux/ip_socket_test_util.cc | 10 + test/syscalls/linux/ip_socket_test_util.h | 25 +++ test/syscalls/linux/proc_net_tcp.cc | 65 ++---- test/syscalls/linux/proc_net_udp.cc | 309 +++++++++++++++++++++++++++++ test/util/fs_util.cc | 9 + test/util/fs_util.h | 3 + 14 files changed, 633 insertions(+), 101 deletions(-) create mode 100644 test/syscalls/linux/proc_net_udp.cc (limited to 'test/syscalls/linux/BUILD') diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index 70ed854a8..c7599d1f6 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -31,7 +31,6 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/log", "//pkg/sentry/context", "//pkg/sentry/fs", diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 9adb23608..5e28982c5 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -17,10 +17,10 @@ package proc import ( "bytes" "fmt" + "io" "time" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -28,9 +28,11 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/usermem" ) // newNet creates a new proc net entry. @@ -57,9 +59,8 @@ func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSo "ptype": newStaticProcInode(ctx, msrc, []byte("Type Device Function")), "route": newStaticProcInode(ctx, msrc, []byte("Iface Destination Gateway Flags RefCnt Use Metric Mask MTU Window IRTT")), "tcp": seqfile.NewSeqFileInode(ctx, &netTCP{k: k}, msrc), - "udp": newStaticProcInode(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops")), - - "unix": seqfile.NewSeqFileInode(ctx, &netUnix{k: k}, msrc), + "udp": seqfile.NewSeqFileInode(ctx, &netUDP{k: k}, msrc), + "unix": seqfile.NewSeqFileInode(ctx, &netUnix{k: k}, msrc), } if s.SupportsIPv6() { @@ -216,7 +217,7 @@ func (n *netUnix) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]s for _, se := range n.k.ListSockets() { s := se.Sock.Get() if s == nil { - log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) continue } sfile := s.(*fs.File) @@ -297,6 +298,42 @@ func (n *netUnix) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]s return data, 0 } +func networkToHost16(n uint16) uint16 { + // n is in network byte order, so is big-endian. The most-significant byte + // should be stored in the lower address. + // + // We manually inline binary.BigEndian.Uint16() because Go does not support + // non-primitive consts, so binary.BigEndian is a (mutable) var, so calls to + // binary.BigEndian.Uint16() require a read of binary.BigEndian and an + // interface method call, defeating inlining. + buf := [2]byte{byte(n >> 8 & 0xff), byte(n & 0xff)} + return usermem.ByteOrder.Uint16(buf[:]) +} + +func writeInetAddr(w io.Writer, a linux.SockAddrInet) { + // linux.SockAddrInet.Port is stored in the network byte order and is + // printed like a number in host byte order. Note that all numbers in host + // byte order are printed with the most-significant byte first when + // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. + port := networkToHost16(a.Port) + + // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order + // (i.e. most-significant byte in index 0). Linux represents this as a + // __be32 which is a typedef for an unsigned int, and is printed with + // %X. This means that for a little-endian machine, Linux prints the + // least-significant byte of the address first. To emulate this, we first + // invert the byte order for the address using usermem.ByteOrder.Uint32, + // which makes it have the equivalent encoding to a __be32 on a little + // endian machine. Note that this operation is a no-op on a big endian + // machine. Then similar to Linux, we format it with %X, which will print + // the most-significant byte of the __be32 address first, which is now + // actually the least-significant byte of the original address in + // linux.SockAddrInet.Addr on little endian machines, due to the conversion. + addr := usermem.ByteOrder.Uint32(a.Addr[:]) + + fmt.Fprintf(w, "%08X:%04X ", addr, port) +} + // netTCP implements seqfile.SeqSource for /proc/net/tcp. // // +stateify savable @@ -311,6 +348,9 @@ func (*netTCP) NeedsUpdate(generation int64) bool { // ReadSeqFileData implements seqfile.SeqSource.ReadSeqFileData. func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]seqfile.SeqData, int64) { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. t := kernel.TaskFromContext(ctx) if h != nil { @@ -321,7 +361,7 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se for _, se := range n.k.ListSockets() { s := se.Sock.Get() if s == nil { - log.Debugf("Couldn't resolve weakref %+v in socket table, racing with destruction?", se.Sock) + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) continue } sfile := s.(*fs.File) @@ -343,27 +383,23 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se // Field: sl; entry number. fmt.Fprintf(&buf, "%4d: ", se.ID) - portBuf := make([]byte, 2) - // Field: local_adddress. var localAddr linux.SockAddrInet - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = *local.(*linux.SockAddrInet) + } } - binary.LittleEndian.PutUint16(portBuf, localAddr.Port) - fmt.Fprintf(&buf, "%08X:%04X ", - binary.LittleEndian.Uint32(localAddr.Addr[:]), - portBuf) + writeInetAddr(&buf, localAddr) // Field: rem_address. var remoteAddr linux.SockAddrInet - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = *remote.(*linux.SockAddrInet) + } } - binary.LittleEndian.PutUint16(portBuf, remoteAddr.Port) - fmt.Fprintf(&buf, "%08X:%04X ", - binary.LittleEndian.Uint32(remoteAddr.Addr[:]), - portBuf) + writeInetAddr(&buf, remoteAddr) // Field: state; socket state. fmt.Fprintf(&buf, "%02X ", sops.State()) @@ -386,7 +422,8 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) fmt.Fprintf(&buf, "%5d ", 0) } else { - fmt.Fprintf(&buf, "%5d ", uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(&buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) } // Field: timeout; number of unanswered 0-window probes. @@ -438,3 +475,125 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se } return data, 0 } + +// netUDP implements seqfile.SeqSource for /proc/net/udp. +// +// +stateify savable +type netUDP struct { + k *kernel.Kernel +} + +// NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. +func (*netUDP) NeedsUpdate(generation int64) bool { + return true +} + +// ReadSeqFileData implements seqfile.SeqSource.ReadSeqFileData. +func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]seqfile.SeqData, int64) { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. + t := kernel.TaskFromContext(ctx) + + if h != nil { + return nil, 0 + } + + var buf bytes.Buffer + for _, se := range n.k.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + continue + } + sfile := s.(*fs.File) + sops, ok := sfile.FileOperations.(socket.Socket) + if !ok { + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + } + if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { + s.DecRef() + // Not udp4 socket. + continue + } + + // For Linux's implementation, see net/ipv4/udp.c:udp4_format_sock(). + + // Field: sl; entry number. + fmt.Fprintf(&buf, "%5d: ", se.ID) + + // Field: local_adddress. + var localAddr linux.SockAddrInet + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = *local.(*linux.SockAddrInet) + } + } + writeInetAddr(&buf, localAddr) + + // Field: rem_address. + var remoteAddr linux.SockAddrInet + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = *remote.(*linux.SockAddrInet) + } + } + writeInetAddr(&buf, remoteAddr) + + // Field: state; socket state. + fmt.Fprintf(&buf, "%02X ", sops.State()) + + // Field: tx_queue, rx_queue; number of packets in the transmit and + // receive queue. Unimplemented. + fmt.Fprintf(&buf, "%08X:%08X ", 0, 0) + + // Field: tr, tm->when. Always 0 for UDP. + fmt.Fprintf(&buf, "%02X:%08X ", 0, 0) + + // Field: retrnsmt. Always 0 for UDP. + fmt.Fprintf(&buf, "%08X ", 0) + + // Field: uid. + uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) + if err != nil { + log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + fmt.Fprintf(&buf, "%5d ", 0) + } else { + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(&buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + } + + // Field: timeout. Always 0 for UDP. + fmt.Fprintf(&buf, "%8d ", 0) + + // Field: inode. + fmt.Fprintf(&buf, "%8d ", sfile.InodeID()) + + // Field: ref; reference count on the socket inode. Don't count the ref + // we obtain while deferencing the weakref to this socket. + fmt.Fprintf(&buf, "%d ", sfile.ReadRefs()-1) + + // Field: Socket struct address. Redacted due to the same reason as + // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. + fmt.Fprintf(&buf, "%#016p ", (*socket.Socket)(nil)) + + // Field: drops; number of dropped packets. Unimplemented. + fmt.Fprintf(&buf, "%d", 0) + + fmt.Fprintf(&buf, "\n") + + s.DecRef() + } + + data := []seqfile.SeqData{ + { + Buf: []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops \n"), + Handle: n, + }, + { + Buf: buf.Bytes(), + Handle: n, + }, + } + return data, 0 +} diff --git a/pkg/sentry/socket/epsocket/epsocket.go b/pkg/sentry/socket/epsocket/epsocket.go index 635042263..def29646e 100644 --- a/pkg/sentry/socket/epsocket/epsocket.go +++ b/pkg/sentry/socket/epsocket/epsocket.go @@ -27,12 +27,14 @@ package epsocket import ( "bytes" "math" + "reflect" "sync" "syscall" "time" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" @@ -52,6 +54,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" + "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/waiter" ) @@ -2421,7 +2424,8 @@ func (s *SocketOperations) State() uint32 { return 0 } - if !s.isPacketBased() { + switch { + case s.skType == linux.SOCK_STREAM && s.protocol == 0 || s.protocol == syscall.IPPROTO_TCP: // TCP socket. switch tcp.EndpointState(s.Endpoint.State()) { case tcp.StateEstablished: @@ -2450,9 +2454,26 @@ func (s *SocketOperations) State() uint32 { // Internal or unknown state. return 0 } + case s.skType == linux.SOCK_DGRAM && s.protocol == 0 || s.protocol == syscall.IPPROTO_UDP: + // UDP socket. + switch udp.EndpointState(s.Endpoint.State()) { + case udp.StateInitial, udp.StateBound, udp.StateClosed: + return linux.TCP_CLOSE + case udp.StateConnected: + return linux.TCP_ESTABLISHED + default: + return 0 + } + case s.skType == linux.SOCK_DGRAM && s.protocol == syscall.IPPROTO_ICMP || s.protocol == syscall.IPPROTO_ICMPV6: + // TODO(b/112063468): Export states for ICMP sockets. + case s.skType == linux.SOCK_RAW: + // TODO(b/112063468): Export states for raw sockets. + default: + // Unknown transport protocol, how did we make this socket? + log.Warningf("Unknown transport protocol for an existing socket: family=%v, type=%v, protocol=%v, internal type %v", s.family, s.skType, s.protocol, reflect.TypeOf(s.Endpoint).Elem()) + return 0 } - // TODO(b/112063468): Export states for UDP, ICMP, and raw sockets. return 0 } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index ac5905772..66455ef46 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -37,13 +37,17 @@ type udpPacket struct { views [8]buffer.View `state:"nosave"` } -type endpointState int +// EndpointState represents the state of a UDP endpoint. +type EndpointState uint32 +// Endpoint states. Note that are represented in a netstack-specific manner and +// may not be meaningful externally. Specifically, they need to be translated to +// Linux's representation for these states if presented to userspace. const ( - stateInitial endpointState = iota - stateBound - stateConnected - stateClosed + StateInitial EndpointState = iota + StateBound + StateConnected + StateClosed ) // endpoint represents a UDP endpoint. This struct serves as the interface @@ -74,7 +78,7 @@ type endpoint struct { mu sync.RWMutex `state:"nosave"` sndBufSize int id stack.TransportEndpointID - state endpointState + state EndpointState bindNICID tcpip.NICID regNICID tcpip.NICID route stack.Route `state:"manual"` @@ -140,7 +144,7 @@ func (e *endpoint) Close() { e.shutdownFlags = tcpip.ShutdownRead | tcpip.ShutdownWrite switch e.state { - case stateBound, stateConnected: + case StateBound, StateConnected: e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) } @@ -163,7 +167,7 @@ func (e *endpoint) Close() { e.route.Release() // Update the state. - e.state = stateClosed + e.state = StateClosed e.mu.Unlock() @@ -211,11 +215,11 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess // Returns true for retry if preparation should be retried. func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpip.Error) { switch e.state { - case stateInitial: - case stateConnected: + case StateInitial: + case StateConnected: return false, nil - case stateBound: + case StateBound: if to == nil { return false, tcpip.ErrDestinationRequired } @@ -232,7 +236,7 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // The state changed when we released the shared locked and re-acquired // it in exclusive mode. Try again. - if e.state != stateInitial { + if e.state != StateInitial { return true, nil } @@ -322,7 +326,7 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha defer e.mu.Unlock() // Recheck state after lock was re-acquired. - if e.state != stateConnected { + if e.state != StateConnected { return 0, nil, tcpip.ErrInvalidEndpointState } } @@ -400,7 +404,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { defer e.mu.Unlock() // We only allow this to be set when we're in the initial state. - if e.state != stateInitial { + if e.state != StateInitial { return tcpip.ErrInvalidEndpointState } @@ -726,7 +730,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - if e.state != stateConnected { + if e.state != StateConnected { return nil } id := stack.TransportEndpointID{} @@ -741,9 +745,9 @@ func (e *endpoint) Disconnect() *tcpip.Error { if err != nil { return err } - e.state = stateBound + e.state = StateBound } else { - e.state = stateInitial + e.state = StateInitial } e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) @@ -772,8 +776,8 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { nicid := addr.NIC var localPort uint16 switch e.state { - case stateInitial: - case stateBound, stateConnected: + case StateInitial: + case StateBound, StateConnected: localPort = e.id.LocalPort if e.bindNICID == 0 { break @@ -801,7 +805,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { RemoteAddress: r.RemoteAddress, } - if e.state == stateInitial { + if e.state == StateInitial { id.LocalAddress = r.LocalAddress } @@ -832,7 +836,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { e.regNICID = nicid e.effectiveNetProtos = netProtos - e.state = stateConnected + e.state = StateConnected e.rcvMu.Lock() e.rcvReady = true @@ -854,7 +858,7 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // A socket in the bound state can still receive multicast messages, // so we need to notify waiters on shutdown. - if e.state != stateBound && e.state != stateConnected { + if e.state != StateBound && e.state != StateConnected { return tcpip.ErrNotConnected } @@ -903,7 +907,7 @@ func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.Networ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { // Don't allow binding once endpoint is not in the initial state // anymore. - if e.state != stateInitial { + if e.state != StateInitial { return tcpip.ErrInvalidEndpointState } @@ -946,7 +950,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { e.effectiveNetProtos = netProtos // Mark endpoint as bound. - e.state = stateBound + e.state = StateBound e.rcvMu.Lock() e.rcvReady = true @@ -989,7 +993,7 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { e.mu.RLock() defer e.mu.RUnlock() - if e.state != stateConnected { + if e.state != StateConnected { return tcpip.FullAddress{}, tcpip.ErrNotConnected } @@ -1069,10 +1073,11 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { } -// State implements socket.Socket.State. +// State implements tcpip.Endpoint.State. func (e *endpoint) State() uint32 { - // TODO(b/112063468): Translate internal state to values returned by Linux. - return 0 + e.mu.Lock() + defer e.mu.Unlock() + return uint32(e.state) } func isBroadcastOrMulticast(a tcpip.Address) bool { diff --git a/pkg/tcpip/transport/udp/endpoint_state.go b/pkg/tcpip/transport/udp/endpoint_state.go index 5cbb56120..be46e6d4e 100644 --- a/pkg/tcpip/transport/udp/endpoint_state.go +++ b/pkg/tcpip/transport/udp/endpoint_state.go @@ -77,7 +77,7 @@ func (e *endpoint) Resume(s *stack.Stack) { } } - if e.state != stateBound && e.state != stateConnected { + if e.state != StateBound && e.state != StateConnected { return } @@ -92,7 +92,7 @@ func (e *endpoint) Resume(s *stack.Stack) { } var err *tcpip.Error - if e.state == stateConnected { + if e.state == StateConnected { e.route, err = e.stack.FindRoute(e.regNICID, e.id.LocalAddress, e.id.RemoteAddress, netProto, e.multicastLoop) if err != nil { panic(err) diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index a874fc9fd..a9edc2c8d 100644 --- a/pkg/tcpip/transport/udp/forwarder.go +++ b/pkg/tcpip/transport/udp/forwarder.go @@ -84,7 +84,7 @@ func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, ep.dstPort = r.id.RemotePort ep.regNICID = r.route.NICID() - ep.state = stateConnected + ep.state = StateConnected ep.rcvMu.Lock() ep.rcvReady = true diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index ccae4925f..6947ddc25 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -691,6 +691,10 @@ syscall_test( syscall_test(test = "//test/syscalls/linux:proc_net_unix_test") +syscall_test(test = "//test/syscalls/linux:proc_net_tcp_test") + +syscall_test(test = "//test/syscalls/linux:proc_net_udp_test") + go_binary( name = "syscall_test_runner", srcs = ["syscall_test_runner.go"], diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 88f3bfcb3..1ce38c929 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -3472,3 +3472,18 @@ cc_binary( "@com_google_googletest//:gtest", ], ) + +cc_binary( + name = "proc_net_udp_test", + testonly = 1, + srcs = ["proc_net_udp.cc"], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + "//test/util:file_descriptor", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_absl//absl/strings", + "@com_google_googletest//:gtest", + ], +) diff --git a/test/syscalls/linux/ip_socket_test_util.cc b/test/syscalls/linux/ip_socket_test_util.cc index c73262e72..410b42a47 100644 --- a/test/syscalls/linux/ip_socket_test_util.cc +++ b/test/syscalls/linux/ip_socket_test_util.cc @@ -23,6 +23,16 @@ namespace gvisor { namespace testing { +uint32_t IPFromInetSockaddr(const struct sockaddr* addr) { + auto* in_addr = reinterpret_cast(addr); + return in_addr->sin_addr.s_addr; +} + +uint16_t PortFromInetSockaddr(const struct sockaddr* addr) { + auto* in_addr = reinterpret_cast(addr); + return ntohs(in_addr->sin_port); +} + PosixErrorOr InterfaceIndex(std::string name) { // TODO(igudger): Consider using netlink. ifreq req = {}; diff --git a/test/syscalls/linux/ip_socket_test_util.h b/test/syscalls/linux/ip_socket_test_util.h index b498a053d..3d36b9620 100644 --- a/test/syscalls/linux/ip_socket_test_util.h +++ b/test/syscalls/linux/ip_socket_test_util.h @@ -26,6 +26,31 @@ namespace gvisor { namespace testing { +// Possible values of the "st" field in a /proc/net/{tcp,udp} entry. Source: +// Linux kernel, include/net/tcp_states.h. +enum { + TCP_ESTABLISHED = 1, + TCP_SYN_SENT, + TCP_SYN_RECV, + TCP_FIN_WAIT1, + TCP_FIN_WAIT2, + TCP_TIME_WAIT, + TCP_CLOSE, + TCP_CLOSE_WAIT, + TCP_LAST_ACK, + TCP_LISTEN, + TCP_CLOSING, + TCP_NEW_SYN_RECV, + + TCP_MAX_STATES +}; + +// Extracts the IP address from an inet sockaddr in network byte order. +uint32_t IPFromInetSockaddr(const struct sockaddr* addr); + +// Extracts the port from an inet sockaddr in host byte order. +uint16_t PortFromInetSockaddr(const struct sockaddr* addr); + // InterfaceIndex returns the index of the named interface. PosixErrorOr InterfaceIndex(std::string name); diff --git a/test/syscalls/linux/proc_net_tcp.cc b/test/syscalls/linux/proc_net_tcp.cc index 498f62d9c..f6d7ad0bb 100644 --- a/test/syscalls/linux/proc_net_tcp.cc +++ b/test/syscalls/linux/proc_net_tcp.cc @@ -38,25 +38,6 @@ constexpr char kProcNetTCPHeader[] = "retrnsmt uid timeout inode " " "; -// Possible values of the "st" field in a /proc/net/tcp entry. Source: Linux -// kernel, include/net/tcp_states.h. -enum { - TCP_ESTABLISHED = 1, - TCP_SYN_SENT, - TCP_SYN_RECV, - TCP_FIN_WAIT1, - TCP_FIN_WAIT2, - TCP_TIME_WAIT, - TCP_CLOSE, - TCP_CLOSE_WAIT, - TCP_LAST_ACK, - TCP_LISTEN, - TCP_CLOSING, - TCP_NEW_SYN_RECV, - - TCP_MAX_STATES -}; - // TCPEntry represents a single entry from /proc/net/tcp. struct TCPEntry { uint32_t local_addr; @@ -70,42 +51,35 @@ struct TCPEntry { uint64_t inode; }; -uint32_t IP(const struct sockaddr* addr) { - auto* in_addr = reinterpret_cast(addr); - return in_addr->sin_addr.s_addr; -} - -uint16_t Port(const struct sockaddr* addr) { - auto* in_addr = reinterpret_cast(addr); - return ntohs(in_addr->sin_port); -} - // Finds the first entry in 'entries' for which 'predicate' returns true. -// Returns true on match, and sets 'match' to point to the matching entry. -bool FindBy(std::vector entries, TCPEntry* match, +// Returns true on match, and sets 'match' to a copy of the matching entry. If +// 'match' is null, it's ignored. +bool FindBy(const std::vector& entries, TCPEntry* match, std::function predicate) { - for (int i = 0; i < entries.size(); ++i) { - if (predicate(entries[i])) { - *match = entries[i]; + for (const TCPEntry& entry : entries) { + if (predicate(entry)) { + if (match != nullptr) { + *match = entry; + } return true; } } return false; } -bool FindByLocalAddr(std::vector entries, TCPEntry* match, +bool FindByLocalAddr(const std::vector& entries, TCPEntry* match, const struct sockaddr* addr) { - uint32_t host = IP(addr); - uint16_t port = Port(addr); + uint32_t host = IPFromInetSockaddr(addr); + uint16_t port = PortFromInetSockaddr(addr); return FindBy(entries, match, [host, port](const TCPEntry& e) { return (e.local_addr == host && e.local_port == port); }); } -bool FindByRemoteAddr(std::vector entries, TCPEntry* match, +bool FindByRemoteAddr(const std::vector& entries, TCPEntry* match, const struct sockaddr* addr) { - uint32_t host = IP(addr); - uint16_t port = Port(addr); + uint32_t host = IPFromInetSockaddr(addr); + uint16_t port = PortFromInetSockaddr(addr); return FindBy(entries, match, [host, port](const TCPEntry& e) { return (e.remote_addr == host && e.remote_port == port); }); @@ -120,7 +94,7 @@ PosixErrorOr> ProcNetTCPEntries() { std::vector entries; std::vector lines = StrSplit(content, '\n'); std::cerr << "" << std::endl; - for (std::string line : lines) { + for (const std::string& line : lines) { std::cerr << line << std::endl; if (!found_header) { @@ -204,9 +178,8 @@ TEST(ProcNetTCP, BindAcceptConnect) { EXPECT_EQ(entries.size(), 2); } - TCPEntry e; - EXPECT_TRUE(FindByLocalAddr(entries, &e, sockets->first_addr())); - EXPECT_TRUE(FindByRemoteAddr(entries, &e, sockets->first_addr())); + EXPECT_TRUE(FindByLocalAddr(entries, nullptr, sockets->first_addr())); + EXPECT_TRUE(FindByRemoteAddr(entries, nullptr, sockets->first_addr())); } TEST(ProcNetTCP, InodeReasonable) { @@ -261,8 +234,8 @@ TEST(ProcNetTCP, State) { FileDescriptor accepted = ASSERT_NO_ERRNO_AND_VALUE(Accept(server->get(), nullptr, nullptr)); - const uint32_t accepted_local_host = IP(&addr); - const uint16_t accepted_local_port = Port(&addr); + const uint32_t accepted_local_host = IPFromInetSockaddr(&addr); + const uint16_t accepted_local_port = PortFromInetSockaddr(&addr); entries = ASSERT_NO_ERRNO_AND_VALUE(ProcNetTCPEntries()); TCPEntry accepted_entry; diff --git a/test/syscalls/linux/proc_net_udp.cc b/test/syscalls/linux/proc_net_udp.cc new file mode 100644 index 000000000..369df8e0e --- /dev/null +++ b/test/syscalls/linux/proc_net_udp.cc @@ -0,0 +1,309 @@ +// Copyright 2019 Google LLC +// +// 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. + +#include +#include +#include +#include + +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "absl/strings/numbers.h" +#include "absl/strings/str_join.h" +#include "absl/strings/str_split.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { +namespace { + +using absl::StrCat; +using absl::StrFormat; +using absl::StrSplit; + +constexpr char kProcNetUDPHeader[] = + " sl local_address rem_address st tx_queue rx_queue tr tm->when " + "retrnsmt uid timeout inode ref pointer drops "; + +// UDPEntry represents a single entry from /proc/net/udp. +struct UDPEntry { + uint32_t local_addr; + uint16_t local_port; + + uint32_t remote_addr; + uint16_t remote_port; + + uint64_t state; + uint64_t uid; + uint64_t inode; +}; + +std::string DescribeFirstInetSocket(const SocketPair& sockets) { + const struct sockaddr* addr = sockets.first_addr(); + return StrFormat("First test socket: fd:%d %8X:%4X", sockets.first_fd(), + IPFromInetSockaddr(addr), PortFromInetSockaddr(addr)); +} + +std::string DescribeSecondInetSocket(const SocketPair& sockets) { + const struct sockaddr* addr = sockets.second_addr(); + return StrFormat("Second test socket fd:%d %8X:%4X", sockets.second_fd(), + IPFromInetSockaddr(addr), PortFromInetSockaddr(addr)); +} + +// Finds the first entry in 'entries' for which 'predicate' returns true. +// Returns true on match, and set 'match' to a copy of the matching entry. If +// 'match' is null, it's ignored. +bool FindBy(const std::vector& entries, UDPEntry* match, + std::function predicate) { + for (const UDPEntry& entry : entries) { + if (predicate(entry)) { + if (match != nullptr) { + *match = entry; + } + return true; + } + } + return false; +} + +bool FindByLocalAddr(const std::vector& entries, UDPEntry* match, + const struct sockaddr* addr) { + uint32_t host = IPFromInetSockaddr(addr); + uint16_t port = PortFromInetSockaddr(addr); + return FindBy(entries, match, [host, port](const UDPEntry& e) { + return (e.local_addr == host && e.local_port == port); + }); +} + +bool FindByRemoteAddr(const std::vector& entries, UDPEntry* match, + const struct sockaddr* addr) { + uint32_t host = IPFromInetSockaddr(addr); + uint16_t port = PortFromInetSockaddr(addr); + return FindBy(entries, match, [host, port](const UDPEntry& e) { + return (e.remote_addr == host && e.remote_port == port); + }); +} + +PosixErrorOr InodeFromSocketFD(int fd) { + ASSIGN_OR_RETURN_ERRNO(struct stat s, Fstat(fd)); + if (!S_ISSOCK(s.st_mode)) { + return PosixError(EINVAL, StrFormat("FD %d is not a socket", fd)); + } + return s.st_ino; +} + +PosixErrorOr FindByFD(const std::vector& entries, + UDPEntry* match, int fd) { + ASSIGN_OR_RETURN_ERRNO(uint64_t inode, InodeFromSocketFD(fd)); + return FindBy(entries, match, + [inode](const UDPEntry& e) { return (e.inode == inode); }); +} + +// Returns a parsed representation of /proc/net/udp entries. +PosixErrorOr> ProcNetUDPEntries() { + std::string content; + RETURN_IF_ERRNO(GetContents("/proc/net/udp", &content)); + + bool found_header = false; + std::vector entries; + std::vector lines = StrSplit(content, '\n'); + std::cerr << "" << std::endl; + for (const std::string& line : lines) { + std::cerr << line << std::endl; + + if (!found_header) { + EXPECT_EQ(line, kProcNetUDPHeader); + found_header = true; + continue; + } + if (line.empty()) { + continue; + } + + // Parse a single entry from /proc/net/udp. + // + // Example entries: + // + // clang-format off + // + // sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops + // 3503: 0100007F:0035 00000000:0000 07 00000000:00000000 00:00000000 00000000 0 0 33317 2 0000000000000000 0 + // 3518: 00000000:0044 00000000:0000 07 00000000:00000000 00:00000000 00000000 0 0 40394 2 0000000000000000 0 + // ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ + // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 + // + // clang-format on + + UDPEntry entry; + std::vector fields = + StrSplit(line, absl::ByAnyChar(": "), absl::SkipEmpty()); + + ASSIGN_OR_RETURN_ERRNO(entry.local_addr, AtoiBase(fields[1], 16)); + ASSIGN_OR_RETURN_ERRNO(entry.local_port, AtoiBase(fields[2], 16)); + + ASSIGN_OR_RETURN_ERRNO(entry.remote_addr, AtoiBase(fields[3], 16)); + ASSIGN_OR_RETURN_ERRNO(entry.remote_port, AtoiBase(fields[4], 16)); + + ASSIGN_OR_RETURN_ERRNO(entry.state, AtoiBase(fields[5], 16)); + ASSIGN_OR_RETURN_ERRNO(entry.uid, Atoi(fields[11])); + ASSIGN_OR_RETURN_ERRNO(entry.inode, Atoi(fields[13])); + + // Linux shares internal data structures between TCP and UDP sockets. The + // proc entries for UDP sockets share some fields with TCP sockets, but + // these fields should always be zero as they're not meaningful for UDP + // sockets. + EXPECT_EQ(fields[8], "00") << StrFormat("sl:%s, tr", fields[0]); + EXPECT_EQ(fields[9], "00000000") << StrFormat("sl:%s, tm->when", fields[0]); + EXPECT_EQ(fields[10], "00000000") + << StrFormat("sl:%s, retrnsmt", fields[0]); + EXPECT_EQ(fields[12], "0") << StrFormat("sl:%s, timeout", fields[0]); + + entries.push_back(entry); + } + std::cerr << "" << std::endl; + + return entries; +} + +TEST(ProcNetUDP, Exists) { + const std::string content = + ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/net/udp")); + const std::string header_line = StrCat(kProcNetUDPHeader, "\n"); + EXPECT_THAT(content, ::testing::StartsWith(header_line)); +} + +TEST(ProcNetUDP, EntryUID) { + auto sockets = + ASSERT_NO_ERRNO_AND_VALUE(IPv4UDPBidirectionalBindSocketPair(0).Create()); + std::vector entries = + ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + UDPEntry e; + ASSERT_TRUE(FindByLocalAddr(entries, &e, sockets->first_addr())) + << DescribeFirstInetSocket(*sockets); + EXPECT_EQ(e.uid, geteuid()); + ASSERT_TRUE(FindByRemoteAddr(entries, &e, sockets->first_addr())) + << DescribeSecondInetSocket(*sockets); + EXPECT_EQ(e.uid, geteuid()); +} + +TEST(ProcNetUDP, FindMutualEntries) { + auto sockets = + ASSERT_NO_ERRNO_AND_VALUE(IPv4UDPBidirectionalBindSocketPair(0).Create()); + std::vector entries = + ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + + EXPECT_TRUE(FindByLocalAddr(entries, nullptr, sockets->first_addr())) + << DescribeFirstInetSocket(*sockets); + EXPECT_TRUE(FindByRemoteAddr(entries, nullptr, sockets->first_addr())) + << DescribeSecondInetSocket(*sockets); + + EXPECT_TRUE(FindByLocalAddr(entries, nullptr, sockets->second_addr())) + << DescribeSecondInetSocket(*sockets); + EXPECT_TRUE(FindByRemoteAddr(entries, nullptr, sockets->second_addr())) + << DescribeFirstInetSocket(*sockets); +} + +TEST(ProcNetUDP, EntriesRemovedOnClose) { + auto sockets = + ASSERT_NO_ERRNO_AND_VALUE(IPv4UDPBidirectionalBindSocketPair(0).Create()); + std::vector entries = + ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + + EXPECT_TRUE(FindByLocalAddr(entries, nullptr, sockets->first_addr())) + << DescribeFirstInetSocket(*sockets); + EXPECT_TRUE(FindByLocalAddr(entries, nullptr, sockets->second_addr())) + << DescribeSecondInetSocket(*sockets); + + EXPECT_THAT(close(sockets->release_first_fd()), SyscallSucceeds()); + entries = ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + // First socket's entry should be gone, but the second socket's entry should + // still exist. + EXPECT_FALSE(FindByLocalAddr(entries, nullptr, sockets->first_addr())) + << DescribeFirstInetSocket(*sockets); + EXPECT_TRUE(FindByLocalAddr(entries, nullptr, sockets->second_addr())) + << DescribeSecondInetSocket(*sockets); + + EXPECT_THAT(close(sockets->release_second_fd()), SyscallSucceeds()); + entries = ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + // Both entries should be gone. + EXPECT_FALSE(FindByLocalAddr(entries, nullptr, sockets->first_addr())) + << DescribeFirstInetSocket(*sockets); + EXPECT_FALSE(FindByLocalAddr(entries, nullptr, sockets->second_addr())) + << DescribeSecondInetSocket(*sockets); +} + +PosixErrorOr> BoundUDPSocket() { + ASSIGN_OR_RETURN_ERRNO(std::unique_ptr socket, + IPv4UDPUnboundSocket(0).Create()); + struct sockaddr_in addr; + addr.sin_family = AF_INET; + addr.sin_addr.s_addr = htonl(INADDR_ANY); + addr.sin_port = 0; + + int res = bind(socket->get(), reinterpret_cast(&addr), + sizeof(addr)); + if (res) { + return PosixError(errno, "bind()"); + } + return socket; +} + +TEST(ProcNetUDP, BoundEntry) { + std::unique_ptr socket = + ASSERT_NO_ERRNO_AND_VALUE(BoundUDPSocket()); + struct sockaddr addr; + socklen_t len = sizeof(addr); + ASSERT_THAT(getsockname(socket->get(), &addr, &len), SyscallSucceeds()); + uint16_t port = PortFromInetSockaddr(&addr); + + std::vector entries = + ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + UDPEntry e; + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(FindByFD(entries, &e, socket->get()))); + EXPECT_EQ(e.local_port, port); + EXPECT_EQ(e.remote_addr, 0); + EXPECT_EQ(e.remote_port, 0); +} + +TEST(ProcNetUDP, BoundSocketStateClosed) { + std::unique_ptr socket = + ASSERT_NO_ERRNO_AND_VALUE(BoundUDPSocket()); + std::vector entries = + ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + UDPEntry e; + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(FindByFD(entries, &e, socket->get()))); + EXPECT_EQ(e.state, TCP_CLOSE); +} + +TEST(ProcNetUDP, ConnectedSocketStateEstablished) { + auto sockets = + ASSERT_NO_ERRNO_AND_VALUE(IPv4UDPBidirectionalBindSocketPair(0).Create()); + std::vector entries = + ASSERT_NO_ERRNO_AND_VALUE(ProcNetUDPEntries()); + + UDPEntry e; + ASSERT_TRUE(FindByLocalAddr(entries, &e, sockets->first_addr())) + << DescribeFirstInetSocket(*sockets); + EXPECT_EQ(e.state, TCP_ESTABLISHED); + + ASSERT_TRUE(FindByLocalAddr(entries, &e, sockets->second_addr())) + << DescribeSecondInetSocket(*sockets); + EXPECT_EQ(e.state, TCP_ESTABLISHED); +} + +} // namespace +} // namespace testing +} // namespace gvisor diff --git a/test/util/fs_util.cc b/test/util/fs_util.cc index ae49725a0..f7d231b14 100644 --- a/test/util/fs_util.cc +++ b/test/util/fs_util.cc @@ -105,6 +105,15 @@ PosixErrorOr Stat(absl::string_view path) { return stat_buf; } +PosixErrorOr Fstat(int fd) { + struct stat stat_buf; + int res = fstat(fd, &stat_buf); + if (res < 0) { + return PosixError(errno, absl::StrCat("fstat ", fd)); + } + return stat_buf; +} + PosixErrorOr Exists(absl::string_view path) { struct stat stat_buf; int res = stat(std::string(path).c_str(), &stat_buf); diff --git a/test/util/fs_util.h b/test/util/fs_util.h index 3969f8309..e5b555891 100644 --- a/test/util/fs_util.h +++ b/test/util/fs_util.h @@ -35,6 +35,9 @@ PosixErrorOr Exists(absl::string_view path); // Returns a stat structure for the given path or an error. PosixErrorOr Stat(absl::string_view path); +// Returns a stat struct for the given fd. +PosixErrorOr Fstat(int fd); + // Deletes the file or directory at path or returns an error. PosixError Delete(absl::string_view path); -- cgit v1.2.3 From 888e87909e1a6c3cf93498e6ecb2d451c7551153 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 30 Aug 2019 15:01:55 -0700 Subject: Add C++ toolchain and fix compile issues. This was accidentally introduced in 31f05d5d4f62c4cd4fe3b95b333d0130aae4b2c1. Fixes #788. PiperOrigin-RevId: 266462843 --- WORKSPACE | 11 +++++++++++ test/syscalls/linux/BUILD | 2 ++ test/syscalls/linux/open.cc | 1 + test/syscalls/linux/pty.cc | 24 ++++++++---------------- test/syscalls/linux/semaphore.cc | 2 ++ 5 files changed, 24 insertions(+), 16 deletions(-) (limited to 'test/syscalls/linux/BUILD') diff --git a/WORKSPACE b/WORKSPACE index 88ba4a2c9..c403e774d 100644 --- a/WORKSPACE +++ b/WORKSPACE @@ -32,6 +32,17 @@ load("@bazel_gazelle//:deps.bzl", "gazelle_dependencies", "go_repository") gazelle_dependencies() +# Load C++ rules. +http_archive( + name = "rules_cc", + sha256 = "67412176974bfce3f4cf8bdaff39784a72ed709fc58def599d1f68710b58d68b", + strip_prefix = "rules_cc-b7fe9697c0c76ab2fd431a891dbb9a6a32ed7c3e", + urls = [ + "https://mirror.bazel.build/github.com/bazelbuild/rules_cc/archive/b7fe9697c0c76ab2fd431a891dbb9a6a32ed7c3e.zip", + "https://github.com/bazelbuild/rules_cc/archive/b7fe9697c0c76ab2fd431a891dbb9a6a32ed7c3e.zip", + ], +) + # Load protobuf dependencies. load("@bazel_tools//tools/build_defs/repo:git.bzl", "git_repository") diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 1ce38c929..bb065aa4f 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1213,6 +1213,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/memory", "@com_google_absl//absl/strings", "@com_google_googletest//:gtest", ], @@ -3362,6 +3363,7 @@ cc_binary( "//test/util:test_util", "//test/util:thread_util", "@com_google_absl//absl/base:core_headers", + "@com_google_absl//absl/memory", "@com_google_absl//absl/synchronization", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index e0525f386..2b1df52ce 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -21,6 +21,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/memory/memory.h" #include "test/syscalls/linux/file_base.h" #include "test/util/capability_util.h" #include "test/util/cleanup.h" diff --git a/test/syscalls/linux/pty.cc b/test/syscalls/linux/pty.cc index bd6907876..c605b6549 100644 --- a/test/syscalls/linux/pty.cc +++ b/test/syscalls/linux/pty.cc @@ -1292,10 +1292,8 @@ TEST_F(JobControlTest, ReleaseTTY) { // Make sure we're ignoring SIGHUP, which will be sent to this process once we // disconnect they TTY. - struct sigaction sa = { - .sa_handler = SIG_IGN, - .sa_flags = 0, - }; + struct sigaction sa = {}; + sa.sa_handler = SIG_IGN; sigemptyset(&sa.sa_mask); struct sigaction old_sa; EXPECT_THAT(sigaction(SIGHUP, &sa, &old_sa), SyscallSucceeds()); @@ -1362,10 +1360,8 @@ TEST_F(JobControlTest, ReleaseTTYSignals) { ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); received = 0; - struct sigaction sa = { - .sa_handler = sig_handler, - .sa_flags = 0, - }; + struct sigaction sa = {}; + sa.sa_handler = sig_handler; sigemptyset(&sa.sa_mask); sigaddset(&sa.sa_mask, SIGHUP); sigaddset(&sa.sa_mask, SIGCONT); @@ -1403,10 +1399,8 @@ TEST_F(JobControlTest, ReleaseTTYSignals) { // Make sure we're ignoring SIGHUP, which will be sent to this process once we // disconnect they TTY. - struct sigaction sighup_sa = { - .sa_handler = SIG_IGN, - .sa_flags = 0, - }; + struct sigaction sighup_sa = {}; + sighup_sa.sa_handler = SIG_IGN; sigemptyset(&sighup_sa.sa_mask); struct sigaction old_sa; EXPECT_THAT(sigaction(SIGHUP, &sighup_sa, &old_sa), SyscallSucceeds()); @@ -1456,10 +1450,8 @@ TEST_F(JobControlTest, SetForegroundProcessGroup) { ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); // Ignore SIGTTOU so that we don't stop ourself when calling tcsetpgrp. - struct sigaction sa = { - .sa_handler = SIG_IGN, - .sa_flags = 0, - }; + struct sigaction sa = {}; + sa.sa_handler = SIG_IGN; sigemptyset(&sa.sa_mask); sigaction(SIGTTOU, &sa, NULL); diff --git a/test/syscalls/linux/semaphore.cc b/test/syscalls/linux/semaphore.cc index 421318fcb..40c57f543 100644 --- a/test/syscalls/linux/semaphore.cc +++ b/test/syscalls/linux/semaphore.cc @@ -15,6 +15,7 @@ #include #include #include + #include #include #include @@ -22,6 +23,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" #include "absl/base/macros.h" +#include "absl/memory/memory.h" #include "absl/synchronization/mutex.h" #include "absl/time/clock.h" #include "test/util/capability_util.h" -- cgit v1.2.3 From 54bf2e8eff4a5e619e7e3abafcda6ffc52d937f2 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Fri, 30 Aug 2019 18:09:37 -0700 Subject: Automated rollback of changelist 261387276 PiperOrigin-RevId: 266491264 --- pkg/sentry/fs/tty/BUILD | 1 - pkg/sentry/fs/tty/dir.go | 3 - pkg/sentry/fs/tty/master.go | 17 +- pkg/sentry/fs/tty/slave.go | 13 +- pkg/sentry/fs/tty/terminal.go | 92 +--------- pkg/sentry/kernel/BUILD | 1 - pkg/sentry/kernel/sessions.go | 12 +- pkg/sentry/kernel/task_start.go | 3 +- pkg/sentry/kernel/thread_group.go | 179 ------------------- pkg/sentry/kernel/tty.go | 28 --- test/syscalls/BUILD | 4 - test/syscalls/linux/BUILD | 19 --- test/syscalls/linux/pty.cc | 351 +++----------------------------------- test/syscalls/linux/pty_root.cc | 68 -------- test/util/BUILD | 11 -- test/util/pty_util.cc | 45 ----- test/util/pty_util.h | 30 ---- 17 files changed, 34 insertions(+), 843 deletions(-) delete mode 100644 pkg/sentry/kernel/tty.go delete mode 100644 test/syscalls/linux/pty_root.cc delete mode 100644 test/util/pty_util.cc delete mode 100644 test/util/pty_util.h (limited to 'test/syscalls/linux/BUILD') diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD index 291164986..5e9327aec 100644 --- a/pkg/sentry/fs/tty/BUILD +++ b/pkg/sentry/fs/tty/BUILD @@ -23,7 +23,6 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go index 2f639c823..1d128532b 100644 --- a/pkg/sentry/fs/tty/dir.go +++ b/pkg/sentry/fs/tty/dir.go @@ -129,9 +129,6 @@ func newDir(ctx context.Context, m *fs.MountSource) *fs.Inode { // Release implements fs.InodeOperations.Release. func (d *dirInodeOperations) Release(ctx context.Context) { - d.mu.Lock() - defer d.mu.Unlock() - d.master.DecRef() if len(d.slaves) != 0 { panic(fmt.Sprintf("devpts directory still contains active terminals: %+v", d)) diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index 19b7557d5..92ec1ca18 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -172,19 +172,6 @@ func (mf *masterFileOperations) Ioctl(ctx context.Context, _ *fs.File, io userme return 0, mf.t.ld.windowSize(ctx, io, args) case linux.TIOCSWINSZ: return 0, mf.t.ld.setWindowSize(ctx, io, args) - case linux.TIOCSCTTY: - // Make the given terminal the controlling terminal of the - // calling process. - return 0, mf.t.setControllingTTY(ctx, io, args, true /* isMaster */) - case linux.TIOCNOTTY: - // Release this process's controlling terminal. - return 0, mf.t.releaseControllingTTY(ctx, io, args, true /* isMaster */) - case linux.TIOCGPGRP: - // Get the foreground process group. - return mf.t.foregroundProcessGroup(ctx, io, args, true /* isMaster */) - case linux.TIOCSPGRP: - // Set the foreground process group. - return mf.t.setForegroundProcessGroup(ctx, io, args, true /* isMaster */) default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY @@ -198,6 +185,8 @@ func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { linux.TCSETS, linux.TCSETSW, linux.TCSETSF, + linux.TIOCGPGRP, + linux.TIOCSPGRP, linux.TIOCGWINSZ, linux.TIOCSWINSZ, linux.TIOCSETD, @@ -211,6 +200,8 @@ func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { linux.TIOCEXCL, linux.TIOCNXCL, linux.TIOCGEXCL, + linux.TIOCNOTTY, + linux.TIOCSCTTY, linux.TIOCGSID, linux.TIOCGETD, linux.TIOCVHANGUP, diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 944c4ada1..e30266404 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -152,16 +152,9 @@ func (sf *slaveFileOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem case linux.TIOCSCTTY: // Make the given terminal the controlling terminal of the // calling process. - return 0, sf.si.t.setControllingTTY(ctx, io, args, false /* isMaster */) - case linux.TIOCNOTTY: - // Release this process's controlling terminal. - return 0, sf.si.t.releaseControllingTTY(ctx, io, args, false /* isMaster */) - case linux.TIOCGPGRP: - // Get the foreground process group. - return sf.si.t.foregroundProcessGroup(ctx, io, args, false /* isMaster */) - case linux.TIOCSPGRP: - // Set the foreground process group. - return sf.si.t.setForegroundProcessGroup(ctx, io, args, false /* isMaster */) + // TODO(b/129283598): Implement once we have support for job + // control. + return 0, nil default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index ff8138820..b7cecb2ed 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -17,10 +17,7 @@ package tty import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" ) // Terminal is a pseudoterminal. @@ -29,100 +26,23 @@ import ( type Terminal struct { refs.AtomicRefCount - // n is the terminal index. It is immutable. + // n is the terminal index. n uint32 - // d is the containing directory. It is immutable. + // d is the containing directory. d *dirInodeOperations - // ld is the line discipline of the terminal. It is immutable. + // ld is the line discipline of the terminal. ld *lineDiscipline - - // masterKTTY contains the controlling process of the master end of - // this terminal. This field is immutable. - masterKTTY *kernel.TTY - - // slaveKTTY contains the controlling process of the slave end of this - // terminal. This field is immutable. - slaveKTTY *kernel.TTY } func newTerminal(ctx context.Context, d *dirInodeOperations, n uint32) *Terminal { termios := linux.DefaultSlaveTermios t := Terminal{ - d: d, - n: n, - ld: newLineDiscipline(termios), - masterKTTY: &kernel.TTY{}, - slaveKTTY: &kernel.TTY{}, + d: d, + n: n, + ld: newLineDiscipline(termios), } t.EnableLeakCheck("tty.Terminal") return &t } - -// setControllingTTY makes tm the controlling terminal of the calling thread -// group. -func (tm *Terminal) setControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { - task := kernel.TaskFromContext(ctx) - if task == nil { - panic("setControllingTTY must be called from a task context") - } - - return task.ThreadGroup().SetControllingTTY(tm.tty(isMaster), args[2].Int()) -} - -// releaseControllingTTY removes tm as the controlling terminal of the calling -// thread group. -func (tm *Terminal) releaseControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { - task := kernel.TaskFromContext(ctx) - if task == nil { - panic("releaseControllingTTY must be called from a task context") - } - - return task.ThreadGroup().ReleaseControllingTTY(tm.tty(isMaster)) -} - -// foregroundProcessGroup gets the process group ID of tm's foreground process. -func (tm *Terminal) foregroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { - task := kernel.TaskFromContext(ctx) - if task == nil { - panic("foregroundProcessGroup must be called from a task context") - } - - ret, err := task.ThreadGroup().ForegroundProcessGroup(tm.tty(isMaster)) - if err != nil { - return 0, err - } - - // Write it out to *arg. - _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(ret), usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err -} - -// foregroundProcessGroup sets tm's foreground process. -func (tm *Terminal) setForegroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { - task := kernel.TaskFromContext(ctx) - if task == nil { - panic("setForegroundProcessGroup must be called from a task context") - } - - // Read in the process group ID. - var pgid int32 - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgid, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - - ret, err := task.ThreadGroup().SetForegroundProcessGroup(tm.tty(isMaster), kernel.ProcessGroupID(pgid)) - return uintptr(ret), err -} - -func (tm *Terminal) tty(isMaster bool) *kernel.TTY { - if isMaster { - return tm.masterKTTY - } - return tm.slaveKTTY -} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 41bee9a22..e61d39c82 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -144,7 +144,6 @@ go_library( "threads.go", "timekeeper.go", "timekeeper_state.go", - "tty.go", "uts_namespace.go", "vdso.go", "version.go", diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go index e5f297478..81fcd8258 100644 --- a/pkg/sentry/kernel/sessions.go +++ b/pkg/sentry/kernel/sessions.go @@ -47,11 +47,6 @@ type Session struct { // The id is immutable. id SessionID - // foreground is the foreground process group. - // - // This is protected by TaskSet.mu. - foreground *ProcessGroup - // ProcessGroups is a list of process groups in this Session. This is // protected by TaskSet.mu. processGroups processGroupList @@ -265,14 +260,12 @@ func (pg *ProcessGroup) SendSignal(info *arch.SignalInfo) error { func (tg *ThreadGroup) CreateSession() error { tg.pidns.owner.mu.Lock() defer tg.pidns.owner.mu.Unlock() - tg.signalHandlers.mu.Lock() - defer tg.signalHandlers.mu.Unlock() return tg.createSession() } // createSession creates a new session for a threadgroup. // -// Precondition: callers must hold TaskSet.mu and the signal mutex for writing. +// Precondition: callers must hold TaskSet.mu for writing. func (tg *ThreadGroup) createSession() error { // Get the ID for this thread in the current namespace. id := tg.pidns.tgids[tg] @@ -353,9 +346,6 @@ func (tg *ThreadGroup) createSession() error { ns.processGroups[ProcessGroupID(local)] = pg } - // Disconnect from the controlling terminal. - tg.tty = nil - return nil } diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index ae6fc4025..d60cd62c7 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -172,10 +172,9 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { if parentPG := tg.parentPG(); parentPG == nil { tg.createSession() } else { - // Inherit the process group and terminal. + // Inherit the process group. parentPG.incRefWithParent(parentPG) tg.processGroup = parentPG - tg.tty = t.parent.tg.tty } } tg.tasks.PushBack(t) diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 0eef24bfb..2a97e3e8e 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -19,13 +19,10 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/syserror" ) // A ThreadGroup is a logical grouping of tasks that has widespread @@ -248,12 +245,6 @@ type ThreadGroup struct { // // mounts is immutable. mounts *fs.MountNamespace - - // tty is the thread group's controlling terminal. If nil, there is no - // controlling terminal. - // - // tty is protected by the signal mutex. - tty *TTY } // newThreadGroup returns a new, empty thread group in PID namespace ns. The @@ -333,176 +324,6 @@ func (tg *ThreadGroup) forEachChildThreadGroupLocked(fn func(*ThreadGroup)) { } } -// SetControllingTTY sets tty as the controlling terminal of tg. -func (tg *ThreadGroup) SetControllingTTY(tty *TTY, arg int32) error { - tty.mu.Lock() - defer tty.mu.Unlock() - - // We might be asked to set the controlling terminal of multiple - // processes, so we lock both the TaskSet and SignalHandlers. - tg.pidns.owner.mu.Lock() - defer tg.pidns.owner.mu.Unlock() - tg.signalHandlers.mu.Lock() - defer tg.signalHandlers.mu.Unlock() - - // "The calling process must be a session leader and not have a - // controlling terminal already." - tty_ioctl(4) - if tg.processGroup.session.leader != tg || tg.tty != nil { - return syserror.EINVAL - } - - // "If this terminal is already the controlling terminal of a different - // session group, then the ioctl fails with EPERM, unless the caller - // has the CAP_SYS_ADMIN capability and arg equals 1, in which case the - // terminal is stolen, and all processes that had it as controlling - // terminal lose it." - tty_ioctl(4) - if tty.tg != nil && tg.processGroup.session != tty.tg.processGroup.session { - if !auth.CredentialsFromContext(tg.leader).HasCapability(linux.CAP_SYS_ADMIN) || arg != 1 { - return syserror.EPERM - } - // Steal the TTY away. Unlike TIOCNOTTY, don't send signals. - for othertg := range tg.pidns.owner.Root.tgids { - // This won't deadlock by locking tg.signalHandlers - // because at this point: - // - We only lock signalHandlers if it's in the same - // session as the tty's controlling thread group. - // - We know that the calling thread group is not in - // the same session as the tty's controlling thread - // group. - if othertg.processGroup.session == tty.tg.processGroup.session { - othertg.signalHandlers.mu.Lock() - othertg.tty = nil - othertg.signalHandlers.mu.Unlock() - } - } - } - - // Set the controlling terminal and foreground process group. - tg.tty = tty - tg.processGroup.session.foreground = tg.processGroup - // Set this as the controlling process of the terminal. - tty.tg = tg - - return nil -} - -// ReleaseControllingTTY gives up tty as the controlling tty of tg. -func (tg *ThreadGroup) ReleaseControllingTTY(tty *TTY) error { - tty.mu.Lock() - defer tty.mu.Unlock() - - // We might be asked to set the controlling terminal of multiple - // processes, so we lock both the TaskSet and SignalHandlers. - tg.pidns.owner.mu.RLock() - defer tg.pidns.owner.mu.RUnlock() - - // Just below, we may re-lock signalHandlers in order to send signals. - // Thus we can't defer Unlock here. - tg.signalHandlers.mu.Lock() - - if tg.tty == nil || tg.tty != tty { - tg.signalHandlers.mu.Unlock() - return syserror.ENOTTY - } - - // "If the process was session leader, then send SIGHUP and SIGCONT to - // the foreground process group and all processes in the current - // session lose their controlling terminal." - tty_ioctl(4) - // Remove tty as the controlling tty for each process in the session, - // then send them SIGHUP and SIGCONT. - - // If we're not the session leader, we don't have to do much. - if tty.tg != tg { - tg.tty = nil - tg.signalHandlers.mu.Unlock() - return nil - } - - tg.signalHandlers.mu.Unlock() - - // We're the session leader. SIGHUP and SIGCONT the foreground process - // group and remove all controlling terminals in the session. - var lastErr error - for othertg := range tg.pidns.owner.Root.tgids { - if othertg.processGroup.session == tg.processGroup.session { - othertg.signalHandlers.mu.Lock() - othertg.tty = nil - if othertg.processGroup == tg.processGroup.session.foreground { - if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGHUP)}, true /* group */); err != nil { - lastErr = err - } - if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGCONT)}, true /* group */); err != nil { - lastErr = err - } - } - othertg.signalHandlers.mu.Unlock() - } - } - - return lastErr -} - -// ForegroundProcessGroup returns the process group ID of the foreground -// process group. -func (tg *ThreadGroup) ForegroundProcessGroup(tty *TTY) (int32, error) { - tty.mu.Lock() - defer tty.mu.Unlock() - - tg.pidns.owner.mu.Lock() - defer tg.pidns.owner.mu.Unlock() - tg.signalHandlers.mu.Lock() - defer tg.signalHandlers.mu.Unlock() - - // "When fd does not refer to the controlling terminal of the calling - // process, -1 is returned" - tcgetpgrp(3) - if tg.tty != tty { - return -1, syserror.ENOTTY - } - - return int32(tg.processGroup.session.foreground.id), nil -} - -// SetForegroundProcessGroup sets the foreground process group of tty to pgid. -func (tg *ThreadGroup) SetForegroundProcessGroup(tty *TTY, pgid ProcessGroupID) (int32, error) { - tty.mu.Lock() - defer tty.mu.Unlock() - - tg.pidns.owner.mu.Lock() - defer tg.pidns.owner.mu.Unlock() - tg.signalHandlers.mu.Lock() - defer tg.signalHandlers.mu.Unlock() - - // TODO(b/129283598): "If tcsetpgrp() is called by a member of a - // background process group in its session, and the calling process is - // not blocking or ignoring SIGTTOU, a SIGTTOU signal is sent to all - // members of this background process group." - - // tty must be the controlling terminal. - if tg.tty != tty { - return -1, syserror.ENOTTY - } - - // pgid must be positive. - if pgid < 0 { - return -1, syserror.EINVAL - } - - // pg must not be empty. Empty process groups are removed from their - // pid namespaces. - pg, ok := tg.pidns.processGroups[pgid] - if !ok { - return -1, syserror.ESRCH - } - - // pg must be part of this process's session. - if tg.processGroup.session != pg.session { - return -1, syserror.EPERM - } - - tg.processGroup.session.foreground.id = pgid - return 0, nil -} - // itimerRealListener implements ktime.Listener for ITIMER_REAL expirations. // // +stateify savable diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go deleted file mode 100644 index 34f84487a..000000000 --- a/pkg/sentry/kernel/tty.go +++ /dev/null @@ -1,28 +0,0 @@ -// 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 kernel - -import "sync" - -// TTY defines the relationship between a thread group and its controlling -// terminal. -// -// +stateify savable -type TTY struct { - mu sync.Mutex `state:"nosave"` - - // tg is protected by mu. - tg *ThreadGroup -} diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 6947ddc25..a8a2e75d3 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -318,10 +318,6 @@ syscall_test( test = "//test/syscalls/linux:pty_test", ) -syscall_test( - test = "//test/syscalls/linux:pty_root_test", -) - syscall_test( add_overlay = True, test = "//test/syscalls/linux:pwritev2_test", diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index bb065aa4f..aeb4b405d 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1278,10 +1278,8 @@ cc_binary( srcs = ["pty.cc"], linkstatic = 1, deps = [ - "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:posix_error", - "//test/util:pty_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", @@ -1293,23 +1291,6 @@ cc_binary( ], ) -cc_binary( - name = "pty_root_test", - testonly = 1, - srcs = ["pty_root.cc"], - linkstatic = 1, - deps = [ - "//test/util:capability_util", - "//test/util:file_descriptor", - "//test/util:posix_error", - "//test/util:pty_util", - "//test/util:test_main", - "//test/util:thread_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_googletest//:gtest", - ], -) - cc_binary( name = "partial_bad_buffer_test", testonly = 1, diff --git a/test/syscalls/linux/pty.cc b/test/syscalls/linux/pty.cc index c605b6549..d1ab4703f 100644 --- a/test/syscalls/linux/pty.cc +++ b/test/syscalls/linux/pty.cc @@ -13,17 +13,13 @@ // limitations under the License. #include -#include #include #include -#include -#include #include #include #include #include #include -#include #include #include @@ -35,10 +31,8 @@ #include "absl/synchronization/notification.h" #include "absl/time/clock.h" #include "absl/time/time.h" -#include "test/util/capability_util.h" #include "test/util/file_descriptor.h" #include "test/util/posix_error.h" -#include "test/util/pty_util.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" @@ -376,6 +370,25 @@ PosixErrorOr PollAndReadFd(int fd, void* buf, size_t count, return PosixError(ETIMEDOUT, "Poll timed out"); } +// Opens the slave end of the passed master as R/W and nonblocking. +PosixErrorOr OpenSlave(const FileDescriptor& master) { + // Get pty index. + int n; + int ret = ioctl(master.get(), TIOCGPTN, &n); + if (ret < 0) { + return PosixError(errno, "ioctl(TIOCGPTN) failed"); + } + + // Unlock pts. + int unlock = 0; + ret = ioctl(master.get(), TIOCSPTLCK, &unlock); + if (ret < 0) { + return PosixError(errno, "ioctl(TIOSPTLCK) failed"); + } + + return Open(absl::StrCat("/dev/pts/", n), O_RDWR | O_NONBLOCK); +} + TEST(BasicPtyTest, StatUnopenedMaster) { struct stat s; ASSERT_THAT(stat("/dev/ptmx", &s), SyscallSucceeds()); @@ -1220,332 +1233,6 @@ TEST_F(PtyTest, SetMasterWindowSize) { EXPECT_EQ(retrieved_ws.ws_col, kCols); } -class JobControlTest : public ::testing::Test { - protected: - void SetUp() override { - master_ = ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/ptmx", O_RDWR | O_NONBLOCK)); - slave_ = ASSERT_NO_ERRNO_AND_VALUE(OpenSlave(master_)); - - // Make this a session leader, which also drops the controlling terminal. - // In the gVisor test environment, this test will be run as the session - // leader already (as the sentry init process). - if (!IsRunningOnGvisor()) { - ASSERT_THAT(setsid(), SyscallSucceeds()); - } - } - - // Master and slave ends of the PTY. Non-blocking. - FileDescriptor master_; - FileDescriptor slave_; -}; - -TEST_F(JobControlTest, SetTTYMaster) { - ASSERT_THAT(ioctl(master_.get(), TIOCSCTTY, 0), SyscallSucceeds()); -} - -TEST_F(JobControlTest, SetTTY) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); -} - -TEST_F(JobControlTest, SetTTYNonLeader) { - // Fork a process that won't be the session leader. - pid_t child = fork(); - if (!child) { - // We shouldn't be able to set the terminal. - TEST_PCHECK(ioctl(slave_.get(), TIOCSCTTY, 0)); - _exit(0); - } - - int wstatus; - ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); - ASSERT_EQ(wstatus, 0); -} - -TEST_F(JobControlTest, SetTTYBadArg) { - // Despite the man page saying arg should be 0 here, Linux doesn't actually - // check. - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 1), SyscallSucceeds()); -} - -TEST_F(JobControlTest, SetTTYDifferentSession) { - SKIP_IF(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); - - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - // Fork, join a new session, and try to steal the parent's controlling - // terminal, which should fail. - pid_t child = fork(); - if (!child) { - TEST_PCHECK(setsid() >= 0); - // We shouldn't be able to steal the terminal. - TEST_PCHECK(ioctl(slave_.get(), TIOCSCTTY, 1)); - _exit(0); - } - - int wstatus; - ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); - ASSERT_EQ(wstatus, 0); -} - -TEST_F(JobControlTest, ReleaseTTY) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - // Make sure we're ignoring SIGHUP, which will be sent to this process once we - // disconnect they TTY. - struct sigaction sa = {}; - sa.sa_handler = SIG_IGN; - sigemptyset(&sa.sa_mask); - struct sigaction old_sa; - EXPECT_THAT(sigaction(SIGHUP, &sa, &old_sa), SyscallSucceeds()); - EXPECT_THAT(ioctl(slave_.get(), TIOCNOTTY), SyscallSucceeds()); - EXPECT_THAT(sigaction(SIGHUP, &old_sa, NULL), SyscallSucceeds()); -} - -TEST_F(JobControlTest, ReleaseUnsetTTY) { - ASSERT_THAT(ioctl(slave_.get(), TIOCNOTTY), SyscallFailsWithErrno(ENOTTY)); -} - -TEST_F(JobControlTest, ReleaseWrongTTY) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - ASSERT_THAT(ioctl(master_.get(), TIOCNOTTY), SyscallFailsWithErrno(ENOTTY)); -} - -TEST_F(JobControlTest, ReleaseTTYNonLeader) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - pid_t child = fork(); - if (!child) { - TEST_PCHECK(!ioctl(slave_.get(), TIOCNOTTY)); - _exit(0); - } - - int wstatus; - ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); - ASSERT_EQ(wstatus, 0); -} - -TEST_F(JobControlTest, ReleaseTTYDifferentSession) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - pid_t child = fork(); - if (!child) { - // Join a new session, then try to disconnect. - TEST_PCHECK(setsid() >= 0); - TEST_PCHECK(ioctl(slave_.get(), TIOCNOTTY)); - _exit(0); - } - - int wstatus; - ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); - ASSERT_EQ(wstatus, 0); -} - -// Used by the child process spawned in ReleaseTTYSignals to track received -// signals. -static int received; - -void sig_handler(int signum) { received |= signum; } - -// When the session leader releases its controlling terminal, the foreground -// process group gets SIGHUP, then SIGCONT. This test: -// - Spawns 2 threads -// - Has thread 1 return 0 if it gets both SIGHUP and SIGCONT -// - Has thread 2 leave the foreground process group, and return non-zero if it -// receives any signals. -// - Has the parent thread release its controlling terminal -// - Checks that thread 1 got both signals -// - Checks that thread 2 didn't get any signals. -TEST_F(JobControlTest, ReleaseTTYSignals) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - received = 0; - struct sigaction sa = {}; - sa.sa_handler = sig_handler; - sigemptyset(&sa.sa_mask); - sigaddset(&sa.sa_mask, SIGHUP); - sigaddset(&sa.sa_mask, SIGCONT); - sigprocmask(SIG_BLOCK, &sa.sa_mask, NULL); - - pid_t same_pgrp_child = fork(); - if (!same_pgrp_child) { - // The child will wait for SIGHUP and SIGCONT, then return 0. It begins with - // SIGHUP and SIGCONT blocked. We install signal handlers for those signals, - // then use sigsuspend to wait for those specific signals. - TEST_PCHECK(!sigaction(SIGHUP, &sa, NULL)); - TEST_PCHECK(!sigaction(SIGCONT, &sa, NULL)); - sigset_t mask; - sigfillset(&mask); - sigdelset(&mask, SIGHUP); - sigdelset(&mask, SIGCONT); - while (received != (SIGHUP | SIGCONT)) { - sigsuspend(&mask); - } - _exit(0); - } - - // We don't want to block these anymore. - sigprocmask(SIG_UNBLOCK, &sa.sa_mask, NULL); - - // This child will return non-zero if either SIGHUP or SIGCONT are received. - pid_t diff_pgrp_child = fork(); - if (!diff_pgrp_child) { - TEST_PCHECK(!setpgid(0, 0)); - TEST_PCHECK(pause()); - _exit(1); - } - - EXPECT_THAT(setpgid(diff_pgrp_child, diff_pgrp_child), SyscallSucceeds()); - - // Make sure we're ignoring SIGHUP, which will be sent to this process once we - // disconnect they TTY. - struct sigaction sighup_sa = {}; - sighup_sa.sa_handler = SIG_IGN; - sigemptyset(&sighup_sa.sa_mask); - struct sigaction old_sa; - EXPECT_THAT(sigaction(SIGHUP, &sighup_sa, &old_sa), SyscallSucceeds()); - - // Release the controlling terminal, sending SIGHUP and SIGCONT to all other - // processes in this process group. - EXPECT_THAT(ioctl(slave_.get(), TIOCNOTTY), SyscallSucceeds()); - - EXPECT_THAT(sigaction(SIGHUP, &old_sa, NULL), SyscallSucceeds()); - - // The child in the same process group will get signaled. - int wstatus; - EXPECT_THAT(waitpid(same_pgrp_child, &wstatus, 0), - SyscallSucceedsWithValue(same_pgrp_child)); - EXPECT_EQ(wstatus, 0); - - // The other child will not get signaled. - EXPECT_THAT(waitpid(diff_pgrp_child, &wstatus, WNOHANG), - SyscallSucceedsWithValue(0)); - EXPECT_THAT(kill(diff_pgrp_child, SIGKILL), SyscallSucceeds()); -} - -TEST_F(JobControlTest, GetForegroundProcessGroup) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - pid_t foreground_pgid; - pid_t pid; - ASSERT_THAT(ioctl(slave_.get(), TIOCGPGRP, &foreground_pgid), - SyscallSucceeds()); - ASSERT_THAT(pid = getpid(), SyscallSucceeds()); - - ASSERT_EQ(foreground_pgid, pid); -} - -TEST_F(JobControlTest, GetForegroundProcessGroupNonControlling) { - // At this point there's no controlling terminal, so TIOCGPGRP should fail. - pid_t foreground_pgid; - ASSERT_THAT(ioctl(slave_.get(), TIOCGPGRP, &foreground_pgid), - SyscallFailsWithErrno(ENOTTY)); -} - -// This test: -// - sets itself as the foreground process group -// - creates a child process in a new process group -// - sets that child as the foreground process group -// - kills its child and sets itself as the foreground process group. -TEST_F(JobControlTest, SetForegroundProcessGroup) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - // Ignore SIGTTOU so that we don't stop ourself when calling tcsetpgrp. - struct sigaction sa = {}; - sa.sa_handler = SIG_IGN; - sigemptyset(&sa.sa_mask); - sigaction(SIGTTOU, &sa, NULL); - - // Set ourself as the foreground process group. - ASSERT_THAT(tcsetpgrp(slave_.get(), getpgid(0)), SyscallSucceeds()); - - // Create a new process that just waits to be signaled. - pid_t child = fork(); - if (!child) { - TEST_PCHECK(!pause()); - // We should never reach this. - _exit(1); - } - - // Make the child its own process group, then make it the controlling process - // group of the terminal. - ASSERT_THAT(setpgid(child, child), SyscallSucceeds()); - ASSERT_THAT(tcsetpgrp(slave_.get(), child), SyscallSucceeds()); - - // Sanity check - we're still the controlling session. - ASSERT_EQ(getsid(0), getsid(child)); - - // Signal the child, wait for it to exit, then retake the terminal. - ASSERT_THAT(kill(child, SIGTERM), SyscallSucceeds()); - int wstatus; - ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); - ASSERT_TRUE(WIFSIGNALED(wstatus)); - ASSERT_EQ(WTERMSIG(wstatus), SIGTERM); - - // Set ourself as the foreground process. - pid_t pgid; - ASSERT_THAT(pgid = getpgid(0), SyscallSucceeds()); - ASSERT_THAT(tcsetpgrp(slave_.get(), pgid), SyscallSucceeds()); -} - -TEST_F(JobControlTest, SetForegroundProcessGroupWrongTTY) { - pid_t pid = getpid(); - ASSERT_THAT(ioctl(slave_.get(), TIOCSPGRP, &pid), - SyscallFailsWithErrno(ENOTTY)); -} - -TEST_F(JobControlTest, SetForegroundProcessGroupNegPgid) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - pid_t pid = -1; - ASSERT_THAT(ioctl(slave_.get(), TIOCSPGRP, &pid), - SyscallFailsWithErrno(EINVAL)); -} - -TEST_F(JobControlTest, SetForegroundProcessGroupEmptyProcessGroup) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - // Create a new process, put it in a new process group, make that group the - // foreground process group, then have the process wait. - pid_t child = fork(); - if (!child) { - TEST_PCHECK(!setpgid(0, 0)); - _exit(0); - } - - // Wait for the child to exit. - int wstatus; - EXPECT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); - // The child's process group doesn't exist anymore - this should fail. - ASSERT_THAT(ioctl(slave_.get(), TIOCSPGRP, &child), - SyscallFailsWithErrno(ESRCH)); -} - -TEST_F(JobControlTest, SetForegroundProcessGroupDifferentSession) { - ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - // Create a new process and put it in a new session. - pid_t child = fork(); - if (!child) { - TEST_PCHECK(setsid() >= 0); - // Tell the parent we're in a new session. - TEST_PCHECK(!raise(SIGSTOP)); - TEST_PCHECK(!pause()); - _exit(1); - } - - // Wait for the child to tell us it's in a new session. - int wstatus; - EXPECT_THAT(waitpid(child, &wstatus, WUNTRACED), - SyscallSucceedsWithValue(child)); - EXPECT_TRUE(WSTOPSIG(wstatus)); - - // Child is in a new session, so we can't make it the foregroup process group. - EXPECT_THAT(ioctl(slave_.get(), TIOCSPGRP, &child), - SyscallFailsWithErrno(EPERM)); - - EXPECT_THAT(kill(child, SIGKILL), SyscallSucceeds()); -} - } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/pty_root.cc b/test/syscalls/linux/pty_root.cc deleted file mode 100644 index d2a321a6e..000000000 --- a/test/syscalls/linux/pty_root.cc +++ /dev/null @@ -1,68 +0,0 @@ -// 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. - -#include -#include - -#include "gtest/gtest.h" -#include "absl/base/macros.h" -#include "test/util/capability_util.h" -#include "test/util/file_descriptor.h" -#include "test/util/posix_error.h" -#include "test/util/pty_util.h" - -namespace gvisor { -namespace testing { - -// These tests should be run as root. -namespace { - -TEST(JobControlRootTest, StealTTY) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); - - // Make this a session leader, which also drops the controlling terminal. - // In the gVisor test environment, this test will be run as the session - // leader already (as the sentry init process). - if (!IsRunningOnGvisor()) { - ASSERT_THAT(setsid(), SyscallSucceeds()); - } - - FileDescriptor master = - ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/ptmx", O_RDWR | O_NONBLOCK)); - FileDescriptor slave = ASSERT_NO_ERRNO_AND_VALUE(OpenSlave(master)); - - // Make slave the controlling terminal. - ASSERT_THAT(ioctl(slave.get(), TIOCSCTTY, 0), SyscallSucceeds()); - - // Fork, join a new session, and try to steal the parent's controlling - // terminal, which should succeed when we have CAP_SYS_ADMIN and pass an arg - // of 1. - pid_t child = fork(); - if (!child) { - TEST_PCHECK(setsid() >= 0); - // We shouldn't be able to steal the terminal with the wrong arg value. - TEST_PCHECK(ioctl(slave.get(), TIOCSCTTY, 0)); - // We should be able to steal it here. - TEST_PCHECK(!ioctl(slave.get(), TIOCSCTTY, 1)); - _exit(0); - } - - int wstatus; - ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); - ASSERT_EQ(wstatus, 0); -} - -} // namespace -} // namespace testing -} // namespace gvisor diff --git a/test/util/BUILD b/test/util/BUILD index cfea029b2..8afd89d8d 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -189,17 +189,6 @@ cc_test( ], ) -cc_library( - name = "pty_util", - testonly = 1, - srcs = ["pty_util.cc"], - hdrs = ["pty_util.h"], - deps = [ - ":file_descriptor", - ":posix_error", - ], -) - cc_library( name = "signal_util", testonly = 1, diff --git a/test/util/pty_util.cc b/test/util/pty_util.cc deleted file mode 100644 index c0fd9a095..000000000 --- a/test/util/pty_util.cc +++ /dev/null @@ -1,45 +0,0 @@ -// Copyright 2019 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. - -#include "test/util/pty_util.h" - -#include -#include - -#include "test/util/file_descriptor.h" -#include "test/util/posix_error.h" - -namespace gvisor { -namespace testing { - -PosixErrorOr OpenSlave(const FileDescriptor& master) { - // Get pty index. - int n; - int ret = ioctl(master.get(), TIOCGPTN, &n); - if (ret < 0) { - return PosixError(errno, "ioctl(TIOCGPTN) failed"); - } - - // Unlock pts. - int unlock = 0; - ret = ioctl(master.get(), TIOCSPTLCK, &unlock); - if (ret < 0) { - return PosixError(errno, "ioctl(TIOSPTLCK) failed"); - } - - return Open(absl::StrCat("/dev/pts/", n), O_RDWR | O_NONBLOCK); -} - -} // namespace testing -} // namespace gvisor diff --git a/test/util/pty_util.h b/test/util/pty_util.h deleted file mode 100644 index 367b14f15..000000000 --- a/test/util/pty_util.h +++ /dev/null @@ -1,30 +0,0 @@ -// Copyright 2019 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. - -#ifndef GVISOR_TEST_UTIL_PTY_UTIL_H_ -#define GVISOR_TEST_UTIL_PTY_UTIL_H_ - -#include "test/util/file_descriptor.h" -#include "test/util/posix_error.h" - -namespace gvisor { -namespace testing { - -// Opens the slave end of the passed master as R/W and nonblocking. -PosixErrorOr OpenSlave(const FileDescriptor& master); - -} // namespace testing -} // namespace gvisor - -#endif // GVISOR_TEST_UTIL_PTY_UTIL_H_ -- cgit v1.2.3 From fbbb2f7ed6a2f735c8e8d48e8b5264d2057e93ad Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Wed, 4 Sep 2019 19:06:44 -0700 Subject: Run proc_net tests. PiperOrigin-RevId: 267280086 --- test/syscalls/BUILD | 2 ++ test/syscalls/linux/BUILD | 1 + test/syscalls/linux/proc_net.cc | 3 +++ 3 files changed, 6 insertions(+) (limited to 'test/syscalls/linux/BUILD') diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 58eb1154a..0135435ea 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -305,6 +305,8 @@ syscall_test( syscall_test(test = "//test/syscalls/linux:proc_pid_uid_gid_map_test") +syscall_test(test = "//test/syscalls/linux:proc_net_test") + syscall_test( size = "medium", test = "//test/syscalls/linux:pselect_test", diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index aeb4b405d..a964bef24 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1521,6 +1521,7 @@ cc_binary( srcs = ["proc_net.cc"], linkstatic = 1, deps = [ + "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", "//test/util:test_main", diff --git a/test/syscalls/linux/proc_net.cc b/test/syscalls/linux/proc_net.cc index 03d0665eb..c097af196 100644 --- a/test/syscalls/linux/proc_net.cc +++ b/test/syscalls/linux/proc_net.cc @@ -14,6 +14,7 @@ #include "gtest/gtest.h" #include "gtest/gtest.h" +#include "test/util/capability_util.h" #include "test/util/file_descriptor.h" #include "test/util/fs_util.h" #include "test/util/test_util.h" @@ -35,6 +36,8 @@ TEST(ProcSysNetIpv4Sack, Exists) { } TEST(ProcSysNetIpv4Sack, CanReadAndWrite) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability((CAP_DAC_OVERRIDE)))); + auto const fd = ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/sys/net/ipv4/tcp_sack", O_RDWR)); -- cgit v1.2.3 From fbdd3ff1da122c1197d2018268b5956a526363d3 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 5 Sep 2019 16:35:23 -0700 Subject: Deflake aio_test. - Most AIO tests call io_setup(nr_events = 128). sizeof(struct io_event) (128*32 = 4096). However, the actual size of the mapping created by io_setup() is determined by: (from fs/aio.c:ioctx_alloc()) /* * We keep track of the number of available ringbuffer slots, to prevent * overflow (reqs_available), and we also use percpu counters for this. * * So since up to half the slots might be on other cpu's percpu counters * and unavailable, double nr_events so userspace sees what they * expected: additionally, we move req_batch slots to/from percpu * counters at a time, so make sure that isn't 0: */ nr_events = max(nr_events, num_possible_cpus() * 4); nr_events *= 2; (from fs/aio.c:aio_setup_ring()) /* Compensate for the ring buffer's head/tail overlap entry */ nr_events += 2; /* 1 is required, 2 for good luck */ size = sizeof(struct aio_ring); size += sizeof(struct io_event) * nr_events; nr_pages = PFN_UP(size); When we mremap() only the first page of a multi-page AIO ring buffer mapping, fs/aio.c:aio_ring_mremap() updates struct kioctx::mmap_base - but struct kioctx::mmap_size is untouched, so sys_io_destroy() => kill_ioctx() vm_unmaps() the mremapped page, plus some number of pages after it. Just get the actual size of the mapping from /proc/self/maps. - Delete test case MremapOver; while it is correct that Linux will not complain if you overwrite the AIO ring buffer with another mapping, it won't actually work in the sense that AIO events will not be written to the new mapping, because Linux stores the struct pages of the ring buffer in struct kioctx::ring_pages and writes to those through kmap() rather than using userspace addresses. - Don't munmap() after mremap(MREMAP_FIXED) returns EFAULT; see new comment in factored-out test case MremapExpansion. PiperOrigin-RevId: 267482903 --- test/syscalls/linux/BUILD | 5 +- test/syscalls/linux/aio.cc | 155 ++++++++++++++++++++---------------------- test/syscalls/linux/mremap.cc | 11 --- test/util/memory_util.h | 12 ++++ 4 files changed, 89 insertions(+), 94 deletions(-) (limited to 'test/syscalls/linux/BUILD') diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index a964bef24..c67ceddb6 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -265,12 +265,15 @@ cc_binary( ], linkstatic = 1, deps = [ - # The heap check doesn't handle mremap properly. + # The heapchecker doesn't recognize that io_destroy munmaps. "@com_google_googletest//:gtest", "@com_google_absl//absl/strings", "//test/util:cleanup", "//test/util:file_descriptor", + "//test/util:fs_util", + "//test/util:memory_util", "//test/util:posix_error", + "//test/util:proc_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", diff --git a/test/syscalls/linux/aio.cc b/test/syscalls/linux/aio.cc index 68dc05417..b27d4e10a 100644 --- a/test/syscalls/linux/aio.cc +++ b/test/syscalls/linux/aio.cc @@ -14,31 +14,57 @@ #include #include -#include #include #include #include #include +#include +#include + #include "gtest/gtest.h" #include "test/syscalls/linux/file_base.h" #include "test/util/cleanup.h" #include "test/util/file_descriptor.h" +#include "test/util/fs_util.h" +#include "test/util/memory_util.h" +#include "test/util/posix_error.h" +#include "test/util/proc_util.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" +using ::testing::_; + namespace gvisor { namespace testing { namespace { +// Returns the size of the VMA containing the given address. +PosixErrorOr VmaSizeAt(uintptr_t addr) { + ASSIGN_OR_RETURN_ERRNO(std::string proc_self_maps, + GetContents("/proc/self/maps")); + ASSIGN_OR_RETURN_ERRNO(auto entries, ParseProcMaps(proc_self_maps)); + // Use binary search to find the first VMA that might contain addr. + ProcMapsEntry target = {}; + target.end = addr; + auto it = + std::upper_bound(entries.begin(), entries.end(), target, + [](const ProcMapsEntry& x, const ProcMapsEntry& y) { + return x.end < y.end; + }); + // Check that it actually contains addr. + if (it == entries.end() || addr < it->start) { + return PosixError(ENOENT, absl::StrCat("no VMA contains address ", addr)); + } + return it->end - it->start; +} + constexpr char kData[] = "hello world!"; int SubmitCtx(aio_context_t ctx, long nr, struct iocb** iocbpp) { return syscall(__NR_io_submit, ctx, nr, iocbpp); } -} // namespace - class AIOTest : public FileTest { public: AIOTest() : ctx_(0) {} @@ -124,10 +150,10 @@ TEST_F(AIOTest, BasicWrite) { EXPECT_EQ(events[0].res, strlen(kData)); // Verify that the file contains the contents. - char verify_buf[32] = {}; - ASSERT_THAT(read(test_file_fd_.get(), &verify_buf[0], strlen(kData)), - SyscallSucceeds()); - EXPECT_EQ(strcmp(kData, &verify_buf[0]), 0); + char verify_buf[sizeof(kData)] = {}; + ASSERT_THAT(read(test_file_fd_.get(), verify_buf, sizeof(kData)), + SyscallSucceedsWithValue(strlen(kData))); + EXPECT_STREQ(verify_buf, kData); } TEST_F(AIOTest, BadWrite) { @@ -220,38 +246,25 @@ TEST_F(AIOTest, CloneVm) { TEST_F(AIOTest, Mremap) { // Setup a context that is 128 entries deep. ASSERT_THAT(SetupContext(128), SyscallSucceeds()); + const size_t ctx_size = + ASSERT_NO_ERRNO_AND_VALUE(VmaSizeAt(reinterpret_cast(ctx_))); struct iocb cb = CreateCallback(); struct iocb* cbs[1] = {&cb}; // Reserve address space for the mremap target so we have something safe to // map over. - // - // N.B. We reserve 2 pages because we'll attempt to remap to 2 pages below. - // That should fail with EFAULT, but will fail with EINVAL if this mmap - // returns the page immediately below ctx_, as - // [new_address, new_address+2*kPageSize) overlaps [ctx_, ctx_+kPageSize). - void* new_address = mmap(nullptr, 2 * kPageSize, PROT_READ, - MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); - ASSERT_THAT(reinterpret_cast(new_address), SyscallSucceeds()); - auto mmap_cleanup = Cleanup([new_address] { - EXPECT_THAT(munmap(new_address, 2 * kPageSize), SyscallSucceeds()); - }); - - // Test that remapping to a larger address fails. - void* res = mremap(reinterpret_cast(ctx_), kPageSize, 2 * kPageSize, - MREMAP_FIXED | MREMAP_MAYMOVE, new_address); - ASSERT_THAT(reinterpret_cast(res), SyscallFailsWithErrno(EFAULT)); + Mapping dst = + ASSERT_NO_ERRNO_AND_VALUE(MmapAnon(ctx_size, PROT_READ, MAP_PRIVATE)); // Remap context 'handle' to a different address. - res = mremap(reinterpret_cast(ctx_), kPageSize, kPageSize, - MREMAP_FIXED | MREMAP_MAYMOVE, new_address); - ASSERT_THAT( - reinterpret_cast(res), - SyscallSucceedsWithValue(reinterpret_cast(new_address))); - mmap_cleanup.Release(); + ASSERT_THAT(Mremap(reinterpret_cast(ctx_), ctx_size, dst.len(), + MREMAP_FIXED | MREMAP_MAYMOVE, dst.ptr()), + IsPosixErrorOkAndHolds(dst.ptr())); aio_context_t old_ctx = ctx_; - ctx_ = reinterpret_cast(new_address); + ctx_ = reinterpret_cast(dst.addr()); + // io_destroy() will unmap dst now. + dst.release(); // Check that submitting the request with the old 'ctx_' fails. ASSERT_THAT(SubmitCtx(old_ctx, 1, cbs), SyscallFailsWithErrno(EINVAL)); @@ -260,18 +273,12 @@ TEST_F(AIOTest, Mremap) { ASSERT_THAT(Submit(1, cbs), SyscallSucceedsWithValue(1)); // Remap again. - new_address = - mmap(nullptr, kPageSize, PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); - ASSERT_THAT(reinterpret_cast(new_address), SyscallSucceeds()); - auto mmap_cleanup2 = Cleanup([new_address] { - EXPECT_THAT(munmap(new_address, kPageSize), SyscallSucceeds()); - }); - res = mremap(reinterpret_cast(ctx_), kPageSize, kPageSize, - MREMAP_FIXED | MREMAP_MAYMOVE, new_address); - ASSERT_THAT(reinterpret_cast(res), - SyscallSucceedsWithValue(reinterpret_cast(new_address))); - mmap_cleanup2.Release(); - ctx_ = reinterpret_cast(new_address); + dst = ASSERT_NO_ERRNO_AND_VALUE(MmapAnon(ctx_size, PROT_READ, MAP_PRIVATE)); + ASSERT_THAT(Mremap(reinterpret_cast(ctx_), ctx_size, dst.len(), + MREMAP_FIXED | MREMAP_MAYMOVE, dst.ptr()), + IsPosixErrorOkAndHolds(dst.ptr())); + ctx_ = reinterpret_cast(dst.addr()); + dst.release(); // Get the reply with yet another 'ctx_' and verify it. struct io_event events[1]; @@ -281,51 +288,33 @@ TEST_F(AIOTest, Mremap) { EXPECT_EQ(events[0].res, strlen(kData)); // Verify that the file contains the contents. - char verify_buf[32] = {}; - ASSERT_THAT(read(test_file_fd_.get(), &verify_buf[0], strlen(kData)), - SyscallSucceeds()); - EXPECT_EQ(strcmp(kData, &verify_buf[0]), 0); + char verify_buf[sizeof(kData)] = {}; + ASSERT_THAT(read(test_file_fd_.get(), verify_buf, sizeof(kData)), + SyscallSucceedsWithValue(strlen(kData))); + EXPECT_STREQ(verify_buf, kData); } -// Tests that AIO context can be replaced with a different mapping at the same -// address and continue working. Don't ask why, but Linux allows it. -TEST_F(AIOTest, MremapOver) { +// Tests that AIO context cannot be expanded with mremap. +TEST_F(AIOTest, MremapExpansion) { // Setup a context that is 128 entries deep. ASSERT_THAT(SetupContext(128), SyscallSucceeds()); + const size_t ctx_size = + ASSERT_NO_ERRNO_AND_VALUE(VmaSizeAt(reinterpret_cast(ctx_))); - struct iocb cb = CreateCallback(); - struct iocb* cbs[1] = {&cb}; - - ASSERT_THAT(Submit(1, cbs), SyscallSucceedsWithValue(1)); - - // Allocate a new VMA, copy 'ctx_' content over, and remap it on top - // of 'ctx_'. - void* new_address = mmap(nullptr, kPageSize, PROT_READ | PROT_WRITE, - MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); - ASSERT_THAT(reinterpret_cast(new_address), SyscallSucceeds()); - auto mmap_cleanup = Cleanup([new_address] { - EXPECT_THAT(munmap(new_address, kPageSize), SyscallSucceeds()); - }); - - memcpy(new_address, reinterpret_cast(ctx_), kPageSize); - void* res = - mremap(new_address, kPageSize, kPageSize, MREMAP_FIXED | MREMAP_MAYMOVE, - reinterpret_cast(ctx_)); - ASSERT_THAT(reinterpret_cast(res), SyscallSucceedsWithValue(ctx_)); - mmap_cleanup.Release(); - - // Everything continues to work just fine. - struct io_event events[1]; - ASSERT_THAT(GetEvents(1, 1, events, nullptr), SyscallSucceedsWithValue(1)); - EXPECT_EQ(events[0].data, 0x123); - EXPECT_EQ(events[0].obj, reinterpret_cast(&cb)); - EXPECT_EQ(events[0].res, strlen(kData)); - - // Verify that the file contains the contents. - char verify_buf[32] = {}; - ASSERT_THAT(read(test_file_fd_.get(), &verify_buf[0], strlen(kData)), - SyscallSucceeds()); - EXPECT_EQ(strcmp(kData, &verify_buf[0]), 0); + // Reserve address space for the mremap target so we have something safe to + // map over. + Mapping dst = ASSERT_NO_ERRNO_AND_VALUE( + MmapAnon(ctx_size + kPageSize, PROT_NONE, MAP_PRIVATE)); + + // Test that remapping to a larger address range fails. + ASSERT_THAT(Mremap(reinterpret_cast(ctx_), ctx_size, dst.len(), + MREMAP_FIXED | MREMAP_MAYMOVE, dst.ptr()), + PosixErrorIs(EFAULT, _)); + + // mm/mremap.c:sys_mremap() => mremap_to() does do_munmap() of the destination + // before it hits the VM_DONTEXPAND check in vma_to_resize(), so we should no + // longer munmap it (another thread may have created a mapping there). + dst.release(); } // Tests that AIO calls fail if context's address is inaccessible. @@ -429,5 +418,7 @@ TEST_P(AIOVectorizedParamTest, BadIOVecs) { INSTANTIATE_TEST_SUITE_P(BadIOVecs, AIOVectorizedParamTest, ::testing::Values(IOCB_CMD_PREADV, IOCB_CMD_PWRITEV)); +} // namespace + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/mremap.cc b/test/syscalls/linux/mremap.cc index 64e435cb7..f0e5f7d82 100644 --- a/test/syscalls/linux/mremap.cc +++ b/test/syscalls/linux/mremap.cc @@ -35,17 +35,6 @@ namespace testing { namespace { -// Wrapper for mremap that returns a PosixErrorOr<>, since the return type of -// void* isn't directly compatible with SyscallSucceeds. -PosixErrorOr Mremap(void* old_address, size_t old_size, size_t new_size, - int flags, void* new_address) { - void* rv = mremap(old_address, old_size, new_size, flags, new_address); - if (rv == MAP_FAILED) { - return PosixError(errno, "mremap failed"); - } - return rv; -} - // Fixture for mremap tests parameterized by mmap flags. using MremapParamTest = ::testing::TestWithParam; diff --git a/test/util/memory_util.h b/test/util/memory_util.h index 190c469b5..e189b73e8 100644 --- a/test/util/memory_util.h +++ b/test/util/memory_util.h @@ -118,6 +118,18 @@ inline PosixErrorOr MmapAnon(size_t length, int prot, int flags) { return Mmap(nullptr, length, prot, flags | MAP_ANONYMOUS, -1, 0); } +// Wrapper for mremap that returns a PosixErrorOr<>, since the return type of +// void* isn't directly compatible with SyscallSucceeds. +inline PosixErrorOr Mremap(void* old_address, size_t old_size, + size_t new_size, int flags, + void* new_address) { + void* rv = mremap(old_address, old_size, new_size, flags, new_address); + if (rv == MAP_FAILED) { + return PosixError(errno, "mremap failed"); + } + return rv; +} + // Returns true if the page containing addr is mapped. inline bool IsMapped(uintptr_t addr) { int const rv = msync(reinterpret_cast(addr & ~(kPageSize - 1)), -- cgit v1.2.3 From 98f7fbb59fc5aca00f47a8145ce1227550869cb8 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Fri, 6 Sep 2019 11:27:44 -0700 Subject: Load C++ rules from @rules_cc See https://github.com/bazelbuild/bazel/issues/8743. This will be required in Bazel 1.0. Protobuf was updated in https://github.com/protocolbuffers/protobuf/commit/bf0c69e1302fe9568fbe310cc54b37d20a9d16a3#diff-96239ee297e0a92ac6ff96a6bc434ef0. GoogleTest was updated in https://github.com/google/googletest/commit/6fd262ecf787d0dc2a91696fd4bf1d3ee1ebfa14. gflags has not yet been updated, so the repo still won't build with --incompatible_load_cc_rules_from_bzl. Tested with buildifier -warnings=native-cc -lint=warn **/BUILD. PiperOrigin-RevId: 267638515 --- WORKSPACE | 21 +++++++++++---------- test/syscalls/linux/BUILD | 1 + test/util/BUILD | 1 + 3 files changed, 13 insertions(+), 10 deletions(-) (limited to 'test/syscalls/linux/BUILD') diff --git a/WORKSPACE b/WORKSPACE index c403e774d..26e76e2d7 100644 --- a/WORKSPACE +++ b/WORKSPACE @@ -44,13 +44,14 @@ http_archive( ) # Load protobuf dependencies. -load("@bazel_tools//tools/build_defs/repo:git.bzl", "git_repository") - -git_repository( +http_archive( name = "com_google_protobuf", - commit = "09745575a923640154bcf307fba8aedff47f240a", - remote = "https://github.com/protocolbuffers/protobuf", - shallow_since = "1558721209 -0700", + sha256 = "532d2575d8c0992065bb19ec5fba13aa3683499726f6055c11b474f91a00bb0c", + strip_prefix = "protobuf-7f520092d9050d96fb4b707ad11a51701af4ce49", + urls = [ + "https://mirror.bazel.build/github.com/protocolbuffers/protobuf/archive/7f520092d9050d96fb4b707ad11a51701af4ce49.zip", + "https://github.com/protocolbuffers/protobuf/archive/7f520092d9050d96fb4b707ad11a51701af4ce49.zip", + ], ) load("@com_google_protobuf//:protobuf_deps.bzl", "protobuf_deps") @@ -242,10 +243,10 @@ http_archive( http_archive( name = "com_google_googletest", - sha256 = "db657310d3c5ca2d3f674e3a4b79718d1d39da70604568ee0568ba8e39065ef4", - strip_prefix = "googletest-31200def0dec8a624c861f919e86e4444e6e6ee7", + sha256 = "0a10bea96d8670e5eef948d79d824162b1577bb7889539e49ec786bfc3e48912", + strip_prefix = "googletest-565f1b848215b77c3732bca345fe76a0431d8b34", urls = [ - "https://mirror.bazel.build/github.com/google/googletest/archive/31200def0dec8a624c861f919e86e4444e6e6ee7.tar.gz", - "https://github.com/google/googletest/archive/31200def0dec8a624c861f919e86e4444e6e6ee7.tar.gz", + "https://mirror.bazel.build/github.com/google/googletest/archive/565f1b848215b77c3732bca345fe76a0431d8b34.tar.gz", + "https://github.com/google/googletest/archive/565f1b848215b77c3732bca345fe76a0431d8b34.tar.gz", ], ) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index c67ceddb6..34057e3d0 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1,3 +1,4 @@ +load("@rules_cc//cc:defs.bzl", "cc_binary", "cc_library") load("//test/syscalls:build_defs.bzl", "select_for_linux") package( diff --git a/test/util/BUILD b/test/util/BUILD index 8afd89d8d..64daa0597 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -1,3 +1,4 @@ +load("@rules_cc//cc:defs.bzl", "cc_library", "cc_test") load("//test/syscalls:build_defs.bzl", "select_for_linux") package( -- cgit v1.2.3 From 7c6ab6a219f37a1d4c18ced4a602458fcf363f85 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Thu, 12 Sep 2019 17:42:14 -0700 Subject: Implement splice methods for pipes and sockets. This also allows the tee(2) implementation to be enabled, since dup can now be properly supported via WriteTo. Note that this change necessitated some minor restructoring with the fs.FileOperations splice methods. If the *fs.File is passed through directly, then only public API methods are accessible, which will deadlock immediately since the locking is already done by fs.Splice. Instead, we pass through an abstract io.Reader or io.Writer, which elide locks and use the underlying fs.FileOperations directly. PiperOrigin-RevId: 268805207 --- pkg/sentry/fs/file.go | 23 +++- pkg/sentry/fs/file_operations.go | 9 +- pkg/sentry/fs/file_overlay.go | 9 +- pkg/sentry/fs/fsutil/file.go | 6 +- pkg/sentry/fs/inotify.go | 5 +- pkg/sentry/fs/splice.go | 162 +++++++++++++------------- pkg/sentry/kernel/pipe/buffer.go | 25 ++++ pkg/sentry/kernel/pipe/pipe.go | 82 +++++++++++--- pkg/sentry/kernel/pipe/reader_writer.go | 76 ++++++++++++- pkg/sentry/socket/epsocket/epsocket.go | 134 +++++++++++++++++++--- pkg/sentry/syscalls/linux/linux64.go | 4 +- pkg/sentry/syscalls/linux/sys_splice.go | 86 +++++++------- pkg/tcpip/header/udp.go | 5 + pkg/tcpip/stack/transport_test.go | 4 +- pkg/tcpip/tcpip.go | 48 ++++---- pkg/tcpip/transport/icmp/endpoint.go | 4 +- pkg/tcpip/transport/raw/endpoint.go | 7 +- pkg/tcpip/transport/tcp/endpoint.go | 68 ++++++----- pkg/tcpip/transport/udp/endpoint.go | 14 +-- test/syscalls/linux/BUILD | 3 + test/syscalls/linux/pipe.cc | 14 +++ test/syscalls/linux/sendfile.cc | 69 ++++++++++++ test/syscalls/linux/splice.cc | 194 +++++++++++++++++++++++++------- 23 files changed, 770 insertions(+), 281 deletions(-) (limited to 'test/syscalls/linux/BUILD') diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index bb8117f89..c0a6e884b 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -515,6 +515,11 @@ type lockedReader struct { // File is the file to read from. File *File + + // Offset is the offset to start at. + // + // This applies only to Read, not ReadAt. + Offset int64 } // Read implements io.Reader.Read. @@ -522,7 +527,8 @@ func (r *lockedReader) Read(buf []byte) (int, error) { if r.Ctx.Interrupted() { return 0, syserror.ErrInterrupted } - n, err := r.File.FileOperations.Read(r.Ctx, r.File, usermem.BytesIOSequence(buf), r.File.offset) + n, err := r.File.FileOperations.Read(r.Ctx, r.File, usermem.BytesIOSequence(buf), r.Offset) + r.Offset += n return int(n), err } @@ -544,11 +550,21 @@ type lockedWriter struct { // File is the file to write to. File *File + + // Offset is the offset to start at. + // + // This applies only to Write, not WriteAt. + Offset int64 } // Write implements io.Writer.Write. func (w *lockedWriter) Write(buf []byte) (int, error) { - return w.WriteAt(buf, w.File.offset) + if w.Ctx.Interrupted() { + return 0, syserror.ErrInterrupted + } + n, err := w.WriteAt(buf, w.Offset) + w.Offset += int64(n) + return int(n), err } // WriteAt implements io.Writer.WriteAt. @@ -562,6 +578,9 @@ func (w *lockedWriter) WriteAt(buf []byte, offset int64) (int, error) { // io.Copy, since our own Write interface does not have this same // contract. Enforce that here. for written < len(buf) { + if w.Ctx.Interrupted() { + return written, syserror.ErrInterrupted + } var n int64 n, err = w.File.FileOperations.Write(w.Ctx, w.File, usermem.BytesIOSequence(buf[written:]), offset+int64(written)) if n > 0 { diff --git a/pkg/sentry/fs/file_operations.go b/pkg/sentry/fs/file_operations.go index d86f5bf45..b88303f17 100644 --- a/pkg/sentry/fs/file_operations.go +++ b/pkg/sentry/fs/file_operations.go @@ -15,6 +15,8 @@ package fs import ( + "io" + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -105,8 +107,11 @@ type FileOperations interface { // on the destination, following by a buffered copy with standard Read // and Write operations. // + // If dup is set, the data should be duplicated into the destination + // and retained. + // // The same preconditions as Read apply. - WriteTo(ctx context.Context, file *File, dst *File, opts SpliceOpts) (int64, error) + WriteTo(ctx context.Context, file *File, dst io.Writer, count int64, dup bool) (int64, error) // Write writes src to file at offset and returns the number of bytes // written which must be greater than or equal to 0. Like Read, file @@ -126,7 +131,7 @@ type FileOperations interface { // source. See WriteTo for details regarding how this is called. // // The same preconditions as Write apply; FileFlags.Write must be set. - ReadFrom(ctx context.Context, file *File, src *File, opts SpliceOpts) (int64, error) + ReadFrom(ctx context.Context, file *File, src io.Reader, count int64) (int64, error) // Fsync writes buffered modifications of file and/or flushes in-flight // operations to backing storage based on syncType. The range to sync is diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index 9820f0b13..225e40186 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -15,6 +15,7 @@ package fs import ( + "io" "sync" "gvisor.dev/gvisor/pkg/refs" @@ -268,9 +269,9 @@ func (f *overlayFileOperations) Read(ctx context.Context, file *File, dst userme } // WriteTo implements FileOperations.WriteTo. -func (f *overlayFileOperations) WriteTo(ctx context.Context, file *File, dst *File, opts SpliceOpts) (n int64, err error) { +func (f *overlayFileOperations) WriteTo(ctx context.Context, file *File, dst io.Writer, count int64, dup bool) (n int64, err error) { err = f.onTop(ctx, file, func(file *File, ops FileOperations) error { - n, err = ops.WriteTo(ctx, file, dst, opts) + n, err = ops.WriteTo(ctx, file, dst, count, dup) return err // Will overwrite itself. }) return @@ -285,9 +286,9 @@ func (f *overlayFileOperations) Write(ctx context.Context, file *File, src userm } // ReadFrom implements FileOperations.ReadFrom. -func (f *overlayFileOperations) ReadFrom(ctx context.Context, file *File, src *File, opts SpliceOpts) (n int64, err error) { +func (f *overlayFileOperations) ReadFrom(ctx context.Context, file *File, src io.Reader, count int64) (n int64, err error) { // See above; f.upper must be non-nil. - return f.upper.FileOperations.ReadFrom(ctx, f.upper, src, opts) + return f.upper.FileOperations.ReadFrom(ctx, f.upper, src, count) } // Fsync implements FileOperations.Fsync. diff --git a/pkg/sentry/fs/fsutil/file.go b/pkg/sentry/fs/fsutil/file.go index 626b9126a..fc5b3b1a1 100644 --- a/pkg/sentry/fs/fsutil/file.go +++ b/pkg/sentry/fs/fsutil/file.go @@ -15,6 +15,8 @@ package fsutil import ( + "io" + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -228,12 +230,12 @@ func (FileNoIoctl) Ioctl(context.Context, *fs.File, usermem.IO, arch.SyscallArgu type FileNoSplice struct{} // WriteTo implements fs.FileOperations.WriteTo. -func (FileNoSplice) WriteTo(context.Context, *fs.File, *fs.File, fs.SpliceOpts) (int64, error) { +func (FileNoSplice) WriteTo(context.Context, *fs.File, io.Writer, int64, bool) (int64, error) { return 0, syserror.ENOSYS } // ReadFrom implements fs.FileOperations.ReadFrom. -func (FileNoSplice) ReadFrom(context.Context, *fs.File, *fs.File, fs.SpliceOpts) (int64, error) { +func (FileNoSplice) ReadFrom(context.Context, *fs.File, io.Reader, int64) (int64, error) { return 0, syserror.ENOSYS } diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go index c7f4e2d13..ba3e0233d 100644 --- a/pkg/sentry/fs/inotify.go +++ b/pkg/sentry/fs/inotify.go @@ -15,6 +15,7 @@ package fs import ( + "io" "sync" "sync/atomic" @@ -172,7 +173,7 @@ func (i *Inotify) Read(ctx context.Context, _ *File, dst usermem.IOSequence, _ i } // WriteTo implements FileOperations.WriteTo. -func (*Inotify) WriteTo(context.Context, *File, *File, SpliceOpts) (int64, error) { +func (*Inotify) WriteTo(context.Context, *File, io.Writer, int64, bool) (int64, error) { return 0, syserror.ENOSYS } @@ -182,7 +183,7 @@ func (*Inotify) Fsync(context.Context, *File, int64, int64, SyncType) error { } // ReadFrom implements FileOperations.ReadFrom. -func (*Inotify) ReadFrom(context.Context, *File, *File, SpliceOpts) (int64, error) { +func (*Inotify) ReadFrom(context.Context, *File, io.Reader, int64) (int64, error) { return 0, syserror.ENOSYS } diff --git a/pkg/sentry/fs/splice.go b/pkg/sentry/fs/splice.go index eed1c2854..b03b7f836 100644 --- a/pkg/sentry/fs/splice.go +++ b/pkg/sentry/fs/splice.go @@ -18,7 +18,6 @@ import ( "io" "sync/atomic" - "gvisor.dev/gvisor/pkg/secio" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/syserror" ) @@ -33,146 +32,131 @@ func Splice(ctx context.Context, dst *File, src *File, opts SpliceOpts) (int64, } // Check whether or not the objects being sliced are stream-oriented - // (i.e. pipes or sockets). If yes, we elide checks and offset locks. - srcPipe := IsPipe(src.Dirent.Inode.StableAttr) || IsSocket(src.Dirent.Inode.StableAttr) - dstPipe := IsPipe(dst.Dirent.Inode.StableAttr) || IsSocket(dst.Dirent.Inode.StableAttr) + // (i.e. pipes or sockets). For all stream-oriented files and files + // where a specific offiset is not request, we acquire the file mutex. + // This has two important side effects. First, it provides the standard + // protection against concurrent writes that would mutate the offset. + // Second, it prevents Splice deadlocks. Only internal anonymous files + // implement the ReadFrom and WriteTo methods directly, and since such + // anonymous files are referred to by a unique fs.File object, we know + // that the file mutex takes strict precedence over internal locks. + // Since we enforce lock ordering here, we can't deadlock by using + // using a file in two different splice operations simultaneously. + srcPipe := !IsRegular(src.Dirent.Inode.StableAttr) + dstPipe := !IsRegular(dst.Dirent.Inode.StableAttr) + dstAppend := !dstPipe && dst.Flags().Append + srcLock := srcPipe || !opts.SrcOffset + dstLock := dstPipe || !opts.DstOffset || dstAppend - if !dstPipe && !opts.DstOffset && !srcPipe && !opts.SrcOffset { + switch { + case srcLock && dstLock: switch { case dst.UniqueID < src.UniqueID: // Acquire dst first. if !dst.mu.Lock(ctx) { return 0, syserror.ErrInterrupted } - defer dst.mu.Unlock() if !src.mu.Lock(ctx) { + dst.mu.Unlock() return 0, syserror.ErrInterrupted } - defer src.mu.Unlock() case dst.UniqueID > src.UniqueID: // Acquire src first. if !src.mu.Lock(ctx) { return 0, syserror.ErrInterrupted } - defer src.mu.Unlock() if !dst.mu.Lock(ctx) { + src.mu.Unlock() return 0, syserror.ErrInterrupted } - defer dst.mu.Unlock() case dst.UniqueID == src.UniqueID: // Acquire only one lock; it's the same file. This is a // bit of a edge case, but presumably it's possible. if !dst.mu.Lock(ctx) { return 0, syserror.ErrInterrupted } - defer dst.mu.Unlock() + srcLock = false // Only need one unlock. } // Use both offsets (locked). opts.DstStart = dst.offset opts.SrcStart = src.offset - } else if !dstPipe && !opts.DstOffset { + case dstLock: // Acquire only dst. if !dst.mu.Lock(ctx) { return 0, syserror.ErrInterrupted } - defer dst.mu.Unlock() opts.DstStart = dst.offset // Safe: locked. - } else if !srcPipe && !opts.SrcOffset { + case srcLock: // Acquire only src. if !src.mu.Lock(ctx) { return 0, syserror.ErrInterrupted } - defer src.mu.Unlock() opts.SrcStart = src.offset // Safe: locked. } - // Check append-only mode and the limit. - if !dstPipe { + var err error + if dstAppend { unlock := dst.Dirent.Inode.lockAppendMu(dst.Flags().Append) defer unlock() - if dst.Flags().Append { - if opts.DstOffset { - // We need to acquire the lock. - if !dst.mu.Lock(ctx) { - return 0, syserror.ErrInterrupted - } - defer dst.mu.Unlock() - } - // Figure out the appropriate offset to use. - if err := dst.offsetForAppend(ctx, &opts.DstStart); err != nil { - return 0, err - } - } + // Figure out the appropriate offset to use. + err = dst.offsetForAppend(ctx, &opts.DstStart) + } + if err == nil && !dstPipe { // Enforce file limits. limit, ok := dst.checkLimit(ctx, opts.DstStart) switch { case ok && limit == 0: - return 0, syserror.ErrExceedsFileSizeLimit + err = syserror.ErrExceedsFileSizeLimit case ok && limit < opts.Length: opts.Length = limit // Cap the write. } } + if err != nil { + if dstLock { + dst.mu.Unlock() + } + if srcLock { + src.mu.Unlock() + } + return 0, err + } - // Attempt to do a WriteTo; this is likely the most efficient. - // - // The underlying implementation may be able to donate buffers. - newOpts := SpliceOpts{ - Length: opts.Length, - SrcStart: opts.SrcStart, - SrcOffset: !srcPipe, - Dup: opts.Dup, - DstStart: opts.DstStart, - DstOffset: !dstPipe, + // Construct readers and writers for the splice. This is used to + // provide a safer locking path for the WriteTo/ReadFrom operations + // (since they will otherwise go through public interface methods which + // conflict with locking done above), and simplifies the fallback path. + w := &lockedWriter{ + Ctx: ctx, + File: dst, + Offset: opts.DstStart, } - n, err := src.FileOperations.WriteTo(ctx, src, dst, newOpts) - if n == 0 && err != nil { - // Attempt as a ReadFrom. If a WriteTo, a ReadFrom may also - // be more efficient than a copy if buffers are cached or readily - // available. (It's unlikely that they can actually be donate - n, err = dst.FileOperations.ReadFrom(ctx, dst, src, newOpts) + r := &lockedReader{ + Ctx: ctx, + File: src, + Offset: opts.SrcStart, } - if n == 0 && err != nil { - // If we've failed up to here, and at least one of the sources - // is a pipe or socket, then we can't properly support dup. - // Return an error indicating that this operation is not - // supported. - if (srcPipe || dstPipe) && newOpts.Dup { - return 0, syserror.EINVAL - } - // We failed to splice the files. But that's fine; we just fall - // back to a slow path in this case. This copies without doing - // any mode changes, so should still be more efficient. - var ( - r io.Reader - w io.Writer - ) - fw := &lockedWriter{ - Ctx: ctx, - File: dst, - } - if newOpts.DstOffset { - // Use the provided offset. - w = secio.NewOffsetWriter(fw, newOpts.DstStart) - } else { - // Writes will proceed with no offset. - w = fw - } - fr := &lockedReader{ - Ctx: ctx, - File: src, - } - if newOpts.SrcOffset { - // Limit to the given offset and length. - r = io.NewSectionReader(fr, opts.SrcStart, opts.Length) - } else { - // Limit just to the given length. - r = &io.LimitedReader{fr, opts.Length} - } + // Attempt to do a WriteTo; this is likely the most efficient. + n, err := src.FileOperations.WriteTo(ctx, src, w, opts.Length, opts.Dup) + if n == 0 && err != nil && err != syserror.ErrWouldBlock && !opts.Dup { + // Attempt as a ReadFrom. If a WriteTo, a ReadFrom may also be + // more efficient than a copy if buffers are cached or readily + // available. (It's unlikely that they can actually be donated). + n, err = dst.FileOperations.ReadFrom(ctx, dst, r, opts.Length) + } - // Copy between the two. - n, err = io.Copy(w, r) + // Support one last fallback option, but only if at least one of + // the source and destination are regular files. This is because + // if we block at some point, we could lose data. If the source is + // not a pipe then reading is not destructive; if the destination + // is a regular file, then it is guaranteed not to block writing. + if n == 0 && err != nil && err != syserror.ErrWouldBlock && !opts.Dup && (!dstPipe || !srcPipe) { + // Fallback to an in-kernel copy. + n, err = io.Copy(w, &io.LimitedReader{ + R: r, + N: opts.Length, + }) } // Update offsets, if required. @@ -185,5 +169,13 @@ func Splice(ctx context.Context, dst *File, src *File, opts SpliceOpts) (int64, } } + // Drop locks. + if dstLock { + dst.mu.Unlock() + } + if srcLock { + src.mu.Unlock() + } + return n, err } diff --git a/pkg/sentry/kernel/pipe/buffer.go b/pkg/sentry/kernel/pipe/buffer.go index 69ef2a720..95bee2d37 100644 --- a/pkg/sentry/kernel/pipe/buffer.go +++ b/pkg/sentry/kernel/pipe/buffer.go @@ -15,6 +15,7 @@ package pipe import ( + "io" "sync" "gvisor.dev/gvisor/pkg/sentry/safemem" @@ -67,6 +68,17 @@ func (b *buffer) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { return n, err } +// WriteFromReader writes to the buffer from an io.Reader. +func (b *buffer) WriteFromReader(r io.Reader, count int64) (int64, error) { + dst := b.data[b.write:] + if count < int64(len(dst)) { + dst = b.data[b.write:][:count] + } + n, err := r.Read(dst) + b.write += n + return int64(n), err +} + // ReadToBlocks implements safemem.Reader.ReadToBlocks. func (b *buffer) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { src := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(b.data[b.read:b.write])) @@ -75,6 +87,19 @@ func (b *buffer) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { return n, err } +// ReadToWriter reads from the buffer into an io.Writer. +func (b *buffer) ReadToWriter(w io.Writer, count int64, dup bool) (int64, error) { + src := b.data[b.read:b.write] + if count < int64(len(src)) { + src = b.data[b.read:][:count] + } + n, err := w.Write(src) + if !dup { + b.read += n + } + return int64(n), err +} + // bufferPool is a pool for buffers. var bufferPool = sync.Pool{ New: func() interface{} { diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 247e2928e..93b50669f 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -23,7 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) @@ -173,13 +172,24 @@ func (p *Pipe) Open(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) *fs.F } } +type readOps struct { + // left returns the bytes remaining. + left func() int64 + + // limit limits subsequence reads. + limit func(int64) + + // read performs the actual read operation. + read func(*buffer) (int64, error) +} + // read reads data from the pipe into dst and returns the number of bytes // read, or returns ErrWouldBlock if the pipe is empty. // // Precondition: this pipe must have readers. -func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error) { +func (p *Pipe) read(ctx context.Context, ops readOps) (int64, error) { // Don't block for a zero-length read even if the pipe is empty. - if dst.NumBytes() == 0 { + if ops.left() == 0 { return 0, nil } @@ -196,12 +206,12 @@ func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error) } // Limit how much we consume. - if dst.NumBytes() > p.size { - dst = dst.TakeFirst64(p.size) + if ops.left() > p.size { + ops.limit(p.size) } done := int64(0) - for dst.NumBytes() > 0 { + for ops.left() > 0 { // Pop the first buffer. first := p.data.Front() if first == nil { @@ -209,10 +219,9 @@ func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error) } // Copy user data. - n, err := dst.CopyOutFrom(ctx, first) + n, err := ops.read(first) done += int64(n) p.size -= n - dst = dst.DropFirst64(n) // Empty buffer? if first.Empty() { @@ -230,12 +239,57 @@ func (p *Pipe) read(ctx context.Context, dst usermem.IOSequence) (int64, error) return done, nil } +// dup duplicates all data from this pipe into the given writer. +// +// There is no blocking behavior implemented here. The writer may propagate +// some blocking error. All the writes must be complete writes. +func (p *Pipe) dup(ctx context.Context, ops readOps) (int64, error) { + p.mu.Lock() + defer p.mu.Unlock() + + // Is the pipe empty? + if p.size == 0 { + if !p.HasWriters() { + // See above. + return 0, nil + } + return 0, syserror.ErrWouldBlock + } + + // Limit how much we consume. + if ops.left() > p.size { + ops.limit(p.size) + } + + done := int64(0) + for buf := p.data.Front(); buf != nil; buf = buf.Next() { + n, err := ops.read(buf) + done += n + if err != nil { + return done, err + } + } + + return done, nil +} + +type writeOps struct { + // left returns the bytes remaining. + left func() int64 + + // limit should limit subsequent writes. + limit func(int64) + + // write should write to the provided buffer. + write func(*buffer) (int64, error) +} + // write writes data from sv into the pipe and returns the number of bytes // written. If no bytes are written because the pipe is full (or has less than // atomicIOBytes free capacity), write returns ErrWouldBlock. // // Precondition: this pipe must have writers. -func (p *Pipe) write(ctx context.Context, src usermem.IOSequence) (int64, error) { +func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) { p.mu.Lock() defer p.mu.Unlock() @@ -246,17 +300,16 @@ func (p *Pipe) write(ctx context.Context, src usermem.IOSequence) (int64, error) // POSIX requires that a write smaller than atomicIOBytes (PIPE_BUF) be // atomic, but requires no atomicity for writes larger than this. - wanted := src.NumBytes() + wanted := ops.left() if avail := p.max - p.size; wanted > avail { if wanted <= p.atomicIOBytes { return 0, syserror.ErrWouldBlock } - // Limit to the available capacity. - src = src.TakeFirst64(avail) + ops.limit(avail) } done := int64(0) - for src.NumBytes() > 0 { + for ops.left() > 0 { // Need a new buffer? last := p.data.Back() if last == nil || last.Full() { @@ -266,10 +319,9 @@ func (p *Pipe) write(ctx context.Context, src usermem.IOSequence) (int64, error) } // Copy user data. - n, err := src.CopyInTo(ctx, last) + n, err := ops.write(last) done += int64(n) p.size += n - src = src.DropFirst64(n) // Handle errors. if err != nil { diff --git a/pkg/sentry/kernel/pipe/reader_writer.go b/pkg/sentry/kernel/pipe/reader_writer.go index f69dbf27b..7c307f013 100644 --- a/pkg/sentry/kernel/pipe/reader_writer.go +++ b/pkg/sentry/kernel/pipe/reader_writer.go @@ -15,6 +15,7 @@ package pipe import ( + "io" "math" "syscall" @@ -55,7 +56,45 @@ func (rw *ReaderWriter) Release() { // Read implements fs.FileOperations.Read. func (rw *ReaderWriter) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) { - n, err := rw.Pipe.read(ctx, dst) + n, err := rw.Pipe.read(ctx, readOps{ + left: func() int64 { + return dst.NumBytes() + }, + limit: func(l int64) { + dst = dst.TakeFirst64(l) + }, + read: func(buf *buffer) (int64, error) { + n, err := dst.CopyOutFrom(ctx, buf) + dst = dst.DropFirst64(n) + return n, err + }, + }) + if n > 0 { + rw.Pipe.Notify(waiter.EventOut) + } + return n, err +} + +// WriteTo implements fs.FileOperations.WriteTo. +func (rw *ReaderWriter) WriteTo(ctx context.Context, _ *fs.File, w io.Writer, count int64, dup bool) (int64, error) { + ops := readOps{ + left: func() int64 { + return count + }, + limit: func(l int64) { + count = l + }, + read: func(buf *buffer) (int64, error) { + n, err := buf.ReadToWriter(w, count, dup) + count -= n + return n, err + }, + } + if dup { + // There is no notification for dup operations. + return rw.Pipe.dup(ctx, ops) + } + n, err := rw.Pipe.read(ctx, ops) if n > 0 { rw.Pipe.Notify(waiter.EventOut) } @@ -64,7 +103,40 @@ func (rw *ReaderWriter) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequ // Write implements fs.FileOperations.Write. func (rw *ReaderWriter) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) { - n, err := rw.Pipe.write(ctx, src) + n, err := rw.Pipe.write(ctx, writeOps{ + left: func() int64 { + return src.NumBytes() + }, + limit: func(l int64) { + src = src.TakeFirst64(l) + }, + write: func(buf *buffer) (int64, error) { + n, err := src.CopyInTo(ctx, buf) + src = src.DropFirst64(n) + return n, err + }, + }) + if n > 0 { + rw.Pipe.Notify(waiter.EventIn) + } + return n, err +} + +// ReadFrom implements fs.FileOperations.WriteTo. +func (rw *ReaderWriter) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader, count int64) (int64, error) { + n, err := rw.Pipe.write(ctx, writeOps{ + left: func() int64 { + return count + }, + limit: func(l int64) { + count = l + }, + write: func(buf *buffer) (int64, error) { + n, err := buf.WriteFromReader(r, count) + count -= n + return n, err + }, + }) if n > 0 { rw.Pipe.Notify(waiter.EventIn) } diff --git a/pkg/sentry/socket/epsocket/epsocket.go b/pkg/sentry/socket/epsocket/epsocket.go index 0e37ce61b..3e05e40fe 100644 --- a/pkg/sentry/socket/epsocket/epsocket.go +++ b/pkg/sentry/socket/epsocket/epsocket.go @@ -26,6 +26,7 @@ package epsocket import ( "bytes" + "io" "math" "reflect" "sync" @@ -227,7 +228,6 @@ type SocketOperations struct { fsutil.FileNoopFlush `state:"nosave"` fsutil.FileNoFsync `state:"nosave"` fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` fsutil.FileUseInodeUnstableAttr `state:"nosave"` socket.SendReceiveTimeout *waiter.Queue @@ -412,17 +412,58 @@ func (s *SocketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS return int64(n), nil } -// ioSequencePayload implements tcpip.Payload. It copies user memory bytes on demand -// based on the requested size. +// WriteTo implements fs.FileOperations.WriteTo. +func (s *SocketOperations) WriteTo(ctx context.Context, _ *fs.File, dst io.Writer, count int64, dup bool) (int64, error) { + s.readMu.Lock() + defer s.readMu.Unlock() + + // Copy as much data as possible. + done := int64(0) + for count > 0 { + // This may return a blocking error. + if err := s.fetchReadView(); err != nil { + return done, err.ToError() + } + + // Write to the underlying file. + n, err := dst.Write(s.readView) + done += int64(n) + count -= int64(n) + if dup { + // That's all we support for dup. This is generally + // supported by any Linux system calls, but the + // expectation is that now a caller will call read to + // actually remove these bytes from the socket. + return done, nil + } + + // Drop that part of the view. + s.readView.TrimFront(n) + if err != nil { + return done, err + } + } + + return done, nil +} + +// ioSequencePayload implements tcpip.Payload. +// +// t copies user memory bytes on demand based on the requested size. type ioSequencePayload struct { ctx context.Context src usermem.IOSequence } -// Get implements tcpip.Payload. -func (i *ioSequencePayload) Get(size int) ([]byte, *tcpip.Error) { - if size > i.Size() { - size = i.Size() +// FullPayload implements tcpip.Payloader.FullPayload +func (i *ioSequencePayload) FullPayload() ([]byte, *tcpip.Error) { + return i.Payload(int(i.src.NumBytes())) +} + +// Payload implements tcpip.Payloader.Payload. +func (i *ioSequencePayload) Payload(size int) ([]byte, *tcpip.Error) { + if max := int(i.src.NumBytes()); size > max { + size = max } v := buffer.NewView(size) if _, err := i.src.CopyIn(i.ctx, v); err != nil { @@ -431,11 +472,6 @@ func (i *ioSequencePayload) Get(size int) ([]byte, *tcpip.Error) { return v, nil } -// Size implements tcpip.Payload. -func (i *ioSequencePayload) Size() int { - return int(i.src.NumBytes()) -} - // DropFirst drops the first n bytes from underlying src. func (i *ioSequencePayload) DropFirst(n int) { i.src = i.src.DropFirst(int(n)) @@ -469,6 +505,76 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO return int64(n), nil } +// readerPayload implements tcpip.Payloader. +// +// It allocates a view and reads from a reader on-demand, based on available +// capacity in the endpoint. +type readerPayload struct { + ctx context.Context + r io.Reader + count int64 + err error +} + +// FullPayload implements tcpip.Payloader.FullPayload. +func (r *readerPayload) FullPayload() ([]byte, *tcpip.Error) { + return r.Payload(int(r.count)) +} + +// Payload implements tcpip.Payloader.Payload. +func (r *readerPayload) Payload(size int) ([]byte, *tcpip.Error) { + if size > int(r.count) { + size = int(r.count) + } + v := buffer.NewView(size) + n, err := r.r.Read(v) + if n > 0 { + // We ignore the error here. It may re-occur on subsequent + // reads, but for now we can enqueue some amount of data. + r.count -= int64(n) + return v[:n], nil + } + if err == syserror.ErrWouldBlock { + return nil, tcpip.ErrWouldBlock + } else if err != nil { + r.err = err // Save for propation. + return nil, tcpip.ErrBadAddress + } + + // There is no data and no error. Return an error, which will propagate + // r.err, which will be nil. This is the desired result: (0, nil). + return nil, tcpip.ErrBadAddress +} + +// ReadFrom implements fs.FileOperations.ReadFrom. +func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader, count int64) (int64, error) { + f := &readerPayload{ctx: ctx, r: r, count: count} + n, resCh, err := s.Endpoint.Write(f, tcpip.WriteOptions{}) + if err == tcpip.ErrWouldBlock { + return 0, syserror.ErrWouldBlock + } + + if resCh != nil { + t := ctx.(*kernel.Task) + if err := t.Block(resCh); err != nil { + return 0, syserr.FromError(err).ToError() + } + + n, _, err = s.Endpoint.Write(f, tcpip.WriteOptions{ + // Reads may be destructive but should be very fast, + // so we can't release the lock while copying data. + Atomic: true, + }) + } + if err == tcpip.ErrWouldBlock { + return n, syserror.ErrWouldBlock + } else if err != nil { + return int64(n), f.err // Propagate error. + } + + return int64(n), nil +} + // Readiness returns a mask of ready events for socket s. func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { r := s.Endpoint.Readiness(mask) @@ -2060,7 +2166,7 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] n, _, err = s.Endpoint.Write(v, opts) } dontWait := flags&linux.MSG_DONTWAIT != 0 - if err == nil && (n >= int64(v.Size()) || dontWait) { + if err == nil && (n >= v.src.NumBytes() || dontWait) { // Complete write. return int(n), nil } @@ -2085,7 +2191,7 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return 0, syserr.TranslateNetstackError(err) } - if err == nil && v.Size() == 0 || err != nil && err != tcpip.ErrWouldBlock { + if err == nil && v.src.NumBytes() == 0 || err != nil && err != tcpip.ErrWouldBlock { return int(total), nil } diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index ed996ba51..150999fb8 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -320,8 +320,8 @@ var AMD64 = &kernel.SyscallTable{ 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 275: syscalls.PartiallySupported("splice", Splice, "Stub implementation.", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 276: syscalls.ErrorWithEvent("tee", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 275: syscalls.Supported("splice", Splice), + 276: syscalls.Supported("tee", Tee), 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), 278: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index 8a98fedcb..f0a292f2f 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -29,9 +29,8 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB total int64 n int64 err error - ch chan struct{} - inW bool - outW bool + inCh chan struct{} + outCh chan struct{} ) for opts.Length > 0 { n, err = fs.Splice(t, outFile, inFile, opts) @@ -43,35 +42,33 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB break } - // Are we a registered waiter? - if ch == nil { - ch = make(chan struct{}, 1) - } - if !inW && !inFile.Flags().NonBlocking { - w, _ := waiter.NewChannelEntry(ch) - inFile.EventRegister(&w, EventMaskRead) - defer inFile.EventUnregister(&w) - inW = true // Registered. - } else if !outW && !outFile.Flags().NonBlocking { - w, _ := waiter.NewChannelEntry(ch) - outFile.EventRegister(&w, EventMaskWrite) - defer outFile.EventUnregister(&w) - outW = true // Registered. - } - - // Was anything registered? If no, everything is non-blocking. - if !inW && !outW { - break - } - - if (!inW || inFile.Readiness(EventMaskRead) != 0) && (!outW || outFile.Readiness(EventMaskWrite) != 0) { - // Something became ready, try again without blocking. - continue + // Note that the blocking behavior here is a bit different than the + // normal pattern. Because we need to have both data to read and data + // to write simultaneously, we actually explicitly block on both of + // these cases in turn before returning to the splice operation. + if inFile.Readiness(EventMaskRead) == 0 { + if inCh == nil { + inCh = make(chan struct{}, 1) + inW, _ := waiter.NewChannelEntry(inCh) + inFile.EventRegister(&inW, EventMaskRead) + defer inFile.EventUnregister(&inW) + continue // Need to refresh readiness. + } + if err = t.Block(inCh); err != nil { + break + } } - - // Block until there's data. - if err = t.Block(ch); err != nil { - break + if outFile.Readiness(EventMaskWrite) == 0 { + if outCh == nil { + outCh = make(chan struct{}, 1) + outW, _ := waiter.NewChannelEntry(outCh) + outFile.EventRegister(&outW, EventMaskWrite) + defer outFile.EventUnregister(&outW) + continue // Need to refresh readiness. + } + if err = t.Block(outCh); err != nil { + break + } } } @@ -149,7 +146,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc Length: count, SrcOffset: true, SrcStart: offset, - }, false) + }, outFile.Flags().NonBlocking) // Copy out the new offset. if _, err := t.CopyOut(offsetAddr, n+offset); err != nil { @@ -159,7 +156,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Send data using splice. n, err = doSplice(t, outFile, inFile, fs.SpliceOpts{ Length: count, - }, false) + }, outFile.Flags().NonBlocking) } // We can only pass a single file to handleIOError, so pick inFile @@ -181,12 +178,6 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, syserror.EINVAL } - // Only non-blocking is meaningful. Note that unlike in Linux, this - // flag is applied consistently. We will have either fully blocking or - // non-blocking behavior below, regardless of the underlying files - // being spliced to. It's unclear if this is a bug or not yet. - nonBlocking := (flags & linux.SPLICE_F_NONBLOCK) != 0 - // Get files. outFile := t.GetFile(outFD) if outFile == nil { @@ -200,6 +191,13 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } defer inFile.DecRef() + // The operation is non-blocking if anything is non-blocking. + // + // N.B. This is a rather simplistic heuristic that avoids some + // poor edge case behavior since the exact semantics here are + // underspecified and vary between versions of Linux itself. + nonBlock := inFile.Flags().NonBlocking || outFile.Flags().NonBlocking || (flags&linux.SPLICE_F_NONBLOCK != 0) + // Construct our options. // // Note that exactly one of the underlying buffers must be a pipe. We @@ -257,7 +255,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } // Splice data. - n, err := doSplice(t, outFile, inFile, opts, nonBlocking) + n, err := doSplice(t, outFile, inFile, opts, nonBlock) // See above; inFile is chosen arbitrarily here. return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "splice", inFile) @@ -275,9 +273,6 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo return 0, nil, syserror.EINVAL } - // Only non-blocking is meaningful. - nonBlocking := (flags & linux.SPLICE_F_NONBLOCK) != 0 - // Get files. outFile := t.GetFile(outFD) if outFile == nil { @@ -301,11 +296,14 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo return 0, nil, syserror.EINVAL } + // The operation is non-blocking if anything is non-blocking. + nonBlock := inFile.Flags().NonBlocking || outFile.Flags().NonBlocking || (flags&linux.SPLICE_F_NONBLOCK != 0) + // Splice data. n, err := doSplice(t, outFile, inFile, fs.SpliceOpts{ Length: count, Dup: true, - }, nonBlocking) + }, nonBlock) // See above; inFile is chosen arbitrarily here. return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "tee", inFile) diff --git a/pkg/tcpip/header/udp.go b/pkg/tcpip/header/udp.go index c1f454805..74412c894 100644 --- a/pkg/tcpip/header/udp.go +++ b/pkg/tcpip/header/udp.go @@ -27,6 +27,11 @@ const ( udpChecksum = 6 ) +const ( + // UDPMaximumPacketSize is the largest possible UDP packet. + UDPMaximumPacketSize = 0xffff +) + // UDPFields contains the fields of a UDP packet. It is used to describe the // fields of a packet that needs to be encoded. type UDPFields struct { diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 87d1e0d0d..847d02982 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -65,13 +65,13 @@ func (*fakeTransportEndpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.Contr return buffer.View{}, tcpip.ControlMessages{}, nil } -func (f *fakeTransportEndpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { if len(f.route.RemoteAddress) == 0 { return 0, nil, tcpip.ErrNoRoute } hdr := buffer.NewPrependable(int(f.route.MaxHeaderLength())) - v, err := p.Get(p.Size()) + v, err := p.FullPayload() if err != nil { return 0, nil, err } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index ebf8a2d04..2534069ab 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -261,31 +261,34 @@ type FullAddress struct { Port uint16 } -// Payload provides an interface around data that is being sent to an endpoint. -// This allows the endpoint to request the amount of data it needs based on -// internal buffers without exposing them. 'p.Get(p.Size())' reads all the data. -type Payload interface { - // Get returns a slice containing exactly 'min(size, p.Size())' bytes. - Get(size int) ([]byte, *Error) - - // Size returns the payload size. - Size() int +// Payloader is an interface that provides data. +// +// This interface allows the endpoint to request the amount of data it needs +// based on internal buffers without exposing them. +type Payloader interface { + // FullPayload returns all available bytes. + FullPayload() ([]byte, *Error) + + // Payload returns a slice containing at most size bytes. + Payload(size int) ([]byte, *Error) } -// SlicePayload implements Payload on top of slices for convenience. +// SlicePayload implements Payloader for slices. +// +// This is typically used for tests. type SlicePayload []byte -// Get implements Payload. -func (s SlicePayload) Get(size int) ([]byte, *Error) { - if size > s.Size() { - size = s.Size() - } - return s[:size], nil +// FullPayload implements Payloader.FullPayload. +func (s SlicePayload) FullPayload() ([]byte, *Error) { + return s, nil } -// Size implements Payload. -func (s SlicePayload) Size() int { - return len(s) +// Payload implements Payloader.Payload. +func (s SlicePayload) Payload(size int) ([]byte, *Error) { + if size > len(s) { + size = len(s) + } + return s[:size], nil } // A ControlMessages contains socket control messages for IP sockets. @@ -338,7 +341,7 @@ type Endpoint interface { // ErrNoLinkAddress and a notification channel is returned for the caller to // block. Channel is closed once address resolution is complete (success or // not). The channel is only non-nil in this case. - Write(Payload, WriteOptions) (int64, <-chan struct{}, *Error) + Write(Payloader, WriteOptions) (int64, <-chan struct{}, *Error) // Peek reads data without consuming it from the endpoint. // @@ -432,6 +435,11 @@ type WriteOptions struct { // EndOfRecord has the same semantics as Linux's MSG_EOR. EndOfRecord bool + + // Atomic means that all data fetched from Payloader must be written to the + // endpoint. If Atomic is false, then data fetched from the Payloader may be + // discarded if available endpoint buffer space is unsufficient. + Atomic bool } // SockOpt represents socket options which values have the int type. diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index e1f622af6..3db060384 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -204,7 +204,7 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // Write writes data to the endpoint's peer. This method does not block // if the data cannot be written. -func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. (This also means that MSG_EOR is a no-op.) if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue @@ -289,7 +289,7 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha } } - v, err := p.Get(p.Size()) + v, err := p.FullPayload() if err != nil { return 0, nil, err } diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 13e17e2a6..cf1c5c433 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -207,7 +207,7 @@ func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMes } // Write implements tcpip.Endpoint.Write. -func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. This also means that MSG_EOR is a no-op. if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue @@ -220,9 +220,8 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (int64 return 0, nil, tcpip.ErrInvalidEndpointState } - payloadBytes, err := payload.Get(payload.Size()) + payloadBytes, err := p.FullPayload() if err != nil { - ep.mu.RUnlock() return 0, nil, err } @@ -230,7 +229,7 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (int64 // destination address, route using that address. if !ep.associated { ip := header.IPv4(payloadBytes) - if !ip.IsValid(payload.Size()) { + if !ip.IsValid(len(payloadBytes)) { ep.mu.RUnlock() return 0, nil, tcpip.ErrInvalidOptionValue } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index ac927569a..dd931f88c 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -806,7 +806,7 @@ func (e *endpoint) isEndpointWritableLocked() (int, *tcpip.Error) { } // Write writes data to the endpoint's peer. -func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // Linux completely ignores any address passed to sendto(2) for TCP sockets // (without the MSG_FASTOPEN flag). Corking is unimplemented, so opts.More // and opts.EndOfRecord are also ignored. @@ -821,47 +821,52 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha return 0, nil, err } - e.sndBufMu.Unlock() - e.mu.RUnlock() - - // Nothing to do if the buffer is empty. - if p.Size() == 0 { - return 0, nil, nil + // We can release locks while copying data. + // + // This is not possible if atomic is set, because we can't allow the + // available buffer space to be consumed by some other caller while we + // are copying data in. + if !opts.Atomic { + e.sndBufMu.Unlock() + e.mu.RUnlock() } - // Copy in memory without holding sndBufMu so that worker goroutine can - // make progress independent of this operation. - v, perr := p.Get(avail) - if perr != nil { + // Fetch data. + v, perr := p.Payload(avail) + if perr != nil || len(v) == 0 { + if opts.Atomic { // See above. + e.sndBufMu.Unlock() + e.mu.RUnlock() + } + // Note that perr may be nil if len(v) == 0. return 0, nil, perr } - e.mu.RLock() - e.sndBufMu.Lock() + if !opts.Atomic { // See above. + e.mu.RLock() + e.sndBufMu.Lock() - // Because we released the lock before copying, check state again - // to make sure the endpoint is still in a valid state for a - // write. - avail, err = e.isEndpointWritableLocked() - if err != nil { - e.sndBufMu.Unlock() - e.mu.RUnlock() - return 0, nil, err - } + // Because we released the lock before copying, check state again + // to make sure the endpoint is still in a valid state for a write. + avail, err = e.isEndpointWritableLocked() + if err != nil { + e.sndBufMu.Unlock() + e.mu.RUnlock() + return 0, nil, err + } - // Discard any excess data copied in due to avail being reduced due to a - // simultaneous write call to the socket. - if avail < len(v) { - v = v[:avail] + // Discard any excess data copied in due to avail being reduced due + // to a simultaneous write call to the socket. + if avail < len(v) { + v = v[:avail] + } } // Add data to the send queue. - l := len(v) s := newSegmentFromView(&e.route, e.id, v) - e.sndBufUsed += l - e.sndBufInQueue += seqnum.Size(l) + e.sndBufUsed += len(v) + e.sndBufInQueue += seqnum.Size(len(v)) e.sndQueue.PushBack(s) - e.sndBufMu.Unlock() // Release the endpoint lock to prevent deadlocks due to lock // order inversion when acquiring workMu. @@ -875,7 +880,8 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha // Let the protocol goroutine do the work. e.sndWaker.Assert() } - return int64(l), nil, nil + + return int64(len(v)), nil, nil } // Peek reads data without consuming it from the endpoint. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index dccb9a7eb..6ac7c067a 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -15,7 +15,6 @@ package udp import ( - "math" "sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -277,17 +276,12 @@ func (e *endpoint) connectRoute(nicid tcpip.NICID, addr tcpip.FullAddress, netPr // Write writes data to the endpoint's peer. This method does not block // if the data cannot be written. -func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. (This also means that MSG_EOR is a no-op.) if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue } - if p.Size() > math.MaxUint16 { - // Payload can't possibly fit in a packet. - return 0, nil, tcpip.ErrMessageTooLong - } - to := opts.To e.mu.RLock() @@ -370,10 +364,14 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha } } - v, err := p.Get(p.Size()) + v, err := p.FullPayload() if err != nil { return 0, nil, err } + if len(v) > header.UDPMaximumPacketSize { + // Payload can't possibly fit in a packet. + return 0, nil, tcpip.ErrMessageTooLong + } ttl := route.DefaultTTL() if header.IsV4MulticastAddress(route.RemoteAddress) || header.IsV6MulticastAddress(route.RemoteAddress) { diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 34057e3d0..df00d2c14 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1867,7 +1867,9 @@ cc_binary( "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", + "//test/util:thread_util", "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], ) @@ -1901,6 +1903,7 @@ cc_binary( "//test/util:test_util", "//test/util:thread_util", "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], ) diff --git a/test/syscalls/linux/pipe.cc b/test/syscalls/linux/pipe.cc index 65afb90f3..10e2a6dfc 100644 --- a/test/syscalls/linux/pipe.cc +++ b/test/syscalls/linux/pipe.cc @@ -168,6 +168,20 @@ TEST_P(PipeTest, Write) { EXPECT_EQ(wbuf, rbuf); } +TEST_P(PipeTest, WritePage) { + SKIP_IF(!CreateBlocking()); + + std::vector wbuf(kPageSize); + RandomizeBuffer(wbuf.data(), wbuf.size()); + std::vector rbuf(wbuf.size()); + + ASSERT_THAT(write(wfd_.get(), wbuf.data(), wbuf.size()), + SyscallSucceedsWithValue(wbuf.size())); + ASSERT_THAT(read(rfd_.get(), rbuf.data(), rbuf.size()), + SyscallSucceedsWithValue(rbuf.size())); + EXPECT_EQ(memcmp(rbuf.data(), wbuf.data(), wbuf.size()), 0); +} + TEST_P(PipeTest, NonBlocking) { SKIP_IF(!CreateNonBlocking()); diff --git a/test/syscalls/linux/sendfile.cc b/test/syscalls/linux/sendfile.cc index 9167ab066..4502e7fb4 100644 --- a/test/syscalls/linux/sendfile.cc +++ b/test/syscalls/linux/sendfile.cc @@ -19,9 +19,12 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" #include "absl/strings/string_view.h" +#include "absl/time/clock.h" +#include "absl/time/time.h" #include "test/util/file_descriptor.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" +#include "test/util/thread_util.h" namespace gvisor { namespace testing { @@ -442,6 +445,72 @@ TEST(SendFileTest, SendToNotARegularFile) { EXPECT_THAT(sendfile(outf.get(), inf.get(), nullptr, 0), SyscallFailsWithErrno(EINVAL)); } + +TEST(SendFileTest, SendPipeWouldBlock) { + // Create temp file. + constexpr char kData[] = + "The fool doth think he is wise, but the wise man knows himself to be a " + "fool."; + constexpr int kDataSize = sizeof(kData) - 1; + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + GetAbsoluteTestTmpdir(), kData, TempPath::kDefaultFileMode)); + + // Open the input file as read only. + const FileDescriptor inf = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDONLY)); + + // Setup the output named pipe. + int fds[2]; + ASSERT_THAT(pipe2(fds, O_NONBLOCK), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Fill up the pipe's buffer. + int pipe_size = -1; + ASSERT_THAT(pipe_size = fcntl(wfd.get(), F_GETPIPE_SZ), SyscallSucceeds()); + std::vector buf(2 * pipe_size); + ASSERT_THAT(write(wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(pipe_size)); + + EXPECT_THAT(sendfile(wfd.get(), inf.get(), nullptr, kDataSize), + SyscallFailsWithErrno(EWOULDBLOCK)); +} + +TEST(SendFileTest, SendPipeBlocks) { + // Create temp file. + constexpr char kData[] = + "The fault, dear Brutus, is not in our stars, but in ourselves."; + constexpr int kDataSize = sizeof(kData) - 1; + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + GetAbsoluteTestTmpdir(), kData, TempPath::kDefaultFileMode)); + + // Open the input file as read only. + const FileDescriptor inf = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDONLY)); + + // Setup the output named pipe. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Fill up the pipe's buffer. + int pipe_size = -1; + ASSERT_THAT(pipe_size = fcntl(wfd.get(), F_GETPIPE_SZ), SyscallSucceeds()); + std::vector buf(pipe_size); + ASSERT_THAT(write(wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(pipe_size)); + + ScopedThread t([&]() { + absl::SleepFor(absl::Milliseconds(100)); + ASSERT_THAT(read(rfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(pipe_size)); + }); + + EXPECT_THAT(sendfile(wfd.get(), inf.get(), nullptr, kDataSize), + SyscallSucceedsWithValue(kDataSize)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/splice.cc b/test/syscalls/linux/splice.cc index e25f264f6..85232cb1f 100644 --- a/test/syscalls/linux/splice.cc +++ b/test/syscalls/linux/splice.cc @@ -14,12 +14,16 @@ #include #include +#include #include +#include #include #include "gmock/gmock.h" #include "gtest/gtest.h" #include "absl/strings/string_view.h" +#include "absl/time/clock.h" +#include "absl/time/time.h" #include "test/util/file_descriptor.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -36,23 +40,23 @@ TEST(SpliceTest, TwoRegularFiles) { const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); // Open the input file as read only. - const FileDescriptor inf = + const FileDescriptor in_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDONLY)); // Open the output file as write only. - const FileDescriptor outf = + const FileDescriptor out_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(out_file.path(), O_WRONLY)); // Verify that it is rejected as expected; regardless of offsets. loff_t in_offset = 0; loff_t out_offset = 0; - EXPECT_THAT(splice(inf.get(), &in_offset, outf.get(), &out_offset, 1, 0), + EXPECT_THAT(splice(in_fd.get(), &in_offset, out_fd.get(), &out_offset, 1, 0), SyscallFailsWithErrno(EINVAL)); - EXPECT_THAT(splice(inf.get(), nullptr, outf.get(), &out_offset, 1, 0), + EXPECT_THAT(splice(in_fd.get(), nullptr, out_fd.get(), &out_offset, 1, 0), SyscallFailsWithErrno(EINVAL)); - EXPECT_THAT(splice(inf.get(), &in_offset, outf.get(), nullptr, 1, 0), + EXPECT_THAT(splice(in_fd.get(), &in_offset, out_fd.get(), nullptr, 1, 0), SyscallFailsWithErrno(EINVAL)); - EXPECT_THAT(splice(inf.get(), nullptr, outf.get(), nullptr, 1, 0), + EXPECT_THAT(splice(in_fd.get(), nullptr, out_fd.get(), nullptr, 1, 0), SyscallFailsWithErrno(EINVAL)); } @@ -75,8 +79,6 @@ TEST(SpliceTest, SamePipe) { } TEST(TeeTest, SamePipe) { - SKIP_IF(IsRunningOnGvisor()); - // Create a new pipe. int fds[2]; ASSERT_THAT(pipe(fds), SyscallSucceeds()); @@ -95,11 +97,9 @@ TEST(TeeTest, SamePipe) { } TEST(TeeTest, RegularFile) { - SKIP_IF(IsRunningOnGvisor()); - // Open some file. const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); - const FileDescriptor inf = + const FileDescriptor in_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); // Create a new pipe. @@ -109,9 +109,9 @@ TEST(TeeTest, RegularFile) { const FileDescriptor wfd(fds[1]); // Attempt to tee from the file. - EXPECT_THAT(tee(inf.get(), wfd.get(), kPageSize, 0), + EXPECT_THAT(tee(in_fd.get(), wfd.get(), kPageSize, 0), SyscallFailsWithErrno(EINVAL)); - EXPECT_THAT(tee(rfd.get(), inf.get(), kPageSize, 0), + EXPECT_THAT(tee(rfd.get(), in_fd.get(), kPageSize, 0), SyscallFailsWithErrno(EINVAL)); } @@ -142,7 +142,7 @@ TEST(SpliceTest, FromEventFD) { constexpr uint64_t kEventFDValue = 1; int efd; ASSERT_THAT(efd = eventfd(kEventFDValue, 0), SyscallSucceeds()); - const FileDescriptor inf(efd); + const FileDescriptor in_fd(efd); // Create a new pipe. int fds[2]; @@ -152,7 +152,7 @@ TEST(SpliceTest, FromEventFD) { // Splice 8-byte eventfd value to pipe. constexpr int kEventFDSize = 8; - EXPECT_THAT(splice(inf.get(), nullptr, wfd.get(), nullptr, kEventFDSize, 0), + EXPECT_THAT(splice(in_fd.get(), nullptr, wfd.get(), nullptr, kEventFDSize, 0), SyscallSucceedsWithValue(kEventFDSize)); // Contents should be equal. @@ -166,7 +166,7 @@ TEST(SpliceTest, FromEventFD) { TEST(SpliceTest, FromEventFDOffset) { int efd; ASSERT_THAT(efd = eventfd(0, 0), SyscallSucceeds()); - const FileDescriptor inf(efd); + const FileDescriptor in_fd(efd); // Create a new pipe. int fds[2]; @@ -179,7 +179,7 @@ TEST(SpliceTest, FromEventFDOffset) { // This is not allowed because eventfd doesn't support pread. constexpr int kEventFDSize = 8; loff_t in_off = 0; - EXPECT_THAT(splice(inf.get(), &in_off, wfd.get(), nullptr, kEventFDSize, 0), + EXPECT_THAT(splice(in_fd.get(), &in_off, wfd.get(), nullptr, kEventFDSize, 0), SyscallFailsWithErrno(EINVAL)); } @@ -200,28 +200,29 @@ TEST(SpliceTest, ToEventFDOffset) { int efd; ASSERT_THAT(efd = eventfd(0, 0), SyscallSucceeds()); - const FileDescriptor outf(efd); + const FileDescriptor out_fd(efd); // Attempt to splice 8-byte eventfd value to pipe with offset. // // This is not allowed because eventfd doesn't support pwrite. loff_t out_off = 0; - EXPECT_THAT(splice(rfd.get(), nullptr, outf.get(), &out_off, kEventFDSize, 0), - SyscallFailsWithErrno(EINVAL)); + EXPECT_THAT( + splice(rfd.get(), nullptr, out_fd.get(), &out_off, kEventFDSize, 0), + SyscallFailsWithErrno(EINVAL)); } TEST(SpliceTest, ToPipe) { // Open the input file. const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); - const FileDescriptor inf = + const FileDescriptor in_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); // Fill with some random data. std::vector buf(kPageSize); RandomizeBuffer(buf.data(), buf.size()); - ASSERT_THAT(write(inf.get(), buf.data(), buf.size()), + ASSERT_THAT(write(in_fd.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(kPageSize)); - ASSERT_THAT(lseek(inf.get(), 0, SEEK_SET), SyscallSucceedsWithValue(0)); + ASSERT_THAT(lseek(in_fd.get(), 0, SEEK_SET), SyscallSucceedsWithValue(0)); // Create a new pipe. int fds[2]; @@ -230,7 +231,7 @@ TEST(SpliceTest, ToPipe) { const FileDescriptor wfd(fds[1]); // Splice to the pipe. - EXPECT_THAT(splice(inf.get(), nullptr, wfd.get(), nullptr, kPageSize, 0), + EXPECT_THAT(splice(in_fd.get(), nullptr, wfd.get(), nullptr, kPageSize, 0), SyscallSucceedsWithValue(kPageSize)); // Contents should be equal. @@ -243,13 +244,13 @@ TEST(SpliceTest, ToPipe) { TEST(SpliceTest, ToPipeOffset) { // Open the input file. const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); - const FileDescriptor inf = + const FileDescriptor in_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); // Fill with some random data. std::vector buf(kPageSize); RandomizeBuffer(buf.data(), buf.size()); - ASSERT_THAT(write(inf.get(), buf.data(), buf.size()), + ASSERT_THAT(write(in_fd.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(kPageSize)); // Create a new pipe. @@ -261,7 +262,7 @@ TEST(SpliceTest, ToPipeOffset) { // Splice to the pipe. loff_t in_offset = kPageSize / 2; EXPECT_THAT( - splice(inf.get(), &in_offset, wfd.get(), nullptr, kPageSize / 2, 0), + splice(in_fd.get(), &in_offset, wfd.get(), nullptr, kPageSize / 2, 0), SyscallSucceedsWithValue(kPageSize / 2)); // Contents should be equal to only the second part. @@ -286,22 +287,22 @@ TEST(SpliceTest, FromPipe) { // Open the input file. const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); - const FileDescriptor outf = + const FileDescriptor out_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(out_file.path(), O_RDWR)); // Splice to the output file. - EXPECT_THAT(splice(rfd.get(), nullptr, outf.get(), nullptr, kPageSize, 0), + EXPECT_THAT(splice(rfd.get(), nullptr, out_fd.get(), nullptr, kPageSize, 0), SyscallSucceedsWithValue(kPageSize)); // The offset of the output should be equal to kPageSize. We assert that and // reset to zero so that we can read the contents and ensure they match. - EXPECT_THAT(lseek(outf.get(), 0, SEEK_CUR), + EXPECT_THAT(lseek(out_fd.get(), 0, SEEK_CUR), SyscallSucceedsWithValue(kPageSize)); - ASSERT_THAT(lseek(outf.get(), 0, SEEK_SET), SyscallSucceedsWithValue(0)); + ASSERT_THAT(lseek(out_fd.get(), 0, SEEK_SET), SyscallSucceedsWithValue(0)); // Contents should be equal. std::vector rbuf(kPageSize); - ASSERT_THAT(read(outf.get(), rbuf.data(), rbuf.size()), + ASSERT_THAT(read(out_fd.get(), rbuf.data(), rbuf.size()), SyscallSucceedsWithValue(kPageSize)); EXPECT_EQ(memcmp(rbuf.data(), buf.data(), buf.size()), 0); } @@ -321,18 +322,19 @@ TEST(SpliceTest, FromPipeOffset) { // Open the input file. const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); - const FileDescriptor outf = + const FileDescriptor out_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(out_file.path(), O_RDWR)); // Splice to the output file. loff_t out_offset = kPageSize / 2; - EXPECT_THAT(splice(rfd.get(), nullptr, outf.get(), &out_offset, kPageSize, 0), - SyscallSucceedsWithValue(kPageSize)); + EXPECT_THAT( + splice(rfd.get(), nullptr, out_fd.get(), &out_offset, kPageSize, 0), + SyscallSucceedsWithValue(kPageSize)); // Content should reflect the splice. We write to a specific offset in the // file, so the internals should now be allocated sparsely. std::vector rbuf(kPageSize); - ASSERT_THAT(read(outf.get(), rbuf.data(), rbuf.size()), + ASSERT_THAT(read(out_fd.get(), rbuf.data(), rbuf.size()), SyscallSucceedsWithValue(kPageSize)); std::vector zbuf(kPageSize / 2); memset(zbuf.data(), 0, zbuf.size()); @@ -404,8 +406,6 @@ TEST(SpliceTest, Blocking) { } TEST(TeeTest, Blocking) { - SKIP_IF(IsRunningOnGvisor()); - // Create two new pipes. int first[2], second[2]; ASSERT_THAT(pipe(first), SyscallSucceeds()); @@ -440,6 +440,49 @@ TEST(TeeTest, Blocking) { EXPECT_EQ(memcmp(rbuf.data(), buf.data(), kPageSize), 0); } +TEST(TeeTest, BlockingWrite) { + // Create two new pipes. + int first[2], second[2]; + ASSERT_THAT(pipe(first), SyscallSucceeds()); + const FileDescriptor rfd1(first[0]); + const FileDescriptor wfd1(first[1]); + ASSERT_THAT(pipe(second), SyscallSucceeds()); + const FileDescriptor rfd2(second[0]); + const FileDescriptor wfd2(second[1]); + + // Make some data available to be read. + std::vector buf1(kPageSize); + RandomizeBuffer(buf1.data(), buf1.size()); + ASSERT_THAT(write(wfd1.get(), buf1.data(), buf1.size()), + SyscallSucceedsWithValue(kPageSize)); + + // Fill up the write pipe's buffer. + int pipe_size = -1; + ASSERT_THAT(pipe_size = fcntl(wfd2.get(), F_GETPIPE_SZ), SyscallSucceeds()); + std::vector buf2(pipe_size); + ASSERT_THAT(write(wfd2.get(), buf2.data(), buf2.size()), + SyscallSucceedsWithValue(pipe_size)); + + ScopedThread t([&]() { + absl::SleepFor(absl::Milliseconds(100)); + ASSERT_THAT(read(rfd2.get(), buf2.data(), buf2.size()), + SyscallSucceedsWithValue(pipe_size)); + }); + + // Attempt a tee immediately; it should block. + EXPECT_THAT(tee(rfd1.get(), wfd2.get(), kPageSize, 0), + SyscallSucceedsWithValue(kPageSize)); + + // Thread should be joinable. + t.Join(); + + // Content should reflect the tee. + std::vector rbuf(kPageSize); + ASSERT_THAT(read(rfd2.get(), rbuf.data(), rbuf.size()), + SyscallSucceedsWithValue(kPageSize)); + EXPECT_EQ(memcmp(rbuf.data(), buf1.data(), kPageSize), 0); +} + TEST(SpliceTest, NonBlocking) { // Create two new pipes. int first[2], second[2]; @@ -457,8 +500,6 @@ TEST(SpliceTest, NonBlocking) { } TEST(TeeTest, NonBlocking) { - SKIP_IF(IsRunningOnGvisor()); - // Create two new pipes. int first[2], second[2]; ASSERT_THAT(pipe(first), SyscallSucceeds()); @@ -473,6 +514,79 @@ TEST(TeeTest, NonBlocking) { SyscallFailsWithErrno(EAGAIN)); } +TEST(TeeTest, MultiPage) { + // Create two new pipes. + int first[2], second[2]; + ASSERT_THAT(pipe(first), SyscallSucceeds()); + const FileDescriptor rfd1(first[0]); + const FileDescriptor wfd1(first[1]); + ASSERT_THAT(pipe(second), SyscallSucceeds()); + const FileDescriptor rfd2(second[0]); + const FileDescriptor wfd2(second[1]); + + // Make some data available to be read. + std::vector wbuf(8 * kPageSize); + RandomizeBuffer(wbuf.data(), wbuf.size()); + ASSERT_THAT(write(wfd1.get(), wbuf.data(), wbuf.size()), + SyscallSucceedsWithValue(wbuf.size())); + + // Attempt a tee immediately; it should complete. + EXPECT_THAT(tee(rfd1.get(), wfd2.get(), wbuf.size(), 0), + SyscallSucceedsWithValue(wbuf.size())); + + // Content should reflect the tee. + std::vector rbuf(wbuf.size()); + ASSERT_THAT(read(rfd2.get(), rbuf.data(), rbuf.size()), + SyscallSucceedsWithValue(rbuf.size())); + EXPECT_EQ(memcmp(rbuf.data(), wbuf.data(), rbuf.size()), 0); + ASSERT_THAT(read(rfd1.get(), rbuf.data(), rbuf.size()), + SyscallSucceedsWithValue(rbuf.size())); + EXPECT_EQ(memcmp(rbuf.data(), wbuf.data(), rbuf.size()), 0); +} + +TEST(SpliceTest, FromPipeMaxFileSize) { + // Create a new pipe. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Fill with some random data. + std::vector buf(kPageSize); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(write(wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(kPageSize)); + + // Open the input file. + const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor out_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(out_file.path(), O_RDWR)); + + EXPECT_THAT(ftruncate(out_fd.get(), 13 << 20), SyscallSucceeds()); + EXPECT_THAT(lseek(out_fd.get(), 0, SEEK_END), + SyscallSucceedsWithValue(13 << 20)); + + // Set our file size limit. + sigset_t set; + sigemptyset(&set); + sigaddset(&set, SIGXFSZ); + TEST_PCHECK(sigprocmask(SIG_BLOCK, &set, nullptr) == 0); + rlimit rlim = {}; + rlim.rlim_cur = rlim.rlim_max = (13 << 20); + EXPECT_THAT(setrlimit(RLIMIT_FSIZE, &rlim), SyscallSucceeds()); + + // Splice to the output file. + EXPECT_THAT( + splice(rfd.get(), nullptr, out_fd.get(), nullptr, 3 * kPageSize, 0), + SyscallFailsWithErrno(EFBIG)); + + // Contents should be equal. + std::vector rbuf(kPageSize); + ASSERT_THAT(read(rfd.get(), rbuf.data(), rbuf.size()), + SyscallSucceedsWithValue(kPageSize)); + EXPECT_EQ(memcmp(rbuf.data(), buf.data(), buf.size()), 0); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 56cb0042181cc4840d8c3995f0970c0c41d4212b Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 16 Sep 2019 11:56:59 -0700 Subject: Migrate from gflags to absl flags absl flags are more modern and we can easily depend on them directly. The repo now successfully builds with --incompatible_load_cc_rules_from_bzl. PiperOrigin-RevId: 269387081 --- WORKSPACE | 10 -------- test/syscalls/linux/BUILD | 12 ++++++++++ test/syscalls/linux/chown.cc | 30 +++++++++++++---------- test/syscalls/linux/fcntl.cc | 47 ++++++++++++++++++++----------------- test/syscalls/linux/kill.cc | 13 +++++----- test/syscalls/linux/link.cc | 6 +++-- test/syscalls/linux/prctl.cc | 9 +++---- test/syscalls/linux/prctl_setuid.cc | 24 ++++++++++++------- test/syscalls/linux/ptrace.cc | 11 +++++---- test/syscalls/linux/sigstop.cc | 7 +++--- test/syscalls/linux/sticky.cc | 31 ++++++++++++++---------- test/syscalls/linux/timers.cc | 7 +++--- test/syscalls/linux/uidgid.cc | 23 +++++++++--------- test/syscalls/linux/vfork.cc | 7 +++--- test/util/BUILD | 3 ++- test/util/test_util.cc | 4 +++- test/util/test_util.h | 1 - 17 files changed, 141 insertions(+), 104 deletions(-) (limited to 'test/syscalls/linux/BUILD') diff --git a/WORKSPACE b/WORKSPACE index 8ce1d42d4..6f62eb73f 100644 --- a/WORKSPACE +++ b/WORKSPACE @@ -221,16 +221,6 @@ go_repository( ) # System Call test dependencies. -http_archive( - name = "com_github_gflags_gflags", - sha256 = "34af2f15cf7367513b352bdcd2493ab14ce43692d2dcd9dfc499492966c64dcf", - strip_prefix = "gflags-2.2.2", - urls = [ - "https://mirror.bazel.build/github.com/gflags/gflags/archive/v2.2.2.tar.gz", - "https://github.com/gflags/gflags/archive/v2.2.2.tar.gz", - ], -) - http_archive( name = "com_google_absl", sha256 = "56775f1283a59e6274c28d99981a9717ff4e0b1161e9129fdb2fcf22531d8d93", diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index df00d2c14..eac32850d 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -394,6 +394,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/synchronization", "@com_google_googletest//:gtest", ], @@ -412,6 +413,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_googletest//:gtest", ], ) @@ -728,6 +730,7 @@ cc_binary( "//test/util:test_util", "//test/util:timer_util", "@com_google_absl//absl/base:core_headers", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/memory", "@com_google_absl//absl/strings", "@com_google_absl//absl/time", @@ -976,6 +979,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/synchronization", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", @@ -996,6 +1000,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/strings", "@com_google_googletest//:gtest", ], @@ -1406,6 +1411,7 @@ cc_binary( "//test/util:posix_error", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_googletest//:gtest", ], ) @@ -1422,6 +1428,7 @@ cc_binary( "//test/util:posix_error", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_googletest//:gtest", ], ) @@ -1605,6 +1612,7 @@ cc_binary( "//test/util:test_util", "//test/util:thread_util", "//test/util:time_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], @@ -1978,6 +1986,7 @@ cc_binary( "//test/util:posix_error", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], @@ -3113,6 +3122,7 @@ cc_binary( "//test/util:signal_util", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], @@ -3192,6 +3202,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/strings", "@com_google_googletest//:gtest", ], @@ -3283,6 +3294,7 @@ cc_binary( "//test/util:multiprocess_util", "//test/util:test_util", "//test/util:time_util", + "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], diff --git a/test/syscalls/linux/chown.cc b/test/syscalls/linux/chown.cc index 2e82f0b3a..7a28b674d 100644 --- a/test/syscalls/linux/chown.cc +++ b/test/syscalls/linux/chown.cc @@ -16,10 +16,12 @@ #include #include #include + #include #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/synchronization/notification.h" #include "test/util/capability_util.h" #include "test/util/file_descriptor.h" @@ -29,9 +31,9 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_int32(scratch_uid1, 65534, "first scratch UID"); -DEFINE_int32(scratch_uid2, 65533, "second scratch UID"); -DEFINE_int32(scratch_gid, 65534, "first scratch GID"); +ABSL_FLAG(int32_t, scratch_uid1, 65534, "first scratch UID"); +ABSL_FLAG(int32_t, scratch_uid2, 65533, "second scratch UID"); +ABSL_FLAG(int32_t, scratch_gid, 65534, "first scratch GID"); namespace gvisor { namespace testing { @@ -100,10 +102,12 @@ TEST_P(ChownParamTest, ChownFilePermissionDenied) { // Change EUID and EGID. // // See note about POSIX below. - EXPECT_THAT(syscall(SYS_setresgid, -1, FLAGS_scratch_gid, -1), - SyscallSucceeds()); - EXPECT_THAT(syscall(SYS_setresuid, -1, FLAGS_scratch_uid1, -1), - SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresgid, -1, absl::GetFlag(FLAGS_scratch_gid), -1), + SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresuid, -1, absl::GetFlag(FLAGS_scratch_uid1), -1), + SyscallSucceeds()); EXPECT_THAT(GetParam()(file.path(), geteuid(), getegid()), PosixErrorIs(EPERM, ::testing::ContainsRegex("chown"))); @@ -125,8 +129,9 @@ TEST_P(ChownParamTest, ChownFileSucceedsAsRoot) { // setresuid syscall. However, we want this thread to have its own set of // credentials different from the parent process, so we use the raw // syscall. - EXPECT_THAT(syscall(SYS_setresuid, -1, FLAGS_scratch_uid2, -1), - SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresuid, -1, absl::GetFlag(FLAGS_scratch_uid2), -1), + SyscallSucceeds()); // Create file and immediately close it. FileDescriptor fd = @@ -143,12 +148,13 @@ TEST_P(ChownParamTest, ChownFileSucceedsAsRoot) { fileCreated.WaitForNotification(); // Set file's owners to someone different. - EXPECT_NO_ERRNO(GetParam()(filename, FLAGS_scratch_uid1, FLAGS_scratch_gid)); + EXPECT_NO_ERRNO(GetParam()(filename, absl::GetFlag(FLAGS_scratch_uid1), + absl::GetFlag(FLAGS_scratch_gid))); struct stat s; EXPECT_THAT(stat(filename.c_str(), &s), SyscallSucceeds()); - EXPECT_EQ(s.st_uid, FLAGS_scratch_uid1); - EXPECT_EQ(s.st_gid, FLAGS_scratch_gid); + EXPECT_EQ(s.st_uid, absl::GetFlag(FLAGS_scratch_uid1)); + EXPECT_EQ(s.st_gid, absl::GetFlag(FLAGS_scratch_gid)); fileChowned.Notify(); } diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 2f8e7c9dd..8a45be12a 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -17,9 +17,12 @@ #include #include +#include + #include "gtest/gtest.h" #include "absl/base/macros.h" #include "absl/base/port.h" +#include "absl/flags/flag.h" #include "absl/memory/memory.h" #include "absl/strings/str_cat.h" #include "absl/time/clock.h" @@ -33,18 +36,19 @@ #include "test/util/test_util.h" #include "test/util/timer_util.h" -DEFINE_string(child_setlock_on, "", - "Contains the path to try to set a file lock on."); -DEFINE_bool(child_setlock_write, false, - "Whether to set a writable lock (otherwise readable)"); -DEFINE_bool(blocking, false, - "Whether to set a blocking lock (otherwise non-blocking)."); -DEFINE_bool(retry_eintr, false, "Whether to retry in the subprocess on EINTR."); -DEFINE_uint64(child_setlock_start, 0, "The value of struct flock start"); -DEFINE_uint64(child_setlock_len, 0, "The value of struct flock len"); -DEFINE_int32(socket_fd, -1, - "A socket to use for communicating more state back " - "to the parent."); +ABSL_FLAG(std::string, child_setlock_on, "", + "Contains the path to try to set a file lock on."); +ABSL_FLAG(bool, child_setlock_write, false, + "Whether to set a writable lock (otherwise readable)"); +ABSL_FLAG(bool, blocking, false, + "Whether to set a blocking lock (otherwise non-blocking)."); +ABSL_FLAG(bool, retry_eintr, false, + "Whether to retry in the subprocess on EINTR."); +ABSL_FLAG(uint64_t, child_setlock_start, 0, "The value of struct flock start"); +ABSL_FLAG(uint64_t, child_setlock_len, 0, "The value of struct flock len"); +ABSL_FLAG(int32_t, socket_fd, -1, + "A socket to use for communicating more state back " + "to the parent."); namespace gvisor { namespace testing { @@ -918,25 +922,26 @@ TEST(FcntlTest, GetOwn) { int main(int argc, char** argv) { gvisor::testing::TestInit(&argc, &argv); - if (!FLAGS_child_setlock_on.empty()) { - int socket_fd = FLAGS_socket_fd; - int fd = open(FLAGS_child_setlock_on.c_str(), O_RDWR, 0666); + const std::string setlock_on = absl::GetFlag(FLAGS_child_setlock_on); + if (!setlock_on.empty()) { + int socket_fd = absl::GetFlag(FLAGS_socket_fd); + int fd = open(setlock_on.c_str(), O_RDWR, 0666); if (fd == -1 && errno != 0) { int err = errno; - std::cerr << "CHILD open " << FLAGS_child_setlock_on << " failed " << err + std::cerr << "CHILD open " << setlock_on << " failed " << err << std::endl; exit(err); } struct flock fl; - if (FLAGS_child_setlock_write) { + if (absl::GetFlag(FLAGS_child_setlock_write)) { fl.l_type = F_WRLCK; } else { fl.l_type = F_RDLCK; } fl.l_whence = SEEK_SET; - fl.l_start = FLAGS_child_setlock_start; - fl.l_len = FLAGS_child_setlock_len; + fl.l_start = absl::GetFlag(FLAGS_child_setlock_start); + fl.l_len = absl::GetFlag(FLAGS_child_setlock_len); // Test the fcntl, no need to log, the error is unambiguously // from fcntl at this point. @@ -946,8 +951,8 @@ int main(int argc, char** argv) { gvisor::testing::MonotonicTimer timer; timer.Start(); do { - ret = fcntl(fd, FLAGS_blocking ? F_SETLKW : F_SETLK, &fl); - } while (FLAGS_retry_eintr && ret == -1 && errno == EINTR); + ret = fcntl(fd, absl::GetFlag(FLAGS_blocking) ? F_SETLKW : F_SETLK, &fl); + } while (absl::GetFlag(FLAGS_retry_eintr) && ret == -1 && errno == EINTR); auto usec = absl::ToInt64Microseconds(timer.Duration()); if (ret == -1 && errno != 0) { diff --git a/test/syscalls/linux/kill.cc b/test/syscalls/linux/kill.cc index 18ad923b8..db29bd59c 100644 --- a/test/syscalls/linux/kill.cc +++ b/test/syscalls/linux/kill.cc @@ -21,6 +21,7 @@ #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/synchronization/mutex.h" #include "absl/time/clock.h" #include "absl/time/time.h" @@ -31,8 +32,8 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_int32(scratch_uid, 65534, "scratch UID"); -DEFINE_int32(scratch_gid, 65534, "scratch GID"); +ABSL_FLAG(int32_t, scratch_uid, 65534, "scratch UID"); +ABSL_FLAG(int32_t, scratch_gid, 65534, "scratch GID"); using ::testing::Ge; @@ -255,8 +256,8 @@ TEST(KillTest, ProcessGroups) { TEST(KillTest, ChildDropsPrivsCannotKill) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SETUID))); - int uid = FLAGS_scratch_uid; - int gid = FLAGS_scratch_gid; + const int uid = absl::GetFlag(FLAGS_scratch_uid); + const int gid = absl::GetFlag(FLAGS_scratch_gid); // Create the child that drops privileges and tries to kill the parent. pid_t pid = fork(); @@ -331,8 +332,8 @@ TEST(KillTest, CanSIGCONTSameSession) { EXPECT_TRUE(WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP) << "status " << status; - int uid = FLAGS_scratch_uid; - int gid = FLAGS_scratch_gid; + const int uid = absl::GetFlag(FLAGS_scratch_uid); + const int gid = absl::GetFlag(FLAGS_scratch_gid); // Drop privileges only in child process, or else this parent process won't be // able to open some log files after the test ends. diff --git a/test/syscalls/linux/link.cc b/test/syscalls/linux/link.cc index a91703070..dd5352954 100644 --- a/test/syscalls/linux/link.cc +++ b/test/syscalls/linux/link.cc @@ -22,6 +22,7 @@ #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/strings/str_cat.h" #include "test/util/capability_util.h" #include "test/util/file_descriptor.h" @@ -31,7 +32,7 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_int32(scratch_uid, 65534, "scratch UID"); +ABSL_FLAG(int32_t, scratch_uid, 65534, "scratch UID"); namespace gvisor { namespace testing { @@ -92,7 +93,8 @@ TEST(LinkTest, PermissionDenied) { // threads have the same UIDs, so using the setuid wrapper sets all threads' // real UID. // Also drops capabilities. - EXPECT_THAT(syscall(SYS_setuid, FLAGS_scratch_uid), SyscallSucceeds()); + EXPECT_THAT(syscall(SYS_setuid, absl::GetFlag(FLAGS_scratch_uid)), + SyscallSucceeds()); EXPECT_THAT(link(oldfile.path().c_str(), newname.c_str()), SyscallFailsWithErrno(EPERM)); diff --git a/test/syscalls/linux/prctl.cc b/test/syscalls/linux/prctl.cc index bd1779557..d07571a5f 100644 --- a/test/syscalls/linux/prctl.cc +++ b/test/syscalls/linux/prctl.cc @@ -21,6 +21,7 @@ #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "test/util/capability_util.h" #include "test/util/cleanup.h" #include "test/util/multiprocess_util.h" @@ -28,9 +29,9 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_bool(prctl_no_new_privs_test_child, false, - "If true, exit with the return value of prctl(PR_GET_NO_NEW_PRIVS) " - "plus an offset (see test source)."); +ABSL_FLAG(bool, prctl_no_new_privs_test_child, false, + "If true, exit with the return value of prctl(PR_GET_NO_NEW_PRIVS) " + "plus an offset (see test source)."); namespace gvisor { namespace testing { @@ -220,7 +221,7 @@ TEST(PrctlTest, RootDumpability) { int main(int argc, char** argv) { gvisor::testing::TestInit(&argc, &argv); - if (FLAGS_prctl_no_new_privs_test_child) { + if (absl::GetFlag(FLAGS_prctl_no_new_privs_test_child)) { exit(gvisor::testing::kPrctlNoNewPrivsTestChildExitBase + prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0)); } diff --git a/test/syscalls/linux/prctl_setuid.cc b/test/syscalls/linux/prctl_setuid.cc index 00dd6523e..30f0d75b3 100644 --- a/test/syscalls/linux/prctl_setuid.cc +++ b/test/syscalls/linux/prctl_setuid.cc @@ -14,9 +14,11 @@ #include #include + #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "test/util/capability_util.h" #include "test/util/logging.h" #include "test/util/multiprocess_util.h" @@ -24,12 +26,12 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_int32(scratch_uid, 65534, "scratch UID"); +ABSL_FLAG(int32_t, scratch_uid, 65534, "scratch UID"); // This flag is used to verify that after an exec PR_GET_KEEPCAPS // returns 0, the return code will be offset by kPrGetKeepCapsExitBase. -DEFINE_bool(prctl_pr_get_keepcaps, false, - "If true the test will verify that prctl with pr_get_keepcaps" - "returns 0. The test will exit with the result of that check."); +ABSL_FLAG(bool, prctl_pr_get_keepcaps, false, + "If true the test will verify that prctl with pr_get_keepcaps" + "returns 0. The test will exit with the result of that check."); // These tests exist seperately from prctl because we need to start // them as root. Setuid() has the behavior that permissions are fully @@ -113,10 +115,12 @@ TEST_F(PrctlKeepCapsSetuidTest, SetUidNoKeepCaps) { // call to only apply to this task. POSIX threads, however, require that // all threads have the same UIDs, so using the setuid wrapper sets all // threads' real UID. - EXPECT_THAT(syscall(SYS_setuid, FLAGS_scratch_uid), SyscallSucceeds()); + EXPECT_THAT(syscall(SYS_setuid, absl::GetFlag(FLAGS_scratch_uid)), + SyscallSucceeds()); // Verify that we changed uid. - EXPECT_THAT(getuid(), SyscallSucceedsWithValue(FLAGS_scratch_uid)); + EXPECT_THAT(getuid(), + SyscallSucceedsWithValue(absl::GetFlag(FLAGS_scratch_uid))); // Verify we lost the capability in the effective set, this always happens. TEST_CHECK(!HaveCapability(CAP_SYS_ADMIN).ValueOrDie()); @@ -157,10 +161,12 @@ TEST_F(PrctlKeepCapsSetuidTest, SetUidKeepCaps) { // call to only apply to this task. POSIX threads, however, require that // all threads have the same UIDs, so using the setuid wrapper sets all // threads' real UID. - EXPECT_THAT(syscall(SYS_setuid, FLAGS_scratch_uid), SyscallSucceeds()); + EXPECT_THAT(syscall(SYS_setuid, absl::GetFlag(FLAGS_scratch_uid)), + SyscallSucceeds()); // Verify that we changed uid. - EXPECT_THAT(getuid(), SyscallSucceedsWithValue(FLAGS_scratch_uid)); + EXPECT_THAT(getuid(), + SyscallSucceedsWithValue(absl::GetFlag(FLAGS_scratch_uid))); // Verify we lost the capability in the effective set, this always happens. TEST_CHECK(!HaveCapability(CAP_SYS_ADMIN).ValueOrDie()); @@ -253,7 +259,7 @@ TEST_F(PrctlKeepCapsSetuidTest, PrGetKeepCaps) { int main(int argc, char** argv) { gvisor::testing::TestInit(&argc, &argv); - if (FLAGS_prctl_pr_get_keepcaps) { + if (absl::GetFlag(FLAGS_prctl_pr_get_keepcaps)) { return gvisor::testing::kPrGetKeepCapsExitBase + prctl(PR_GET_KEEPCAPS, 0, 0, 0, 0); } diff --git a/test/syscalls/linux/ptrace.cc b/test/syscalls/linux/ptrace.cc index abf2b1a04..8f3800380 100644 --- a/test/syscalls/linux/ptrace.cc +++ b/test/syscalls/linux/ptrace.cc @@ -27,6 +27,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/logging.h" @@ -36,10 +37,10 @@ #include "test/util/thread_util.h" #include "test/util/time_util.h" -DEFINE_bool(ptrace_test_execve_child, false, - "If true, run the " - "PtraceExecveTest_Execve_GetRegs_PeekUser_SIGKILL_TraceClone_" - "TraceExit child workload."); +ABSL_FLAG(bool, ptrace_test_execve_child, false, + "If true, run the " + "PtraceExecveTest_Execve_GetRegs_PeekUser_SIGKILL_TraceClone_" + "TraceExit child workload."); namespace gvisor { namespace testing { @@ -1206,7 +1207,7 @@ TEST(PtraceTest, SeizeSetOptions) { int main(int argc, char** argv) { gvisor::testing::TestInit(&argc, &argv); - if (FLAGS_ptrace_test_execve_child) { + if (absl::GetFlag(FLAGS_ptrace_test_execve_child)) { gvisor::testing::RunExecveChild(); } diff --git a/test/syscalls/linux/sigstop.cc b/test/syscalls/linux/sigstop.cc index 9c7210e17..7db57d968 100644 --- a/test/syscalls/linux/sigstop.cc +++ b/test/syscalls/linux/sigstop.cc @@ -17,6 +17,7 @@ #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/multiprocess_util.h" @@ -24,8 +25,8 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_bool(sigstop_test_child, false, - "If true, run the SigstopTest child workload."); +ABSL_FLAG(bool, sigstop_test_child, false, + "If true, run the SigstopTest child workload."); namespace gvisor { namespace testing { @@ -141,7 +142,7 @@ void RunChild() { int main(int argc, char** argv) { gvisor::testing::TestInit(&argc, &argv); - if (FLAGS_sigstop_test_child) { + if (absl::GetFlag(FLAGS_sigstop_test_child)) { gvisor::testing::RunChild(); return 1; } diff --git a/test/syscalls/linux/sticky.cc b/test/syscalls/linux/sticky.cc index 59fb5dfe6..7e73325bf 100644 --- a/test/syscalls/linux/sticky.cc +++ b/test/syscalls/linux/sticky.cc @@ -17,9 +17,11 @@ #include #include #include + #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "test/util/capability_util.h" #include "test/util/file_descriptor.h" #include "test/util/fs_util.h" @@ -27,8 +29,8 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_int32(scratch_uid, 65534, "first scratch UID"); -DEFINE_int32(scratch_gid, 65534, "first scratch GID"); +ABSL_FLAG(int32_t, scratch_uid, 65534, "first scratch UID"); +ABSL_FLAG(int32_t, scratch_gid, 65534, "first scratch GID"); namespace gvisor { namespace testing { @@ -52,10 +54,12 @@ TEST(StickyTest, StickyBitPermDenied) { } // Change EUID and EGID. - EXPECT_THAT(syscall(SYS_setresgid, -1, FLAGS_scratch_gid, -1), - SyscallSucceeds()); - EXPECT_THAT(syscall(SYS_setresuid, -1, FLAGS_scratch_uid, -1), - SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresgid, -1, absl::GetFlag(FLAGS_scratch_gid), -1), + SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresuid, -1, absl::GetFlag(FLAGS_scratch_uid), -1), + SyscallSucceeds()); EXPECT_THAT(rmdir(path.c_str()), SyscallFailsWithErrno(EPERM)); }); @@ -78,8 +82,9 @@ TEST(StickyTest, StickyBitSameUID) { } // Change EGID. - EXPECT_THAT(syscall(SYS_setresgid, -1, FLAGS_scratch_gid, -1), - SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresgid, -1, absl::GetFlag(FLAGS_scratch_gid), -1), + SyscallSucceeds()); // We still have the same EUID. EXPECT_THAT(rmdir(path.c_str()), SyscallSucceeds()); @@ -101,10 +106,12 @@ TEST(StickyTest, StickyBitCapFOWNER) { EXPECT_THAT(prctl(PR_SET_KEEPCAPS, 1, 0, 0, 0), SyscallSucceeds()); // Change EUID and EGID. - EXPECT_THAT(syscall(SYS_setresgid, -1, FLAGS_scratch_gid, -1), - SyscallSucceeds()); - EXPECT_THAT(syscall(SYS_setresuid, -1, FLAGS_scratch_uid, -1), - SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresgid, -1, absl::GetFlag(FLAGS_scratch_gid), -1), + SyscallSucceeds()); + EXPECT_THAT( + syscall(SYS_setresuid, -1, absl::GetFlag(FLAGS_scratch_uid), -1), + SyscallSucceeds()); EXPECT_NO_ERRNO(SetCapability(CAP_FOWNER, true)); EXPECT_THAT(rmdir(path.c_str()), SyscallSucceeds()); diff --git a/test/syscalls/linux/timers.cc b/test/syscalls/linux/timers.cc index fd42e81e1..3db18d7ac 100644 --- a/test/syscalls/linux/timers.cc +++ b/test/syscalls/linux/timers.cc @@ -23,6 +23,7 @@ #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/cleanup.h" @@ -33,8 +34,8 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_bool(timers_test_sleep, false, - "If true, sleep forever instead of running tests."); +ABSL_FLAG(bool, timers_test_sleep, false, + "If true, sleep forever instead of running tests."); using ::testing::_; using ::testing::AnyOf; @@ -635,7 +636,7 @@ TEST(IntervalTimerTest, IgnoredSignalCountsAsOverrun) { int main(int argc, char** argv) { gvisor::testing::TestInit(&argc, &argv); - if (FLAGS_timers_test_sleep) { + if (absl::GetFlag(FLAGS_timers_test_sleep)) { while (true) { absl::SleepFor(absl::Seconds(10)); } diff --git a/test/syscalls/linux/uidgid.cc b/test/syscalls/linux/uidgid.cc index bf1ca8679..d48453a93 100644 --- a/test/syscalls/linux/uidgid.cc +++ b/test/syscalls/linux/uidgid.cc @@ -18,6 +18,7 @@ #include #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/strings/str_cat.h" #include "absl/strings/str_join.h" #include "test/util/capability_util.h" @@ -25,10 +26,10 @@ #include "test/util/test_util.h" #include "test/util/thread_util.h" -DEFINE_int32(scratch_uid1, 65534, "first scratch UID"); -DEFINE_int32(scratch_uid2, 65533, "second scratch UID"); -DEFINE_int32(scratch_gid1, 65534, "first scratch GID"); -DEFINE_int32(scratch_gid2, 65533, "second scratch GID"); +ABSL_FLAG(int32_t, scratch_uid1, 65534, "first scratch UID"); +ABSL_FLAG(int32_t, scratch_uid2, 65533, "second scratch UID"); +ABSL_FLAG(int32_t, scratch_gid1, 65534, "first scratch GID"); +ABSL_FLAG(int32_t, scratch_gid2, 65533, "second scratch GID"); using ::testing::UnorderedElementsAreArray; @@ -146,7 +147,7 @@ TEST(UidGidRootTest, Setuid) { // real UID. EXPECT_THAT(syscall(SYS_setuid, -1), SyscallFailsWithErrno(EINVAL)); - const uid_t uid = FLAGS_scratch_uid1; + const uid_t uid = absl::GetFlag(FLAGS_scratch_uid1); EXPECT_THAT(syscall(SYS_setuid, uid), SyscallSucceeds()); // "If the effective UID of the caller is root (more precisely: if the // caller has the CAP_SETUID capability), the real UID and saved set-user-ID @@ -160,7 +161,7 @@ TEST(UidGidRootTest, Setgid) { EXPECT_THAT(setgid(-1), SyscallFailsWithErrno(EINVAL)); - const gid_t gid = FLAGS_scratch_gid1; + const gid_t gid = absl::GetFlag(FLAGS_scratch_gid1); ASSERT_THAT(setgid(gid), SyscallSucceeds()); EXPECT_NO_ERRNO(CheckGIDs(gid, gid, gid)); } @@ -168,7 +169,7 @@ TEST(UidGidRootTest, Setgid) { TEST(UidGidRootTest, SetgidNotFromThreadGroupLeader) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(IsRoot())); - const gid_t gid = FLAGS_scratch_gid1; + const gid_t gid = absl::GetFlag(FLAGS_scratch_gid1); // NOTE(b/64676707): Do setgid in a separate thread so that we can test if // info.si_pid is set correctly. ScopedThread([gid] { ASSERT_THAT(setgid(gid), SyscallSucceeds()); }); @@ -189,8 +190,8 @@ TEST(UidGidRootTest, Setreuid) { // cannot be opened by the `uid` set below after the test. After calling // setuid(non-zero-UID), there is no way to get root privileges back. ScopedThread([&] { - const uid_t ruid = FLAGS_scratch_uid1; - const uid_t euid = FLAGS_scratch_uid2; + const uid_t ruid = absl::GetFlag(FLAGS_scratch_uid1); + const uid_t euid = absl::GetFlag(FLAGS_scratch_uid2); // Use syscall instead of glibc setuid wrapper because we want this setuid // call to only apply to this task. posix threads, however, require that all @@ -211,8 +212,8 @@ TEST(UidGidRootTest, Setregid) { EXPECT_THAT(setregid(-1, -1), SyscallSucceeds()); EXPECT_NO_ERRNO(CheckGIDs(0, 0, 0)); - const gid_t rgid = FLAGS_scratch_gid1; - const gid_t egid = FLAGS_scratch_gid2; + const gid_t rgid = absl::GetFlag(FLAGS_scratch_gid1); + const gid_t egid = absl::GetFlag(FLAGS_scratch_gid2); ASSERT_THAT(setregid(rgid, egid), SyscallSucceeds()); EXPECT_NO_ERRNO(CheckGIDs(rgid, egid, egid)); } diff --git a/test/syscalls/linux/vfork.cc b/test/syscalls/linux/vfork.cc index f67b06f37..0aaba482d 100644 --- a/test/syscalls/linux/vfork.cc +++ b/test/syscalls/linux/vfork.cc @@ -22,14 +22,15 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/flags/flag.h" #include "absl/time/time.h" #include "test/util/logging.h" #include "test/util/multiprocess_util.h" #include "test/util/test_util.h" #include "test/util/time_util.h" -DEFINE_bool(vfork_test_child, false, - "If true, run the VforkTest child workload."); +ABSL_FLAG(bool, vfork_test_child, false, + "If true, run the VforkTest child workload."); namespace gvisor { namespace testing { @@ -186,7 +187,7 @@ int RunChild() { int main(int argc, char** argv) { gvisor::testing::TestInit(&argc, &argv); - if (FLAGS_vfork_test_child) { + if (absl::GetFlag(FLAGS_vfork_test_child)) { return gvisor::testing::RunChild(); } diff --git a/test/util/BUILD b/test/util/BUILD index 64daa0597..52f8b9e1f 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -228,8 +228,9 @@ cc_library( ":logging", ":posix_error", ":save_util", - "@com_github_gflags_gflags//:gflags", "@com_google_absl//absl/base:core_headers", + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/flags:parse", "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", "@com_google_absl//absl/time", diff --git a/test/util/test_util.cc b/test/util/test_util.cc index e42bba04a..ba0dcf7d0 100644 --- a/test/util/test_util.cc +++ b/test/util/test_util.cc @@ -28,6 +28,8 @@ #include #include "absl/base/attributes.h" +#include "absl/flags/flag.h" // IWYU pragma: keep +#include "absl/flags/parse.h" // IWYU pragma: keep #include "absl/strings/numbers.h" #include "absl/strings/str_cat.h" #include "absl/strings/str_split.h" @@ -224,7 +226,7 @@ bool Equivalent(uint64_t current, uint64_t target, double tolerance) { void TestInit(int* argc, char*** argv) { ::testing::InitGoogleTest(argc, *argv); - ::gflags::ParseCommandLineFlags(argc, argv, true); + ::absl::ParseCommandLine(*argc, *argv); // Always mask SIGPIPE as it's common and tests aren't expected to handle it. struct sigaction sa = {}; diff --git a/test/util/test_util.h b/test/util/test_util.h index cdbe8bfd1..b9d2dc2ba 100644 --- a/test/util/test_util.h +++ b/test/util/test_util.h @@ -185,7 +185,6 @@ #include #include -#include #include "gmock/gmock.h" #include "absl/strings/str_cat.h" #include "absl/strings/str_format.h" -- cgit v1.2.3 From c98e7f0d19478ca57ba8c96444f225784035321e Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Wed, 18 Sep 2019 15:15:16 -0700 Subject: Signalfd support Note that the exact semantics for these signalfds are slightly different from Linux. These signalfds are bound to the process at creation time. Reads, polls, etc. are all associated with signals directed at that task. In Linux, all signalfd operations are associated with current, regardless of where the signalfd originated. In practice, this should not be an issue given how signalfds are used. In order to fix this however, we will need to plumb the context through all the event APIs. This gets complicated really quickly, because the waiter APIs are all netstack-specific, and not generally exposed to the context. Probably not worthwhile fixing immediately. PiperOrigin-RevId: 269901749 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/signalfd.go | 45 +++++ pkg/sentry/kernel/signalfd/BUILD | 22 +++ pkg/sentry/kernel/signalfd/signalfd.go | 137 +++++++++++++ pkg/sentry/kernel/task.go | 8 + pkg/sentry/kernel/task_signals.go | 18 ++ pkg/sentry/syscalls/linux/BUILD | 1 + pkg/sentry/syscalls/linux/linux64.go | 4 +- pkg/sentry/syscalls/linux/sys_signal.go | 77 ++++++++ test/syscalls/linux/BUILD | 18 ++ test/syscalls/linux/signalfd.cc | 333 ++++++++++++++++++++++++++++++++ 11 files changed, 662 insertions(+), 2 deletions(-) create mode 100644 pkg/abi/linux/signalfd.go create mode 100644 pkg/sentry/kernel/signalfd/BUILD create mode 100644 pkg/sentry/kernel/signalfd/signalfd.go create mode 100644 test/syscalls/linux/signalfd.cc (limited to 'test/syscalls/linux/BUILD') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 39c92bb33..f45934466 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -46,6 +46,7 @@ go_library( "sem.go", "shm.go", "signal.go", + "signalfd.go", "socket.go", "splice.go", "tcp.go", diff --git a/pkg/abi/linux/signalfd.go b/pkg/abi/linux/signalfd.go new file mode 100644 index 000000000..85fad9956 --- /dev/null +++ b/pkg/abi/linux/signalfd.go @@ -0,0 +1,45 @@ +// Copyright 2019 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 linux + +const ( + // SFD_NONBLOCK is a signalfd(2) flag. + SFD_NONBLOCK = 00004000 + + // SFD_CLOEXEC is a signalfd(2) flag. + SFD_CLOEXEC = 02000000 +) + +// SignalfdSiginfo is the siginfo encoding for signalfds. +type SignalfdSiginfo struct { + Signo uint32 + Errno int32 + Code int32 + PID uint32 + UID uint32 + FD int32 + TID uint32 + Band uint32 + Overrun uint32 + TrapNo uint32 + Status int32 + Int int32 + Ptr uint64 + UTime uint64 + STime uint64 + Addr uint64 + AddrLSB uint16 + _ [48]uint8 +} diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD new file mode 100644 index 000000000..50b69d154 --- /dev/null +++ b/pkg/sentry/kernel/signalfd/BUILD @@ -0,0 +1,22 @@ +package(licenses = ["notice"]) + +load("//tools/go_stateify:defs.bzl", "go_library") + +go_library( + name = "signalfd", + srcs = ["signalfd.go"], + importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/signalfd", + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/binary", + "//pkg/sentry/context", + "//pkg/sentry/fs", + "//pkg/sentry/fs/anon", + "//pkg/sentry/fs/fsutil", + "//pkg/sentry/kernel", + "//pkg/sentry/usermem", + "//pkg/syserror", + "//pkg/waiter", + ], +) diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go new file mode 100644 index 000000000..06fd5ec88 --- /dev/null +++ b/pkg/sentry/kernel/signalfd/signalfd.go @@ -0,0 +1,137 @@ +// Copyright 2019 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 signalfd provides an implementation of signal file descriptors. +package signalfd + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fs/anon" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/waiter" +) + +// SignalOperations represent a file with signalfd semantics. +// +// +stateify savable +type SignalOperations struct { + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FilePipeSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoSplice `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileUseInodeUnstableAttr `state:"nosave"` + + // target is the original task target. + // + // The semantics here are a bit broken. Linux will always use current + // for all reads, regardless of where the signalfd originated. We can't + // do exactly that because we need to plumb the context through + // EventRegister in order to support proper blocking behavior. This + // will undoubtedly become very complicated quickly. + target *kernel.Task + + // mu protects below. + mu sync.Mutex `state:"nosave"` + + // mask is the signal mask. Protected by mu. + mask linux.SignalSet +} + +// New creates a new signalfd object with the supplied mask. +func New(ctx context.Context, mask linux.SignalSet) (*fs.File, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + // No task context? Not valid. + return nil, syserror.EINVAL + } + // name matches fs/signalfd.c:signalfd4. + dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "anon_inode:[signalfd]") + return fs.NewFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}, &SignalOperations{ + target: t, + mask: mask, + }), nil +} + +// Release implements fs.FileOperations.Release. +func (s *SignalOperations) Release() {} + +// Mask returns the signal mask. +func (s *SignalOperations) Mask() linux.SignalSet { + s.mu.Lock() + mask := s.mask + s.mu.Unlock() + return mask +} + +// SetMask sets the signal mask. +func (s *SignalOperations) SetMask(mask linux.SignalSet) { + s.mu.Lock() + s.mask = mask + s.mu.Unlock() +} + +// Read implements fs.FileOperations.Read. +func (s *SignalOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) { + // Attempt to dequeue relevant signals. + info, err := s.target.Sigtimedwait(s.Mask(), 0) + if err != nil { + // There must be no signal available. + return 0, syserror.ErrWouldBlock + } + + // Copy out the signal info using the specified format. + var buf [128]byte + binary.Marshal(buf[:0], usermem.ByteOrder, &linux.SignalfdSiginfo{ + Signo: uint32(info.Signo), + Errno: info.Errno, + Code: info.Code, + PID: uint32(info.Pid()), + UID: uint32(info.Uid()), + Status: info.Status(), + Overrun: uint32(info.Overrun()), + Addr: info.Addr(), + }) + n, err := dst.CopyOut(ctx, buf[:]) + return int64(n), err +} + +// Readiness implements waiter.Waitable.Readiness. +func (s *SignalOperations) Readiness(mask waiter.EventMask) waiter.EventMask { + return mask & waiter.EventIn +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (s *SignalOperations) EventRegister(entry *waiter.Entry, _ waiter.EventMask) { + // Register for the signal set; ignore the passed events. + s.target.SignalRegister(entry, waiter.EventMask(s.Mask())) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (s *SignalOperations) EventUnregister(entry *waiter.Entry) { + // Unregister the original entry. + s.target.SignalUnregister(entry) +} diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index e91f82bb3..c82ef5486 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/waiter" "gvisor.dev/gvisor/third_party/gvsync" ) @@ -133,6 +134,13 @@ type Task struct { // signalStack is exclusive to the task goroutine. signalStack arch.SignalStack + // signalQueue is a set of registered waiters for signal-related events. + // + // signalQueue is protected by the signalMutex. Note that the task does + // not implement all queue methods, specifically the readiness checks. + // The task only broadcast a notification on signal delivery. + signalQueue waiter.Queue `state:"zerovalue"` + // If groupStopPending is true, the task should participate in a group // stop in the interrupt path. // diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 266959a07..39cd1340d 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -28,6 +28,7 @@ import ( ucspb "gvisor.dev/gvisor/pkg/sentry/kernel/uncaught_signal_go_proto" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/waiter" ) // SignalAction is an internal signal action. @@ -497,6 +498,9 @@ func (tg *ThreadGroup) applySignalSideEffectsLocked(sig linux.Signal) { // // Preconditions: The signal mutex must be locked. func (t *Task) canReceiveSignalLocked(sig linux.Signal) bool { + // Notify that the signal is queued. + t.signalQueue.Notify(waiter.EventMask(linux.MakeSignalSet(sig))) + // - Do not choose tasks that are blocking the signal. if linux.SignalSetOf(sig)&t.signalMask != 0 { return false @@ -1108,3 +1112,17 @@ func (*runInterruptAfterSignalDeliveryStop) execute(t *Task) taskRunState { t.tg.signalHandlers.mu.Unlock() return t.deliverSignal(info, act) } + +// SignalRegister registers a waiter for pending signals. +func (t *Task) SignalRegister(e *waiter.Entry, mask waiter.EventMask) { + t.tg.signalHandlers.mu.Lock() + t.signalQueue.EventRegister(e, mask) + t.tg.signalHandlers.mu.Unlock() +} + +// SignalUnregister unregisters a waiter for pending signals. +func (t *Task) SignalUnregister(e *waiter.Entry) { + t.tg.signalHandlers.mu.Lock() + t.signalQueue.EventUnregister(e) + t.tg.signalHandlers.mu.Unlock() +} diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 33a40b9c6..e76ee27d2 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -74,6 +74,7 @@ go_library( "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/sched", "//pkg/sentry/kernel/shm", + "//pkg/sentry/kernel/signalfd", "//pkg/sentry/kernel/time", "//pkg/sentry/limits", "//pkg/sentry/memmap", diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index 150999fb8..18d24ab61 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -327,14 +327,14 @@ var AMD64 = &kernel.SyscallTable{ 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) 280: syscalls.Supported("utimensat", Utimensat), 281: syscalls.Supported("epoll_pwait", EpollPwait), - 282: syscalls.ErrorWithEvent("signalfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426) + 282: syscalls.PartiallySupported("signalfd", Signalfd, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), 283: syscalls.Supported("timerfd_create", TimerfdCreate), 284: syscalls.Supported("eventfd", Eventfd), 285: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), 286: syscalls.Supported("timerfd_settime", TimerfdSettime), 287: syscalls.Supported("timerfd_gettime", TimerfdGettime), 288: syscalls.Supported("accept4", Accept4), - 289: syscalls.ErrorWithEvent("signalfd4", syserror.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426) + 289: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), 290: syscalls.Supported("eventfd2", Eventfd2), 291: syscalls.Supported("epoll_create1", EpollCreate1), 292: syscalls.Supported("dup3", Dup3), diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index 0104a94c0..fb6efd5d8 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -20,7 +20,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/signalfd" + "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) @@ -506,3 +509,77 @@ func RestartSyscall(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne t.Debugf("Restart block missing in restart_syscall(2). Did ptrace inject a return value of ERESTART_RESTARTBLOCK?") return 0, nil, syserror.EINTR } + +// sharedSignalfd is shared between the two calls. +func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize uint, flags int32) (uintptr, *kernel.SyscallControl, error) { + // Copy in the signal mask. + mask, err := copyInSigSet(t, sigset, sigsetsize) + if err != nil { + return 0, nil, err + } + + // Always check for valid flags, even if not creating. + if flags&^(linux.SFD_NONBLOCK|linux.SFD_CLOEXEC) != 0 { + return 0, nil, syserror.EINVAL + } + + // Is this a change to an existing signalfd? + // + // The spec indicates that this should adjust the mask. + if fd != -1 { + file := t.GetFile(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Is this a signalfd? + if s, ok := file.FileOperations.(*signalfd.SignalOperations); ok { + s.SetMask(mask) + return 0, nil, nil + } + + // Not a signalfd. + return 0, nil, syserror.EINVAL + } + + // Create a new file. + file, err := signalfd.New(t, mask) + if err != nil { + return 0, nil, err + } + defer file.DecRef() + + // Set appropriate flags. + file.SetFlags(fs.SettableFileFlags{ + NonBlocking: flags&linux.SFD_NONBLOCK != 0, + }) + + // Create a new descriptor. + fd, err = t.NewFDFrom(0, file, kernel.FDFlags{ + CloseOnExec: flags&linux.SFD_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + + // Done. + return uintptr(fd), nil, nil +} + +// Signalfd implements the linux syscall signalfd(2). +func Signalfd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + sigset := args[1].Pointer() + sigsetsize := args[2].SizeT() + return sharedSignalfd(t, fd, sigset, sigsetsize, 0) +} + +// Signalfd4 implements the linux syscall signalfd4(2). +func Signalfd4(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + sigset := args[1].Pointer() + sigsetsize := args[2].SizeT() + flags := args[3].Int() + return sharedSignalfd(t, fd, sigset, sigsetsize, flags) +} diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index eac32850d..56fe7be37 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1963,6 +1963,24 @@ cc_binary( ], ) +cc_binary( + name = "signalfd_test", + testonly = 1, + srcs = ["signalfd.cc"], + linkstatic = 1, + deps = [ + "//test/util:file_descriptor", + "//test/util:logging", + "//test/util:posix_error", + "//test/util:signal_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_absl//absl/synchronization", + "@com_google_googletest//:gtest", + ], +) + cc_binary( name = "sigprocmask_test", testonly = 1, diff --git a/test/syscalls/linux/signalfd.cc b/test/syscalls/linux/signalfd.cc new file mode 100644 index 000000000..54c598627 --- /dev/null +++ b/test/syscalls/linux/signalfd.cc @@ -0,0 +1,333 @@ +// Copyright 2019 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. + +#include +#include +#include +#include +#include +#include +#include + +#include +#include + +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "absl/synchronization/mutex.h" +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" +#include "test/util/signal_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +using ::testing::KilledBySignal; + +namespace gvisor { +namespace testing { + +namespace { + +constexpr int kSigno = SIGUSR1; +constexpr int kSignoAlt = SIGUSR2; + +// Returns a new signalfd. +inline PosixErrorOr NewSignalFD(sigset_t* mask, int flags = 0) { + int fd = signalfd(-1, mask, flags); + MaybeSave(); + if (fd < 0) { + return PosixError(errno, "signalfd"); + } + return FileDescriptor(fd); +} + +TEST(Signalfd, Basic) { + // Create the signalfd. + sigset_t mask; + sigemptyset(&mask); + sigaddset(&mask, kSigno); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, 0)); + + // Deliver the blocked signal. + const auto scoped_sigmask = + ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_BLOCK, kSigno)); + ASSERT_THAT(tgkill(getpid(), gettid(), kSigno), SyscallSucceeds()); + + // We should now read the signal. + struct signalfd_siginfo rbuf; + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallSucceedsWithValue(sizeof(rbuf))); + EXPECT_EQ(rbuf.ssi_signo, kSigno); +} + +TEST(Signalfd, MaskWorks) { + // Create two signalfds with different masks. + sigset_t mask1, mask2; + sigemptyset(&mask1); + sigemptyset(&mask2); + sigaddset(&mask1, kSigno); + sigaddset(&mask2, kSignoAlt); + FileDescriptor fd1 = ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask1, 0)); + FileDescriptor fd2 = ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask2, 0)); + + // Deliver the two signals. + const auto scoped_sigmask1 = + ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_BLOCK, kSigno)); + const auto scoped_sigmask2 = + ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_BLOCK, kSignoAlt)); + ASSERT_THAT(tgkill(getpid(), gettid(), kSigno), SyscallSucceeds()); + ASSERT_THAT(tgkill(getpid(), gettid(), kSignoAlt), SyscallSucceeds()); + + // We should see the signals on the appropriate signalfds. + // + // We read in the opposite order as the signals deliver above, to ensure that + // we don't happen to read the correct signal from the correct signalfd. + struct signalfd_siginfo rbuf1, rbuf2; + ASSERT_THAT(read(fd2.get(), &rbuf2, sizeof(rbuf2)), + SyscallSucceedsWithValue(sizeof(rbuf2))); + EXPECT_EQ(rbuf2.ssi_signo, kSignoAlt); + ASSERT_THAT(read(fd1.get(), &rbuf1, sizeof(rbuf1)), + SyscallSucceedsWithValue(sizeof(rbuf1))); + EXPECT_EQ(rbuf1.ssi_signo, kSigno); +} + +TEST(Signalfd, Cloexec) { + // Exec tests confirm that O_CLOEXEC has the intended effect. We just create a + // signalfd with the appropriate flag here and assert that the FD has it set. + sigset_t mask; + sigemptyset(&mask); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, SFD_CLOEXEC)); + EXPECT_THAT(fcntl(fd.get(), F_GETFD), SyscallSucceedsWithValue(FD_CLOEXEC)); +} + +TEST(Signalfd, Blocking) { + // Create the signalfd in blocking mode. + sigset_t mask; + sigemptyset(&mask); + sigaddset(&mask, kSigno); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, 0)); + + // Shared tid variable. + absl::Mutex mu; + bool has_tid; + pid_t tid; + + // Start a thread reading. + ScopedThread t([&] { + // Copy the tid and notify the caller. + { + absl::MutexLock ml(&mu); + tid = gettid(); + has_tid = true; + } + + // Read the signal from the signalfd. + struct signalfd_siginfo rbuf; + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallSucceedsWithValue(sizeof(rbuf))); + EXPECT_EQ(rbuf.ssi_signo, kSigno); + }); + + // Wait until blocked. + absl::MutexLock ml(&mu); + mu.Await(absl::Condition(&has_tid)); + + // Deliver the signal to either the waiting thread, or + // to this thread. N.B. this is a bug in the core gVisor + // behavior for signalfd, and needs to be fixed. + // + // See gvisor.dev/issue/139. + if (IsRunningOnGvisor()) { + ASSERT_THAT(tgkill(getpid(), gettid(), kSigno), SyscallSucceeds()); + } else { + ASSERT_THAT(tgkill(getpid(), tid, kSigno), SyscallSucceeds()); + } + + // Ensure that it was received. + t.Join(); +} + +TEST(Signalfd, ThreadGroup) { + // Create the signalfd in blocking mode. + sigset_t mask; + sigemptyset(&mask); + sigaddset(&mask, kSigno); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, 0)); + + // Shared variable. + absl::Mutex mu; + bool first = false; + bool second = false; + + // Start a thread reading. + ScopedThread t([&] { + // Read the signal from the signalfd. + struct signalfd_siginfo rbuf; + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallSucceedsWithValue(sizeof(rbuf))); + EXPECT_EQ(rbuf.ssi_signo, kSigno); + + // Wait for the other thread. + absl::MutexLock ml(&mu); + first = true; + mu.Await(absl::Condition(&second)); + }); + + // Deliver the signal to the threadgroup. + ASSERT_THAT(kill(getpid(), kSigno), SyscallSucceeds()); + + // Wait for the first thread to process. + { + absl::MutexLock ml(&mu); + mu.Await(absl::Condition(&first)); + } + + // Deliver to the thread group again (other thread still exists). + ASSERT_THAT(kill(getpid(), kSigno), SyscallSucceeds()); + + // Ensure that we can also receive it. + struct signalfd_siginfo rbuf; + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallSucceedsWithValue(sizeof(rbuf))); + EXPECT_EQ(rbuf.ssi_signo, kSigno); + + // Mark the test as done. + { + absl::MutexLock ml(&mu); + second = true; + } + + // The other thread should be joinable. + t.Join(); +} + +TEST(Signalfd, Nonblock) { + // Create the signalfd in non-blocking mode. + sigset_t mask; + sigemptyset(&mask); + sigaddset(&mask, kSigno); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, SFD_NONBLOCK)); + + // We should return if we attempt to read. + struct signalfd_siginfo rbuf; + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallFailsWithErrno(EWOULDBLOCK)); + + // Block and deliver the signal. + const auto scoped_sigmask = + ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_BLOCK, kSigno)); + ASSERT_THAT(tgkill(getpid(), gettid(), kSigno), SyscallSucceeds()); + + // Ensure that a read actually works. + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallSucceedsWithValue(sizeof(rbuf))); + EXPECT_EQ(rbuf.ssi_signo, kSigno); + + // Should block again. + EXPECT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallFailsWithErrno(EWOULDBLOCK)); +} + +TEST(Signalfd, SetMask) { + // Create the signalfd matching nothing. + sigset_t mask; + sigemptyset(&mask); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, SFD_NONBLOCK)); + + // Block and deliver a signal. + const auto scoped_sigmask = + ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_BLOCK, kSigno)); + ASSERT_THAT(tgkill(getpid(), gettid(), kSigno), SyscallSucceeds()); + + // We should have nothing. + struct signalfd_siginfo rbuf; + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallFailsWithErrno(EWOULDBLOCK)); + + // Change the signal mask. + sigaddset(&mask, kSigno); + ASSERT_THAT(signalfd(fd.get(), &mask, 0), SyscallSucceeds()); + + // We should now have the signal. + ASSERT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallSucceedsWithValue(sizeof(rbuf))); + EXPECT_EQ(rbuf.ssi_signo, kSigno); +} + +TEST(Signalfd, Poll) { + // Create the signalfd. + sigset_t mask; + sigemptyset(&mask); + sigaddset(&mask, kSigno); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, 0)); + + // Block the signal, and start a thread to deliver it. + const auto scoped_sigmask = + ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_BLOCK, kSigno)); + pid_t orig_tid = gettid(); + ScopedThread t([&] { + absl::SleepFor(absl::Seconds(5)); + ASSERT_THAT(tgkill(getpid(), orig_tid, kSigno), SyscallSucceeds()); + }); + + // Start polling for the signal. We expect that it is not available at the + // outset, but then becomes available when the signal is sent. We give a + // timeout of 10000ms (or the delay above + 5 seconds of additional grace + // time). + struct pollfd poll_fd = {fd.get(), POLLIN, 0}; + EXPECT_THAT(RetryEINTR(poll)(&poll_fd, 1, 10000), + SyscallSucceedsWithValue(1)); + + // Actually read the signal to prevent delivery. + struct signalfd_siginfo rbuf; + EXPECT_THAT(read(fd.get(), &rbuf, sizeof(rbuf)), + SyscallSucceedsWithValue(sizeof(rbuf))); +} + +TEST(Signalfd, KillStillKills) { + sigset_t mask; + sigemptyset(&mask); + sigaddset(&mask, SIGKILL); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NewSignalFD(&mask, SFD_CLOEXEC)); + + // Just because there is a signalfd, we shouldn't see any change in behavior + // for unblockable signals. It's easier to test this with SIGKILL. + const auto scoped_sigmask = + ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_BLOCK, SIGKILL)); + EXPECT_EXIT(tgkill(getpid(), gettid(), SIGKILL), KilledBySignal(SIGKILL), ""); +} + +} // namespace + +} // namespace testing +} // namespace gvisor + +int main(int argc, char** argv) { + // These tests depend on delivering signals. Block them up front so that all + // other threads created by TestInit will also have them blocked, and they + // will not interface with the rest of the test. + sigset_t set; + sigemptyset(&set); + sigaddset(&set, gvisor::testing::kSigno); + sigaddset(&set, gvisor::testing::kSignoAlt); + TEST_PCHECK(sigprocmask(SIG_BLOCK, &set, nullptr) == 0); + + gvisor::testing::TestInit(&argc, &argv); + + return RUN_ALL_TESTS(); +} -- cgit v1.2.3 From 0a8a75f3dabaf5c097710c3bb961b67b8ed653b5 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 19 Sep 2019 11:35:27 -0700 Subject: Job control: controlling TTYs and foreground process groups. Adresses a deadlock with the rolled back change: https://github.com/google/gvisor/commit/b6a5b950d28e0b474fdad160b88bc15314cf9259 Creating a session from an orphaned process group was causing a lock to be acquired twice by a single goroutine. This behavior is addressed, and a test (OrphanRegression) has been added to pty.cc. Implemented the following ioctls: - TIOCSCTTY - set controlling TTY - TIOCNOTTY - remove controlling tty, maybe signal some other processes - TIOCGPGRP - get foreground process group. Also enables tcgetpgrp(). - TIOCSPGRP - set foreground process group. Also enabled tcsetpgrp(). Next steps are to actually turn terminal-generated control characters (e.g. C^c) into signals to the proper process groups, and to send SIGTTOU and SIGTTIN when appropriate. PiperOrigin-RevId: 270088599 --- pkg/sentry/fs/tty/BUILD | 1 + pkg/sentry/fs/tty/dir.go | 3 + pkg/sentry/fs/tty/master.go | 17 +- pkg/sentry/fs/tty/slave.go | 13 +- pkg/sentry/fs/tty/terminal.go | 92 ++++++++- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/sessions.go | 20 +- pkg/sentry/kernel/task_start.go | 3 +- pkg/sentry/kernel/thread_group.go | 179 +++++++++++++++++ pkg/sentry/kernel/tty.go | 28 +++ test/syscalls/BUILD | 4 + test/syscalls/linux/BUILD | 19 ++ test/syscalls/linux/pty.cc | 393 ++++++++++++++++++++++++++++++++++++-- test/syscalls/linux/pty_root.cc | 68 +++++++ test/util/BUILD | 11 ++ test/util/pty_util.cc | 45 +++++ test/util/pty_util.h | 30 +++ 17 files changed, 892 insertions(+), 35 deletions(-) create mode 100644 pkg/sentry/kernel/tty.go create mode 100644 test/syscalls/linux/pty_root.cc create mode 100644 test/util/pty_util.cc create mode 100644 test/util/pty_util.h (limited to 'test/syscalls/linux/BUILD') diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD index d799de748..25811f668 100644 --- a/pkg/sentry/fs/tty/BUILD +++ b/pkg/sentry/fs/tty/BUILD @@ -25,6 +25,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go index 1d128532b..2f639c823 100644 --- a/pkg/sentry/fs/tty/dir.go +++ b/pkg/sentry/fs/tty/dir.go @@ -129,6 +129,9 @@ func newDir(ctx context.Context, m *fs.MountSource) *fs.Inode { // Release implements fs.InodeOperations.Release. func (d *dirInodeOperations) Release(ctx context.Context) { + d.mu.Lock() + defer d.mu.Unlock() + d.master.DecRef() if len(d.slaves) != 0 { panic(fmt.Sprintf("devpts directory still contains active terminals: %+v", d)) diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index 92ec1ca18..19b7557d5 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -172,6 +172,19 @@ func (mf *masterFileOperations) Ioctl(ctx context.Context, _ *fs.File, io userme return 0, mf.t.ld.windowSize(ctx, io, args) case linux.TIOCSWINSZ: return 0, mf.t.ld.setWindowSize(ctx, io, args) + case linux.TIOCSCTTY: + // Make the given terminal the controlling terminal of the + // calling process. + return 0, mf.t.setControllingTTY(ctx, io, args, true /* isMaster */) + case linux.TIOCNOTTY: + // Release this process's controlling terminal. + return 0, mf.t.releaseControllingTTY(ctx, io, args, true /* isMaster */) + case linux.TIOCGPGRP: + // Get the foreground process group. + return mf.t.foregroundProcessGroup(ctx, io, args, true /* isMaster */) + case linux.TIOCSPGRP: + // Set the foreground process group. + return mf.t.setForegroundProcessGroup(ctx, io, args, true /* isMaster */) default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY @@ -185,8 +198,6 @@ func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { linux.TCSETS, linux.TCSETSW, linux.TCSETSF, - linux.TIOCGPGRP, - linux.TIOCSPGRP, linux.TIOCGWINSZ, linux.TIOCSWINSZ, linux.TIOCSETD, @@ -200,8 +211,6 @@ func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { linux.TIOCEXCL, linux.TIOCNXCL, linux.TIOCGEXCL, - linux.TIOCNOTTY, - linux.TIOCSCTTY, linux.TIOCGSID, linux.TIOCGETD, linux.TIOCVHANGUP, diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index e30266404..944c4ada1 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -152,9 +152,16 @@ func (sf *slaveFileOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem case linux.TIOCSCTTY: // Make the given terminal the controlling terminal of the // calling process. - // TODO(b/129283598): Implement once we have support for job - // control. - return 0, nil + return 0, sf.si.t.setControllingTTY(ctx, io, args, false /* isMaster */) + case linux.TIOCNOTTY: + // Release this process's controlling terminal. + return 0, sf.si.t.releaseControllingTTY(ctx, io, args, false /* isMaster */) + case linux.TIOCGPGRP: + // Get the foreground process group. + return sf.si.t.foregroundProcessGroup(ctx, io, args, false /* isMaster */) + case linux.TIOCSPGRP: + // Set the foreground process group. + return sf.si.t.setForegroundProcessGroup(ctx, io, args, false /* isMaster */) default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index b7cecb2ed..ff8138820 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -17,7 +17,10 @@ package tty import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" ) // Terminal is a pseudoterminal. @@ -26,23 +29,100 @@ import ( type Terminal struct { refs.AtomicRefCount - // n is the terminal index. + // n is the terminal index. It is immutable. n uint32 - // d is the containing directory. + // d is the containing directory. It is immutable. d *dirInodeOperations - // ld is the line discipline of the terminal. + // ld is the line discipline of the terminal. It is immutable. ld *lineDiscipline + + // masterKTTY contains the controlling process of the master end of + // this terminal. This field is immutable. + masterKTTY *kernel.TTY + + // slaveKTTY contains the controlling process of the slave end of this + // terminal. This field is immutable. + slaveKTTY *kernel.TTY } func newTerminal(ctx context.Context, d *dirInodeOperations, n uint32) *Terminal { termios := linux.DefaultSlaveTermios t := Terminal{ - d: d, - n: n, - ld: newLineDiscipline(termios), + d: d, + n: n, + ld: newLineDiscipline(termios), + masterKTTY: &kernel.TTY{}, + slaveKTTY: &kernel.TTY{}, } t.EnableLeakCheck("tty.Terminal") return &t } + +// setControllingTTY makes tm the controlling terminal of the calling thread +// group. +func (tm *Terminal) setControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("setControllingTTY must be called from a task context") + } + + return task.ThreadGroup().SetControllingTTY(tm.tty(isMaster), args[2].Int()) +} + +// releaseControllingTTY removes tm as the controlling terminal of the calling +// thread group. +func (tm *Terminal) releaseControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("releaseControllingTTY must be called from a task context") + } + + return task.ThreadGroup().ReleaseControllingTTY(tm.tty(isMaster)) +} + +// foregroundProcessGroup gets the process group ID of tm's foreground process. +func (tm *Terminal) foregroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("foregroundProcessGroup must be called from a task context") + } + + ret, err := task.ThreadGroup().ForegroundProcessGroup(tm.tty(isMaster)) + if err != nil { + return 0, err + } + + // Write it out to *arg. + _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(ret), usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err +} + +// foregroundProcessGroup sets tm's foreground process. +func (tm *Terminal) setForegroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("setForegroundProcessGroup must be called from a task context") + } + + // Read in the process group ID. + var pgid int32 + if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgid, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + + ret, err := task.ThreadGroup().SetForegroundProcessGroup(tm.tty(isMaster), kernel.ProcessGroupID(pgid)) + return uintptr(ret), err +} + +func (tm *Terminal) tty(isMaster bool) *kernel.TTY { + if isMaster { + return tm.masterKTTY + } + return tm.slaveKTTY +} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index e964a991b..eaccfd02d 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -145,6 +145,7 @@ go_library( "threads.go", "timekeeper.go", "timekeeper_state.go", + "tty.go", "uts_namespace.go", "vdso.go", "version.go", diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go index 81fcd8258..047b5214d 100644 --- a/pkg/sentry/kernel/sessions.go +++ b/pkg/sentry/kernel/sessions.go @@ -47,6 +47,11 @@ type Session struct { // The id is immutable. id SessionID + // foreground is the foreground process group. + // + // This is protected by TaskSet.mu. + foreground *ProcessGroup + // ProcessGroups is a list of process groups in this Session. This is // protected by TaskSet.mu. processGroups processGroupList @@ -260,12 +265,14 @@ func (pg *ProcessGroup) SendSignal(info *arch.SignalInfo) error { func (tg *ThreadGroup) CreateSession() error { tg.pidns.owner.mu.Lock() defer tg.pidns.owner.mu.Unlock() + tg.signalHandlers.mu.Lock() + defer tg.signalHandlers.mu.Unlock() return tg.createSession() } // createSession creates a new session for a threadgroup. // -// Precondition: callers must hold TaskSet.mu for writing. +// Precondition: callers must hold TaskSet.mu and the signal mutex for writing. func (tg *ThreadGroup) createSession() error { // Get the ID for this thread in the current namespace. id := tg.pidns.tgids[tg] @@ -321,8 +328,14 @@ func (tg *ThreadGroup) createSession() error { childTG.processGroup.incRefWithParent(pg) childTG.processGroup.decRefWithParent(oldParentPG) }) - tg.processGroup.decRefWithParent(oldParentPG) + // If tg.processGroup is an orphan, decRefWithParent will lock + // the signal mutex of each thread group in tg.processGroup. + // However, tg's signal mutex may already be locked at this + // point. We change tg's process group before calling + // decRefWithParent to avoid locking tg's signal mutex twice. + oldPG := tg.processGroup tg.processGroup = pg + oldPG.decRefWithParent(oldParentPG) } else { // The current process group may be nil only in the case of an // unparented thread group (i.e. the init process). This would @@ -346,6 +359,9 @@ func (tg *ThreadGroup) createSession() error { ns.processGroups[ProcessGroupID(local)] = pg } + // Disconnect from the controlling terminal. + tg.tty = nil + return nil } diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index d60cd62c7..ae6fc4025 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -172,9 +172,10 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { if parentPG := tg.parentPG(); parentPG == nil { tg.createSession() } else { - // Inherit the process group. + // Inherit the process group and terminal. parentPG.incRefWithParent(parentPG) tg.processGroup = parentPG + tg.tty = t.parent.tg.tty } } tg.tasks.PushBack(t) diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 2a97e3e8e..0eef24bfb 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -19,10 +19,13 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/syserror" ) // A ThreadGroup is a logical grouping of tasks that has widespread @@ -245,6 +248,12 @@ type ThreadGroup struct { // // mounts is immutable. mounts *fs.MountNamespace + + // tty is the thread group's controlling terminal. If nil, there is no + // controlling terminal. + // + // tty is protected by the signal mutex. + tty *TTY } // newThreadGroup returns a new, empty thread group in PID namespace ns. The @@ -324,6 +333,176 @@ func (tg *ThreadGroup) forEachChildThreadGroupLocked(fn func(*ThreadGroup)) { } } +// SetControllingTTY sets tty as the controlling terminal of tg. +func (tg *ThreadGroup) SetControllingTTY(tty *TTY, arg int32) error { + tty.mu.Lock() + defer tty.mu.Unlock() + + // We might be asked to set the controlling terminal of multiple + // processes, so we lock both the TaskSet and SignalHandlers. + tg.pidns.owner.mu.Lock() + defer tg.pidns.owner.mu.Unlock() + tg.signalHandlers.mu.Lock() + defer tg.signalHandlers.mu.Unlock() + + // "The calling process must be a session leader and not have a + // controlling terminal already." - tty_ioctl(4) + if tg.processGroup.session.leader != tg || tg.tty != nil { + return syserror.EINVAL + } + + // "If this terminal is already the controlling terminal of a different + // session group, then the ioctl fails with EPERM, unless the caller + // has the CAP_SYS_ADMIN capability and arg equals 1, in which case the + // terminal is stolen, and all processes that had it as controlling + // terminal lose it." - tty_ioctl(4) + if tty.tg != nil && tg.processGroup.session != tty.tg.processGroup.session { + if !auth.CredentialsFromContext(tg.leader).HasCapability(linux.CAP_SYS_ADMIN) || arg != 1 { + return syserror.EPERM + } + // Steal the TTY away. Unlike TIOCNOTTY, don't send signals. + for othertg := range tg.pidns.owner.Root.tgids { + // This won't deadlock by locking tg.signalHandlers + // because at this point: + // - We only lock signalHandlers if it's in the same + // session as the tty's controlling thread group. + // - We know that the calling thread group is not in + // the same session as the tty's controlling thread + // group. + if othertg.processGroup.session == tty.tg.processGroup.session { + othertg.signalHandlers.mu.Lock() + othertg.tty = nil + othertg.signalHandlers.mu.Unlock() + } + } + } + + // Set the controlling terminal and foreground process group. + tg.tty = tty + tg.processGroup.session.foreground = tg.processGroup + // Set this as the controlling process of the terminal. + tty.tg = tg + + return nil +} + +// ReleaseControllingTTY gives up tty as the controlling tty of tg. +func (tg *ThreadGroup) ReleaseControllingTTY(tty *TTY) error { + tty.mu.Lock() + defer tty.mu.Unlock() + + // We might be asked to set the controlling terminal of multiple + // processes, so we lock both the TaskSet and SignalHandlers. + tg.pidns.owner.mu.RLock() + defer tg.pidns.owner.mu.RUnlock() + + // Just below, we may re-lock signalHandlers in order to send signals. + // Thus we can't defer Unlock here. + tg.signalHandlers.mu.Lock() + + if tg.tty == nil || tg.tty != tty { + tg.signalHandlers.mu.Unlock() + return syserror.ENOTTY + } + + // "If the process was session leader, then send SIGHUP and SIGCONT to + // the foreground process group and all processes in the current + // session lose their controlling terminal." - tty_ioctl(4) + // Remove tty as the controlling tty for each process in the session, + // then send them SIGHUP and SIGCONT. + + // If we're not the session leader, we don't have to do much. + if tty.tg != tg { + tg.tty = nil + tg.signalHandlers.mu.Unlock() + return nil + } + + tg.signalHandlers.mu.Unlock() + + // We're the session leader. SIGHUP and SIGCONT the foreground process + // group and remove all controlling terminals in the session. + var lastErr error + for othertg := range tg.pidns.owner.Root.tgids { + if othertg.processGroup.session == tg.processGroup.session { + othertg.signalHandlers.mu.Lock() + othertg.tty = nil + if othertg.processGroup == tg.processGroup.session.foreground { + if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGHUP)}, true /* group */); err != nil { + lastErr = err + } + if err := othertg.leader.sendSignalLocked(&arch.SignalInfo{Signo: int32(linux.SIGCONT)}, true /* group */); err != nil { + lastErr = err + } + } + othertg.signalHandlers.mu.Unlock() + } + } + + return lastErr +} + +// ForegroundProcessGroup returns the process group ID of the foreground +// process group. +func (tg *ThreadGroup) ForegroundProcessGroup(tty *TTY) (int32, error) { + tty.mu.Lock() + defer tty.mu.Unlock() + + tg.pidns.owner.mu.Lock() + defer tg.pidns.owner.mu.Unlock() + tg.signalHandlers.mu.Lock() + defer tg.signalHandlers.mu.Unlock() + + // "When fd does not refer to the controlling terminal of the calling + // process, -1 is returned" - tcgetpgrp(3) + if tg.tty != tty { + return -1, syserror.ENOTTY + } + + return int32(tg.processGroup.session.foreground.id), nil +} + +// SetForegroundProcessGroup sets the foreground process group of tty to pgid. +func (tg *ThreadGroup) SetForegroundProcessGroup(tty *TTY, pgid ProcessGroupID) (int32, error) { + tty.mu.Lock() + defer tty.mu.Unlock() + + tg.pidns.owner.mu.Lock() + defer tg.pidns.owner.mu.Unlock() + tg.signalHandlers.mu.Lock() + defer tg.signalHandlers.mu.Unlock() + + // TODO(b/129283598): "If tcsetpgrp() is called by a member of a + // background process group in its session, and the calling process is + // not blocking or ignoring SIGTTOU, a SIGTTOU signal is sent to all + // members of this background process group." + + // tty must be the controlling terminal. + if tg.tty != tty { + return -1, syserror.ENOTTY + } + + // pgid must be positive. + if pgid < 0 { + return -1, syserror.EINVAL + } + + // pg must not be empty. Empty process groups are removed from their + // pid namespaces. + pg, ok := tg.pidns.processGroups[pgid] + if !ok { + return -1, syserror.ESRCH + } + + // pg must be part of this process's session. + if tg.processGroup.session != pg.session { + return -1, syserror.EPERM + } + + tg.processGroup.session.foreground.id = pgid + return 0, nil +} + // itimerRealListener implements ktime.Listener for ITIMER_REAL expirations. // // +stateify savable diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go new file mode 100644 index 000000000..34f84487a --- /dev/null +++ b/pkg/sentry/kernel/tty.go @@ -0,0 +1,28 @@ +// 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 kernel + +import "sync" + +// TTY defines the relationship between a thread group and its controlling +// terminal. +// +// +stateify savable +type TTY struct { + mu sync.Mutex `state:"nosave"` + + // tg is protected by mu. + tg *ThreadGroup +} diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 0135435ea..63e4c63dd 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -320,6 +320,10 @@ syscall_test( test = "//test/syscalls/linux:pty_test", ) +syscall_test( + test = "//test/syscalls/linux:pty_root_test", +) + syscall_test( add_overlay = True, test = "//test/syscalls/linux:pwritev2_test", diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 56fe7be37..a4cebf46f 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1287,8 +1287,10 @@ cc_binary( srcs = ["pty.cc"], linkstatic = 1, deps = [ + "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:posix_error", + "//test/util:pty_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", @@ -1300,6 +1302,23 @@ cc_binary( ], ) +cc_binary( + name = "pty_root_test", + testonly = 1, + srcs = ["pty_root.cc"], + linkstatic = 1, + deps = [ + "//test/util:capability_util", + "//test/util:file_descriptor", + "//test/util:posix_error", + "//test/util:pty_util", + "//test/util:test_main", + "//test/util:thread_util", + "@com_google_absl//absl/base:core_headers", + "@com_google_googletest//:gtest", + ], +) + cc_binary( name = "partial_bad_buffer_test", testonly = 1, diff --git a/test/syscalls/linux/pty.cc b/test/syscalls/linux/pty.cc index d1ab4703f..286388316 100644 --- a/test/syscalls/linux/pty.cc +++ b/test/syscalls/linux/pty.cc @@ -13,13 +13,17 @@ // limitations under the License. #include +#include #include #include +#include +#include #include #include #include #include #include +#include #include #include @@ -31,8 +35,10 @@ #include "absl/synchronization/notification.h" #include "absl/time/clock.h" #include "absl/time/time.h" +#include "test/util/capability_util.h" #include "test/util/file_descriptor.h" #include "test/util/posix_error.h" +#include "test/util/pty_util.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" @@ -370,25 +376,6 @@ PosixErrorOr PollAndReadFd(int fd, void* buf, size_t count, return PosixError(ETIMEDOUT, "Poll timed out"); } -// Opens the slave end of the passed master as R/W and nonblocking. -PosixErrorOr OpenSlave(const FileDescriptor& master) { - // Get pty index. - int n; - int ret = ioctl(master.get(), TIOCGPTN, &n); - if (ret < 0) { - return PosixError(errno, "ioctl(TIOCGPTN) failed"); - } - - // Unlock pts. - int unlock = 0; - ret = ioctl(master.get(), TIOCSPTLCK, &unlock); - if (ret < 0) { - return PosixError(errno, "ioctl(TIOSPTLCK) failed"); - } - - return Open(absl::StrCat("/dev/pts/", n), O_RDWR | O_NONBLOCK); -} - TEST(BasicPtyTest, StatUnopenedMaster) { struct stat s; ASSERT_THAT(stat("/dev/ptmx", &s), SyscallSucceeds()); @@ -1233,6 +1220,374 @@ TEST_F(PtyTest, SetMasterWindowSize) { EXPECT_EQ(retrieved_ws.ws_col, kCols); } +class JobControlTest : public ::testing::Test { + protected: + void SetUp() override { + master_ = ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/ptmx", O_RDWR | O_NONBLOCK)); + slave_ = ASSERT_NO_ERRNO_AND_VALUE(OpenSlave(master_)); + + // Make this a session leader, which also drops the controlling terminal. + // In the gVisor test environment, this test will be run as the session + // leader already (as the sentry init process). + if (!IsRunningOnGvisor()) { + ASSERT_THAT(setsid(), SyscallSucceeds()); + } + } + + // Master and slave ends of the PTY. Non-blocking. + FileDescriptor master_; + FileDescriptor slave_; +}; + +TEST_F(JobControlTest, SetTTYMaster) { + ASSERT_THAT(ioctl(master_.get(), TIOCSCTTY, 0), SyscallSucceeds()); +} + +TEST_F(JobControlTest, SetTTY) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); +} + +TEST_F(JobControlTest, SetTTYNonLeader) { + // Fork a process that won't be the session leader. + pid_t child = fork(); + if (!child) { + // We shouldn't be able to set the terminal. + TEST_PCHECK(ioctl(slave_.get(), TIOCSCTTY, 0)); + _exit(0); + } + + int wstatus; + ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + ASSERT_EQ(wstatus, 0); +} + +TEST_F(JobControlTest, SetTTYBadArg) { + // Despite the man page saying arg should be 0 here, Linux doesn't actually + // check. + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 1), SyscallSucceeds()); +} + +TEST_F(JobControlTest, SetTTYDifferentSession) { + SKIP_IF(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + // Fork, join a new session, and try to steal the parent's controlling + // terminal, which should fail. + pid_t child = fork(); + if (!child) { + TEST_PCHECK(setsid() >= 0); + // We shouldn't be able to steal the terminal. + TEST_PCHECK(ioctl(slave_.get(), TIOCSCTTY, 1)); + _exit(0); + } + + int wstatus; + ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + ASSERT_EQ(wstatus, 0); +} + +TEST_F(JobControlTest, ReleaseTTY) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + // Make sure we're ignoring SIGHUP, which will be sent to this process once we + // disconnect they TTY. + struct sigaction sa = { + .sa_handler = SIG_IGN, + .sa_flags = 0, + }; + sigemptyset(&sa.sa_mask); + struct sigaction old_sa; + EXPECT_THAT(sigaction(SIGHUP, &sa, &old_sa), SyscallSucceeds()); + EXPECT_THAT(ioctl(slave_.get(), TIOCNOTTY), SyscallSucceeds()); + EXPECT_THAT(sigaction(SIGHUP, &old_sa, NULL), SyscallSucceeds()); +} + +TEST_F(JobControlTest, ReleaseUnsetTTY) { + ASSERT_THAT(ioctl(slave_.get(), TIOCNOTTY), SyscallFailsWithErrno(ENOTTY)); +} + +TEST_F(JobControlTest, ReleaseWrongTTY) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + ASSERT_THAT(ioctl(master_.get(), TIOCNOTTY), SyscallFailsWithErrno(ENOTTY)); +} + +TEST_F(JobControlTest, ReleaseTTYNonLeader) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + pid_t child = fork(); + if (!child) { + TEST_PCHECK(!ioctl(slave_.get(), TIOCNOTTY)); + _exit(0); + } + + int wstatus; + ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + ASSERT_EQ(wstatus, 0); +} + +TEST_F(JobControlTest, ReleaseTTYDifferentSession) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + pid_t child = fork(); + if (!child) { + // Join a new session, then try to disconnect. + TEST_PCHECK(setsid() >= 0); + TEST_PCHECK(ioctl(slave_.get(), TIOCNOTTY)); + _exit(0); + } + + int wstatus; + ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + ASSERT_EQ(wstatus, 0); +} + +// Used by the child process spawned in ReleaseTTYSignals to track received +// signals. +static int received; + +void sig_handler(int signum) { received |= signum; } + +// When the session leader releases its controlling terminal, the foreground +// process group gets SIGHUP, then SIGCONT. This test: +// - Spawns 2 threads +// - Has thread 1 return 0 if it gets both SIGHUP and SIGCONT +// - Has thread 2 leave the foreground process group, and return non-zero if it +// receives any signals. +// - Has the parent thread release its controlling terminal +// - Checks that thread 1 got both signals +// - Checks that thread 2 didn't get any signals. +TEST_F(JobControlTest, ReleaseTTYSignals) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + received = 0; + struct sigaction sa = { + .sa_handler = sig_handler, + .sa_flags = 0, + }; + sigemptyset(&sa.sa_mask); + sigaddset(&sa.sa_mask, SIGHUP); + sigaddset(&sa.sa_mask, SIGCONT); + sigprocmask(SIG_BLOCK, &sa.sa_mask, NULL); + + pid_t same_pgrp_child = fork(); + if (!same_pgrp_child) { + // The child will wait for SIGHUP and SIGCONT, then return 0. It begins with + // SIGHUP and SIGCONT blocked. We install signal handlers for those signals, + // then use sigsuspend to wait for those specific signals. + TEST_PCHECK(!sigaction(SIGHUP, &sa, NULL)); + TEST_PCHECK(!sigaction(SIGCONT, &sa, NULL)); + sigset_t mask; + sigfillset(&mask); + sigdelset(&mask, SIGHUP); + sigdelset(&mask, SIGCONT); + while (received != (SIGHUP | SIGCONT)) { + sigsuspend(&mask); + } + _exit(0); + } + + // We don't want to block these anymore. + sigprocmask(SIG_UNBLOCK, &sa.sa_mask, NULL); + + // This child will return non-zero if either SIGHUP or SIGCONT are received. + pid_t diff_pgrp_child = fork(); + if (!diff_pgrp_child) { + TEST_PCHECK(!setpgid(0, 0)); + TEST_PCHECK(pause()); + _exit(1); + } + + EXPECT_THAT(setpgid(diff_pgrp_child, diff_pgrp_child), SyscallSucceeds()); + + // Make sure we're ignoring SIGHUP, which will be sent to this process once we + // disconnect they TTY. + struct sigaction sighup_sa = { + .sa_handler = SIG_IGN, + .sa_flags = 0, + }; + sigemptyset(&sighup_sa.sa_mask); + struct sigaction old_sa; + EXPECT_THAT(sigaction(SIGHUP, &sighup_sa, &old_sa), SyscallSucceeds()); + + // Release the controlling terminal, sending SIGHUP and SIGCONT to all other + // processes in this process group. + EXPECT_THAT(ioctl(slave_.get(), TIOCNOTTY), SyscallSucceeds()); + + EXPECT_THAT(sigaction(SIGHUP, &old_sa, NULL), SyscallSucceeds()); + + // The child in the same process group will get signaled. + int wstatus; + EXPECT_THAT(waitpid(same_pgrp_child, &wstatus, 0), + SyscallSucceedsWithValue(same_pgrp_child)); + EXPECT_EQ(wstatus, 0); + + // The other child will not get signaled. + EXPECT_THAT(waitpid(diff_pgrp_child, &wstatus, WNOHANG), + SyscallSucceedsWithValue(0)); + EXPECT_THAT(kill(diff_pgrp_child, SIGKILL), SyscallSucceeds()); +} + +TEST_F(JobControlTest, GetForegroundProcessGroup) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + pid_t foreground_pgid; + pid_t pid; + ASSERT_THAT(ioctl(slave_.get(), TIOCGPGRP, &foreground_pgid), + SyscallSucceeds()); + ASSERT_THAT(pid = getpid(), SyscallSucceeds()); + + ASSERT_EQ(foreground_pgid, pid); +} + +TEST_F(JobControlTest, GetForegroundProcessGroupNonControlling) { + // At this point there's no controlling terminal, so TIOCGPGRP should fail. + pid_t foreground_pgid; + ASSERT_THAT(ioctl(slave_.get(), TIOCGPGRP, &foreground_pgid), + SyscallFailsWithErrno(ENOTTY)); +} + +// This test: +// - sets itself as the foreground process group +// - creates a child process in a new process group +// - sets that child as the foreground process group +// - kills its child and sets itself as the foreground process group. +TEST_F(JobControlTest, SetForegroundProcessGroup) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + // Ignore SIGTTOU so that we don't stop ourself when calling tcsetpgrp. + struct sigaction sa = { + .sa_handler = SIG_IGN, + .sa_flags = 0, + }; + sigemptyset(&sa.sa_mask); + sigaction(SIGTTOU, &sa, NULL); + + // Set ourself as the foreground process group. + ASSERT_THAT(tcsetpgrp(slave_.get(), getpgid(0)), SyscallSucceeds()); + + // Create a new process that just waits to be signaled. + pid_t child = fork(); + if (!child) { + TEST_PCHECK(!pause()); + // We should never reach this. + _exit(1); + } + + // Make the child its own process group, then make it the controlling process + // group of the terminal. + ASSERT_THAT(setpgid(child, child), SyscallSucceeds()); + ASSERT_THAT(tcsetpgrp(slave_.get(), child), SyscallSucceeds()); + + // Sanity check - we're still the controlling session. + ASSERT_EQ(getsid(0), getsid(child)); + + // Signal the child, wait for it to exit, then retake the terminal. + ASSERT_THAT(kill(child, SIGTERM), SyscallSucceeds()); + int wstatus; + ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + ASSERT_TRUE(WIFSIGNALED(wstatus)); + ASSERT_EQ(WTERMSIG(wstatus), SIGTERM); + + // Set ourself as the foreground process. + pid_t pgid; + ASSERT_THAT(pgid = getpgid(0), SyscallSucceeds()); + ASSERT_THAT(tcsetpgrp(slave_.get(), pgid), SyscallSucceeds()); +} + +TEST_F(JobControlTest, SetForegroundProcessGroupWrongTTY) { + pid_t pid = getpid(); + ASSERT_THAT(ioctl(slave_.get(), TIOCSPGRP, &pid), + SyscallFailsWithErrno(ENOTTY)); +} + +TEST_F(JobControlTest, SetForegroundProcessGroupNegPgid) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + pid_t pid = -1; + ASSERT_THAT(ioctl(slave_.get(), TIOCSPGRP, &pid), + SyscallFailsWithErrno(EINVAL)); +} + +TEST_F(JobControlTest, SetForegroundProcessGroupEmptyProcessGroup) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + // Create a new process, put it in a new process group, make that group the + // foreground process group, then have the process wait. + pid_t child = fork(); + if (!child) { + TEST_PCHECK(!setpgid(0, 0)); + _exit(0); + } + + // Wait for the child to exit. + int wstatus; + EXPECT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + // The child's process group doesn't exist anymore - this should fail. + ASSERT_THAT(ioctl(slave_.get(), TIOCSPGRP, &child), + SyscallFailsWithErrno(ESRCH)); +} + +TEST_F(JobControlTest, SetForegroundProcessGroupDifferentSession) { + ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + // Create a new process and put it in a new session. + pid_t child = fork(); + if (!child) { + TEST_PCHECK(setsid() >= 0); + // Tell the parent we're in a new session. + TEST_PCHECK(!raise(SIGSTOP)); + TEST_PCHECK(!pause()); + _exit(1); + } + + // Wait for the child to tell us it's in a new session. + int wstatus; + EXPECT_THAT(waitpid(child, &wstatus, WUNTRACED), + SyscallSucceedsWithValue(child)); + EXPECT_TRUE(WSTOPSIG(wstatus)); + + // Child is in a new session, so we can't make it the foregroup process group. + EXPECT_THAT(ioctl(slave_.get(), TIOCSPGRP, &child), + SyscallFailsWithErrno(EPERM)); + + EXPECT_THAT(kill(child, SIGKILL), SyscallSucceeds()); +} + +// Verify that we don't hang when creating a new session from an orphaned +// process group (b/139968068). Calling setsid() creates an orphaned process +// group, as process groups that contain the session's leading process are +// orphans. +// +// We create 2 sessions in this test. The init process in gVisor is considered +// not to be an orphan (see sessions.go), so we have to create a session from +// which to create a session. The latter session is being created from an +// orphaned process group. +TEST_F(JobControlTest, OrphanRegression) { + pid_t session_2_leader = fork(); + if (!session_2_leader) { + TEST_PCHECK(setsid() >= 0); + + pid_t session_3_leader = fork(); + if (!session_3_leader) { + TEST_PCHECK(setsid() >= 0); + + _exit(0); + } + + int wstatus; + TEST_PCHECK(waitpid(session_3_leader, &wstatus, 0) == session_3_leader); + TEST_PCHECK(wstatus == 0); + + _exit(0); + } + + int wstatus; + ASSERT_THAT(waitpid(session_2_leader, &wstatus, 0), + SyscallSucceedsWithValue(session_2_leader)); + ASSERT_EQ(wstatus, 0); +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/pty_root.cc b/test/syscalls/linux/pty_root.cc new file mode 100644 index 000000000..14a4af980 --- /dev/null +++ b/test/syscalls/linux/pty_root.cc @@ -0,0 +1,68 @@ +// 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. + +#include +#include + +#include "gtest/gtest.h" +#include "absl/base/macros.h" +#include "test/util/capability_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" +#include "test/util/pty_util.h" + +namespace gvisor { +namespace testing { + +// These tests should be run as root. +namespace { + +TEST(JobControlRootTest, StealTTY) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + // Make this a session leader, which also drops the controlling terminal. + // In the gVisor test environment, this test will be run as the session + // leader already (as the sentry init process). + if (!IsRunningOnGvisor()) { + ASSERT_THAT(setsid(), SyscallSucceeds()); + } + + FileDescriptor master = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/ptmx", O_RDWR | O_NONBLOCK)); + FileDescriptor slave = ASSERT_NO_ERRNO_AND_VALUE(OpenSlave(master)); + + // Make slave the controlling terminal. + ASSERT_THAT(ioctl(slave.get(), TIOCSCTTY, 0), SyscallSucceeds()); + + // Fork, join a new session, and try to steal the parent's controlling + // terminal, which should succeed when we have CAP_SYS_ADMIN and pass an arg + // of 1. + pid_t child = fork(); + if (!child) { + ASSERT_THAT(setsid(), SyscallSucceeds()); + // We shouldn't be able to steal the terminal with the wrong arg value. + TEST_PCHECK(ioctl(slave.get(), TIOCSCTTY, 0)); + // We should be able to steal it here. + TEST_PCHECK(!ioctl(slave.get(), TIOCSCTTY, 1)); + _exit(0); + } + + int wstatus; + ASSERT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + ASSERT_EQ(wstatus, 0); +} + +} // namespace +} // namespace testing +} // namespace gvisor diff --git a/test/util/BUILD b/test/util/BUILD index 52f8b9e1f..25ed9c944 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -190,6 +190,17 @@ cc_test( ], ) +cc_library( + name = "pty_util", + testonly = 1, + srcs = ["pty_util.cc"], + hdrs = ["pty_util.h"], + deps = [ + ":file_descriptor", + ":posix_error", + ], +) + cc_library( name = "signal_util", testonly = 1, diff --git a/test/util/pty_util.cc b/test/util/pty_util.cc new file mode 100644 index 000000000..c0fd9a095 --- /dev/null +++ b/test/util/pty_util.cc @@ -0,0 +1,45 @@ +// Copyright 2019 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. + +#include "test/util/pty_util.h" + +#include +#include + +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" + +namespace gvisor { +namespace testing { + +PosixErrorOr OpenSlave(const FileDescriptor& master) { + // Get pty index. + int n; + int ret = ioctl(master.get(), TIOCGPTN, &n); + if (ret < 0) { + return PosixError(errno, "ioctl(TIOCGPTN) failed"); + } + + // Unlock pts. + int unlock = 0; + ret = ioctl(master.get(), TIOCSPTLCK, &unlock); + if (ret < 0) { + return PosixError(errno, "ioctl(TIOSPTLCK) failed"); + } + + return Open(absl::StrCat("/dev/pts/", n), O_RDWR | O_NONBLOCK); +} + +} // namespace testing +} // namespace gvisor diff --git a/test/util/pty_util.h b/test/util/pty_util.h new file mode 100644 index 000000000..367b14f15 --- /dev/null +++ b/test/util/pty_util.h @@ -0,0 +1,30 @@ +// Copyright 2019 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. + +#ifndef GVISOR_TEST_UTIL_PTY_UTIL_H_ +#define GVISOR_TEST_UTIL_PTY_UTIL_H_ + +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" + +namespace gvisor { +namespace testing { + +// Opens the slave end of the passed master as R/W and nonblocking. +PosixErrorOr OpenSlave(const FileDescriptor& master); + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_UTIL_PTY_UTIL_H_ -- cgit v1.2.3 From 502f8f238ea58c4828e528e563d8dbd419faeea7 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 24 Sep 2019 13:25:25 -0700 Subject: Stub out readahead implementation. Closes #261 PiperOrigin-RevId: 270973347 --- pkg/sentry/syscalls/linux/linux64.go | 2 +- pkg/sentry/syscalls/linux/sys_read.go | 33 +++++++++++++ test/syscalls/BUILD | 5 ++ test/syscalls/linux/BUILD | 14 ++++++ test/syscalls/linux/readahead.cc | 91 +++++++++++++++++++++++++++++++++++ 5 files changed, 144 insertions(+), 1 deletion(-) create mode 100644 test/syscalls/linux/readahead.cc (limited to 'test/syscalls/linux/BUILD') diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index 18d24ab61..61acd0abd 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -232,7 +232,7 @@ var AMD64 = &kernel.SyscallTable{ 184: syscalls.Error("tuxcall", syserror.ENOSYS, "Not implemented in Linux.", nil), 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), - 187: syscalls.ErrorWithEvent("readahead", syserror.ENOSYS, "", []string{"gvisor.dev/issue/261"}), // TODO(b/29351341) + 187: syscalls.Supported("readahead", Readahead), 188: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index 3ab54271c..cd31e0649 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -72,6 +72,39 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "read", file) } +// Readahead implements readahead(2). +func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + offset := args[1].Int64() + size := args[2].SizeT() + + file := t.GetFile(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the file is readable. + if !file.Flags().Read { + return 0, nil, syserror.EBADF + } + + // Check that the size is valid. + if int(size) < 0 { + return 0, nil, syserror.EINVAL + } + + // Check that the offset is legitimate. + if offset < 0 { + return 0, nil, syserror.EINVAL + } + + // Return EINVAL; if the underlying file type does not support readahead, + // then Linux will return EINVAL to indicate as much. In the future, we + // may extend this function to actually support readahead hints. + return 0, nil, syserror.EINVAL +} + // Pread64 implements linux syscall pread64(2). func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 63e4c63dd..341e6b252 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -345,6 +345,11 @@ syscall_test( test = "//test/syscalls/linux:read_test", ) +syscall_test( + add_overlay = True, + test = "//test/syscalls/linux:readahead_test", +) + syscall_test( size = "medium", shard_count = 5, diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index a4cebf46f..28b23ce58 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1735,6 +1735,20 @@ cc_binary( ], ) +cc_binary( + name = "readahead_test", + testonly = 1, + srcs = ["readahead.cc"], + linkstatic = 1, + deps = [ + "//test/util:file_descriptor", + "//test/util:temp_path", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_googletest//:gtest", + ], +) + cc_binary( name = "readv_test", testonly = 1, diff --git a/test/syscalls/linux/readahead.cc b/test/syscalls/linux/readahead.cc new file mode 100644 index 000000000..09703b5c1 --- /dev/null +++ b/test/syscalls/linux/readahead.cc @@ -0,0 +1,91 @@ +// Copyright 2019 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. + +#include +#include + +#include "gtest/gtest.h" +#include "test/util/file_descriptor.h" +#include "test/util/temp_path.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +TEST(ReadaheadTest, InvalidFD) { + EXPECT_THAT(readahead(-1, 1, 1), SyscallFailsWithErrno(EBADF)); +} + +TEST(ReadaheadTest, InvalidOffset) { + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); + EXPECT_THAT(readahead(fd.get(), -1, 1), SyscallFailsWithErrno(EINVAL)); +} + +TEST(ReadaheadTest, ValidOffset) { + constexpr char kData[] = "123"; + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + GetAbsoluteTestTmpdir(), kData, TempPath::kDefaultFileMode)); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); + + // N.B. The implementation of readahead is filesystem-specific, and a file + // backed by ram may return EINVAL because there is nothing to be read. + EXPECT_THAT(readahead(fd.get(), 1, 1), AnyOf(SyscallSucceedsWithValue(0), + SyscallFailsWithErrno(EINVAL))); +} + +TEST(ReadaheadTest, PastEnd) { + constexpr char kData[] = "123"; + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + GetAbsoluteTestTmpdir(), kData, TempPath::kDefaultFileMode)); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); + // See above. + EXPECT_THAT(readahead(fd.get(), 2, 2), AnyOf(SyscallSucceedsWithValue(0), + SyscallFailsWithErrno(EINVAL))); +} + +TEST(ReadaheadTest, CrossesEnd) { + constexpr char kData[] = "123"; + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + GetAbsoluteTestTmpdir(), kData, TempPath::kDefaultFileMode)); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); + // See above. + EXPECT_THAT(readahead(fd.get(), 4, 2), AnyOf(SyscallSucceedsWithValue(0), + SyscallFailsWithErrno(EINVAL))); +} + +TEST(ReadaheadTest, WriteOnly) { + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_WRONLY)); + EXPECT_THAT(readahead(fd.get(), 0, 1), SyscallFailsWithErrno(EBADF)); +} + +TEST(ReadaheadTest, InvalidSize) { + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDWR)); + EXPECT_THAT(readahead(fd.get(), 0, -1), SyscallFailsWithErrno(EINVAL)); +} + +} // namespace + +} // namespace testing +} // namespace gvisor -- cgit v1.2.3 From abbee5615f4480d8a41b4cf63839d2ab13b19abf Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 27 Sep 2019 14:12:35 -0700 Subject: Implement SO_BINDTODEVICE sockopt PiperOrigin-RevId: 271644926 --- pkg/sentry/socket/epsocket/epsocket.go | 20 ++ pkg/sentry/syscalls/linux/sys_socket.go | 2 +- pkg/tcpip/ports/ports.go | 114 ++++-- pkg/tcpip/ports/ports_test.go | 113 +++++- pkg/tcpip/stack/BUILD | 4 + pkg/tcpip/stack/nic.go | 15 +- pkg/tcpip/stack/stack.go | 58 +--- pkg/tcpip/stack/transport_demuxer.go | 227 +++++++----- pkg/tcpip/stack/transport_demuxer_test.go | 352 +++++++++++++++++++ pkg/tcpip/stack/transport_test.go | 5 +- pkg/tcpip/tcpip.go | 4 + pkg/tcpip/transport/icmp/endpoint.go | 6 +- pkg/tcpip/transport/tcp/accept.go | 2 +- pkg/tcpip/transport/tcp/endpoint.go | 56 ++- pkg/tcpip/transport/tcp/tcp_test.go | 116 +++++++ pkg/tcpip/transport/tcp/testing/context/context.go | 26 +- pkg/tcpip/transport/udp/BUILD | 1 + pkg/tcpip/transport/udp/endpoint.go | 42 ++- pkg/tcpip/transport/udp/forwarder.go | 2 +- pkg/tcpip/transport/udp/udp_test.go | 120 +++---- test/syscalls/linux/BUILD | 75 ++++ test/syscalls/linux/socket_bind_to_device.cc | 314 +++++++++++++++++ .../linux/socket_bind_to_device_distribution.cc | 381 +++++++++++++++++++++ .../linux/socket_bind_to_device_sequence.cc | 316 +++++++++++++++++ test/syscalls/linux/socket_bind_to_device_util.cc | 75 ++++ test/syscalls/linux/socket_bind_to_device_util.h | 67 ++++ test/syscalls/linux/uidgid.cc | 25 +- test/util/BUILD | 11 + test/util/uid_util.cc | 44 +++ test/util/uid_util.h | 29 ++ 30 files changed, 2308 insertions(+), 314 deletions(-) create mode 100644 pkg/tcpip/stack/transport_demuxer_test.go create mode 100644 test/syscalls/linux/socket_bind_to_device.cc create mode 100644 test/syscalls/linux/socket_bind_to_device_distribution.cc create mode 100644 test/syscalls/linux/socket_bind_to_device_sequence.cc create mode 100644 test/syscalls/linux/socket_bind_to_device_util.cc create mode 100644 test/syscalls/linux/socket_bind_to_device_util.h create mode 100644 test/util/uid_util.cc create mode 100644 test/util/uid_util.h (limited to 'test/syscalls/linux/BUILD') diff --git a/pkg/sentry/socket/epsocket/epsocket.go b/pkg/sentry/socket/epsocket/epsocket.go index 3e66f9cbb..5812085fa 100644 --- a/pkg/sentry/socket/epsocket/epsocket.go +++ b/pkg/sentry/socket/epsocket/epsocket.go @@ -942,6 +942,19 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family return int32(v), nil + case linux.SO_BINDTODEVICE: + var v tcpip.BindToDeviceOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + if len(v) == 0 { + return []byte{}, nil + } + if outLen < linux.IFNAMSIZ { + return nil, syserr.ErrInvalidArgument + } + return append([]byte(v), 0), nil + case linux.SO_BROADCAST: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument @@ -1305,6 +1318,13 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i v := usermem.ByteOrder.Uint32(optVal) return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReusePortOption(v))) + case linux.SO_BINDTODEVICE: + n := bytes.IndexByte(optVal, 0) + if n == -1 { + n = len(optVal) + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(optVal[:n]))) + case linux.SO_BROADCAST: if len(optVal) < sizeOfInt32 { return syserr.ErrInvalidArgument diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 3bac4d90d..b5a72ce63 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -531,7 +531,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, syserror.ENOTSOCK } - if optLen <= 0 { + if optLen < 0 { return 0, nil, syserror.EINVAL } if optLen > maxOptLen { diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index 315780c0c..40e202717 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -47,43 +47,76 @@ type portNode struct { refs int } -// bindAddresses is a set of IP addresses. -type bindAddresses map[tcpip.Address]portNode - -// isAvailable checks whether an IP address is available to bind to. -func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool) bool { - if addr == anyIPAddress { - if len(b) == 0 { - return true - } +// deviceNode is never empty. When it has no elements, it is removed from the +// map that references it. +type deviceNode map[tcpip.NICID]portNode + +// isAvailable checks whether binding is possible by device. If not binding to a +// device, check against all portNodes. If binding to a specific device, check +// against the unspecified device and the provided device. +func (d deviceNode) isAvailable(reuse bool, bindToDevice tcpip.NICID) bool { + if bindToDevice == 0 { + // Trying to binding all devices. if !reuse { + // Can't bind because the (addr,port) is already bound. return false } - for _, n := range b { - if !n.reuse { + for _, p := range d { + if !p.reuse { + // Can't bind because the (addr,port) was previously bound without reuse. return false } } return true } - // If all addresses for this portDescriptor are already bound, no - // address is available. - if n, ok := b[anyIPAddress]; ok { - if !reuse { + if p, ok := d[0]; ok { + if !reuse || !p.reuse { return false } - if !n.reuse { + } + + if p, ok := d[bindToDevice]; ok { + if !reuse || !p.reuse { return false } } - if n, ok := b[addr]; ok { - if !reuse { + return true +} + +// bindAddresses is a set of IP addresses. +type bindAddresses map[tcpip.Address]deviceNode + +// isAvailable checks whether an IP address is available to bind to. If the +// address is the "any" address, check all other addresses. Otherwise, just +// check against the "any" address and the provided address. +func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool, bindToDevice tcpip.NICID) bool { + if addr == anyIPAddress { + // If binding to the "any" address then check that there are no conflicts + // with all addresses. + for _, d := range b { + if !d.isAvailable(reuse, bindToDevice) { + return false + } + } + return true + } + + // Check that there is no conflict with the "any" address. + if d, ok := b[anyIPAddress]; ok { + if !d.isAvailable(reuse, bindToDevice) { + return false + } + } + + // Check that this is no conflict with the provided address. + if d, ok := b[addr]; ok { + if !d.isAvailable(reuse, bindToDevice) { return false } - return n.reuse } + return true } @@ -116,17 +149,17 @@ func (s *PortManager) PickEphemeralPort(testPort func(p uint16) (bool, *tcpip.Er } // IsPortAvailable tests if the given port is available on all given protocols. -func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { +func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { s.mu.Lock() defer s.mu.Unlock() - return s.isPortAvailableLocked(networks, transport, addr, port, reuse) + return s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) } -func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { +func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { for _, network := range networks { desc := portDescriptor{network, transport, port} if addrs, ok := s.allocatedPorts[desc]; ok { - if !addrs.isAvailable(addr, reuse) { + if !addrs.isAvailable(addr, reuse, bindToDevice) { return false } } @@ -138,14 +171,14 @@ func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumb // reserved by another endpoint. If port is zero, ReservePort will search for // an unreserved ephemeral port and reserve it, returning its value in the // "port" return value. -func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) (reservedPort uint16, err *tcpip.Error) { +func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() // If a port is specified, just try to reserve it for all network // protocols. if port != 0 { - if !s.reserveSpecificPort(networks, transport, addr, port, reuse) { + if !s.reserveSpecificPort(networks, transport, addr, port, reuse, bindToDevice) { return 0, tcpip.ErrPortInUse } return port, nil @@ -153,13 +186,13 @@ func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transp // A port wasn't specified, so try to find one. return s.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { - return s.reserveSpecificPort(networks, transport, addr, p, reuse), nil + return s.reserveSpecificPort(networks, transport, addr, p, reuse, bindToDevice), nil }) } // reserveSpecificPort tries to reserve the given port on all given protocols. -func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { - if !s.isPortAvailableLocked(networks, transport, addr, port, reuse) { +func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { + if !s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) { return false } @@ -171,11 +204,16 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber m = make(bindAddresses) s.allocatedPorts[desc] = m } - if n, ok := m[addr]; ok { + d, ok := m[addr] + if !ok { + d = make(deviceNode) + m[addr] = d + } + if n, ok := d[bindToDevice]; ok { n.refs++ - m[addr] = n + d[bindToDevice] = n } else { - m[addr] = portNode{reuse: reuse, refs: 1} + d[bindToDevice] = portNode{reuse: reuse, refs: 1} } } @@ -184,22 +222,28 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber // ReleasePort releases the reservation on a port/IP combination so that it can // be reserved by other endpoints. -func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16) { +func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, bindToDevice tcpip.NICID) { s.mu.Lock() defer s.mu.Unlock() for _, network := range networks { desc := portDescriptor{network, transport, port} if m, ok := s.allocatedPorts[desc]; ok { - n, ok := m[addr] + d, ok := m[addr] + if !ok { + continue + } + n, ok := d[bindToDevice] if !ok { continue } n.refs-- + d[bindToDevice] = n if n.refs == 0 { + delete(d, bindToDevice) + } + if len(d) == 0 { delete(m, addr) - } else { - m[addr] = n } if len(m) == 0 { delete(s.allocatedPorts, desc) diff --git a/pkg/tcpip/ports/ports_test.go b/pkg/tcpip/ports/ports_test.go index 689401661..a67e283f1 100644 --- a/pkg/tcpip/ports/ports_test.go +++ b/pkg/tcpip/ports/ports_test.go @@ -34,6 +34,7 @@ type portReserveTestAction struct { want *tcpip.Error reuse bool release bool + device tcpip.NICID } func TestPortReservation(t *testing.T) { @@ -100,6 +101,112 @@ func TestPortReservation(t *testing.T) { {port: 24, ip: anyIPAddress, release: true}, {port: 24, ip: anyIPAddress, reuse: false, want: nil}, }, + }, { + tname: "bind twice with device fails", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 3, want: nil}, + {port: 24, ip: fakeIPAddress, device: 3, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 1, want: nil}, + {port: 24, ip: fakeIPAddress, device: 2, want: nil}, + }, + }, { + tname: "bind to device and then without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, want: nil}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + }, + }, { + tname: "binding with reuse and device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "mixing reuse and not reuse by binding to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: nil}, + }, + }, { + tname: "can't bind to 0 after mixing reuse and not reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind and release", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + + // Release the bind to device 0 and try again. + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil}, + }, + }, { + tname: "bind twice with reuse once", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "release an unreserved device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil}, + // The below don't exist. + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil, release: true}, + {port: 9999, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + // Release all. + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil, release: true}, + }, }, } { t.Run(test.tname, func(t *testing.T) { @@ -108,12 +215,12 @@ func TestPortReservation(t *testing.T) { for _, test := range test.actions { if test.release { - pm.ReleasePort(net, fakeTransNumber, test.ip, test.port) + pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.device) continue } - gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse) + gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse, test.device) if err != test.want { - t.Fatalf("ReservePort(.., .., %s, %d, %t) = %v, want %v", test.ip, test.port, test.release, err, test.want) + t.Fatalf("ReservePort(.., .., %s, %d, %t, %d) = %v, want %v", test.ip, test.port, test.reuse, test.device, err, test.want) } if test.port == 0 && (gotPort == 0 || gotPort < FirstEphemeral) { t.Fatalf("ReservePort(.., .., .., 0) = %d, want port number >= %d to be picked", gotPort, FirstEphemeral) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 28c49e8ff..3842f1f7d 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -54,6 +54,7 @@ go_test( size = "small", srcs = [ "stack_test.go", + "transport_demuxer_test.go", "transport_test.go", ], deps = [ @@ -64,6 +65,9 @@ go_test( "//pkg/tcpip/iptables", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", + "//pkg/tcpip/network/ipv4", + "//pkg/tcpip/network/ipv6", + "//pkg/tcpip/transport/udp", "//pkg/waiter", ], ) diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 0e8a23f00..f6106f762 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -34,8 +34,6 @@ type NIC struct { linkEP LinkEndpoint loopback bool - demux *transportDemuxer - mu sync.RWMutex spoofing bool promiscuous bool @@ -85,7 +83,6 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback name: name, linkEP: ep, loopback: loopback, - demux: newTransportDemuxer(stack), primary: make(map[tcpip.NetworkProtocolNumber]*ilist.List), endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint), mcastJoins: make(map[NetworkEndpointID]int32), @@ -707,9 +704,7 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // Raw socket packets are delivered based solely on the transport // protocol number. We do not inspect the payload to ensure it's // validly formed. - if !n.demux.deliverRawPacket(r, protocol, netHeader, vv) { - n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) - } + n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) if len(vv.First()) < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() @@ -723,9 +718,6 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} - if n.demux.deliverPacket(r, protocol, netHeader, vv, id) { - return - } if n.stack.demux.deliverPacket(r, protocol, netHeader, vv, id) { return } @@ -767,10 +759,7 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp } id := TransportEndpointID{srcPort, local, dstPort, remote} - if n.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { - return - } - if n.stack.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { + if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, vv, id) { return } } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 18d1704a5..6a8079823 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -1033,73 +1033,27 @@ func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep. // transport dispatcher. Received packets that match the provided id will be // delivered to the given endpoint; specifying a nic is optional, but // nic-specific IDs have precedence over global ones. -func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { - if nicID == 0 { - return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) +func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort, bindToDevice) } // UnregisterTransportEndpoint removes the endpoint with the given id from the // stack transport dispatcher. -func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) { - if nicID == 0 { - s.demux.unregisterEndpoint(netProtos, protocol, id, ep) - return - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterEndpoint(netProtos, protocol, id, ep) - } +func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { + s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) } // RegisterRawTransportEndpoint registers the given endpoint with the stack // transport dispatcher. Received packets that match the provided transport // protocol will be delivered to the given endpoint. func (s *Stack) RegisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error { - if nicID == 0 { - return s.demux.registerRawEndpoint(netProto, transProto, ep) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerRawEndpoint(netProto, transProto, ep) + return s.demux.registerRawEndpoint(netProto, transProto, ep) } // UnregisterRawTransportEndpoint removes the endpoint for the transport // protocol from the stack transport dispatcher. func (s *Stack) UnregisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) { - if nicID == 0 { - s.demux.unregisterRawEndpoint(netProto, transProto, ep) - return - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterRawEndpoint(netProto, transProto, ep) - } + s.demux.unregisterRawEndpoint(netProto, transProto, ep) } // RegisterRestoredEndpoint records e as an endpoint that has been restored on diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index cf8a6d129..8c768c299 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -35,25 +35,109 @@ type protocolIDs struct { type transportEndpoints struct { // mu protects all fields of the transportEndpoints. mu sync.RWMutex - endpoints map[TransportEndpointID]TransportEndpoint + endpoints map[TransportEndpointID]*endpointsByNic // rawEndpoints contains endpoints for raw sockets, which receive all // traffic of a given protocol regardless of port. rawEndpoints []RawTransportEndpoint } +type endpointsByNic struct { + mu sync.RWMutex + endpoints map[tcpip.NICID]*multiPortEndpoint + // seed is a random secret for a jenkins hash. + seed uint32 +} + +// HandlePacket is called by the stack when new packets arrive to this transport +// endpoint. +func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { + epsByNic.mu.RLock() + + mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] + if !ok { + if mpep, ok = epsByNic.endpoints[0]; !ok { + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return + } + } + + // If this is a broadcast or multicast datagram, deliver the datagram to all + // endpoints bound to the right device. + if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) { + mpep.handlePacketAll(r, id, vv) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return + } + + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, vv) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. +} + +// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. +func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) { + epsByNic.mu.RLock() + defer epsByNic.mu.RUnlock() + + mpep, ok := epsByNic.endpoints[n.ID()] + if !ok { + mpep, ok = epsByNic.endpoints[0] + } + if !ok { + return + } + + // TODO(eyalsoha): Why don't we look at id to see if this packet needs to + // broadcast like we are doing with handlePacket above? + + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandleControlPacket(id, typ, extra, vv) +} + +// registerEndpoint returns true if it succeeds. It fails and returns +// false if ep already has an element with the same key. +func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + + if multiPortEp, ok := epsByNic.endpoints[bindToDevice]; ok { + // There was already a bind. + return multiPortEp.singleRegisterEndpoint(t, reusePort) + } + + // This is a new binding. + multiPortEp := &multiPortEndpoint{} + multiPortEp.endpointsMap = make(map[TransportEndpoint]int) + multiPortEp.reuse = reusePort + epsByNic.endpoints[bindToDevice] = multiPortEp + return multiPortEp.singleRegisterEndpoint(t, reusePort) +} + +// unregisterEndpoint returns true if endpointsByNic has to be unregistered. +func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint) bool { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + multiPortEp, ok := epsByNic.endpoints[bindToDevice] + if !ok { + return false + } + if multiPortEp.unregisterEndpoint(t) { + delete(epsByNic.endpoints, bindToDevice) + } + return len(epsByNic.endpoints) == 0 +} + // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. -func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint) { +func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { eps.mu.Lock() defer eps.mu.Unlock() - e, ok := eps.endpoints[id] + epsByNic, ok := eps.endpoints[id] if !ok { return } - if multiPortEp, ok := e.(*multiPortEndpoint); ok { - if !multiPortEp.unregisterEndpoint(ep) { - return - } + if !epsByNic.unregisterEndpoint(bindToDevice, ep) { + return } delete(eps.endpoints, id) } @@ -75,7 +159,7 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { for netProto := range stack.networkProtocols { for proto := range stack.transportProtocols { d.protocol[protocolIDs{netProto, proto}] = &transportEndpoints{ - endpoints: make(map[TransportEndpointID]TransportEndpoint), + endpoints: make(map[TransportEndpointID]*endpointsByNic), } } } @@ -85,10 +169,10 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { // registerEndpoint registers the given endpoint with the dispatcher such that // packets that match the endpoint ID are delivered to it. -func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { for i, n := range netProtos { - if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort); err != nil { - d.unregisterEndpoint(netProtos[:i], protocol, id, ep) + if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort, bindToDevice); err != nil { + d.unregisterEndpoint(netProtos[:i], protocol, id, ep, bindToDevice) return err } } @@ -97,13 +181,14 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum } // multiPortEndpoint is a container for TransportEndpoints which are bound to -// the same pair of address and port. +// the same pair of address and port. endpointsArr always has at least one +// element. type multiPortEndpoint struct { mu sync.RWMutex endpointsArr []TransportEndpoint endpointsMap map[TransportEndpoint]int - // seed is a random secret for a jenkins hash. - seed uint32 + // reuse indicates if more than one endpoint is allowed. + reuse bool } // reciprocalScale scales a value into range [0, n). @@ -117,9 +202,10 @@ func reciprocalScale(val, n uint32) uint32 { // selectEndpoint calculates a hash of destination and source addresses and // ports then uses it to select a socket. In this case, all packets from one // address will be sent to same endpoint. -func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEndpoint { - ep.mu.RLock() - defer ep.mu.RUnlock() +func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32) TransportEndpoint { + if len(mpep.endpointsArr) == 1 { + return mpep.endpointsArr[0] + } payload := []byte{ byte(id.LocalPort), @@ -128,51 +214,50 @@ func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEnd byte(id.RemotePort >> 8), } - h := jenkins.Sum32(ep.seed) + h := jenkins.Sum32(seed) h.Write(payload) h.Write([]byte(id.LocalAddress)) h.Write([]byte(id.RemoteAddress)) hash := h.Sum32() - idx := reciprocalScale(hash, uint32(len(ep.endpointsArr))) - return ep.endpointsArr[idx] + idx := reciprocalScale(hash, uint32(len(mpep.endpointsArr))) + return mpep.endpointsArr[idx] } -// HandlePacket is called by the stack when new packets arrive to this transport -// endpoint. -func (ep *multiPortEndpoint) HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { - // If this is a broadcast or multicast datagram, deliver the datagram to all - // endpoints managed by ep. - if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) { - for i, endpoint := range ep.endpointsArr { - // HandlePacket modifies vv, so each endpoint needs its own copy. - if i == len(ep.endpointsArr)-1 { - endpoint.HandlePacket(r, id, vv) - break - } - vvCopy := buffer.NewView(vv.Size()) - copy(vvCopy, vv.ToView()) - endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) +func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { + ep.mu.RLock() + for i, endpoint := range ep.endpointsArr { + // HandlePacket modifies vv, so each endpoint needs its own copy except for + // the final one. + if i == len(ep.endpointsArr)-1 { + endpoint.HandlePacket(r, id, vv) + break } - } else { - ep.selectEndpoint(id).HandlePacket(r, id, vv) + vvCopy := buffer.NewView(vv.Size()) + copy(vvCopy, vv.ToView()) + endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) } + ep.mu.RUnlock() // Don't use defer for performance reasons. } -// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (ep *multiPortEndpoint) HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) { - ep.selectEndpoint(id).HandleControlPacket(id, typ, extra, vv) -} - -func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint) { +// singleRegisterEndpoint tries to add an endpoint to the multiPortEndpoint +// list. The list might be empty already. +func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, reusePort bool) *tcpip.Error { ep.mu.Lock() defer ep.mu.Unlock() - // A new endpoint is added into endpointsArr and its index there is - // saved in endpointsMap. This will allows to remove endpoint from - // the array fast. + if len(ep.endpointsArr) > 0 { + // If it was previously bound, we need to check if we can bind again. + if !ep.reuse || !reusePort { + return tcpip.ErrPortInUse + } + } + + // A new endpoint is added into endpointsArr and its index there is saved in + // endpointsMap. This will allow us to remove endpoint from the array fast. ep.endpointsMap[t] = len(ep.endpointsArr) ep.endpointsArr = append(ep.endpointsArr, t) + return nil } // unregisterEndpoint returns true if multiPortEndpoint has to be unregistered. @@ -197,53 +282,41 @@ func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint) bool { return true } -func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { if id.RemotePort != 0 { + // TODO(eyalsoha): Why? reusePort = false } eps, ok := d.protocol[protocolIDs{netProto, protocol}] if !ok { - return nil + return tcpip.ErrUnknownProtocol } eps.mu.Lock() defer eps.mu.Unlock() - var multiPortEp *multiPortEndpoint - if _, ok := eps.endpoints[id]; ok { - if !reusePort { - return tcpip.ErrPortInUse - } - multiPortEp, ok = eps.endpoints[id].(*multiPortEndpoint) - if !ok { - return tcpip.ErrPortInUse - } + if epsByNic, ok := eps.endpoints[id]; ok { + // There was already a binding. + return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) } - if reusePort { - if multiPortEp == nil { - multiPortEp = &multiPortEndpoint{} - multiPortEp.endpointsMap = make(map[TransportEndpoint]int) - multiPortEp.seed = rand.Uint32() - eps.endpoints[id] = multiPortEp - } - - multiPortEp.singleRegisterEndpoint(ep) - - return nil + // This is a new binding. + epsByNic := &endpointsByNic{ + endpoints: make(map[tcpip.NICID]*multiPortEndpoint), + seed: rand.Uint32(), } - eps.endpoints[id] = ep + eps.endpoints[id] = epsByNic - return nil + return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) } // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. -func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) { +func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { for _, n := range netProtos { if eps, ok := d.protocol[protocolIDs{n, protocol}]; ok { - eps.unregisterEndpoint(id, ep) + eps.unregisterEndpoint(id, ep, bindToDevice) } } } @@ -273,7 +346,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto // If the packet is a broadcast, then find all matching transport endpoints. // Otherwise, try to find a single matching transport endpoint. - destEps := make([]TransportEndpoint, 0, 1) + destEps := make([]*endpointsByNic, 0, 1) eps.mu.RLock() if protocol == header.UDPProtocolNumber && id.LocalAddress == header.IPv4Broadcast { @@ -299,7 +372,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto // Deliver the packet. for _, ep := range destEps { - ep.HandlePacket(r, id, vv) + ep.handlePacket(r, id, vv) } return true @@ -331,7 +404,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr // deliverControlPacket attempts to deliver the given control packet. Returns // true if it found an endpoint, false otherwise. -func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{net, trans}] if !ok { return false @@ -348,12 +421,12 @@ func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, } // Deliver the packet. - ep.HandleControlPacket(id, typ, extra, vv) + ep.handleControlPacket(n, id, typ, extra, vv) return true } -func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) TransportEndpoint { +func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) *endpointsByNic { // Try to find a match with the id as provided. if ep, ok := eps.endpoints[id]; ok { return ep diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go new file mode 100644 index 000000000..210233dc0 --- /dev/null +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -0,0 +1,352 @@ +// 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 stack_test + +import ( + "math" + "math/rand" + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/udp" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + stackV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" + testV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" + + stackAddr = "\x0a\x00\x00\x01" + stackPort = 1234 + testPort = 4096 +) + +type testContext struct { + t *testing.T + linkEPs map[string]*channel.Endpoint + s *stack.Stack + + ep tcpip.Endpoint + wq waiter.Queue +} + +func (c *testContext) cleanup() { + if c.ep != nil { + c.ep.Close() + } +} + +func (c *testContext) createV6Endpoint(v6only bool) { + var err *tcpip.Error + c.ep, err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + + var v tcpip.V6OnlyOption + if v6only { + v = 1 + } + if err := c.ep.SetSockOpt(v); err != nil { + c.t.Fatalf("SetSockOpt failed: %v", err) + } +} + +// newDualTestContextMultiNic creates the testing context and also linkEpNames +// named NICs. +func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + linkEPs := make(map[string]*channel.Endpoint) + for i, linkEpName := range linkEpNames { + channelEP := channel.New(256, mtu, "") + nicid := tcpip.NICID(i + 1) + if err := s.CreateNamedNIC(nicid, linkEpName, channelEP); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + linkEPs[linkEpName] = channelEP + + if err := s.AddAddress(nicid, ipv4.ProtocolNumber, stackAddr); err != nil { + t.Fatalf("AddAddress IPv4 failed: %v", err) + } + + if err := s.AddAddress(nicid, ipv6.ProtocolNumber, stackV6Addr); err != nil { + t.Fatalf("AddAddress IPv6 failed: %v", err) + } + } + + s.SetRouteTable([]tcpip.Route{ + { + Destination: header.IPv4EmptySubnet, + NIC: 1, + }, + { + Destination: header.IPv6EmptySubnet, + NIC: 1, + }, + }) + + return &testContext{ + t: t, + s: s, + linkEPs: linkEPs, + } +} + +type headers struct { + srcPort uint16 + dstPort uint16 +} + +func newPayload() []byte { + b := make([]byte, 30+rand.Intn(100)) + for i := range b { + b[i] = byte(rand.Intn(256)) + } + return b +} + +func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) { + // Allocate a buffer for data and headers. + buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) + copy(buf[len(buf)-len(payload):], payload) + + // Initialize the IP header. + ip := header.IPv6(buf) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(header.UDPMinimumSize + len(payload)), + NextHeader: uint8(udp.ProtocolNumber), + HopLimit: 65, + SrcAddr: testV6Addr, + DstAddr: stackV6Addr, + }) + + // Initialize the UDP header. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.Encode(&header.UDPFields{ + SrcPort: h.srcPort, + DstPort: h.dstPort, + Length: uint16(header.UDPMinimumSize + len(payload)), + }) + + // Calculate the UDP pseudo-header checksum. + xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testV6Addr, stackV6Addr, uint16(len(u))) + + // Calculate the UDP checksum and set it. + xsum = header.Checksum(payload, xsum) + u.SetChecksum(^u.CalculateChecksum(xsum)) + + // Inject packet. + c.linkEPs[linkEpName].Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) +} + +func TestTransportDemuxerRegister(t *testing.T) { + for _, test := range []struct { + name string + proto tcpip.NetworkProtocolNumber + want *tcpip.Error + }{ + {"failure", ipv6.ProtocolNumber, tcpip.ErrUnknownProtocol}, + {"success", ipv4.ProtocolNumber, nil}, + } { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + if got, want := s.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{test.proto}, udp.ProtocolNumber, stack.TransportEndpointID{}, nil, false, 0), test.want; got != want { + t.Fatalf("s.RegisterTransportEndpoint(...) = %v, want %v", got, want) + } + }) + } +} + +// TestReuseBindToDevice injects varied packets on input devices and checks that +// the distribution of packets received matches expectations. +func TestDistribution(t *testing.T) { + type endpointSockopts struct { + reuse int + bindToDevice string + } + for _, test := range []struct { + name string + // endpoints will received the inject packets. + endpoints []endpointSockopts + // wantedDistribution is the wanted ratio of packets received on each + // endpoint for each NIC on which packets are injected. + wantedDistributions map[string][]float64 + }{ + { + "BindPortReuse", + // 5 endpoints that all have reuse set. + []endpointSockopts{ + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed evenly. + "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2}, + }, + }, + { + "BindToDevice", + // 3 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{0, "dev0"}, + endpointSockopts{0, "dev1"}, + endpointSockopts{0, "dev2"}, + }, + map[string][]float64{ + // Injected packets on dev0 go only to the endpoint bound to dev0. + "dev0": []float64{1, 0, 0}, + // Injected packets on dev1 go only to the endpoint bound to dev1. + "dev1": []float64{0, 1, 0}, + // Injected packets on dev2 go only to the endpoint bound to dev2. + "dev2": []float64{0, 0, 1}, + }, + }, + { + "ReuseAndBindToDevice", + // 6 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed among endpoints bound to + // dev0. + "dev0": []float64{0.5, 0.5, 0, 0, 0, 0}, + // Injected packets on dev1 get distributed among endpoints bound to + // dev1 or unbound. + "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, + // Injected packets on dev999 go only to the unbound. + "dev999": []float64{0, 0, 0, 0, 0, 1}, + }, + }, + } { + t.Run(test.name, func(t *testing.T) { + for device, wantedDistribution := range test.wantedDistributions { + t.Run(device, func(t *testing.T) { + var devices []string + for d := range test.wantedDistributions { + devices = append(devices, d) + } + c := newDualTestContextMultiNic(t, defaultMTU, devices) + defer c.cleanup() + + c.createV6Endpoint(false) + + eps := make(map[tcpip.Endpoint]int) + + pollChannel := make(chan tcpip.Endpoint) + for i, endpoint := range test.endpoints { + // Try to receive the data. + wq := waiter.Queue{} + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + defer close(ch) + + var err *tcpip.Error + ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + eps[ep] = i + + go func(ep tcpip.Endpoint) { + for range ch { + pollChannel <- ep + } + }(ep) + + defer ep.Close() + reusePortOption := tcpip.ReusePortOption(endpoint.reuse) + if err := ep.SetSockOpt(reusePortOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", reusePortOption, i, err) + } + bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice) + if err := ep.SetSockOpt(bindToDeviceOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", bindToDeviceOption, i, err) + } + if err := ep.Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { + t.Fatalf("ep.Bind(...) on endpoint %d failed: %v", i, err) + } + } + + npackets := 100000 + nports := 10000 + if got, want := len(test.endpoints), len(wantedDistribution); got != want { + t.Fatalf("got len(test.endpoints) = %d, want %d", got, want) + } + ports := make(map[uint16]tcpip.Endpoint) + stats := make(map[tcpip.Endpoint]int) + for i := 0; i < npackets; i++ { + // Send a packet. + port := uint16(i % nports) + payload := newPayload() + c.sendV6Packet(payload, + &headers{ + srcPort: testPort + port, + dstPort: stackPort}, + device) + + var addr tcpip.FullAddress + ep := <-pollChannel + _, _, err := ep.Read(&addr) + if err != nil { + c.t.Fatalf("Read on endpoint %d failed: %v", eps[ep], err) + } + stats[ep]++ + if i < nports { + ports[uint16(i)] = ep + } else { + // Check that all packets from one client are handled by the same + // socket. + if want, got := ports[port], ep; want != got { + t.Fatalf("Packet sent on port %d expected on endpoint %d but received on endpoint %d", port, eps[want], eps[got]) + } + } + } + + // Check that a packet distribution is as expected. + for ep, i := range eps { + wantedRatio := wantedDistribution[i] + wantedRecv := wantedRatio * float64(npackets) + actualRecv := stats[ep] + actualRatio := float64(stats[ep]) / float64(npackets) + // The deviation is less than 10%. + if math.Abs(actualRatio-wantedRatio) > 0.05 { + t.Errorf("wanted about %.0f%% (%.0f of %d) packets to arrive on endpoint %d, got %.0f%% (%d of %d)", wantedRatio*100, wantedRecv, npackets, i, actualRatio*100, actualRecv, npackets) + } + } + }) + } + }) + } +} diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 56e8a5d9b..842a16277 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -127,7 +127,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Try to register so that we can start receiving packets. f.id.RemoteAddress = addr.Addr - err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false) + err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false /* reuse */, 0 /* bindToDevice */) if err != nil { return err } @@ -168,7 +168,8 @@ func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error { fakeTransNumber, stack.TransportEndpointID{LocalAddress: a.Addr}, f, - false, + false, /* reuse */ + 0, /* bindtoDevice */ ); err != nil { return err } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index c021c67ac..faaa4a4e3 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -495,6 +495,10 @@ type ReuseAddressOption int // to be bound to an identical socket address. type ReusePortOption int +// BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets +// should bind only on a specific NIC. +type BindToDeviceOption string + // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. type QuickAckOption int diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index a111fdb2a..a3a910d41 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -104,7 +104,7 @@ func (e *endpoint) Close() { e.shutdownFlags = tcpip.ShutdownRead | tcpip.ShutdownWrite switch e.state { case stateBound, stateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.transProto, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.transProto, e.id, e, 0 /* bindToDevice */) } // Close the receive list and drain it. @@ -543,14 +543,14 @@ func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.Networ if id.LocalPort != 0 { // The endpoint already has a local port, just attempt to // register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false /* reuse */, 0 /* bindToDevice */) return id, err } // We need to find a port for the endpoint. _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { id.LocalPort = p - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false /* reuse */, 0 /* bindtodevice */) switch err { case nil: return true, nil diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 0802e984e..3ae4a5426 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -242,7 +242,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.id, n, n.reusePort); err != nil { + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.id, n, n.reusePort, n.bindToDevice); err != nil { n.Close() return nil, err } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 35b489c68..a1cd0d481 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -280,6 +280,9 @@ type endpoint struct { // reusePort is set to true if SO_REUSEPORT is enabled. reusePort bool + // bindToDevice is set to the NIC on which to bind or disabled if 0. + bindToDevice tcpip.NICID + // delay enables Nagle's algorithm. // // delay is a boolean (0 is false) and must be accessed atomically. @@ -564,11 +567,11 @@ func (e *endpoint) Close() { // in Listen() when trying to register. if e.state == StateListen && e.isPortReserved { if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.isRegistered = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -625,12 +628,12 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.isRegistered = false } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -1060,6 +1063,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicid, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicid + return nil + } + } + return tcpip.ErrUnknownDevice + case tcpip.QuickAckOption: if v == 0 { atomic.StoreUint32(&e.slowAck, 1) @@ -1260,6 +1278,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = "" + return nil + case *tcpip.QuickAckOption: *o = 1 if v := atomic.LoadUint32(&e.slowAck); v != 0 { @@ -1466,7 +1494,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er if e.id.LocalPort != 0 { // The endpoint is bound to a port, attempt to register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice) if err != nil { return err } @@ -1480,13 +1508,15 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er if sameAddr && p == e.id.RemotePort { return false, nil } - if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false) { + // reusePort is false below because connect cannot reuse a port even if + // reusePort was set. + if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false /* reusePort */, e.bindToDevice) { return false, nil } id := e.id id.LocalPort = p - switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) { + switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) { case nil: e.id = id return true, nil @@ -1504,7 +1534,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er // before Connect: in such a case we don't want to hold on to // reservations anymore. if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -1648,7 +1678,7 @@ func (e *endpoint) Listen(backlog int) (err *tcpip.Error) { } // Register the endpoint. - if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort); err != nil { + if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice); err != nil { return err } @@ -1729,7 +1759,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { } } - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort, e.bindToDevice) if err != nil { return err } @@ -1739,16 +1769,16 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { e.id.LocalPort = port // Any failures beyond this point must remove the port registration. - defer func() { + defer func(bindToDevice tcpip.NICID) { if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port) + e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, bindToDevice) e.isPortReserved = false e.effectiveNetProtos = nil e.id.LocalPort = 0 e.id.LocalAddress = "" e.boundNICID = 0 } - }() + }(e.bindToDevice) // If an address is specified, we must ensure that it's one of our // local addresses. diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 2be094876..089826a88 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -465,6 +465,66 @@ func TestSimpleReceive(t *testing.T) { ) } +func TestConnectBindToDevice(t *testing.T) { + for _, test := range []struct { + name string + device string + want tcp.EndpointState + }{ + {"RightDevice", "nic1", tcp.StateEstablished}, + {"WrongDevice", "nic2", tcp.StateSynSent}, + {"AnyDevice", "", tcp.StateEstablished}, + } { + t.Run(test.name, func(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + bindToDevice := tcpip.BindToDeviceOption(test.device) + c.EP.SetSockOpt(bindToDevice) + // Start connection attempt. + waitEntry, _ := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&waitEntry, waiter.EventOut) + defer c.WQ.EventUnregister(&waitEntry) + + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet. + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + ), + ) + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.SendPacket(nil, &context.Headers{ + SrcPort: tcpHdr.DestinationPort(), + DstPort: tcpHdr.SourcePort(), + Flags: header.TCPFlagSyn | header.TCPFlagAck, + SeqNum: iss, + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + TCPOpts: nil, + }) + + c.GetPacket() + if got, want := tcp.EndpointState(c.EP.State()), test.want; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + }) + } +} + func TestOutOfOrderReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -2970,6 +3030,62 @@ func TestMinMaxBufferSizes(t *testing.T) { checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) } +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}}) + + ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) + } + defer ep.Close() + + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } + + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) + } + + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s + } + + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) + } +} + func makeStack() (*stack.Stack, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index d3f1d2cdf..ef823e4ae 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -158,7 +158,14 @@ func New(t *testing.T, mtu uint32) *Context { if testing.Verbose() { wep = sniffer.New(ep) } - if err := s.CreateNIC(1, wep); err != nil { + if err := s.CreateNamedNIC(1, "nic1", wep); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + wep2 := stack.LinkEndpoint(channel.New(1000, mtu, "")) + if testing.Verbose() { + wep2 = sniffer.New(channel.New(1000, mtu, "")) + } + if err := s.CreateNamedNIC(2, "nic2", wep2); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -588,12 +595,8 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) c.Port = tcpHdr.SourcePort() } -// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends -// the specified option bytes as the Option field in the initial SYN packet. -// -// It also sets the receive buffer for the endpoint to the specified -// value in epRcvBuf. -func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int, options []byte) { +// Create creates a TCP endpoint. +func (c *Context) Create(epRcvBuf int) { // Create TCP endpoint. var err *tcpip.Error c.EP, err = c.s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) @@ -606,6 +609,15 @@ func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum. c.t.Fatalf("SetSockOpt failed failed: %v", err) } } +} + +// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends +// the specified option bytes as the Option field in the initial SYN packet. +// +// It also sets the receive buffer for the endpoint to the specified +// value in epRcvBuf. +func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int, options []byte) { + c.Create(epRcvBuf) c.Connect(iss, rcvWnd, options) } diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index c1ca22b35..7a635ab8d 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -52,6 +52,7 @@ go_test( "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", + "//pkg/tcpip/link/loopback", "//pkg/tcpip/link/sniffer", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 0bec7e62d..52f5af777 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -88,6 +88,7 @@ type endpoint struct { multicastNICID tcpip.NICID multicastLoop bool reusePort bool + bindToDevice tcpip.NICID broadcast bool // shutdownFlags represent the current shutdown state of the endpoint. @@ -144,8 +145,8 @@ func (e *endpoint) Close() { switch e.state { case StateBound, StateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) } for _, mem := range e.multicastMemberships { @@ -551,6 +552,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.reusePort = v != 0 e.mu.Unlock() + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicid, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicid + return nil + } + } + return tcpip.ErrUnknownDevice + case tcpip.BroadcastOption: e.mu.Lock() e.broadcast = v != 0 @@ -646,6 +662,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = tcpip.BindToDeviceOption("") + return nil + case *tcpip.KeepaliveEnabledOption: *o = 0 return nil @@ -753,12 +779,12 @@ func (e *endpoint) Disconnect() *tcpip.Error { } else { if e.id.LocalPort != 0 { // Release the ephemeral port. - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) } e.state = StateInitial } - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.id = id e.route.Release() e.route = stack.Route{} @@ -835,7 +861,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Remove the old registration. if e.id.LocalPort != 0 { - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) } e.id = id @@ -898,16 +924,16 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { if e.id.LocalPort == 0 { - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort, e.bindToDevice) if err != nil { return id, err } id.LocalPort = port } - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort) + e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.bindToDevice) } return id, err } diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index a9edc2c8d..2d0bc5221 100644 --- a/pkg/tcpip/transport/udp/forwarder.go +++ b/pkg/tcpip/transport/udp/forwarder.go @@ -74,7 +74,7 @@ func (r *ForwarderRequest) ID() stack.TransportEndpointID { // CreateEndpoint creates a connected UDP endpoint for the session request. func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { ep := newEndpoint(r.stack, r.route.NetProto, queue) - if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort); err != nil { + if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort, ep.bindToDevice); err != nil { ep.Close() return nil, err } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 2ec27be4d..5059ca22d 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -17,7 +17,6 @@ package udp_test import ( "bytes" "fmt" - "math" "math/rand" "testing" "time" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/checker" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" @@ -476,87 +476,59 @@ func newMinPayload(minSize int) []byte { return b } -func TestBindPortReuse(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - var eps [5]tcpip.Endpoint - reusePortOpt := tcpip.ReusePortOption(1) - - pollChannel := make(chan tcpip.Endpoint) - for i := 0; i < len(eps); i++ { - // Try to receive the data. - wq := waiter.Queue{} - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - defer close(ch) - - var err *tcpip.Error - eps[i], err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) - if err != nil { - c.t.Fatalf("NewEndpoint failed: %v", err) - } - - go func(ep tcpip.Endpoint) { - for range ch { - pollChannel <- ep - } - }(eps[i]) +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) - defer eps[i].Close() - if err := eps[i].SetSockOpt(reusePortOpt); err != nil { - c.t.Fatalf("SetSockOpt failed failed: %v", err) - } - if err := eps[i].Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) - } + ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) } + defer ep.Close() - npackets := 100000 - nports := 10000 - ports := make(map[uint16]tcpip.Endpoint) - stats := make(map[tcpip.Endpoint]int) - for i := 0; i < npackets; i++ { - // Send a packet. - port := uint16(i % nports) - payload := newPayload() - c.injectV6Packet(payload, &header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: testV6Addr, Port: testPort + port}, - dstAddr: tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}, - }) + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } - var addr tcpip.FullAddress - ep := <-pollChannel - _, _, err := ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - stats[ep]++ - if i < nports { - ports[uint16(i)] = ep - } else { - // Check that all packets from one client are handled - // by the same socket. - if ports[port] != ep { - t.Fatalf("Port mismatch") - } - } + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) } - if len(stats) != len(eps) { - t.Fatalf("Only %d(expected %d) sockets received packets", len(stats), len(eps)) + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s } - // Check that a packet distribution is fair between sockets. - for _, c := range stats { - n := float64(npackets) / float64(len(eps)) - // The deviation is less than 10%. - if math.Abs(float64(c)-n) > n/10 { - t.Fatal(c, n) - } + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) } } diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 28b23ce58..e645eebfa 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2463,6 +2463,63 @@ cc_binary( ], ) +cc_binary( + name = "socket_bind_to_device_test", + testonly = 1, + srcs = [ + "socket_bind_to_device.cc", + ], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_bind_to_device_util", + ":socket_test_util", + "//test/util:capability_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_googletest//:gtest", + ], +) + +cc_binary( + name = "socket_bind_to_device_sequence_test", + testonly = 1, + srcs = [ + "socket_bind_to_device_sequence.cc", + ], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_bind_to_device_util", + ":socket_test_util", + "//test/util:capability_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_googletest//:gtest", + ], +) + +cc_binary( + name = "socket_bind_to_device_distribution_test", + testonly = 1, + srcs = [ + "socket_bind_to_device_distribution.cc", + ], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_bind_to_device_util", + ":socket_test_util", + "//test/util:capability_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_googletest//:gtest", + ], +) + cc_binary( name = "socket_ip_udp_loopback_non_blocking_test", testonly = 1, @@ -2740,6 +2797,23 @@ cc_library( alwayslink = 1, ) +cc_library( + name = "socket_bind_to_device_util", + testonly = 1, + srcs = [ + "socket_bind_to_device_util.cc", + ], + hdrs = [ + "socket_bind_to_device_util.h", + ], + deps = [ + "//test/util:test_util", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + ], + alwayslink = 1, +) + cc_binary( name = "socket_stream_local_test", testonly = 1, @@ -3253,6 +3327,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "//test/util:uid_util", "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/strings", "@com_google_googletest//:gtest", diff --git a/test/syscalls/linux/socket_bind_to_device.cc b/test/syscalls/linux/socket_bind_to_device.cc new file mode 100644 index 000000000..d20821cac --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device.cc @@ -0,0 +1,314 @@ +// Copyright 2019 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. + +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_bind_to_device_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +namespace gvisor { +namespace testing { + +using std::string; + +// Test fixture for SO_BINDTODEVICE tests. +class BindToDeviceTest : public ::testing::TestWithParam { + protected: + void SetUp() override { + printf("Testing case: %s\n", GetParam().description.c_str()); + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) + << "CAP_NET_RAW is required to use SO_BINDTODEVICE"; + + interface_name_ = "eth1"; + auto interface_names = GetInterfaceNames(); + if (interface_names.find(interface_name_) == interface_names.end()) { + // Need a tunnel. + tunnel_ = ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New()); + interface_name_ = tunnel_->GetName(); + ASSERT_FALSE(interface_name_.empty()); + } + socket_ = ASSERT_NO_ERRNO_AND_VALUE(GetParam().Create()); + } + + string interface_name() const { return interface_name_; } + + int socket_fd() const { return socket_->get(); } + + private: + std::unique_ptr tunnel_; + string interface_name_; + std::unique_ptr socket_; +}; + +constexpr char kIllegalIfnameChar = '/'; + +// Tests getsockopt of the default value. +TEST_P(BindToDeviceTest, GetsockoptDefault) { + char name_buffer[IFNAMSIZ * 2]; + char original_name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Read the default SO_BINDTODEVICE. + memset(original_name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + for (size_t i = 0; i <= sizeof(name_buffer); i++) { + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = i; + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, &name_buffer_size), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(name_buffer_size, 0); + EXPECT_EQ(memcmp(name_buffer, original_name_buffer, sizeof(name_buffer)), + 0); + } +} + +// Tests setsockopt of invalid device name. +TEST_P(BindToDeviceTest, SetsockoptInvalidDeviceName) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Set an invalid device name. + memset(name_buffer, kIllegalIfnameChar, 5); + name_buffer_size = 5; + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + name_buffer_size), + SyscallFailsWithErrno(ENODEV)); +} + +// Tests setsockopt of a buffer with a valid device name but not +// null-terminated, with different sizes of buffer. +TEST_P(BindToDeviceTest, SetsockoptValidDeviceNameWithoutNullTermination) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + strncpy(name_buffer, interface_name().c_str(), interface_name().size() + 1); + // Intentionally overwrite the null at the end. + memset(name_buffer + interface_name().size(), kIllegalIfnameChar, + sizeof(name_buffer) - interface_name().size()); + for (size_t i = 1; i <= sizeof(name_buffer); i++) { + name_buffer_size = i; + SCOPED_TRACE(absl::StrCat("Buffer size: ", i)); + // It should only work if the size provided is exactly right. + if (name_buffer_size == interface_name().size()) { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallSucceeds()); + } else { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallFailsWithErrno(ENODEV)); + } + } +} + +// Tests setsockopt of a buffer with a valid device name and null-terminated, +// with different sizes of buffer. +TEST_P(BindToDeviceTest, SetsockoptValidDeviceNameWithNullTermination) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + strncpy(name_buffer, interface_name().c_str(), interface_name().size() + 1); + // Don't overwrite the null at the end. + memset(name_buffer + interface_name().size() + 1, kIllegalIfnameChar, + sizeof(name_buffer) - interface_name().size() - 1); + for (size_t i = 1; i <= sizeof(name_buffer); i++) { + name_buffer_size = i; + SCOPED_TRACE(absl::StrCat("Buffer size: ", i)); + // It should only work if the size provided is at least the right size. + if (name_buffer_size >= interface_name().size()) { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallSucceeds()); + } else { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallFailsWithErrno(ENODEV)); + } + } +} + +// Tests that setsockopt of an invalid device name doesn't unset the previous +// valid setsockopt. +TEST_P(BindToDeviceTest, SetsockoptValidThenInvalid) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + ASSERT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + + // Write unsuccessfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = 5; + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallFailsWithErrno(ENODEV)); + + // Read it back successfully, it's unchanged. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); +} + +// Tests that setsockopt of zero-length string correctly unsets the previous +// value. +TEST_P(BindToDeviceTest, SetsockoptValidThenClear) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + + // Clear it successfully. + name_buffer_size = 0; + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + name_buffer_size), + SyscallSucceeds()); + + // Read it back successfully, it's cleared. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, 0); +} + +// Tests that setsockopt of empty string correctly unsets the previous +// value. +TEST_P(BindToDeviceTest, SetsockoptValidThenClearWithNull) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + + // Clear it successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer[0] = 0; + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + name_buffer_size), + SyscallSucceeds()); + + // Read it back successfully, it's cleared. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, 0); +} + +// Tests getsockopt with different buffer sizes. +TEST_P(BindToDeviceTest, GetsockoptDevice) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + ASSERT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back at various buffer sizes. + for (size_t i = 0; i <= sizeof(name_buffer); i++) { + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = i; + SCOPED_TRACE(absl::StrCat("Buffer size: ", i)); + // Linux only allows a buffer at least IFNAMSIZ, even if less would suffice + // for this interface name. + if (name_buffer_size >= IFNAMSIZ) { + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + } else { + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, &name_buffer_size), + SyscallFailsWithErrno(EINVAL)); + EXPECT_EQ(name_buffer_size, i); + } + } +} + +INSTANTIATE_TEST_SUITE_P(BindToDeviceTest, BindToDeviceTest, + ::testing::Values(IPv4UDPUnboundSocket(0), + IPv4TCPUnboundSocket(0))); + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_distribution.cc b/test/syscalls/linux/socket_bind_to_device_distribution.cc new file mode 100644 index 000000000..4d2400328 --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_distribution.cc @@ -0,0 +1,381 @@ +// Copyright 2019 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. + +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_bind_to_device_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +namespace gvisor { +namespace testing { + +using std::string; +using std::vector; + +struct EndpointConfig { + std::string bind_to_device; + double expected_ratio; +}; + +struct DistributionTestCase { + std::string name; + std::vector endpoints; +}; + +struct ListenerConnector { + TestAddress listener; + TestAddress connector; +}; + +// Test fixture for SO_BINDTODEVICE tests the distribution of packets received +// with varying SO_BINDTODEVICE settings. +class BindToDeviceDistributionTest + : public ::testing::TestWithParam< + ::testing::tuple> { + protected: + void SetUp() override { + printf("Testing case: %s, listener=%s, connector=%s\n", + ::testing::get<1>(GetParam()).name.c_str(), + ::testing::get<0>(GetParam()).listener.description.c_str(), + ::testing::get<0>(GetParam()).connector.description.c_str()); + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) + << "CAP_NET_RAW is required to use SO_BINDTODEVICE"; + } +}; + +PosixErrorOr AddrPort(int family, sockaddr_storage const& addr) { + switch (family) { + case AF_INET: + return static_cast( + reinterpret_cast(&addr)->sin_port); + case AF_INET6: + return static_cast( + reinterpret_cast(&addr)->sin6_port); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +PosixError SetAddrPort(int family, sockaddr_storage* addr, uint16_t port) { + switch (family) { + case AF_INET: + reinterpret_cast(addr)->sin_port = port; + return NoError(); + case AF_INET6: + reinterpret_cast(addr)->sin6_port = port; + return NoError(); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +// Binds sockets to different devices and then creates many TCP connections. +// Checks that the distribution of connections received on the sockets matches +// the expectation. +TEST_P(BindToDeviceDistributionTest, Tcp) { + auto const& [listener_connector, test] = GetParam(); + + TestAddress const& listener = listener_connector.listener; + TestAddress const& connector = listener_connector.connector; + sockaddr_storage listen_addr = listener.addr; + sockaddr_storage conn_addr = connector.addr; + + auto interface_names = GetInterfaceNames(); + + // Create the listening sockets. + std::vector listener_fds; + std::vector> all_tunnels; + for (auto const& endpoint : test.endpoints) { + if (!endpoint.bind_to_device.empty() && + interface_names.find(endpoint.bind_to_device) == + interface_names.end()) { + all_tunnels.push_back( + ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New(endpoint.bind_to_device))); + interface_names.insert(endpoint.bind_to_device); + } + + listener_fds.push_back(ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP))); + int fd = listener_fds.back().get(); + + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, + endpoint.bind_to_device.c_str(), + endpoint.bind_to_device.size() + 1), + SyscallSucceeds()); + ASSERT_THAT( + bind(fd, reinterpret_cast(&listen_addr), listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(fd, 40), SyscallSucceeds()); + + // On the first bind we need to determine which port was bound. + if (listener_fds.size() > 1) { + continue; + } + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT( + getsockname(listener_fds[0].get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + } + + constexpr int kConnectAttempts = 10000; + std::atomic connects_received = ATOMIC_VAR_INIT(0); + std::vector accept_counts(listener_fds.size(), 0); + std::vector> listen_threads( + listener_fds.size()); + + for (int i = 0; i < listener_fds.size(); i++) { + listen_threads[i] = absl::make_unique( + [&listener_fds, &accept_counts, &connects_received, i, + kConnectAttempts]() { + do { + auto fd = Accept(listener_fds[i].get(), nullptr, nullptr); + if (!fd.ok()) { + // Another thread has shutdown our read side causing the accept to + // fail. + ASSERT_GE(connects_received, kConnectAttempts) + << "errno = " << fd.error(); + return; + } + // Receive some data from a socket to be sure that the connect() + // system call has been completed on another side. + int data; + EXPECT_THAT( + RetryEINTR(recv)(fd.ValueOrDie().get(), &data, sizeof(data), 0), + SyscallSucceedsWithValue(sizeof(data))); + accept_counts[i]++; + } while (++connects_received < kConnectAttempts); + + // Shutdown all sockets to wake up other threads. + for (auto const& listener_fd : listener_fds) { + shutdown(listener_fd.get(), SHUT_RDWR); + } + }); + } + + for (int i = 0; i < kConnectAttempts; i++) { + FileDescriptor const fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + ASSERT_THAT( + RetryEINTR(connect)(fd.get(), reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + EXPECT_THAT(RetryEINTR(send)(fd.get(), &i, sizeof(i), 0), + SyscallSucceedsWithValue(sizeof(i))); + } + + // Join threads to be sure that all connections have been counted. + for (auto const& listen_thread : listen_threads) { + listen_thread->Join(); + } + // Check that connections are distributed correctly among listening sockets. + for (int i = 0; i < accept_counts.size(); i++) { + EXPECT_THAT( + accept_counts[i], + EquivalentWithin(static_cast(kConnectAttempts * + test.endpoints[i].expected_ratio), + 0.10)) + << "endpoint " << i << " got the wrong number of packets"; + } +} + +// Binds sockets to different devices and then sends many UDP packets. Checks +// that the distribution of packets received on the sockets matches the +// expectation. +TEST_P(BindToDeviceDistributionTest, Udp) { + auto const& [listener_connector, test] = GetParam(); + + TestAddress const& listener = listener_connector.listener; + TestAddress const& connector = listener_connector.connector; + sockaddr_storage listen_addr = listener.addr; + sockaddr_storage conn_addr = connector.addr; + + auto interface_names = GetInterfaceNames(); + + // Create the listening socket. + std::vector listener_fds; + std::vector> all_tunnels; + for (auto const& endpoint : test.endpoints) { + if (!endpoint.bind_to_device.empty() && + interface_names.find(endpoint.bind_to_device) == + interface_names.end()) { + all_tunnels.push_back( + ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New(endpoint.bind_to_device))); + interface_names.insert(endpoint.bind_to_device); + } + + listener_fds.push_back( + ASSERT_NO_ERRNO_AND_VALUE(Socket(listener.family(), SOCK_DGRAM, 0))); + int fd = listener_fds.back().get(); + + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, + endpoint.bind_to_device.c_str(), + endpoint.bind_to_device.size() + 1), + SyscallSucceeds()); + ASSERT_THAT( + bind(fd, reinterpret_cast(&listen_addr), listener.addr_len), + SyscallSucceeds()); + + // On the first bind we need to determine which port was bound. + if (listener_fds.size() > 1) { + continue; + } + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT( + getsockname(listener_fds[0].get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + ASSERT_NO_ERRNO(SetAddrPort(listener.family(), &listen_addr, port)); + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + } + + constexpr int kConnectAttempts = 10000; + std::atomic packets_received = ATOMIC_VAR_INIT(0); + std::vector packets_per_socket(listener_fds.size(), 0); + std::vector> receiver_threads( + listener_fds.size()); + + for (int i = 0; i < listener_fds.size(); i++) { + receiver_threads[i] = absl::make_unique( + [&listener_fds, &packets_per_socket, &packets_received, i]() { + do { + struct sockaddr_storage addr = {}; + socklen_t addrlen = sizeof(addr); + int data; + + auto ret = RetryEINTR(recvfrom)( + listener_fds[i].get(), &data, sizeof(data), 0, + reinterpret_cast(&addr), &addrlen); + + if (packets_received < kConnectAttempts) { + ASSERT_THAT(ret, SyscallSucceedsWithValue(sizeof(data))); + } + + if (ret != sizeof(data)) { + // Another thread may have shutdown our read side causing the + // recvfrom to fail. + break; + } + + packets_received++; + packets_per_socket[i]++; + + // A response is required to synchronize with the main thread, + // otherwise the main thread can send more than can fit into receive + // queues. + EXPECT_THAT(RetryEINTR(sendto)( + listener_fds[i].get(), &data, sizeof(data), 0, + reinterpret_cast(&addr), addrlen), + SyscallSucceedsWithValue(sizeof(data))); + } while (packets_received < kConnectAttempts); + + // Shutdown all sockets to wake up other threads. + for (auto const& listener_fd : listener_fds) { + shutdown(listener_fd.get(), SHUT_RDWR); + } + }); + } + + for (int i = 0; i < kConnectAttempts; i++) { + FileDescriptor const fd = + ASSERT_NO_ERRNO_AND_VALUE(Socket(connector.family(), SOCK_DGRAM, 0)); + EXPECT_THAT(RetryEINTR(sendto)(fd.get(), &i, sizeof(i), 0, + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceedsWithValue(sizeof(i))); + int data; + EXPECT_THAT(RetryEINTR(recv)(fd.get(), &data, sizeof(data), 0), + SyscallSucceedsWithValue(sizeof(data))); + } + + // Join threads to be sure that all connections have been counted. + for (auto const& receiver_thread : receiver_threads) { + receiver_thread->Join(); + } + // Check that packets are distributed correctly among listening sockets. + for (int i = 0; i < packets_per_socket.size(); i++) { + EXPECT_THAT( + packets_per_socket[i], + EquivalentWithin(static_cast(kConnectAttempts * + test.endpoints[i].expected_ratio), + 0.10)) + << "endpoint " << i << " got the wrong number of packets"; + } +} + +std::vector GetDistributionTestCases() { + return std::vector{ + {"Even distribution among sockets not bound to device", + {{"", 1. / 3}, {"", 1. / 3}, {"", 1. / 3}}}, + {"Sockets bound to other interfaces get no packets", + {{"eth1", 0}, {"", 1. / 2}, {"", 1. / 2}}}, + {"Bound has priority over unbound", {{"eth1", 0}, {"", 0}, {"lo", 1}}}, + {"Even distribution among sockets bound to device", + {{"eth1", 0}, {"lo", 1. / 2}, {"lo", 1. / 2}}}, + }; +} + +INSTANTIATE_TEST_SUITE_P( + BindToDeviceTest, BindToDeviceDistributionTest, + ::testing::Combine(::testing::Values( + // Listeners bound to IPv4 addresses refuse + // connections using IPv6 addresses. + ListenerConnector{V4Any(), V4Loopback()}, + ListenerConnector{V4Loopback(), V4MappedLoopback()}), + ::testing::ValuesIn(GetDistributionTestCases()))); + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_sequence.cc b/test/syscalls/linux/socket_bind_to_device_sequence.cc new file mode 100644 index 000000000..a7365d139 --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_sequence.cc @@ -0,0 +1,316 @@ +// Copyright 2019 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. + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_bind_to_device_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +namespace gvisor { +namespace testing { + +using std::string; +using std::vector; + +// Test fixture for SO_BINDTODEVICE tests the results of sequences of socket +// binding. +class BindToDeviceSequenceTest : public ::testing::TestWithParam { + protected: + void SetUp() override { + printf("Testing case: %s\n", GetParam().description.c_str()); + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) + << "CAP_NET_RAW is required to use SO_BINDTODEVICE"; + socket_factory_ = GetParam(); + + interface_names_ = GetInterfaceNames(); + } + + PosixErrorOr> NewSocket() const { + return socket_factory_.Create(); + } + + // Gets a device by device_id. If the device_id has been seen before, returns + // the previously returned device. If not, finds or creates a new device. + // Returns an empty string on failure. + void GetDevice(int device_id, string *device_name) { + auto device = devices_.find(device_id); + if (device != devices_.end()) { + *device_name = device->second; + return; + } + + // Need to pick a new device. Try ethernet first. + *device_name = absl::StrCat("eth", next_unused_eth_); + if (interface_names_.find(*device_name) != interface_names_.end()) { + devices_[device_id] = *device_name; + next_unused_eth_++; + return; + } + + // Need to make a new tunnel device. gVisor tests should have enough + // ethernet devices to never reach here. + ASSERT_FALSE(IsRunningOnGvisor()); + // Need a tunnel. + tunnels_.push_back(ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New())); + devices_[device_id] = tunnels_.back()->GetName(); + *device_name = devices_[device_id]; + } + + // Release the socket + void ReleaseSocket(int socket_id) { + // Close the socket that was made in a previous action. The socket_id + // indicates which socket to close based on index into the list of actions. + sockets_to_close_.erase(socket_id); + } + + // Bind a socket with the reuse option and bind_to_device options. Checks + // that all steps succeed and that the bind command's error matches want. + // Sets the socket_id to uniquely identify the socket bound if it is not + // nullptr. + void BindSocket(bool reuse, int device_id = 0, int want = 0, + int *socket_id = nullptr) { + next_socket_id_++; + sockets_to_close_[next_socket_id_] = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + auto socket_fd = sockets_to_close_[next_socket_id_]->get(); + if (socket_id != nullptr) { + *socket_id = next_socket_id_; + } + + // If reuse is indicated, do that. + if (reuse) { + EXPECT_THAT(setsockopt(socket_fd, SOL_SOCKET, SO_REUSEPORT, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceedsWithValue(0)); + } + + // If the device is non-zero, bind to that device. + if (device_id != 0) { + string device_name; + ASSERT_NO_FATAL_FAILURE(GetDevice(device_id, &device_name)); + EXPECT_THAT(setsockopt(socket_fd, SOL_SOCKET, SO_BINDTODEVICE, + device_name.c_str(), device_name.size() + 1), + SyscallSucceedsWithValue(0)); + char get_device[100]; + socklen_t get_device_size = 100; + EXPECT_THAT(getsockopt(socket_fd, SOL_SOCKET, SO_BINDTODEVICE, get_device, + &get_device_size), + SyscallSucceedsWithValue(0)); + } + + struct sockaddr_in addr = {}; + addr.sin_family = AF_INET; + addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK); + addr.sin_port = port_; + if (want == 0) { + ASSERT_THAT( + bind(socket_fd, reinterpret_cast(&addr), + sizeof(addr)), + SyscallSucceeds()); + } else { + ASSERT_THAT( + bind(socket_fd, reinterpret_cast(&addr), + sizeof(addr)), + SyscallFailsWithErrno(want)); + } + + if (port_ == 0) { + // We don't yet know what port we'll be using so we need to fetch it and + // remember it for future commands. + socklen_t addr_size = sizeof(addr); + ASSERT_THAT( + getsockname(socket_fd, reinterpret_cast(&addr), + &addr_size), + SyscallSucceeds()); + port_ = addr.sin_port; + } + } + + private: + SocketKind socket_factory_; + // devices maps from the device id in the test case to the name of the device. + std::unordered_map devices_; + // These are the tunnels that were created for the test and will be destroyed + // by the destructor. + vector> tunnels_; + // A list of all interface names before the test started. + std::unordered_set interface_names_; + // The next ethernet device to use when requested a device. + int next_unused_eth_ = 1; + // The port for all tests. Originally 0 (any) and later set to the port that + // all further commands will use. + in_port_t port_ = 0; + // sockets_to_close_ is a map from action index to the socket that was + // created. + std::unordered_map> + sockets_to_close_; + int next_socket_id_ = 0; +}; + +TEST_P(BindToDeviceSequenceTest, BindTwiceWithDeviceFails) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 3)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 3, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindToDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 1)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 2)); +} + +TEST_P(BindToDeviceSequenceTest, BindToDeviceAndThenWithoutDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindWithoutDevice) { + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ false)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindWithDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 456, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 789, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindWithReuse) { + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true, /* bind_to_device */ 0)); +} + +TEST_P(BindToDeviceSequenceTest, BindingWithReuseAndDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 456)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 789)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 999, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, MixingReuseAndNotReuseByBindingToDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 456, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 789, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 999, 0)); +} + +TEST_P(BindToDeviceSequenceTest, CannotBindTo0AfterMixingReuseAndNotReuse) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 456)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindAndRelease) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + int to_release; + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, 0, &to_release)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 345, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 789)); + // Release the bind to device 0 and try again. + ASSERT_NO_FATAL_FAILURE(ReleaseSocket(to_release)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 345)); +} + +TEST_P(BindToDeviceSequenceTest, BindTwiceWithReuseOnce) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +INSTANTIATE_TEST_SUITE_P(BindToDeviceTest, BindToDeviceSequenceTest, + ::testing::Values(IPv4UDPUnboundSocket(0), + IPv4TCPUnboundSocket(0))); + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_util.cc b/test/syscalls/linux/socket_bind_to_device_util.cc new file mode 100644 index 000000000..f4ee775bd --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_util.cc @@ -0,0 +1,75 @@ +// Copyright 2019 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. + +#include "test/syscalls/linux/socket_bind_to_device_util.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +using std::string; + +PosixErrorOr> Tunnel::New(string tunnel_name) { + int fd; + RETURN_ERROR_IF_SYSCALL_FAIL(fd = open("/dev/net/tun", O_RDWR)); + + // Using `new` to access a non-public constructor. + auto new_tunnel = absl::WrapUnique(new Tunnel(fd)); + + ifreq ifr = {}; + ifr.ifr_flags = IFF_TUN; + strncpy(ifr.ifr_name, tunnel_name.c_str(), sizeof(ifr.ifr_name)); + + RETURN_ERROR_IF_SYSCALL_FAIL(ioctl(fd, TUNSETIFF, &ifr)); + new_tunnel->name_ = ifr.ifr_name; + return new_tunnel; +} + +std::unordered_set GetInterfaceNames() { + struct if_nameindex* interfaces = if_nameindex(); + std::unordered_set names; + if (interfaces == nullptr) { + return names; + } + for (auto interface = interfaces; + interface->if_index != 0 || interface->if_name != nullptr; interface++) { + names.insert(interface->if_name); + } + if_freenameindex(interfaces); + return names; +} + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_util.h b/test/syscalls/linux/socket_bind_to_device_util.h new file mode 100644 index 000000000..f941ccc86 --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_util.h @@ -0,0 +1,67 @@ +// Copyright 2019 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. + +#ifndef GVISOR_TEST_SYSCALLS_SOCKET_BIND_TO_DEVICE_UTILS_H_ +#define GVISOR_TEST_SYSCALLS_SOCKET_BIND_TO_DEVICE_UTILS_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/memory/memory.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +class Tunnel { + public: + static PosixErrorOr> New( + std::string tunnel_name = ""); + const std::string& GetName() const { return name_; } + + ~Tunnel() { + if (fd_ != -1) { + close(fd_); + } + } + + private: + Tunnel(int fd) : fd_(fd) {} + int fd_ = -1; + std::string name_; +}; + +std::unordered_set GetInterfaceNames(); + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_SYSCALLS_SOCKET_BIND_TO_DEVICE_UTILS_H_ diff --git a/test/syscalls/linux/uidgid.cc b/test/syscalls/linux/uidgid.cc index d48453a93..6218fbce1 100644 --- a/test/syscalls/linux/uidgid.cc +++ b/test/syscalls/linux/uidgid.cc @@ -25,6 +25,7 @@ #include "test/util/posix_error.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" +#include "test/util/uid_util.h" ABSL_FLAG(int32_t, scratch_uid1, 65534, "first scratch UID"); ABSL_FLAG(int32_t, scratch_uid2, 65533, "second scratch UID"); @@ -68,30 +69,6 @@ TEST(UidGidTest, Getgroups) { // here; see the setgroups test below. } -// If the caller's real/effective/saved user/group IDs are all 0, IsRoot returns -// true. Otherwise IsRoot logs an explanatory message and returns false. -PosixErrorOr IsRoot() { - uid_t ruid, euid, suid; - int rc = getresuid(&ruid, &euid, &suid); - MaybeSave(); - if (rc < 0) { - return PosixError(errno, "getresuid"); - } - if (ruid != 0 || euid != 0 || suid != 0) { - return false; - } - gid_t rgid, egid, sgid; - rc = getresgid(&rgid, &egid, &sgid); - MaybeSave(); - if (rc < 0) { - return PosixError(errno, "getresgid"); - } - if (rgid != 0 || egid != 0 || sgid != 0) { - return false; - } - return true; -} - // Checks that the calling process' real/effective/saved user IDs are // ruid/euid/suid respectively. PosixError CheckUIDs(uid_t ruid, uid_t euid, uid_t suid) { diff --git a/test/util/BUILD b/test/util/BUILD index 25ed9c944..5d2a9cc2c 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -324,3 +324,14 @@ cc_library( ":test_util", ], ) + +cc_library( + name = "uid_util", + testonly = 1, + srcs = ["uid_util.cc"], + hdrs = ["uid_util.h"], + deps = [ + ":posix_error", + ":save_util", + ], +) diff --git a/test/util/uid_util.cc b/test/util/uid_util.cc new file mode 100644 index 000000000..b131b4b99 --- /dev/null +++ b/test/util/uid_util.cc @@ -0,0 +1,44 @@ +// 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. + +#include "test/util/posix_error.h" +#include "test/util/save_util.h" + +namespace gvisor { +namespace testing { + +PosixErrorOr IsRoot() { + uid_t ruid, euid, suid; + int rc = getresuid(&ruid, &euid, &suid); + MaybeSave(); + if (rc < 0) { + return PosixError(errno, "getresuid"); + } + if (ruid != 0 || euid != 0 || suid != 0) { + return false; + } + gid_t rgid, egid, sgid; + rc = getresgid(&rgid, &egid, &sgid); + MaybeSave(); + if (rc < 0) { + return PosixError(errno, "getresgid"); + } + if (rgid != 0 || egid != 0 || sgid != 0) { + return false; + } + return true; +} + +} // namespace testing +} // namespace gvisor diff --git a/test/util/uid_util.h b/test/util/uid_util.h new file mode 100644 index 000000000..2cd387fb0 --- /dev/null +++ b/test/util/uid_util.h @@ -0,0 +1,29 @@ +// 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. + +#ifndef GVISOR_TEST_SYSCALLS_UID_UTIL_H_ +#define GVISOR_TEST_SYSCALLS_UID_UTIL_H_ + +#include "test/util/posix_error.h" + +namespace gvisor { +namespace testing { + +// Returns true if the caller's real/effective/saved user/group IDs are all 0. +PosixErrorOr IsRoot(); + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_SYSCALLS_UID_UTIL_H_ -- cgit v1.2.3 From c8bb20865d5f2b391b670b391d72e574002363e3 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 27 Sep 2019 15:57:16 -0700 Subject: Automated rollback of changelist 256276198 PiperOrigin-RevId: 271665517 --- test/syscalls/linux/BUILD | 2 -- 1 file changed, 2 deletions(-) (limited to 'test/syscalls/linux/BUILD') diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index e645eebfa..d5a2b7725 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -3237,8 +3237,6 @@ cc_binary( testonly = 1, srcs = ["timers.cc"], linkstatic = 1, - # FIXME(b/136599201) - tags = ["flaky"], deps = [ "//test/util:cleanup", "//test/util:logging", -- cgit v1.2.3