diff options
47 files changed, 713 insertions, 94 deletions
diff --git a/pkg/sentry/fs/tmpfs/inode_file.go b/pkg/sentry/fs/tmpfs/inode_file.go index dabc10662..25abbc151 100644 --- a/pkg/sentry/fs/tmpfs/inode_file.go +++ b/pkg/sentry/fs/tmpfs/inode_file.go @@ -17,6 +17,7 @@ package tmpfs import ( "fmt" "io" + "math" "time" "gvisor.dev/gvisor/pkg/abi/linux" @@ -444,10 +445,15 @@ func (rw *fileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) defer rw.f.dataMu.Unlock() // Compute the range to write. - end := fs.WriteEndOffset(rw.offset, int64(srcs.NumBytes())) - if end == rw.offset { // srcs.NumBytes() == 0? + if srcs.NumBytes() == 0 { + // Nothing to do. return 0, nil } + end := fs.WriteEndOffset(rw.offset, int64(srcs.NumBytes())) + if end == math.MaxInt64 { + // Overflow. + return 0, syserror.EINVAL + } // Check if seals prevent either file growth or all writes. switch { diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go index 6b078cd1e..f6da41c27 100644 --- a/pkg/sentry/platform/ring0/aarch64.go +++ b/pkg/sentry/platform/ring0/aarch64.go @@ -88,14 +88,14 @@ const ( El0Sync_undef El0Sync_dbg El0Sync_inv - VirtualizationException _NR_INTERRUPTS ) // System call vectors. const ( - Syscall Vector = El0Sync_svc - PageFault Vector = El0Sync_da + Syscall Vector = El0Sync_svc + PageFault Vector = El0Sync_da + VirtualizationException Vector = El0Error ) // VirtualAddressBits returns the number bits available for virtual addresses. diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 679842288..baa6c4910 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -601,7 +601,19 @@ TEXT ·El0_fiq(SB),NOSPLIT,$0 B ·Shutdown(SB) TEXT ·El0_error(SB),NOSPLIT,$0 - B ·Shutdown(SB) + KERNEL_ENTRY_FROM_EL0 + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + WORD $0xd538601a //MRS FAR_EL1, R26 + + MOVD R26, CPU_FAULT_ADDR(RSV_REG) + + MOVD $1, R3 + MOVD R3, CPU_ERROR_TYPE(RSV_REG) // Set error type to user. + + MOVD $VirtualizationException, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + + B ·Halt(SB) TEXT ·El0_sync_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go index 8c960c749..057fb5c69 100644 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -85,6 +85,7 @@ func Emit(w io.Writer) { fmt.Fprintf(w, "#define PageFault 0x%02x\n", PageFault) fmt.Fprintf(w, "#define Syscall 0x%02x\n", Syscall) + fmt.Fprintf(w, "#define VirtualizationException 0x%02x\n", VirtualizationException) p := &syscall.PtraceRegs{} fmt.Fprintf(w, "\n// Ptrace registers.\n") diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 8619cc506..049d04bf2 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1260,6 +1260,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_DEFER_ACCEPT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TCPDeferAcceptOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(time.Duration(v) / time.Second), nil + default: emitUnimplementedEventTCP(t, name) } @@ -1713,6 +1725,16 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * v := usermem.ByteOrder.Uint32(optVal) return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPLingerTimeoutOption(time.Second * time.Duration(v)))) + case linux.TCP_DEFER_ACCEPT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + v := int32(usermem.ByteOrder.Uint32(optVal)) + if v < 0 { + v = 0 + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPDeferAcceptOption(time.Second * time.Duration(v)))) + case linux.TCP_REPAIR_OPTIONS: t.Kernel().EmitUnimplementedEvent(t) diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index f43d6c155..fd642834b 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -25,6 +25,10 @@ import ( // doSplice implements a blocking splice operation. func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonBlocking bool) (int64, error) { + if opts.Length < 0 || opts.SrcStart < 0 || opts.DstStart < 0 { + return 0, syserror.EINVAL + } + var ( total int64 n int64 @@ -82,11 +86,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc offsetAddr := args[2].Pointer() count := int64(args[3].SizeT()) - // Don't send a negative number of bytes. - if count < 0 { - return 0, nil, syserror.EINVAL - } - // Get files. inFile := t.GetFile(inFD) if inFile == nil { @@ -136,11 +135,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } - // The offset must be valid. - if offset < 0 { - return 0, nil, syserror.EINVAL - } - // Do the splice. n, err = doSplice(t, outFile, inFile, fs.SpliceOpts{ Length: count, @@ -227,6 +221,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if _, err := t.CopyIn(outOffset, &offset); err != nil { return 0, nil, err } + // Use the destination offset. opts.DstOffset = true opts.DstStart = offset @@ -244,6 +239,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if _, err := t.CopyIn(inOffset, &offset); err != nil { return 0, nil, err } + // Use the source offset. opts.SrcOffset = true opts.SrcStart = offset diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index 245694118..31294345d 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -167,8 +167,8 @@ type NDPDispatcher interface { // reason, such as the address being removed). If an error occured // during DAD, err will be set and resolved must be ignored. // - // This function is permitted to block indefinitely without interfering - // with the stack's operation. + // This function is not permitted to block indefinitely. This function + // is also not permitted to call into the stack. OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) // OnDefaultRouterDiscovered will be called when a new default router is @@ -538,29 +538,11 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address) *tcpip.Error { r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, snmc, ndp.nic.linkEP.LinkAddress(), ref, false, false) defer r.Release() - linkAddr := ndp.nic.linkEP.LinkAddress() - isValidLinkAddr := header.IsValidUnicastEthernetAddress(linkAddr) - ndpNSSize := header.ICMPv6NeighborSolicitMinimumSize - if isValidLinkAddr { - // Only include a Source Link Layer Address option if the NIC has a valid - // link layer address. - // - // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by - // LinkEndpoint.LinkAddress) before reaching this point. - ndpNSSize += header.NDPLinkLayerAddressSize - } - - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + ndpNSSize) - pkt := header.ICMPv6(hdr.Prepend(ndpNSSize)) + hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborSolicitMinimumSize) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) pkt.SetType(header.ICMPv6NeighborSolicit) ns := header.NDPNeighborSolicit(pkt.NDPPayload()) ns.SetTargetAddress(addr) - - if isValidLinkAddr { - ns.Options().Serialize(header.NDPOptionsSerializer{ - header.NDPSourceLinkLayerAddressOption(linkAddr), - }) - } pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) sent := r.Stats().ICMP.V6PacketsSent @@ -607,8 +589,8 @@ func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { delete(ndp.dad, addr) // Let the integrator know DAD did not resolve. - if ndp.nic.stack.ndpDisp != nil { - go ndp.nic.stack.ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, false, nil) + if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, false, nil) } } diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 726468e41..bc7cfbcb4 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -413,14 +413,18 @@ func TestDADResolve(t *testing.T) { t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) } - // Check NDP packet. + // Check NDP NS packet. + // + // As per RFC 4861 section 4.3, a possible option is the Source Link + // Layer option, but this option MUST NOT be included when the source + // address of the packet is the unspecified address. checker.IPv6(t, p.Pkt.Header.View().ToVectorisedView().First(), + checker.SrcAddr(header.IPv6Any), + checker.DstAddr(header.SolicitedNodeAddr(addr1)), checker.TTL(header.NDPHopLimit), checker.NDPNS( checker.NDPNSTargetAddress(addr1), - checker.NDPNSOptions([]header.NDPOption{ - header.NDPSourceLinkLayerAddressOption(linkAddr1), - }), + checker.NDPNSOptions(nil), )) } }) @@ -497,7 +501,7 @@ func TestDADFail(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { ndpDisp := ndpDispatcher{ - dadC: make(chan ndpDADEvent), + dadC: make(chan ndpDADEvent, 1), } ndpConfigs := stack.DefaultNDPConfigurations() opts := stack.Options{ @@ -576,7 +580,7 @@ func TestDADFail(t *testing.T) { // removed. func TestDADStop(t *testing.T) { ndpDisp := ndpDispatcher{ - dadC: make(chan ndpDADEvent), + dadC: make(chan ndpDADEvent, 1), } ndpConfigs := stack.NDPConfigurations{ RetransmitTimer: time.Second, diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 59c9b3fb0..0fa141d58 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -626,6 +626,12 @@ type TCPLingerTimeoutOption time.Duration // before being marked closed. type TCPTimeWaitTimeoutOption time.Duration +// TCPDeferAcceptOption is used by SetSockOpt/GetSockOpt to allow a +// accept to return a completed connection only when there is data to be +// read. This usually means the listening socket will drop the final ACK +// for a handshake till the specified timeout until a segment with data arrives. +type TCPDeferAcceptOption time.Duration + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default // TTL value for multicast messages. The default is 1. type MulticastTTLOption uint8 diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 4acd9fb9a..272e8f570 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -57,6 +57,7 @@ go_library( imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ + "//pkg/log", "//pkg/rand", "//pkg/sleep", "//pkg/sync", @@ -90,6 +91,7 @@ go_test( tags = ["flaky"], deps = [ ":tcp", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/checker", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index d469758eb..6101f2945 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -222,13 +222,13 @@ func (l *listenContext) isCookieValid(id stack.TransportEndpointID, cookie seqnu // createConnectingEndpoint creates a new endpoint in a connecting state, with // the connection parameters given by the arguments. -func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, irs seqnum.Value, rcvdSynOpts *header.TCPSynOptions) (*endpoint, *tcpip.Error) { +func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, irs seqnum.Value, rcvdSynOpts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { // Create a new endpoint. netProto := l.netProto if netProto == 0 { netProto = s.route.NetProto } - n := newEndpoint(l.stack, netProto, nil) + n := newEndpoint(l.stack, netProto, queue) n.v6only = l.v6only n.ID = s.id n.boundNICID = s.route.NICID() @@ -273,16 +273,17 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i // createEndpoint creates a new endpoint in connected state and then performs // the TCP 3-way handshake. -func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions) (*endpoint, *tcpip.Error) { +func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { // Create new endpoint. irs := s.sequenceNumber isn := generateSecureISN(s.id, l.stack.Seed()) - ep, err := l.createConnectingEndpoint(s, isn, irs, opts) + ep, err := l.createConnectingEndpoint(s, isn, irs, opts, queue) if err != nil { return nil, err } // listenEP is nil when listenContext is used by tcp.Forwarder. + deferAccept := time.Duration(0) if l.listenEP != nil { l.listenEP.mu.Lock() if l.listenEP.EndpointState() != StateListen { @@ -290,13 +291,12 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head return nil, tcpip.ErrConnectionAborted } l.addPendingEndpoint(ep) + deferAccept = l.listenEP.deferAccept l.listenEP.mu.Unlock() } // Perform the 3-way handshake. - h := newHandshake(ep, seqnum.Size(ep.initialReceiveWindow())) - - h.resetToSynRcvd(isn, irs, opts) + h := newPassiveHandshake(ep, seqnum.Size(ep.initialReceiveWindow()), isn, irs, opts, deferAccept) if err := h.execute(); err != nil { ep.Close() if l.listenEP != nil { @@ -377,16 +377,14 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header defer e.decSynRcvdCount() defer s.decRef() - n, err := ctx.createEndpointAndPerformHandshake(s, opts) + n, err := ctx.createEndpointAndPerformHandshake(s, opts, &waiter.Queue{}) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() e.stats.FailedConnectionAttempts.Increment() return } ctx.removePendingEndpoint(n) - // Start the protocol goroutine. - wq := &waiter.Queue{} - n.startAcceptedLoop(wq) + n.startAcceptedLoop() e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() e.deliverAccepted(n) @@ -546,7 +544,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { rcvdSynOptions.TSEcr = s.parsedOptions.TSEcr } - n, err := ctx.createConnectingEndpoint(s, s.ackNumber-1, s.sequenceNumber-1, rcvdSynOptions) + n, err := ctx.createConnectingEndpoint(s, s.ackNumber-1, s.sequenceNumber-1, rcvdSynOptions, &waiter.Queue{}) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() e.stats.FailedConnectionAttempts.Increment() @@ -576,8 +574,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // space available in the backlog. // Start the protocol goroutine. - wq := &waiter.Queue{} - n.startAcceptedLoop(wq) + n.startAcceptedLoop() e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() go e.deliverAccepted(n) } diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 4e3c5419c..9ff7ac261 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -86,6 +86,19 @@ type handshake struct { // rcvWndScale is the receive window scale, as defined in RFC 1323. rcvWndScale int + + // startTime is the time at which the first SYN/SYN-ACK was sent. + startTime time.Time + + // deferAccept if non-zero will drop the final ACK for a passive + // handshake till an ACK segment with data is received or the timeout is + // hit. + deferAccept time.Duration + + // acked is true if the the final ACK for a 3-way handshake has + // been received. This is required to stop retransmitting the + // original SYN-ACK when deferAccept is enabled. + acked bool } func newHandshake(ep *endpoint, rcvWnd seqnum.Size) handshake { @@ -112,6 +125,12 @@ func newHandshake(ep *endpoint, rcvWnd seqnum.Size) handshake { return h } +func newPassiveHandshake(ep *endpoint, rcvWnd seqnum.Size, isn, irs seqnum.Value, opts *header.TCPSynOptions, deferAccept time.Duration) handshake { + h := newHandshake(ep, rcvWnd) + h.resetToSynRcvd(isn, irs, opts, deferAccept) + return h +} + // FindWndScale determines the window scale to use for the given maximum window // size. func FindWndScale(wnd seqnum.Size) int { @@ -181,7 +200,7 @@ func (h *handshake) effectiveRcvWndScale() uint8 { // resetToSynRcvd resets the state of the handshake object to the SYN-RCVD // state. -func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *header.TCPSynOptions) { +func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *header.TCPSynOptions, deferAccept time.Duration) { h.active = false h.state = handshakeSynRcvd h.flags = header.TCPFlagSyn | header.TCPFlagAck @@ -189,6 +208,7 @@ func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *hea h.ackNum = irs + 1 h.mss = opts.MSS h.sndWndScale = opts.WS + h.deferAccept = deferAccept h.ep.mu.Lock() h.ep.setEndpointState(StateSynRecv) h.ep.mu.Unlock() @@ -352,6 +372,14 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { // We have previously received (and acknowledged) the peer's SYN. If the // peer acknowledges our SYN, the handshake is completed. if s.flagIsSet(header.TCPFlagAck) { + // If deferAccept is not zero and this is a bare ACK and the + // timeout is not hit then drop the ACK. + if h.deferAccept != 0 && s.data.Size() == 0 && time.Since(h.startTime) < h.deferAccept { + h.acked = true + h.ep.stack.Stats().DroppedPackets.Increment() + return nil + } + // If the timestamp option is negotiated and the segment does // not carry a timestamp option then the segment must be dropped // as per https://tools.ietf.org/html/rfc7323#section-3.2. @@ -365,10 +393,16 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { h.ep.updateRecentTimestamp(s.parsedOptions.TSVal, h.ackNum, s.sequenceNumber) } h.state = handshakeCompleted + h.ep.mu.Lock() h.ep.transitionToStateEstablishedLocked(h) + // If the segment has data then requeue it for the receiver + // to process it again once main loop is started. + if s.data.Size() > 0 { + s.incRef() + h.ep.enqueueSegment(s) + } h.ep.mu.Unlock() - return nil } @@ -471,6 +505,7 @@ func (h *handshake) execute() *tcpip.Error { } } + h.startTime = time.Now() // Initialize the resend timer. resendWaker := sleep.Waker{} timeOut := time.Duration(time.Second) @@ -524,11 +559,21 @@ func (h *handshake) execute() *tcpip.Error { switch index, _ := s.Fetch(true); index { case wakerForResend: timeOut *= 2 - if timeOut > 60*time.Second { + if timeOut > MaxRTO { return tcpip.ErrTimeout } rt.Reset(timeOut) - h.ep.sendSynTCP(&h.ep.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + // Resend the SYN/SYN-ACK only if the following conditions hold. + // - It's an active handshake (deferAccept does not apply) + // - It's a passive handshake and we have not yet got the final-ACK. + // - It's a passive handshake and we got an ACK but deferAccept is + // enabled and we are now past the deferAccept duration. + // The last is required to provide a way for the peer to complete + // the connection with another ACK or data (as ACKs are never + // retransmitted on their own). + if h.active || !h.acked || h.deferAccept != 0 && time.Since(h.startTime) > h.deferAccept { + h.ep.sendSynTCP(&h.ep.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + } case wakerForNotification: n := h.ep.fetchNotifications() diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 13718ff55..b5a8e15ee 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -498,6 +498,13 @@ type endpoint struct { // without any data being acked. userTimeout time.Duration + // deferAccept if non-zero specifies a user specified time during + // which the final ACK of a handshake will be dropped provided the + // ACK is a bare ACK and carries no data. If the timeout is crossed then + // the bare ACK is accepted and the connection is delivered to the + // listener. + deferAccept time.Duration + // pendingAccepted is a synchronization primitive used to track number // of connections that are queued up to be delivered to the accepted // channel. We use this to ensure that all goroutines blocked on writing @@ -1574,6 +1581,15 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.TCPDeferAcceptOption: + e.mu.Lock() + if time.Duration(v) > MaxRTO { + v = tcpip.TCPDeferAcceptOption(MaxRTO) + } + e.deferAccept = time.Duration(v) + e.mu.Unlock() + return nil + default: return nil } @@ -1798,6 +1814,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case *tcpip.TCPDeferAcceptOption: + e.mu.Lock() + *o = tcpip.TCPDeferAcceptOption(e.deferAccept) + e.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -2025,8 +2047,14 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // work mutex is available. if e.workMu.TryLock() { e.mu.Lock() - e.resetConnectionLocked(tcpip.ErrConnectionAborted) - e.notifyProtocolGoroutine(notifyTickleWorker) + // We need to double check here to make + // sure worker has not transitioned the + // endpoint out of a connected state + // before trying to send a reset. + if e.EndpointState().connected() { + e.resetConnectionLocked(tcpip.ErrConnectionAborted) + e.notifyProtocolGoroutine(notifyTickleWorker) + } e.mu.Unlock() e.workMu.Unlock() } else { @@ -2149,9 +2177,8 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { // startAcceptedLoop sets up required state and starts a goroutine with the // main loop for accepted connections. -func (e *endpoint) startAcceptedLoop(waiterQueue *waiter.Queue) { +func (e *endpoint) startAcceptedLoop() { e.mu.Lock() - e.waiterQueue = waiterQueue e.workerRunning = true e.mu.Unlock() wakerInitDone := make(chan struct{}) @@ -2177,7 +2204,6 @@ func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { default: return nil, nil, tcpip.ErrWouldBlock } - return n, n.waiterQueue, nil } diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go index 7eb613be5..c9ee5bf06 100644 --- a/pkg/tcpip/transport/tcp/forwarder.go +++ b/pkg/tcpip/transport/tcp/forwarder.go @@ -157,13 +157,13 @@ func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, TSVal: r.synOptions.TSVal, TSEcr: r.synOptions.TSEcr, SACKPermitted: r.synOptions.SACKPermitted, - }) + }, queue) if err != nil { return nil, err } // Start the protocol goroutine. - ep.startAcceptedLoop(queue) + ep.startAcceptedLoop() return ep, nil } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index df2fb1071..2c1505067 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -21,6 +21,7 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/checker" @@ -6787,3 +6788,183 @@ func TestIncreaseWindowOnBufferResize(t *testing.T) { ), ) } + +func TestTCPDeferAccept(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10); err != nil { + t.Fatal("Listen failed:", err) + } + + const tcpDeferAccept = 1 * time.Second + if err := c.EP.SetSockOpt(tcpip.TCPDeferAcceptOption(tcpDeferAccept)); err != nil { + t.Fatalf("c.EP.SetSockOpt(TCPDeferAcceptOption(%s) failed: %v", tcpDeferAccept, err) + } + + irs, iss := executeHandshake(t, c, context.TestPort, false /* synCookiesInUse */) + + if _, _, err := c.EP.Accept(); err != tcpip.ErrWouldBlock { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: %s", err, tcpip.ErrWouldBlock) + } + + // Send data. This should result in an acceptable endpoint. + c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + }) + + // Receive ACK for the data we sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) + + // Give a bit of time for the socket to be delivered to the accept queue. + time.Sleep(50 * time.Millisecond) + aep, _, err := c.EP.Accept() + if err != nil { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: nil", err) + } + + aep.Close() + // Closing aep without reading the data should trigger a RST. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) +} + +func TestTCPDeferAcceptTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10); err != nil { + t.Fatal("Listen failed:", err) + } + + const tcpDeferAccept = 1 * time.Second + if err := c.EP.SetSockOpt(tcpip.TCPDeferAcceptOption(tcpDeferAccept)); err != nil { + t.Fatalf("c.EP.SetSockOpt(TCPDeferAcceptOption(%s) failed: %v", tcpDeferAccept, err) + } + + irs, iss := executeHandshake(t, c, context.TestPort, false /* synCookiesInUse */) + + if _, _, err := c.EP.Accept(); err != tcpip.ErrWouldBlock { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: %s", err, tcpip.ErrWouldBlock) + } + + // Sleep for a little of the tcpDeferAccept timeout. + time.Sleep(tcpDeferAccept + 100*time.Millisecond) + + // On timeout expiry we should get a SYN-ACK retransmission. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), + checker.AckNum(uint32(irs)+1))) + + // Send data. This should result in an acceptable endpoint. + c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + }) + + // Receive ACK for the data we sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) + + // Give sometime for the endpoint to be delivered to the accept queue. + time.Sleep(50 * time.Millisecond) + aep, _, err := c.EP.Accept() + if err != nil { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: nil", err) + } + + aep.Close() + // Closing aep without reading the data should trigger a RST. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) +} + +func TestResetDuringClose(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + iss := seqnum.Value(789) + c.CreateConnected(iss, 30000, -1 /* epRecvBuf */) + // Send some data to make sure there is some unread + // data to trigger a reset on c.Close. + irs := c.IRS + c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: iss.Add(1), + AckNum: irs.Add(1), + RcvWnd: 30000, + }) + + // Receive ACK for the data we sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(irs.Add(1))), + checker.AckNum(uint32(iss.Add(5))))) + + // Close in a separate goroutine so that we can trigger + // a race with the RST we send below. This should not + // panic due to the route being released depeding on + // whether Close() sends an active RST or the RST sent + // below is processed by the worker first. + var wg sync.WaitGroup + + wg.Add(1) + go func() { + defer wg.Done() + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + SeqNum: iss.Add(5), + AckNum: c.IRS.Add(5), + RcvWnd: 30000, + Flags: header.TCPFlagRst, + }) + }() + + wg.Add(1) + go func() { + defer wg.Done() + c.EP.Close() + }() + + wg.Wait() +} diff --git a/scripts/iptables_tests.sh b/scripts/iptables_tests.sh index c47cbd675..3069d8628 100755 --- a/scripts/iptables_tests.sh +++ b/scripts/iptables_tests.sh @@ -19,9 +19,9 @@ source $(dirname $0)/common.sh install_runsc_for_test iptables # Build the docker image for the test. -run //test/iptables/runner --norun +run //test/iptables/runner-image --norun # TODO(gvisor.dev/issue/170): Also test this on runsc once iptables are better # supported test //test/iptables:iptables_test "--test_arg=--runtime=runc" \ - "--test_arg=--image=bazel/test/iptables/runner:runner" + "--test_arg=--image=bazel/test/iptables/runner:runner-image" diff --git a/test/iptables/README.md b/test/iptables/README.md index 9f8e34420..8f61b4c41 100644 --- a/test/iptables/README.md +++ b/test/iptables/README.md @@ -28,7 +28,7 @@ Your test is now runnable with bazel! Build the testing Docker container: ```bash -$ bazel run //test/iptables/runner -- --norun +$ bazel run //test/iptables/runner-image -- --norun ``` Run an individual test via: diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 679a29bef..41909582a 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -30,7 +30,7 @@ import ( const timeout = 18 * time.Second -var image = flag.String("image", "bazel/test/iptables/runner:runner", "image to run tests in") +var image = flag.String("image", "bazel/test/iptables/runner:runner-image", "image to run tests in") type result struct { output string diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 74bf068ec..6f57c9755 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -754,6 +754,7 @@ cc_binary( ":socket_test_util", "//test/util:cleanup", "//test/util:eventfd_util", + "//test/util:fs_util", "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:save_util", @@ -1391,6 +1392,7 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "//test/util:fs_util", "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", @@ -2173,6 +2175,7 @@ cc_library( ":socket_test_util", "//test/util:test_util", "//test/util:thread_util", + "@com_google_absl//absl/memory", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 4f3aa81d6..421c15b87 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -31,6 +31,7 @@ #include "test/syscalls/linux/socket_test_util.h" #include "test/util/cleanup.h" #include "test/util/eventfd_util.h" +#include "test/util/fs_util.h" #include "test/util/multiprocess_util.h" #include "test/util/posix_error.h" #include "test/util/save_util.h" @@ -55,10 +56,6 @@ ABSL_FLAG(int32_t, socket_fd, -1, namespace gvisor { namespace testing { -// O_LARGEFILE as defined by Linux. glibc tries to be clever by setting it to 0 -// because "it isn't needed", even though Linux can return it via F_GETFL. -constexpr int kOLargeFile = 00100000; - class FcntlLockTest : public ::testing::Test { public: void SetUp() override { diff --git a/test/syscalls/linux/pipe.cc b/test/syscalls/linux/pipe.cc index ac9b21b24..d8e19e910 100644 --- a/test/syscalls/linux/pipe.cc +++ b/test/syscalls/linux/pipe.cc @@ -25,6 +25,7 @@ #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/file_descriptor.h" +#include "test/util/fs_util.h" #include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -144,11 +145,10 @@ TEST_P(PipeTest, Flags) { if (IsNamedPipe()) { // May be stubbed to zero; define locally. - constexpr int kLargefile = 0100000; EXPECT_THAT(fcntl(rfd_.get(), F_GETFL), - SyscallSucceedsWithValue(kLargefile | O_RDONLY)); + SyscallSucceedsWithValue(kOLargeFile | O_RDONLY)); EXPECT_THAT(fcntl(wfd_.get(), F_GETFL), - SyscallSucceedsWithValue(kLargefile | O_WRONLY)); + SyscallSucceedsWithValue(kOLargeFile | O_WRONLY)); } else { EXPECT_THAT(fcntl(rfd_.get(), F_GETFL), SyscallSucceedsWithValue(O_RDONLY)); EXPECT_THAT(fcntl(wfd_.get(), F_GETFL), SyscallSucceedsWithValue(O_WRONLY)); diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index bf9bb45d3..a03c1e43d 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -100,18 +100,6 @@ namespace { #define SUID_DUMP_ROOT 2 #endif /* SUID_DUMP_ROOT */ -// O_LARGEFILE as defined by Linux. glibc tries to be clever by setting it to 0 -// because "it isn't needed", even though Linux can return it via F_GETFL. -#if defined(__x86_64__) || defined(__i386__) -constexpr int kOLargeFile = 00100000; -#elif __aarch64__ -// The value originate from the Linux -// kernel's arch/arm64/include/uapi/asm/fcntl.h. -constexpr int kOLargeFile = 00400000; -#else -#error "Unknown architecture" -#endif - #if defined(__x86_64__) || defined(__i386__) // This list of "required" fields is taken from reading the file // arch/x86/kernel/cpu/proc.c and seeing which fields will be unconditionally diff --git a/test/syscalls/linux/socket_abstract.cc b/test/syscalls/linux/socket_abstract.cc index 715d87b76..00999f192 100644 --- a/test/syscalls/linux/socket_abstract.cc +++ b/test/syscalls/linux/socket_abstract.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVec<SocketPairKind>( @@ -43,5 +44,6 @@ INSTANTIATE_TEST_SUITE_P( AbstractUnixSockets, UnixSocketPairCmsgTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_filesystem.cc b/test/syscalls/linux/socket_filesystem.cc index 74e262959..287359363 100644 --- a/test/syscalls/linux/socket_filesystem.cc +++ b/test/syscalls/linux/socket_filesystem.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVec<SocketPairKind>( @@ -43,5 +44,6 @@ INSTANTIATE_TEST_SUITE_P( FilesystemUnixSockets, UnixSocketPairCmsgTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index 2f9821555..3bf7081b9 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -828,6 +828,164 @@ TEST_P(SocketInetLoopbackTest, AcceptedInheritsTCPUserTimeout) { EXPECT_EQ(get, kUserTimeout); } +// TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not +// saved. Enable S/R once issue is fixed. +TEST_P(SocketInetLoopbackTest, TCPDeferAccept_NoRandomSave) { + // TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not + // saved. Enable S/R issue is fixed. + DisableSave ds; + + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast<sockaddr*>(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast<sockaddr*>(&listen_addr), &addrlen), + SyscallSucceeds()); + + const uint16_t port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Set the TCP_DEFER_ACCEPT on the listening socket. + constexpr int kTCPDeferAccept = 3; + ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, + &kTCPDeferAccept, sizeof(kTCPDeferAccept)), + SyscallSucceeds()); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast<sockaddr*>(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Set the listening socket to nonblock so that we can verify that there is no + // connection in queue despite the connect above succeeding since the peer has + // sent no data and TCP_DEFER_ACCEPT is set on the listening socket. Set the + // FD to O_NONBLOCK. + int opts; + ASSERT_THAT(opts = fcntl(listen_fd.get(), F_GETFL), SyscallSucceeds()); + opts |= O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + ASSERT_THAT(accept(listen_fd.get(), nullptr, nullptr), + SyscallFailsWithErrno(EWOULDBLOCK)); + + // Set FD back to blocking. + opts &= ~O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + // Now write some data to the socket. + int data = 0; + ASSERT_THAT(RetryEINTR(write)(conn_fd.get(), &data, sizeof(data)), + SyscallSucceedsWithValue(sizeof(data))); + + // This should now cause the connection to complete and be delivered to the + // accept socket. + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + + // Verify that the accepted socket returns the data written. + int get = -1; + ASSERT_THAT(RetryEINTR(recv)(accepted.get(), &get, sizeof(get), 0), + SyscallSucceedsWithValue(sizeof(get))); + + EXPECT_EQ(get, data); +} + +// TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not +// saved. Enable S/R once issue is fixed. +TEST_P(SocketInetLoopbackTest, TCPDeferAcceptTimeout_NoRandomSave) { + // TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not + // saved. Enable S/R once issue is fixed. + DisableSave ds; + + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast<sockaddr*>(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast<sockaddr*>(&listen_addr), &addrlen), + SyscallSucceeds()); + + const uint16_t port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Set the TCP_DEFER_ACCEPT on the listening socket. + constexpr int kTCPDeferAccept = 3; + ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, + &kTCPDeferAccept, sizeof(kTCPDeferAccept)), + SyscallSucceeds()); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast<sockaddr*>(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Set the listening socket to nonblock so that we can verify that there is no + // connection in queue despite the connect above succeeding since the peer has + // sent no data and TCP_DEFER_ACCEPT is set on the listening socket. Set the + // FD to O_NONBLOCK. + int opts; + ASSERT_THAT(opts = fcntl(listen_fd.get(), F_GETFL), SyscallSucceeds()); + opts |= O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + // Verify that there is no acceptable connection before TCP_DEFER_ACCEPT + // timeout is hit. + absl::SleepFor(absl::Seconds(kTCPDeferAccept - 1)); + ASSERT_THAT(accept(listen_fd.get(), nullptr, nullptr), + SyscallFailsWithErrno(EWOULDBLOCK)); + + // Set FD back to blocking. + opts &= ~O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + // Now sleep for a little over the TCP_DEFER_ACCEPT duration. When the timeout + // is hit a SYN-ACK should be retransmitted by the listener as a last ditch + // attempt to complete the connection with or without data. + absl::SleepFor(absl::Seconds(2)); + + // Verify that we have a connection that can be accepted even though no + // data was written. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); +} + INSTANTIATE_TEST_SUITE_P( All, SocketInetLoopbackTest, ::testing::Values( diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc index 57ce8e169..27779e47c 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic.cc @@ -24,6 +24,7 @@ #include <sys/un.h> #include "gtest/gtest.h" +#include "absl/memory/memory.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/syscalls/linux/socket_test_util.h" @@ -875,5 +876,37 @@ TEST_P(TCPSocketPairTest, SetTCPUserTimeoutAboveZero) { EXPECT_EQ(get, kAbove); } +TEST_P(TCPSocketPairTest, TCPResetDuringClose_NoRandomSave) { + DisableSave ds; // Too many syscalls. + constexpr int kThreadCount = 1000; + std::unique_ptr<ScopedThread> instances[kThreadCount]; + for (int i = 0; i < kThreadCount; i++) { + instances[i] = absl::make_unique<ScopedThread>([&]() { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + ScopedThread t([&]() { + // Close one end to trigger sending of a FIN. + struct pollfd poll_fd = {sockets->second_fd(), POLLIN | POLLHUP, 0}; + // Wait up to 20 seconds for the data. + constexpr int kPollTimeoutMs = 20000; + ASSERT_THAT(RetryEINTR(poll)(&poll_fd, 1, kPollTimeoutMs), + SyscallSucceedsWithValue(1)); + ASSERT_THAT(close(sockets->release_second_fd()), SyscallSucceeds()); + }); + + // Send some data then close. + constexpr char kStr[] = "abc"; + ASSERT_THAT(write(sockets->first_fd(), kStr, 3), + SyscallSucceedsWithValue(3)); + absl::SleepFor(absl::Milliseconds(10)); + ASSERT_THAT(close(sockets->release_first_fd()), SyscallSucceeds()); + t.Join(); + }); + } + for (int i = 0; i < kThreadCount; i++) { + instances[i]->Join(); + } +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc b/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc index d11f7cc23..4e79d21f4 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVecToVec<SocketPairKind>( @@ -39,5 +40,6 @@ INSTANTIATE_TEST_SUITE_P( AllTCPSockets, TCPSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc b/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc index fcd20102f..f996b93d2 100644 --- a/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc +++ b/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVecToVec<SocketPairKind>( @@ -39,5 +40,6 @@ INSTANTIATE_TEST_SUITE_P( BlockingTCPSockets, BlockingStreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc b/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc index 63a05b799..ffa377210 100644 --- a/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc +++ b/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVecToVec<SocketPairKind>( @@ -38,5 +39,6 @@ INSTANTIATE_TEST_SUITE_P( NonBlockingTCPSockets, NonBlockingSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ip_udp_loopback.cc b/test/syscalls/linux/socket_ip_udp_loopback.cc index 1df74a348..c7fa44884 100644 --- a/test/syscalls/linux/socket_ip_udp_loopback.cc +++ b/test/syscalls/linux/socket_ip_udp_loopback.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return { @@ -44,5 +45,6 @@ INSTANTIATE_TEST_SUITE_P( AllUDPSockets, UDPSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ip_udp_loopback_blocking.cc b/test/syscalls/linux/socket_ip_udp_loopback_blocking.cc index 1e259efa7..d6925a8df 100644 --- a/test/syscalls/linux/socket_ip_udp_loopback_blocking.cc +++ b/test/syscalls/linux/socket_ip_udp_loopback_blocking.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return { @@ -33,5 +34,6 @@ INSTANTIATE_TEST_SUITE_P( BlockingUDPSockets, BlockingNonStreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ip_udp_loopback_nonblock.cc b/test/syscalls/linux/socket_ip_udp_loopback_nonblock.cc index 74cbd326d..d675eddc6 100644 --- a/test/syscalls/linux/socket_ip_udp_loopback_nonblock.cc +++ b/test/syscalls/linux/socket_ip_udp_loopback_nonblock.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return { @@ -33,5 +34,6 @@ INSTANTIATE_TEST_SUITE_P( NonBlockingUDPSockets, NonBlockingSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_abstract_nonblock.cc b/test/syscalls/linux/socket_unix_abstract_nonblock.cc index be31ab2a7..8bef76b67 100644 --- a/test/syscalls/linux/socket_unix_abstract_nonblock.cc +++ b/test/syscalls/linux/socket_unix_abstract_nonblock.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVec<SocketPairKind>( @@ -33,5 +34,6 @@ INSTANTIATE_TEST_SUITE_P( NonBlockingAbstractUnixSockets, NonBlockingSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_blocking_local.cc b/test/syscalls/linux/socket_unix_blocking_local.cc index 6f84221b2..77cb8c6d6 100644 --- a/test/syscalls/linux/socket_unix_blocking_local.cc +++ b/test/syscalls/linux/socket_unix_blocking_local.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return VecCat<SocketPairKind>( @@ -39,5 +40,6 @@ INSTANTIATE_TEST_SUITE_P( NonBlockingUnixDomainSockets, BlockingSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_dgram_local.cc b/test/syscalls/linux/socket_unix_dgram_local.cc index 9134fcdf7..31d2d5216 100644 --- a/test/syscalls/linux/socket_unix_dgram_local.cc +++ b/test/syscalls/linux/socket_unix_dgram_local.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return VecCat<SocketPairKind>(VecCat<SocketPairKind>( @@ -52,5 +53,6 @@ INSTANTIATE_TEST_SUITE_P( DgramUnixSockets, NonStreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_domain.cc b/test/syscalls/linux/socket_unix_domain.cc index fa3efc7f8..f7dff8b4d 100644 --- a/test/syscalls/linux/socket_unix_domain.cc +++ b/test/syscalls/linux/socket_unix_domain.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVec<SocketPairKind>( @@ -33,5 +34,6 @@ INSTANTIATE_TEST_SUITE_P( AllUnixDomainSockets, AllSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_filesystem_nonblock.cc b/test/syscalls/linux/socket_unix_filesystem_nonblock.cc index 8ba7af971..6700b4d90 100644 --- a/test/syscalls/linux/socket_unix_filesystem_nonblock.cc +++ b/test/syscalls/linux/socket_unix_filesystem_nonblock.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVec<SocketPairKind>( @@ -33,5 +34,6 @@ INSTANTIATE_TEST_SUITE_P( NonBlockingFilesystemUnixSockets, NonBlockingSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc b/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc index 8855d5001..fddcdf1c5 100644 --- a/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc +++ b/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return VecCat<SocketPairKind>( @@ -36,5 +37,6 @@ INSTANTIATE_TEST_SUITE_P( BlockingNonStreamUnixSockets, BlockingNonStreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_pair.cc b/test/syscalls/linux/socket_unix_pair.cc index 411fb4518..85999db04 100644 --- a/test/syscalls/linux/socket_unix_pair.cc +++ b/test/syscalls/linux/socket_unix_pair.cc @@ -22,6 +22,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return VecCat<SocketPairKind>(ApplyVec<SocketPairKind>( @@ -38,5 +39,6 @@ INSTANTIATE_TEST_SUITE_P( AllUnixDomainSockets, UnixSocketPairCmsgTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_pair_nonblock.cc b/test/syscalls/linux/socket_unix_pair_nonblock.cc index 3135d325f..281410a9a 100644 --- a/test/syscalls/linux/socket_unix_pair_nonblock.cc +++ b/test/syscalls/linux/socket_unix_pair_nonblock.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return ApplyVec<SocketPairKind>( @@ -33,5 +34,6 @@ INSTANTIATE_TEST_SUITE_P( NonBlockingUnixSockets, NonBlockingSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_seqpacket_local.cc b/test/syscalls/linux/socket_unix_seqpacket_local.cc index dff75a532..69a5f150d 100644 --- a/test/syscalls/linux/socket_unix_seqpacket_local.cc +++ b/test/syscalls/linux/socket_unix_seqpacket_local.cc @@ -23,6 +23,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return VecCat<SocketPairKind>(VecCat<SocketPairKind>( @@ -52,5 +53,6 @@ INSTANTIATE_TEST_SUITE_P( SeqpacketUnixSockets, UnixNonStreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_stream_blocking_local.cc b/test/syscalls/linux/socket_unix_stream_blocking_local.cc index 08e579ba7..8429bd429 100644 --- a/test/syscalls/linux/socket_unix_stream_blocking_local.cc +++ b/test/syscalls/linux/socket_unix_stream_blocking_local.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return { @@ -34,5 +35,6 @@ INSTANTIATE_TEST_SUITE_P( BlockingStreamUnixSockets, BlockingStreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_stream_local.cc b/test/syscalls/linux/socket_unix_stream_local.cc index 65eef1a81..a7e3449a9 100644 --- a/test/syscalls/linux/socket_unix_stream_local.cc +++ b/test/syscalls/linux/socket_unix_stream_local.cc @@ -21,6 +21,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return VecCat<SocketPairKind>( @@ -42,5 +43,6 @@ INSTANTIATE_TEST_SUITE_P( StreamUnixSockets, StreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_unix_stream_nonblock_local.cc b/test/syscalls/linux/socket_unix_stream_nonblock_local.cc index 1936aa135..4b763c8e2 100644 --- a/test/syscalls/linux/socket_unix_stream_nonblock_local.cc +++ b/test/syscalls/linux/socket_unix_stream_nonblock_local.cc @@ -20,6 +20,7 @@ namespace gvisor { namespace testing { +namespace { std::vector<SocketPairKind> GetSocketPairs() { return { @@ -33,5 +34,6 @@ INSTANTIATE_TEST_SUITE_P( NonBlockingStreamUnixSockets, NonBlockingStreamSocketPairTest, ::testing::ValuesIn(IncludeReversals(GetSocketPairs()))); +} // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/splice.cc b/test/syscalls/linux/splice.cc index 85232cb1f..faa1247f6 100644 --- a/test/syscalls/linux/splice.cc +++ b/test/syscalls/linux/splice.cc @@ -60,6 +60,62 @@ TEST(SpliceTest, TwoRegularFiles) { SyscallFailsWithErrno(EINVAL)); } +int memfd_create(const std::string& name, unsigned int flags) { + return syscall(__NR_memfd_create, name.c_str(), flags); +} + +TEST(SpliceTest, NegativeOffset) { + // Create a new pipe. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Fill the pipe. + std::vector<char> buf(kPageSize); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(write(wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(kPageSize)); + + // Open the output file as write only. + int fd; + EXPECT_THAT(fd = memfd_create("negative", 0), SyscallSucceeds()); + const FileDescriptor out_fd(fd); + + loff_t out_offset = 0xffffffffffffffffull; + constexpr int kSize = 2; + EXPECT_THAT(splice(rfd.get(), nullptr, out_fd.get(), &out_offset, kSize, 0), + SyscallFailsWithErrno(EINVAL)); +} + +// Write offset + size overflows int64. +// +// This is a regression test for b/148041624. +TEST(SpliceTest, WriteOverflow) { + // Create a new pipe. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Fill the pipe. + std::vector<char> buf(kPageSize); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(write(wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(kPageSize)); + + // Open the output file. + int fd; + EXPECT_THAT(fd = memfd_create("overflow", 0), SyscallSucceeds()); + const FileDescriptor out_fd(fd); + + // out_offset + kSize overflows INT64_MAX. + loff_t out_offset = 0x7ffffffffffffffeull; + constexpr int kSize = 3; + EXPECT_THAT(splice(rfd.get(), nullptr, out_fd.get(), &out_offset, kSize, 0), + SyscallFailsWithErrno(EINVAL)); +} + TEST(SpliceTest, SamePipe) { // Create a new pipe. int fds[2]; diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index 33a5ac66c..525ccbd88 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -1286,6 +1286,59 @@ TEST_P(SimpleTcpSocketTest, SetTCPUserTimeout) { EXPECT_EQ(get, kTCPUserTimeout); } +TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptNeg) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + // -ve TCP_DEFER_ACCEPT is same as setting it to zero. + constexpr int kNeg = -1; + EXPECT_THAT( + setsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &kNeg, sizeof(kNeg)), + SyscallSucceeds()); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); +} + +TEST_P(SimpleTcpSocketTest, GetTCPDeferAcceptDefault) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); +} + +TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptGreaterThanZero) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + // kTCPDeferAccept is in seconds. + // NOTE: linux translates seconds to # of retries and back from + // #of retries to seconds. Which means only certain values + // translate back exactly. That's why we use 3 here, a value of + // 5 will result in us getting back 7 instead of 5 in the + // getsockopt. + constexpr int kTCPDeferAccept = 3; + ASSERT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, + &kTCPDeferAccept, sizeof(kTCPDeferAccept)), + SyscallSucceeds()); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kTCPDeferAccept); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); diff --git a/test/util/fs_util.h b/test/util/fs_util.h index ee1b341d7..caf19b24d 100644 --- a/test/util/fs_util.h +++ b/test/util/fs_util.h @@ -26,6 +26,17 @@ namespace gvisor { namespace testing { + +// O_LARGEFILE as defined by Linux. glibc tries to be clever by setting it to 0 +// because "it isn't needed", even though Linux can return it via F_GETFL. +#if defined(__x86_64__) +constexpr int kOLargeFile = 00100000; +#elif defined(__aarch64__) +constexpr int kOLargeFile = 00400000; +#else +#error "Unknown architecture" +#endif + // Returns a status or the current working directory. PosixErrorOr<std::string> GetCWD(); |