diff options
Diffstat (limited to 'pkg/tcpip/transport')
24 files changed, 1485 insertions, 10272 deletions
diff --git a/pkg/tcpip/transport/icmp/BUILD b/pkg/tcpip/transport/icmp/BUILD deleted file mode 100644 index d78a162b8..000000000 --- a/pkg/tcpip/transport/icmp/BUILD +++ /dev/null @@ -1,48 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") - -go_template_instance( - name = "icmp_packet_list", - out = "icmp_packet_list.go", - package = "icmp", - prefix = "icmpPacket", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*icmpPacket", - "Linker": "*icmpPacket", - }, -) - -go_library( - name = "icmp", - srcs = [ - "endpoint.go", - "endpoint_state.go", - "icmp_packet_list.go", - "protocol.go", - ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/icmp", - imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], - visibility = ["//visibility:public"], - deps = [ - "//pkg/sleep", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/header", - "//pkg/tcpip/iptables", - "//pkg/tcpip/stack", - "//pkg/tcpip/transport/raw", - "//pkg/tcpip/transport/tcp", - "//pkg/waiter", - ], -) - -filegroup( - name = "autogen", - srcs = [ - "icmp_packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/icmp/icmp_packet_list.go b/pkg/tcpip/transport/icmp/icmp_packet_list.go new file mode 100755 index 000000000..1b35e5b4a --- /dev/null +++ b/pkg/tcpip/transport/icmp/icmp_packet_list.go @@ -0,0 +1,173 @@ +package icmp + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type icmpPacketElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (icmpPacketElementMapper) linkerFor(elem *icmpPacket) *icmpPacket { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type icmpPacketList struct { + head *icmpPacket + tail *icmpPacket +} + +// Reset resets list l to the empty state. +func (l *icmpPacketList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *icmpPacketList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *icmpPacketList) Front() *icmpPacket { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *icmpPacketList) Back() *icmpPacket { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *icmpPacketList) PushFront(e *icmpPacket) { + icmpPacketElementMapper{}.linkerFor(e).SetNext(l.head) + icmpPacketElementMapper{}.linkerFor(e).SetPrev(nil) + + if l.head != nil { + icmpPacketElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *icmpPacketList) PushBack(e *icmpPacket) { + icmpPacketElementMapper{}.linkerFor(e).SetNext(nil) + icmpPacketElementMapper{}.linkerFor(e).SetPrev(l.tail) + + if l.tail != nil { + icmpPacketElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *icmpPacketList) PushBackList(m *icmpPacketList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + icmpPacketElementMapper{}.linkerFor(l.tail).SetNext(m.head) + icmpPacketElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *icmpPacketList) InsertAfter(b, e *icmpPacket) { + a := icmpPacketElementMapper{}.linkerFor(b).Next() + icmpPacketElementMapper{}.linkerFor(e).SetNext(a) + icmpPacketElementMapper{}.linkerFor(e).SetPrev(b) + icmpPacketElementMapper{}.linkerFor(b).SetNext(e) + + if a != nil { + icmpPacketElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *icmpPacketList) InsertBefore(a, e *icmpPacket) { + b := icmpPacketElementMapper{}.linkerFor(a).Prev() + icmpPacketElementMapper{}.linkerFor(e).SetNext(a) + icmpPacketElementMapper{}.linkerFor(e).SetPrev(b) + icmpPacketElementMapper{}.linkerFor(a).SetPrev(e) + + if b != nil { + icmpPacketElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *icmpPacketList) Remove(e *icmpPacket) { + prev := icmpPacketElementMapper{}.linkerFor(e).Prev() + next := icmpPacketElementMapper{}.linkerFor(e).Next() + + if prev != nil { + icmpPacketElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + icmpPacketElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type icmpPacketEntry struct { + next *icmpPacket + prev *icmpPacket +} + +// Next returns the entry that follows e in the list. +func (e *icmpPacketEntry) Next() *icmpPacket { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *icmpPacketEntry) Prev() *icmpPacket { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *icmpPacketEntry) SetNext(elem *icmpPacket) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *icmpPacketEntry) SetPrev(elem *icmpPacket) { + e.prev = elem +} diff --git a/pkg/tcpip/transport/icmp/icmp_state_autogen.go b/pkg/tcpip/transport/icmp/icmp_state_autogen.go new file mode 100755 index 000000000..5caa1cd64 --- /dev/null +++ b/pkg/tcpip/transport/icmp/icmp_state_autogen.go @@ -0,0 +1,98 @@ +// automatically generated by stateify. + +package icmp + +import ( + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) + +func (x *icmpPacket) beforeSave() {} +func (x *icmpPacket) save(m state.Map) { + x.beforeSave() + var data buffer.VectorisedView = x.saveData() + m.SaveValue("data", data) + m.Save("icmpPacketEntry", &x.icmpPacketEntry) + m.Save("senderAddress", &x.senderAddress) + m.Save("timestamp", &x.timestamp) +} + +func (x *icmpPacket) afterLoad() {} +func (x *icmpPacket) load(m state.Map) { + m.Load("icmpPacketEntry", &x.icmpPacketEntry) + m.Load("senderAddress", &x.senderAddress) + m.Load("timestamp", &x.timestamp) + m.LoadValue("data", new(buffer.VectorisedView), func(y interface{}) { x.loadData(y.(buffer.VectorisedView)) }) +} + +func (x *endpoint) save(m state.Map) { + x.beforeSave() + var rcvBufSizeMax int = x.saveRcvBufSizeMax() + m.SaveValue("rcvBufSizeMax", rcvBufSizeMax) + m.Save("netProto", &x.netProto) + m.Save("transProto", &x.transProto) + m.Save("waiterQueue", &x.waiterQueue) + m.Save("rcvReady", &x.rcvReady) + m.Save("rcvList", &x.rcvList) + m.Save("rcvBufSize", &x.rcvBufSize) + m.Save("rcvClosed", &x.rcvClosed) + m.Save("sndBufSize", &x.sndBufSize) + m.Save("shutdownFlags", &x.shutdownFlags) + m.Save("id", &x.id) + m.Save("state", &x.state) + m.Save("bindNICID", &x.bindNICID) + m.Save("bindAddr", &x.bindAddr) + m.Save("regNICID", &x.regNICID) +} + +func (x *endpoint) load(m state.Map) { + m.Load("netProto", &x.netProto) + m.Load("transProto", &x.transProto) + m.Load("waiterQueue", &x.waiterQueue) + m.Load("rcvReady", &x.rcvReady) + m.Load("rcvList", &x.rcvList) + m.Load("rcvBufSize", &x.rcvBufSize) + m.Load("rcvClosed", &x.rcvClosed) + m.Load("sndBufSize", &x.sndBufSize) + m.Load("shutdownFlags", &x.shutdownFlags) + m.Load("id", &x.id) + m.Load("state", &x.state) + m.Load("bindNICID", &x.bindNICID) + m.Load("bindAddr", &x.bindAddr) + m.Load("regNICID", &x.regNICID) + m.LoadValue("rcvBufSizeMax", new(int), func(y interface{}) { x.loadRcvBufSizeMax(y.(int)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *icmpPacketList) beforeSave() {} +func (x *icmpPacketList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *icmpPacketList) afterLoad() {} +func (x *icmpPacketList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *icmpPacketEntry) beforeSave() {} +func (x *icmpPacketEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *icmpPacketEntry) afterLoad() {} +func (x *icmpPacketEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("icmp.icmpPacket", (*icmpPacket)(nil), state.Fns{Save: (*icmpPacket).save, Load: (*icmpPacket).load}) + state.Register("icmp.endpoint", (*endpoint)(nil), state.Fns{Save: (*endpoint).save, Load: (*endpoint).load}) + state.Register("icmp.icmpPacketList", (*icmpPacketList)(nil), state.Fns{Save: (*icmpPacketList).save, Load: (*icmpPacketList).load}) + state.Register("icmp.icmpPacketEntry", (*icmpPacketEntry)(nil), state.Fns{Save: (*icmpPacketEntry).save, Load: (*icmpPacketEntry).load}) +} diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD deleted file mode 100644 index 7241f6c19..000000000 --- a/pkg/tcpip/transport/raw/BUILD +++ /dev/null @@ -1,47 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") - -go_template_instance( - name = "packet_list", - out = "packet_list.go", - package = "raw", - prefix = "packet", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*packet", - "Linker": "*packet", - }, -) - -go_library( - name = "raw", - srcs = [ - "endpoint.go", - "endpoint_state.go", - "packet_list.go", - "protocol.go", - ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/raw", - imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], - visibility = ["//visibility:public"], - deps = [ - "//pkg/log", - "//pkg/sleep", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/header", - "//pkg/tcpip/iptables", - "//pkg/tcpip/stack", - "//pkg/waiter", - ], -) - -filegroup( - name = "autogen", - srcs = [ - "packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/raw/packet_list.go b/pkg/tcpip/transport/raw/packet_list.go new file mode 100755 index 000000000..2e9074934 --- /dev/null +++ b/pkg/tcpip/transport/raw/packet_list.go @@ -0,0 +1,173 @@ +package raw + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type packetElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (packetElementMapper) linkerFor(elem *packet) *packet { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type packetList struct { + head *packet + tail *packet +} + +// Reset resets list l to the empty state. +func (l *packetList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *packetList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *packetList) Front() *packet { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *packetList) Back() *packet { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *packetList) PushFront(e *packet) { + packetElementMapper{}.linkerFor(e).SetNext(l.head) + packetElementMapper{}.linkerFor(e).SetPrev(nil) + + if l.head != nil { + packetElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *packetList) PushBack(e *packet) { + packetElementMapper{}.linkerFor(e).SetNext(nil) + packetElementMapper{}.linkerFor(e).SetPrev(l.tail) + + if l.tail != nil { + packetElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *packetList) PushBackList(m *packetList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + packetElementMapper{}.linkerFor(l.tail).SetNext(m.head) + packetElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *packetList) InsertAfter(b, e *packet) { + a := packetElementMapper{}.linkerFor(b).Next() + packetElementMapper{}.linkerFor(e).SetNext(a) + packetElementMapper{}.linkerFor(e).SetPrev(b) + packetElementMapper{}.linkerFor(b).SetNext(e) + + if a != nil { + packetElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *packetList) InsertBefore(a, e *packet) { + b := packetElementMapper{}.linkerFor(a).Prev() + packetElementMapper{}.linkerFor(e).SetNext(a) + packetElementMapper{}.linkerFor(e).SetPrev(b) + packetElementMapper{}.linkerFor(a).SetPrev(e) + + if b != nil { + packetElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *packetList) Remove(e *packet) { + prev := packetElementMapper{}.linkerFor(e).Prev() + next := packetElementMapper{}.linkerFor(e).Next() + + if prev != nil { + packetElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + packetElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type packetEntry struct { + next *packet + prev *packet +} + +// Next returns the entry that follows e in the list. +func (e *packetEntry) Next() *packet { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *packetEntry) Prev() *packet { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *packetEntry) SetNext(elem *packet) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *packetEntry) SetPrev(elem *packet) { + e.prev = elem +} diff --git a/pkg/tcpip/transport/raw/raw_state_autogen.go b/pkg/tcpip/transport/raw/raw_state_autogen.go new file mode 100755 index 000000000..947fcc1bb --- /dev/null +++ b/pkg/tcpip/transport/raw/raw_state_autogen.go @@ -0,0 +1,98 @@ +// automatically generated by stateify. + +package raw + +import ( + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) + +func (x *packet) beforeSave() {} +func (x *packet) save(m state.Map) { + x.beforeSave() + var data buffer.VectorisedView = x.saveData() + m.SaveValue("data", data) + m.Save("packetEntry", &x.packetEntry) + m.Save("timestampNS", &x.timestampNS) + m.Save("senderAddr", &x.senderAddr) +} + +func (x *packet) afterLoad() {} +func (x *packet) load(m state.Map) { + m.Load("packetEntry", &x.packetEntry) + m.Load("timestampNS", &x.timestampNS) + m.Load("senderAddr", &x.senderAddr) + m.LoadValue("data", new(buffer.VectorisedView), func(y interface{}) { x.loadData(y.(buffer.VectorisedView)) }) +} + +func (x *endpoint) save(m state.Map) { + x.beforeSave() + var rcvBufSizeMax int = x.saveRcvBufSizeMax() + m.SaveValue("rcvBufSizeMax", rcvBufSizeMax) + m.Save("netProto", &x.netProto) + m.Save("transProto", &x.transProto) + m.Save("waiterQueue", &x.waiterQueue) + m.Save("associated", &x.associated) + m.Save("rcvList", &x.rcvList) + m.Save("rcvBufSize", &x.rcvBufSize) + m.Save("rcvClosed", &x.rcvClosed) + m.Save("sndBufSize", &x.sndBufSize) + m.Save("closed", &x.closed) + m.Save("connected", &x.connected) + m.Save("bound", &x.bound) + m.Save("registeredNIC", &x.registeredNIC) + m.Save("boundNIC", &x.boundNIC) + m.Save("boundAddr", &x.boundAddr) +} + +func (x *endpoint) load(m state.Map) { + m.Load("netProto", &x.netProto) + m.Load("transProto", &x.transProto) + m.Load("waiterQueue", &x.waiterQueue) + m.Load("associated", &x.associated) + m.Load("rcvList", &x.rcvList) + m.Load("rcvBufSize", &x.rcvBufSize) + m.Load("rcvClosed", &x.rcvClosed) + m.Load("sndBufSize", &x.sndBufSize) + m.Load("closed", &x.closed) + m.Load("connected", &x.connected) + m.Load("bound", &x.bound) + m.Load("registeredNIC", &x.registeredNIC) + m.Load("boundNIC", &x.boundNIC) + m.Load("boundAddr", &x.boundAddr) + m.LoadValue("rcvBufSizeMax", new(int), func(y interface{}) { x.loadRcvBufSizeMax(y.(int)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *packetList) beforeSave() {} +func (x *packetList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *packetList) afterLoad() {} +func (x *packetList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *packetEntry) beforeSave() {} +func (x *packetEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *packetEntry) afterLoad() {} +func (x *packetEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("raw.packet", (*packet)(nil), state.Fns{Save: (*packet).save, Load: (*packet).load}) + state.Register("raw.endpoint", (*endpoint)(nil), state.Fns{Save: (*endpoint).save, Load: (*endpoint).load}) + state.Register("raw.packetList", (*packetList)(nil), state.Fns{Save: (*packetList).save, Load: (*packetList).load}) + state.Register("raw.packetEntry", (*packetEntry)(nil), state.Fns{Save: (*packetEntry).save, Load: (*packetEntry).load}) +} diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD deleted file mode 100644 index 39a839ab7..000000000 --- a/pkg/tcpip/transport/tcp/BUILD +++ /dev/null @@ -1,100 +0,0 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") - -go_template_instance( - name = "tcp_segment_list", - out = "tcp_segment_list.go", - package = "tcp", - prefix = "segment", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*segment", - "Linker": "*segment", - }, -) - -go_library( - name = "tcp", - srcs = [ - "accept.go", - "connect.go", - "cubic.go", - "cubic_state.go", - "endpoint.go", - "endpoint_state.go", - "forwarder.go", - "protocol.go", - "rcv.go", - "reno.go", - "sack.go", - "sack_scoreboard.go", - "segment.go", - "segment_heap.go", - "segment_queue.go", - "segment_state.go", - "snd.go", - "snd_state.go", - "tcp_segment_list.go", - "timer.go", - ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcp", - imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], - visibility = ["//visibility:public"], - deps = [ - "//pkg/rand", - "//pkg/sleep", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/header", - "//pkg/tcpip/iptables", - "//pkg/tcpip/seqnum", - "//pkg/tcpip/stack", - "//pkg/tcpip/transport/raw", - "//pkg/tmutex", - "//pkg/waiter", - "@com_github_google_btree//:go_default_library", - ], -) - -filegroup( - name = "autogen", - srcs = [ - "tcp_segment_list.go", - ], - visibility = ["//:sandbox"], -) - -go_test( - name = "tcp_test", - size = "small", - srcs = [ - "dual_stack_test.go", - "sack_scoreboard_test.go", - "tcp_noracedetector_test.go", - "tcp_sack_test.go", - "tcp_test.go", - "tcp_timestamp_test.go", - ], - # FIXME(b/68809571) - tags = ["flaky"], - deps = [ - ":tcp", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/checker", - "//pkg/tcpip/header", - "//pkg/tcpip/link/loopback", - "//pkg/tcpip/link/sniffer", - "//pkg/tcpip/network/ipv4", - "//pkg/tcpip/network/ipv6", - "//pkg/tcpip/ports", - "//pkg/tcpip/seqnum", - "//pkg/tcpip/stack", - "//pkg/tcpip/transport/tcp/testing/context", - "//pkg/waiter", - ], -) diff --git a/pkg/tcpip/transport/tcp/dual_stack_test.go b/pkg/tcpip/transport/tcp/dual_stack_test.go deleted file mode 100644 index c54610a87..000000000 --- a/pkg/tcpip/transport/tcp/dual_stack_test.go +++ /dev/null @@ -1,658 +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 tcp_test - -import ( - "testing" - "time" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/checker" - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" - "gvisor.dev/gvisor/pkg/tcpip/seqnum" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestV4MappedConnectOnV6Only(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(true) - - // Start connection attempt, it must fail. - err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestV4MappedAddr, Port: context.TestPort}) - if err != tcpip.ErrNoRoute { - t.Fatalf("Unexpected return value from Connect: %v", err) - } -} - -func testV4Connect(t *testing.T, c *context.Context) { - // Start connection attempt. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventOut) - defer c.WQ.EventUnregister(&we) - - err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestV4MappedAddr, Port: context.TestPort}) - if 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), - ), - ) - - tcp := header.TCP(header.IPv4(b).Payload()) - c.IRS = seqnum.Value(tcp.SequenceNumber()) - - iss := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: tcp.DestinationPort(), - DstPort: tcp.SourcePort(), - Flags: header.TCPFlagSyn | header.TCPFlagAck, - SeqNum: iss, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Receive ACK packet. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), - ), - ) - - // Wait for connection to be established. - select { - case <-ch: - err = c.EP.GetSockOpt(tcpip.ErrorOption{}) - if err != nil { - t.Fatalf("Unexpected error when connecting: %v", err) - } - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for connection") - } -} - -func TestV4MappedConnect(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Test the connection request. - testV4Connect(t, c) -} - -func TestV4ConnectWhenBoundToWildcard(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test the connection request. - testV4Connect(t, c) -} - -func TestV4ConnectWhenBoundToV4MappedWildcard(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind to v4 mapped wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Addr: context.V4MappedWildcardAddr}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test the connection request. - testV4Connect(t, c) -} - -func TestV4ConnectWhenBoundToV4Mapped(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind to v4 mapped address. - if err := c.EP.Bind(tcpip.FullAddress{Addr: context.StackV4MappedAddr}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test the connection request. - testV4Connect(t, c) -} - -func testV6Connect(t *testing.T, c *context.Context) { - // Start connection attempt to IPv6 address. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventOut) - defer c.WQ.EventUnregister(&we) - - err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestV6Addr, Port: context.TestPort}) - if err != tcpip.ErrConnectStarted { - t.Fatalf("Unexpected return value from Connect: %v", err) - } - - // Receive SYN packet. - b := c.GetV6Packet() - checker.IPv6(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagSyn), - ), - ) - - tcp := header.TCP(header.IPv6(b).Payload()) - c.IRS = seqnum.Value(tcp.SequenceNumber()) - - iss := seqnum.Value(789) - c.SendV6Packet(nil, &context.Headers{ - SrcPort: tcp.DestinationPort(), - DstPort: tcp.SourcePort(), - Flags: header.TCPFlagSyn | header.TCPFlagAck, - SeqNum: iss, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Receive ACK packet. - checker.IPv6(t, c.GetV6Packet(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), - ), - ) - - // Wait for connection to be established. - select { - case <-ch: - err = c.EP.GetSockOpt(tcpip.ErrorOption{}) - if err != nil { - t.Fatalf("Unexpected error when connecting: %v", err) - } - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for connection") - } -} - -func TestV6Connect(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Test the connection request. - testV6Connect(t, c) -} - -func TestV6ConnectV6Only(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(true) - - // Test the connection request. - testV6Connect(t, c) -} - -func TestV6ConnectWhenBoundToWildcard(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test the connection request. - testV6Connect(t, c) -} - -func TestV6ConnectWhenBoundToLocalAddress(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind to local address. - if err := c.EP.Bind(tcpip.FullAddress{Addr: context.StackV6Addr}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test the connection request. - testV6Connect(t, c) -} - -func TestV4RefuseOnV6Only(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(true) - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Start listening. - if err := c.EP.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Send a SYN request. - irs := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: irs, - RcvWnd: 30000, - }) - - // Receive the RST reply. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.SrcPort(context.StackPort), - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.AckNum(uint32(irs)+1), - ), - ) -} - -func TestV6RefuseOnBoundToV4Mapped(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind and listen. - if err := c.EP.Bind(tcpip.FullAddress{Addr: context.V4MappedWildcardAddr, Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := c.EP.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Send a SYN request. - irs := seqnum.Value(789) - c.SendV6Packet(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: irs, - RcvWnd: 30000, - }) - - // Receive the RST reply. - checker.IPv6(t, c.GetV6Packet(), - checker.TCP( - checker.SrcPort(context.StackPort), - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.AckNum(uint32(irs)+1), - ), - ) -} - -func testV4Accept(t *testing.T, c *context.Context) { - c.SetGSOEnabled(true) - defer c.SetGSOEnabled(false) - - // Start listening. - if err := c.EP.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Send a SYN request. - irs := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: irs, - RcvWnd: 30000, - }) - - // Receive the SYN-ACK reply. - b := c.GetPacket() - tcp := header.TCP(header.IPv4(b).Payload()) - iss := seqnum.Value(tcp.SequenceNumber()) - checker.IPv4(t, b, - checker.TCP( - checker.SrcPort(context.StackPort), - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1), - ), - ) - - // Send ACK. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagAck, - SeqNum: irs + 1, - AckNum: iss + 1, - RcvWnd: 30000, - }) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - nep, _, err := c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - nep, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - // Make sure we get the same error when calling the original ep and the - // new one. This validates that v4-mapped endpoints are still able to - // query the V6Only flag, whereas pure v4 endpoints are not. - var v tcpip.V6OnlyOption - expected := c.EP.GetSockOpt(&v) - if err := nep.GetSockOpt(&v); err != expected { - t.Fatalf("GetSockOpt returned unexpected value: got %v, want %v", err, expected) - } - - // Check the peer address. - addr, err := nep.GetRemoteAddress() - if err != nil { - t.Fatalf("GetRemoteAddress failed failed: %v", err) - } - - if addr.Addr != context.TestAddr { - t.Fatalf("Unexpected remote address: got %v, want %v", addr.Addr, context.TestAddr) - } - - data := "Don't panic" - nep.Write(tcpip.SlicePayload(buffer.NewViewFromBytes([]byte(data))), tcpip.WriteOptions{}) - b = c.GetPacket() - tcp = header.TCP(header.IPv4(b).Payload()) - if string(tcp.Payload()) != data { - t.Fatalf("Unexpected data: got %v, want %v", string(tcp.Payload()), data) - } -} - -func TestV4AcceptOnV6(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testV4Accept(t, c) -} - -func TestV4AcceptOnBoundToV4MappedWildcard(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind to v4 mapped wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Addr: context.V4MappedWildcardAddr, Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testV4Accept(t, c) -} - -func TestV4AcceptOnBoundToV4Mapped(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind and listen. - if err := c.EP.Bind(tcpip.FullAddress{Addr: context.StackV4MappedAddr, Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testV4Accept(t, c) -} - -func TestV6AcceptOnV6(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateV6Endpoint(false) - - // Bind and listen. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := c.EP.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Send a SYN request. - irs := seqnum.Value(789) - c.SendV6Packet(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: irs, - RcvWnd: 30000, - }) - - // Receive the SYN-ACK reply. - b := c.GetV6Packet() - tcp := header.TCP(header.IPv6(b).Payload()) - iss := seqnum.Value(tcp.SequenceNumber()) - checker.IPv6(t, b, - checker.TCP( - checker.SrcPort(context.StackPort), - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1), - ), - ) - - // Send ACK. - c.SendV6Packet(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagAck, - SeqNum: irs + 1, - AckNum: iss + 1, - RcvWnd: 30000, - }) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - nep, _, err := c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - nep, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - // Make sure we can still query the v6 only status of the new endpoint, - // that is, that it is in fact a v6 socket. - var v tcpip.V6OnlyOption - if err := nep.GetSockOpt(&v); err != nil { - t.Fatalf("GetSockOpt failed failed: %v", err) - } - - // Check the peer address. - addr, err := nep.GetRemoteAddress() - if err != nil { - t.Fatalf("GetRemoteAddress failed failed: %v", err) - } - - if addr.Addr != context.TestV6Addr { - t.Fatalf("Unexpected remote address: got %v, want %v", addr.Addr, context.TestV6Addr) - } -} - -func TestV4AcceptOnV4(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testV4Accept(t, c) -} - -func testV4ListenClose(t *testing.T, c *context.Context) { - // Set the SynRcvd threshold to zero to force a syn cookie based accept - // to happen. - saved := tcp.SynRcvdCountThreshold - defer func() { - tcp.SynRcvdCountThreshold = saved - }() - tcp.SynRcvdCountThreshold = 0 - const n = uint16(32) - - // Start listening. - if err := c.EP.Listen(int(tcp.SynRcvdCountThreshold + 1)); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - irs := seqnum.Value(789) - for i := uint16(0); i < n; i++ { - // Send a SYN request. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort + i, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: irs, - RcvWnd: 30000, - }) - } - - // Each of these ACK's will cause a syn-cookie based connection to be - // accepted and delivered to the listening endpoint. - for i := uint16(0); i < n; i++ { - b := c.GetPacket() - tcp := header.TCP(header.IPv4(b).Payload()) - iss := seqnum.Value(tcp.SequenceNumber()) - // Send ACK. - c.SendPacket(nil, &context.Headers{ - SrcPort: tcp.DestinationPort(), - DstPort: context.StackPort, - Flags: header.TCPFlagAck, - SeqNum: irs + 1, - AckNum: iss + 1, - RcvWnd: 30000, - }) - } - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - nep, _, err := c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - nep, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(10 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - nep.Close() - c.EP.Close() -} - -func TestV4ListenCloseOnV4(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testV4ListenClose(t, c) -} diff --git a/pkg/tcpip/transport/tcp/sack_scoreboard_test.go b/pkg/tcpip/transport/tcp/sack_scoreboard_test.go deleted file mode 100644 index b4e5ba0df..000000000 --- a/pkg/tcpip/transport/tcp/sack_scoreboard_test.go +++ /dev/null @@ -1,249 +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 tcp_test - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/seqnum" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" -) - -const smss = 1500 - -func initScoreboard(blocks []header.SACKBlock, iss seqnum.Value) *tcp.SACKScoreboard { - s := tcp.NewSACKScoreboard(smss, iss) - for _, blk := range blocks { - s.Insert(blk) - } - return s -} - -func TestSACKScoreboardIsSACKED(t *testing.T) { - type blockTest struct { - block header.SACKBlock - sacked bool - } - testCases := []struct { - comment string - scoreboardBlocks []header.SACKBlock - blockTests []blockTest - iss seqnum.Value - }{ - { - "Test holes and unsacked SACK blocks in SACKed ranges and insertion of overlapping SACK blocks", - []header.SACKBlock{{10, 20}, {10, 30}, {30, 40}, {41, 50}, {5, 10}, {1, 50}, {111, 120}, {101, 110}, {52, 120}}, - []blockTest{ - {header.SACKBlock{15, 21}, true}, - {header.SACKBlock{200, 201}, false}, - {header.SACKBlock{50, 51}, false}, - {header.SACKBlock{53, 120}, true}, - }, - 0, - }, - { - "Test disjoint SACKBlocks", - []header.SACKBlock{{2288624809, 2288810057}, {2288811477, 2288838565}}, - []blockTest{ - {header.SACKBlock{2288624809, 2288810057}, true}, - {header.SACKBlock{2288811477, 2288838565}, true}, - {header.SACKBlock{2288810057, 2288811477}, false}, - }, - 2288624809, - }, - { - "Test sequence number wrap around", - []header.SACKBlock{{4294254144, 225652}, {5340409, 5350509}}, - []blockTest{ - {header.SACKBlock{4294254144, 4294254145}, true}, - {header.SACKBlock{4294254143, 4294254144}, false}, - {header.SACKBlock{4294254144, 1}, true}, - {header.SACKBlock{225652, 5350509}, false}, - {header.SACKBlock{5340409, 5350509}, true}, - {header.SACKBlock{5350509, 5350609}, false}, - }, - 4294254144, - }, - { - "Test disjoint SACKBlocks out of order", - []header.SACKBlock{{827450276, 827454536}, {827426028, 827428868}}, - []blockTest{ - {header.SACKBlock{827426028, 827428867}, true}, - {header.SACKBlock{827450168, 827450275}, false}, - }, - 827426000, - }, - } - for _, tc := range testCases { - sb := initScoreboard(tc.scoreboardBlocks, tc.iss) - for _, blkTest := range tc.blockTests { - if want, got := blkTest.sacked, sb.IsSACKED(blkTest.block); got != want { - t.Errorf("%s: s.IsSACKED(%v) = %v, want %v", tc.comment, blkTest.block, got, want) - } - } - } -} - -func TestSACKScoreboardIsRangeLost(t *testing.T) { - s := tcp.NewSACKScoreboard(10, 0) - s.Insert(header.SACKBlock{1, 25}) - s.Insert(header.SACKBlock{25, 50}) - s.Insert(header.SACKBlock{51, 100}) - s.Insert(header.SACKBlock{111, 120}) - s.Insert(header.SACKBlock{101, 110}) - s.Insert(header.SACKBlock{121, 141}) - s.Insert(header.SACKBlock{145, 146}) - s.Insert(header.SACKBlock{147, 148}) - s.Insert(header.SACKBlock{149, 150}) - s.Insert(header.SACKBlock{153, 154}) - s.Insert(header.SACKBlock{155, 156}) - testCases := []struct { - block header.SACKBlock - lost bool - }{ - // Block not covered by SACK block and has more than - // nDupAckThreshold discontiguous SACK blocks after it as well - // as (nDupAckThreshold -1) * 10 (smss) bytes that have been - // SACKED above the sequence number covered by this block. - {block: header.SACKBlock{0, 1}, lost: true}, - - // These blocks have all been SACKed and should not be - // considered lost. - {block: header.SACKBlock{1, 2}, lost: false}, - {block: header.SACKBlock{25, 26}, lost: false}, - {block: header.SACKBlock{1, 45}, lost: false}, - - // Same as the first case above. - {block: header.SACKBlock{50, 51}, lost: true}, - - // This block has been SACKed and should not be considered lost. - {block: header.SACKBlock{119, 120}, lost: false}, - - // This one should return true because there are > - // (nDupAckThreshold - 1) * 10 (smss) bytes that have been - // sacked above this sequence number. - {block: header.SACKBlock{120, 121}, lost: true}, - - // This block has been SACKed and should not be considered lost. - {block: header.SACKBlock{125, 126}, lost: false}, - - // This block has not been SACKed and there are nDupAckThreshold - // number of SACKed blocks after it. - {block: header.SACKBlock{141, 145}, lost: true}, - - // This block has not been SACKed and there are less than - // nDupAckThreshold SACKed sequences after it. - {block: header.SACKBlock{151, 152}, lost: false}, - } - for _, tc := range testCases { - if want, got := tc.lost, s.IsRangeLost(tc.block); got != want { - t.Errorf("s.IsRangeLost(%v) = %v, want %v", tc.block, got, want) - } - } -} - -func TestSACKScoreboardIsLost(t *testing.T) { - s := tcp.NewSACKScoreboard(10, 0) - s.Insert(header.SACKBlock{1, 25}) - s.Insert(header.SACKBlock{25, 50}) - s.Insert(header.SACKBlock{51, 100}) - s.Insert(header.SACKBlock{111, 120}) - s.Insert(header.SACKBlock{101, 110}) - s.Insert(header.SACKBlock{121, 141}) - s.Insert(header.SACKBlock{121, 141}) - s.Insert(header.SACKBlock{145, 146}) - s.Insert(header.SACKBlock{147, 148}) - s.Insert(header.SACKBlock{149, 150}) - s.Insert(header.SACKBlock{153, 154}) - s.Insert(header.SACKBlock{155, 156}) - testCases := []struct { - seq seqnum.Value - lost bool - }{ - // Sequence number not covered by SACK block and has more than - // nDupAckThreshold discontiguous SACK blocks after it as well - // as (nDupAckThreshold -1) * 10 (smss) bytes that have been - // SACKED above the sequence number. - {seq: 0, lost: true}, - - // These sequence numbers have all been SACKed and should not be - // considered lost. - {seq: 1, lost: false}, - {seq: 25, lost: false}, - {seq: 45, lost: false}, - - // Same as first case above. - {seq: 50, lost: true}, - - // This block has been SACKed and should not be considered lost. - {seq: 119, lost: false}, - - // This one should return true because there are > - // (nDupAckThreshold - 1) * 10 (smss) bytes that have been - // sacked above this sequence number. - {seq: 120, lost: true}, - - // This sequence number has been SACKed and should not be - // considered lost. - {seq: 125, lost: false}, - - // This sequence number has not been SACKed and there are - // nDupAckThreshold number of SACKed blocks after it. - {seq: 141, lost: true}, - - // This sequence number has not been SACKed and there are less - // than nDupAckThreshold SACKed sequences after it. - {seq: 151, lost: false}, - } - for _, tc := range testCases { - if want, got := tc.lost, s.IsLost(tc.seq); got != want { - t.Errorf("s.IsLost(%v) = %v, want %v", tc.seq, got, want) - } - } -} - -func TestSACKScoreboardDelete(t *testing.T) { - blocks := []header.SACKBlock{{4294254144, 225652}, {5340409, 5350509}} - s := initScoreboard(blocks, 4294254143) - s.Delete(5340408) - if s.Empty() { - t.Fatalf("s.Empty() = true, want false") - } - if got, want := s.Sacked(), blocks[1].Start.Size(blocks[1].End); got != want { - t.Fatalf("incorrect sacked bytes in scoreboard got: %v, want: %v", got, want) - } - s.Delete(5340410) - if s.Empty() { - t.Fatal("s.Empty() = true, want false") - } - newSB := header.SACKBlock{5340410, 5350509} - if !s.IsSACKED(newSB) { - t.Fatalf("s.IsSACKED(%v) = false, want true, scoreboard: %v", newSB, s) - } - s.Delete(5350509) - lastOctet := header.SACKBlock{5350508, 5350509} - if s.IsSACKED(lastOctet) { - t.Fatalf("s.IsSACKED(%v) = false, want true", lastOctet) - } - - s.Delete(5350510) - if !s.Empty() { - t.Fatal("s.Empty() = false, want true") - } - if got, want := s.Sacked(), seqnum.Size(0); got != want { - t.Fatalf("incorrect sacked bytes in scoreboard got: %v, want: %v", got, want) - } -} diff --git a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go deleted file mode 100644 index 272bbcdbd..000000000 --- a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go +++ /dev/null @@ -1,519 +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. -// -// These tests are flaky when run under the go race detector due to some -// iterations taking long enough that the retransmit timer can kick in causing -// the congestion window measurements to fail due to extra packets etc. -// -// +build !race - -package tcp_test - -import ( - "fmt" - "math" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context" -) - -func TestFastRecovery(t *testing.T) { - maxPayload := 32 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - const iterations = 7 - data := buffer.NewView(2 * maxPayload * (tcp.InitialCwnd << (iterations + 1))) - for i := range data { - data[i] = byte(i) - } - - // Write all the data in one shot. Packets will only be written at the - // MTU size though. - if _, _, err := c.EP.Write(tcpip.SlicePayload(data), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Do slow start for a few iterations. - expected := tcp.InitialCwnd - bytesRead := 0 - for i := 0; i < iterations; i++ { - expected = tcp.InitialCwnd << uint(i) - if i > 0 { - // Acknowledge all the data received so far if not on - // first iteration. - c.SendAck(790, bytesRead) - } - - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd.", 50*time.Millisecond) - } - - // Send 3 duplicate acks. This should force an immediate retransmit of - // the pending packet and put the sender into fast recovery. - rtxOffset := bytesRead - maxPayload*expected - for i := 0; i < 3; i++ { - c.SendAck(790, rtxOffset) - } - - // Receive the retransmitted packet. - c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload) - - if got, want := c.Stack().Stats().TCP.FastRetransmit.Value(), uint64(1); got != want { - t.Errorf("got stats.TCP.FastRetransmit.Value = %v, want = %v", got, want) - } - - if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(1); got != want { - t.Errorf("got stats.TCP.Retransmit.Value = %v, want = %v", got, want) - } - - if got, want := c.Stack().Stats().TCP.FastRecovery.Value(), uint64(1); got != want { - t.Errorf("got stats.TCP.FastRecovery.Value = %v, want = %v", got, want) - } - - // Now send 7 mode duplicate acks. Each of these should cause a window - // inflation by 1 and cause the sender to send an extra packet. - for i := 0; i < 7; i++ { - c.SendAck(790, rtxOffset) - } - - recover := bytesRead - - // Ensure no new packets arrive. - c.CheckNoPacketTimeout("More packets received than expected during recovery after dupacks for this cwnd.", - 50*time.Millisecond) - - // Acknowledge half of the pending data. - rtxOffset = bytesRead - expected*maxPayload/2 - c.SendAck(790, rtxOffset) - - // Receive the retransmit due to partial ack. - c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload) - - if got, want := c.Stack().Stats().TCP.FastRetransmit.Value(), uint64(2); got != want { - t.Errorf("got stats.TCP.FastRetransmit.Value = %v, want = %v", got, want) - } - - if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(2); got != want { - t.Errorf("got stats.TCP.Retransmit.Value = %v, want = %v", got, want) - } - - // Receive the 10 extra packets that should have been released due to - // the congestion window inflation in recovery. - for i := 0; i < 10; i++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // A partial ACK during recovery should reduce congestion window by the - // number acked. Since we had "expected" packets outstanding before sending - // partial ack and we acked expected/2 , the cwnd and outstanding should - // be expected/2 + 10 (7 dupAcks + 3 for the original 3 dupacks that triggered - // fast recovery). Which means the sender should not send any more packets - // till we ack this one. - c.CheckNoPacketTimeout("More packets received than expected during recovery after partial ack for this cwnd.", - 50*time.Millisecond) - - // Acknowledge all pending data to recover point. - c.SendAck(790, recover) - - // At this point, the cwnd should reset to expected/2 and there are 10 - // packets outstanding. - // - // NOTE: Technically netstack is incorrect in that we adjust the cwnd on - // the same segment that takes us out of recovery. But because of that - // the actual cwnd at exit of recovery will be expected/2 + 1 as we - // acked a cwnd worth of packets which will increase the cwnd further by - // 1 in congestion avoidance. - // - // Now in the first iteration since there are 10 packets outstanding. - // We would expect to get expected/2 +1 - 10 packets. But subsequent - // iterations will send us expected/2 + 1 + 1 (per iteration). - expected = expected/2 + 1 - 10 - for i := 0; i < iterations; i++ { - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout(fmt.Sprintf("More packets received(after deflation) than expected %d for this cwnd.", expected), 50*time.Millisecond) - - // Acknowledge all the data received so far. - c.SendAck(790, bytesRead) - - // In cogestion avoidance, the packets trains increase by 1 in - // each iteration. - if i == 0 { - // After the first iteration we expect to get the full - // congestion window worth of packets in every - // iteration. - expected += 10 - } - expected++ - } -} - -func TestExponentialIncreaseDuringSlowStart(t *testing.T) { - maxPayload := 32 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - const iterations = 7 - data := buffer.NewView(maxPayload * (tcp.InitialCwnd << (iterations + 1))) - for i := range data { - data[i] = byte(i) - } - - // Write all the data in one shot. Packets will only be written at the - // MTU size though. - if _, _, err := c.EP.Write(tcpip.SlicePayload(data), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - expected := tcp.InitialCwnd - bytesRead := 0 - for i := 0; i < iterations; i++ { - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd.", 50*time.Millisecond) - - // Acknowledge all the data received so far. - c.SendAck(790, bytesRead) - - // Double the number of expected packets for the next iteration. - expected *= 2 - } -} - -func TestCongestionAvoidance(t *testing.T) { - maxPayload := 32 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - const iterations = 7 - data := buffer.NewView(2 * maxPayload * (tcp.InitialCwnd << (iterations + 1))) - for i := range data { - data[i] = byte(i) - } - - // Write all the data in one shot. Packets will only be written at the - // MTU size though. - if _, _, err := c.EP.Write(tcpip.SlicePayload(data), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Do slow start for a few iterations. - expected := tcp.InitialCwnd - bytesRead := 0 - for i := 0; i < iterations; i++ { - expected = tcp.InitialCwnd << uint(i) - if i > 0 { - // Acknowledge all the data received so far if not on - // first iteration. - c.SendAck(790, bytesRead) - } - - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd (slow start phase).", 50*time.Millisecond) - } - - // Don't acknowledge the first packet of the last packet train. Let's - // wait for them to time out, which will trigger a restart of slow - // start, and initialization of ssthresh to cwnd/2. - rtxOffset := bytesRead - maxPayload*expected - c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload) - - // Acknowledge all the data received so far. - c.SendAck(790, bytesRead) - - // This part is tricky: when the timeout happened, we had "expected" - // packets pending, cwnd reset to 1, and ssthresh set to expected/2. - // By acknowledging "expected" packets, the slow-start part will - // increase cwnd to expected/2 (which "consumes" expected/2-1 of the - // acknowledgements), then the congestion avoidance part will consume - // an extra expected/2 acks to take cwnd to expected/2 + 1. One ack - // remains in the "ack count" (which will cause cwnd to be incremented - // once it reaches cwnd acks). - // - // So we're straight into congestion avoidance with cwnd set to - // expected/2 + 1. - // - // Check that packets trains of cwnd packets are sent, and that cwnd is - // incremented by 1 after we acknowledge each packet. - expected = expected/2 + 1 - for i := 0; i < iterations; i++ { - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd (congestion avoidance phase).", 50*time.Millisecond) - - // Acknowledge all the data received so far. - c.SendAck(790, bytesRead) - - // In cogestion avoidance, the packets trains increase by 1 in - // each iteration. - expected++ - } -} - -// cubicCwnd returns an estimate of a cubic window given the -// originalCwnd, wMax, last congestion event time and sRTT. -func cubicCwnd(origCwnd int, wMax int, congEventTime time.Time, sRTT time.Duration) int { - cwnd := float64(origCwnd) - // We wait 50ms between each iteration so sRTT as computed by cubic - // should be close to 50ms. - elapsed := (time.Since(congEventTime) + sRTT).Seconds() - k := math.Cbrt(float64(wMax) * 0.3 / 0.7) - wtRTT := 0.4*math.Pow(elapsed-k, 3) + float64(wMax) - cwnd += (wtRTT - cwnd) / cwnd - return int(cwnd) -} - -func TestCubicCongestionAvoidance(t *testing.T) { - maxPayload := 32 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - enableCUBIC(t, c) - - c.CreateConnected(789, 30000, nil) - - const iterations = 7 - data := buffer.NewView(2 * maxPayload * (tcp.InitialCwnd << (iterations + 1))) - - for i := range data { - data[i] = byte(i) - } - - // Write all the data in one shot. Packets will only be written at the - // MTU size though. - if _, _, err := c.EP.Write(tcpip.SlicePayload(data), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Do slow start for a few iterations. - expected := tcp.InitialCwnd - bytesRead := 0 - for i := 0; i < iterations; i++ { - expected = tcp.InitialCwnd << uint(i) - if i > 0 { - // Acknowledge all the data received so far if not on - // first iteration. - c.SendAck(790, bytesRead) - } - - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd (during slow-start phase).", 50*time.Millisecond) - } - - // Don't acknowledge the first packet of the last packet train. Let's - // wait for them to time out, which will trigger a restart of slow - // start, and initialization of ssthresh to cwnd * 0.7. - rtxOffset := bytesRead - maxPayload*expected - c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload) - - // Acknowledge all pending data. - c.SendAck(790, bytesRead) - - // Store away the time we sent the ACK and assuming a 200ms RTO - // we estimate that the sender will have an RTO 200ms from now - // and go back into slow start. - packetDropTime := time.Now().Add(200 * time.Millisecond) - - // This part is tricky: when the timeout happened, we had "expected" - // packets pending, cwnd reset to 1, and ssthresh set to expected * 0.7. - // By acknowledging "expected" packets, the slow-start part will - // increase cwnd to expected/2 essentially putting the connection - // straight into congestion avoidance. - wMax := expected - // Lower expected as per cubic spec after a congestion event. - expected = int(float64(expected) * 0.7) - cwnd := expected - for i := 0; i < iterations; i++ { - // Cubic grows window independent of ACKs. Cubic Window growth - // is a function of time elapsed since last congestion event. - // As a result the congestion window does not grow - // deterministically in response to ACKs. - // - // We need to roughly estimate what the cwnd of the sender is - // based on when we sent the dupacks. - cwnd := cubicCwnd(cwnd, wMax, packetDropTime, 50*time.Millisecond) - - packetsExpected := cwnd - for j := 0; j < packetsExpected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - t.Logf("expected packets received, next trying to receive any extra packets that may come") - - // If our estimate was correct there should be no more pending packets. - // We attempt to read a packet a few times with a short sleep in between - // to ensure that we don't see the sender send any unexpected packets. - unexpectedPackets := 0 - for { - gotPacket := c.ReceiveNonBlockingAndCheckPacket(data, bytesRead, maxPayload) - if !gotPacket { - break - } - bytesRead += maxPayload - unexpectedPackets++ - time.Sleep(1 * time.Millisecond) - } - if unexpectedPackets != 0 { - t.Fatalf("received %d unexpected packets for iteration %d", unexpectedPackets, i) - } - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd(congestion avoidance)", 5*time.Millisecond) - - // Acknowledge all the data received so far. - c.SendAck(790, bytesRead) - } -} - -func TestRetransmit(t *testing.T) { - maxPayload := 32 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - const iterations = 7 - data := buffer.NewView(maxPayload * (tcp.InitialCwnd << (iterations + 1))) - for i := range data { - data[i] = byte(i) - } - - // Write all the data in two shots. Packets will only be written at the - // MTU size though. - half := data[:len(data)/2] - if _, _, err := c.EP.Write(tcpip.SlicePayload(half), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - half = data[len(data)/2:] - if _, _, err := c.EP.Write(tcpip.SlicePayload(half), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Do slow start for a few iterations. - expected := tcp.InitialCwnd - bytesRead := 0 - for i := 0; i < iterations; i++ { - expected = tcp.InitialCwnd << uint(i) - if i > 0 { - // Acknowledge all the data received so far if not on - // first iteration. - c.SendAck(790, bytesRead) - } - - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacket(data, bytesRead, maxPayload) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd.", 50*time.Millisecond) - } - - // Wait for a timeout and retransmit. - rtxOffset := bytesRead - maxPayload*expected - c.ReceiveAndCheckPacket(data, rtxOffset, maxPayload) - - if got, want := c.Stack().Stats().TCP.Timeouts.Value(), uint64(1); got != want { - t.Errorf("got stats.TCP.Timeouts.Value = %v, want = %v", got, want) - } - - if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(1); got != want { - t.Errorf("got stats.TCP.Retransmits.Value = %v, want = %v", got, want) - } - - if got, want := c.Stack().Stats().TCP.SlowStartRetransmits.Value(), uint64(1); got != want { - t.Errorf("got stats.TCP.SlowStartRetransmits.Value = %v, want = %v", got, want) - } - - // Acknowledge half of the pending data. - rtxOffset = bytesRead - expected*maxPayload/2 - c.SendAck(790, rtxOffset) - - // Receive the remaining data, making sure that acknowledged data is not - // retransmitted. - for offset := rtxOffset; offset < len(data); offset += maxPayload { - c.ReceiveAndCheckPacket(data, offset, maxPayload) - c.SendAck(790, offset+maxPayload) - } - - c.CheckNoPacketTimeout("More packets received than expected for this cwnd.", 50*time.Millisecond) -} diff --git a/pkg/tcpip/transport/tcp/tcp_sack_test.go b/pkg/tcpip/transport/tcp/tcp_sack_test.go deleted file mode 100644 index 4e7f1a740..000000000 --- a/pkg/tcpip/transport/tcp/tcp_sack_test.go +++ /dev/null @@ -1,564 +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 tcp_test - -import ( - "fmt" - "log" - "reflect" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/seqnum" - "gvisor.dev/gvisor/pkg/tcpip/stack" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context" -) - -// createConnectedWithSACKPermittedOption creates and connects c.ep with the -// SACKPermitted option enabled if the stack in the context has the SACK support -// enabled. -func createConnectedWithSACKPermittedOption(c *context.Context) *context.RawEndpoint { - return c.CreateConnectedWithOptions(header.TCPSynOptions{SACKPermitted: c.SACKEnabled()}) -} - -// createConnectedWithSACKAndTS creates and connects c.ep with the SACK & TS -// option enabled if the stack in the context has SACK and TS enabled. -func createConnectedWithSACKAndTS(c *context.Context) *context.RawEndpoint { - return c.CreateConnectedWithOptions(header.TCPSynOptions{SACKPermitted: c.SACKEnabled(), TS: true}) -} - -func setStackSACKPermitted(t *testing.T, c *context.Context, enable bool) { - t.Helper() - if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enable)); err != nil { - t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, SACKEnabled(%v) = %v", enable, err) - } -} - -// TestSackPermittedConnect establishes a connection with the SACK option -// enabled. -func TestSackPermittedConnect(t *testing.T) { - for _, sackEnabled := range []bool{false, true} { - t.Run(fmt.Sprintf("stack.sackEnabled: %v", sackEnabled), func(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - setStackSACKPermitted(t, c, sackEnabled) - rep := createConnectedWithSACKPermittedOption(c) - data := []byte{1, 2, 3} - - rep.SendPacket(data, nil) - savedSeqNum := rep.NextSeqNum - rep.VerifyACKNoSACK() - - // Make an out of order packet and send it. - rep.NextSeqNum += 3 - sackBlocks := []header.SACKBlock{ - {rep.NextSeqNum, rep.NextSeqNum.Add(seqnum.Size(len(data)))}, - } - rep.SendPacket(data, nil) - - // Restore the saved sequence number so that the - // VerifyXXX calls use the right sequence number for - // checking ACK numbers. - rep.NextSeqNum = savedSeqNum - if sackEnabled { - rep.VerifyACKHasSACK(sackBlocks) - } else { - rep.VerifyACKNoSACK() - } - - // Send the missing segment. - rep.SendPacket(data, nil) - // The ACK should contain the cumulative ACK for all 9 - // bytes sent and no SACK blocks. - rep.NextSeqNum += 3 - // Check that no SACK block is returned in the ACK. - rep.VerifyACKNoSACK() - }) - } -} - -// TestSackDisabledConnect establishes a connection with the SACK option -// disabled and verifies that no SACKs are sent for out of order segments. -func TestSackDisabledConnect(t *testing.T) { - for _, sackEnabled := range []bool{false, true} { - t.Run(fmt.Sprintf("sackEnabled: %v", sackEnabled), func(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - setStackSACKPermitted(t, c, sackEnabled) - - rep := c.CreateConnectedWithOptions(header.TCPSynOptions{}) - - data := []byte{1, 2, 3} - - rep.SendPacket(data, nil) - savedSeqNum := rep.NextSeqNum - rep.VerifyACKNoSACK() - - // Make an out of order packet and send it. - rep.NextSeqNum += 3 - rep.SendPacket(data, nil) - - // The ACK should contain the older sequence number and - // no SACK blocks. - rep.NextSeqNum = savedSeqNum - rep.VerifyACKNoSACK() - - // Send the missing segment. - rep.SendPacket(data, nil) - // The ACK should contain the cumulative ACK for all 9 - // bytes sent and no SACK blocks. - rep.NextSeqNum += 3 - // Check that no SACK block is returned in the ACK. - rep.VerifyACKNoSACK() - }) - } -} - -// TestSackPermittedAccept accepts and establishes a connection with the -// SACKPermitted option enabled if the connection request specifies the -// SACKPermitted option. In case of SYN cookies SACK should be disabled as we -// don't encode the SACK information in the cookie. -func TestSackPermittedAccept(t *testing.T) { - type testCase struct { - cookieEnabled bool - sackPermitted bool - wndScale int - wndSize uint16 - } - - testCases := []testCase{ - // When cookie is used window scaling is disabled. - {true, false, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, true, 5, 0x8000}, // 0x8000 * 2^5 = 1<<20 = 1MB window (the default). - } - savedSynCountThreshold := tcp.SynRcvdCountThreshold - defer func() { - tcp.SynRcvdCountThreshold = savedSynCountThreshold - }() - for _, tc := range testCases { - t.Run(fmt.Sprintf("test: %#v", tc), func(t *testing.T) { - if tc.cookieEnabled { - tcp.SynRcvdCountThreshold = 0 - } else { - tcp.SynRcvdCountThreshold = savedSynCountThreshold - } - for _, sackEnabled := range []bool{false, true} { - t.Run(fmt.Sprintf("test stack.sackEnabled: %v", sackEnabled), func(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - setStackSACKPermitted(t, c, sackEnabled) - - rep := c.AcceptWithOptions(tc.wndScale, header.TCPSynOptions{MSS: defaultIPv4MSS, SACKPermitted: tc.sackPermitted}) - // Now verify no SACK blocks are - // received when sack is disabled. - data := []byte{1, 2, 3} - rep.SendPacket(data, nil) - rep.VerifyACKNoSACK() - - savedSeqNum := rep.NextSeqNum - - // Make an out of order packet and send - // it. - rep.NextSeqNum += 3 - sackBlocks := []header.SACKBlock{ - {rep.NextSeqNum, rep.NextSeqNum.Add(seqnum.Size(len(data)))}, - } - rep.SendPacket(data, nil) - - // The ACK should contain the older - // sequence number. - rep.NextSeqNum = savedSeqNum - if sackEnabled && tc.sackPermitted { - rep.VerifyACKHasSACK(sackBlocks) - } else { - rep.VerifyACKNoSACK() - } - - // Send the missing segment. - rep.SendPacket(data, nil) - // The ACK should contain the cumulative - // ACK for all 9 bytes sent and no SACK - // blocks. - rep.NextSeqNum += 3 - // Check that no SACK block is returned - // in the ACK. - rep.VerifyACKNoSACK() - }) - } - }) - } -} - -// TestSackDisabledAccept accepts and establishes a connection with -// the SACKPermitted option disabled and verifies that no SACKs are -// sent for out of order packets. -func TestSackDisabledAccept(t *testing.T) { - type testCase struct { - cookieEnabled bool - wndScale int - wndSize uint16 - } - - testCases := []testCase{ - // When cookie is used window scaling is disabled. - {true, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, 5, 0x8000}, // 0x8000 * 2^5 = 1<<20 = 1MB window (the default). - } - savedSynCountThreshold := tcp.SynRcvdCountThreshold - defer func() { - tcp.SynRcvdCountThreshold = savedSynCountThreshold - }() - for _, tc := range testCases { - t.Run(fmt.Sprintf("test: %#v", tc), func(t *testing.T) { - if tc.cookieEnabled { - tcp.SynRcvdCountThreshold = 0 - } else { - tcp.SynRcvdCountThreshold = savedSynCountThreshold - } - for _, sackEnabled := range []bool{false, true} { - t.Run(fmt.Sprintf("test: sackEnabled: %v", sackEnabled), func(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - setStackSACKPermitted(t, c, sackEnabled) - - rep := c.AcceptWithOptions(tc.wndScale, header.TCPSynOptions{MSS: defaultIPv4MSS}) - - // Now verify no SACK blocks are - // received when sack is disabled. - data := []byte{1, 2, 3} - rep.SendPacket(data, nil) - rep.VerifyACKNoSACK() - savedSeqNum := rep.NextSeqNum - - // Make an out of order packet and send - // it. - rep.NextSeqNum += 3 - rep.SendPacket(data, nil) - - // The ACK should contain the older - // sequence number and no SACK blocks. - rep.NextSeqNum = savedSeqNum - rep.VerifyACKNoSACK() - - // Send the missing segment. - rep.SendPacket(data, nil) - // The ACK should contain the cumulative - // ACK for all 9 bytes sent and no SACK - // blocks. - rep.NextSeqNum += 3 - // Check that no SACK block is returned - // in the ACK. - rep.VerifyACKNoSACK() - }) - } - }) - } -} - -func TestUpdateSACKBlocks(t *testing.T) { - testCases := []struct { - segStart seqnum.Value - segEnd seqnum.Value - rcvNxt seqnum.Value - sackBlocks []header.SACKBlock - updated []header.SACKBlock - }{ - // Trivial cases where current SACK block list is empty and we - // have an out of order delivery. - {10, 11, 2, []header.SACKBlock{}, []header.SACKBlock{{10, 11}}}, - {10, 12, 2, []header.SACKBlock{}, []header.SACKBlock{{10, 12}}}, - {10, 20, 2, []header.SACKBlock{}, []header.SACKBlock{{10, 20}}}, - - // Cases where current SACK block list is not empty and we have - // an out of order delivery. Tests that the updated SACK block - // list has the first block as the one that contains the new - // SACK block representing the segment that was just delivered. - {10, 11, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{10, 11}, {12, 20}}}, - {24, 30, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{24, 30}, {12, 20}}}, - {24, 30, 9, []header.SACKBlock{{12, 20}, {32, 40}}, []header.SACKBlock{{24, 30}, {12, 20}, {32, 40}}}, - - // Ensure that we only retain header.MaxSACKBlocks and drop the - // oldest one if adding a new block exceeds - // header.MaxSACKBlocks. - {24, 30, 9, - []header.SACKBlock{{12, 20}, {32, 40}, {42, 50}, {52, 60}, {62, 70}, {72, 80}}, - []header.SACKBlock{{24, 30}, {12, 20}, {32, 40}, {42, 50}, {52, 60}, {62, 70}}}, - - // Cases where segment extends an existing SACK block. - {10, 12, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{10, 20}}}, - {10, 22, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{10, 22}}}, - {10, 22, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{10, 22}}}, - {15, 22, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{12, 22}}}, - {15, 25, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{12, 25}}}, - {11, 25, 9, []header.SACKBlock{{12, 20}}, []header.SACKBlock{{11, 25}}}, - {10, 12, 9, []header.SACKBlock{{12, 20}, {32, 40}}, []header.SACKBlock{{10, 20}, {32, 40}}}, - {10, 22, 9, []header.SACKBlock{{12, 20}, {32, 40}}, []header.SACKBlock{{10, 22}, {32, 40}}}, - {10, 22, 9, []header.SACKBlock{{12, 20}, {32, 40}}, []header.SACKBlock{{10, 22}, {32, 40}}}, - {15, 22, 9, []header.SACKBlock{{12, 20}, {32, 40}}, []header.SACKBlock{{12, 22}, {32, 40}}}, - {15, 25, 9, []header.SACKBlock{{12, 20}, {32, 40}}, []header.SACKBlock{{12, 25}, {32, 40}}}, - {11, 25, 9, []header.SACKBlock{{12, 20}, {32, 40}}, []header.SACKBlock{{11, 25}, {32, 40}}}, - - // Cases where segment contains rcvNxt. - {10, 20, 15, []header.SACKBlock{{20, 30}, {40, 50}}, []header.SACKBlock{{40, 50}}}, - } - - for _, tc := range testCases { - var sack tcp.SACKInfo - copy(sack.Blocks[:], tc.sackBlocks) - sack.NumBlocks = len(tc.sackBlocks) - tcp.UpdateSACKBlocks(&sack, tc.segStart, tc.segEnd, tc.rcvNxt) - if got, want := sack.Blocks[:sack.NumBlocks], tc.updated; !reflect.DeepEqual(got, want) { - t.Errorf("UpdateSACKBlocks(%v, %v, %v, %v), got: %v, want: %v", tc.sackBlocks, tc.segStart, tc.segEnd, tc.rcvNxt, got, want) - } - - } -} - -func TestTrimSackBlockList(t *testing.T) { - testCases := []struct { - rcvNxt seqnum.Value - sackBlocks []header.SACKBlock - trimmed []header.SACKBlock - }{ - // Simple cases where we trim whole entries. - {2, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}}, - {21, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{{22, 30}, {32, 40}}}, - {31, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{{32, 40}}}, - {40, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{}}, - // Cases where we need to update a block. - {12, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{{12, 20}, {22, 30}, {32, 40}}}, - {23, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{{23, 30}, {32, 40}}}, - {33, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{{33, 40}}}, - {41, []header.SACKBlock{{10, 20}, {22, 30}, {32, 40}}, []header.SACKBlock{}}, - } - for _, tc := range testCases { - var sack tcp.SACKInfo - copy(sack.Blocks[:], tc.sackBlocks) - sack.NumBlocks = len(tc.sackBlocks) - tcp.TrimSACKBlockList(&sack, tc.rcvNxt) - if got, want := sack.Blocks[:sack.NumBlocks], tc.trimmed; !reflect.DeepEqual(got, want) { - t.Errorf("TrimSackBlockList(%v, %v), got: %v, want: %v", tc.sackBlocks, tc.rcvNxt, got, want) - } - } -} - -func TestSACKRecovery(t *testing.T) { - const maxPayload = 10 - // See: tcp.makeOptions for why tsOptionSize is set to 12 here. - const tsOptionSize = 12 - // Enabling SACK means the payload size is reduced to account - // for the extra space required for the TCP options. - // - // We increase the MTU by 40 bytes to account for SACK and Timestamp - // options. - const maxTCPOptionSize = 40 - - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxTCPOptionSize+maxPayload)) - defer c.Cleanup() - - c.Stack().AddTCPProbe(func(s stack.TCPEndpointState) { - // We use log.Printf instead of t.Logf here because this probe - // can fire even when the test function has finished. This is - // because closing the endpoint in cleanup() does not mean the - // actual worker loop terminates immediately as it still has to - // do a full TCP shutdown. But this test can finish running - // before the shutdown is done. Using t.Logf in such a case - // causes the test to panic due to logging after test finished. - log.Printf("state: %+v\n", s) - }) - setStackSACKPermitted(t, c, true) - createConnectedWithSACKAndTS(c) - - const iterations = 7 - data := buffer.NewView(2 * maxPayload * (tcp.InitialCwnd << (iterations + 1))) - for i := range data { - data[i] = byte(i) - } - - // Write all the data in one shot. Packets will only be written at the - // MTU size though. - if _, _, err := c.EP.Write(tcpip.SlicePayload(data), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Do slow start for a few iterations. - expected := tcp.InitialCwnd - bytesRead := 0 - for i := 0; i < iterations; i++ { - expected = tcp.InitialCwnd << uint(i) - if i > 0 { - // Acknowledge all the data received so far if not on - // first iteration. - c.SendAck(790, bytesRead) - } - - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacketWithOptions(data, bytesRead, maxPayload, tsOptionSize) - bytesRead += maxPayload - } - - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout("More packets received than expected for this cwnd.", 50*time.Millisecond) - } - - // Send 3 duplicate acks. This should force an immediate retransmit of - // the pending packet and put the sender into fast recovery. - rtxOffset := bytesRead - maxPayload*expected - start := c.IRS.Add(seqnum.Size(rtxOffset) + 30 + 1) - end := start.Add(10) - for i := 0; i < 3; i++ { - c.SendAckWithSACK(790, rtxOffset, []header.SACKBlock{{start, end}}) - end = end.Add(10) - } - - // Receive the retransmitted packet. - c.ReceiveAndCheckPacketWithOptions(data, rtxOffset, maxPayload, tsOptionSize) - - tcpStats := c.Stack().Stats().TCP - stats := []struct { - stat *tcpip.StatCounter - name string - want uint64 - }{ - {tcpStats.FastRetransmit, "stats.TCP.FastRetransmit", 1}, - {tcpStats.Retransmits, "stats.TCP.Retransmits", 1}, - {tcpStats.SACKRecovery, "stats.TCP.SACKRecovery", 1}, - {tcpStats.FastRecovery, "stats.TCP.FastRecovery", 0}, - } - for _, s := range stats { - if got, want := s.stat.Value(), s.want; got != want { - t.Errorf("got %s.Value() = %v, want = %v", s.name, got, want) - } - } - - // Now send 7 mode duplicate ACKs. In SACK TCP dupAcks do not cause - // window inflation and sending of packets is completely handled by the - // SACK Recovery algorithm. We should see no packets being released, as - // the cwnd at this point after entering recovery should be half of the - // outstanding number of packets in flight. - for i := 0; i < 7; i++ { - c.SendAckWithSACK(790, rtxOffset, []header.SACKBlock{{start, end}}) - end = end.Add(10) - } - - recover := bytesRead - - // Ensure no new packets arrive. - c.CheckNoPacketTimeout("More packets received than expected during recovery after dupacks for this cwnd.", - 50*time.Millisecond) - - // Acknowledge half of the pending data. This along with the 10 sacked - // segments above should reduce the outstanding below the current - // congestion window allowing the sender to transmit data. - rtxOffset = bytesRead - expected*maxPayload/2 - - // Now send a partial ACK w/ a SACK block that indicates that the next 3 - // segments are lost and we have received 6 segments after the lost - // segments. This should cause the sender to immediately transmit all 3 - // segments in response to this ACK unlike in FastRecovery where only 1 - // segment is retransmitted per ACK. - start = c.IRS.Add(seqnum.Size(rtxOffset) + 30 + 1) - end = start.Add(60) - c.SendAckWithSACK(790, rtxOffset, []header.SACKBlock{{start, end}}) - - // At this point, we acked expected/2 packets and we SACKED 6 packets and - // 3 segments were considered lost due to the SACK block we sent. - // - // So total packets outstanding can be calculated as follows after 7 - // iterations of slow start -> 10/20/40/80/160/320/640. So expected - // should be 640 at start, then we went to recover at which point the - // cwnd should be set to 320 + 3 (for the 3 dupAcks which have left the - // network). - // Outstanding at this point after acking half the window - // (320 packets) will be: - // outstanding = 640-320-6(due to SACK block)-3 = 311 - // - // The last 3 is due to the fact that the first 3 packets after - // rtxOffset will be considered lost due to the SACK blocks sent. - // Receive the retransmit due to partial ack. - - c.ReceiveAndCheckPacketWithOptions(data, rtxOffset, maxPayload, tsOptionSize) - // Receive the 2 extra packets that should have been retransmitted as - // those should be considered lost and immediately retransmitted based - // on the SACK information in the previous ACK sent above. - for i := 0; i < 2; i++ { - c.ReceiveAndCheckPacketWithOptions(data, rtxOffset+maxPayload*(i+1), maxPayload, tsOptionSize) - } - - // Now we should get 9 more new unsent packets as the cwnd is 323 and - // outstanding is 311. - for i := 0; i < 9; i++ { - c.ReceiveAndCheckPacketWithOptions(data, bytesRead, maxPayload, tsOptionSize) - bytesRead += maxPayload - } - - // In SACK recovery only the first segment is fast retransmitted when - // entering recovery. - if got, want := c.Stack().Stats().TCP.FastRetransmit.Value(), uint64(1); got != want { - t.Errorf("got stats.TCP.FastRetransmit.Value = %v, want = %v", got, want) - } - - if got, want := c.Stack().Stats().TCP.Retransmits.Value(), uint64(4); got != want { - t.Errorf("got stats.TCP.Retransmits.Value = %v, want = %v", got, want) - } - - c.CheckNoPacketTimeout("More packets received than expected during recovery after partial ack for this cwnd.", 50*time.Millisecond) - - // Acknowledge all pending data to recover point. - c.SendAck(790, recover) - - // At this point, the cwnd should reset to expected/2 and there are 9 - // packets outstanding. - // - // Now in the first iteration since there are 9 packets outstanding. - // We would expect to get expected/2 - 9 packets. But subsequent - // iterations will send us expected/2 + 1 (per iteration). - expected = expected/2 - 9 - for i := 0; i < iterations; i++ { - // Read all packets expected on this iteration. Don't - // acknowledge any of them just yet, so that we can measure the - // congestion window. - for j := 0; j < expected; j++ { - c.ReceiveAndCheckPacketWithOptions(data, bytesRead, maxPayload, tsOptionSize) - bytesRead += maxPayload - } - // Check we don't receive any more packets on this iteration. - // The timeout can't be too high or we'll trigger a timeout. - c.CheckNoPacketTimeout(fmt.Sprintf("More packets received(after deflation) than expected %d for this cwnd and iteration: %d.", expected, i), 50*time.Millisecond) - - // Acknowledge all the data received so far. - c.SendAck(790, bytesRead) - - // In cogestion avoidance, the packets trains increase by 1 in - // each iteration. - if i == 0 { - // After the first iteration we expect to get the full - // congestion window worth of packets in every - // iteration. - expected += 9 - } - expected++ - } -} diff --git a/pkg/tcpip/transport/tcp/tcp_segment_list.go b/pkg/tcpip/transport/tcp/tcp_segment_list.go new file mode 100755 index 000000000..029f98a11 --- /dev/null +++ b/pkg/tcpip/transport/tcp/tcp_segment_list.go @@ -0,0 +1,173 @@ +package tcp + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type segmentElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (segmentElementMapper) linkerFor(elem *segment) *segment { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type segmentList struct { + head *segment + tail *segment +} + +// Reset resets list l to the empty state. +func (l *segmentList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *segmentList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *segmentList) Front() *segment { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *segmentList) Back() *segment { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *segmentList) PushFront(e *segment) { + segmentElementMapper{}.linkerFor(e).SetNext(l.head) + segmentElementMapper{}.linkerFor(e).SetPrev(nil) + + if l.head != nil { + segmentElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *segmentList) PushBack(e *segment) { + segmentElementMapper{}.linkerFor(e).SetNext(nil) + segmentElementMapper{}.linkerFor(e).SetPrev(l.tail) + + if l.tail != nil { + segmentElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *segmentList) PushBackList(m *segmentList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + segmentElementMapper{}.linkerFor(l.tail).SetNext(m.head) + segmentElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *segmentList) InsertAfter(b, e *segment) { + a := segmentElementMapper{}.linkerFor(b).Next() + segmentElementMapper{}.linkerFor(e).SetNext(a) + segmentElementMapper{}.linkerFor(e).SetPrev(b) + segmentElementMapper{}.linkerFor(b).SetNext(e) + + if a != nil { + segmentElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *segmentList) InsertBefore(a, e *segment) { + b := segmentElementMapper{}.linkerFor(a).Prev() + segmentElementMapper{}.linkerFor(e).SetNext(a) + segmentElementMapper{}.linkerFor(e).SetPrev(b) + segmentElementMapper{}.linkerFor(a).SetPrev(e) + + if b != nil { + segmentElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *segmentList) Remove(e *segment) { + prev := segmentElementMapper{}.linkerFor(e).Prev() + next := segmentElementMapper{}.linkerFor(e).Next() + + if prev != nil { + segmentElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + segmentElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type segmentEntry struct { + next *segment + prev *segment +} + +// Next returns the entry that follows e in the list. +func (e *segmentEntry) Next() *segment { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *segmentEntry) Prev() *segment { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *segmentEntry) SetNext(elem *segment) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *segmentEntry) SetPrev(elem *segment) { + e.prev = elem +} diff --git a/pkg/tcpip/transport/tcp/tcp_state_autogen.go b/pkg/tcpip/transport/tcp/tcp_state_autogen.go new file mode 100755 index 000000000..d92c1b67d --- /dev/null +++ b/pkg/tcpip/transport/tcp/tcp_state_autogen.go @@ -0,0 +1,469 @@ +// automatically generated by stateify. + +package tcp + +import ( + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) + +func (x *cubicState) beforeSave() {} +func (x *cubicState) save(m state.Map) { + x.beforeSave() + var t unixTime = x.saveT() + m.SaveValue("t", t) + m.Save("wLastMax", &x.wLastMax) + m.Save("wMax", &x.wMax) + m.Save("numCongestionEvents", &x.numCongestionEvents) + m.Save("c", &x.c) + m.Save("k", &x.k) + m.Save("beta", &x.beta) + m.Save("wC", &x.wC) + m.Save("wEst", &x.wEst) + m.Save("s", &x.s) +} + +func (x *cubicState) afterLoad() {} +func (x *cubicState) load(m state.Map) { + m.Load("wLastMax", &x.wLastMax) + m.Load("wMax", &x.wMax) + m.Load("numCongestionEvents", &x.numCongestionEvents) + m.Load("c", &x.c) + m.Load("k", &x.k) + m.Load("beta", &x.beta) + m.Load("wC", &x.wC) + m.Load("wEst", &x.wEst) + m.Load("s", &x.s) + m.LoadValue("t", new(unixTime), func(y interface{}) { x.loadT(y.(unixTime)) }) +} + +func (x *SACKInfo) beforeSave() {} +func (x *SACKInfo) save(m state.Map) { + x.beforeSave() + m.Save("Blocks", &x.Blocks) + m.Save("NumBlocks", &x.NumBlocks) +} + +func (x *SACKInfo) afterLoad() {} +func (x *SACKInfo) load(m state.Map) { + m.Load("Blocks", &x.Blocks) + m.Load("NumBlocks", &x.NumBlocks) +} + +func (x *rcvBufAutoTuneParams) beforeSave() {} +func (x *rcvBufAutoTuneParams) save(m state.Map) { + x.beforeSave() + var measureTime unixTime = x.saveMeasureTime() + m.SaveValue("measureTime", measureTime) + var rttMeasureTime unixTime = x.saveRttMeasureTime() + m.SaveValue("rttMeasureTime", rttMeasureTime) + m.Save("copied", &x.copied) + m.Save("prevCopied", &x.prevCopied) + m.Save("rtt", &x.rtt) + m.Save("rttMeasureSeqNumber", &x.rttMeasureSeqNumber) + m.Save("disabled", &x.disabled) +} + +func (x *rcvBufAutoTuneParams) afterLoad() {} +func (x *rcvBufAutoTuneParams) load(m state.Map) { + m.Load("copied", &x.copied) + m.Load("prevCopied", &x.prevCopied) + m.Load("rtt", &x.rtt) + m.Load("rttMeasureSeqNumber", &x.rttMeasureSeqNumber) + m.Load("disabled", &x.disabled) + m.LoadValue("measureTime", new(unixTime), func(y interface{}) { x.loadMeasureTime(y.(unixTime)) }) + m.LoadValue("rttMeasureTime", new(unixTime), func(y interface{}) { x.loadRttMeasureTime(y.(unixTime)) }) +} + +func (x *endpoint) save(m state.Map) { + x.beforeSave() + var lastError string = x.saveLastError() + m.SaveValue("lastError", lastError) + var state EndpointState = x.saveState() + m.SaveValue("state", state) + var hardError string = x.saveHardError() + m.SaveValue("hardError", hardError) + var acceptedChan []*endpoint = x.saveAcceptedChan() + m.SaveValue("acceptedChan", acceptedChan) + m.Save("netProto", &x.netProto) + m.Save("waiterQueue", &x.waiterQueue) + m.Save("rcvList", &x.rcvList) + m.Save("rcvClosed", &x.rcvClosed) + m.Save("rcvBufSize", &x.rcvBufSize) + m.Save("rcvBufUsed", &x.rcvBufUsed) + m.Save("rcvAutoParams", &x.rcvAutoParams) + m.Save("zeroWindow", &x.zeroWindow) + m.Save("id", &x.id) + m.Save("isRegistered", &x.isRegistered) + m.Save("v6only", &x.v6only) + m.Save("isConnectNotified", &x.isConnectNotified) + m.Save("broadcast", &x.broadcast) + m.Save("workerRunning", &x.workerRunning) + m.Save("workerCleanup", &x.workerCleanup) + m.Save("sendTSOk", &x.sendTSOk) + m.Save("recentTS", &x.recentTS) + m.Save("tsOffset", &x.tsOffset) + m.Save("shutdownFlags", &x.shutdownFlags) + m.Save("sackPermitted", &x.sackPermitted) + m.Save("sack", &x.sack) + m.Save("reusePort", &x.reusePort) + m.Save("delay", &x.delay) + m.Save("cork", &x.cork) + m.Save("scoreboard", &x.scoreboard) + m.Save("reuseAddr", &x.reuseAddr) + m.Save("slowAck", &x.slowAck) + m.Save("segmentQueue", &x.segmentQueue) + m.Save("synRcvdCount", &x.synRcvdCount) + m.Save("userMSS", &x.userMSS) + m.Save("sndBufSize", &x.sndBufSize) + m.Save("sndBufUsed", &x.sndBufUsed) + m.Save("sndClosed", &x.sndClosed) + m.Save("sndBufInQueue", &x.sndBufInQueue) + m.Save("sndQueue", &x.sndQueue) + m.Save("cc", &x.cc) + m.Save("packetTooBigCount", &x.packetTooBigCount) + m.Save("sndMTU", &x.sndMTU) + m.Save("keepalive", &x.keepalive) + m.Save("rcv", &x.rcv) + m.Save("snd", &x.snd) + m.Save("bindAddress", &x.bindAddress) + m.Save("connectingAddress", &x.connectingAddress) + m.Save("amss", &x.amss) + m.Save("gso", &x.gso) +} + +func (x *endpoint) load(m state.Map) { + m.Load("netProto", &x.netProto) + m.LoadWait("waiterQueue", &x.waiterQueue) + m.LoadWait("rcvList", &x.rcvList) + m.Load("rcvClosed", &x.rcvClosed) + m.Load("rcvBufSize", &x.rcvBufSize) + m.Load("rcvBufUsed", &x.rcvBufUsed) + m.Load("rcvAutoParams", &x.rcvAutoParams) + m.Load("zeroWindow", &x.zeroWindow) + m.Load("id", &x.id) + m.Load("isRegistered", &x.isRegistered) + m.Load("v6only", &x.v6only) + m.Load("isConnectNotified", &x.isConnectNotified) + m.Load("broadcast", &x.broadcast) + m.Load("workerRunning", &x.workerRunning) + m.Load("workerCleanup", &x.workerCleanup) + m.Load("sendTSOk", &x.sendTSOk) + m.Load("recentTS", &x.recentTS) + m.Load("tsOffset", &x.tsOffset) + m.Load("shutdownFlags", &x.shutdownFlags) + m.Load("sackPermitted", &x.sackPermitted) + m.Load("sack", &x.sack) + m.Load("reusePort", &x.reusePort) + m.Load("delay", &x.delay) + m.Load("cork", &x.cork) + m.Load("scoreboard", &x.scoreboard) + m.Load("reuseAddr", &x.reuseAddr) + m.Load("slowAck", &x.slowAck) + m.LoadWait("segmentQueue", &x.segmentQueue) + m.Load("synRcvdCount", &x.synRcvdCount) + m.Load("userMSS", &x.userMSS) + m.Load("sndBufSize", &x.sndBufSize) + m.Load("sndBufUsed", &x.sndBufUsed) + m.Load("sndClosed", &x.sndClosed) + m.Load("sndBufInQueue", &x.sndBufInQueue) + m.LoadWait("sndQueue", &x.sndQueue) + m.Load("cc", &x.cc) + m.Load("packetTooBigCount", &x.packetTooBigCount) + m.Load("sndMTU", &x.sndMTU) + m.Load("keepalive", &x.keepalive) + m.LoadWait("rcv", &x.rcv) + m.LoadWait("snd", &x.snd) + m.Load("bindAddress", &x.bindAddress) + m.Load("connectingAddress", &x.connectingAddress) + m.Load("amss", &x.amss) + m.Load("gso", &x.gso) + m.LoadValue("lastError", new(string), func(y interface{}) { x.loadLastError(y.(string)) }) + m.LoadValue("state", new(EndpointState), func(y interface{}) { x.loadState(y.(EndpointState)) }) + m.LoadValue("hardError", new(string), func(y interface{}) { x.loadHardError(y.(string)) }) + m.LoadValue("acceptedChan", new([]*endpoint), func(y interface{}) { x.loadAcceptedChan(y.([]*endpoint)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *keepalive) beforeSave() {} +func (x *keepalive) save(m state.Map) { + x.beforeSave() + m.Save("enabled", &x.enabled) + m.Save("idle", &x.idle) + m.Save("interval", &x.interval) + m.Save("count", &x.count) + m.Save("unacked", &x.unacked) +} + +func (x *keepalive) afterLoad() {} +func (x *keepalive) load(m state.Map) { + m.Load("enabled", &x.enabled) + m.Load("idle", &x.idle) + m.Load("interval", &x.interval) + m.Load("count", &x.count) + m.Load("unacked", &x.unacked) +} + +func (x *receiver) beforeSave() {} +func (x *receiver) save(m state.Map) { + x.beforeSave() + m.Save("ep", &x.ep) + m.Save("rcvNxt", &x.rcvNxt) + m.Save("rcvAcc", &x.rcvAcc) + m.Save("rcvWnd", &x.rcvWnd) + m.Save("rcvWndScale", &x.rcvWndScale) + m.Save("closed", &x.closed) + m.Save("pendingRcvdSegments", &x.pendingRcvdSegments) + m.Save("pendingBufUsed", &x.pendingBufUsed) + m.Save("pendingBufSize", &x.pendingBufSize) +} + +func (x *receiver) afterLoad() {} +func (x *receiver) load(m state.Map) { + m.Load("ep", &x.ep) + m.Load("rcvNxt", &x.rcvNxt) + m.Load("rcvAcc", &x.rcvAcc) + m.Load("rcvWnd", &x.rcvWnd) + m.Load("rcvWndScale", &x.rcvWndScale) + m.Load("closed", &x.closed) + m.Load("pendingRcvdSegments", &x.pendingRcvdSegments) + m.Load("pendingBufUsed", &x.pendingBufUsed) + m.Load("pendingBufSize", &x.pendingBufSize) +} + +func (x *renoState) beforeSave() {} +func (x *renoState) save(m state.Map) { + x.beforeSave() + m.Save("s", &x.s) +} + +func (x *renoState) afterLoad() {} +func (x *renoState) load(m state.Map) { + m.Load("s", &x.s) +} + +func (x *SACKScoreboard) beforeSave() {} +func (x *SACKScoreboard) save(m state.Map) { + x.beforeSave() + m.Save("smss", &x.smss) + m.Save("maxSACKED", &x.maxSACKED) +} + +func (x *SACKScoreboard) afterLoad() {} +func (x *SACKScoreboard) load(m state.Map) { + m.Load("smss", &x.smss) + m.Load("maxSACKED", &x.maxSACKED) +} + +func (x *segment) beforeSave() {} +func (x *segment) save(m state.Map) { + x.beforeSave() + var data buffer.VectorisedView = x.saveData() + m.SaveValue("data", data) + var options []byte = x.saveOptions() + m.SaveValue("options", options) + var rcvdTime unixTime = x.saveRcvdTime() + m.SaveValue("rcvdTime", rcvdTime) + var xmitTime unixTime = x.saveXmitTime() + m.SaveValue("xmitTime", xmitTime) + m.Save("segmentEntry", &x.segmentEntry) + m.Save("refCnt", &x.refCnt) + m.Save("viewToDeliver", &x.viewToDeliver) + m.Save("sequenceNumber", &x.sequenceNumber) + m.Save("ackNumber", &x.ackNumber) + m.Save("flags", &x.flags) + m.Save("window", &x.window) + m.Save("csum", &x.csum) + m.Save("csumValid", &x.csumValid) + m.Save("parsedOptions", &x.parsedOptions) + m.Save("hasNewSACKInfo", &x.hasNewSACKInfo) +} + +func (x *segment) afterLoad() {} +func (x *segment) load(m state.Map) { + m.Load("segmentEntry", &x.segmentEntry) + m.Load("refCnt", &x.refCnt) + m.Load("viewToDeliver", &x.viewToDeliver) + m.Load("sequenceNumber", &x.sequenceNumber) + m.Load("ackNumber", &x.ackNumber) + m.Load("flags", &x.flags) + m.Load("window", &x.window) + m.Load("csum", &x.csum) + m.Load("csumValid", &x.csumValid) + m.Load("parsedOptions", &x.parsedOptions) + m.Load("hasNewSACKInfo", &x.hasNewSACKInfo) + m.LoadValue("data", new(buffer.VectorisedView), func(y interface{}) { x.loadData(y.(buffer.VectorisedView)) }) + m.LoadValue("options", new([]byte), func(y interface{}) { x.loadOptions(y.([]byte)) }) + m.LoadValue("rcvdTime", new(unixTime), func(y interface{}) { x.loadRcvdTime(y.(unixTime)) }) + m.LoadValue("xmitTime", new(unixTime), func(y interface{}) { x.loadXmitTime(y.(unixTime)) }) +} + +func (x *segmentQueue) beforeSave() {} +func (x *segmentQueue) save(m state.Map) { + x.beforeSave() + m.Save("list", &x.list) + m.Save("limit", &x.limit) + m.Save("used", &x.used) +} + +func (x *segmentQueue) afterLoad() {} +func (x *segmentQueue) load(m state.Map) { + m.LoadWait("list", &x.list) + m.Load("limit", &x.limit) + m.Load("used", &x.used) +} + +func (x *sender) beforeSave() {} +func (x *sender) save(m state.Map) { + x.beforeSave() + var lastSendTime unixTime = x.saveLastSendTime() + m.SaveValue("lastSendTime", lastSendTime) + var rttMeasureTime unixTime = x.saveRttMeasureTime() + m.SaveValue("rttMeasureTime", rttMeasureTime) + m.Save("ep", &x.ep) + m.Save("dupAckCount", &x.dupAckCount) + m.Save("fr", &x.fr) + m.Save("sndCwnd", &x.sndCwnd) + m.Save("sndSsthresh", &x.sndSsthresh) + m.Save("sndCAAckCount", &x.sndCAAckCount) + m.Save("outstanding", &x.outstanding) + m.Save("sndWnd", &x.sndWnd) + m.Save("sndUna", &x.sndUna) + m.Save("sndNxt", &x.sndNxt) + m.Save("sndNxtList", &x.sndNxtList) + m.Save("rttMeasureSeqNum", &x.rttMeasureSeqNum) + m.Save("closed", &x.closed) + m.Save("writeNext", &x.writeNext) + m.Save("writeList", &x.writeList) + m.Save("rtt", &x.rtt) + m.Save("rto", &x.rto) + m.Save("maxPayloadSize", &x.maxPayloadSize) + m.Save("gso", &x.gso) + m.Save("sndWndScale", &x.sndWndScale) + m.Save("maxSentAck", &x.maxSentAck) + m.Save("state", &x.state) + m.Save("cc", &x.cc) +} + +func (x *sender) load(m state.Map) { + m.Load("ep", &x.ep) + m.Load("dupAckCount", &x.dupAckCount) + m.Load("fr", &x.fr) + m.Load("sndCwnd", &x.sndCwnd) + m.Load("sndSsthresh", &x.sndSsthresh) + m.Load("sndCAAckCount", &x.sndCAAckCount) + m.Load("outstanding", &x.outstanding) + m.Load("sndWnd", &x.sndWnd) + m.Load("sndUna", &x.sndUna) + m.Load("sndNxt", &x.sndNxt) + m.Load("sndNxtList", &x.sndNxtList) + m.Load("rttMeasureSeqNum", &x.rttMeasureSeqNum) + m.Load("closed", &x.closed) + m.Load("writeNext", &x.writeNext) + m.Load("writeList", &x.writeList) + m.Load("rtt", &x.rtt) + m.Load("rto", &x.rto) + m.Load("maxPayloadSize", &x.maxPayloadSize) + m.Load("gso", &x.gso) + m.Load("sndWndScale", &x.sndWndScale) + m.Load("maxSentAck", &x.maxSentAck) + m.Load("state", &x.state) + m.Load("cc", &x.cc) + m.LoadValue("lastSendTime", new(unixTime), func(y interface{}) { x.loadLastSendTime(y.(unixTime)) }) + m.LoadValue("rttMeasureTime", new(unixTime), func(y interface{}) { x.loadRttMeasureTime(y.(unixTime)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *rtt) beforeSave() {} +func (x *rtt) save(m state.Map) { + x.beforeSave() + m.Save("srtt", &x.srtt) + m.Save("rttvar", &x.rttvar) + m.Save("srttInited", &x.srttInited) +} + +func (x *rtt) afterLoad() {} +func (x *rtt) load(m state.Map) { + m.Load("srtt", &x.srtt) + m.Load("rttvar", &x.rttvar) + m.Load("srttInited", &x.srttInited) +} + +func (x *fastRecovery) beforeSave() {} +func (x *fastRecovery) save(m state.Map) { + x.beforeSave() + m.Save("active", &x.active) + m.Save("first", &x.first) + m.Save("last", &x.last) + m.Save("maxCwnd", &x.maxCwnd) + m.Save("highRxt", &x.highRxt) + m.Save("rescueRxt", &x.rescueRxt) +} + +func (x *fastRecovery) afterLoad() {} +func (x *fastRecovery) load(m state.Map) { + m.Load("active", &x.active) + m.Load("first", &x.first) + m.Load("last", &x.last) + m.Load("maxCwnd", &x.maxCwnd) + m.Load("highRxt", &x.highRxt) + m.Load("rescueRxt", &x.rescueRxt) +} + +func (x *unixTime) beforeSave() {} +func (x *unixTime) save(m state.Map) { + x.beforeSave() + m.Save("second", &x.second) + m.Save("nano", &x.nano) +} + +func (x *unixTime) afterLoad() {} +func (x *unixTime) load(m state.Map) { + m.Load("second", &x.second) + m.Load("nano", &x.nano) +} + +func (x *segmentList) beforeSave() {} +func (x *segmentList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *segmentList) afterLoad() {} +func (x *segmentList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *segmentEntry) beforeSave() {} +func (x *segmentEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *segmentEntry) afterLoad() {} +func (x *segmentEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("tcp.cubicState", (*cubicState)(nil), state.Fns{Save: (*cubicState).save, Load: (*cubicState).load}) + state.Register("tcp.SACKInfo", (*SACKInfo)(nil), state.Fns{Save: (*SACKInfo).save, Load: (*SACKInfo).load}) + state.Register("tcp.rcvBufAutoTuneParams", (*rcvBufAutoTuneParams)(nil), state.Fns{Save: (*rcvBufAutoTuneParams).save, Load: (*rcvBufAutoTuneParams).load}) + state.Register("tcp.endpoint", (*endpoint)(nil), state.Fns{Save: (*endpoint).save, Load: (*endpoint).load}) + state.Register("tcp.keepalive", (*keepalive)(nil), state.Fns{Save: (*keepalive).save, Load: (*keepalive).load}) + state.Register("tcp.receiver", (*receiver)(nil), state.Fns{Save: (*receiver).save, Load: (*receiver).load}) + state.Register("tcp.renoState", (*renoState)(nil), state.Fns{Save: (*renoState).save, Load: (*renoState).load}) + state.Register("tcp.SACKScoreboard", (*SACKScoreboard)(nil), state.Fns{Save: (*SACKScoreboard).save, Load: (*SACKScoreboard).load}) + state.Register("tcp.segment", (*segment)(nil), state.Fns{Save: (*segment).save, Load: (*segment).load}) + state.Register("tcp.segmentQueue", (*segmentQueue)(nil), state.Fns{Save: (*segmentQueue).save, Load: (*segmentQueue).load}) + state.Register("tcp.sender", (*sender)(nil), state.Fns{Save: (*sender).save, Load: (*sender).load}) + state.Register("tcp.rtt", (*rtt)(nil), state.Fns{Save: (*rtt).save, Load: (*rtt).load}) + state.Register("tcp.fastRecovery", (*fastRecovery)(nil), state.Fns{Save: (*fastRecovery).save, Load: (*fastRecovery).load}) + state.Register("tcp.unixTime", (*unixTime)(nil), state.Fns{Save: (*unixTime).save, Load: (*unixTime).load}) + state.Register("tcp.segmentList", (*segmentList)(nil), state.Fns{Save: (*segmentList).save, Load: (*segmentList).load}) + state.Register("tcp.segmentEntry", (*segmentEntry)(nil), state.Fns{Save: (*segmentEntry).save, Load: (*segmentEntry).load}) +} diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go deleted file mode 100644 index 32bb45224..000000000 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ /dev/null @@ -1,4370 +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 tcp_test - -import ( - "bytes" - "fmt" - "math" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/checker" - "gvisor.dev/gvisor/pkg/tcpip/header" - "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" - "gvisor.dev/gvisor/pkg/tcpip/ports" - "gvisor.dev/gvisor/pkg/tcpip/seqnum" - "gvisor.dev/gvisor/pkg/tcpip/stack" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context" - "gvisor.dev/gvisor/pkg/waiter" -) - -const ( - // defaultMTU is the MTU, in bytes, used throughout the tests, except - // where another value is explicitly used. It is chosen to match the MTU - // of loopback interfaces on linux systems. - defaultMTU = 65535 - - // defaultIPv4MSS is the MSS sent by the network stack in SYN/SYN-ACK for an - // IPv4 endpoint when the MTU is set to defaultMTU in the test. - defaultIPv4MSS = defaultMTU - header.IPv4MinimumSize - header.TCPMinimumSize -) - -func TestGiveUpConnect(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - var wq waiter.Queue - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - // Register for notification, then start connection attempt. - waitEntry, notifyCh := waiter.NewChannelEntry(nil) - wq.EventRegister(&waitEntry, waiter.EventOut) - defer wq.EventUnregister(&waitEntry) - - if err := ep.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { - t.Fatalf("got ep.Connect(...) = %v, want = %v", err, tcpip.ErrConnectStarted) - } - - // Close the connection, wait for completion. - ep.Close() - - // Wait for ep to become writable. - <-notifyCh - if err := ep.GetSockOpt(tcpip.ErrorOption{}); err != tcpip.ErrAborted { - t.Fatalf("got ep.GetSockOpt(tcpip.ErrorOption{}) = %v, want = %v", err, tcpip.ErrAborted) - } -} - -func TestConnectIncrementActiveConnection(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - stats := c.Stack().Stats() - want := stats.TCP.ActiveConnectionOpenings.Value() + 1 - - c.CreateConnected(789, 30000, nil) - if got := stats.TCP.ActiveConnectionOpenings.Value(); got != want { - t.Errorf("got stats.TCP.ActtiveConnectionOpenings.Value() = %v, want = %v", got, want) - } -} - -func TestConnectDoesNotIncrementFailedConnectionAttempts(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - stats := c.Stack().Stats() - want := stats.TCP.FailedConnectionAttempts.Value() - - c.CreateConnected(789, 30000, nil) - if got := stats.TCP.FailedConnectionAttempts.Value(); got != want { - t.Errorf("got stats.TCP.FailedConnectionOpenings.Value() = %v, want = %v", got, want) - } -} - -func TestActiveFailedConnectionAttemptIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - stats := c.Stack().Stats() - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - c.EP = ep - want := stats.TCP.FailedConnectionAttempts.Value() + 1 - - if err := c.EP.Connect(tcpip.FullAddress{NIC: 2, Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrNoRoute { - t.Errorf("got c.EP.Connect(...) = %v, want = %v", err, tcpip.ErrNoRoute) - } - - if got := stats.TCP.FailedConnectionAttempts.Value(); got != want { - t.Errorf("got stats.TCP.FailedConnectionAttempts.Value() = %v, want = %v", got, want) - } -} - -func TestTCPSegmentsSentIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - stats := c.Stack().Stats() - // SYN and ACK - want := stats.TCP.SegmentsSent.Value() + 2 - c.CreateConnected(789, 30000, nil) - - if got := stats.TCP.SegmentsSent.Value(); got != want { - t.Errorf("got stats.TCP.SegmentsSent.Value() = %v, want = %v", got, want) - } -} - -func TestTCPResetsSentIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - stats := c.Stack().Stats() - wq := &waiter.Queue{} - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - want := stats.TCP.SegmentsSent.Value() + 1 - - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Send a SYN request. - iss := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: iss, - }) - - // Receive the SYN-ACK reply. - b := c.GetPacket() - tcpHdr := header.TCP(header.IPv4(b).Payload()) - c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) - - ackHeaders := &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagAck, - SeqNum: iss + 1, - // If the AckNum is not the increment of the last sequence number, a RST - // segment is sent back in response. - AckNum: c.IRS + 2, - } - - // Send ACK. - c.SendPacket(nil, ackHeaders) - - c.GetPacket() - if got := stats.TCP.ResetsSent.Value(); got != want { - t.Errorf("got stats.TCP.ResetsSent.Value() = %v, want = %v", got, want) - } -} - -// TestTCPResetSentForACKWhenNotUsingSynCookies checks that the stack generates -// a RST if an ACK is received on the listening socket for which there is no -// active handshake in progress and we are not using SYN cookies. -func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - wq := &waiter.Queue{} - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Send a SYN request. - iss := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: iss, - }) - - // Receive the SYN-ACK reply. - b := c.GetPacket() - tcpHdr := header.TCP(header.IPv4(b).Payload()) - c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) - - ackHeaders := &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagAck, - SeqNum: iss + 1, - AckNum: c.IRS + 1, - } - - // Send ACK. - c.SendPacket(nil, ackHeaders) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - - c.EP, _, err = ep.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - c.EP, _, err = ep.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - c.EP.Close() - checker.IPv4(t, c.GetPacket(), checker.TCP( - checker.SrcPort(context.StackPort), - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), - checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) - - finHeaders := &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: iss + 1, - AckNum: c.IRS + 2, - } - - c.SendPacket(nil, finHeaders) - - // Get the ACK to the FIN we just sent. - c.GetPacket() - - // Now resend the same ACK, this ACK should generate a RST as there - // should be no endpoint in SYN-RCVD state and we are not using - // syn-cookies yet. The reason we send the same ACK is we need a valid - // cookie(IRS) generated by the netstack without which the ACK will be - // rejected. - c.SendPacket(nil, ackHeaders) - - checker.IPv4(t, c.GetPacket(), checker.TCP( - checker.SrcPort(context.StackPort), - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) -} - -func TestTCPResetsReceivedIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - stats := c.Stack().Stats() - want := stats.TCP.ResetsReceived.Value() + 1 - iss := seqnum.Value(789) - rcvWnd := seqnum.Size(30000) - c.CreateConnected(iss, rcvWnd, nil) - - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - SeqNum: iss.Add(1), - AckNum: c.IRS.Add(1), - RcvWnd: rcvWnd, - Flags: header.TCPFlagRst, - }) - - if got := stats.TCP.ResetsReceived.Value(); got != want { - t.Errorf("got stats.TCP.ResetsReceived.Value() = %v, want = %v", got, want) - } -} - -func TestTCPResetsDoNotGenerateResets(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - stats := c.Stack().Stats() - want := stats.TCP.ResetsReceived.Value() + 1 - iss := seqnum.Value(789) - rcvWnd := seqnum.Size(30000) - c.CreateConnected(iss, rcvWnd, nil) - - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - SeqNum: iss.Add(1), - AckNum: c.IRS.Add(1), - RcvWnd: rcvWnd, - Flags: header.TCPFlagRst, - }) - - if got := stats.TCP.ResetsReceived.Value(); got != want { - t.Errorf("got stats.TCP.ResetsReceived.Value() = %v, want = %v", got, want) - } - c.CheckNoPacketTimeout("got an unexpected packet", 100*time.Millisecond) -} - -func TestActiveHandshake(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) -} - -func TestNonBlockingClose(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - ep := c.EP - c.EP = nil - - // Close the endpoint and measure how long it takes. - t0 := time.Now() - ep.Close() - if diff := time.Now().Sub(t0); diff > 3*time.Second { - t.Fatalf("Took too long to close: %v", diff) - } -} - -func TestConnectResetAfterClose(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - ep := c.EP - c.EP = nil - - // Close the endpoint, make sure we get a FIN segment, then acknowledge - // to complete closure of sender, but don't send our own FIN. - ep.Close() - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Wait for the ep to give up waiting for a FIN, and send a RST. - time.Sleep(3 * time.Second) - for { - b := c.GetPacket() - tcpHdr := header.TCP(header.IPv4(b).Payload()) - if tcpHdr.Flags() == header.TCPFlagAck|header.TCPFlagFin { - // This is a retransmit of the FIN, ignore it. - continue - } - - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), - ), - ) - break - } -} - -func TestSimpleReceive(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - data := []byte{1, 2, 3} - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // Receive data. - v, _, err := c.EP.Read(nil) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - if !bytes.Equal(data, v) { - t.Fatalf("got data = %v, want = %v", v, data) - } - - // Check that ACK is received. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestOutOfOrderReceive(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - // Send second half of data first, with seqnum 3 ahead of expected. - data := []byte{1, 2, 3, 4, 5, 6} - c.SendPacket(data[3:], &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 793, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Check that we get an ACK specifying which seqnum is expected. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - - // Wait 200ms and check that no data has been received. - time.Sleep(200 * time.Millisecond) - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - // Send the first 3 bytes now. - c.SendPacket(data[:3], &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Receive data. - read := make([]byte, 0, 6) - for len(read) < len(data) { - v, _, err := c.EP.Read(nil) - if err != nil { - if err == tcpip.ErrWouldBlock { - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(5 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - continue - } - t.Fatalf("Read failed: %v", err) - } - - read = append(read, v...) - } - - // Check that we received the data in proper order. - if !bytes.Equal(data, read) { - t.Fatalf("got data = %v, want = %v", read, data) - } - - // Check that the whole data is acknowledged. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestOutOfOrderFlood(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create a new connection with initial window size of 10. - opt := tcpip.ReceiveBufferSizeOption(10) - c.CreateConnected(789, 30000, &opt) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - // Send 100 packets before the actual one that is expected. - data := []byte{1, 2, 3, 4, 5, 6} - for i := 0; i < 100; i++ { - c.SendPacket(data[3:], &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 796, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - } - - // Send packet with seqnum 793. It must be discarded because the - // out-of-order buffer was filled by the previous packets. - c.SendPacket(data[3:], &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 793, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - - // Now send the expected packet, seqnum 790. - c.SendPacket(data[:3], &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Check that only packet 790 is acknowledged. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(793), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestRstOnCloseWithUnreadData(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - data := []byte{1, 2, 3} - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(3 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // Check that ACK is received, this happens regardless of the read. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - - // Now that we know we have unread data, let's just close the connection - // and verify that netstack sends an RST rather than a FIN. - c.EP.Close() - - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), - // We shouldn't consume a sequence number on RST. - checker.SeqNum(uint32(c.IRS)+1), - )) - // The RST puts the endpoint into an error state. - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - // This final ACK should be ignored because an ACK on a reset doesn't mean - // anything. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790 + len(data)), - AckNum: c.IRS.Add(seqnum.Size(2)), - RcvWnd: 30000, - }) -} - -func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - data := []byte{1, 2, 3} - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(3 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // Check that ACK is received, this happens regardless of the read. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - - // Cause a FIN to be generated. - c.EP.Shutdown(tcpip.ShutdownWrite) - - // Make sure we get the FIN but DON't ACK IT. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - checker.SeqNum(uint32(c.IRS)+1), - )) - - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateFinWait1; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - // Cause a RST to be generated by closing the read end now since we have - // unread data. - c.EP.Shutdown(tcpip.ShutdownRead) - - // Make sure we get the RST - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), - // We shouldn't consume a sequence number on RST. - checker.SeqNum(uint32(c.IRS)+1), - )) - // The RST puts the endpoint into an error state. - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - // The ACK to the FIN should now be rejected since the connection has been - // closed by a RST. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790 + len(data)), - AckNum: c.IRS.Add(seqnum.Size(2)), - RcvWnd: 30000, - }) -} - -func TestShutdownRead(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - if err := c.EP.Shutdown(tcpip.ShutdownRead); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrClosedForReceive { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrClosedForReceive) - } -} - -func TestFullWindowReceive(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - opt := tcpip.ReceiveBufferSizeOption(10) - c.CreateConnected(789, 30000, &opt) - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - _, _, err := c.EP.Read(nil) - if err != tcpip.ErrWouldBlock { - t.Fatalf("Read failed: %v", err) - } - - // Fill up the window. - data := []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10} - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(5 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // Check that data is acknowledged, and window goes to zero. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - checker.Window(0), - ), - ) - - // Receive data and check it. - v, _, err := c.EP.Read(nil) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - if !bytes.Equal(data, v) { - t.Fatalf("got data = %v, want = %v", v, data) - } - - // Check that we get an ACK for the newly non-zero window. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - checker.Window(10), - ), - ) -} - -func TestNoWindowShrinking(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Start off with a window size of 10, then shrink it to 5. - opt := tcpip.ReceiveBufferSizeOption(10) - c.CreateConnected(789, 30000, &opt) - - opt = 5 - if err := c.EP.SetSockOpt(opt); err != nil { - t.Fatalf("SetSockOpt failed: %v", err) - } - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - // Send 3 bytes, check that the peer acknowledges them. - data := []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10} - c.SendPacket(data[:3], &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(5 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // Check that data is acknowledged, and that window doesn't go to zero - // just yet because it was previously set to 10. It must go to 7 now. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(793), - checker.TCPFlags(header.TCPFlagAck), - checker.Window(7), - ), - ) - - // Send 7 more bytes, check that the window fills up. - c.SendPacket(data[3:], &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 793, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - select { - case <-ch: - case <-time.After(5 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - checker.Window(0), - ), - ) - - // Receive data and check it. - read := make([]byte, 0, 10) - for len(read) < len(data) { - v, _, err := c.EP.Read(nil) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - read = append(read, v...) - } - - if !bytes.Equal(data, read) { - t.Fatalf("got data = %v, want = %v", read, data) - } - - // Check that we get an ACK for the newly non-zero window, which is the - // new size. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), - checker.TCPFlags(header.TCPFlagAck), - checker.Window(5), - ), - ) -} - -func TestSimpleSend(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Check that data is received. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - if p := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(data, p) { - t.Fatalf("got data = %v, want = %v", p, data) - } - - // Acknowledge the data. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1 + seqnum.Size(len(data))), - RcvWnd: 30000, - }) -} - -func TestZeroWindowSend(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 0, nil) - - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}) - if err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Since the window is currently zero, check that no packet is received. - c.CheckNoPacket("Packet received when window is zero") - - // Open up the window. Data should be received now. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Check that data is received. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - if p := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(data, p) { - t.Fatalf("got data = %v, want = %v", p, data) - } - - // Acknowledge the data. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1 + seqnum.Size(len(data))), - RcvWnd: 30000, - }) -} - -func TestScaledWindowConnect(t *testing.T) { - // This test ensures that window scaling is used when the peer - // does advertise it and connection is established with Connect(). - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Set the window size greater than the maximum non-scaled window. - opt := tcpip.ReceiveBufferSizeOption(65535 * 3) - c.CreateConnectedWithRawOptions(789, 30000, &opt, []byte{ - header.TCPOptionWS, 3, 0, header.TCPOptionNOP, - }) - - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Check that data is received, and that advertised window is 0xbfff, - // that is, that it is scaled. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xbfff), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) -} - -func TestNonScaledWindowConnect(t *testing.T) { - // This test ensures that window scaling is not used when the peer - // doesn't advertise it and connection is established with Connect(). - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Set the window size greater than the maximum non-scaled window. - opt := tcpip.ReceiveBufferSizeOption(65535 * 3) - c.CreateConnected(789, 30000, &opt) - - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Check that data is received, and that advertised window is 0xffff, - // that is, that it's not scaled. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xffff), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) -} - -func TestScaledWindowAccept(t *testing.T) { - // This test ensures that window scaling is used when the peer - // does advertise it and connection is established with Accept(). - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create EP and start listening. - wq := &waiter.Queue{} - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - - // Set the window size greater than the maximum non-scaled window. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(65535 * 3)); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) - } - - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Do 3-way handshake. - c.PassiveConnectWithOptions(100, 2, header.TCPSynOptions{MSS: defaultIPv4MSS}) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - - c.EP, _, err = ep.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - c.EP, _, err = ep.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Check that data is received, and that advertised window is 0xbfff, - // that is, that it is scaled. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xbfff), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) -} - -func TestNonScaledWindowAccept(t *testing.T) { - // This test ensures that window scaling is not used when the peer - // doesn't advertise it and connection is established with Accept(). - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create EP and start listening. - wq := &waiter.Queue{} - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - - // Set the window size greater than the maximum non-scaled window. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(65535 * 3)); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) - } - - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Do 3-way handshake w/ window scaling disabled. The SYN-ACK to the SYN - // should not carry the window scaling option. - c.PassiveConnect(100, -1, header.TCPSynOptions{MSS: defaultIPv4MSS}) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - - c.EP, _, err = ep.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - c.EP, _, err = ep.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Check that data is received, and that advertised window is 0xffff, - // that is, that it's not scaled. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xffff), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) -} - -func TestZeroScaledWindowReceive(t *testing.T) { - // This test ensures that the endpoint sends a non-zero window size - // advertisement when the scaled window transitions from 0 to non-zero, - // but the actual window (not scaled) hasn't gotten to zero. - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Set the window size such that a window scale of 4 will be used. - const wnd = 65535 * 10 - const ws = uint32(4) - opt := tcpip.ReceiveBufferSizeOption(wnd) - c.CreateConnectedWithRawOptions(789, 30000, &opt, []byte{ - header.TCPOptionWS, 3, 0, header.TCPOptionNOP, - }) - - // Write chunks of 50000 bytes. - remain := wnd - sent := 0 - data := make([]byte, 50000) - for remain > len(data) { - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790 + sent), - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - sent += len(data) - remain -= len(data) - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(remain>>ws)), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - } - - // Make the window non-zero, but the scaled window zero. - if remain >= 16 { - data = data[:remain-15] - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790 + sent), - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - sent += len(data) - remain -= len(data) - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(0), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - } - - // Read some data. An ack should be sent in response to that. - v, _, err := c.EP.Read(nil) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(len(v)>>ws)), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestSegmentMerging(t *testing.T) { - tests := []struct { - name string - stop func(tcpip.Endpoint) - resume func(tcpip.Endpoint) - }{ - { - "stop work", - func(ep tcpip.Endpoint) { - ep.(interface{ StopWork() }).StopWork() - }, - func(ep tcpip.Endpoint) { - ep.(interface{ ResumeWork() }).ResumeWork() - }, - }, - { - "cork", - func(ep tcpip.Endpoint) { - ep.SetSockOpt(tcpip.CorkOption(1)) - }, - func(ep tcpip.Endpoint) { - ep.SetSockOpt(tcpip.CorkOption(0)) - }, - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Prevent the endpoint from processing packets. - test.stop(c.EP) - - var allData []byte - for i, data := range [][]byte{{1, 2, 3, 4}, {5, 6, 7}, {8, 9}, {10}, {11}} { - allData = append(allData, data...) - view := buffer.NewViewFromBytes(data) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write #%d failed: %v", i+1, err) - } - } - - // Let the endpoint process the segments that we just sent. - test.resume(c.EP) - - // Check that data is received. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(allData)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - if got := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(got, allData) { - t.Fatalf("got data = %v, want = %v", got, allData) - } - - // Acknowledge the data. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1 + seqnum.Size(len(allData))), - RcvWnd: 30000, - }) - }) - } -} - -func TestDelay(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - c.EP.SetSockOpt(tcpip.DelayOption(1)) - - var allData []byte - for i, data := range [][]byte{{0}, {1, 2, 3, 4}, {5, 6, 7}, {8, 9}, {10}, {11}} { - allData = append(allData, data...) - view := buffer.NewViewFromBytes(data) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write #%d failed: %v", i+1, err) - } - } - - seq := c.IRS.Add(1) - for _, want := range [][]byte{allData[:1], allData[1:]} { - // Check that data is received. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(want)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - if got := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(got, want) { - t.Fatalf("got data = %v, want = %v", got, want) - } - - seq = seq.Add(seqnum.Size(len(want))) - // Acknowledge the data. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: seq, - RcvWnd: 30000, - }) - } -} - -func TestUndelay(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - c.EP.SetSockOpt(tcpip.DelayOption(1)) - - allData := [][]byte{{0}, {1, 2, 3}} - for i, data := range allData { - view := buffer.NewViewFromBytes(data) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write #%d failed: %v", i+1, err) - } - } - - seq := c.IRS.Add(1) - - // Check that data is received. - first := c.GetPacket() - checker.IPv4(t, first, - checker.PayloadLen(len(allData[0])+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - if got, want := first[header.IPv4MinimumSize+header.TCPMinimumSize:], allData[0]; !bytes.Equal(got, want) { - t.Fatalf("got first packet's data = %v, want = %v", got, want) - } - - seq = seq.Add(seqnum.Size(len(allData[0]))) - - // Check that we don't get the second packet yet. - c.CheckNoPacketTimeout("delayed second packet transmitted", 100*time.Millisecond) - - c.EP.SetSockOpt(tcpip.DelayOption(0)) - - // Check that data is received. - second := c.GetPacket() - checker.IPv4(t, second, - checker.PayloadLen(len(allData[1])+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - if got, want := second[header.IPv4MinimumSize+header.TCPMinimumSize:], allData[1]; !bytes.Equal(got, want) { - t.Fatalf("got second packet's data = %v, want = %v", got, want) - } - - seq = seq.Add(seqnum.Size(len(allData[1]))) - - // Acknowledge the data. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: seq, - RcvWnd: 30000, - }) -} - -func TestMSSNotDelayed(t *testing.T) { - tests := []struct { - name string - fn func(tcpip.Endpoint) - }{ - {"no-op", func(tcpip.Endpoint) {}}, - {"delay", func(ep tcpip.Endpoint) { ep.SetSockOpt(tcpip.DelayOption(1)) }}, - {"cork", func(ep tcpip.Endpoint) { ep.SetSockOpt(tcpip.CorkOption(1)) }}, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - const maxPayload = 100 - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnectedWithRawOptions(789, 30000, nil, []byte{ - header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), - }) - - test.fn(c.EP) - - allData := [][]byte{{0}, make([]byte, maxPayload), make([]byte, maxPayload)} - for i, data := range allData { - view := buffer.NewViewFromBytes(data) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write #%d failed: %v", i+1, err) - } - } - - seq := c.IRS.Add(1) - - for i, data := range allData { - // Check that data is received. - packet := c.GetPacket() - checker.IPv4(t, packet, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - if got, want := packet[header.IPv4MinimumSize+header.TCPMinimumSize:], data; !bytes.Equal(got, want) { - t.Fatalf("got packet #%d's data = %v, want = %v", i+1, got, want) - } - - seq = seq.Add(seqnum.Size(len(data))) - } - - // Acknowledge the data. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: seq, - RcvWnd: 30000, - }) - }) - } -} - -func testBrokenUpWrite(t *testing.T, c *context.Context, maxPayload int) { - payloadMultiplier := 10 - dataLen := payloadMultiplier * maxPayload - data := make([]byte, dataLen) - for i := range data { - data[i] = byte(i) - } - - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Check that data is received in chunks. - bytesReceived := 0 - numPackets := 0 - for bytesReceived != dataLen { - b := c.GetPacket() - numPackets++ - tcpHdr := header.TCP(header.IPv4(b).Payload()) - payloadLen := len(tcpHdr.Payload()) - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1+uint32(bytesReceived)), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - pdata := data[bytesReceived : bytesReceived+payloadLen] - if p := tcpHdr.Payload(); !bytes.Equal(pdata, p) { - t.Fatalf("got data = %v, want = %v", p, pdata) - } - bytesReceived += payloadLen - var options []byte - if c.TimeStampEnabled { - // If timestamp option is enabled, echo back the timestamp and increment - // the TSEcr value included in the packet and send that back as the TSVal. - parsedOpts := tcpHdr.ParsedOptions() - tsOpt := [12]byte{header.TCPOptionNOP, header.TCPOptionNOP} - header.EncodeTSOption(parsedOpts.TSEcr+1, parsedOpts.TSVal, tsOpt[2:]) - options = tsOpt[:] - } - // Acknowledge the data. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1 + seqnum.Size(bytesReceived)), - RcvWnd: 30000, - TCPOpts: options, - }) - } - if numPackets == 1 { - t.Fatalf("expected write to be broken up into multiple packets, but got 1 packet") - } -} - -func TestSendGreaterThanMTU(t *testing.T) { - const maxPayload = 100 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - testBrokenUpWrite(t, c, maxPayload) -} - -func TestActiveSendMSSLessThanMTU(t *testing.T) { - const maxPayload = 100 - c := context.New(t, 65535) - defer c.Cleanup() - - c.CreateConnectedWithRawOptions(789, 30000, nil, []byte{ - header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), - }) - testBrokenUpWrite(t, c, maxPayload) -} - -func TestPassiveSendMSSLessThanMTU(t *testing.T) { - const maxPayload = 100 - const mtu = 1200 - c := context.New(t, mtu) - defer c.Cleanup() - - // Create EP and start listening. - wq := &waiter.Queue{} - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - - // Set the buffer size to a deterministic size so that we can check the - // window scaling option. - const rcvBufferSize = 0x20000 - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(rcvBufferSize)); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) - } - - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Do 3-way handshake. - c.PassiveConnect(maxPayload, -1, header.TCPSynOptions{MSS: mtu - header.IPv4MinimumSize - header.TCPMinimumSize}) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - - c.EP, _, err = ep.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - c.EP, _, err = ep.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - // Check that data gets properly segmented. - testBrokenUpWrite(t, c, maxPayload) -} - -func TestSynCookiePassiveSendMSSLessThanMTU(t *testing.T) { - const maxPayload = 536 - const mtu = 2000 - c := context.New(t, mtu) - defer c.Cleanup() - - // Set the SynRcvd threshold to zero to force a syn cookie based accept - // to happen. - saved := tcp.SynRcvdCountThreshold - defer func() { - tcp.SynRcvdCountThreshold = saved - }() - tcp.SynRcvdCountThreshold = 0 - - // Create EP and start listening. - wq := &waiter.Queue{} - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Do 3-way handshake. - c.PassiveConnect(maxPayload, -1, header.TCPSynOptions{MSS: mtu - header.IPv4MinimumSize - header.TCPMinimumSize}) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - - c.EP, _, err = ep.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - c.EP, _, err = ep.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - // Check that data gets properly segmented. - testBrokenUpWrite(t, c, maxPayload) -} - -func TestForwarderSendMSSLessThanMTU(t *testing.T) { - const maxPayload = 100 - const mtu = 1200 - c := context.New(t, mtu) - defer c.Cleanup() - - s := c.Stack() - ch := make(chan *tcpip.Error, 1) - f := tcp.NewForwarder(s, 65536, 10, func(r *tcp.ForwarderRequest) { - var err *tcpip.Error - c.EP, err = r.CreateEndpoint(&c.WQ) - ch <- err - }) - s.SetTransportProtocolHandler(tcp.ProtocolNumber, f.HandlePacket) - - // Do 3-way handshake. - c.PassiveConnect(maxPayload, -1, header.TCPSynOptions{MSS: mtu - header.IPv4MinimumSize - header.TCPMinimumSize}) - - // Wait for connection to be available. - select { - case err := <-ch: - if err != nil { - t.Fatalf("Error creating endpoint: %v", err) - } - case <-time.After(2 * time.Second): - t.Fatalf("Timed out waiting for connection") - } - - // Check that data gets properly segmented. - testBrokenUpWrite(t, c, maxPayload) -} - -func TestSynOptionsOnActiveConnect(t *testing.T) { - const mtu = 1400 - c := context.New(t, mtu) - defer c.Cleanup() - - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - // Set the buffer size to a deterministic size so that we can check the - // window scaling option. - const rcvBufferSize = 0x20000 - const wndScale = 2 - if err := c.EP.SetSockOpt(tcpip.ReceiveBufferSizeOption(rcvBufferSize)); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) - } - - // Start connection attempt. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventOut) - defer c.WQ.EventUnregister(&we) - - if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { - t.Fatalf("got c.EP.Connect(...) = %v, want = %v", err, tcpip.ErrConnectStarted) - } - - // Receive SYN packet. - b := c.GetPacket() - mss := uint16(mtu - header.IPv4MinimumSize - header.TCPMinimumSize) - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagSyn), - checker.TCPSynOptions(header.TCPSynOptions{MSS: mss, WS: wndScale}), - ), - ) - - tcpHdr := header.TCP(header.IPv4(b).Payload()) - c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) - - // Wait for retransmit. - time.Sleep(1 * time.Second) - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagSyn), - checker.SrcPort(tcpHdr.SourcePort()), - checker.SeqNum(tcpHdr.SequenceNumber()), - checker.TCPSynOptions(header.TCPSynOptions{MSS: mss, WS: wndScale}), - ), - ) - - // Send SYN-ACK. - iss := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: tcpHdr.DestinationPort(), - DstPort: tcpHdr.SourcePort(), - Flags: header.TCPFlagSyn | header.TCPFlagAck, - SeqNum: iss, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - // Receive ACK packet. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), - ), - ) - - // Wait for connection to be established. - select { - case <-ch: - if err := c.EP.GetSockOpt(tcpip.ErrorOption{}); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) - } - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for connection") - } -} - -func TestCloseListener(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create listener. - var wq waiter.Queue - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - if err := ep.Bind(tcpip.FullAddress{}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Close the listener and measure how long it takes. - t0 := time.Now() - ep.Close() - if diff := time.Now().Sub(t0); diff > 3*time.Second { - t.Fatalf("Took too long to close: %v", diff) - } -} - -func TestReceiveOnResetConnection(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Send RST segment. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagRst, - SeqNum: 790, - RcvWnd: 30000, - }) - - // Try to read. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - -loop: - for { - switch _, _, err := c.EP.Read(nil); err { - case tcpip.ErrWouldBlock: - select { - case <-ch: - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for reset to arrive") - } - case tcpip.ErrConnectionReset: - break loop - default: - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrConnectionReset) - } - } -} - -func TestSendOnResetConnection(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Send RST segment. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagRst, - SeqNum: 790, - RcvWnd: 30000, - }) - - // Wait for the RST to be received. - time.Sleep(1 * time.Second) - - // Try to write. - view := buffer.NewView(10) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != tcpip.ErrConnectionReset { - t.Fatalf("got c.EP.Write(...) = %v, want = %v", err, tcpip.ErrConnectionReset) - } -} - -func TestFinImmediately(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Shutdown immediately, check that we get a FIN. - if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - - // Ack and send FIN as well. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 790, - AckNum: c.IRS.Add(2), - RcvWnd: 30000, - }) - - // Check that the stack acks the FIN. - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestFinRetransmit(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Shutdown immediately, check that we get a FIN. - if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - - // Don't acknowledge yet. We should get a retransmit of the FIN. - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - - // Ack and send FIN as well. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 790, - AckNum: c.IRS.Add(2), - RcvWnd: 30000, - }) - - // Check that the stack acks the FIN. - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestFinWithNoPendingData(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Write something out, and have it acknowledged. - view := buffer.NewView(10) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - next := uint32(c.IRS) + 1 - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - next += uint32(len(view)) - - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - // Shutdown, check that we get a FIN. - if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - next++ - - // Ack and send FIN as well. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - // Check that the stack acks the FIN. - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestFinWithPendingDataCwndFull(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Write enough segments to fill the congestion window before ACK'ing - // any of them. - view := buffer.NewView(10) - for i := tcp.InitialCwnd; i > 0; i-- { - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - } - - next := uint32(c.IRS) + 1 - for i := tcp.InitialCwnd; i > 0; i-- { - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - next += uint32(len(view)) - } - - // Shutdown the connection, check that the FIN segment isn't sent - // because the congestion window doesn't allow it. Wait until a - // retransmit is received. - if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - // Send the ACK that will allow the FIN to be sent as well. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - next++ - - // Send a FIN that acknowledges everything. Get an ACK back. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestFinWithPendingData(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Write something out, and acknowledge it to get cwnd to 2. - view := buffer.NewView(10) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - next := uint32(c.IRS) + 1 - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - next += uint32(len(view)) - - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - // Write new data, but don't acknowledge it. - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - next += uint32(len(view)) - - // Shutdown the connection, check that we do get a FIN. - if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - next++ - - // Send a FIN that acknowledges everything. Get an ACK back. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), - checker.TCPFlags(header.TCPFlagAck), - ), - ) -} - -func TestFinWithPartialAck(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Write something out, and acknowledge it to get cwnd to 2. Also send - // FIN from the test side. - view := buffer.NewView(10) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - next := uint32(c.IRS) + 1 - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - next += uint32(len(view)) - - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - // Check that we get an ACK for the fin. - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - // Write new data, but don't acknowledge it. - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - next += uint32(len(view)) - - // Shutdown the connection, check that we do get a FIN. - if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - next++ - - // Send an ACK for the data, but not for the FIN yet. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 791, - AckNum: seqnum.Value(next - 1), - RcvWnd: 30000, - }) - - // Check that we don't get a retransmit of the FIN. - c.CheckNoPacketTimeout("FIN retransmitted when data was ack'd", 100*time.Millisecond) - - // Ack the FIN. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 791, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) -} - -func TestUpdateListenBacklog(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create listener. - var wq waiter.Queue - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - if err := ep.Bind(tcpip.FullAddress{}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Update the backlog with another Listen() on the same endpoint. - if err := ep.Listen(20); err != nil { - t.Fatalf("Listen failed to update backlog: %v", err) - } - - ep.Close() -} - -func scaledSendWindow(t *testing.T, scale uint8) { - // This test ensures that the endpoint is using the right scaling by - // sending a buffer that is larger than the window size, and ensuring - // that the endpoint doesn't send more than allowed. - c := context.New(t, defaultMTU) - defer c.Cleanup() - - maxPayload := defaultMTU - header.IPv4MinimumSize - header.TCPMinimumSize - c.CreateConnectedWithRawOptions(789, 0, nil, []byte{ - header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), - header.TCPOptionWS, 3, scale, header.TCPOptionNOP, - }) - - // Open up the window with a scaled value. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 1, - }) - - // Send some data. Check that it's capped by the window size. - view := buffer.NewView(65535) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Check that only data that fits in the scaled window is sent. - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen((1<<scale)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - // Reset the connection to free resources. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagRst, - SeqNum: 790, - }) -} - -func TestScaledSendWindow(t *testing.T) { - for scale := uint8(0); scale <= 14; scale++ { - scaledSendWindow(t, scale) - } -} - -func TestReceivedValidSegmentCountIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - c.CreateConnected(789, 30000, nil) - stats := c.Stack().Stats() - want := stats.TCP.ValidSegmentsReceived.Value() + 1 - - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790), - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - if got := stats.TCP.ValidSegmentsReceived.Value(); got != want { - t.Errorf("got stats.TCP.ValidSegmentsReceived.Value() = %v, want = %v", got, want) - } -} - -func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - c.CreateConnected(789, 30000, nil) - stats := c.Stack().Stats() - want := stats.TCP.InvalidSegmentsReceived.Value() + 1 - vv := c.BuildSegment(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790), - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] - tcpbuf[header.TCPDataOffset] = ((header.TCPMinimumSize - 1) / 4) << 4 - - c.SendSegment(vv) - - if got := stats.TCP.InvalidSegmentsReceived.Value(); got != want { - t.Errorf("got stats.TCP.InvalidSegmentsReceived.Value() = %v, want = %v", got, want) - } -} - -func TestReceivedIncorrectChecksumIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - c.CreateConnected(789, 30000, nil) - stats := c.Stack().Stats() - want := stats.TCP.ChecksumErrors.Value() + 1 - vv := c.BuildSegment([]byte{0x1, 0x2, 0x3}, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790), - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] - // Overwrite a byte in the payload which should cause checksum - // verification to fail. - tcpbuf[(tcpbuf[header.TCPDataOffset]>>4)*4] = 0x4 - - c.SendSegment(vv) - - if got := stats.TCP.ChecksumErrors.Value(); got != want { - t.Errorf("got stats.TCP.ChecksumErrors.Value() = %d, want = %d", got, want) - } -} - -func TestReceivedSegmentQueuing(t *testing.T) { - // This test sends 200 segments containing a few bytes each to an - // endpoint and checks that they're all received and acknowledged by - // the endpoint, that is, that none of the segments are dropped by - // internal queues. - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - // Send 200 segments. - data := []byte{1, 2, 3} - for i := 0; i < 200; i++ { - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seqnum.Value(790 + i*len(data)), - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - } - - // Receive ACKs for all segments. - last := seqnum.Value(790 + 200*len(data)) - for { - b := c.GetPacket() - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - tcpHdr := header.TCP(header.IPv4(b).Payload()) - ack := seqnum.Value(tcpHdr.AckNumber()) - if ack == last { - break - } - - if last.LessThan(ack) { - t.Fatalf("Acknowledge (%v) beyond the expected (%v)", ack, last) - } - } -} - -func TestReadAfterClosedState(t *testing.T) { - // This test ensures that calling Read() or Peek() after the endpoint - // has transitioned to closedState still works if there is pending - // data. To transition to stateClosed without calling Close(), we must - // shutdown the send path and the peer must send its own FIN. - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - // Shutdown immediately for write, check that we get a FIN. - if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) - } - - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - ), - ) - - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateFinWait1; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - // Send some data and acknowledge the FIN. - data := []byte{1, 2, 3} - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck | header.TCPFlagFin, - SeqNum: 790, - AckNum: c.IRS.Add(2), - RcvWnd: 30000, - }) - - // Check that ACK is received. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(uint32(791+len(data))), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - - // Give the stack the chance to transition to closed state. Note that since - // both the sender and receiver are now closed, we effectively skip the - // TIME-WAIT state. - time.Sleep(1 * time.Second) - - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateClose; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // Check that peek works. - peekBuf := make([]byte, 10) - n, _, err := c.EP.Peek([][]byte{peekBuf}) - if err != nil { - t.Fatalf("Peek failed: %v", err) - } - - peekBuf = peekBuf[:n] - if !bytes.Equal(data, peekBuf) { - t.Fatalf("got data = %v, want = %v", peekBuf, data) - } - - // Receive data. - v, _, err := c.EP.Read(nil) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - if !bytes.Equal(data, v) { - t.Fatalf("got data = %v, want = %v", v, data) - } - - // Now that we drained the queue, check that functions fail with the - // right error code. - if _, _, err := c.EP.Read(nil); err != tcpip.ErrClosedForReceive { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrClosedForReceive) - } - - if _, _, err := c.EP.Peek([][]byte{peekBuf}); err != tcpip.ErrClosedForReceive { - t.Fatalf("got c.EP.Peek(...) = %v, want = %v", err, tcpip.ErrClosedForReceive) - } -} - -func TestReusePort(t *testing.T) { - // This test ensures that ports are immediately available for reuse - // after Close on the endpoints using them returns. - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // First case, just an endpoint that was bound. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - c.EP.Close() - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - c.EP.Close() - - // Second case, an endpoint that was bound and is connecting.. - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { - t.Fatalf("got c.EP.Connect(...) = %v, want = %v", err, tcpip.ErrConnectStarted) - } - c.EP.Close() - - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - c.EP.Close() - - // Third case, an endpoint that was bound and is listening. - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - if err := c.EP.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - c.EP.Close() - - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - if err := c.EP.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } -} - -func checkRecvBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { - t.Helper() - - var s tcpip.ReceiveBufferSizeOption - if err := ep.GetSockOpt(&s); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) - } - - if int(s) != v { - t.Fatalf("got receive buffer size = %v, want = %v", s, v) - } -} - -func checkSendBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { - t.Helper() - - var s tcpip.SendBufferSizeOption - if err := ep.GetSockOpt(&s); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) - } - - if int(s) != v { - t.Fatalf("got send buffer size = %v, want = %v", s, v) - } -} - -func TestDefaultBufferSizes(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) - - // Check the default values. - ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - defer func() { - if ep != nil { - ep.Close() - } - }() - - checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize) - checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) - - // Change the default send buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{1, tcp.DefaultSendBufferSize * 2, tcp.DefaultSendBufferSize * 20}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - ep.Close() - ep, err = s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - - checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*2) - checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) - - // Change the default receive buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, tcp.DefaultReceiveBufferSize * 3, tcp.DefaultReceiveBufferSize * 30}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - ep.Close() - ep, err = s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - - checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*2) - checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*3) -} - -func TestMinMaxBufferSizes(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) - - // Check the default values. - ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) - if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) - } - defer ep.Close() - - // Change the min/max values for send/receive - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{200, tcp.DefaultReceiveBufferSize * 2, tcp.DefaultReceiveBufferSize * 20}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{300, tcp.DefaultSendBufferSize * 3, tcp.DefaultSendBufferSize * 30}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - // Set values below the min. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(199)); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) - } - - checkRecvBufferSize(t, ep, 200) - - if err := ep.SetSockOpt(tcpip.SendBufferSizeOption(299)); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) - } - - checkSendBufferSize(t, ep, 300) - - // Set values above the max. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(1 + tcp.DefaultReceiveBufferSize*20)); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) - } - - checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20) - - if err := ep.SetSockOpt(tcpip.SendBufferSizeOption(1 + tcp.DefaultSendBufferSize*30)); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) - } - - checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) -} - -func makeStack() (*stack.Stack, *tcpip.Error) { - s := stack.New([]string{ - ipv4.ProtocolName, - ipv6.ProtocolName, - }, []string{tcp.ProtocolName}, stack.Options{}) - - id := loopback.New() - if testing.Verbose() { - id = sniffer.New(id) - } - - if err := s.CreateNIC(1, id); err != nil { - return nil, err - } - - for _, ct := range []struct { - number tcpip.NetworkProtocolNumber - address tcpip.Address - }{ - {ipv4.ProtocolNumber, context.StackAddr}, - {ipv6.ProtocolNumber, context.StackV6Addr}, - } { - if err := s.AddAddress(1, ct.number, ct.address); err != nil { - return nil, err - } - } - - s.SetRouteTable([]tcpip.Route{ - { - Destination: header.IPv4EmptySubnet, - NIC: 1, - }, - { - Destination: header.IPv6EmptySubnet, - NIC: 1, - }, - }) - - return s, nil -} - -func TestSelfConnect(t *testing.T) { - // This test ensures that intentional self-connects work. In particular, - // it checks that if an endpoint binds to say 127.0.0.1:1000 then - // connects to 127.0.0.1:1000, then it will be connected to itself, and - // is able to send and receive data through the same endpoint. - s, err := makeStack() - if err != nil { - t.Fatal(err) - } - - var wq waiter.Queue - ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Register for notification, then start connection attempt. - waitEntry, notifyCh := waiter.NewChannelEntry(nil) - wq.EventRegister(&waitEntry, waiter.EventOut) - defer wq.EventUnregister(&waitEntry) - - if err := ep.Connect(tcpip.FullAddress{Addr: context.StackAddr, Port: context.StackPort}); err != tcpip.ErrConnectStarted { - t.Fatalf("got ep.Connect(...) = %v, want = %v", err, tcpip.ErrConnectStarted) - } - - <-notifyCh - if err := ep.GetSockOpt(tcpip.ErrorOption{}); err != nil { - t.Fatalf("Connect failed: %v", err) - } - - // Write something. - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - if _, _, err := ep.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - // Read back what was written. - wq.EventUnregister(&waitEntry) - wq.EventRegister(&waitEntry, waiter.EventIn) - rd, _, err := ep.Read(nil) - if err != nil { - if err != tcpip.ErrWouldBlock { - t.Fatalf("Read failed: %v", err) - } - <-notifyCh - rd, _, err = ep.Read(nil) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - } - - if !bytes.Equal(data, rd) { - t.Fatalf("got data = %v, want = %v", rd, data) - } -} - -func TestConnectAvoidsBoundPorts(t *testing.T) { - addressTypes := func(t *testing.T, network string) []string { - switch network { - case "ipv4": - return []string{"v4"} - case "ipv6": - return []string{"v6"} - case "dual": - return []string{"v6", "mapped"} - default: - t.Fatalf("unknown network: '%s'", network) - } - - panic("unreachable") - } - - address := func(t *testing.T, addressType string, isAny bool) tcpip.Address { - switch addressType { - case "v4": - if isAny { - return "" - } - return context.StackAddr - case "v6": - if isAny { - return "" - } - return context.StackV6Addr - case "mapped": - if isAny { - return context.V4MappedWildcardAddr - } - return context.StackV4MappedAddr - default: - t.Fatalf("unknown address type: '%s'", addressType) - } - - panic("unreachable") - } - // This test ensures that Endpoint.Connect doesn't select already-bound ports. - networks := []string{"ipv4", "ipv6", "dual"} - for _, exhaustedNetwork := range networks { - t.Run(fmt.Sprintf("exhaustedNetwork=%s", exhaustedNetwork), func(t *testing.T) { - for _, exhaustedAddressType := range addressTypes(t, exhaustedNetwork) { - t.Run(fmt.Sprintf("exhaustedAddressType=%s", exhaustedAddressType), func(t *testing.T) { - for _, isAny := range []bool{false, true} { - t.Run(fmt.Sprintf("isAny=%t", isAny), func(t *testing.T) { - for _, candidateNetwork := range networks { - t.Run(fmt.Sprintf("candidateNetwork=%s", candidateNetwork), func(t *testing.T) { - for _, candidateAddressType := range addressTypes(t, candidateNetwork) { - t.Run(fmt.Sprintf("candidateAddressType=%s", candidateAddressType), func(t *testing.T) { - s, err := makeStack() - if err != nil { - t.Fatal(err) - } - - var wq waiter.Queue - var eps []tcpip.Endpoint - defer func() { - for _, ep := range eps { - ep.Close() - } - }() - makeEP := func(network string) tcpip.Endpoint { - var networkProtocolNumber tcpip.NetworkProtocolNumber - switch network { - case "ipv4": - networkProtocolNumber = ipv4.ProtocolNumber - case "ipv6", "dual": - networkProtocolNumber = ipv6.ProtocolNumber - default: - t.Fatalf("unknown network: '%s'", network) - } - ep, err := s.NewEndpoint(tcp.ProtocolNumber, networkProtocolNumber, &wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - eps = append(eps, ep) - switch network { - case "ipv4": - case "ipv6": - if err := ep.SetSockOpt(tcpip.V6OnlyOption(1)); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(1)) failed: %v", err) - } - case "dual": - if err := ep.SetSockOpt(tcpip.V6OnlyOption(0)); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(0)) failed: %v", err) - } - default: - t.Fatalf("unknown network: '%s'", network) - } - return ep - } - - var v4reserved, v6reserved bool - switch exhaustedAddressType { - case "v4", "mapped": - v4reserved = true - case "v6": - v6reserved = true - // Dual stack sockets bound to v6 any reserve on v4 as - // well. - if isAny { - switch exhaustedNetwork { - case "ipv6": - case "dual": - v4reserved = true - default: - t.Fatalf("unknown address type: '%s'", exhaustedNetwork) - } - } - default: - t.Fatalf("unknown address type: '%s'", exhaustedAddressType) - } - var collides bool - switch candidateAddressType { - case "v4", "mapped": - collides = v4reserved - case "v6": - collides = v6reserved - default: - t.Fatalf("unknown address type: '%s'", candidateAddressType) - } - - for i := ports.FirstEphemeral; i <= math.MaxUint16; i++ { - if makeEP(exhaustedNetwork).Bind(tcpip.FullAddress{Addr: address(t, exhaustedAddressType, isAny), Port: uint16(i)}); err != nil { - t.Fatalf("Bind(%d) failed: %v", i, err) - } - } - want := tcpip.ErrConnectStarted - if collides { - want = tcpip.ErrNoPortAvailable - } - if err := makeEP(candidateNetwork).Connect(tcpip.FullAddress{Addr: address(t, candidateAddressType, false), Port: 31337}); err != want { - t.Fatalf("got ep.Connect(..) = %v, want = %v", err, want) - } - }) - } - }) - } - }) - } - }) - } - }) - } -} - -func TestPathMTUDiscovery(t *testing.T) { - // This test verifies the stack retransmits packets after it receives an - // ICMP packet indicating that the path MTU has been exceeded. - c := context.New(t, 1500) - defer c.Cleanup() - - // Create new connection with MSS of 1460. - const maxPayload = 1500 - header.TCPMinimumSize - header.IPv4MinimumSize - c.CreateConnectedWithRawOptions(789, 30000, nil, []byte{ - header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), - }) - - // Send 3200 bytes of data. - const writeSize = 3200 - data := buffer.NewView(writeSize) - for i := range data { - data[i] = byte(i) - } - - if _, _, err := c.EP.Write(tcpip.SlicePayload(data), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - receivePackets := func(c *context.Context, sizes []int, which int, seqNum uint32) []byte { - var ret []byte - for i, size := range sizes { - p := c.GetPacket() - if i == which { - ret = p - } - checker.IPv4(t, p, - checker.PayloadLen(size+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(seqNum), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - seqNum += uint32(size) - } - return ret - } - - // Receive three packets. - sizes := []int{maxPayload, maxPayload, writeSize - 2*maxPayload} - first := receivePackets(c, sizes, 0, uint32(c.IRS)+1) - - // Send "packet too big" messages back to netstack. - const newMTU = 1200 - const newMaxPayload = newMTU - header.IPv4MinimumSize - header.TCPMinimumSize - mtu := []byte{0, 0, newMTU / 256, newMTU % 256} - c.SendICMPPacket(header.ICMPv4DstUnreachable, header.ICMPv4FragmentationNeeded, mtu, first, newMTU) - - // See retransmitted packets. None exceeding the new max. - sizes = []int{newMaxPayload, maxPayload - newMaxPayload, newMaxPayload, maxPayload - newMaxPayload, writeSize - 2*maxPayload} - receivePackets(c, sizes, -1, uint32(c.IRS)+1) -} - -func TestTCPEndpointProbe(t *testing.T) { - c := context.New(t, 1500) - defer c.Cleanup() - - invoked := make(chan struct{}) - c.Stack().AddTCPProbe(func(state stack.TCPEndpointState) { - // Validate that the endpoint ID is what we expect. - // - // We don't do an extensive validation of every field but a - // basic sanity test. - if got, want := state.ID.LocalAddress, tcpip.Address(context.StackAddr); got != want { - t.Fatalf("got LocalAddress: %q, want: %q", got, want) - } - if got, want := state.ID.LocalPort, c.Port; got != want { - t.Fatalf("got LocalPort: %d, want: %d", got, want) - } - if got, want := state.ID.RemoteAddress, tcpip.Address(context.TestAddr); got != want { - t.Fatalf("got RemoteAddress: %q, want: %q", got, want) - } - if got, want := state.ID.RemotePort, uint16(context.TestPort); got != want { - t.Fatalf("got RemotePort: %d, want: %d", got, want) - } - - invoked <- struct{}{} - }) - - c.CreateConnected(789, 30000, nil) - - data := []byte{1, 2, 3} - c.SendPacket(data, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - }) - - select { - case <-invoked: - case <-time.After(100 * time.Millisecond): - t.Fatalf("TCP Probe function was not called") - } -} - -func TestStackSetCongestionControl(t *testing.T) { - testCases := []struct { - cc tcpip.CongestionControlOption - err *tcpip.Error - }{ - {"reno", nil}, - {"cubic", nil}, - {"blahblah", tcpip.ErrNoSuchFile}, - } - - for _, tc := range testCases { - t.Run(fmt.Sprintf("SetTransportProtocolOption(.., %v)", tc.cc), func(t *testing.T) { - c := context.New(t, 1500) - defer c.Cleanup() - - s := c.Stack() - - var oldCC tcpip.CongestionControlOption - if err := s.TransportProtocolOption(tcp.ProtocolNumber, &oldCC); err != nil { - t.Fatalf("s.TransportProtocolOption(%v, %v) = %v", tcp.ProtocolNumber, &oldCC, err) - } - - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tc.cc); err != tc.err { - t.Fatalf("s.SetTransportProtocolOption(%v, %v) = %v, want %v", tcp.ProtocolNumber, tc.cc, err, tc.err) - } - - var cc tcpip.CongestionControlOption - if err := s.TransportProtocolOption(tcp.ProtocolNumber, &cc); err != nil { - t.Fatalf("s.TransportProtocolOption(%v, %v) = %v", tcp.ProtocolNumber, &cc, err) - } - - got, want := cc, oldCC - // If SetTransportProtocolOption is expected to succeed - // then the returned value for congestion control should - // match the one specified in the - // SetTransportProtocolOption call above, else it should - // be what it was before the call to - // SetTransportProtocolOption. - if tc.err == nil { - want = tc.cc - } - if got != want { - t.Fatalf("got congestion control: %v, want: %v", got, want) - } - }) - } -} - -func TestStackAvailableCongestionControl(t *testing.T) { - c := context.New(t, 1500) - defer c.Cleanup() - - s := c.Stack() - - // Query permitted congestion control algorithms. - var aCC tcpip.AvailableCongestionControlOption - if err := s.TransportProtocolOption(tcp.ProtocolNumber, &aCC); err != nil { - t.Fatalf("s.TransportProtocolOption(%v, %v) = %v", tcp.ProtocolNumber, &aCC, err) - } - if got, want := aCC, tcpip.AvailableCongestionControlOption("reno cubic"); got != want { - t.Fatalf("got tcpip.AvailableCongestionControlOption: %v, want: %v", got, want) - } -} - -func TestStackSetAvailableCongestionControl(t *testing.T) { - c := context.New(t, 1500) - defer c.Cleanup() - - s := c.Stack() - - // Setting AvailableCongestionControlOption should fail. - aCC := tcpip.AvailableCongestionControlOption("xyz") - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, &aCC); err == nil { - t.Fatalf("s.TransportProtocolOption(%v, %v) = nil, want non-nil", tcp.ProtocolNumber, &aCC) - } - - // Verify that we still get the expected list of congestion control options. - var cc tcpip.AvailableCongestionControlOption - if err := s.TransportProtocolOption(tcp.ProtocolNumber, &cc); err != nil { - t.Fatalf("s.TransportProtocolOption(%v, %v) = %v", tcp.ProtocolNumber, &cc, err) - } - if got, want := cc, tcpip.AvailableCongestionControlOption("reno cubic"); got != want { - t.Fatalf("got tcpip.AvailableCongestionControlOption: %v, want: %v", got, want) - } -} - -func TestEndpointSetCongestionControl(t *testing.T) { - testCases := []struct { - cc tcpip.CongestionControlOption - err *tcpip.Error - }{ - {"reno", nil}, - {"cubic", nil}, - {"blahblah", tcpip.ErrNoSuchFile}, - } - - for _, connected := range []bool{false, true} { - for _, tc := range testCases { - t.Run(fmt.Sprintf("SetSockOpt(.., %v) w/ connected = %v", tc.cc, connected), func(t *testing.T) { - c := context.New(t, 1500) - defer c.Cleanup() - - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - var oldCC tcpip.CongestionControlOption - if err := c.EP.GetSockOpt(&oldCC); err != nil { - t.Fatalf("c.EP.SockOpt(%v) = %v", &oldCC, err) - } - - if connected { - c.Connect(789 /* iss */, 32768 /* rcvWnd */, nil) - } - - if err := c.EP.SetSockOpt(tc.cc); err != tc.err { - t.Fatalf("c.EP.SetSockOpt(%v) = %v, want %v", tc.cc, err, tc.err) - } - - var cc tcpip.CongestionControlOption - if err := c.EP.GetSockOpt(&cc); err != nil { - t.Fatalf("c.EP.SockOpt(%v) = %v", &cc, err) - } - - got, want := cc, oldCC - // If SetSockOpt is expected to succeed then the - // returned value for congestion control should match - // the one specified in the SetSockOpt above, else it - // should be what it was before the call to SetSockOpt. - if tc.err == nil { - want = tc.cc - } - if got != want { - t.Fatalf("got congestion control: %v, want: %v", got, want) - } - }) - } - } -} - -func enableCUBIC(t *testing.T, c *context.Context) { - t.Helper() - opt := tcpip.CongestionControlOption("cubic") - if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, opt); err != nil { - t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, %v = %v", opt, err) - } -} - -func TestKeepalive(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnected(789, 30000, nil) - - c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) - c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(10 * time.Millisecond)) - c.EP.SetSockOpt(tcpip.KeepaliveCountOption(5)) - c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) - - // 5 unacked keepalives are sent. ACK each one, and check that the - // connection stays alive after 5. - for i := 0; i < 10; i++ { - b := c.GetPacket() - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)), - checker.AckNum(uint32(790)), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - - // Acknowledge the keepalive. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: c.IRS, - RcvWnd: 30000, - }) - } - - // Check that the connection is still alive. - if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) - } - - // Send some data and wait before ACKing it. Keepalives should be disabled - // during this period. - view := buffer.NewView(3) - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Write failed: %v", err) - } - - next := uint32(c.IRS) + 1 - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - // Wait for the packet to be retransmitted. Verify that no keepalives - // were sent. - checker.IPv4(t, c.GetPacket(), - checker.PayloadLen(len(view)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagPsh), - ), - ) - c.CheckNoPacket("Keepalive packet received while unACKed data is pending") - - next += uint32(len(view)) - - // Send ACK. Keepalives should start sending again. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: 790, - AckNum: seqnum.Value(next), - RcvWnd: 30000, - }) - - // Now receive 5 keepalives, but don't ACK them. The connection - // should be reset after 5. - for i := 0; i < 5; i++ { - b := c.GetPacket() - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next-1)), - checker.AckNum(uint32(790)), - checker.TCPFlags(header.TCPFlagAck), - ), - ) - } - - // The connection should be terminated after 5 unacked keepalives. - checker.IPv4(t, c.GetPacket(), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next)), - checker.AckNum(uint32(790)), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), - ), - ) - - if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) - } -} - -func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCookieInUse bool) (irs, iss seqnum.Value) { - // Send a SYN request. - irs = seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: srcPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: irs, - RcvWnd: 30000, - }) - - // Receive the SYN-ACK reply.w - b := c.GetPacket() - tcp := header.TCP(header.IPv4(b).Payload()) - iss = seqnum.Value(tcp.SequenceNumber()) - tcpCheckers := []checker.TransportChecker{ - checker.SrcPort(context.StackPort), - checker.DstPort(srcPort), - checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), - } - - if synCookieInUse { - // When cookies are in use window scaling is disabled. - tcpCheckers = append(tcpCheckers, checker.TCPSynOptions(header.TCPSynOptions{ - WS: -1, - MSS: c.MSSWithoutOptions(), - })) - } - - checker.IPv4(t, b, checker.TCP(tcpCheckers...)) - - // Send ACK. - c.SendPacket(nil, &context.Headers{ - SrcPort: srcPort, - DstPort: context.StackPort, - Flags: header.TCPFlagAck, - SeqNum: irs + 1, - AckNum: iss + 1, - RcvWnd: 30000, - }) - return irs, iss -} - -// TestListenBacklogFull tests that netstack does not complete handshakes if the -// listen backlog for the endpoint is full. -func TestListenBacklogFull(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - // Start listening. - listenBacklog := 2 - if err := c.EP.Listen(listenBacklog); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - for i := 0; i < listenBacklog; i++ { - executeHandshake(t, c, context.TestPort+uint16(i), false /*synCookieInUse */) - } - - time.Sleep(50 * time.Millisecond) - - // Now execute send one more SYN. The stack should not respond as the backlog - // is full at this point. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort + 2, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: seqnum.Value(789), - RcvWnd: 30000, - }) - c.CheckNoPacketTimeout("unexpected packet received", 50*time.Millisecond) - - // Try to accept the connections in the backlog. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - for i := 0; i < listenBacklog; i++ { - _, _, err = c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - _, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - } - - // Now verify that there are no more connections that can be accepted. - _, _, err = c.EP.Accept() - if err != tcpip.ErrWouldBlock { - select { - case <-ch: - t.Fatalf("unexpected endpoint delivered on Accept: %+v", c.EP) - case <-time.After(1 * time.Second): - } - } - - // Now a new handshake must succeed. - executeHandshake(t, c, context.TestPort+2, false /*synCookieInUse */) - - newEP, _, err := c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - newEP, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - // Now verify that the TCP socket is usable and in a connected state. - data := "Don't panic" - newEP.Write(tcpip.SlicePayload(buffer.NewViewFromBytes([]byte(data))), tcpip.WriteOptions{}) - b := c.GetPacket() - tcp := header.TCP(header.IPv4(b).Payload()) - if string(tcp.Payload()) != data { - t.Fatalf("Unexpected data: got %v, want %v", string(tcp.Payload()), data) - } -} - -func TestListenSynRcvdQueueFull(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - // Start listening. - listenBacklog := 1 - if err := c.EP.Listen(listenBacklog); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - // Send two SYN's the first one should get a SYN-ACK, the - // second one should not get any response and is dropped as - // the synRcvd count will be equal to backlog. - irs := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: seqnum.Value(789), - RcvWnd: 30000, - }) - - // Receive the SYN-ACK reply. - b := c.GetPacket() - tcp := header.TCP(header.IPv4(b).Payload()) - iss := seqnum.Value(tcp.SequenceNumber()) - tcpCheckers := []checker.TransportChecker{ - checker.SrcPort(context.StackPort), - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), - } - checker.IPv4(t, b, checker.TCP(tcpCheckers...)) - - // Now execute send one more SYN. The stack should not respond as the backlog - // is full at this point. - // - // NOTE: we did not complete the handshake for the previous one so the - // accept backlog should be empty and there should be one connection in - // synRcvd state. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort + 1, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: seqnum.Value(889), - RcvWnd: 30000, - }) - c.CheckNoPacketTimeout("unexpected packet received", 50*time.Millisecond) - - // Now complete the previous connection and verify that there is a connection - // to accept. - // Send ACK. - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagAck, - SeqNum: irs + 1, - AckNum: iss + 1, - RcvWnd: 30000, - }) - - // Try to accept the connections in the backlog. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - newEP, _, err := c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - newEP, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - // Now verify that the TCP socket is usable and in a connected state. - data := "Don't panic" - newEP.Write(tcpip.SlicePayload(buffer.NewViewFromBytes([]byte(data))), tcpip.WriteOptions{}) - pkt := c.GetPacket() - tcp = header.TCP(header.IPv4(pkt).Payload()) - if string(tcp.Payload()) != data { - t.Fatalf("Unexpected data: got %v, want %v", string(tcp.Payload()), data) - } -} - -func TestListenBacklogFullSynCookieInUse(t *testing.T) { - saved := tcp.SynRcvdCountThreshold - defer func() { - tcp.SynRcvdCountThreshold = saved - }() - tcp.SynRcvdCountThreshold = 1 - - c := context.New(t, defaultMTU) - defer c.Cleanup() - - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - // Bind to wildcard. - if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - // Start listening. - listenBacklog := 1 - portOffset := uint16(0) - if err := c.EP.Listen(listenBacklog); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - executeHandshake(t, c, context.TestPort+portOffset, false) - portOffset++ - // Wait for this to be delivered to the accept queue. - time.Sleep(50 * time.Millisecond) - - // Send a SYN request. - irs := seqnum.Value(789) - c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: irs, - RcvWnd: 30000, - }) - // The Syn should be dropped as the endpoint's backlog is full. - c.CheckNoPacketTimeout("unexpected packet received", 50*time.Millisecond) - - // Verify that there is only one acceptable connection at this point. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - _, _, err = c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - _, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - // Now verify that there are no more connections that can be accepted. - _, _, err = c.EP.Accept() - if err != tcpip.ErrWouldBlock { - select { - case <-ch: - t.Fatalf("unexpected endpoint delivered on Accept: %+v", c.EP) - case <-time.After(1 * time.Second): - } - } -} - -func TestPassiveConnectionAttemptIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - c.EP = ep - if err := ep.Bind(tcpip.FullAddress{Addr: context.StackAddr, Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - if got, want := tcp.EndpointState(ep.State()), tcp.StateBound; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - if err := c.EP.Listen(1); err != nil { - t.Fatalf("Listen failed: %v", err) - } - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateListen; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - stats := c.Stack().Stats() - want := stats.TCP.PassiveConnectionOpenings.Value() + 1 - - srcPort := uint16(context.TestPort) - executeHandshake(t, c, srcPort+1, false) - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - // Verify that there is only one acceptable connection at this point. - _, _, err = c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - _, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - - if got := stats.TCP.PassiveConnectionOpenings.Value(); got != want { - t.Errorf("got stats.TCP.PassiveConnectionOpenings.Value() = %v, want = %v", got, want) - } -} - -func TestPassiveFailedConnectionAttemptIncrement(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - stats := c.Stack().Stats() - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - c.EP = ep - if err := c.EP.Bind(tcpip.FullAddress{Addr: context.StackAddr, Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - if err := c.EP.Listen(1); err != nil { - t.Fatalf("Listen failed: %v", err) - } - - srcPort := uint16(context.TestPort) - // Now attempt a handshakes it will fill up the accept backlog. - executeHandshake(t, c, srcPort, false) - - // Give time for the final ACK to be processed as otherwise the next handshake could - // get accepted before the previous one based on goroutine scheduling. - time.Sleep(50 * time.Millisecond) - - want := stats.TCP.ListenOverflowSynDrop.Value() + 1 - - // Now we will send one more SYN and this one should get dropped - // Send a SYN request. - c.SendPacket(nil, &context.Headers{ - SrcPort: srcPort + 2, - DstPort: context.StackPort, - Flags: header.TCPFlagSyn, - SeqNum: seqnum.Value(789), - RcvWnd: 30000, - }) - - time.Sleep(50 * time.Millisecond) - if got := stats.TCP.ListenOverflowSynDrop.Value(); got != want { - t.Errorf("got stats.TCP.ListenOverflowSynDrop.Value() = %v, want = %v", got, want) - } - - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - // Now check that there is one acceptable connections. - _, _, err = c.EP.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - _, _, err = c.EP.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } -} - -func TestEndpointBindListenAcceptState(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - wq := &waiter.Queue{} - ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - - if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) - } - if got, want := tcp.EndpointState(ep.State()), tcp.StateBound; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) - } - if got, want := tcp.EndpointState(ep.State()), tcp.StateListen; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - c.PassiveConnectWithOptions(100, 5, header.TCPSynOptions{MSS: defaultIPv4MSS}) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - - aep, _, err := ep.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - aep, _, err = ep.Accept() - if err != nil { - t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for accept") - } - } - if got, want := tcp.EndpointState(aep.State()), tcp.StateEstablished; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - // Listening endpoint remains in listen state. - if got, want := tcp.EndpointState(ep.State()), tcp.StateListen; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - ep.Close() - // Give worker goroutines time to receive the close notification. - time.Sleep(1 * time.Second) - if got, want := tcp.EndpointState(ep.State()), tcp.StateClose; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - // Accepted endpoint remains open when the listen endpoint is closed. - if got, want := tcp.EndpointState(aep.State()), tcp.StateEstablished; got != want { - t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - -} - -// This test verifies that the auto tuning does not grow the receive buffer if -// the application is not reading the data actively. -func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { - const mtu = 1500 - const mss = mtu - header.IPv4MinimumSize - header.TCPMinimumSize - - c := context.New(t, mtu) - defer c.Cleanup() - - stk := c.Stack() - // Set lower limits for auto-tuning tests. This is required because the - // test stops the worker which can cause packets to be dropped because - // the segment queue holding unprocessed packets is limited to 500. - const receiveBufferSize = 80 << 10 // 80KB. - const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, receiveBufferSize, maxReceiveBufferSize}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - // Enable auto-tuning. - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - // Change the expected window scale to match the value needed for the - // maximum buffer size defined above. - c.WindowScale = uint8(tcp.FindWndScale(maxReceiveBufferSize)) - - rawEP := c.CreateConnectedWithOptions(header.TCPSynOptions{TS: true, WS: 4}) - - // NOTE: The timestamp values in the sent packets are meaningless to the - // peer so we just increment the timestamp value by 1 every batch as we - // are not really using them for anything. Send a single byte to verify - // the advertised window. - tsVal := rawEP.TSVal + 1 - - // Introduce a 25ms latency by delaying the first byte. - latency := 25 * time.Millisecond - time.Sleep(latency) - rawEP.SendPacketWithTS([]byte{1}, tsVal) - - // Verify that the ACK has the expected window. - wantRcvWnd := receiveBufferSize - wantRcvWnd = (wantRcvWnd >> uint32(c.WindowScale)) - rawEP.VerifyACKRcvWnd(uint16(wantRcvWnd - 1)) - time.Sleep(25 * time.Millisecond) - - // Allocate a large enough payload for the test. - b := make([]byte, int(receiveBufferSize)*2) - offset := 0 - payloadSize := receiveBufferSize - 1 - worker := (c.EP).(interface { - StopWork() - ResumeWork() - }) - tsVal++ - - // Stop the worker goroutine. - worker.StopWork() - start := offset - end := offset + payloadSize - packetsSent := 0 - for ; start < end; start += mss { - rawEP.SendPacketWithTS(b[start:start+mss], tsVal) - packetsSent++ - } - // Resume the worker so that it only sees the packets once all of them - // are waiting to be read. - worker.ResumeWork() - - // Since we read no bytes the window should goto zero till the - // application reads some of the data. - // Discard all intermediate acks except the last one. - if packetsSent > 100 { - for i := 0; i < (packetsSent / 100); i++ { - _ = c.GetPacket() - } - } - rawEP.VerifyACKRcvWnd(0) - - time.Sleep(25 * time.Millisecond) - // Verify that sending more data when window is closed is dropped and - // not acked. - rawEP.SendPacketWithTS(b[start:start+mss], tsVal) - - // Verify that the stack sends us back an ACK with the sequence number - // of the last packet sent indicating it was dropped. - p := c.GetPacket() - checker.IPv4(t, p, checker.TCP( - checker.AckNum(uint32(rawEP.NextSeqNum)-uint32(mss)), - checker.Window(0), - )) - - // Now read all the data from the endpoint and verify that advertised - // window increases to the full available buffer size. - for { - _, _, err := c.EP.Read(nil) - if err == tcpip.ErrWouldBlock { - break - } - } - - // Verify that we receive a non-zero window update ACK. When running - // under thread santizer this test can end up sending more than 1 - // ack, 1 for the non-zero window - p = c.GetPacket() - checker.IPv4(t, p, checker.TCP( - checker.AckNum(uint32(rawEP.NextSeqNum)-uint32(mss)), - func(t *testing.T, h header.Transport) { - tcp, ok := h.(header.TCP) - if !ok { - return - } - if w := tcp.WindowSize(); w == 0 || w > uint16(wantRcvWnd) { - t.Errorf("expected a non-zero window: got %d, want <= wantRcvWnd", w, wantRcvWnd) - } - }, - )) -} - -// This test verifies that the auto tuning does not grow the receive buffer if -// the application is not reading the data actively. -func TestReceiveBufferAutoTuning(t *testing.T) { - const mtu = 1500 - const mss = mtu - header.IPv4MinimumSize - header.TCPMinimumSize - - c := context.New(t, mtu) - defer c.Cleanup() - - // Enable Auto-tuning. - stk := c.Stack() - // Set lower limits for auto-tuning tests. This is required because the - // test stops the worker which can cause packets to be dropped because - // the segment queue holding unprocessed packets is limited to 500. - const receiveBufferSize = 80 << 10 // 80KB. - const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, receiveBufferSize, maxReceiveBufferSize}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - // Enable auto-tuning. - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - // Change the expected window scale to match the value needed for the - // maximum buffer size used by stack. - c.WindowScale = uint8(tcp.FindWndScale(maxReceiveBufferSize)) - - rawEP := c.CreateConnectedWithOptions(header.TCPSynOptions{TS: true, WS: 4}) - - wantRcvWnd := receiveBufferSize - scaleRcvWnd := func(rcvWnd int) uint16 { - return uint16(rcvWnd >> uint16(c.WindowScale)) - } - // Allocate a large array to send to the endpoint. - b := make([]byte, receiveBufferSize*48) - - // In every iteration we will send double the number of bytes sent in - // the previous iteration and read the same from the app. The received - // window should grow by at least 2x of bytes read by the app in every - // RTT. - offset := 0 - payloadSize := receiveBufferSize / 8 - worker := (c.EP).(interface { - StopWork() - ResumeWork() - }) - tsVal := rawEP.TSVal - // We are going to do our own computation of what the moderated receive - // buffer should be based on sent/copied data per RTT and verify that - // the advertised window by the stack matches our calculations. - prevCopied := 0 - done := false - latency := 1 * time.Millisecond - for i := 0; !done; i++ { - tsVal++ - - // Stop the worker goroutine. - worker.StopWork() - start := offset - end := offset + payloadSize - totalSent := 0 - packetsSent := 0 - for ; start < end; start += mss { - rawEP.SendPacketWithTS(b[start:start+mss], tsVal) - totalSent += mss - packetsSent++ - } - // Resume it so that it only sees the packets once all of them - // are waiting to be read. - worker.ResumeWork() - - // Give 1ms for the worker to process the packets. - time.Sleep(1 * time.Millisecond) - - // Verify that the advertised window on the ACK is reduced by - // the total bytes sent. - expectedWnd := wantRcvWnd - totalSent - if packetsSent > 100 { - for i := 0; i < (packetsSent / 100); i++ { - _ = c.GetPacket() - } - } - rawEP.VerifyACKRcvWnd(scaleRcvWnd(expectedWnd)) - - // Now read all the data from the endpoint and invoke the - // moderation API to allow for receive buffer auto-tuning - // to happen before we measure the new window. - totalCopied := 0 - for { - b, _, err := c.EP.Read(nil) - if err == tcpip.ErrWouldBlock { - break - } - totalCopied += len(b) - } - - // Invoke the moderation API. This is required for auto-tuning - // to happen. This method is normally expected to be invoked - // from a higher layer than tcpip.Endpoint. So we simulate - // copying to user-space by invoking it explicitly here. - c.EP.ModerateRecvBuf(totalCopied) - - // Now send a keep-alive packet to trigger an ACK so that we can - // measure the new window. - rawEP.NextSeqNum-- - rawEP.SendPacketWithTS(nil, tsVal) - rawEP.NextSeqNum++ - - if i == 0 { - // In the first iteration the receiver based RTT is not - // yet known as a result the moderation code should not - // increase the advertised window. - rawEP.VerifyACKRcvWnd(scaleRcvWnd(wantRcvWnd)) - prevCopied = totalCopied - } else { - rttCopied := totalCopied - if i == 1 { - // The moderation code accumulates copied bytes till - // RTT is established. So add in the bytes sent in - // the first iteration to the total bytes for this - // RTT. - rttCopied += prevCopied - // Now reset it to the initial value used by the - // auto tuning logic. - prevCopied = tcp.InitialCwnd * mss * 2 - } - newWnd := rttCopied<<1 + 16*mss - grow := (newWnd * (rttCopied - prevCopied)) / prevCopied - newWnd += (grow << 1) - if newWnd > maxReceiveBufferSize { - newWnd = maxReceiveBufferSize - done = true - } - rawEP.VerifyACKRcvWnd(scaleRcvWnd(newWnd)) - wantRcvWnd = newWnd - prevCopied = rttCopied - // Increase the latency after first two iterations to - // establish a low RTT value in the receiver since it - // only tracks the lowest value. This ensures that when - // ModerateRcvBuf is called the elapsed time is always > - // rtt. Without this the test is flaky due to delays due - // to scheduling/wakeup etc. - latency += 50 * time.Millisecond - } - time.Sleep(latency) - offset += payloadSize - payloadSize *= 2 - } -} diff --git a/pkg/tcpip/transport/tcp/tcp_timestamp_test.go b/pkg/tcpip/transport/tcp/tcp_timestamp_test.go deleted file mode 100644 index a641e953d..000000000 --- a/pkg/tcpip/transport/tcp/tcp_timestamp_test.go +++ /dev/null @@ -1,295 +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 tcp_test - -import ( - "bytes" - "math/rand" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/checker" - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context" - "gvisor.dev/gvisor/pkg/waiter" -) - -// createConnectedWithTimestampOption creates and connects c.ep with the -// timestamp option enabled. -func createConnectedWithTimestampOption(c *context.Context) *context.RawEndpoint { - return c.CreateConnectedWithOptions(header.TCPSynOptions{TS: true, TSVal: 1}) -} - -// TestTimeStampEnabledConnect tests that netstack sends the timestamp option on -// an active connect and sets the TS Echo Reply fields correctly when the -// SYN-ACK also indicates support for the TS option and provides a TSVal. -func TestTimeStampEnabledConnect(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - rep := createConnectedWithTimestampOption(c) - - // Register for read and validate that we have data to read. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - // The following tests ensure that TS option once enabled behaves - // correctly as described in - // https://tools.ietf.org/html/rfc7323#section-4.3. - // - // We are not testing delayed ACKs here, but we do test out of order - // packet delivery and filling the sequence number hole created due to - // the out of order packet. - // - // The test also verifies that the sequence numbers and timestamps are - // as expected. - data := []byte{1, 2, 3} - - // First we increment tsVal by a small amount. - tsVal := rep.TSVal + 100 - rep.SendPacketWithTS(data, tsVal) - rep.VerifyACKWithTS(tsVal) - - // Next we send an out of order packet. - rep.NextSeqNum += 3 - tsVal += 200 - rep.SendPacketWithTS(data, tsVal) - - // The ACK should contain the original sequenceNumber and an older TS. - rep.NextSeqNum -= 6 - rep.VerifyACKWithTS(tsVal - 200) - - // Next we fill the hole and the returned ACK should contain the - // cumulative sequence number acking all data sent till now and have the - // latest timestamp sent below in its TSEcr field. - tsVal -= 100 - rep.SendPacketWithTS(data, tsVal) - rep.NextSeqNum += 3 - rep.VerifyACKWithTS(tsVal) - - // Increment tsVal by a large value that doesn't result in a wrap around. - tsVal += 0x7fffffff - rep.SendPacketWithTS(data, tsVal) - rep.VerifyACKWithTS(tsVal) - - // Increment tsVal again by a large value which should cause the - // timestamp value to wrap around. The returned ACK should contain the - // wrapped around timestamp in its tsEcr field and not the tsVal from - // the previous packet sent above. - tsVal += 0x7fffffff - rep.SendPacketWithTS(data, tsVal) - rep.VerifyACKWithTS(tsVal) - - select { - case <-ch: - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // There should be 5 views to read and each of them should - // contain the same data. - for i := 0; i < 5; i++ { - got, _, err := c.EP.Read(nil) - if err != nil { - t.Fatalf("Unexpected error from Read: %v", err) - } - if want := data; bytes.Compare(got, want) != 0 { - t.Fatalf("Data is different: got: %v, want: %v", got, want) - } - } -} - -// TestTimeStampDisabledConnect tests that netstack sends timestamp option on an -// active connect but if the SYN-ACK doesn't specify the TS option then -// timestamp option is not enabled and future packets do not contain a -// timestamp. -func TestTimeStampDisabledConnect(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() - - c.CreateConnectedWithOptions(header.TCPSynOptions{}) -} - -func timeStampEnabledAccept(t *testing.T, cookieEnabled bool, wndScale int, wndSize uint16) { - savedSynCountThreshold := tcp.SynRcvdCountThreshold - defer func() { - tcp.SynRcvdCountThreshold = savedSynCountThreshold - }() - - if cookieEnabled { - tcp.SynRcvdCountThreshold = 0 - } - c := context.New(t, defaultMTU) - defer c.Cleanup() - - t.Logf("Test w/ CookieEnabled = %v", cookieEnabled) - tsVal := rand.Uint32() - c.AcceptWithOptions(wndScale, header.TCPSynOptions{MSS: defaultIPv4MSS, TS: true, TSVal: tsVal}) - - // Now send some data and validate that timestamp is echoed correctly in the ACK. - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Unexpected error from Write: %v", err) - } - - // Check that data is received and that the timestamp option TSEcr field - // matches the expected value. - b := c.GetPacket() - checker.IPv4(t, b, - // Add 12 bytes for the timestamp option + 2 NOPs to align at 4 - // byte boundary. - checker.PayloadLen(len(data)+header.TCPMinimumSize+12), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(wndSize), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - checker.TCPTimestampChecker(true, 0, tsVal+1), - ), - ) -} - -// TestTimeStampEnabledAccept tests that if the SYN on a passive connect -// specifies the Timestamp option then the Timestamp option is sent on a SYN-ACK -// and echoes the tsVal field of the original SYN in the tcEcr field of the -// SYN-ACK. We cover the cases where SYN cookies are enabled/disabled and verify -// that Timestamp option is enabled in both cases if requested in the original -// SYN. -func TestTimeStampEnabledAccept(t *testing.T) { - testCases := []struct { - cookieEnabled bool - wndScale int - wndSize uint16 - }{ - {true, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, 5, 0x8000}, // DefaultReceiveBufferSize is 1MB >> 5. - } - for _, tc := range testCases { - timeStampEnabledAccept(t, tc.cookieEnabled, tc.wndScale, tc.wndSize) - } -} - -func timeStampDisabledAccept(t *testing.T, cookieEnabled bool, wndScale int, wndSize uint16) { - savedSynCountThreshold := tcp.SynRcvdCountThreshold - defer func() { - tcp.SynRcvdCountThreshold = savedSynCountThreshold - }() - if cookieEnabled { - tcp.SynRcvdCountThreshold = 0 - } - - c := context.New(t, defaultMTU) - defer c.Cleanup() - - t.Logf("Test w/ CookieEnabled = %v", cookieEnabled) - c.AcceptWithOptions(wndScale, header.TCPSynOptions{MSS: defaultIPv4MSS}) - - // Now send some data with the accepted connection endpoint and validate - // that no timestamp option is sent in the TCP segment. - data := []byte{1, 2, 3} - view := buffer.NewView(len(data)) - copy(view, data) - - if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { - t.Fatalf("Unexpected error from Write: %v", err) - } - - // Check that data is received and that the timestamp option is disabled - // when SYN cookies are enabled/disabled. - b := c.GetPacket() - checker.IPv4(t, b, - checker.PayloadLen(len(data)+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(wndSize), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - checker.TCPTimestampChecker(false, 0, 0), - ), - ) -} - -// TestTimeStampDisabledAccept tests that Timestamp option is not used when the -// peer doesn't advertise it and connection is established with Accept(). -func TestTimeStampDisabledAccept(t *testing.T) { - testCases := []struct { - cookieEnabled bool - wndScale int - wndSize uint16 - }{ - {true, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, 5, 0x8000}, // DefaultReceiveBufferSize is 1MB >> 5. - } - for _, tc := range testCases { - timeStampDisabledAccept(t, tc.cookieEnabled, tc.wndScale, tc.wndSize) - } -} - -func TestSendGreaterThanMTUWithOptions(t *testing.T) { - const maxPayload = 100 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - createConnectedWithTimestampOption(c) - testBrokenUpWrite(t, c, maxPayload) -} - -func TestSegmentNotDroppedWhenTimestampMissing(t *testing.T) { - const maxPayload = 100 - c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) - defer c.Cleanup() - - rep := createConnectedWithTimestampOption(c) - - // Register for read. - we, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&we, waiter.EventIn) - defer c.WQ.EventUnregister(&we) - - droppedPacketsStat := c.Stack().Stats().DroppedPackets - droppedPackets := droppedPacketsStat.Value() - data := []byte{1, 2, 3} - // Send a packet with no TCP options/timestamp. - rep.SendPacket(data, nil) - - select { - case <-ch: - case <-time.After(1 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") - } - - // Assert that DroppedPackets was not incremented. - if got, want := droppedPacketsStat.Value(), droppedPackets; got != want { - t.Fatalf("incorrect number of dropped packets, got: %v, want: %v", got, want) - } - - // Issue a read and we should data. - got, _, err := c.EP.Read(nil) - if err != nil { - t.Fatalf("Unexpected error from Read: %v", err) - } - if want := data; bytes.Compare(got, want) != 0 { - t.Fatalf("Data is different: got: %v, want: %v", got, want) - } -} diff --git a/pkg/tcpip/transport/tcp/testing/context/BUILD b/pkg/tcpip/transport/tcp/testing/context/BUILD deleted file mode 100644 index 19b0d31c5..000000000 --- a/pkg/tcpip/transport/tcp/testing/context/BUILD +++ /dev/null @@ -1,27 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "context", - testonly = 1, - srcs = ["context.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context", - visibility = [ - "//:sandbox", - ], - deps = [ - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/checker", - "//pkg/tcpip/header", - "//pkg/tcpip/link/channel", - "//pkg/tcpip/link/sniffer", - "//pkg/tcpip/network/ipv4", - "//pkg/tcpip/network/ipv6", - "//pkg/tcpip/seqnum", - "//pkg/tcpip/stack", - "//pkg/tcpip/transport/tcp", - "//pkg/waiter", - ], -) diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go deleted file mode 100644 index 16783e716..000000000 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ /dev/null @@ -1,1046 +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 context provides a test context for use in tcp tests. It also -// provides helper methods to assert/check certain behaviours. -package context - -import ( - "bytes" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "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/sniffer" - "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" - "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" - "gvisor.dev/gvisor/pkg/tcpip/seqnum" - "gvisor.dev/gvisor/pkg/tcpip/stack" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" - "gvisor.dev/gvisor/pkg/waiter" -) - -const ( - // StackAddr is the IPv4 address assigned to the stack. - StackAddr = "\x0a\x00\x00\x01" - - // StackPort is used as the listening port in tests for passive - // connects. - StackPort = 1234 - - // TestAddr is the source address for packets sent to the stack via the - // link layer endpoint. - TestAddr = "\x0a\x00\x00\x02" - - // TestPort is the TCP port used for packets sent to the stack - // via the link layer endpoint. - TestPort = 4096 - - // StackV6Addr is the IPv6 address assigned to the stack. - StackV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" - - // TestV6Addr is the source address for packets sent to the stack via - // the link layer endpoint. - TestV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" - - // StackV4MappedAddr is StackAddr as a mapped v6 address. - StackV4MappedAddr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff" + StackAddr - - // TestV4MappedAddr is TestAddr as a mapped v6 address. - TestV4MappedAddr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff" + TestAddr - - // V4MappedWildcardAddr is the mapped v6 representation of 0.0.0.0. - V4MappedWildcardAddr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00" - - // testInitialSequenceNumber is the initial sequence number sent in packets that - // are sent in response to a SYN or in the initial SYN sent to the stack. - testInitialSequenceNumber = 789 -) - -// Headers is used to represent the TCP header fields when building a -// new packet. -type Headers struct { - // SrcPort holds the src port value to be used in the packet. - SrcPort uint16 - - // DstPort holds the destination port value to be used in the packet. - DstPort uint16 - - // SeqNum is the value of the sequence number field in the TCP header. - SeqNum seqnum.Value - - // AckNum represents the acknowledgement number field in the TCP header. - AckNum seqnum.Value - - // Flags are the TCP flags in the TCP header. - Flags int - - // RcvWnd is the window to be advertised in the ReceiveWindow field of - // the TCP header. - RcvWnd seqnum.Size - - // TCPOpts holds the options to be sent in the option field of the TCP - // header. - TCPOpts []byte -} - -// Context provides an initialized Network stack and a link layer endpoint -// for use in TCP tests. -type Context struct { - t *testing.T - linkEP *channel.Endpoint - s *stack.Stack - - // IRS holds the initial sequence number in the SYN sent by endpoint in - // case of an active connect or the sequence number sent by the endpoint - // in the SYN-ACK sent in response to a SYN when listening in passive - // mode. - IRS seqnum.Value - - // Port holds the port bound by EP below in case of an active connect or - // the listening port number in case of a passive connect. - Port uint16 - - // EP is the test endpoint in the stack owned by this context. This endpoint - // is used in various tests to either initiate an active connect or is used - // as a passive listening endpoint to accept inbound connections. - EP tcpip.Endpoint - - // Wq is the wait queue associated with EP and is used to block for events - // on EP. - WQ waiter.Queue - - // TimeStampEnabled is true if ep is connected with the timestamp option - // enabled. - TimeStampEnabled bool - - // WindowScale is the expected window scale in SYN packets sent by - // the stack. - WindowScale uint8 -} - -// New allocates and initializes a test context containing a new -// stack and a link-layer endpoint. -func New(t *testing.T, mtu uint32) *Context { - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) - - // Allow minimum send/receive buffer sizes to be 1 during tests. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{1, tcp.DefaultSendBufferSize, 10 * tcp.DefaultSendBufferSize}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, tcp.DefaultReceiveBufferSize, 10 * tcp.DefaultReceiveBufferSize}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) - } - - // Some of the congestion control tests send up to 640 packets, we so - // set the channel size to 1000. - ep := channel.New(1000, mtu, "") - wep := stack.LinkEndpoint(ep) - if testing.Verbose() { - wep = sniffer.New(ep) - } - if err := s.CreateNIC(1, wep); err != nil { - t.Fatalf("CreateNIC failed: %v", err) - } - - if err := s.AddAddress(1, ipv4.ProtocolNumber, StackAddr); err != nil { - t.Fatalf("AddAddress failed: %v", err) - } - - if err := s.AddAddress(1, ipv6.ProtocolNumber, StackV6Addr); err != nil { - t.Fatalf("AddAddress failed: %v", err) - } - - s.SetRouteTable([]tcpip.Route{ - { - Destination: header.IPv4EmptySubnet, - NIC: 1, - }, - { - Destination: header.IPv6EmptySubnet, - NIC: 1, - }, - }) - - return &Context{ - t: t, - s: s, - linkEP: ep, - WindowScale: uint8(tcp.FindWndScale(tcp.DefaultReceiveBufferSize)), - } -} - -// Cleanup closes the context endpoint if required. -func (c *Context) Cleanup() { - if c.EP != nil { - c.EP.Close() - } -} - -// Stack returns a reference to the stack in the Context. -func (c *Context) Stack() *stack.Stack { - return c.s -} - -// CheckNoPacketTimeout verifies that no packet is received during the time -// specified by wait. -func (c *Context) CheckNoPacketTimeout(errMsg string, wait time.Duration) { - c.t.Helper() - - select { - case <-c.linkEP.C: - c.t.Fatal(errMsg) - - case <-time.After(wait): - } -} - -// CheckNoPacket verifies that no packet is received for 1 second. -func (c *Context) CheckNoPacket(errMsg string) { - c.CheckNoPacketTimeout(errMsg, 1*time.Second) -} - -// GetPacket reads a packet from the link layer endpoint and verifies -// that it is an IPv4 packet with the expected source and destination -// addresses. It will fail with an error if no packet is received for -// 2 seconds. -func (c *Context) GetPacket() []byte { - select { - case p := <-c.linkEP.C: - if p.Proto != ipv4.ProtocolNumber { - c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) - } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) - - if p.GSO != nil && p.GSO.L3HdrLen != header.IPv4MinimumSize { - c.t.Errorf("L3HdrLen %v (expected %v)", p.GSO.L3HdrLen, header.IPv4MinimumSize) - } - - checker.IPv4(c.t, b, checker.SrcAddr(StackAddr), checker.DstAddr(TestAddr)) - return b - - case <-time.After(2 * time.Second): - c.t.Fatalf("Packet wasn't written out") - } - - return nil -} - -// GetPacketNonBlocking reads a packet from the link layer endpoint -// and verifies that it is an IPv4 packet with the expected source -// and destination address. If no packet is available it will return -// nil immediately. -func (c *Context) GetPacketNonBlocking() []byte { - select { - case p := <-c.linkEP.C: - if p.Proto != ipv4.ProtocolNumber { - c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) - } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) - - checker.IPv4(c.t, b, checker.SrcAddr(StackAddr), checker.DstAddr(TestAddr)) - return b - default: - return nil - } -} - -// SendICMPPacket builds and sends an ICMPv4 packet via the link layer endpoint. -func (c *Context) SendICMPPacket(typ header.ICMPv4Type, code uint8, p1, p2 []byte, maxTotalSize int) { - // Allocate a buffer data and headers. - buf := buffer.NewView(header.IPv4MinimumSize + header.ICMPv4PayloadOffset + len(p2)) - if len(buf) > maxTotalSize { - buf = buf[:maxTotalSize] - } - - ip := header.IPv4(buf) - ip.Encode(&header.IPv4Fields{ - IHL: header.IPv4MinimumSize, - TotalLength: uint16(len(buf)), - TTL: 65, - Protocol: uint8(header.ICMPv4ProtocolNumber), - SrcAddr: TestAddr, - DstAddr: StackAddr, - }) - ip.SetChecksum(^ip.CalculateChecksum()) - - icmp := header.ICMPv4(buf[header.IPv4MinimumSize:]) - icmp.SetType(typ) - icmp.SetCode(code) - const icmpv4VariableHeaderOffset = 4 - copy(icmp[icmpv4VariableHeaderOffset:], p1) - copy(icmp[header.ICMPv4PayloadOffset:], p2) - - // Inject packet. - c.linkEP.Inject(ipv4.ProtocolNumber, buf.ToVectorisedView()) -} - -// BuildSegment builds a TCP segment based on the given Headers and payload. -func (c *Context) BuildSegment(payload []byte, h *Headers) buffer.VectorisedView { - // Allocate a buffer for data and headers. - buf := buffer.NewView(header.TCPMinimumSize + header.IPv4MinimumSize + len(h.TCPOpts) + len(payload)) - copy(buf[len(buf)-len(payload):], payload) - copy(buf[len(buf)-len(payload)-len(h.TCPOpts):], h.TCPOpts) - - // Initialize the IP header. - ip := header.IPv4(buf) - ip.Encode(&header.IPv4Fields{ - IHL: header.IPv4MinimumSize, - TotalLength: uint16(len(buf)), - TTL: 65, - Protocol: uint8(tcp.ProtocolNumber), - SrcAddr: TestAddr, - DstAddr: StackAddr, - }) - ip.SetChecksum(^ip.CalculateChecksum()) - - // Initialize the TCP header. - t := header.TCP(buf[header.IPv4MinimumSize:]) - t.Encode(&header.TCPFields{ - SrcPort: h.SrcPort, - DstPort: h.DstPort, - SeqNum: uint32(h.SeqNum), - AckNum: uint32(h.AckNum), - DataOffset: uint8(header.TCPMinimumSize + len(h.TCPOpts)), - Flags: uint8(h.Flags), - WindowSize: uint16(h.RcvWnd), - }) - - // Calculate the TCP pseudo-header checksum. - xsum := header.PseudoHeaderChecksum(tcp.ProtocolNumber, TestAddr, StackAddr, uint16(len(t))) - - // Calculate the TCP checksum and set it. - xsum = header.Checksum(payload, xsum) - t.SetChecksum(^t.CalculateChecksum(xsum)) - - // Inject packet. - return buf.ToVectorisedView() -} - -// SendSegment sends a TCP segment that has already been built and written to a -// buffer.VectorisedView. -func (c *Context) SendSegment(s buffer.VectorisedView) { - c.linkEP.Inject(ipv4.ProtocolNumber, s) -} - -// SendPacket builds and sends a TCP segment(with the provided payload & TCP -// headers) in an IPv4 packet via the link layer endpoint. -func (c *Context) SendPacket(payload []byte, h *Headers) { - c.linkEP.Inject(ipv4.ProtocolNumber, c.BuildSegment(payload, h)) -} - -// SendAck sends an ACK packet. -func (c *Context) SendAck(seq seqnum.Value, bytesReceived int) { - c.SendAckWithSACK(seq, bytesReceived, nil) -} - -// SendAckWithSACK sends an ACK packet which includes the sackBlocks specified. -func (c *Context) SendAckWithSACK(seq seqnum.Value, bytesReceived int, sackBlocks []header.SACKBlock) { - options := make([]byte, 40) - offset := 0 - if len(sackBlocks) > 0 { - offset += header.EncodeNOP(options[offset:]) - offset += header.EncodeNOP(options[offset:]) - offset += header.EncodeSACKBlocks(sackBlocks, options[offset:]) - } - - c.SendPacket(nil, &Headers{ - SrcPort: TestPort, - DstPort: c.Port, - Flags: header.TCPFlagAck, - SeqNum: seq, - AckNum: c.IRS.Add(1 + seqnum.Size(bytesReceived)), - RcvWnd: 30000, - TCPOpts: options[:offset], - }) -} - -// ReceiveAndCheckPacket reads a packet from the link layer endpoint and -// verifies that the packet packet payload of packet matches the slice -// of data indicated by offset & size. -func (c *Context) ReceiveAndCheckPacket(data []byte, offset, size int) { - c.ReceiveAndCheckPacketWithOptions(data, offset, size, 0) -} - -// ReceiveAndCheckPacketWithOptions reads a packet from the link layer endpoint -// and verifies that the packet packet payload of packet matches the slice of -// data indicated by offset & size and skips optlen bytes in addition to the IP -// TCP headers when comparing the data. -func (c *Context) ReceiveAndCheckPacketWithOptions(data []byte, offset, size, optlen int) { - b := c.GetPacket() - checker.IPv4(c.t, b, - checker.PayloadLen(size+header.TCPMinimumSize+optlen), - checker.TCP( - checker.DstPort(TestPort), - checker.SeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), - checker.AckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - pdata := data[offset:][:size] - if p := b[header.IPv4MinimumSize+header.TCPMinimumSize+optlen:]; bytes.Compare(pdata, p) != 0 { - c.t.Fatalf("Data is different: expected %v, got %v", pdata, p) - } -} - -// ReceiveNonBlockingAndCheckPacket reads a packet from the link layer endpoint -// and verifies that the packet packet payload of packet matches the slice of -// data indicated by offset & size. It returns true if a packet was received and -// processed. -func (c *Context) ReceiveNonBlockingAndCheckPacket(data []byte, offset, size int) bool { - b := c.GetPacketNonBlocking() - if b == nil { - return false - } - checker.IPv4(c.t, b, - checker.PayloadLen(size+header.TCPMinimumSize), - checker.TCP( - checker.DstPort(TestPort), - checker.SeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), - checker.AckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), - checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), - ), - ) - - pdata := data[offset:][:size] - if p := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; bytes.Compare(pdata, p) != 0 { - c.t.Fatalf("Data is different: expected %v, got %v", pdata, p) - } - return true -} - -// CreateV6Endpoint creates and initializes c.ep as a IPv6 Endpoint. If v6Only -// is true then it sets the IP_V6ONLY option on the socket to make it a IPv6 -// only endpoint instead of a default dual stack socket. -func (c *Context) CreateV6Endpoint(v6only bool) { - var err *tcpip.Error - c.EP, err = c.s.NewEndpoint(tcp.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 failed: %v", err) - } -} - -// GetV6Packet reads a single packet from the link layer endpoint of the context -// and asserts that it is an IPv6 Packet with the expected src/dest addresses. -func (c *Context) GetV6Packet() []byte { - select { - case p := <-c.linkEP.C: - if p.Proto != ipv6.ProtocolNumber { - c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv6.ProtocolNumber) - } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) - - checker.IPv6(c.t, b, checker.SrcAddr(StackV6Addr), checker.DstAddr(TestV6Addr)) - return b - - case <-time.After(2 * time.Second): - c.t.Fatalf("Packet wasn't written out") - } - - return nil -} - -// SendV6Packet builds and sends an IPv6 Packet via the link layer endpoint of -// the context. -func (c *Context) SendV6Packet(payload []byte, h *Headers) { - // Allocate a buffer for data and headers. - buf := buffer.NewView(header.TCPMinimumSize + 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.TCPMinimumSize + len(payload)), - NextHeader: uint8(tcp.ProtocolNumber), - HopLimit: 65, - SrcAddr: TestV6Addr, - DstAddr: StackV6Addr, - }) - - // Initialize the TCP header. - t := header.TCP(buf[header.IPv6MinimumSize:]) - t.Encode(&header.TCPFields{ - SrcPort: h.SrcPort, - DstPort: h.DstPort, - SeqNum: uint32(h.SeqNum), - AckNum: uint32(h.AckNum), - DataOffset: header.TCPMinimumSize, - Flags: uint8(h.Flags), - WindowSize: uint16(h.RcvWnd), - }) - - // Calculate the TCP pseudo-header checksum. - xsum := header.PseudoHeaderChecksum(tcp.ProtocolNumber, TestV6Addr, StackV6Addr, uint16(len(t))) - - // Calculate the TCP checksum and set it. - xsum = header.Checksum(payload, xsum) - t.SetChecksum(^t.CalculateChecksum(xsum)) - - // Inject packet. - c.linkEP.Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) -} - -// CreateConnected creates a connected TCP endpoint. -func (c *Context) CreateConnected(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf *tcpip.ReceiveBufferSizeOption) { - c.CreateConnectedWithRawOptions(iss, rcvWnd, epRcvBuf, nil) -} - -// Connect performs the 3-way handshake for c.EP with the provided Initial -// Sequence Number (iss) and receive window(rcvWnd) and any options if -// specified. -// -// It also sets the receive buffer for the endpoint to the specified -// value in epRcvBuf. -// -// PreCondition: c.EP must already be created. -func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) { - // Start connection attempt. - waitEntry, notifyCh := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&waitEntry, waiter.EventOut) - defer c.WQ.EventUnregister(&waitEntry) - - if err := c.EP.Connect(tcpip.FullAddress{Addr: TestAddr, Port: TestPort}); err != tcpip.ErrConnectStarted { - c.t.Fatalf("Unexpected return value from Connect: %v", err) - } - - // Receive SYN packet. - b := c.GetPacket() - checker.IPv4(c.t, b, - checker.TCP( - checker.DstPort(TestPort), - checker.TCPFlags(header.TCPFlagSyn), - ), - ) - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { - c.t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) - } - - tcpHdr := header.TCP(header.IPv4(b).Payload()) - c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) - - c.SendPacket(nil, &Headers{ - SrcPort: tcpHdr.DestinationPort(), - DstPort: tcpHdr.SourcePort(), - Flags: header.TCPFlagSyn | header.TCPFlagAck, - SeqNum: iss, - AckNum: c.IRS.Add(1), - RcvWnd: rcvWnd, - TCPOpts: options, - }) - - // Receive ACK packet. - checker.IPv4(c.t, c.GetPacket(), - checker.TCP( - checker.DstPort(TestPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), - ), - ) - - // Wait for connection to be established. - select { - case <-notifyCh: - if err := c.EP.GetSockOpt(tcpip.ErrorOption{}); err != nil { - c.t.Fatalf("Unexpected error when connecting: %v", err) - } - case <-time.After(1 * time.Second): - c.t.Fatalf("Timed out waiting for connection") - } - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateEstablished; got != want { - c.t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) - } - - 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 *tcpip.ReceiveBufferSizeOption, options []byte) { - // Create TCP endpoint. - var err *tcpip.Error - c.EP, err = c.s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - c.t.Fatalf("NewEndpoint failed: %v", err) - } - - if epRcvBuf != nil { - if err := c.EP.SetSockOpt(*epRcvBuf); err != nil { - c.t.Fatalf("SetSockOpt failed failed: %v", err) - } - } - c.Connect(iss, rcvWnd, options) -} - -// RawEndpoint is just a small wrapper around a TCP endpoint's state to make -// sending data and ACK packets easy while being able to manipulate the sequence -// numbers and timestamp values as needed. -type RawEndpoint struct { - C *Context - SrcPort uint16 - DstPort uint16 - Flags int - NextSeqNum seqnum.Value - AckNum seqnum.Value - WndSize seqnum.Size - RecentTS uint32 // Stores the latest timestamp to echo back. - TSVal uint32 // TSVal stores the last timestamp sent by this endpoint. - - // SackPermitted is true if SACKPermitted option was negotiated for this endpoint. - SACKPermitted bool -} - -// SendPacketWithTS embeds the provided tsVal in the Timestamp option -// for the packet to be sent out. -func (r *RawEndpoint) SendPacketWithTS(payload []byte, tsVal uint32) { - r.TSVal = tsVal - tsOpt := [12]byte{header.TCPOptionNOP, header.TCPOptionNOP} - header.EncodeTSOption(r.TSVal, r.RecentTS, tsOpt[2:]) - r.SendPacket(payload, tsOpt[:]) -} - -// SendPacket is a small wrapper function to build and send packets. -func (r *RawEndpoint) SendPacket(payload []byte, opts []byte) { - packetHeaders := &Headers{ - SrcPort: r.SrcPort, - DstPort: r.DstPort, - Flags: r.Flags, - SeqNum: r.NextSeqNum, - AckNum: r.AckNum, - RcvWnd: r.WndSize, - TCPOpts: opts, - } - r.C.SendPacket(payload, packetHeaders) - r.NextSeqNum = r.NextSeqNum.Add(seqnum.Size(len(payload))) -} - -// VerifyACKWithTS verifies that the tsEcr field in the ack matches the provided -// tsVal. -func (r *RawEndpoint) VerifyACKWithTS(tsVal uint32) { - // Read ACK and verify that tsEcr of ACK packet is [1,2,3,4] - ackPacket := r.C.GetPacket() - checker.IPv4(r.C.t, ackPacket, - checker.TCP( - checker.DstPort(r.SrcPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), - checker.TCPTimestampChecker(true, 0, tsVal), - ), - ) - // Store the parsed TSVal from the ack as recentTS. - tcpSeg := header.TCP(header.IPv4(ackPacket).Payload()) - opts := tcpSeg.ParsedOptions() - r.RecentTS = opts.TSVal -} - -// VerifyACKRcvWnd verifies that the window advertised by the incoming ACK -// matches the provided rcvWnd. -func (r *RawEndpoint) VerifyACKRcvWnd(rcvWnd uint16) { - ackPacket := r.C.GetPacket() - checker.IPv4(r.C.t, ackPacket, - checker.TCP( - checker.DstPort(r.SrcPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), - checker.Window(rcvWnd), - ), - ) -} - -// VerifyACKNoSACK verifies that the ACK does not contain a SACK block. -func (r *RawEndpoint) VerifyACKNoSACK() { - r.VerifyACKHasSACK(nil) -} - -// VerifyACKHasSACK verifies that the ACK contains the specified SACKBlocks. -func (r *RawEndpoint) VerifyACKHasSACK(sackBlocks []header.SACKBlock) { - // Read ACK and verify that the TCP options in the segment do - // not contain a SACK block. - ackPacket := r.C.GetPacket() - checker.IPv4(r.C.t, ackPacket, - checker.TCP( - checker.DstPort(r.SrcPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), - checker.TCPSACKBlockChecker(sackBlocks), - ), - ) -} - -// CreateConnectedWithOptions creates and connects c.ep with the specified TCP -// options enabled and returns a RawEndpoint which represents the other end of -// the connection. -// -// It also verifies where required(eg.Timestamp) that the ACK to the SYN-ACK -// does not carry an option that was not requested. -func (c *Context) CreateConnectedWithOptions(wantOptions header.TCPSynOptions) *RawEndpoint { - var err *tcpip.Error - c.EP, err = c.s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) - if err != nil { - c.t.Fatalf("c.s.NewEndpoint(tcp, ipv4...) = %v", err) - } - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateInitial; got != want { - c.t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) - } - - // Start connection attempt. - waitEntry, notifyCh := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&waitEntry, waiter.EventOut) - defer c.WQ.EventUnregister(&waitEntry) - - testFullAddr := tcpip.FullAddress{Addr: TestAddr, Port: TestPort} - err = c.EP.Connect(testFullAddr) - if err != tcpip.ErrConnectStarted { - c.t.Fatalf("c.ep.Connect(%v) = %v", testFullAddr, err) - } - // Receive SYN packet. - b := c.GetPacket() - // Validate that the syn has the timestamp option and a valid - // TS value. - mss := uint16(c.linkEP.MTU() - header.IPv4MinimumSize - header.TCPMinimumSize) - - checker.IPv4(c.t, b, - checker.TCP( - checker.DstPort(TestPort), - checker.TCPFlags(header.TCPFlagSyn), - checker.TCPSynOptions(header.TCPSynOptions{ - MSS: mss, - TS: true, - WS: int(c.WindowScale), - SACKPermitted: c.SACKEnabled(), - }), - ), - ) - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { - c.t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) - } - - tcpSeg := header.TCP(header.IPv4(b).Payload()) - synOptions := header.ParseSynOptions(tcpSeg.Options(), false) - - // Build options w/ tsVal to be sent in the SYN-ACK. - synAckOptions := make([]byte, header.TCPOptionsMaximumSize) - offset := 0 - if wantOptions.WS != -1 { - offset += header.EncodeWSOption(wantOptions.WS, synAckOptions[offset:]) - } - if wantOptions.TS { - offset += header.EncodeTSOption(wantOptions.TSVal, synOptions.TSVal, synAckOptions[offset:]) - } - if wantOptions.SACKPermitted { - offset += header.EncodeSACKPermittedOption(synAckOptions[offset:]) - } - - offset += header.AddTCPOptionPadding(synAckOptions, offset) - - // Build SYN-ACK. - c.IRS = seqnum.Value(tcpSeg.SequenceNumber()) - iss := seqnum.Value(testInitialSequenceNumber) - c.SendPacket(nil, &Headers{ - SrcPort: tcpSeg.DestinationPort(), - DstPort: tcpSeg.SourcePort(), - Flags: header.TCPFlagSyn | header.TCPFlagAck, - SeqNum: iss, - AckNum: c.IRS.Add(1), - RcvWnd: 30000, - TCPOpts: synAckOptions[:offset], - }) - - // Read ACK. - ackPacket := c.GetPacket() - - // Verify TCP header fields. - tcpCheckers := []checker.TransportChecker{ - checker.DstPort(TestPort), - checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS) + 1), - checker.AckNum(uint32(iss) + 1), - } - - // Verify that tsEcr of ACK packet is wantOptions.TSVal if the - // timestamp option was enabled, if not then we verify that - // there is no timestamp in the ACK packet. - if wantOptions.TS { - tcpCheckers = append(tcpCheckers, checker.TCPTimestampChecker(true, 0, wantOptions.TSVal)) - } else { - tcpCheckers = append(tcpCheckers, checker.TCPTimestampChecker(false, 0, 0)) - } - - checker.IPv4(c.t, ackPacket, checker.TCP(tcpCheckers...)) - - ackSeg := header.TCP(header.IPv4(ackPacket).Payload()) - ackOptions := ackSeg.ParsedOptions() - - // Wait for connection to be established. - select { - case <-notifyCh: - err = c.EP.GetSockOpt(tcpip.ErrorOption{}) - if err != nil { - c.t.Fatalf("Unexpected error when connecting: %v", err) - } - case <-time.After(1 * time.Second): - c.t.Fatalf("Timed out waiting for connection") - } - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateEstablished; got != want { - c.t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) - } - - // Store the source port in use by the endpoint. - c.Port = tcpSeg.SourcePort() - - // Mark in context that timestamp option is enabled for this endpoint. - c.TimeStampEnabled = true - - return &RawEndpoint{ - C: c, - SrcPort: tcpSeg.DestinationPort(), - DstPort: tcpSeg.SourcePort(), - Flags: header.TCPFlagAck | header.TCPFlagPsh, - NextSeqNum: iss + 1, - AckNum: c.IRS.Add(1), - WndSize: 30000, - RecentTS: ackOptions.TSVal, - TSVal: wantOptions.TSVal, - SACKPermitted: wantOptions.SACKPermitted, - } -} - -// AcceptWithOptions initializes a listening endpoint and connects to it with the -// provided options enabled. It also verifies that the SYN-ACK has the expected -// values for the provided options. -// -// The function returns a RawEndpoint representing the other end of the accepted -// endpoint. -func (c *Context) AcceptWithOptions(wndScale int, synOptions header.TCPSynOptions) *RawEndpoint { - // Create EP and start listening. - wq := &waiter.Queue{} - ep, err := c.s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) - if err != nil { - c.t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - - if err := ep.Bind(tcpip.FullAddress{Port: StackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - if got, want := tcp.EndpointState(ep.State()), tcp.StateBound; got != want { - c.t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - if err := ep.Listen(10); err != nil { - c.t.Fatalf("Listen failed: %v", err) - } - if got, want := tcp.EndpointState(ep.State()), tcp.StateListen; got != want { - c.t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - rep := c.PassiveConnectWithOptions(100, wndScale, synOptions) - - // Try to accept the connection. - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - - c.EP, _, err = ep.Accept() - if err == tcpip.ErrWouldBlock { - // Wait for connection to be established. - select { - case <-ch: - c.EP, _, err = ep.Accept() - if err != nil { - c.t.Fatalf("Accept failed: %v", err) - } - - case <-time.After(1 * time.Second): - c.t.Fatalf("Timed out waiting for accept") - } - } - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateEstablished; got != want { - c.t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) - } - - return rep -} - -// PassiveConnect just disables WindowScaling and delegates the call to -// PassiveConnectWithOptions. -func (c *Context) PassiveConnect(maxPayload, wndScale int, synOptions header.TCPSynOptions) { - synOptions.WS = -1 - c.PassiveConnectWithOptions(maxPayload, wndScale, synOptions) -} - -// PassiveConnectWithOptions initiates a new connection (with the specified TCP -// options enabled) to the port on which the Context.ep is listening for new -// connections. It also validates that the SYN-ACK has the expected values for -// the enabled options. -// -// NOTE: MSS is not a negotiated option and it can be asymmetric -// in each direction. This function uses the maxPayload to set the MSS to be -// sent to the peer on a connect and validates that the MSS in the SYN-ACK -// response is equal to the MTU - (tcphdr len + iphdr len). -// -// wndScale is the expected window scale in the SYN-ACK and synOptions.WS is the -// value of the window scaling option to be sent in the SYN. If synOptions.WS > -// 0 then we send the WindowScale option. -func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions header.TCPSynOptions) *RawEndpoint { - opts := make([]byte, header.TCPOptionsMaximumSize) - offset := 0 - offset += header.EncodeMSSOption(uint32(maxPayload), opts) - - if synOptions.WS >= 0 { - offset += header.EncodeWSOption(3, opts[offset:]) - } - if synOptions.TS { - offset += header.EncodeTSOption(synOptions.TSVal, synOptions.TSEcr, opts[offset:]) - } - - if synOptions.SACKPermitted { - offset += header.EncodeSACKPermittedOption(opts[offset:]) - } - - paddingToAdd := 4 - offset%4 - // Now add any padding bytes that might be required to quad align the - // options. - for i := offset; i < offset+paddingToAdd; i++ { - opts[i] = header.TCPOptionNOP - } - offset += paddingToAdd - - // Send a SYN request. - iss := seqnum.Value(testInitialSequenceNumber) - c.SendPacket(nil, &Headers{ - SrcPort: TestPort, - DstPort: StackPort, - Flags: header.TCPFlagSyn, - SeqNum: iss, - RcvWnd: 30000, - TCPOpts: opts[:offset], - }) - - // Receive the SYN-ACK reply. Make sure MSS and other expected options - // are present. - b := c.GetPacket() - tcp := header.TCP(header.IPv4(b).Payload()) - c.IRS = seqnum.Value(tcp.SequenceNumber()) - - tcpCheckers := []checker.TransportChecker{ - checker.SrcPort(StackPort), - checker.DstPort(TestPort), - checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(iss) + 1), - checker.TCPSynOptions(header.TCPSynOptions{MSS: synOptions.MSS, WS: wndScale, SACKPermitted: synOptions.SACKPermitted && c.SACKEnabled()}), - } - - // If TS option was enabled in the original SYN then add a checker to - // validate the Timestamp option in the SYN-ACK. - if synOptions.TS { - tcpCheckers = append(tcpCheckers, checker.TCPTimestampChecker(synOptions.TS, 0, synOptions.TSVal)) - } else { - tcpCheckers = append(tcpCheckers, checker.TCPTimestampChecker(false, 0, 0)) - } - - checker.IPv4(c.t, b, checker.TCP(tcpCheckers...)) - rcvWnd := seqnum.Size(30000) - ackHeaders := &Headers{ - SrcPort: TestPort, - DstPort: StackPort, - Flags: header.TCPFlagAck, - SeqNum: iss + 1, - AckNum: c.IRS + 1, - RcvWnd: rcvWnd, - } - - // If WS was expected to be in effect then scale the advertised window - // correspondingly. - if synOptions.WS > 0 { - ackHeaders.RcvWnd = rcvWnd >> byte(synOptions.WS) - } - - parsedOpts := tcp.ParsedOptions() - if synOptions.TS { - // Echo the tsVal back to the peer in the tsEcr field of the - // timestamp option. - // Increment TSVal by 1 from the value sent in the SYN and echo - // the TSVal in the SYN-ACK in the TSEcr field. - opts := [12]byte{header.TCPOptionNOP, header.TCPOptionNOP} - header.EncodeTSOption(synOptions.TSVal+1, parsedOpts.TSVal, opts[2:]) - ackHeaders.TCPOpts = opts[:] - } - - // Send ACK. - c.SendPacket(nil, ackHeaders) - - c.Port = StackPort - - return &RawEndpoint{ - C: c, - SrcPort: TestPort, - DstPort: StackPort, - Flags: header.TCPFlagPsh | header.TCPFlagAck, - NextSeqNum: iss + 1, - AckNum: c.IRS + 1, - WndSize: rcvWnd, - SACKPermitted: synOptions.SACKPermitted && c.SACKEnabled(), - RecentTS: parsedOpts.TSVal, - TSVal: synOptions.TSVal + 1, - } -} - -// SACKEnabled returns true if the TCP Protocol option SACKEnabled is set to true -// for the Stack in the context. -func (c *Context) SACKEnabled() bool { - var v tcp.SACKEnabled - if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &v); err != nil { - // Stack doesn't support SACK. So just return. - return false - } - return bool(v) -} - -// SetGSOEnabled enables or disables generic segmentation offload. -func (c *Context) SetGSOEnabled(enable bool) { - c.linkEP.GSO = enable -} - -// MSSWithoutOptions returns the value for the MSS used by the stack when no -// options are in use. -func (c *Context) MSSWithoutOptions() uint16 { - return uint16(c.linkEP.MTU() - header.IPv4MinimumSize - header.TCPMinimumSize) -} diff --git a/pkg/tcpip/transport/tcpconntrack/BUILD b/pkg/tcpip/transport/tcpconntrack/BUILD deleted file mode 100644 index 43fcc27f0..000000000 --- a/pkg/tcpip/transport/tcpconntrack/BUILD +++ /dev/null @@ -1,25 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "tcpconntrack", - srcs = ["tcp_conntrack.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcpconntrack", - visibility = ["//visibility:public"], - deps = [ - "//pkg/tcpip/header", - "//pkg/tcpip/seqnum", - ], -) - -go_test( - name = "tcpconntrack_test", - size = "small", - srcs = ["tcp_conntrack_test.go"], - deps = [ - ":tcpconntrack", - "//pkg/tcpip/header", - ], -) diff --git a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go b/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go deleted file mode 100644 index 93712cd45..000000000 --- a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go +++ /dev/null @@ -1,349 +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 tcpconntrack implements a TCP connection tracking object. It allows -// users with access to a segment stream to figure out when a connection is -// established, reset, and closed (and in the last case, who closed first). -package tcpconntrack - -import ( - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/seqnum" -) - -// Result is returned when the state of a TCB is updated in response to an -// inbound or outbound segment. -type Result int - -const ( - // ResultDrop indicates that the segment should be dropped. - ResultDrop Result = iota - - // ResultConnecting indicates that the connection remains in a - // connecting state. - ResultConnecting - - // ResultAlive indicates that the connection remains alive (connected). - ResultAlive - - // ResultReset indicates that the connection was reset. - ResultReset - - // ResultClosedByPeer indicates that the connection was gracefully - // closed, and the inbound stream was closed first. - ResultClosedByPeer - - // ResultClosedBySelf indicates that the connection was gracefully - // closed, and the outbound stream was closed first. - ResultClosedBySelf -) - -// TCB is a TCP Control Block. It holds state necessary to keep track of a TCP -// connection and inform the caller when the connection has been closed. -type TCB struct { - inbound stream - outbound stream - - // State handlers. - handlerInbound func(*TCB, header.TCP) Result - handlerOutbound func(*TCB, header.TCP) Result - - // firstFin holds a pointer to the first stream to send a FIN. - firstFin *stream - - // state is the current state of the stream. - state Result -} - -// Init initializes the state of the TCB according to the initial SYN. -func (t *TCB) Init(initialSyn header.TCP) Result { - t.handlerInbound = synSentStateInbound - t.handlerOutbound = synSentStateOutbound - - iss := seqnum.Value(initialSyn.SequenceNumber()) - t.outbound.una = iss - t.outbound.nxt = iss.Add(logicalLen(initialSyn)) - t.outbound.end = t.outbound.nxt - - // Even though "end" is a sequence number, we don't know the initial - // receive sequence number yet, so we store the window size until we get - // a SYN from the peer. - t.inbound.una = 0 - t.inbound.nxt = 0 - t.inbound.end = seqnum.Value(initialSyn.WindowSize()) - t.state = ResultConnecting - return t.state -} - -// UpdateStateInbound updates the state of the TCB based on the supplied inbound -// segment. -func (t *TCB) UpdateStateInbound(tcp header.TCP) Result { - st := t.handlerInbound(t, tcp) - if st != ResultDrop { - t.state = st - } - return st -} - -// UpdateStateOutbound updates the state of the TCB based on the supplied -// outbound segment. -func (t *TCB) UpdateStateOutbound(tcp header.TCP) Result { - st := t.handlerOutbound(t, tcp) - if st != ResultDrop { - t.state = st - } - return st -} - -// IsAlive returns true as long as the connection is established(Alive) -// or connecting state. -func (t *TCB) IsAlive() bool { - return !t.inbound.rstSeen && !t.outbound.rstSeen && (!t.inbound.closed() || !t.outbound.closed()) -} - -// OutboundSendSequenceNumber returns the snd.NXT for the outbound stream. -func (t *TCB) OutboundSendSequenceNumber() seqnum.Value { - return t.outbound.nxt -} - -// InboundSendSequenceNumber returns the snd.NXT for the inbound stream. -func (t *TCB) InboundSendSequenceNumber() seqnum.Value { - return t.inbound.nxt -} - -// adapResult modifies the supplied "Result" according to the state of the TCB; -// if r is anything other than "Alive", or if one of the streams isn't closed -// yet, it is returned unmodified. Otherwise it's converted to either -// ClosedBySelf or ClosedByPeer depending on which stream was closed first. -func (t *TCB) adaptResult(r Result) Result { - // Check the unmodified case. - if r != ResultAlive || !t.inbound.closed() || !t.outbound.closed() { - return r - } - - // Find out which was closed first. - if t.firstFin == &t.outbound { - return ResultClosedBySelf - } - - return ResultClosedByPeer -} - -// synSentStateInbound is the state handler for inbound segments when the -// connection is in SYN-SENT state. -func synSentStateInbound(t *TCB, tcp header.TCP) Result { - flags := tcp.Flags() - ackPresent := flags&header.TCPFlagAck != 0 - ack := seqnum.Value(tcp.AckNumber()) - - // Ignore segment if ack is present but not acceptable. - if ackPresent && !(ack-1).InRange(t.outbound.una, t.outbound.nxt) { - return ResultConnecting - } - - // If reset is specified, we will let the packet through no matter what - // but we will also destroy the connection if the ACK is present (and - // implicitly acceptable). - if flags&header.TCPFlagRst != 0 { - if ackPresent { - t.inbound.rstSeen = true - return ResultReset - } - return ResultConnecting - } - - // Ignore segment if SYN is not set. - if flags&header.TCPFlagSyn == 0 { - return ResultConnecting - } - - // Update state informed by this SYN. - irs := seqnum.Value(tcp.SequenceNumber()) - t.inbound.una = irs - t.inbound.nxt = irs.Add(logicalLen(tcp)) - t.inbound.end += irs - - t.outbound.end = t.outbound.una.Add(seqnum.Size(tcp.WindowSize())) - - // If the ACK was set (it is acceptable), update our unacknowledgement - // tracking. - if ackPresent { - // Advance the "una" and "end" indices of the outbound stream. - if t.outbound.una.LessThan(ack) { - t.outbound.una = ack - } - - if end := ack.Add(seqnum.Size(tcp.WindowSize())); t.outbound.end.LessThan(end) { - t.outbound.end = end - } - } - - // Update handlers so that new calls will be handled by new state. - t.handlerInbound = allOtherInbound - t.handlerOutbound = allOtherOutbound - - return ResultAlive -} - -// synSentStateOutbound is the state handler for outbound segments when the -// connection is in SYN-SENT state. -func synSentStateOutbound(t *TCB, tcp header.TCP) Result { - // Drop outbound segments that aren't retransmits of the original one. - if tcp.Flags() != header.TCPFlagSyn || - tcp.SequenceNumber() != uint32(t.outbound.una) { - return ResultDrop - } - - // Update the receive window. We only remember the largest value seen. - if wnd := seqnum.Value(tcp.WindowSize()); wnd > t.inbound.end { - t.inbound.end = wnd - } - - return ResultConnecting -} - -// update updates the state of inbound and outbound streams, given the supplied -// inbound segment. For outbound segments, this same function can be called with -// swapped inbound/outbound streams. -func update(tcp header.TCP, inbound, outbound *stream, firstFin **stream) Result { - // Ignore segments out of the window. - s := seqnum.Value(tcp.SequenceNumber()) - if !inbound.acceptable(s, dataLen(tcp)) { - return ResultAlive - } - - flags := tcp.Flags() - if flags&header.TCPFlagRst != 0 { - inbound.rstSeen = true - return ResultReset - } - - // Ignore segments that don't have the ACK flag, and those with the SYN - // flag. - if flags&header.TCPFlagAck == 0 || flags&header.TCPFlagSyn != 0 { - return ResultAlive - } - - // Ignore segments that acknowledge not yet sent data. - ack := seqnum.Value(tcp.AckNumber()) - if outbound.nxt.LessThan(ack) { - return ResultAlive - } - - // Advance the "una" and "end" indices of the outbound stream. - if outbound.una.LessThan(ack) { - outbound.una = ack - } - - if end := ack.Add(seqnum.Size(tcp.WindowSize())); outbound.end.LessThan(end) { - outbound.end = end - } - - // Advance the "nxt" index of the inbound stream. - end := s.Add(logicalLen(tcp)) - if inbound.nxt.LessThan(end) { - inbound.nxt = end - } - - // Note the index of the FIN segment. And stash away a pointer to the - // first stream to see a FIN. - if flags&header.TCPFlagFin != 0 && !inbound.finSeen { - inbound.finSeen = true - inbound.fin = end - 1 - - if *firstFin == nil { - *firstFin = inbound - } - } - - return ResultAlive -} - -// allOtherInbound is the state handler for inbound segments in all states -// except SYN-SENT. -func allOtherInbound(t *TCB, tcp header.TCP) Result { - return t.adaptResult(update(tcp, &t.inbound, &t.outbound, &t.firstFin)) -} - -// allOtherOutbound is the state handler for outbound segments in all states -// except SYN-SENT. -func allOtherOutbound(t *TCB, tcp header.TCP) Result { - return t.adaptResult(update(tcp, &t.outbound, &t.inbound, &t.firstFin)) -} - -// streams holds the state of a TCP unidirectional stream. -type stream struct { - // The interval [una, end) is the allowed interval as defined by the - // receiver, i.e., anything less than una has already been acknowledged - // and anything greater than or equal to end is beyond the receiver - // window. The interval [una, nxt) is the acknowledgable range, whose - // right edge indicates the sequence number of the next byte to be sent - // by the sender, i.e., anything greater than or equal to nxt hasn't - // been sent yet. - una seqnum.Value - nxt seqnum.Value - end seqnum.Value - - // finSeen indicates if a FIN has already been sent on this stream. - finSeen bool - - // fin is the sequence number of the FIN. It is only valid after finSeen - // is set to true. - fin seqnum.Value - - // rstSeen indicates if a RST has already been sent on this stream. - rstSeen bool -} - -// acceptable determines if the segment with the given sequence number and data -// length is acceptable, i.e., if it's within the [una, end) window or, in case -// the window is zero, if it's a packet with no payload and sequence number -// equal to una. -func (s *stream) acceptable(segSeq seqnum.Value, segLen seqnum.Size) bool { - wnd := s.una.Size(s.end) - if wnd == 0 { - return segLen == 0 && segSeq == s.una - } - - // Make sure [segSeq, seqSeq+segLen) is non-empty. - if segLen == 0 { - segLen = 1 - } - - return seqnum.Overlap(s.una, wnd, segSeq, segLen) -} - -// closed determines if the stream has already been closed. This happens when -// a FIN has been set by the sender and acknowledged by the receiver. -func (s *stream) closed() bool { - return s.finSeen && s.fin.LessThan(s.una) -} - -// dataLen returns the length of the TCP segment payload. -func dataLen(tcp header.TCP) seqnum.Size { - return seqnum.Size(len(tcp) - int(tcp.DataOffset())) -} - -// logicalLen calculates the logical length of the TCP segment. -func logicalLen(tcp header.TCP) seqnum.Size { - l := dataLen(tcp) - flags := tcp.Flags() - if flags&header.TCPFlagSyn != 0 { - l++ - } - if flags&header.TCPFlagFin != 0 { - l++ - } - return l -} diff --git a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack_test.go b/pkg/tcpip/transport/tcpconntrack/tcp_conntrack_test.go deleted file mode 100644 index 5e271b7ca..000000000 --- a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack_test.go +++ /dev/null @@ -1,511 +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 tcpconntrack_test - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcpconntrack" -) - -// connected creates a connection tracker TCB and sets it to a connected state -// by performing a 3-way handshake. -func connected(t *testing.T, iss, irs uint32, isw, irw uint16) *tcpconntrack.TCB { - // Send SYN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: iss, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: irw, - }) - - tcb := tcpconntrack.TCB{} - tcb.Init(tcp) - - // Receive SYN-ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: irs, - AckNum: iss + 1, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn | header.TCPFlagAck, - WindowSize: isw, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Send ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: iss + 1, - AckNum: irs + 1, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: irw, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - return &tcb -} - -func TestConnectionRefused(t *testing.T) { - // Send SYN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 30000, - }) - - tcb := tcpconntrack.TCB{} - tcb.Init(tcp) - - // Receive RST. - tcp.Encode(&header.TCPFields{ - SeqNum: 789, - AckNum: 1235, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagRst | header.TCPFlagAck, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultReset { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultReset) - } -} - -func TestConnectionRefusedInSynRcvd(t *testing.T) { - // Send SYN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 30000, - }) - - tcb := tcpconntrack.TCB{} - tcb.Init(tcp) - - // Receive SYN. - tcp.Encode(&header.TCPFields{ - SeqNum: 789, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Receive RST with no ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 790, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagRst, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultReset { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultReset) - } -} - -func TestConnectionResetInSynRcvd(t *testing.T) { - // Send SYN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 30000, - }) - - tcb := tcpconntrack.TCB{} - tcb.Init(tcp) - - // Receive SYN. - tcp.Encode(&header.TCPFields{ - SeqNum: 789, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Send RST with no ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 1235, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagRst, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultReset { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultReset) - } -} - -func TestRetransmitOnSynSent(t *testing.T) { - // Send initial SYN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 30000, - }) - - tcb := tcpconntrack.TCB{} - tcb.Init(tcp) - - // Retransmit the same SYN. - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultConnecting { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultConnecting) - } -} - -func TestRetransmitOnSynRcvd(t *testing.T) { - // Send initial SYN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 30000, - }) - - tcb := tcpconntrack.TCB{} - tcb.Init(tcp) - - // Receive SYN. This will cause the state to go to SYN-RCVD. - tcp.Encode(&header.TCPFields{ - SeqNum: 789, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Retransmit the original SYN. - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Transmit a SYN-ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 790, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn | header.TCPFlagAck, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } -} - -func TestClosedBySelf(t *testing.T) { - tcb := connected(t, 1234, 789, 30000, 50000) - - // Send FIN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 1235, - AckNum: 790, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck | header.TCPFlagFin, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Receive FIN/ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 790, - AckNum: 1236, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck | header.TCPFlagFin, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Send ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 1236, - AckNum: 791, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultClosedBySelf { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultClosedBySelf) - } -} - -func TestClosedByPeer(t *testing.T) { - tcb := connected(t, 1234, 789, 30000, 50000) - - // Receive FIN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 790, - AckNum: 1235, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck | header.TCPFlagFin, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Send FIN/ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 1235, - AckNum: 791, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck | header.TCPFlagFin, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Receive ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 791, - AckNum: 1236, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultClosedByPeer { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultClosedByPeer) - } -} - -func TestSendAndReceiveDataClosedBySelf(t *testing.T) { - sseq := uint32(1234) - rseq := uint32(789) - tcb := connected(t, sseq, rseq, 30000, 50000) - sseq++ - rseq++ - - // Send some data. - tcp := make(header.TCP, header.TCPMinimumSize+1024) - - for i := uint32(0); i < 10; i++ { - // Send some data. - tcp.Encode(&header.TCPFields{ - SeqNum: sseq, - AckNum: rseq, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 30000, - }) - sseq += uint32(len(tcp)) - header.TCPMinimumSize - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Receive ack for data. - tcp.Encode(&header.TCPFields{ - SeqNum: rseq, - AckNum: sseq, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp[:header.TCPMinimumSize]); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - } - - for i := uint32(0); i < 10; i++ { - // Receive some data. - tcp.Encode(&header.TCPFields{ - SeqNum: rseq, - AckNum: sseq, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 50000, - }) - rseq += uint32(len(tcp)) - header.TCPMinimumSize - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Send ack for data. - tcp.Encode(&header.TCPFields{ - SeqNum: sseq, - AckNum: rseq, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp[:header.TCPMinimumSize]); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - } - - // Send FIN. - tcp = tcp[:header.TCPMinimumSize] - tcp.Encode(&header.TCPFields{ - SeqNum: sseq, - AckNum: rseq, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck | header.TCPFlagFin, - WindowSize: 30000, - }) - sseq++ - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Receive FIN/ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: rseq, - AckNum: sseq, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck | header.TCPFlagFin, - WindowSize: 50000, - }) - rseq++ - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Send ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: sseq, - AckNum: rseq, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultClosedBySelf { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultClosedBySelf) - } -} - -func TestIgnoreBadResetOnSynSent(t *testing.T) { - // Send SYN. - tcp := make(header.TCP, header.TCPMinimumSize) - tcp.Encode(&header.TCPFields{ - SeqNum: 1234, - AckNum: 0, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn, - WindowSize: 30000, - }) - - tcb := tcpconntrack.TCB{} - tcb.Init(tcp) - - // Receive a RST with a bad ACK, it should not cause the connection to - // be reset. - acks := []uint32{1234, 1236, 1000, 5000} - flags := []uint8{header.TCPFlagRst, header.TCPFlagRst | header.TCPFlagAck} - for _, a := range acks { - for _, f := range flags { - tcp.Encode(&header.TCPFields{ - SeqNum: 789, - AckNum: a, - DataOffset: header.TCPMinimumSize, - Flags: f, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultConnecting { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - } - } - - // Complete the handshake. - // Receive SYN-ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 789, - AckNum: 1235, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagSyn | header.TCPFlagAck, - WindowSize: 50000, - }) - - if r := tcb.UpdateStateInbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } - - // Send ACK. - tcp.Encode(&header.TCPFields{ - SeqNum: 1235, - AckNum: 790, - DataOffset: header.TCPMinimumSize, - Flags: header.TCPFlagAck, - WindowSize: 30000, - }) - - if r := tcb.UpdateStateOutbound(tcp); r != tcpconntrack.ResultAlive { - t.Fatalf("Bad result: got %v, want %v", r, tcpconntrack.ResultAlive) - } -} diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD deleted file mode 100644 index c1ca22b35..000000000 --- a/pkg/tcpip/transport/udp/BUILD +++ /dev/null @@ -1,69 +0,0 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") - -go_template_instance( - name = "udp_packet_list", - out = "udp_packet_list.go", - package = "udp", - prefix = "udpPacket", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*udpPacket", - "Linker": "*udpPacket", - }, -) - -go_library( - name = "udp", - srcs = [ - "endpoint.go", - "endpoint_state.go", - "forwarder.go", - "protocol.go", - "udp_packet_list.go", - ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/udp", - imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], - visibility = ["//visibility:public"], - deps = [ - "//pkg/sleep", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/header", - "//pkg/tcpip/iptables", - "//pkg/tcpip/stack", - "//pkg/tcpip/transport/raw", - "//pkg/waiter", - ], -) - -go_test( - name = "udp_x_test", - size = "small", - srcs = ["udp_test.go"], - deps = [ - ":udp", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/checker", - "//pkg/tcpip/header", - "//pkg/tcpip/link/channel", - "//pkg/tcpip/link/sniffer", - "//pkg/tcpip/network/ipv4", - "//pkg/tcpip/network/ipv6", - "//pkg/tcpip/stack", - "//pkg/waiter", - ], -) - -filegroup( - name = "autogen", - srcs = [ - "udp_packet_list.go", - ], - visibility = ["//:sandbox"], -) diff --git a/pkg/tcpip/transport/udp/udp_packet_list.go b/pkg/tcpip/transport/udp/udp_packet_list.go new file mode 100755 index 000000000..673a9373b --- /dev/null +++ b/pkg/tcpip/transport/udp/udp_packet_list.go @@ -0,0 +1,173 @@ +package udp + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type udpPacketElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (udpPacketElementMapper) linkerFor(elem *udpPacket) *udpPacket { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type udpPacketList struct { + head *udpPacket + tail *udpPacket +} + +// Reset resets list l to the empty state. +func (l *udpPacketList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *udpPacketList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *udpPacketList) Front() *udpPacket { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *udpPacketList) Back() *udpPacket { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *udpPacketList) PushFront(e *udpPacket) { + udpPacketElementMapper{}.linkerFor(e).SetNext(l.head) + udpPacketElementMapper{}.linkerFor(e).SetPrev(nil) + + if l.head != nil { + udpPacketElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *udpPacketList) PushBack(e *udpPacket) { + udpPacketElementMapper{}.linkerFor(e).SetNext(nil) + udpPacketElementMapper{}.linkerFor(e).SetPrev(l.tail) + + if l.tail != nil { + udpPacketElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *udpPacketList) PushBackList(m *udpPacketList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + udpPacketElementMapper{}.linkerFor(l.tail).SetNext(m.head) + udpPacketElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *udpPacketList) InsertAfter(b, e *udpPacket) { + a := udpPacketElementMapper{}.linkerFor(b).Next() + udpPacketElementMapper{}.linkerFor(e).SetNext(a) + udpPacketElementMapper{}.linkerFor(e).SetPrev(b) + udpPacketElementMapper{}.linkerFor(b).SetNext(e) + + if a != nil { + udpPacketElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *udpPacketList) InsertBefore(a, e *udpPacket) { + b := udpPacketElementMapper{}.linkerFor(a).Prev() + udpPacketElementMapper{}.linkerFor(e).SetNext(a) + udpPacketElementMapper{}.linkerFor(e).SetPrev(b) + udpPacketElementMapper{}.linkerFor(a).SetPrev(e) + + if b != nil { + udpPacketElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *udpPacketList) Remove(e *udpPacket) { + prev := udpPacketElementMapper{}.linkerFor(e).Prev() + next := udpPacketElementMapper{}.linkerFor(e).Next() + + if prev != nil { + udpPacketElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + udpPacketElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type udpPacketEntry struct { + next *udpPacket + prev *udpPacket +} + +// Next returns the entry that follows e in the list. +func (e *udpPacketEntry) Next() *udpPacket { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *udpPacketEntry) Prev() *udpPacket { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *udpPacketEntry) SetNext(elem *udpPacket) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *udpPacketEntry) SetPrev(elem *udpPacket) { + e.prev = elem +} diff --git a/pkg/tcpip/transport/udp/udp_state_autogen.go b/pkg/tcpip/transport/udp/udp_state_autogen.go new file mode 100755 index 000000000..eb9d419ef --- /dev/null +++ b/pkg/tcpip/transport/udp/udp_state_autogen.go @@ -0,0 +1,128 @@ +// automatically generated by stateify. + +package udp + +import ( + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) + +func (x *udpPacket) beforeSave() {} +func (x *udpPacket) save(m state.Map) { + x.beforeSave() + var data buffer.VectorisedView = x.saveData() + m.SaveValue("data", data) + m.Save("udpPacketEntry", &x.udpPacketEntry) + m.Save("senderAddress", &x.senderAddress) + m.Save("timestamp", &x.timestamp) +} + +func (x *udpPacket) afterLoad() {} +func (x *udpPacket) load(m state.Map) { + m.Load("udpPacketEntry", &x.udpPacketEntry) + m.Load("senderAddress", &x.senderAddress) + m.Load("timestamp", &x.timestamp) + m.LoadValue("data", new(buffer.VectorisedView), func(y interface{}) { x.loadData(y.(buffer.VectorisedView)) }) +} + +func (x *endpoint) save(m state.Map) { + x.beforeSave() + var rcvBufSizeMax int = x.saveRcvBufSizeMax() + m.SaveValue("rcvBufSizeMax", rcvBufSizeMax) + m.Save("netProto", &x.netProto) + m.Save("waiterQueue", &x.waiterQueue) + m.Save("rcvReady", &x.rcvReady) + m.Save("rcvList", &x.rcvList) + m.Save("rcvBufSize", &x.rcvBufSize) + m.Save("rcvClosed", &x.rcvClosed) + m.Save("sndBufSize", &x.sndBufSize) + m.Save("id", &x.id) + m.Save("state", &x.state) + m.Save("bindNICID", &x.bindNICID) + m.Save("regNICID", &x.regNICID) + m.Save("dstPort", &x.dstPort) + m.Save("v6only", &x.v6only) + m.Save("multicastTTL", &x.multicastTTL) + m.Save("multicastAddr", &x.multicastAddr) + m.Save("multicastNICID", &x.multicastNICID) + m.Save("multicastLoop", &x.multicastLoop) + m.Save("reusePort", &x.reusePort) + m.Save("broadcast", &x.broadcast) + m.Save("shutdownFlags", &x.shutdownFlags) + m.Save("multicastMemberships", &x.multicastMemberships) + m.Save("effectiveNetProtos", &x.effectiveNetProtos) +} + +func (x *endpoint) load(m state.Map) { + m.Load("netProto", &x.netProto) + m.Load("waiterQueue", &x.waiterQueue) + m.Load("rcvReady", &x.rcvReady) + m.Load("rcvList", &x.rcvList) + m.Load("rcvBufSize", &x.rcvBufSize) + m.Load("rcvClosed", &x.rcvClosed) + m.Load("sndBufSize", &x.sndBufSize) + m.Load("id", &x.id) + m.Load("state", &x.state) + m.Load("bindNICID", &x.bindNICID) + m.Load("regNICID", &x.regNICID) + m.Load("dstPort", &x.dstPort) + m.Load("v6only", &x.v6only) + m.Load("multicastTTL", &x.multicastTTL) + m.Load("multicastAddr", &x.multicastAddr) + m.Load("multicastNICID", &x.multicastNICID) + m.Load("multicastLoop", &x.multicastLoop) + m.Load("reusePort", &x.reusePort) + m.Load("broadcast", &x.broadcast) + m.Load("shutdownFlags", &x.shutdownFlags) + m.Load("multicastMemberships", &x.multicastMemberships) + m.Load("effectiveNetProtos", &x.effectiveNetProtos) + m.LoadValue("rcvBufSizeMax", new(int), func(y interface{}) { x.loadRcvBufSizeMax(y.(int)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *multicastMembership) beforeSave() {} +func (x *multicastMembership) save(m state.Map) { + x.beforeSave() + m.Save("nicID", &x.nicID) + m.Save("multicastAddr", &x.multicastAddr) +} + +func (x *multicastMembership) afterLoad() {} +func (x *multicastMembership) load(m state.Map) { + m.Load("nicID", &x.nicID) + m.Load("multicastAddr", &x.multicastAddr) +} + +func (x *udpPacketList) beforeSave() {} +func (x *udpPacketList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *udpPacketList) afterLoad() {} +func (x *udpPacketList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *udpPacketEntry) beforeSave() {} +func (x *udpPacketEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *udpPacketEntry) afterLoad() {} +func (x *udpPacketEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("udp.udpPacket", (*udpPacket)(nil), state.Fns{Save: (*udpPacket).save, Load: (*udpPacket).load}) + state.Register("udp.endpoint", (*endpoint)(nil), state.Fns{Save: (*endpoint).save, Load: (*endpoint).load}) + state.Register("udp.multicastMembership", (*multicastMembership)(nil), state.Fns{Save: (*multicastMembership).save, Load: (*multicastMembership).load}) + state.Register("udp.udpPacketList", (*udpPacketList)(nil), state.Fns{Save: (*udpPacketList).save, Load: (*udpPacketList).load}) + state.Register("udp.udpPacketEntry", (*udpPacketEntry)(nil), state.Fns{Save: (*udpPacketEntry).save, Load: (*udpPacketEntry).load}) +} diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go deleted file mode 100644 index c6deab892..000000000 --- a/pkg/tcpip/transport/udp/udp_test.go +++ /dev/null @@ -1,1395 +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 udp_test - -import ( - "bytes" - "fmt" - "math" - "math/rand" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "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/sniffer" - "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" -) - -// Addresses and ports used for testing. It is recommended that tests stick to -// using these addresses as it allows using the testFlow helper. -// Naming rules: 'stack*'' denotes local addresses and ports, while 'test*' -// represents the remote endpoint. -const ( - v4MappedAddrPrefix = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff" - 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" - stackV4MappedAddr = v4MappedAddrPrefix + stackAddr - testV4MappedAddr = v4MappedAddrPrefix + testAddr - multicastV4MappedAddr = v4MappedAddrPrefix + multicastAddr - broadcastV4MappedAddr = v4MappedAddrPrefix + broadcastAddr - v4MappedWildcardAddr = v4MappedAddrPrefix + "\x00\x00\x00\x00" - - stackAddr = "\x0a\x00\x00\x01" - stackPort = 1234 - testAddr = "\x0a\x00\x00\x02" - testPort = 4096 - multicastAddr = "\xe8\x2b\xd3\xea" - multicastV6Addr = "\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" - broadcastAddr = header.IPv4Broadcast - - // defaultMTU is the MTU, in bytes, used throughout the tests, except - // where another value is explicitly used. It is chosen to match the MTU - // of loopback interfaces on linux systems. - defaultMTU = 65536 -) - -// header4Tuple stores the 4-tuple {src-IP, src-port, dst-IP, dst-port} used in -// a packet header. These values are used to populate a header or verify one. -// Note that because they are used in packet headers, the addresses are never in -// a V4-mapped format. -type header4Tuple struct { - srcAddr tcpip.FullAddress - dstAddr tcpip.FullAddress -} - -// testFlow implements a helper type used for sending and receiving test -// packets. A given test flow value defines 1) the socket endpoint used for the -// test and 2) the type of packet send or received on the endpoint. E.g., a -// multicastV6Only flow is a V6 multicast packet passing through a V6-only -// endpoint. The type provides helper methods to characterize the flow (e.g., -// isV4) as well as return a proper header4Tuple for it. -type testFlow int - -const ( - unicastV4 testFlow = iota // V4 unicast on a V4 socket - unicastV4in6 // V4-mapped unicast on a V6-dual socket - unicastV6 // V6 unicast on a V6 socket - unicastV6Only // V6 unicast on a V6-only socket - multicastV4 // V4 multicast on a V4 socket - multicastV4in6 // V4-mapped multicast on a V6-dual socket - multicastV6 // V6 multicast on a V6 socket - multicastV6Only // V6 multicast on a V6-only socket - broadcast // V4 broadcast on a V4 socket - broadcastIn6 // V4-mapped broadcast on a V6-dual socket -) - -func (flow testFlow) String() string { - switch flow { - case unicastV4: - return "unicastV4" - case unicastV6: - return "unicastV6" - case unicastV6Only: - return "unicastV6Only" - case unicastV4in6: - return "unicastV4in6" - case multicastV4: - return "multicastV4" - case multicastV6: - return "multicastV6" - case multicastV6Only: - return "multicastV6Only" - case multicastV4in6: - return "multicastV4in6" - case broadcast: - return "broadcast" - case broadcastIn6: - return "broadcastIn6" - default: - return "unknown" - } -} - -// packetDirection explains if a flow is incoming (read) or outgoing (write). -type packetDirection int - -const ( - incoming packetDirection = iota - outgoing -) - -// header4Tuple returns the header4Tuple for the given flow and direction. Note -// that the tuple contains no mapped addresses as those only exist at the socket -// level but not at the packet header level. -func (flow testFlow) header4Tuple(d packetDirection) header4Tuple { - var h header4Tuple - if flow.isV4() { - if d == outgoing { - h = header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: stackAddr, Port: stackPort}, - dstAddr: tcpip.FullAddress{Addr: testAddr, Port: testPort}, - } - } else { - h = header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: testAddr, Port: testPort}, - dstAddr: tcpip.FullAddress{Addr: stackAddr, Port: stackPort}, - } - } - if flow.isMulticast() { - h.dstAddr.Addr = multicastAddr - } else if flow.isBroadcast() { - h.dstAddr.Addr = broadcastAddr - } - } else { // IPv6 - if d == outgoing { - h = header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}, - dstAddr: tcpip.FullAddress{Addr: testV6Addr, Port: testPort}, - } - } else { - h = header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: testV6Addr, Port: testPort}, - dstAddr: tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}, - } - } - if flow.isMulticast() { - h.dstAddr.Addr = multicastV6Addr - } - } - return h -} - -func (flow testFlow) getMcastAddr() tcpip.Address { - if flow.isV4() { - return multicastAddr - } - return multicastV6Addr -} - -// mapAddrIfApplicable converts the given V4 address into its V4-mapped version -// if it is applicable to the flow. -func (flow testFlow) mapAddrIfApplicable(v4Addr tcpip.Address) tcpip.Address { - if flow.isMapped() { - return v4MappedAddrPrefix + v4Addr - } - return v4Addr -} - -// netProto returns the protocol number used for the network packet. -func (flow testFlow) netProto() tcpip.NetworkProtocolNumber { - if flow.isV4() { - return ipv4.ProtocolNumber - } - return ipv6.ProtocolNumber -} - -// sockProto returns the protocol number used when creating the socket -// endpoint for this flow. -func (flow testFlow) sockProto() tcpip.NetworkProtocolNumber { - switch flow { - case unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, multicastV6Only, broadcastIn6: - return ipv6.ProtocolNumber - case unicastV4, multicastV4, broadcast: - return ipv4.ProtocolNumber - default: - panic(fmt.Sprintf("invalid testFlow given: %d", flow)) - } -} - -func (flow testFlow) checkerFn() func(*testing.T, []byte, ...checker.NetworkChecker) { - if flow.isV4() { - return checker.IPv4 - } - return checker.IPv6 -} - -func (flow testFlow) isV6() bool { return !flow.isV4() } -func (flow testFlow) isV4() bool { - return flow.sockProto() == ipv4.ProtocolNumber || flow.isMapped() -} - -func (flow testFlow) isV6Only() bool { - switch flow { - case unicastV6Only, multicastV6Only: - return true - case unicastV4, unicastV4in6, unicastV6, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6: - return false - default: - panic(fmt.Sprintf("invalid testFlow given: %d", flow)) - } -} - -func (flow testFlow) isMulticast() bool { - switch flow { - case multicastV4, multicastV4in6, multicastV6, multicastV6Only: - return true - case unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6: - return false - default: - panic(fmt.Sprintf("invalid testFlow given: %d", flow)) - } -} - -func (flow testFlow) isBroadcast() bool { - switch flow { - case broadcast, broadcastIn6: - return true - case unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, multicastV6Only: - return false - default: - panic(fmt.Sprintf("invalid testFlow given: %d", flow)) - } -} - -func (flow testFlow) isMapped() bool { - switch flow { - case unicastV4in6, multicastV4in6, broadcastIn6: - return true - case unicastV4, unicastV6, unicastV6Only, multicastV4, multicastV6, multicastV6Only, broadcast: - return false - default: - panic(fmt.Sprintf("invalid testFlow given: %d", flow)) - } -} - -type testContext struct { - t *testing.T - linkEP *channel.Endpoint - s *stack.Stack - - ep tcpip.Endpoint - wq waiter.Queue -} - -func newDualTestContext(t *testing.T, mtu uint32) *testContext { - t.Helper() - - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{udp.ProtocolName}, stack.Options{}) - ep := channel.New(256, mtu, "") - wep := stack.LinkEndpoint(ep) - - if testing.Verbose() { - wep = sniffer.New(ep) - } - if err := s.CreateNIC(1, wep); err != nil { - t.Fatalf("CreateNIC failed: %v", err) - } - - if err := s.AddAddress(1, ipv4.ProtocolNumber, stackAddr); err != nil { - t.Fatalf("AddAddress failed: %v", err) - } - - if err := s.AddAddress(1, ipv6.ProtocolNumber, stackV6Addr); err != nil { - t.Fatalf("AddAddress failed: %v", err) - } - - s.SetRouteTable([]tcpip.Route{ - { - Destination: header.IPv4EmptySubnet, - NIC: 1, - }, - { - Destination: header.IPv6EmptySubnet, - NIC: 1, - }, - }) - - return &testContext{ - t: t, - s: s, - linkEP: ep, - } -} - -func (c *testContext) cleanup() { - if c.ep != nil { - c.ep.Close() - } -} - -func (c *testContext) createEndpoint(proto tcpip.NetworkProtocolNumber) { - c.t.Helper() - - var err *tcpip.Error - c.ep, err = c.s.NewEndpoint(udp.ProtocolNumber, proto, &c.wq) - if err != nil { - c.t.Fatal("NewEndpoint failed: ", err) - } -} - -func (c *testContext) createEndpointForFlow(flow testFlow) { - c.t.Helper() - - c.createEndpoint(flow.sockProto()) - if flow.isV6Only() { - if err := c.ep.SetSockOpt(tcpip.V6OnlyOption(1)); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) - } - } else if flow.isBroadcast() { - if err := c.ep.SetSockOpt(tcpip.BroadcastOption(1)); err != nil { - c.t.Fatal("SetSockOpt failed:", err) - } - } -} - -// getPacketAndVerify reads a packet from the link endpoint and verifies the -// header against expected values from the given test flow. In addition, it -// calls any extra checker functions provided. -func (c *testContext) getPacketAndVerify(flow testFlow, checkers ...checker.NetworkChecker) []byte { - c.t.Helper() - - select { - case p := <-c.linkEP.C: - if p.Proto != flow.netProto() { - c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, flow.netProto()) - } - b := make([]byte, len(p.Header)+len(p.Payload)) - copy(b, p.Header) - copy(b[len(p.Header):], p.Payload) - - h := flow.header4Tuple(outgoing) - checkers := append( - checkers, - checker.SrcAddr(h.srcAddr.Addr), - checker.DstAddr(h.dstAddr.Addr), - checker.UDP(checker.DstPort(h.dstAddr.Port)), - ) - flow.checkerFn()(c.t, b, checkers...) - return b - - case <-time.After(2 * time.Second): - c.t.Fatalf("Packet wasn't written out") - } - - return nil -} - -// injectPacket creates a packet of the given flow and with the given payload, -// and injects it into the link endpoint. -func (c *testContext) injectPacket(flow testFlow, payload []byte) { - c.t.Helper() - - h := flow.header4Tuple(incoming) - if flow.isV4() { - c.injectV4Packet(payload, &h) - } else { - c.injectV6Packet(payload, &h) - } -} - -// injectV6Packet creates a V6 test packet with the given payload and header -// values, and injects it into the link endpoint. -func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple) { - // 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: h.srcAddr.Addr, - DstAddr: h.dstAddr.Addr, - }) - - // Initialize the UDP header. - u := header.UDP(buf[header.IPv6MinimumSize:]) - u.Encode(&header.UDPFields{ - SrcPort: h.srcAddr.Port, - DstPort: h.dstAddr.Port, - Length: uint16(header.UDPMinimumSize + len(payload)), - }) - - // Calculate the UDP pseudo-header checksum. - xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, h.srcAddr.Addr, h.dstAddr.Addr, uint16(len(u))) - - // Calculate the UDP checksum and set it. - xsum = header.Checksum(payload, xsum) - u.SetChecksum(^u.CalculateChecksum(xsum)) - - // Inject packet. - c.linkEP.Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) -} - -// injectV6Packet creates a V4 test packet with the given payload and header -// values, and injects it into the link endpoint. -func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple) { - // Allocate a buffer for data and headers. - buf := buffer.NewView(header.UDPMinimumSize + header.IPv4MinimumSize + len(payload)) - copy(buf[len(buf)-len(payload):], payload) - - // Initialize the IP header. - ip := header.IPv4(buf) - ip.Encode(&header.IPv4Fields{ - IHL: header.IPv4MinimumSize, - TotalLength: uint16(len(buf)), - TTL: 65, - Protocol: uint8(udp.ProtocolNumber), - SrcAddr: h.srcAddr.Addr, - DstAddr: h.dstAddr.Addr, - }) - ip.SetChecksum(^ip.CalculateChecksum()) - - // Initialize the UDP header. - u := header.UDP(buf[header.IPv4MinimumSize:]) - u.Encode(&header.UDPFields{ - SrcPort: h.srcAddr.Port, - DstPort: h.dstAddr.Port, - Length: uint16(header.UDPMinimumSize + len(payload)), - }) - - // Calculate the UDP pseudo-header checksum. - xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, h.srcAddr.Addr, h.dstAddr.Addr, uint16(len(u))) - - // Calculate the UDP checksum and set it. - xsum = header.Checksum(payload, xsum) - u.SetChecksum(^u.CalculateChecksum(xsum)) - - // Inject packet. - c.linkEP.Inject(ipv4.ProtocolNumber, buf.ToVectorisedView()) -} - -func newPayload() []byte { - return newMinPayload(30) -} - -func newMinPayload(minSize int) []byte { - b := make([]byte, minSize+rand.Intn(100)) - for i := range b { - b[i] = byte(rand.Intn(256)) - } - 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]) - - 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) - } - } - - 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}, - }) - - 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") - } - } - } - - if len(stats) != len(eps) { - t.Fatalf("Only %d(expected %d) sockets received packets", len(stats), len(eps)) - } - - // 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) - } - } -} - -// testRead sends a packet of the given test flow into the stack by injecting it -// into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness. -func testRead(c *testContext, flow testFlow) { - c.t.Helper() - - payload := newPayload() - c.injectPacket(flow, payload) - - // Try to receive the data. - we, ch := waiter.NewChannelEntry(nil) - c.wq.EventRegister(&we, waiter.EventIn) - defer c.wq.EventUnregister(&we) - - var addr tcpip.FullAddress - v, _, err := c.ep.Read(&addr) - if err == tcpip.ErrWouldBlock { - // Wait for data to become available. - select { - case <-ch: - v, _, err = c.ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - - case <-time.After(1 * time.Second): - c.t.Fatalf("Timed out waiting for data") - } - } - - // Check the peer address. - h := flow.header4Tuple(incoming) - if addr.Addr != h.srcAddr.Addr { - c.t.Fatalf("Unexpected remote address: got %v, want %v", addr.Addr, h.srcAddr) - } - - // Check the payload. - if !bytes.Equal(payload, v) { - c.t.Fatalf("Bad payload: got %x, want %x", v, payload) - } -} - -func TestBindEphemeralPort(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - if err := c.ep.Bind(tcpip.FullAddress{}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) - } -} - -func TestBindReservedPort(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) - } - - addr, err := c.ep.GetLocalAddress() - if err != nil { - t.Fatalf("GetLocalAddress failed: %v", err) - } - - // We can't bind the address reserved by the connected endpoint above. - { - ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - if got, want := ep.Bind(addr), tcpip.ErrPortInUse; got != want { - t.Fatalf("got ep.Bind(...) = %v, want = %v", got, want) - } - } - - func() { - ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &c.wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - // We can't bind ipv4-any on the port reserved by the connected endpoint - // above, since the endpoint is dual-stack. - if got, want := ep.Bind(tcpip.FullAddress{Port: addr.Port}), tcpip.ErrPortInUse; got != want { - t.Fatalf("got ep.Bind(...) = %v, want = %v", got, want) - } - // We can bind an ipv4 address on this port, though. - if err := ep.Bind(tcpip.FullAddress{Addr: stackAddr, Port: addr.Port}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) - } - }() - - // Once the connected endpoint releases its port reservation, we are able to - // bind ipv4-any once again. - c.ep.Close() - func() { - ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &c.wq) - if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) - } - defer ep.Close() - if err := ep.Bind(tcpip.FullAddress{Port: addr.Port}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) - } - }() -} - -func TestV4ReadOnV6(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(unicastV4in6) - - // Bind to wildcard. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testRead(c, unicastV4in6) -} - -func TestV4ReadOnBoundToV4MappedWildcard(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(unicastV4in6) - - // Bind to v4 mapped wildcard. - if err := c.ep.Bind(tcpip.FullAddress{Addr: v4MappedWildcardAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testRead(c, unicastV4in6) -} - -func TestV4ReadOnBoundToV4Mapped(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(unicastV4in6) - - // Bind to local address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: stackV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testRead(c, unicastV4in6) -} - -func TestV6ReadOnV6(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(unicastV6) - - // Bind to wildcard. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testRead(c, unicastV6) -} - -func TestV4ReadOnV4(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(unicastV4) - - // Bind to wildcard. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - // Test acceptance. - testRead(c, unicastV4) -} - -// TestReadOnBoundToMulticast checks that an endpoint can bind to a multicast -// address and receive data sent to that address. -func TestReadOnBoundToMulticast(t *testing.T) { - // FIXME(b/128189410): multicastV4in6 currently doesn't work as - // AddMembershipOption doesn't handle V4in6 addresses. - for _, flow := range []testFlow{multicastV4, multicastV6, multicastV6Only} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to multicast address. - mcastAddr := flow.mapAddrIfApplicable(flow.getMcastAddr()) - if err := c.ep.Bind(tcpip.FullAddress{Addr: mcastAddr, Port: stackPort}); err != nil { - c.t.Fatal("Bind failed:", err) - } - - // Join multicast group. - ifoptSet := tcpip.AddMembershipOption{NIC: 1, MulticastAddr: mcastAddr} - if err := c.ep.SetSockOpt(ifoptSet); err != nil { - c.t.Fatal("SetSockOpt failed:", err) - } - - testRead(c, flow) - }) - } -} - -// TestV4ReadOnBoundToBroadcast checks that an endpoint can bind to a broadcast -// address and receive broadcast data on it. -func TestV4ReadOnBoundToBroadcast(t *testing.T) { - for _, flow := range []testFlow{broadcast, broadcastIn6} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to broadcast address. - bcastAddr := flow.mapAddrIfApplicable(broadcastAddr) - if err := c.ep.Bind(tcpip.FullAddress{Addr: bcastAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %s", err) - } - - // Test acceptance. - testRead(c, flow) - }) - } -} - -// testFailingWrite sends a packet of the given test flow into the UDP endpoint -// and verifies it fails with the provided error code. -func testFailingWrite(c *testContext, flow testFlow, wantErr *tcpip.Error) { - c.t.Helper() - - h := flow.header4Tuple(outgoing) - writeDstAddr := flow.mapAddrIfApplicable(h.dstAddr.Addr) - - payload := buffer.View(newPayload()) - _, _, gotErr := c.ep.Write(tcpip.SlicePayload(payload), tcpip.WriteOptions{ - To: &tcpip.FullAddress{Addr: writeDstAddr, Port: h.dstAddr.Port}, - }) - if gotErr != wantErr { - c.t.Fatalf("Write returned unexpected error: got %v, want %v", gotErr, wantErr) - } -} - -// testWrite sends a packet of the given test flow from the UDP endpoint to the -// flow's destination address:port. It then receives it from the link endpoint -// and verifies its correctness including any additional checker functions -// provided. -func testWrite(c *testContext, flow testFlow, checkers ...checker.NetworkChecker) uint16 { - c.t.Helper() - return testWriteInternal(c, flow, true, checkers...) -} - -// testWriteWithoutDestination sends a packet of the given test flow from the -// UDP endpoint without giving a destination address:port. It then receives it -// from the link endpoint and verifies its correctness including any additional -// checker functions provided. -func testWriteWithoutDestination(c *testContext, flow testFlow, checkers ...checker.NetworkChecker) uint16 { - c.t.Helper() - return testWriteInternal(c, flow, false, checkers...) -} - -func testWriteInternal(c *testContext, flow testFlow, setDest bool, checkers ...checker.NetworkChecker) uint16 { - c.t.Helper() - - writeOpts := tcpip.WriteOptions{} - if setDest { - h := flow.header4Tuple(outgoing) - writeDstAddr := flow.mapAddrIfApplicable(h.dstAddr.Addr) - writeOpts = tcpip.WriteOptions{ - To: &tcpip.FullAddress{Addr: writeDstAddr, Port: h.dstAddr.Port}, - } - } - payload := buffer.View(newPayload()) - n, _, err := c.ep.Write(tcpip.SlicePayload(payload), writeOpts) - if err != nil { - c.t.Fatalf("Write failed: %v", err) - } - if n != int64(len(payload)) { - c.t.Fatalf("Bad number of bytes written: got %v, want %v", n, len(payload)) - } - - // Received the packet and check the payload. - b := c.getPacketAndVerify(flow, checkers...) - var udp header.UDP - if flow.isV4() { - udp = header.UDP(header.IPv4(b).Payload()) - } else { - udp = header.UDP(header.IPv6(b).Payload()) - } - if !bytes.Equal(payload, udp.Payload()) { - c.t.Fatalf("Bad payload: got %x, want %x", udp.Payload(), payload) - } - - return udp.SourcePort() -} - -func testDualWrite(c *testContext) uint16 { - c.t.Helper() - - v4Port := testWrite(c, unicastV4in6) - v6Port := testWrite(c, unicastV6) - if v4Port != v6Port { - c.t.Fatalf("expected v4 and v6 ports to be equal: got v4Port = %d, v6Port = %d", v4Port, v6Port) - } - - return v4Port -} - -func TestDualWriteUnbound(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - testDualWrite(c) -} - -func TestDualWriteBoundToWildcard(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - // Bind to wildcard. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - p := testDualWrite(c) - if p != stackPort { - c.t.Fatalf("Bad port: got %v, want %v", p, stackPort) - } -} - -func TestDualWriteConnectedToV6(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - // Connect to v6 address. - if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - testWrite(c, unicastV6) - - // Write to V4 mapped address. - testFailingWrite(c, unicastV4in6, tcpip.ErrNetworkUnreachable) -} - -func TestDualWriteConnectedToV4Mapped(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - // Connect to v4 mapped address. - if err := c.ep.Connect(tcpip.FullAddress{Addr: testV4MappedAddr, Port: testPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - testWrite(c, unicastV4in6) - - // Write to v6 address. - testFailingWrite(c, unicastV6, tcpip.ErrInvalidEndpointState) -} - -func TestV4WriteOnV6Only(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(unicastV6Only) - - // Write to V4 mapped address. - testFailingWrite(c, unicastV4in6, tcpip.ErrNoRoute) -} - -func TestV6WriteOnBoundToV4Mapped(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - // Bind to v4 mapped address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: stackV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - // Write to v6 address. - testFailingWrite(c, unicastV6, tcpip.ErrInvalidEndpointState) -} - -func TestV6WriteOnConnected(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - // Connect to v6 address. - if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) - } - - testWriteWithoutDestination(c, unicastV6) -} - -func TestV4WriteOnConnected(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - // Connect to v4 mapped address. - if err := c.ep.Connect(tcpip.FullAddress{Addr: testV4MappedAddr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) - } - - testWriteWithoutDestination(c, unicastV4) -} - -// TestWriteOnBoundToV4Multicast checks that we can send packets out of a socket -// that is bound to a V4 multicast address. -func TestWriteOnBoundToV4Multicast(t *testing.T) { - for _, flow := range []testFlow{unicastV4, multicastV4, broadcast} { - t.Run(fmt.Sprintf("%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to V4 mcast address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: multicastAddr, Port: stackPort}); err != nil { - c.t.Fatal("Bind failed:", err) - } - - testWrite(c, flow) - }) - } -} - -// TestWriteOnBoundToV4MappedMulticast checks that we can send packets out of a -// socket that is bound to a V4-mapped multicast address. -func TestWriteOnBoundToV4MappedMulticast(t *testing.T) { - for _, flow := range []testFlow{unicastV4in6, multicastV4in6, broadcastIn6} { - t.Run(fmt.Sprintf("%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to V4Mapped mcast address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: multicastV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %s", err) - } - - testWrite(c, flow) - }) - } -} - -// TestWriteOnBoundToV6Multicast checks that we can send packets out of a -// socket that is bound to a V6 multicast address. -func TestWriteOnBoundToV6Multicast(t *testing.T) { - for _, flow := range []testFlow{unicastV6, multicastV6} { - t.Run(fmt.Sprintf("%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to V6 mcast address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: multicastV6Addr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %s", err) - } - - testWrite(c, flow) - }) - } -} - -// TestWriteOnBoundToV6Multicast checks that we can send packets out of a -// V6-only socket that is bound to a V6 multicast address. -func TestWriteOnBoundToV6OnlyMulticast(t *testing.T) { - for _, flow := range []testFlow{unicastV6Only, multicastV6Only} { - t.Run(fmt.Sprintf("%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to V6 mcast address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: multicastV6Addr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %s", err) - } - - testWrite(c, flow) - }) - } -} - -// TestWriteOnBoundToBroadcast checks that we can send packets out of a -// socket that is bound to the broadcast address. -func TestWriteOnBoundToBroadcast(t *testing.T) { - for _, flow := range []testFlow{unicastV4, multicastV4, broadcast} { - t.Run(fmt.Sprintf("%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to V4 broadcast address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: broadcastAddr, Port: stackPort}); err != nil { - c.t.Fatal("Bind failed:", err) - } - - testWrite(c, flow) - }) - } -} - -// TestWriteOnBoundToV4MappedBroadcast checks that we can send packets out of a -// socket that is bound to the V4-mapped broadcast address. -func TestWriteOnBoundToV4MappedBroadcast(t *testing.T) { - for _, flow := range []testFlow{unicastV4in6, multicastV4in6, broadcastIn6} { - t.Run(fmt.Sprintf("%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Bind to V4Mapped mcast address. - if err := c.ep.Bind(tcpip.FullAddress{Addr: broadcastV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %s", err) - } - - testWrite(c, flow) - }) - } -} - -func TestReadIncrementsPacketsReceived(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - // Create IPv4 UDP endpoint - c.createEndpoint(ipv6.ProtocolNumber) - - // Bind to wildcard. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) - } - - testRead(c, unicastV4) - - var want uint64 = 1 - if got := c.s.Stats().UDP.PacketsReceived.Value(); got != want { - c.t.Fatalf("Read did not increment PacketsReceived: got %v, want %v", got, want) - } -} - -func TestWriteIncrementsPacketsSent(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - testDualWrite(c) - - var want uint64 = 2 - if got := c.s.Stats().UDP.PacketsSent.Value(); got != want { - c.t.Fatalf("Write did not increment PacketsSent: got %v, want %v", got, want) - } -} - -func TestTTL(t *testing.T) { - for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - const multicastTTL = 42 - if err := c.ep.SetSockOpt(tcpip.MulticastTTLOption(multicastTTL)); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) - } - - var wantTTL uint8 - if flow.isMulticast() { - wantTTL = multicastTTL - } else { - var p stack.NetworkProtocol - if flow.isV4() { - p = ipv4.NewProtocol() - } else { - p = ipv6.NewProtocol() - } - ep, err := p.NewEndpoint(0, tcpip.AddressWithPrefix{}, nil, nil, nil) - if err != nil { - t.Fatal(err) - } - wantTTL = ep.DefaultTTL() - ep.Close() - } - - testWrite(c, flow, checker.TTL(wantTTL)) - }) - } -} - -func TestMulticastInterfaceOption(t *testing.T) { - for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - for _, bindTyp := range []string{"bound", "unbound"} { - t.Run(bindTyp, func(t *testing.T) { - for _, optTyp := range []string{"use local-addr", "use NICID", "use local-addr and NIC"} { - t.Run(optTyp, func(t *testing.T) { - h := flow.header4Tuple(outgoing) - mcastAddr := h.dstAddr.Addr - localIfAddr := h.srcAddr.Addr - - var ifoptSet tcpip.MulticastInterfaceOption - switch optTyp { - case "use local-addr": - ifoptSet.InterfaceAddr = localIfAddr - case "use NICID": - ifoptSet.NIC = 1 - case "use local-addr and NIC": - ifoptSet.InterfaceAddr = localIfAddr - ifoptSet.NIC = 1 - default: - t.Fatal("unknown test variant") - } - - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(flow.sockProto()) - - if bindTyp == "bound" { - // Bind the socket by connecting to the multicast address. - // This may have an influence on how the multicast interface - // is set. - addr := tcpip.FullAddress{ - Addr: flow.mapAddrIfApplicable(mcastAddr), - Port: stackPort, - } - if err := c.ep.Connect(addr); err != nil { - c.t.Fatalf("Connect failed: %v", err) - } - } - - if err := c.ep.SetSockOpt(ifoptSet); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) - } - - // Verify multicast interface addr and NIC were set correctly. - // Note that NIC must be 1 since this is our outgoing interface. - ifoptWant := tcpip.MulticastInterfaceOption{NIC: 1, InterfaceAddr: ifoptSet.InterfaceAddr} - var ifoptGot tcpip.MulticastInterfaceOption - if err := c.ep.GetSockOpt(&ifoptGot); err != nil { - c.t.Fatalf("GetSockOpt failed: %v", err) - } - if ifoptGot != ifoptWant { - c.t.Errorf("got GetSockOpt() = %#v, want = %#v", ifoptGot, ifoptWant) - } - }) - } - }) - } - }) - } -} - -// TestV4UnknownDestination verifies that we generate an ICMPv4 Destination -// Unreachable message when a udp datagram is received on ports for which there -// is no bound udp socket. -func TestV4UnknownDestination(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - testCases := []struct { - flow testFlow - icmpRequired bool - // largePayload if true, will result in a payload large enough - // so that the final generated IPv4 packet is larger than - // header.IPv4MinimumProcessableDatagramSize. - largePayload bool - }{ - {unicastV4, true, false}, - {unicastV4, true, true}, - {multicastV4, false, false}, - {multicastV4, false, true}, - {broadcast, false, false}, - {broadcast, false, true}, - } - for _, tc := range testCases { - t.Run(fmt.Sprintf("flow:%s icmpRequired:%t largePayload:%t", tc.flow, tc.icmpRequired, tc.largePayload), func(t *testing.T) { - payload := newPayload() - if tc.largePayload { - payload = newMinPayload(576) - } - c.injectPacket(tc.flow, payload) - if !tc.icmpRequired { - select { - case p := <-c.linkEP.C: - t.Fatalf("unexpected packet received: %+v", p) - case <-time.After(1 * time.Second): - return - } - } - - select { - case p := <-c.linkEP.C: - var pkt []byte - pkt = append(pkt, p.Header...) - pkt = append(pkt, p.Payload...) - if got, want := len(pkt), header.IPv4MinimumProcessableDatagramSize; got > want { - t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) - } - - hdr := header.IPv4(pkt) - checker.IPv4(t, hdr, checker.ICMPv4( - checker.ICMPv4Type(header.ICMPv4DstUnreachable), - checker.ICMPv4Code(header.ICMPv4PortUnreachable))) - - icmpPkt := header.ICMPv4(hdr.Payload()) - payloadIPHeader := header.IPv4(icmpPkt.Payload()) - wantLen := len(payload) - if tc.largePayload { - wantLen = header.IPv4MinimumProcessableDatagramSize - header.IPv4MinimumSize*2 - header.ICMPv4MinimumSize - header.UDPMinimumSize - } - - // In case of large payloads the IP packet may be truncated. Update - // the length field before retrieving the udp datagram payload. - payloadIPHeader.SetTotalLength(uint16(wantLen + header.UDPMinimumSize + header.IPv4MinimumSize)) - - origDgram := header.UDP(payloadIPHeader.Payload()) - if got, want := len(origDgram.Payload()), wantLen; got != want { - t.Fatalf("unexpected payload length got: %d, want: %d", got, want) - } - if got, want := origDgram.Payload(), payload[:wantLen]; !bytes.Equal(got, want) { - t.Fatalf("unexpected payload got: %d, want: %d", got, want) - } - case <-time.After(1 * time.Second): - t.Fatalf("packet wasn't written out") - } - }) - } -} - -// TestV6UnknownDestination verifies that we generate an ICMPv6 Destination -// Unreachable message when a udp datagram is received on ports for which there -// is no bound udp socket. -func TestV6UnknownDestination(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - testCases := []struct { - flow testFlow - icmpRequired bool - // largePayload if true will result in a payload large enough to - // create an IPv6 packet > header.IPv6MinimumMTU bytes. - largePayload bool - }{ - {unicastV6, true, false}, - {unicastV6, true, true}, - {multicastV6, false, false}, - {multicastV6, false, true}, - } - for _, tc := range testCases { - t.Run(fmt.Sprintf("flow:%s icmpRequired:%t largePayload:%t", tc.flow, tc.icmpRequired, tc.largePayload), func(t *testing.T) { - payload := newPayload() - if tc.largePayload { - payload = newMinPayload(1280) - } - c.injectPacket(tc.flow, payload) - if !tc.icmpRequired { - select { - case p := <-c.linkEP.C: - t.Fatalf("unexpected packet received: %+v", p) - case <-time.After(1 * time.Second): - return - } - } - - select { - case p := <-c.linkEP.C: - var pkt []byte - pkt = append(pkt, p.Header...) - pkt = append(pkt, p.Payload...) - if got, want := len(pkt), header.IPv6MinimumMTU; got > want { - t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) - } - - hdr := header.IPv6(pkt) - checker.IPv6(t, hdr, checker.ICMPv6( - checker.ICMPv6Type(header.ICMPv6DstUnreachable), - checker.ICMPv6Code(header.ICMPv6PortUnreachable))) - - icmpPkt := header.ICMPv6(hdr.Payload()) - payloadIPHeader := header.IPv6(icmpPkt.Payload()) - wantLen := len(payload) - if tc.largePayload { - wantLen = header.IPv6MinimumMTU - header.IPv6MinimumSize*2 - header.ICMPv6MinimumSize - header.UDPMinimumSize - } - // In case of large payloads the IP packet may be truncated. Update - // the length field before retrieving the udp datagram payload. - payloadIPHeader.SetPayloadLength(uint16(wantLen + header.UDPMinimumSize)) - - origDgram := header.UDP(payloadIPHeader.Payload()) - if got, want := len(origDgram.Payload()), wantLen; got != want { - t.Fatalf("unexpected payload length got: %d, want: %d", got, want) - } - if got, want := origDgram.Payload(), payload[:wantLen]; !bytes.Equal(got, want) { - t.Fatalf("unexpected payload got: %v, want: %v", got, want) - } - case <-time.After(1 * time.Second): - t.Fatalf("packet wasn't written out") - } - }) - } -} |