diff options
author | gVisor bot <gvisor-bot@google.com> | 2020-06-18 13:09:58 +0000 |
---|---|---|
committer | gVisor bot <gvisor-bot@google.com> | 2020-06-18 13:09:58 +0000 |
commit | 356e6b7c281e4da83b1b2e0316b3131f2833ed19 (patch) | |
tree | 8089b06024da825f8147905ae5ddac21b7b55e14 | |
parent | 359e4c56571e18e9e6b2618cdd46293aa271666c (diff) | |
parent | 07ff909e76d8233827e705476ec116fc2cecec2f (diff) |
Merge release-20200608.0-82-g07ff909e7 (automated)
-rw-r--r-- | pkg/abi/linux/linux_abi_autogen_unsafe.go | 10 | ||||
-rw-r--r-- | pkg/sentry/socket/netstack/stack.go | 12 | ||||
-rw-r--r-- | pkg/tcpip/tcpip.go | 26 | ||||
-rw-r--r-- | pkg/tcpip/transport/raw/endpoint.go | 72 | ||||
-rw-r--r-- | pkg/tcpip/transport/raw/raw_state_autogen.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/connect.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/endpoint.go | 16 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/endpoint_state.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/protocol.go | 59 | ||||
-rw-r--r-- | pkg/tcpip/transport/udp/endpoint.go | 57 | ||||
-rw-r--r-- | pkg/tcpip/transport/udp/protocol.go | 70 | ||||
-rw-r--r-- | pkg/tcpip/transport/udp/udp_state_autogen.go | 2 | ||||
-rw-r--r-- | runsc/boot/loader.go | 6 |
13 files changed, 256 insertions, 80 deletions
diff --git a/pkg/abi/linux/linux_abi_autogen_unsafe.go b/pkg/abi/linux/linux_abi_autogen_unsafe.go index 8bc0ec04c..ff38e37d3 100644 --- a/pkg/abi/linux/linux_abi_autogen_unsafe.go +++ b/pkg/abi/linux/linux_abi_autogen_unsafe.go @@ -124,12 +124,12 @@ func (s *Statx) UnmarshalBytes(src []byte) { // Packed implements marshal.Marshallable.Packed. //go:nosplit func (s *Statx) Packed() bool { - return s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() + return s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() } // MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. func (s *Statx) MarshalUnsafe(dst []byte) { - if s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { + if s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() { safecopy.CopyIn(dst, unsafe.Pointer(s)) } else { s.MarshalBytes(dst) @@ -138,7 +138,7 @@ func (s *Statx) MarshalUnsafe(dst []byte) { // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. func (s *Statx) UnmarshalUnsafe(src []byte) { - if s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() { + if s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { safecopy.CopyOut(unsafe.Pointer(s), src) } else { s.UnmarshalBytes(src) @@ -178,7 +178,7 @@ func (s *Statx) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) { // CopyIn implements marshal.Marshallable.CopyIn. //go:nosplit func (s *Statx) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { - if !s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() { + if !s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { // Type Statx doesn't have a packed layout in memory, fall back to UnmarshalBytes. buf := task.CopyScratchBuffer(s.SizeBytes()) // escapes: okay. length, err := task.CopyInBytes(addr, buf) // escapes: okay. @@ -204,7 +204,7 @@ func (s *Statx) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { // WriteTo implements io.WriterTo.WriteTo. func (s *Statx) WriteTo(w io.Writer) (int64, error) { - if !s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { + if !s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() { // Type Statx doesn't have a packed layout in memory, fall back to MarshalBytes. buf := make([]byte, s.SizeBytes()) s.MarshalBytes(buf) diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 9b44c2b89..f97f9b6f3 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -144,7 +144,7 @@ func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { // TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { - var rs tcp.ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &rs) return inet.TCPBufferSize{ Min: rs.Min, @@ -155,7 +155,7 @@ func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { // SetTCPReceiveBufferSize implements inet.Stack.SetTCPReceiveBufferSize. func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { - rs := tcp.ReceiveBufferSizeOption{ + rs := tcpip.StackReceiveBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -165,7 +165,7 @@ func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { // TCPSendBufferSize implements inet.Stack.TCPSendBufferSize. func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { - var ss tcp.SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &ss) return inet.TCPBufferSize{ Min: ss.Min, @@ -176,7 +176,7 @@ func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { // SetTCPSendBufferSize implements inet.Stack.SetTCPSendBufferSize. func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { - ss := tcp.SendBufferSizeOption{ + ss := tcpip.StackSendBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -186,14 +186,14 @@ func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { // TCPSACKEnabled implements inet.Stack.TCPSACKEnabled. func (s *Stack) TCPSACKEnabled() (bool, error) { - var sack tcp.SACKEnabled + var sack tcpip.StackSACKEnabled err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &sack) return bool(sack), syserr.TranslateNetstackError(err).ToError() } // SetTCPSACKEnabled implements inet.Stack.SetTCPSACKEnabled. func (s *Stack) SetTCPSACKEnabled(enabled bool) error { - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enabled))).ToError() + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(enabled))).ToError() } // Statistics implements inet.Stack.Statistics. diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index b7b227328..3ad130b23 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -813,6 +813,32 @@ type OutOfBandInlineOption int // a default TTL. type DefaultTTLOption uint8 +// StackSACKEnabled is used by stack.(*Stack).TransportProtocolOption to +// enable/disable SACK support in TCP. See: https://tools.ietf.org/html/rfc2018. +type StackSACKEnabled bool + +// StackDelayEnabled is used by stack.(Stack*).TransportProtocolOption to +// enable/disable Nagle's algorithm in TCP. +type StackDelayEnabled bool + +// StackSendBufferSizeOption is used by stack.(Stack*).TransportProtocolOption +// to get/set the default, min and max send buffer sizes. +type StackSendBufferSizeOption struct { + Min int + Default int + Max int +} + +// StackReceiveBufferSizeOption is used by +// stack.(Stack*).TransportProtocolOption to get/set the default, min and max +// receive buffer sizes. +type StackReceiveBufferSizeOption struct { + Min int + Default int + Max int +} + +// // IPPacketInfo is the message struture for IP_PKTINFO. // // +stateify savable diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index a406d815e..6a7977259 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -26,6 +26,8 @@ package raw import ( + "fmt" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -66,16 +68,17 @@ type endpoint struct { // protected by rcvMu. rcvMu sync.Mutex `state:"nosave"` rcvList rawPacketList - rcvBufSizeMax int `state:".(int)"` rcvBufSize int + rcvBufSizeMax int `state:".(int)"` rcvClosed bool // The following fields are protected by mu. - mu sync.RWMutex `state:"nosave"` - sndBufSize int - closed bool - connected bool - bound bool + mu sync.RWMutex `state:"nosave"` + sndBufSize int + sndBufSizeMax int + closed bool + connected bool + bound bool // route is the route to a remote network endpoint. It is set via // Connect(), and is valid only when conneted is true. route stack.Route `state:"manual"` @@ -103,10 +106,21 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt }, waiterQueue: waiterQueue, rcvBufSizeMax: 32 * 1024, - sndBufSize: 32 * 1024, + sndBufSizeMax: 32 * 1024, associated: associated, } + // Override with stack defaults. + var ss tcpip.StackSendBufferSizeOption + if err := s.TransportProtocolOption(transProto, &ss); err == nil { + e.sndBufSizeMax = ss.Default + } + + var rs tcpip.StackReceiveBufferSizeOption + if err := s.TransportProtocolOption(transProto, &rs); err == nil { + e.rcvBufSizeMax = rs.Default + } + // Unassociated endpoints are write-only and users call Write() with IP // headers included. Because they're write-only, We don't need to // register with the stack. @@ -523,7 +537,46 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.SendBufferSizeOption: + // Make sure the send buffer size is within the min and max + // allowed. + var ss tcpip.StackSendBufferSizeOption + if err := e.stack.TransportProtocolOption(e.TransProto, &ss); err != nil { + panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, ss, err)) + } + if v > ss.Max { + v = ss.Max + } + if v < ss.Min { + v = ss.Min + } + e.mu.Lock() + e.sndBufSizeMax = v + e.mu.Unlock() + return nil + + case tcpip.ReceiveBufferSizeOption: + // Make sure the receive buffer size is within the min and max + // allowed. + var rs tcpip.StackReceiveBufferSizeOption + if err := e.stack.TransportProtocolOption(e.TransProto, &rs); err != nil { + panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, rs, err)) + } + if v > rs.Max { + v = rs.Max + } + if v < rs.Min { + v = rs.Min + } + e.rcvMu.Lock() + e.rcvBufSizeMax = v + e.rcvMu.Unlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. @@ -563,7 +616,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { case tcpip.SendBufferSizeOption: e.mu.Lock() - v := e.sndBufSize + v := e.sndBufSizeMax e.mu.Unlock() return v, nil @@ -636,7 +689,6 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() - e.rcvMu.Unlock() e.stats.PacketsReceived.Increment() // Notify waiters that there's data to be read. diff --git a/pkg/tcpip/transport/raw/raw_state_autogen.go b/pkg/tcpip/transport/raw/raw_state_autogen.go index c755fa847..875ebda36 100644 --- a/pkg/tcpip/transport/raw/raw_state_autogen.go +++ b/pkg/tcpip/transport/raw/raw_state_autogen.go @@ -36,6 +36,7 @@ func (x *endpoint) save(m state.Map) { m.Save("rcvBufSize", &x.rcvBufSize) m.Save("rcvClosed", &x.rcvClosed) m.Save("sndBufSize", &x.sndBufSize) + m.Save("sndBufSizeMax", &x.sndBufSizeMax) m.Save("closed", &x.closed) m.Save("connected", &x.connected) m.Save("bound", &x.bound) @@ -50,6 +51,7 @@ func (x *endpoint) load(m state.Map) { m.Load("rcvBufSize", &x.rcvBufSize) m.Load("rcvClosed", &x.rcvClosed) m.Load("sndBufSize", &x.sndBufSize) + m.Load("sndBufSizeMax", &x.sndBufSizeMax) m.Load("closed", &x.closed) m.Load("connected", &x.connected) m.Load("bound", &x.bound) diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 9d4dce826..377643b82 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -521,7 +521,7 @@ func (h *handshake) execute() *tcpip.Error { s.AddWaker(&h.ep.newSegmentWaker, wakerForNewSegment) defer s.Done() - var sackEnabled SACKEnabled + var sackEnabled tcpip.StackSACKEnabled if err := h.ep.stack.TransportProtocolOption(ProtocolNumber, &sackEnabled); err != nil { // If stack returned an error when checking for SACKEnabled // status then just default to switching off SACK negotiation. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index f225b00e7..10df2bcd5 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -851,12 +851,12 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue maxSynRetries: DefaultSynRetries, } - var ss SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { e.sndBufSize = ss.Default } - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { e.rcvBufSize = rs.Default } @@ -871,7 +871,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.rcvAutoParams.disabled = !bool(mrb) } - var de DelayEnabled + var de tcpip.StackDelayEnabled if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { e.SetSockOptBool(tcpip.DelayOption, true) } @@ -1588,7 +1588,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min { v = rs.Min @@ -1638,7 +1638,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if v < ss.Min { v = ss.Min @@ -1678,7 +1678,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return tcpip.ErrInvalidOptionValue } } - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min/2 { v = rs.Min / 2 @@ -2609,7 +2609,7 @@ func (e *endpoint) receiveBufferSize() int { } func (e *endpoint) maxReceiveBufferSize() int { - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { // As a fallback return the hardcoded max buffer size. return MaxBufferSize @@ -2690,7 +2690,7 @@ func timeStampOffset() uint32 { // if the SYN options indicate that the SACK option was negotiated and the TCP // stack is configured to enable TCP SACK option. func (e *endpoint) maybeEnableSACKPermitted(synOpts *header.TCPSynOptions) { - var v SACKEnabled + var v tcpip.StackSACKEnabled if err := e.stack.TransportProtocolOption(ProtocolNumber, &v); err != nil { // Stack doesn't support SACK. So just return. return diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index cbb779666..0bebec2d1 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -186,7 +186,7 @@ func (e *endpoint) Resume(s *stack.Stack) { epState := e.origEndpointState switch epState { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: - var ss SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if e.sndBufSize < ss.Min || e.sndBufSize > ss.Max { panic(fmt.Sprintf("endpoint.sndBufSize %d is outside the min and max allowed [%d, %d]", e.sndBufSize, ss.Min, ss.Max)) diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 73b8a6782..3cff55afa 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -71,29 +71,6 @@ const ( DefaultSynRetries = 6 ) -// SACKEnabled option can be used to enable SACK support in the TCP -// protocol. See: https://tools.ietf.org/html/rfc2018. -type SACKEnabled bool - -// DelayEnabled option can be used to enable Nagle's algorithm in the TCP protocol. -type DelayEnabled bool - -// SendBufferSizeOption allows the default, min and max send buffer sizes for -// TCP endpoints to be queried or configured. -type SendBufferSizeOption struct { - Min int - Default int - Max int -} - -// ReceiveBufferSizeOption allows the default, min and max receive buffer size -// for TCP endpoints to be queried or configured. -type ReceiveBufferSizeOption struct { - Min int - Default int - Max int -} - const ( ccReno = "reno" ccCubic = "cubic" @@ -160,8 +137,8 @@ type protocol struct { mu sync.RWMutex sackEnabled bool delayEnabled bool - sendBufferSize SendBufferSizeOption - recvBufferSize ReceiveBufferSizeOption + sendBufferSize tcpip.StackSendBufferSizeOption + recvBufferSize tcpip.StackReceiveBufferSizeOption congestionControl string availableCongestionControl []string moderateReceiveBuffer bool @@ -272,19 +249,19 @@ func replyWithReset(s *segment, tos, ttl uint8) { // SetOption implements stack.TransportProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { switch v := option.(type) { - case SACKEnabled: + case tcpip.StackSACKEnabled: p.mu.Lock() p.sackEnabled = bool(v) p.mu.Unlock() return nil - case DelayEnabled: + case tcpip.StackDelayEnabled: p.mu.Lock() p.delayEnabled = bool(v) p.mu.Unlock() return nil - case SendBufferSizeOption: + case tcpip.StackSendBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -293,7 +270,7 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil - case ReceiveBufferSizeOption: + case tcpip.StackReceiveBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -386,25 +363,25 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { // Option implements stack.TransportProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { switch v := option.(type) { - case *SACKEnabled: + case *tcpip.StackSACKEnabled: p.mu.RLock() - *v = SACKEnabled(p.sackEnabled) + *v = tcpip.StackSACKEnabled(p.sackEnabled) p.mu.RUnlock() return nil - case *DelayEnabled: + case *tcpip.StackDelayEnabled: p.mu.RLock() - *v = DelayEnabled(p.delayEnabled) + *v = tcpip.StackDelayEnabled(p.delayEnabled) p.mu.RUnlock() return nil - case *SendBufferSizeOption: + case *tcpip.StackSendBufferSizeOption: p.mu.RLock() *v = p.sendBufferSize p.mu.RUnlock() return nil - case *ReceiveBufferSizeOption: + case *tcpip.StackReceiveBufferSizeOption: p.mu.RLock() *v = p.recvBufferSize p.mu.RUnlock() @@ -514,8 +491,16 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a TCP transport protocol. func NewProtocol() stack.TransportProtocol { return &protocol{ - sendBufferSize: SendBufferSizeOption{MinBufferSize, DefaultSendBufferSize, MaxBufferSize}, - recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, + sendBufferSize: tcpip.StackSendBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultSendBufferSize, + Max: MaxBufferSize, + }, + recvBufferSize: tcpip.StackReceiveBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultReceiveBufferSize, + Max: MaxBufferSize, + }, congestionControl: ccReno, availableCongestionControl: []string{ccReno, ccCubic}, tcpLingerTimeout: DefaultTCPLingerTimeout, diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index df5efbf6a..f51988047 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -15,6 +15,8 @@ package udp import ( + "fmt" + "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -94,6 +96,7 @@ type endpoint struct { // The following fields are protected by the mu mutex. mu sync.RWMutex `state:"nosave"` sndBufSize int + sndBufSizeMax int state EndpointState route stack.Route `state:"manual"` dstPort uint16 @@ -159,7 +162,7 @@ type multicastMembership struct { } func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint { - return &endpoint{ + e := &endpoint{ stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{ NetProto: netProto, @@ -181,10 +184,23 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue multicastTTL: 1, multicastLoop: true, rcvBufSizeMax: 32 * 1024, - sndBufSize: 32 * 1024, + sndBufSizeMax: 32 * 1024, state: StateInitial, uniqueID: s.UniqueID(), } + + // Override with stack defaults. + var ss tcpip.StackSendBufferSizeOption + if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { + e.sndBufSizeMax = ss.Default + } + + var rs tcpip.StackReceiveBufferSizeOption + if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + e.rcvBufSizeMax = rs.Default + } + + return e } // UniqueID implements stack.TransportEndpoint.UniqueID. @@ -611,8 +627,43 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.mu.Unlock() case tcpip.ReceiveBufferSizeOption: + // Make sure the receive buffer size is within the min and max + // allowed. + var rs tcpip.StackReceiveBufferSizeOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, rs, err)) + } + + if v < rs.Min { + v = rs.Min + } + if v > rs.Max { + v = rs.Max + } + + e.mu.Lock() + e.rcvBufSizeMax = v + e.mu.Unlock() + return nil case tcpip.SendBufferSizeOption: + // Make sure the send buffer size is within the min and max + // allowed. + var ss tcpip.StackSendBufferSizeOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, ss, err)) + } + + if v < ss.Min { + v = ss.Min + } + if v > ss.Max { + v = ss.Max + } + e.mu.Lock() + e.sndBufSizeMax = v + e.mu.Unlock() + return nil } return nil @@ -861,7 +912,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { case tcpip.SendBufferSizeOption: e.mu.Lock() - v := e.sndBufSize + v := e.sndBufSizeMax e.mu.Unlock() return v, nil diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 4218e7d03..fc93f93c0 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -21,6 +21,7 @@ package udp import ( + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -32,9 +33,27 @@ import ( const ( // ProtocolNumber is the udp protocol number. ProtocolNumber = header.UDPProtocolNumber + + // MinBufferSize is the smallest size of a receive or send buffer. + MinBufferSize = 4 << 10 // 4KiB bytes. + + // DefaultSendBufferSize is the default size of the send buffer for + // an endpoint. + DefaultSendBufferSize = 32 << 10 // 32KiB + + // DefaultReceiveBufferSize is the default size of the receive buffer + // for an endpoint. + DefaultReceiveBufferSize = 32 << 10 // 32KiB + + // MaxBufferSize is the largest size a receive/send buffer can grow to. + MaxBufferSize = 4 << 20 // 4MiB ) -type protocol struct{} +type protocol struct { + mu sync.RWMutex + sendBufferSize tcpip.StackSendBufferSizeOption + recvBufferSize tcpip.StackReceiveBufferSizeOption +} // Number returns the udp protocol number. func (*protocol) Number() tcpip.TransportProtocolNumber { @@ -183,13 +202,49 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } // SetOption implements stack.TransportProtocol.SetOption. -func (*protocol) SetOption(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption +func (p *protocol) SetOption(option interface{}) *tcpip.Error { + switch v := option.(type) { + case tcpip.StackSendBufferSizeOption: + if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + p.mu.Lock() + p.sendBufferSize = v + p.mu.Unlock() + return nil + + case tcpip.StackReceiveBufferSizeOption: + if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + p.mu.Lock() + p.recvBufferSize = v + p.mu.Unlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // Option implements stack.TransportProtocol.Option. -func (*protocol) Option(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption +func (p *protocol) Option(option interface{}) *tcpip.Error { + switch v := option.(type) { + case *tcpip.StackSendBufferSizeOption: + p.mu.RLock() + *v = p.sendBufferSize + p.mu.RUnlock() + return nil + + case *tcpip.StackReceiveBufferSizeOption: + p.mu.RLock() + *v = p.recvBufferSize + p.mu.RUnlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // Close implements stack.TransportProtocol.Close. @@ -212,5 +267,8 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a UDP transport protocol. func NewProtocol() stack.TransportProtocol { - return &protocol{} + return &protocol{ + sendBufferSize: tcpip.StackSendBufferSizeOption{Min: MinBufferSize, Default: DefaultSendBufferSize, Max: MaxBufferSize}, + recvBufferSize: tcpip.StackReceiveBufferSizeOption{Min: MinBufferSize, Default: DefaultReceiveBufferSize, Max: MaxBufferSize}, + } } diff --git a/pkg/tcpip/transport/udp/udp_state_autogen.go b/pkg/tcpip/transport/udp/udp_state_autogen.go index a5c31e4fd..0dba38a58 100644 --- a/pkg/tcpip/transport/udp/udp_state_autogen.go +++ b/pkg/tcpip/transport/udp/udp_state_autogen.go @@ -43,6 +43,7 @@ func (x *endpoint) save(m state.Map) { m.Save("rcvBufSize", &x.rcvBufSize) m.Save("rcvClosed", &x.rcvClosed) m.Save("sndBufSize", &x.sndBufSize) + m.Save("sndBufSizeMax", &x.sndBufSizeMax) m.Save("state", &x.state) m.Save("dstPort", &x.dstPort) m.Save("v6only", &x.v6only) @@ -75,6 +76,7 @@ func (x *endpoint) load(m state.Map) { m.Load("rcvBufSize", &x.rcvBufSize) m.Load("rcvClosed", &x.rcvClosed) m.Load("sndBufSize", &x.sndBufSize) + m.Load("sndBufSizeMax", &x.sndBufSizeMax) m.Load("state", &x.state) m.Load("dstPort", &x.dstPort) m.Load("v6only", &x.v6only) diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index b05a8bd45..c6efcdc83 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -1058,8 +1058,8 @@ func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (in })} // Enable SACK Recovery. - if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(true)); err != nil { - return nil, fmt.Errorf("failed to enable SACK: %v", err) + if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(true)); err != nil { + return nil, fmt.Errorf("failed to enable SACK: %s", err) } // Set default TTLs as required by socket/netstack. @@ -1068,7 +1068,7 @@ func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (in // Enable Receive Buffer Auto-Tuning. if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { - return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) + return nil, fmt.Errorf("SetTransportProtocolOption failed: %s", err) } s.FillIPTablesMetadata() |