diff options
Diffstat (limited to 'pkg/tcpip/tcpip.go')
| -rw-r--r-- | pkg/tcpip/tcpip.go | 194 |
1 files changed, 123 insertions, 71 deletions
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 0e944712f..1ca4088c9 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -323,11 +323,26 @@ type ControlMessages struct { // TOS is the IPv4 type of service of the associated packet. TOS uint8 - // HasTClass indicates whether Tclass is valid/set. + // HasTClass indicates whether TClass is valid/set. HasTClass bool - // Tclass is the IPv6 traffic class of the associated packet. - TClass int32 + // TClass is the IPv6 traffic class of the associated packet. + TClass uint32 + + // HasIPPacketInfo indicates whether PacketInfo is set. + HasIPPacketInfo bool + + // PacketInfo holds interface and address data on an incoming packet. + PacketInfo IPPacketInfo +} + +// PacketOwner is used to get UID and GID of the packet. +type PacketOwner interface { + // UID returns UID of the packet. + UID() uint32 + + // GID returns GID of the packet. + GID() uint32 } // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp) @@ -335,9 +350,15 @@ type ControlMessages struct { // networking stack. type Endpoint interface { // Close puts the endpoint in a closed state and frees all resources - // associated with it. + // associated with it. Close initiates the teardown process, the + // Endpoint may not be fully closed when Close returns. Close() + // Abort initiates an expedited endpoint teardown. As compared to + // Close, Abort prioritizes closing the Endpoint quickly over cleanly. + // Abort is best effort; implementing Abort with Close is acceptable. + Abort() + // Read reads data from the endpoint and optionally returns the sender. // // This method does not block if there is no data pending. It will also @@ -458,6 +479,9 @@ type Endpoint interface { // Stats returns a reference to the endpoint stats. Stats() EndpointStats + + // SetOwner sets the task owner to the endpoint owner. + SetOwner(owner PacketOwner) } // EndpointInfo is the interface implemented by each endpoint info struct. @@ -496,9 +520,56 @@ type WriteOptions struct { type SockOptBool int const ( + // BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether + // datagram sockets are allowed to send packets to a broadcast address. + BroadcastOption SockOptBool = iota + + // CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be + // held until segments are full by the TCP transport protocol. + CorkOption + + // DelayOption is used by SetSockOpt/GetSockOpt to specify if data + // should be sent out immediately by the transport protocol. For TCP, + // it determines if the Nagle algorithm is on or off. + DelayOption + + // KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether + // TCP keepalive is enabled for this socket. + KeepaliveEnabledOption + + // MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether + // multicast packets sent over a non-loopback interface will be looped back. + MulticastLoopOption + + // PasscredOption is used by SetSockOpt/GetSockOpt to specify whether + // SCM_CREDENTIALS socket control messages are enabled. + // + // Only supported on Unix sockets. + PasscredOption + + // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. + QuickAckOption + + // ReceiveTClassOption is used by SetSockOpt/GetSockOpt to specify if the + // IPV6_TCLASS ancillary message is passed with incoming packets. + ReceiveTClassOption + // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS // ancillary message is passed with incoming packets. - ReceiveTOSOption SockOptBool = iota + ReceiveTOSOption + + // ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify + // if more inforamtion is provided with incoming packets such + // as interface index and address. + ReceiveIPPacketInfoOption + + // ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() + // should allow reuse of local address. + ReuseAddressOption + + // ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets + // to be bound to an identical socket address. + ReusePortOption // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 // socket is to be restricted to sending and receiving IPv6 packets only. @@ -509,9 +580,30 @@ const ( type SockOptInt int const ( + // KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number + // of un-ACKed TCP keepalives that will be sent before the connection is + // closed. + KeepaliveCountOption SockOptInt = iota + + // IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS + // for all subsequent outgoing IPv4 packets from the endpoint. + IPv4TOSOption + + // IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS + // for all subsequent outgoing IPv6 packets from the endpoint. + IPv6TrafficClassOption + + // MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current + // Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. + MaxSegOption + + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default + // TTL value for multicast messages. The default is 1. + MulticastTTLOption + // ReceiveQueueSizeOption is used in GetSockOptInt to specify that the // number of unread bytes in the input buffer should be returned. - ReceiveQueueSizeOption SockOptInt = iota + ReceiveQueueSizeOption // SendBufferSizeOption is used by SetSockOptInt/GetSockOptInt to // specify the send buffer size option. @@ -525,44 +617,21 @@ const ( // number of unread bytes in the output buffer should be returned. SendQueueSizeOption - // DelayOption is used by SetSockOpt/GetSockOpt to specify if data - // should be sent out immediately by the transport protocol. For TCP, - // it determines if the Nagle algorithm is on or off. - DelayOption - - // TODO(b/137664753): convert all int socket options to be handled via - // GetSockOptInt. + // TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop + // limit value for unicast messages. The default is protocol specific. + // + // A zero value indicates the default. + TTLOption ) // ErrorOption is used in GetSockOpt to specify that the last error reported by // the endpoint should be cleared and returned. type ErrorOption struct{} -// CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be -// held until segments are full by the TCP transport protocol. -type CorkOption int - -// ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() -// should allow reuse of local address. -type ReuseAddressOption int - -// ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets -// to be bound to an identical socket address. -type ReusePortOption int - // BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets // should bind only on a specific NIC. type BindToDeviceOption NICID -// QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. -type QuickAckOption int - -// PasscredOption is used by SetSockOpt/GetSockOpt to specify whether -// SCM_CREDENTIALS socket control messages are enabled. -// -// Only supported on Unix sockets. -type PasscredOption int - // TCPInfoOption is used by GetSockOpt to expose TCP statistics. // // TODO(b/64800844): Add and populate stat fields. @@ -571,10 +640,6 @@ type TCPInfoOption struct { RTTVar time.Duration } -// KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether -// TCP keepalive is enabled for this socket. -type KeepaliveEnabledOption int - // KeepaliveIdleOption is used by SetSockOpt/GetSockOpt to specify the time a // connection must remain idle before the first TCP keepalive packet is sent. // Once this time is reached, KeepaliveIntervalOption is used instead. @@ -584,11 +649,6 @@ type KeepaliveIdleOption time.Duration // interval between sending TCP keepalive packets. type KeepaliveIntervalOption time.Duration -// KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number -// of un-ACKed TCP keepalives that will be sent before the connection is -// closed. -type KeepaliveCountOption int - // TCPUserTimeoutOption is used by SetSockOpt/GetSockOpt to specify a user // specified timeout for a given TCP connection. // See: RFC5482 for details. @@ -602,20 +662,9 @@ type CongestionControlOption string // control algorithms. type AvailableCongestionControlOption string -// ModerateReceiveBufferOption allows the caller to enable/disable TCP receive // buffer moderation. type ModerateReceiveBufferOption bool -// MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current -// Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. -type MaxSegOption int - -// TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop -// limit value for unicast messages. The default is protocol specific. -// -// A zero value indicates the default. -type TTLOption uint8 - // TCPLingerTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the // maximum duration for which a socket lingers in the TCP_FIN_WAIT_2 state // before being marked closed. @@ -632,9 +681,14 @@ type TCPTimeWaitTimeoutOption time.Duration // for a handshake till the specified timeout until a segment with data arrives. type TCPDeferAcceptOption time.Duration -// MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default -// TTL value for multicast messages. The default is 1. -type MulticastTTLOption uint8 +// TCPMinRTOOption is use by SetSockOpt/GetSockOpt to allow overriding +// default MinRTO used by the Stack. +type TCPMinRTOOption time.Duration + +// TCPSynRcvdCountThresholdOption is used by SetSockOpt/GetSockOpt to specify +// the number of endpoints that can be in SYN-RCVD state before the stack +// switches to using SYN cookies. +type TCPSynRcvdCountThresholdOption uint64 // MulticastInterfaceOption is used by SetSockOpt/GetSockOpt to specify a // default interface for multicast. @@ -643,10 +697,6 @@ type MulticastInterfaceOption struct { InterfaceAddr Address } -// MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether -// multicast packets sent over a non-loopback interface will be looped back. -type MulticastLoopOption bool - // MembershipOption is used by SetSockOpt/GetSockOpt as an argument to // AddMembershipOption and RemoveMembershipOption. type MembershipOption struct { @@ -669,21 +719,23 @@ type RemoveMembershipOption MembershipOption // TCP out-of-band data is delivered along with the normal in-band data. type OutOfBandInlineOption int -// BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether -// datagram sockets are allowed to send packets to a broadcast address. -type BroadcastOption int - // DefaultTTLOption is used by stack.(*Stack).NetworkProtocolOption to specify // a default TTL. type DefaultTTLOption uint8 -// IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS -// for all subsequent outgoing IPv4 packets from the endpoint. -type IPv4TOSOption uint8 +// IPPacketInfo is the message struture for IP_PKTINFO. +// +// +stateify savable +type IPPacketInfo struct { + // NIC is the ID of the NIC to be used. + NIC NICID + + // LocalAddr is the local address. + LocalAddr Address -// IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS -// for all subsequent outgoing IPv6 packets from the endpoint. -type IPv6TrafficClassOption uint8 + // DestinationAddr is the destination address. + DestinationAddr Address +} // Route is a row in the routing table. It specifies through which NIC (and // gateway) sets of packets should be routed. A row is considered viable if the |