diff options
Diffstat (limited to 'pkg/tcpip/tcpip.go')
| -rw-r--r-- | pkg/tcpip/tcpip.go | 357 |
1 files changed, 221 insertions, 136 deletions
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 3ab2b7654..f798056c0 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -31,6 +31,7 @@ package tcpip import ( "errors" "fmt" + "io" "math/bits" "reflect" "strconv" @@ -39,7 +40,6 @@ import ( "time" "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/waiter" ) @@ -49,8 +49,9 @@ const ipv4AddressSize = 4 // Error represents an error in the netstack error space. Using a special type // ensures that errors outside of this space are not accidentally introduced. // -// Note: to support save / restore, it is important that all tcpip errors have -// distinct error messages. +// All errors must have unique msg strings. +// +// +stateify savable type Error struct { msg string @@ -112,6 +113,7 @@ var ( ErrNotPermitted = &Error{msg: "operation not permitted"} ErrAddressFamilyNotSupported = &Error{msg: "address family not supported by protocol"} ErrMalformedHeader = &Error{msg: "header is malformed"} + ErrBadBuffer = &Error{msg: "bad buffer"} ) var messageToError map[string]*Error @@ -161,6 +163,7 @@ func StringToError(s string) *Error { ErrNotPermitted, ErrAddressFamilyNotSupported, ErrMalformedHeader, + ErrBadBuffer, } messageToError = make(map[string]*Error) @@ -247,6 +250,54 @@ func (a Address) WithPrefix() AddressWithPrefix { } } +// Unspecified returns true if the address is unspecified. +func (a Address) Unspecified() bool { + for _, b := range a { + if b != 0 { + return false + } + } + return true +} + +// MatchingPrefix returns the matching prefix length in bits. +// +// Panics if b and a have different lengths. +func (a Address) MatchingPrefix(b Address) uint8 { + const bitsInAByte = 8 + + if len(a) != len(b) { + panic(fmt.Sprintf("addresses %s and %s do not have the same length", a, b)) + } + + var prefix uint8 + for i := range a { + aByte := a[i] + bByte := b[i] + + if aByte == bByte { + prefix += bitsInAByte + continue + } + + // Count the remaining matching bits in the byte from MSbit to LSBbit. + mask := uint8(1) << (bitsInAByte - 1) + for { + if aByte&mask == bByte&mask { + prefix++ + mask >>= 1 + continue + } + + break + } + + break + } + + return prefix +} + // AddressMask is a bitmask for an address. type AddressMask string @@ -447,6 +498,21 @@ func (s SlicePayload) Payload(size int) ([]byte, *Error) { return s[:size], nil } +var _ io.Writer = (*SliceWriter)(nil) + +// SliceWriter implements io.Writer for slices. +type SliceWriter []byte + +// Write implements io.Writer.Write. +func (s *SliceWriter) Write(b []byte) (int, error) { + n := copy(*s, b) + *s = (*s)[n:] + if n < len(b) { + return n, io.ErrShortWrite + } + return n, nil +} + // A ControlMessages contains socket control messages for IP sockets. // // +stateify savable @@ -481,6 +547,17 @@ type ControlMessages struct { // PacketInfo holds interface and address data on an incoming packet. PacketInfo IPPacketInfo + + // HasOriginalDestinationAddress indicates whether OriginalDstAddress is + // set. + HasOriginalDstAddress bool + + // OriginalDestinationAddress holds the original destination address + // and port of the incoming packet. + OriginalDstAddress FullAddress + + // SockErr is the dequeued socket error on recvmsg(MSG_ERRQUEUE). + SockErr *SockError } // PacketOwner is used to get UID and GID of the packet. @@ -492,6 +569,40 @@ type PacketOwner interface { GID() uint32 } +// ReadOptions contains options for Endpoint.Read. +type ReadOptions struct { + // Peek indicates whether this read is a peek. + Peek bool + + // NeedRemoteAddr indicates whether to return the remote address, if + // supported. + NeedRemoteAddr bool + + // NeedLinkPacketInfo indicates whether to return the link-layer information, + // if supported. + NeedLinkPacketInfo bool +} + +// ReadResult represents result for a successful Endpoint.Read. +type ReadResult struct { + // Count is the number of bytes received and written to the buffer. + Count int + + // Total is the number of bytes of the received packet. This can be used to + // determine whether the read is truncated. + Total int + + // ControlMessages is the control messages received. + ControlMessages ControlMessages + + // RemoteAddr is the remote address if ReadOptions.NeedAddr is true. + RemoteAddr FullAddress + + // LinkPacketInfo is the link-layer information of the received packet if + // ReadOptions.NeedLinkPacketInfo is true. + LinkPacketInfo LinkPacketInfo +} + // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp) // that exposes functionality like read, write, connect, etc. to users of the // networking stack. @@ -506,11 +617,15 @@ type Endpoint interface { // Abort is best effort; implementing Abort with Close is acceptable. Abort() - // Read reads data from the endpoint and optionally returns the sender. + // Read reads data from the endpoint and optionally writes to dst. + // + // This method does not block if there is no data pending; in this case, + // ErrWouldBlock is returned. // - // This method does not block if there is no data pending. It will also - // either return an error or data, never both. - Read(*FullAddress) (buffer.View, ControlMessages, *Error) + // If non-zero number of bytes are successfully read and written to dst, err + // must be nil. Otherwise, if dst failed to write anything, ErrBadBuffer + // should be returned. + Read(dst io.Writer, count int, opts ReadOptions) (res ReadResult, err *Error) // Write writes data to the endpoint's peer. This method does not block if // the data cannot be written. @@ -532,11 +647,6 @@ type Endpoint interface { // not). The channel is only non-nil in this case. Write(Payloader, WriteOptions) (int64, <-chan struct{}, *Error) - // Peek reads data without consuming it from the endpoint. - // - // This method does not block if there is no data pending. - Peek([][]byte) (int64, ControlMessages, *Error) - // Connect connects the endpoint to its peer. Specifying a NIC is // optional. // @@ -593,10 +703,6 @@ type Endpoint interface { // SetSockOpt sets a socket option. SetSockOpt(opt SettableSocketOption) *Error - // SetSockOptBool sets a socket option, for simple cases where a value - // has the bool type. - SetSockOptBool(opt SockOptBool, v bool) *Error - // SetSockOptInt sets a socket option, for simple cases where a value // has the int type. SetSockOptInt(opt SockOptInt, v int) *Error @@ -604,10 +710,6 @@ type Endpoint interface { // GetSockOpt gets a socket option. GetSockOpt(opt GettableSocketOption) *Error - // GetSockOptBool gets a socket option for simple cases where a return - // value has the bool type. - GetSockOptBool(SockOptBool) (bool, *Error) - // GetSockOptInt gets a socket option for simple cases where a return // value has the int type. GetSockOptInt(SockOptInt) (int, *Error) @@ -634,6 +736,10 @@ type Endpoint interface { // LastError clears and returns the last error reported by the endpoint. LastError() *Error + + // SocketOptions returns the structure which contains all the socket + // level options. + SocketOptions() *SocketOptions } // LinkPacketInfo holds Link layer information for a received packet. @@ -647,17 +753,6 @@ type LinkPacketInfo struct { PktType PacketType } -// PacketEndpoint are additional methods that are only implemented by Packet -// endpoints. -type PacketEndpoint interface { - // ReadPacket reads a datagram/packet from the endpoint and optionally - // returns the sender and additional LinkPacketInfo. - // - // This method does not block if there is no data pending. It will also - // either return an error or data, never both. - ReadPacket(*FullAddress, *LinkPacketInfo) (buffer.View, ControlMessages, *Error) -} - // EndpointInfo is the interface implemented by each endpoint info struct. type EndpointInfo interface { // IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo @@ -690,84 +785,6 @@ type WriteOptions struct { Atomic bool } -// SockOptBool represents socket options which values have the bool type. -type SockOptBool int - -const ( - // BroadcastOption is used by SetSockOptBool/GetSockOptBool to specify - // whether datagram sockets are allowed to send packets to a broadcast - // address. - BroadcastOption SockOptBool = iota - - // CorkOption is used by SetSockOptBool/GetSockOptBool to specify if - // data should be held until segments are full by the TCP transport - // protocol. - CorkOption - - // DelayOption is used by SetSockOptBool/GetSockOptBool 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 SetSockOptBool/GetSockOptBool to - // specify whether TCP keepalive is enabled for this socket. - KeepaliveEnabledOption - - // MulticastLoopOption is used by SetSockOptBool/GetSockOptBool to - // specify whether multicast packets sent over a non-loopback interface - // will be looped back. - MulticastLoopOption - - // NoChecksumOption is used by SetSockOptBool/GetSockOptBool to specify - // whether UDP checksum is disabled for this socket. - NoChecksumOption - - // PasscredOption is used by SetSockOptBool/GetSockOptBool to specify - // whether SCM_CREDENTIALS socket control messages are enabled. - // - // Only supported on Unix sockets. - PasscredOption - - // QuickAckOption is stubbed out in SetSockOptBool/GetSockOptBool. - QuickAckOption - - // ReceiveTClassOption is used by SetSockOptBool/GetSockOptBool to - // specify if the IPV6_TCLASS ancillary message is passed with incoming - // packets. - ReceiveTClassOption - - // ReceiveTOSOption is used by SetSockOptBool/GetSockOptBool to specify - // if the TOS ancillary message is passed with incoming packets. - ReceiveTOSOption - - // ReceiveIPPacketInfoOption is used by SetSockOptBool/GetSockOptBool to - // specify if more inforamtion is provided with incoming packets such as - // interface index and address. - ReceiveIPPacketInfoOption - - // ReuseAddressOption is used by SetSockOptBool/GetSockOptBool to - // specify whether Bind() should allow reuse of local address. - ReuseAddressOption - - // ReusePortOption is used by SetSockOptBool/GetSockOptBool to permit - // multiple sockets to be bound to an identical socket address. - ReusePortOption - - // V6OnlyOption is used by SetSockOptBool/GetSockOptBool to specify - // whether an IPv6 socket is to be restricted to sending and receiving - // IPv6 packets only. - V6OnlyOption - - // IPHdrIncludedOption is used by SetSockOpt to indicate for a raw - // endpoint that all packets being written have an IP header and the - // endpoint should not attach an IP header. - IPHdrIncludedOption - - // AcceptConnOption is used by GetSockOptBool to indicate if the - // socket is a listening socket. - AcceptConnOption -) - // SockOptInt represents socket options which values have the int type. type SockOptInt int @@ -977,14 +994,6 @@ type SettableSocketOption interface { isSettableSocketOption() } -// BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets -// should bind only on a specific NIC. -type BindToDeviceOption NICID - -func (*BindToDeviceOption) isGettableSocketOption() {} - -func (*BindToDeviceOption) isSettableSocketOption() {} - // TCPInfoOption is used by GetSockOpt to expose TCP statistics. // // TODO(b/64800844): Add and populate stat fields. @@ -1159,14 +1168,6 @@ type RemoveMembershipOption MembershipOption func (*RemoveMembershipOption) isSettableSocketOption() {} -// OutOfBandInlineOption is used by SetSockOpt/GetSockOpt to specify whether -// TCP out-of-band data is delivered along with the normal in-band data. -type OutOfBandInlineOption int - -func (*OutOfBandInlineOption) isGettableSocketOption() {} - -func (*OutOfBandInlineOption) isSettableSocketOption() {} - // SocketDetachFilterOption is used by SetSockOpt to detach a previously attached // classic BPF filter on a given endpoint. type SocketDetachFilterOption int @@ -1216,10 +1217,6 @@ type LingerOption struct { Timeout time.Duration } -func (*LingerOption) isGettableSocketOption() {} - -func (*LingerOption) isSettableSocketOption() {} - // IPPacketInfo is the message structure for IP_PKTINFO. // // +stateify savable @@ -1390,6 +1387,18 @@ type ICMPv6PacketStats struct { // RedirectMsg is the total number of ICMPv6 redirect message packets // counted. RedirectMsg *StatCounter + + // MulticastListenerQuery is the total number of Multicast Listener Query + // messages counted. + MulticastListenerQuery *StatCounter + + // MulticastListenerReport is the total number of Multicast Listener Report + // messages counted. + MulticastListenerReport *StatCounter + + // MulticastListenerDone is the total number of Multicast Listener Done + // messages counted. + MulticastListenerDone *StatCounter } // ICMPv4SentPacketStats collects outbound ICMPv4-specific stats. @@ -1431,6 +1440,10 @@ type ICMPv6SentPacketStats struct { type ICMPv6ReceivedPacketStats struct { ICMPv6PacketStats + // Unrecognized is the total number of ICMPv6 packets received that the + // transport layer does not know how to parse. + Unrecognized *StatCounter + // Invalid is the total number of ICMPv6 packets received that the // transport layer could not parse. Invalid *StatCounter @@ -1440,33 +1453,102 @@ type ICMPv6ReceivedPacketStats struct { RouterOnlyPacketsDroppedByHost *StatCounter } -// ICMPStats collects ICMP-specific stats (both v4 and v6). -type ICMPStats struct { +// ICMPv4Stats collects ICMPv4-specific stats. +type ICMPv4Stats struct { // ICMPv4SentPacketStats contains counts of sent packets by ICMPv4 packet type // and a single count of packets which failed to write to the link // layer. - V4PacketsSent ICMPv4SentPacketStats + PacketsSent ICMPv4SentPacketStats // ICMPv4ReceivedPacketStats contains counts of received packets by ICMPv4 // packet type and a single count of invalid packets received. - V4PacketsReceived ICMPv4ReceivedPacketStats + PacketsReceived ICMPv4ReceivedPacketStats +} +// ICMPv6Stats collects ICMPv6-specific stats. +type ICMPv6Stats struct { // ICMPv6SentPacketStats contains counts of sent packets by ICMPv6 packet type // and a single count of packets which failed to write to the link // layer. - V6PacketsSent ICMPv6SentPacketStats + PacketsSent ICMPv6SentPacketStats // ICMPv6ReceivedPacketStats contains counts of received packets by ICMPv6 // packet type and a single count of invalid packets received. - V6PacketsReceived ICMPv6ReceivedPacketStats + PacketsReceived ICMPv6ReceivedPacketStats +} + +// ICMPStats collects ICMP-specific stats (both v4 and v6). +type ICMPStats struct { + // V4 contains the ICMPv4-specifics stats. + V4 ICMPv4Stats + + // V6 contains the ICMPv4-specifics stats. + V6 ICMPv6Stats +} + +// IGMPPacketStats enumerates counts for all IGMP packet types. +type IGMPPacketStats struct { + // MembershipQuery is the total number of Membership Query messages counted. + MembershipQuery *StatCounter + + // V1MembershipReport is the total number of Version 1 Membership Report + // messages counted. + V1MembershipReport *StatCounter + + // V2MembershipReport is the total number of Version 2 Membership Report + // messages counted. + V2MembershipReport *StatCounter + + // LeaveGroup is the total number of Leave Group messages counted. + LeaveGroup *StatCounter +} + +// IGMPSentPacketStats collects outbound IGMP-specific stats. +type IGMPSentPacketStats struct { + IGMPPacketStats + + // Dropped is the total number of IGMP packets dropped. + Dropped *StatCounter +} + +// IGMPReceivedPacketStats collects inbound IGMP-specific stats. +type IGMPReceivedPacketStats struct { + IGMPPacketStats + + // Invalid is the total number of IGMP packets received that IGMP could not + // parse. + Invalid *StatCounter + + // ChecksumErrors is the total number of IGMP packets dropped due to bad + // checksums. + ChecksumErrors *StatCounter + + // Unrecognized is the total number of unrecognized messages counted, these + // are silently ignored for forward-compatibilty. + Unrecognized *StatCounter +} + +// IGMPStats colelcts IGMP-specific stats. +type IGMPStats struct { + // IGMPSentPacketStats contains counts of sent packets by IGMP packet type + // and a single count of invalid packets received. + PacketsSent IGMPSentPacketStats + + // IGMPReceivedPacketStats contains counts of received packets by IGMP packet + // type and a single count of invalid packets received. + PacketsReceived IGMPReceivedPacketStats } // IPStats collects IP-specific stats (both v4 and v6). type IPStats struct { // PacketsReceived is the total number of IP packets received from the - // link layer in nic.DeliverNetworkPacket. + // link layer. PacketsReceived *StatCounter + // DisabledPacketsReceived is the total number of IP packets received from the + // link layer when the IP layer is disabled. + DisabledPacketsReceived *StatCounter + // InvalidDestinationAddressesReceived is the total number of IP packets // received with an unknown or invalid destination address. InvalidDestinationAddressesReceived *StatCounter @@ -1662,6 +1744,9 @@ type Stats struct { // ICMP breaks out ICMP-specific stats (both v4 and v6). ICMP ICMPStats + // IGMP breaks out IGMP-specific stats. + IGMP IGMPStats + // IP breaks out IP-specific stats (both v4 and v6). IP IPStats |