inet.af/netstack@v0.0.0-20220214151720-7585b01ddccf/tcpip/network/ipv4/icmp.go (about)

     1  // Copyright 2021 The gVisor Authors.
     2  //
     3  // Licensed under the Apache License, Version 2.0 (the "License");
     4  // you may not use this file except in compliance with the License.
     5  // You may obtain a copy of the License at
     6  //
     7  //     http://www.apache.org/licenses/LICENSE-2.0
     8  //
     9  // Unless required by applicable law or agreed to in writing, software
    10  // distributed under the License is distributed on an "AS IS" BASIS,
    11  // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    12  // See the License for the specific language governing permissions and
    13  // limitations under the License.
    14  
    15  package ipv4
    16  
    17  import (
    18  	"fmt"
    19  
    20  	"inet.af/netstack/tcpip"
    21  	"inet.af/netstack/tcpip/buffer"
    22  	"inet.af/netstack/tcpip/header"
    23  	"inet.af/netstack/tcpip/stack"
    24  )
    25  
    26  // icmpv4DestinationUnreachableSockError is a general ICMPv4 Destination
    27  // Unreachable error.
    28  //
    29  // +stateify savable
    30  type icmpv4DestinationUnreachableSockError struct{}
    31  
    32  // Origin implements tcpip.SockErrorCause.
    33  func (*icmpv4DestinationUnreachableSockError) Origin() tcpip.SockErrOrigin {
    34  	return tcpip.SockExtErrorOriginICMP
    35  }
    36  
    37  // Type implements tcpip.SockErrorCause.
    38  func (*icmpv4DestinationUnreachableSockError) Type() uint8 {
    39  	return uint8(header.ICMPv4DstUnreachable)
    40  }
    41  
    42  // Info implements tcpip.SockErrorCause.
    43  func (*icmpv4DestinationUnreachableSockError) Info() uint32 {
    44  	return 0
    45  }
    46  
    47  var _ stack.TransportError = (*icmpv4DestinationHostUnreachableSockError)(nil)
    48  
    49  // icmpv4DestinationHostUnreachableSockError is an ICMPv4 Destination Host
    50  // Unreachable error.
    51  //
    52  // It indicates that a packet was not able to reach the destination host.
    53  //
    54  // +stateify savable
    55  type icmpv4DestinationHostUnreachableSockError struct {
    56  	icmpv4DestinationUnreachableSockError
    57  }
    58  
    59  // Code implements tcpip.SockErrorCause.
    60  func (*icmpv4DestinationHostUnreachableSockError) Code() uint8 {
    61  	return uint8(header.ICMPv4HostUnreachable)
    62  }
    63  
    64  // Kind implements stack.TransportError.
    65  func (*icmpv4DestinationHostUnreachableSockError) Kind() stack.TransportErrorKind {
    66  	return stack.DestinationHostUnreachableTransportError
    67  }
    68  
    69  var _ stack.TransportError = (*icmpv4DestinationPortUnreachableSockError)(nil)
    70  
    71  // icmpv4DestinationPortUnreachableSockError is an ICMPv4 Destination Port
    72  // Unreachable error.
    73  //
    74  // It indicates that a packet reached the destination host, but the transport
    75  // protocol was not active on the destination port.
    76  //
    77  // +stateify savable
    78  type icmpv4DestinationPortUnreachableSockError struct {
    79  	icmpv4DestinationUnreachableSockError
    80  }
    81  
    82  // Code implements tcpip.SockErrorCause.
    83  func (*icmpv4DestinationPortUnreachableSockError) Code() uint8 {
    84  	return uint8(header.ICMPv4PortUnreachable)
    85  }
    86  
    87  // Kind implements stack.TransportError.
    88  func (*icmpv4DestinationPortUnreachableSockError) Kind() stack.TransportErrorKind {
    89  	return stack.DestinationPortUnreachableTransportError
    90  }
    91  
    92  var _ stack.TransportError = (*icmpv4FragmentationNeededSockError)(nil)
    93  
    94  // icmpv4FragmentationNeededSockError is an ICMPv4 Destination Unreachable error
    95  // due to fragmentation being required but the packet was set to not be
    96  // fragmented.
    97  //
    98  // It indicates that a link exists on the path to the destination with an MTU
    99  // that is too small to carry the packet.
   100  //
   101  // +stateify savable
   102  type icmpv4FragmentationNeededSockError struct {
   103  	icmpv4DestinationUnreachableSockError
   104  
   105  	mtu uint32
   106  }
   107  
   108  // Code implements tcpip.SockErrorCause.
   109  func (*icmpv4FragmentationNeededSockError) Code() uint8 {
   110  	return uint8(header.ICMPv4FragmentationNeeded)
   111  }
   112  
   113  // Info implements tcpip.SockErrorCause.
   114  func (e *icmpv4FragmentationNeededSockError) Info() uint32 {
   115  	return e.mtu
   116  }
   117  
   118  // Kind implements stack.TransportError.
   119  func (*icmpv4FragmentationNeededSockError) Kind() stack.TransportErrorKind {
   120  	return stack.PacketTooBigTransportError
   121  }
   122  
   123  func (e *endpoint) checkLocalAddress(addr tcpip.Address) bool {
   124  	if e.nic.Spoofing() {
   125  		return true
   126  	}
   127  
   128  	if addressEndpoint := e.AcquireAssignedAddress(addr, false, stack.NeverPrimaryEndpoint); addressEndpoint != nil {
   129  		addressEndpoint.DecRef()
   130  		return true
   131  	}
   132  	return false
   133  }
   134  
   135  // handleControl handles the case when an ICMP error packet contains the headers
   136  // of the original packet that caused the ICMP one to be sent. This information
   137  // is used to find out which transport endpoint must be notified about the ICMP
   138  // packet. We only expect the payload, not the enclosing ICMP packet.
   139  func (e *endpoint) handleControl(errInfo stack.TransportError, pkt *stack.PacketBuffer) {
   140  	h, ok := pkt.Data().PullUp(header.IPv4MinimumSize)
   141  	if !ok {
   142  		return
   143  	}
   144  	hdr := header.IPv4(h)
   145  
   146  	// We don't use IsValid() here because ICMP only requires that the IP
   147  	// header plus 8 bytes of the transport header be included. So it's
   148  	// likely that it is truncated, which would cause IsValid to return
   149  	// false.
   150  	//
   151  	// Drop packet if it doesn't have the basic IPv4 header or if the
   152  	// original source address doesn't match an address we own.
   153  	srcAddr := hdr.SourceAddress()
   154  	if !e.checkLocalAddress(srcAddr) {
   155  		return
   156  	}
   157  
   158  	hlen := int(hdr.HeaderLength())
   159  	if pkt.Data().Size() < hlen || hdr.FragmentOffset() != 0 {
   160  		// We won't be able to handle this if it doesn't contain the
   161  		// full IPv4 header, or if it's a fragment not at offset 0
   162  		// (because it won't have the transport header).
   163  		return
   164  	}
   165  
   166  	// Keep needed information before trimming header.
   167  	p := hdr.TransportProtocol()
   168  	dstAddr := hdr.DestinationAddress()
   169  	// Skip the ip header, then deliver the error.
   170  	if _, ok := pkt.Data().Consume(hlen); !ok {
   171  		panic(fmt.Sprintf("could not consume the IP header of %d bytes", hlen))
   172  	}
   173  	e.dispatcher.DeliverTransportError(srcAddr, dstAddr, ProtocolNumber, p, errInfo, pkt)
   174  }
   175  
   176  func (e *endpoint) handleICMP(pkt *stack.PacketBuffer) {
   177  	received := e.stats.icmp.packetsReceived
   178  	h := header.ICMPv4(pkt.TransportHeader().View())
   179  	if len(h) < header.ICMPv4MinimumSize {
   180  		received.invalid.Increment()
   181  		return
   182  	}
   183  
   184  	// Only do in-stack processing if the checksum is correct.
   185  	if header.Checksum(h, pkt.Data().AsRange().Checksum()) != 0xffff {
   186  		received.invalid.Increment()
   187  		// It's possible that a raw socket expects to receive this regardless
   188  		// of checksum errors. If it's an echo request we know it's safe because
   189  		// we are the only handler, however other types do not cope well with
   190  		// packets with checksum errors.
   191  		switch h.Type() {
   192  		case header.ICMPv4Echo:
   193  			e.dispatcher.DeliverTransportPacket(header.ICMPv4ProtocolNumber, pkt)
   194  		}
   195  		return
   196  	}
   197  
   198  	iph := header.IPv4(pkt.NetworkHeader().View())
   199  	var newOptions header.IPv4Options
   200  	if opts := iph.Options(); len(opts) != 0 {
   201  		// RFC 1122 section 3.2.2.6 (page 43) (and similar for other round trip
   202  		// type ICMP packets):
   203  		//    If a Record Route and/or Time Stamp option is received in an
   204  		//    ICMP Echo Request, this option (these options) SHOULD be
   205  		//    updated to include the current host and included in the IP
   206  		//    header of the Echo Reply message, without "truncation".
   207  		//    Thus, the recorded route will be for the entire round trip.
   208  		//
   209  		// So we need to let the option processor know how it should handle them.
   210  		var op optionsUsage
   211  		if h.Type() == header.ICMPv4Echo {
   212  			op = &optionUsageEcho{}
   213  		} else {
   214  			op = &optionUsageReceive{}
   215  		}
   216  		var optProblem *header.IPv4OptParameterProblem
   217  		newOptions, _, optProblem = e.processIPOptions(pkt, opts, op)
   218  		if optProblem != nil {
   219  			if optProblem.NeedICMP {
   220  				_ = e.protocol.returnError(&icmpReasonParamProblem{
   221  					pointer: optProblem.Pointer,
   222  				}, pkt)
   223  				e.stats.ip.MalformedPacketsReceived.Increment()
   224  			}
   225  			return
   226  		}
   227  		copied := copy(opts, newOptions)
   228  		if copied != len(newOptions) {
   229  			panic(fmt.Sprintf("copied %d bytes of new options, expected %d bytes", copied, len(newOptions)))
   230  		}
   231  		for i := copied; i < len(opts); i++ {
   232  			// Pad with 0 (EOL). RFC 791 page 23 says "The padding is zero".
   233  			opts[i] = byte(header.IPv4OptionListEndType)
   234  		}
   235  	}
   236  
   237  	// TODO(b/112892170): Meaningfully handle all ICMP types.
   238  	switch h.Type() {
   239  	case header.ICMPv4Echo:
   240  		received.echoRequest.Increment()
   241  
   242  		// DeliverTransportPacket may modify pkt so don't use it beyond
   243  		// this point. Make a deep copy of the data before pkt gets sent as we will
   244  		// be modifying fields. Both the ICMP header (with its type modified to
   245  		// EchoReply) and payload are reused in the reply packet.
   246  		//
   247  		// TODO(gvisor.dev/issue/4399): The copy may not be needed if there are no
   248  		// waiting endpoints. Consider moving responsibility for doing the copy to
   249  		// DeliverTransportPacket so that is is only done when needed.
   250  		replyData := stack.PayloadSince(pkt.TransportHeader())
   251  		ipHdr := header.IPv4(pkt.NetworkHeader().View())
   252  		localAddressBroadcast := pkt.NetworkPacketInfo.LocalAddressBroadcast
   253  
   254  		// It's possible that a raw socket expects to receive this.
   255  		e.dispatcher.DeliverTransportPacket(header.ICMPv4ProtocolNumber, pkt)
   256  		pkt = nil
   257  
   258  		// Take the base of the incoming request IP header but replace the options.
   259  		replyHeaderLength := uint8(header.IPv4MinimumSize + len(newOptions))
   260  		replyIPHdr := header.IPv4(append(iph[:header.IPv4MinimumSize:header.IPv4MinimumSize], newOptions...))
   261  		replyIPHdr.SetHeaderLength(replyHeaderLength)
   262  
   263  		// As per RFC 1122 section 3.2.1.3, when a host sends any datagram, the IP
   264  		// source address MUST be one of its own IP addresses (but not a broadcast
   265  		// or multicast address).
   266  		localAddr := ipHdr.DestinationAddress()
   267  		if localAddressBroadcast || header.IsV4MulticastAddress(localAddr) {
   268  			localAddr = ""
   269  		}
   270  
   271  		r, err := e.protocol.stack.FindRoute(e.nic.ID(), localAddr, ipHdr.SourceAddress(), ProtocolNumber, false /* multicastLoop */)
   272  		if err != nil {
   273  			// If we cannot find a route to the destination, silently drop the packet.
   274  			return
   275  		}
   276  		defer r.Release()
   277  
   278  		sent := e.stats.icmp.packetsSent
   279  		if !e.protocol.allowICMPReply(header.ICMPv4EchoReply, header.ICMPv4UnusedCode) {
   280  			sent.rateLimited.Increment()
   281  			return
   282  		}
   283  
   284  		// TODO(gvisor.dev/issue/3810:) When adding protocol numbers into the
   285  		// header information, we may have to change this code to handle the
   286  		// ICMP header no longer being in the data buffer.
   287  
   288  		// Because IP and ICMP are so closely intertwined, we need to handcraft our
   289  		// IP header to be able to follow RFC 792. The wording on page 13 is as
   290  		// follows:
   291  		//   IP Fields:
   292  		//   Addresses
   293  		//     The address of the source in an echo message will be the
   294  		//     destination of the echo reply message.  To form an echo reply
   295  		//     message, the source and destination addresses are simply reversed,
   296  		//     the type code changed to 0, and the checksum recomputed.
   297  		//
   298  		// This was interpreted by early implementors to mean that all options must
   299  		// be copied from the echo request IP header to the echo reply IP header
   300  		// and this behaviour is still relied upon by some applications.
   301  		//
   302  		// Create a copy of the IP header we received, options and all, and change
   303  		// The fields we need to alter.
   304  		//
   305  		// We need to produce the entire packet in the data segment in order to
   306  		// use WriteHeaderIncludedPacket(). WriteHeaderIncludedPacket sets the
   307  		// total length and the header checksum so we don't need to set those here.
   308  		replyIPHdr.SetSourceAddress(r.LocalAddress())
   309  		replyIPHdr.SetDestinationAddress(r.RemoteAddress())
   310  		replyIPHdr.SetTTL(r.DefaultTTL())
   311  
   312  		replyICMPHdr := header.ICMPv4(replyData)
   313  		replyICMPHdr.SetType(header.ICMPv4EchoReply)
   314  		replyICMPHdr.SetChecksum(0)
   315  		replyICMPHdr.SetChecksum(^header.Checksum(replyData, 0))
   316  
   317  		replyVV := buffer.View(replyIPHdr).ToVectorisedView()
   318  		replyVV.AppendView(replyData)
   319  		replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
   320  			ReserveHeaderBytes: int(r.MaxHeaderLength()),
   321  			Data:               replyVV,
   322  		})
   323  		defer replyPkt.DecRef()
   324  		replyPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber
   325  
   326  		if err := r.WriteHeaderIncludedPacket(replyPkt); err != nil {
   327  			sent.dropped.Increment()
   328  			return
   329  		}
   330  		sent.echoReply.Increment()
   331  
   332  	case header.ICMPv4EchoReply:
   333  		received.echoReply.Increment()
   334  
   335  		// ICMP sockets expect the ICMP header to be present, so we don't consume
   336  		// the ICMP header.
   337  		e.dispatcher.DeliverTransportPacket(header.ICMPv4ProtocolNumber, pkt)
   338  
   339  	case header.ICMPv4DstUnreachable:
   340  		received.dstUnreachable.Increment()
   341  
   342  		mtu := h.MTU()
   343  		code := h.Code()
   344  		switch code {
   345  		case header.ICMPv4HostUnreachable:
   346  			e.handleControl(&icmpv4DestinationHostUnreachableSockError{}, pkt)
   347  		case header.ICMPv4PortUnreachable:
   348  			e.handleControl(&icmpv4DestinationPortUnreachableSockError{}, pkt)
   349  		case header.ICMPv4FragmentationNeeded:
   350  			networkMTU, err := calculateNetworkMTU(uint32(mtu), header.IPv4MinimumSize)
   351  			if err != nil {
   352  				networkMTU = 0
   353  			}
   354  			e.handleControl(&icmpv4FragmentationNeededSockError{mtu: networkMTU}, pkt)
   355  		}
   356  	case header.ICMPv4SrcQuench:
   357  		received.srcQuench.Increment()
   358  
   359  	case header.ICMPv4Redirect:
   360  		received.redirect.Increment()
   361  
   362  	case header.ICMPv4TimeExceeded:
   363  		received.timeExceeded.Increment()
   364  
   365  	case header.ICMPv4ParamProblem:
   366  		received.paramProblem.Increment()
   367  
   368  	case header.ICMPv4Timestamp:
   369  		received.timestamp.Increment()
   370  
   371  	case header.ICMPv4TimestampReply:
   372  		received.timestampReply.Increment()
   373  
   374  	case header.ICMPv4InfoRequest:
   375  		received.infoRequest.Increment()
   376  
   377  	case header.ICMPv4InfoReply:
   378  		received.infoReply.Increment()
   379  
   380  	default:
   381  		received.invalid.Increment()
   382  	}
   383  }
   384  
   385  // ======= ICMP Error packet generation =========
   386  
   387  // icmpReason is a marker interface for IPv4 specific ICMP errors.
   388  type icmpReason interface {
   389  	isICMPReason()
   390  	// isForwarding indicates whether or not the error arose while attempting to
   391  	// forward a packet.
   392  	isForwarding() bool
   393  }
   394  
   395  // icmpReasonPortUnreachable is an error where the transport protocol has no
   396  // listener and no alternative means to inform the sender.
   397  type icmpReasonPortUnreachable struct{}
   398  
   399  func (*icmpReasonPortUnreachable) isICMPReason() {}
   400  func (*icmpReasonPortUnreachable) isForwarding() bool {
   401  	return false
   402  }
   403  
   404  // icmpReasonProtoUnreachable is an error where the transport protocol is
   405  // not supported.
   406  type icmpReasonProtoUnreachable struct{}
   407  
   408  func (*icmpReasonProtoUnreachable) isICMPReason() {}
   409  func (*icmpReasonProtoUnreachable) isForwarding() bool {
   410  	return false
   411  }
   412  
   413  // icmpReasonTTLExceeded is an error where a packet's time to live exceeded in
   414  // transit to its final destination, as per RFC 792 page 6, Time Exceeded
   415  // Message.
   416  type icmpReasonTTLExceeded struct{}
   417  
   418  func (*icmpReasonTTLExceeded) isICMPReason() {}
   419  func (*icmpReasonTTLExceeded) isForwarding() bool {
   420  	// If we hit a TTL Exceeded error, then we know we are operating as a router.
   421  	// As per RFC 792 page 6, Time Exceeded Message,
   422  	//
   423  	//   If the gateway processing a datagram finds the time to live field
   424  	//   is zero it must discard the datagram.  The gateway may also notify
   425  	//   the source host via the time exceeded message.
   426  	return true
   427  }
   428  
   429  // icmpReasonReassemblyTimeout is an error where insufficient fragments are
   430  // received to complete reassembly of a packet within a configured time after
   431  // the reception of the first-arriving fragment of that packet.
   432  type icmpReasonReassemblyTimeout struct{}
   433  
   434  func (*icmpReasonReassemblyTimeout) isICMPReason() {}
   435  func (*icmpReasonReassemblyTimeout) isForwarding() bool {
   436  	return false
   437  }
   438  
   439  // icmpReasonParamProblem is an error to use to request a Parameter Problem
   440  // message to be sent.
   441  type icmpReasonParamProblem struct {
   442  	pointer    byte
   443  	forwarding bool
   444  }
   445  
   446  func (*icmpReasonParamProblem) isICMPReason() {}
   447  func (r *icmpReasonParamProblem) isForwarding() bool {
   448  	return r.forwarding
   449  }
   450  
   451  // icmpReasonNetworkUnreachable is an error in which the network specified in
   452  // the internet destination field of the datagram is unreachable.
   453  type icmpReasonNetworkUnreachable struct{}
   454  
   455  func (*icmpReasonNetworkUnreachable) isICMPReason() {}
   456  func (*icmpReasonNetworkUnreachable) isForwarding() bool {
   457  	// If we hit a Net Unreachable error, then we know we are operating as
   458  	// a router. As per RFC 792 page 5, Destination Unreachable Message,
   459  	//
   460  	//  If, according to the information in the gateway's routing tables,
   461  	//  the network specified in the internet destination field of a
   462  	//  datagram is unreachable, e.g., the distance to the network is
   463  	//  infinity, the gateway may send a destination unreachable message to
   464  	//  the internet source host of the datagram.
   465  	return true
   466  }
   467  
   468  // icmpReasonFragmentationNeeded is an error where a packet requires
   469  // fragmentation while also having the Don't Fragment flag set, as per RFC 792
   470  // page 3, Destination Unreachable Message.
   471  type icmpReasonFragmentationNeeded struct{}
   472  
   473  func (*icmpReasonFragmentationNeeded) isICMPReason() {}
   474  func (*icmpReasonFragmentationNeeded) isForwarding() bool {
   475  	// If we hit a Don't Fragment error, then we know we are operating as a router.
   476  	// As per RFC 792 page 4, Destination Unreachable Message,
   477  	//
   478  	//   Another case is when a datagram must be fragmented to be forwarded by a
   479  	//   gateway yet the Don't Fragment flag is on. In this case the gateway must
   480  	//   discard the datagram and may return a destination unreachable message.
   481  	return true
   482  }
   483  
   484  // icmpReasonHostUnreachable is an error in which the host specified in the
   485  // internet destination field of the datagram is unreachable.
   486  type icmpReasonHostUnreachable struct{}
   487  
   488  func (*icmpReasonHostUnreachable) isICMPReason() {}
   489  func (*icmpReasonHostUnreachable) isForwarding() bool {
   490  	// If we hit a Host Unreachable error, then we know we are operating as a
   491  	// router. As per RFC 792 page 5, Destination Unreachable Message,
   492  	//
   493  	//   In addition, in some networks, the gateway may be able to determine
   494  	//   if the internet destination host is unreachable.  Gateways in these
   495  	//   networks may send destination unreachable messages to the source host
   496  	//   when the destination host is unreachable.
   497  	return true
   498  }
   499  
   500  // returnError takes an error descriptor and generates the appropriate ICMP
   501  // error packet for IPv4 and sends it back to the remote device that sent
   502  // the problematic packet. It incorporates as much of that packet as
   503  // possible as well as any error metadata as is available. returnError
   504  // expects pkt to hold a valid IPv4 packet as per the wire format.
   505  func (p *protocol) returnError(reason icmpReason, pkt *stack.PacketBuffer) tcpip.Error {
   506  	origIPHdr := header.IPv4(pkt.NetworkHeader().View())
   507  	origIPHdrSrc := origIPHdr.SourceAddress()
   508  	origIPHdrDst := origIPHdr.DestinationAddress()
   509  
   510  	// We check we are responding only when we are allowed to.
   511  	// See RFC 1812 section 4.3.2.7 (shown below).
   512  	//
   513  	// =========
   514  	// 4.3.2.7 When Not to Send ICMP Errors
   515  	//
   516  	//  An ICMP error message MUST NOT be sent as the result of receiving:
   517  	//
   518  	//  o An ICMP error message, or
   519  	//
   520  	//  o A packet which fails the IP header validation tests described in
   521  	//    Section [5.2.2] (except where that section specifically permits
   522  	//    the sending of an ICMP error message), or
   523  	//
   524  	//  o A packet destined to an IP broadcast or IP multicast address, or
   525  	//
   526  	//  o A packet sent as a Link Layer broadcast or multicast, or
   527  	//
   528  	//  o Any fragment of a datagram other then the first fragment (i.e., a
   529  	// packet for which the fragment offset in the IP header is nonzero).
   530  	//
   531  	// TODO(gvisor.dev/issues/4058): Make sure we don't send ICMP errors in
   532  	// response to a non-initial fragment, but it currently can not happen.
   533  	if pkt.NetworkPacketInfo.LocalAddressBroadcast || header.IsV4MulticastAddress(origIPHdrDst) || origIPHdrSrc == header.IPv4Any {
   534  		return nil
   535  	}
   536  
   537  	// If we are operating as a router/gateway, don't use the packet's destination
   538  	// address as the response's source address as we should not not own the
   539  	// destination address of a packet we are forwarding.
   540  	localAddr := origIPHdrDst
   541  	if reason.isForwarding() {
   542  		localAddr = ""
   543  	}
   544  
   545  	// Even if we were able to receive a packet from some remote, we may not have
   546  	// a route to it - the remote may be blocked via routing rules. We must always
   547  	// consult our routing table and find a route to the remote before sending any
   548  	// packet.
   549  	route, err := p.stack.FindRoute(pkt.NICID, localAddr, origIPHdrSrc, ProtocolNumber, false /* multicastLoop */)
   550  	if err != nil {
   551  		return err
   552  	}
   553  	defer route.Release()
   554  
   555  	p.mu.Lock()
   556  	// We retrieve an endpoint using the newly constructed route's NICID rather
   557  	// than the packet's NICID. The packet's NICID corresponds to the NIC on
   558  	// which it arrived, which isn't necessarily the same as the NIC on which it
   559  	// will be transmitted. On the other hand, the route's NIC *is* guaranteed
   560  	// to be the NIC on which the packet will be transmitted.
   561  	netEP, ok := p.mu.eps[route.NICID()]
   562  	p.mu.Unlock()
   563  	if !ok {
   564  		return &tcpip.ErrNotConnected{}
   565  	}
   566  
   567  	transportHeader := pkt.TransportHeader().View()
   568  
   569  	// Don't respond to icmp error packets.
   570  	if origIPHdr.Protocol() == uint8(header.ICMPv4ProtocolNumber) {
   571  		// We need to decide to explicitly name the packets we can respond to or
   572  		// the ones we can not respond to. The decision is somewhat arbitrary and
   573  		// if problems arise this could be reversed. It was judged less of a breach
   574  		// of protocol to not respond to unknown non-error packets than to respond
   575  		// to unknown error packets so we take the first approach.
   576  		if len(transportHeader) < header.ICMPv4MinimumSize {
   577  			// The packet is malformed.
   578  			return nil
   579  		}
   580  		switch header.ICMPv4(transportHeader).Type() {
   581  		case
   582  			header.ICMPv4EchoReply,
   583  			header.ICMPv4Echo,
   584  			header.ICMPv4Timestamp,
   585  			header.ICMPv4TimestampReply,
   586  			header.ICMPv4InfoRequest,
   587  			header.ICMPv4InfoReply:
   588  		default:
   589  			// Assume any type we don't know about may be an error type.
   590  			return nil
   591  		}
   592  	}
   593  
   594  	sent := netEP.stats.icmp.packetsSent
   595  	icmpType, icmpCode, counter, pointer := func() (header.ICMPv4Type, header.ICMPv4Code, tcpip.MultiCounterStat, byte) {
   596  		switch reason := reason.(type) {
   597  		case *icmpReasonPortUnreachable:
   598  			return header.ICMPv4DstUnreachable, header.ICMPv4PortUnreachable, sent.dstUnreachable, 0
   599  		case *icmpReasonProtoUnreachable:
   600  			return header.ICMPv4DstUnreachable, header.ICMPv4ProtoUnreachable, sent.dstUnreachable, 0
   601  		case *icmpReasonNetworkUnreachable:
   602  			return header.ICMPv4DstUnreachable, header.ICMPv4NetUnreachable, sent.dstUnreachable, 0
   603  		case *icmpReasonHostUnreachable:
   604  			return header.ICMPv4DstUnreachable, header.ICMPv4HostUnreachable, sent.dstUnreachable, 0
   605  		case *icmpReasonFragmentationNeeded:
   606  			return header.ICMPv4DstUnreachable, header.ICMPv4FragmentationNeeded, sent.dstUnreachable, 0
   607  		case *icmpReasonTTLExceeded:
   608  			return header.ICMPv4TimeExceeded, header.ICMPv4TTLExceeded, sent.timeExceeded, 0
   609  		case *icmpReasonReassemblyTimeout:
   610  			return header.ICMPv4TimeExceeded, header.ICMPv4ReassemblyTimeout, sent.timeExceeded, 0
   611  		case *icmpReasonParamProblem:
   612  			return header.ICMPv4ParamProblem, header.ICMPv4UnusedCode, sent.paramProblem, reason.pointer
   613  		default:
   614  			panic(fmt.Sprintf("unsupported ICMP type %T", reason))
   615  		}
   616  	}()
   617  
   618  	if !p.allowICMPReply(icmpType, icmpCode) {
   619  		sent.rateLimited.Increment()
   620  		return nil
   621  	}
   622  
   623  	// Now work out how much of the triggering packet we should return.
   624  	// As per RFC 1812 Section 4.3.2.3
   625  	//
   626  	//   ICMP datagram SHOULD contain as much of the original
   627  	//   datagram as possible without the length of the ICMP
   628  	//   datagram exceeding 576 bytes.
   629  	//
   630  	// NOTE: The above RFC referenced is different from the original
   631  	// recommendation in RFC 1122 and RFC 792 where it mentioned that at
   632  	// least 8 bytes of the payload must be included. Today linux and other
   633  	// systems implement the RFC 1812 definition and not the original
   634  	// requirement. We treat 8 bytes as the minimum but will try send more.
   635  	mtu := int(route.MTU())
   636  	const maxIPData = header.IPv4MinimumProcessableDatagramSize - header.IPv4MinimumSize
   637  	if mtu > maxIPData {
   638  		mtu = maxIPData
   639  	}
   640  	available := mtu - header.ICMPv4MinimumSize
   641  
   642  	if available < len(origIPHdr)+header.ICMPv4MinimumErrorPayloadSize {
   643  		return nil
   644  	}
   645  
   646  	payloadLen := len(origIPHdr) + transportHeader.Size() + pkt.Data().Size()
   647  	if payloadLen > available {
   648  		payloadLen = available
   649  	}
   650  
   651  	// The buffers used by pkt may be used elsewhere in the system.
   652  	// For example, an AF_RAW or AF_PACKET socket may use what the transport
   653  	// protocol considers an unreachable destination. Thus we deep copy pkt to
   654  	// prevent multiple ownership and SR errors. The new copy is a vectorized
   655  	// view with the entire incoming IP packet reassembled and truncated as
   656  	// required. This is now the payload of the new ICMP packet and no longer
   657  	// considered a packet in its own right.
   658  	newHeader := append(buffer.View(nil), origIPHdr...)
   659  	newHeader = append(newHeader, transportHeader...)
   660  	payload := newHeader.ToVectorisedView()
   661  	if dataCap := payloadLen - payload.Size(); dataCap > 0 {
   662  		payload.AppendView(pkt.Data().AsRange().Capped(dataCap).ToOwnedView())
   663  	} else {
   664  		payload.CapLength(payloadLen)
   665  	}
   666  
   667  	icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
   668  		ReserveHeaderBytes: int(route.MaxHeaderLength()) + header.ICMPv4MinimumSize,
   669  		Data:               payload,
   670  	})
   671  	defer icmpPkt.DecRef()
   672  
   673  	icmpPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber
   674  
   675  	icmpHdr := header.ICMPv4(icmpPkt.TransportHeader().Push(header.ICMPv4MinimumSize))
   676  	icmpHdr.SetCode(icmpCode)
   677  	icmpHdr.SetType(icmpType)
   678  	icmpHdr.SetPointer(pointer)
   679  	icmpHdr.SetChecksum(header.ICMPv4Checksum(icmpHdr, icmpPkt.Data().AsRange().Checksum()))
   680  
   681  	if err := route.WritePacket(
   682  		stack.NetworkHeaderParams{
   683  			Protocol: header.ICMPv4ProtocolNumber,
   684  			TTL:      route.DefaultTTL(),
   685  			TOS:      stack.DefaultTOS,
   686  		},
   687  		icmpPkt,
   688  	); err != nil {
   689  		sent.dropped.Increment()
   690  		return err
   691  	}
   692  	counter.Increment()
   693  	return nil
   694  }
   695  
   696  // OnReassemblyTimeout implements fragmentation.TimeoutHandler.
   697  func (p *protocol) OnReassemblyTimeout(pkt *stack.PacketBuffer) {
   698  	// OnReassemblyTimeout sends a Time Exceeded Message, as per RFC 792:
   699  	//
   700  	//   If a host reassembling a fragmented datagram cannot complete the
   701  	//   reassembly due to missing fragments within its time limit it discards the
   702  	//   datagram, and it may send a time exceeded message.
   703  	//
   704  	//   If fragment zero is not available then no time exceeded need be sent at
   705  	//   all.
   706  	if pkt != nil {
   707  		p.returnError(&icmpReasonReassemblyTimeout{}, pkt)
   708  	}
   709  }