diff options
-rw-r--r-- | pkg/tcpip/buffer/view.go | 18 | ||||
-rw-r--r-- | pkg/tcpip/header/ipv4.go | 71 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/icmp.go | 84 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/ipv4.go | 2 |
4 files changed, 133 insertions, 42 deletions
diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index ea0c5413d..8db70a700 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -84,8 +84,8 @@ type VectorisedView struct { size int } -// NewVectorisedView creates a new vectorised view from an already-allocated slice -// of View and sets its size. +// NewVectorisedView creates a new vectorised view from an already-allocated +// slice of View and sets its size. func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } @@ -170,8 +170,9 @@ func (vv *VectorisedView) CapLength(length int) { } // Clone returns a clone of this VectorisedView. -// If the buffer argument is large enough to contain all the Views of this VectorisedView, -// the method will avoid allocations and use the buffer to store the Views of the clone. +// If the buffer argument is large enough to contain all the Views of this +// VectorisedView, the method will avoid allocations and use the buffer to +// store the Views of the clone. func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } @@ -209,7 +210,8 @@ func (vv *VectorisedView) PullUp(count int) (View, bool) { return newFirst, true } -// Size returns the size in bytes of the entire content stored in the vectorised view. +// Size returns the size in bytes of the entire content stored in the +// vectorised view. func (vv *VectorisedView) Size() int { return vv.size } @@ -222,6 +224,12 @@ func (vv *VectorisedView) ToView() View { if len(vv.views) == 1 { return vv.views[0] } + return vv.ToOwnedView() +} + +// ToOwnedView returns a single view containing the content of the vectorised +// view that vv does not own. +func (vv *VectorisedView) ToOwnedView() View { u := make([]byte, 0, vv.size) for _, v := range vv.views { u = append(u, v...) diff --git a/pkg/tcpip/header/ipv4.go b/pkg/tcpip/header/ipv4.go index b07d9991d..4c6e4be64 100644 --- a/pkg/tcpip/header/ipv4.go +++ b/pkg/tcpip/header/ipv4.go @@ -16,10 +16,29 @@ package header import ( "encoding/binary" + "fmt" "gvisor.dev/gvisor/pkg/tcpip" ) +// RFC 971 defines the fields of the IPv4 header on page 11 using the following +// diagram: ("Figure 4") +// 0 1 2 3 +// 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// |Version| IHL |Type of Service| Total Length | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Identification |Flags| Fragment Offset | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Time to Live | Protocol | Header Checksum | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Source Address | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Destination Address | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Options | Padding | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// const ( versIHL = 0 tos = 1 @@ -33,6 +52,7 @@ const ( checksum = 10 srcAddr = 12 dstAddr = 16 + options = 20 ) // IPv4Fields contains the fields of an IPv4 packet. It is used to describe the @@ -76,7 +96,8 @@ type IPv4Fields struct { // IPv4 represents an ipv4 header stored in a byte array. // Most of the methods of IPv4 access to the underlying slice without // checking the boundaries and could panic because of 'index out of range'. -// Always call IsValid() to validate an instance of IPv4 before using other methods. +// Always call IsValid() to validate an instance of IPv4 before using other +// methods. type IPv4 []byte const ( @@ -151,13 +172,44 @@ func IPVersion(b []byte) int { if len(b) < versIHL+1 { return -1 } - return int(b[versIHL] >> 4) + return int(b[versIHL] >> ipVersionShift) } +// RFC 791 page 11 shows the header length (IHL) is in the lower 4 bits +// of the first byte, and is counted in multiples of 4 bytes. +// +// 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// |Version| IHL |Type of Service| Total Length | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// (...) +// Version: 4 bits +// The Version field indicates the format of the internet header. This +// document describes version 4. +// +// IHL: 4 bits +// Internet Header Length is the length of the internet header in 32 +// bit words, and thus points to the beginning of the data. Note that +// the minimum value for a correct header is 5. +// +const ( + ipVersionShift = 4 + ipIHLMask = 0x0f + IPv4IHLStride = 4 +) + // HeaderLength returns the value of the "header length" field of the ipv4 // header. The length returned is in bytes. func (b IPv4) HeaderLength() uint8 { - return (b[versIHL] & 0xf) * 4 + return (b[versIHL] & ipIHLMask) * IPv4IHLStride +} + +// SetHeaderLength sets the value of the "Internet Header Length" field. +func (b IPv4) SetHeaderLength(hdrLen uint8) { + if hdrLen > IPv4MaximumHeaderSize { + panic(fmt.Sprintf("got IPv4 Header size = %d, want <= %d", hdrLen, IPv4MaximumHeaderSize)) + } + b[versIHL] = (IPv4Version << ipVersionShift) | ((hdrLen / IPv4IHLStride) & ipIHLMask) } // ID returns the value of the identifier field of the ipv4 header. @@ -211,6 +263,12 @@ func (b IPv4) DestinationAddress() tcpip.Address { return tcpip.Address(b[dstAddr : dstAddr+IPv4AddressSize]) } +// Options returns a a buffer holding the options. +func (b IPv4) Options() []byte { + hdrLen := b.HeaderLength() + return b[options:hdrLen:hdrLen] +} + // TransportProtocol implements Network.TransportProtocol. func (b IPv4) TransportProtocol() tcpip.TransportProtocolNumber { return tcpip.TransportProtocolNumber(b.Protocol()) @@ -236,6 +294,11 @@ func (b IPv4) SetTOS(v uint8, _ uint32) { b[tos] = v } +// SetTTL sets the "Time to Live" field of the IPv4 header. +func (b IPv4) SetTTL(v byte) { + b[ttl] = v +} + // SetTotalLength sets the "total length" field of the ipv4 header. func (b IPv4) SetTotalLength(totalLength uint16) { binary.BigEndian.PutUint16(b[IPv4TotalLenOffset:], totalLength) @@ -276,7 +339,7 @@ func (b IPv4) CalculateChecksum() uint16 { // Encode encodes all the fields of the ipv4 header. func (b IPv4) Encode(i *IPv4Fields) { - b[versIHL] = (4 << 4) | ((i.IHL / 4) & 0xf) + b.SetHeaderLength(i.IHL) b[tos] = i.TOS b.SetTotalLength(i.TotalLength) binary.BigEndian.PutUint16(b[id:], i.ID) diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index a8985ff5d..eab9a530c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -79,27 +79,27 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { received.Echo.Increment() // Only send a reply if the checksum is valid. - wantChecksum := h.Checksum() - // Reset the checksum field to 0 to can calculate the proper - // checksum. We'll have to reset this before we hand the packet - // off. + headerChecksum := h.Checksum() h.SetChecksum(0) - gotChecksum := ^header.ChecksumVV(pkt.Data, 0 /* initial */) - if gotChecksum != wantChecksum { - // It's possible that a raw socket expects to receive this. - h.SetChecksum(wantChecksum) + calculatedChecksum := ^header.ChecksumVV(pkt.Data, 0 /* initial */) + h.SetChecksum(headerChecksum) + if calculatedChecksum != headerChecksum { + // It's possible that a raw socket still expects to receive this. e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) received.Invalid.Increment() return } - // Make a copy of data before pkt gets sent to raw socket. - // DeliverTransportPacket will take ownership of pkt. - replyData := pkt.Data.Clone(nil) - replyData.TrimFront(header.ICMPv4MinimumSize) + // DeliverTransportPacket will take ownership of pkt so don't use it beyond + // this point. Make a deep copy of the data before pkt gets sent as we will + // be modifying fields. + // + // TODO(gvisor.dev/issue/4399): The copy may not be needed if there are no + // waiting endpoints. Consider moving responsibility for doing the copy to + // DeliverTransportPacket so that is is only done when needed. + replyData := pkt.Data.ToOwnedView() + replyIPHdr := header.IPv4(append(buffer.View(nil), pkt.NetworkHeader().View()...)) - // It's possible that a raw socket expects to receive this. - h.SetChecksum(wantChecksum) e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) remoteLinkAddr := r.RemoteLinkAddress @@ -122,29 +122,49 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { // Use the remote link address from the incoming packet. r.ResolveWith(remoteLinkAddr) - // Prepare a reply packet. - icmpHdr := make(header.ICMPv4, header.ICMPv4MinimumSize) - copy(icmpHdr, h) - icmpHdr.SetType(header.ICMPv4EchoReply) - icmpHdr.SetChecksum(0) - icmpHdr.SetChecksum(^header.Checksum(icmpHdr, header.ChecksumVV(replyData, 0))) - dataVV := buffer.View(icmpHdr).ToVectorisedView() - dataVV.Append(replyData) + // TODO(gvisor.dev/issue/3810:) When adding protocol numbers into the + // header information, we may have to change this code to handle the + // ICMP header no longer being in the data buffer. + + // Because IP and ICMP are so closely intertwined, we need to handcraft our + // IP header to be able to follow RFC 792. The wording on page 13 is as + // follows: + // IP Fields: + // Addresses + // The address of the source in an echo message will be the + // destination of the echo reply message. To form an echo reply + // message, the source and destination addresses are simply reversed, + // the type code changed to 0, and the checksum recomputed. + // + // This was interpreted by early implementors to mean that all options must + // be copied from the echo request IP header to the echo reply IP header + // and this behaviour is still relied upon by some applications. + // + // Create a copy of the IP header we received, options and all, and change + // The fields we need to alter. + // + // We need to produce the entire packet in the data segment in order to + // use WriteHeaderIncludedPacket(). + replyIPHdr.SetSourceAddress(r.LocalAddress) + replyIPHdr.SetDestinationAddress(r.RemoteAddress) + replyIPHdr.SetTTL(r.DefaultTTL()) + + replyICMPHdr := header.ICMPv4(replyData) + replyICMPHdr.SetType(header.ICMPv4EchoReply) + replyICMPHdr.SetChecksum(0) + replyICMPHdr.SetChecksum(^header.Checksum(replyData, 0)) + + replyVV := buffer.View(replyIPHdr).ToVectorisedView() + replyVV.AppendView(replyData) replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()), - Data: dataVV, + Data: replyVV, }) - // TODO(gvisor.dev/issue/3810): When adding protocol numbers into the header - // information we will have to change this code to handle the ICMP header - // no longer being in the data buffer. replyPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber - // Send out the reply packet. + + // The checksum will be calculated so we don't need to do it here. sent := stats.ICMP.V4PacketsSent - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ - Protocol: header.ICMPv4ProtocolNumber, - TTL: r.DefaultTTL(), - TOS: stack.DefaultTOS, - }, replyPkt); err != nil { + if err := r.WriteHeaderIncludedPacket(replyPkt); err != nil { sent.Dropped.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 1f6e14c3f..a2be64fb8 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -174,7 +174,7 @@ func (e *endpoint) MTU() uint32 { // MaxHeaderLength returns the maximum length needed by ipv4 headers (and // underlying protocols). func (e *endpoint) MaxHeaderLength() uint16 { - return e.linkEP.MaxHeaderLength() + header.IPv4MinimumSize + return e.linkEP.MaxHeaderLength() + header.IPv4MaximumHeaderSize } // GSOMaxSize returns the maximum GSO packet size. |