Merge release-20201019.0-56-g0a035a101 (automated)

6 files changed, 148 insertions, 15 deletions

diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go
index bb31ef61a..936601287 100644
--- a/pkg/tcpip/network/fragmentation/fragmentation.go
+++ b/pkg/tcpip/network/fragmentation/fragmentation.go
@@ -136,8 +136,16 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea
 // proto is the protocol number marked in the fragment being processed. It has
 // to be given here outside of the FragmentID struct because IPv6 should not use
 // the protocol to identify a fragment.
+//
+// releaseCB is a callback that will run when the fragment reassembly of a
+// packet is complete or cancelled. releaseCB take a a boolean argument which is
+// true iff the reassembly is cancelled due to timeout. releaseCB should be
+// passed only with the first fragment of a packet. If more than one releaseCB
+// are passed for the same packet, only the first releaseCB will be saved for
+// the packet and the succeeding ones will be dropped by running them
+// immediately with a false argument.
 func (f *Fragmentation) Process(
-	id FragmentID, first, last uint16, more bool, proto uint8, vv buffer.VectorisedView) (
+	id FragmentID, first, last uint16, more bool, proto uint8, vv buffer.VectorisedView, releaseCB func(bool)) (
 	buffer.VectorisedView, uint8, bool, error) {
 	if first > last {
 		return buffer.VectorisedView{}, 0, false, fmt.Errorf("first=%d is greater than last=%d: %w", first, last, ErrInvalidArgs)
@@ -171,6 +179,12 @@ func (f *Fragmentation) Process(
 			f.releaseReassemblersLocked()
 		}
 	}
+	if releaseCB != nil {
+		if !r.setCallback(releaseCB) {
+			// We got a duplicate callback. Release it immediately.
+			releaseCB(false /* timedOut */)
+		}
+	}
 	f.mu.Unlock()
 
 	res, firstFragmentProto, done, consumed, err := r.process(first, last, more, proto, vv)
@@ -178,14 +192,14 @@ func (f *Fragmentation) Process(
 		// We probably got an invalid sequence of fragments. Just
 		// discard the reassembler and move on.
 		f.mu.Lock()
-		f.release(r)
+		f.release(r, false /* timedOut */)
 		f.mu.Unlock()
 		return buffer.VectorisedView{}, 0, false, fmt.Errorf("fragmentation processing error: %w", err)
 	}
 	f.mu.Lock()
 	f.size += consumed
 	if done {
-		f.release(r)
+		f.release(r, false /* timedOut */)
 	}
 	// Evict reassemblers if we are consuming more memory than highLimit until
 	// we reach lowLimit.
@@ -195,14 +209,14 @@ func (f *Fragmentation) Process(
 			if tail == nil {
 				break
 			}
-			f.release(tail)
+			f.release(tail, false /* timedOut */)
 		}
 	}
 	f.mu.Unlock()
 	return res, firstFragmentProto, done, nil
 }
 
-func (f *Fragmentation) release(r *reassembler) {
+func (f *Fragmentation) release(r *reassembler, timedOut bool) {
 	// Before releasing a fragment we need to check if r is already marked as done.
 	// Otherwise, we would delete it twice.
 	if r.checkDoneOrMark() {
@@ -216,6 +230,8 @@ func (f *Fragmentation) release(r *reassembler) {
 		log.Printf("memory counter < 0 (%d), this is an accounting bug that requires investigation", f.size)
 		f.size = 0
 	}
+
+	r.release(timedOut) // releaseCB may run.
 }
 
 // releaseReassemblersLocked releases already-expired reassemblers, then
@@ -238,7 +254,7 @@ func (f *Fragmentation) releaseReassemblersLocked() {
 			break
 		}
 		// If the oldest reassembler has already expired, release it.
-		f.release(r)
+		f.release(r, true /* timedOut*/)
 	}
 }
 
diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go
index 9bb051a30..c0cc0bde0 100644
--- a/pkg/tcpip/network/fragmentation/reassembler.go
+++ b/pkg/tcpip/network/fragmentation/reassembler.go
@@ -41,6 +41,7 @@ type reassembler struct {
 	heap         fragHeap
 	done         bool
 	creationTime int64
+	callback     func(bool)
 }
 
 func newReassembler(id FragmentID, clock tcpip.Clock) *reassembler {
@@ -123,3 +124,24 @@ func (r *reassembler) checkDoneOrMark() bool {
 	r.mu.Unlock()
 	return prev
 }
+
+func (r *reassembler) setCallback(c func(bool)) bool {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	if r.callback != nil {
+		return false
+	}
+	r.callback = c
+	return true
+}
+
+func (r *reassembler) release(timedOut bool) {
+	r.mu.Lock()
+	callback := r.callback
+	r.callback = nil
+	r.mu.Unlock()
+
+	if callback != nil {
+		callback(timedOut)
+	}
+}
diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go
index ac66403fa..1edb6de6b 100644
--- a/pkg/tcpip/network/ipv4/icmp.go
+++ b/pkg/tcpip/network/ipv4/icmp.go
@@ -237,6 +237,13 @@ type icmpReasonProtoUnreachable struct{}
 
 func (*icmpReasonProtoUnreachable) isICMPReason() {}
 
+// icmpReasonReassemblyTimeout is an error where insufficient fragments are
+// received to complete reassembly of a packet within a configured time after
+// the reception of the first-arriving fragment of that packet.
+type icmpReasonReassemblyTimeout struct{}
+
+func (*icmpReasonReassemblyTimeout) isICMPReason() {}
+
 // returnError takes an error descriptor and generates the appropriate ICMP
 // error packet for IPv4 and sends it back to the remote device that sent
 // the problematic packet. It incorporates as much of that packet as
@@ -377,17 +384,24 @@ func (p *protocol) returnError(r *stack.Route, reason icmpReason, pkt *stack.Pac
 	icmpPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber
 
 	icmpHdr := header.ICMPv4(icmpPkt.TransportHeader().Push(header.ICMPv4MinimumSize))
+	var counter *tcpip.StatCounter
 	switch reason.(type) {
 	case *icmpReasonPortUnreachable:
+		icmpHdr.SetType(header.ICMPv4DstUnreachable)
 		icmpHdr.SetCode(header.ICMPv4PortUnreachable)
+		counter = sent.DstUnreachable
 	case *icmpReasonProtoUnreachable:
+		icmpHdr.SetType(header.ICMPv4DstUnreachable)
 		icmpHdr.SetCode(header.ICMPv4ProtoUnreachable)
+		counter = sent.DstUnreachable
+	case *icmpReasonReassemblyTimeout:
+		icmpHdr.SetType(header.ICMPv4TimeExceeded)
+		icmpHdr.SetCode(header.ICMPv4ReassemblyTimeout)
+		counter = sent.TimeExceeded
 	default:
 		panic(fmt.Sprintf("unsupported ICMP type %T", reason))
 	}
-	icmpHdr.SetType(header.ICMPv4DstUnreachable)
 	icmpHdr.SetChecksum(header.ICMPv4Checksum(icmpHdr, icmpPkt.Data))
-	counter := sent.DstUnreachable
 
 	if err := route.WritePacket(
 		nil, /* gso */
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 41e548c85..e6b842d61 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -38,7 +38,7 @@ const (
 	// Considering that it is an old recommendation, we use the same reassembly
 	// timeout that linux defines, which is 30 seconds:
 	// https://github.com/torvalds/linux/blob/47ec5303d73ea344e84f46660fff693c57641386/include/net/ip.h#L138
-	reassembleTimeout = 30 * time.Second
+	ReassembleTimeout = 30 * time.Second
 
 	// ProtocolNumber is the ipv4 protocol number.
 	ProtocolNumber = header.IPv4ProtocolNumber
@@ -520,6 +520,28 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 			r.Stats().IP.MalformedFragmentsReceived.Increment()
 			return
 		}
+
+		// Set up a callback in case we need to send a Time Exceeded Message, as per
+		// RFC 792:
+		//
+		//   If a host reassembling a fragmented datagram cannot complete the
+		//   reassembly due to missing fragments within its time limit it discards
+		//   the datagram, and it may send a time exceeded message.
+		//
+		//   If fragment zero is not available then no time exceeded need be sent at
+		//   all.
+		var releaseCB func(bool)
+		if start == 0 {
+			pkt := pkt.Clone()
+			r := r.Clone()
+			releaseCB = func(timedOut bool) {
+				if timedOut {
+					_ = e.protocol.returnError(&r, &icmpReasonReassemblyTimeout{}, pkt)
+				}
+				r.Release()
+			}
+		}
+
 		var ready bool
 		var err error
 		proto := h.Protocol()
@@ -537,6 +559,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 			h.More(),
 			proto,
 			pkt.Data,
+			releaseCB,
 		)
 		if err != nil {
 			r.Stats().IP.MalformedPacketsReceived.Increment()
@@ -856,7 +879,7 @@ func NewProtocol(s *stack.Stack) stack.NetworkProtocol {
 		ids:           ids,
 		hashIV:        hashIV,
 		defaultTTL:    DefaultTTL,
-		fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, reassembleTimeout, s.Clock()),
+		fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock()),
 	}
 }
 
diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go
index b10249c9e..3c15e41a7 100644
--- a/pkg/tcpip/network/ipv6/icmp.go
+++ b/pkg/tcpip/network/ipv6/icmp.go
@@ -744,6 +744,13 @@ type icmpReasonPortUnreachable struct{}
 
 func (*icmpReasonPortUnreachable) isICMPReason() {}
 
+// icmpReasonReassemblyTimeout is an error where insufficient fragments are
+// received to complete reassembly of a packet within a configured time after
+// the reception of the first-arriving fragment of that packet.
+type icmpReasonReassemblyTimeout struct{}
+
+func (*icmpReasonReassemblyTimeout) isICMPReason() {}
+
 // returnError takes an error descriptor and generates the appropriate ICMP
 // error packet for IPv6 and sends it.
 func (p *protocol) returnError(r *stack.Route, reason icmpReason, pkt *stack.PacketBuffer) *tcpip.Error {
@@ -836,7 +843,9 @@ func (p *protocol) returnError(r *stack.Route, reason icmpReason, pkt *stack.Pac
 	if payloadLen > available {
 		payloadLen = available
 	}
-	payload := buffer.NewVectorisedView(pkt.Size(), pkt.Views())
+	payload := network.ToVectorisedView()
+	payload.AppendView(transport)
+	payload.Append(pkt.Data)
 	payload.CapLength(payloadLen)
 
 	newPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{
@@ -857,6 +866,10 @@ func (p *protocol) returnError(r *stack.Route, reason icmpReason, pkt *stack.Pac
 		icmpHdr.SetType(header.ICMPv6DstUnreachable)
 		icmpHdr.SetCode(header.ICMPv6PortUnreachable)
 		counter = sent.DstUnreachable
+	case *icmpReasonReassemblyTimeout:
+		icmpHdr.SetType(header.ICMPv6TimeExceeded)
+		icmpHdr.SetCode(header.ICMPv6ReassemblyTimeout)
+		counter = sent.TimeExceeded
 	default:
 		panic(fmt.Sprintf("unsupported ICMP type %T", reason))
 	}
diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index 3c2578343..1e38f3a9d 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -41,7 +41,7 @@ const (
 	//
 	// Linux also uses 60 seconds for reassembly timeout:
 	// https://github.com/torvalds/linux/blob/47ec5303d73ea344e84f46660fff693c57641386/include/net/ipv6.h#L456
-	reassembleTimeout = 60 * time.Second
+	ReassembleTimeout = 60 * time.Second
 
 	// ProtocolNumber is the ipv6 protocol number.
 	ProtocolNumber = header.IPv6ProtocolNumber
@@ -777,6 +777,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 				continue
 			}
 
+			fragmentFieldOffset := it.ParseOffset()
+
 			// Don't consume the iterator if we have the first fragment because we
 			// will use it to validate that the first fragment holds the upper layer
 			// header.
@@ -834,17 +836,59 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 				return
 			}
 
+			// As per RFC 2460 Section 4.5:
+			//
+			//    If the length of a fragment, as derived from the fragment packet's
+			//    Payload Length field, is not a multiple of 8 octets and the M flag
+			//    of that fragment is 1, then that fragment must be discarded and an
+			//    ICMP Parameter Problem, Code 0, message should be sent to the source
+			//    of the fragment, pointing to the Payload Length field of the
+			//    fragment packet.
+			if extHdr.More() && fragmentPayloadLen%header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit != 0 {
+				r.Stats().IP.MalformedPacketsReceived.Increment()
+				r.Stats().IP.MalformedFragmentsReceived.Increment()
+				_ = e.protocol.returnError(r, &icmpReasonParameterProblem{
+					code:    header.ICMPv6ErroneousHeader,
+					pointer: header.IPv6PayloadLenOffset,
+				}, pkt)
+				return
+			}
+
 			// The packet is a fragment, let's try to reassemble it.
 			start := extHdr.FragmentOffset() * header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit
 
-			// Drop the fragment if the size of the reassembled payload would exceed
-			// the maximum payload size.
+			// As per RFC 2460 Section 4.5:
+			//
+			//    If the length and offset of a fragment are such that the Payload
+			//    Length of the packet reassembled from that fragment would exceed
+			//    65,535 octets, then that fragment must be discarded and an ICMP
+			//    Parameter Problem, Code 0, message should be sent to the source of
+			//    the fragment, pointing to the Fragment Offset field of the fragment
+			//    packet.
 			if int(start)+fragmentPayloadLen > header.IPv6MaximumPayloadSize {
 				r.Stats().IP.MalformedPacketsReceived.Increment()
 				r.Stats().IP.MalformedFragmentsReceived.Increment()
+				_ = e.protocol.returnError(r, &icmpReasonParameterProblem{
+					code:    header.ICMPv6ErroneousHeader,
+					pointer: fragmentFieldOffset,
+				}, pkt)
 				return
 			}
 
+			// Set up a callback in case we need to send a Time Exceeded Message as
+			// per RFC 2460 Section 4.5.
+			var releaseCB func(bool)
+			if start == 0 {
+				pkt := pkt.Clone()
+				r := r.Clone()
+				releaseCB = func(timedOut bool) {
+					if timedOut {
+						_ = e.protocol.returnError(&r, &icmpReasonReassemblyTimeout{}, pkt)
+					}
+					r.Release()
+				}
+			}
+
 			// Note that pkt doesn't have its transport header set after reassembly,
 			// and won't until DeliverNetworkPacket sets it.
 			data, proto, ready, err := e.protocol.fragmentation.Process(
@@ -860,6 +904,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) {
 				extHdr.More(),
 				uint8(rawPayload.Identifier),
 				rawPayload.Buf,
+				releaseCB,
 			)
 			if err != nil {
 				r.Stats().IP.MalformedPacketsReceived.Increment()
@@ -1535,7 +1580,7 @@ func NewProtocolWithOptions(opts Options) stack.NetworkProtocolFactory {
 	return func(s *stack.Stack) stack.NetworkProtocol {
 		p := &protocol{
 			stack:         s,
-			fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, reassembleTimeout, s.Clock()),
+			fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock()),
 			ids:           ids,
 			hashIV:        hashIV,