Files
echolot/server/internal/dataplane/frag_linux_test.go
T
mrambossekandClaude Fable 5 a7dccf7da2
server-release / image (push) Successful in 15s
server-test / test (push) Successful in 32s
server-release / release (push) Successful in 32s
frag_send: crafted IP fragments, so ordering can be tested and not just delivery
Letting the kernel fragment an oversized datagram answers one question — do
fragments get through. It cannot answer the more interesting one, because the
kernel always emits them in order, first one first.

The classic middlebox fault is exactly about that ordering. Only the first
fragment carries the UDP header, and therefore the ports; a stateful firewall
or NAT that has not seen it has no flow to match the rest against, and many
drop them. That is invisible to any in-order test and shows up in the field as
"large DNS answers fail on this network" or "the tunnel breaks when the MTU
drops" — it works until the network reorders, then fails intermittently, which
is the hardest kind of fault to chase.

So the server now builds the fragments itself (raw socket, IP_HDRINCL) and
controls their order: in_order as a baseline, reversed, and first-fragment-last.
The datagram is assembled and signed whole before being cut up, so what the
client reassembles is indistinguishable from an ordinary packet — otherwise it
would be measuring our sender rather than the path.

Two details that would silently produce wrong answers:
  - The UDP checksum is computed rather than left zero. A zero-checksum datagram
    is dropped by some middleboxes, and that drop would be recorded as a
    fragmentation failure, which is the wrong conclusion entirely.
  - Fragment offsets are in 8-byte units, so non-final fragments are rounded to
    a multiple of 8. A 100-byte fragment is not an error, it is a datagram no
    host will ever reassemble.

frag-send is advertised only when a raw socket can actually be opened — checked
by opening one, since a permission model has more ways to say no than a
capability bit has to say yes.

Fragment header arithmetic is unit-tested (reassembly coverage, MF flags, shared
IP ID, 8-byte offsets, checksum verification), cross-compiled and run on Linux
since the code is build-tagged.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-01 13:45:09 +02:00

161 lines
5.4 KiB
Go

// SPDX-FileCopyrightText: 2026 Echolot contributors
// SPDX-License-Identifier: GPL-3.0-or-later
//go:build linux
package dataplane
import (
"encoding/binary"
"net/netip"
"testing"
)
// Fragment headers are the kind of thing that is either exactly right or silently useless: a
// wrong offset unit, a missing MF bit or a bad checksum produces packets that leave the machine
// and are dropped by the receiver's IP stack without a word. Nothing downstream would notice —
// the client would simply record "fragments do not get through", which is a wrong answer rather
// than a missing one. Hence these check the bytes.
func testAddrs() (netip.AddrPort, netip.AddrPort) {
return netip.MustParseAddrPort("192.0.2.1:8442"), netip.MustParseAddrPort("198.51.100.9:41000")
}
func TestSplitCoversThePayloadExactlyOnce(t *testing.T) {
src, dst := testAddrs()
udp := buildUDP(src, dst, make([]byte, 2000))
frags := splitIPv4(src.Addr(), dst.Addr(), udp, 576, 0x1234)
if len(frags) < 3 {
t.Fatalf("expected several fragments for %d bytes, got %d", len(udp), len(frags))
}
// Reassemble the way a receiver would: place each fragment's payload at its offset.
rebuilt := make([]byte, len(udp))
covered := make([]bool, len(udp))
for _, f := range frags {
flagsOff := binary.BigEndian.Uint16(f[6:8])
off := int(flagsOff&0x1FFF) * 8
body := f[20:]
if off+len(body) > len(udp) {
t.Fatalf("fragment at offset %d overruns the datagram", off)
}
for i, b := range body {
if covered[off+i] {
t.Fatalf("byte %d delivered twice", off+i)
}
covered[off+i] = true
rebuilt[off+i] = b
}
}
for i, c := range covered {
if !c {
t.Fatalf("byte %d was never sent", i)
}
}
for i := range udp {
if rebuilt[i] != udp[i] {
t.Fatalf("reassembled byte %d differs", i)
}
}
}
func TestFragmentHeadersAreWellFormed(t *testing.T) {
src, dst := testAddrs()
udp := buildUDP(src, dst, make([]byte, 3000))
frags := splitIPv4(src.Addr(), dst.Addr(), udp, 800, 0xBEEF)
for i, f := range frags {
if got := f[0]; got != 0x45 {
t.Errorf("fragment %d: version/IHL = %#x, want 0x45", i, got)
}
if got := f[9]; got != 17 {
t.Errorf("fragment %d: protocol = %d, want 17 (UDP)", i, got)
}
if got := binary.BigEndian.Uint16(f[4:6]); got != 0xBEEF {
t.Errorf("fragment %d: IP ID = %#x — all fragments of one datagram must share it", i, got)
}
if got := binary.BigEndian.Uint16(f[2:4]); int(got) != len(f) {
t.Errorf("fragment %d: total length = %d, actual %d", i, got, len(f))
}
flagsOff := binary.BigEndian.Uint16(f[6:8])
mf := flagsOff&0x2000 != 0
wantMF := i < len(frags)-1
if mf != wantMF {
t.Errorf("fragment %d: MF = %v, want %v", i, mf, wantMF)
}
}
}
// Offsets are counted in 8-byte units, so every fragment but the last must be a multiple of 8.
// A 100-byte "fragment size" that silently becomes 100 bytes on the wire produces a datagram no
// host will ever reassemble.
func TestNonFinalFragmentsAreEightByteMultiples(t *testing.T) {
src, dst := testAddrs()
udp := buildUDP(src, dst, make([]byte, 2500))
for _, size := range []int{8, 100, 576, 999, 1400} {
frags := splitIPv4(src.Addr(), dst.Addr(), udp, (size/8)*8, 1)
for i, f := range frags[:len(frags)-1] {
if body := len(f) - 20; body%8 != 0 {
t.Errorf("size %d: non-final fragment %d carries %d bytes, not a multiple of 8",
size, i, body)
}
}
}
}
// The UDP checksum is optional in IPv4, and sending zero would be less code — but a
// zero-checksum datagram is dropped by some middleboxes, and that drop would be recorded as a
// fragmentation failure. So it must be present and correct.
func TestUDPChecksumVerifies(t *testing.T) {
src, dst := testAddrs()
for _, n := range []int{0, 1, 7, 8, 100, 1001} { // odd lengths exercise the tail-byte path
udp := buildUDP(src, dst, make([]byte, n))
if got := binary.BigEndian.Uint16(udp[6:8]); got == 0 {
t.Fatalf("payload %d: checksum is zero, which means 'not computed'", n)
}
if sum := verifyUDPChecksum(src.Addr(), dst.Addr(), udp); sum != 0xFFFF {
t.Errorf("payload %d: checksum does not verify (one's complement sum %#x)", n, sum)
}
if got := binary.BigEndian.Uint16(udp[4:6]); int(got) != len(udp) {
t.Errorf("payload %d: UDP length field %d, actual %d", n, got, len(udp))
}
}
}
func TestUDPPortsComeFromTheSessionAddresses(t *testing.T) {
src, dst := testAddrs()
udp := buildUDP(src, dst, []byte("x"))
if got := binary.BigEndian.Uint16(udp[0:2]); got != src.Port() {
t.Errorf("source port = %d, want %d", got, src.Port())
}
// The destination port must be the client's observed source port, or the datagram arrives
// at the machine and is discarded before any socket sees it.
if got := binary.BigEndian.Uint16(udp[2:4]); got != dst.Port() {
t.Errorf("destination port = %d, want %d", got, dst.Port())
}
}
// Recomputes the one's complement sum over the pseudo-header and datagram; a correct checksum
// makes the total 0xFFFF.
func verifyUDPChecksum(src, dst netip.Addr, udp []byte) uint16 {
var sum uint32
s4, d4 := src.Unmap().As4(), dst.Unmap().As4()
for _, b := range [][]byte{s4[:], d4[:]} {
sum += uint32(binary.BigEndian.Uint16(b[0:2]))
sum += uint32(binary.BigEndian.Uint16(b[2:4]))
}
sum += 17
sum += uint32(len(udp))
for i := 0; i+1 < len(udp); i += 2 {
sum += uint32(binary.BigEndian.Uint16(udp[i : i+2]))
}
if len(udp)%2 == 1 {
sum += uint32(udp[len(udp)-1]) << 8
}
for sum>>16 != 0 {
sum = (sum & 0xFFFF) + (sum >> 16)
}
return uint16(sum)
}