Partial

RFC 3948 - UDP Encapsulation of IPsec ESP Packets

Every requirement this repository extracted from RFC 3948, the tests bound to it, and what a reader has verified about them. This summary is enrolled and gated by ./le rfc check.

Overview

Positive

what Ze has

Tested both ways61.5%8 of 13 binding obligations

a positive test proves Ze does what the requirement demands and a negative one proves it refuses what the requirement forbids

One polarity plus reason30.8%4 of 13 binding obligations

the requirement admits no counter-case, so one polarity plus a recorded reason is the whole proof available for it

One polarity, unexcused0.0%0 of 13 binding obligations

one direction is tested, the other is neither tested nor excused, and nothing states which

Proven by a recorded break3.8%1 of 26 tagged units

a red was observed once under a recorded procedure, and the unit, the claim and the producer it rested on still hash to what was recorded. The break is not re-run. A test pair is not a proof until one has been observed

Neutral

measures that are neither good news nor bad

Gated MUSTs14of 18 this summary declares

MUST-level requirements the gate HOLDS. A population, not a result: the shares beside it are what says how Ze stands

Out of scope1of 14 gated MUSTs

a {not-applicable} annotation says the obligation does not bind Ze. Scope, not coverage: it is in no share below

Negative

what Ze owes

No test at all7.7%1 of 13 binding obligations

no test carries the requirement id, whether or not a gap states why

The 4 shares marked as a part above are the whole of the 13 obligations that bind Ze: they add to 100%. Proven by a recorded break is a share of TAGGED UNITS, a different population, so it is not one of them.

How to read the colors

A color names what the measure MEANS, not how well Ze scores on it. Green is a good outcome at any value, red is a bad one, and neither a population nor a scope count is an outcome, so both take no color. The number under the label is what says how far Ze has got.

Card	Tone here	Why that color
Gated MUSTs	neutral	no color: a population is a scale, and a larger one is neither good news nor bad. It is the accounting total
Out of scope	neutral	no color: an obligation that never bound Ze is neither an achievement nor a failure, and counting it either way would be a claim
Tested both ways	ok	green at every value: a test pair is the outcome this gate exists to produce, and the share under the label is what says how far Ze has got
One polarity plus reason	ok	green at every value: where no counter-case exists, one polarity IS the complete answer, and a recorded reason is what the gate demands beside it
One polarity, unexcused	ok	green at zero, RED above it: half a proof with no reason for the other half
No test at all	bad	green at zero, RED above it: a binding obligation nothing exercises is a claim with nothing behind it, whether or not a reason is stated
Proven by a recorded break	ok	green at every value: an observed break is the outcome the discrimination gate exists to produce. The denominator is TAGGED UNITS, not obligations, so this share is not one of the parts above
Audit verdicts	warn	RED on the first weak, wrong or unimplemented verdict, amber while a verdict is no longer current or a gated MUST is unjudged, green when every one is judged sound and current

At a glance

Field	Value
Public status	Partial
Enrolment	Enrolled
Requirements	18
Gated MUST-level	14
Obligations that bind Ze	13
Not applicable, so out of scope	1
Declared gaps	1
Gated with no test	0
Nightly-only evidence	2
Test tags	26
Tagged units	26
Recorded audit verdicts	0
Discrimination records	1
Summary	`rfc/short/rfc3948.md`
Requirement shard	`rfc/requirements/rfc3948.md`
RFC text	`rfc/full/rfc3948.txt`

Enrolment

Enrolled: UDP Encapsulation of IPsec ESP Packets (NAT-Traversal): eleven MUST-level requirements. Eight are met with tags in internal/component/ike: 2.1-1 (the ESP SPI is never zero), 2.1-3 (demultiplex on port 4500 -- a four-zero-byte marker is IKE, a single 0xFF byte is a NAT keepalive, otherwise ESP), 2.2-1 (the non-ESP marker of four zero bytes is prepended to IKE), 1-1 (a tunnel-mode client supports tunnel mode: every Child SA that negotiated no USE_TRANSPORT_MODE installs as tunnel, and a peer request the operator did not configure cannot change that), and 4-3 (a received NAT-keepalive reaches no SA, so it is never read as evidence the connection is live) carry positive+negative tags; 2.1-2 (the ESP SA uses UDP port 4500 for encapsulation), 4-1 (NAT keepalives are sent at a conservative sub-binding-timeout interval) and 2.3-2 (the keepalive payload is one octet of 0xFF) are {single-polarity: positive} with new tests. The last three ids were extracted by the 2026-08-31 extraction sign-off, which found their sites unmapped. 3.1.2-1 (ESP-in-UDP encapsulation and transport-mode checksum fixup), 3.1.2-2 (inner checksum handling on decapsulation), and 2.1-4 (do not depend on a zero UDP checksum) are {not-applicable}: ze delegates UDP-ESP encapsulation and decapsulation to the kernel via XFRM_ENCAP_ESPINUDP and never inspects the UDP checksum.

What the public ledger says

Status: Partial

What the ledger says is covered:

NAT-T non-ESP marker, UDP 4500 encapsulation, NAT keepalive, XFRM UDP encap attributes.

What the ledger says remains

Section 5.1 tunnel mode conflict (RFC3948-5.1-1) is a gap. Ze assigns no inner address to a remote peer, so it devises no way of preventing two peers behind one NAT from reaching it with the same self-chosen inner address. The section's RECOMMENDED remedy is a locally unique address per peer, and the allocator for it is written and unreached: Pool.Allocate (internal/component/ike/eap/pool.go) has no non-test caller, and registerIKE (internal/component/ike/engine/register.go) discards the pool it builds. No engine code constructs a Configuration payload, so ze sends no CFG_REPLY. Closing it is plan/spec-ike-virtual-ip-assignment.md.

Coverage

Bucket	Count	What it counts
Positive and negative tests	8	one part of the gated population
Annotated instead of tested	6	one part of the gated population
One polarity only	0	one part of the gated population
No test and no annotation	0	one part of the gated population
Evidence that runs nightly only	2	an overlay: each of these is also counted by the part it falls in
Gated MUST-level requirements	14	every gated MUST falls in exactly one bucket above

Positive and negative tests (8): RFC3948-2.1-1, RFC3948-3.1.2-1, RFC3948-3.1.2-2, RFC3948-2.2-1, RFC3948-2.1-3, RFC3948-1-1, RFC3948-4-3, RFC3948-5.2-1

Annotated instead of tested (6): RFC3948-2.1-2, RFC3948-2.1-4, RFC3948-4-1, RFC3948-2.3-2, RFC3948-3.1.1-1, RFC3948-5.1-1

Evidence that runs nightly only (2): RFC3948-3.1.2-1, RFC3948-3.1.2-2

Requirements

Requirement	Level	Section	Tests
`RFC3948-2.1-1`	ESP SPI MUST NOT be zero (zero is reserved for the Non-ESP Marker to distinguish IKE from ESP on port 4500) (S2.1)
	MUST NOT	2.1 - UDP-Encapsulated ESP Header Format	positive`unit/verifyTestGenerateESPSPI` negative`unit/verifyTestGenerateESPSPI`
`RFC3948-2.1-2`	Source and destination ports MUST match the ports used by IKE (port 4500 after port float) (S2.1)
	MUST	2.1 - UDP-Encapsulated ESP Header Format	positive`unit/verifyTestChildSANATTEncapPorts` negative`no test`no negative test {single-polarity}: the Child SA install unconditionally sets both UDP-encap ports to the fixed IKE NAT-T port 4500 (internal/component/ike/engine/child.go:237-238,265-266); no ze code path produces a non-4500 encap port, so there is no mismatched-port case to reject
`RFC3948-3.1.2-1`	Transport mode decapsulation MUST do one of three things with the inner TCP/UDP checksum: recompute it incrementally from the addresses received via IKE, recompute it in full, or zero a UDP checksum and flag the stack that a TCP one need not be computed (S3.1.2)
	MUST	3.1.2 - Transport Mode Decapsulation NAT Procedure	positive`interop/nightlycheckNATTTransportInnerChecksum` negative`interop/nightlycheckNATTTunnelInnerChecksum` nightly-only: every test bound to this requirement runs in the scheduled workflow alone, so nothing here is proven on the merge path
`RFC3948-3.1.2-2`	Tunnel mode TCP checksums MUST be verified (S3.1.2)
	MUST	3.1.2 - Transport Mode Decapsulation NAT Procedure	positive`interop/nightlycheckNATTTunnelInnerChecksum` negative`interop/nightlycheckNATTTransportInnerChecksum` nightly-only: every test bound to this requirement runs in the scheduled workflow alone, so nothing here is proven on the merge path
`RFC3948-2.2-1`	Non-ESP Marker (4 zero bytes) MUST be prepended to IKE packets on port 4500 (S2.2)
	MUST	2.2 - IKE Header Format for Port 4500	positive`unit/verifyTestNonESPMarker` negative`unit/verifyTestNonESPMarkerESPPacket`
`RFC3948-2.1-3`	Receiver MUST demultiplex by inspecting first 4 bytes after UDP header: zero = IKE, non-zero = ESP, 1-byte 0xFF = keepalive (S2.1, S2.2, S2.3)
	MUST	2.1 - UDP-Encapsulated ESP Header Format	positive`unit/verifyTestIsNATKeepalive` positive`unit/verifyTestNonESPMarker` negative`unit/verifyTestIsNATKeepalive` negative`unit/verifyTestNonESPMarkerESPPacket`
`RFC3948-2.1-4`	Receivers MUST NOT depend on the UDP checksum being zero (S2.1)
	MUST NOT	2.1 - UDP-Encapsulated ESP Header Format	positive`no test`no positive test negative`no test`no negative test {not-applicable}: ze reads IKE/ESP demux from the OS UDP socket (internal/component/ike/engine/register.go:421) and never inspects the UDP checksum, so no ze code path can depend on it; ESP-in-UDP checksum handling is the kernel's
`RFC3948-4-1`	Keepalive interval MUST be shorter than the NAT binding timeout (S4)
	MUST	4 - NAT Keepalive Procedure	positive`unit/verifyTestKeepaliveDefaultInterval` positive`unit/verifyTestNATKeepalive` negative`no test`no negative test {single-polarity}: the keepalive interval is a fixed conservative 20s constant (internal/component/ike/transport/keepalive.go:13), well under a typical NAT UDP binding lifetime; there is no dynamic binding-timeout query to drive a negative
`RFC3948-1-1`	IPsec tunnel mode clients MUST support tunnel mode (S1)
	MUST	1 - Introduction	positive`unit/verifyTestChildSAInstallsInDataplane` positive`unit/verifyTestTunnelModeIsTheChildSADefault` negative`unit/verifyTestTunnelModeIsTheChildSADefault`
`RFC3948-2.3-2`	The NAT-keepalive sender MUST use a one-octet-long payload with the value 0xFF (S2.3)
	MUST	2.3 - NAT-Keepalive Packet Format	positive`unit/verifyTestNATKeepalive` negative`no test`no negative test {single-polarity}: the keepalive payload is a fixed one-octet constant (internal/component/ike/transport/keepalive.go:14,48) that no input can vary, so there is no non-conforming sender case to drive
`RFC3948-4-3`	Reception of NAT-keepalive packets MUST NOT be used to detect whether a connection is live (S4)
	MUST NOT	4 - NAT Keepalive Procedure	positive`unit/verifyTestNATKeepaliveReachesNoSA` negative`unit/verifyTestNATKeepaliveIsNeverDeliveredToASession` negative`unit/verifyTestNATKeepaliveReachesNoSA`
`RFC3948-3.1.1-1`	Tunnel mode decapsulation, depending on local policy, MUST do one of three things: check the inner source address against the policy, check it against the address assigned to the peer, or NAT the packet (S3.1.1)
	MUST	3.1.1 - Tunnel Mode Decapsulation NAT Procedure	positive`unit/verifyTestChildInboundPolicyDefinesTheValidInnerSourceSpace` positive`unit/verifyTestChildSAInboundPolicyUsesNegotiatedTS` negative`no test`no negative test {single-polarity}: ze takes the first option by installing the inbound policy with the negotiated remote traffic selector as its source selector (childPolicyParams, internal/component/ike/engine/child.go), and the drop of a mismatched inner packet is the kernel's, so ze holds no rejecting branch
`RFC3948-5.1-1`	Implementors MUST devise ways of preventing two remote peers from reaching one security gateway with overlapping inner addresses (S5.1)
	MUST	5.1 - Tunnel Mode Conflict	positive`no test`no positive test negative`no test`no negative test {gap}: ze devises no such way, because it assigns no inner address at all. The section's RECOMMENDED remedy is to give each remote peer a locally unique address, and the allocator for it exists and is unreached: Pool.Allocate (internal/component/ike/eap/pool.go) leases a unique address and refuses on exhaustion with ErrPoolExhausted, yet its only callers are pool_test.go and pool_release_test.go. registerIKE (internal/component/ike/engine/register.go) builds the pool from the remote-access config and then discards it at a bare `_ = ipPool`, so no lease ever reaches a peer. No engine code constructs a wire.PayloadCP either, so ze sends no CFG_REPLY and a client keeps whatever inner address it chose for itself. Two peers behind one NAT that both chose 10.1.2.3 would therefore present the gateway with the ambiguity the section names, and ze holds nothing that prevents it. Closing this is plan/spec-ike-virtual-ip-assignment.md
`RFC3948-5.2-1`	Implementations MUST handle conflicting connections from clients behind one NAT, either by disallowing conflicting connections or by other means (S5.2)
	MUST	5.2 - Transport Mode Conflict	positive`unit/verifyTestPolicyOwnerSeparatesDistinctSelectors` negative`unit/verifyTestPolicyOwnerRefusesASecondPeerOnOneSelector`
`RFC3948-2.1-5`	IPv4 UDP checksum SHOULD be zero on transmit (S2.1)
	SHOULD	2.1 - UDP-Encapsulated ESP Header Format	positive`no test`no positive test negative`no test`no negative test
`RFC3948-2.3-1`	NAT-keepalive packets (1 byte 0xFF) SHOULD be ignored by the receiver (S2.3)
	SHOULD	2.3 - NAT-Keepalive Packet Format	positive`no test`no positive test negative`no test`no negative test
`RFC3948-3.1.2-3`	TCP checksum verification MAY be skipped for transport mode only if the packet is integrity-protected (S3.1.2)
	MAY	3.1.2 - Transport Mode Decapsulation NAT Procedure	positive`no test`no positive test negative`no test`no negative test
`RFC3948-4-2`	Keepalive interval is implementation-specific; typical 20-30 seconds (S4)
	MAY	4 - NAT Keepalive Procedure	positive`no test`no positive test negative`no test`no negative test

Gaps and untested MUSTs

Requirement	State	Reason
`RFC3948-2.1-4` Receivers MUST NOT depend on the UDP checksum being zero (S2.1)	no test	no test carries this requirement id; annotated {not-applicable}: ze reads IKE/ESP demux from the OS UDP socket (internal/component/ike/engine/register.go:421) and never inspects the UDP checksum, so no ze code path can depend on it; ESP-in-UDP checksum handling is the kernel's
`RFC3948-5.1-1` Implementors MUST devise ways of preventing two remote peers from reaching one security gateway with overlapping inner addresses (S5.1)	{gap}, no test	ze devises no such way, because it assigns no inner address at all. The section's RECOMMENDED remedy is to give each remote peer a locally unique address, and the allocator for it exists and is unreached: Pool.Allocate (internal/component/ike/eap/pool.go) leases a unique address and refuses on exhaustion with ErrPoolExhausted, yet its only callers are pool_test.go and pool_release_test.go. registerIKE (internal/component/ike/engine/register.go) builds the pool from the remote-access config and then discards it at a bare `_ = ipPool`, so no lease ever reaches a peer. No engine code constructs a wire.PayloadCP either, so ze sends no CFG_REPLY and a client keeps whatever inner address it chose for itself. Two peers behind one NAT that both chose 10.1.2.3 would therefore present the gateway with the ambiguity the section names, and ze holds nothing that prevents it. Closing this is plan/spec-ike-virtual-ip-assignment.md

Proof state

A tagged unit reads unproven where no discrimination record exists for it: nothing in this tree has been observed to break it, so the claim its tag makes is unproven.

`RFC3948-2.1-1`

ESP SPI MUST NOT be zero (zero is reserved for the Non-ESP Marker to distinguish IKE from ESP on port 4500) (S2.1)