[levy]

Stacked on #1388 (the IGMP and MLD repairs, topic/standards-tests-igmp-mld-level2-fixes). Merge #1388 first. This pull request targets that branch, so the diff shows only the ND commits.

The level 2 standards tests of IPv6 Neighbor Discovery and address autoconfiguration (tests/protocol/nd) found eleven defects in the model, against RFC 4861 and RFC 4862. doc/project/evidence/model/nd/results.md on master lists them as model gaps, and 15 protocol tests fail on master because of them. This pull request repairs ten of the defects, one defect in each commit. The eleventh, gap 9, is half repaired by #1388 and half open. After it, 36 of the 37 ND tests pass.

The defects have an effect on almost every IPv6 simulation:

• Every IPv6 datagram without a hop limit from its transport left with the constant 30. RFC 4861 specifies the CurHopLimit of the interface: 64, or the value of the router.
• A router advertised the hop limit 30 and always an MTU option of 1280. A host stored the MTU and the RetransTimer of its router, but did not use them, and Duplicate Address Detection read the Retrans Timer field in milliseconds as seconds.
• A router answered only the first Router Solicitation, and its first periodic advertisement came 198 to 600 seconds after the start, not within 16 seconds.
• A host dropped a whole Router Advertisement during Duplicate Address Detection, so on a link with two routers it could stay without a default router.
• A Redirect carried no option: no target link-layer address and no part of the invoking packet.
• An address from Ipv6NetworkConfigurator was valid at once and was never tested for a duplicate, and an address whose valid lifetime ended stayed in use.

The repairs are in Ipv6, Ipv6InterfaceData, Ipv6NeighbourDiscovery, the ICMPv6 serializer and Ipv6NetworkConfigurator. Each repair commit carries its source change, the test that it makes pass, its paragraph of the release note doc/whatsnew/incompatible/ipv6-neighbor-discovery-corrections.txt, and the fingerprint values that it moves. The statistical results are on the branch with the same name in the statistics repository, built on the statistics branch of #1388.

Why stacked. On master alone, these repairs make eight tests of other suites fail. The ND solicited-node change lets an MLDv2 host send Reports before the first MLDv1 Query, and an MLDv1 router stops on them in three MLD tests. The DAD of configured addresses keeps an address tentative for about 2 to 3 seconds: two MLD start-up tests fail, because the Mld Query timer starts before the router can send, and three IPv6 IPsec tests fail, because their flows start at 1 second. #1388 repairs both MLD causes and carries the solicited-node change, so this branch is built on it and does not repeat that change.

Gap 9 is half open. #1388 makes an interface join the solicited-node group of each address. But it also holds every MLD message until the link-local address is tested, so the Report for the group of a tentative address leaves after the solicitation of Duplicate Address Detection, and Rfc4862GroupsBeforeDad still fails. RFC 9777 section 5.2.14, and RFC 3590 for MLDv1, send that Report from the unspecified address. That repair needs the source in Ipv6 and in MLD, and routers that ignore such Reports; it is left for a later pull request. The test stays a plain FAIL with its new reason.

Risk: the trajectories of most IPv6 simulations change. 62 fingerprint configurations and 59 statistical results move, in examples/bgpv4, eigrp, inet, ipv6, manetrouting, ospfv3, pim and rip. Two defaults of Ipv6RoutingTable change: advCurHopLimit from 30 to 64, and advLinkMtu from 1280 to 0 (no MTU option). A simulation that needs the MTU option must set advLinkMtu. A configured global address is now usable about one to two seconds after the boot of its node, after Duplicate Address Detection. Set dupAddrDetectTransmits to 0 to keep the old behavior; the IPv6 IPsec tests now do this.

#1266 (icmpv6: keep one Router Discovery process per interface) changes Ipv6NeighbourDiscovery.cc too. The second of the two to merge must rebase on the first.

Commits, in the order to read them

Each fix commit in src moves behavior. The tests commits change no model code: the first gives Rfc4862GroupsBeforeDad its new reason, and the others prepare the next repair, each for a test whose timing the repair changes. The plan and doc commits change no behavior.

• c6d2dc2b48 plan: add: ND level 2 repairs, one gap for each commit
• bf67a8a702 tests: change: Rfc4862GroupsBeforeDad names the MLD Report that waits for the link-local address (the new reason of the open half of gap 9)
• 3cf34d9394 ipv6: fix: a datagram takes the CurHopLimit of its outgoing interface (gap 1)
• 4f6c18213c networklayer: fix: a router advertises the hop limit 64 and no MTU option (gap 2, before gap 3: with the old MTU default, gap 3 alone would limit every host to 1280 octets)
• 8d9f9a9e8d networklayer: fix: a host fragments to the MTU that its router advertises (gap 3)
• 555cdd0678 networklayer: fix: address resolution waits the RetransTimer of its router (gap 4)
• 2787442783 icmpv6: fix: a router answers every Router Solicitation (gap 5)
• 59d5ef4270 tests: fix: the initial advertisements may come exactly 16 s apart (prepares gap 6)
• 9931374d01 icmpv6: fix: the first advertisements of a router come within 16 s (gap 6, with the seed of two MLDv1 module tests)
• f84db43f6a tests: fix: the answer of R1 comes during the address detection of A (prepares gap 7)
• 9d94072e30 icmpv6: fix: a host takes its routers from advertisements during DAD (gap 7)
• a554b72f20 tests: fix: R1 knows the link-layer address of R2 before the Redirect (prepares gap 8)
• 55aa88fc4b icmpv6: fix: a Redirect carries the target address and the invoking packet (gap 8)
• 95456c13dc tests: fix: the IPv6 IPsec scenarios run without Duplicate Address Detection (prepares gap 10)
• e3ff43da45 networklayer: fix: a node tests the addresses that its configurator gives (gap 10)
• ce926baec8 networklayer: fix: an address is removed when its valid lifetime ends (gap 11)
• 8631810bdc plan: change: ND gaps 1 to 11 are repaired
• eadff0948b doc: change: ND results, ledger and matrix after the repairs (a new run of the ND tests at 8631810bdc)
• b43d0ca975 plan: change: ND level 2 repairs are done
• cc4d7ec3c7 plan: location: move the ND repair plan to done

Architectural surface:

• Packet content: the hop limit of IPv6 datagrams; the Cur Hop Limit, MTU option and Retrans Timer of Router Advertisements; the new options of a Redirect, with the new message class Ipv6NdRedirectedHeader in Ipv6NdMessage.msg and its serialization in Icmpv6HeaderSerializer.
• Configuration surface: the defaults of advCurHopLimit (64) and advLinkMtu (0) of Ipv6RoutingTable, and the default of AdvRetransTimer in Ipv6InterfaceData (0, so that an advertisement leaves the field unspecified). No parameter is new or removed.
• Contracts: Ipv6InterfaceData keeps the CurHopLimit, LinkMTU and RetransTimer of an interface, and the outgoing interface writes the hop limit of a datagram. An address of the configurator is tentative until Duplicate Address Detection ends. The ipv6AddressRemoved signal now also fires when a valid lifetime ends.
• Tests of other suites: the six IPv6 IPsec tests set dupAddrDetectTransmits = 0, as the common mockups of the IPsec checks now say. The module tests MLD_host_basic and MLD_host_groupstates pin seed 0, because the new random delays of the first Router Advertisements swap their two hosts at seed 1. IPv6_RA_MTU_option sets advLinkMtu = 1280.
• No feature descriptor changes, no sealed path, and no new AV-* or NV-* row.

Evidence

All builds are clean builds (make MODE=<mode> clean, then make MODE=<mode>), with OMNeT++ 6.x and clang on Linux. The commands are those of opp_repl, as on master. The results are at the tip of this branch, that is with #1388.

Command	Result
opp_run_protocol_tests -m debug -w '^tests/protocol/nd$' at the tip	37 tests: 36 PASS, 1 FAIL, Rfc4862GroupsBeforeDad (gap 9). Master: 15 FAIL
opp_run_protocol_tests -m debug -w '^tests/protocol/mld$' at the tip, and the same for igmp and ipsec	MLD 47 PASS, IGMP 43 PASS, IPsec 28 PASS and its 3 expected FAIL
opp_run_protocol_tests -m debug at the tip	825 tests: 454 PASS, 306 expected SKIP (the packetdrill tests need inet-gpl), 45 expected FAIL, 20 FAIL. The 20 are the 10 open RIP tests, Rfc4862GroupsBeforeDad, and 9 tests of tests/protocol/ipv6 (Rfc4443*, Rfc8200*) that fail on master too and that master repaired after the base of #1388. No other test fails.
opp_run_module_tests -m debug at the tip	759 PASS
opp_run_fingerprint_tests -m release at the tip	1840 tests: 1837 PASS, 2 expected ERROR, 1 FAIL: examples/inet/gsandcs SAT, which depends on the z3 version of the machine and does not move with this branch
inet_run_statistical_tests -m release, in a copy of the container of the GitHub job (clang-16), with the statistics branch (see the note below the table)	929 tests: 844 PASS, 10 expected SKIP, 1 expected ERROR, 73 FAIL and 1 ERROR. These 74 configurations fail on #1388 and on master too (802.11, 10BASE-T1S, DHCP and other areas). Three master failures pass with this statistics branch: ipv6/mipv6 Handover, ipv6/mipv6roaming and ipv6/pmipv6, whose results this branch regenerates. No other configuration fails.

The container of the GitHub job has no opp_repl. So the statistical runs used INET's own copy of the test framework, with the commits of #1388 and of this branch replayed onto master 664b0f8940, before master removed that copy. The src trees are the same as on this branch. On GitHub, the job runs opp_run_statistical_tests.

Baseline updates. Each commit got its fingerprint values in tests/fingerprint/store.json from a clean release build at that commit, on the base of #1388 (opp_update_fingerprint_test_results -m release -w <dirs> -i <ini files> -c <configs>, for the configurations that the branch moves). Each statistics commit was made in the container of the GitHub job at the INET commit that it names. Each commit message names the configurations.

Commit	Fingerprint configurations	Statistical results
3cf34d9394 CurHopLimit	14 (~tND only)	0
4f6c18213c router defaults	36	33
555cdd0678 RetransTimer	32 (~tND only)	0
2787442783 answers to every solicitation	7	7
9931374d01 initial advertising interval	56	53
9d94072e30 advertisements during DAD	2	2
55aa88fc4b Redirect options	3	3
e3ff43da45 detection of configured addresses	39	39

The other commits move no fingerprint and no statistical result. The statistics branch is topic/standards-tests-nd-level2-fixes of inet-framework/statistics: 6 commits on topic/standards-tests-igmp-mld-level2-fixes of #1388.

Not verified here:

• The feature tests, the validation tests and the chart tests. The CI of this pull request runs them.
• AdvReachableTime has the same unit error as the Retrans Timer had. No test covers it, and this branch does not repair it.
• The second half of gap 9 (the Report from the unspecified address, RFC 3590).

The plan, with the decisions of each gap and of the stack, is plan/done/nd-level2-fixes.md.

🤖 Generated with Claude Code