dhcp: decode and encode option 41 (NIS servers)

[dylanpulver]

dhcp4msg.options is an RFC 2132 ordered table of option code → wire format, and _register_options() turns it into the decode and encode maps. Codes missing from it fall through dhcpmsg.decode() to array8.

Option 41 is absent. dhcp4msg.py:78-80 runs 40 → 85 → 42: the slot between NIS_DOMAIN (40) and NTP_SERVERS (42) holds NDS_SERVERS, which is code 85.

RFC 2132 §8.2: "The code for this option is 41. Its minimum length is 4, and the length MUST be a multiple of 4." That is byte-for-byte the same shape as NTP_SERVERS in §8.3, which makes 42 an exact control. Feeding the same eight bytes (192.168.1.10, 192.168.1.11) under codes 41, 42 and 85 into dhcp4msg(buf=...).decode():

before                                                  after
nds_servers  ['192.168.1.10', '192.168.1.11']           ['192.168.1.10', '192.168.1.11']
nis_servers  [192, 168, 1, 10, 192, 168, 1, 11]   <--   ['192.168.1.10', '192.168.1.11']
ntp_servers  ['192.168.1.10', '192.168.1.11']           ['192.168.1.10', '192.168.1.11']

decode_map[41]   None                                   CodeMapping(name='nis_servers', format='ip4list')
encode_map       nis_servers=False                      nis_servers=True

The fix is additive: NDS_SERVERS is correct in its own right (RFC 2241 §3 gives code 85 the same format), so nothing is removed, and the second test row pins that.

grep -rn "nds_servers\|NDS_SERVERS" . returns two hits repo-wide: the enum definition and that one table line. grep -rn "nis_servers" tests/ returns none. The only NIS in the tests is test_parser.py:355-356, where Option.NIS_SERVERS appears inside an asserted parameter_list — that is option 55's array of requested codes, decoded by array8, which never touches the decode map. The tests prove the client can ask for NIS servers; nothing proves it can read the answer.

Mutants

• A — revert the table line, keep both new test rows: nis_servers fails with KeyError: 'nis_servers' (it cannot be encoded).
• B — treat it as a typo and replace NDS_SERVERS with NIS_SERVERS rather than adding: nds_servers fails with KeyError: 'nds_servers'. That is what pins the fix as additive rather than a substitution.

What I ran, and what I could not

The two rows go into the existing test_encode_decode_options table. I could not run the repo's suite. make test is Linux-only and tests/test_linux/ cannot even be collected on macOS — pyroute2/ext/rawsocket.py:3 fails with ImportError: cannot import name 'AF_PACKET' from 'socket'. So I drove the parametrized table directly instead: a small runner reads the @pytest.mark.parametrize tuple out of test_encode.py with ast.literal_eval and executes the same test body. All 13 cases pass with the change; the mutants above are the same runner on the same table. Every claim about the repo's own nox sessions is therefore untested by me — CI is the first real run.

pre-commit run --files pyroute2/dhcp/dhcp4msg.py tests/test_linux/test_dhcp/test_encode.py passes (isort, black, flake8, whitespace, end-of-file).

Also not done: no live DHCP server exchange, and I did not touch Lease, which has no nis_servers accessor. Related but deliberately left alone: leases.py:173 annotates domain_search as list[str], but option 119 is likewise absent from this table so it decodes via array8 to a list of ints — that one needs an RFC 3397 name-compression decoder, not a table entry, so it is not in this PR.

Not from a bug report; found checking protocol constant tables against their RFCs. AI assistance: this change was written with an AI coding assistant.