netlink: answer non-dump RTM_GETLINK for lo

[congwang-mk]

Fixes #239

Problem

Only the dump forms of RTM_GETLINK and RTM_GETADDR were synthesized, so a lookup by name or index got EOPNOTSUPP. iproute2 resolves a device name with a non-dump RTM_GETLINK carrying IFLA_IFNAME before it does anything else, and reads any error as a missing device:

sandlock run -r /usr -r /proc -- ip link show lo      -> Device "lo" does not exist.
sandlock run -r /usr -r /proc -- ip addr show dev lo  -> Device "lo" does not exist.

Fix

• parse_request keeps the ifinfomsg index and the IFLA_IFNAME or IFLA_ALT_IFNAME attribute (the kernel accepts both).
• A non-dump RTM_GETLINK follows rtnl_getlink: a positive index wins over the name, a request with neither is EINVAL, anything but lo (index 1) is ENODEV. A match is a single RTM_NEWLINK without NLM_F_MULTI and without NLMSG_DONE.
• The reply is acked when the request sets NLM_F_ACK. libnl sets that flag on every request and waits for the ack, so it would hang otherwise. This was not tested against libnl itself.
• Non-dump RTM_GETADDR stays EOPNOTSUPP even though the issue title mentions it: that is what the kernel returns for AF_UNSPEC and AF_INET, and ip addr show dev lo only failed because of the name lookup.

After the fix, in the sandbox:

ip link show lo      -> 1: lo: <LOOPBACK,UP> mtu 65536 ...   rc=0
ip addr show dev lo  -> lo with 127.0.0.1/8 and ::1/128      rc=0
ip link show eth0    -> Device "eth0" does not exist.        rc=1
ip -br link / addr   -> unchanged

Testing

• Unit tests for the parser (index and name kept, IFLA_ALT_IFNAME, truncated attribute ignored) and for the lookup (by name, by index, index precedence, ENODEV, EINVAL, ack).
• Integration test link_lookup_by_name_finds_only_lo sends the request over a raw netlink socket from Python, so it does not need ip installed.
• Locally: netlink unit tests 16 of 16, test_netlink_virt integration tests 16 of 16. The full suite was not run locally.

🤖 Generated with Claude Code