cli(ipv6): a link-local address is not a global one, and the box refuses it #2406

Merged
jaylouisw merged 1 commit from rev/1121-1013 into main 2026-11-21 10:42:14 +00:00
Owner

Question. What does an IOS-like box do with ipv6 address fe80::1/64 - a link-local address handed to the global form of the command?

Evidence.

  • Fast Reroute, Configure, Verify, and Troubleshoot IPv6 Addressing: ipv6 address FE80::1/64 -> % Link local requires link-local keyword; ipv6 address 2001::2 link-local -> % Invalid link-local address; its running-config keeps ipv6 address FE80::1 link-local (no prefix length).
  • Cisco Community td-p/3038567 (C887VAM, IOS 15.5(3)M5): the same sentence from the keyword in the wrong place.
  • Cisco Community td-p/4678356 (SD-WAN template): ipv6 address fe80::1234/16 refused, component response quoting the same sentence.
  • Cisco IOS IPv6 Command Reference (ipv6-i1.html): the global form and ipv6 address ... link-local are separate grammar; a link-local must be configured with the link-local command.
  • RFC 4291 2.5.6 (fe80::/10); PacketLife and Cisco Community td-p/854688 for the reverse rule.

Verdict. The box refuses it. This console stored fe80::1/64 as the interface's global address and advertised fe80::/64 in its router advertisements with A=1 - a SLAAC promise no box makes.

Changed. engine/cli.js _ipv6Iface(): fe80::/10 in the global form -> % Link local requires link-local keyword plus one console line naming the MAC-derived link-local; link-local with a global address -> % Invalid link-local address; the keyword form with a link-local -> the console line only (this model keeps no manual link-local). Nothing is stored in any case. engine/ra.js: addressPrefixes() skips a link-local prefix (guards boards saved before the door). New test/ipv6-address-link-local.test.js (41 assertions); docs/CLI.md IPv6 section updated; new research/ipv6-address-link-local-refusal.md.

Tests. npm test: 445 suites, 15,024 assertions, 0 failed. node scripts/syntax-check.js: all files OK.

Unresolved. Whether the box also prints the caret line, and where the caret sits (the SD-WAN page shows both renderings); the caret for a prefix length after the keyword; a trailing non-keyword word is still ignored; modelling a manual link-local is a feature, not this fix; the rule implemented is the IOS one (Juniper accepts a link-local in the ordinary address statement).

**Question.** What does an IOS-like box do with `ipv6 address fe80::1/64` - a link-local address handed to the *global* form of the command? **Evidence.** - Fast Reroute, *Configure, Verify, and Troubleshoot IPv6 Addressing*: `ipv6 address FE80::1/64` -> `% Link local requires link-local keyword`; `ipv6 address 2001::2 link-local` -> `% Invalid link-local address`; its running-config keeps `ipv6 address FE80::1 link-local` (no prefix length). - Cisco Community td-p/3038567 (C887VAM, IOS 15.5(3)M5): the same sentence from the keyword in the wrong place. - Cisco Community td-p/4678356 (SD-WAN template): `ipv6 address fe80::1234/16` refused, component response quoting the same sentence. - Cisco IOS IPv6 Command Reference (`ipv6-i1.html`): the global form and `ipv6 address ... link-local` are separate grammar; a link-local must be configured with the `link-local` command. - RFC 4291 2.5.6 (`fe80::/10`); PacketLife and Cisco Community td-p/854688 for the reverse rule. **Verdict.** The box refuses it. This console stored `fe80::1/64` as the interface's global address and advertised `fe80::/64` in its router advertisements with A=1 - a SLAAC promise no box makes. **Changed.** `engine/cli.js` `_ipv6Iface()`: fe80::/10 in the global form -> `% Link local requires link-local keyword` plus one console line naming the MAC-derived link-local; `link-local` with a global address -> `% Invalid link-local address`; the keyword form with a link-local -> the console line only (this model keeps no manual link-local). Nothing is stored in any case. `engine/ra.js`: `addressPrefixes()` skips a link-local prefix (guards boards saved before the door). New `test/ipv6-address-link-local.test.js` (41 assertions); `docs/CLI.md` IPv6 section updated; new `research/ipv6-address-link-local-refusal.md`. **Tests.** `npm test`: 445 suites, 15,024 assertions, 0 failed. `node scripts/syntax-check.js`: all files OK. **Unresolved.** Whether the box also prints the caret line, and where the caret sits (the SD-WAN page shows both renderings); the caret for a prefix length after the keyword; a trailing non-keyword word is still ignored; modelling a manual link-local is a feature, not this fix; the rule implemented is the IOS one (Juniper accepts a link-local in the ordinary address statement).
cli(ipv6): a link-local address is not a global one, and the box refuses it
All checks were successful
CI / Classify the change (pull_request) Successful in 7s
CI / Server smoke and Docker build (pull_request) Successful in 27s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 6s
CI / Test (Node 20) (pull_request) Successful in 57s
CI / Test (Node 18) (pull_request) Successful in 58s
CI / Test (Node 22) (pull_request) Successful in 57s
CI / Arm auto-merge (pull_request) Successful in 3s
dfd4478995
`ipv6 address fe80::1/64` was stored as the interface's global address and then
advertised as `fe80::/64` with the autonomous flag set, so hosts on the wire were
told to form link-local addresses from it. IOS keeps a link-local in a different
form (`ipv6 address <addr> link-local`) and answers the mix-up with
`% Link local requires link-local keyword`; a global address given the keyword
draws `% Invalid link-local address`. Both sentences are captured, and the engine
now prints them and stores nothing. The keyword form is answered with a console
line naming the MAC-derived link-local, because this model has no manual one.
`ra.addressPrefixes()` also drops a link-local prefix, so an old board cannot put
one back on the wire.

research/ipv6-address-link-local-refusal.md has the captures and what is open.
forgejo-actions scheduled this pull request to auto merge when all checks succeed 2026-11-21 10:25:48 +00:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge!2406
No description provided.