Troubleshooting hosts and LANs
Apply a layered method to IPv6 host and LAN faults, from link to application. Interpret ICMPv6 error messages and codes (RFC 4443) and recognise rate limiting. Diagnose Path MTU Discovery failures (RFC 8201) and calculate MTU, MSS and ping sizes. Find and fix six common LAN faults in a namespace lab.
Take the quiz · pass mark 70% · 9 questions
Explanation
Level 3 has covered how hosts and LANs should work. This lesson turns that into a troubleshooting method you can use on a support call, in a NOC or in the practical part of the Level 3 exam.
A layered method
Work from the bottom up and prove each layer before moving on. Most IPv6 LAN faults are found in the first five steps.
| Step | Question | Commands (Linux; Windows equivalent) |
|---|---|---|
| 1. Link | Is the interface up, in the right VLAN, with IPv6 enabled? | ip link, sysctl net.ipv6.conf.eth0.disable_ipv6; Get-NetAdapter, Get-NetAdapterBinding -ComponentID ms_tcpip6 |
| 2. Link-local | Does the host have a link-local address, not tentative or dadfailed? | ip -6 addr; Get-NetIPAddress -AddressFamily IPv6 |
| 3. Router discovery | Does an RA arrive, from the right router, with the expected flags and prefixes? | rdisc6 eth0, capture on type 134; netsh interface ipv6 show interface |
| 4. Addresses and routes | Are global addresses preferred (not deprecated)? Is there a default route via a link-local address? | ip -6 addr, ip -6 route, ip -6 route get <dst>; Get-NetRoute -AddressFamily IPv6 |
| 5. Neighbours | Can the host resolve the router (REACHABLE/STALE, not FAILED)? | ip -6 neigh, ndisc6 <addr> eth0; Get-NetNeighbor -AddressFamily IPv6 |
| 6. Beyond the router | Does ping -6 reach the router, then an Internet address? Where does traceroute -6 stop? | ping -6, traceroute -6, mtr -6 |
| 7. Path MTU | Do large packets pass? | tracepath -6, ping -6 -s 1452 -M do |
| 8. DNS | Do names resolve over IPv6, with correct AAAA records? | dig -6, resolvectl status; Resolve-DnsName -Type AAAA |
| 9. Application | Does the service listen on IPv6, and do firewalls allow it? | ss -6ltn, curl -6 -v; Test-NetConnection -ComputerName <name> -Port 443 |
Capture early. One tcpdump -ni eth0 icmp6 running while you test usually shows the fault faster than reading configuration files.
ICMPv6 errors (RFC 4443)
| Type | Message | Codes |
|---|---|---|
| 1 | Destination Unreachable | 0 no route, 1 administratively prohibited, 2 beyond scope of source, 3 address unreachable, 4 port unreachable, 5 source address failed ingress/egress policy, 6 reject route to destination |
| 2 | Packet Too Big | Carries the MTU of the next hop |
| 3 | Time Exceeded | 0 hop limit exceeded in transit (what traceroute uses), 1 fragment reassembly time exceeded |
| 4 | Parameter Problem | 0 erroneous header field, 1 unrecognised Next Header, 2 unrecognised IPv6 option |
| 128 / 129 | Echo Request / Reply |
Reading the code saves time:
- Code 3, address unreachable, comes from the last router, after its address resolution for the destination failed (L3.01). The host is down, in another VLAN, or not answering NS.
- Code 1 or code 6 means a filter or a reject route, not a broken link.
- Code 5 means a source address filter (for example BCP 38 at the ISP) dropped the packet, often because the host used the wrong source address in a multihomed network (L3.04).
Rate limiting. RFC 4443 requires nodes to limit the rate of ICMPv6 error messages and recommends a token bucket. Linux limits most ICMPv6 errors per destination (net.ipv6.icmp.ratelimit, 100 ms by default), but excludes Packet Too Big. Routers rate-limit too. Missing hops (* * *) in a traceroute or missing replies in a fast ping can be rate limiting, not loss.
Do not filter ICMPv6 blindly. ND, MLD and PMTUD all depend on it. RFC 4890 lists which ICMPv6 messages firewalls must allow; Level 5 covers it in detail.
Path MTU Discovery (RFC 8201)
IPv6 routers do not fragment. If a packet is too big for the next link, the router drops it and sends Packet Too Big (type 2) with the next-hop MTU. The source lowers its Path MTU for that destination and resends smaller packets. RFC 8200 requires every link to carry at least 1280 bytes, and RFC 8201 says a node must not reduce its Path MTU below that. A node may try a larger PMTU again no sooner than 5 minutes after a PTB, and should wait 10 minutes by default.
PMTU black holes. If a firewall drops PTB messages, small packets work and large ones disappear. The classic symptom: TCP connects, small pages load, but large downloads, TLS handshakes with big certificates, or SSH sessions after login hang. Diagnose with:
ping -6 -c3 -s 1452 -M do 2001:db8:f000::80 # 1452 + 8 + 40 = 1500 bytes
ping -6 -c3 -s 1232 -M do 2001:db8:f000::80 # 1280 bytes, always fits
tracepath -6 2001:db8:f000::80 # shows "pmtu" per hop
ip -6 route get 2001:db8:f000::80 # shows a cached "mtu" after a PTB
Fixes, in order of preference: let PTB through every firewall on the path; set a correct MTU on the tunnel or PPPoE link (and advertise it in the RA MTU option); clamp the TCP MSS on the router as a workaround for TCP only.
TCP MSS over IPv6 = MTU − 40 (IPv6 header) − 20 (TCP header). Common values:
| Link | MTU | IPv6 TCP MSS |
|---|---|---|
| Ethernet | 1500 | 1440 |
| PPPoE | 1492 | 1432 |
| 6in4 tunnel over 1500 | 1480 | 1420 |
| Minimum IPv6 MTU | 1280 | 1220 |
Six common LAN faults
| Fault | What you see | Fix |
|---|---|---|
No RA (radvd down, RA disabled, router accept_ra-style misconfiguration) | Only link-local addresses, no default route | Restore RAs on the router |
| Rogue RA | Second default route via an unknown fe80:: address, unexpected prefix | Find and stop the sender (MAC in the capture); RA Guard on switches |
| Duplicate address | dadfailed on a static address | Find the other owner (NA in capture), renumber |
| ICMPv6 dropped on the host | Neighbour entries stay INCOMPLETE or FAILED; the router cannot reach the host | Allow ND types 133 to 137 in the host firewall |
| PMTU black hole | Small packets pass, large ones vanish | Allow PTB, fix MTU, clamp MSS |
| Wrong or missing DNS | Ping by address works, names fail | Correct RDNSS / DHCPv6 DNS, resolver ACLs |
Drawing diagram…
Worked examples
Example 1: Ping size for a 1500-byte packet
ICMPv6 Echo has an 8-byte header and the IPv6 header is 40 bytes, so a 1500-byte packet needs -s 1500 − 48 = 1452.
Example 2: MSS on PPPoE
PPPoE reduces the MTU to 1492. The IPv6 TCP MSS is 1492 − 40 − 20 = 1432.
Example 3: Tunnel inside the network
A 6in4 tunnel (20-byte IPv4 header) on a 1500-byte path has MTU 1480. Hosts behind it should use an MSS of 1420. If the router advertises MTU 1480 in its RAs, hosts use it from the start.
Example 4: When will the host try a larger MTU?
A host received a PTB at 10:00:00 local time. RFC 8201 lets it try a larger PMTU no earlier than 10:05:00 (5 minutes), and by default it waits 10 minutes, until 10:10:00.
Example 5: Reading a Destination Unreachable
A capture on the customer's side shows ICMPv6 type 1 code 3 from 3fff:a00:1003::1. The hypervisor's router tried to resolve the VPS address and got no NA: the VPS is down, its address is wrong, or its firewall drops NS.
Lab
Goal: Diagnose and fix six faults on a namespace LAN using the layered method.
You need: The L3.01 LAN with radvd on r1 (L3.02), plus a DNS server namespace dns (2001:db8:3000:f00::53) and a web server namespace web (2001:db8:f000::80) behind a second router r2, connected to r1 by a point-to-point link. Work in pairs: one person injects a fault using the instructor notes below without telling the other which one, and the other diagnoses it. Swap roles after three faults.
For each fault:
- Record the user-visible symptom (what fails, what works).
- Work through the method table and record the first step that fails, with the command output.
- Capture the relevant traffic and annotate it.
- Fix the fault and show the symptom is gone.
- Write one line: root cause, evidence, fix.
For the PMTU fault, show the difference between ping -6 -s 1300 -M do and ping -6 -s 1400 -M do from web to h1, and between curl of a small and a large file from h1.
Instructor notes (fault injection)
Inject one fault at a time and undo it before the next:
- No RA: stop radvd on
r1, then restarth1's interface so its old addresses go. - Rogue RA: in
h2, run radvd advertisingfdf1:a861:26b6:99::/64withAdvDefaultPreference high. - Duplicate address: add
h1's static address toh2beforeh1boots its interface. - ICMPv6 dropped on the host: in
h1,ip6tables -A INPUT -p icmpv6 --icmpv6-type neighbour-solicitation -j DROP, then flush the router's neighbour cache. - PMTU black hole: set MTU 1400 on both ends of the
r1–r2link, and inr2runip6tables -A OUTPUT -p icmpv6 --icmpv6-type packet-too-big -j DROP. The small-MTU link must not be next to either endpoint, because TCP's MSS negotiation would otherwise hide the problem. - DNS: change radvd's RDNSS on
r1to an address where no resolver runs (2001:db8:3000:f00::99).
What to submit
A one-page report per fault with the five items above, plus the time you took per fault.
Summary
- Troubleshoot bottom up: link, link-local and DAD, RA, addresses and routes, neighbours, beyond the router, PMTU, DNS, application. Capture early.
- ICMPv6 codes point at causes: code 3 = address resolution failed at the last router, 1 and 6 = filtering or reject, 5 = source address policy. Errors are rate-limited.
- PMTUD (RFC 8201): PTB must pass. Minimum MTU 1280. Ping
-s= MTU − 48. IPv6 TCP MSS = MTU − 60. - Six common faults: no RA, rogue RA, duplicate address, ICMPv6 dropped on the host, PMTU black hole, DNS.
- The Level 3 exam includes a 30-minute break/fix practical like this lab.
Quiz imported · 9 questions · pass mark 70%. Lab questions are not auto-marked.