AS12637 SEEWEB SEEWEB s.r.l.

AS overview

Test target: 2001:4b78::22 · dns2.seeweb.it · address space: mixed · prefixes announced: 10

prefix(es): 2a05:b6c1::/32, 2a02:e9::/32, 2a02:ea::/32, 2a0d:b000::/29, 2a04:2700::/29, 2a05:b6c0::/32, 2a0b:fd40::/29, 2001:4b78::/29 (+2 more)

Targets & per-vantage-point probe results (10 target(s) across 3 vantage point(s))

Source codes (hover any badge for full description): CUR=curated · NS/SOA/MX=DNS records · SPF=SPF TXT · DRV=holder-derived (www, ns1, mail, gw, …) · ATL=RIPE Atlas · PDB=PeeringDB · WHOIS=RIPE whois · RDAP=RIPE RDAP · HIT=IPv6 Hitlist · RTR=in-prefix path hop (router) · PRB=static prefix probe · SYN=synthetic prefix::1 fallback

Target IP / hostname	Source	NL host (go6lab)					ITA host (Karsolink)					SLO host (6connect)
Target IP / hostname	Source	ping	1500B	:80	:443	reached	ping	1500B	:80	:443	reached	ping	1500B	:80	:443	reached
`2001:4b78:1:20::28` smtp-avas.seeweb.it	MX	✓	✓	?	?	✓	✓	✓	refused	refused	✓	✓	✓	refused	refused	✓
`2001:4b78:1:3::140` dns1.seeweb.it	drv	✓	✓	?	?	✓	✓	✓	refused	refused	✓	✓	✓	refused	refused	✓
`2001:4b78:1:3::143` ns3.seeweb.it	drv	✓	✓	?	?	✓	✓	✓	refused	refused	✓	✓	✓	refused	refused	✓
`2001:4b78:1:3::4` ns2.seeweb.it	drv	✓	✓	?	?	✓	✓	✓	refused	refused	✓	✓	✓	refused	refused	✓
`2001:4b78:1:a::75` anubi.seeweb.it	SPF	✓	✓	open	open	✓	✓	✓	open	open	✓	✓	✓	open	open	✓
`2001:4b78:1:f:25::138` rah-b.seeweb.it	MX	✓	✓	?	?	-	✓	✓	?	?	-	✓	✓	?	?	-
`2001:4b78:1:f::132` git.seeweb.it	drv	✓	✓	?	?	-	✓	✓	?	?	-	✓	✓	?	?	-
`2001:4b78::22` dns2.seeweb.it	NS	✓	✓	?	?	✓	✓	✓	refused	refused	✓	✓	✓	refused	refused	✓
`2001:4b78::5666` monitor.seeweb.it	drv	✓	✓	?	open	-	✓	✓	?	open	-	✓	✓	?	open	-
`2001:4b7a:100:1::1` ns3.freedns.si	NS	✓	✓	?	?	✓	✓	✓	refused	refused	✓	✓	✓	refused	refused	✓

MIX-IT / NAMEX / MINAP / TOP-IX / VSIX / EQUINIX-MILAN / BGPX-ROME / NAMEX-BARI / PCIX / DECIX-PALERMO peering

MIX-IT / NAMEX / MINAP / TOP-IX / VSIX / EQUINIX-MILAN / BGPX-ROME / NAMEX-BARI / PCIX / DECIX-PALERMO member - Open peering, peering member, since 2011-07-01. IPv6 LAN: 2001:7f8:b:100:1d1:a5d1:2637:86. Locations: -.

RPKI & ASPA

10 of 10 prefix(es) covered by a valid ROA.

Prefix	RPKI state	Reason	Covering VRP(s)
`2a05:b6c1::/32`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /29 maxLen 48 (ripe)
`2a02:e9::/32`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /32 maxLen 32 (ripe)
`2a02:ea::/32`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /32 maxLen 48 (ripe)
`2a0d:b000::/29`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /29 maxLen 48 (ripe)
`2a04:2700::/29`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /29 maxLen 48 (ripe)
`2a05:b6c0::/32`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /29 maxLen 48 (ripe)
`2a0b:fd40::/29`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /29 maxLen 48 (ripe)
`2001:4b78::/29`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /29 maxLen 48 (ripe)
`2001:4b78::/32`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /32 maxLen 48 (ripe); AS12637 /29 maxLen 48 (ripe)
`2a05:be80::/29`	✓ valid	matched VRP (correct origin, length within maxLength)	AS12637 /29 maxLen 48 (ripe)

ASPA: this AS has published an ASPA record listing 5 upstream provider(s) that are authorized to propagate routes from it: AS174, AS1299, AS2914, AS3303, AS50673. Routes from this AS that arrive via any other upstream are considered ASPA-invalid by validators that enforce ASPA.

Random-IP probe (yarrp) into the AS' prefixes

At least one router inside AS12637's announced prefix replied during the random-target probe. Hop(s): 2001:4b7a:0:f000::2. These are candidate targets for direct testing.

ICMPv6 Type 2 (Packet Too Big) acceptance - active test

Vantage points disagree about Type 2 acceptance - transit ASes on one path may be filtering Type 2 even though the destination's stack accepts it on another path:

NL host (go6lab): Type 2 NOT honored. Forged PTB had no effect (filtered en route or stack ignored it). (ICMPv6 Echo + forged PTB)
ITA host (Karsolink): inconclusive: no Echo Reply to 1300-byte probe. (ICMPv6 Echo + forged PTB)
SLO host (6connect): Type 2 honored. Forged PTB accepted; next response shrunk. (ICMPv6 Echo + forged PTB)

Why they disagree - path-asymmetric Type 2 filtering: Both vantages used the ICMPv6 Echo + forged-PTB method on the same destination IP, but reached different verdicts. The destination's stack is the same in both cases - so the difference is in the path: a transit AS on one route filters our forged Type 2 (or its return ICMPv6 carrying the embedded packet) while the other route delivers it intact. The vantage that says honored shows what the destination's stack actually does; the one that says not honored shows that the path between us and them drops Type 2.

Failure detail — what to grep in your logs

vantage	our source	your target	method	result	tested (UTC)
go6lab	`2a00:8642:42::75`	`2001:4b78::22`	`icmp6-echo`	not_honored	2026-05-30T12:41:46Z
karsolink	`2a12:d8c0:105a:9001::a154`	`2001:4b78::22`	`icmp6-echo`	no_echo	2026-05-30T12:42:56Z
odin	`2607:fae0:a000::42`	`2001:4b78::22`	`icmp6-echo`	honored	2026-05-30T12:41:39Z

Attempt log (go6lab):

icmp6-echo → not_honored (size_before=1360, size_after=1360)
dns-tcp → no_tcp: no DNS response data
tls → no_tcp: TLS connect failed
http → no_tcp: tcp/80 not open

Attempt log (karsolink):

icmp6-echo → no_echo: no Echo Reply to 1300-byte probe
dns-tcp → inconclusive (size_before=156, size_after=None): max DNS-over-TCP segment 156B; not big enough to test PMTU shrink
tls → no_tcp: TLS connect failed
http → no_tcp: tcp/80 not open

Attempt log (odin):

icmp6-echo → honored (size_before=1460, size_after=None)
dns-tcp → inconclusive (size_before=156, size_after=None): max DNS-over-TCP segment 156B; not big enough to test PMTU shrink
tls → no_tcp: TLS connect failed
http → no_tcp: tcp/80 not open

To match the corresponding ICMPv6 packet on your side (host firewall, AS edge, or transit tap), look for our PTBs around the timestamps above:

sudo tcpdump -i any -n -e 'icmp6 and ip6[40] = 2 and (src host 2607:fae0:a000::42 or src host 2a00:8642:42::75 or src host 2a12:d8c0:105a:9001::a154)'

If you see our PTBs arriving but the destination's TCP/Echo flow does not shrink, the drop is in the destination kernel (cause 3 below). If you don't see them at all, drop is upstream of you (cause 2). If you only see them from one of our two source IPs, the drop is path-asymmetric — one transit on the asymmetric route is filtering, the other is not.

What does "Type 2 not honored" actually mean? — click to expand

What this test does

Using the icmp6-echo method, we send a 1500-byte ICMPv6 Echo Request, then a forged ICMPv6 Type 2 (Packet Too Big) declaring path MTU=1280, and observe whether the next Echo Reply arrives split into IPv6 fragments. RFC 4890 requires hosts and intermediate networks not to filter ICMPv6 Type 2; the destination's TCP/UDP stack must act on a received PTB by lowering its Path MTU cache for that destination, which makes subsequent segments smaller.

What we measured

The TCP segment size your server emitted before our forged Type 2 was 1360 B; after, it was 1360 B. No change. RFC 4890 ("Type 2 messages MUST NOT be filtered") expects subsequent segments to shrink to fit a Path MTU of 1280 B.

Three plausible causes

Your host firewall is dropping ICMPv6 Type 2 inbound. Many default firewall rule sets only allow Echo Request/Reply and Neighbor Discovery, silently dropping all other ICMPv6 types - including Packet Too Big.
An upstream / transit network is dropping ICMPv6 Type 2 before it reaches you. Some transit ASes filter ICMPv6 messages other than Echo at the edge. The forged PTB never arrives, so your stack never has a chance to act on it.
Your kernel is ignoring the PTB. Linux / BSD stacks normally accept ICMPv6 PTB and update the route cache, but a few sysctls (or a hardened kernel) can be configured to ignore PMTU updates - typically as part of an over-aggressive anti-spoofing or uRPF policy.

How to check & fix (Linux examples)

1. Confirm Type 2 is not blocked at the host firewall:

sudo ip6tables -L INPUT -nv | grep -iE 'icmpv6|packet-too-big'
sudo nft list ruleset 2>/dev/null | grep -A1 'icmpv6'

If you see rules dropping ICMPv6 unconditionally, change them to permit at least icmpv6 type packet-too-big (and destination-unreachable, time-exceeded, parameter-problem per RFC 4890).

2. Confirm the kernel accepts incoming PTB:

sudo sysctl net.ipv6.conf.all.accept_redirects net.ipv4.ip_no_pmtu_disc net.ipv6.route.mtu_expires

The defaults (accept_redirects=1, ip_no_pmtu_disc=0) are the right values for honoring PTB.

3. Live trace: while we have an open TCP flow with a small MSS (we run our test from 2607:fae0:a000::42 on odin, 2a00:8642:42::75 on go6lab and 2a12:d8c0:105a:9001::a154 on karsolink), watch for our forged Type 2 arriving on your interface:

sudo tcpdump -i any -n -e 'icmp6 and ip6[40] = 2 and (src host 2607:fae0:a000::42 or src host 2a00:8642:42::75 or src host 2a12:d8c0:105a:9001::a154)'

If you see our PTBs arriving but TCP segments still stay big, the drop is in your kernel or NIC offload (cause 3). If you don't see them at all, the drop is upstream of you (cause 2) - ask your upstream(s) to permit ICMPv6 Type 2.

4. If your test target above (2026-05-30T12:41:46Z) is a host that you don't own (e.g. a third-party DNS / TLS server you happen to operate prefixes for), the verdict reflects that specific host's behaviour - try the test against a server you do own and we'll happily re-run.

RFC 4890 references: §4.3.1 (Packet Too Big - MUST NOT be dropped), and RFC 8201 for the broader PMTUD requirement.

Where the path divergence is

Per-vantage probe + verdict (headline):

go6lab: probe icmp6-echo → verdict not_honored
karsolink: probe icmp6-echo → verdict no_echo
odin: probe icmp6-echo → verdict honored

Full per-method matrix (all four methods run at each vantage):

vantage	icmp6-echo	dns-tcp	tls	http
go6lab	not_honored	no_tcp	no_tcp	no_tcp
karsolink	no_echo	inconclusive	no_tcp	no_tcp
odin	honored	inconclusive	no_tcp	no_tcp

These vantage-level disagreements are rooted somewhere in the forward paths. Joining each per-vantage traceroute against the IP→AS lookup from our global yarrp mesh, the first hop where the paths land in different ASes is the most likely site of the offending filter / unreachable AS / Type 2 drop.

Suspect transit AS (ranked by how often they appear at the divergence point on the path of the worse-classifying vantage): AS12637.

From go6lab (open/Type-2=not_honored) the path enters an opaque hop at hop 9; from odin (open/Type-2=honored) the same hop is in AS12637. Last common AS: AS12637. The disagreement is most likely rooted in one of those two transit ASes.
From karsolink (open/Type-2=no_echo) the path enters an opaque hop at hop 9; from odin (open/Type-2=honored) the same hop is in AS12637. Last common AS: AS12637. The disagreement is most likely rooted in one of those two transit ASes.

Diagnosis is path-level, not packet-level: it tells you which transit AS is the prime suspect, not exactly which firewall rule is to blame. Use the tracepath6 output below (when available) for per-hop PMTU evidence on the same path.

✨ Diagnostic interpretation

An operator-targeted diagnostic interpretation can be generated for this AS on demand. The pipeline holds the raw verdicts and the per-hop walk; an AI pass synthesises where to start looking. Click the button below to run it.

Per-hop PTB acceptance walk

From the local vantage, we walk the forward path hop-by-hop, sending each hop a 1500-byte ICMPv6 Echo, then a forged PTB (MTU=1280) sourced from us, then another 1500-byte Echo. If the second reply arrives fragmented or smaller, that hop honoured the PTB. If unchanged, it didn't. The first ✗ in an otherwise-✓ path is the most likely filter location. Cross-country aggregation: see the global PTB filter atlas for transit ASes ranked by filter rate across all measurements.

#	hop IP	AS	Holder	PTB acceptance
1	2607:fae0:a000::2	AS8038	6CONNECT - 6connect, Inc., US	- skipped (CoPP)
2	2607:fae0:a000:2::2	AS8038	6CONNECT - 6connect, Inc., US	- skipped (CoPP)
3	*	-		- (no IP)
4	*	-		- (no IP)
5	*	-		- (no IP)
6	*	-		- (no IP)
7	*	-		- (no IP)
8	2001:7f8:b:100:1d1:a5d1:2637:86	-	NA	? no_response
9	2001:4b78:2:a::8	AS12637	SEEWEB - SEEWEB s.r.l., IT	? no_response
10	2001:4b78::22	AS12637	SEEWEB - SEEWEB s.r.l., IT	? no_response

Tests host-mode PTB acceptance (PTBs aimed at the hop itself). A router that honours PTBs to itself can still be filtering PTBs transiting through it; this is one indicator, not proof of full PTB transparency. Hops in CoPP-rate-limit ranges are skipped to avoid false signals.

tracepath6 per-hop PMTU drilldown

For each target where Type 2 verdicts disagreed across vantages, tracepath -6 ran from every vantage to capture per-hop PMTU evolution along that vantage's actual forward path. A pmtu change entry on a hop means that hop's router generated a PTB and we observed the shrink; absence of any change combined with a not_honored verdict suggests a router somewhere downstream is silently dropping >MTU packets (an RFC 4890 violation) rather than sending a PTB.

Target `2001:4b78::22`

go6lab — verdict not_honored

verdict = not_honored; final pmtu = 1500

#	hop IP	pmtu change
1	<span class='muted'>no reply</span>
2	2a00:8642:1000:f000::1
3	2a00:8642:dc1:2::b
4	2001:7f8:1::a501:2637:1
5	2001:4b78:2:a::8
6	2001:4b78::22

karsolink — verdict no_echo

verdict = no_echo; final pmtu = 1500

#	hop IP	pmtu change
1	<span class='muted'>no reply</span>
2	2a12:d8c0:109f:121::a1
3	2a12:d8c0:101f:103::1
4	2001:7f8:b:100:1d1:a5d1:2637:86
5	2001:4b78:2:a::8
6	2001:4b78::22

odin — verdict honored