Tor 0.4.9.17 security release: IPv6 /64 DoS fix for relays amid ongoing DDoS
Tor's 13 November release counts IPv6 clients by /64 block, fixes TROVE-2026-031 and TROVE-2026-075, and asks relay operators to upgrade now.
· 6 min read
Written by AI from the sources linked below. How we use AI
On this page
The Tor Project released Tor 0.4.9.17 on 13 November 2026, a security release that changes how relays count IPv6 clients while the network is under what the project calls an ongoing DDoS attack. Relay operators are asked to upgrade as soon as possible. It is the fifth 0.4.9.x release in about seven weeks, and the project says more are likely.
Key points
- Tor 0.4.9.17 was announced on 13 November 2026 in a Tor Project forum post (by user dgoulet), which says that it "comes in rapid response to the ongoing DDoS attack on the network".
- The headline fix (TROVE-2026-031) makes relays track IPv6 clients by their /64 prefix instead of by individual address, so a flood from one IPv6 block counts as one client for DoS defences.
- The release also rejects malformed hostname records in RESOLVED cells (TROVE-2026-075), removes the keys of directory authority "bastet" over a potential compromise, and changes the IP address of the "maatuska" authority.
- The Tor Project says it is "actively evaluating additional defenses" and expects further releases soon; no action beyond upgrading is needed after the update.
- Our sources are Tor's own announcement and release notes; we found no independent outlet with figures on the attack's size, so none are given here.
What to do
- Relay operators: upgrade to 0.4.9.17 as soon as your distribution or the Tor Project's package repository offers it. The announcement says no further action item is required afterwards. Check with
tor --version. - Directory authority operators: the release changes the maatuska address and removes the bastet keys, so authority operators should read the release notes. Ordinary relays and clients do not edit anything by hand.
- Clients and onion service operators: this release is aimed mostly at relays. Two of the minor fixes touch clients (see below), so updating is sensible but is not described as urgent for them.
- Reading statistics: if you track relay or bridge client counts from extra-info documents, expect the unique-client numbers to drop, because an IPv6 /64 is now counted once.
What happened
On 13 November 2026 the Tor Project posted the 0.4.9.17 announcement on its forum. It says the release "includes a series of security fixes" and that one in particular matters for relay DoS defences related to IPv6 handling. It describes the situation as "a massive flood of client circuits" overwhelming relays across the network, "affecting most relay operators". The project thanks the operators who have been helping on the tor-relays mailing list and tells them to keep doing so.
The release notes in Tor's repository give the details:
- IPv6 clients tracked by /64 (TROVE-2026-031, bug 41344). Tor's DoS defences watch how many connections and circuits each client opens. Before this change, an IPv6 client was identified by its single address. Since a network operator normally controls a whole /64 block (about 18 quintillion addresses), an attacker with one block could appear as a huge number of separate, individually well-behaved clients. Counting the whole /64 as one client closes that gap. The notes say the bug dates to Tor 0.3.3.2-alpha.
- Client counts in statistics. Relays and bridges now report an IPv6 /64 as a single client in their extra-info documents (ticket 41344).
- RESOLVED cells (TROVE-2026-075, bug 41464). Tor now rejects any RESOLVED cell with a hostname record that is not a valid DNS name. The fix is marked as a bugfix on 0.1.1.1-alpha, so the gap is very old.
- Stream isolation (bug 41474). For the HTTPTunnelPort, header values longer than 251 bytes had their recorded length truncated to 8 bits, so two different isolation values whose lengths agreed modulo 256 could be treated as equal and share a circuit.
- ntor-v3 (bug 41475). Relays now reject ntor-v3 subprotocol requests whose entries are not sorted and deduplicated, as was intended.
- Client directory fetches (bug 41476). A latent assertion that could fire on an indirect directory request to a relay with ORPort 0 is avoided. The notes say the current consensus has no such relays.
- Directory authorities. The keys for the authority bastet are removed "due to a potential compromise", and can be reinstated later with new keys. The authority maatuska changes IP address (ticket 41487). Fallback directories and GeoIP data were regenerated on 13 November.
The announcement calls the release "an emergency release affecting mostly relays". We could not find a public statement saying who is behind the attack, how large it is, or what the potential compromise of bastet involved. Those details are unconfirmed.
Background
0.4.9.17 is the latest in a run of fast releases. The release notes and an oss-security post show the sequence:
- 0.4.9.13 (around 23 September 2026): an earlier security release; we did not read its notes and say nothing more about it.
- 0.4.9.14 (7 October 2026): "Another week, another security release", with major bugfixes across relay, client, onion service and authority code, including a conflux fix (bug 41328) and fixes for the MiddleOnly flag (TROVE-2026-067) and expired family certificates (TROVE-2026-062).
- 0.4.9.15 (28 October 2026): new DoS defences against large volumes of directory requests, including per-circuit rate limiting of BEGIN_DIR cells on directory caches, plus a sticky HSDir flag that survives restarts. It also made new protocol capabilities required, which the notes say would stop Tor 0.4.8 and earlier working on the network.
- 0.4.9.16 (4 November 2026): an emergency release for directory authorities only. The notes say old clients "did not, in fact, correctly handle" unparseable microdescriptors, so the required "Desc" and "Microdesc" protocol versions were reverted from 3 to 2 because they were making 0.4.8.x clients exit.
- 0.4.9.17 (13 November 2026): the release covered here.
Tor has faced denial-of-service pressure for years. The project's DoS specification describes relays as exposed to many resource attacks (memory, CPU, bandwidth, channels, circuits and onion service introductions) and notes that its C implementation does not monitor CPU use directly: it limits protocol resources, such as circuit extensions, that cause CPU load. Separately, a relay operator group, 1AEO, published its own measurements of a circuit-building DoS from late June 2026 on its guard relays, about 10 percent of the network by its account. It reports open circuits peaking at 18.5 times baseline from 25 June. That is one operator's self-published data from an earlier wave, not a measurement of the current attack, and the operator itself notes that timing alone does not establish a common attacker.
How it works
A Tor circuit is built through several relays, each of which must do cryptographic handshakes and then keep state for the circuit. If an attacker opens very many circuits and leaves them idle, relays burn CPU and memory while carrying almost no traffic. That is why the announcement talks about a flood of circuits rather than a flood of bandwidth.
Rate limits per client address are the usual defence, and that is where IPv6 matters. A relay that treats each IPv6 address as a separate client can be fooled by a source that rotates through addresses inside its own block. Grouping by /64 means the limit applies to the block as a whole. The trade-off, which the notes do not discuss, is that unrelated users sharing one /64 would share a limit; that is also how the statistics change. We have not seen a published analysis of that effect.
What it means
For relay operators, the practical message is to patch now and keep patching: the cadence of releases since late September means the stable version they run may be out of date within days. For the network, the fixes target stability under load; the notes do not claim they stop the attack.
For users, nothing needs doing to browse. Relay overload can still mean slower or less reliable connections and onion service reachability problems, but we found no official figures on user impact in this episode. The authority changes are also a reminder that the directory authorities are the network's trust anchor: removing a key "due to a potential compromise" is a precaution that the project has not explained in the notes.
What we don't know yet
- How large the attack is, how long it has lasted and whether one party is behind it. The announcement gives none of this.
- What the "potential compromise" of the bastet authority was.
- Whether grouping IPv6 clients by /64 is enough, or what the "additional defenses" will be.
- Whether users have seen measurable slowdowns.
What to watch next
- The next 0.4.9.x release, which the project says to expect soon.
- Release notes for new TROVE identifiers, and any explanation of bastet.
- Posts on the Tor Project forum and the tor-relays mailing list from relay operators.
Sources
Written by
The official account of the people who run audiotale.net: announcements, conversation starters and the newsroom's drafts. Posts written with AI say so. Every other account here is a real person.
Read next
uBlock Origin GHSA-cmqw-9r2w-6wv2: element picker XSS fixed in 1.75.4
A web page could drive uBlock Origin's element picker and add trusted filters with one click; the High-severity flaw is patched, so check you run 1.75.4 or newer.
audiotale.net team6 min read
Cloudflare MOLE: post-quantum Privacy Pass design with 4.3KB to 12KB tokens, not yet deployed
Cloudflare researchers propose MOLE, a post-quantum framework for anonymous rate-limiting tokens, with estimated sizes far below a 2023 attempt but no proofs or standard yet.
audiotale.net team5 min read

Harvard breach added to Have I Been Pwned: 850,500 alumni and donor emails, what to do
Have I Been Pwned has added the Harvard alumni and fundraising breach, with about 850,500 addresses and details such as donations and family members.
audiotale.net team6 min read