[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

[tor-relays] Re: The attacker is back and bypassed the new anti-DoS mitigation



Hey

I experienced another Tor relay failure today, November 10, 2026. I thought the kernel logs might be useful in the context of the ongoing attacks.

My relay, Atomirotta, is running Tor 0.4.9.15 on FreeBSD 15.1-RELEASE-p3, with 1 GB RAM and 1 GB swap.

Importantly, the relay did not have the HSDir flag when this happened.

At 15:24:52 EET (13:24:52 UTC), the FreeBSD kernel killed the Tor process due to memory pressure.

Relevant kernel messages:

15:24:46 swap_pager: indefinite wait buffer
15:24:52 [zone: mbuf_jumbo_page] kern.ipc.nmbjumbop limit reached
15:24:52 pid 53847 (tor), jid 0, uid 256,
         was killed: failed to reclaim memory
15:24:53 Limiting tcp reset response from 20773
         to 190 packets/sec
15:24:54 Limiting tcp reset response from 7996
         to 193 packets/sec
15:24:55 Limiting tcp reset response from 5800
         to 203 packets/sec

The TCP reset rate remained elevated for more than a minute after Tor was killed, gradually decreasing.

Network buffer statistics collected after the failure, before restarting Tor:

326/30497/30823/30823
4k (page size) jumbo clusters
(current/cache/total/max)

13059472/0/0
requests for jumbo clusters denied
(4k/9k/16k)

Configured limits:

kern.ipc.nmbjumbop: 30823
kern.ipc.nmbclusters: 61647

The 4K jumbo-cluster allocation denial counter had reached 13,059,472. This is a cumulative kernel counter since boot, not a measurement of requests during this particular incident.

After restarting Tor, the 4K jumbo-cluster statistics dropped to:

339/9/348/30823
(current/cache/total/max)

The denial counter did not increase between these two measurements.

Traffic immediately before the failure (vnStat):

Time    RX          TX          Total
14:00   13.70 GiB   13.88 GiB   27.58 GiB
15:00    6.29 GiB    6.29 GiB   12.58 GiB
16:00   19.47 MiB   14.27 MiB   33.74 MiB

Traffic was approximately symmetrical at the hourly level. The Tor process was killed about 25 minutes into the 15:00 hour.

The VPS itself did not reboot and had an uptime of over 36 days. Swap usage measured after the failure was only 11 MB, so I cannot confirm peak swap consumption at the time of failure.

I have also experienced previous Tor processes being killed by the FreeBSD kernel under memory pressure on October 18 and October 26.

I cannot conclusively attribute today's incident to the same attacker, and I do not have BEGIN_DIR or onionskin counters from the event. However, the combination of kernel memory pressure, exhausted network-buffer allocation capacity, and the large TCP-reset burst seems worth investigating.

Has anyone else running FreeBSD Tor relays observed similar nmbjumbop exhaustion or memory-reclaim failures during these floods?

Are there any recommended mitigations or additional diagnostics for this specific failure pattern?

Regards,
Kullinaama
_______________________________________________
tor-relays mailing list -- tor-relays@xxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to tor-relays-leave@xxxxxxxxxxxxxxxxxxxx