Few tools for a Tor relay.

The scripts ipv4-rules.sh and ipv6-rules.sh protect a Tor relay against DDoS attacks¹ at the IP network layer, as seen in this metrics:

An older example is here.

¹ see ticket 40636 and 40093 of the Tor project.

Mark an IP address as malicious if its connection attempts over a short time period exceed a given threshold. Block that IP for a much longer time period.

Therefore a simple network rule won't make it. However ipset helps to achieve the goal. Further considerations:

• never touch established connections
• try to not overblock

Install jq, ipset and iptables, e.g. for Debian or Ubuntu do:

sudo apt update
sudo apt install -y jq ipset iptables

Download the DDoS prevention scripts

wget -q https://raw.githubusercontent.com/toralf/torutils/main/ipv4-rules.sh -O ipv4-rules.sh
wget -q https://raw.githubusercontent.com/toralf/torutils/main/ipv6-rules.sh -O ipv6-rules.sh
chmod +x ./ipv4-rules.sh ./ipv6-rules.sh

Make a backup of the current iptables filter table:

sudo /usr/sbin/iptables-save >./rules.v4
sudo /usr/sbin/ip6tables-save >./rules.v6

Run a quick test

sudo ./ipv4-rules.sh test
sudo ./ipv6-rules.sh test

Best is to stop the Tor service(s) now. Then flush the connection tracking table

sudo /usr/sbin/conntrack -F

and (re-)start the Tor service. Check that your ssh login and your other services are still working. Watch the iptables live statistics by:

sudo watch -t ./ipv4-rules.sh # replace 4 with 6 for IPv6

If something failed then restore the backuped state:

sudo ./ipv4-rules.sh stop
sudo ./ipv6-rules.sh stop
sudo /usr/sbin/iptables-restore <./rules.v4
sudo /usr/sbin/ip6tables-restore <./rules.v6

But if everything looks uk, run the script again, but now with the parameter start:

sudo ./ipv4-rules.sh start
sudo ./ipv6-rules.sh start

Finally, ensure, that the package iptables-persistent is either de-installed or at least disabled and create cron jobs (via crontab -e)e.g.:

# DDoS prevention
@reboot /root/ipv4-rules.sh start; /root/ipv6-rules.sh start

# have recent ipset data after a reboot available
@hourly /root/ipv4-rules.sh save; /root/ipv6-rules.sh save

# update Tor authorities
@daily  /root/ipv4-rules.sh update; /root/ipv6-rules.sh update

issue reports about any findings are welcome.

The DDoS scripts creates generic filter rules for the local network, ICMP, ssh, DHCP and (if defined) additional services. Then this rule set is applied ¹:

1. trust connection attempt from trusted Tor authorities/Snowflake servers
2. block the source ² for 24 hours if the connection attempt rate from the source to the Tor port exceeds > 4/min ³ within last 2 minutes
3. ignore the connection attempt if there are already 4 established connections to the Tor port
4. accept the connection attempt to the Tor port

¹ for IPv4 with "source" a single ip address is meant, for IPv6 a /80 block is meant

² the value is derived from ticket 40636

³ for IPv4 look here, for IPv6 here

If the parsing of the Tor config and/or of the SSH config fails then:

1. define the local running relay/s explicitly at the command line after the keyword start, e.g.:

   sudo ./ipv4-rules.sh start 1.2.3.4:443 5.6.7.8:9001

2. -or- define them as environment variables, e.g.:

   sudo CONFIGURED_RELAYS="5.6.7.8:9001 1.2.3.4:443" ./ipv4-rules.sh start

   (CONFIGURED_RELAYS6 for IPv6).

A command line argument takes precedence over the corresponding environment variable.

Allow inbound traffic to additional address:port destinations by e.g.:

export ADD_LOCAL_SERVICES="2.71.82.81:828 3.141.59.26:53"
export ADD_LOCAL_SERVICES6="[cafe::abba]:1234"

A slightly different syntax (as the separator > is used instead : to highlight that the address is src, and the port is dst) is used for ADD_REMOTE_SERVICES to allow inbound traffic, e.g.:

export ADD_LOCAL_SERVICES="4.3.2.1>4711"
export ADD_LOCAL_SERVICES6="[cafe::abba]>4711"

The above allows traffic from the specified remote address to the local port 4711.

The script sets few sysctl values. If unwanted then please comment out the call of setSysctlValues(). If Hetzners system monitor isn't used, then comment out addHetzner(). To append rules onto existing iptables rule set (overwrite is the default) please comment out the call clearRules().

The script metrics.sh exports DDoS metrics into a Prometheus-formatted file. The scrape of it is handled by node_exporter. More details plus few Grafana dashboards are here.

Graphs¹ of rx/tx packets, traffic and socket counts from 5th, 6th and 7th of Nov show the results for few DDoS attacks over 3 days for 2 relays. A more heavier attack was observed at 12th of Nov. A periodic drop down of the socket count metric, vanishing over time, appeared at 5th of Dec. Current attacks e.g. at the 7th of March are still handled well. Few more helper scripts were developed to analyze the attack vector. Look here for details.

¹ using sysstat, created e.g. by

# create the SVG file
svg=/tmp/graph.svg
sadf -g -t /var/log/sa/sa${DAY:-`date +%d`} -O skipempty,oneday -- -n DEV,SOCK,SOCK6 --iface=enp8s0 >$svg
# fix SVG canvas size
h=$(tail -n 2 $svg | head -n 1 | cut -f 5 -d ' ')
sed -i -e "s,height=\"[0-9]*\",height=\"$h\"," $svg
# display it
firefox $svg

I used this Ansible role to deploy and configure Tor relays and Snowflake stahndalone proxies.

info.py gives a summary of all Tor related TCP connections, e.g.:

sudo ./info.py --address 127.0.0.1 --ctrlport 9051

 ORport 9051  0.4.8.0-alpha-dev   uptime: 02:58:04   flags: Fast, Guard, Running, Stable, V2Dir, Valid
+------------------------------+-------+-------+
| Type                         |  IPv4 |  IPv6 |
+------------------------------+-------+-------+
| Inbound to our OR from relay |  2304 |   885 |
| Inbound to our OR from other |  3188 |    68 |
| Inbound to our ControlPort   |       |     1 |
| Outbound to relay OR         |  2551 |   629 |
| Outbound to relay non-OR     |       |       |
| Outbound exit traffic        |       |       |
| Outbound unknown             |    16 |     4 |
+------------------------------+-------+-------+
| Total                        |  8059 |  1587 |
+------------------------------+-------+-------+
 relay OR connections  6369
 relay OR ips          5753
    3 inbound v4 with > 2 connections each

If your Tor relay is an Exit then ps.py gives live statistics about those network connections:

sudo ./ps.py --address 127.0.0.1 --ctrlport 9051

orstatus.py prints the closing reason to stdout, orstatus-stats.sh prints/plots statistics (see this example) from that.

orstatus.py --ctrlport 9051 --address 127.0.0.1 >>/tmp/orstatus &
sleep 3600
orstatus-stats.sh /tmp/orstatus

key-expires.py helps to maintain Tor offline keys. It returns the expiration time in seconds of the mid-term signing key. A cronjob for that task could look like this:

seconds=$(sudo ./key-expires.py /var/lib/tor/keys/ed25519_signing_cert)
days=$((seconds / 86400))
[[ $days -lt 23 ]] && echo "Tor signing key expires in less than $days day(s)"

If Tor metrics are enabled then this 1-liner does a similar job (replace 9052 with the metrics port):

date -d@$(curl -s localhost:9052/metrics | grep "^tor_relay_signing_cert_expiry_timestamp" | awk '{ print $2 }')

An open Tor control port is needed to query the Tor process via API. Configure it in torrc, e.g.:

ControlPort 127.0.0.1:9051

The python library Stem is needed. Install it either by your package manager, e.g. for Ubuntu:

sudo apt install python3-stem

-or- (better) use the more recent git sources, e.g.:

git clone https://github.com/torproject/stem.git
export PYTHONPATH=$PWD/stem

The script watch.sh helps to constantly monitor the host and Tor log files. It sends findings via mailx.

log=/tmp/${0##*/}.log

# watch syslog
/opt/torutils/watch.sh /var/log/messages /opt/torutils/watch-messages.txt &>>$log &
# watch Tor
/opt/torutils/watch.sh /var/log/tor/notice.log /opt/torutils/watch-tor.txt -v &>>$log &