Linux: how to get network stats on tcp (ie excluding udp)? - linux

Question

I've been using /sys/class/net/eno1/statistics/rx_bytes and tx_bytes to gather stats on my network interface. The trouble is, that network has a device (a Silicon Dust HDHOMERUN HDTV tuner) which constantly streams UDP packets at a very high rate that I don't want to monitor. I'd like to remove that traffic from the monitor - perhaps by only looking at TCP packets.
Is there any way to separate out the TCP and UDP stats?
netstat -st gives some info but it's somewhat cryptic - just how big is a 'segment'? The MTU? The man page is silent on that.
$ netstat -st | grep 'segments received'
25449056 segments received
1683 bad segments received
$ netstat -st | grep 'segments sent out'
37860139 segments sent out

Answer 1

Based on this answer from serverfault. If you are using iptables you can add a rule to each of the INPUT and OUTPUT chains which will count every packet which carries TCP in the payload. It is possible that you will need to invoke every iptables command with sudo.
Create the rules:
# Match all TCP-carrying packets incoming on 'eno1' iface
iptables -I INPUT -i eno1 -p tcp
# Match all TCP-carrying packets outgoing through 'eno1' iface
iptables -I OUTPUT -o eno1 -p tcp
Afterwards, you can use iptables -nvxL INPUT or OUTPUT to be presented with the number of bytes processed by the rule:
Chain INPUT (policy ACCEPT 9387 packets, 7868103 bytes)
pkts bytes target prot opt in out source destination
10582 9874623 tcp -- eno1 * 0.0.0.0/0 0.0.0.0/0
In case you already have other rules defined it might be handy to create a separate chain entirely. This is also described in the answer i referenced, though you also want the -i and -o options in the in/out chains respectively. These allow you to filter on a single interface (use -i for INPUT and -o for OUTPUT).
iptables -N count_in # create custom chain named 'count_in'
iptables -A count_in -j RETURN # append RETURN action to chain 'count_in'
iptables -I INPUT -j count_in # insert chain at the top of chain INPUT
iptables -I count_in 1 -i eno1 -p tcp # insert rule that matches all tcp packets on eno1
# and has no action (does nothing)
iptables -nvxL count_in # list chain 'count_in' rules
I am not sure whether the "bytes" counter includes the IP header, or just the TCP segment bytes but it is still probably the closest metric to what you want to measure (TCP-only rx/tx bytes).
Additionally keep in mind that oftentimes rules defined with iptables are not actually saved and will get deleted on a system reboot. To enable them persistently on every reboot you may use the iptables-save and iptables-restore commands. To learn their usage you should probably look in your Linux distro's documentation as well as iptables manual.
Finally, AFAIK iptables is considered legacy by now and it is being slowly replaced by nftables. I myself still have iptables installed in my system by default. If you want to switch/are already using nftables, then you need to translate above commands to the syntax supported by the nft command. There is a utility called iptables-translate available which may help with this. It's purpose is to translate old iptables commands to equivalent nft commands. I mention this mostly for the sake of completeness, you should be just fine using iptables for your particular task if you have it installed.

Answer 2

You can use iptraf-ng.
Install with:
sudo apt install iptraf-ng
This will give you statistics per protocol (IPv4/IPv6/TCP/UDP/ICMP/...) on a specific interface:
sudo iptraf-ng -d eth0
You can also use this to have details per ports:
sudo iptraf-ng -s eth0