A modern alternative network traffic sniffer inspired by bandwhich(Rust) and nethogs(C ).
sniffer.mov
sniffer is designed for network troubleshooting. It can be started at any time to analyze the processes or connections causing increases in network traffic without loading any kernel modules. By the way, the TUI of it is responsive that can fit with terminals of all sizes automatically.
sniffer manipulates gopacket to sniff the interfaces and record packets' info. gopacket wraps the Golang port of libpacp
library, and provides some additional features. One of the projects that inspired the sniffer is bandwhich
, which has a sophisticated interface and multiple ways to display data, but it does not support BPF filters. Another one is nethlogs
, which supports BPF filters, but can only view data by process, without connections or remote address perspective. sniffer combines the advantages of those two projects also adhering a new Plot mode.
Connections and Process Matching
On Linux, sniffer refers to the ways in which the ss tool used, obtaining the connections of the ESTABLISHED
state by netlink socket. Since that approach is more efficient than reading the /proc/net/*
files directly. But both need to aggregate and calculate the network traffic of the process by matching the inode
information under /proc/${pid}/fd
.
On macOS, the lsof command is invoked, which relies on capturing the command output for analyzing process connections information. And sniffer manipulates the API provided by gopsutil directly on Windows.
sniffer relies on the libpcap
library to capture user-level packets hence you need to have it installed first.
Debian/Ubuntu
$ sudo apt-get install libpcap-dev
CentOS/Fedora
$ sudo yum install libpcap libpcap-devel
MacOS
$ brew install libpcap
Windows
Windows need to have npcap installed for capturing packets.
After that, install sniffer by go get
command.
$ go get -u github.com/chenjiandongx/sniffer
❯ sniffer -h
# A modern alternative network traffic sniffer.
Usage:
sniffer [flags]
Examples:
# processes mode for pid 1024,2048 in MB unit
$ sniffer -p 1024 -p 2048 -m 2 -u MB
# only capture the TCP protocol packets with lo,eth prefixed devices
$ sniffer -b tcp -d lo -d eth
Flags:
-a, --all-devices listen all devices if present
-b, --bpf string specify string pcap filter with the BPF syntax (default "tcp or udp")
-d, --devices-prefix stringArray prefixed devices to monitor (default [en,lo,eth,em,bond])
-h, --help help for sniffer
-i, --interval int interval for refresh rate in seconds (default 1)
-l, --list list all devices name
-m, --mode int view mode of sniffer (0: bytes 1: packets 2: processes)
-n, --no-dns-resolve disable the DNS resolution
-p, --pids int32Slice pids to watch, empty stands for all pids (default [])
-u, --unit string unit of traffic stats, optional: B, Kb, KB, Mb, MB, Gb, GB (default "KB")
-v, --version version for sniffer
Hotkeys
Keys | Description |
---|---|
Space | pause refreshing |
Tab | rearrange tables |
s | switch next view mode |
q / Ctrl C | quit |
iperf is a tool for active measurements of the maximum achievable bandwidth on IP networks. Next we use this tool to forge massive packets on the lo
device.
$ iperf -s -p 5001
$ iperf -c localhost --parallel 40 -i 1 -t 2000
sniffer vs bandwhich vs nethogs
As you can see, CPU overheads bandwhich > sniffer > nethogs
, memory overheads sniffer > nethogs > bandwhich
.
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME COMMAND
128405 root 20 0 210168 5184 3596 S 31.0 0.3 1:21.69 bandwhich
128596 root 20 0 1449872 21912 8512 S 20.7 1.1 0:28.54 sniffer
128415 root 20 0 18936 7464 6900 S 5.7 0.4 0:11.56 nethogs
See what stats they show, sniffer and bandwhich output are very approximate(~ 2.5GB/s). netlogs can only handles packets 1.122GB/s.
# sniffer
Total </ Connections: 80 Up: 2.5GBps Down: 0.0GBps />
# bandwhich
Total Up / Down: 2.5GiBps / 964Bps
# nethogs
TOTAL 1.122 0.000 GB/sec
Bytes Mode: display traffic stats in bytes by the Table widget.
Packets Mode: display traffic stats in packets by the Table widget.
Processes Mode: display traffic stats groups by process using Plot widget.
MIT ©chenjiandongx