In the course of setting up my own OpenStack cloud, I found myself working up and down through multiple layers of network virtualization. Needless to say, I ran into plenty of network issues. To solve them, I researched, gathering small bits of knowledge from many different websites (Stack Overflow, I’m looking at you), forums, and IRC channels. Once my cloud was up and running I lost track of a lot of this information. But, recently, I encountered a new issue and found myself digging through my old bookmarks. That’s why I’ve decided to write up a consolidated record of the knowledge and tools I’ve gathered for inspecting network traffic and answering that age old question: “Why can’t I ping that server?”
A lot of this applies for many Linux and Unix environments, but the specific commands I’m providing for reference are intended for Ubuntu Linux (tested on 12.04 and 16.04).
Let’s start with the basics. The commands many folks have known for ages. Just in case there’s one you’re less familiar with, give these a skim anyway.
ifconfig. I’ve been using this one for years to find out what my IP address is. It’s good for a couple other things as well.
# Display information on all adapters. ifconfig # Display information on a specific adapter. ifconfig eth0 # Bring an adapter up. sudo ifconfig eth0 up # Bring an adapter down. sudo ifconfig eth0 down
ifconfig is considered outdated by many. I don’t doubt that the
ip command which supersedes it is better and more powerful, but I still think it will take a very long time for
ifconfig to die out completely. Regardless, it’s good to be familiar with how it’s used.
# Display address information for all adapters. ip addr show # Display address information for a specific adapter. ip addr show eth0 # Bring an adapter up. sudo ip link set eth0 up # Bring an adapter down. sudo ip link set eth0 down
route command lets you see the system’s routing table. This is how it determines which interface to send traffic out on depending on the destination address.
# Display the system routing table route
My troubleshooting generally has to do with being unable to contact a server, but at times DNS resolution is the issue. In these cases,
nslookup is great for checking whether or not you’re able to resolve DNS queries.
# Look up the IP address of the given name using the default DNS server. nslookup google.com # Look up the IP address of the given name using the specified DNS server. nslookup google.com 18.104.22.168
I generally use
netstat on the server side of the connection when I need to make sure that my server is actually running and listening on the correct port. You can also use this command to show packet statistics on various protocols. This can be a huge help in identifying lower level networking issues.
# Display all listening TCP ports. # -p shows process information for each listener (requires sudo). # -l shows only listening sockets. # -n shows numeric IP addresses and port numbers. # -t shows TCP listeners. You can use -u to see UDP listeners instead. sudo netstat -plnt # Displays statistics for various protocols. netstat -s
watch command has nothing to do with networking, but I often use it to monitor the output of the above commands. What it does is periodically re-run a command and display its output on a fresh screen. There is also a parameter to have
watch highlight any changes in the command’s output.
# Watch a command watch <command> # Specify the refresh interval in seconds (default is 2 seconds) watch -n 1 <command> # Highlight changes watch -d <command>
These are the basic commands I use to check on the status and general configuration of a machine that I am troubleshooting. It can also be handy to reference these files:
# Local hostname resolution. /etc/hosts # Configured DNS servers. /etc/resolv.conf
Step one should always be to use these tools to ensure you have everything configured correctly. If you see that your adapter has an APIPA address (one of those weird looking 169.254.X.X addresses) you’ll know pretty quickly why nothing is working – you haven’t gotten a valid IP address from a DHCP server. If you’re sure everything is configured correctly but you’re still having issues. It’s time to move on to the next tool.
The above tools are really for inspecting your configuration, not inspecting network traffic. If you’re pretty sure your network settings are correct, it’s time to watch some of the actual traffic going across the wire. To do that, I use
tcpdump puts the network interface it’s monitoring into promiscuous mode, you’ll always need to use
sudo when executing it. Beyond that, there is one parameter you’ll always want to pass, the
-i parameter, to tell
tcpdump which network interface to watch. You can pass
-i any to watch all interfaces, but generally you’re going to be focusing on a specific one.
The absolute simplest test you can do is to use tcpdump on both your sending and receiving machines to watch ICMP traffic as you have them try to ping each other.
# Watch all ICMP traffic on the specified interface. # -n displays IP addresses instead of resolving host names. # -i specifies the interface to listen on. # -p specifies the protocol to filter for. sudo tcpdump -ni eth0 -p icmp
If you’re debugging a more complex issue than basic connectivity, you’re going to need a different filter. However, even the broadest filter should still exclude the SSH traffic that is going back and forth between your local machine and the machine you’re debugging. The simplest way to do that is to ignore all traffic on port 22.
# Display all traffic except for SSH. sudo tcpdump -ni eth0 not port 22
tcpdump should give you a lot of insight when inspecting network traffic. It might still be necessary to dig deeper, though. Digging deeper for me means inspecting the
iptables firewall and monitoring the progress of individual packets through its complex system of tables and chains. In fact, troubleshooting
iptables is so complicated and annoying that I’m going to write a separate post about it. I will, however, leave you with a short list of issues I’ve seen via
tcpdump that I needed an understanding of
iptables to understand and resolve:
- Traffic being sent out a public interface with an improperly NATed source IP.
- Outbound traffic that never receives a response from the server.
- Inbound traffic that passes the public interface but never makes it to the correct private interface.
- Communication that works initially but suddenly inbound packets stop getting received.
The decisions about which packets to accept, where to route them, and whether or not to NAT the source or destination addresses all happen in
iptables, and when I finish my follow-up post, I will tell you all I’ve learned about it (hint: it’s way more than I ever wanted to know).
What other techniques do you use for inspecting network traffic in your virtual environments? Let me know in the comments, and help this post be the oasis of information that saves many some poor souls from pulling all their hair out!