Shortly after Christmas I opened the ssh port 22 at home to access my network from the Internet, while commuting or traveling. To secure the open port I installed fail2ban on the machine I redirected the port to. As soon as the jail was enabled in the file /etc/fail2ban/jail.conf and fail2ban was restarted it blocks the source IP for 5 minutes after 6 failed login attempts by default. If I remember right, it took only a few minutes and the first IP was blocked. I decided, that 5 minutes is too short and extended the blocking to one day, as well as the number of allowed login attempts, which I reduced to 3.
Since I use ssh-keys, I also decided to deny root login at all and password logins via pam by modifying /etc/ssh/sshd_config:
However, this caused some trouble with fail2ban. The log entries of failed login attempts in /var/log/auth.log now looked differently. And soon there was someone (more likely a bot?) trying to login 1960 times within 1:15 hours from the IP address 188.8.131.52. Therefore, I added the following lines to failregex in /etc/fail2ban/filter.d/sshd.conf
^%(__prefix_line)sConnection closed by \[preauth\]\s*$
^%(__prefix_line)sReceived disconnect from : 11: Bye Bye \[preauth\]\s*$
Now fail2ban blocks the login attempts again.
Before I turned off pam authentication, I searched the Internet for how to log passwords and came across this article on a Symantec site using so called honeypots to attract hackers with a weak system. This inspired me to evaluate at least the data I can extract from my auth.log files. I wrote a perl script, which creates a SQLite database of the failed logins and an R script using ggplot2 to display the results, which is even possible without using fail2ban. To find out the subnet and country of the source IP I found a function for R here, which I used for whois lookups. There are other possibilities to visualize the data, too, for example on the fail2ban webpage itself.
There were 4628 failed logins at total from 1005 IP addresses (457 subnets) with 296 usernames. And here are my results:
Timeseries for the one month my port is open now:
The usernames used. Note that since I deny password logins usernames are only logged when they are before connecting (e.g. ssh admin@<host> for command line ssh). Note the x-axis are scaled logarithmic and only a subset of the total usernames, subnets and countries are shown:
The subnets, from which most attacks came:
And finally the map (with this worldmap) with the number of attacks color coded (again logarithmic):
The strange colorbar is somehow related to cairoDevice. However, fonts are much better with it than with the default png device.
Both scripts I wrote are available on GitHub.
Another related link from Cisco: http://www.cisco.com/web/about/security/intelligence/ssh-security.html.