Server hardening. The very words conjure up images of tempering soft steel into an unbreakable blade, or taking soft clay and firing it in a kiln, producing a hardened vessel that will last many years. Indeed, server hardening is very much like that. Putting an unprotected server out on the Internet is like putting chum in the ocean water you are swimming in—it won't be long and you'll have a lot of excited sharks circling you, and the outcome is unlikely to be good. Everyone knows it, but sometimes under the pressure of deadlines, not to mention the inevitable push from the business interests to prioritize those things with more immediate visibility and that add to the bottom line, it can be difficult to keep up with even what threats you need to mitigate, much less the best techniques to use to do so. This is how corners get cut—corners that increase our risk of catastrophe.
This isn't entirely inexcusable. A sysadmin must necessarily be a jack of all trades, and security is only one responsibility that must be considered, and not the one most likely to cause immediate pain. Even in organizations that have dedicated security staff, those parts of the organization dedicated to it often spend their time keeping up with the nitty gritty of the latest exploits and can't know the stack they are protecting as well as those who are knee deep in maintaining it. The more specialized and diversified the separate organizations, the more isolated each group becomes from the big picture. Without the big picture, sensible trade-offs between security and functionality are harder to make. Since a deep and thorough knowledge of the technology stack along with the business it serves is necessary to do a thorough job with security, it sometimes seems nearly hopeless.
A truly comprehensive work on server hardening would be beyond the scope not only of a single article, but a single (very large) book, yet all is not lost. It is true that there can be no "one true hardening procedure" due to the many and varied environments, technologies and purposes to which those technologies are put, but it is also true that you can develop a methodology for governing those technologies and the processes that put the technology to use that can guide you toward a sane setup. You can boil down the essentials to a few principles that you then can apply across the board. In this article, I explore some examples of application.
I also should say that server hardening, in itself, is almost a useless endeavor if you are going to undercut yourself with lazy choices like passwords of "abc123" or lack a holistic approach to security in the environment. Insecure coding practices can mean that the one hole you open is gaping, and users e-mailing passwords can negate all your hard work. The human element is key, and that means fostering security consciousness at all steps of the process. Security that is bolted on instead of baked in will never be as complete or as easy to maintain, but when you don't have executive support for organizational standards, bolting it on may be the best you can do. You can sleep well though knowing that at least the Linux server for which you are responsible is in fact properly if not exhaustively secured.
The single most important principle of server hardening is this: minimize your attack surface. The reason is simple and intuitive: a smaller target is harder to hit. Applying this principle across all facets of the server is essential. This begins with installing only the specific packages and software that are exactly necessary for the business purpose of the server and the minimal set of management and maintenance packages. Everything present must be vetted and trusted and maintained. Every line of code that can be run is another potential exploit on your system, and what is not installed can not be used against you. Every distribution and service of which I am aware has an option for a minimal install, and this is always where you should begin.
The second most important principle is like it: secure that which must be exposed. This likewise spans the environment from physical access to the hardware, to encrypting everything that you can everywhere—at rest on the disk, on the network and everywhere in between. For the physical location of the server, locks, biometrics, access logs—all the tools you can bring to bear to controlling and recording who gains physical access to your server are good things, because physical access, an accessible BIOS and a bootable USB drive are just one combination that can mean that your server might as well have grown legs and walked away with all your data on it. Rogue, hidden wireless SSIDs broadcast from a USB device can exist for some time before being stumbled upon.
For the purposes of this article though, I'm going to make a few assumptions that will shrink the topics to cover a bit. Let's assume you are putting a new Linux-based server on a cloud service like AWS or Rackspace. What do you need to do first? Since this is in someone else's data center, and you already have vetted the physical security practices of the provider (right?), you begin with your distribution of choice and a minimal install—just enough to boot and start SSH so you can access your shiny new server.
Within the parameters of this example scenario, there are levels of concern that differ depending on the purpose of the server, ranging from "this is a toy I'm playing with, and I don't care what happens to it" all the way to "governments will topple and masses of people die if this information is leaked", and although a different level of paranoia and effort needs to be applied in each case, the principles remain the same. Even if you don't care what ultimately happens to the server, you still don't want it joining a botnet and contributing to Internet Mayhem. If you don't care, you are bad and you should feel bad. If you are setting up a server for the latter purpose, you are probably more expert than myself and have no reason to be reading this article, so let's split the difference and assume that should your server be cracked, embarrassment, brand damage and loss of revenue (along with your job) will ensue.
In any of these cases, the very first thing to do is tighten your network access. If the hosting provider provides a mechanism for this, like Amazon's "Zones", use it, but don't stop there. Underneath securing what must be exposed is another principle: layers within layers containing hurdle after hurdle. Increase the effort required to reach the final destination, and you reduce the number that are willing and able to reach it. Zones, or network firewalls, can fail due to bugs, mistakes and who knows what factors that could come into play. Maximizing redundancy and backup systems in the case of failure is a good in itself. All of the most celebrated data thefts have happened when not just some but all of the advice contained in this article was ignored, and if only one hurdle had required some effort to surmount, it is likely that those responsible would have moved on to someone else with lower hanging fruit. Don't be the lower hanging fruit. You don't always have to outrun the bear.
The first principle, that which is not present (installed or running) can not be used against you, requires that you ensure you've both closed down and turned off all unnecessary services and ports in all runlevels and made them inaccessible via your server's firewall, in addition to whatever other firewalling you are doing on the network. This can be done via your distribution's tools or simply by editing filenames in /etc/rcX.d directories. If you aren't sure if you need something, turn it off, reboot, and see what breaks.
But, before doing the above, make sure you have an emergency console back door first! This won't be the last time you need it. When just beginning to tinker with securing a server, it is likely you will lock yourself out more than once. If your provider doesn't provide a console that works when the network is inaccessible, the next best thing is to take an image and roll back if the server goes dark.
I suggest first doing two things: running
ps -ef and making sure
you understand what all running processes are doing, and
lsof -ni |
grep LISTEN to make sure you understand why all the listening ports
are open, and that the process you expect has opened them.
For instance, on one of my servers running WordPress, the results are these:
# ps -ef | grep -v \] | wc -l 39
I won't list out all of my process names, but after pulling out all the kernel processes, I have 39 other processes running, and I know exactly what all of them are and why they are running. Next I examine:
# lsof -ni | grep LISTEN mysqld 1638 mysql 10u IPv4 10579 0t0 TCP 127.0.0.1:mysql (LISTEN) sshd 1952 root 3u IPv4 11571 0t0 TCP *:ssh (LISTEN) sshd 1952 root 4u IPv6 11573 0t0 TCP *:ssh (LISTEN) nginx 2319 root 7u IPv4 12400 0t0 TCP *:http (LISTEN) nginx 2319 root 8u IPv4 12401 0t0 TCP *:https (LISTEN) nginx 2319 root 9u IPv6 12402 0t0 TCP *:http (LISTEN) nginx 2320 www-data 7u IPv4 12400 0t0 TCP *:http (LISTEN) nginx 2320 www-data 8u IPv4 12401 0t0 TCP *:https (LISTEN) nginx 2320 www-data 9u IPv6 12402 0t0 TCP *:http (LISTEN)
This is exactly as I expect, and it's the minimal set of ports necessary for the purpose of the server (to run WordPress).
Now, to make sure only the necessary ports are open, you need to tune your firewall. Most hosting providers, if you use one of their templates, will by default have all rules set to "accept". This is bad. This defies the second principle: whatever must be exposed must be secured. If, by some accident of nature, some software opened a port you did not expect, you need to make sure it will be inaccessible.
-- I was cloud before cloud was cool. Not in the sense of being an amorphous collection of loosely related molecules with indeterminate borders -- or maybe I am. Holla @geek_king, http://twitter.com/geek_king
- Android Browser Security--What You Haven't Been Told
- Epiq Solutions' Sidekiq M.2
- Readers' Choice Awards 2013
- The Many Paths to a Solution
- Nativ Disc
- Download "Linux Management with Red Hat Satellite: Measuring Business Impact and ROI"
- Synopsys' Coverity
- Securing the Programmer
- RPi-Powered pi-topCEED Makes the Case as a Low-Cost Modular Learning Desktop
- Returning Values from Bash Functions