Slashdot Mirror
Protecting Your Enterprise Network from Vendor App Servers?
anomaly wonders: "I work for a company with a large IT infrastructure. We have lots of applications in our environment. For a number of applications, vendors provide the apps, and provide core support to those app servers. Our vendors are notorious for demanding superuser access to the boxes that support their applications. To protect our enterprise network from attacks allowed in by well-meaning but less-than-perfectly-competent vendors, we have set up a quarantined network for each vendor. This works well when the model is ASP-like and all of the components live on a single box, but fails when the application needs to be connected to one or more enterprise applications (RDBMS, smtp, they want backup, etc) or when it needs to be connected to lots of target systems inside our environment on lots of different ports. How can I restrict a vendor/application server's access to our enterprise network while still providing platforms to make the applications productive for our user community?"
"Frequently vendors can't restrict their applications to run on a limited set of ports. Most of the time they stare blankly when we want their application to run as something less than superuser.
Our biggest challenge is keeping track of all of the dependencies and managing what ports need to be allowed to which destinations. Of course, when security is tight our business-types say 'you're breaking my application.'
What can you suggest about how to provide access to applications, patch/protect the OS on the app server, and protect the enterprise network? What does your organization do?"
Our biggest challenge is keeping track of all of the dependencies and managing what ports need to be allowed to which destinations. Of course, when security is tight our business-types say 'you're breaking my application.'
What can you suggest about how to provide access to applications, patch/protect the OS on the app server, and protect the enterprise network? What does your organization do?"
You need a Layer 7 Firewall, a.k.a. an application level firewall. Something like Zorp is a good start, but you probably need something with a bit more intelligence about the applications you are talking about though.
L7 Firewalls usually get a bad rap because they tend to be pretty fussy in setup, something you can't really avoind with this kind of stuff. Also, if I was in your shoes, I would learn to stop worrying and start loving tight-ass SLA's...
People who think they know everything are a great annoyance to those of us who do.
Well... don't get EDS to work on it!
Did he inhale?
Seriously.
I built a network for vendor lab equipment that was it's own netowrk, connected to the main network via a file server with 2 nics and NO routing. Therefore, file could be transfered from the lab network to the file server, and then from there to any of the main networks, of course after they had been scanned for viruses and such, since anti virus software usually would not run on their equipment. My rules were firm and backed by higher-ups - if you couldnt get your equipment to work in the enivroment given, or with only minimal flexibility from me (for example i'd write scripts to move their files automatically to the main, backed up network or something simple like that) - then we will find another vendor to provide us a solution that fits. Period. I never had any problem with a vendor once they heard the terms. They won't make money off us if they can't make the system work in our environment.
Don't Tread on Me
Our vendors are notorious for demanding superuser access to the boxes that support their applications.
There are a few things you can do in this case:
1. Get a different vendor (if the application truly does require su for vendor support).
2. Reqest different support staff from the vendor (do this if the app doesn't require su but the support staff is too lazy).
3. Learn how to support the app in-house
I am very serious on these points. You do not want to give root access to people that should not need it. If they say they need it, they have an awful application.
The place I work has a vendor area fenced off from the rest of the server room, and the vendors only have access to what they need. If they need more privileges, the IT guys watch them like a hawk until they're done.
if the app NEEDS to run, you put it on a DMZ and let the world have at it. If they want internal access....make an effort to secure it and when they say they can't do that, get it in writing -- email will do if you've already got that -- and make sure you've secured everything you can. Not much else you CAN do, unless you're the boss.
FreeBSD for the impatient.
Only give temporary access to allow agreed changes to take place. Superuser isn't required to diagnose problems, only to fix them. This also gives some stability as 3rd parties don't enter and alter things as they please.
The disadvantage is that there is an operational overhead -- but then there should be because if its a pain to change things then less gets changed.
Frequently vendors can't restrict their applications to run on a limited set of ports. Most of the time they stare blankly when we want their application to run as something less than superuser.
Simple answer: Don't let them. This is standard operating procedure in the financial services industry. Do you think that a bank would EVER let a non-employee or non-contractor access any bank system whatsoever? Especially remotely? If these companies want to do business with you, they WILL play by your rules or you'll pick one of their competitors products. In my experience, the only companies that required remote access to their systems were ones that A. Didn't have a fully working product, and B. Had to have the developer log in constantly to patch binaries with the latest bug fixes just to get a semi-working product. These are not the type of companies you want to do business with in the first place.
Speaking as a sysadmin for a smallish financial services company that processes around 1.2 million transactions a month, I would NEVER allow any vendor remote access to our network. It just wouldn't happen. Even if I did want to give them access, there are rules and regulations that strictly forbid my giving them access. If they give you a hard time, make something up about a security audit or something.
Seriously, you are asking for trouble if you let them have access. Who's going to take the blame when one of their developers logs in and wipes out all of your company's data? Chances are they'll blame it on you and you'll be in trouble for their mistake.
"When the president does it, that means it's not illegal." - Richard M. Nixon
What can you suggest?
Find some better vendors?
If a vendor is unsympathetic to your concerns, it's up to you to find an alternative or work around them. As you explain, the second option is not always possible when they require access to a number of services at a fundamental level. The worst cases of this occur when you have one or two vendors to pick from for a given application. My suggestion is then to design the application yourself within your security parameters and functionality requirements -- as many people do not have that capability within their own ASP (otherwise they'd do it already) you might want to use something like Sourceforge and contract a team overseas to do it cheaply, supervising the project from here and optionally open-sourcing it after it's built. Then you've got something designed to your parameters without support or upgrade costs especially if the community digs what you've built.
Try not. Do or do not, there is no try.
-- Dr. Spock, stardate 2822-3.
Can you not just write a contract that holds them completely liable for all damages, losses, downtime, etc caused by that machine? Then give them the option of whether or not they need to be a superuser that badly.
Seriously ... . I'm a sysadmin for a medium sized accountancy company, and you wouldn't believe the number of vendors I've had to beat off. I end up talking them down to lower levels of access, all the while listening to the lusers talk down to me as if they knew more about running it on *my* network than the local sysadmin.
In the long run though, I've figured that these lusers are going to be more trouble than they're worth, and am talking the boss into replacing the NetApps and Junipers (routers) with Gentoo Linux boxen. So in short, don't protect, just reject!
I have a similar configuration between two in house networks I use the firebox X700 in routing mode it allows me to open only the ports needed to be routed between the networks and also allows me to monitor all traffic to keep an eye on the test network side.
Create a dedicated VPN. The misbehaving app server sees only this VPN. Everything it needs to access has a presence thereon, carefully firewalled to allow the relevant ports to open. Everything it doesn't need to access, is not even on the network.
I'm an IT guy who does applications managment for an Air Force network and this is a problem that we have run into as well.
;)*
Exchange Service accounts, NetIQ accounts running as Domain Admins etc.
This is generally not a good idea, but for your average IT admin superuser domain level rights were the only way to have those programs function properly for the longest time. However, one simply solution that I have just recently tested with our NetIQ application managment Server is to create the local service account on say...our SMS server not as a domain admin, but just as a local admin. If I do this I can avoid giving blanket access to the entire domain to that program, and I can control which servers have that particular service account on them.
Obviously its not a total fix, although luckily for us, server 2003 has eliminated a lot of the most common "null session" issues that were such a pain to hunt down and change manually.
The only other thing I could suggest is creating a test-bed network and installing one of the suspect application servers at a time and using a sniffer capture program to see which ports have what sort of protocol activity over them. Document the information, test several application servers at a time, shut down uneeded ports and then implement it on your network. *with approval of course
I do hope that I haven't completely misunderstood your problem.
We force the vendors to enter via VPN. we use the VPN gateway to restrict each vendor account's access to only the IP addresses of the systems they need access to. Further, we occasionally use a packet sniffer to watch certain vendors.
We disable the account by default and require them to contact us and tell us what they are doing (change control) before letting them in.
Works for us.
I work for an ISP, and we do not give anyone outside the system/network administration team root access to anything if at all possible. We have given network vendors (Redback and Ascend/Lucent) remote access to a device a couple of times (we changed the passwords on the particular device before and after), and we have one web application that we had to give the vendor root access for setup (but that was on a dedicated server).
Our antispam server is a dedicated server, but it still was screwed up once by the vendor. We were having a problem (only half our mail was actually being processed by the filters), and management directed that I go ahead and give the vendor support person access to the server (as the user the software runs under). He logged in to look at it, and within minutes the system was broke and mail was queueing up because the antispam software shut down and wouldn't restart. He had seen something unrelated to our problem that he thought didn't look right and decided to "fix" it for us. As soon as I got him to change it back, I told him to log out and removed his access (and they won't ever get it again).
After that, the only other time we considered allowing a vendor access was on a problem case that was over a year old that we had to have fixed ASAP. However, in that case, we were NOT going to allow remote access; the vendor was going to have to send somebody to us and we'd sit him down in front of the console (which would be logging) with us looking over his shoulder.
The only people that know your systems and network are your people (and you'll make enough mistakes). Vendors should not have access to change things at any point; at most they should get a "view" type account (they can look all they want but they can't touch). If something needs changing, they need to tell you and you can make the change (after evaluating it and having a back-out plan). For complex systems, you never end up installing software exactly the way the vendor specified; they are not going to know or understand changes you made for your local configuration (and how such changes affect other services and systems). Even a well-meaning "fix" can cause serious problems, and since it'll be your job to fix them, you better know what was done.
From the vender side.
First: It's amazing how many people haven't got a CLUE. "Don't give it to them or get another vender" isn't an option. When you're running a 24x7x365 system and expect sub 1 hour response time from your venders (our customers do) you're going to have to give some.
All of the boxes we hook up have phone lines into them. If security is an issue on dial-in access, we set it up as follows:
Modem is enabled dial-out only, no dial in access.
If the box dials home, we contact the administrator, identify who we are and which box has troubles and have them enable access to the server/hardware and reset the root password temporarily on the box. This occurs only if it's something we haven't already configured to work with sudo.
We then keep logs of the entire session, then email it to the customer when we're done.
When we're done, we have them reset the root password and disable dial-in on the modem again.
If we require access to a network resource from the machine, we're onsite with the customer shoulder surfing.
We do this at secure military sites, financial institutions, etc. and have yet to have an issue in 20 years.
I disagree with what you say, but I'll defend your right to say it to the death - Voltaire
1) write in to the SLA policy-- a SU agreement with liablity
2) use (as mentioned) a L7 switch (Telena has one, too)
3) give temporary access, and make sure you check for root kits everytime with a script
4) tell your management just how expensive it is to have so many vendor's spoons in the soup and how potentially destabilizing it is to do this
5) use smart token card access coupled to your own CA; Tie the proximity of the card via RFID to a pacemaker attached directly to the aorta. If they lose it, they die. Simple.
6) partition roots across servers. Get an SNMP trap when they logon to keep track of them. Set a script against cron to send an additional alarm when they're on for more than a few minutes or upload more than a few megabytes through specific ports (indicating massive changes rather than remote control screen delta)
7) ask for one of their children for hostage use
---- Teach Peace. It's Cheaper Than War.
With the internet and remote admin tools coming / have come to full potential, vendors keep coming to ask for this more and more.
:: Speed on Highway and Number and Color of Cars traveling in pack/line. Some great laughs later over some good wine.
I have been a software vendor for many years. With software that is correctly build and tested, the need have access to a forgien box is about 0.
That is not say that there were not times I wished I had access. Mainly language translation, English-French with neither tech being able to speak the other and a translator that did not understand techincal terms. Think Baud Rate and Partity Bit
Even when I went on site, I generally never touched a machine. I always had the local operator or manager do the actual typing. This way, they learned the what and how along with me, and how to fix it.
But back to today's world, IF and I MEAN IF the higher levels of company force/black mail you into a corner , setup a connection PC. Vendor connects to there, and that is terminal of them in your system. Log every thing. Turn off the PC when you can not be watching what happening.
The FreeBSD operating system provides the jail(2) system call and the jail(1) command for imprisoning a process and its future decendants. The jail facility is based on the chroot(2) implementation, but prevents well-documented means to escape chroot confinement, offering partitioning of the file system, process, and networking namespaces. The facility removes all super-user privileges that would affect objects not entirely inside the jail.
For more information read:
Then you make your policy strictly exclusionary. And when they say "BUT I NEED THIS!", you say, "Ok, fill out a form 23" or whatever the form is. They'll learn quickly that they aren't going to get many of them approved and they'll start putting them in only when they really need them.
I do security
No just threaten to call IBM global outsourcing. That usually gets the venders to play nicely. But if you Really get IBM inside your company you will reach new levels on non-productivity. Just because they manage the largest global corporations doesn't mean they are good.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
...since VMS has a finer grained privilege system which has no such thing as "superuser" (though heavily priv'd users are possible). In addition there are third party and builtin controls available that allow things like limiting which files/directories/devices even super-priv'd users may get to, may allow certain programs to have privs without the maintenance accounts having them, hiding some storage from programs, etc. etc. Also while a program can be permitted access to kernel space, the privs to do that are distinct from those allowing filesystem access or network access. You can (and should) ask why any extra privs are needed, and can track their usage. Modern VMS runs only on Alpha or IA64, good performing processors but not x86. On the other hand the fact that the underlying instruction sets used are not x86 means that everyman's buffer overflow exploits tend not to be engineered for the machine. That programs typically never run in fully priv'd environments (but are given privs they need for legitimate use only) limits too what mischief buffer/heap overflows and the like may do. /dev/kmem, and expect to spend a lot longer setting all that up than you would in VMS, which ships secure out of the box.
If you need to run more safely in x86, I would suggest having a look at SELinux (from NSA) which gives some better controls than usual. Pay special attention to protections of devices, especially some of the more dangerous ones like
I work at for a large government contractor, where root privilage is truely guarded to only those who need to know as there is data that absolutely need to be protected. Allowing random people from random companies access is not an option. It is actually a criminal offense to let someone else use your account who does not also have an account on the system (it is still a security violation even if they do have an account and is possibly subject to disciplanary action, up to and including criminal punishment depending on what it was being used for).
Basically, the only people who have root are the actual system administrators. And that is even locked down in the sense that we are supposed to log in with our regular account first and "su" to root, or otherwise need to call up security and let them know what system you need to log into directly as root (login and audit logs are checked on a regular basis for any discrepencies).
Basically, like I said, if they need root, they go thru one of us to do something.
We were all warned a long time ago that MS products sucked, remember the Magic 8 Ball said, "Outlook not so good"
There is big marketing battle just behind you!
"Evil thrives when good men do nothing"
As one of those vendors for governments, I have to constantly deal with some moron admins who refuse to give me any access to the machine, yet CONSTANTLY call for support.
I don't generally need superuser access on the machine, since generally the only thing that gets screwed up is the data in the RDBMS (you know, user error).
I had one propellerhead go around in my database deleting tables and columns he felt they didn't need. He told me on the phone "we don't use timestamps here". One of those slam your head on the desk conversations. These are civil servants with lifetime jobs, and maybe they knew all about VAX in 1970, but goddamn if they aren't dense.
They tend to think that RAID is a magical "never need to backup ever" solution. I just love it when they call me up after their second RAID-5 drive failed, and I ask them when they last did a backup - and they go "uhhhh we don't need to backup we gots RAID".
Then I explain how RAID has nothing to do with archival or backups, etc, etc.. And I pull out the backup I made last time they had a major upgrade and tell them they have to reenter every parking ticket for the last 8 months, and they threaten and bitch how it's my fault and I tell them I'm not their admin, and if they really want to go to their bosses and fess up how incompetent they are they can go ahead.
Frankly, I'd love for some more competent clients. Of all of them, I can think of one who has any clue what to do with a computer.
But then sometimes they call with a problem that requires fixing on the machine. I'm not going to sit on the phone talking them through shit, I'm not going to email them scripts or code, etc. More than once I've had to tell them that if they don't give me access it wont get fixed.
If it's a problem for you, give them superuser rights when they need it, when they're done doing maintainance, take it away.
I don't need no instructions to know how to rock!!!!
You do NOT let outside users have access to production systems as root (or as any other user). When they ask, you tell them to stop smoking crack and make them tell you how to fix it. When they ask again, instead of root, give them the middle finger, and them make them tell you how to fix it.
You need to learn to do things on your own. Calling up the vendor and handing them root to fix your problems merely makes you vendor dependant, and in my opinion any SysAdmin who does that should be fired. If I were a manager, why would I continue to employ SysAdmins who only call the vendor every time there is a problem? I would be thinking to myself that not only am I paying this bozo a huge sallary, but I'll be paying for the support contract forever as well, and not just for one system, but for however many app servers you've got. Why not just hire someone for minimum wage to dial the vendor when there's a problem than pay someone who is suposedly highly trained but can do nothing for themselves?
I'm not a manager, but I am the lead SysAdmin for an Internet services company. We have about 40 servers and about half as many networking devices (managed switches, firewalls, load balancers, etc). Whenever we get a new type of device we do end up calling the vendor quite a bit, but we always make sure they teach us the solution and we impliment it. Over six months or so as I learn all the ins and outs and bugs of the device we no longer need to contact the vendor. I have a team of three people, and if one of them gave a vendor the root or enable password on any of our devices I'd have them fired for network security breach, and if they weren't taking proper steps to learn a new device on their own I'd have them fired for incompetence and replaced. There are too many smart people out there to keep employed dumb ones, especially for the price tag SysAdmins deserve.
Let me qualify explain that statement. I support a small web-based application. It'll win on Unix or Windows (under Apache & PHP) I don't care about that stuff. But what does piss me off is when we have to port the app to every DB MS under the sun. Our support costs go through the roof just because the customer wants it their app DB.
Well, try explaining to IT (not a smart bunch, after all they dropped out of CS (or never even tried)) that their DB du jour is not up to the task of our RDBMS. True 90% SQL compliance is a good start, but there it is impossible to create a full-featured app and still be completely agnostic.
Examples: MySQL is the worst. The last insert ID is a horrible concept that is not portible. Try finding it in Informix, it is impossible to find, but exists. Thent here is the MySQL timestamp 'feature' - the first timestanp field (and only the first) when UPDATEd is updated to NOW() regardless of whether or not you include it in the updated feilds.
The only reason why they want the DB changed is because they have lazy IT departments that want to do nothing. Their IT staff does not understand the complexities of SQL DBs to them it is just an DB, and "thay're all compabible through SQL" yeah, right. "What about ODBC?" ODBC is fine for basic record operations, but the moment you try to do anyhing advanced, you're out of luck.
I used MySQL as an example, but I've worked with several other Databases, and there is no Holy Grail of DB abstraction. There's much more to databases than inserting updating, deleting and selecting rows, depsite what IT thinks.
Slashdot's rate-of-post filter: Preventing you from posting too many great ideas at once.
1. Smaller niche vendors usually don't have the hardware or manpower to simulate your system. You went with them because they were half the price of dealing with IBM, remember? :)
2. Often, those applications that require Superuser for the installation are third party applications. Install it yourself. You may have more trouble on the outset, but you'll know what's going on with your machine.
3. Often, vendors have their programmers as installers. The bad news is that you see the problems you do with the installations. The good news is that they'll know exactly why they need root - and they'll tell you what they need done. This might need to be a tag team installation, of course.
4. Remember, you can always invite them up there if you want to pay them for their time. Remote access is requested because it's cheaper. Alternatively, put in the contract that you want installation instructions. It will take more time, but you're always welcome to pay more.
Many of these vendor problems are reduced to cost-cutting measures. If you want to pay more, then vendors will be willing to oblige.
Your vendors can already run arbitrary code on your systems. You must trust them.
The standard practice of firewalls to restrict access, and lower priviledge accounts are all good (and important). But the proper protection should have been negotiated into the service contract in terms of access the vendor requires and liability. Durring that negotiation process the technical authority should have considered the security concerns (and added costs). If the technical authority was inept the best you can do is minimize the risks now, and use this example to raise security concerns for the next contracts.
What if you set up your Linux system with User Mode Linux, or your FreeBSD system with FreeBSD jails, like modern hosting companies provide. This will provide your external customer/vendor with root access, but within a locked in virtual server. If your external vendor wants to maintain their database installation, fine: they have root on their own "machine", on their own IP, and there is very little risk to the larger system with real root.
May not apply to *nix but for Windows servers, I answer requests like this with Remote Control on a Terminal session. If a vendor needs to get "Administator" level access for some reason, I push them through Terminal Services then remote control their session so I or someone else can watch what they do. If the vendor is smart (haven't found one yet) they are aware that someone can watch what they are doing, but they aren't so I usually get to watch them do all kinds of boneheaded things while they take a guess and check approach to fixing a problem. The downside for me is I have to take notes so when the vendor finally disconnects I can undo anythign they did that was outside the scope of their repair.
Which makes me wonder why I can't record an RDP session somehow. This would serve two purposes: a) it would let me replay what a vendor had done later when it is more convenient and b) it would give me some proof if I later have to go blast a vendor for doing something absurdly stupid.
The same question for VNC or really any kind of remote management tool that is likely to be used on the Windows platform. Can any of them be logged and/or replayed?
- JoeShmoe
.
-- I wonder which will go down in history as the bigger failure: the War on Drugs or the War on Filesharing
Dual auditing. My activity is logged by a system I have no control over, and that other admin is logged by my system. It's true there are ways I could cover my tracks but it would be apparent I had hidden something even if they couldn't tell what I hid. At which point my ass would no longer be covered (logging has advantages for both sides in that sense), and I'd be asking if you want fries with that.
That's how my shop does it at least.
All's true that is mistrusted