Tracy R Reed wrote:
> John H. Robinson, IV wrote:
> 
> >https: An email goes out to the admin group stating that the https dd
> >not start, and needs to be started manually. That works. good enough. At
> >least for ACS at UCSD.
> 
> That doesn't work for us, didn't work for MP3.com, and doesn't work for 
> hardly any of the ecommerce sites

You asked for Real Life example. I gave you one. It is also Good Enough.
Not my fault that your load balancers can't redirect traffic away from a
downed https host.

> Break-ins are pretty rare for reasonably admin'd Linux boxes.
> Break-ins of webservers with SSL on them even more rare.

I think we both know that https s nothing magical, so saying that https
somehow adds a layer of protection to the *server* is misdirection at
best. SSL protects the data in transit, nothing else.

Now, if your server offered *only* https (and not even ssh) *AND* the
https was set up to check client certificates (nothing that any general
publuc e-commerce site would ever do) and limited communication to only
known clients, then yes. I would accept the argument that https is more
secure (server speaking) than http.

I don't think you can name one site that actually checks client
certificats on their https.


> Let's talk about some of the REAL threats we face:
> 
> 1. Password guessing.

Encryption of data tapes won't to a thing about this.

> 2. Poorly written web apps.

Ibid.

> 3. ssh/httpd/whatever buffer overflow.

Ibid.


Okay, so we see that the real threats have nothing to do with tapes (or
even disks) in transit. So taking the time and effort in your *DISASTER
RECOVERY* tapes to encrypt adds another layer to restore functionality
to the dataserver. As you pointed out, MP3.com lost money evey minute
that a single https serving server was down, compund that by decrypting
the data tapes. Including running around to each to enter the
passphrase.

I am ignoring the time to get the passphrases, as I assume that it is
done in parallel with the tapes themselves. Let's hope the DR plan
mentions where the passphrases are and how to get them.

I would not want to be the one to sell to the big bosses: ``We MUST have
unecrypted secret keys! The time that one system is down, we will lose
big $$! We MUST encrypt the disaster recovery tapes! This will cause our
entire data center to be down a significant period of time, losing us
even more bigger $$$!''


> If it is easy to mitigate the risk due to a swiped SSL key or swapped
> out password I do it 

It is, actually. Especially the latter: encrypted swap.
http://linux.ioerror.us/2006/09/encrypting-your-swap-partition-on-fedora-core/

It will likely mean a reboot. Doing it one sever at a time, or during
your regularly scheduled maintenance window and you gain a good amount
of security with little real cost.

The other way is a bt more difficult. If I were to architecture something
like that, it would include: load balancer + a number of https servers,
each of which is on a UPS. This way they won't all go down on a
powerfalure. If it is long enoug, then, yes, they will all go down. I
would architecture it so the time for power downage would be Suffcient
(however long that might be. Site dependent, of course). When an https
server went down, the load balancer would take it out of rotation.
Either nagios or the server (upon reboot) would issues a notice to the
admin team, begging for attention. If two went down, that notice would
become quite critical. Someone should respond in some way before the
last https went down. I would have the https servers on remote KVMs, so
anadmin could fix the problem from their home. No need to drive in.

Not that hard anymore, is it?


> but leaving the site down for extended periods of time or having to
> build a special install of our OS which implements swap encryption
> just isn't worth it in practice.

What is so special about it? It is Just Plain Smart. I mean, you are
taking significant amounts of $$ for being down, right?

> >In your MySQL application, liberal use of AES_ENCRYPT and EAS_DECRYPT
> >would do wonders for protecting sensitive data.
> 
> I don't think mysql 4.1.20 has AES_ENCRYPT etc. But if we can ever make 
> the jump to mysql5 we will look into it. We currently use a sort of 
> helper program to encrypt the data.

Since the data is already encrypted, no need to encrypt it onto tape,
right? Good. I think we can agree on that one.

> A mysql version migration can be a major undertaking. Not gonna happen
> overnight.

I agree. That is quite the undertaking, as you have to test all of your
applications against it and such. However, you are already using a
helper app to encrypt to disk, so no need to upgrade just for that.

> Encrypted swap doesn't really buy us much as far as security goes as a
> number of other things could.

I agree. It only helps in the case of a stolen or shipped disk.

> And as I mentioned above encrypting the SSL key isn't practical for
> us.

I think you missed the point of the SSL key. The theory was: if you can
have a mechanism for having an encrypted SSL key, then you can use that
mechanism for other applications that need to decrypt their data files
(that are storing to the disk encrypted). This is different from
encrypted partions, as an encrypted parttion that is dumped to tape is
no longer encrypted.

The idea was to encrypt the sensitive data, so t is encrypted on disk
and on tape, so you don't need the overhead of decrypting in case of
Disaster Recovery, and yet stll protect yourself in the case of stolen,
lost, or misdirected disaster recovery tapes.


Is this more difficult than simply encrypting to tape? Yes. It will also
protect you from a lot more stuff, too. Such as the case of missing,
stollen, misdirected, or accdentally released hard drives.


> As a sysadmin/technology person I am really trying to get out of the 
> habit of saying "ALL you have to do is..." and trivializing other 
> peoples technical issues.

You can make reasonable assumptions as to the technical issues. If your
assumptions are wrong, revise as updated information becomes available.
I find that most people are rather myoptic (myself included) and like to
see their own little world. It does take some effort on each persons
part to shake out of that viewpoint.

-john


-- 
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to