Re: Compromised postgresql instances

On 06/08/2018 06:13 PM, Andrew Gierth wrote:
>>>>>> "Tom" == Tom Lane <tgl@sss.pgh.pa.us> writes:
>   > Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
>   >> Please cite actual instances of such reports. Vague queries like
>   >> this help nobody.
>
> We do also get them on the IRC channel every once in a while, not
> very frequently but enough to notice (maybe 2-3 so far this year?).
>
>   Tom> Unless there's some evidence that these attacks are getting in
>   Tom> through a heretofore unknown PG security vulnerability, rather
>   Tom> than user misconfiguration (such as weak/no password), I'm not
>   Tom> sure what the security list would have to offer. Right now it
>   Tom> seems like Steve's move to try to gather more evidence is quite
>   Tom> the right thing to do.
>
> Right. All the instances on IRC that I'm personally aware of have
> followed this pattern: either the user has used "host all all 0.0.0.0/0
> trust", or they used "host all all 0.0.0.0/0 md5" where the password for
> the postgres user has been one where it's plausible that a simple
> automated dictionary attack could have guessed it (e.g. "654321" in one
> report).
>
> Excuses for doing this have varied (but I'm pretty sure I've heard "we
> put that in while trying to fix a problem and forgot to take it out"
> more than once - so it's worth a reminder that one should never, ever,
> suggest this on any help forum).
>
> The reports are similar enough and generic enough that it seems pretty
> certain that the scanning and subsequent compromise is all automated;
> some reports have included a (failed) attempt to use local root exploits
> to escalate from the postgres user to root access. Compromised systems
> have been reportedly used as DDoS traffic sources and for cryptocurrency
> mining, but obviously other uses can't be ruled out. I don't know of any
> reports of data being exfiltrated or modified, but that doesn't mean it
> doesn't happen.
>
> I have (private) logs of the channel going back a while, but I haven't
> made any attempt to track how often it happens - the above is basically
> all from memory.
>



Well, I guess every service is going to be vulnerable to 
misconfiguration. The fact that they are automated (oresumably knowing 
how to speak the wire protocol or using a library that does) is a bit 
scary, but not surprising.


My view is that generally DB servers should not be reachable from the 
internet in the first place, certainly not without a client cert, but 
ideally not at all. Then misconfigurations like those Andrew refers to 
above would not be so lethal.

cheers

andrew

-- 
Andrew Dunstan                https://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services