Posted by Doug Huff on seclists.org:

I have two independent sources claiming known SQLi vulnerabilities in MtGox.

One of said SQLi vulnerabilties was confirmed to be patched on the 16th. The other was not patched, to anyone’s knowledge, at the time of the market crash and database leak. The one that was not patched could have plausibly been used to dump the user table.

The details follow in these chat logs. POC for the referenced xss+csrf is also provided. Whether or not it is still an issue is not known for sure at this time as the site cannot be accessed.

It has also been found that MtGox exposes it’s admin user interface even if a user does not have the admin flag set on their account. As of now it is thought that most actions attempted to be used will throw permission errors. Once again. This cannot be confirmed at this time.

From #bitcoin-hax log:

[21:22:13] <Wasplet2> so, for some background, i’ve been going around to every bitcoin-related site i can find and doing some cursory analysis (looking for sqli, xss, lack of csrf tokens, and authorization issues primarily) … in all cases but mtgox, the bugs have been reported and fixed. never had a chance to report my findings in mtgox

[21:23:07] <Wasplet2> the findings on mtgox were pretty vast: no csrf tokens on transactions, xss all over the place, sqli in a few places. but a couple days after i initially poked at it, the xss and csrf stuff was fixed — apparently someone else had found something, and that made someone at mtgox look over the code and fix those things

[21:23:49] <Wasplet2> so i was left with two issues: a sqli in claims, and a sqli in getActivity … both would make it trivial to, well, get just about any info you wanted.

[21:24:18] <ius> Was getActivity blind too?

[21:24:29] <Wasplet2> no — you could get a full sql query out of it.

[21:24:35] <ius> Convenient.

Which means…

[21:36:28] <jrmithdobbs> so basically, the most likely timeline is this:

[21:37:20] <jrmithdobbs> someone else found these sqli issues, probably weeks ago.

[21:37:32] <jrmithdobbs> they’ve been holding onto this to see what they could do with it.

[21:38:46] <jrmithdobbs> they took their full db dump on the 17th (this has been confirmed by people who have confirmed their passwords that had been changed on the 17th)

[21:39:20] <ius> Yeah, sounds right. I can vouch for >=16th

[21:39:28] <jrmithdobbs> figured the jig was up and concocted a plan to try and get btc out of the exchange

[21:39:52] <jrmithdobbs> they used the sqli to add the huge sum to one of the accounts with a horrible password that was plain md5

[21:39:58] <jrmithdobbs> crashed the market

So…

[21:45:56] <jrmithdobbs> so more plausible is that they used the select injection to get the userdb and found a user with a bad password that was plain md5 (or username == password, etc) that had the most coins?

[21:46:13] <Wasplet2> jrmithdobbs: maybe. the other option is that they simply found a different vulnerability

[21:46:26] <Wasplet2> i mean, really, i found both of those within _maybe_ 15 minutes.

[21:46:36] <PovAddict> given the fact that you could add money to yourself via sqli, I really don’t think someone had that many coins in his account