Last year I found a design error in the MD5 version of the hashing method used by Mastercard Internet Gateway Service. The flaw allows modification of transaction amount. They have awarded me with a bounty for reporting it. This year, they have switched to HMAC-SHA256, but this one also has a flaw (and no response from MasterCard).
If you just want to know what the bug is, just skip to the Flaw part.
What is MIGS?
When you pay on a website, the website owner usually just connects their system to an intermediate payment gateway (you will be forwarded to another website). This payment gateway then connects to several payments system available in a country. For credit card payment, many gateways will connect to another gateway (one of them is MIGS) which works with many banks to provide 3DSecure service.
How does it work?
The payment flow is usually like this if you use MIGS:
- You select items from an online store (merchant)
- You enter your credit card number on the website
- The card number, amount, etc is then signed and returned to the browser which will auto POST to intermediate payment gateway
- The intermediate payment gateway will convert the format to the one requested by MIGS, sign it (with MIGS key), and return it to the browser. Again this will auto POST, this time to MIGS server.
- If 3D secure not requested, then go to step 6. If 3D secure is requested, MIGS will redirect the request to the bank that issues the card, the bank will ask for an OTP, and then it will generate HTML that will auto POST data to MIGS
- MIGS will return a signed data to the browser, and will auto POST the data back to the intermediate Gateway
- Intermediate Gateway will check if the data is valid or not based on the signature. If it is not valid, then error page will be generated
- Based on MIGS response, payment gateway will forward the status to the merchant
Notice that instead of communicating directly between servers, communications are done via user’s browser, but everything is signed. In theory, if the signing process and verification process is correct then everything will be fine. Unfortunately, this is not always the case.
Flaw in the MIGS MD5 Hashing
This bug is extremely simple. The hashing method used is:
MD5(Secret + Data)
But it was not vulnerable to hash length extension attack (some checks were done to prevent this). The data is created like this: for every query parameter that starts with vpc_, sort it, then concatenate the values only, without delimiter. For example, if we have this data:
Name: Joe Amount: 10000 Card: 1234567890123456
vpc_Amount=10000 vpc_Card=1234567890123456 vpc_Name=Joe
Get the values, and concatenate it:
Note that if I change the parameters:
vpc_Amount=1 vpc_B=0000 vpc_Card=1234567890123456 vpc_Name=Joe
Get the values, and concatenate it:
The MD5 value is still the same. So basically, when the data is being sent to MIGS, we can just insert additional parameter after the amount to eat the last digits, or to the front to eat the first digits, the amount will be slashed, and you can pay a 2000 USD MacBook with 2 USD.
Intermediate gateways and merchant can work around this bug by always checking that the amount returned by MIGS is indeed the same as the amount requested.
MasterCard rewarded me with 8500 USD for this bug.
Flaw in the HMAC-SHA256 Hashing
The new HMAC-SHA256 has a flaw that can be exploited if we can inject invalid values to intermediate payment gateways. I have tested that at least one payment gateway (Fusion Payments) have this bug. I was rewarded 500 USD from Fusion Payments. It may affect other Payment gateways that connect to MIGS.
In the new version, they have added delimiters (&) between fields, added field names and not just values, and used HMAC-SHA256. For the same data above, the hashed data is:
We can’t shift anything, everything should be fine. But what happens if a value contains & or = or other special characters?
Reading this documentation, it says that:
Note: The values in all name value pairs should NOT be URL encoded for the purpose of hashing.
The “NOT” is my emphasis. It means that if we have these fields:
Amount=100 Card=1234 CVV=555
It will be hashed as:
And if we have this (amount contains the & and =)
It will be hashed as:
The same as before. Still not really a problem at this point.
Of course, I thought that may be the documentation is wrong, may be it should be encoded. But I have checked the behavior of the MIGS server, and the behavior is as documented. May be they don’t want to deal with different encodings (such as + instead of %20).
There doesn’t seem to be any problem with that, any invalid values will be checked by MIGS and will cause an error (for example invalid amount above will be rejected).
To exploit this we need to construct a string which will be a valid request, and also a valid MIGS server response. We don’t need to contact MIGS server at all, we are forcing the client to sign a valid data for themselves.
A basic request looks like this:
and a basic response from the server will look like this:
In the Fusion Payment’s case, the exploit is done by injecting vpc_CardSecurityCode (CVV)
The client/payment gateway will generate the correct hash for this string
Now we can post this data back to the client itself (without ever going to MIGS server), but we change it slightly so that the client will read the correct variables (most client will only check for
- This will be hashed the same as the previous data
- The client will ignore vpc_AccessCode and the value inside it
- The client will process the vpc_TxnResponseCode, etc and assume the transaction is valid
It can be said that this is a MIGS client bug, but the hashing method chosen by MasterCard allows this to happen, had the value been encoded, this bug will not be possible.
Response from MIGS
MasterCard did not respond to this bug in the HMAC-SHA256. When reporting I have CC-ed it to several persons that handled the previous bug. None of the emails bounced. Not even a “we are checking this” email from them. They also have my Facebook in case they need to contact me (this is from the interaction about the MD5 bug).
Some people are sneaky and will try to deny that they have received a bug report, so now when reporting a bug, I put it in a password protected post (that is why you can see several password-protected posts in this blog). So far at least 3 views from MasterCard IP address (3 views that enter the password). They have to type in a password to read the report, so it is impossible for them to accidentally click it without reading it. I have nagged them every week for a reply.
My expectation was that they would try to warn everyone connecting to their system to check and filter for injections.
Flaws In Payment Gateways
As an extra note: even though payment gateways handle money, they are not as secure as people think. During my pentests I found several flaws in the design of the payment protocol on several intermediate gateways. Unfortunately, I can’t go into detail on this one(when I say “pentests”, it means something under NDA).
I also found flaws in the implementation. For example Hash Length Extension Attack, XML signature verification error, etc. One of the simplest bugs that I found is in Fusion Payments. The first bug that I found was: they didn’t even check the signature from MIGS. That means we can just alter the data returned by MIGS and mark the transaction as successful. This just means changing a single character from F (false) to 0 (success).
So basically we can just enter any credit card number, got a failed response from MIGS, change it, and suddenly payment is successful. This is a 20 million USD company, and I got 400 USD for this bug. This is not the first payment gateway that had this flaw, during my pentest I found this exact bug in another payment gateway. Despite the relatively low amount of bounty, Fusion Payments is currently the only payment gateway that I contacted that is very clear in their bug bounty program, and is very quick in responding my emails and fixing their bugs.
Payment gateways are not as secure as you think. With the relatively low bounty (and in several cases that I have reported: 0 USD), I am wondering how many people already exploited bugs in payment gateways.