This is kinda cool. One way of enumerating usernames is to try a username against a login screen and have the error message tell you “That username doesn’t exist.” Or try to create a new account and have the system tell you “That username already exists.” But if a site is coded properly, it won’t give you that kind of info, making username enumeration (ie. figuring out valid and existing usernames) harder. So how about figuring them out with a timing attack?
When a username and password are submitted to a site for checking, they’re sent to a database and the dbms needs to find the username, and when it finds the row with the username, it checks the password hash against what exists in the database. However if the username doesn’t exist, the dbms doesn’t need to bother checking the password hash. It can just return the generic fail message. This small difference can be seen in the response time. In a recent test, I created a list of 50 usernames and 5 were known good. I interspersed the valid usernames in with all the invalid ones. I used the same password for every attempt, and ran them through Burp Intruder. The result was that the five good returned the slowest response times. There was one invalid password mixed in, but out of the six slowest responses, my five valid usernames were right there. Knowing this, I could do some open source searches for potential usernames and test them against a login screen. I did also test usernames of varying length and it didn’t change the results. Just in case of having a list with mostly valid usernames, I could also pad it with likely garbage usernames, things like “aaaaaaaaa” or “nekdhspfacshabdfks”. This one will be fun to try again in future assessments.