Perhaps the safer bet is to continue along your original idea of tweaking auth. In general it may not be a good idea to have validators access the record before validation, it could cause DoS problems.
Massimo On Sep 28, 9:47 pm, Dave <[email protected]> wrote: > I struggled with that a little bit as I was following the application > flow particularly in the login method. All the other methods (well, > except the reset request method) are after login and there is already > a user object. That's why it was so easy for me to just put > onvalidate modifications in the change pass and register methods. > > The flow tripped me up a bit on the login side. Since we haven't > tried to select the user record from the database at the time of > validation, we don't know if the user exists at all. The user lookup > happens in the model code a bit farther down... well, actually pretty > much immediately after the form is accepted post-validation. > > However.... If we could pass the entire form object to the validator, > the validator could then fetch the user row from the database and > perform the necessary initialization of salt & algorithm for this > user's stored password. Like you, I'm not sure if that's cleaner or > not though... > > Something I was aiming for when I started looking at this was to be > able to easily change the hash algorithm used without requiring all > users to reset their passwords. I made my edits with this in mind. > When doing the check password routine I extract the algorithm and salt > from the stored password string. That means whatever the "preference" > is for hashing, the stored password hash algorithm is honored. When a > password is encrypted, as in register or change_password, the > preferred hashing preferences are honored. > > That mindset is important to me for two reasons. First, my day-job is > information security. Several times in the recent past I've had to > tell people "don't use ____ anymore because it's weak". The > application I'm working on now has about 1300 users and the passwords > were hashed with sha1. My preference is nothing less than sha256. > This approach will allow users to authenticate with what's in place, > but when they change their password it gets stored with the new hash > algo. > > On Sep 28, 9:46 pm, Massimo Di Pierro <[email protected]> > wrote: > > > > > > > > > Anyway, I have not looked into this with sufficient detail to know > > what is cleaner (or dirtier). > > I we pass an optional record argument to validate we will need to > > change sqlhtml as well. > > > Massimo > > > On Sep 28, 8:11 pm, Dave <[email protected]> wrote: > > > > That may be an even better way to accomplish this task. I actually > > > considered using some sort of self-reference first, but didn't find it > > > there. > > > > On Sep 28, 9:00 pm, Massimo Di Pierro <[email protected]> > > > wrote: > > > > > I was thinking about this... how about editing the validate function > > > > in dal.py and allow it to pass the current record to the validators? > > > > We cannot change the APIs but a validator is an object so we can just > > > > pass the current record by attaching it as an attribute to the > > > > validator object. CRYPT could check if the current record exists and > > > > use it to extract the salt. In this case you woud change only the > > > > validate function in dal.py and CRYPT. > > > > > On Sep 28, 6:39 pm, Dave <[email protected]> wrote: > > > > > > I've made this work. I still need a little bit of time to make it > > > > > backward compatible with non-salted passwords though. > > > > > > First things first, I had to disable the CRYPT validator. Although it > > > > > makes handling passwords easier, encrypting them at the validator > > > > > level really limits a lot of account enforcement options. > > > > > > Next I created an extra utility library with two methods: > > > > > crypt(password='', algorithm='sha256', salt='') and > > > > > check_password(plain_password='', encrypted_password=''). The crypt > > > > > function is pretty simple. It returns a hexdigest given a specified > > > > > algorithm, string and salt. If no salt is specified, a 16 character > > > > > salt is randomly generated. > > > > > > Finally I had to do some tweaking in the Auth class. First I changed > > > > > the login method to check the password with my check_password instead > > > > > of a simple string compare. Next I had to add a "onvalidate" method > > > > > to the 'register' and 'change_password' methods to encrypt the > > > > > password. The change_password was actually a little more involved > > > > > because I had to customize the old_password validator to use my > > > > > check_password method. > > > > > > I'm not quite ready to share the diff files because I want to re-work > > > > > it so that it's reverse compatible with auth_user data created with > > > > > the default settings. > > > > > > On Sep 21, 1:05 pm, Dave <[email protected]> wrote: > > > > > > > Well clearly I've sparked plenty of discussion. I am working on > > > > > > this > > > > > > to fit my app need. Once I have a working model that doesn't break > > > > > > other applications that use the default hashing and CRYPT functions > > > > > > I'll post my work. As others have commented, the typical way for > > > > > > storing the password would be {algorithm}$salt$hash. I have no > > > > > > problem with that. The previous developer of the app I am working > > > > > > on > > > > > > just chose to store thesaltin a separate field in the table. It's > > > > > > fairly trivial for me to convert the 1500 or so user password > > > > > > strings. > > > > > > > Stay tuned and I'll post something later this week or next. > > > > > > > On Sep 20, 11:36 pm, Massimo Di Pierro <[email protected]> > > > > > > wrote: > > > > > > > > This will be useful put presents a technical difficulty because > > > > > > > of the > > > > > > > way CRYPT works. CRYPT is the validator that check is a password > > > > > > > is > > > > > > > valid, and it does not know what is stored in db therefore it > > > > > > > does not > > > > > > > know thesalt. Anyway, let me know if you have a suggestion. > > > > > > > > Massimo > > > > > > > > On Sep 20, 9:25 pm, Dave <[email protected]> wrote: > > > > > > > > > I have just started using web2py but already, I'm quite > > > > > > > > impressed. In > > > > > > > > the past couple days I've already rolled out an entire site > > > > > > > > rewrite > > > > > > > > and I'm working on my second project. > > > > > > > > > The project I'm working on right now is currently in PHP. I > > > > > > > > was in > > > > > > > > the process of converting it to a Java / Spring MVC project > > > > > > > > when I > > > > > > > > discovered web2py and decided that'd be a much easier, simpler > > > > > > > > and > > > > > > > > quicker way to roll the app. > > > > > > > > > So, let me get to my point.. The current application utilizes > > > > > > > > the php > > > > > > > > sha1() function with aper-usersaltstored in the database. > > > > > > > > Thesalt > > > > > > > > is randomly generated each time the password is changed. This > > > > > > > > is > > > > > > > > similar to the default configuration on most linux boxes. > > > > > > > > > I need to make some changes to the Auth class to support the > > > > > > > > per- > > > > > > > > record passwordsaltinstead of application-widesalt. Does it > > > > > > > > make > > > > > > > > sense for me to provide my edits as part of the project, in case > > > > > > > > someone else thinks the functionality is useful? I plan on > > > > > > > > basically > > > > > > > > checking to see if there is a 'salt' field in the user auth > > > > > > > > table, and > > > > > > > > if so, append that to the plain text password before passing it > > > > > > > > to the > > > > > > > > appropriate hashlib function. > > > > > > > > > Thoughts?
