When I posted my previous post on accounts on Google+ I received a very interesting suggestions that I would like to bring to the attention of more people. Andrew Cooks pointed out that what LastPass (and other password managers) really need, is a way to specify the password policy programmatically, rather than crowdsourcing this data as LastPass is doing right now.
There are already a number of cross-product specifications of fixed-path files used to describe parameters such as
sitemap.xml. While cleaning up my blog’s image serving I also found that there is a
rules.abe file used by the NoScript extension for Firefox. In this optic, adding a new
password-policy.txt file to define some parameters for the password policy of the website.
Things like the minimum and maximum length of the password, which characters are allowed, whether it is case sensitive or not. These are important information that all the password managers need to know, and as I said not all websites make it clear to the user either. I’ll recount two different horror stories, one in the past and one more recent, that show how that is important.
The first story is from probably almost ten years ago or so. I registered with the Italian postal service. I selected a “strong” (not really) password, 11 characters long. It was not really dictionary-based, but it was close enough if you knew my passwords’ pattern. Anyway, I liked the idea of having the long password. I signed up for it, I logged back in, everything worked. Until a few months later, when I decided I wanted to fetch that particular mailbox from GMail — yes, the Italian postal service gives you an email box, no I don’t want to comment further on that.
What happens is that the moment I tried to set up the mail fetching on GMail, it kept failing authentication. And I’m sure I used the right password that I’ve insisted using up to that point! I log in on the website just fine with it, so what gives? A quick check at the password that my browser (I think Firefox at the time) think is the password of that website shows me the truth: the password I’ve been using to log in does not match the one I tried to use from GMail: the last character is not there. Some further inspection of the postal service website shows that the password fields, both in the password change and login (and I assumed at the time the registration page for obvious reasons), set a
maxlength value to 10. So of course, as long as I typed or pasted the password in the field, the way I typed it when I registered, it worked perfectly fine, but when I tried to login out of band (through POP3) it used the password as I intended, and failed.
A similar, more recent story happened with LastMinute. I went to change my password, in my recent spree of updating all my passwords, even for accounts not in use (mostly to make sure that they don’t get leaked and allow people to do crazy stuff to me). My default password generator on LastPass is set to generate 32-characters passwords. But that did not work for LastMinute, or rather, it appeared to. It let me change my password just fine, but when I tried logging back in, well, it did not work. Yes, this is the reason that I try to log back in after generating the password, I’ve seen that happening before. In this case, the problem was to be found in the length of the password.
But just having a proper range for the password length wouldn’t be enough. Other details that would be useful are for instance the allowed symbols; I have found that sometimes I need to either generate a number of passwords to find one that does not have one of the disallowed symbols but still has some, or give up on the symbols altogether and ask LastPass to generate only letters and numbers. Or having a way to tell that the password is case sensitive or not — because if it is not, what I do is disable the generation of one set of letters, so that it randomises them better.
But there is more metadata that could be of use there — things like which domains should the password be used with, for instance. Right now LastPass has a (limited) predefined list of equivalent domains, and hostnames that need to match exactly (so that bugs.gentoo.org and forums.gentoo.org are given different passwords), while it’s up to you to build the rest of the database. Even for the Amazon domains, the list is not comprehensive and I had to add quite a few when logging in the Italian and UK stores.
Of course if you were just to tell that your website uses the same password as, say, google.com, you’re going to have a problem. What you need is a reciprocal indication that both sites think the other is equivalent, basically serving the same identical file. This makes the implementation a bit more complex but it should not be too difficult as those kind of sites tend to at least share
robots.txt (okay, not in the case of Amazon), so distributing one more file should not be that difficult.
I’m not sure if anybody is going to work on implementing this, or writing a proper specification for it, rather than a vague rant on a blog, but hope can’t die, right?
Update 2021-07-22: as it turns out, this idea was implemented, at the end. By Apple, of all companies. Including a repository for users to contribute the configuration that is not present in the actual sites.
I like the idea, but* Some rules are going to be hard to express in parseable form, such as “at least 1 number, 1 lowercase letter, 1 uppercase letter” (it may be a stupid rule, but it’s a common one).* The “this domain uses the same password as on domain foo” looks like a trivial exploit: just setup a website that asks to login using the same login/password as gmail.com and voilà.
You seem not to have read my post in full then, as I noted that exactly the case:> Of course if you were just to tell that your website uses the same password as, say, google.com, you’re going to have a problem. What you need is a reciprocal indication that both sites think the other is equivalent, basically serving the same identical file.Regarding the 1 number, 1 lowercase letter, 1 uppercase letter… that’s not as difficult to express as you think it is. You just need to be able to express valid character classes supported, and then say “you need at least this many character classes”.This would also allow to express whether foreign character classes are supported (say Japanese characters).