- cross-posted to:
- caffeitalia@feddit.it
- genzedong@lemmygrad.ml
I don't really understand what any of this means, maybe someone can explain it for me, I'm a little nervous to keep browsing on lemmygrad if this can apparently be exploited thru comments and posts or something?
there's disagreement about what's happening several comments down so an explanation would be appreciated
I tested it on hexbear, it does seem to get invalidated on password change. I saw the relevant GitHub commit last night but I didn't read into the exact implementation. They might somehow be adding in a password hash to the mix? Or maintaining a blocklist of invalidated jwts but that would be ugly
But yeah. I kinda get why JWTs are like that now, but Lemmy isn't a massively distributed system and the tokens are valid for a nearly indefinite period lol. And baked into request URLs
It seems like by the time you implement all the shit people are suggesting in the GitHub you've completely defeated any simplicity JWTs once had and would be better off just tracking it all in the db
Even if you did, the JWT would still be valid if your hash changed, it would just mean that any future issued JWTs would be different. It does really sound like however it is implemented, it is being used wrong. That said, I've been doing web-dev for 2 years, and my brain melted when I looked at the lemmy stack, so what do I know?
Is that not the norm? I store JWTs as html only cookies with a certain expiry date. The cookie is then sent in every request to the backend.
One more thing I wanted to ask, maybe you can help me. I am looking at the network inspector to too see what the hexbear requests look like. Where the hell is the post data coming in from? All I can see is requests for images and svgs. Seemingly no requests to any posts api. Pretty sure I have no filters set. Am I beeing a total noob here?
Edit: Ah, it's coming in via websocket. Weird. Ewww.
The websocket thing is going away when we update to the latest version (soon, I think the rebase is already done). you can look at basically any federated lemmy instance if you want to see what the network traffic looks like for the HTTP API, its pretty straightforward iirc, and there is some documentation, though it leaves a lot to be desired lol
What I mean about the cookie is that they are (or were? haven't checked recently) literally encoding it into the URL iirc, like instead of just sending the cookie along in the headers or putting it in the request body somewhere the URL would be /api/v3/endpoint?auth=<your-JWT-here>
And then some error pages would have the URL in the error message, so you had users posting their whole tokens when they asked for support
Not sure if that's fixed or not
Oh dear. That is bad.
indeed.
For the record, it doesn't look like you are doing that any more. JWTs are sent in the cookies.
You must be looking at hexbear
lemmygrad.ml is running the latest upstream (0.18.2) and it is being sent in the URL for each API call
Oh man. That is shocking
https://github.com/LemmyNet/lemmy/issues/3499
I'm still at a bit of a loss as to how a jwt token can be invalidated by a user changing their password. Surely this means making some database query on every request, the sort of thing you are trying to avoid by using jwts in the first place.
Do you know anything about how this is achieved in Lemmy?
Yeah, I think they are hitting the db.
https://github.com/LemmyNet/lemmy/pull/1493
If I'm understanding correctly, they are storing the last password change timestamp in the db: local_user.validator_time and then when they fetch the logged-in user details for a request they compare the timestamp of the token to that validator_time and reject the jwt if it's greater.
I don't think lemmy is using jwt because they really needed the low overhead, most of these requests need to hit the db regardless, they are (IMO) just using it because it was simple to use initially.
This does make me wonder if there are some API requests which don't call check_validator_time() and would still be usable after a pw change