Closed
Description
I am running the Rails 5.0.0 beta1 and after deploying to production have noticed something odd. My application is running being an Nginx reverse proxy which decrypts SSL.
Every time I try to submit a form, a ActionController::InvalidAuthenticityToken is raised. I have managed to narrow this down to line 399 in ActionController. The problem is that the request origin has https://
as the protocol and the request base_url has http://
as the protocol.
I have tried setting the X-Forwarded-Proto
header and using config.force_ssl = true
but neither made any difference.
I have not yet figured out a way to isolate a failing test case but am happy to look into it if anyone has any suggestions.
Activity
tpbowden commentedon Jan 8, 2016
Fix by adding more headers in Nginx (X-Forwarded-Ssl on, X-Forwarded-Port 443 and X-Forwarded-Host hostname)
afair commentedon Jan 19, 2016
I also ran into this issue. I'll describe it differently to help with future searches.
Submission of any form to Nginx proxying to a Puma unix socket running a Rails 5.0.0.beta1 app responds with:
and throws:
even though the App's CSRF is set up correctly. The app works in development (puma or thin, without Nginx), and worked in Nginx/Puma/Rails-4.2. I don't know if this is specific to using Puma, but seems like the same problem could exist with other servers (unicorn, passenger, etc.)
I don't know if this behavior change is an error or intended.
I can confirm @tpbowden's solution worked for me as well. In my nginx configuration, I setup the app to proxy to puma listening on a unix socket. Here are the relevant parts of the configuration.
tpbowden commentedon Jan 19, 2016
The issue isn't related to Puma and would happen to any web server running behind a reverse proxy which isn't passing the correct headers through. The way I managed to figure out how this issue was being caused was by following the error down the stack trace.
The root of the issue is the code in
request.origin == request.base_url
, which compares theOrigin
header's value against the base_url which is build up by aRack::Response
. There are 3 parts to this value, thescheme
, thehost
and theport
. All three of these are calculated based on a series of headers which not present by default when using a reverse proxy.You can follow the source for
Rack::Request#base_url
here and it will become clear by looking in your web logs which headers are missing.tpbowden commentedon Jan 19, 2016
I've just done a bit more research and it is definitely the intended behaviour. It can be disabled in your app config using
config.action_controller.forgery_protection_origin_check = false
. It is enabled by default however.jonnyparris commentedon Mar 11, 2016
@tpbowden Did you find a solution that doesn't disable forgery protection in the end? I think I'm facing the same problem and really struggling to make 'clean' progress.
tpbowden commentedon Mar 11, 2016
Yes the problem was not passing enough headers through my reverse proxy (nginx) for Rails/Rack to determine whether the request was originally made using https or not (since it was already decrypted by the time it hit Rails)
jonnyparris commentedon Mar 11, 2016
ahh, so how did you find out what was missing...and how did you add the missing headers? Thanks so much for the quick reply!
tpbowden commentedon Mar 11, 2016
If you look at @afair's comment above he was referenced a correct Nginx config, you just need to add the x-forwarded headers related to ssl. I figured out which ones were missing by digging though the source code of Rack's request object to find out how it determine's the protocol of the request.
jonnyparris commentedon Mar 11, 2016
Alas @afair's config doesn't work for me. Moreover, my logs have been annoyingly sparse since adding my Puma config (but that's another story.)
Thanks anyway, I'll think on it.
jonnyparris commentedon Mar 12, 2016
I narrowed my issue down to nginx not playing well with 2 virtual hosts running at the same time. I'm trying to run a staging VPS and a production VPS using jwilder's nginx-proxy.
Everything works fine using the standard generated conf file when only staging OR production is running. However I run into the
ActionController::InvalidAuthenticityToken
errors on form submissions when both are running at once.I haven't solved it yet, but thought I'd leave this here in case anyone else follows a similar trail of blunders.
straight-shoota commentedon Feb 7, 2017
Thanks @tpbowden and @afair I was desperatly trying to make any sense of this oddly failing CSRF protection... then I found this issue 👍
tripitakit commentedon Feb 17, 2017
Many thanks @tpbowden and @afair, you've made my day!
68 remaining items