FastCGI redirect loop with uWSGI since 6.1

[hrkfdn]

Hey,

I just upgraded my system to OpenBSD 6.1 and have been running into problems with my Flask application deployed via uWSGI and FastCGI sockets.

Whenever I try to request the application I end up with a recursive redirect loop:

--2017-05-06 11:40:04--  http://foo.bar/
Resolving foo.bar (foo.bar)... x.x.x.x
Connecting to foo.bar (foo.bar)|x.x.x.x|:80... connected.
HTTP request sent, awaiting response...
  HTTP/1.1 301 Moved Permanently
  Connection: keep-alive
  Content-Type: text/html; charset=utf-8
  Date: Sat, 06 May 2017 09:40:04 GMT
  Location: http://foo.bar/
  Server: OpenBSD httpd
  Transfer-Encoding: chunked
Location: http://foo.bar/ [following]

As you can see I request the root path and get redirected to it. If I run uWSGI in HTTP mode without httpd in between, this does not happen, so I'm assuming it's not a webapp or uWSGI problem. Additionally, this worked
fine in 6.0.

Has something changed in the way paths are passed to FastCGI changed between 6.0 and 6.1?

[hrkfdn]

Did some research: This happens, because PATH_INFO is "" when an application mounted at the root path of a domain is requested.

In that case, it should probably be set to "/", otherwise Flask/Werkzeug will respond with an HTTP 301 to /.

Because PATH_INFO is still passed as "" after the redirect, although "/" was requested, an endless redirect loop occurs.

I am unsure, if this is an httpd or Werkzeug error.

[throwable-one]

Faced same problem with httpd--fastcgi--uwsgi(flask). Django survives it somehow, but not Flask.
app.url_map.strict_slashes = False fixes issue for Flask.

Since browser explictily sends "/", should not httpd(8) set PATH_INFO="/" for fastcgi?
From the other hand, app is mounted to "/", so slash could be treated as mountpoint, not relative path.

[hrkfdn]

I would opt for the first option as it seems to fix many problems and not cause any. But that's just me. I tried convincing many of the parties involved (Flask/Werkzeug, httpd), but unfortunately interest is limited. For now, I just keep my local installation patched.

[flipchan]

hello, thanks alot! i had the exact same problem, app.url_map.strict_slashes = False solved

[artsi0m]

Thanks in advance for opening this issue and providing solution for Flask

[sysfu]

I'm trying to resolve an infinite redirect loop for a WordPress install on httpd. The web server is running on localhost behind relayd which terminates the TLS connections.

It appears to be the same issue detailed above, however I can't find the equivalent WordPress setting to Flasks' app.url_map.strict_slashes = False

Online people have suggested modifying the WordPress Permalinks settings to eliminate the trailing slash. I tried selecting the first option called 'Plain' which displays links in this format with no trailing slash, but that does not resolve the issue. http://www.reopensd.org/?p=123

The only way that I can log into WordPress admin dashboard for now is by adding the following settings to wp-config.php. I can then login to domain.tld/wp-admin/ however the page rendering is barebones text.

define( 'WP_HOME', 'http://www.reopensd.org' );
define( 'WP_SITEURL', 'http://www.reopensd.org' );

If I remove those my browser gives me the infinite redirect loop

If I remove relayd from the mix and serve WordPress on straight port 80 via httpd, the web site loads and performs as expected.