Lighttpd

I'll keep updating this ticket with new data as the issue seems serious, so from what i discovered after some code edits is that the path that returns HTTP-400 for bad request is broken in .49 and .50, while in .45 it's working fine.

The problem manifests that

con->http_status = 400;
buffer_free(fcgi_env);
return HANDLER_FINISHED;

Which can be found in mod_fastcgi.c no longer causes the server to return http-400, the server just goes into infinite loop

In .45 - fcgi_create_env() will return HANDLER_ERROR, on error. In .49/.50 it'll return HANDLER_FINISHED on error, which is causing that loop because when changed - the error is gone. Now the question is why http_cgi_headers(..) is not able to correctly read _GET payload larger than 32kb

Ok, so to fix the bug it's necessary to return HANDLER_ERROR, instead of HANDLER_FINISHED for error(s) in fcgi_create_env.

Problem with sending fcgi-data larger than 32k is that the function fcgi_env_add(...) relies on constant:

#define FCGI_MAX_LENGTH 0xffff

After increasing it the server can't process large messages anyway as fcgi header size is limited to 16bit... not sure why it seems to be limited to 32k instead of 64k, but maybe it might be possible to send multiple headers...

Anyway would be nice to add some warning too when server.max-request-size or server.max-request-field-size exceeds 31k maybe

Thanks for the repro. I'll have more time to dig into it this weekend.

...
Now the question is why http_cgi_headers(..) is not able to correctly read _GET payload larger than 32kb
...
After increasing it the server can't process large messages anyway as fcgi header size is limited to 16bit... not sure why it seems to be limited to 32k instead of 64k, but maybe it might be possible to send multiple headers...

lighttpd provides QUERY_STRING and REQUEST_URI in the CGI environment, so 32k + 32k = 64k.