Jay Doane created COUCHDB-3049:
----------------------------------

             Summary: inet:sockname called on ssl socket
                 Key: COUCHDB-3049
                 URL: https://issues.apache.org/jira/browse/COUCHDB-3049
             Project: CouchDB
          Issue Type: Bug
          Components: HTTP Interface
            Reporter: Jay Doane


I configured couchdb to accept ssl requests on port 6894 and put it behind 
haproxy, but the health checks were failing because of this error:

req_err(1419482684) unknown_error : function_clause    
[<<"prim_inet:sockname/1">>,<<"chttpd:absolute_uri/2 
L541">>,<<"cloudant_plugin_chttpd:update_absolute_uri/1 
L78">>,<<"cloudant_plugin_chttpd:before_request/1 L20">>,<<"lists:foldl/3 
L1261">>,<<"chttpd_plugin:before_request/1 L30">>,<<"chttpd:before_request/1 
L252">>,<<"chttpd:handle_request_int/1 L225">>]

I traced the problem to calling inet:sockname on an ssl socket, which can occur 
in a couple places:
https://github.com/apache/couchdb-chttpd/blob/master/src/chttpd.erl#L549
https://github.com/apache/couchdb-couch/blob/master/src/couch_httpd.erl#L516
However, I didn't do a thorough audit to ensure there are no other places where 
an inet function might be called on an ssl socket.

In any case, a simple workaround is to change the "option httpchk" line in 
haproxy.cfg to something like this:
option httpchk GET /_up HTTP/1.0\r\nHost:\ some.host
which will avoid dropping into that particular case where the crashing code is 
called.

The simplest solution is to create a helper function that distinguishes between 
tcp and ssl sockets:

sockname(Socket) when is_pid(Socket) ->
    inet:sockname(Socket);
sockname(Socket) when is_tuple(Socket) -> % is_tuple since ssl_api.hrl is not 
exposed
    ssl:sockname(Socket).

although that adds some more code duplication unless it's done in one location 
like couch_util.

I'm inclined to just open 2 separate nearly identical PRs since there's already 
an incredible amount of duplicated code between those 2 modules, it seems a 
shame to go against the grain ;)




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to