[Rspamd-Users] errno as HTTP Error code

[Steve Sturges (ststurge)]

On Mar 8, 2021, at 3:42 PM, Vsevolod Stakhov <vsevolod at rspamd.com> wrote:
> 
> On 08/03/2021 19:53, Vsevolod Stakhov wrote:
>> On 08/03/2021 19:47, Steve Sturges (ststurge) via Users wrote:
>>> Hi all-
>>> 
>>> Looking at rspamd 2.7 and I was trying to track down a recent issue where I got an unexpected HTTP error code back from rspamd's HTTP API when redis was having issues. Trying to trace this back from the messages in rspamd.log, I have a question about the use of errno in this code:
>>> 
>>> static void
>>> rspamd_http_event_handler (int fd, short what, gpointer ud)
>>> {
>>> ...
>>>        else if (r == 0) {
>>>            /* We can still call http parser */
>>>            http_parser_execute (&priv->parser, &priv->parser_cb, d, r);
>>> 
>>>            if (!conn->finished) {
>>>                err = g_error_new (HTTP_ERROR,
>>>                        errno,
>>>                        "IO read error: unexpected EOF");
>>>                conn->error_handler (conn, err);
>>>                g_error_free (err);
>>>            }
>>>            REF_RELEASE (pbuf);
>>>            rspamd_http_connection_unref (conn);
>>> 
>>>            return;
>>>        }
>>>        else {
>>>            if (!priv->ssl) {
>>>                err = g_error_new (HTTP_ERROR,
>>>                        errno,
>>>                        "IO read error: %s",
>>>                        strerror (errno));
>>>                conn->error_handler (conn, err);
>>>                g_error_free (err);
>>>            }
>>> 
>>>            REF_RELEASE (pbuf);
>>>            rspamd_http_connection_unref (conn);
>>> 
>>>            return;
>>>        }
>>>    }
>>> ...
>>> 
>>> 
>>> In other areas of this function, g_error_new is invoked with a specific HTTP error code — for example, 413, 400, 500 (500 + internal parser errno).  In the case above, where it is using errno directly that then becomes the actual HTTP code in the HTTP response generated in rspamd_http_router_error_handler().  There are other cases as well, farther down in the function above that use ETIMEDOUT directly.
>>> 
>>> Should those errno values should be mapped to HTTP response codes, probably 503 or perhaps 408 or 504?
>>> 
>>> Cheers
>>> -steve
>>> 
>> 
>> Yes, that should be fixed, thanks.
>> 
> 
> I have fixed this and various other inconsistencies I have found in this
> changeset:
> https://github.com/rspamd/rspamd/commit/b1e10ef14ebb91fa2ca1b61494e25fa099e85429

Thank you for the quick turn-around!