On 24 Jan 2022, at 16:45, Yann Ylavic <ylavic....@gmail.com> wrote: >> True - changed to AP_FILTER_ERROR in r1897418, which is the correct code for >> this. > > Would it work if ssl_hook_process_connection() handled only EAGAIN and > otherwise always returned DECLINED (expecting that the usual > processing would re-catch the error and still behave correctly)?
It’s not ideal. This means on error, we would then attempt to read again (this time a blocking read), and then possibly error again (but maybe not). The existing behaviour where we read and then throw away the error result is definitely a bug. Regards, Graham —
