Errors on the initial protocol message do not log the IP of the sender anymore. I.e. doing $ echo "GOOD MORNING" >/dev/tcp/somehost.somedomain/22 results in sshd[28192]: Bad protocol version identification 'GOOD MORNING' from UNKNOWN This appears to be due to the fact that sock_in gets closed before get_remote_ipaddr() has a chance to find out who is at the remote end. Apparently, this worked somehow at least in openssh-3.6p1, perhaps the IP caching was different then. Since the process will exit immediately afterwards anyway, maybe there is no need to close these two sockets? They don't get closed on other codepaths with similar functionality (e.g after the "scanned from .. Don't panic."-piece).
Created attachment 1201 [details] minimal patch
Comment on attachment 1201 [details] minimal patch While the diff looks reasonable to I can't see this behaviour with 4.4p1 (that area of code was somewhat restructured). Can you reproduce with 4.4p1?
Appears indeed to have been fixed in 4.4. Sorry for not checking earlier.. Jan
Close resolved bugs after release.