On 2017-03-04, Kai Krakow <hurikha...@gmail.com> wrote: > Am Sat, 04 Mar 2017 08:02:11 +0000 schrieb "J. Roeleveld" > <jo...@antarean.org>: > >> >> >Normally, when things are working but idle, the TCP connection to 445 >> >shows an SMB echo request/rseponse transaction once per minute. When >> >it fails, the TCP connection evidently got dropped, and the Windows >> >machine repeatedly shuts down new ones: >> > >> >The failure mode looks like this in wireshark: >> > >> > Gentoo Windows >> > >> > -> SYN -> 445 >> > <- SYN/ACK <- 445 >> > -> ACK -> 445 >> > -> SMB[echo req] -> 445 >> > <- RST <- 445 >> > >> >[that repeats 800 times per second for long periods of time] >> > >> >Then at some point, it starts to work: >> > >> > -> SYN -> 445 >> > <- SYN/ACK <- 445 >> > -> ACK -> 445 >> > -> SMB[proto neg req] -> 445 >> > <- SMB[proto neg rsp] <- 445 >> > -> SMB[ses setup req] -> 445 >> > <- SMB[ses setup rsp] <- 445 >> > ... >> >> >Sometimes the umount times out and "fails" because the "host is >> >down", and when that happens, it seems like it immediately starts to >> >work again. :/ >> >> Are other hosts linux or windows?
Other Linux and Windows clients don't seem to be having this problem. >> Maybe a dodgy switch forgetting the correct path? I don't think so. I can ping the host while the CIFS subsystem says "host is down". If the switch is forgetting the path, who's sending back the SYN/ACK and the RST > Or an MTU problem... Is there a router in the path? Nope. I'm going to try to set up a Wireshark capture in ring-buffer mode and somehow detect the failure and stop the capture... -- Grant
