Interesting - checking some of the other places I tested from that did not fail - one used 2.6.27, the other used 2.6.35. Well, hang on a second - I just went back through my notes from last week and it looks like I also saw one hang with the 2.6.35 site as a destination. I'll do some more testing with this site. Curious - the problem only seems to happen when copying from one particular source. Copying the same material from different sources seemed to work properly. Other than bulk testing, is there a footprint to this problem I can look for to compare a before/after scenario? Thanks - Greg -----Original Message----- From: Jan Engelhardt [mailto:jengelh@xxxxxxxxxx] Sent: Wednesday, April 13, 2011 7:27 AM To: Greg Scott Cc: netfilter@xxxxxxxxxxxxxxx Subject: Re: Very strange ftp mget problem On Wednesday 2011-04-13 14:06, Greg Scott wrote: >I'm posting here because maybe ip_conntrack_ftp might have a problem. Or it >could be hardware... > >An AS/400 behind the firewall starts up a batch job running a script that does >an ftp mget of several dozen or more files from an ftp site. The mget hangs >at random times, sometimes almost immediately, other times after copying Âas >many as a few hundred files. >[...] >Both fw1 and fw2 are HP minitowers, about 4 years old by now. I forget the >exact hardware models. Fw1 is running kernel 2.6.18 and fw2 uses kernel 2.6.25 >or so. Your posting of "2.6.18" and "2.6.25" and "random times" and "almost immediately" (but not _totally immediately_ which is important) hints towards this being old kernels having a problem with all the *ACK features of TCP. Been there, seen it, and it was resolved by 2.6.25's timeframe even here. ÿô.nlj·Ÿ®‰†+%ŠË±é¥Šwÿº{.nlj·§z×–×þ)íè jg¬±¨ ¶‰šŽŠÝjÿ¾«þG«é¸ ¢·¦j:+v‰¨Šwèm¶Ÿÿþø®w¥þŠà£¢·hšâÿ†Ù
