On Wed, 23 Jan 2008, Ilpo Järvinen wrote: > On Wed, 23 Jan 2008, Dave Young wrote: > > > On Jan 23, 2008 3:41 PM, Ilpo Järvinen wrote: > > > > > > On Tue, 22 Jan 2008, David Miller wrote: > > > > > > > From: "Dave Young" > > > > Date: Wed, 23 Jan 2008 09:44:30 +0800 > > > > > > > > > On Jan 22, 2008 6:47 PM, Ilpo Järvinen wrote: > > > > > > [PATCH] [TCP]: debug S+L > > > > > > > > > > Thanks, If there's new findings I will let you know. > > > > > > > > Thanks for helping with this bug Dave. > > > > > > I noticed btw that there thing might (is likely to) spuriously trigger at > > > WARN_ON(sacked != tp->sacked_out); because those won't be equal when SACK > > > is not enabled. If that does happen too often, I send a fixed patch for > > > it, yet, the fact that I print print tp->rx_opt.sack_ok allows > > > identification of those cases already as it's zero when SACK is not > > > enabled. > > > > > > Just ask if you need the updated debug patch. > > > > Thanks, please send, I would like to get it. > > There you go. I fixed non-SACK case by adding tcp_is_sack checks there and > also added two verifys to tcp_ack to see if there's corruption outside of > TCP. There's some discussion about a problem that is very likely the same as in here (sorry for not remembering to cc you in there due to rapid progress): http://marc.info/?t=120107174200003&r=1&w=2 -- i.