Messages in this thread | | | Date | Fri, 9 Jan 2009 23:50:10 +0100 | From | Willy Tarreau <> | Subject | Re: [PATCH] tcp: splice as many packets as possible at once |
| |
On Sat, Jan 10, 2009 at 01:42:58AM +0300, Evgeniy Polyakov wrote: > On Fri, Jan 09, 2009 at 11:17:44PM +0100, Willy Tarreau (w@1wt.eu) wrote: > > However I'm OK for the !timeo before release_sock/lock_sock. I just > > don't know if we can put the rest of the if above or not. I don't > > know what changes we're supposed to collect by doing release_sock/ > > lock_sock before the if(). > > Not to interrupt the discussion, but for the clarification, that > release_sock/lock_sock is used to process the backlog accumulated while > socket was locked. And while dropping additional pair before the final > release is ok, but moving this itself should be thought of twice.
Nice, thanks Evgeniy. So it makes sense to move only the !timeo test above since it's not dependant on the socket, and leave the rest of the test where it currently is. That's what Eric has proposed in his latest patch.
Well, I'm now trying to educate myself on the send part. It's still not very clear to me and I'd like to understand a little bit better why we have this corruption problem and why there is a difference between sending segments from memory and sending them from another socket where they were already waiting.
I think I'll put printks everywhere and see what I can observe. Knowing about the GSO/SG workaround already helps me enable/disable the bug.
Regards, Willy
| |