2015-06-17 8:45 GMT+03:00 Landry Breuil <lan...@rhaalovely.net>: > On Wed, Jun 17, 2015 at 03:03:21AM +0300, Sergey Ryazanov wrote: >> Hello Stuart, >> >> xl2tpd could not correctly handle fragmented ppp frame while read it >> from pty if one of consequent read operation return an error. >> That's happening because xl2tpd too often cleans the buffer and >> assumes that each read frame operation is performed for new frame, >> since xl2tpd uses single buffer for pty and for socket read operations. >> >> The attached patch moves all pty related buffers into _call_ context >> and corrects calculation of buffer position, what fixes issue with >> handling of fragmented frames. Now we can use MTU > 1000. I tested >> these changes with load (L2TP tunnel without IPSec) and all seems work >> normally. No more "Protocol-Reject" frames. >> >> I attach two patches: >> xl2tpd-1.3.1-fix-ppp-fragmentation.patch - against xl2tpd-1.3.1 >> sources, implements fix >> ports-net-xl2tpd-fix-ppp-fragmentation.patch - integrates patch to >> ports tree and also corrects README: enlarge MTU/MRU values in >> example config and remove note about fragmented frames issues. >> >> Anyway, even with this change and MTU ~ 1400, overall performance is >> nasty: only 2 mbps of upload rate on my 3GHz Intel CPU. > > OOoooh i have to test this... i never managed to get a reliable xl2tp > setup with an openbsd npppd server... > Have you talked about it with upstream?
Do you mean discuss speed issue with xl2tpd developers? -- Sergey