Found this one on comp.unix.internals. I thought it was appropriate for this list. --- begin article --- >From comp.unix.internals Tue Aug 2 08:25:28 1994 >From: ruppin@bimacs.BITNET (Adi Ruppin) Newsgroups: comp.unix.internals Subject: SunOs bug Keywords: SunOs, bug, TCP/IP, accept Message-Id: <4859@bimacs.BITNET> Date: 31 Jul 94 19:11:15 GMT Organization: Math & CS, BarIlan U, Ramat-Gan, Israel Lines: 31 Hello all, I have encountered another serious bug in SunOs 4.1.1 thru 4.1.3U_1 that bothers me a great deal: It seems that when you send data through sockets by opening them, sending the buffer and then closing them, if you repeat this process quickly enough, the messages do not get to their destination in the correct order. This bug does not appear to exist on Solaris 2.0, OSF or Ultrix. I tried playing with the LINGER parameter and other parameters but with no luck. According to my examinations, the bug lies on the server side. It seems that if you send messages quickly, SunOs transposes the messages it queues on the connections linked list 'so_q'. I tried to make a patch in the kernel by ubstituting the kernel 'soqinsque()' function with another, and some change did appeared, but still it wasn't completely ok. (the soqinsque() function is called by the sonewconn() function, and it links a new connection to the so_q linked list). My assumption is that Sun continued to develop its kernel independently, without noticing that system 4.4's developers *did* fix this bug, judging from the look of their soninsque() routine and the comment over it that says that their function works "properly"... I'll be glad to hear from anyone who knows how to fix this bug, or who can bring this bug to Sun's attention. After all, it's a very basic bug in their old operating system. ruppin@bimacs.cs.biu.ac.il Please reply by email also. --- end of article --- G.J.W. Hagenaars | postmaster National Capital Freenet xx247@freenet.carleton.ca | Remembering Mike Carty 1968-1994 gert-jan@stack.urc.tue.nl | I'm Dutch, what's _your_ excuse?