rhealey@ub.d.umn.edu (Rob Healey) (03/07/89)
Hello again. I've been trying to get NNTP to run properly under WINS TCP/IP 2.1 and SVR3.2. I've run into a VERY bizarre problem. First the background: NNTP is compiled to run under INETD, more exactly tcplisten, this means that tcplisten does all the nasty socket work and NNTP only has to deal with stdin, stdout and stderr. OK, now for the WEIRD part. NNTP connection is made by client wanting to post an article, NNTP forks, child execl's inews and posts article successfully. What the parent does is far from normal: It seems to have lost it's kid! The parent does a wait and immediatly returns with error 10, "No childred to wait on". Well, I put a 1 minute sleep in the child code, did a ps -efl and both nntpd and it's kid showed up with all the numbers, PID & PPID, correct. So, how is it that a parent can lose it's child in a wait when the child hasn't exited yet???? Inquiring minds WANT to know... Is there some problem of kid ID getting lost when a parent or grandparent is disassociated from a tty? On other NNTP news, NNTP 1.5 to be exact, the server seems to feed remote rn's as long as they aren't other 3b2's, the problem with the 3b2's is that the client rrn dies with "socket operation attempted on non-socket". I need to get the posting/xfer side of NNTP working so it'll be a while before I tackle client rn's running on 3b2's. Could a person in the know enlighten me as to just how braindamaged WINS 2.1's socket implementation is? Under 1.1 I got a client bind nameserver library to work, under 2.1, barf-o-rama. Am I wasting my time trying to get all this network code to work??? Frustrated with 2.1, -Rob Healey rhealey@ub.d.umn.edu