home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!dtix!darwin.sura.net!zaphod.mps.ohio-state.edu!usc!rpi!batcomputer!cornell!seth!rcbc
- From: rcbc@seth.cs.cornell.edu (Robert Cooper)
- Newsgroups: comp.sys.isis
- Subject: Re: problems with ISISV2.1 on IBM RS/6000 workstations.
- Message-ID: <RCBC.92Aug17135532@seth.cs.cornell.edu>
- Date: 17 Aug 92 18:55:32 GMT
- References: <1992Aug14.152930.5359@puma.ATL.GE.COM>
- Sender: rcbc@cs.cornell.edu (Robert Cooper)
- Reply-To: rcbc@cs.cornell.edu
- Organization: Cornell Univ. CS Dept, Ithaca NY 14853
- Lines: 65
- In-Reply-To: gthaker@fergie.atl.ge.com's message of 14 Aug 92 15:29:30 GMT
-
- In article <1992Aug14.152930.5359@puma.ATL.GE.COM> gthaker@fergie.atl.ge.com (Gautam H Thaker) writes:
-
- We use ISISV3.0.x in most situations, but for various reasons we have to
- use ISISV2.1 on an IBM RS/6000 machine on one project. Things are not
- working quite right on this IBM machine. The symptom is that socket files
- created as /tmp/Clxyzq do not seem to get deleted when sockets are closed.
- This requires manual deleting of these files every time. This even happens
- after one uses the "cmd" program. Attempts to use the cmd program a 2nd
- time fail and another file with names such as "/tmp/Cl1234" is seen.
-
- I do not remember seeing this problem on V2.1, and on V3.0 it does not
- occur (even on AIX). I suspect the problem is due to the following known
- bug in V2.1 that causes the protections on the named sockets to be messed up:
-
- =============================================================
- 2) Another bug (NOT ONLY on SGI systems) is related to the
- permissions on a socket created by protos during start-up:
- Line 120 of pr_client.c needs to be changed as shown below:
- (remove the 0x)
-
- 120,c120
- > chmod(ux_addr.sun_path, 0666);
- ---
- < chmod(ux_addr.sun_path, 0x666);
-
-
- 3) As discovered incidental to #2 above, another umask problem:
-
- in xmgr.c:Line 195: chmod(logfile, 666);
- should be changed to: chmod(logfile, 0666);
- ==============================================================
-
- This is an excerpt from the file pub/comp.sys.isis/v21.bugs
- available from ftp@cs.cornell.edu by anonymous FTP. This bug might have
- caused the protections to be such that Isis could not delete the named
- socket at startup. (Note: Isis does not delete the named socket when a
- program exits, but just before creating the named socket next time it
- tries to delete any existing socket of the same name.)
-
- Gautam then writes:
-
- Another anamoly is that in the name "/tmp/Cl<pid>" pid only shows up as
- four digit number even though the actual pid would be a 5 digit number.
- I looked at the code of ISISV2.1 where these files are created, but nothing
- there explains this 5 digit to 4 digit translation.
-
- Anyone experience anything like this on ISISV2.1 and IBM machines? Is there
- a fix? Is this an OS problem?
-
- This problem is triggered by a change to the sockaddr structure
- in a recent release of AIX. The bug is an incorrect calculation
- of the socket address length. Here is the fix to clib/cl_isis.c:
-
- 321,322c321,322
- < if(bind(isis_socket, (struct sockaddr*)&ux_addr,
- < strlen(ux_addr.sun_path) + sizeof(ux_addr.sun_family)) == -1)
- ---
- > if(bind(isis_socket, (struct sockaddr*)&ux_addr,
- > sizeof(struct sockaddr_un)) == -1)
-
- This problem has been fixed in V3.0 of Isis for some time.
-
- -- Robert Cooper
- --
- --
-