From cse.psu.edu!9fans-outgoing-owner Wed Jun 12 14:35:21 1996 Received: from colossus.cse.psu.edu ([130.203.1.2]) by cannon.ecf.toronto.edu with SMTP id <4489>; Wed, 12 Jun 1996 14:35:10 -0400 Received: by colossus.cse.psu.edu id <78361>; Wed, 12 Jun 1996 14:03:30 -0400 Received: from plan9.cs.york.ac.uk ([144.32.33.120]) by colossus.cse.psu.edu with SMTP id <78368>; Wed, 12 Jun 1996 14:02:41 -0400 From: forsyth@plan9.cs.york.ac.uk To: 9fans@cse.psu.edu Date: Wed, 12 Jun 1996 08:53:05 -0400 subject: bug in aux/listen Message-Id: <96Jun12.140241edt.78368@colossus.cse.psu.edu> Sender: owner-9fans@cse.psu.edu Precedence: bulk Reply-To: 9fans@cse.psu.edu Status: RO aux/listen does an rfork sharing memory to create the listener for each service. that also ensures that Subsequent forks by the parent will then propagate the shared data and bss between children. [rfork(2)] aux/listen later forks a process to handle each incoming call. that doesn't specify RFMEM, but is still sharing memory. perhaps that was expected, but in any case, the later becomenone() calls newns, which uses bio to read /lib/namespace, and malloc is called, but the libc.h version doesn't expect to share memory; mayhem ensues. there are safe versions of malloc (for instance in /sys/src/cmd/ndb), or one could rewrite listen in Alef, but for now it seems easier just to be a bit less efficient. listen doesn't seem to rely on sharing memory. forse% diff $home/cd/cmd/aux/listen.c /sys/src/cmd/aux/listen.c 100c100 < switch(rfork(RFFDG|RFPROC|RFMEM)){ --- > switch(rfork(RFFDG|RFPROC)){ /* had RFMEM */