Subject: Argh - memory leaks in shells on NetBSD 1.0
To: None <current-users@NetBSD.ORG>
From: Sean Doran <email@example.com>
Date: 05/17/1995 02:24:02
I'm having a problem with NetBSD-1.0 on a SPARCStation 2.
I built rc (ftp://viz.tamu.edu/rc/) version 1.4 (and also
1.5) with little difficult, and noticed after some time
that the RSS of running rc processes were enormously large.
After some tracking around I discovered several things:
-- running ps l$pid in succession would show
an increase of 32 or more in the RSS each time
-- running ps l$pid | cat (or any other pipeline)
would increase the RSS by even more
-- it didn't matter what compiled the binary;
the memory leak was not dependent on the compiler,
which is what the rc maintainer types thought
-- a binary built under SunOS 4.1.3_U1 and compiled
statically would not leak under SunOS, but
would leak under NetBSD
-- /bin/sh (statically linked) from SunOS leaks under NetBSD
Things I haven't tested:
statically linked /bin/csh or bash from SunOS
bash as shipped, built on NetBSD (though I suppose
Chet would have noticed any leak like this, yes?)
other variants (mcsh, from der Mouse)
mostly due to frustration and business tempered with a bit of laziness.
The current hypothesis is that there may be a bug in
the sbrk system call or something wrong with the StunOS
compatibility code, however it clearly appears to be a
problem in the interaction between kernel and software,
rather than in the library or in the software itself.
I figure someone may know or recognize something I can't
see for myself.
Any help in squashing this memory leak so that I don't
see things like this:
UID PID PPID CPU PRI NI VSZ RSS WCHAN STAT TT TIME COMMAND
X 1935 1887 1 10 0 212 30988 wait I+ p0 0:06.77 /bin/rc -i
would be *much* appreciated!