Re: OSF/1 binaries on linux/axp bug (fs/exec.c)
Jim Gettys (jg@pa.dec.com)
Thu, 22 Oct 1998 13:07:23 -0700
> Sender: owner-linux-kernel@vger.rutgers.edu
> From: cadams@ro.com (Chris Adams)
> Date: 22 Oct 1998 10:28:30 -0500
> To: linux-kernel@vger.rutgers.edu
> Newsgroups: local.list.linux.kernel
> Subject: Re: OSF/1 binaries on linux/axp bug (fs/exec.c)
> -----
> Once upon a time, Bob McElrath <mcelrath@draal.physics.wisc.edu> said:
> >The linux/axp implementation of loading OSF/1 binaries seems to have a bug.
> >The loading works fine, but generates the following message:
> >
> > 513:/sbin/loader: Fatal Error: loader cannot be invoked directly
> >
> >After perusing fs/exec.c I beleve the reason to be that /sbin/loader gets
> >run twice, once for the original application, and a second time
> >*for*itself*. ps reveals:
> >
> >(0)<mcelrath@draal:/home/mcelrath> netscape &
> >[1] 512
> >513:/sbin/loader: Fatal Error: loader cannot be invoked directly
> >(0)<mcelrath@draal:/home/mcelrath> ps aux | grep loader
> >mcelrath 512 0.6 6.5 22920 16688 ? S 11:48 0:01 /sbin/loader
> >mcelrath 513 0.0 0.0 0 0 ? Z 11:48 0:00 (loader <zombie>)
> >
> >Thus, I surmise that /sbin/loader is executed a *second* time, with itself
> >as its argument, and subsequently fails and gives the above message.
> >
> >Can someone more familiar with the data structures involved in loading and
> >executing binaries possibly look at this and eliminate the second execution
> >of /sbin/loader? I can test any patches that anyone comes up with.
>
> The second execute is because netscape re-executes itself to get
> asynchronous DNS lookups. The standard resolver libraries only support
> synchronous DNS lookups, so to allow the user to interrupt them, netscape
> starts a second process it can signal. I suspect the problem lies in the
> fact that the ARGV string shows /sbin/loader instead of netscape, so when
> netscape looks at its ARGV to find its own name and run it, it instead
> tries to run /sbin/loader, which doesn't work.
>
> I never tried to figure out how to fix it - I just turned off async DNS
> lookups (put "export MOZILLA_NO_ASYNC_DNS=True" in your .profile to do
> this).
> --
Not (yet) knowing my way around the insides of Linux, I can't directly
help, but the following (courtesy of a friend at Netscape) may help.
Netscape forks and (potentially forks again) and talks to the forked process
via pipes to do the asynchronous lookup. Strange, but true...
There are comments in the mozilla code about having trouble changing the
name of the subprocess on some platforms that may be related.
The (somewhat strange) code can be found in: mozilla/network/main/unix-dns.c.
see: http://cvs-mirror.mozilla.org/webtools/lxr/source/network/main/unix-dns.c
for exact code.
- JIm
- Jim
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/