[prev in list] [next in list] [prev in thread] [next in thread] 

List:       ipfilter
Subject:    Re: crash on solaris9/sparc with 4.1.31
From:       Joseph Tam <tam () math ! ubc ! ca>
Date:       2008-12-16 1:30:35
Message-ID: 18831_1229391482_49470679_18831_1793_1_Pine.GSO.4.64.0812151717480.2685 () mnc ! zngu ! hop ! pn
[Download RAW message or body]


Thanks, Darren.  Helpful as always.

> > gcc -fno-hosted
> >
> >     Shouldn't the default gcc flags have "-fno-hosted" for both
> >     pfil and ipf to avoid these error message when loading the
> >     kernel modules:
> >
> >         unix: WARNING: mod_load: cannot load module 'pfil'
> >         unix: /kernel/drv/sparcv9/pfil: undefined symbol
> >         unix:  'memset'
> 
> Actually, I think the problem is the missing -ffreestanding.

The man pages for my gcc says that "-ffreestanding" is synonymous
with "-fno-hosted" but I like your version better since it's more
descriptive.

> > Missing /usr/xpg4/bin/grep
> 
> So in that case the actual statement needs to look like this:
> MIOCPULLUP:sh=/usr/ccs/bin/nm -p /dev/ksyms|grep miocpullup >/dev/null
> 2>&1; if ...
> 
> Does that work for you?

If it works for you, it works for me.  You can even specify "/usr/bin/grep"
since SUNWcsu is not what you would call "optional".

What I essentially did was find all instances of /usr/xpg4/bin/grep
(and it always used with the -q flag) and replaced it with the "grep
>/dev/null" construct.  I took this cue from the GNU grep man pages

      -s, --no-messages
           Suppress error messages about nonexistent or unreadable
           files.   Portability note: unlike GNU grep, traditional
           grep did not conform to  POSIX.2,  because  traditional
           grep  lacked a -q option and its -s option behaved like
           GNU grep's -q option.  Shell  scripts  intended  to  be
           portable  to  traditional grep should avoid both -q and
           -s and should redirect output to /dev/null instead.

Don't forget about

 	pfil/SunOS/postinstall

> > arc4random.c
> > ...
> >    It looks like some bug in the system header rather than ipfilter
> >     (or perhaps I'm using the wrong OS specific gcc).  Whatever the
> >     reason, since it doesn't seem to be required, can't this include
> >     statemen just be removed?
> 
> I've got a funny feeling that it is needed, somewhere else...
> It feels like I should upload a 4.1.32-RC0.

I sympathize with your attempt at trying to maintain a working code
base across many architectures, OS's, versions, and compilers.

> > Kernel crash?
> >
> >     When should IPF be recompiled?  Every time "uname -v" changes, or
> >     perhaps only when the major patch# changes?
> 
> To be safe, every time "uname -v" changes.

What a pain.  But then again, one of the Solaris10 patch bundles I tried
completely rendered the system inoperable.

> To diagnose crashes on Solaris 8 and 9, download this package:
> http://coombs.anu.edu.au/~avalon/CTEact-8.5.17.tar.gz
> http://coombs.anu.edu.au/~avalon/CTEactx-8.5.17.tar.gz
> 
> .. install that and email me the act.# file from /var/crash/foo.

I'll have a look at it and send you the results.  I am re-introducing
firewall rules back into ipf.conf to see if I can narrow the rule that
triggers the crash.

Joseph Tam <tam@math.ubc.ca>
[prev in list] [next in list] [prev in thread] [next in thread]