Coda File System

Re: codasrv crash on netbsd/sparc64 3.0

From: Sean Caron <caron.sean_at_gmail.com>
Date: Tue, 25 Apr 2006 15:58:58 -0400
Jan

Awesome! It seems to work now -- codasrv starts up without any faults that I
can
see.

(This is reverting back to the default header file, and just stripping out
that little bit
of code that's got the MMAP dependency on rwcdb_file.c)

I'll try running a NetBSD/sparc[64] or NetBSD/macppc client against my
server
when I get home in a couple of hours and see how that goes. Hopefully it
will be
fairly well-behaved -- you guys will probably hear from me again sometime
soon if
not :)

This was a first install -- /vice was created from scratch, so there was no
existing
prot_users.cdb file. After running vice-setup & starting codasrv, it seems
to be
about 4K, though that may very well be the minimum allocation unit size on
the
disk.

Thanks everyone on this list for your help & support getting coda running
thus far.

Regards, Sean
scaron_at_umich.edu

On 4/25/06, Jan Harkes <jaharkes_at_cs.cmu.edu> wrote:
>
> On Tue, Apr 25, 2006 at 02:59:14PM -0400, Sean Caron wrote:
> > Using 'pdbtool list' also gives segmentation fault...
> >
> > Should I try to apply this patch to rwcdb_file.c in addition to the
> earlier
> > rwcdb_pack.h patch and see what happens? Or do you guys think there is
> > another issue afoot?
>
> Well the segfault made it all clear. The mmap is failing, i.e. returns
> the 32-bit unsigned '-1', but the code is checking for MAP_FAILED, which
> I guess is defined as a 64-bit unsigned representation of '-1'.o
>
> You should be able to revert the change to rwcdb_pack.h, and only use
> the fix for rwcdb_file.c. Since it simply avoids using mmap we won't
> have to figure out how to correctly handle the 32 vs. 64 bit issues
> when it is failing.
>
> In any case, it is kind of curious that mmap fails. Do you have
> /vice/db/prot_users.cdb and what size is it?
>
> Jan
>
Received on 2006-04-25 16:01:21