arlad dies

[Love]

André Zúquete <avz at sabrina.inesc.pt> writes:

> |Ok, I think I have found the problem. Arlad and xfs get out of sync since
> |xfs seem to be happy updating the since of the file.
> |
> |Note that I haven't tried this patch since I'm out in the
> |bandwidth-backwaters/forest (vistiting my parents). I haven't hade chance
> |to check if it compiles or try run the regression tests.
> 
> It compiles and it runs. Tkx for the help!

But after talking to Assar yestoday we came to the conclusion that the
patch is wrong. It will break other things (like callbacks and the syscall
sequence open("foo", O_CREATE), write(), utimes(), fchmod(), close()).

xfs need to update the filelength.

Now I would like to know more to figure out why this is happening. I have
tried to reproduce it on a same type of computer (RH6.2, same kernel,
makedepends) and I also tried to write a small program that did the
syscalls in the same order as you strace did it. Can you provide debugging
information ?

I'm intressed in the values of size and entry->status.Length when it
crashes, also a backtrace from gdb would be valuable. Xfs and arlad trace
(started with ``fs arladeb all ; fs xfsdebug all'') might be useful.

Love