On Fri, Mar 31, 2000 at 02:05:23PM -0600, Steve Lord wrote:
> > hi,
> >
> > first the good news: compiles out of the box...
> > now the bad news: i've got a 1gb partition with xfs and my homedir
> > (something just under 1gb data). so first test is to copy the homedir to
> > the xfs filesystem... works great until the filesystem is about halfway
> > full. then nothing else happens. the cp is still in the process list but
> > sleeping and never woke up until i killed it (i waited roughly 5h)
> > btw. as long as the cp was running kswapd was eating quite a bit of cpu
> > time, but hey... didn't crash so far ;-)
> >
> > the snapshot i used was the tarball from the ftp-server and a following
> > cvs update (done today 2p.m. gmt)
> >
> > anyone any ideas?
> >
> > bernd
> >
>
> How many files are we talking about here? I regularly copy kernel trees
> into XFS. Is it possible that if you multiply the number of files you
> have in your directory by 64K you get close to filling the 1Gbyte?
> There is a bug where preallocation of space does not get pruned down
> until unmount time, and the is another bug (an infinite loop) in the
> out of space case.
I see the same problem (cp hanging) when I copy a tree with all rfcs
(suse/doc2/rfc.rpm on a suse CD) onto a XFS partition.
>
> Alternatively:
>
> Did you compile with CONFIG_PAGE_BUF_META turned on? You could try running
> without it - all the hard hangs I have seen have had this enabled.
>
> Also, do you have 1 cpu or more in your box?
>
> If you have SMP hardware you could turn on the nmi_watchdog (see
> Documentation/nmi_watchdog.txt) and run ksymoops on the output if
> you get any.
nmi_watchdog should normally on by default in 2.3 ATM.
He probably has a UP machine.
> If this is a really hard hang I suspect the kdb debugger will not
> help diagnose the problem.
It just oopsed in the rfc cp with a NULL pointer dereference in sys_open.
Should be reproducable (it is here at least)
-Andi
|