infinite loop when copying to ext2fs

[Martin Cracauer]

Kris Kennaway wrote on Mon, Jun 16, 2008 at 11:27:53AM +0200: 
> Martin Cracauer wrote:
> >Kris Kennaway wrote on Sat, Mar 01, 2008 at 10:22:26PM +0100: 
> >>Jakub Siroky wrote:
> >>>I've just confirmed the same situation on 6.2-RELEASE amd64/GENERIC. I
> >>>did not noticed it before because I started using ext2fs extensively
> >>>some months ago.
> >>>
> >>>Regards,
> >>>Jakub
> >>>
> >>>On Sat, 19 Jan 2008 16:44:34 +0100
> >>>Kris Kennaway <kris at FreeBSD.org> wrote:
> >>>
> >>>>Kris Kennaway wrote:
> >>>>>Jakub Siroky wrote:
> >>>>>>I have two large ext2fs partitions (368 and 313GB) to hold data
> >>>>>>shared between several OSes. While there were no problems on
> >>>>>>6-STABLE branch I was quite disappointed after upgrade to
> >>>>>>7-STABLE. Whenever I copy/write to ext2fs partition the system
> >>>>>>freezes totally without crashdump. So I set debugging settings to
> >>>>>>kernel config (DEBUG,WITNESS,..) and in console I reproduced error
> >>>>>>situation ending with full screen of unstoppable running text with
> >>>>>>lot of memory addresses and a few recognisable words: 'new block
> >>>>>>bit set for ext already' - again with no crashdump. Then I have
> >>>>>>formatted 1GB partition with ext2fs and the problem on this small
> >>>>>>partition appears only sometimes. 
> >>>>>OK, I am able to reproduce this.
> >>>>>
> >>>>>Kris
> >>>>>
> >>>>Is anyone able to look at this?  I could not spot a candidate change 
> >>>>that has not been merged to 6.x.
> >>>>
> >>>>Kris
> >>>
> >>Sounds like it may have been broken by the change to ext2_bitops.h by 
> >>cracauer.  Can you confirm whether backing out 1.2.2.1 fixes it?
> >
> >I don't think my change can cause a new endless loop.
> >
> >I only reversed the order of tests to ensure we don't overrun a page
> >bounddary (into possibly unmapped space).
> >
> >-       while(*p == ~0U && ofs < sz) {
> >+       while(ofs < sz && *p == ~0U) {
> >
> >It is, however, likely that the code was buggy in the first place.
> >Linux has replaced all this (the allocation code).
> >
> >Also note that the code I fixed is amd64 only.  If the endless loop
> >appears on i386 it's something else.
> >
> >Martin
> 
> It is amd64 only.  I am able to reproduce using the method in the 
> original mails, can you?

Didn't try yet, but I did get a probably unrelated panic on ext2fs
just last week :-) I'll fire it up this week.

How big does the partition have to be to show the problem in this bug?

Martin
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Martin Cracauer <cracauer at cons.org>   http://www.cons.org/cracauer/
FreeBSD - where you want to go, today.      http://www.freebsd.org/