Jorgen Lundman wrote:
> 
> I finished the crypt-fs hack into vnd a while back, but have found it
> can sometime panic/freeze/die in most spectacular ways, on the i386 arch
> whereas the sparc one just ticks along (so far, may be coincidence - it
> is a lot slower).
> 
> Sticking a collection of large files (20Megs) on it and executing:
> 
> find . -name \*.cliff -exec rm -f {} \;
> for i in `find . -type f -print`
> do
> cp $i $i.cliff
> done
> 
> is quaranteed to kill it, for me atleast.
> 
> I've spent the last three days picking apart my code, and that of vnd,
> with a tootpick (just exaggerating) only to find it appears to be
> related to:
> 
> http://www.NetBSD.org/cgi-bin/query-pr-single.pl?number=10731
> and 12189
> 
> in that if I turn off my code/encryption it still happens.
> 
> I've tried mounting it with "-o sync" and it does take noticably more
> effort for it to die, but running my kill script enough times will get
> it.
> 
> This is in a NetBSD-1.5 release kernel environment. Has any work been
> done to vnd since then, or is it still just as (possibly) broken?
> 
> Thanks,

To reply to myself. 1.5.2 crashes just as badly and there is no change
in -current's vnd.c so I assume that too is broken.

I've ported my crypt changes into ccd, farily easily, and that runs like
a charm. I've run my kill.sh script about 25 times.

If anyone else is seeking something like it, send me a message and I'll
forward you the source and diff.

Lund


-- 
Jorgen "Lord" Lundman <lundman@lundman.net>
Technology Manager, Unix Administrator
Phone: +44 (0)20-86591860  Mobile: +44 (0)79-58642918
Pager: 07958642918@one2one.net
"Rare is the person who can weigh the faults of others 
 without putting his thumb on the scales": Byron J. Langenfeld