Subject: Re: safety of UFS_DIRHASH
To: None <firstname.lastname@example.org>
From: Stephen M. Rumble <email@example.com>
Date: 02/16/2007 10:42:27
Quoting <Blair Sadewitz <firstname.lastname@example.org>:
> I noticed that there was a bug fix for this recently. Exactly how
> dangerous is it now? Using it seems to significantly increase LFS
> performance (metadata-related?), though I haven't done benchmarks yet.
> Unfortunately, it still doesn't help the "use up all the CPU while
> untarring pkgsrc" problem. :( much.
UFS_DIRHASH was initially marked experimental with two known issues:
1) FFS_EI didn't work
2) LFS didn't work
1) was fixed recently due to a FreeBSD developer's observation.
2) I never saw an explicit fix for, but apparently it works now.
However, Thor found another problem a while back (sorry for the trouble!):
This is the dangerous bug. The other two were obvious breakage
(directory listings didn't work). I don't think anything was done to
fix it, but perhaps the newlock2 code takes care of the appropriate
locking. Can anybody more knowledgeable comment? If newlock2 doesn't
already do it, could we not simply wrap the pool calls with a lock?