[OpenAFS-devel] Linux 2.6.12 kernel BUG at fs/namei.c:1189

[Rainer Toebbicke]

chas williams - CONTRACTOR wrote:

> 
> the only objection is that this patch seems to know too much about
> the behavior of the linux kernel.  interestingly enough lock_kernel()
> is recursive, and AFS_GLOCK() is not.  i imagine this is what keeps
> other filesystems from deadlocking in the same way.


Well, the way afs_put_inode clicks into standard kernel business is 
pretty much kernel-specific. It was instructive to try and understand 
the difference in Irix, OSF and AIX just to see that things are not 
always obvious (and clean).

A relatively kernel-unspecific way would be to queue an event that is 
handled "later" by somebody running near top level. The question is a) 
how to queue without a lock and b) whether sometimes things must 
happen "now" rather than "later". But that sounds like an added 
complexity with its own probability of going wrong eventually. A lot 
of the AFS client assumes things about the inner working of the kernel 
in that area so I wouldn't be too difficult about that (except for the 
compare for "kswapd" of course which I admit is extremely poor).

BTW: lock_kernel() is not only "recursive", it also gets dropped and 
reacquired when you give up control (e.g. sleep).  Its semantics are 
quite different from the AFS global lock.

-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland
Phone: +41 22 767 8985       Fax: +41 22 767 7155