Re: CVS commit: src/sys/fs/tmpfs

[yamt%mwd.biglobe.ne.jp@localhost (YAMAMOTO Takashi)]

hi,

> On Fri Apr 24 2009 at 18:00:45 +0300, Antti Kantee wrote:
>> On Sat Apr 11 2009 at 20:42:59 +0000, Andrew Doran wrote:
>> > On Sat, Apr 11, 2009 at 12:21:57AM +0000, Perry E. Metzger wrote:
>> > 
>> > > Modified Files:
>> > >  src/sys/fs/tmpfs: tmpfs_vnops.c
>> > > 
>> > > Log Message:
>> > > SAVENAME was not set for rename and delete as required
>> > > 
>> > > Patch from christos, fixes pr 41183
>> > 
>> > Now it leaks pathname buffers.
>> > 
>> > Who reviewed this change and who okayed it?
>> 
>> Hmm, does this work correctly if you find the component via the
>> cache_lookup() path?
> 
> Ok, I dug into this a little.  Short answer: no, but ...
> 
> It seems that cache_lookup() always returns false if MAKEENTRY is not set.
> However, it first does the lookup and removes the entry.  Does anyone
> know why it then returns false and forces a relookup?  Now in the
> case of tmpfs we always get 1 cache lookup and 2 full lookups for each
> remove/rename operation.

because it's what ufs expects. :-)

i introduced cache_lookup_raw for nfs, which doesn't want these behaviours.
i wanted to replace cache_lookup but some people prefered the current one.

YAMAMOTO Takashi