[Pvfs2-developers] patch: namei bug fixes

[Phil Carns]

> Hi Phil,
> 
> Good idea to look at the other file systems.  My (admittedly limited)  
> understanding of disconnected dentries is based on the Documentation/ 
> filesystems/Exporting doc, which may be a bit outdated.  It suggests  
> that lookup should return whatever d_splice_alias returns (assuming  the 
> inode exists), which may return NULL when the dentry is  disconnected?
> 
> Also, it seems clear that ENOENT cases require returning NULL.
> 
> -sam

Ok, I think we are converging on something here :)  Your latest namei3 
patch seems to do what you describe, except for the part about returning 
whatever d_splice_alias() returns.  I experimented by modifying line 213 
of namei.c after applying your patch, so that rather than returning 
"NULL", it returns "res".

On 2.6 kernels, this still passes all LTP tests and handles deleting 
renamed directories without complaint.  On 2.4 kernels it causes a 
kernel crash when deleting renamed directories.  I think both of these 
results confirm what we see from other file systems: on 2.6 we should 
pass along the return code of d_splice_alias(), while on 2.4 we should 
always return NULL.

I doubt that there is a feature test for this- kinda hard to tell when 
there is a semantic change to the vfs lookup function.

Do you think that we should go with your patch plus a "#ifdef 
PVFS2_LINUX_KERNEL_2_4" wrapper to control whether it returns NULL or 
res on that one line?

-Phil