[Pvfs2-developers] NULL op pointer in dbpf

[Sam Lang]

On Oct 12, 2006, at 12:45 PM, Pete Wyckoff wrote:

> I could use some help tracking this down.  Just updated from the
> head into my local working copy and find that my standard tests
> break.
>
> On the server, in readdir, we end up in dbpf_keyval_iterate.  It
> calls dbpf_op_init_queued_or_immediate, which because we are using
> immediate completion, does not grab a queued op structure, and does
> not modify the passed-in pointer q_op_pp.

Unfortunately, the immediate completion path hasn't been tested  
recently.  Out of curiosity, have you always used it?

>
> Then dbpf_keyval_iterate calls on down into dbpf_queue_or_service
> with q_op_p == NULL.  The first statement of that code does:
>
>     if( coll_p->immediate_completion &&
>            (DBPF_OP_IS_KEYVAL(op_p->type) || DBPF_OP_IS_DSPACE(op_p- 
> >type)))
>
> but the op type is KEYVAL_ITERATE, so the condition fails.  Is this
> the problem?

Yeah I would bet.

>
> Sam, I recall you replaced the ranges of op types with lots
> of explicit equality comparisons for some reason.  Did you leave
> this one out on purpose?  If not, does that indicate we should maybe
> try to bug-proof these comparisons a bit?  Maybe
>
>     #define KEYVAL 1<<6
>     #define DSPACE 1<<5
>     #define BSTREAM 1<<4
>
>     enum op_type {
> 	KEYVAL_READ = KEYVAL | 1
> 	KEYVAL_WRITE = KEYVAL | 2
> 	KEYVAL_ITERATE = KEYVAL | 3
> 	..
>     }
>
>     #define DBPF_OP_IS_KEYVAL(type) ((type) & KEYVAL)
>

I like this idea.  Do you want to fix it?  I can do it if you would  
prefer.

-sam

> Or tell me if I'm barking up the wrong tree.
>
> 		-- Pete
> _______________________________________________
> Pvfs2-developers mailing list
> Pvfs2-developers at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
>