dtrace discuss - Feb 2009 - Basic dtrace question

Hey all,

I''m trying to debug some nfs client/server issues and have made a  
dtrace script that outputs nfs3_bio actions (ie, when the page isn''t  
already cached and a IO has to go back to the NFS server) and by and  
large that part is working. What I''m trying to do is make the  
distinction between a IO that is the result of a NFS read-ahead and  
one that isn''t and reflect that in the output.

I''m probably not groking how the stack works when seeing if  
fbt::nfs3_readahead:entry is involved in the IO''s birth prior to  
nfs3_bio being triggered... any suggestions on how to get this? the  
script below only occasionally shows a 1 if nfs3_readahead was involved.

#!/usr/sbin/dtrace -s

#pragma D option quiet

io::nfs3_bio:start
{
         self->trace = 1;
         self->now = timestamp;
         self->bi = *args[0];
         self->fi = *args[2];
	self->readahead = 0;
}

fbt::nfs3_readahead:entry
/self->trace/
{
         self->readahead = 1;
}

io::nfs3_bio:done
/self->trace/
{

         printf("%Y\t%d\t%s\t%d\t%d\tREADAHEAD=%d\n",
                 walltimestamp,
                 (timestamp - self->now) / 1000,
                 self->fi.fi_pathname,
                 self->fi.fi_offset,
                 self->bi.b_bcount,
                 self->readahead
         );

         stack();
}