dtrace discuss - Sep 2005 - io provider and files in a forceddirectio mounted filesystem

The following script is used as a first attempt to discover IO patterns in a 
dbase setup:

#------------------------------------------------------------------
#pragma D option dynvarsize=128m

dtrace:::BEGIN
{
}

pid$target::kaio:entry
{
        self->doit = 1;
}

pid$target::_aiodone:return
{
        self->doit = 0;
}


io:::start
/self->doit || execname == "oracle"/
{
        block[args[2]->fi_pathname, args[0]->b_blkno ] = timestamp;
}

io:::done
/block[args[2]->fi_pathname, args[0]->b_blkno]/
{
        self->action = (args[0]->b_flags & B_READ) ? "R" :
"W";
        @countem[args[1]->dev_pathname, args[2]->fi_pathname,
self->action ] =
count();
        @elapsed[args[1]->dev_pathname, args[2]->fi_pathname,
self->action ] =
sum(timestamp - block[args[2]->fi_pathname, args[0]->b_blkno]);
}

dtrace:::END
{
        printa("count: %64s %40s %1s %@14d\n", @countem);
        normalize(@elapsed, 1000);
        printa("elapsed: %64s %40s %1s %@14d\n", @elapsed);
}
#------------------------------------------------------------------

The Oracle dbase its files all reside in filesystems. The redo is on a directio 
mounted filesystem. 
The output I get when running this script tells me all the filenames (including 
the device names) except for the redolog files!

Remounting the filesystem that holds the redolog files without the directio 
option however ''solves'' this. Without the directio option the
redolog files are
listed by name too.

Why is it that mounting a filesystem with the direcio option disables dtrace its
ability to show the filename?


Any clues, or can this be considered a ''undocumented feature''?

BTW $target refers to the PID of the redologwriter: the only process that is 
writing to these redolog files. All other processes related to Oracle are all 
known in the kernel under the execname "oracle"

Thanks
Paul