On Fri, Apr 20, 2012 at 1:09 AM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> Note that we do *not* need to bother with fput_light() - even if it does
> fput(), that fput() won't usually be the final one.
Ack. Most of the time the fput_light()->fput will just decrement the use count.
> We also get something else out of that - AFAICS, the kludge in __scm_destroy()
> can be killed after that. We did it to prevent recursion on fput(), right?
> Now that recursion will be gone...
Hmm.. That points out that we may have a *lot* of these pending final
fput's, though. So the deferral queueing should be fairly light. What
were your particular plans for it?
This actually sounds like a fairly good usage-case for Oleg's new
task_work_add() thing. That would defer the final fput, but at the
same time guarantee that it gets done before returning to user space -
in case there are any issues with synchronous actions. Have you looked
at Oleg's series? You weren't cc'd because it didn't affect you, but
look at lkml for "task_work_add()" to find it.
NOTE! If pure kernel threads do fput() deferral (and maybe they do -
I'm thinking nfsd etc), then the task-work thing might need some extra
thought.
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
