Re: [PATCH v4 3/3] RDS: make sure not to loop forever inside rds_send_xmit

[Sowmini Varadhan]

On (04/07/15 17:26), David Miller wrote:
> >  	/*
> > +	 * we record the send generation after doing the xmit acquire.
> > +	 * if someone else manages to jump in and do some work, we'll use
> > +	 * this to avoid a goto restart farther down.
> > +	 *
> > +	 * we don't need a lock because the counter is only incremented
> > +	 * while we have the in_xmit bit held.
> > +	 */
> > +	conn->c_send_gen++;
> > +	send_gen = conn->c_send_gen;
> 
> This increment does need to either be changed to be an atomic_t
> or covered by a lock.
> 
> Otherwise two concurrent callers can both try to increment it at
> the same time, and it only effectively increments once.  That's
> corrupted state and will break all of the new logic added here.
I'm afraid I dont follow what race condiiton you are seeing? Prior
to this line, the "acquire_in_xmit" check would have only allowed
one thread to successfully increment c_send_gen, right? What did I
miss?


> Still very unhappy with this patch series submission, as I still find
> new problems every time I look at these changes.  Are you evaluating
> them and double checking all of the claims in the commit log message
> and comments, and logic in these, or are you just passing them off
> upstream after testing and leaving the checking to people like me?

I am sorry about the aggravation this is causing. But I'm trying to do
the right thing here, and fix someone else's lapse in failing to
properly commit things that were recognized as upstream issues.

Thus all I'm left with is visual inspection, and to run whatever
regression tests we have for these commits. 

--Sowmini

[I had to truncate recipient list, because it does not make it past
rds-devel moderator otherwise]
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/