11.0 stuck on high network load

[Slawa Olhovchenkov]

On Wed, Oct 12, 2016 at 11:19:48AM +0200, Julien Charbon wrote:

> > if INP_WLOCK is like spinlock -- this is dead lock.
> > if INP_WLOCK is like mutex -- thread1 resheduled.
> 
>  Thanks, I understand you question now.  No an interrupt cannot bypass a
> lock:  Here INP_WLOCK is like mutex -- thread1 resheduled.

Thanks, nice.

> >>> As I remeber race created by call tcp_twstart() at time of end
> >>> tcp_close(), at path sofree()-tcp_usr_detach() and unexpected
> >>> INP_TIMEWAIT state in the tcp_usr_detach(). INP_TIMEWAIT set in tcp_twstart()
> >>
> >>  Exactly, thus the current fix is:  If you already have the INP_DROPPED
> >> flag set you are not allowed to call tcp_twstart(), actually it is a
> >> good candidate for a new INVARIANT.  Let me add that.
> >>
> >>> After check source code I am found invocation of tcp_twstart() in
> >>> sys/netinet/tcp_stacks/fastpath.c, sys/netinet/tcp_input.c,
> >>> sys/dev/cxgb/ulp/tom/cxgb_cpl_io.c, sys/dev/cxgbe/tom/t4_cpl_io.c.
> >>>
> >>> Invocation from sys/netinet/tcp_stacks/fastpath.c and
> >>> sys/netinet/tcp_input.c guarded by INP_WLOCK in tcp_input(), and now
> >>> will be OK.
> >>>
> >>> Invocation from sys/dev/cxgb/ulp/tom/cxgb_cpl_io.c and
> >>> sys/dev/cxgbe/tom/t4_cpl_io.c is not clear to me, I am see independed
> >>> INP_WLOCK. Is this OK?
> >>>
> >>> Can be thread A wants do_peer_close() directed from chelsio IRQ
> >>> handler, bypass tcp_input()?
> >>
> >>  If you look carefully INP_WLOCK is used in cxgb_cpl_io.c and
> >> t4_cpl_io.c before calling tcp_twstart().
> > 
> > Yes, and you remeber: sys/netinet/tcp_subr.c
> > 
> >   1535  struct tcpcb *
> >   1536  tcp_close(struct tcpcb *tp)
> >   1537  {
> > ...
> >   1569                  INP_WUNLOCK(inp);
> >   1570                  ACCEPT_LOCK();
> >   1571                  SOCK_LOCK(so);
> >   1572                  so->so_state &= ~SS_PROTOREF;
> >   1573                  sofree(so);
> >   1574                  return (NULL);
> > 
> > sofree() call tcp_usr_detach() and in tcp_usr_detach() we have
> > unexpected INP_TIMEWAIT.
> 
>  I see, thus just for the context:  The TCP stack in sys/dev/cxgb* is a
> TOE (TCP Offload Engine?) TCP stack for Chelsio NICs, it is a
> separate/side TCP stack that is used only with TCP_OFFLOAD option.
> 
>  This TOE TCP stack actually has its own set of detach()/input()
> functions and seems to check INP_DROPPED flag properly.  I guess @np
> check fixes in socket TCP stack and decides which one can also impact
> the Chelsio TOE TCP stack.  Some bugs are only in socket TCP stack, some
> are only in TOE TCP stack.

I am fear about other direction -- setting INP_TIMEWAIT in Chelsio TOE
TCP stack and impact this to
tcp_timer_2msl()/tcp_close()/sofree()/tcp_usr_detach() path.