> On 05 Mar 11:27, Ananyev, Konstantin wrote:
> >
> >
> > >
> > > On Thu, Mar 05, 2020 at 01:33:49AM +0800, ZY Qiu wrote:
> > > > When compiling with -O0,
> > > > the compiler does not optimize two memory accesses into one.
> > > > Leads to accessing a null pointer when queue post Rx burst callback
> > > > removal while traffic is running.
> > > > See rte_eth_tx_burst function.
> > > >
> > > > Signed-off-by: ZY Qiu <tgw_t...@tencent.com>
> > > > ---
> > > > lib/librte_ethdev/rte_ethdev.h | 6 ++----
> > > > 1 file changed, 2 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/lib/librte_ethdev/rte_ethdev.h
> > > > b/lib/librte_ethdev/rte_ethdev.h
> > > > index d1a593ad1..35eb580ff 100644
> > > > --- a/lib/librte_ethdev/rte_ethdev.h
> > > > +++ b/lib/librte_ethdev/rte_ethdev.h
> > > > @@ -4388,10 +4388,8 @@ rte_eth_rx_burst(uint16_t port_id, uint16_t
> > > > queue_id,
> > > > rx_pkts, nb_pkts);
> > > >
> > > > #ifdef RTE_ETHDEV_RXTX_CALLBACKS
> > > > - if (unlikely(dev->post_rx_burst_cbs[queue_id] != NULL)) {
> > > > - struct rte_eth_rxtx_callback *cb =
> > > > - dev->post_rx_burst_cbs[queue_id];
> > > > -
> > > > + struct rte_eth_rxtx_callback *cb =
> > > > dev->post_rx_burst_cbs[queue_id];
> > > > + if (unlikely(cb != NULL)) {
> > > > do {
> > > > nb_rx = cb->fn.rx(port_id, queue_id, rx_pkts,
> > > > nb_rx,
> > > > nb_pkts, cb->param);
> > > > --
> > > > 2.17.1
> > > While I don't have an issue with this fix, can you explain as to why this
> > > is a
> > > problem that needs to be fixed? Normally TOCTOU issues are flagged and
> > > fixed for external resources e.g. files, that can be modified between
> > > check
> > > and use, but this is just referencing internal data in the program itself,
> > > so I'm wondering what the risk is? From a security viewpoint if an
> > > attacker
> > > can modify the function pointers in our code, is it not already "game
> > > over"
> > > for keeping the running program safe?
> > >
> >
> > Right now RX/TX cb functions are not protected by any sync mechanism.
> > So while dataplane thread can do RX/TX control threads supposed to
> > be able to add/remove callbacks.
> > I am agree with Stephen here, we probably need either (volatile *)
> > or compiler_barrier() here.
> >
> >
> For my opinion,
> the key question here is if the abstract layer code has to be thread safe
> or application
> developer look after thread safe of key data structure ?
>
> 1. Single thread case :
> Current code has no issue even compiler behavior is different with
> -O0 or O3.
> -O3 merge 2 loads into 1, -O0 still use 2 loads.
>
> 2. Multiple thread case:
> As Konstantin said, there is no sync primitive to protect cb
> pointer at all.
> Because of X86 64bit memory access is atomic, then, -O3 and -O0
> will lead to totally different result.
> I don’t think that's a fix because a Fix cannot depend on
> specific Arch is strong memory order or weak memory order.
>
> Volatile or memory barrier may not fix this with a general style for
> multi-threads.
Can you elaborate why?
From my perspective compiler_barrier seems enough here.
>
> I will suggest add comment to clarify the scenario and let developer make
> decision.
>
> Regards
