Re: [PATCH v2 9/9] netfilter: Replace smp_read_barrier_depends() with lockless_dereference()

[Andres Freund <andres@anarazel.de>]

Hi,

On 2014-11-21 16:57:00 -0500, Pranith Kumar wrote:
> On Fri, Nov 21, 2014 at 11:12 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > On Fri, 2014-11-21 at 10:06 -0500, Pranith Kumar wrote:
> >> Recently lockless_dereference() was added which can be used in place of
> >> hard-coding smp_read_barrier_depends(). The following PATCH makes the change.
> >>
> >> Signed-off-by: Pranith Kumar <bobby.prani@gmail.com>
> >> ---
> >>  net/ipv4/netfilter/arp_tables.c | 3 +--
> >>  net/ipv4/netfilter/ip_tables.c  | 3 +--
> >>  net/ipv6/netfilter/ip6_tables.c | 3 +--
> >>  3 files changed, 3 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/net/ipv4/netfilter/arp_tables.c b/net/ipv4/netfilter/arp_tables.c
> >> index f95b6f9..fc7533d 100644
> >> --- a/net/ipv4/netfilter/arp_tables.c
> >> +++ b/net/ipv4/netfilter/arp_tables.c
> >> @@ -270,12 +270,11 @@ unsigned int arpt_do_table(struct sk_buff *skb,
> >>
> >>       local_bh_disable();
> >>       addend = xt_write_recseq_begin();
> >> -     private = table->private;
> >>       /*
> >>        * Ensure we load private-> members after we've fetched the base
> >>        * pointer.
> >>        */
> >> -     smp_read_barrier_depends();
> >> +     private = lockless_dereference(table->private);
> >>       table_base = private->entries[smp_processor_id()];
> >>
> >
> >
> > Please carefully read the code, before and after your change, then
> > you'll see this change broke the code.
> >
> > Problem is that a bug like that can be really hard to diagnose and fix
> > later, so really you have to be very careful doing these mechanical
> > changes.
> >
> > IMO, current code+comment is better than with this
> > lockless_dereference() which in this particular case obfuscates the
> > code. more than anything.
> >
> > In this case we do have a lock (sort of), so lockless_dereference() is
> > quite misleading.
> >
> 
> Hi Eric,
> 
> Thanks for looking at this patch.
> 
> I've been scratching my head since morning trying to find out what was
> so obviously wrong with this patch. Alas, I don't see what you do.

Afaics the read_barrier_depends protected the load from private->entries[x]
earlier, not the load of table->private itself.

Greetings,

Andres Freund