> On 21 September 2016 at 23:02 Florian Westphal <fw@xxxxxxxxx> wrote:
>
>
> Fabian Frederick <fabf@xxxxxxxxx> wrote:
> > Since commit f330a7fdbe16
> > ("netfilter: conntrack: get rid of conntrack timer")
> >
> > closed connections remain longer in /proc/net/nf_conntrack
> >
> > Running current kernel; just after boot:
> > cat /proc/net/nf_conntrack | wc -l = 5
> > 4 minutes required to clean up the table.
>
> We should reap the stale entries while iterating, just like
> we do for ctnetlink interface.
>
> Can you try this patch?
>
> diff --git a/net/netfilter/nf_conntrack_standalone.c
> b/net/netfilter/nf_conntrack_standalone.c
> --- a/net/netfilter/nf_conntrack_standalone.c
> +++ b/net/netfilter/nf_conntrack_standalone.c
> @@ -212,6 +212,11 @@ static int ct_seq_show(struct seq_file *s, void *v)
> if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
> return 0;
>
> + if (nf_ct_should_gc(ct)) {
> + nf_ct_kill(ct);
> + goto release;
> + }
> +
> /* we only want to print DIR_ORIGINAL */
> if (NF_CT_DIRECTION(hash))
> goto release;
>
Hello Florian,
First problem is solved: table gets cleared 3 minutes earlier
but I still have kmemleak before running the following:
echo scan > /sys/kernel/debug/kmemleak
cat /sys/kernel/debug/kmemleak
Nothing
echo scan > /sys/kernel/debug/kmemleak
cat /sys/kernel/debug/kmemleak
-> rsyslogd
I talked about false positive because everything is cleared later.
Note that problem appears only in a VM which is really slow due to
ksoftirqd eating lot of CPU for an unknown reason. Maybe you should test
somewhere else before applying.
Regards,
Fabian
> > Going back to kernel version before commit above there are
> > no connections after some seconds.
> >
> > Referring to the commit changelog this was an expected behaviour but
> > it results in temporary kmemleak reports:
>
> I don't see kmemleak complaints on my test vm, I'm reluctant to
> turn it off.
>
> Can you explain why we see such 'false positive'?
>
> The conntracks should still be referenced, as they
> are in main table.
>
> > unreferenced object 0xffff88003b0e6600 (size 248):
> > comm "rsyslogd", pid 1595, jiffies 4294741312 (age 7.343s)
> > ...
> > backtrace:
> > [] kmemleak_alloc+0x23/0x40
> > [] kmem_cache_alloc+0xd9/0x180
> > [] __nf_conntrack_alloc.isra.50+0x48/0x170
> > [] nf_conntrack_in+0x3a2/0x5f0
> > [] ipv4_conntrack_local+0x40/0x50
> > [] nf_iterate+0x5d/0x70
> > [] nf_hook_slow+0x5f/0xb0
> > [] __ip_local_out+0xad/0xe0
> > [] ip_local_out+0x17/0x40
> > [] ip_send_skb+0x14/0x40
> > [] udp_send_skb+0x91/0x260
> > [] udp_sendmsg+0x2f5/0x950
> > [] inet_sendmsg+0x60/0x90
> > [] sock_sendmsg+0x33/0x40
> > [] SYSC_sendto+0xee/0x160
> > [] SyS_sendto+0x9/0x10
> >
> > (248 bytes being an nf_conn structure)
> >
> > Those structures being cleared in gc_worker() later on we can't talk
> > about unreferenced object so this patch uses kmemleak_not_leak() to
> > prevent those warnings.
>
> If thats the case, why is kmemleak complaining? Are you sure this
> is a false positive?
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
