On 9/30/25 9:19 PM, Florian Westphal wrote:
> Eric Woudstra <[email protected]> wrote:
>> Add a script to test various scenarios where a bridge is involved
>> in the fastpath. It runs tests in the forward path, and also in
>> a bridged path.
>
> Why is this still an RFC, what is missing to appy this?
Changes in the patchset "conntrack: bridge: add double vlan, pppoe and
pppoe-in-q" has lead to changes in this script. I'm waiting for that
patch-set is to be accepted. Then I will send this script without the
rfc tag.
> Also:
>
> PASS: forward, without vlan-device, without vlan encap, client1,
> without fastpath
>
> net/bridge/br_private.h:1627 suspicious rcu_dereference_protected() usage!
>
> other info that might help us debug this:
>
> rcu_scheduler_active = 2, debug_locks = 1
> 7 locks held by socat/410:
> #0: ffff88800d7a9c90 (sk_lock-AF_INET){+.+.}-{0:0}, at:
> inet_stream_connect+0x43/0xa0
> #1: ffffffff9a779900 (rcu_read_lock){....}-{1:3}, at:
> __ip_queue_xmit+0x62/0x1830
> #2: ffffffff9a779900 (rcu_read_lock){....}-{1:3}, at: ip_output+0x57/0x3c0
> #3: ffffffff9a779900 (rcu_read_lock){....}-{1:3}, at:
> ip_finish_output2+0x263/0x17d0
> #4: ffffffff9a779900 (rcu_read_lock){....}-{1:3}, at:
> process_backlog+0x38a/0x14b0
> #5: ffffffff9a779900 (rcu_read_lock){....}-{1:3}, at:
> netif_receive_skb_internal+0x83/0x330
> #6: ffffffff9a779900 (rcu_read_lock){....}-{1:3}, at:
> nf_hook.constprop.0+0x8a/0x440
>
> stack backtrace:
> CPU: 0 UID: 0 PID: 410 Comm: socat Not tainted 6.17.0-rc7-virtme #1
> PREEMPT(full)
> Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
> Call Trace:
> <IRQ>
> dump_stack_lvl+0x6f/0xb0
> lockdep_rcu_suspicious.cold+0x4f/0xb1
> br_vlan_fill_forward_path_pvid+0x32c/0x410 [bridge]
> br_fill_forward_path+0x7a/0x4d0 [bridge]
> ...
>
> I did not see a mention of this, nor a bug fix.
>
> Its a pre-existing bug, br_vlan_fill_forward_path_pvid uses
> br_vlan_group() instead of _rcu version.
>
> Will you send a patch for this?
I had this as part of an upcoming patch-set, but I will remove it from
there and send it as a separate patch. I assume this can go to nf
instead of nf-next, as it is a bugfix?
