[BUG] af_packet FANOUT and device dismantle

[BUG] af_packet FANOUT and device dismantle

[Eric Dumazet]

Hi David

When you added fanout in commit dc99f600698dcac,
it seems a device dismantle is not properly handled.

packet_notifier() does properly finds all sockets attached to the
device and we call __unregister_prot_hook()

But the actual dev_remove_pack() is called when the last socket attached
to the FANOUT group is _closed_ , possibly minutes after the device
disappeared.

So we trigger the BUG_ON(!list_empty(&dev->ptype_all)); in
netdev_run_todo(), that came with linux-4.0 in 
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=7866a621043fbaca3d7389e9b9f69dd1a2e5a855

Fix would be to call dev_remote_pack() from __fanout_unlink() when
num_members becomes 0.

Let me know if you agree with this, thanks.

Re: [BUG] af_packet FANOUT and device dismantle

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 01 Sep 2016 17:04:43 -0700

> When you added fanout in commit dc99f600698dcac,
> it seems a device dismantle is not properly handled.
> 
> packet_notifier() does properly finds all sockets attached to the
> device and we call __unregister_prot_hook()
> 
> But the actual dev_remove_pack() is called when the last socket attached
> to the FANOUT group is _closed_ , possibly minutes after the device
> disappeared.
> 
> So we trigger the BUG_ON(!list_empty(&dev->ptype_all)); in
> netdev_run_todo(), that came with linux-4.0 in 
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=7866a621043fbaca3d7389e9b9f69dd1a2e5a855
> 
> Fix would be to call dev_remote_pack() from __fanout_unlink() when
> num_members becomes 0.
> 
> Let me know if you agree with this, thanks.

I completely agree with your analysis, good catch.