Re: What is the purpose of dev->gflags?

[Vladimir Oltean <olteanv@gmail.com>]

On Mon, Apr 11, 2022 at 06:27:54PM +0200, Nicolas Dichtel wrote:
> Same here. Some complex path are called (eg. dev_change_rx_flags =>
> ops->ndo_change_rx_flags() => vlan_dev_change_rx_flags => dev_set_allmulti =>
> __dev_set_allmulti => etc).
> Maybe you made an audit to check that other flags cannot be changed. But, if it
> changes in the future, we will miss them here.

I guess I just don't see what other dev->flags that aren't masked out
from netdev notifier calls may or should change during the call to
__dev_set_allmulti(), regardless of the complexity or depth of the
call path.

And the commit that added the __dev_notify_flags() call said "dev:
always advertise rx_flags changes via netlink" (i.e. the function was
called for its rtmsg_ifinfo() part, not for its call_netdevice_notifiers()
part).

There *was* no call to dev_notify_flags prior to that commit, and it
didn't give a reason for voluntarily going through the netdev notifiers,
either.

> Did you see a bug? What is the issue?

I didn't see any bug, as mentioned I was trying to follow how
dev->gflags is used (see title) and stumbled upon this strange pattern
while doing so. dev->gflags is not updated from dev_set_allmulti()
almost by definition, otherwise in-kernel drivers wouldn't have a way to
update IFF_ALLMULTI without user space becoming aware of it.

The reason for emailing you to was to understand the intention, I do
understand that the code has went through changes since 2013 and that
a more in-depth audit is still needed before making any change.