Re: [PATCH net] bonding: Fix bonding crash

Fri, Sep 02, 2016 at 03:33:20PM CEST, eric.dumazet@gmail.com wrote:

On Fri, 2016-09-02 at 09:52 +0200, Jiri Pirko wrote:

quoted

No, please, don't make bonding a spacial citizen introducing this.
Please handle the issue inside the bonding code, like we do for the rest
of master devices (and how it was once done for bonding). Thanks.

I do not feel this netdev_is_rx_handler_busy() use is special to
bonding.

It makes sense to use it early, to avoid complex rollback, once various
events have been sent all over.

bond_enslave() is 447 lines long already.

You perfectly know how hard it is to 'handle the issue inside the
bonding code' as you chose to not fix bonding and write team instead.

So Mahesh patch makes perfect sense to me. It exactly fixes a stupid
bond_enslave() behavior, trying to set rx_handler way too late.

By doing sanity checks before any action, we simply do not have to add
complex rollback.

Understand the reason. All I say is we tried hard (I surely did) in
the past to remove bonding specific quirks and now we are adding one.
And the simple reason is that the code is such a mess.

Just use team instead and you'll be fine. You can "google" it :)