Re: [PATCH net] net, sched: fix soft lockup in tc_classify

[Daniel Borkmann <daniel@iogearbox.net>]

On 12/21/2016 07:51 PM, Cong Wang wrote:
> On Wed, Dec 21, 2016 at 9:04 AM, Daniel Borkmann <daniel@iogearbox.net> wrote:
>> What happens is that in tc_ctl_tfilter(), thread A allocates a new
>> tp, initializes it, sets tp_created to 1, and calls into tp->ops->change()
>> with it. In that classifier callback we had to unlock/lock the rtnl
>> mutex and returned with -EAGAIN. One reason why we need to drop there
>> is, for example, that we need to request an action module to be loaded.
>
> Excellent catch!
>
> But why do we have to replay the request here? Shouldn't we just return
> EAGAIN to user-space and let user-space decide to retry or not?
> Replaying is the root of the evil here.

Right, this behavior is already pretty old (2005), see history
tree 8d7c694553dc ("[PKT_SCHED]: act_api.c: drop rtnl for loading
modules") and 437293de63d8 ("[PKT_SCHED]: cls_api.c: drop rtnl
for loading modules"), some binaries could rely on that behavior
in one way or another I'd presume.