Re: [PATCH v7] vsock: add vsock device

[Ian Campbell <ian.campbell@docker.com>]

>> > This stops one side from causing unbounded resource consumption.
>>
>> This is now parsable (thanks) but it's not clear to me how it stops
>> this from happening.
>>
>> I think what I am missing is the bit which says explicitly "and under
>> these circumstances you should stop processing the ring entirely" (as
>> opposed to dumping traffic on the floor)
>
> This is actually a general spec issue: we don't specify
> what should devices do when they encounter an invalid request,
> and that is not good: in particular, some devices just
> cause the guest to exit.

My concern isn't about invalid requests or really even error
conditions, but the ongoing reliability of the protocol under normal
use. The protocol here relies on both ends behaving in a certain way
when the rings are filling up to avoid a deadlock.

Ian.