On 1/24/25 10:31 AM, Jeff Layton wrote:
Stepping back, when we do find failing callbacks, should we be doing more to alert the admin? What would be an appropriate response? Ratelimited pr_notice()? Conditional tracepoints? Something else?
If we can identify a reasonable action that an admin can take, then pr_ratelimited (or once per client instance) to report a decoding failure makes sense. Report the IP address of the client that is sending us weird crap. For the more conventional session failures, I'm less enthusiastic about logging these because frequently they are the result of transport failures (or a suspended client, or ...). We'll have to sort through the various cases to understand where a logged error might be helpful/actionable. -- Chuck Lever
