On Thu, Feb 16, 2017 at 11:08 AM, <l...@pengaru.com> wrote: > On Thu, Feb 16, 2017 at 10:52:19AM -0500, Soheil Hassas Yeganeh wrote: >> On Thu, Feb 16, 2017 at 10:50 AM, Soheil Hassas Yeganeh >> <soh...@google.com> wrote: >> > Thank you Vito for the report. >> > >> > The patch you cited actually resolves a similar backward compatibility >> > problem for traceroute. >> > >> > I suspect the problem here is that there's a local error queued on the >> > error queue after an ICMP message. ping apparently expect the >> > sk->sk_err to be set for the local errors as well, and hence the >> > error. Ideally, ping should read the error queue if there an EPOLLERR, >> > because local errors never sk->sk_err on their own. That is, if we >> > have >> >> [oops] That is, if we have only one local error on the error queue, we >> cannot rely on having an error on recvmsg (i.e., sk->sk_err being set) >> even in 4.9. >> >> <snip> > > Hi Soheil, > > This doesn't appear to be trivially reproducible here by just running ping > as it were originally discovered. I'll see if I can reliably cause the > malfunction somehow, but until then I can't meaningfully test patches. > > Perhaps a form of fault injection would make more sense if there's a > reasonable idea of what this is stemming from?
I tried to generate different ICMP errors as well as local errors, but unfortunately haven't been able to reproduce the problem. > I've opened an issue with iputils on github in the event that this is found > to be a ping bug. Your input might be helpful there as well: > https://github.com/iputils/iputils/issues/74 Sent a pull request. Although, we might want to at least confirm the userspace patch fixes the issue in ping. Thanks! Soheil > Thanks, > Vito Caputo
