NFD

I captured a Wireshark trace on the router at one incident.

From the attached trace, we can see:

1. In packet 8, the end host tcp4://128.196.203.35:35306 sends a prefix registration command.
2. In packet 10, the router sends an Interest to udp4://199.175.53.34:6363 to retrieve the signing certificate.
3. In packet 11, the certificate is retrieved successfully.
4. In packet 12, the router returns a 400 error to the end host.
5. 50 seconds later, in packet 20, the end host sends a new prefix registration command.
6. In packet 21, the router returns a 200 successful response to the end host.

On another long-running end host, I observe that 400 error occurs once an hour:

sunny@vps3:~$ grep 'fail to propagate' /var/log/ndn/nfd.log | awk '$1>1492664400 {print $1}' | awk 'NR>1{print int($1)-last} {last=int($1)}'
3651
3650
3651
3651
3651
3651
3651
3651
3651

I know of two places where constant "one hour" is used:

• Default value of ValidatorConfig's keyTimestampTtl is set to one hour.
• FreshnessPeriod of IdentityCertificate is set to one hour.

I guess the IdentityCertificate FreshnessPeriod is the deciding factor: whenever a certificate retrieval is needed, the registration command fails.

I believe this is a bug in Dispatcher::processControlCommandInterest:

shared_ptr<ControlParameters> parameters;
try {
  parameters = parser(pc);
}
catch (const tlv::Error&) {
  return;
}

AcceptContinuation accept = bind(accepted, _1, prefix, interest, parameters.get());
RejectContinuation reject = bind(rejected, _1, interest);
authorization(prefix, interest, parameters.get(), accept, reject);

If authorization function is asynchronous, the ControlParameters object is deallocated when Dispatcher::processControlCommandInterest returns, and thus parameters.get() passed to accept is a dangling pointer.

https://gerrit.named-data.net/3857