ntptime: Reject short NTP replies.

[pablogventura]

Summary

• Fixes ntptime spoof or bad response shape #1136: time() used recv(48) and unpacked msg[40:44] with no length check. A short datagram could raise a confusing error or yield a bad timestamp.
• Require len(msg) >= 48, otherwise OSError(-1) (same family as a socket timeout). Still uses recv(48).
• Do not compare the UDP source address to the send target (can differ in complex network topologies).
• Kiss of Death / zero timestamps are unchanged (see ntptime: Does not handle KoD (Kiss of Death) resonse. #604).

Test plan

• micropython/net/ntptime/test_ntptime.py (valid 48-byte reply, short payload), wired into package tests CI.

[pablogventura]

ntptime.mpy (mpy-cross, compiled as ntptime.py):

Version	bytes	vs master
master	563	-
this commit	600	+37

[dpgeorge]

Thanks for the PR.

I'm not sure it's wise to check the address of the UDP sender via recvfrom(). In complex network topologies it could be that the IP address differs between the sendto and recvfrom.

See eg this implementation which indeed does not check the address: https://github.com/vicharl/ntpclient/blob/master/ntpclient.c#L326-L333

Also might be worth seeing how this library handles that scenario: https://github.com/ntpsec/ntpsec

The other change to reject short replies seems fine.

[pablogventura]

Thanks @dpgeorge - dropped the recvfrom() source-address check and kept rejecting short replies with OSError(-1). Still uses recv(48).

Addressed in 1aa3dfc.

ntptime.mpy (mpy-cross, compiled as ntptime.py):

Version	bytes	vs master
master	563	-
previous on this branch	600	+37
this commit	584	+21

[dpgeorge]

Thanks for updating.