On Thu, Jun 4, 2020 at 22:18:10 +0300, Adrian Bunk wrote: > On Thu, Jun 04, 2020 at 07:55:32PM +0200, Julien Cristau wrote: > > Control: tag -1 moreinfo > > > > On Thu, Jun 04, 2020 at 07:59:22PM +0300, Adrian Bunk wrote: > > > FAIL: dtls_hello_random_value > > > ============================= > > > > > > testing: default > > > client:156: client: Handshake failed: Resource temporarily unavailable, > > > try again. > > > server:271: server: Handshake has failed: The TLS connection was > > > non-properly terminated. > > > > > > FAIL dtls_hello_random_value (exit status: 1) > > > > > This particular test seems to use a AF_UNIX socketpair, I'm not sure > > how it would fail due to network setup? > > I copied the log from the 3.6.13-4 arm64 failure. > This specific test passes in the other logs. > > The binary-all log (non-conova) of 3.6.13-4 has no test failures, > but the build fails due to timeout after 150 minutes without output. > > My gut feeling is that there might be an (unrelated) problem with > the #961889 fix, but this is just a guess. > The arch:all failure is at grnet and that buildd has both v4 and v6 addresses, so presumably unrelated.
Which leaves us with the armel failure (on arm-conova-03) which does look related to the lack of public v4 address, e.g.: > FAIL: sni-hostname.sh > ===================== > > Checking SNI hostname in gnutls-cli > Echo Server listening on IPv6 :: port 26448...done > Could not connect to 127.0.0.1:26448: Connection refused > Failure: 1. handshake should have succeeded! > Exiting via signal 15 > FAIL sni-hostname.sh (exit status: 1) If we listen on :: and then try to connect to 127.0.0.1, that won't work. And indeed gnutls-serv seems to call getaddrinfo with node == NULL and hints.ai_flags == AI_PASSIVE|AI_ADDRCONFIG, to figure out what address to listen on. If the host has no non-local ipv4 address, that getaddrinfo call returns :: but not 0.0.0.0; and then the test hardcodes 127.0.0.1 as the address for gnutls-cli to connect to, and sadness ensues. Cheers, Julien
