[Dnsmasq-discuss] [PATCH] Refuse to start with EADDRINUSE in --bind-dynamic mode

[Simon Kelley]

Isn't this sufficient to fix the problem?

Not calling die() when bind-dynamic is set is intended to handle the 
case that  bind returns EADDRNOTAVAIL because you've configured 
--listen-address=1.2.3.4 but there's not a local interface with that 
address. dnsmasq runs anyway in the expectation that such an interface 
will appear in the future and a socket will be bound then.

I don't think there's a die()/syslog() conflict at all.


diff --git a/src/network.c b/src/network.c
index ca9fada..db1d528 100644
--- a/src/network.c
+++ b/src/network.c
@@ -924,7 +924,7 @@ static int make_sock(union mysockaddr *addr, int 
type, int dienow)
         {
           /* failure to bind addresses given by --listen-address at 
this point
              is OK if we're doing bind-dynamic */
-         if (!option_bool(OPT_CLEVERBIND))
+         if (!option_bool(OPT_CLEVERBIND) || errno == EADDRINUSE)
             die(s, daemon->addrbuff, EC_BADNET);
         }
        else

Cheers,

Simon.

On 22/11/2023 19:27, Petr Menšík wrote:
> Hello everyone,
> 
> I have received error report RHEL-16398 [1], which I think makes sense 
> to fix even in the lastest version. I believe it allows non-intentional 
> another instance running without error. What is worse, it does not even 
> show any warning that initialization is incomplete.
> 
> Of course the problem at start is those errors happen in time when no 
> log is available. I think that can be fixed easily by using stderr at 
> that time. That is patch #1.
> 
> Second makes EADDRNOTAVAIL bind errors still hidden, but prints all 
> other errors at least to stderr. On a system with systemd that should 
> make it present in journalctl -u dnsmasq anyway. EADDRINUSE is made 
> fatal, because that would not be usually handled by new addresses added 
> later. If there is a need to start another dnsmasq instance without TCP 
> listeners, I think that should be specified more explicitly. Makes 
> EADDRINUSE fatal the same way as with --bind-interfaces.
> 
> Would you find any other errors, which should be hidden or made fatal? 
> What would you think of those changes?
> 
> 1. https://issues.redhat.com/browse/RHEL-16398
> 
> 
> _______________________________________________
> Dnsmasq-discuss mailing list
> Dnsmasq-discuss at lists.thekelleys.org.uk
> https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss