[Dnsmasq-discuss] [PATCH] Do not shorten a host's infinite lease to the client's requested time

[Simon Kelley]

I'm not sure why the existing behaviour is a problem, and I'm concerned 
by the comment in the github bug:

Apparently, iOS rejects infinite leases on mobile Wi-Fi connections.

which, if true, would make the new behaviour a problem. Unfortunately, 
Google AI seems to have decided that infinite leases are a problem, 
based solely on that comment :)

I have no Apple devices, so can't test this.

Happy to take the patch if it can be explained to me how this solves a 
real problem, and we are sure it doesn't add an new problem.

It's also worth noting that if it's 50/50 which behaviour is "correct", 
there's a built-in bias towards the _existing_ behaviour. Gratuitously 
changing such subtle details risks breaking existing installations which 
rely on the old behaviour and their owners almost certainly don't realise.

Simon.




On 09.08.2026 09:32, Dominik Derigs wrote:
> Hi Simon,
> 
> a `dhcp-host` reservation with `infinite` silently ends up with a finite 
> lease
> as soon as the client sends option 51. calc_time() treats the configured 
> lease
> time as an upper bound, and `infinite` is an unbounded one, so whatever the
> client asks for is shorter and wins. Apple devices request 7776000 
> seconds in
> every DHCPREQUEST, so every iPhone/iPad/Watch with a reservation is 
> affected.
> 
> The attached patch honors `infinite` literally when it comes from `dhcp- 
> host`:
> for a reserved address there is nothing to be gained from expiring it 
> early.
> `dhcp-range` is left alone - there, granting a client's shorter request 
> does
> return pool addresses sooner.
> 
> Reproducer, with a client sending option 51 = 7776000 for its reserved 
> address:
> 
>    dnsmasq -d --port=0 --dhcp-authoritative \
>      --dhcp-range=10.99.0.100,10.99.0.200,255.255.255.0,1h \
>      --dhcp-host=02:aa:bb:cc:dd:ee,10.99.0.9,testhost,infinite
> 
> The DHCPACK lease time is 7776000 before and 4294967295 with the patch. A
> finite `dhcp-host` lease time and pool addresses behave exactly as 
> before, in
> both directions.
> 
> Downstream report: https://github.com/pi-hole/FTL/issues/3014
> 
> Best regards,
> Dominik