[Dnsmasq-discuss] [PATCH] DHCPv6 network range is not checked well with dhcp-sequential-ip

[Geert Stappers]

On Tue, Oct 07, 2025 at 07:18:27PM +0200, Petr Menšík via Dnsmasq-discuss wrote:
> Hello Simon and everyone,
> 
> I am sending this patch on behalf of Tomáš Korbař, who worked on
> debugging the issue and even written a test [1] to verify it works. Test
> is written in our shell based beakerlib, but I think is readable enough
> even in basic form.

I'm sending this reminder.
Because there was no-follow-up yet.

 
> Currentdhcp-range does not work well in case DHCP client would refuse the
> offered address. When this happens, current dnsmasq does not even stop at
> maximal address specified. This change adds similar check like when no
> dhcp-sequential-ip is used. Tomáš tested it already, but would anyone mind
> testing it also? If there is no regression or need to adjust it, would you
> mind merging it? Thank you in advance! Cheers, Petr
> 1. https://gitlab.com/redhat/centos-stream/tests/dnsmasq/-/merge_requests/6
> 
> -- 
> Petr Menšík
> Senior Software Engieer, RHEL
> Red Hat, https://www.redhat.com/
> PGP: DFCF908DB7C87E8E529925BC4931CA5B6C9FC5CB

> From 913472864bdcd5b7bb01c31ff27706b6dfbb99a2 Mon Sep 17 00:00:00 2001
> From: Tomas Korbar <tkorbar at redhat.com>
> Date: Tue, 7 Oct 2025 19:12:00 +0200
> Subject: [PATCH] Fix obeying dhcp-range if dhcp-sequential-ip and IPV6 is used
> 
> Previously starting position for searching the next free address
> in IPV6 range was no checked whether it is in bounds. This
> commit adds modulo operation like when dhcp-sequential-ip is
> not used to ensure dnsmasq does not offer out of range address.
> ---
>  src/dhcp6.c | 17 ++++++++---------
>  1 file changed, 8 insertions(+), 9 deletions(-)
> 
> diff --git a/src/dhcp6.c b/src/dhcp6.c
> index 482caf7d..99cc77b2 100644
> --- a/src/dhcp6.c
> +++ b/src/dhcp6.c
> @@ -522,27 +522,26 @@ struct dhcp_context *address6_allocate(struct dhcp_context *context,  unsigned c
>  	continue;
>        else
>  	{ 
> +	  u64 range = 1 + addr6part(&c->end6) - addr6part(&c->start6);
> +	  u64 offset = c->addr_epoch;
>  	  if (!temp_addr && option_bool(OPT_CONSEC_ADDR))
>  	    {
>  	      /* seed is largest extant lease addr in this context,
>  		 skip addresses equal to the number of addresses rejected
>  		 by clients. This should avoid the same client being offered the same
>  		 address after it has rjected it. */
> -	      start = lease_find_max_addr6(c) + 1 + serial + c->addr_epoch;
> +	      offset += lease_find_max_addr6(c) - addr6part(&c->start6) + 1 + serial;
>  	      if (c->addr_epoch)
>  		c->addr_epoch--;
>  	    }
>  	  else
>  	    {
> -	      u64 range = 1 + addr6part(&c->end6) - addr6part(&c->start6);
> -	      u64 offset = j + c->addr_epoch;
> -
> -	      /* don't divide by zero if range is whole 2^64 */
> -	      if (range != 0)
> -		offset = offset % range;
> -
> -	      start = addr6part(&c->start6) + offset;
> +	      offset += j;
>  	    }
> +	  /* don't divide by zero if range is whole 2^64 */
> +	  if (range != 0)
> +		offset = offset % range;
> +	  start = addr6part(&c->start6) + offset;
>  
>  	  /* iterate until we find a free address. */
>  	  addr = start;
> -- 
> 2.51.0
> 

Groeten
Geert Stappers
Picking up some patches
-- 
Silence is hard to parse