Re: [PATCH net] rds: tcp: hold the net_device across ipv6_chk_addr() in rds_tcp_laddr_check()

[Allison Henderson <achender@kernel.org>]

On Thu, 2026-07-09 at 00:44 -0700, Xiang Mei wrote:
> rds_tcp_laddr_check() looks up a scoped IPv6 interface with
> dev_get_by_index_rcu(), drops the RCU read-side lock, and only then
> passes the bare struct net_device * into ipv6_chk_addr().
> 
> dev_get_by_index_rcu() only keeps the device alive within the same RCU
> read-side section. After rcu_read_unlock(), a concurrent RTM_DELLINK can
> free the net_device; ipv6_chk_addr() then dereferences the stale pointer
> in __ipv6_chk_addr_and_flags() (e.g. l3mdev_master_dev_rcu(dev)), reading
> freed memory.
> 
> Take a reference with dev_hold() before dropping the RCU lock and release
> it with dev_put() after ipv6_chk_addr(), so the device cannot be freed
> while in use. dev_put(NULL) is a no-op, so the scope_id == 0 path is
> unaffected.
> 
>   BUG: KASAN: slab-use-after-free in __ipv6_chk_addr_and_flags (... net/ipv6/addrconf.c:1998)
>   Read of size 8 at addr ffff8880106ec000 by task exploit/153
>   Call Trace:
>    ...
>    kasan_report (mm/kasan/report.c:595)
>    __ipv6_chk_addr_and_flags (... net/ipv6/addrconf.c:1998)
>    ipv6_chk_addr (net/ipv6/addrconf.c:2031 net/ipv6/addrconf.c:1972)
>    rds_tcp_laddr_check (net/rds/tcp.c:370)
>    rds_bind (net/rds/bind.c:248)
>    __sys_bind (net/socket.c:1920)
>    __x64_sys_bind (net/socket.c:1956)
>    do_syscall_64 (arch/x86/entry/syscall_64.c:63)
>    entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:121)
> 
> Fixes: eee2fa6ab322 ("rds: Changing IP address internal representation to struct in6_addr")
> Reported-by: Weiming Shi <bestswngs@gmail.com>
> Assisted-by: Claude:claude-opus-4-8
> Signed-off-by: Xiang Mei <xmei5@asu.edu>
> ---
>  net/rds/tcp.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/net/rds/tcp.c b/net/rds/tcp.c
> index a1de114d5e2e..204dcdc33c27 100644
> --- a/net/rds/tcp.c
> +++ b/net/rds/tcp.c
> @@ -363,12 +363,16 @@ int rds_tcp_laddr_check(struct net *net, const struct in6_addr *addr,
>  			rcu_read_unlock();
>  			return -EADDRNOTAVAIL;
>  		}
> +		dev_hold(dev);
>  		rcu_read_unlock();
>  	}
>  #if IS_ENABLED(CONFIG_IPV6)
>  	ret = ipv6_chk_addr(net, addr, dev, 0);
> +	dev_put(dev);
Hi Xiang,

Thanks for working on this.  It looks like this patch generated
some CI warnings for dev_hold/put that should be addressed:
https://patchwork.kernel.org/project/netdevbpf/patch/20260709074459.326345-1-xmei5@asu.edu/

Per the comments from include/linux/netdevice.h:4546, netdev_hold/put
replaced dev_hold/put, which takes a netdevice_tracker pointer.  But I
think the easier way to do this may be just to widen the rcu locks already
in this function.  Just move rcu_read_lock(); above the scope_id check,
and make sure it's released before any returns.  That should get away from 
having to use the hold/puts all together.

	rcu_read_lock();
	if (scope_id != 0) {
		dev = dev_get_by_index_rcu(net, scope_id);
		if (!dev) {
			rcu_read_unlock();
			return -EADDRNOTAVAIL;
		}
	}
#if IS_ENABLED(CONFIG_IPV6)
	ret = ipv6_chk_addr(net, addr, dev, 0);
	if (ret) {
		rcu_read_unlock();
		return 0;
	}
#endif
	rcu_read_unlock();
	return -EADDRNOTAVAIL;

Also, this patch conflicts with another submitted patch: 
[net,v2] rds: Fix inet6_addr_lst NULL dereference when IPv6 is disabled:
https://patchwork.kernel.org/project/netdevbpf/patch/20260709162723.367523-1-Ilia.Gavrilov@infotecs.ru/

So, please rebase on Ilia's v2, and mention the dependency in your change 
log so the stable folks pick them up in the right order.  Both fixes carry 
the same Fixes: tag so they should travel together.

Thank you!
Allison

>  	if (ret)
>  		return 0;
> +#else
> +	dev_put(dev);
>  #endif
>  	return -EADDRNOTAVAIL;
>  }