Re: [PATCH net-next v3] net: dropreason: add SKB_DROP_REASON_IP_TTL_EXCEEDED

[Jiayuan Chen <jiayuan.chen@linux.dev>]

on 9/4/26 11:01 AM, Junjie Cao wrote:
> The forwarding paths report an expired TTL or hop limit as
> SKB_DROP_REASON_IP_INHDR, the reason otherwise used for a header that is
> malformed (ip_input.c, exthdrs.c, br_netfilter). Nothing else in the drop
> path separates the two: IPSTATS_MIB_INHDRERRORS covers both, and the TTL
> check runs before NF_INET_FORWARD, so netfilter tracing stops at
> PREROUTING and never sees the drop.
>
> The Fedora bug linked below shows how that reads in practice. The
> reporter took kfree_skb(reason=IP_INHDR, loc=ip_forward) to mean the
> software header checksum check had failed, and worked through RX checksum
> offload, tc csum actions and both libvirt firewall backends before the
> drops turned out to be replies arriving with TTL 1. ip_forward() never
> verifies the header checksum; that runs earlier, in ip_rcv_core(), and
> reports IP_CSUM.
>
> TTL expiry is not a corner case -- every traceroute through a Linux
> router goes through too_many_hops.
>
> The three loopback hop limit checks in exthdrs.c drop with no reason at
> all; give them the new one.
>
> IPSTATS_MIB_INHDRERRORS stays as it is: RFC 1213 counts time-to-live
> exceeded under ipInHdrErrors. The drop reason has no such constraint.
>
> Link: https://bugzilla.redhat.com/show_bug.cgi?id=2517131
> Signed-off-by: Junjie Cao <junjie.cao@intel.com>


Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev>

[...]

>   /**
> @@ -606,6 +607,11 @@ enum skb_drop_reason {
>   	SKB_DROP_REASON_PSP_OUTPUT,
>   	/** @SKB_DROP_REASON_RECURSION_LIMIT: Dead loop on virtual device. */
>   	SKB_DROP_REASON_RECURSION_LIMIT,
> +	/**
> +	 * @SKB_DROP_REASON_IP_TTL_EXCEEDED: IPv4 TTL or IPv6 hop limit hit
> +	 * zero (see IPSTATS_MIB_INHDRERRORS)


nit: "<= 1" would be more accurate than "hit zero".


> +	 */
> +	SKB_DROP_REASON_IP_TTL_EXCEEDED,
>   	/**
>   	 * @SKB_DROP_REASON_MAX: the maximum of core drop reasons, which
>   	 * shouldn't be used as a real 'reason' - only for tracing code gen
>