Re: [PATCH v6 6/6] selftests/bpf: tc_tunnel - validate decap GSO and encapsulation state

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Pointer arithmetic `kskb->head + kskb->end` causes build failures and runtime logic errors on 32-bit kernels.
- [Medium] Missing kernel feature causes test to fail instead of skip, breaking CI on older kernels.
--

commit f73fb4725baa7b601a1b82c32e5095c6b4294f5c
Author: Nick Hudson <nhudson@akamai.com>

selftests/bpf: tc_tunnel - validate decap GSO and encapsulation state

This commit enhances the tc_tunnel selftest to validate post-decap tunnel
state for both GSO and non-GSO packets, and checks IPXIP decap for IPIP/SIT
paths.

> diff --git a/tools/testing/selftests/bpf/progs/test_tc_tunnel.c b/tools/testing/selftests/bpf/progs/test_tc_tunnel.c
> index 7376df405a6b..853bca962910 100644
> --- a/tools/testing/selftests/bpf/progs/test_tc_tunnel.c
> +++ b/tools/testing/selftests/bpf/progs/test_tc_tunnel.c
[ ... ]
> @@ -621,6 +645,11 @@ static int decap_internal(struct __sk_buff *skb, int off, int len, char proto)
>  		break;
>  	case IPPROTO_GRE:
>  		olen += sizeof(struct gre_hdr);
> +		if (!bpf_core_enum_value_exists(enum bpf_adj_room_flags,
> +						BPF_F_ADJ_ROOM_DECAP_L4_GRE))
> +			return TC_ACT_SHOT;

If the kernel is missing support for these new decapsulation flags, will
returning TC_ACT_SHOT cause the test harness to time out and fail the test?

Since selftests often run on older kernels, should missing features gracefully
skip the test instead of failing it to prevent false positives and CI
breakages on stable or LTS branches?

[ ... ]
> @@ -655,6 +688,40 @@ static int decap_internal(struct __sk_buff *skb, int off, int len, char proto)
>  	if (bpf_skb_adjust_room(skb, -olen, BPF_ADJ_ROOM_MAC, flags))
>  		return TC_ACT_SHOT;
>  
> +	kskb = bpf_cast_to_kern_ctx(skb);
> +	shinfo = bpf_core_cast(kskb->head + kskb->end, struct skb_shared_info);

Will this arithmetic work correctly on 32-bit kernels?

On 64-bit architectures, sk_buff_data_t is an integer offset, so
kskb->head + kskb->end works. However, on 32-bit platforms, sk_buff_data_t is
a pointer (unsigned char *), making the addition of two pointers invalid C and
causing build failures.

Furthermore, even if cross-compiled with a 64-bit vmlinux.h, adding
kskb->head to an absolute kskb->end pointer would yield an invalid address at
runtime.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260504101759.3319427-1-nhudson@akamai.com?part=6