On Mon, 27 Apr 2026 12:41:53 +0200 Robin Jarry <[email protected]> wrote:
> Instead of guessing what are the proper header lengths, pass > a rte_net_hdr_lens struct to rte_net_get_ptype and use it to get the > proper header lengths/offsets in tap_verify_csum. > > This allows supporting stacked VLAN/QinQ tags and IPv6 extensions. > > Signed-off-by: Robin Jarry <[email protected]> > --- AI review spotted some things. But it looks like more of a bug in rte_net_get_ptype Now let me check the upstream source for the IPv6 header context:Found the bug. Let me verify the IPv6 payload_len semantics:That confirms the bug. `payload_len` is documented as "IP payload size, including ext. headers" — exactly what I needed. --- **Review of [PATCH dpdk v2] net/tap: use offsets provided by rte_net_get_ptype** Error: The IPv6 packet length check double-counts extension header bytes. `rte_ipv6_hdr.payload_len` is defined as "IP payload size, including ext. headers" (rte_ip6.h:487), meaning it already includes extension header bytes counted from just after the 40-byte fixed IPv6 header. But `hlen->l3_len` for `RTE_PTYPE_L3_IPV6_EXT` is set by `rte_net_get_ptype` to `40 + extension_headers_size` (rte_net.c:439). Adding them together counts the extension bytes twice: if (hlen->l2_len + hlen->l3_len + rte_be_to_cpu_16(iph->payload_len) > rte_pktmbuf_data_len(mbuf)) return; For any IPv6 packet with extension headers this check will fire spuriously, causing `tap_verify_csum` to return early without setting checksum offload flags — the very case this patch is trying to fix. The pre-existing code was correct: it used `sizeof(struct rte_ipv6_hdr)` (40) as the L3 term, leaving `payload_len` to account for extensions. The fix is to keep that same constant here: if (hlen->l2_len + sizeof(struct rte_ipv6_hdr) + rte_be_to_cpu_16(iph->payload_len) > rte_pktmbuf_data_len(mbuf)) return;
