From 5c2c3c38be396257a6a2e55bd601a12bb9781507 Mon Sep 17 00:00:00 2001 From: Paolo Abeni Date: Mon, 2 Feb 2026 12:43:14 +0100 Subject: [PATCH 1/2] net: gro: fix outer network offset The udp GRO complete stage assumes that all the packets inserted the RX have the `encapsulation` flag zeroed. Such assumption is not true, as a few H/W NICs can set such flag when H/W offloading the checksum for an UDP encapsulated traffic, the tun driver can inject GSO packets with UDP encapsulation and the problematic layout can also be created via a veth based setup. Due to the above, in the problematic scenarios, udp4_gro_complete() uses the wrong network offset (inner instead of outer) to compute the outer UDP header pseudo checksum, leading to csum validation errors later on in packet processing. Address the issue always clearing the encapsulation flag at GRO completion time. Such flag will be set again as needed for encapsulated packets by udp_gro_complete(). Fixes: 5ef31ea5d053 ("net: gro: fix udp bad offset in socket lookup by adding {inner_}network_offset to napi_gro_cb") Reviewed-by: Willem de Bruijn Signed-off-by: Paolo Abeni Reviewed-by: Eric Dumazet Link: https://patch.msgid.link/562638dbebb3b15424220e26a180274b387e2a88.1770032084.git.pabeni@redhat.com Signed-off-by: Jakub Kicinski --- net/core/gro.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/net/core/gro.c b/net/core/gro.c index 76f9c3712422..482fa7d7f598 100644 --- a/net/core/gro.c +++ b/net/core/gro.c @@ -265,6 +265,8 @@ static void gro_complete(struct gro_node *gro, struct sk_buff *skb) goto out; } + /* NICs can feed encapsulated packets into GRO */ + skb->encapsulation = 0; rcu_read_lock(); list_for_each_entry_rcu(ptype, head, list) { if (ptype->type != type || !ptype->callbacks.gro_complete) From bee60ce21b751275b3a7766f614373ef02dde512 Mon Sep 17 00:00:00 2001 From: Paolo Abeni Date: Mon, 2 Feb 2026 12:43:15 +0100 Subject: [PATCH 2/2] selftest: net: add a test-case for encap segmentation after GRO We had a few patches in this area and no explicit coverage so far. The test case covers the scenario addressed by the previous fix; reusing the existing udpgro_fwd.sh script to leverage part of the of the virtual network setup, even if such script is possibly not a perfect fit. Note that the mentioned script already contains several shellcheck violation; this patch does not fix the existing code, just avoids adding more issues in the new one. Reviewed-by: Willem de Bruijn Signed-off-by: Paolo Abeni Link: https://patch.msgid.link/768ca132af81e83856e34d3105b86c37e566a7ad.1770032084.git.pabeni@redhat.com Signed-off-by: Jakub Kicinski --- tools/testing/selftests/net/udpgro_fwd.sh | 64 +++++++++++++++++++++++ 1 file changed, 64 insertions(+) diff --git a/tools/testing/selftests/net/udpgro_fwd.sh b/tools/testing/selftests/net/udpgro_fwd.sh index a39fdc4aa2ff..9b722c1e4b0f 100755 --- a/tools/testing/selftests/net/udpgro_fwd.sh +++ b/tools/testing/selftests/net/udpgro_fwd.sh @@ -162,6 +162,39 @@ run_test() { echo " ok" } +run_test_csum() { + local -r msg="$1" + local -r dst="$2" + local csum_error_filter=UdpInCsumErrors + local csum_errors + + printf "%-40s" "$msg" + + is_ipv6 "$dst" && csum_error_filter=Udp6InCsumErrors + + ip netns exec "$NS_DST" iperf3 -s -1 >/dev/null & + wait_local_port_listen "$NS_DST" 5201 tcp + local spid="$!" + ip netns exec "$NS_SRC" iperf3 -c "$dst" -t 2 >/dev/null + local retc="$?" + wait "$spid" + local rets="$?" + if [ "$rets" -ne 0 ] || [ "$retc" -ne 0 ]; then + echo " fail client exit code $retc, server $rets" + ret=1 + return + fi + + csum_errors=$(ip netns exec "$NS_DST" nstat -as "$csum_error_filter" | + grep "$csum_error_filter" | awk '{print $2}') + if [ -n "$csum_errors" ] && [ "$csum_errors" -gt 0 ]; then + echo " fail - csum error on receive $csum_errors, expected 0" + ret=1 + return + fi + echo " ok" +} + run_bench() { local -r msg=$1 local -r dst=$2 @@ -260,6 +293,37 @@ for family in 4 6; do ip netns exec $NS_SRC $PING -q -c 1 $OL_NET$DST_NAT >/dev/null run_test "GRO fwd over UDP tunnel" $OL_NET$DST_NAT 10 10 $OL_NET$DST cleanup + + # force segmentation and re-aggregation + create_vxlan_pair + ip netns exec "$NS_DST" ethtool -K veth"$DST" generic-receive-offload on + ip netns exec "$NS_SRC" ethtool -K veth"$SRC" tso off + ip -n "$NS_SRC" link set dev veth"$SRC" mtu 1430 + + # forward to a 2nd veth pair + ip -n "$NS_DST" link add br0 type bridge + ip -n "$NS_DST" link set dev veth"$DST" master br0 + + # segment the aggregated TSO packet, without csum offload + ip -n "$NS_DST" link add veth_segment type veth peer veth_rx + for FEATURE in tso tx-udp-segmentation tx-checksumming; do + ip netns exec "$NS_DST" ethtool -K veth_segment "$FEATURE" off + done + ip -n "$NS_DST" link set dev veth_segment master br0 up + ip -n "$NS_DST" link set dev br0 up + ip -n "$NS_DST" link set dev veth_rx up + + # move the lower layer IP in the last added veth + for ADDR in "$BM_NET_V4$DST/24" "$BM_NET_V6$DST/64"; do + # the dad argument will let iproute emit a unharmful warning + # with ipv4 addresses + ip -n "$NS_DST" addr del dev veth"$DST" "$ADDR" + ip -n "$NS_DST" addr add dev veth_rx "$ADDR" \ + nodad 2>/dev/null + done + + run_test_csum "GSO after GRO" "$OL_NET$DST" + cleanup done exit $ret