Rather than drop all pending Tx offloads on recirculation,
preserve inner offloads (and mark packet with outer Tx offloads)
when parsing the packet again.
Fixes: c6538b443984 ("dpif-netdev: Fix crash due to tunnel offloading on
recirculation.")
Fixes: 084c8087292c ("userspace: Support VXLAN and GENEVE TSO.")
Signed-off-by: David Marchand <[email protected]>
---
Changes since v1:
- rebased,
- dropped API change on miniflow_extract(), rely on tunnel offloading
flag presence instead,
- introduced dp_packet_reset_outer_offsets,
---
lib/dp-packet.h | 23 +++++++++++------------
lib/dpif-netdev.c | 27 ---------------------------
lib/flow.c | 34 ++++++++++++++++++++++++++++------
3 files changed, 39 insertions(+), 45 deletions(-)
diff --git a/lib/dp-packet.h b/lib/dp-packet.h
index f94a82b07c..0f487a4283 100644
--- a/lib/dp-packet.h
+++ b/lib/dp-packet.h
@@ -210,6 +210,7 @@ static inline void dp_packet_set_tso_segsz(struct dp_packet
*, uint16_t);
void *dp_packet_resize_l2(struct dp_packet *, int increment);
void *dp_packet_resize_l2_5(struct dp_packet *, int increment);
static inline void *dp_packet_eth(const struct dp_packet *);
+static inline void dp_packet_reset_outer_offsets(struct dp_packet *);
static inline void dp_packet_reset_offsets(struct dp_packet *);
static inline void dp_packet_reset_offload(struct dp_packet *);
static inline uint16_t dp_packet_l2_pad_size(const struct dp_packet *);
@@ -433,15 +434,22 @@ dp_packet_eth(const struct dp_packet *b)
? dp_packet_data(b) : NULL;
}
-/* Resets all layer offsets. 'l3' offset must be set before 'l2' can be
- * retrieved. */
+/* Resets all outer layer offsets. */
static inline void
-dp_packet_reset_offsets(struct dp_packet *b)
+dp_packet_reset_outer_offsets(struct dp_packet *b)
{
b->l2_pad_size = 0;
b->l2_5_ofs = UINT16_MAX;
b->l3_ofs = UINT16_MAX;
b->l4_ofs = UINT16_MAX;
+}
+
+/* Resets all layer offsets. 'l3' offset must be set before 'l2' can be
+ * retrieved. */
+static inline void
+dp_packet_reset_offsets(struct dp_packet *b)
+{
+ dp_packet_reset_outer_offsets(b);
b->inner_l3_ofs = UINT16_MAX;
b->inner_l4_ofs = UINT16_MAX;
}
@@ -1316,15 +1324,6 @@ dp_packet_hwol_set_tunnel_gre(struct dp_packet *b)
*dp_packet_ol_flags_ptr(b) |= DP_PACKET_OL_TX_TUNNEL_GRE;
}
-/* Clears tunnel offloading marks. */
-static inline void
-dp_packet_hwol_reset_tunnel(struct dp_packet *b)
-{
- *dp_packet_ol_flags_ptr(b) &= ~(DP_PACKET_OL_TX_TUNNEL_VXLAN |
- DP_PACKET_OL_TX_TUNNEL_GRE |
- DP_PACKET_OL_TX_TUNNEL_GENEVE);
-}
-
/* Mark packet 'b' as a tunnel packet with outer IPv4 header. */
static inline void
dp_packet_hwol_set_tx_outer_ipv4(struct dp_packet *b)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index b572fab23d..11f82210a8 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -115,7 +115,6 @@ COVERAGE_DEFINE(datapath_drop_lock_error);
COVERAGE_DEFINE(datapath_drop_userspace_action_error);
COVERAGE_DEFINE(datapath_drop_tunnel_push_error);
COVERAGE_DEFINE(datapath_drop_tunnel_pop_error);
-COVERAGE_DEFINE(datapath_drop_tunnel_tso_recirc);
COVERAGE_DEFINE(datapath_drop_recirc_error);
COVERAGE_DEFINE(datapath_drop_invalid_port);
COVERAGE_DEFINE(datapath_drop_invalid_bond);
@@ -8923,32 +8922,6 @@ static void
dp_netdev_recirculate(struct dp_netdev_pmd_thread *pmd,
struct dp_packet_batch *packets)
{
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
- size_t i, size = dp_packet_batch_size(packets);
- struct dp_packet *packet;
-
- DP_PACKET_BATCH_REFILL_FOR_EACH (i, size, packet, packets) {
- if (dp_packet_hwol_is_tunnel(packet)) {
- if (dp_packet_hwol_is_tso(packet)) {
- /* Can't perform GSO in the middle of a pipeline. */
- COVERAGE_INC(datapath_drop_tunnel_tso_recirc);
- dp_packet_delete(packet);
- VLOG_WARN_RL(&rl, "Recirculating tunnel packets with "
- "TSO is not supported");
- continue;
- }
- /* Have to fix all the checksums before re-parsing, because the
- * packet will be treated as having a single set of headers. */
- dp_packet_ol_send_prepare(packet, 0);
- /* This packet must not be marked with anything tunnel-related. */
- dp_packet_hwol_reset_tunnel(packet);
- /* Clear inner offsets. Other ones are collateral, but they will
- * be re-initialized on re-parsing. */
- dp_packet_reset_offsets(packet);
- }
- dp_packet_batch_refill(packets, packet, i);
- }
-
dp_netdev_input__(pmd, packets, true, 0);
}
diff --git a/lib/flow.c b/lib/flow.c
index 0eb34892f2..93ee50f8df 100644
--- a/lib/flow.c
+++ b/lib/flow.c
@@ -804,6 +804,7 @@ miniflow_extract(struct dp_packet *packet, struct miniflow
*dst)
uint8_t nw_frag, nw_tos, nw_ttl, nw_proto;
uint8_t *ct_nw_proto_p = NULL;
ovs_be16 ct_tp_src = 0, ct_tp_dst = 0;
+ bool tunneling;
/* Metadata. */
if (flow_tnl_dst_is_set(&md->tunnel)) {
@@ -857,7 +858,13 @@ miniflow_extract(struct dp_packet *packet, struct miniflow
*dst)
/* Initialize packet's layer pointer and offsets. */
frame = data;
- dp_packet_reset_offsets(packet);
+ tunneling = dp_packet_hwol_is_tunnel(packet);
+ if (tunneling) {
+ /* Preserve inner offsets from previous circulation. */
+ dp_packet_reset_outer_offsets(packet);
+ } else {
+ dp_packet_reset_offsets(packet);
+ }
if (packet_type == htonl(PT_ETH)) {
/* Must have full Ethernet header to proceed. */
@@ -936,9 +943,16 @@ miniflow_extract(struct dp_packet *packet, struct miniflow
*dst)
nw_proto = nh->ip_proto;
nw_frag = ipv4_get_nw_frag(nh);
data_pull(&data, &size, ip_len);
- dp_packet_hwol_set_tx_ipv4(packet);
- if (dp_packet_ip_checksum_good(packet)) {
- dp_packet_hwol_set_tx_ip_csum(packet);
+ if (tunneling) {
+ dp_packet_hwol_set_tx_outer_ipv4(packet);
+ if (dp_packet_ip_checksum_good(packet)) {
+ dp_packet_hwol_set_tx_outer_ipv4_csum(packet);
+ }
+ } else {
+ dp_packet_hwol_set_tx_ipv4(packet);
+ if (dp_packet_ip_checksum_good(packet)) {
+ dp_packet_hwol_set_tx_ip_csum(packet);
+ }
}
} else if (dl_type == htons(ETH_TYPE_IPV6)) {
const struct ovs_16aligned_ip6_hdr *nh = data;
@@ -953,7 +967,11 @@ miniflow_extract(struct dp_packet *packet, struct miniflow
*dst)
}
data_pull(&data, &size, sizeof *nh);
- dp_packet_hwol_set_tx_ipv6(packet);
+ if (tunneling) {
+ dp_packet_hwol_set_tx_outer_ipv6(packet);
+ } else {
+ dp_packet_hwol_set_tx_ipv6(packet);
+ }
plen = ntohs(nh->ip6_plen);
dp_packet_set_l2_pad_size(packet, size - plen);
size = plen; /* Never pull padding. */
@@ -1078,7 +1096,11 @@ miniflow_extract(struct dp_packet *packet, struct
miniflow *dst)
dp_packet_ol_l4_csum_check_partial(packet);
if (dp_packet_l4_checksum_good(packet)
|| dp_packet_ol_l4_csum_partial(packet)) {
- dp_packet_hwol_set_csum_udp(packet);
+ if (tunneling) {
+ dp_packet_hwol_set_outer_udp_csum(packet);
+ } else {
+ dp_packet_hwol_set_csum_udp(packet);
+ }
}
}
} else if (OVS_LIKELY(nw_proto == IPPROTO_SCTP)) {
--
2.47.1
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
