examples/ip_frag: fix Tx un-fragmented packets
authorKonstantin Ananyev <konstantin.ananyev@intel.com>
Thu, 18 Jul 2019 10:11:12 +0000 (11:11 +0100)
committerThomas Monjalon <thomas@monjalon.net>
Thu, 18 Jul 2019 21:05:36 +0000 (23:05 +0200)
With latest changes l3fwd_simple_forward() blindly set
(PKT_TX_IPV4 | PKT_TX_IP_CKSUM) ol_flags for all IPv4 packets.
Though for un-fragmented packets we also do have to set l3_len
to make HW IP cksum offload to work properly.
That causes HW/PMD to drop or generate invalid packets.
Though for un-fragmented packets we don't need to regenerate
IPv4 cksum, as L3 header is not modified.
Fix by setting ol_flags only when required.

Fixes: 16863bbb4a41 ("examples/ip_fragmentation: enable IP checksum offload")

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
examples/ip_fragmentation/main.c

index 1b51bf2..d03e93c 100644 (file)
@@ -245,8 +245,10 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
        uint8_t ipv6;
        uint16_t port_out;
        int32_t len2;
+       uint64_t ol_flags;
 
        ipv6 = 0;
+       ol_flags = 0;
        rxq = &qconf->rx_queue_list[queueid];
 
        /* by default, send everything back to the source port */
@@ -289,6 +291,9 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
                        /* Free input packet */
                        rte_pktmbuf_free(m);
 
+                       /* request HW to regenerate IPv4 cksum */
+                       ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
+
                        /* If we fail to fragment the packet */
                        if (unlikely (len2 < 0))
                                return;
@@ -348,11 +353,13 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
                        rte_panic("No headroom in mbuf.\n");
                }
 
+               m->ol_flags |= ol_flags;
                m->l2_len = sizeof(struct rte_ether_hdr);
 
                /* 02:00:00:00:00:xx */
                d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
-               *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
+               *((uint64_t *)d_addr_bytes) = 0x000000000002 +
+                       ((uint64_t)port_out << 40);
 
                /* src addr */
                rte_ether_addr_copy(&ports_eth_addr[port_out],
@@ -363,7 +370,6 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
                } else {
                        eth_hdr->ether_type =
                                rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
-                       m->ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
                }
        }