examples/ip_fragmentation: fix fail to send un-fragmented packets
Checks
Commit Message
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 HWi/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.
Another small fix - don't set IPv4/IPV6 ether type for unknown
packet types.
Fixes: 16863bbb4a41 ("examples/ip_fragmentation: enable IP checksum offload")
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
examples/ip_fragmentation/main.c | 29 +++++++++++++++--------------
1 file changed, 15 insertions(+), 14 deletions(-)
Comments
10/07/2019 16:16, Konstantin Ananyev:
> 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 HWi/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.
> Another small fix - don't set IPv4/IPV6 ether type for unknown
> packet types.
>
> Fixes: 16863bbb4a41 ("examples/ip_fragmentation: enable IP checksum offload")
Shouldn't it be 2 separate fixes?
@@ -242,16 +242,21 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
{
struct rx_queue *rxq;
uint32_t i, len, next_hop;
- uint8_t ipv6;
- uint16_t port_out;
+ uint16_t port_out, ether_type;
int32_t len2;
+ uint64_t ol_flags;
+ const struct rte_ether_hdr *eth;
- ipv6 = 0;
+ ol_flags = 0;
rxq = &qconf->rx_queue_list[queueid];
/* by default, send everything back to the source port */
port_out = port_in;
+ /* save ether type of the incoming packet */
+ eth = rte_pktmbuf_mtod(m, const struct rte_ether_hdr *);
+ ether_type = eth->ether_type;
+
/* Remove the Ethernet header and trailer from the input packet */
rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
@@ -289,6 +294,9 @@ 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;
@@ -297,8 +305,6 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
/* if this is an IPv6 packet */
struct rte_ipv6_hdr *ip_hdr;
- ipv6 = 1;
-
/* Read the lookup key (i.e. ip_dst) from the input packet */
ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv6_hdr *);
@@ -348,23 +354,18 @@ 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 = ð_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],
ð_hdr->s_addr);
- if (ipv6) {
- eth_hdr->ether_type =
- rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6);
- } else {
- eth_hdr->ether_type =
- rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
- m->ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
- }
+ eth_hdr->ether_type = ether_type;
}
len += len2;