The pcap PMD assumed that every error from pcap_sendpacket()
was from back pressure. But in the case of bad mbuf or file system
full or other hard errors, the packet needs to be dropped.
Bugzilla ID: 2132
Fixes: cf91cce204ea ("net/pcap: fix error accounting and backpressure on Tx")
Reported-by: Anthropic OSS scanner (ANT-2026-DLCI8IK6)
Signed-off-by: Stephen Hemminger <[email protected]>
---
drivers/net/pcap/pcap_ethdev.c | 20 +++++++++-----------
1 file changed, 9 insertions(+), 11 deletions(-)
diff --git a/drivers/net/pcap/pcap_ethdev.c b/drivers/net/pcap/pcap_ethdev.c
index fb18f0932b..e7a8f007e0 100644
--- a/drivers/net/pcap/pcap_ethdev.c
+++ b/drivers/net/pcap/pcap_ethdev.c
@@ -607,16 +607,6 @@ eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t
nb_pkts)
/*
* Send a burst of packets to a pcap device.
- *
- * On Linux, pcap_sendpacket() calls send() on a blocking PF_PACKET
- * socket with default kernel buffer sizes and no TX ring (PACKET_TX_RING).
- * The send() call only blocks when the kernel socket send buffer is full,
- * providing limited backpressure.
- *
- * On error, pcap_sendpacket() returns non-zero and the loop breaks,
- * leaving remaining packets unsent.
- *
- * Bottom line: backpressure is not an error.
*/
static uint16_t
eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
@@ -666,10 +656,18 @@ eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t
nb_pkts)
data = rte_pktmbuf_read(mbuf, 0, len, temp_data);
RTE_ASSERT(data != NULL);
+ errno = 0;
if (unlikely(pcap_sendpacket(pcap, data, len) != 0)) {
+ /* retry on soft errors */
+ if (errno == EINTR || errno == EAGAIN ||
+ errno == EWOULDBLOCK)
+ break;
+
/* Assume failure is backpressure */
PMD_LOG(ERR, "pcap_sendpacket() failed: %s",
pcap_geterr(pcap));
- break;
+ rte_pktmbuf_free(mbuf);
+ tx_queue->tx_stat.err_pkts++;
+ continue;
}
num_tx++;
tx_bytes += len;
--
2.53.0