[PATCH 2/3] net/bonding: prevent crash on Rx/Tx from secondary process

[Stephen Hemminger]

The bonding PMD's secondary process attach path registered the
ethdev but never installed rx_pkt_burst or tx_pkt_burst, leaving
both as NULL. Any rx_burst or tx_burst call from a secondary
process therefore crashed with a NULL pointer dereference.

Fully sharing bonding state across processes would be
overly complex. Instead, install blackhole burst functions
in the secondary so the data path is safe by default. Rx returns 0
and Tx frees the mbufs and reports them as transmitted,
matching /dev/null semantics so applications do not spin
retrying. Each stub logs once at NOTICE level on first use.

Also reject bond mode changes from a secondary process.
rte_eth_bond_mode_set() is callable from secondary, and
without this guard it would overwrite the secondary's safe
burst stubs with real per-mode functions whose internal
state is not valid outside the primary, reintroducing the
crash by another path.

This keeps secondary support available for the more
common use cases of procinfo and packet capture.

Bugzilla ID: 1698
Fixes: 4852aa8f6e21 ("drivers/net: enable hotplug on secondary process")
Cc: stable at dpdk.org

Signed-off-by: Stephen Hemminger <stephen at networkplumber.org>
---
 drivers/net/bonding/rte_eth_bond_pmd.c | 43 +++++++++++++++++++++++++-
 1 file changed, 42 insertions(+), 1 deletion(-)

diff --git a/drivers/net/bonding/rte_eth_bond_pmd.c b/drivers/net/bonding/rte_eth_bond_pmd.c
index 96725071da..6a42257d2b 100644
--- a/drivers/net/bonding/rte_eth_bond_pmd.c
+++ b/drivers/net/bonding/rte_eth_bond_pmd.c
@@ -56,6 +56,33 @@ get_vlan_offset(struct rte_ether_hdr *eth_hdr, uint16_t *proto)
 	return vlan_offset;
 }
 
+static uint16_t
+bond_ethdev_rx_secondary(void *queue __rte_unused,
+			 struct rte_mbuf **bufs __rte_unused, uint16_t nb_pkts __rte_unused)
+{
+	static bool once = true;
+
+	if (once) {
+		/* once per process is enough of a notice */
+		RTE_BOND_LOG(NOTICE, "receive not supported in secondary");
+		once = false;
+	}
+	return 0;
+}
+
+static uint16_t
+bond_ethdev_tx_secondary(void *queue __rte_unused, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+	static bool once = true;
+
+	if (once) {
+		RTE_BOND_LOG(NOTICE, "transmit not supported in secondary");
+		once = false;
+	}
+	rte_pktmbuf_free_bulk(bufs, nb_pkts);
+	return nb_pkts;
+}
+
 static uint16_t
 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 {
@@ -1599,6 +1626,11 @@ bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, uint8_t mode)
 {
 	struct bond_dev_private *internals;
 
+	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
+		RTE_BOND_LOG(ERR, "Setting mode in secondary not allowed");
+		return -1;
+	}
+
 	internals = eth_dev->data->dev_private;
 
 	switch (mode) {
@@ -3799,9 +3831,18 @@ bond_probe(struct rte_vdev_device *dev)
 			RTE_BOND_LOG(ERR, "Failed to probe %s", name);
 			return -1;
 		}
-		/* TODO: request info from primary to set up Rx and Tx */
+
 		eth_dev->dev_ops = &default_dev_ops;
 		eth_dev->device = &dev->device;
+
+		/*
+		 * Propagation of bond mode would require adding
+		 * MP client/server support and lots of error handling.
+		 *
+		 * For now just install a black hole.
+		 */
+		eth_dev->tx_pkt_burst = bond_ethdev_tx_secondary;
+		eth_dev->rx_pkt_burst = bond_ethdev_rx_secondary;
 		rte_eth_dev_probing_finish(eth_dev);
 		return 0;
 	}
-- 
2.53.0