netdev-offload-dpdk: Lock rte_flow map access.

Add a lock to access the ufid to rte_flow map.  This will protect it
from concurrent write accesses when multiple threads attempt it.

At this point, the reason for taking the lock is not to fullfill the
needs of the DPDK offload implementation anymore. Rewrite the comments
to reflect this change. The lock is still needed to protect against
changes to netdev port mapping.

Signed-off-by: Gaetan Rivet <grive@u256.net>
Reviewed-by: Eli Britstein <elibr@nvidia.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>

2 files changed, 61 insertions, 8 deletions

diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 7e93e9250..8eef3fda6 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -2557,7 +2557,7 @@ mark_to_flow_disassociate(struct dp_netdev_pmd_thread *pmd,
         port = netdev_ports_get(in_port, dpif_type_str);
         if (port) {
             /* Taking a global 'port_mutex' to fulfill thread safety
-             * restrictions for the netdev-offload-dpdk module. */
+             * restrictions regarding netdev port mapping. */
             ovs_mutex_lock(&pmd->dp->port_mutex);
             ret = netdev_flow_del(port, &flow->mega_ufid, NULL);
             ovs_mutex_unlock(&pmd->dp->port_mutex);
@@ -2726,8 +2726,8 @@ dp_netdev_flow_offload_put(struct dp_offload_flow_item *offload)
         goto err_free;
     }
 
-    /* Taking a global 'port_mutex' to fulfill thread safety restrictions for
-     * the netdev-offload-dpdk module. */
+    /* Taking a global 'port_mutex' to fulfill thread safety
+     * restrictions regarding the netdev port mapping. */
     ovs_mutex_lock(&pmd->dp->port_mutex);
     ret = netdev_flow_put(port, &offload->match,
                           CONST_CAST(struct nlattr *, offload->actions),
@@ -3566,7 +3566,7 @@ dpif_netdev_get_flow_offload_status(const struct dp_netdev *dp,
     }
     ofpbuf_use_stack(&buf, &act_buf, sizeof act_buf);
     /* Taking a global 'port_mutex' to fulfill thread safety
-     * restrictions for the netdev-offload-dpdk module.
+     * restrictions regarding netdev port mapping.
      *
      * XXX: Main thread will try to pause/stop all revalidators during datapath
      *      reconfiguration via datapath purge callback (dp_purge_cb) while
diff --git a/lib/netdev-offload-dpdk.c b/lib/netdev-offload-dpdk.c
index 1b92063c0..7a3d6b3de 100644
--- a/lib/netdev-offload-dpdk.c
+++ b/lib/netdev-offload-dpdk.c
@@ -41,9 +41,6 @@ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(600, 600);
  *
  * Below API is NOT thread safe in following terms:
  *
- *  - The caller must be sure that none of these functions will be called
- *    simultaneously.  Even for different 'netdev's.
- *
  *  - The caller must be sure that 'netdev' will not be destructed/deallocated.
  *
  *  - The caller must be sure that 'netdev' configuration will not be changed.
@@ -70,6 +67,7 @@ struct ufid_to_rte_flow_data {
 struct netdev_offload_dpdk_data {
     struct cmap ufid_to_rte_flow;
     uint64_t *rte_flow_counters;
+    struct ovs_mutex map_lock;
 };
 
 static int
@@ -78,6 +76,7 @@ offload_data_init(struct netdev *netdev)
     struct netdev_offload_dpdk_data *data;
 
     data = xzalloc(sizeof *data);
+    ovs_mutex_init(&data->map_lock);
     cmap_init(&data->ufid_to_rte_flow);
     data->rte_flow_counters = xcalloc(netdev_offload_thread_nb(),
                                       sizeof *data->rte_flow_counters);
@@ -90,6 +89,7 @@ offload_data_init(struct netdev *netdev)
 static void
 offload_data_destroy__(struct netdev_offload_dpdk_data *data)
 {
+    ovs_mutex_destroy(&data->map_lock);
     free(data->rte_flow_counters);
     free(data);
 }
@@ -121,6 +121,34 @@ offload_data_destroy(struct netdev *netdev)
     ovsrcu_set(&netdev->hw_info.offload_data, NULL);
 }
 
+static void
+offload_data_lock(struct netdev *netdev)
+    OVS_NO_THREAD_SAFETY_ANALYSIS
+{
+    struct netdev_offload_dpdk_data *data;
+
+    data = (struct netdev_offload_dpdk_data *)
+        ovsrcu_get(void *, &netdev->hw_info.offload_data);
+    if (!data) {
+        return;
+    }
+    ovs_mutex_lock(&data->map_lock);
+}
+
+static void
+offload_data_unlock(struct netdev *netdev)
+    OVS_NO_THREAD_SAFETY_ANALYSIS
+{
+    struct netdev_offload_dpdk_data *data;
+
+    data = (struct netdev_offload_dpdk_data *)
+        ovsrcu_get(void *, &netdev->hw_info.offload_data);
+    if (!data) {
+        return;
+    }
+    ovs_mutex_unlock(&data->map_lock);
+}
+
 static struct cmap *
 offload_data_map(struct netdev *netdev)
 {
@@ -159,6 +187,24 @@ ufid_to_rte_flow_data_find(struct netdev *netdev,
     return NULL;
 }
 
+/* Find rte_flow with @ufid, lock-protected. */
+static struct ufid_to_rte_flow_data *
+ufid_to_rte_flow_data_find_protected(struct netdev *netdev,
+                                     const ovs_u128 *ufid)
+{
+    size_t hash = hash_bytes(ufid, sizeof *ufid, 0);
+    struct ufid_to_rte_flow_data *data;
+    struct cmap *map = offload_data_map(netdev);
+
+    CMAP_FOR_EACH_WITH_HASH_PROTECTED (data, node, hash, map) {
+        if (ovs_u128_equals(*ufid, data->ufid)) {
+            return data;
+        }
+    }
+
+    return NULL;
+}
+
 static inline struct ufid_to_rte_flow_data *
 ufid_to_rte_flow_associate(const ovs_u128 *ufid, struct netdev *netdev,
                            struct netdev *physdev, struct rte_flow *rte_flow,
@@ -175,13 +221,15 @@ ufid_to_rte_flow_associate(const ovs_u128 *ufid, struct netdev *netdev,
 
     data = xzalloc(sizeof *data);
 
+    offload_data_lock(netdev);
+
     /*
      * We should not simply overwrite an existing rte flow.
      * We should have deleted it first before re-adding it.
      * Thus, if following assert triggers, something is wrong:
      * the rte_flow is not destroyed.
      */
-    data_prev = ufid_to_rte_flow_data_find(netdev, ufid, false);
+    data_prev = ufid_to_rte_flow_data_find_protected(netdev, ufid);
     if (data_prev) {
         ovs_assert(data_prev->rte_flow == NULL);
     }
@@ -193,6 +241,8 @@ ufid_to_rte_flow_associate(const ovs_u128 *ufid, struct netdev *netdev,
     data->actions_offloaded = actions_offloaded;
 
     cmap_insert(map, CONST_CAST(struct cmap_node *, &data->node), hash);
+
+    offload_data_unlock(netdev);
     return data;
 }
 
@@ -206,7 +256,10 @@ ufid_to_rte_flow_disassociate(struct ufid_to_rte_flow_data *data)
         return;
     }
 
+    offload_data_lock(data->netdev);
     cmap_remove(map, CONST_CAST(struct cmap_node *, &data->node), hash);
+    offload_data_unlock(data->netdev);
+
     if (data->netdev != data->physdev) {
         netdev_close(data->netdev);
     }