xdp: handle "bad" messages in knot_xsk_sendmmsg()

diff --git a/src/libknot/xdp/af_xdp.c b/src/libknot/xdp/af_xdp.c
index 583238ac51ee943bfd0de538cc4451add17787c9..395cd8b234f1a2a401f2b597ed366244c4a50d60 100644 (file)
--- a/src/libknot/xdp/af_xdp.c
+++ b/src/libknot/xdp/af_xdp.c
@@ -377,13 +377,12 @@ static void xsk_sendmsg_ipv6(struct knot_xsk_socket *socket, const knot_xsk_msg_
        };
 }
 
-static void xsk_sendmsg_error(struct knot_xsk_socket *socket, const knot_xsk_msg_t *msg,
-                              uint32_t index)
+static void tx_free_relative(struct xsk_umem_info *umem, uint64_t addr_relative)
 {
-       *xsk_ring_prod__tx_desc(&socket->tx, index) = (struct xdp_desc){
-               .addr = msg,
-               .len = 0,
-       };
+       /* The address may not point to *start* of buffer, but `/` solves that. */
+       uint64_t index = addr_relative / FRAME_SIZE;
+       assert(index < UMEM_FRAME_COUNT);
+       umem->tx_free_indices[umem->tx_free_count++] = index;
 }
 
 _public_
@@ -393,30 +392,29 @@ int knot_xsk_sendmmsg(struct knot_xsk_socket *socket, const knot_xsk_msg_t msgs[
                return KNOT_EINVAL;
        }
 
-       uint32_t idx = 0;
-       const uint32_t reserved = xsk_ring_prod__reserve(&socket->tx, count, &idx);
+       // FIXME: explain why we do this by hand!
+       uint32_t idx = socket->tx.cached_prod;
 
-       for (uint32_t i = 0; i < reserved; ++i) {
+       for (uint32_t i = 0; i < count; ++i) {
                const knot_xsk_msg_t *msg = &msgs[i];
 
-               switch (msg->ip_from.ss_family) {
-               case AF_INET:
-                       xsk_sendmsg_ipv4(socket, msg, idx);
-                       break;
-               case AF_INET6:
-                       xsk_sendmsg_ipv6(socket, msg, idx);
-                       break;
-               default:
-                       xsk_sendmsg_error(socket, msg, idx);
-                       break;
+               if (msg->payload.iov_len && msg->ip_from.ss_family == AF_INET) {
+                       xsk_sendmsg_ipv4(socket, msg, idx++);
+               } else if (msg->payload.iov_len && msg->ip_from.ss_family == AF_INET6) {
+                       xsk_sendmsg_ipv6(socket, msg, idx++);
+               } else {
+                       /* Some problem; we just ignore this message. */
+                       uint64_t addr_relative = (uint8_t *)msg->payload.iov_base
+                                               - socket->umem->frames->bytes;
+                       tx_free_relative(socket->umem, addr_relative);
                }
-
-               idx++;
        }
 
-       xsk_ring_prod__submit(&socket->tx, reserved);
+       *sent = idx - socket->tx.cached_prod;
+       assert(*sent <= count);
+       socket->tx.cached_prod = idx;
+       xsk_ring_prod__submit(&socket->tx, *sent);
        socket->kernel_needs_wakeup = true;
-       *sent = reserved;
 
        return KNOT_EOK;
 }
@@ -432,10 +430,7 @@ void knot_xsk_prepare_alloc(struct knot_xsk_socket *socket)
        assert(umem->tx_free_count + completed <= UMEM_FRAME_COUNT_TX);
        for (int i = 0; i < completed; ++i, ++idx_cq) {
                uint64_t addr_relative = *xsk_ring_cons__comp_addr(cq, idx_cq);
-               /* The address may not point to *start* of buffer, but `/` solves that. */
-               uint64_t index = addr_relative / FRAME_SIZE;
-               assert(index < UMEM_FRAME_COUNT);
-               umem->tx_free_indices[umem->tx_free_count++] = index;
+               tx_free_relative(umem, addr_relative);
        }
        xsk_ring_cons__release(cq, completed);
 }