net/virtio: fix mbuf data and packet length mismatch
authorMarvin Liu <yong.liu@intel.com>
Mon, 23 Sep 2019 14:05:11 +0000 (22:05 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 7 Oct 2019 13:00:57 +0000 (15:00 +0200)
If reserve virtio header room by function rte_pktmbuf_prepend, both
segment data length and packet length of mbuf will be increased.
Data length will be equal to descriptor length, while packet length
should be decreased as virtio-net header won't be taken into packet.
Thus will cause mismatch in mbuf structure. Fix this issue by access
mbuf data directly and increase descriptor length if it is needed.

Fixes: 58169a9c8153 ("net/virtio: support Tx checksum offload")
Fixes: 892dc798fa9c ("net/virtio: implement Tx path for packed queues")
Fixes: 4905ed3a523f ("net/virtio: optimize Tx enqueue for packed ring")
Fixes: e5f456a98d3c ("net/virtio: support in-order Rx and Tx")
Cc: stable@dpdk.org
Reported-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Marvin Liu <yong.liu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
drivers/net/virtio/virtio_rxtx.c

index 42f2beb..929aa4c 100644 (file)
@@ -640,9 +640,8 @@ virtqueue_enqueue_xmit_inorder(struct virtnet_tx *txvq,
                dxp->ndescs = 1;
                virtio_update_packet_stats(&txvq->stats, cookies[i]);
 
-               hdr = (struct virtio_net_hdr *)
-                       rte_pktmbuf_prepend(cookies[i], head_size);
-               cookies[i]->pkt_len -= head_size;
+               hdr = (struct virtio_net_hdr *)(char *)cookies[i]->buf_addr +
+                       cookies[i]->data_off - head_size;
 
                /* if offload disabled, hdr is not zeroed yet, do it now */
                if (!vq->hw->has_tx_offload)
@@ -651,9 +650,10 @@ virtqueue_enqueue_xmit_inorder(struct virtnet_tx *txvq,
                        virtqueue_xmit_offload(hdr, cookies[i], true);
 
                start_dp[idx].addr  = VIRTIO_MBUF_DATA_DMA_ADDR(cookies[i], vq);
-               start_dp[idx].len   = cookies[i]->data_len;
+               start_dp[idx].len   = cookies[i]->data_len + head_size;
                start_dp[idx].flags = 0;
 
+
                vq_update_avail_ring(vq, idx);
 
                idx++;
@@ -687,9 +687,8 @@ virtqueue_enqueue_xmit_packed_fast(struct virtnet_tx *txvq,
        flags = vq->vq_packed.cached_flags;
 
        /* prepend cannot fail, checked by caller */
-       hdr = (struct virtio_net_hdr *)
-               rte_pktmbuf_prepend(cookie, head_size);
-       cookie->pkt_len -= head_size;
+       hdr = (struct virtio_net_hdr *)(char *)cookie->buf_addr +
+               cookie->data_off - head_size;
 
        /* if offload disabled, hdr is not zeroed yet, do it now */
        if (!vq->hw->has_tx_offload)
@@ -698,7 +697,7 @@ virtqueue_enqueue_xmit_packed_fast(struct virtnet_tx *txvq,
                virtqueue_xmit_offload(hdr, cookie, true);
 
        dp->addr = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq);
-       dp->len  = cookie->data_len;
+       dp->len  = cookie->data_len + head_size;
        dp->id   = id;
 
        if (++vq->vq_avail_idx >= vq->vq_nentries) {
@@ -730,6 +729,7 @@ virtqueue_enqueue_xmit_packed(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
        uint16_t head_size = vq->hw->vtnet_hdr_size;
        struct virtio_net_hdr *hdr;
        uint16_t prev;
+       bool prepend_header = false;
 
        id = in_order ? vq->vq_avail_idx : vq->vq_desc_head_idx;
 
@@ -748,12 +748,9 @@ virtqueue_enqueue_xmit_packed(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
 
        if (can_push) {
                /* prepend cannot fail, checked by caller */
-               hdr = (struct virtio_net_hdr *)
-                       rte_pktmbuf_prepend(cookie, head_size);
-               /* rte_pktmbuf_prepend() counts the hdr size to the pkt length,
-                * which is wrong. Below subtract restores correct pkt size.
-                */
-               cookie->pkt_len -= head_size;
+               hdr = (struct virtio_net_hdr *)(char *)cookie->buf_addr +
+                       cookie->data_off - head_size;
+               prepend_header = true;
 
                /* if offload disabled, it is not zeroed below, do it now */
                if (!vq->hw->has_tx_offload)
@@ -781,6 +778,11 @@ virtqueue_enqueue_xmit_packed(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
 
                start_dp[idx].addr = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq);
                start_dp[idx].len  = cookie->data_len;
+               if (prepend_header) {
+                       start_dp[idx].len += head_size;
+                       prepend_header = false;
+               }
+
                if (likely(idx != head_idx)) {
                        flags = cookie->next ? VRING_DESC_F_NEXT : 0;
                        flags |= vq->vq_packed.cached_flags;
@@ -822,6 +824,7 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
        uint16_t seg_num = cookie->nb_segs;
        uint16_t head_idx, idx;
        uint16_t head_size = vq->hw->vtnet_hdr_size;
+       bool prepend_header = false;
        struct virtio_net_hdr *hdr;
 
        head_idx = vq->vq_desc_head_idx;
@@ -837,12 +840,9 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
 
        if (can_push) {
                /* prepend cannot fail, checked by caller */
-               hdr = (struct virtio_net_hdr *)
-                       rte_pktmbuf_prepend(cookie, head_size);
-               /* rte_pktmbuf_prepend() counts the hdr size to the pkt length,
-                * which is wrong. Below subtract restores correct pkt size.
-                */
-               cookie->pkt_len -= head_size;
+               hdr = (struct virtio_net_hdr *)(char *)cookie->buf_addr +
+                       cookie->data_off - head_size;
+               prepend_header = true;
 
                /* if offload disabled, it is not zeroed below, do it now */
                if (!vq->hw->has_tx_offload)
@@ -881,6 +881,10 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
        do {
                start_dp[idx].addr  = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq);
                start_dp[idx].len   = cookie->data_len;
+               if (prepend_header) {
+                       start_dp[idx].len += head_size;
+                       prepend_header = false;
+               }
                start_dp[idx].flags = cookie->next ? VRING_DESC_F_NEXT : 0;
                idx = start_dp[idx].next;
        } while ((cookie = cookie->next) != NULL);