net/virtio: fix packet corruption
authorOlivier Matz <olivier.matz@6wind.com>
Tue, 19 Jul 2016 12:31:59 +0000 (14:31 +0200)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Thu, 21 Jul 2016 22:27:29 +0000 (00:27 +0200)
The support of virtio-user changed the way the mbuf dma address is
retrieved, using a physical address in case of virtio-pci and a virtual
address in case of virtio-user.

This change introduced some possible memory corruption in packets,
replacing:
  m->buf_physaddr + RTE_PKTMBUF_HEADROOM
by:
  m->buf_physaddr + m->data_off     (through a macro)

This patch fixes this issue, restoring the original behavior.

By the way, it also rework the macros, adding a "VIRTIO_" prefix and
API comments.

Fixes: f24f8f9fee8a ("net/virtio: allow virtual address to fill vring descriptors")

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
drivers/net/virtio/virtio_ethdev.c
drivers/net/virtio/virtio_rxtx.c
drivers/net/virtio/virtio_rxtx_simple.c
drivers/net/virtio/virtqueue.h

index 850e3ba..fcc996e 100644 (file)
@@ -454,7 +454,7 @@ int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 
        /* For virtio-user case (that is when dev->pci_dev is NULL), we use
         * virtual address. And we need properly set _offset_, please see
-        * MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
+        * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
         */
        if (dev->pci_dev)
                vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
index a27208e..9aba044 100644 (file)
@@ -193,7 +193,8 @@ virtqueue_enqueue_recv_refill(struct virtqueue *vq, struct rte_mbuf *cookie)
 
        start_dp = vq->vq_ring.desc;
        start_dp[idx].addr =
-               MBUF_DATA_DMA_ADDR(cookie, vq->offset) - hw->vtnet_hdr_size;
+               VIRTIO_MBUF_ADDR(cookie, vq) +
+               RTE_PKTMBUF_HEADROOM - hw->vtnet_hdr_size;
        start_dp[idx].len =
                cookie->buf_len - RTE_PKTMBUF_HEADROOM + hw->vtnet_hdr_size;
        start_dp[idx].flags =  VRING_DESC_F_WRITE;
@@ -265,7 +266,7 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
        }
 
        do {
-               start_dp[idx].addr  = MBUF_DATA_DMA_ADDR(cookie, vq->offset);
+               start_dp[idx].addr  = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq);
                start_dp[idx].len   = cookie->data_len;
                start_dp[idx].flags = cookie->next ? VRING_DESC_F_NEXT : 0;
                idx = start_dp[idx].next;
index d8fcc15..6517aa8 100644 (file)
@@ -80,8 +80,9 @@ virtqueue_enqueue_recv_refill_simple(struct virtqueue *vq,
        vq->sw_ring[desc_idx] = cookie;
 
        start_dp = vq->vq_ring.desc;
-       start_dp[desc_idx].addr = MBUF_DATA_DMA_ADDR(cookie, vq->offset) -
-                                 vq->hw->vtnet_hdr_size;
+       start_dp[desc_idx].addr =
+               VIRTIO_MBUF_ADDR(cookie, vq) +
+               RTE_PKTMBUF_HEADROOM - vq->hw->vtnet_hdr_size;
        start_dp[desc_idx].len = cookie->buf_len -
                RTE_PKTMBUF_HEADROOM + vq->hw->vtnet_hdr_size;
 
@@ -120,8 +121,8 @@ virtio_rxq_rearm_vec(struct virtnet_rx *rxvq)
                *(uint64_t *)p = rxvq->mbuf_initializer;
 
                start_dp[i].addr =
-                       MBUF_DATA_DMA_ADDR(sw_ring[i], vq->offset) -
-                       vq->hw->vtnet_hdr_size;
+                       VIRTIO_MBUF_ADDR(sw_ring[i], vq) +
+                       RTE_PKTMBUF_HEADROOM - vq->hw->vtnet_hdr_size;
                start_dp[i].len = sw_ring[i]->buf_len -
                        RTE_PKTMBUF_HEADROOM + vq->hw->vtnet_hdr_size;
        }
@@ -371,7 +372,7 @@ virtio_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
                        vq->vq_descx[desc_idx + i].cookie = tx_pkts[i];
                for (i = 0; i < nb_tail; i++) {
                        start_dp[desc_idx].addr =
-                               MBUF_DATA_DMA_ADDR(*tx_pkts, vq->offset);
+                               VIRTIO_MBUF_DATA_DMA_ADDR(*tx_pkts, vq);
                        start_dp[desc_idx].len = (*tx_pkts)->pkt_len;
                        tx_pkts++;
                        desc_idx++;
@@ -383,7 +384,7 @@ virtio_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
                vq->vq_descx[desc_idx + i].cookie = tx_pkts[i];
        for (i = 0; i < nb_commit; i++) {
                start_dp[desc_idx].addr =
-                       MBUF_DATA_DMA_ADDR(*tx_pkts, vq->offset);
+                       VIRTIO_MBUF_DATA_DMA_ADDR(*tx_pkts, vq);
                start_dp[desc_idx].len = (*tx_pkts)->pkt_len;
                tx_pkts++;
                desc_idx++;
index 455aaaf..c452d04 100644 (file)
@@ -67,12 +67,21 @@ struct rte_mbuf;
 #define VIRTQUEUE_MAX_NAME_SZ 32
 
 #ifdef RTE_VIRTIO_USER
-#define MBUF_DATA_DMA_ADDR(mb, offset) \
-       ((uint64_t)((uintptr_t)(*(void **)((uintptr_t)mb + offset)) \
-                       + (mb)->data_off))
-#else /* RTE_VIRTIO_USER */
-#define MBUF_DATA_DMA_ADDR(mb, offset) rte_mbuf_data_dma_addr(mb)
-#endif /* RTE_VIRTIO_USER */
+/**
+ * Return the physical address (or virtual address in case of
+ * virtio-user) of mbuf data buffer.
+ */
+#define VIRTIO_MBUF_ADDR(mb, vq) (*(uint64_t *)((uintptr_t)(mb) + (vq)->offset))
+#else
+#define VIRTIO_MBUF_ADDR(mb, vq) ((mb)->buf_physaddr)
+#endif
+
+/**
+ * Return the physical address (or virtual address in case of
+ * virtio-user) of mbuf data buffer, taking care of mbuf data offset
+ */
+#define VIRTIO_MBUF_DATA_DMA_ADDR(mb, vq) \
+       (VIRTIO_MBUF_ADDR(mb, vq) + (mb)->data_off)
 
 #define VTNET_SQ_RQ_QUEUE_IDX 0
 #define VTNET_SQ_TQ_QUEUE_IDX 1
@@ -182,8 +191,8 @@ struct virtqueue {
        void *vq_ring_virt_mem;  /**< linear address of vring*/
        unsigned int vq_ring_size;
 
-       phys_addr_t vq_ring_mem; /**< physical address of vring */
-                               /**< use virtual address for virtio-user. */
+       phys_addr_t vq_ring_mem; /**< physical address of vring,
+                                 * or virtual address for virtio-user. */
 
        /**
         * Head of the free chain in the descriptor table. If