vhost: postpone rings addresses translation
authorMaxime Coquelin <maxime.coquelin@redhat.com>
Thu, 5 Oct 2017 08:36:23 +0000 (10:36 +0200)
committerYuanhan Liu <yliu@fridaylinux.org>
Tue, 10 Oct 2017 13:52:27 +0000 (15:52 +0200)
This patch postpones rings addresses translations and checks, as
addresses sent by the master shuld not be interpreted as long as
ring is not started and enabled[0].

When protocol features aren't negotiated, the ring is started in
enabled state, so the addresses translations are postponed to
vhost_user_set_vring_kick().
Otherwise, it is postponed to when ring is enabled, in
vhost_user_set_vring_enable().

[0]: http://lists.nongnu.org/archive/html/qemu-devel/2017-05/msg04355.html

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
lib/librte_vhost/vhost.h
lib/librte_vhost/vhost_user.c

index 2ea6c46..b873cce 100644 (file)
@@ -125,6 +125,7 @@ struct vhost_virtqueue {
 
        struct vring_used_elem  *shadow_used_ring;
        uint16_t                shadow_used_idx;
+       struct vhost_vring_addr ring_addrs;
 
        struct batch_copy_elem  *batch_copy_elems;
        uint16_t                batch_copy_nb_elems;
index 34086a7..b2159e0 100644 (file)
@@ -356,6 +356,7 @@ static int
 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
 {
        struct vhost_virtqueue *vq;
+       struct vhost_vring_addr *addr = &msg->payload.addr;
 
        if (dev->mem == NULL)
                return -1;
@@ -363,35 +364,50 @@ vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
        /* addr->index refers to the queue index. The txq 1, rxq is 0. */
        vq = dev->virtqueue[msg->payload.addr.index];
 
+       /*
+        * Rings addresses should not be interpreted as long as the ring is not
+        * started and enabled
+        */
+       memcpy(&vq->ring_addrs, addr, sizeof(*addr));
+
+       return 0;
+}
+
+static struct virtio_net *
+translate_ring_addresses(struct virtio_net *dev, int vq_index)
+{
+       struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
+       struct vhost_vring_addr *addr = &vq->ring_addrs;
+
        /* The addresses are converted from QEMU virtual to Vhost virtual. */
        vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
-                       msg->payload.addr.desc_user_addr);
+                       addr->desc_user_addr);
        if (vq->desc == 0) {
                RTE_LOG(ERR, VHOST_CONFIG,
                        "(%d) failed to find desc ring address.\n",
                        dev->vid);
-               return -1;
+               return NULL;
        }
 
-       dev = numa_realloc(dev, msg->payload.addr.index);
-       vq = dev->virtqueue[msg->payload.addr.index];
+       dev = numa_realloc(dev, vq_index);
+       vq = dev->virtqueue[vq_index];
 
        vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
-                       msg->payload.addr.avail_user_addr);
+                       addr->avail_user_addr);
        if (vq->avail == 0) {
                RTE_LOG(ERR, VHOST_CONFIG,
                        "(%d) failed to find avail ring address.\n",
                        dev->vid);
-               return -1;
+               return NULL;
        }
 
        vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
-                       msg->payload.addr.used_user_addr);
+                       addr->used_user_addr);
        if (vq->used == 0) {
                RTE_LOG(ERR, VHOST_CONFIG,
                        "(%d) failed to find used ring address.\n",
                        dev->vid);
-               return -1;
+               return NULL;
        }
 
        if (vq->last_used_idx != vq->used->idx) {
@@ -403,7 +419,7 @@ vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
                vq->last_avail_idx = vq->used->idx;
        }
 
-       vq->log_guest_addr = msg->payload.addr.log_guest_addr;
+       vq->log_guest_addr = addr->log_guest_addr;
 
        LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
                        dev->vid, vq->desc);
@@ -414,7 +430,7 @@ vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
        LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
                        dev->vid, vq->log_guest_addr);
 
-       return 0;
+       return dev;
 }
 
 /*
@@ -685,10 +701,11 @@ vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 }
 
 static void
-vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
+vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 {
        struct vhost_vring_file file;
        struct vhost_virtqueue *vq;
+       struct virtio_net *dev = *pdev;
 
        file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
        if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
@@ -698,6 +715,16 @@ vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
        RTE_LOG(INFO, VHOST_CONFIG,
                "vring kick idx:%d file:%d\n", file.index, file.fd);
 
+       /*
+        * Interpret ring addresses only when ring is started and enabled.
+        * This is now if protocol features aren't supported.
+        */
+       if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
+               *pdev = dev = translate_ring_addresses(dev, file.index);
+               if (!dev)
+                       return;
+       }
+
        vq = dev->virtqueue[file.index];
 
        /*
@@ -778,15 +805,29 @@ vhost_user_get_vring_base(struct virtio_net *dev,
  * enable the virtio queue pair.
  */
 static int
-vhost_user_set_vring_enable(struct virtio_net *dev,
+vhost_user_set_vring_enable(struct virtio_net **pdev,
                            VhostUserMsg *msg)
 {
+       struct virtio_net *dev = *pdev;
        int enable = (int)msg->payload.state.num;
 
        RTE_LOG(INFO, VHOST_CONFIG,
                "set queue enable: %d to qp idx: %d\n",
                enable, msg->payload.state.index);
 
+       /*
+        * Interpret ring addresses only when ring is started and enabled.
+        * This is now if protocol features are supported.
+        */
+       if (enable && (dev->features &
+                               (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
+               dev = translate_ring_addresses(dev, msg->payload.state.index);
+               if (!dev)
+                       return -1;
+
+               *pdev = dev;
+       }
+
        if (dev->notify_ops->vring_state_changed)
                dev->notify_ops->vring_state_changed(dev->vid,
                                msg->payload.state.index, enable);
@@ -1155,7 +1196,7 @@ vhost_user_msg_handler(int vid, int fd)
                break;
 
        case VHOST_USER_SET_VRING_KICK:
-               vhost_user_set_vring_kick(dev, &msg);
+               vhost_user_set_vring_kick(&dev, &msg);
                break;
        case VHOST_USER_SET_VRING_CALL:
                vhost_user_set_vring_call(dev, &msg);
@@ -1174,7 +1215,7 @@ vhost_user_msg_handler(int vid, int fd)
                break;
 
        case VHOST_USER_SET_VRING_ENABLE:
-               vhost_user_set_vring_enable(dev, &msg);
+               vhost_user_set_vring_enable(&dev, &msg);
                break;
        case VHOST_USER_SEND_RARP:
                vhost_user_send_rarp(dev, &msg);