macvtap: add GSO/csum offload support
authorArnd Bergmann <arnd@arndb.de>
Thu, 18 Feb 2010 05:48:17 +0000 (05:48 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 18 Feb 2010 22:08:38 +0000 (14:08 -0800)
Added flags field to macvtap_queue to enable/disable processing of
virtio_net_hdr via IFF_VNET_HDR. This flag is checked to prepend virtio_net_hdr
in the receive path and process/skip virtio_net_hdr in the send path.

Original patch by Sridhar, further changes by Arnd.

Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/macvtap.c

index e354501..55ceae0 100644 (file)
@@ -17,6 +17,7 @@
 #include <net/net_namespace.h>
 #include <net/rtnetlink.h>
 #include <net/sock.h>
+#include <linux/virtio_net.h>
 
 /*
  * A macvtap queue is the central object of this driver, it connects
@@ -37,6 +38,7 @@ struct macvtap_queue {
        struct socket sock;
        struct macvlan_dev *vlan;
        struct file *file;
+       unsigned int flags;
 };
 
 static struct proto macvtap_proto = {
@@ -276,6 +278,7 @@ static int macvtap_open(struct inode *inode, struct file *file)
        q->sock.ops = &macvtap_socket_ops;
        sock_init_data(&q->sock, &q->sk);
        q->sk.sk_write_space = macvtap_sock_write_space;
+       q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 
        err = macvtap_set_queue(dev, file, q);
        if (err)
@@ -318,6 +321,111 @@ out:
        return mask;
 }
 
+static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
+                                               size_t len, size_t linear,
+                                               int noblock, int *err)
+{
+       struct sk_buff *skb;
+
+       /* Under a page?  Don't bother with paged skb. */
+       if (prepad + len < PAGE_SIZE || !linear)
+               linear = len;
+
+       skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
+                                  err);
+       if (!skb)
+               return NULL;
+
+       skb_reserve(skb, prepad);
+       skb_put(skb, linear);
+       skb->data_len = len - linear;
+       skb->len += len - linear;
+
+       return skb;
+}
+
+/*
+ * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
+ * be shared with the tun/tap driver.
+ */
+static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
+                                    struct virtio_net_hdr *vnet_hdr)
+{
+       unsigned short gso_type = 0;
+       if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+               switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
+               case VIRTIO_NET_HDR_GSO_TCPV4:
+                       gso_type = SKB_GSO_TCPV4;
+                       break;
+               case VIRTIO_NET_HDR_GSO_TCPV6:
+                       gso_type = SKB_GSO_TCPV6;
+                       break;
+               case VIRTIO_NET_HDR_GSO_UDP:
+                       gso_type = SKB_GSO_UDP;
+                       break;
+               default:
+                       return -EINVAL;
+               }
+
+               if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
+                       gso_type |= SKB_GSO_TCP_ECN;
+
+               if (vnet_hdr->gso_size == 0)
+                       return -EINVAL;
+       }
+
+       if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
+               if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
+                                         vnet_hdr->csum_offset))
+                       return -EINVAL;
+       }
+
+       if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+               skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
+               skb_shinfo(skb)->gso_type = gso_type;
+
+               /* Header must be checked, and gso_segs computed. */
+               skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
+               skb_shinfo(skb)->gso_segs = 0;
+       }
+       return 0;
+}
+
+static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
+                                  struct virtio_net_hdr *vnet_hdr)
+{
+       memset(vnet_hdr, 0, sizeof(*vnet_hdr));
+
+       if (skb_is_gso(skb)) {
+               struct skb_shared_info *sinfo = skb_shinfo(skb);
+
+               /* This is a hint as to how much should be linear. */
+               vnet_hdr->hdr_len = skb_headlen(skb);
+               vnet_hdr->gso_size = sinfo->gso_size;
+               if (sinfo->gso_type & SKB_GSO_TCPV4)
+                       vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+               else if (sinfo->gso_type & SKB_GSO_TCPV6)
+                       vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
+               else if (sinfo->gso_type & SKB_GSO_UDP)
+                       vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
+               else
+                       BUG();
+               if (sinfo->gso_type & SKB_GSO_TCP_ECN)
+                       vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
+       } else
+               vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
+
+       if (skb->ip_summed == CHECKSUM_PARTIAL) {
+               vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
+               vnet_hdr->csum_start = skb->csum_start -
+                                       skb_headroom(skb);
+               vnet_hdr->csum_offset = skb->csum_offset;
+       } /* else everything is zero */
+
+       return 0;
+}
+
+
 /* Get packet from user space buffer */
 static ssize_t macvtap_get_user(struct macvtap_queue *q,
                                const struct iovec *iv, size_t count,
@@ -327,22 +435,53 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q,
        struct macvlan_dev *vlan;
        size_t len = count;
        int err;
+       struct virtio_net_hdr vnet_hdr = { 0 };
+       int vnet_hdr_len = 0;
+
+       if (q->flags & IFF_VNET_HDR) {
+               vnet_hdr_len = sizeof(vnet_hdr);
+
+               err = -EINVAL;
+               if ((len -= vnet_hdr_len) < 0)
+                       goto err;
+
+               err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
+                                          vnet_hdr_len);
+               if (err < 0)
+                       goto err;
+               if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
+                    vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
+                                                       vnet_hdr.hdr_len)
+                       vnet_hdr.hdr_len = vnet_hdr.csum_start +
+                                               vnet_hdr.csum_offset + 2;
+               err = -EINVAL;
+               if (vnet_hdr.hdr_len > len)
+                       goto err;
+       }
 
+       err = -EINVAL;
        if (unlikely(len < ETH_HLEN))
-               return -EINVAL;
+               goto err;
 
-       skb = sock_alloc_send_skb(&q->sk, NET_IP_ALIGN + len, noblock, &err);
+       skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
+                               noblock, &err);
        if (!skb)
                goto err;
 
-       skb_reserve(skb, NET_IP_ALIGN);
-       skb_put(skb, count);
-
-       err = skb_copy_datagram_from_iovec(skb, 0, iv, 0, len);
+       err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
        if (err)
-               goto err;
+               goto err_kfree;
 
        skb_set_network_header(skb, ETH_HLEN);
+       skb_reset_mac_header(skb);
+       skb->protocol = eth_hdr(skb)->h_proto;
+
+       if (vnet_hdr_len) {
+               err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
+               if (err)
+                       goto err_kfree;
+       }
+
        rcu_read_lock_bh();
        vlan = rcu_dereference(q->vlan);
        if (vlan)
@@ -353,15 +492,16 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q,
 
        return count;
 
+err_kfree:
+       kfree_skb(skb);
+
 err:
        rcu_read_lock_bh();
        vlan = rcu_dereference(q->vlan);
        if (vlan)
-               macvlan_count_rx(q->vlan, 0, false, false);
+               netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++;
        rcu_read_unlock_bh();
 
-       kfree_skb(skb);
-
        return err;
 }
 
@@ -384,10 +524,25 @@ static ssize_t macvtap_put_user(struct macvtap_queue *q,
 {
        struct macvlan_dev *vlan;
        int ret;
+       int vnet_hdr_len = 0;
+
+       if (q->flags & IFF_VNET_HDR) {
+               struct virtio_net_hdr vnet_hdr;
+               vnet_hdr_len = sizeof (vnet_hdr);
+               if ((len -= vnet_hdr_len) < 0)
+                       return -EINVAL;
+
+               ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
+               if (ret)
+                       return ret;
+
+               if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, vnet_hdr_len))
+                       return -EFAULT;
+       }
 
        len = min_t(int, skb->len, len);
 
-       ret = skb_copy_datagram_const_iovec(skb, 0, iv, 0, len);
+       ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
 
        rcu_read_lock_bh();
        vlan = rcu_dereference(q->vlan);
@@ -395,7 +550,7 @@ static ssize_t macvtap_put_user(struct macvtap_queue *q,
                macvlan_count_rx(vlan, len, ret == 0, 0);
        rcu_read_unlock_bh();
 
-       return ret ? ret : len;
+       return ret ? ret : (len + vnet_hdr_len);
 }
 
 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
@@ -473,9 +628,14 @@ static long macvtap_ioctl(struct file *file, unsigned int cmd,
                /* ignore the name, just look at flags */
                if (get_user(u, &ifr->ifr_flags))
                        return -EFAULT;
-               if (u != (IFF_TAP | IFF_NO_PI))
-                       return -EINVAL;
-               return 0;
+
+               ret = 0;
+               if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
+                       ret = -EINVAL;
+               else
+                       q->flags = u;
+
+               return ret;
 
        case TUNGETIFF:
                rcu_read_lock_bh();
@@ -489,13 +649,13 @@ static long macvtap_ioctl(struct file *file, unsigned int cmd,
 
                ret = 0;
                if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) ||
-                   put_user((TUN_TAP_DEV | TUN_NO_PI), &ifr->ifr_flags))
+                   put_user(q->flags, &ifr->ifr_flags))
                        ret = -EFAULT;
                dev_put(vlan->dev);
                return ret;
 
        case TUNGETFEATURES:
-               if (put_user((IFF_TAP | IFF_NO_PI), up))
+               if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
                        return -EFAULT;
                return 0;
 
@@ -509,15 +669,13 @@ static long macvtap_ioctl(struct file *file, unsigned int cmd,
        case TUNSETOFFLOAD:
                /* let the user check for future flags */
                if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
-                         TUN_F_TSO_ECN | TUN_F_UFO))
-                       return -EINVAL;
-
-               /* TODO: add support for these, so far we don't
-                        support any offload */
-               if (arg & (TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
-                        TUN_F_TSO_ECN | TUN_F_UFO))
+                           TUN_F_TSO_ECN | TUN_F_UFO))
                        return -EINVAL;
 
+               /* TODO: only accept frames with the features that
+                        got enabled for forwarded frames */
+               if (!(q->flags & IFF_VNET_HDR))
+                       return  -EINVAL;
                return 0;
 
        default: