Feedback welcome, as always!
(There's been talk of a virtualization git tree, in which case there'll
be a decent home for these patches soon).
Cheers,
Rusty.
=Add feature and GSO support to virtio net driver.
If you don't do GSO, you can simply ignore the first sg element of
every outgoing packet, and tack a dummy one on every incoming.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
---
drivers/net/virtio_net.c | 130 ++++++++++++++++++++++++++++++++++++++++----
include/linux/virtio_net.h | 25 +++++++-
2 files changed, 143 insertions(+), 12 deletions(-)
==================================================================---
a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -16,12 +16,13 @@
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
-//#define DEBUG
+#define DEBUG
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/module.h>
#include <linux/virtio.h>
#include <linux/scatterlist.h>
+#include <linux/virtio_net.h>
/* FIXME: Make dynamic */
#define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
@@ -40,6 +41,18 @@ struct virtnet_info
struct sk_buff_head send;
};
+static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb)
+{
+ return (struct virtio_net_hdr *)skb->cb;
+}
+
+static void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb)
+{
+ sg->page = virt_to_page(skb_vnet_hdr(skb));
+ sg->offset = offset_in_page(skb_vnet_hdr(skb));
+ sg->length = sizeof(struct virtio_net_hdr);
+}
+
static bool skb_xmit_done(struct virtqueue *vq)
{
struct virtnet_info *vi = vq->priv;
@@ -52,12 +65,14 @@ static void receive_skb(struct net_devic
static void receive_skb(struct net_device *dev, struct sk_buff *skb,
unsigned len)
{
- if (unlikely(len < ETH_HLEN)) {
+ struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
+
+ if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
pr_debug("%s: short packet %i\n", dev->name, len);
dev->stats.rx_length_errors++;
- dev_kfree_skb(skb);
- return;
- }
+ goto drop;
+ }
+ len -= sizeof(struct virtio_net_hdr);
BUG_ON(len > MAX_PACKET_LEN);
skb_trim(skb, len);
@@ -66,13 +81,70 @@ static void receive_skb(struct net_devic
ntohs(skb->protocol), skb->len, skb->pkt_type);
dev->stats.rx_bytes += skb->len;
dev->stats.rx_packets++;
+
+ if (hdr->flags & VIRTIO_NET_F_NEEDS_CSUM) {
+ pr_debug("Needs csum!\n");
+ skb->ip_summed = CHECKSUM_PARTIAL;
+ skb->csum_start = hdr->csum_start;
+ skb->csum_offset = hdr->csum_offset;
+ if (skb->csum_start > skb->len - 2
+ || skb->csum_offset > skb->len - 2) {
+ if (net_ratelimit())
+ printk(KERN_WARNING "%s: csum=%u/%u len=%u\n",
+ dev->name, skb->csum_start,
+ skb->csum_offset, skb->len);
+ goto frame_err;
+ }
+ }
+
+ if (hdr->gso_type != VIRTIO_NET_GSO_NONE) {
+ pr_debug("GSO!\n");
+ switch (hdr->gso_type) {
+ case VIRTIO_NET_GSO_TCP:
+ skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
+ break;
+ case VIRTIO_NET_GSO_TCP_ECN:
+ skb_shinfo(skb)->gso_type = SKB_GSO_TCP_ECN;
+ break;
+ case VIRTIO_NET_GSO_UDP:
+ skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
+ break;
+ case VIRTIO_NET_GSO_TCPV6:
+ skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
+ break;
+ default:
+ if (net_ratelimit())
+ printk(KERN_WARNING "%s: bad gso type %u.\n",
+ dev->name, hdr->gso_type);
+ goto frame_err;
+ }
+
+ skb_shinfo(skb)->gso_size = hdr->gso_size;
+ if (skb_shinfo(skb)->gso_size == 0) {
+ if (net_ratelimit())
+ printk(KERN_WARNING "%s: zero gso size.\n",
+ dev->name);
+ goto frame_err;
+ }
+
+ /* Header must be checked, and gso_segs computed. */
+ skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
+ skb_shinfo(skb)->gso_segs = 0;
+ }
+
netif_receive_skb(skb);
+ return;
+
+frame_err:
+ dev->stats.rx_frame_errors++;
+drop:
+ dev_kfree_skb(skb);
}
static void try_fill_recv(struct virtnet_info *vi)
{
struct sk_buff *skb;
- struct scatterlist sg[MAX_SKB_FRAGS];
+ struct scatterlist sg[1+MAX_SKB_FRAGS];
int num, err;
for (;;) {
@@ -81,7 +153,8 @@ static void try_fill_recv(struct virtnet
break;
skb_put(skb, MAX_PACKET_LEN);
- num = skb_to_sgvec(skb, sg, 0, skb->len);
+ vnet_hdr_to_sg(sg, skb);
+ num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
skb_queue_head(&vi->recv, skb);
err = vi->vq_recv->ops->add_buf(vi->vq_recv, sg, 0, num, skb);
@@ -161,7 +234,8 @@ static int start_xmit(struct sk_buff *sk
{
struct virtnet_info *vi = netdev_priv(dev);
int num, err;
- struct scatterlist sg[MAX_SKB_FRAGS];
+ struct scatterlist sg[1+MAX_SKB_FRAGS];
+ struct virtio_net_hdr *hdr;
const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
pr_debug("%s: xmit %p %02x:%02x:%02x:%02x:%02x:%02x\n",
@@ -170,7 +244,41 @@ static int start_xmit(struct sk_buff *sk
free_old_xmit_skbs(vi);
- num = skb_to_sgvec(skb, sg, 0, skb->len);
+ /* Encode metadata header at front. */
+ hdr = skb_vnet_hdr(skb);
+ if (skb->ip_summed == CHECKSUM_PARTIAL) {
+ hdr->flags = VIRTIO_NET_F_NEEDS_CSUM;
+ hdr->csum_start = skb->csum_start - skb_headroom(skb);
+ hdr->csum_offset = skb->csum_offset;
+ } else {
+ hdr->flags = 0;
+ hdr->csum_offset = hdr->csum_start = 0;
+ }
+
+ if (skb_is_gso(skb)) {
+ printk("xmit: gso size %u len %u (%u/%u/%u)\n",
+ skb_shinfo(skb)->gso_size, skb->len,
+ skb_transport_header(skb) - skb->head,
+ skb_network_header(skb) - skb->head,
+ skb->data - skb->head);
+ hdr->gso_size = skb_shinfo(skb)->gso_size;
+ if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
+ hdr->gso_type = VIRTIO_NET_GSO_TCP_ECN;
+ else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
+ hdr->gso_type = VIRTIO_NET_GSO_TCP;
+ else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
+ hdr->gso_type = VIRTIO_NET_GSO_TCPV6;
+ else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
+ hdr->gso_type = VIRTIO_NET_GSO_UDP;
+ else
+ BUG();
+ } else {
+ hdr->gso_type = VIRTIO_NET_GSO_NONE;
+ hdr->gso_size = 0;
+ }
+
+ vnet_hdr_to_sg(sg, skb);
+ num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
__skb_queue_head(&vi->send, skb);
err = vi->vq_send->ops->add_buf(vi->vq_send, sg, num, 0, skb);
if (err) {
@@ -220,7 +328,8 @@ struct net_device *virtnet_probe(struct
struct net_device *virtnet_probe(struct virtqueue *vq_recv,
struct virtqueue *vq_send,
struct device *device,
- const u8 mac[ETH_ALEN])
+ const u8 mac[ETH_ALEN],
+ unsigned long features)
{
int err;
struct net_device *dev;
@@ -239,6 +348,7 @@ struct net_device *virtnet_probe(struct
dev->poll = virtnet_poll;
dev->hard_start_xmit = start_xmit;
dev->weight = 16;
+ dev->features = features;
SET_NETDEV_DEV(dev, device);
vi = netdev_priv(dev);
==================================================================---
a/include/linux/virtio_net.h
+++ b/include/linux/virtio_net.h
@@ -1,6 +1,26 @@
#ifndef _LINUX_VIRTIO_NET_H
#define _LINUX_VIRTIO_NET_H
#include <linux/types.h>
+
+/* This is the first element of the scatter-gather list.
+ * If you don't specify GSO or CSUM features, you can simply ignore the
+ * header. */
+struct virtio_net_hdr
+{
+#define VIRTIO_NET_F_NEEDS_CSUM 1 // Use csum_start, csum_offset
+ __u16 flags;
+#define VIRTIO_NET_GSO_NONE 0 // Not a GSO frame
+#define VIRTIO_NET_GSO_TCP 1 // GSO frame, IPv4 TCP (TSO)
+#define VIRTIO_NET_GSO_TCP_ECN 2 // GSO frame, IPv4 TCP w/ ECN
+#define VIRTIO_NET_GSO_UDP 3 // GSO frame, IPv4 UDP (UFO)
+#define VIRTIO_NET_GSO_TCPV6 4 // GSO frame, IPv6 TCP
+ __u16 gso_type;
+ __u16 gso_size;
+ __u16 csum_start;
+ __u16 csum_offset;
+};
+
+#ifdef __KERNEL__
#include <linux/etherdevice.h>
struct device;
struct net_device;
@@ -9,7 +29,8 @@ struct net_device *virtnet_probe(struct
struct net_device *virtnet_probe(struct virtqueue *vq_recv,
struct virtqueue *vq_send,
struct device *dev,
- const u8 mac[ETH_ALEN]);
+ const u8 mac[ETH_ALEN],
+ unsigned long features);
void virtnet_remove(struct net_device *dev);
-
+#endif /* __KERNEL__ */
#endif /* _LINUX_VIRTIO_NET_H */
