drivers/net/virtio_net.c

   1 /* A simple network driver using virtio.
   2  *
   3  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19 //#define DEBUG
  20 #include <linux/netdevice.h>
  21 #include <linux/etherdevice.h>
  22 #include <linux/module.h>
  23 #include <linux/virtio.h>
  24 #include <linux/virtio_net.h>
  25 #include <linux/scatterlist.h>
  26
  27 static int csum = 1, gso = 1;
  28 module_param(csum, bool, 0444);
  29 module_param(gso, bool, 0444);
  30
  31 /* FIXME: MTU in config. */
  32 #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
  33
  34 struct virtnet_info
  35 {
  36         struct virtio_device *vdev;
  37         struct virtqueue *rvq, *svq;
  38         struct net_device *dev;
  39         struct napi_struct napi;
  40
  41         /* Number of input buffers, and max we've ever had. */
  42         unsigned int num, max;
  43
  44         /* Receive & send queues. */
  45         struct sk_buff_head recv;
  46         struct sk_buff_head send;
  47 };
  48
  49 static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb)
  50 {
  51         return (struct virtio_net_hdr *)skb->cb;
  52 }
  53
  54 static inline void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb)
  55 {
  56         sg_init_one(sg, skb_vnet_hdr(skb), sizeof(struct virtio_net_hdr));
  57 }
  58
  59 static void skb_xmit_done(struct virtqueue *rvq)
  60 {
  61         struct virtnet_info *vi = rvq->vdev->priv;
  62
  63         /* In case we were waiting for output buffers. */
  64         netif_wake_queue(vi->dev);
  65 }
  66
  67 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
  68                         unsigned len)
  69 {
  70         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
  71
  72         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
  73                 pr_debug("%s: short packet %i\n", dev->name, len);
  74                 dev->stats.rx_length_errors++;
  75                 goto drop;
  76         }
  77         len -= sizeof(struct virtio_net_hdr);
  78         BUG_ON(len > MAX_PACKET_LEN);
  79
  80         skb_trim(skb, len);
  81         skb->protocol = eth_type_trans(skb, dev);
  82         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
  83                  ntohs(skb->protocol), skb->len, skb->pkt_type);
  84         dev->stats.rx_bytes += skb->len;
  85         dev->stats.rx_packets++;
  86
  87         if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
  88                 pr_debug("Needs csum!\n");
  89                 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
  90                         goto frame_err;
  91         }
  92
  93         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
  94                 pr_debug("GSO!\n");
  95                 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
  96                 case VIRTIO_NET_HDR_GSO_TCPV4:
  97                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
  98                         break;
  99                 case VIRTIO_NET_HDR_GSO_UDP:
 100                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 101                         break;
 102                 case VIRTIO_NET_HDR_GSO_TCPV6:
 103                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 104                         break;
 105                 default:
 106                         if (net_ratelimit())
 107                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 108                                        dev->name, hdr->gso_type);
 109                         goto frame_err;
 110                 }
 111
 112                 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 113                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 114
 115                 skb_shinfo(skb)->gso_size = hdr->gso_size;
 116                 if (skb_shinfo(skb)->gso_size == 0) {
 117                         if (net_ratelimit())
 118                                 printk(KERN_WARNING "%s: zero gso size.\n",
 119                                        dev->name);
 120                         goto frame_err;
 121                 }
 122
 123                 /* Header must be checked, and gso_segs computed. */
 124                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 125                 skb_shinfo(skb)->gso_segs = 0;
 126         }
 127
 128         netif_receive_skb(skb);
 129         return;
 130
 131 frame_err:
 132         dev->stats.rx_frame_errors++;
 133 drop:
 134         dev_kfree_skb(skb);
 135 }
 136
 137 static void try_fill_recv(struct virtnet_info *vi)
 138 {
 139         struct sk_buff *skb;
 140         struct scatterlist sg[1+MAX_SKB_FRAGS];
 141         int num, err;
 142
 143         sg_init_table(sg, 1+MAX_SKB_FRAGS);
 144         for (;;) {
 145                 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
 146                 if (unlikely(!skb))
 147                         break;
 148
 149                 skb_put(skb, MAX_PACKET_LEN);
 150                 vnet_hdr_to_sg(sg, skb);
 151                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 152                 skb_queue_head(&vi->recv, skb);
 153
 154                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 155                 if (err) {
 156                         skb_unlink(skb, &vi->recv);
 157                         kfree_skb(skb);
 158                         break;
 159                 }
 160                 vi->num++;
 161         }
 162         if (unlikely(vi->num > vi->max))
 163                 vi->max = vi->num;
 164         vi->rvq->vq_ops->kick(vi->rvq);
 165 }
 166
 167 static void skb_recv_done(struct virtqueue *rvq)
 168 {
 169         struct virtnet_info *vi = rvq->vdev->priv;
 170         /* Schedule NAPI, Suppress further interrupts if successful. */
 171         if (netif_rx_schedule_prep(vi->dev, &vi->napi)) {
 172                 rvq->vq_ops->disable_cb(rvq);
 173                 __netif_rx_schedule(vi->dev, &vi->napi);
 174         }
 175 }
 176
 177 static int virtnet_poll(struct napi_struct *napi, int budget)
 178 {
 179         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 180         struct sk_buff *skb = NULL;
 181         unsigned int len, received = 0;
 182
 183 again:
 184         while (received < budget &&
 185                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 186                 __skb_unlink(skb, &vi->recv);
 187                 receive_skb(vi->dev, skb, len);
 188                 vi->num--;
 189                 received++;
 190         }
 191
 192         /* FIXME: If we oom and completely run out of inbufs, we need
 193          * to start a timer trying to fill more. */
 194         if (vi->num < vi->max / 2)
 195                 try_fill_recv(vi);
 196
 197         /* Out of packets? */
 198         if (received < budget) {
 199                 netif_rx_complete(vi->dev, napi);
 200                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 201                     && netif_rx_reschedule(vi->dev, napi))
 202                         goto again;
 203         }
 204
 205         return received;
 206 }
 207
 208 static void free_old_xmit_skbs(struct virtnet_info *vi)
 209 {
 210         struct sk_buff *skb;
 211         unsigned int len;
 212
 213         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 214                 pr_debug("Sent skb %p\n", skb);
 215                 __skb_unlink(skb, &vi->send);
 216                 vi->dev->stats.tx_bytes += len;
 217                 vi->dev->stats.tx_packets++;
 218                 kfree_skb(skb);
 219         }
 220 }
 221
 222 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
 223 {
 224         struct virtnet_info *vi = netdev_priv(dev);
 225         int num, err;
 226         struct scatterlist sg[1+MAX_SKB_FRAGS];
 227         struct virtio_net_hdr *hdr;
 228         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 229         DECLARE_MAC_BUF(mac);
 230
 231         sg_init_table(sg, 1+MAX_SKB_FRAGS);
 232
 233         pr_debug("%s: xmit %p %s\n", dev->name, skb, print_mac(mac, dest));
 234
 235         free_old_xmit_skbs(vi);
 236
 237         /* Encode metadata header at front. */
 238         hdr = skb_vnet_hdr(skb);
 239         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 240                 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 241                 hdr->csum_start = skb->csum_start - skb_headroom(skb);
 242                 hdr->csum_offset = skb->csum_offset;
 243         } else {
 244                 hdr->flags = 0;
 245                 hdr->csum_offset = hdr->csum_start = 0;
 246         }
 247
 248         if (skb_is_gso(skb)) {
 249                 hdr->hdr_len = skb_transport_header(skb) - skb->data;
 250                 hdr->gso_size = skb_shinfo(skb)->gso_size;
 251                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 252                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 253                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 254                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 255                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 256                         hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 257                 else
 258                         BUG();
 259                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 260                         hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 261         } else {
 262                 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 263                 hdr->gso_size = hdr->hdr_len = 0;
 264         }
 265
 266         vnet_hdr_to_sg(sg, skb);
 267         num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 268         __skb_queue_head(&vi->send, skb);
 269         err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
 270         if (err) {
 271                 pr_debug("%s: virtio not prepared to send\n", dev->name);
 272                 skb_unlink(skb, &vi->send);
 273                 netif_stop_queue(dev);
 274                 return NETDEV_TX_BUSY;
 275         }
 276         vi->svq->vq_ops->kick(vi->svq);
 277
 278         return 0;
 279 }
 280
 281 static int virtnet_open(struct net_device *dev)
 282 {
 283         struct virtnet_info *vi = netdev_priv(dev);
 284
 285         napi_enable(&vi->napi);
 286         return 0;
 287 }
 288
 289 static int virtnet_close(struct net_device *dev)
 290 {
 291         struct virtnet_info *vi = netdev_priv(dev);
 292
 293         napi_disable(&vi->napi);
 294
 295         return 0;
 296 }
 297
 298 static int virtnet_probe(struct virtio_device *vdev)
 299 {
 300         int err;
 301         struct net_device *dev;
 302         struct virtnet_info *vi;
 303
 304         /* Allocate ourselves a network device with room for our info */
 305         dev = alloc_etherdev(sizeof(struct virtnet_info));
 306         if (!dev)
 307                 return -ENOMEM;
 308
 309         /* Set up network device as normal. */
 310         dev->open = virtnet_open;
 311         dev->stop = virtnet_close;
 312         dev->hard_start_xmit = start_xmit;
 313         dev->features = NETIF_F_HIGHDMA;
 314         SET_NETDEV_DEV(dev, &vdev->dev);
 315
 316         /* Do we support "hardware" checksums? */
 317         if (csum && vdev->config->feature(vdev, VIRTIO_NET_F_CSUM)) {
 318                 /* This opens up the world of extra features. */
 319                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 320                 if (gso && vdev->config->feature(vdev, VIRTIO_NET_F_GSO)) {
 321                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 322                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 323                 }
 324         }
 325
 326         /* Configuration may specify what MAC to use.  Otherwise random. */
 327         if (vdev->config->feature(vdev, VIRTIO_NET_F_MAC)) {
 328                 vdev->config->get(vdev,
 329                                   offsetof(struct virtio_net_config, mac),
 330                                   dev->dev_addr, dev->addr_len);
 331         } else
 332                 random_ether_addr(dev->dev_addr);
 333
 334         /* Set up our device-specific information */
 335         vi = netdev_priv(dev);
 336         netif_napi_add(dev, &vi->napi, virtnet_poll, 16);
 337         vi->dev = dev;
 338         vi->vdev = vdev;
 339
 340         /* We expect two virtqueues, receive then send. */
 341         vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
 342         if (IS_ERR(vi->rvq)) {
 343                 err = PTR_ERR(vi->rvq);
 344                 goto free;
 345         }
 346
 347         vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
 348         if (IS_ERR(vi->svq)) {
 349                 err = PTR_ERR(vi->svq);
 350                 goto free_recv;
 351         }
 352
 353         /* Initialize our empty receive and send queues. */
 354         skb_queue_head_init(&vi->recv);
 355         skb_queue_head_init(&vi->send);
 356
 357         err = register_netdev(dev);
 358         if (err) {
 359                 pr_debug("virtio_net: registering device failed\n");
 360                 goto free_send;
 361         }
 362
 363         /* Last of all, set up some receive buffers. */
 364         try_fill_recv(vi);
 365
 366         /* If we didn't even get one input buffer, we're useless. */
 367         if (vi->num == 0) {
 368                 err = -ENOMEM;
 369                 goto unregister;
 370         }
 371
 372         pr_debug("virtnet: registered device %s\n", dev->name);
 373         vdev->priv = vi;
 374         return 0;
 375
 376 unregister:
 377         unregister_netdev(dev);
 378 free_send:
 379         vdev->config->del_vq(vi->svq);
 380 free_recv:
 381         vdev->config->del_vq(vi->rvq);
 382 free:
 383         free_netdev(dev);
 384         return err;
 385 }
 386
 387 static void virtnet_remove(struct virtio_device *vdev)
 388 {
 389         struct virtnet_info *vi = vdev->priv;
 390         struct sk_buff *skb;
 391
 392         /* Stop all the virtqueues. */
 393         vdev->config->reset(vdev);
 394
 395         /* Free our skbs in send and recv queues, if any. */
 396         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 397                 kfree_skb(skb);
 398                 vi->num--;
 399         }
 400         while ((skb = __skb_dequeue(&vi->send)) != NULL)
 401                 kfree_skb(skb);
 402
 403         BUG_ON(vi->num != 0);
 404
 405         vdev->config->del_vq(vi->svq);
 406         vdev->config->del_vq(vi->rvq);
 407         unregister_netdev(vi->dev);
 408         free_netdev(vi->dev);
 409 }
 410
 411 static struct virtio_device_id id_table[] = {
 412         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
 413         { 0 },
 414 };
 415
 416 static struct virtio_driver virtio_net = {
 417         .driver.name =  KBUILD_MODNAME,
 418         .driver.owner = THIS_MODULE,
 419         .id_table =     id_table,
 420         .probe =        virtnet_probe,
 421         .remove =       __devexit_p(virtnet_remove),
 422 };
 423
 424 static int __init init(void)
 425 {
 426         return register_virtio_driver(&virtio_net);
 427 }
 428
 429 static void __exit fini(void)
 430 {
 431         unregister_virtio_driver(&virtio_net);
 432 }
 433 module_init(init);
 434 module_exit(fini);
 435
 436 MODULE_DEVICE_TABLE(virtio, id_table);
 437 MODULE_DESCRIPTION("Virtio network driver");
 438 MODULE_LICENSE("GPL");