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 /* FIXME: MTU in config. */
  28 #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
  29
  30 struct virtnet_info
  31 {
  32         struct virtio_device *vdev;
  33         struct virtqueue *rvq, *svq;
  34         struct net_device *dev;
  35         struct napi_struct napi;
  36
  37         /* Number of input buffers, and max we've ever had. */
  38         unsigned int num, max;
  39
  40         /* Receive & send queues. */
  41         struct sk_buff_head recv;
  42         struct sk_buff_head send;
  43 };
  44
  45 static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb)
  46 {
  47         return (struct virtio_net_hdr *)skb->cb;
  48 }
  49
  50 static inline void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb)
  51 {
  52         sg_init_one(sg, skb_vnet_hdr(skb), sizeof(struct virtio_net_hdr));
  53 }
  54
  55 static void skb_xmit_done(struct virtqueue *rvq)
  56 {
  57         struct virtnet_info *vi = rvq->vdev->priv;
  58
  59         /* In case we were waiting for output buffers. */
  60         netif_wake_queue(vi->dev);
  61 }
  62
  63 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
  64                         unsigned len)
  65 {
  66         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
  67
  68         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
  69                 pr_debug("%s: short packet %i\n", dev->name, len);
  70                 dev->stats.rx_length_errors++;
  71                 goto drop;
  72         }
  73         len -= sizeof(struct virtio_net_hdr);
  74         BUG_ON(len > MAX_PACKET_LEN);
  75
  76         skb_trim(skb, len);
  77         skb->protocol = eth_type_trans(skb, dev);
  78         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
  79                  ntohs(skb->protocol), skb->len, skb->pkt_type);
  80         dev->stats.rx_bytes += skb->len;
  81         dev->stats.rx_packets++;
  82
  83         if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
  84                 pr_debug("Needs csum!\n");
  85                 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
  86                         goto frame_err;
  87         }
  88
  89         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
  90                 pr_debug("GSO!\n");
  91                 switch (hdr->gso_type) {
  92                 case VIRTIO_NET_HDR_GSO_TCPV4:
  93                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
  94                         break;
  95                 case VIRTIO_NET_HDR_GSO_TCPV4_ECN:
  96                         skb_shinfo(skb)->gso_type = SKB_GSO_TCP_ECN;
  97                         break;
  98                 case VIRTIO_NET_HDR_GSO_UDP:
  99                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 100                         break;
 101                 case VIRTIO_NET_HDR_GSO_TCPV6:
 102                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 103                         break;
 104                 default:
 105                         if (net_ratelimit())
 106                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 107                                        dev->name, hdr->gso_type);
 108                         goto frame_err;
 109                 }
 110
 111                 skb_shinfo(skb)->gso_size = hdr->gso_size;
 112                 if (skb_shinfo(skb)->gso_size == 0) {
 113                         if (net_ratelimit())
 114                                 printk(KERN_WARNING "%s: zero gso size.\n",
 115                                        dev->name);
 116                         goto frame_err;
 117                 }
 118
 119                 /* Header must be checked, and gso_segs computed. */
 120                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 121                 skb_shinfo(skb)->gso_segs = 0;
 122         }
 123
 124         netif_receive_skb(skb);
 125         return;
 126
 127 frame_err:
 128         dev->stats.rx_frame_errors++;
 129 drop:
 130         dev_kfree_skb(skb);
 131 }
 132
 133 static void try_fill_recv(struct virtnet_info *vi)
 134 {
 135         struct sk_buff *skb;
 136         struct scatterlist sg[1+MAX_SKB_FRAGS];
 137         int num, err;
 138
 139         sg_init_table(sg, 1+MAX_SKB_FRAGS);
 140         for (;;) {
 141                 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
 142                 if (unlikely(!skb))
 143                         break;
 144
 145                 skb_put(skb, MAX_PACKET_LEN);
 146                 vnet_hdr_to_sg(sg, skb);
 147                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 148                 skb_queue_head(&vi->recv, skb);
 149
 150                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 151                 if (err) {
 152                         skb_unlink(skb, &vi->recv);
 153                         kfree_skb(skb);
 154                         break;
 155                 }
 156                 vi->num++;
 157         }
 158         if (unlikely(vi->num > vi->max))
 159                 vi->max = vi->num;
 160         vi->rvq->vq_ops->kick(vi->rvq);
 161 }
 162
 163 static void skb_recv_done(struct virtqueue *rvq)
 164 {
 165         struct virtnet_info *vi = rvq->vdev->priv;
 166         /* Schedule NAPI, Suppress further interrupts if successful. */
 167         if (netif_rx_schedule_prep(vi->dev, &vi->napi)) {
 168                 rvq->vq_ops->disable_cb(rvq);
 169                 __netif_rx_schedule(vi->dev, &vi->napi);
 170         }
 171 }
 172
 173 static int virtnet_poll(struct napi_struct *napi, int budget)
 174 {
 175         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 176         struct sk_buff *skb = NULL;
 177         unsigned int len, received = 0;
 178
 179 again:
 180         while (received < budget &&
 181                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 182                 __skb_unlink(skb, &vi->recv);
 183                 receive_skb(vi->dev, skb, len);
 184                 vi->num--;
 185                 received++;
 186         }
 187
 188         /* FIXME: If we oom and completely run out of inbufs, we need
 189          * to start a timer trying to fill more. */
 190         if (vi->num < vi->max / 2)
 191                 try_fill_recv(vi);
 192
 193         /* Out of packets? */
 194         if (received < budget) {
 195                 netif_rx_complete(vi->dev, napi);
 196                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 197                     && netif_rx_reschedule(vi->dev, napi))
 198                         goto again;
 199         }
 200
 201         return received;
 202 }
 203
 204 static void free_old_xmit_skbs(struct virtnet_info *vi)
 205 {
 206         struct sk_buff *skb;
 207         unsigned int len;
 208
 209         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 210                 pr_debug("Sent skb %p\n", skb);
 211                 __skb_unlink(skb, &vi->send);
 212                 vi->dev->stats.tx_bytes += len;
 213                 vi->dev->stats.tx_packets++;
 214                 kfree_skb(skb);
 215         }
 216 }
 217
 218 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
 219 {
 220         struct virtnet_info *vi = netdev_priv(dev);
 221         int num, err;
 222         struct scatterlist sg[1+MAX_SKB_FRAGS];
 223         struct virtio_net_hdr *hdr;
 224         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 225         DECLARE_MAC_BUF(mac);
 226
 227         sg_init_table(sg, 1+MAX_SKB_FRAGS);
 228
 229         pr_debug("%s: xmit %p %s\n", dev->name, skb, print_mac(mac, dest));
 230
 231         free_old_xmit_skbs(vi);
 232
 233         /* Encode metadata header at front. */
 234         hdr = skb_vnet_hdr(skb);
 235         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 236                 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 237                 hdr->csum_start = skb->csum_start - skb_headroom(skb);
 238                 hdr->csum_offset = skb->csum_offset;
 239         } else {
 240                 hdr->flags = 0;
 241                 hdr->csum_offset = hdr->csum_start = 0;
 242         }
 243
 244         if (skb_is_gso(skb)) {
 245                 hdr->hdr_len = skb_transport_header(skb) - skb->data;
 246                 hdr->gso_size = skb_shinfo(skb)->gso_size;
 247                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 248                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4_ECN;
 249                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 250                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 251                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 252                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 253                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 254                         hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 255                 else
 256                         BUG();
 257         } else {
 258                 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 259                 hdr->gso_size = hdr->hdr_len = 0;
 260         }
 261
 262         vnet_hdr_to_sg(sg, skb);
 263         num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 264         __skb_queue_head(&vi->send, skb);
 265         err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
 266         if (err) {
 267                 pr_debug("%s: virtio not prepared to send\n", dev->name);
 268                 skb_unlink(skb, &vi->send);
 269                 netif_stop_queue(dev);
 270                 return NETDEV_TX_BUSY;
 271         }
 272         vi->svq->vq_ops->kick(vi->svq);
 273
 274         return 0;
 275 }
 276
 277 static int virtnet_open(struct net_device *dev)
 278 {
 279         struct virtnet_info *vi = netdev_priv(dev);
 280
 281         try_fill_recv(vi);
 282
 283         /* If we didn't even get one input buffer, we're useless. */
 284         if (vi->num == 0)
 285                 return -ENOMEM;
 286
 287         napi_enable(&vi->napi);
 288         return 0;
 289 }
 290
 291 static int virtnet_close(struct net_device *dev)
 292 {
 293         struct virtnet_info *vi = netdev_priv(dev);
 294         struct sk_buff *skb;
 295
 296         napi_disable(&vi->napi);
 297
 298         /* networking core has neutered skb_xmit_done/skb_recv_done, so don't
 299          * worry about races vs. get(). */
 300         vi->rvq->vq_ops->shutdown(vi->rvq);
 301         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 302                 kfree_skb(skb);
 303                 vi->num--;
 304         }
 305         vi->svq->vq_ops->shutdown(vi->svq);
 306         while ((skb = __skb_dequeue(&vi->send)) != NULL)
 307                 kfree_skb(skb);
 308
 309         BUG_ON(vi->num != 0);
 310         return 0;
 311 }
 312
 313 static int virtnet_probe(struct virtio_device *vdev)
 314 {
 315         int err;
 316         struct net_device *dev;
 317         struct virtnet_info *vi;
 318
 319         /* Allocate ourselves a network device with room for our info */
 320         dev = alloc_etherdev(sizeof(struct virtnet_info));
 321         if (!dev)
 322                 return -ENOMEM;
 323
 324         /* Set up network device as normal. */
 325         ether_setup(dev);
 326         dev->open = virtnet_open;
 327         dev->stop = virtnet_close;
 328         dev->hard_start_xmit = start_xmit;
 329         dev->features = NETIF_F_HIGHDMA;
 330         SET_NETDEV_DEV(dev, &vdev->dev);
 331
 332         /* Do we support "hardware" checksums? */
 333         if (vdev->config->feature(vdev, VIRTIO_NET_F_NO_CSUM)) {
 334                 /* This opens up the world of extra features. */
 335                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 336                 if (vdev->config->feature(vdev, VIRTIO_NET_F_TSO4))
 337                         dev->features |= NETIF_F_TSO;
 338                 if (vdev->config->feature(vdev, VIRTIO_NET_F_UFO))
 339                         dev->features |= NETIF_F_UFO;
 340                 if (vdev->config->feature(vdev, VIRTIO_NET_F_TSO4_ECN))
 341                         dev->features |= NETIF_F_TSO_ECN;
 342                 if (vdev->config->feature(vdev, VIRTIO_NET_F_TSO6))
 343                         dev->features |= NETIF_F_TSO6;
 344         }
 345
 346         /* Configuration may specify what MAC to use.  Otherwise random. */
 347         if (vdev->config->feature(vdev, VIRTIO_NET_F_MAC)) {
 348                 vdev->config->get(vdev,
 349                                   offsetof(struct virtio_net_config, mac),
 350                                   dev->dev_addr, dev->addr_len);
 351         } else
 352                 random_ether_addr(dev->dev_addr);
 353
 354         /* Set up our device-specific information */
 355         vi = netdev_priv(dev);
 356         netif_napi_add(dev, &vi->napi, virtnet_poll, 16);
 357         vi->dev = dev;
 358         vi->vdev = vdev;
 359
 360         /* We expect two virtqueues, receive then send. */
 361         vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
 362         if (IS_ERR(vi->rvq)) {
 363                 err = PTR_ERR(vi->rvq);
 364                 goto free;
 365         }
 366
 367         vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
 368         if (IS_ERR(vi->svq)) {
 369                 err = PTR_ERR(vi->svq);
 370                 goto free_recv;
 371         }
 372
 373         /* Initialize our empty receive and send queues. */
 374         skb_queue_head_init(&vi->recv);
 375         skb_queue_head_init(&vi->send);
 376
 377         err = register_netdev(dev);
 378         if (err) {
 379                 pr_debug("virtio_net: registering device failed\n");
 380                 goto free_send;
 381         }
 382         pr_debug("virtnet: registered device %s\n", dev->name);
 383         vdev->priv = vi;
 384         return 0;
 385
 386 free_send:
 387         vdev->config->del_vq(vi->svq);
 388 free_recv:
 389         vdev->config->del_vq(vi->rvq);
 390 free:
 391         free_netdev(dev);
 392         return err;
 393 }
 394
 395 static void virtnet_remove(struct virtio_device *vdev)
 396 {
 397         struct virtnet_info *vi = vdev->priv;
 398
 399         vdev->config->del_vq(vi->svq);
 400         vdev->config->del_vq(vi->rvq);
 401         unregister_netdev(vi->dev);
 402         free_netdev(vi->dev);
 403 }
 404
 405 static struct virtio_device_id id_table[] = {
 406         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
 407         { 0 },
 408 };
 409
 410 static struct virtio_driver virtio_net = {
 411         .driver.name =  KBUILD_MODNAME,
 412         .driver.owner = THIS_MODULE,
 413         .id_table =     id_table,
 414         .probe =        virtnet_probe,
 415         .remove =       __devexit_p(virtnet_remove),
 416 };
 417
 418 static int __init init(void)
 419 {
 420         return register_virtio_driver(&virtio_net);
 421 }
 422
 423 static void __exit fini(void)
 424 {
 425         unregister_virtio_driver(&virtio_net);
 426 }
 427 module_init(init);
 428 module_exit(fini);
 429
 430 MODULE_DEVICE_TABLE(virtio, id_table);
 431 MODULE_DESCRIPTION("Virtio network driver");
 432 MODULE_LICENSE("GPL");