drivers/net/virtio_net.c

   1 /* A 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/ethtool.h>
  23 #include <linux/module.h>
  24 #include <linux/virtio.h>
  25 #include <linux/virtio_net.h>
  26 #include <linux/scatterlist.h>
  27 #include <linux/if_vlan.h>
  28
  29 static int napi_weight = 128;
  30 module_param(napi_weight, int, 0444);
  31
  32 static int csum = 1, gso = 1;
  33 module_param(csum, bool, 0444);
  34 module_param(gso, bool, 0444);
  35
  36 /* FIXME: MTU in config. */
  37 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
  38 #define GOOD_COPY_LEN   128
  39
  40 #define VIRTNET_SEND_COMMAND_SG_MAX    2
  41
  42 struct virtnet_info
  43 {
  44         struct virtio_device *vdev;
  45         struct virtqueue *rvq, *svq, *cvq;
  46         struct net_device *dev;
  47         struct napi_struct napi;
  48         unsigned int status;
  49
  50         /* Number of input buffers, and max we've ever had. */
  51         unsigned int num, max;
  52
  53         /* I like... big packets and I cannot lie! */
  54         bool big_packets;
  55
  56         /* Host will merge rx buffers for big packets (shake it! shake it!) */
  57         bool mergeable_rx_bufs;
  58
  59         /* Receive & send queues. */
  60         struct sk_buff_head recv;
  61         struct sk_buff_head send;
  62
  63         /* Work struct for refilling if we run low on memory. */
  64         struct delayed_work refill;
  65
  66         /* Chain pages by the private ptr. */
  67         struct page *pages;
  68 };
  69
  70 struct skb_vnet_hdr {
  71         union {
  72                 struct virtio_net_hdr hdr;
  73                 struct virtio_net_hdr_mrg_rxbuf mhdr;
  74         };
  75         unsigned int num_sg;
  76 };
  77
  78 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
  79 {
  80         return (struct skb_vnet_hdr *)skb->cb;
  81 }
  82
  83 static void give_a_page(struct virtnet_info *vi, struct page *page)
  84 {
  85         page->private = (unsigned long)vi->pages;
  86         vi->pages = page;
  87 }
  88
  89 static void trim_pages(struct virtnet_info *vi, struct sk_buff *skb)
  90 {
  91         unsigned int i;
  92
  93         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
  94                 give_a_page(vi, skb_shinfo(skb)->frags[i].page);
  95         skb_shinfo(skb)->nr_frags = 0;
  96         skb->data_len = 0;
  97 }
  98
  99 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
 100 {
 101         struct page *p = vi->pages;
 102
 103         if (p)
 104                 vi->pages = (struct page *)p->private;
 105         else
 106                 p = alloc_page(gfp_mask);
 107         return p;
 108 }
 109
 110 static void skb_xmit_done(struct virtqueue *svq)
 111 {
 112         struct virtnet_info *vi = svq->vdev->priv;
 113
 114         /* Suppress further interrupts. */
 115         svq->vq_ops->disable_cb(svq);
 116
 117         /* We were probably waiting for more output buffers. */
 118         netif_wake_queue(vi->dev);
 119 }
 120
 121 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
 122                         unsigned len)
 123 {
 124         struct virtnet_info *vi = netdev_priv(dev);
 125         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 126         int err;
 127         int i;
 128
 129         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 130                 pr_debug("%s: short packet %i\n", dev->name, len);
 131                 dev->stats.rx_length_errors++;
 132                 goto drop;
 133         }
 134
 135         if (vi->mergeable_rx_bufs) {
 136                 unsigned int copy;
 137                 char *p = page_address(skb_shinfo(skb)->frags[0].page);
 138
 139                 if (len > PAGE_SIZE)
 140                         len = PAGE_SIZE;
 141                 len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
 142
 143                 memcpy(&hdr->mhdr, p, sizeof(hdr->mhdr));
 144                 p += sizeof(hdr->mhdr);
 145
 146                 copy = len;
 147                 if (copy > skb_tailroom(skb))
 148                         copy = skb_tailroom(skb);
 149
 150                 memcpy(skb_put(skb, copy), p, copy);
 151
 152                 len -= copy;
 153
 154                 if (!len) {
 155                         give_a_page(vi, skb_shinfo(skb)->frags[0].page);
 156                         skb_shinfo(skb)->nr_frags--;
 157                 } else {
 158                         skb_shinfo(skb)->frags[0].page_offset +=
 159                                 sizeof(hdr->mhdr) + copy;
 160                         skb_shinfo(skb)->frags[0].size = len;
 161                         skb->data_len += len;
 162                         skb->len += len;
 163                 }
 164
 165                 while (--hdr->mhdr.num_buffers) {
 166                         struct sk_buff *nskb;
 167
 168                         i = skb_shinfo(skb)->nr_frags;
 169                         if (i >= MAX_SKB_FRAGS) {
 170                                 pr_debug("%s: packet too long %d\n", dev->name,
 171                                          len);
 172                                 dev->stats.rx_length_errors++;
 173                                 goto drop;
 174                         }
 175
 176                         nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
 177                         if (!nskb) {
 178                                 pr_debug("%s: rx error: %d buffers missing\n",
 179                                          dev->name, hdr->mhdr.num_buffers);
 180                                 dev->stats.rx_length_errors++;
 181                                 goto drop;
 182                         }
 183
 184                         __skb_unlink(nskb, &vi->recv);
 185                         vi->num--;
 186
 187                         skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
 188                         skb_shinfo(nskb)->nr_frags = 0;
 189                         kfree_skb(nskb);
 190
 191                         if (len > PAGE_SIZE)
 192                                 len = PAGE_SIZE;
 193
 194                         skb_shinfo(skb)->frags[i].size = len;
 195                         skb_shinfo(skb)->nr_frags++;
 196                         skb->data_len += len;
 197                         skb->len += len;
 198                 }
 199         } else {
 200                 len -= sizeof(hdr->hdr);
 201
 202                 if (len <= MAX_PACKET_LEN)
 203                         trim_pages(vi, skb);
 204
 205                 err = pskb_trim(skb, len);
 206                 if (err) {
 207                         pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
 208                                  len, err);
 209                         dev->stats.rx_dropped++;
 210                         goto drop;
 211                 }
 212         }
 213
 214         skb->truesize += skb->data_len;
 215         dev->stats.rx_bytes += skb->len;
 216         dev->stats.rx_packets++;
 217
 218         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 219                 pr_debug("Needs csum!\n");
 220                 if (!skb_partial_csum_set(skb,
 221                                           hdr->hdr.csum_start,
 222                                           hdr->hdr.csum_offset))
 223                         goto frame_err;
 224         }
 225
 226         skb->protocol = eth_type_trans(skb, dev);
 227         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 228                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 229
 230         if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 231                 pr_debug("GSO!\n");
 232                 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 233                 case VIRTIO_NET_HDR_GSO_TCPV4:
 234                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 235                         break;
 236                 case VIRTIO_NET_HDR_GSO_UDP:
 237                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 238                         break;
 239                 case VIRTIO_NET_HDR_GSO_TCPV6:
 240                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 241                         break;
 242                 default:
 243                         if (net_ratelimit())
 244                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 245                                        dev->name, hdr->hdr.gso_type);
 246                         goto frame_err;
 247                 }
 248
 249                 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 250                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 251
 252                 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
 253                 if (skb_shinfo(skb)->gso_size == 0) {
 254                         if (net_ratelimit())
 255                                 printk(KERN_WARNING "%s: zero gso size.\n",
 256                                        dev->name);
 257                         goto frame_err;
 258                 }
 259
 260                 /* Header must be checked, and gso_segs computed. */
 261                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 262                 skb_shinfo(skb)->gso_segs = 0;
 263         }
 264
 265         netif_receive_skb(skb);
 266         return;
 267
 268 frame_err:
 269         dev->stats.rx_frame_errors++;
 270 drop:
 271         dev_kfree_skb(skb);
 272 }
 273
 274 static bool try_fill_recv_maxbufs(struct virtnet_info *vi, gfp_t gfp)
 275 {
 276         struct sk_buff *skb;
 277         struct scatterlist sg[2+MAX_SKB_FRAGS];
 278         int num, err, i;
 279         bool oom = false;
 280
 281         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 282         do {
 283                 struct skb_vnet_hdr *hdr;
 284
 285                 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN + NET_IP_ALIGN);
 286                 if (unlikely(!skb)) {
 287                         oom = true;
 288                         break;
 289                 }
 290
 291                 skb_reserve(skb, NET_IP_ALIGN);
 292                 skb_put(skb, MAX_PACKET_LEN);
 293
 294                 hdr = skb_vnet_hdr(skb);
 295                 sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
 296
 297                 if (vi->big_packets) {
 298                         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 299                                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 300                                 f->page = get_a_page(vi, gfp);
 301                                 if (!f->page)
 302                                         break;
 303
 304                                 f->page_offset = 0;
 305                                 f->size = PAGE_SIZE;
 306
 307                                 skb->data_len += PAGE_SIZE;
 308                                 skb->len += PAGE_SIZE;
 309
 310                                 skb_shinfo(skb)->nr_frags++;
 311                         }
 312                 }
 313
 314                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 315                 skb_queue_head(&vi->recv, skb);
 316
 317                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 318                 if (err < 0) {
 319                         skb_unlink(skb, &vi->recv);
 320                         trim_pages(vi, skb);
 321                         kfree_skb(skb);
 322                         break;
 323                 }
 324                 vi->num++;
 325         } while (err >= num);
 326         if (unlikely(vi->num > vi->max))
 327                 vi->max = vi->num;
 328         vi->rvq->vq_ops->kick(vi->rvq);
 329         return !oom;
 330 }
 331
 332 /* Returns false if we couldn't fill entirely (OOM). */
 333 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
 334 {
 335         struct sk_buff *skb;
 336         struct scatterlist sg[1];
 337         int err;
 338         bool oom = false;
 339
 340         if (!vi->mergeable_rx_bufs)
 341                 return try_fill_recv_maxbufs(vi, gfp);
 342
 343         do {
 344                 skb_frag_t *f;
 345
 346                 skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
 347                 if (unlikely(!skb)) {
 348                         oom = true;
 349                         break;
 350                 }
 351
 352                 skb_reserve(skb, NET_IP_ALIGN);
 353
 354                 f = &skb_shinfo(skb)->frags[0];
 355                 f->page = get_a_page(vi, gfp);
 356                 if (!f->page) {
 357                         oom = true;
 358                         kfree_skb(skb);
 359                         break;
 360                 }
 361
 362                 f->page_offset = 0;
 363                 f->size = PAGE_SIZE;
 364
 365                 skb_shinfo(skb)->nr_frags++;
 366
 367                 sg_init_one(sg, page_address(f->page), PAGE_SIZE);
 368                 skb_queue_head(&vi->recv, skb);
 369
 370                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
 371                 if (err < 0) {
 372                         skb_unlink(skb, &vi->recv);
 373                         kfree_skb(skb);
 374                         break;
 375                 }
 376                 vi->num++;
 377         } while (err > 0);
 378         if (unlikely(vi->num > vi->max))
 379                 vi->max = vi->num;
 380         vi->rvq->vq_ops->kick(vi->rvq);
 381         return !oom;
 382 }
 383
 384 static void skb_recv_done(struct virtqueue *rvq)
 385 {
 386         struct virtnet_info *vi = rvq->vdev->priv;
 387         /* Schedule NAPI, Suppress further interrupts if successful. */
 388         if (napi_schedule_prep(&vi->napi)) {
 389                 rvq->vq_ops->disable_cb(rvq);
 390                 __napi_schedule(&vi->napi);
 391         }
 392 }
 393
 394 static void refill_work(struct work_struct *work)
 395 {
 396         struct virtnet_info *vi;
 397         bool still_empty;
 398
 399         vi = container_of(work, struct virtnet_info, refill.work);
 400         napi_disable(&vi->napi);
 401         still_empty = !try_fill_recv(vi, GFP_KERNEL);
 402         napi_enable(&vi->napi);
 403
 404         /* In theory, this can happen: if we don't get any buffers in
 405          * we will *never* try to fill again. */
 406         if (still_empty)
 407                 schedule_delayed_work(&vi->refill, HZ/2);
 408 }
 409
 410 static int virtnet_poll(struct napi_struct *napi, int budget)
 411 {
 412         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 413         struct sk_buff *skb = NULL;
 414         unsigned int len, received = 0;
 415
 416 again:
 417         while (received < budget &&
 418                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 419                 __skb_unlink(skb, &vi->recv);
 420                 receive_skb(vi->dev, skb, len);
 421                 vi->num--;
 422                 received++;
 423         }
 424
 425         if (vi->num < vi->max / 2) {
 426                 if (!try_fill_recv(vi, GFP_ATOMIC))
 427                         schedule_delayed_work(&vi->refill, 0);
 428         }
 429
 430         /* Out of packets? */
 431         if (received < budget) {
 432                 napi_complete(napi);
 433                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 434                     && napi_schedule_prep(napi)) {
 435                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 436                         __napi_schedule(napi);
 437                         goto again;
 438                 }
 439         }
 440
 441         return received;
 442 }
 443
 444 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
 445 {
 446         struct sk_buff *skb;
 447         unsigned int len, tot_sgs = 0;
 448
 449         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 450                 pr_debug("Sent skb %p\n", skb);
 451                 __skb_unlink(skb, &vi->send);
 452                 vi->dev->stats.tx_bytes += skb->len;
 453                 vi->dev->stats.tx_packets++;
 454                 tot_sgs += skb_vnet_hdr(skb)->num_sg;
 455                 dev_kfree_skb_any(skb);
 456         }
 457         return tot_sgs;
 458 }
 459
 460 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 461 {
 462         struct scatterlist sg[2+MAX_SKB_FRAGS];
 463         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 464         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 465
 466         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 467
 468         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 469
 470         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 471                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 472                 hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
 473                 hdr->hdr.csum_offset = skb->csum_offset;
 474         } else {
 475                 hdr->hdr.flags = 0;
 476                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
 477         }
 478
 479         if (skb_is_gso(skb)) {
 480                 hdr->hdr.hdr_len = skb_headlen(skb);
 481                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
 482                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 483                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 484                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 485                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 486                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 487                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 488                 else
 489                         BUG();
 490                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 491                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 492         } else {
 493                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 494                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
 495         }
 496
 497         hdr->mhdr.num_buffers = 0;
 498
 499         /* Encode metadata header at front. */
 500         if (vi->mergeable_rx_bufs)
 501                 sg_set_buf(sg, &hdr->mhdr, sizeof(hdr->mhdr));
 502         else
 503                 sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
 504
 505         hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 506         return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
 507 }
 508
 509 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 510 {
 511         struct virtnet_info *vi = netdev_priv(dev);
 512         int capacity;
 513
 514 again:
 515         /* Free up any pending old buffers before queueing new ones. */
 516         free_old_xmit_skbs(vi);
 517
 518         /* Try to transmit */
 519         capacity = xmit_skb(vi, skb);
 520
 521         /* This can happen with OOM and indirect buffers. */
 522         if (unlikely(capacity < 0)) {
 523                 netif_stop_queue(dev);
 524                 dev_warn(&dev->dev, "Unexpected full queue\n");
 525                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 526                         vi->svq->vq_ops->disable_cb(vi->svq);
 527                         netif_start_queue(dev);
 528                         goto again;
 529                 }
 530                 return NETDEV_TX_BUSY;
 531         }
 532         vi->svq->vq_ops->kick(vi->svq);
 533
 534         /*
 535          * Put new one in send queue.  You'd expect we'd need this before
 536          * xmit_skb calls add_buf(), since the callback can be triggered
 537          * immediately after that.  But since the callback just triggers
 538          * another call back here, normal network xmit locking prevents the
 539          * race.
 540          */
 541         __skb_queue_head(&vi->send, skb);
 542
 543         /* Don't wait up for transmitted skbs to be freed. */
 544         skb_orphan(skb);
 545         nf_reset(skb);
 546
 547         /* Apparently nice girls don't return TX_BUSY; stop the queue
 548          * before it gets out of hand.  Naturally, this wastes entries. */
 549         if (capacity < 2+MAX_SKB_FRAGS) {
 550                 netif_stop_queue(dev);
 551                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 552                         /* More just got used, free them then recheck. */
 553                         capacity += free_old_xmit_skbs(vi);
 554                         if (capacity >= 2+MAX_SKB_FRAGS) {
 555                                 netif_start_queue(dev);
 556                                 vi->svq->vq_ops->disable_cb(vi->svq);
 557                         }
 558                 }
 559         }
 560
 561         return NETDEV_TX_OK;
 562 }
 563
 564 static int virtnet_set_mac_address(struct net_device *dev, void *p)
 565 {
 566         struct virtnet_info *vi = netdev_priv(dev);
 567         struct virtio_device *vdev = vi->vdev;
 568         int ret;
 569
 570         ret = eth_mac_addr(dev, p);
 571         if (ret)
 572                 return ret;
 573
 574         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
 575                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 576                                   dev->dev_addr, dev->addr_len);
 577
 578         return 0;
 579 }
 580
 581 #ifdef CONFIG_NET_POLL_CONTROLLER
 582 static void virtnet_netpoll(struct net_device *dev)
 583 {
 584         struct virtnet_info *vi = netdev_priv(dev);
 585
 586         napi_schedule(&vi->napi);
 587 }
 588 #endif
 589
 590 static int virtnet_open(struct net_device *dev)
 591 {
 592         struct virtnet_info *vi = netdev_priv(dev);
 593
 594         napi_enable(&vi->napi);
 595
 596         /* If all buffers were filled by other side before we napi_enabled, we
 597          * won't get another interrupt, so process any outstanding packets
 598          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 599          * We synchronize against interrupts via NAPI_STATE_SCHED */
 600         if (napi_schedule_prep(&vi->napi)) {
 601                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 602                 __napi_schedule(&vi->napi);
 603         }
 604         return 0;
 605 }
 606
 607 /*
 608  * Send command via the control virtqueue and check status.  Commands
 609  * supported by the hypervisor, as indicated by feature bits, should
 610  * never fail unless improperly formated.
 611  */
 612 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 613                                  struct scatterlist *data, int out, int in)
 614 {
 615         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 616         struct virtio_net_ctrl_hdr ctrl;
 617         virtio_net_ctrl_ack status = ~0;
 618         unsigned int tmp;
 619         int i;
 620
 621         /* Caller should know better */
 622         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 623                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 624
 625         out++; /* Add header */
 626         in++; /* Add return status */
 627
 628         ctrl.class = class;
 629         ctrl.cmd = cmd;
 630
 631         sg_init_table(sg, out + in);
 632
 633         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 634         for_each_sg(data, s, out + in - 2, i)
 635                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 636         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 637
 638         BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
 639
 640         vi->cvq->vq_ops->kick(vi->cvq);
 641
 642         /*
 643          * Spin for a response, the kick causes an ioport write, trapping
 644          * into the hypervisor, so the request should be handled immediately.
 645          */
 646         while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
 647                 cpu_relax();
 648
 649         return status == VIRTIO_NET_OK;
 650 }
 651
 652 static int virtnet_close(struct net_device *dev)
 653 {
 654         struct virtnet_info *vi = netdev_priv(dev);
 655
 656         napi_disable(&vi->napi);
 657
 658         return 0;
 659 }
 660
 661 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 662 {
 663         struct virtnet_info *vi = netdev_priv(dev);
 664         struct virtio_device *vdev = vi->vdev;
 665
 666         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 667                 return -ENOSYS;
 668
 669         return ethtool_op_set_tx_hw_csum(dev, data);
 670 }
 671
 672 static void virtnet_set_rx_mode(struct net_device *dev)
 673 {
 674         struct virtnet_info *vi = netdev_priv(dev);
 675         struct scatterlist sg[2];
 676         u8 promisc, allmulti;
 677         struct virtio_net_ctrl_mac *mac_data;
 678         struct dev_addr_list *addr;
 679         struct netdev_hw_addr *ha;
 680         void *buf;
 681         int i;
 682
 683         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 684         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 685                 return;
 686
 687         promisc = ((dev->flags & IFF_PROMISC) != 0);
 688         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 689
 690         sg_init_one(sg, &promisc, sizeof(promisc));
 691
 692         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 693                                   VIRTIO_NET_CTRL_RX_PROMISC,
 694                                   sg, 1, 0))
 695                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 696                          promisc ? "en" : "dis");
 697
 698         sg_init_one(sg, &allmulti, sizeof(allmulti));
 699
 700         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 701                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 702                                   sg, 1, 0))
 703                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 704                          allmulti ? "en" : "dis");
 705
 706         /* MAC filter - use one buffer for both lists */
 707         mac_data = buf = kzalloc(((dev->uc.count + dev->mc_count) * ETH_ALEN) +
 708                                  (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 709         if (!buf) {
 710                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 711                 return;
 712         }
 713
 714         sg_init_table(sg, 2);
 715
 716         /* Store the unicast list and count in the front of the buffer */
 717         mac_data->entries = dev->uc.count;
 718         i = 0;
 719         list_for_each_entry(ha, &dev->uc.list, list)
 720                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 721
 722         sg_set_buf(&sg[0], mac_data,
 723                    sizeof(mac_data->entries) + (dev->uc.count * ETH_ALEN));
 724
 725         /* multicast list and count fill the end */
 726         mac_data = (void *)&mac_data->macs[dev->uc.count][0];
 727
 728         mac_data->entries = dev->mc_count;
 729         addr = dev->mc_list;
 730         for (i = 0; i < dev->mc_count; i++, addr = addr->next)
 731                 memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
 732
 733         sg_set_buf(&sg[1], mac_data,
 734                    sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
 735
 736         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 737                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 738                                   sg, 2, 0))
 739                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 740
 741         kfree(buf);
 742 }
 743
 744 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
 745 {
 746         struct virtnet_info *vi = netdev_priv(dev);
 747         struct scatterlist sg;
 748
 749         sg_init_one(&sg, &vid, sizeof(vid));
 750
 751         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 752                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
 753                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 754 }
 755
 756 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 757 {
 758         struct virtnet_info *vi = netdev_priv(dev);
 759         struct scatterlist sg;
 760
 761         sg_init_one(&sg, &vid, sizeof(vid));
 762
 763         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 764                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 765                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 766 }
 767
 768 static const struct ethtool_ops virtnet_ethtool_ops = {
 769         .set_tx_csum = virtnet_set_tx_csum,
 770         .set_sg = ethtool_op_set_sg,
 771         .set_tso = ethtool_op_set_tso,
 772         .set_ufo = ethtool_op_set_ufo,
 773         .get_link = ethtool_op_get_link,
 774 };
 775
 776 #define MIN_MTU 68
 777 #define MAX_MTU 65535
 778
 779 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 780 {
 781         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 782                 return -EINVAL;
 783         dev->mtu = new_mtu;
 784         return 0;
 785 }
 786
 787 static const struct net_device_ops virtnet_netdev = {
 788         .ndo_open            = virtnet_open,
 789         .ndo_stop            = virtnet_close,
 790         .ndo_start_xmit      = start_xmit,
 791         .ndo_validate_addr   = eth_validate_addr,
 792         .ndo_set_mac_address = virtnet_set_mac_address,
 793         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 794         .ndo_change_mtu      = virtnet_change_mtu,
 795         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 796         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 797 #ifdef CONFIG_NET_POLL_CONTROLLER
 798         .ndo_poll_controller = virtnet_netpoll,
 799 #endif
 800 };
 801
 802 static void virtnet_update_status(struct virtnet_info *vi)
 803 {
 804         u16 v;
 805
 806         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
 807                 return;
 808
 809         vi->vdev->config->get(vi->vdev,
 810                               offsetof(struct virtio_net_config, status),
 811                               &v, sizeof(v));
 812
 813         /* Ignore unknown (future) status bits */
 814         v &= VIRTIO_NET_S_LINK_UP;
 815
 816         if (vi->status == v)
 817                 return;
 818
 819         vi->status = v;
 820
 821         if (vi->status & VIRTIO_NET_S_LINK_UP) {
 822                 netif_carrier_on(vi->dev);
 823                 netif_wake_queue(vi->dev);
 824         } else {
 825                 netif_carrier_off(vi->dev);
 826                 netif_stop_queue(vi->dev);
 827         }
 828 }
 829
 830 static void virtnet_config_changed(struct virtio_device *vdev)
 831 {
 832         struct virtnet_info *vi = vdev->priv;
 833
 834         virtnet_update_status(vi);
 835 }
 836
 837 static int virtnet_probe(struct virtio_device *vdev)
 838 {
 839         int err;
 840         struct net_device *dev;
 841         struct virtnet_info *vi;
 842         struct virtqueue *vqs[3];
 843         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
 844         const char *names[] = { "input", "output", "control" };
 845         int nvqs;
 846
 847         /* Allocate ourselves a network device with room for our info */
 848         dev = alloc_etherdev(sizeof(struct virtnet_info));
 849         if (!dev)
 850                 return -ENOMEM;
 851
 852         /* Set up network device as normal. */
 853         dev->netdev_ops = &virtnet_netdev;
 854         dev->features = NETIF_F_HIGHDMA;
 855         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 856         SET_NETDEV_DEV(dev, &vdev->dev);
 857
 858         /* Do we support "hardware" checksums? */
 859         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 860                 /* This opens up the world of extra features. */
 861                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 862                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 863                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 864                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 865                 }
 866                 /* Individual feature bits: what can host handle? */
 867                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 868                         dev->features |= NETIF_F_TSO;
 869                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 870                         dev->features |= NETIF_F_TSO6;
 871                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 872                         dev->features |= NETIF_F_TSO_ECN;
 873                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 874                         dev->features |= NETIF_F_UFO;
 875         }
 876
 877         /* Configuration may specify what MAC to use.  Otherwise random. */
 878         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 879                 vdev->config->get(vdev,
 880                                   offsetof(struct virtio_net_config, mac),
 881                                   dev->dev_addr, dev->addr_len);
 882         } else
 883                 random_ether_addr(dev->dev_addr);
 884
 885         /* Set up our device-specific information */
 886         vi = netdev_priv(dev);
 887         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 888         vi->dev = dev;
 889         vi->vdev = vdev;
 890         vdev->priv = vi;
 891         vi->pages = NULL;
 892         INIT_DELAYED_WORK(&vi->refill, refill_work);
 893
 894         /* If we can receive ANY GSO packets, we must allocate large ones. */
 895         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 896             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 897             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 898                 vi->big_packets = true;
 899
 900         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 901                 vi->mergeable_rx_bufs = true;
 902
 903         /* We expect two virtqueues, receive then send,
 904          * and optionally control. */
 905         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
 906
 907         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
 908         if (err)
 909                 goto free;
 910
 911         vi->rvq = vqs[0];
 912         vi->svq = vqs[1];
 913
 914         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
 915                 vi->cvq = vqs[2];
 916
 917                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
 918                         dev->features |= NETIF_F_HW_VLAN_FILTER;
 919         }
 920
 921         /* Initialize our empty receive and send queues. */
 922         skb_queue_head_init(&vi->recv);
 923         skb_queue_head_init(&vi->send);
 924
 925         err = register_netdev(dev);
 926         if (err) {
 927                 pr_debug("virtio_net: registering device failed\n");
 928                 goto free_vqs;
 929         }
 930
 931         /* Last of all, set up some receive buffers. */
 932         try_fill_recv(vi, GFP_KERNEL);
 933
 934         /* If we didn't even get one input buffer, we're useless. */
 935         if (vi->num == 0) {
 936                 err = -ENOMEM;
 937                 goto unregister;
 938         }
 939
 940         vi->status = VIRTIO_NET_S_LINK_UP;
 941         virtnet_update_status(vi);
 942         netif_carrier_on(dev);
 943
 944         pr_debug("virtnet: registered device %s\n", dev->name);
 945         return 0;
 946
 947 unregister:
 948         unregister_netdev(dev);
 949         cancel_delayed_work_sync(&vi->refill);
 950 free_vqs:
 951         vdev->config->del_vqs(vdev);
 952 free:
 953         free_netdev(dev);
 954         return err;
 955 }
 956
 957 static void __devexit virtnet_remove(struct virtio_device *vdev)
 958 {
 959         struct virtnet_info *vi = vdev->priv;
 960         struct sk_buff *skb;
 961
 962         /* Stop all the virtqueues. */
 963         vdev->config->reset(vdev);
 964
 965         /* Free our skbs in send and recv queues, if any. */
 966         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 967                 kfree_skb(skb);
 968                 vi->num--;
 969         }
 970         __skb_queue_purge(&vi->send);
 971
 972         BUG_ON(vi->num != 0);
 973
 974         unregister_netdev(vi->dev);
 975         cancel_delayed_work_sync(&vi->refill);
 976
 977         vdev->config->del_vqs(vi->vdev);
 978
 979         while (vi->pages)
 980                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
 981
 982         free_netdev(vi->dev);
 983 }
 984
 985 static struct virtio_device_id id_table[] = {
 986         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
 987         { 0 },
 988 };
 989
 990 static unsigned int features[] = {
 991         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
 992         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
 993         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
 994         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
 995         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
 996         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
 997         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
 998 };
 999
1000 static struct virtio_driver virtio_net_driver = {
1001         .feature_table = features,
1002         .feature_table_size = ARRAY_SIZE(features),
1003         .driver.name =  KBUILD_MODNAME,
1004         .driver.owner = THIS_MODULE,
1005         .id_table =     id_table,
1006         .probe =        virtnet_probe,
1007         .remove =       __devexit_p(virtnet_remove),
1008         .config_changed = virtnet_config_changed,
1009 };
1010
1011 static int __init init(void)
1012 {
1013         return register_virtio_driver(&virtio_net_driver);
1014 }
1015
1016 static void __exit fini(void)
1017 {
1018         unregister_virtio_driver(&virtio_net_driver);
1019 }
1020 module_init(init);
1021 module_exit(fini);
1022
1023 MODULE_DEVICE_TABLE(virtio, id_table);
1024 MODULE_DESCRIPTION("Virtio network driver");
1025 MODULE_LICENSE("GPL");