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         try_fill_recv(vi, GFP_KERNEL);
 402         still_empty = (vi->num == 0);
 403         napi_enable(&vi->napi);
 404
 405         /* In theory, this can happen: if we don't get any buffers in
 406          * we will *never* try to fill again. */
 407         if (still_empty)
 408                 schedule_delayed_work(&vi->refill, HZ/2);
 409 }
 410
 411 static int virtnet_poll(struct napi_struct *napi, int budget)
 412 {
 413         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 414         struct sk_buff *skb = NULL;
 415         unsigned int len, received = 0;
 416
 417 again:
 418         while (received < budget &&
 419                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 420                 __skb_unlink(skb, &vi->recv);
 421                 receive_skb(vi->dev, skb, len);
 422                 vi->num--;
 423                 received++;
 424         }
 425
 426         if (vi->num < vi->max / 2) {
 427                 if (!try_fill_recv(vi, GFP_ATOMIC))
 428                         schedule_delayed_work(&vi->refill, 0);
 429         }
 430
 431         /* Out of packets? */
 432         if (received < budget) {
 433                 napi_complete(napi);
 434                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 435                     && napi_schedule_prep(napi)) {
 436                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 437                         __napi_schedule(napi);
 438                         goto again;
 439                 }
 440         }
 441
 442         return received;
 443 }
 444
 445 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
 446 {
 447         struct sk_buff *skb;
 448         unsigned int len, tot_sgs = 0;
 449
 450         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 451                 pr_debug("Sent skb %p\n", skb);
 452                 __skb_unlink(skb, &vi->send);
 453                 vi->dev->stats.tx_bytes += skb->len;
 454                 vi->dev->stats.tx_packets++;
 455                 tot_sgs += skb_vnet_hdr(skb)->num_sg;
 456                 dev_kfree_skb_any(skb);
 457         }
 458         return tot_sgs;
 459 }
 460
 461 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 462 {
 463         struct scatterlist sg[2+MAX_SKB_FRAGS];
 464         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 465         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 466
 467         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 468
 469         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 470
 471         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 472                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 473                 hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
 474                 hdr->hdr.csum_offset = skb->csum_offset;
 475         } else {
 476                 hdr->hdr.flags = 0;
 477                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
 478         }
 479
 480         if (skb_is_gso(skb)) {
 481                 hdr->hdr.hdr_len = skb_headlen(skb);
 482                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
 483                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 484                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 485                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 486                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 487                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 488                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 489                 else
 490                         BUG();
 491                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 492                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 493         } else {
 494                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 495                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
 496         }
 497
 498         hdr->mhdr.num_buffers = 0;
 499
 500         /* Encode metadata header at front. */
 501         if (vi->mergeable_rx_bufs)
 502                 sg_set_buf(sg, &hdr->mhdr, sizeof(hdr->mhdr));
 503         else
 504                 sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
 505
 506         hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 507         return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
 508 }
 509
 510 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 511 {
 512         struct virtnet_info *vi = netdev_priv(dev);
 513         int capacity;
 514
 515 again:
 516         /* Free up any pending old buffers before queueing new ones. */
 517         free_old_xmit_skbs(vi);
 518
 519         /* Try to transmit */
 520         capacity = xmit_skb(vi, skb);
 521
 522         /* This can happen with OOM and indirect buffers. */
 523         if (unlikely(capacity < 0)) {
 524                 netif_stop_queue(dev);
 525                 dev_warn(&dev->dev, "Unexpected full queue\n");
 526                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 527                         vi->svq->vq_ops->disable_cb(vi->svq);
 528                         netif_start_queue(dev);
 529                         goto again;
 530                 }
 531                 return NETDEV_TX_BUSY;
 532         }
 533         vi->svq->vq_ops->kick(vi->svq);
 534
 535         /*
 536          * Put new one in send queue.  You'd expect we'd need this before
 537          * xmit_skb calls add_buf(), since the callback can be triggered
 538          * immediately after that.  But since the callback just triggers
 539          * another call back here, normal network xmit locking prevents the
 540          * race.
 541          */
 542         __skb_queue_head(&vi->send, skb);
 543
 544         /* Don't wait up for transmitted skbs to be freed. */
 545         skb_orphan(skb);
 546         nf_reset(skb);
 547
 548         /* Apparently nice girls don't return TX_BUSY; stop the queue
 549          * before it gets out of hand.  Naturally, this wastes entries. */
 550         if (capacity < 2+MAX_SKB_FRAGS) {
 551                 netif_stop_queue(dev);
 552                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 553                         /* More just got used, free them then recheck. */
 554                         capacity += free_old_xmit_skbs(vi);
 555                         if (capacity >= 2+MAX_SKB_FRAGS) {
 556                                 netif_start_queue(dev);
 557                                 vi->svq->vq_ops->disable_cb(vi->svq);
 558                         }
 559                 }
 560         }
 561
 562         return NETDEV_TX_OK;
 563 }
 564
 565 static int virtnet_set_mac_address(struct net_device *dev, void *p)
 566 {
 567         struct virtnet_info *vi = netdev_priv(dev);
 568         struct virtio_device *vdev = vi->vdev;
 569         int ret;
 570
 571         ret = eth_mac_addr(dev, p);
 572         if (ret)
 573                 return ret;
 574
 575         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
 576                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 577                                   dev->dev_addr, dev->addr_len);
 578
 579         return 0;
 580 }
 581
 582 #ifdef CONFIG_NET_POLL_CONTROLLER
 583 static void virtnet_netpoll(struct net_device *dev)
 584 {
 585         struct virtnet_info *vi = netdev_priv(dev);
 586
 587         napi_schedule(&vi->napi);
 588 }
 589 #endif
 590
 591 static int virtnet_open(struct net_device *dev)
 592 {
 593         struct virtnet_info *vi = netdev_priv(dev);
 594
 595         napi_enable(&vi->napi);
 596
 597         /* If all buffers were filled by other side before we napi_enabled, we
 598          * won't get another interrupt, so process any outstanding packets
 599          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 600          * We synchronize against interrupts via NAPI_STATE_SCHED */
 601         if (napi_schedule_prep(&vi->napi)) {
 602                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 603                 __napi_schedule(&vi->napi);
 604         }
 605         return 0;
 606 }
 607
 608 /*
 609  * Send command via the control virtqueue and check status.  Commands
 610  * supported by the hypervisor, as indicated by feature bits, should
 611  * never fail unless improperly formated.
 612  */
 613 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 614                                  struct scatterlist *data, int out, int in)
 615 {
 616         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 617         struct virtio_net_ctrl_hdr ctrl;
 618         virtio_net_ctrl_ack status = ~0;
 619         unsigned int tmp;
 620         int i;
 621
 622         /* Caller should know better */
 623         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 624                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 625
 626         out++; /* Add header */
 627         in++; /* Add return status */
 628
 629         ctrl.class = class;
 630         ctrl.cmd = cmd;
 631
 632         sg_init_table(sg, out + in);
 633
 634         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 635         for_each_sg(data, s, out + in - 2, i)
 636                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 637         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 638
 639         BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
 640
 641         vi->cvq->vq_ops->kick(vi->cvq);
 642
 643         /*
 644          * Spin for a response, the kick causes an ioport write, trapping
 645          * into the hypervisor, so the request should be handled immediately.
 646          */
 647         while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
 648                 cpu_relax();
 649
 650         return status == VIRTIO_NET_OK;
 651 }
 652
 653 static int virtnet_close(struct net_device *dev)
 654 {
 655         struct virtnet_info *vi = netdev_priv(dev);
 656
 657         napi_disable(&vi->napi);
 658
 659         return 0;
 660 }
 661
 662 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 663 {
 664         struct virtnet_info *vi = netdev_priv(dev);
 665         struct virtio_device *vdev = vi->vdev;
 666
 667         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 668                 return -ENOSYS;
 669
 670         return ethtool_op_set_tx_hw_csum(dev, data);
 671 }
 672
 673 static void virtnet_set_rx_mode(struct net_device *dev)
 674 {
 675         struct virtnet_info *vi = netdev_priv(dev);
 676         struct scatterlist sg[2];
 677         u8 promisc, allmulti;
 678         struct virtio_net_ctrl_mac *mac_data;
 679         struct dev_addr_list *addr;
 680         struct netdev_hw_addr *ha;
 681         void *buf;
 682         int i;
 683
 684         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 685         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 686                 return;
 687
 688         promisc = ((dev->flags & IFF_PROMISC) != 0);
 689         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 690
 691         sg_init_one(sg, &promisc, sizeof(promisc));
 692
 693         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 694                                   VIRTIO_NET_CTRL_RX_PROMISC,
 695                                   sg, 1, 0))
 696                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 697                          promisc ? "en" : "dis");
 698
 699         sg_init_one(sg, &allmulti, sizeof(allmulti));
 700
 701         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 702                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 703                                   sg, 1, 0))
 704                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 705                          allmulti ? "en" : "dis");
 706
 707         /* MAC filter - use one buffer for both lists */
 708         mac_data = buf = kzalloc(((dev->uc.count + dev->mc_count) * ETH_ALEN) +
 709                                  (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 710         if (!buf) {
 711                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 712                 return;
 713         }
 714
 715         sg_init_table(sg, 2);
 716
 717         /* Store the unicast list and count in the front of the buffer */
 718         mac_data->entries = dev->uc.count;
 719         i = 0;
 720         list_for_each_entry(ha, &dev->uc.list, list)
 721                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 722
 723         sg_set_buf(&sg[0], mac_data,
 724                    sizeof(mac_data->entries) + (dev->uc.count * ETH_ALEN));
 725
 726         /* multicast list and count fill the end */
 727         mac_data = (void *)&mac_data->macs[dev->uc.count][0];
 728
 729         mac_data->entries = dev->mc_count;
 730         addr = dev->mc_list;
 731         for (i = 0; i < dev->mc_count; i++, addr = addr->next)
 732                 memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
 733
 734         sg_set_buf(&sg[1], mac_data,
 735                    sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
 736
 737         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 738                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 739                                   sg, 2, 0))
 740                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 741
 742         kfree(buf);
 743 }
 744
 745 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
 746 {
 747         struct virtnet_info *vi = netdev_priv(dev);
 748         struct scatterlist sg;
 749
 750         sg_init_one(&sg, &vid, sizeof(vid));
 751
 752         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 753                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
 754                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 755 }
 756
 757 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 758 {
 759         struct virtnet_info *vi = netdev_priv(dev);
 760         struct scatterlist sg;
 761
 762         sg_init_one(&sg, &vid, sizeof(vid));
 763
 764         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 765                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 766                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 767 }
 768
 769 static const struct ethtool_ops virtnet_ethtool_ops = {
 770         .set_tx_csum = virtnet_set_tx_csum,
 771         .set_sg = ethtool_op_set_sg,
 772         .set_tso = ethtool_op_set_tso,
 773         .set_ufo = ethtool_op_set_ufo,
 774         .get_link = ethtool_op_get_link,
 775 };
 776
 777 #define MIN_MTU 68
 778 #define MAX_MTU 65535
 779
 780 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 781 {
 782         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 783                 return -EINVAL;
 784         dev->mtu = new_mtu;
 785         return 0;
 786 }
 787
 788 static const struct net_device_ops virtnet_netdev = {
 789         .ndo_open            = virtnet_open,
 790         .ndo_stop            = virtnet_close,
 791         .ndo_start_xmit      = start_xmit,
 792         .ndo_validate_addr   = eth_validate_addr,
 793         .ndo_set_mac_address = virtnet_set_mac_address,
 794         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 795         .ndo_change_mtu      = virtnet_change_mtu,
 796         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 797         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 798 #ifdef CONFIG_NET_POLL_CONTROLLER
 799         .ndo_poll_controller = virtnet_netpoll,
 800 #endif
 801 };
 802
 803 static void virtnet_update_status(struct virtnet_info *vi)
 804 {
 805         u16 v;
 806
 807         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
 808                 return;
 809
 810         vi->vdev->config->get(vi->vdev,
 811                               offsetof(struct virtio_net_config, status),
 812                               &v, sizeof(v));
 813
 814         /* Ignore unknown (future) status bits */
 815         v &= VIRTIO_NET_S_LINK_UP;
 816
 817         if (vi->status == v)
 818                 return;
 819
 820         vi->status = v;
 821
 822         if (vi->status & VIRTIO_NET_S_LINK_UP) {
 823                 netif_carrier_on(vi->dev);
 824                 netif_wake_queue(vi->dev);
 825         } else {
 826                 netif_carrier_off(vi->dev);
 827                 netif_stop_queue(vi->dev);
 828         }
 829 }
 830
 831 static void virtnet_config_changed(struct virtio_device *vdev)
 832 {
 833         struct virtnet_info *vi = vdev->priv;
 834
 835         virtnet_update_status(vi);
 836 }
 837
 838 static int virtnet_probe(struct virtio_device *vdev)
 839 {
 840         int err;
 841         struct net_device *dev;
 842         struct virtnet_info *vi;
 843         struct virtqueue *vqs[3];
 844         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
 845         const char *names[] = { "input", "output", "control" };
 846         int nvqs;
 847
 848         /* Allocate ourselves a network device with room for our info */
 849         dev = alloc_etherdev(sizeof(struct virtnet_info));
 850         if (!dev)
 851                 return -ENOMEM;
 852
 853         /* Set up network device as normal. */
 854         dev->netdev_ops = &virtnet_netdev;
 855         dev->features = NETIF_F_HIGHDMA;
 856         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 857         SET_NETDEV_DEV(dev, &vdev->dev);
 858
 859         /* Do we support "hardware" checksums? */
 860         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 861                 /* This opens up the world of extra features. */
 862                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 863                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 864                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 865                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 866                 }
 867                 /* Individual feature bits: what can host handle? */
 868                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 869                         dev->features |= NETIF_F_TSO;
 870                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 871                         dev->features |= NETIF_F_TSO6;
 872                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 873                         dev->features |= NETIF_F_TSO_ECN;
 874                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 875                         dev->features |= NETIF_F_UFO;
 876         }
 877
 878         /* Configuration may specify what MAC to use.  Otherwise random. */
 879         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 880                 vdev->config->get(vdev,
 881                                   offsetof(struct virtio_net_config, mac),
 882                                   dev->dev_addr, dev->addr_len);
 883         } else
 884                 random_ether_addr(dev->dev_addr);
 885
 886         /* Set up our device-specific information */
 887         vi = netdev_priv(dev);
 888         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 889         vi->dev = dev;
 890         vi->vdev = vdev;
 891         vdev->priv = vi;
 892         vi->pages = NULL;
 893         INIT_DELAYED_WORK(&vi->refill, refill_work);
 894
 895         /* If we can receive ANY GSO packets, we must allocate large ones. */
 896         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 897             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 898             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 899                 vi->big_packets = true;
 900
 901         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 902                 vi->mergeable_rx_bufs = true;
 903
 904         /* We expect two virtqueues, receive then send,
 905          * and optionally control. */
 906         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
 907
 908         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
 909         if (err)
 910                 goto free;
 911
 912         vi->rvq = vqs[0];
 913         vi->svq = vqs[1];
 914
 915         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
 916                 vi->cvq = vqs[2];
 917
 918                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
 919                         dev->features |= NETIF_F_HW_VLAN_FILTER;
 920         }
 921
 922         /* Initialize our empty receive and send queues. */
 923         skb_queue_head_init(&vi->recv);
 924         skb_queue_head_init(&vi->send);
 925
 926         err = register_netdev(dev);
 927         if (err) {
 928                 pr_debug("virtio_net: registering device failed\n");
 929                 goto free_vqs;
 930         }
 931
 932         /* Last of all, set up some receive buffers. */
 933         try_fill_recv(vi, GFP_KERNEL);
 934
 935         /* If we didn't even get one input buffer, we're useless. */
 936         if (vi->num == 0) {
 937                 err = -ENOMEM;
 938                 goto unregister;
 939         }
 940
 941         vi->status = VIRTIO_NET_S_LINK_UP;
 942         virtnet_update_status(vi);
 943         netif_carrier_on(dev);
 944
 945         pr_debug("virtnet: registered device %s\n", dev->name);
 946         return 0;
 947
 948 unregister:
 949         unregister_netdev(dev);
 950         cancel_delayed_work_sync(&vi->refill);
 951 free_vqs:
 952         vdev->config->del_vqs(vdev);
 953 free:
 954         free_netdev(dev);
 955         return err;
 956 }
 957
 958 static void __devexit virtnet_remove(struct virtio_device *vdev)
 959 {
 960         struct virtnet_info *vi = vdev->priv;
 961         struct sk_buff *skb;
 962
 963         /* Stop all the virtqueues. */
 964         vdev->config->reset(vdev);
 965
 966         /* Free our skbs in send and recv queues, if any. */
 967         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 968                 kfree_skb(skb);
 969                 vi->num--;
 970         }
 971         __skb_queue_purge(&vi->send);
 972
 973         BUG_ON(vi->num != 0);
 974
 975         unregister_netdev(vi->dev);
 976         cancel_delayed_work_sync(&vi->refill);
 977
 978         vdev->config->del_vqs(vi->vdev);
 979
 980         while (vi->pages)
 981                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
 982
 983         free_netdev(vi->dev);
 984 }
 985
 986 static struct virtio_device_id id_table[] = {
 987         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
 988         { 0 },
 989 };
 990
 991 static unsigned int features[] = {
 992         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
 993         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
 994         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
 995         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
 996         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
 997         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
 998         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
 999 };
1000
1001 static struct virtio_driver virtio_net = {
1002         .feature_table = features,
1003         .feature_table_size = ARRAY_SIZE(features),
1004         .driver.name =  KBUILD_MODNAME,
1005         .driver.owner = THIS_MODULE,
1006         .id_table =     id_table,
1007         .probe =        virtnet_probe,
1008         .remove =       __devexit_p(virtnet_remove),
1009         .config_changed = virtnet_config_changed,
1010 };
1011
1012 static int __init init(void)
1013 {
1014         return register_virtio_driver(&virtio_net);
1015 }
1016
1017 static void __exit fini(void)
1018 {
1019         unregister_virtio_driver(&virtio_net);
1020 }
1021 module_init(init);
1022 module_exit(fini);
1023
1024 MODULE_DEVICE_TABLE(virtio, id_table);
1025 MODULE_DESCRIPTION("Virtio network driver");
1026 MODULE_LICENSE("GPL");