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/ethtool.h>
  23 #include <linux/module.h>
  24 #include <linux/virtio.h>
  25 #include <linux/virtio_ids.h>
  26 #include <linux/virtio_net.h>
  27 #include <linux/scatterlist.h>
  28 #include <linux/if_vlan.h>
  29
  30 static int napi_weight = 128;
  31 module_param(napi_weight, int, 0444);
  32
  33 static int csum = 1, gso = 1;
  34 module_param(csum, bool, 0444);
  35 module_param(gso, bool, 0444);
  36
  37 /* FIXME: MTU in config. */
  38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
  39 #define GOOD_COPY_LEN   128
  40
  41 #define VIRTNET_SEND_COMMAND_SG_MAX    2
  42
  43 struct virtnet_info
  44 {
  45         struct virtio_device *vdev;
  46         struct virtqueue *rvq, *svq, *cvq;
  47         struct net_device *dev;
  48         struct napi_struct napi;
  49         unsigned int status;
  50
  51         /* If we need to free in a timer, this is it. */
  52         struct timer_list xmit_free_timer;
  53
  54         /* Number of input buffers, and max we've ever had. */
  55         unsigned int num, max;
  56
  57         /* For cleaning up after transmission. */
  58         struct tasklet_struct tasklet;
  59         bool free_in_tasklet;
  60
  61         /* I like... big packets and I cannot lie! */
  62         bool big_packets;
  63
  64         /* Host will merge rx buffers for big packets (shake it! shake it!) */
  65         bool mergeable_rx_bufs;
  66
  67         /* Receive & send queues. */
  68         struct sk_buff_head recv;
  69         struct sk_buff_head send;
  70
  71         /* Work struct for refilling if we run low on memory. */
  72         struct delayed_work refill;
  73
  74         /* Chain pages by the private ptr. */
  75         struct page *pages;
  76 };
  77
  78 static inline void *skb_vnet_hdr(struct sk_buff *skb)
  79 {
  80         return (struct virtio_net_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         if (vi->free_in_tasklet)
 121                 tasklet_schedule(&vi->tasklet);
 122 }
 123
 124 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
 125                         unsigned len)
 126 {
 127         struct virtnet_info *vi = netdev_priv(dev);
 128         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 129         int err;
 130         int i;
 131
 132         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 133                 pr_debug("%s: short packet %i\n", dev->name, len);
 134                 dev->stats.rx_length_errors++;
 135                 goto drop;
 136         }
 137
 138         if (vi->mergeable_rx_bufs) {
 139                 struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 140                 unsigned int copy;
 141                 char *p = page_address(skb_shinfo(skb)->frags[0].page);
 142
 143                 if (len > PAGE_SIZE)
 144                         len = PAGE_SIZE;
 145                 len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
 146
 147                 memcpy(hdr, p, sizeof(*mhdr));
 148                 p += sizeof(*mhdr);
 149
 150                 copy = len;
 151                 if (copy > skb_tailroom(skb))
 152                         copy = skb_tailroom(skb);
 153
 154                 memcpy(skb_put(skb, copy), p, copy);
 155
 156                 len -= copy;
 157
 158                 if (!len) {
 159                         give_a_page(vi, skb_shinfo(skb)->frags[0].page);
 160                         skb_shinfo(skb)->nr_frags--;
 161                 } else {
 162                         skb_shinfo(skb)->frags[0].page_offset +=
 163                                 sizeof(*mhdr) + copy;
 164                         skb_shinfo(skb)->frags[0].size = len;
 165                         skb->data_len += len;
 166                         skb->len += len;
 167                 }
 168
 169                 while (--mhdr->num_buffers) {
 170                         struct sk_buff *nskb;
 171
 172                         i = skb_shinfo(skb)->nr_frags;
 173                         if (i >= MAX_SKB_FRAGS) {
 174                                 pr_debug("%s: packet too long %d\n", dev->name,
 175                                          len);
 176                                 dev->stats.rx_length_errors++;
 177                                 goto drop;
 178                         }
 179
 180                         nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
 181                         if (!nskb) {
 182                                 pr_debug("%s: rx error: %d buffers missing\n",
 183                                          dev->name, mhdr->num_buffers);
 184                                 dev->stats.rx_length_errors++;
 185                                 goto drop;
 186                         }
 187
 188                         __skb_unlink(nskb, &vi->recv);
 189                         vi->num--;
 190
 191                         skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
 192                         skb_shinfo(nskb)->nr_frags = 0;
 193                         kfree_skb(nskb);
 194
 195                         if (len > PAGE_SIZE)
 196                                 len = PAGE_SIZE;
 197
 198                         skb_shinfo(skb)->frags[i].size = len;
 199                         skb_shinfo(skb)->nr_frags++;
 200                         skb->data_len += len;
 201                         skb->len += len;
 202                 }
 203         } else {
 204                 len -= sizeof(struct virtio_net_hdr);
 205
 206                 if (len <= MAX_PACKET_LEN)
 207                         trim_pages(vi, skb);
 208
 209                 err = pskb_trim(skb, len);
 210                 if (err) {
 211                         pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
 212                                  len, err);
 213                         dev->stats.rx_dropped++;
 214                         goto drop;
 215                 }
 216         }
 217
 218         skb->truesize += skb->data_len;
 219         dev->stats.rx_bytes += skb->len;
 220         dev->stats.rx_packets++;
 221
 222         if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 223                 pr_debug("Needs csum!\n");
 224                 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
 225                         goto frame_err;
 226         }
 227
 228         skb->protocol = eth_type_trans(skb, dev);
 229         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 230                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 231
 232         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 233                 pr_debug("GSO!\n");
 234                 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 235                 case VIRTIO_NET_HDR_GSO_TCPV4:
 236                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 237                         break;
 238                 case VIRTIO_NET_HDR_GSO_UDP:
 239                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 240                         break;
 241                 case VIRTIO_NET_HDR_GSO_TCPV6:
 242                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 243                         break;
 244                 default:
 245                         if (net_ratelimit())
 246                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 247                                        dev->name, hdr->gso_type);
 248                         goto frame_err;
 249                 }
 250
 251                 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 252                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 253
 254                 skb_shinfo(skb)->gso_size = hdr->gso_size;
 255                 if (skb_shinfo(skb)->gso_size == 0) {
 256                         if (net_ratelimit())
 257                                 printk(KERN_WARNING "%s: zero gso size.\n",
 258                                        dev->name);
 259                         goto frame_err;
 260                 }
 261
 262                 /* Header must be checked, and gso_segs computed. */
 263                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 264                 skb_shinfo(skb)->gso_segs = 0;
 265         }
 266
 267         netif_receive_skb(skb);
 268         return;
 269
 270 frame_err:
 271         dev->stats.rx_frame_errors++;
 272 drop:
 273         dev_kfree_skb(skb);
 274 }
 275
 276 static bool try_fill_recv_maxbufs(struct virtnet_info *vi, gfp_t gfp)
 277 {
 278         struct sk_buff *skb;
 279         struct scatterlist sg[2+MAX_SKB_FRAGS];
 280         int num, err, i;
 281         bool oom = false;
 282
 283         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 284         for (;;) {
 285                 struct virtio_net_hdr *hdr;
 286
 287                 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN + NET_IP_ALIGN);
 288                 if (unlikely(!skb)) {
 289                         oom = true;
 290                         break;
 291                 }
 292
 293                 skb_reserve(skb, NET_IP_ALIGN);
 294                 skb_put(skb, MAX_PACKET_LEN);
 295
 296                 hdr = skb_vnet_hdr(skb);
 297                 sg_set_buf(sg, hdr, sizeof(*hdr));
 298
 299                 if (vi->big_packets) {
 300                         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 301                                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 302                                 f->page = get_a_page(vi, gfp);
 303                                 if (!f->page)
 304                                         break;
 305
 306                                 f->page_offset = 0;
 307                                 f->size = PAGE_SIZE;
 308
 309                                 skb->data_len += PAGE_SIZE;
 310                                 skb->len += PAGE_SIZE;
 311
 312                                 skb_shinfo(skb)->nr_frags++;
 313                         }
 314                 }
 315
 316                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 317                 skb_queue_head(&vi->recv, skb);
 318
 319                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 320                 if (err < 0) {
 321                         skb_unlink(skb, &vi->recv);
 322                         trim_pages(vi, skb);
 323                         kfree_skb(skb);
 324                         break;
 325                 }
 326                 vi->num++;
 327         }
 328         if (unlikely(vi->num > vi->max))
 329                 vi->max = vi->num;
 330         vi->rvq->vq_ops->kick(vi->rvq);
 331         return !oom;
 332 }
 333
 334 /* Returns false if we couldn't fill entirely (OOM). */
 335 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
 336 {
 337         struct sk_buff *skb;
 338         struct scatterlist sg[1];
 339         int err;
 340         bool oom = false;
 341
 342         if (!vi->mergeable_rx_bufs)
 343                 return try_fill_recv_maxbufs(vi, gfp);
 344
 345         for (;;) {
 346                 skb_frag_t *f;
 347
 348                 skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
 349                 if (unlikely(!skb)) {
 350                         oom = true;
 351                         break;
 352                 }
 353
 354                 skb_reserve(skb, NET_IP_ALIGN);
 355
 356                 f = &skb_shinfo(skb)->frags[0];
 357                 f->page = get_a_page(vi, gfp);
 358                 if (!f->page) {
 359                         oom = true;
 360                         kfree_skb(skb);
 361                         break;
 362                 }
 363
 364                 f->page_offset = 0;
 365                 f->size = PAGE_SIZE;
 366
 367                 skb_shinfo(skb)->nr_frags++;
 368
 369                 sg_init_one(sg, page_address(f->page), PAGE_SIZE);
 370                 skb_queue_head(&vi->recv, skb);
 371
 372                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
 373                 if (err < 0) {
 374                         skb_unlink(skb, &vi->recv);
 375                         kfree_skb(skb);
 376                         break;
 377                 }
 378                 vi->num++;
 379         }
 380         if (unlikely(vi->num > vi->max))
 381                 vi->max = vi->num;
 382         vi->rvq->vq_ops->kick(vi->rvq);
 383         return !oom;
 384 }
 385
 386 static void skb_recv_done(struct virtqueue *rvq)
 387 {
 388         struct virtnet_info *vi = rvq->vdev->priv;
 389         /* Schedule NAPI, Suppress further interrupts if successful. */
 390         if (napi_schedule_prep(&vi->napi)) {
 391                 rvq->vq_ops->disable_cb(rvq);
 392                 __napi_schedule(&vi->napi);
 393         }
 394 }
 395
 396 static void refill_work(struct work_struct *work)
 397 {
 398         struct virtnet_info *vi;
 399         bool still_empty;
 400
 401         vi = container_of(work, struct virtnet_info, refill.work);
 402         napi_disable(&vi->napi);
 403         try_fill_recv(vi, GFP_KERNEL);
 404         still_empty = (vi->num == 0);
 405         napi_enable(&vi->napi);
 406
 407         /* In theory, this can happen: if we don't get any buffers in
 408          * we will *never* try to fill again. */
 409         if (still_empty)
 410                 schedule_delayed_work(&vi->refill, HZ/2);
 411 }
 412
 413 static int virtnet_poll(struct napi_struct *napi, int budget)
 414 {
 415         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 416         struct sk_buff *skb = NULL;
 417         unsigned int len, received = 0;
 418
 419 again:
 420         while (received < budget &&
 421                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 422                 __skb_unlink(skb, &vi->recv);
 423                 receive_skb(vi->dev, skb, len);
 424                 vi->num--;
 425                 received++;
 426         }
 427
 428         if (vi->num < vi->max / 2) {
 429                 if (!try_fill_recv(vi, GFP_ATOMIC))
 430                         schedule_delayed_work(&vi->refill, 0);
 431         }
 432
 433         /* Out of packets? */
 434         if (received < budget) {
 435                 napi_complete(napi);
 436                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 437                     && napi_schedule_prep(napi)) {
 438                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 439                         __napi_schedule(napi);
 440                         goto again;
 441                 }
 442         }
 443
 444         return received;
 445 }
 446
 447 static void free_old_xmit_skbs(struct virtnet_info *vi)
 448 {
 449         struct sk_buff *skb;
 450         unsigned int len;
 451
 452         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 453                 pr_debug("Sent skb %p\n", skb);
 454                 __skb_unlink(skb, &vi->send);
 455                 vi->dev->stats.tx_bytes += skb->len;
 456                 vi->dev->stats.tx_packets++;
 457                 kfree_skb(skb);
 458         }
 459 }
 460
 461 /* If the virtio transport doesn't always notify us when all in-flight packets
 462  * are consumed, we fall back to using this function on a timer to free them. */
 463 static void xmit_free(unsigned long data)
 464 {
 465         struct virtnet_info *vi = (void *)data;
 466
 467         netif_tx_lock(vi->dev);
 468
 469         free_old_xmit_skbs(vi);
 470
 471         if (!skb_queue_empty(&vi->send))
 472                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 473
 474         netif_tx_unlock(vi->dev);
 475 }
 476
 477 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 478 {
 479         int num, err;
 480         struct scatterlist sg[2+MAX_SKB_FRAGS];
 481         struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 482         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 483         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 484
 485         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 486
 487         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 488
 489         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 490                 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 491                 hdr->csum_start = skb->csum_start - skb_headroom(skb);
 492                 hdr->csum_offset = skb->csum_offset;
 493         } else {
 494                 hdr->flags = 0;
 495                 hdr->csum_offset = hdr->csum_start = 0;
 496         }
 497
 498         if (skb_is_gso(skb)) {
 499                 hdr->hdr_len = skb_headlen(skb);
 500                 hdr->gso_size = skb_shinfo(skb)->gso_size;
 501                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 502                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 503                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 504                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 505                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 506                         hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 507                 else
 508                         BUG();
 509                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 510                         hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 511         } else {
 512                 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 513                 hdr->gso_size = hdr->hdr_len = 0;
 514         }
 515
 516         mhdr->num_buffers = 0;
 517
 518         /* Encode metadata header at front. */
 519         if (vi->mergeable_rx_bufs)
 520                 sg_set_buf(sg, mhdr, sizeof(*mhdr));
 521         else
 522                 sg_set_buf(sg, hdr, sizeof(*hdr));
 523
 524         num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 525
 526         err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
 527         if (err >= 0 && !vi->free_in_tasklet) {
 528                 /* Don't wait up for transmitted skbs to be freed. */
 529                 skb_orphan(skb);
 530                 nf_reset(skb);
 531                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 532         }
 533
 534         return err;
 535 }
 536
 537 static void xmit_tasklet(unsigned long data)
 538 {
 539         struct virtnet_info *vi = (void *)data;
 540
 541         netif_tx_lock_bh(vi->dev);
 542         free_old_xmit_skbs(vi);
 543         netif_tx_unlock_bh(vi->dev);
 544 }
 545
 546 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 547 {
 548         struct virtnet_info *vi = netdev_priv(dev);
 549
 550 again:
 551         /* Free up any pending old buffers before queueing new ones. */
 552         free_old_xmit_skbs(vi);
 553
 554         /* Put new one in send queue and do transmit */
 555         __skb_queue_head(&vi->send, skb);
 556         if (likely(xmit_skb(vi, skb) >= 0)) {
 557                 vi->svq->vq_ops->kick(vi->svq);
 558                 return NETDEV_TX_OK;
 559         }
 560
 561         /* Ring too full for this packet, remove it from queue again. */
 562         pr_debug("%s: virtio not prepared to send\n", dev->name);
 563         __skb_unlink(skb, &vi->send);
 564         netif_stop_queue(dev);
 565
 566         /* Activate callback for using skbs: if this returns false it
 567          * means some were used in the meantime. */
 568         if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 569                 vi->svq->vq_ops->disable_cb(vi->svq);
 570                 netif_start_queue(dev);
 571                 goto again;
 572         }
 573         return NETDEV_TX_BUSY;
 574 }
 575
 576 static int virtnet_set_mac_address(struct net_device *dev, void *p)
 577 {
 578         struct virtnet_info *vi = netdev_priv(dev);
 579         struct virtio_device *vdev = vi->vdev;
 580         int ret;
 581
 582         ret = eth_mac_addr(dev, p);
 583         if (ret)
 584                 return ret;
 585
 586         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
 587                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 588                                   dev->dev_addr, dev->addr_len);
 589
 590         return 0;
 591 }
 592
 593 #ifdef CONFIG_NET_POLL_CONTROLLER
 594 static void virtnet_netpoll(struct net_device *dev)
 595 {
 596         struct virtnet_info *vi = netdev_priv(dev);
 597
 598         napi_schedule(&vi->napi);
 599 }
 600 #endif
 601
 602 static int virtnet_open(struct net_device *dev)
 603 {
 604         struct virtnet_info *vi = netdev_priv(dev);
 605
 606         napi_enable(&vi->napi);
 607
 608         /* If all buffers were filled by other side before we napi_enabled, we
 609          * won't get another interrupt, so process any outstanding packets
 610          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 611          * We synchronize against interrupts via NAPI_STATE_SCHED */
 612         if (napi_schedule_prep(&vi->napi)) {
 613                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 614                 __napi_schedule(&vi->napi);
 615         }
 616         return 0;
 617 }
 618
 619 /*
 620  * Send command via the control virtqueue and check status.  Commands
 621  * supported by the hypervisor, as indicated by feature bits, should
 622  * never fail unless improperly formated.
 623  */
 624 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 625                                  struct scatterlist *data, int out, int in)
 626 {
 627         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 628         struct virtio_net_ctrl_hdr ctrl;
 629         virtio_net_ctrl_ack status = ~0;
 630         unsigned int tmp;
 631         int i;
 632
 633         /* Caller should know better */
 634         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 635                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 636
 637         out++; /* Add header */
 638         in++; /* Add return status */
 639
 640         ctrl.class = class;
 641         ctrl.cmd = cmd;
 642
 643         sg_init_table(sg, out + in);
 644
 645         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 646         for_each_sg(data, s, out + in - 2, i)
 647                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 648         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 649
 650         BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
 651
 652         vi->cvq->vq_ops->kick(vi->cvq);
 653
 654         /*
 655          * Spin for a response, the kick causes an ioport write, trapping
 656          * into the hypervisor, so the request should be handled immediately.
 657          */
 658         while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
 659                 cpu_relax();
 660
 661         return status == VIRTIO_NET_OK;
 662 }
 663
 664 static int virtnet_close(struct net_device *dev)
 665 {
 666         struct virtnet_info *vi = netdev_priv(dev);
 667
 668         napi_disable(&vi->napi);
 669
 670         return 0;
 671 }
 672
 673 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 674 {
 675         struct virtnet_info *vi = netdev_priv(dev);
 676         struct virtio_device *vdev = vi->vdev;
 677
 678         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 679                 return -ENOSYS;
 680
 681         return ethtool_op_set_tx_hw_csum(dev, data);
 682 }
 683
 684 static void virtnet_set_rx_mode(struct net_device *dev)
 685 {
 686         struct virtnet_info *vi = netdev_priv(dev);
 687         struct scatterlist sg[2];
 688         u8 promisc, allmulti;
 689         struct virtio_net_ctrl_mac *mac_data;
 690         struct dev_addr_list *addr;
 691         struct netdev_hw_addr *ha;
 692         void *buf;
 693         int i;
 694
 695         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 696         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 697                 return;
 698
 699         promisc = ((dev->flags & IFF_PROMISC) != 0);
 700         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 701
 702         sg_init_one(sg, &promisc, sizeof(promisc));
 703
 704         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 705                                   VIRTIO_NET_CTRL_RX_PROMISC,
 706                                   sg, 1, 0))
 707                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 708                          promisc ? "en" : "dis");
 709
 710         sg_init_one(sg, &allmulti, sizeof(allmulti));
 711
 712         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 713                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 714                                   sg, 1, 0))
 715                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 716                          allmulti ? "en" : "dis");
 717
 718         /* MAC filter - use one buffer for both lists */
 719         mac_data = buf = kzalloc(((dev->uc.count + dev->mc_count) * ETH_ALEN) +
 720                                  (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 721         if (!buf) {
 722                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 723                 return;
 724         }
 725
 726         sg_init_table(sg, 2);
 727
 728         /* Store the unicast list and count in the front of the buffer */
 729         mac_data->entries = dev->uc.count;
 730         i = 0;
 731         list_for_each_entry(ha, &dev->uc.list, list)
 732                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 733
 734         sg_set_buf(&sg[0], mac_data,
 735                    sizeof(mac_data->entries) + (dev->uc.count * ETH_ALEN));
 736
 737         /* multicast list and count fill the end */
 738         mac_data = (void *)&mac_data->macs[dev->uc.count][0];
 739
 740         mac_data->entries = dev->mc_count;
 741         addr = dev->mc_list;
 742         for (i = 0; i < dev->mc_count; i++, addr = addr->next)
 743                 memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
 744
 745         sg_set_buf(&sg[1], mac_data,
 746                    sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
 747
 748         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 749                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 750                                   sg, 2, 0))
 751                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 752
 753         kfree(buf);
 754 }
 755
 756 static void virtnet_vlan_rx_add_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_ADD, &sg, 1, 0))
 765                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 766 }
 767
 768 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 769 {
 770         struct virtnet_info *vi = netdev_priv(dev);
 771         struct scatterlist sg;
 772
 773         sg_init_one(&sg, &vid, sizeof(vid));
 774
 775         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 776                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 777                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 778 }
 779
 780 static const struct ethtool_ops virtnet_ethtool_ops = {
 781         .set_tx_csum = virtnet_set_tx_csum,
 782         .set_sg = ethtool_op_set_sg,
 783         .set_tso = ethtool_op_set_tso,
 784         .set_ufo = ethtool_op_set_ufo,
 785         .get_link = ethtool_op_get_link,
 786 };
 787
 788 #define MIN_MTU 68
 789 #define MAX_MTU 65535
 790
 791 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 792 {
 793         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 794                 return -EINVAL;
 795         dev->mtu = new_mtu;
 796         return 0;
 797 }
 798
 799 static const struct net_device_ops virtnet_netdev = {
 800         .ndo_open            = virtnet_open,
 801         .ndo_stop            = virtnet_close,
 802         .ndo_start_xmit      = start_xmit,
 803         .ndo_validate_addr   = eth_validate_addr,
 804         .ndo_set_mac_address = virtnet_set_mac_address,
 805         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 806         .ndo_change_mtu      = virtnet_change_mtu,
 807         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 808         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 809 #ifdef CONFIG_NET_POLL_CONTROLLER
 810         .ndo_poll_controller = virtnet_netpoll,
 811 #endif
 812 };
 813
 814 static void virtnet_update_status(struct virtnet_info *vi)
 815 {
 816         u16 v;
 817
 818         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
 819                 return;
 820
 821         vi->vdev->config->get(vi->vdev,
 822                               offsetof(struct virtio_net_config, status),
 823                               &v, sizeof(v));
 824
 825         /* Ignore unknown (future) status bits */
 826         v &= VIRTIO_NET_S_LINK_UP;
 827
 828         if (vi->status == v)
 829                 return;
 830
 831         vi->status = v;
 832
 833         if (vi->status & VIRTIO_NET_S_LINK_UP) {
 834                 netif_carrier_on(vi->dev);
 835                 netif_wake_queue(vi->dev);
 836         } else {
 837                 netif_carrier_off(vi->dev);
 838                 netif_stop_queue(vi->dev);
 839         }
 840 }
 841
 842 static void virtnet_config_changed(struct virtio_device *vdev)
 843 {
 844         struct virtnet_info *vi = vdev->priv;
 845
 846         virtnet_update_status(vi);
 847 }
 848
 849 static int virtnet_probe(struct virtio_device *vdev)
 850 {
 851         int err;
 852         struct net_device *dev;
 853         struct virtnet_info *vi;
 854         struct virtqueue *vqs[3];
 855         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
 856         const char *names[] = { "input", "output", "control" };
 857         int nvqs;
 858
 859         /* Allocate ourselves a network device with room for our info */
 860         dev = alloc_etherdev(sizeof(struct virtnet_info));
 861         if (!dev)
 862                 return -ENOMEM;
 863
 864         /* Set up network device as normal. */
 865         dev->netdev_ops = &virtnet_netdev;
 866         dev->features = NETIF_F_HIGHDMA;
 867         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 868         SET_NETDEV_DEV(dev, &vdev->dev);
 869
 870         /* Do we support "hardware" checksums? */
 871         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 872                 /* This opens up the world of extra features. */
 873                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 874                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 875                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 876                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 877                 }
 878                 /* Individual feature bits: what can host handle? */
 879                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 880                         dev->features |= NETIF_F_TSO;
 881                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 882                         dev->features |= NETIF_F_TSO6;
 883                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 884                         dev->features |= NETIF_F_TSO_ECN;
 885                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 886                         dev->features |= NETIF_F_UFO;
 887         }
 888
 889         /* Configuration may specify what MAC to use.  Otherwise random. */
 890         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 891                 vdev->config->get(vdev,
 892                                   offsetof(struct virtio_net_config, mac),
 893                                   dev->dev_addr, dev->addr_len);
 894         } else
 895                 random_ether_addr(dev->dev_addr);
 896
 897         /* Set up our device-specific information */
 898         vi = netdev_priv(dev);
 899         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 900         vi->dev = dev;
 901         vi->vdev = vdev;
 902         vdev->priv = vi;
 903         vi->pages = NULL;
 904         INIT_DELAYED_WORK(&vi->refill, refill_work);
 905
 906         /* If they give us a callback when all buffers are done, we don't need
 907          * the timer. */
 908         vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
 909
 910         /* If we can receive ANY GSO packets, we must allocate large ones. */
 911         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 912             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 913             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 914                 vi->big_packets = true;
 915
 916         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 917                 vi->mergeable_rx_bufs = true;
 918
 919         /* We expect two virtqueues, receive then send,
 920          * and optionally control. */
 921         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
 922
 923         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
 924         if (err)
 925                 goto free;
 926
 927         vi->rvq = vqs[0];
 928         vi->svq = vqs[1];
 929
 930         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
 931                 vi->cvq = vqs[2];
 932
 933                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
 934                         dev->features |= NETIF_F_HW_VLAN_FILTER;
 935         }
 936
 937         /* Initialize our empty receive and send queues. */
 938         skb_queue_head_init(&vi->recv);
 939         skb_queue_head_init(&vi->send);
 940
 941         tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
 942
 943         if (!vi->free_in_tasklet)
 944                 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
 945
 946         err = register_netdev(dev);
 947         if (err) {
 948                 pr_debug("virtio_net: registering device failed\n");
 949                 goto free_vqs;
 950         }
 951
 952         /* Last of all, set up some receive buffers. */
 953         try_fill_recv(vi, GFP_KERNEL);
 954
 955         /* If we didn't even get one input buffer, we're useless. */
 956         if (vi->num == 0) {
 957                 err = -ENOMEM;
 958                 goto unregister;
 959         }
 960
 961         vi->status = VIRTIO_NET_S_LINK_UP;
 962         virtnet_update_status(vi);
 963         netif_carrier_on(dev);
 964
 965         pr_debug("virtnet: registered device %s\n", dev->name);
 966         return 0;
 967
 968 unregister:
 969         unregister_netdev(dev);
 970         cancel_delayed_work_sync(&vi->refill);
 971 free_vqs:
 972         vdev->config->del_vqs(vdev);
 973 free:
 974         free_netdev(dev);
 975         return err;
 976 }
 977
 978 static void virtnet_remove(struct virtio_device *vdev)
 979 {
 980         struct virtnet_info *vi = vdev->priv;
 981         struct sk_buff *skb;
 982
 983         /* Stop all the virtqueues. */
 984         vdev->config->reset(vdev);
 985
 986         if (!vi->free_in_tasklet)
 987                 del_timer_sync(&vi->xmit_free_timer);
 988
 989         /* Free our skbs in send and recv queues, if any. */
 990         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 991                 kfree_skb(skb);
 992                 vi->num--;
 993         }
 994         __skb_queue_purge(&vi->send);
 995
 996         BUG_ON(vi->num != 0);
 997
 998         unregister_netdev(vi->dev);
 999         cancel_delayed_work_sync(&vi->refill);
1000
1001         vdev->config->del_vqs(vi->vdev);
1002
1003         while (vi->pages)
1004                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1005
1006         free_netdev(vi->dev);
1007 }
1008
1009 static struct virtio_device_id id_table[] = {
1010         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1011         { 0 },
1012 };
1013
1014 static unsigned int features[] = {
1015         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1016         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1017         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1018         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1019         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1020         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1021         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1022         VIRTIO_F_NOTIFY_ON_EMPTY,
1023 };
1024
1025 static struct virtio_driver virtio_net = {
1026         .feature_table = features,
1027         .feature_table_size = ARRAY_SIZE(features),
1028         .driver.name =  KBUILD_MODNAME,
1029         .driver.owner = THIS_MODULE,
1030         .id_table =     id_table,
1031         .probe =        virtnet_probe,
1032         .remove =       __devexit_p(virtnet_remove),
1033         .config_changed = virtnet_config_changed,
1034 };
1035
1036 static int __init init(void)
1037 {
1038         return register_virtio_driver(&virtio_net);
1039 }
1040
1041 static void __exit fini(void)
1042 {
1043         unregister_virtio_driver(&virtio_net);
1044 }
1045 module_init(init);
1046 module_exit(fini);
1047
1048 MODULE_DEVICE_TABLE(virtio, id_table);
1049 MODULE_DESCRIPTION("Virtio network driver");
1050 MODULE_LICENSE("GPL");