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