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