SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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_ip_align(vi->dev, MAX_PACKET_LEN);
 287                 if (unlikely(!skb)) {
 288                         oom = true;
 289                         break;
 290                 }
 291
 292                 skb_put(skb, MAX_PACKET_LEN);
 293
 294                 hdr = skb_vnet_hdr(skb);
 295                 sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
 296
 297                 if (vi->big_packets) {
 298                         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 299                                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 300                                 f->page = get_a_page(vi, gfp);
 301                                 if (!f->page)
 302                                         break;
 303
 304                                 f->page_offset = 0;
 305                                 f->size = PAGE_SIZE;
 306
 307                                 skb->data_len += PAGE_SIZE;
 308                                 skb->len += PAGE_SIZE;
 309
 310                                 skb_shinfo(skb)->nr_frags++;
 311                         }
 312                 }
 313
 314                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 315                 skb_queue_head(&vi->recv, skb);
 316
 317                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 318                 if (err < 0) {
 319                         skb_unlink(skb, &vi->recv);
 320                         trim_pages(vi, skb);
 321                         kfree_skb(skb);
 322                         break;
 323                 }
 324                 vi->num++;
 325         } while (err >= num);
 326         if (unlikely(vi->num > vi->max))
 327                 vi->max = vi->num;
 328         vi->rvq->vq_ops->kick(vi->rvq);
 329         return !oom;
 330 }
 331
 332 /* Returns false if we couldn't fill entirely (OOM). */
 333 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
 334 {
 335         struct sk_buff *skb;
 336         struct scatterlist sg[1];
 337         int err;
 338         bool oom = false;
 339
 340         if (!vi->mergeable_rx_bufs)
 341                 return try_fill_recv_maxbufs(vi, gfp);
 342
 343         do {
 344                 skb_frag_t *f;
 345
 346                 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
 347                 if (unlikely(!skb)) {
 348                         oom = true;
 349                         break;
 350                 }
 351
 352                 f = &skb_shinfo(skb)->frags[0];
 353                 f->page = get_a_page(vi, gfp);
 354                 if (!f->page) {
 355                         oom = true;
 356                         kfree_skb(skb);
 357                         break;
 358                 }
 359
 360                 f->page_offset = 0;
 361                 f->size = PAGE_SIZE;
 362
 363                 skb_shinfo(skb)->nr_frags++;
 364
 365                 sg_init_one(sg, page_address(f->page), PAGE_SIZE);
 366                 skb_queue_head(&vi->recv, skb);
 367
 368                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
 369                 if (err < 0) {
 370                         skb_unlink(skb, &vi->recv);
 371                         kfree_skb(skb);
 372                         break;
 373                 }
 374                 vi->num++;
 375         } while (err > 0);
 376         if (unlikely(vi->num > vi->max))
 377                 vi->max = vi->num;
 378         vi->rvq->vq_ops->kick(vi->rvq);
 379         return !oom;
 380 }
 381
 382 static void skb_recv_done(struct virtqueue *rvq)
 383 {
 384         struct virtnet_info *vi = rvq->vdev->priv;
 385         /* Schedule NAPI, Suppress further interrupts if successful. */
 386         if (napi_schedule_prep(&vi->napi)) {
 387                 rvq->vq_ops->disable_cb(rvq);
 388                 __napi_schedule(&vi->napi);
 389         }
 390 }
 391
 392 static void refill_work(struct work_struct *work)
 393 {
 394         struct virtnet_info *vi;
 395         bool still_empty;
 396
 397         vi = container_of(work, struct virtnet_info, refill.work);
 398         napi_disable(&vi->napi);
 399         try_fill_recv(vi, GFP_KERNEL);
 400         still_empty = (vi->num == 0);
 401         napi_enable(&vi->napi);
 402
 403         /* In theory, this can happen: if we don't get any buffers in
 404          * we will *never* try to fill again. */
 405         if (still_empty)
 406                 schedule_delayed_work(&vi->refill, HZ/2);
 407 }
 408
 409 static int virtnet_poll(struct napi_struct *napi, int budget)
 410 {
 411         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 412         struct sk_buff *skb = NULL;
 413         unsigned int len, received = 0;
 414
 415 again:
 416         while (received < budget &&
 417                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 418                 __skb_unlink(skb, &vi->recv);
 419                 receive_skb(vi->dev, skb, len);
 420                 vi->num--;
 421                 received++;
 422         }
 423
 424         if (vi->num < vi->max / 2) {
 425                 if (!try_fill_recv(vi, GFP_ATOMIC))
 426                         schedule_delayed_work(&vi->refill, 0);
 427         }
 428
 429         /* Out of packets? */
 430         if (received < budget) {
 431                 napi_complete(napi);
 432                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 433                     && napi_schedule_prep(napi)) {
 434                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 435                         __napi_schedule(napi);
 436                         goto again;
 437                 }
 438         }
 439
 440         return received;
 441 }
 442
 443 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
 444 {
 445         struct sk_buff *skb;
 446         unsigned int len, tot_sgs = 0;
 447
 448         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 449                 pr_debug("Sent skb %p\n", skb);
 450                 __skb_unlink(skb, &vi->send);
 451                 vi->dev->stats.tx_bytes += skb->len;
 452                 vi->dev->stats.tx_packets++;
 453                 tot_sgs += skb_vnet_hdr(skb)->num_sg;
 454                 dev_kfree_skb_any(skb);
 455         }
 456         return tot_sgs;
 457 }
 458
 459 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 460 {
 461         struct scatterlist sg[2+MAX_SKB_FRAGS];
 462         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 463         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 464
 465         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 466
 467         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 468
 469         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 470                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 471                 hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
 472                 hdr->hdr.csum_offset = skb->csum_offset;
 473         } else {
 474                 hdr->hdr.flags = 0;
 475                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
 476         }
 477
 478         if (skb_is_gso(skb)) {
 479                 hdr->hdr.hdr_len = skb_headlen(skb);
 480                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
 481                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 482                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 483                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 484                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 485                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 486                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 487                 else
 488                         BUG();
 489                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 490                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 491         } else {
 492                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 493                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
 494         }
 495
 496         hdr->mhdr.num_buffers = 0;
 497
 498         /* Encode metadata header at front. */
 499         if (vi->mergeable_rx_bufs)
 500                 sg_set_buf(sg, &hdr->mhdr, sizeof(hdr->mhdr));
 501         else
 502                 sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
 503
 504         hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 505         return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
 506 }
 507
 508 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 509 {
 510         struct virtnet_info *vi = netdev_priv(dev);
 511         int capacity;
 512
 513 again:
 514         /* Free up any pending old buffers before queueing new ones. */
 515         free_old_xmit_skbs(vi);
 516
 517         /* Try to transmit */
 518         capacity = xmit_skb(vi, skb);
 519
 520         /* This can happen with OOM and indirect buffers. */
 521         if (unlikely(capacity < 0)) {
 522                 netif_stop_queue(dev);
 523                 dev_warn(&dev->dev, "Unexpected full queue\n");
 524                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 525                         vi->svq->vq_ops->disable_cb(vi->svq);
 526                         netif_start_queue(dev);
 527                         goto again;
 528                 }
 529                 return NETDEV_TX_BUSY;
 530         }
 531         vi->svq->vq_ops->kick(vi->svq);
 532
 533         /*
 534          * Put new one in send queue.  You'd expect we'd need this before
 535          * xmit_skb calls add_buf(), since the callback can be triggered
 536          * immediately after that.  But since the callback just triggers
 537          * another call back here, normal network xmit locking prevents the
 538          * race.
 539          */
 540         __skb_queue_head(&vi->send, skb);
 541
 542         /* Don't wait up for transmitted skbs to be freed. */
 543         skb_orphan(skb);
 544         nf_reset(skb);
 545
 546         /* Apparently nice girls don't return TX_BUSY; stop the queue
 547          * before it gets out of hand.  Naturally, this wastes entries. */
 548         if (capacity < 2+MAX_SKB_FRAGS) {
 549                 netif_stop_queue(dev);
 550                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 551                         /* More just got used, free them then recheck. */
 552                         capacity += free_old_xmit_skbs(vi);
 553                         if (capacity >= 2+MAX_SKB_FRAGS) {
 554                                 netif_start_queue(dev);
 555                                 vi->svq->vq_ops->disable_cb(vi->svq);
 556                         }
 557                 }
 558         }
 559
 560         return NETDEV_TX_OK;
 561 }
 562
 563 static int virtnet_set_mac_address(struct net_device *dev, void *p)
 564 {
 565         struct virtnet_info *vi = netdev_priv(dev);
 566         struct virtio_device *vdev = vi->vdev;
 567         int ret;
 568
 569         ret = eth_mac_addr(dev, p);
 570         if (ret)
 571                 return ret;
 572
 573         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
 574                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 575                                   dev->dev_addr, dev->addr_len);
 576
 577         return 0;
 578 }
 579
 580 #ifdef CONFIG_NET_POLL_CONTROLLER
 581 static void virtnet_netpoll(struct net_device *dev)
 582 {
 583         struct virtnet_info *vi = netdev_priv(dev);
 584
 585         napi_schedule(&vi->napi);
 586 }
 587 #endif
 588
 589 static int virtnet_open(struct net_device *dev)
 590 {
 591         struct virtnet_info *vi = netdev_priv(dev);
 592
 593         napi_enable(&vi->napi);
 594
 595         /* If all buffers were filled by other side before we napi_enabled, we
 596          * won't get another interrupt, so process any outstanding packets
 597          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 598          * We synchronize against interrupts via NAPI_STATE_SCHED */
 599         if (napi_schedule_prep(&vi->napi)) {
 600                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 601                 __napi_schedule(&vi->napi);
 602         }
 603         return 0;
 604 }
 605
 606 /*
 607  * Send command via the control virtqueue and check status.  Commands
 608  * supported by the hypervisor, as indicated by feature bits, should
 609  * never fail unless improperly formated.
 610  */
 611 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 612                                  struct scatterlist *data, int out, int in)
 613 {
 614         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 615         struct virtio_net_ctrl_hdr ctrl;
 616         virtio_net_ctrl_ack status = ~0;
 617         unsigned int tmp;
 618         int i;
 619
 620         /* Caller should know better */
 621         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 622                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 623
 624         out++; /* Add header */
 625         in++; /* Add return status */
 626
 627         ctrl.class = class;
 628         ctrl.cmd = cmd;
 629
 630         sg_init_table(sg, out + in);
 631
 632         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 633         for_each_sg(data, s, out + in - 2, i)
 634                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 635         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 636
 637         BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
 638
 639         vi->cvq->vq_ops->kick(vi->cvq);
 640
 641         /*
 642          * Spin for a response, the kick causes an ioport write, trapping
 643          * into the hypervisor, so the request should be handled immediately.
 644          */
 645         while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
 646                 cpu_relax();
 647
 648         return status == VIRTIO_NET_OK;
 649 }
 650
 651 static int virtnet_close(struct net_device *dev)
 652 {
 653         struct virtnet_info *vi = netdev_priv(dev);
 654
 655         napi_disable(&vi->napi);
 656
 657         return 0;
 658 }
 659
 660 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 661 {
 662         struct virtnet_info *vi = netdev_priv(dev);
 663         struct virtio_device *vdev = vi->vdev;
 664
 665         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 666                 return -ENOSYS;
 667
 668         return ethtool_op_set_tx_hw_csum(dev, data);
 669 }
 670
 671 static void virtnet_set_rx_mode(struct net_device *dev)
 672 {
 673         struct virtnet_info *vi = netdev_priv(dev);
 674         struct scatterlist sg[2];
 675         u8 promisc, allmulti;
 676         struct virtio_net_ctrl_mac *mac_data;
 677         struct dev_addr_list *addr;
 678         struct netdev_hw_addr *ha;
 679         void *buf;
 680         int i;
 681
 682         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 683         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 684                 return;
 685
 686         promisc = ((dev->flags & IFF_PROMISC) != 0);
 687         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 688
 689         sg_init_one(sg, &promisc, sizeof(promisc));
 690
 691         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 692                                   VIRTIO_NET_CTRL_RX_PROMISC,
 693                                   sg, 1, 0))
 694                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 695                          promisc ? "en" : "dis");
 696
 697         sg_init_one(sg, &allmulti, sizeof(allmulti));
 698
 699         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 700                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 701                                   sg, 1, 0))
 702                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 703                          allmulti ? "en" : "dis");
 704
 705         /* MAC filter - use one buffer for both lists */
 706         mac_data = buf = kzalloc(((dev->uc.count + dev->mc_count) * ETH_ALEN) +
 707                                  (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 708         if (!buf) {
 709                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 710                 return;
 711         }
 712
 713         sg_init_table(sg, 2);
 714
 715         /* Store the unicast list and count in the front of the buffer */
 716         mac_data->entries = dev->uc.count;
 717         i = 0;
 718         list_for_each_entry(ha, &dev->uc.list, list)
 719                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 720
 721         sg_set_buf(&sg[0], mac_data,
 722                    sizeof(mac_data->entries) + (dev->uc.count * ETH_ALEN));
 723
 724         /* multicast list and count fill the end */
 725         mac_data = (void *)&mac_data->macs[dev->uc.count][0];
 726
 727         mac_data->entries = dev->mc_count;
 728         addr = dev->mc_list;
 729         for (i = 0; i < dev->mc_count; i++, addr = addr->next)
 730                 memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
 731
 732         sg_set_buf(&sg[1], mac_data,
 733                    sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
 734
 735         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 736                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 737                                   sg, 2, 0))
 738                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 739
 740         kfree(buf);
 741 }
 742
 743 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
 744 {
 745         struct virtnet_info *vi = netdev_priv(dev);
 746         struct scatterlist sg;
 747
 748         sg_init_one(&sg, &vid, sizeof(vid));
 749
 750         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 751                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
 752                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 753 }
 754
 755 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 756 {
 757         struct virtnet_info *vi = netdev_priv(dev);
 758         struct scatterlist sg;
 759
 760         sg_init_one(&sg, &vid, sizeof(vid));
 761
 762         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 763                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 764                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 765 }
 766
 767 static const struct ethtool_ops virtnet_ethtool_ops = {
 768         .set_tx_csum = virtnet_set_tx_csum,
 769         .set_sg = ethtool_op_set_sg,
 770         .set_tso = ethtool_op_set_tso,
 771         .set_ufo = ethtool_op_set_ufo,
 772         .get_link = ethtool_op_get_link,
 773 };
 774
 775 #define MIN_MTU 68
 776 #define MAX_MTU 65535
 777
 778 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 779 {
 780         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 781                 return -EINVAL;
 782         dev->mtu = new_mtu;
 783         return 0;
 784 }
 785
 786 static const struct net_device_ops virtnet_netdev = {
 787         .ndo_open            = virtnet_open,
 788         .ndo_stop            = virtnet_close,
 789         .ndo_start_xmit      = start_xmit,
 790         .ndo_validate_addr   = eth_validate_addr,
 791         .ndo_set_mac_address = virtnet_set_mac_address,
 792         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 793         .ndo_change_mtu      = virtnet_change_mtu,
 794         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 795         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 796 #ifdef CONFIG_NET_POLL_CONTROLLER
 797         .ndo_poll_controller = virtnet_netpoll,
 798 #endif
 799 };
 800
 801 static void virtnet_update_status(struct virtnet_info *vi)
 802 {
 803         u16 v;
 804
 805         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
 806                 return;
 807
 808         vi->vdev->config->get(vi->vdev,
 809                               offsetof(struct virtio_net_config, status),
 810                               &v, sizeof(v));
 811
 812         /* Ignore unknown (future) status bits */
 813         v &= VIRTIO_NET_S_LINK_UP;
 814
 815         if (vi->status == v)
 816                 return;
 817
 818         vi->status = v;
 819
 820         if (vi->status & VIRTIO_NET_S_LINK_UP) {
 821                 netif_carrier_on(vi->dev);
 822                 netif_wake_queue(vi->dev);
 823         } else {
 824                 netif_carrier_off(vi->dev);
 825                 netif_stop_queue(vi->dev);
 826         }
 827 }
 828
 829 static void virtnet_config_changed(struct virtio_device *vdev)
 830 {
 831         struct virtnet_info *vi = vdev->priv;
 832
 833         virtnet_update_status(vi);
 834 }
 835
 836 static int virtnet_probe(struct virtio_device *vdev)
 837 {
 838         int err;
 839         struct net_device *dev;
 840         struct virtnet_info *vi;
 841         struct virtqueue *vqs[3];
 842         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
 843         const char *names[] = { "input", "output", "control" };
 844         int nvqs;
 845
 846         /* Allocate ourselves a network device with room for our info */
 847         dev = alloc_etherdev(sizeof(struct virtnet_info));
 848         if (!dev)
 849                 return -ENOMEM;
 850
 851         /* Set up network device as normal. */
 852         dev->netdev_ops = &virtnet_netdev;
 853         dev->features = NETIF_F_HIGHDMA;
 854         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 855         SET_NETDEV_DEV(dev, &vdev->dev);
 856
 857         /* Do we support "hardware" checksums? */
 858         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 859                 /* This opens up the world of extra features. */
 860                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 861                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 862                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 863                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 864                 }
 865                 /* Individual feature bits: what can host handle? */
 866                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 867                         dev->features |= NETIF_F_TSO;
 868                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 869                         dev->features |= NETIF_F_TSO6;
 870                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 871                         dev->features |= NETIF_F_TSO_ECN;
 872                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 873                         dev->features |= NETIF_F_UFO;
 874         }
 875
 876         /* Configuration may specify what MAC to use.  Otherwise random. */
 877         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 878                 vdev->config->get(vdev,
 879                                   offsetof(struct virtio_net_config, mac),
 880                                   dev->dev_addr, dev->addr_len);
 881         } else
 882                 random_ether_addr(dev->dev_addr);
 883
 884         /* Set up our device-specific information */
 885         vi = netdev_priv(dev);
 886         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 887         vi->dev = dev;
 888         vi->vdev = vdev;
 889         vdev->priv = vi;
 890         vi->pages = NULL;
 891         INIT_DELAYED_WORK(&vi->refill, refill_work);
 892
 893         /* If we can receive ANY GSO packets, we must allocate large ones. */
 894         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 895             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 896             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 897                 vi->big_packets = true;
 898
 899         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 900                 vi->mergeable_rx_bufs = true;
 901
 902         /* We expect two virtqueues, receive then send,
 903          * and optionally control. */
 904         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
 905
 906         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
 907         if (err)
 908                 goto free;
 909
 910         vi->rvq = vqs[0];
 911         vi->svq = vqs[1];
 912
 913         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
 914                 vi->cvq = vqs[2];
 915
 916                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
 917                         dev->features |= NETIF_F_HW_VLAN_FILTER;
 918         }
 919
 920         /* Initialize our empty receive and send queues. */
 921         skb_queue_head_init(&vi->recv);
 922         skb_queue_head_init(&vi->send);
 923
 924         err = register_netdev(dev);
 925         if (err) {
 926                 pr_debug("virtio_net: registering device failed\n");
 927                 goto free_vqs;
 928         }
 929
 930         /* Last of all, set up some receive buffers. */
 931         try_fill_recv(vi, GFP_KERNEL);
 932
 933         /* If we didn't even get one input buffer, we're useless. */
 934         if (vi->num == 0) {
 935                 err = -ENOMEM;
 936                 goto unregister;
 937         }
 938
 939         vi->status = VIRTIO_NET_S_LINK_UP;
 940         virtnet_update_status(vi);
 941         netif_carrier_on(dev);
 942
 943         pr_debug("virtnet: registered device %s\n", dev->name);
 944         return 0;
 945
 946 unregister:
 947         unregister_netdev(dev);
 948         cancel_delayed_work_sync(&vi->refill);
 949 free_vqs:
 950         vdev->config->del_vqs(vdev);
 951 free:
 952         free_netdev(dev);
 953         return err;
 954 }
 955
 956 static void __devexit virtnet_remove(struct virtio_device *vdev)
 957 {
 958         struct virtnet_info *vi = vdev->priv;
 959         struct sk_buff *skb;
 960
 961         /* Stop all the virtqueues. */
 962         vdev->config->reset(vdev);
 963
 964         /* Free our skbs in send and recv queues, if any. */
 965         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 966                 kfree_skb(skb);
 967                 vi->num--;
 968         }
 969         __skb_queue_purge(&vi->send);
 970
 971         BUG_ON(vi->num != 0);
 972
 973         unregister_netdev(vi->dev);
 974         cancel_delayed_work_sync(&vi->refill);
 975
 976         vdev->config->del_vqs(vi->vdev);
 977
 978         while (vi->pages)
 979                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
 980
 981         free_netdev(vi->dev);
 982 }
 983
 984 static struct virtio_device_id id_table[] = {
 985         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
 986         { 0 },
 987 };
 988
 989 static unsigned int features[] = {
 990         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
 991         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
 992         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
 993         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
 994         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
 995         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
 996         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
 997 };
 998
 999 static struct virtio_driver virtio_net = {
1000         .feature_table = features,
1001         .feature_table_size = ARRAY_SIZE(features),
1002         .driver.name =  KBUILD_MODNAME,
1003         .driver.owner = THIS_MODULE,
1004         .id_table =     id_table,
1005         .probe =        virtnet_probe,
1006         .remove =       __devexit_p(virtnet_remove),
1007         .config_changed = virtnet_config_changed,
1008 };
1009
1010 static int __init init(void)
1011 {
1012         return register_virtio_driver(&virtio_net);
1013 }
1014
1015 static void __exit fini(void)
1016 {
1017         unregister_virtio_driver(&virtio_net);
1018 }
1019 module_init(init);
1020 module_exit(fini);
1021
1022 MODULE_DEVICE_TABLE(virtio, id_table);
1023 MODULE_DESCRIPTION("Virtio network driver");
1024 MODULE_LICENSE("GPL");