drivers/vhost/net.c

   1 /* Copyright (C) 2009 Red Hat, Inc.
   2  * Author: Michael S. Tsirkin <mst@redhat.com>
   3  *
   4  * This work is licensed under the terms of the GNU GPL, version 2.
   5  *
   6  * virtio-net server in host kernel.
   7  */
   8
   9 #include <linux/compat.h>
  10 #include <linux/eventfd.h>
  11 #include <linux/vhost.h>
  12 #include <linux/virtio_net.h>
  13 #include <linux/miscdevice.h>
  14 #include <linux/module.h>
  15 #include <linux/moduleparam.h>
  16 #include <linux/mutex.h>
  17 #include <linux/workqueue.h>
  18 #include <linux/file.h>
  19 #include <linux/slab.h>
  20 #include <linux/vmalloc.h>
  21
  22 #include <linux/net.h>
  23 #include <linux/if_packet.h>
  24 #include <linux/if_arp.h>
  25 #include <linux/if_tun.h>
  26 #include <linux/if_macvlan.h>
  27 #include <linux/if_vlan.h>
  28
  29 #include <net/sock.h>
  30
  31 #include "vhost.h"
  32
  33 static int experimental_zcopytx = 1;
  34 module_param(experimental_zcopytx, int, 0444);
  35 MODULE_PARM_DESC(experimental_zcopytx, "Enable Zero Copy TX;"
  36                                        " 1 -Enable; 0 - Disable");
  37
  38 /* Max number of bytes transferred before requeueing the job.
  39  * Using this limit prevents one virtqueue from starving others. */
  40 #define VHOST_NET_WEIGHT 0x80000
  41
  42 /* MAX number of TX used buffers for outstanding zerocopy */
  43 #define VHOST_MAX_PEND 128
  44 #define VHOST_GOODCOPY_LEN 256
  45
  46 /*
  47  * For transmit, used buffer len is unused; we override it to track buffer
  48  * status internally; used for zerocopy tx only.
  49  */
  50 /* Lower device DMA failed */
  51 #define VHOST_DMA_FAILED_LEN    ((__force __virtio32)3)
  52 /* Lower device DMA done */
  53 #define VHOST_DMA_DONE_LEN      ((__force __virtio32)2)
  54 /* Lower device DMA in progress */
  55 #define VHOST_DMA_IN_PROGRESS   ((__force __virtio32)1)
  56 /* Buffer unused */
  57 #define VHOST_DMA_CLEAR_LEN     ((__force __virtio32)0)
  58
  59 #define VHOST_DMA_IS_DONE(len) ((__force u32)(len) >= (__force u32)VHOST_DMA_DONE_LEN)
  60
  61 enum {
  62         VHOST_NET_FEATURES = VHOST_FEATURES |
  63                          (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) |
  64                          (1ULL << VIRTIO_NET_F_MRG_RXBUF) |
  65                          (1ULL << VIRTIO_F_IOMMU_PLATFORM)
  66 };
  67
  68 enum {
  69         VHOST_NET_VQ_RX = 0,
  70         VHOST_NET_VQ_TX = 1,
  71         VHOST_NET_VQ_MAX = 2,
  72 };
  73
  74 struct vhost_net_ubuf_ref {
  75         /* refcount follows semantics similar to kref:
  76          *  0: object is released
  77          *  1: no outstanding ubufs
  78          * >1: outstanding ubufs
  79          */
  80         atomic_t refcount;
  81         wait_queue_head_t wait;
  82         struct vhost_virtqueue *vq;
  83 };
  84
  85 struct vhost_net_virtqueue {
  86         struct vhost_virtqueue vq;
  87         size_t vhost_hlen;
  88         size_t sock_hlen;
  89         /* vhost zerocopy support fields below: */
  90         /* last used idx for outstanding DMA zerocopy buffers */
  91         int upend_idx;
  92         /* first used idx for DMA done zerocopy buffers */
  93         int done_idx;
  94         /* an array of userspace buffers info */
  95         struct ubuf_info *ubuf_info;
  96         /* Reference counting for outstanding ubufs.
  97          * Protected by vq mutex. Writers must also take device mutex. */
  98         struct vhost_net_ubuf_ref *ubufs;
  99 };
 100
 101 struct vhost_net {
 102         struct vhost_dev dev;
 103         struct vhost_net_virtqueue vqs[VHOST_NET_VQ_MAX];
 104         struct vhost_poll poll[VHOST_NET_VQ_MAX];
 105         /* Number of TX recently submitted.
 106          * Protected by tx vq lock. */
 107         unsigned tx_packets;
 108         /* Number of times zerocopy TX recently failed.
 109          * Protected by tx vq lock. */
 110         unsigned tx_zcopy_err;
 111         /* Flush in progress. Protected by tx vq lock. */
 112         bool tx_flush;
 113 };
 114
 115 static unsigned vhost_net_zcopy_mask __read_mostly;
 116
 117 static void vhost_net_enable_zcopy(int vq)
 118 {
 119         vhost_net_zcopy_mask |= 0x1 << vq;
 120 }
 121
 122 static struct vhost_net_ubuf_ref *
 123 vhost_net_ubuf_alloc(struct vhost_virtqueue *vq, bool zcopy)
 124 {
 125         struct vhost_net_ubuf_ref *ubufs;
 126         /* No zero copy backend? Nothing to count. */
 127         if (!zcopy)
 128                 return NULL;
 129         ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL);
 130         if (!ubufs)
 131                 return ERR_PTR(-ENOMEM);
 132         atomic_set(&ubufs->refcount, 1);
 133         init_waitqueue_head(&ubufs->wait);
 134         ubufs->vq = vq;
 135         return ubufs;
 136 }
 137
 138 static int vhost_net_ubuf_put(struct vhost_net_ubuf_ref *ubufs)
 139 {
 140         int r = atomic_sub_return(1, &ubufs->refcount);
 141         if (unlikely(!r))
 142                 wake_up(&ubufs->wait);
 143         return r;
 144 }
 145
 146 static void vhost_net_ubuf_put_and_wait(struct vhost_net_ubuf_ref *ubufs)
 147 {
 148         vhost_net_ubuf_put(ubufs);
 149         wait_event(ubufs->wait, !atomic_read(&ubufs->refcount));
 150 }
 151
 152 static void vhost_net_ubuf_put_wait_and_free(struct vhost_net_ubuf_ref *ubufs)
 153 {
 154         vhost_net_ubuf_put_and_wait(ubufs);
 155         kfree(ubufs);
 156 }
 157
 158 static void vhost_net_clear_ubuf_info(struct vhost_net *n)
 159 {
 160         int i;
 161
 162         for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
 163                 kfree(n->vqs[i].ubuf_info);
 164                 n->vqs[i].ubuf_info = NULL;
 165         }
 166 }
 167
 168 static int vhost_net_set_ubuf_info(struct vhost_net *n)
 169 {
 170         bool zcopy;
 171         int i;
 172
 173         for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
 174                 zcopy = vhost_net_zcopy_mask & (0x1 << i);
 175                 if (!zcopy)
 176                         continue;
 177                 n->vqs[i].ubuf_info = kmalloc(sizeof(*n->vqs[i].ubuf_info) *
 178                                               UIO_MAXIOV, GFP_KERNEL);
 179                 if  (!n->vqs[i].ubuf_info)
 180                         goto err;
 181         }
 182         return 0;
 183
 184 err:
 185         vhost_net_clear_ubuf_info(n);
 186         return -ENOMEM;
 187 }
 188
 189 static void vhost_net_vq_reset(struct vhost_net *n)
 190 {
 191         int i;
 192
 193         vhost_net_clear_ubuf_info(n);
 194
 195         for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
 196                 n->vqs[i].done_idx = 0;
 197                 n->vqs[i].upend_idx = 0;
 198                 n->vqs[i].ubufs = NULL;
 199                 n->vqs[i].vhost_hlen = 0;
 200                 n->vqs[i].sock_hlen = 0;
 201         }
 202
 203 }
 204
 205 static void vhost_net_tx_packet(struct vhost_net *net)
 206 {
 207         ++net->tx_packets;
 208         if (net->tx_packets < 1024)
 209                 return;
 210         net->tx_packets = 0;
 211         net->tx_zcopy_err = 0;
 212 }
 213
 214 static void vhost_net_tx_err(struct vhost_net *net)
 215 {
 216         ++net->tx_zcopy_err;
 217 }
 218
 219 static bool vhost_net_tx_select_zcopy(struct vhost_net *net)
 220 {
 221         /* TX flush waits for outstanding DMAs to be done.
 222          * Don't start new DMAs.
 223          */
 224         return !net->tx_flush &&
 225                 net->tx_packets / 64 >= net->tx_zcopy_err;
 226 }
 227
 228 static bool vhost_sock_zcopy(struct socket *sock)
 229 {
 230         return unlikely(experimental_zcopytx) &&
 231                 sock_flag(sock->sk, SOCK_ZEROCOPY);
 232 }
 233
 234 /* In case of DMA done not in order in lower device driver for some reason.
 235  * upend_idx is used to track end of used idx, done_idx is used to track head
 236  * of used idx. Once lower device DMA done contiguously, we will signal KVM
 237  * guest used idx.
 238  */
 239 static void vhost_zerocopy_signal_used(struct vhost_net *net,
 240                                        struct vhost_virtqueue *vq)
 241 {
 242         struct vhost_net_virtqueue *nvq =
 243                 container_of(vq, struct vhost_net_virtqueue, vq);
 244         int i, add;
 245         int j = 0;
 246
 247         for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) {
 248                 if (vq->heads[i].len == VHOST_DMA_FAILED_LEN)
 249                         vhost_net_tx_err(net);
 250                 if (VHOST_DMA_IS_DONE(vq->heads[i].len)) {
 251                         vq->heads[i].len = VHOST_DMA_CLEAR_LEN;
 252                         ++j;
 253                 } else
 254                         break;
 255         }
 256         while (j) {
 257                 add = min(UIO_MAXIOV - nvq->done_idx, j);
 258                 vhost_add_used_and_signal_n(vq->dev, vq,
 259                                             &vq->heads[nvq->done_idx], add);
 260                 nvq->done_idx = (nvq->done_idx + add) % UIO_MAXIOV;
 261                 j -= add;
 262         }
 263 }
 264
 265 static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success)
 266 {
 267         struct vhost_net_ubuf_ref *ubufs = ubuf->ctx;
 268         struct vhost_virtqueue *vq = ubufs->vq;
 269         int cnt;
 270
 271         rcu_read_lock_bh();
 272
 273         /* set len to mark this desc buffers done DMA */
 274         vq->heads[ubuf->desc].len = success ?
 275                 VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN;
 276         cnt = vhost_net_ubuf_put(ubufs);
 277
 278         /*
 279          * Trigger polling thread if guest stopped submitting new buffers:
 280          * in this case, the refcount after decrement will eventually reach 1.
 281          * We also trigger polling periodically after each 16 packets
 282          * (the value 16 here is more or less arbitrary, it's tuned to trigger
 283          * less than 10% of times).
 284          */
 285         if (cnt <= 1 || !(cnt % 16))
 286                 vhost_poll_queue(&vq->poll);
 287
 288         rcu_read_unlock_bh();
 289 }
 290
 291 static inline unsigned long busy_clock(void)
 292 {
 293         return local_clock() >> 10;
 294 }
 295
 296 static bool vhost_can_busy_poll(struct vhost_dev *dev,
 297                                 unsigned long endtime)
 298 {
 299         return likely(!need_resched()) &&
 300                likely(!time_after(busy_clock(), endtime)) &&
 301                likely(!signal_pending(current)) &&
 302                !vhost_has_work(dev);
 303 }
 304
 305 static int vhost_net_tx_get_vq_desc(struct vhost_net *net,
 306                                     struct vhost_virtqueue *vq,
 307                                     struct iovec iov[], unsigned int iov_size,
 308                                     unsigned int *out_num, unsigned int *in_num)
 309 {
 310         unsigned long uninitialized_var(endtime);
 311         int r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 312                                   out_num, in_num, NULL, NULL);
 313
 314         if (r == vq->num && vq->busyloop_timeout) {
 315                 preempt_disable();
 316                 endtime = busy_clock() + vq->busyloop_timeout;
 317                 while (vhost_can_busy_poll(vq->dev, endtime) &&
 318                        vhost_vq_avail_empty(vq->dev, vq))
 319                         cpu_relax_lowlatency();
 320                 preempt_enable();
 321                 r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 322                                       out_num, in_num, NULL, NULL);
 323         }
 324
 325         return r;
 326 }
 327
 328 /* Expects to be always run from workqueue - which acts as
 329  * read-size critical section for our kind of RCU. */
 330 static void handle_tx(struct vhost_net *net)
 331 {
 332         struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
 333         struct vhost_virtqueue *vq = &nvq->vq;
 334         unsigned out, in;
 335         int head;
 336         struct msghdr msg = {
 337                 .msg_name = NULL,
 338                 .msg_namelen = 0,
 339                 .msg_control = NULL,
 340                 .msg_controllen = 0,
 341                 .msg_flags = MSG_DONTWAIT,
 342         };
 343         size_t len, total_len = 0;
 344         int err;
 345         size_t hdr_size;
 346         struct socket *sock;
 347         struct vhost_net_ubuf_ref *uninitialized_var(ubufs);
 348         bool zcopy, zcopy_used;
 349
 350         mutex_lock(&vq->mutex);
 351         sock = vq->private_data;
 352         if (!sock)
 353                 goto out;
 354
 355         if (!vq_iotlb_prefetch(vq))
 356                 goto out;
 357
 358         vhost_disable_notify(&net->dev, vq);
 359
 360         hdr_size = nvq->vhost_hlen;
 361         zcopy = nvq->ubufs;
 362
 363         for (;;) {
 364                 /* Release DMAs done buffers first */
 365                 if (zcopy)
 366                         vhost_zerocopy_signal_used(net, vq);
 367
 368                 /* If more outstanding DMAs, queue the work.
 369                  * Handle upend_idx wrap around
 370                  */
 371                 if (unlikely((nvq->upend_idx + vq->num - VHOST_MAX_PEND)
 372                               % UIO_MAXIOV == nvq->done_idx))
 373                         break;
 374
 375                 head = vhost_net_tx_get_vq_desc(net, vq, vq->iov,
 376                                                 ARRAY_SIZE(vq->iov),
 377                                                 &out, &in);
 378                 /* On error, stop handling until the next kick. */
 379                 if (unlikely(head < 0))
 380                         break;
 381                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
 382                 if (head == vq->num) {
 383                         if (unlikely(vhost_enable_notify(&net->dev, vq))) {
 384                                 vhost_disable_notify(&net->dev, vq);
 385                                 continue;
 386                         }
 387                         break;
 388                 }
 389                 if (in) {
 390                         vq_err(vq, "Unexpected descriptor format for TX: "
 391                                "out %d, int %d\n", out, in);
 392                         break;
 393                 }
 394                 /* Skip header. TODO: support TSO. */
 395                 len = iov_length(vq->iov, out);
 396                 iov_iter_init(&msg.msg_iter, WRITE, vq->iov, out, len);
 397                 iov_iter_advance(&msg.msg_iter, hdr_size);
 398                 /* Sanity check */
 399                 if (!msg_data_left(&msg)) {
 400                         vq_err(vq, "Unexpected header len for TX: "
 401                                "%zd expected %zd\n",
 402                                len, hdr_size);
 403                         break;
 404                 }
 405                 len = msg_data_left(&msg);
 406
 407                 zcopy_used = zcopy && len >= VHOST_GOODCOPY_LEN
 408                                    && (nvq->upend_idx + 1) % UIO_MAXIOV !=
 409                                       nvq->done_idx
 410                                    && vhost_net_tx_select_zcopy(net);
 411
 412                 /* use msg_control to pass vhost zerocopy ubuf info to skb */
 413                 if (zcopy_used) {
 414                         struct ubuf_info *ubuf;
 415                         ubuf = nvq->ubuf_info + nvq->upend_idx;
 416
 417                         vq->heads[nvq->upend_idx].id = cpu_to_vhost32(vq, head);
 418                         vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS;
 419                         ubuf->callback = vhost_zerocopy_callback;
 420                         ubuf->ctx = nvq->ubufs;
 421                         ubuf->desc = nvq->upend_idx;
 422                         msg.msg_control = ubuf;
 423                         msg.msg_controllen = sizeof(ubuf);
 424                         ubufs = nvq->ubufs;
 425                         atomic_inc(&ubufs->refcount);
 426                         nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV;
 427                 } else {
 428                         msg.msg_control = NULL;
 429                         ubufs = NULL;
 430                 }
 431                 /* TODO: Check specific error and bomb out unless ENOBUFS? */
 432                 err = sock->ops->sendmsg(sock, &msg, len);
 433                 if (unlikely(err < 0)) {
 434                         if (zcopy_used) {
 435                                 vhost_net_ubuf_put(ubufs);
 436                                 nvq->upend_idx = ((unsigned)nvq->upend_idx - 1)
 437                                         % UIO_MAXIOV;
 438                         }
 439                         vhost_discard_vq_desc(vq, 1);
 440                         break;
 441                 }
 442                 if (err != len)
 443                         pr_debug("Truncated TX packet: "
 444                                  " len %d != %zd\n", err, len);
 445                 if (!zcopy_used)
 446                         vhost_add_used_and_signal(&net->dev, vq, head, 0);
 447                 else
 448                         vhost_zerocopy_signal_used(net, vq);
 449                 total_len += len;
 450                 vhost_net_tx_packet(net);
 451                 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
 452                         vhost_poll_queue(&vq->poll);
 453                         break;
 454                 }
 455         }
 456 out:
 457         mutex_unlock(&vq->mutex);
 458 }
 459
 460 static int peek_head_len(struct sock *sk)
 461 {
 462         struct sk_buff *head;
 463         int len = 0;
 464         unsigned long flags;
 465
 466         spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
 467         head = skb_peek(&sk->sk_receive_queue);
 468         if (likely(head)) {
 469                 len = head->len;
 470                 if (skb_vlan_tag_present(head))
 471                         len += VLAN_HLEN;
 472         }
 473
 474         spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags);
 475         return len;
 476 }
 477
 478 static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk)
 479 {
 480         struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
 481         struct vhost_virtqueue *vq = &nvq->vq;
 482         unsigned long uninitialized_var(endtime);
 483         int len = peek_head_len(sk);
 484
 485         if (!len && vq->busyloop_timeout) {
 486                 /* Both tx vq and rx socket were polled here */
 487                 mutex_lock(&vq->mutex);
 488                 vhost_disable_notify(&net->dev, vq);
 489
 490                 preempt_disable();
 491                 endtime = busy_clock() + vq->busyloop_timeout;
 492
 493                 while (vhost_can_busy_poll(&net->dev, endtime) &&
 494                        skb_queue_empty(&sk->sk_receive_queue) &&
 495                        vhost_vq_avail_empty(&net->dev, vq))
 496                         cpu_relax_lowlatency();
 497
 498                 preempt_enable();
 499
 500                 if (vhost_enable_notify(&net->dev, vq))
 501                         vhost_poll_queue(&vq->poll);
 502                 mutex_unlock(&vq->mutex);
 503
 504                 len = peek_head_len(sk);
 505         }
 506
 507         return len;
 508 }
 509
 510 /* This is a multi-buffer version of vhost_get_desc, that works if
 511  *      vq has read descriptors only.
 512  * @vq          - the relevant virtqueue
 513  * @datalen     - data length we'll be reading
 514  * @iovcount    - returned count of io vectors we fill
 515  * @log         - vhost log
 516  * @log_num     - log offset
 517  * @quota       - headcount quota, 1 for big buffer
 518  *      returns number of buffer heads allocated, negative on error
 519  */
 520 static int get_rx_bufs(struct vhost_virtqueue *vq,
 521                        struct vring_used_elem *heads,
 522                        int datalen,
 523                        unsigned *iovcount,
 524                        struct vhost_log *log,
 525                        unsigned *log_num,
 526                        unsigned int quota)
 527 {
 528         unsigned int out, in;
 529         int seg = 0;
 530         int headcount = 0;
 531         unsigned d;
 532         int r, nlogs = 0;
 533         /* len is always initialized before use since we are always called with
 534          * datalen > 0.
 535          */
 536         u32 uninitialized_var(len);
 537
 538         while (datalen > 0 && headcount < quota) {
 539                 if (unlikely(seg >= UIO_MAXIOV)) {
 540                         r = -ENOBUFS;
 541                         goto err;
 542                 }
 543                 r = vhost_get_vq_desc(vq, vq->iov + seg,
 544                                       ARRAY_SIZE(vq->iov) - seg, &out,
 545                                       &in, log, log_num);
 546                 if (unlikely(r < 0))
 547                         goto err;
 548
 549                 d = r;
 550                 if (d == vq->num) {
 551                         r = 0;
 552                         goto err;
 553                 }
 554                 if (unlikely(out || in <= 0)) {
 555                         vq_err(vq, "unexpected descriptor format for RX: "
 556                                 "out %d, in %d\n", out, in);
 557                         r = -EINVAL;
 558                         goto err;
 559                 }
 560                 if (unlikely(log)) {
 561                         nlogs += *log_num;
 562                         log += *log_num;
 563                 }
 564                 heads[headcount].id = cpu_to_vhost32(vq, d);
 565                 len = iov_length(vq->iov + seg, in);
 566                 heads[headcount].len = cpu_to_vhost32(vq, len);
 567                 datalen -= len;
 568                 ++headcount;
 569                 seg += in;
 570         }
 571         heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen);
 572         *iovcount = seg;
 573         if (unlikely(log))
 574                 *log_num = nlogs;
 575
 576         /* Detect overrun */
 577         if (unlikely(datalen > 0)) {
 578                 r = UIO_MAXIOV + 1;
 579                 goto err;
 580         }
 581         return headcount;
 582 err:
 583         vhost_discard_vq_desc(vq, headcount);
 584         return r;
 585 }
 586
 587 /* Expects to be always run from workqueue - which acts as
 588  * read-size critical section for our kind of RCU. */
 589 static void handle_rx(struct vhost_net *net)
 590 {
 591         struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX];
 592         struct vhost_virtqueue *vq = &nvq->vq;
 593         unsigned uninitialized_var(in), log;
 594         struct vhost_log *vq_log;
 595         struct msghdr msg = {
 596                 .msg_name = NULL,
 597                 .msg_namelen = 0,
 598                 .msg_control = NULL, /* FIXME: get and handle RX aux data. */
 599                 .msg_controllen = 0,
 600                 .msg_flags = MSG_DONTWAIT,
 601         };
 602         struct virtio_net_hdr hdr = {
 603                 .flags = 0,
 604                 .gso_type = VIRTIO_NET_HDR_GSO_NONE
 605         };
 606         size_t total_len = 0;
 607         int err, mergeable;
 608         s16 headcount;
 609         size_t vhost_hlen, sock_hlen;
 610         size_t vhost_len, sock_len;
 611         struct socket *sock;
 612         struct iov_iter fixup;
 613         __virtio16 num_buffers;
 614
 615         mutex_lock(&vq->mutex);
 616         sock = vq->private_data;
 617         if (!sock)
 618                 goto out;
 619
 620         if (!vq_iotlb_prefetch(vq))
 621                 goto out;
 622
 623         vhost_disable_notify(&net->dev, vq);
 624
 625         vhost_hlen = nvq->vhost_hlen;
 626         sock_hlen = nvq->sock_hlen;
 627
 628         vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ?
 629                 vq->log : NULL;
 630         mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF);
 631
 632         while ((sock_len = vhost_net_rx_peek_head_len(net, sock->sk))) {
 633                 sock_len += sock_hlen;
 634                 vhost_len = sock_len + vhost_hlen;
 635                 headcount = get_rx_bufs(vq, vq->heads, vhost_len,
 636                                         &in, vq_log, &log,
 637                                         likely(mergeable) ? UIO_MAXIOV : 1);
 638                 /* On error, stop handling until the next kick. */
 639                 if (unlikely(headcount < 0))
 640                         break;
 641                 /* On overrun, truncate and discard */
 642                 if (unlikely(headcount > UIO_MAXIOV)) {
 643                         iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1);
 644                         err = sock->ops->recvmsg(sock, &msg,
 645                                                  1, MSG_DONTWAIT | MSG_TRUNC);
 646                         pr_debug("Discarded rx packet: len %zd\n", sock_len);
 647                         continue;
 648                 }
 649                 /* OK, now we need to know about added descriptors. */
 650                 if (!headcount) {
 651                         if (unlikely(vhost_enable_notify(&net->dev, vq))) {
 652                                 /* They have slipped one in as we were
 653                                  * doing that: check again. */
 654                                 vhost_disable_notify(&net->dev, vq);
 655                                 continue;
 656                         }
 657                         /* Nothing new?  Wait for eventfd to tell us
 658                          * they refilled. */
 659                         break;
 660                 }
 661                 /* We don't need to be notified again. */
 662                 iov_iter_init(&msg.msg_iter, READ, vq->iov, in, vhost_len);
 663                 fixup = msg.msg_iter;
 664                 if (unlikely((vhost_hlen))) {
 665                         /* We will supply the header ourselves
 666                          * TODO: support TSO.
 667                          */
 668                         iov_iter_advance(&msg.msg_iter, vhost_hlen);
 669                 }
 670                 err = sock->ops->recvmsg(sock, &msg,
 671                                          sock_len, MSG_DONTWAIT | MSG_TRUNC);
 672                 /* Userspace might have consumed the packet meanwhile:
 673                  * it's not supposed to do this usually, but might be hard
 674                  * to prevent. Discard data we got (if any) and keep going. */
 675                 if (unlikely(err != sock_len)) {
 676                         pr_debug("Discarded rx packet: "
 677                                  " len %d, expected %zd\n", err, sock_len);
 678                         vhost_discard_vq_desc(vq, headcount);
 679                         continue;
 680                 }
 681                 /* Supply virtio_net_hdr if VHOST_NET_F_VIRTIO_NET_HDR */
 682                 if (unlikely(vhost_hlen)) {
 683                         if (copy_to_iter(&hdr, sizeof(hdr),
 684                                          &fixup) != sizeof(hdr)) {
 685                                 vq_err(vq, "Unable to write vnet_hdr "
 686                                        "at addr %p\n", vq->iov->iov_base);
 687                                 break;
 688                         }
 689                 } else {
 690                         /* Header came from socket; we'll need to patch
 691                          * ->num_buffers over if VIRTIO_NET_F_MRG_RXBUF
 692                          */
 693                         iov_iter_advance(&fixup, sizeof(hdr));
 694                 }
 695                 /* TODO: Should check and handle checksum. */
 696
 697                 num_buffers = cpu_to_vhost16(vq, headcount);
 698                 if (likely(mergeable) &&
 699                     copy_to_iter(&num_buffers, sizeof num_buffers,
 700                                  &fixup) != sizeof num_buffers) {
 701                         vq_err(vq, "Failed num_buffers write");
 702                         vhost_discard_vq_desc(vq, headcount);
 703                         break;
 704                 }
 705                 vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
 706                                             headcount);
 707                 if (unlikely(vq_log))
 708                         vhost_log_write(vq, vq_log, log, vhost_len);
 709                 total_len += vhost_len;
 710                 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
 711                         vhost_poll_queue(&vq->poll);
 712                         break;
 713                 }
 714         }
 715 out:
 716         mutex_unlock(&vq->mutex);
 717 }
 718
 719 static void handle_tx_kick(struct vhost_work *work)
 720 {
 721         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 722                                                   poll.work);
 723         struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
 724
 725         handle_tx(net);
 726 }
 727
 728 static void handle_rx_kick(struct vhost_work *work)
 729 {
 730         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 731                                                   poll.work);
 732         struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
 733
 734         handle_rx(net);
 735 }
 736
 737 static void handle_tx_net(struct vhost_work *work)
 738 {
 739         struct vhost_net *net = container_of(work, struct vhost_net,
 740                                              poll[VHOST_NET_VQ_TX].work);
 741         handle_tx(net);
 742 }
 743
 744 static void handle_rx_net(struct vhost_work *work)
 745 {
 746         struct vhost_net *net = container_of(work, struct vhost_net,
 747                                              poll[VHOST_NET_VQ_RX].work);
 748         handle_rx(net);
 749 }
 750
 751 static int vhost_net_open(struct inode *inode, struct file *f)
 752 {
 753         struct vhost_net *n;
 754         struct vhost_dev *dev;
 755         struct vhost_virtqueue **vqs;
 756         int i;
 757
 758         n = kmalloc(sizeof *n, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
 759         if (!n) {
 760                 n = vmalloc(sizeof *n);
 761                 if (!n)
 762                         return -ENOMEM;
 763         }
 764         vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL);
 765         if (!vqs) {
 766                 kvfree(n);
 767                 return -ENOMEM;
 768         }
 769
 770         dev = &n->dev;
 771         vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq;
 772         vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq;
 773         n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick;
 774         n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick;
 775         for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
 776                 n->vqs[i].ubufs = NULL;
 777                 n->vqs[i].ubuf_info = NULL;
 778                 n->vqs[i].upend_idx = 0;
 779                 n->vqs[i].done_idx = 0;
 780                 n->vqs[i].vhost_hlen = 0;
 781                 n->vqs[i].sock_hlen = 0;
 782         }
 783         vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX);
 784
 785         vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev);
 786         vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev);
 787
 788         f->private_data = n;
 789
 790         return 0;
 791 }
 792
 793 static void vhost_net_disable_vq(struct vhost_net *n,
 794                                  struct vhost_virtqueue *vq)
 795 {
 796         struct vhost_net_virtqueue *nvq =
 797                 container_of(vq, struct vhost_net_virtqueue, vq);
 798         struct vhost_poll *poll = n->poll + (nvq - n->vqs);
 799         if (!vq->private_data)
 800                 return;
 801         vhost_poll_stop(poll);
 802 }
 803
 804 static int vhost_net_enable_vq(struct vhost_net *n,
 805                                 struct vhost_virtqueue *vq)
 806 {
 807         struct vhost_net_virtqueue *nvq =
 808                 container_of(vq, struct vhost_net_virtqueue, vq);
 809         struct vhost_poll *poll = n->poll + (nvq - n->vqs);
 810         struct socket *sock;
 811
 812         sock = vq->private_data;
 813         if (!sock)
 814                 return 0;
 815
 816         return vhost_poll_start(poll, sock->file);
 817 }
 818
 819 static struct socket *vhost_net_stop_vq(struct vhost_net *n,
 820                                         struct vhost_virtqueue *vq)
 821 {
 822         struct socket *sock;
 823
 824         mutex_lock(&vq->mutex);
 825         sock = vq->private_data;
 826         vhost_net_disable_vq(n, vq);
 827         vq->private_data = NULL;
 828         mutex_unlock(&vq->mutex);
 829         return sock;
 830 }
 831
 832 static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock,
 833                            struct socket **rx_sock)
 834 {
 835         *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq);
 836         *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq);
 837 }
 838
 839 static void vhost_net_flush_vq(struct vhost_net *n, int index)
 840 {
 841         vhost_poll_flush(n->poll + index);
 842         vhost_poll_flush(&n->vqs[index].vq.poll);
 843 }
 844
 845 static void vhost_net_flush(struct vhost_net *n)
 846 {
 847         vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
 848         vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
 849         if (n->vqs[VHOST_NET_VQ_TX].ubufs) {
 850                 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
 851                 n->tx_flush = true;
 852                 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
 853                 /* Wait for all lower device DMAs done. */
 854                 vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs);
 855                 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
 856                 n->tx_flush = false;
 857                 atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1);
 858                 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
 859         }
 860 }
 861
 862 static int vhost_net_release(struct inode *inode, struct file *f)
 863 {
 864         struct vhost_net *n = f->private_data;
 865         struct socket *tx_sock;
 866         struct socket *rx_sock;
 867
 868         vhost_net_stop(n, &tx_sock, &rx_sock);
 869         vhost_net_flush(n);
 870         vhost_dev_stop(&n->dev);
 871         vhost_dev_cleanup(&n->dev, false);
 872         vhost_net_vq_reset(n);
 873         if (tx_sock)
 874                 sockfd_put(tx_sock);
 875         if (rx_sock)
 876                 sockfd_put(rx_sock);
 877         /* Make sure no callbacks are outstanding */
 878         synchronize_rcu_bh();
 879         /* We do an extra flush before freeing memory,
 880          * since jobs can re-queue themselves. */
 881         vhost_net_flush(n);
 882         kfree(n->dev.vqs);
 883         kvfree(n);
 884         return 0;
 885 }
 886
 887 static struct socket *get_raw_socket(int fd)
 888 {
 889         struct {
 890                 struct sockaddr_ll sa;
 891                 char  buf[MAX_ADDR_LEN];
 892         } uaddr;
 893         int uaddr_len = sizeof uaddr, r;
 894         struct socket *sock = sockfd_lookup(fd, &r);
 895
 896         if (!sock)
 897                 return ERR_PTR(-ENOTSOCK);
 898
 899         /* Parameter checking */
 900         if (sock->sk->sk_type != SOCK_RAW) {
 901                 r = -ESOCKTNOSUPPORT;
 902                 goto err;
 903         }
 904
 905         r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa,
 906                                &uaddr_len, 0);
 907         if (r)
 908                 goto err;
 909
 910         if (uaddr.sa.sll_family != AF_PACKET) {
 911                 r = -EPFNOSUPPORT;
 912                 goto err;
 913         }
 914         return sock;
 915 err:
 916         sockfd_put(sock);
 917         return ERR_PTR(r);
 918 }
 919
 920 static struct socket *get_tap_socket(int fd)
 921 {
 922         struct file *file = fget(fd);
 923         struct socket *sock;
 924
 925         if (!file)
 926                 return ERR_PTR(-EBADF);
 927         sock = tun_get_socket(file);
 928         if (!IS_ERR(sock))
 929                 return sock;
 930         sock = macvtap_get_socket(file);
 931         if (IS_ERR(sock))
 932                 fput(file);
 933         return sock;
 934 }
 935
 936 static struct socket *get_socket(int fd)
 937 {
 938         struct socket *sock;
 939
 940         /* special case to disable backend */
 941         if (fd == -1)
 942                 return NULL;
 943         sock = get_raw_socket(fd);
 944         if (!IS_ERR(sock))
 945                 return sock;
 946         sock = get_tap_socket(fd);
 947         if (!IS_ERR(sock))
 948                 return sock;
 949         return ERR_PTR(-ENOTSOCK);
 950 }
 951
 952 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
 953 {
 954         struct socket *sock, *oldsock;
 955         struct vhost_virtqueue *vq;
 956         struct vhost_net_virtqueue *nvq;
 957         struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL;
 958         int r;
 959
 960         mutex_lock(&n->dev.mutex);
 961         r = vhost_dev_check_owner(&n->dev);
 962         if (r)
 963                 goto err;
 964
 965         if (index >= VHOST_NET_VQ_MAX) {
 966                 r = -ENOBUFS;
 967                 goto err;
 968         }
 969         vq = &n->vqs[index].vq;
 970         nvq = &n->vqs[index];
 971         mutex_lock(&vq->mutex);
 972
 973         /* Verify that ring has been setup correctly. */
 974         if (!vhost_vq_access_ok(vq)) {
 975                 r = -EFAULT;
 976                 goto err_vq;
 977         }
 978         sock = get_socket(fd);
 979         if (IS_ERR(sock)) {
 980                 r = PTR_ERR(sock);
 981                 goto err_vq;
 982         }
 983
 984         /* start polling new socket */
 985         oldsock = vq->private_data;
 986         if (sock != oldsock) {
 987                 ubufs = vhost_net_ubuf_alloc(vq,
 988                                              sock && vhost_sock_zcopy(sock));
 989                 if (IS_ERR(ubufs)) {
 990                         r = PTR_ERR(ubufs);
 991                         goto err_ubufs;
 992                 }
 993
 994                 vhost_net_disable_vq(n, vq);
 995                 vq->private_data = sock;
 996                 r = vhost_vq_init_access(vq);
 997                 if (r)
 998                         goto err_used;
 999                 r = vhost_net_enable_vq(n, vq);
1000                 if (r)
1001                         goto err_used;
1002
1003                 oldubufs = nvq->ubufs;
1004                 nvq->ubufs = ubufs;
1005
1006                 n->tx_packets = 0;
1007                 n->tx_zcopy_err = 0;
1008                 n->tx_flush = false;
1009         }
1010
1011         mutex_unlock(&vq->mutex);
1012
1013         if (oldubufs) {
1014                 vhost_net_ubuf_put_wait_and_free(oldubufs);
1015                 mutex_lock(&vq->mutex);
1016                 vhost_zerocopy_signal_used(n, vq);
1017                 mutex_unlock(&vq->mutex);
1018         }
1019
1020         if (oldsock) {
1021                 vhost_net_flush_vq(n, index);
1022                 sockfd_put(oldsock);
1023         }
1024
1025         mutex_unlock(&n->dev.mutex);
1026         return 0;
1027
1028 err_used:
1029         vq->private_data = oldsock;
1030         vhost_net_enable_vq(n, vq);
1031         if (ubufs)
1032                 vhost_net_ubuf_put_wait_and_free(ubufs);
1033 err_ubufs:
1034         sockfd_put(sock);
1035 err_vq:
1036         mutex_unlock(&vq->mutex);
1037 err:
1038         mutex_unlock(&n->dev.mutex);
1039         return r;
1040 }
1041
1042 static long vhost_net_reset_owner(struct vhost_net *n)
1043 {
1044         struct socket *tx_sock = NULL;
1045         struct socket *rx_sock = NULL;
1046         long err;
1047         struct vhost_umem *umem;
1048
1049         mutex_lock(&n->dev.mutex);
1050         err = vhost_dev_check_owner(&n->dev);
1051         if (err)
1052                 goto done;
1053         umem = vhost_dev_reset_owner_prepare();
1054         if (!umem) {
1055                 err = -ENOMEM;
1056                 goto done;
1057         }
1058         vhost_net_stop(n, &tx_sock, &rx_sock);
1059         vhost_net_flush(n);
1060         vhost_dev_reset_owner(&n->dev, umem);
1061         vhost_net_vq_reset(n);
1062 done:
1063         mutex_unlock(&n->dev.mutex);
1064         if (tx_sock)
1065                 sockfd_put(tx_sock);
1066         if (rx_sock)
1067                 sockfd_put(rx_sock);
1068         return err;
1069 }
1070
1071 static int vhost_net_set_features(struct vhost_net *n, u64 features)
1072 {
1073         size_t vhost_hlen, sock_hlen, hdr_len;
1074         int i;
1075
1076         hdr_len = (features & ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
1077                                (1ULL << VIRTIO_F_VERSION_1))) ?
1078                         sizeof(struct virtio_net_hdr_mrg_rxbuf) :
1079                         sizeof(struct virtio_net_hdr);
1080         if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) {
1081                 /* vhost provides vnet_hdr */
1082                 vhost_hlen = hdr_len;
1083                 sock_hlen = 0;
1084         } else {
1085                 /* socket provides vnet_hdr */
1086                 vhost_hlen = 0;
1087                 sock_hlen = hdr_len;
1088         }
1089         mutex_lock(&n->dev.mutex);
1090         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1091             !vhost_log_access_ok(&n->dev))
1092                 goto out_unlock;
1093
1094         if ((features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))) {
1095                 if (vhost_init_device_iotlb(&n->dev, true))
1096                         goto out_unlock;
1097         }
1098
1099         for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
1100                 mutex_lock(&n->vqs[i].vq.mutex);
1101                 n->vqs[i].vq.acked_features = features;
1102                 n->vqs[i].vhost_hlen = vhost_hlen;
1103                 n->vqs[i].sock_hlen = sock_hlen;
1104                 mutex_unlock(&n->vqs[i].vq.mutex);
1105         }
1106         mutex_unlock(&n->dev.mutex);
1107         return 0;
1108
1109 out_unlock:
1110         mutex_unlock(&n->dev.mutex);
1111         return -EFAULT;
1112 }
1113
1114 static long vhost_net_set_owner(struct vhost_net *n)
1115 {
1116         int r;
1117
1118         mutex_lock(&n->dev.mutex);
1119         if (vhost_dev_has_owner(&n->dev)) {
1120                 r = -EBUSY;
1121                 goto out;
1122         }
1123         r = vhost_net_set_ubuf_info(n);
1124         if (r)
1125                 goto out;
1126         r = vhost_dev_set_owner(&n->dev);
1127         if (r)
1128                 vhost_net_clear_ubuf_info(n);
1129         vhost_net_flush(n);
1130 out:
1131         mutex_unlock(&n->dev.mutex);
1132         return r;
1133 }
1134
1135 static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
1136                             unsigned long arg)
1137 {
1138         struct vhost_net *n = f->private_data;
1139         void __user *argp = (void __user *)arg;
1140         u64 __user *featurep = argp;
1141         struct vhost_vring_file backend;
1142         u64 features;
1143         int r;
1144
1145         switch (ioctl) {
1146         case VHOST_NET_SET_BACKEND:
1147                 if (copy_from_user(&backend, argp, sizeof backend))
1148                         return -EFAULT;
1149                 return vhost_net_set_backend(n, backend.index, backend.fd);
1150         case VHOST_GET_FEATURES:
1151                 features = VHOST_NET_FEATURES;
1152                 if (copy_to_user(featurep, &features, sizeof features))
1153                         return -EFAULT;
1154                 return 0;
1155         case VHOST_SET_FEATURES:
1156                 if (copy_from_user(&features, featurep, sizeof features))
1157                         return -EFAULT;
1158                 if (features & ~VHOST_NET_FEATURES)
1159                         return -EOPNOTSUPP;
1160                 return vhost_net_set_features(n, features);
1161         case VHOST_RESET_OWNER:
1162                 return vhost_net_reset_owner(n);
1163         case VHOST_SET_OWNER:
1164                 return vhost_net_set_owner(n);
1165         default:
1166                 mutex_lock(&n->dev.mutex);
1167                 r = vhost_dev_ioctl(&n->dev, ioctl, argp);
1168                 if (r == -ENOIOCTLCMD)
1169                         r = vhost_vring_ioctl(&n->dev, ioctl, argp);
1170                 else
1171                         vhost_net_flush(n);
1172                 mutex_unlock(&n->dev.mutex);
1173                 return r;
1174         }
1175 }
1176
1177 #ifdef CONFIG_COMPAT
1178 static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl,
1179                                    unsigned long arg)
1180 {
1181         return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1182 }
1183 #endif
1184
1185 static ssize_t vhost_net_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1186 {
1187         struct file *file = iocb->ki_filp;
1188         struct vhost_net *n = file->private_data;
1189         struct vhost_dev *dev = &n->dev;
1190         int noblock = file->f_flags & O_NONBLOCK;
1191
1192         return vhost_chr_read_iter(dev, to, noblock);
1193 }
1194
1195 static ssize_t vhost_net_chr_write_iter(struct kiocb *iocb,
1196                                         struct iov_iter *from)
1197 {
1198         struct file *file = iocb->ki_filp;
1199         struct vhost_net *n = file->private_data;
1200         struct vhost_dev *dev = &n->dev;
1201
1202         return vhost_chr_write_iter(dev, from);
1203 }
1204
1205 static unsigned int vhost_net_chr_poll(struct file *file, poll_table *wait)
1206 {
1207         struct vhost_net *n = file->private_data;
1208         struct vhost_dev *dev = &n->dev;
1209
1210         return vhost_chr_poll(file, dev, wait);
1211 }
1212
1213 static const struct file_operations vhost_net_fops = {
1214         .owner          = THIS_MODULE,
1215         .release        = vhost_net_release,
1216         .read_iter      = vhost_net_chr_read_iter,
1217         .write_iter     = vhost_net_chr_write_iter,
1218         .poll           = vhost_net_chr_poll,
1219         .unlocked_ioctl = vhost_net_ioctl,
1220 #ifdef CONFIG_COMPAT
1221         .compat_ioctl   = vhost_net_compat_ioctl,
1222 #endif
1223         .open           = vhost_net_open,
1224         .llseek         = noop_llseek,
1225 };
1226
1227 static struct miscdevice vhost_net_misc = {
1228         .minor = VHOST_NET_MINOR,
1229         .name = "vhost-net",
1230         .fops = &vhost_net_fops,
1231 };
1232
1233 static int vhost_net_init(void)
1234 {
1235         if (experimental_zcopytx)
1236                 vhost_net_enable_zcopy(VHOST_NET_VQ_TX);
1237         return misc_register(&vhost_net_misc);
1238 }
1239 module_init(vhost_net_init);
1240
1241 static void vhost_net_exit(void)
1242 {
1243         misc_deregister(&vhost_net_misc);
1244 }
1245 module_exit(vhost_net_exit);
1246
1247 MODULE_VERSION("0.0.1");
1248 MODULE_LICENSE("GPL v2");
1249 MODULE_AUTHOR("Michael S. Tsirkin");
1250 MODULE_DESCRIPTION("Host kernel accelerator for virtio net");
1251 MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR);
1252 MODULE_ALIAS("devname:vhost-net");